app/dma-perf: introduce dma-perf application
PATCH 1/1 app/dma-perf: introduce dma-perf application
There are many high-performance DMA devices supported in DPDK now, and these DMA devices can also be integrated into other modules of DPDK as accelerators, such as Vhost. Before integrating DMA into applications, developers need to know the performance of these DMA devices in various scenarios and the performance of CPUs in the same scenario, such as different buffer lengths. Only in this way can we know the target performance of the application accelerated by using them. This patch introduces a high-performance testing tool, which supports comparing the performance of CPU and DMA in different scenarios automatically with a pre-set config file. Memory Copy performance test are supported for now.
--- app/meson.build | 1 + app/test-dma-perf/benchmark.c | 467 ++++++++++++++++++++++++++++++++++ app/test-dma-perf/config.ini | 56 ++++ app/test-dma-perf/main.c | 445 ++++++++++++++++++++++++++++++++ app/test-dma-perf/main.h | 56 ++++ app/test-dma-perf/meson.build | 17 ++ 6 files changed, 1042 insertions(+) create mode 100644 app/test-dma-perf/benchmark.c create mode 100644 app/test-dma-perf/config.ini create mode 100644 app/test-dma-perf/main.c create mode 100644 app/test-dma-perf/main.h create mode 100644 app/test-dma-perf/meson.build
app/meson.build +1 -0
@@ -19,6 +19,7 @@ apps = [ 'test-cmdline', 'test-compress-perf', 'test-crypto-perf', + 'test-dma-perf', 'test-eventdev', 'test-fib', 'test-flow-perf',
app/test-dma-perf/benchmark.c new file +467 -0
467 lines not on this page · raw message
app/test-dma-perf/config.ini new file +56 -0
56 lines not on this page · raw message
app/test-dma-perf/main.c new file +445 -0
445 lines not on this page · raw message
app/test-dma-perf/main.h new file +56 -0
56 lines not on this page · raw message
app/test-dma-perf/meson.build new file +17 -0
@@ -0,0 +1,17 @@ +# SPDX-License-Identifier: BSD-3-Clause +# Copyright(c) 2019-2023 Intel Corporation + +# meson file, for building this app as part of a main DPDK build. + +if is_windows + build = false + reason = 'not supported on Windows' + subdir_done() +endif + +deps += ['dmadev', 'mbuf', 'cfgfile'] + +sources = files( + 'main.c', + 'benchmark.c', +)
Signature · 1 line
2.35.1
PATCH v2 1/1 app/dma-perf: introduce dma-perf application
There are many high-performance DMA devices supported in DPDK now, and these DMA devices can also be integrated into other modules of DPDK as accelerators, such as Vhost. Before integrating DMA into applications, developers need to know the performance of these DMA devices in various scenarios and the performance of CPUs in the same scenario, such as different buffer lengths. Only in this way can we know the target performance of the application accelerated by using them. This patch introduces a high-performance testing tool, which supports comparing the performance of CPU and DMA in different scenarios automatically with a pre-set config file. Memory Copy performance test are supported for now.
---
v2:
added lcore/dmadev designation;
added error case process;
removed worker_threads parameter from config.ini; improved the logs;
improved config file;
app/meson.build | 1 + app/test-dma-perf/benchmark.c | 471 ++++++++++++++++++++++++++++ app/test-dma-perf/config.ini | 59 ++++ app/test-dma-perf/main.c | 567 ++++++++++++++++++++++++++++++++++ app/test-dma-perf/main.h | 69 +++++ app/test-dma-perf/meson.build | 17 + 6 files changed, 1184 insertions(+) create mode 100644 app/test-dma-perf/benchmark.c create mode 100644 app/test-dma-perf/config.ini create mode 100644 app/test-dma-perf/main.c create mode 100644 app/test-dma-perf/main.h create mode 100644 app/test-dma-perf/meson.build
app/meson.build +1 -0
@@ -19,6 +19,7 @@ apps = [ 'test-cmdline', 'test-compress-perf', 'test-crypto-perf', + 'test-dma-perf', 'test-eventdev', 'test-fib', 'test-flow-perf',
app/test-dma-perf/benchmark.c new file +471 -0
471 lines not on this page · raw message
app/test-dma-perf/config.ini new file +59 -0
59 lines not on this page · raw message
app/test-dma-perf/main.c new file +567 -0
567 lines not on this page · raw message
app/test-dma-perf/main.h new file +69 -0
69 lines not on this page · raw message
app/test-dma-perf/meson.build new file +17 -0
@@ -0,0 +1,17 @@ +# SPDX-License-Identifier: BSD-3-Clause +# Copyright(c) 2019-2023 Intel Corporation + +# meson file, for building this app as part of a main DPDK build. + +if is_windows + build = false + reason = 'not supported on Windows' + subdir_done() +endif + +deps += ['dmadev', 'mbuf', 'cfgfile'] + +sources = files( + 'main.c', + 'benchmark.c', +)
--
2.35.1
PATCH v3 1/1 app/dma-perf: introduce dma-perf application
In reply to Cheng JiangThere are many high-performance DMA devices supported in DPDK now, and these DMA devices can also be integrated into other modules of DPDK as accelerators, such as Vhost. Before integrating DMA into applications, developers need to know the performance of these DMA devices in various scenarios and the performance of CPUs in the same scenario, such as different buffer lengths. Only in this way can we know the target performance of the application accelerated by using them. This patch introduces a high-performance testing tool, which supports comparing the performance of CPU and DMA in different scenarios automatically with a pre-set config file. Memory Copy performance test are supported for now.
---
v3:
fixed some typos;
v2:
added lcore/dmadev designation;
added error case process;
removed worker_threads parameter from config.ini; improved the logs;
improved config file;
app/meson.build | 1 + app/test-dma-perf/benchmark.c | 471 ++++++++++++++++++++++++++++ app/test-dma-perf/config.ini | 59 ++++ app/test-dma-perf/main.c | 567 ++++++++++++++++++++++++++++++++++ app/test-dma-perf/main.h | 69 +++++ app/test-dma-perf/meson.build | 17 + 6 files changed, 1184 insertions(+) create mode 100644 app/test-dma-perf/benchmark.c create mode 100644 app/test-dma-perf/config.ini create mode 100644 app/test-dma-perf/main.c create mode 100644 app/test-dma-perf/main.h create mode 100644 app/test-dma-perf/meson.build
app/meson.build +1 -0
@@ -19,6 +19,7 @@ apps = [ 'test-cmdline', 'test-compress-perf', 'test-crypto-perf', + 'test-dma-perf', 'test-eventdev', 'test-fib', 'test-flow-perf',
app/test-dma-perf/benchmark.c new file +471 -0
471 lines not on this page · raw message
app/test-dma-perf/config.ini new file +59 -0
59 lines not on this page · raw message
app/test-dma-perf/main.c new file +567 -0
567 lines not on this page · raw message
app/test-dma-perf/main.h new file +69 -0
69 lines not on this page · raw message
app/test-dma-perf/meson.build new file +17 -0
@@ -0,0 +1,17 @@ +# SPDX-License-Identifier: BSD-3-Clause +# Copyright(c) 2019-2023 Intel Corporation + +# meson file, for building this app as part of a main DPDK build. + +if is_windows + build = false + reason = 'not supported on Windows' + subdir_done() +endif + +deps += ['dmadev', 'mbuf', 'cfgfile'] + +sources = files( + 'main.c', + 'benchmark.c', +)
--
2.35.1
PATCH v4 1/1 app/dma-perf: introduce dma-perf application
In reply to Cheng JiangThere are many high-performance DMA devices supported in DPDK now, and these DMA devices can also be integrated into other modules of DPDK as accelerators, such as Vhost. Before integrating DMA into applications, developers need to know the performance of these DMA devices in various scenarios and the performance of CPUs in the same scenario, such as different buffer lengths. Only in this way can we know the target performance of the application accelerated by using them. This patch introduces a high-performance testing tool, which supports comparing the performance of CPU and DMA in different scenarios automatically with a pre-set config file. Memory Copy performance test are supported for now.
---
v4:
fixed inaccuracy of the memory footprint display;
v3:
fixed some typos;
v2:
added lcore/dmadev designation;
added error case process;
removed worker_threads parameter from config.ini; improved the logs;
improved config file;
app/meson.build | 1 + app/test-dma-perf/benchmark.c | 471 ++++++++++++++++++++++++++++ app/test-dma-perf/config.ini | 59 ++++ app/test-dma-perf/main.c | 567 ++++++++++++++++++++++++++++++++++ app/test-dma-perf/main.h | 69 +++++ app/test-dma-perf/meson.build | 17 + 6 files changed, 1184 insertions(+) create mode 100644 app/test-dma-perf/benchmark.c create mode 100644 app/test-dma-perf/config.ini create mode 100644 app/test-dma-perf/main.c create mode 100644 app/test-dma-perf/main.h create mode 100644 app/test-dma-perf/meson.build
app/meson.build +1 -0
@@ -19,6 +19,7 @@ apps = [ 'test-cmdline', 'test-compress-perf', 'test-crypto-perf', + 'test-dma-perf', 'test-eventdev', 'test-fib', 'test-flow-perf',
app/test-dma-perf/benchmark.c new file +471 -0
471 lines not on this page · raw message
app/test-dma-perf/config.ini new file +59 -0
59 lines not on this page · raw message
app/test-dma-perf/main.c new file +567 -0
567 lines not on this page · raw message
app/test-dma-perf/main.h new file +69 -0
69 lines not on this page · raw message
app/test-dma-perf/meson.build new file +17 -0
@@ -0,0 +1,17 @@ +# SPDX-License-Identifier: BSD-3-Clause +# Copyright(c) 2019-2023 Intel Corporation + +# meson file, for building this app as part of a main DPDK build. + +if is_windows + build = false + reason = 'not supported on Windows' + subdir_done() +endif + +deps += ['dmadev', 'mbuf', 'cfgfile'] + +sources = files( + 'main.c', + 'benchmark.c', +)
--
2.40.1
Quoted 22 lines
-----Original Message-----
From: Jiang, Cheng1 <[email protected]>
Sent: Thursday, June 8, 2023 1:03 PM
To: [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; Ma, WenwuX <[email protected]>; Wang, YuanX <[email protected]>; He, Xingguang <[email protected]>; Jiang, Cheng1 <[email protected]>
Subject: [PATCH v4] app/dma-perf: introduce dma-perf applicationThere are many high-performance DMA devices supported in DPDK now, and these DMA devices can also be integrated into other modules of DPDK as accelerators, such as Vhost. Before integrating DMA into applications, developers need to know the performance of these DMA devices in various scenarios and the performance of CPUs in the same scenario, such as different buffer lengths. Only in this way can we know the target performance of the application accelerated by using them. This patch introduces a high-performance testing tool, which supports comparing the performance of CPU and DMA in different scenarios automatically with a pre-set config file. Memory Copy performance test are supported for now.
Quoted 4 lines
Signed-off-by Cheng Jiang <[email protected]>Signed-off-by Jiayu Hu <[email protected]>Signed-off-by Yuan Wang <[email protected]>Acked-by Morten Brørup <[email protected]>
Some LONG_LINE warning could be fixed if it's friendly for searching logs.
I didn't look into all details but overall it seems good.
Quoted 25 lines
---
v4:
fixed inaccuracy of the memory footprint display;
v3:
fixed some typos;
v2:
added lcore/dmadev designation;
added error case process;
removedworker_threadsparameter from config.ini; improved the logs;
improved config file;app/meson.build | 1 + app/test-dma-perf/benchmark.c | 471 ++++++++++++++++++++++++++++ app/test-dma-perf/config.ini | 59 ++++ app/test-dma-perf/main.c | 567 ++++++++++++++++++++++++++++++++++ app/test-dma-perf/main.h | 69 +++++ app/test-dma-perf/meson.build | 17 + 6 files changed, 1184 insertions(+) create mode 100644 app/test-dma-perf/benchmark.c create mode 100644 app/test-dma-perf/config.ini create mode 100644 app/test-dma-perf/main.c create mode 100644 app/test-dma-perf/main.h create mode 100644 app/test-dma-perf/meson.build
Hi Chenbo,
Quoted 40 lines
-----Original Message-----
From: Xia, Chenbo <[email protected]>
Sent: Thursday, June 8, 2023 4:28 PM
To: Jiang, Cheng1 <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; Ma, WenwuX <[email protected]>; Wang, YuanX <[email protected]>; He, Xingguang <[email protected]>
Subject: RE: [PATCH v4] app/dma-perf: introduce dma-perf applicationQuoted 8 lines
-----Original Message-----
From: Jiang, Cheng1 <[email protected]>
Sent: Thursday, June 8, 2023 1:03 PM
To: [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; Ma, WenwuX <[email protected]>; Wang,YuanX
Quoted 4 lines
<[email protected]>; He, Xingguang <[email protected]>; Jiang, Cheng1 <[email protected]>
Subject: [PATCH v4] app/dma-perf: introduce dma-perf applicationQuoted 10 lines
There are many high-performance DMA devices supported in DPDK now, and these DMA devices can also be integrated into other modules of DPDK as accelerators, such as Vhost. Before integrating DMA into applications, developers need to know the performance of these DMA devices in various scenarios and the performance of CPUs in the same scenario, such as different buffer lengths. Only in this way can we know the target performance of the application accelerated by using them. This patch introduces a high-performance testing tool, which supports comparing the performance of CPU and DMA in different scenarios automatically with a pre-set config file. Memory Copy performance test are
supported for now.
Quoted 5 lines
Signed-off-by Cheng Jiang <[email protected]>Signed-off-by Jiayu Hu <[email protected]>Signed-off-by Yuan Wang <[email protected]>Acked-by Morten Brørup <[email protected]>
Quoted 5 lines
Some
LONG_LINEwarning could be fixed if it's friendly for searching logs.I didn't look into all details but overall it seems good.
Acked-by Chenbo Xia <[email protected]>
Sure, I will fix it in the next version.
Thanks a lot.
Cheng
Quoted 25 lines
Quoted 25 lines
---
v4:
fixed inaccuracy of the memory footprint display;
v3:
fixed some typos;
v2:
added lcore/dmadev designation;
added error case process;
removedworker_threadsparameter from config.ini; improved the logs;
improved config file;app/meson.build | 1 + app/test-dma-perf/benchmark.c | 471 ++++++++++++++++++++++++++++ app/test-dma-perf/config.ini | 59 ++++ app/test-dma-perf/main.c | 567 ++++++++++++++++++++++++++++++++++ app/test-dma-perf/main.h | 69 +++++ app/test-dma-perf/meson.build | 17 + 6 files changed, 1184 insertions(+) create mode 100644 app/test-dma-perf/benchmark.c create mode 100644app/test-dma-perf/config.ini create mode 100644
app/test-dma-perf/main.ccreate mode 100644app/test-dma-perf/main.hcreate mode 100644 app/test-dma-perf/meson.build
PATCH v5 1/1 app/dma-perf: introduce dma-perf application
In reply to Cheng JiangThere are many high-performance DMA devices supported in DPDK now, and these DMA devices can also be integrated into other modules of DPDK as accelerators, such as Vhost. Before integrating DMA into applications, developers need to know the performance of these DMA devices in various scenarios and the performance of CPUs in the same scenario, such as different buffer lengths. Only in this way can we know the target performance of the application accelerated by using them. This patch introduces a high-performance testing tool, which supports comparing the performance of CPU and DMA in different scenarios automatically with a pre-set config file. Memory Copy performance test are supported for now.
---
v5:
fixed some LONG_LINE warnings;
v4:
fixed inaccuracy of the memory footprint display;
v3:
fixed some typos;
v2:
added lcore/dmadev designation;
added error case process;
removed worker_threads parameter from config.ini; improved the logs;
improved config file;
app/meson.build | 1 + app/test-dma-perf/benchmark.c | 472 ++++++++++++++++++++++++++++ app/test-dma-perf/config.ini | 59 ++++ app/test-dma-perf/main.c | 569 ++++++++++++++++++++++++++++++++++ app/test-dma-perf/main.h | 69 +++++ app/test-dma-perf/meson.build | 17 + 6 files changed, 1187 insertions(+) create mode 100644 app/test-dma-perf/benchmark.c create mode 100644 app/test-dma-perf/config.ini create mode 100644 app/test-dma-perf/main.c create mode 100644 app/test-dma-perf/main.h create mode 100644 app/test-dma-perf/meson.build
app/meson.build +1 -0
@@ -19,6 +19,7 @@ apps = [ 'test-cmdline', 'test-compress-perf', 'test-crypto-perf', + 'test-dma-perf', 'test-eventdev', 'test-fib', 'test-flow-perf',
app/test-dma-perf/benchmark.c new file +472 -0
472 lines not on this page · raw message
app/test-dma-perf/config.ini new file +59 -0
59 lines not on this page · raw message
app/test-dma-perf/main.c new file +569 -0
569 lines not on this page · raw message
app/test-dma-perf/main.h new file +69 -0
69 lines not on this page · raw message
app/test-dma-perf/meson.build new file +17 -0
@@ -0,0 +1,17 @@ +# SPDX-License-Identifier: BSD-3-Clause +# Copyright(c) 2019-2023 Intel Corporation + +# meson file, for building this app as part of a main DPDK build. + +if is_windows + build = false + reason = 'not supported on Windows' + subdir_done() +endif + +deps += ['dmadev', 'mbuf', 'cfgfile'] + +sources = files( + 'main.c', + 'benchmark.c', +)
--
2.40.1
Hi,
Thanks for adding the app. Few comments inline. Please check.
Thanks,
Anoob
Quoted 56 lines
-----Original Message-----
From: Cheng Jiang <[email protected]>
Sent: Thursday, June 8, 2023 2:14 PM
To: [email protected]; [email protected]; [email protected]; [email protected]
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; Cheng Jiang <[email protected]>
Subject: [EXT] [PATCH v5] app/dma-perf: introduce dma-perf applicationExternal Email
----------------------------------------------------------------------
There are many high-performance DMA devices supported in DPDK now, and these DMA devices can also be integrated into other modules of DPDK as accelerators, such as Vhost. Before integrating DMA into applications, developers need to know the performance of these DMA devices in various scenarios and the performance of CPUs in the same scenario, such as different buffer lengths. Only in this way can we know the target performance of the application accelerated by using them. This patch introduces a high-performance testing tool, which supports comparing the performance of CPU and DMA in different scenarios automatically with a pre- set config file. Memory Copy performance test are supported for now.Signed-off-by Cheng Jiang <[email protected]>Signed-off-by Jiayu Hu <[email protected]>Signed-off-by Yuan Wang <[email protected]>Acked-by Morten Brørup <[email protected]>Acked-by Chenbo Xia <[email protected]>---
v5:
fixed someLONG_LINEwarnings;
v4:
fixed inaccuracy of the memory footprint display;
v3:
fixed some typos;
v2:
added lcore/dmadev designation;
added error case process;
removedworker_threadsparameter from config.ini; improved the logs;
improved config file;app/meson.build | 1 + app/test-dma-perf/benchmark.c | 472 ++++++++++++++++++++++++++++ app/test-dma-perf/config.ini | 59 ++++ app/test-dma-perf/main.c | 569 ++++++++++++++++++++++++++++++++++ app/test-dma-perf/main.h | 69 +++++ app/test-dma-perf/meson.build | 17 + 6 files changed, 1187 insertions(+) create mode 100644 app/test-dma-perf/benchmark.c create mode 100644app/test-dma-perf/config.ini create mode 100644 app/test-dma-
perf/main.ccreate mode 100644app/test-dma-perf/main.hcreate mode 100644 app/test-dma-perf/meson.build
<snip>
app/test-dma-perf/benchmark.c line 129
Quoted 7 lines
app/test-dma-perf/benchmark.c new file +7
-0+ +/* Configuration of device. */ +static void +configure_dmadev_queue (uint32_t dev_id,uint32_t ring_size) { +uint16_t vchan =0 ; +struct rte_dma_info info; +struct rte_dma_conf dev_config = { .nb_vchans =1 };
[Anoob] Is it possible to use more vchans? The code launches as many threads as the number of dma devices. Instead it should be total number of vchans.
Quoted 21 lines
app/test-dma-perf/benchmark.c new file +14
-0+struct rte_dma_vchan_conf qconf = { + .direction =RTE_DMA_DIR_MEM_TO_MEM , + .nb_desc = ring_size + }; + +if (rte_dma_configure (dev_id, &dev_config) !=0 ) +rte_exit (EXIT_FAILURE ,"Error with dma configure.\n" ); + +if (rte_dma_vchan_setup (dev_id, vchan, &qconf) !=0 ) +rte_exit (EXIT_FAILURE ,"Error with queue configuration.\n" ); + +rte_dma_info_get (dev_id, &info); +if (info.nb_vchans !=1 ) + rte_exit(EXIT_FAILURE, "Error, no configured queuesreported on device id. %u\n",
app/test-dma-perf/benchmark.c new file +6
-0+ dev_id); + +if (rte_dma_start (dev_id) !=0 ) +rte_exit (EXIT_FAILURE ,"Error with dma start.\n" ); } + +
<snip>
app/test-dma-perf/benchmark.c line 195
Quoted 9 lines
app/test-dma-perf/benchmark.c new file +9
-0+static inline int +do_dma_mem_copy (void *p) +{ +uint16_t *para_idx = (uint16_t *)p; +volatile struct lcore_params *para = worker_params[*para_idx]; +volatile struct worker_info *worker_info = &(para->worker_info); +uint16_t dev_id = para->dev_id; +uint32_t nr_buf = para->nr_buf; +uint16_t kick_batch = para->kick_batch;
[Anoob] Some of these variables can be made const. Since this is fast path, might be beneficial doing that way.
Quoted 43 lines
app/test-dma-perf/benchmark.c new file +24
-0+uint32_t buf_size = para->buf_size; +struct rte_mbuf **srcs = para->srcs; +struct rte_mbuf **dsts = para->dsts; +int64_t async_cnt =0 ; +int nr_cpl =0 ; +uint32_t i; +uint32_t poll_cnt =0 ; + + worker_info->stop_flag = false; + worker_info->ready_flag = true; + +while (!worker_info->start_flag) + ; + +while (1 ) { +for (i =0 ; i < nr_buf; i++) { +if (unlikely (rte_dma_copy (dev_id, +0 , +rte_pktmbuf_iova (srcs[i]), +rte_pktmbuf_iova (dsts[i]), + buf_size, +0 ) <0 )) { +rte_dma_submit (dev_id,0 ); + while (rte_dma_burst_capacity(dev_id, 0) ==0) {
(quoted diff) +1
-0+ nr_cpl = rte_dma_completed(dev_id,0,
MAX_DMA_CPL_NB,app/test-dma-perf/benchmark.c new file +11
-0+ NULL, NULL); + async_cnt -= nr_cpl; + worker_info->total_cpl += nr_cpl; + } +if (rte_dma_copy (dev_id, +0 , +rte_pktmbuf_iova (srcs[i]), +rte_pktmbuf_iova (dsts[i]), + buf_size, +0 ) <0 ) { + printf("enqueue fail again at %u\n",i);
(quoted diff) +1
-0+ printf("space:%d\n",
rte_dma_burst_capacity(dev_id, 0));(quoted diff) +1
-0+ rte_exit(EXIT_FAILURE, "DMAenqueue failed\n");
Review
Quoted 1 line
(quoted diff) +1
-0+ }
[Anoob] Only if the API returns -ENOSPC we should retry submitting, right? Other errors should be treated as fatal errors.
Do we need to use rte_dma_burst_capacity() API?
Can't we try something like,
dma_copy:
ret = rte_dma_copy(dev_id, 0, rte_pktmbuf_iova(srcs[i]), rte_pktmbuf_iova(dsts[i]), buf_size, 0);
if (unlikely (ret < 0) {
if (ret == -ENOSPC) {
rte_dma_submit(dev_id, 0);
/* DMA completed & other handling */
goto dma_copy;
} else {
/* Error exit */
}
}Review
Quoted 10 lines
app/test-dma-perf/benchmark.c new file +7
-0+ } + async_cnt++; + +if ((async_cnt % kick_batch) ==0 ) { +rte_dma_submit (dev_id,0 ); +/* add a poll to avoid ring full */ + nr_cpl = rte_dma_completed(dev_id, 0,
MAX_DMA_CPL_NB, NULL, NULL);(quoted diff) +2
-0+ async_cnt -= nr_cpl; + worker_info->total_cpl += nr_cpl;
[Anoob] Above code can be made as a static inline function so that in cases rte_dma_copy returns -ENOSPC, same static inline can be called.
<snip>
Hi,
Thanks for your comments, the replies are inline.
Thanks,
Cheng
Quoted 86 lines
-----Original Message-----
From: Anoob Joseph <[email protected]>
Sent: Friday, June 9, 2023 7:44 PM
To: Jiang, Cheng1 <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; Ma, WenwuX <[email protected]>; Wang, YuanX <[email protected]>; He, Xingguang <[email protected]>; Jerin Jacob Kollanukkaran <[email protected]>; Vamsi Krishna Attunuru <[email protected]>; Amit Prakash Shukla <[email protected]>; Satha Koteswara Rao Kottidi <[email protected]>; Gowrishankar Muthukrishnan <[email protected]>; Vidya Sagar Velumuri <[email protected]>
Subject: RE: [EXT] [PATCH v5] app/dma-perf: introduce dma-perf applicationHi,
Thanks for adding the app. Few comments inline. Please check.
Thanks,
AnoobQuoted 10 lines
-----Original Message-----
From: Cheng Jiang <[email protected]>
Sent: Thursday, June 8, 2023 2:14 PM
To: [email protected]; [email protected]; [email protected]; [email protected]
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; Cheng Jiang <[email protected]>
Subject: [EXT] [PATCH v5] app/dma-perf: introduce dma-perf applicationQuoted 4 lines
External Email
----------------------------------------------------------------------
There are many high-performance DMA devices supported in DPDK now,and
Quoted 9 lines
these DMA devices can also be integrated into other modules of DPDK as accelerators, such as Vhost. Before integrating DMA into applications, developers need to know the performance of these DMA devices in various scenarios and the performance of CPUs in the same scenario, such as different buffer lengths. Only in this way can we know the target performance of the application accelerated by using them. This patch introduces a high-performance testing tool, which supports comparing the performance of CPU and DMA in different scenarios automatically with a pre- set config file. Memory Copy performance test
are supported for now.
Quoted 20 lines
Signed-off-by Cheng Jiang <[email protected]>Signed-off-by Jiayu Hu <[email protected]>Signed-off-by Yuan Wang <[email protected]>Acked-by Morten Brørup <[email protected]>Acked-by Chenbo Xia <[email protected]>---
v5:
fixed someLONG_LINEwarnings;
v4:
fixed inaccuracy of the memory footprint display;
v3:
fixed some typos;
v2:
added lcore/dmadev designation;
added error case process;
removedworker_threadsparameter from config.ini; improved the logs;
improved config file;Quoted 9 lines
app/meson.build | 1 + app/test-dma-perf/benchmark.c | 472 ++++++++++++++++++++++++++++ app/test-dma-perf/config.ini | 59 ++++ app/test-dma-perf/main.c | 569 ++++++++++++++++++++++++++++++++++ app/test-dma-perf/main.h | 69 +++++ app/test-dma-perf/meson.build | 17 + 6 files changed, 1187 insertions(+) create mode 100644 app/test-dma-perf/benchmark.c create mode100644
Quoted 4 lines
app/test-dma-perf/config.ini create mode 100644 app/test-dma-
perf/main.ccreate mode 100644app/test-dma-perf/main.hcreate mode 100644 app/test-dma-perf/meson.build<snip>
Quoted 11 lines
app/test-dma-perf/benchmark.c line 129
Quoted 7 lines
app/test-dma-perf/benchmark.c new file +7
-0+ +/* Configuration of device. */ +static void +configure_dmadev_queue (uint32_t dev_id,uint32_t ring_size) { +uint16_t vchan =0 ; +struct rte_dma_info info; +struct rte_dma_conf dev_config = { .nb_vchans =1 };[Anoob] Is it possible to use more vchans? The code launches as many threads as the number of dma devices. Instead it should be total number of vchans.
[Cheng] Really good suggestion. This is feasible, but in the initial stage, we want to keep things simple. Perhaps in the future, we can add a parameter to configure the number of vchans for each device and then launch the corresponding number of threads for each vchan.
Quoted 25 lines
Quoted 22 lines
app/test-dma-perf/benchmark.c new file +14
-0+struct rte_dma_vchan_conf qconf = { + .direction =RTE_DMA_DIR_MEM_TO_MEM , + .nb_desc = ring_size + }; + +if (rte_dma_configure (dev_id, &dev_config) !=0 ) +rte_exit (EXIT_FAILURE ,"Error with dma configure.\n" ); + +if (rte_dma_vchan_setup (dev_id, vchan, &qconf) !=0 ) +rte_exit (EXIT_FAILURE ,"Error with queue configuration.\n" ); + +rte_dma_info_get (dev_id, &info); +if (info.nb_vchans !=1 ) + rte_exit(EXIT_FAILURE, "Error, no configured queuesreported on device id. %u\n",
app/test-dma-perf/benchmark.c new file +6
-0+ dev_id); + +if (rte_dma_start (dev_id) !=0 ) +rte_exit (EXIT_FAILURE ,"Error with dma start.\n" ); } + +<snip>
Quoted 12 lines
app/test-dma-perf/benchmark.c line 195
Quoted 9 lines
app/test-dma-perf/benchmark.c new file +9
-0+static inline int +do_dma_mem_copy (void *p) +{ +uint16_t *para_idx = (uint16_t *)p; +volatile struct lcore_params *para = worker_params[*para_idx]; +volatile struct worker_info *worker_info = &(para->worker_info); +uint16_t dev_id = para->dev_id; +uint32_t nr_buf = para->nr_buf; +uint16_t kick_batch = para->kick_batch;[Anoob] Some of these variables can be made const. Since this is fast path, might be beneficial doing that way.
[Cheng] Good idea, I'll improve it in the next version.
Quoted 54 lines
Quoted 44 lines
app/test-dma-perf/benchmark.c new file +24
-0+uint32_t buf_size = para->buf_size; +struct rte_mbuf **srcs = para->srcs; +struct rte_mbuf **dsts = para->dsts; +int64_t async_cnt =0 ; +int nr_cpl =0 ; +uint32_t i; +uint32_t poll_cnt =0 ; + + worker_info->stop_flag = false; + worker_info->ready_flag = true; + +while (!worker_info->start_flag) + ; + +while (1 ) { +for (i =0 ; i < nr_buf; i++) { +if (unlikely (rte_dma_copy (dev_id, +0 , +rte_pktmbuf_iova (srcs[i]), +rte_pktmbuf_iova (dsts[i]), + buf_size, +0 ) <0 )) { +rte_dma_submit (dev_id,0 ); + while (rte_dma_burst_capacity(dev_id, 0) ==0) {
(quoted diff) +1
-0+ nr_cpl = rte_dma_completed(dev_id,0,
MAX_DMA_CPL_NB,app/test-dma-perf/benchmark.c new file +11
-0+ NULL, NULL); + async_cnt -= nr_cpl; + worker_info->total_cpl += nr_cpl; + } +if (rte_dma_copy (dev_id, +0 , +rte_pktmbuf_iova (srcs[i]), +rte_pktmbuf_iova (dsts[i]), + buf_size, +0 ) <0 ) { + printf("enqueue fail again at %u\n",i);
(quoted diff) +1
-0+ printf("space:%d\n",
rte_dma_burst_capacity(dev_id, 0));(quoted diff) +1
-0+ rte_exit(EXIT_FAILURE, "DMAenqueue failed\n");
Review
Quoted 1 line
(quoted diff) +1
-0+ }[Anoob] Only if the API returns
-ENOSPCwe should retry submitting, right? Other errors should be treated as fatal errors.Do we need to use
rte_dma_burst_capacity()API?Can't we try something like,
dma_copy:
Quoted 13 lines
ret = rte_dma_copy(dev_id, 0, rte_pktmbuf_iova(srcs[i]),
rte_pktmbuf_iova(dsts[i]), buf_size, 0);
if (unlikely (ret < 0) {
if (ret == -ENOSPC) {
rte_dma_submit(dev_id, 0);
/* DMA completed & other handling */
goto dma_copy;
} else {
/* Error exit */
}
}[Cheng] Good idea, we don't have to check the capacity explicitly. I think your implementation is more clear, thanks. I will fix it in the next version.
Quoted 5 lines
Quoted 5 lines
app/test-dma-perf/benchmark.c new file +5
-0+ } + async_cnt++; + +if ((async_cnt % kick_batch) ==0 ) { +rte_dma_submit (dev_id,0 );
Quoted 9 lines
Review
Quoted 5 lines
(quoted diff) +2
-0+ /* add a poll to avoid ring full */ + nr_cpl = rte_dma_completed(dev_id, 0,
MAX_DMA_CPL_NB, NULL, NULL);(quoted diff) +2
-0+ async_cnt -= nr_cpl; + worker_info->total_cpl += nr_cpl;[Anoob] Above code can be made as a static inline function so that in cases
rte_dma_copyreturns-ENOSPC, same static inline can be called.
[Cheng] sure, got it. Thanks!
Quoted 1 line
<snip>
Quoted 58 lines
-----Original Message-----
From: Cheng Jiang <[email protected]>
Sent: Thursday, June 8, 2023 2:14 PM
To: [email protected]; [email protected]; [email protected]; [email protected]
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; Cheng Jiang <[email protected]>
Subject: [EXT] [PATCH v5] app/dma-perf: introduce dma-perf applicationExternal Email
----------------------------------------------------------------------
There are many high-performance DMA devices supported in DPDK now, and these DMA devices can also be integrated into other modules of DPDK as accelerators, such as Vhost. Before integrating DMA into applications, developers need to know the performance of these DMA devices in various scenarios and the performance of CPUs in the same scenario, such as different buffer lengths. Only in this way can we know the target performance of the application accelerated by using them. This patch introduces a high-performance testing tool, which supports comparing the performance of CPU and DMA in different scenarios automatically with a pre- set config file. Memory Copy performance test are supported for now.Signed-off-by Cheng Jiang <[email protected]>Signed-off-by Jiayu Hu <[email protected]>Signed-off-by Yuan Wang <[email protected]>Acked-by Morten Brørup <[email protected]>Acked-by Chenbo Xia <[email protected]>---
v5:
fixed someLONG_LINEwarnings;
v4:
fixed inaccuracy of the memory footprint display;
v3:
fixed some typos;
v2:
added lcore/dmadev designation;
added error case process;
removedworker_threadsparameter from config.ini; improved the logs;
improved config file;app/meson.build | 1 + app/test-dma-perf/benchmark.c | 472 ++++++++++++++++++++++++++++ app/test-dma-perf/config.ini | 59 ++++ app/test-dma-perf/main.c | 569 ++++++++++++++++++++++++++++++++++ app/test-dma-perf/main.h | 69 +++++ app/test-dma-perf/meson.build | 17 + 6 files changed, 1187 insertions(+) create mode 100644 app/test-dma-perf/benchmark.c create mode 100644app/test-dma-perf/config.ini create mode 100644 app/test-dma-
perf/main.ccreate mode 100644app/test-dma-perf/main.hcreate mode 100644 app/test-dma-perf/meson.build
<snip>
Quoted 52 lines
app/test-dma-perf/benchmark.c new file +34
-0+ +static inline int +do_dma_mem_copy (void *p) +{ +uint16_t *para_idx = (uint16_t *)p; +volatile struct lcore_params *para = worker_params[*para_idx]; +volatile struct worker_info *worker_info = &(para->worker_info); +uint16_t dev_id = para->dev_id; +uint32_t nr_buf = para->nr_buf; +uint16_t kick_batch = para->kick_batch; +uint32_t buf_size = para->buf_size; +struct rte_mbuf **srcs = para->srcs; +struct rte_mbuf **dsts = para->dsts; +int64_t async_cnt =0 ; +int nr_cpl =0 ; +uint32_t i; +uint32_t poll_cnt =0 ; + + worker_info->stop_flag = false; + worker_info->ready_flag = true; + +while (!worker_info->start_flag) + ; + +while (1 ) { +for (i =0 ; i < nr_buf; i++) { +if (unlikely (rte_dma_copy (dev_id, +0 , +rte_pktmbuf_iova (srcs[i]), +rte_pktmbuf_iova (dsts[i]), + buf_size, +0 ) <0 )) { +rte_dma_submit (dev_id,0 ); + while (rte_dma_burst_capacity(dev_id, 0) ==0) {
(quoted diff) +1
-0+ nr_cpl = rte_dma_completed(dev_id,0,
MAX_DMA_CPL_NB,app/test-dma-perf/benchmark.c new file +11
-0+ NULL, NULL); + async_cnt -= nr_cpl; + worker_info->total_cpl += nr_cpl; + } +if (rte_dma_copy (dev_id, +0 , +rte_pktmbuf_iova (srcs[i]), +rte_pktmbuf_iova (dsts[i]), + buf_size, +0 ) <0 ) { + printf("enqueue fail again at %u\n",i);
(quoted diff) +1
-0+ printf("space:%d\n",
rte_dma_burst_capacity(dev_id, 0));(quoted diff) +1
-0+ rte_exit(EXIT_FAILURE, "DMA
Quoted 1 line
enqueue failed\n");
[Amit]: On all success or failure exits, please call rte_dma_stop and rte_dma_close to exit cleanly.
Quoted 28 lines
app/test-dma-perf/benchmark.c new file +8
-0+ } + } + async_cnt++; + +if ((async_cnt % kick_batch) ==0 ) { +rte_dma_submit (dev_id,0 ); +/* add a poll to avoid ring full */ + nr_cpl = rte_dma_completed(dev_id, 0,
MAX_DMA_CPL_NB, NULL, NULL);app/test-dma-perf/benchmark.c new file +12
-0+ async_cnt -= nr_cpl; + worker_info->total_cpl += nr_cpl; + } + } + +if (worker_info->stop_flag) +break ; + } + +rte_dma_submit (dev_id,0 ); +while ((async_cnt >0 ) && (poll_cnt++ <POLL_MAX )) { + nr_cpl = rte_dma_completed(dev_id, 0,
MAX_DMA_CPL_NB, NULL, NULL);app/test-dma-perf/benchmark.c new file +6
-0+ async_cnt -= nr_cpl; + } + +return 0 ; +} +
<snip>
Quoted 119 lines
app/test-dma-perf/benchmark.c new file +11
-0+ +void +mem_copy_benchmark (struct test_configure *cfg,bool is_dma) { +uint16_t i; +uint32_t offset; +unsigned int lcore_id =0 ; +struct rte_mbuf **srcs =NULL , **dsts =NULL ; +struct lcore_dma_map_t *ldm = &cfg->lcore_dma_map; +unsigned int buf_size = cfg->buf_size.cur; +uint16_t kick_batch = cfg->kick_batch.cur; + uint32_t nr_buf = cfg->nr_buf = (cfg->mem_size.cur * 1024 * 1024) /(
cfg->buf_size.cur* 2);app/test-dma-perf/benchmark.c new file +27
-0+uint16_t nb_workers = ldm->cnt; +uint16_t test_secs = cfg->test_secs; +float memory; +uint32_t avg_cycles =0 ; +float mops; +float bandwidth; + +if (setup_memory_env (cfg, &srcs, &dsts) <0 ) + goto out; + +if (is_dma) +if (config_dmadevs (cfg) <0 ) + goto out; + +if (cfg->cache_flush) { +cache_flush_buf (srcs, buf_size, nr_buf); +cache_flush_buf (dsts, buf_size, nr_buf); +rte_mb (); + } + +printf ("Start testing....\n" ); + +for (i =0 ; i < nb_workers; i++) { + lcore_id = ldm->lcores[i]; + offset = nr_buf / nb_workers * i; + + worker_params[i] = rte_malloc(NULL, sizeof(struct
lcore_params), 0);(quoted diff) +2
-0+ if (!worker_params[i]) { + printf("lcore parameters malloc failure for lcore%d\n",
lcore_id);app/test-dma-perf/benchmark.c new file +4
-0+break ; + } +if (is_dma) { + worker_params[i]->dma_name = ldm-Quoted 1 line
dma_names[i];
app/test-dma-perf/benchmark.c new file +5
-0+ worker_params[i]->dev_id = ldm->dma_ids[i]; + worker_params[i]->kick_batch = kick_batch; + } + worker_params[i]->worker_id = i; + worker_params[i]->nr_buf = (uint32_t)(nr_buf /
nb_workers);app/test-dma-perf/benchmark.c new file +9
-0+ worker_params[i]->buf_size = buf_size; + worker_params[i]->test_secs = test_secs; + worker_params[i]->srcs = srcs + offset; + worker_params[i]->dsts = dsts + offset; + worker_params[i]->scenario_id = cfg->scenario_id; + worker_params[i]->lcore_id = lcore_id; + +if (is_dma) + rte_eal_remote_launch(do_dma_mem_copy, (void*)(&i),
lcore_id);(quoted diff) +2
-0+ else + rte_eal_remote_launch(do_cpu_mem_copy, (void*)(&i),
lcore_id);app/test-dma-perf/benchmark.c new file +6
-0+ } + +while (1 ) { +bool ready = true; +for (i =0 ; i < nb_workers; i++) { + if (worker_params[i]->worker_info.ready_flag ==false) {
app/test-dma-perf/benchmark.c new file +19
-0+ ready =0 ; +break ; + } + } +if (ready) +break ; + } + +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.start_flag = true; + +usleep (TEST_WAIT_U_SECOND ); +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.test_cpl = +worker_params[i]->worker_info.total_cpl; + +usleep (test_secs *1000 *1000 ); +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.test_cpl =worker_params[i]->worker_info.total_cpl -
(quoted diff) +1
-0+ worker_params[i]-Quoted 1 line
worker_info.test_cpl;
app/test-dma-perf/benchmark.c new file +12
-0+ +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.stop_flag = true; + +rte_eal_mp_wait_lcore (); + +for (i =0 ; i < nb_workers; i++) { +calc_result (buf_size, nr_buf, nb_workers, test_secs, + worker_params[i]->worker_info.test_cpl, + &memory, &avg_cycles, &bandwidth, &mops); + output_result(cfg->scenario_id, worker_params[i]->lcore_id, + worker_params[i]->dma_name,
avg_cycles,buf_size,(quoted diff) +1
-0+ nr_buf / nb_workers, memory,bandwidth, mops,
is_dma);app/test-dma-perf/benchmark.c new file +8
-0+ } + +out: +/* free env */ +if (srcs) +rte_pktmbuf_free_bulk (srcs, nr_buf); +if (dsts) +rte_pktmbuf_free_bulk (dsts, nr_buf);
app/test-dma-perf/benchmark.c line 469
Quoted 10 lines
app/test-dma-perf/benchmark.c new file +10
-0+ +if (src_pool) +rte_mempool_free (src_pool); +if (dst_pool) +rte_mempool_free (dst_pool); + +if (is_dma) { +for (i =0 ; i < nb_workers; i++) { +printf ("Stopping dmadev %d\n" , ldm->dma_ids[i]); +rte_dma_stop (ldm->dma_ids[i]);
[Amit]: Below rte_dma_stop please call rte_dma_close for clean exit.
<snip>
Quoted 25 lines
app/test-dma-perf/main.h new file +1
-0+#endif /* _MAIN_H_ */ app/test-dma- renamed from app/test-dma-perf/meson.build +0
-0perf/meson.build new file mode 100644 index 0000000000..bd6c264002
app/test-dma-perf/meson.build new file +17
-1@@ -0,0 +1,17 @@ +# SPDX-License-Identifier: BSD-3-Clause # Copyright(c) 2019-2023 Intel +Corporation + +# meson file, for building this app as part of a main DPDK build. + +if is_windows + build = false + reason = 'not supported on Windows' + subdir_done() +endif + +deps += ['dmadev', 'mbuf', 'cfgfile'] + +sources = files( + 'main.c', + 'benchmark.c', +) --2.40.1
Hi,
Thanks for your comments, replies are inline.
Thanks,
Cheng
Quoted 125 lines
-----Original Message-----
From: Amit Prakash Shukla <[email protected]>
Sent: Friday, June 9, 2023 10:03 PM
To: Jiang, Cheng1 <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; Ma, WenwuX <[email protected]>; Wang, YuanX <[email protected]>; He, Xingguang <[email protected]>; Jerin Jacob Kollanukkaran <[email protected]>; Anoob Joseph <[email protected]>
Subject: RE: [EXT] [PATCH v5] app/dma-perf: introduce dma-perf applicationQuoted 10 lines
-----Original Message-----
From: Cheng Jiang <[email protected]>
Sent: Thursday, June 8, 2023 2:14 PM
To: [email protected]; [email protected]; [email protected]; [email protected]
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; Cheng Jiang <[email protected]>
Subject: [EXT] [PATCH v5] app/dma-perf: introduce dma-perf applicationQuoted 4 lines
External Email
----------------------------------------------------------------------
There are many high-performance DMA devices supported in DPDK now,and
Quoted 9 lines
these DMA devices can also be integrated into other modules of DPDK as accelerators, such as Vhost. Before integrating DMA into applications, developers need to know the performance of these DMA devices in various scenarios and the performance of CPUs in the same scenario, such as different buffer lengths. Only in this way can we know the target performance of the application accelerated by using them. This patch introduces a high-performance testing tool, which supports comparing the performance of CPU and DMA in different scenarios automatically with a pre- set config file. Memory Copy performance test
are supported for now.
Quoted 20 lines
Signed-off-by Cheng Jiang <[email protected]>Signed-off-by Jiayu Hu <[email protected]>Signed-off-by Yuan Wang <[email protected]>Acked-by Morten Brørup <[email protected]>Acked-by Chenbo Xia <[email protected]>---
v5:
fixed someLONG_LINEwarnings;
v4:
fixed inaccuracy of the memory footprint display;
v3:
fixed some typos;
v2:
added lcore/dmadev designation;
added error case process;
removedworker_threadsparameter from config.ini; improved the logs;
improved config file;Quoted 9 lines
app/meson.build | 1 + app/test-dma-perf/benchmark.c | 472 ++++++++++++++++++++++++++++ app/test-dma-perf/config.ini | 59 ++++ app/test-dma-perf/main.c | 569 ++++++++++++++++++++++++++++++++++ app/test-dma-perf/main.h | 69 +++++ app/test-dma-perf/meson.build | 17 + 6 files changed, 1187 insertions(+) create mode 100644 app/test-dma-perf/benchmark.c create mode100644
Quoted 4 lines
app/test-dma-perf/config.ini create mode 100644 app/test-dma-
perf/main.ccreate mode 100644app/test-dma-perf/main.hcreate mode 100644 app/test-dma-perf/meson.build<snip>
Quoted 48 lines
app/test-dma-perf/benchmark.c new file +34
-0+ +static inline int +do_dma_mem_copy (void *p) +{ +uint16_t *para_idx = (uint16_t *)p; +volatile struct lcore_params *para = worker_params[*para_idx]; +volatile struct worker_info *worker_info = &(para->worker_info); +uint16_t dev_id = para->dev_id; +uint32_t nr_buf = para->nr_buf; +uint16_t kick_batch = para->kick_batch; +uint32_t buf_size = para->buf_size; +struct rte_mbuf **srcs = para->srcs; +struct rte_mbuf **dsts = para->dsts; +int64_t async_cnt =0 ; +int nr_cpl =0 ; +uint32_t i; +uint32_t poll_cnt =0 ; + + worker_info->stop_flag = false; + worker_info->ready_flag = true; + +while (!worker_info->start_flag) + ; + +while (1 ) { +for (i =0 ; i < nr_buf; i++) { +if (unlikely (rte_dma_copy (dev_id, +0 , +rte_pktmbuf_iova (srcs[i]), +rte_pktmbuf_iova (dsts[i]), + buf_size, +0 ) <0 )) { +rte_dma_submit (dev_id,0 ); + while (rte_dma_burst_capacity(dev_id, 0) ==0) {
(quoted diff) +1
-0+ nr_cpl = rte_dma_completed(dev_id,0,
MAX_DMA_CPL_NB,app/test-dma-perf/benchmark.c new file +11
-0+ NULL, NULL); + async_cnt -= nr_cpl; + worker_info->total_cpl += nr_cpl; + } +if (rte_dma_copy (dev_id, +0 , +rte_pktmbuf_iova (srcs[i]), +rte_pktmbuf_iova (dsts[i]), + buf_size, +0 ) <0 ) { + printf("enqueue fail again at %u\n",
Quoted 8 lines
Quoted 5 lines
i);
(quoted diff) +1
-0+ printf("space:%d\n",
rte_dma_burst_capacity(dev_id, 0));(quoted diff) +1
-0+ rte_exit(EXIT_FAILURE, "DMAenqueue failed\n");
[Amit]: On all success or failure exits, please call
rte_dma_stopandrte_dma_closeto exit cleanly.
[Cheng] Got it. Thanks, I'll fix it in the next version.
Quoted 155 lines
Quoted 29 lines
app/test-dma-perf/benchmark.c new file +8
-0+ } + } + async_cnt++; + +if ((async_cnt % kick_batch) ==0 ) { +rte_dma_submit (dev_id,0 ); +/* add a poll to avoid ring full */ + nr_cpl = rte_dma_completed(dev_id, 0,
MAX_DMA_CPL_NB, NULL, NULL);app/test-dma-perf/benchmark.c new file +12
-0+ async_cnt -= nr_cpl; + worker_info->total_cpl += nr_cpl; + } + } + +if (worker_info->stop_flag) +break ; + } + +rte_dma_submit (dev_id,0 ); +while ((async_cnt >0 ) && (poll_cnt++ <POLL_MAX )) { + nr_cpl = rte_dma_completed(dev_id, 0,
MAX_DMA_CPL_NB, NULL, NULL);app/test-dma-perf/benchmark.c new file +6
-0+ async_cnt -= nr_cpl; + } + +return 0 ; +} +<snip>
Quoted 123 lines
app/test-dma-perf/benchmark.c new file +11
-0+ +void +mem_copy_benchmark (struct test_configure *cfg,bool is_dma) { +uint16_t i; +uint32_t offset; +unsigned int lcore_id =0 ; +struct rte_mbuf **srcs =NULL , **dsts =NULL ; +struct lcore_dma_map_t *ldm = &cfg->lcore_dma_map; +unsigned int buf_size = cfg->buf_size.cur; +uint16_t kick_batch = cfg->kick_batch.cur; + uint32_t nr_buf = cfg->nr_buf = (cfg->mem_size.cur * 1024 * 1024) /(
cfg->buf_size.cur* 2);app/test-dma-perf/benchmark.c new file +27
-0+uint16_t nb_workers = ldm->cnt; +uint16_t test_secs = cfg->test_secs; +float memory; +uint32_t avg_cycles =0 ; +float mops; +float bandwidth; + +if (setup_memory_env (cfg, &srcs, &dsts) <0 ) + goto out; + +if (is_dma) +if (config_dmadevs (cfg) <0 ) + goto out; + +if (cfg->cache_flush) { +cache_flush_buf (srcs, buf_size, nr_buf); +cache_flush_buf (dsts, buf_size, nr_buf); +rte_mb (); + } + +printf ("Start testing....\n" ); + +for (i =0 ; i < nb_workers; i++) { + lcore_id = ldm->lcores[i]; + offset = nr_buf / nb_workers * i; + + worker_params[i] = rte_malloc(NULL, sizeof(struct
lcore_params), 0);(quoted diff) +2
-0+ if (!worker_params[i]) { + printf("lcore parameters malloc failure for lcore%d\n",
lcore_id);app/test-dma-perf/benchmark.c new file +4
-0+break ; + } +if (is_dma) { + worker_params[i]->dma_name = ldm-Quoted 1 line
dma_names[i];
app/test-dma-perf/benchmark.c new file +5
-0+ worker_params[i]->dev_id = ldm->dma_ids[i]; + worker_params[i]->kick_batch = kick_batch; + } + worker_params[i]->worker_id = i; + worker_params[i]->nr_buf = (uint32_t)(nr_buf /
nb_workers);app/test-dma-perf/benchmark.c new file +9
-0+ worker_params[i]->buf_size = buf_size; + worker_params[i]->test_secs = test_secs; + worker_params[i]->srcs = srcs + offset; + worker_params[i]->dsts = dsts + offset; + worker_params[i]->scenario_id = cfg->scenario_id; + worker_params[i]->lcore_id = lcore_id; + +if (is_dma) + rte_eal_remote_launch(do_dma_mem_copy, (void*)(&i),
lcore_id);(quoted diff) +2
-0+ else + rte_eal_remote_launch(do_cpu_mem_copy, (void*)(&i),
lcore_id);app/test-dma-perf/benchmark.c new file +6
-0+ } + +while (1 ) { +bool ready = true; +for (i =0 ; i < nb_workers; i++) { + if (worker_params[i]->worker_info.ready_flag ==false) {
app/test-dma-perf/benchmark.c new file +19
-0+ ready =0 ; +break ; + } + } +if (ready) +break ; + } + +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.start_flag = true; + +usleep (TEST_WAIT_U_SECOND ); +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.test_cpl = +worker_params[i]->worker_info.total_cpl; + +usleep (test_secs *1000 *1000 ); +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.test_cpl =worker_params[i]->worker_info.total_cpl -
(quoted diff) +1
-0+ worker_params[i]-Quoted 1 line
worker_info.test_cpl;
app/test-dma-perf/benchmark.c new file +12
-0+ +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.stop_flag = true; + +rte_eal_mp_wait_lcore (); + +for (i =0 ; i < nb_workers; i++) { +calc_result (buf_size, nr_buf, nb_workers, test_secs, + worker_params[i]->worker_info.test_cpl, + &memory, &avg_cycles, &bandwidth, &mops); + output_result(cfg->scenario_id, worker_params[i]->lcore_id, + worker_params[i]->dma_name,
avg_cycles,buf_size,(quoted diff) +1
-0+ nr_buf / nb_workers, memory,bandwidth, mops,
is_dma);app/test-dma-perf/benchmark.c new file +12
-0+ } + +out: +/* free env */ +if (srcs) +rte_pktmbuf_free_bulk (srcs, nr_buf); +if (dsts) +rte_pktmbuf_free_bulk (dsts, nr_buf); + +if (src_pool) +rte_mempool_free (src_pool); +if (dst_pool)
Quoted 9 lines
app/test-dma-perf/benchmark.c line 469
Quoted 6 lines
app/test-dma-perf/benchmark.c new file +6
-0+rte_mempool_free (dst_pool); + +if (is_dma) { +for (i =0 ; i < nb_workers; i++) { +printf ("Stopping dmadev %d\n" , ldm->dma_ids[i]); +rte_dma_stop (ldm->dma_ids[i]);[Amit]: Below
rte_dma_stopplease callrte_dma_closefor clean exit.
[Cheng] Sure, I'll fix it in the next version.
Quoted 27 lines
<snip>
Quoted 25 lines
app/test-dma-perf/main.h new file +1
-0+#endif /* _MAIN_H_ */ app/test-dma- renamed from app/test-dma-perf/meson.build +0
-0perf/meson.build new file mode 100644 index 0000000000..bd6c264002
app/test-dma-perf/meson.build new file +17
-1@@ -0,0 +1,17 @@ +# SPDX-License-Identifier: BSD-3-Clause # Copyright(c) 2019-2023 +Intel Corporation + +# meson file, for building this app as part of a main DPDK build. + +if is_windows + build = false + reason = 'not supported on Windows' + subdir_done() +endif + +deps += ['dmadev', 'mbuf', 'cfgfile'] + +sources = files( + 'main.c', + 'benchmark.c', +) --2.40.1
Hi,
Replies are inline.
Quoted 161 lines
-----Original Message-----
From: Jiang, Cheng1
Sent: Monday, June 12, 2023 4:27 PM
To: Amit Prakash Shukla <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; Ma, WenwuX <[email protected]>; Wang, YuanX <[email protected]>; He, Xingguang <[email protected]>; Jerin Jacob Kollanukkaran <[email protected]>; Anoob Joseph <[email protected]>
Subject: RE: [EXT] [PATCH v5] app/dma-perf: introduce dma-perf applicationHi,
Thanks for your comments, replies are inline.
Thanks,
ChengQuoted 8 lines
-----Original Message-----
From: Amit Prakash Shukla <[email protected]>
Sent: Friday, June 9, 2023 10:03 PM
To: Jiang, Cheng1 <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; Ma, WenwuX <[email protected]>; Wang,YuanX
Quoted 15 lines
<[email protected]>; He, Xingguang <[email protected]>; Jerin Jacob Kollanukkaran <[email protected]>; Anoob Joseph <[email protected]>
Subject: RE: [EXT] [PATCH v5] app/dma-perf: introduce dma-perf applicationQuoted 7 lines
-----Original Message-----
From: Cheng Jiang <[email protected]>
Sent: Thursday, June 8, 2023 2:14 PM
To: [email protected]; [email protected]; [email protected]; [email protected]
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected];Quoted 107 lines
Quoted 9 lines
Cheng Jiang <[email protected]>
Subject: [EXT] [PATCH v5] app/dma-perf: introduce dma-perf applicationExternal Email
--------------------------------------------------------------------
-- There are many high-performance DMA devices supported in DPDK now,and
Quoted 10 lines
these DMA devices can also be integrated into other modules of DPDK as accelerators, such as Vhost. Before integrating DMA into applications, developers need to know the performance of these DMA devices in various scenarios and the performance of CPUs in the same scenario, such as different buffer lengths. Only in this way can we know the target performance of the application accelerated by using them. This patch introduces a high-performance testing tool, which supports comparing the performance of CPU and DMA in different scenarios automatically with a pre- set config file. Memory Copy performance test
are supported for now.
Quoted 28 lines
Signed-off-by Cheng Jiang <[email protected]>Signed-off-by Jiayu Hu <[email protected]>Signed-off-by Yuan Wang <[email protected]>Acked-by Morten Brørup <[email protected]>Acked-by Chenbo Xia <[email protected]>---
v5:
fixed someLONG_LINEwarnings;
v4:
fixed inaccuracy of the memory footprint display;
v3:
fixed some typos;
v2:
added lcore/dmadev designation;
added error case process;
removedworker_threadsparameter from config.ini; improved the logs;
improved config file;app/meson.build | 1 + app/test-dma-perf/benchmark.c | 472 ++++++++++++++++++++++++++++ app/test-dma-perf/config.ini | 59 ++++ app/test-dma-perf/main.c | 569 ++++++++++++++++++++++++++++++++++ app/test-dma-perf/main.h | 69 +++++ app/test-dma-perf/meson.build | 17 + 6 files changed, 1187 insertions(+) create mode 100644Quoted 1 line
app/test-dma-perf/benchmark.ccreate mode100644
Quoted 5 lines
app/test-dma-perf/config.ini create mode 100644 app/test-dma-
perf/main.ccreate mode 100644app/test-dma-perf/main.hcreate mode
100644 app/test-dma-perf/meson.build<snip>
Quoted 48 lines
app/test-dma-perf/benchmark.c new file +34
-0+ +static inline int +do_dma_mem_copy (void *p) +{ +uint16_t *para_idx = (uint16_t *)p; +volatile struct lcore_params *para = worker_params[*para_idx]; +volatile struct worker_info *worker_info = &(para->worker_info); +uint16_t dev_id = para->dev_id; +uint32_t nr_buf = para->nr_buf; +uint16_t kick_batch = para->kick_batch; +uint32_t buf_size = para->buf_size; +struct rte_mbuf **srcs = para->srcs; +struct rte_mbuf **dsts = para->dsts; +int64_t async_cnt =0 ; +int nr_cpl =0 ; +uint32_t i; +uint32_t poll_cnt =0 ; + + worker_info->stop_flag = false; + worker_info->ready_flag = true; + +while (!worker_info->start_flag) + ; + +while (1 ) { +for (i =0 ; i < nr_buf; i++) { +if (unlikely (rte_dma_copy (dev_id, +0 , +rte_pktmbuf_iova (srcs[i]), +rte_pktmbuf_iova (dsts[i]), + buf_size, +0 ) <0 )) { +rte_dma_submit (dev_id,0 ); + while (rte_dma_burst_capacity(dev_id, 0) ==0) {
(quoted diff) +1
-0+ nr_cpl = rte_dma_completed(dev_id,0,
MAX_DMA_CPL_NB,app/test-dma-perf/benchmark.c new file +11
-0+ NULL, NULL); + async_cnt -= nr_cpl; + worker_info->total_cpl += nr_cpl; + } +if (rte_dma_copy (dev_id, +0 , +rte_pktmbuf_iova (srcs[i]), +rte_pktmbuf_iova (dsts[i]), + buf_size, +0 ) <0 ) { + printf("enqueue fail again at %u\n",Quoted 8 lines
Quoted 5 lines
i);
(quoted diff) +1
-0+ printf("space:%d\n",
rte_dma_burst_capacity(dev_id, 0));(quoted diff) +1
-0+ rte_exit(EXIT_FAILURE, "DMAenqueue failed\n");
[Amit]: On all success or failure exits, please call
rte_dma_stopandrte_dma_closeto exit cleanly.
Quoted 1 line
[Cheng] Got it. Thanks, I'll fix it in the next version.
[Cheng] Hi, I take a look into it, and find out rte_exit() will call rte_dma_close(), so I think there is no need to do it explicitly here. What do you think?
Quoted 167 lines
Quoted 157 lines
Quoted 30 lines
app/test-dma-perf/benchmark.c new file +8
-0+ } + } + async_cnt++; + +if ((async_cnt % kick_batch) ==0 ) { +rte_dma_submit (dev_id,0 ); +/* add a poll to avoid ring full */ + nr_cpl = rte_dma_completed(dev_id, 0,
MAX_DMA_CPL_NB, NULL, NULL);app/test-dma-perf/benchmark.c new file +12
-0+ async_cnt -= nr_cpl; + worker_info->total_cpl += nr_cpl; + } + } + +if (worker_info->stop_flag) +break ; + } + +rte_dma_submit (dev_id,0 ); +while ((async_cnt >0 ) && (poll_cnt++ <POLL_MAX )) { + nr_cpl = rte_dma_completed(dev_id, 0,
MAX_DMA_CPL_NB, NULL, NULL);app/test-dma-perf/benchmark.c new file +6
-0+ async_cnt -= nr_cpl; + } + +return 0 ; +} +<snip>
Quoted 124 lines
app/test-dma-perf/benchmark.c new file +12
-0+ +void +mem_copy_benchmark (struct test_configure *cfg,bool is_dma) { +uint16_t i; +uint32_t offset; +unsigned int lcore_id =0 ; +struct rte_mbuf **srcs =NULL , **dsts =NULL ; +struct lcore_dma_map_t *ldm = &cfg->lcore_dma_map; +unsigned int buf_size = cfg->buf_size.cur; +uint16_t kick_batch = cfg->kick_batch.cur; +uint32_t nr_buf = cfg->nr_buf = (cfg->mem_size.cur *1024 *1024 ) +/(
cfg->buf_size.cur* 2);app/test-dma-perf/benchmark.c new file +27
-0+uint16_t nb_workers = ldm->cnt; +uint16_t test_secs = cfg->test_secs; +float memory; +uint32_t avg_cycles =0 ; +float mops; +float bandwidth; + +if (setup_memory_env (cfg, &srcs, &dsts) <0 ) + goto out; + +if (is_dma) +if (config_dmadevs (cfg) <0 ) + goto out; + +if (cfg->cache_flush) { +cache_flush_buf (srcs, buf_size, nr_buf); +cache_flush_buf (dsts, buf_size, nr_buf); +rte_mb (); + } + +printf ("Start testing....\n" ); + +for (i =0 ; i < nb_workers; i++) { + lcore_id = ldm->lcores[i]; + offset = nr_buf / nb_workers * i; + + worker_params[i] = rte_malloc(NULL, sizeof(struct
lcore_params), 0);(quoted diff) +2
-0+ if (!worker_params[i]) { + printf("lcore parameters malloc failure for lcore%d\n",
lcore_id);app/test-dma-perf/benchmark.c new file +4
-0+break ; + } +if (is_dma) { + worker_params[i]->dma_name = ldm-Quoted 1 line
dma_names[i];
app/test-dma-perf/benchmark.c new file +5
-0+ worker_params[i]->dev_id = ldm->dma_ids[i]; + worker_params[i]->kick_batch = kick_batch; + } + worker_params[i]->worker_id = i; + worker_params[i]->nr_buf = (uint32_t)(nr_buf /
nb_workers);app/test-dma-perf/benchmark.c new file +9
-0+ worker_params[i]->buf_size = buf_size; + worker_params[i]->test_secs = test_secs; + worker_params[i]->srcs = srcs + offset; + worker_params[i]->dsts = dsts + offset; + worker_params[i]->scenario_id = cfg->scenario_id; + worker_params[i]->lcore_id = lcore_id; + +if (is_dma) + rte_eal_remote_launch(do_dma_mem_copy, (void*)(&i),
lcore_id);(quoted diff) +2
-0+ else + rte_eal_remote_launch(do_cpu_mem_copy, (void*)(&i),
lcore_id);app/test-dma-perf/benchmark.c new file +6
-0+ } + +while (1 ) { +bool ready = true; +for (i =0 ; i < nb_workers; i++) { + if (worker_params[i]->worker_info.ready_flag ==false) {
app/test-dma-perf/benchmark.c new file +19
-0+ ready =0 ; +break ; + } + } +if (ready) +break ; + } + +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.start_flag = true; + +usleep (TEST_WAIT_U_SECOND ); +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.test_cpl = +worker_params[i]->worker_info.total_cpl; + +usleep (test_secs *1000 *1000 ); +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.test_cpl =worker_params[i]->worker_info.total_cpl -
(quoted diff) +1
-0+ worker_params[i]-Quoted 1 line
worker_info.test_cpl;
app/test-dma-perf/benchmark.c new file +12
-0+ +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.stop_flag = true; + +rte_eal_mp_wait_lcore (); + +for (i =0 ; i < nb_workers; i++) { +calc_result (buf_size, nr_buf, nb_workers, test_secs, + worker_params[i]->worker_info.test_cpl, + &memory, &avg_cycles, &bandwidth, &mops); + output_result(cfg->scenario_id, worker_params[i]->lcore_id, + worker_params[i]->dma_name,
avg_cycles,buf_size,(quoted diff) +1
-0+ nr_buf / nb_workers, memory,bandwidth, mops,
is_dma);app/test-dma-perf/benchmark.c new file +12
-0+ } + +out: +/* free env */ +if (srcs) +rte_pktmbuf_free_bulk (srcs, nr_buf); +if (dsts) +rte_pktmbuf_free_bulk (dsts, nr_buf); + +if (src_pool) +rte_mempool_free (src_pool); +if (dst_pool)Quoted 9 lines
app/test-dma-perf/benchmark.c line 469
Quoted 6 lines
app/test-dma-perf/benchmark.c new file +6
-0+rte_mempool_free (dst_pool); + +if (is_dma) { +for (i =0 ; i < nb_workers; i++) { +printf ("Stopping dmadev %d\n" , ldm->dma_ids[i]); +rte_dma_stop (ldm->dma_ids[i]);[Amit]: Below
rte_dma_stopplease callrte_dma_closefor clean exit.
Quoted 1 line
[Cheng] Sure, I'll fix it in the next version.
[Cheng] We are not planning to exit here, we are just going to do the next test, so we shouldn't call the rte_dma_close(). Just clarify. And when we finished all the test, we will call the rte_eal_cleanup() which will call the rte_dma_close(). Thanks!
Quoted 28 lines
Quoted 28 lines
<snip>
Quoted 25 lines
app/test-dma-perf/main.h new file +1
-0+#endif /* _MAIN_H_ */ app/test-dma- renamed from app/test-dma-perf/meson.build +0
-0perf/meson.build new file mode 100644 index 0000000000..bd6c264002
app/test-dma-perf/meson.build new file +17
-1@@ -0,0 +1,17 @@ +# SPDX-License-Identifier: BSD-3-Clause # Copyright(c) 2019-2023 +Intel Corporation + +# meson file, for building this app as part of a main DPDK build. + +if is_windows + build = false + reason = 'not supported on Windows' + subdir_done() +endif + +deps += ['dmadev', 'mbuf', 'cfgfile'] + +sources = files( + 'main.c', + 'benchmark.c', +) --2.40.1
Hi Cheng,
My replies are inline.
Thanks,
Amit Shukla
Quoted 183 lines
-----Original Message-----
From: Jiang, Cheng1 <[email protected]>
Sent: Tuesday, June 13, 2023 10:21 AM
To: Amit Prakash Shukla <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; Ma, WenwuX <[email protected]>; Wang, YuanX <[email protected]>; He, Xingguang <[email protected]>; Jerin Jacob Kollanukkaran <[email protected]>; Anoob Joseph <[email protected]>
Subject: RE: [EXT] [PATCH v5] app/dma-perf: introduce dma-perf applicationHi,
Replies are inline.
Quoted 8 lines
-----Original Message-----
From: Jiang, Cheng1
Sent: Monday, June 12, 2023 4:27 PM
To: Amit Prakash Shukla <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; Ma, WenwuX <[email protected]>; Wang,YuanX
Quoted 1 line
<[email protected]>; He, Xingguang <[email protected]>;
Jerin
Quoted 29 lines
Jacob Kollanukkaran <[email protected]>; Anoob Joseph <[email protected]>
Subject: RE: [EXT] [PATCH v5] app/dma-perf: introduce dma-perf applicationHi,
Thanks for your comments, replies are inline.
Thanks,
ChengQuoted 8 lines
-----Original Message-----
From: Amit Prakash Shukla <[email protected]>
Sent: Friday, June 9, 2023 10:03 PM
To: Jiang, Cheng1 <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; Ma, WenwuX <[email protected]>; Wang,YuanX
Quoted 8 lines
<[email protected]>; He, Xingguang <[email protected]>; Jerin Jacob Kollanukkaran <[email protected]>; Anoob Joseph <[email protected]>
Subject: RE: [EXT] [PATCH v5] app/dma-perf: introduce dma-perf applicationQuoted 52 lines
Quoted 7 lines
Quoted 7 lines
-----Original Message-----
From: Cheng Jiang <[email protected]>
Sent: Thursday, June 8, 2023 2:14 PM
To: [email protected]; [email protected]; [email protected]; [email protected]
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected];Quoted 44 lines
Quoted 10 lines
Cheng Jiang <[email protected]>
Subject: [EXT] [PATCH v5] app/dma-perf: introduce dma-perf applicationExternal Email
------------------------------------------------------------------
--
-- There are many high-performance DMA devices supported in DPDK now,and
Quoted 10 lines
these DMA devices can also be integrated into other modules of DPDK as accelerators, such as Vhost. Before integrating DMA into applications, developers need to know the performance of these DMA devices in various scenarios and the performance of CPUs in the same scenario, such as different buffer lengths. Only in this way can we know the target performance of the application accelerated by using them. This patch introduces a high-performance testing tool, which supports comparing the performance of CPU and DMA in different scenarios automatically with a pre- set config file. Memory Copy performance test
are supported for now.
Quoted 22 lines
Signed-off-by Cheng Jiang <[email protected]>Signed-off-by Jiayu Hu <[email protected]>Signed-off-by Yuan Wang <[email protected]>Acked-by Morten Brørup <[email protected]>Acked-by Chenbo Xia <[email protected]>---
v5:
fixed someLONG_LINEwarnings;
v4:
fixed inaccuracy of the memory footprint display;
v3:
fixed some typos;
v2:
added lcore/dmadev designation;
added error case process;
removedworker_threadsparameter from config.ini; improved the logs;
improved config file;app/meson.build | 1 + app/test-dma-perf/benchmark.c | 472++++++++++++++++++++++++++++
Quoted 73 lines
Quoted 72 lines
Quoted 7 lines
app/test-dma-perf/config.ini | 59 ++++
app/test-dma-perf/main.c | 569 ++++++++++++++++++++++++++++++++++ app/test-dma-perf/main.h | 69 +++++ app/test-dma-perf/meson.build | 17 + 6 files changed, 1187 insertions(+) create mode 100644
app/test-dma-perf/benchmark.ccreate mode100644
Quoted 5 lines
app/test-dma-perf/config.ini create mode 100644 app/test-dma-
perf/main.ccreate mode 100644app/test-dma-perf/main.hcreate mode
100644 app/test-dma-perf/meson.build<snip>
Quoted 52 lines
app/test-dma-perf/benchmark.c new file +34
-0+ +static inline int +do_dma_mem_copy (void *p) +{ +uint16_t *para_idx = (uint16_t *)p; +volatile struct lcore_params *para = worker_params[*para_idx]; +volatile struct worker_info *worker_info = &(para->worker_info); +uint16_t dev_id = para->dev_id; +uint32_t nr_buf = para->nr_buf; +uint16_t kick_batch = para->kick_batch; +uint32_t buf_size = para->buf_size; +struct rte_mbuf **srcs = para->srcs; +struct rte_mbuf **dsts = para->dsts; +int64_t async_cnt =0 ; +int nr_cpl =0 ; +uint32_t i; +uint32_t poll_cnt =0 ; + + worker_info->stop_flag = false; + worker_info->ready_flag = true; + +while (!worker_info->start_flag) + ; + +while (1 ) { +for (i =0 ; i < nr_buf; i++) { +if (unlikely (rte_dma_copy (dev_id, +0 , +rte_pktmbuf_iova (srcs[i]), +rte_pktmbuf_iova (dsts[i]), + buf_size, +0 ) <0 )) { +rte_dma_submit (dev_id,0 ); + while (rte_dma_burst_capacity(dev_id, 0) ==0) {
(quoted diff) +1
-0+ nr_cpl = rte_dma_completed(dev_id,0,
MAX_DMA_CPL_NB,app/test-dma-perf/benchmark.c new file +11
-0+ NULL, NULL); + async_cnt -= nr_cpl; + worker_info->total_cpl += nr_cpl; + } +if (rte_dma_copy (dev_id, +0 , +rte_pktmbuf_iova (srcs[i]), +rte_pktmbuf_iova (dsts[i]), + buf_size, +0 ) <0 ) { + printf("enqueue fail again at %u\n",i);
(quoted diff) +1
-0+ printf("space:%d\n",
rte_dma_burst_capacity(dev_id, 0));(quoted diff) +1
-0+ rte_exit(EXIT_FAILURE, "DMAQuoted 1 line
enqueue failed\n");
[Amit]: On all success or failure exits, please call
rte_dma_stopandrte_dma_closeto exit cleanly.
Quoted 5 lines
Quoted 1 line
[Cheng] Got it. Thanks, I'll fix it in the next version.
[Cheng] Hi, I take a look into it, and find out
rte_exit()will callrte_dma_close(), so I think there is no need to do it explicitly here. What do you think?
[Amit]: I agree, not required to explicitly call rte_dma_close here.
Quoted 168 lines
Quoted 168 lines
Quoted 167 lines
Quoted 31 lines
app/test-dma-perf/benchmark.c new file +8
-0+ } + } + async_cnt++; + +if ((async_cnt % kick_batch) ==0 ) { +rte_dma_submit (dev_id,0 ); +/* add a poll to avoid ring full */ + nr_cpl = rte_dma_completed(dev_id, 0,
MAX_DMA_CPL_NB, NULL, NULL);app/test-dma-perf/benchmark.c new file +12
-0+ async_cnt -= nr_cpl; + worker_info->total_cpl += nr_cpl; + } + } + +if (worker_info->stop_flag) +break ; + } + +rte_dma_submit (dev_id,0 ); +while ((async_cnt >0 ) && (poll_cnt++ <POLL_MAX )) { + nr_cpl = rte_dma_completed(dev_id, 0,
MAX_DMA_CPL_NB, NULL, NULL);app/test-dma-perf/benchmark.c new file +6
-0+ async_cnt -= nr_cpl; + } + +return 0 ; +} +<snip>
Quoted 120 lines
app/test-dma-perf/benchmark.c new file +12
-0+ +void +mem_copy_benchmark (struct test_configure *cfg,bool is_dma) { +uint16_t i; +uint32_t offset; +unsigned int lcore_id =0 ; +struct rte_mbuf **srcs =NULL , **dsts =NULL ; +struct lcore_dma_map_t *ldm = &cfg->lcore_dma_map; +unsigned int buf_size = cfg->buf_size.cur; +uint16_t kick_batch = cfg->kick_batch.cur; +uint32_t nr_buf = cfg->nr_buf = (cfg->mem_size.cur *1024 * +1024) /(
cfg->buf_size.cur* 2);app/test-dma-perf/benchmark.c new file +27
-0+uint16_t nb_workers = ldm->cnt; +uint16_t test_secs = cfg->test_secs; +float memory; +uint32_t avg_cycles =0 ; +float mops; +float bandwidth; + +if (setup_memory_env (cfg, &srcs, &dsts) <0 ) + goto out; + +if (is_dma) +if (config_dmadevs (cfg) <0 ) + goto out; + +if (cfg->cache_flush) { +cache_flush_buf (srcs, buf_size, nr_buf); +cache_flush_buf (dsts, buf_size, nr_buf); +rte_mb (); + } + +printf ("Start testing....\n" ); + +for (i =0 ; i < nb_workers; i++) { + lcore_id = ldm->lcores[i]; + offset = nr_buf / nb_workers * i; + + worker_params[i] = rte_malloc(NULL, sizeof(struct
lcore_params), 0);(quoted diff) +2
-0+ if (!worker_params[i]) { + printf("lcore parameters malloc failure for lcore%d\n",
lcore_id);app/test-dma-perf/benchmark.c new file +4
-0+break ; + } +if (is_dma) { + worker_params[i]->dma_name = ldm-Quoted 1 line
dma_names[i];
app/test-dma-perf/benchmark.c new file +5
-0+ worker_params[i]->dev_id = ldm->dma_ids[i]; + worker_params[i]->kick_batch = kick_batch; + } + worker_params[i]->worker_id = i; + worker_params[i]->nr_buf = (uint32_t)(nr_buf /
nb_workers);app/test-dma-perf/benchmark.c new file +9
-0+ worker_params[i]->buf_size = buf_size; + worker_params[i]->test_secs = test_secs; + worker_params[i]->srcs = srcs + offset; + worker_params[i]->dsts = dsts + offset; + worker_params[i]->scenario_id = cfg->scenario_id; + worker_params[i]->lcore_id = lcore_id; + +if (is_dma) + rte_eal_remote_launch(do_dma_mem_copy, (void*)(&i),
lcore_id);(quoted diff) +2
-0+ else + rte_eal_remote_launch(do_cpu_mem_copy, (void*)(&i),
lcore_id);app/test-dma-perf/benchmark.c new file +6
-0+ } + +while (1 ) { +bool ready = true; +for (i =0 ; i < nb_workers; i++) { + if (worker_params[i]->worker_info.ready_flag ==false) {
app/test-dma-perf/benchmark.c new file +19
-0+ ready =0 ; +break ; + } + } +if (ready) +break ; + } + +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.start_flag = true; + +usleep (TEST_WAIT_U_SECOND ); +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.test_cpl = +worker_params[i]->worker_info.total_cpl; + +usleep (test_secs *1000 *1000 ); +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.test_cpl =worker_params[i]->worker_info.total_cpl -
(quoted diff) +1
-0+ worker_params[i]-Quoted 1 line
worker_info.test_cpl;
app/test-dma-perf/benchmark.c new file +12
-0+ +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.stop_flag = true; + +rte_eal_mp_wait_lcore (); + +for (i =0 ; i < nb_workers; i++) { +calc_result (buf_size, nr_buf, nb_workers, test_secs, + worker_params[i]->worker_info.test_cpl, + &memory, &avg_cycles, &bandwidth, &mops); + output_result(cfg->scenario_id, worker_params[i]->lcore_id, + worker_params[i]->dma_name,
avg_cycles,buf_size,(quoted diff) +1
-0+ nr_buf / nb_workers, memory,bandwidth, mops,
is_dma);app/test-dma-perf/benchmark.c new file +8
-0+ } + +out: +/* free env */ +if (srcs) +rte_pktmbuf_free_bulk (srcs, nr_buf); +if (dsts) +rte_pktmbuf_free_bulk (dsts, nr_buf);app/test-dma-perf/benchmark.c line 469
Quoted 10 lines
app/test-dma-perf/benchmark.c new file +10
-0+ +if (src_pool) +rte_mempool_free (src_pool); +if (dst_pool) +rte_mempool_free (dst_pool); + +if (is_dma) { +for (i =0 ; i < nb_workers; i++) { +printf ("Stopping dmadev %d\n" , ldm->dma_ids[i]); +rte_dma_stop (ldm->dma_ids[i]);[Amit]: Below
rte_dma_stopplease callrte_dma_closefor clean exit.
Quoted 6 lines
Quoted 1 line
[Cheng] Sure, I'll fix it in the next version.
[Cheng] We are not planning to exit here, we are just going to do the next test, so we shouldn't call the
rte_dma_close(). Just clarify. And when we finished all the test, we will call therte_eal_cleanup()which will call therte_dma_close(). Thanks!
[Amit]: Sure, thanks.
Quoted 29 lines
Quoted 29 lines
Quoted 29 lines
<snip>
Quoted 25 lines
app/test-dma-perf/main.h new file +1
-0+#endif /* _MAIN_H_ */ app/test-dma- renamed from app/test-dma-perf/meson.build +0
-0perf/meson.build new file mode 100644 index 0000000000..bd6c264002
app/test-dma-perf/meson.build new file +17
-1@@ -0,0 +1,17 @@ +# SPDX-License-Identifier: BSD-3-Clause # Copyright(c) 2019-2023 +Intel Corporation + +# meson file, for building this app as part of a main DPDK build. + +if is_windows + build = false + reason = 'not supported on Windows' + subdir_done() +endif + +deps += ['dmadev', 'mbuf', 'cfgfile'] + +sources = files( + 'main.c', + 'benchmark.c', +) --2.40.1
PATCH v6 1/1 app/dma-perf: introduce dma-perf application
In reply to Cheng JiangThere are many high-performance DMA devices supported in DPDK now, and these DMA devices can also be integrated into other modules of DPDK as accelerators, such as Vhost. Before integrating DMA into applications, developers need to know the performance of these DMA devices in various scenarios and the performance of CPUs in the same scenario, such as different buffer lengths. Only in this way can we know the target performance of the application accelerated by using them. This patch introduces a high-performance testing tool, which supports comparing the performance of CPU and DMA in different scenarios automatically with a pre-set config file. Memory Copy performance test are supported for now.
---
v6:
improved code based on Anoob's comments; fixed some code structure issues;
v5:
fixed some LONG_LINE warnings;
v4:
fixed inaccuracy of the memory footprint display;
v3:
fixed some typos;
v2:
added lcore/dmadev designation;
added error case process;
removed worker_threads parameter from config.ini; improved the logs;
improved config file;
app/meson.build | 1 + app/test-dma-perf/benchmark.c | 477 ++++++++++++++++++++++++++++ app/test-dma-perf/config.ini | 59 ++++ app/test-dma-perf/main.c | 569 ++++++++++++++++++++++++++++++++++ app/test-dma-perf/main.h | 69 +++++ app/test-dma-perf/meson.build | 17 + 6 files changed, 1192 insertions(+) create mode 100644 app/test-dma-perf/benchmark.c create mode 100644 app/test-dma-perf/config.ini create mode 100644 app/test-dma-perf/main.c create mode 100644 app/test-dma-perf/main.h create mode 100644 app/test-dma-perf/meson.build
app/meson.build +1 -0
@@ -19,6 +19,7 @@ apps = [ 'test-cmdline', 'test-compress-perf', 'test-crypto-perf', + 'test-dma-perf', 'test-eventdev', 'test-fib', 'test-flow-perf',
app/test-dma-perf/benchmark.c new file +477 -0
477 lines not on this page · raw message
app/test-dma-perf/config.ini new file +59 -0
59 lines not on this page · raw message
app/test-dma-perf/main.c new file +569 -0
569 lines not on this page · raw message
app/test-dma-perf/main.h new file +69 -0
69 lines not on this page · raw message
app/test-dma-perf/meson.build new file +17 -0
@@ -0,0 +1,17 @@ +# SPDX-License-Identifier: BSD-3-Clause +# Copyright(c) 2019-2023 Intel Corporation + +# meson file, for building this app as part of a main DPDK build. + +if is_windows + build = false + reason = 'not supported on Windows' + subdir_done() +endif + +deps += ['dmadev', 'mbuf', 'cfgfile'] + +sources = files( + 'main.c', + 'benchmark.c', +)
--
2.40.1
Hi Cheng,
Few comments inline. Please check.
Thanks,
Dengdui
Quoted 68 lines · Cheng Jiang
There are many high-performance DMA devices supported in DPDK now, and these DMA devices can also be integrated into other modules of DPDK as accelerators, such as Vhost. Before integrating DMA into applications, developers need to know the performance of these DMA devices in various scenarios and the performance of CPUs in the same scenario, such as different buffer lengths. Only in this way can we know the target performance of the application accelerated by using them. This patch introduces a high-performance testing tool, which supports comparing the performance of CPU and DMA in different scenarios automatically with a pre-set config file. Memory Copy performance test are supported for now.
Signed-off-by Cheng Jiang <[email protected]>Signed-off-by Jiayu Hu <[email protected]>Signed-off-by Yuan Wang <[email protected]>Acked-by Morten Brørup <[email protected]>Acked-by Chenbo Xia <[email protected]>---
v6:
improved code based on Anoob's comments; fixed some code structure issues;
v5:
fixed someLONG_LINEwarnings;
v4:
fixed inaccuracy of the memory footprint display;
v3:
fixed some typos;
v2:
added lcore/dmadev designation;
added error case process;
removedworker_threadsparameter from config.ini; improved the logs;
improved config file;app/meson.build | 1 + app/test-dma-perf/benchmark.c | 477 ++++++++++++++++++++++++++++ app/test-dma-perf/config.ini | 59 ++++ app/test-dma-perf/main.c | 569 ++++++++++++++++++++++++++++++++++ app/test-dma-perf/main.h | 69 +++++ app/test-dma-perf/meson.build | 17 + 6 files changed, 1192 insertions(+) create mode 100644 app/test-dma-perf/benchmark.c create mode 100644 app/test-dma-perf/config.ini create mode 100644 app/test-dma-perf/main.c create mode 100644 app/test-dma-perf/main.h create mode 100644 app/test-dma-perf/meson.buildapp/meson.build +1
-0@@ -19,6 +19,7 @@ apps = [ 'test-cmdline', 'test-compress-perf', 'test-crypto-perf', + 'test-dma-perf', 'test-eventdev', 'test-fib', 'test-flow-perf',app/test-dma-perf/benchmark.c new file +4
-0@@ -0,0 +1,477 @@ +/* SPDX-License-Identifier: BSD-3-Clause +* Copyright(c) 2023 Intel Corporation +*/ +
<snip>
Quoted 14 lines
app/test-dma-perf/benchmark.c new file +14
-0+static inline int +__rte_format_printf(3, 4) +print_err(const char *func, int lineno, const char *format, ...) +{ + va_list ap; + int ret; + + ret = fprintf(stderr, "In %s:%d - ", func, lineno); + va_start(ap, format); + ret += vfprintf(stderr, format, ap); + va_end(ap); + + return ret; +}
app/test-dma-perf/benchmark.c line 78
Quoted 10 lines
app/test-dma-perf/benchmark.c new file +10
-0+ +static inline void +calc_result (uint32_t buf_size,uint32_t nr_buf,uint16_t nb_workers,uint16_t test_secs, +uint32_t total_cnt,float *memory,uint32_t *ave_cycle, +float *bandwidth,float *mops) +{ + *memory = (float )(buf_size * (nr_buf / nb_workers) *2 ) / (1024 *1024 ); + *ave_cycle = test_secs *rte_get_timer_hz () / total_cnt; + *bandwidth = (buf_size *8 * (rte_get_timer_hz () / (float )*ave_cycle)) /1000000000 ; + *mops = (float )rte_get_timer_hz () / *ave_cycle /1000000 ;
The value of ave_cycle may be 0.
*mops = (float)(total_cnt / test_secs) / 1000000; ?
Quoted 44 lines
app/test-dma-perf/benchmark.c new file +44
-0+} + +static void +output_result (uint8_t scenario_id,uint32_t lcore_id,char *dma_name,uint64_t ave_cycle, +uint32_t buf_size,uint32_t nr_buf,float memory, +float bandwidth,float mops,bool is_dma) +{ +if (is_dma) +printf ("lcore %u, DMA %s:\n" , lcore_id, dma_name); +else +printf ("lcore %u\n" , lcore_id); + +printf ("average cycles/op: %" PRIu64", buffer size: %u, nr_buf: %u, memory: %.2lfMB, frequency: %" PRIu64".\n" , + ave_cycle, buf_size, nr_buf, memory,rte_get_timer_hz ()); +printf ("Average bandwidth: %.3lfGbps, MOps: %.3lf\n" , bandwidth, mops); + +if (is_dma) +snprintf (output_str[lcore_id],MAX_OUTPUT_STR_LEN ,CSV_LINE_DMA_FMT , + scenario_id, lcore_id, dma_name, buf_size, + nr_buf, memory, ave_cycle, bandwidth, mops); +else +snprintf (output_str[lcore_id],MAX_OUTPUT_STR_LEN ,CSV_LINE_CPU_FMT , + scenario_id, lcore_id, buf_size, + nr_buf, memory, ave_cycle, bandwidth, mops); +} + +static inline void +cache_flush_buf (__maybe_unused struct rte_mbuf **array, +__maybe_unused uint32_t buf_size, +__maybe_unused uint32_t nr_buf) +{ +#ifdef RTE_ARCH_X86_64 +char *data; +struct rte_mbuf **srcs = array; +uint32_t i, offset; + +for (i =0 ; i < nr_buf; i++) { + data = rte_pktmbuf_mtod(srcs[i], char *); +for (offset =0 ; offset < buf_size; offset +=64 ) +__builtin_ia32_clflush (data + offset); + } +#endif +} +
<snip>
Quoted 7 lines
app/test-dma-perf/benchmark.c new file +7
-0+ +/* Parse the argument given in the command line of the application */ +static int +append_eal_args (int argc,char **argv,const char *eal_args,char **new_argv) +{ +int i; +char *tokens[MAX_EAL_PARAM_NB ];
app/test-dma-perf/benchmark.c
Quoted 10 lines
app/test-dma-perf/benchmark.c new file +10
-0+char args[MAX_EAL_PARAM_LEN ] = {0 }; +int token_nb, new_argc =0 ; + +for (i =0 ; i < argc; i++) { +if ((strcmp (argv[i],CMDLINE_CONFIG_ARG ) ==0 ) || + (strcmp (argv[i],CMDLINE_RESULT_ARG ) ==0 )) { + i++; +continue ; + } +strlcpy (new_argv[new_argc], argv[i],sizeof (new_argv[new_argc]));
The type_of argv[new_argc] is *char. Cannot use sizeof().
strcpy(new_argv[new_argc], argv[i]); or strlcpy(new_argv[new_argc], argv[i], MAX_EAL_PARAM_LEN); ?
Quoted 133 lines
app/test-dma-perf/benchmark.c new file +133
-0+ new_argc++; + } + +if (eal_args) { +strlcpy (args, eal_args,sizeof (args)); + token_nb =rte_strsplit (args,strlen (args), + tokens,MAX_EAL_PARAM_NB ,' ' ); +for (i =0 ; i < token_nb; i++) +strcpy (new_argv[new_argc++], tokens[i]); + } + +return new_argc; +} + +int +main (int argc,char *argv[]) +{ +int ret; +uint16_t case_nb; +uint32_t i, nb_lcores; +pid_t cpid, wpid; +int wstatus; +char args[MAX_EAL_PARAM_NB ][MAX_EAL_PARAM_LEN ]; +char *pargs[MAX_EAL_PARAM_NB ]; +char *cfg_path_ptr =NULL ; +char *rst_path_ptr =NULL ; +char rst_path[PATH_MAX ]; +int new_argc; +bool is_first_case = true; + +memset (args,0 ,sizeof (args)); + +for (i =0 ; i <RTE_DIM (pargs); i++) + pargs[i] = args[i]; + +for (i =0 ; i < (uint32_t )argc; i++) { +if (strncmp (argv[i],CMDLINE_CONFIG_ARG ,MAX_LONG_OPT_SZ ) ==0 ) + cfg_path_ptr = argv[i +1 ]; +if (strncmp (argv[i],CMDLINE_RESULT_ARG ,MAX_LONG_OPT_SZ ) ==0 ) + rst_path_ptr = argv[i +1 ]; + } +if (cfg_path_ptr ==NULL ) { +printf ("Config file not assigned.\n" ); +return -1 ; + } +if (rst_path_ptr ==NULL ) { +strlcpy (rst_path, cfg_path_ptr,PATH_MAX ); +strcat (strtok (basename (rst_path),"." ),"_result.csv" ); + rst_path_ptr = rst_path; + } + + case_nb =load_configs (cfg_path_ptr); + fd =fopen (rst_path_ptr,"w" ); +if (fd ==NULL ) { +printf ("Open output CSV file error.\n" ); +return -1 ; + } +fclose (fd); + +for (i =0 ; i < case_nb; i++) { +if (test_cases[i].test_type ==TEST_TYPE_NONE ) { +printf ("No test type in test case %d.\n\n" , i +1 ); +continue ; + } +if (!test_cases[i].is_valid) { +printf ("Invalid test case %d.\n\n" , i +1 ); +continue ; + } + + cpid =fork (); +if (cpid <0 ) { +printf ("Fork case %d failed.\n" , i +1 ); +exit (EXIT_FAILURE ); + }else if (cpid ==0 ) { +printf ("\nRunning case %u\n\n" , i +1 ); + + new_argc =append_eal_args (argc, argv, test_cases[i].eal_args, pargs); + ret =rte_eal_init (new_argc, pargs); +if (ret <0 ) +rte_exit (EXIT_FAILURE ,"Invalid EAL arguments\n" ); + +/* Check lcores. */ + nb_lcores =rte_lcore_count (); +if (nb_lcores <2 ) +rte_exit (EXIT_FAILURE , +"There should be at least 2 worker lcores.\n" ); + + fd =fopen (rst_path_ptr,"a" ); +if (!fd) { +printf ("Open output CSV file error.\n" ); +return 0 ; + } + +if (is_first_case) { +output_env_info (); + is_first_case = false; + } +run_test (i +1 , &test_cases[i]); + +/* clean up the EAL */ +rte_eal_cleanup (); + +fclose (fd); + +printf ("\nCase %u completed.\n\n" , i +1 ); + +exit (EXIT_SUCCESS ); + }else { + wpid =waitpid (cpid, &wstatus,0 ); +if (wpid ==-1 ) { +printf ("waitpid error.\n" ); +exit (EXIT_FAILURE ); + } + +if (WIFEXITED (wstatus)) +printf ("Case process exited. status %d\n\n" , +WEXITSTATUS (wstatus)); +else if (WIFSIGNALED (wstatus)) +printf ("Case process killed by signal %d\n\n" , +WTERMSIG (wstatus)); +else if (WIFSTOPPED (wstatus)) +printf ("Case process stopped by signal %d\n\n" , +WSTOPSIG (wstatus)); +else if (WIFCONTINUED (wstatus)) +printf ("Case process continued.\n\n" ); +else +printf ("Case process unknown terminated.\n\n" ); + } + } + +printf ("Bye...\n" ); +return 0 ; +}
<snip>
Hi,
Thanks for your comments, replies are inline.
Thanks a again,
Cheng
Quoted 116 lines
-----Original Message-----
From: huangdengdui <[email protected]>
Sent: Tuesday, June 13, 2023 8:55 PM
To: Jiang, Cheng1 <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>; [email protected]; [email protected]
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; Ma, WenwuX <[email protected]>; Wang, YuanX <[email protected]>; He, Xingguang <[email protected]>
Subject: Re: [PATCH v6] app/dma-perf: introduce dma-perf applicationHi Cheng,
Few comments inline. Please check.
Thanks,
DengduiQuoted 1 line · Cheng Jiang
There are many high-performance DMA devices supported in DPDK now,
and
Quoted 9 lines
these DMA devices can also be integrated into other modules of DPDK as accelerators, such as Vhost. Before integrating DMA into applications, developers need to know the performance of these DMA devices in various scenarios and the performance of CPUs in the same scenario, such as different buffer lengths. Only in this way can we know the target performance of the application accelerated by using them. This patch introduces a high-performance testing tool, which supports comparing the performance of CPU and DMA in different scenarios automatically with a pre-set config file. Memory Copy performance test are
supported for now.
Quoted 23 lines
Signed-off-by Cheng Jiang <[email protected]>Signed-off-by Jiayu Hu <[email protected]>Signed-off-by Yuan Wang <[email protected]>Acked-by Morten Brørup <[email protected]>Acked-by Chenbo Xia <[email protected]>---
v6:
improved code based on Anoob's comments; fixed some code structure issues;
v5:
fixed someLONG_LINEwarnings;
v4:
fixed inaccuracy of the memory footprint display;
v3:
fixed some typos;
v2:
added lcore/dmadev designation;
added error case process;
removedworker_threadsparameter from config.ini; improved the logs;
improved config file;Quoted 4 lines
app/meson.build | 1 + app/test-dma-perf/benchmark.c | 477 ++++++++++++++++++++++++++++ app/test-dma-perf/config.ini | 59 ++++ app/test-dma-perf/main.c | 569++++++++++++++++++++++++++++++++++
Quoted 4 lines
app/test-dma-perf/main.h | 69 +++++ app/test-dma-perf/meson.build | 17 + 6 files changed, 1192 insertions(+) create mode 100644 app/test-dma-perf/benchmark.c create mode100644
Quoted 25 lines
app/test-dma-perf/config.ini create mode 100644
app/test-dma-perf/main.ccreate mode 100644app/test-dma-perf/main.hcreate mode 100644 app/test-dma-perf/meson.buildapp/meson.build index renamed from app/meson.build +0
-0
74d2420f67..4fc1a83eba100644app/meson.build +1
-0@@ -19,6 +19,7 @@ apps = [ 'test-cmdline', 'test-compress-perf', 'test-crypto-perf', + 'test-dma-perf', 'test-eventdev', 'test-fib', 'test-flow-perf',(quoted diff) +0
-0
b/app/test-dma-perf/benchmark.cnew file mode 100644 index 0000000000..bc1ca82297app/test-dma-perf/benchmark.c new file +3
-0@@ -0,0 +1,477 @@ +/* SPDX-License-Identifier: BSD-3-Clause +* Copyright(c) 2023 Intel Corporation */ +<snip>
Quoted 6 lines
app/test-dma-perf/benchmark.c new file +6
-0+static inline int +__rte_format_printf(3, 4) +print_err(const char *func, int lineno, const char *format, ...) { + va_list ap; + int ret; +app/test-dma-perf/benchmark.c
Quoted 10 lines
app/test-dma-perf/benchmark.c new file +10
-0+ ret =fprintf (stderr,"In %s:%d - " , func, lineno); +va_start (ap, format); + ret +=vfprintf (stderr, format, ap); +va_end (ap); + +return ret; +} + +static inline void +calc_result(uint32_t buf_size, uint32_t nr_buf, uint16_t nb_workers,
uint16_ttest_secs,app/test-dma-perf/benchmark.c
Quoted 1 line
app/test-dma-perf/benchmark.c new file +1
-0+ uint32_t total_cnt, float *memory, uint32_t
*ave_cycle,app/test-dma-perf/benchmark.c
Quoted 3 lines
app/test-dma-perf/benchmark.c new file +3
-0+ float *bandwidth, float *mops) +{ + *memory = (float)(buf_size * (nr_buf / nb_workers) * 2) / (1024 *1024);
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 2 lines
app/test-dma-perf/benchmark.c new file +2
-0+ *ave_cycle = test_secs *rte_get_timer_hz () / total_cnt; + *bandwidth = (buf_size *8 * (rte_get_timer_hz () / (float )*ave_cycle))/ 1000000000;
app/test-dma-perf/benchmark.c line 78
Quoted 1 line
app/test-dma-perf/benchmark.c new file +1
-0+ *mops = (float )rte_get_timer_hz () / *ave_cycle /1000000 ;The value of
ave_cyclemay be 0.*mops = (float)(total_cnt /
test_secs) / 1000000; ?
OK, it makes sense to me. I'll fix it in the next version.
Quoted 71 lines
app/test-dma-perf/benchmark.c
Quoted 5 lines
app/test-dma-perf/benchmark.c new file +4
-0+} + +static void +output_result(uint8_t scenario_id, uint32_t lcore_id, char *dma_name,
uint64_tave_cycle,app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+ uint32_t buf_size, uint32_t nr_buf, float memory, + float bandwidth, float mops, bool is_dma) { + if (is_dma) + printf("lcore %u, DMA %s:\n", lcore_id, dma_name); + else + printf("lcore %u\n", lcore_id); + + printf("average cycles/op: %" PRIu64 ", buffer size: %u, nr_buf: %u,memory: %.2lfMB, frequency: %" PRIu64 ".\n",
app/test-dma-perf/benchmark.c
Quoted 1 line
app/test-dma-perf/benchmark.c new file +1
-0+ ave_cycle, buf_size, nr_buf, memory,
rte_get_timer_hz());app/test-dma-perf/benchmark.c
Quoted 5 lines
app/test-dma-perf/benchmark.c new file +5
-0+printf ("Average bandwidth: %.3lfGbps, MOps: %.3lf\n" , bandwidth, +mops); + +if (is_dma) + snprintf(output_str[lcore_id], MAX_OUTPUT_STR_LEN,
CSV_LINE_DMA_FMT,app/test-dma-perf/benchmark.c
Quoted 4 lines
app/test-dma-perf/benchmark.c new file +4
-0+ scenario_id, lcore_id, dma_name, buf_size, + nr_buf, memory, ave_cycle, bandwidth, mops); + else + snprintf(output_str[lcore_id], MAX_OUTPUT_STR_LEN,
CSV_LINE_CPU_FMT,Quoted 21 lines
app/test-dma-perf/benchmark.c new file +21
-0+ scenario_id, lcore_id, buf_size, + nr_buf, memory, ave_cycle, bandwidth, mops); } + +static inline void +cache_flush_buf (__maybe_unused struct rte_mbuf **array, +__maybe_unused uint32_t buf_size, +__maybe_unused uint32_t nr_buf) +{ +#ifdef RTE_ARCH_X86_64 +char *data; +struct rte_mbuf **srcs = array; +uint32_t i, offset; + +for (i =0 ; i < nr_buf; i++) { + data = rte_pktmbuf_mtod(srcs[i], char *); +for (offset =0 ; offset < buf_size; offset +=64 ) +__builtin_ia32_clflush (data + offset); + } +#endif +} +<snip>
app/test-dma-perf/benchmark.c
Quoted 12 lines
app/test-dma-perf/benchmark.c new file +12
-0+ +/* Parse the argument given in the command line of the application */ +static int append_eal_args (int argc,char **argv,const char +*eal_args,char **new_argv) { +int i; +char *tokens[MAX_EAL_PARAM_NB ]; +char args[MAX_EAL_PARAM_LEN ] = {0 }; +int token_nb, new_argc =0 ; + +for (i =0 ; i < argc; i++) { +if ((strcmp (argv[i],CMDLINE_CONFIG_ARG ) ==0 ) || + (strcmp (argv[i],CMDLINE_RESULT_ARG ) ==0 )){
app/test-dma-perf/benchmark.c
Quoted 4 lines
app/test-dma-perf/benchmark.c new file +4
-0+ i++; +continue ; + } + strlcpy(new_argv[new_argc], argv[i],
sizeof(new_argv[new_argc]));
Quoted 4 lines
The
type_ofargv[new_argc] is *char. Cannot usesizeof().
strcpy(new_argv[new_argc], argv[i]); orstrlcpy(new_argv[new_argc], argv[i], MAX_EAL_PARAM_LEN); ?
Yes, it's a mistake, thanks for pointing out! I'll fix it in the next version.
Quoted 143 lines
Quoted 28 lines
app/test-dma-perf/benchmark.c new file +27
-0+ new_argc++; + } + +if (eal_args) { +strlcpy (args, eal_args,sizeof (args)); + token_nb =rte_strsplit (args,strlen (args), + tokens,MAX_EAL_PARAM_NB ,' ' ); +for (i =0 ; i < token_nb; i++) +strcpy (new_argv[new_argc++], tokens[i]); + } + +return new_argc; +} + +int +main (int argc,char *argv[]) +{ +int ret; +uint16_t case_nb; +uint32_t i, nb_lcores; +pid_t cpid, wpid; +int wstatus; +char args[MAX_EAL_PARAM_NB ][MAX_EAL_PARAM_LEN ]; +char *pargs[MAX_EAL_PARAM_NB ]; +char *cfg_path_ptr =NULL ; +char *rst_path_ptr =NULL ; +char rst_path[PATH_MAX ];app/test-dma-perf/benchmark.c
Quoted 10 lines
app/test-dma-perf/benchmark.c new file +10
-0+int new_argc; +bool is_first_case = true; + +memset (args,0 ,sizeof (args)); + +for (i =0 ; i <RTE_DIM (pargs); i++) + pargs[i] = args[i]; + +for (i =0 ; i < (uint32_t )argc; i++) { + if (strncmp(argv[i], CMDLINE_CONFIG_ARG,
MAX_LONG_OPT_SZ) == 0)app/test-dma-perf/benchmark.c
Quoted 2 lines
app/test-dma-perf/benchmark.c new file +2
-0+ cfg_path_ptr = argv[i +1 ]; + if (strncmp(argv[i], CMDLINE_RESULT_ARG,
MAX_LONG_OPT_SZ) == 0)Quoted 28 lines
app/test-dma-perf/benchmark.c new file +28
-0+ rst_path_ptr = argv[i +1 ]; + } +if (cfg_path_ptr ==NULL ) { +printf ("Config file not assigned.\n" ); +return -1 ; + } +if (rst_path_ptr ==NULL ) { +strlcpy (rst_path, cfg_path_ptr,PATH_MAX ); +strcat (strtok (basename (rst_path),"." ),"_result.csv" ); + rst_path_ptr = rst_path; + } + + case_nb =load_configs (cfg_path_ptr); + fd =fopen (rst_path_ptr,"w" ); +if (fd ==NULL ) { +printf ("Open output CSV file error.\n" ); +return -1 ; + } +fclose (fd); + +for (i =0 ; i < case_nb; i++) { +if (test_cases[i].test_type ==TEST_TYPE_NONE ) { +printf ("No test type in test case %d.\n\n" , i +1 ); +continue ; + } +if (!test_cases[i].is_valid) { +printf ("Invalid test case %d.\n\n" , i +1 ); +continue ;app/test-dma-perf/benchmark.c
Quoted 10 lines
app/test-dma-perf/benchmark.c new file +10
-0+ } + + cpid =fork (); +if (cpid <0 ) { +printf ("Fork case %d failed.\n" , i +1 ); +exit (EXIT_FAILURE ); + }else if (cpid ==0 ) { +printf ("\nRunning case %u\n\n" , i +1 ); + + new_argc = append_eal_args(argc, argv,test_cases[i].eal_args, pargs);
app/test-dma-perf/benchmark.c
Quoted 3 lines
app/test-dma-perf/benchmark.c new file +3
-0+ ret =rte_eal_init (new_argc, pargs); +if (ret <0 ) + rte_exit(EXIT_FAILURE, "Invalid EALarguments\n");
app/test-dma-perf/benchmark.c
Quoted 6 lines
app/test-dma-perf/benchmark.c new file +6
-0+ +/* Check lcores. */ + nb_lcores =rte_lcore_count (); +if (nb_lcores <2 ) + rte_exit(EXIT_FAILURE, + "There should be at least 2 workerlcores.\n");
Quoted 26 lines
app/test-dma-perf/benchmark.c new file +26
-0+ + fd =fopen (rst_path_ptr,"a" ); +if (!fd) { +printf ("Open output CSV file error.\n" ); +return 0 ; + } + +if (is_first_case) { +output_env_info (); + is_first_case = false; + } +run_test (i +1 , &test_cases[i]); + +/* clean up the EAL */ +rte_eal_cleanup (); + +fclose (fd); + +printf ("\nCase %u completed.\n\n" , i +1 ); + +exit (EXIT_SUCCESS ); + }else { + wpid =waitpid (cpid, &wstatus,0 ); +if (wpid ==-1 ) { +printf ("waitpid error.\n" ); +exit (EXIT_FAILURE );app/test-dma-perf/benchmark.c
Quoted 10 lines
app/test-dma-perf/benchmark.c new file +10
-0+ } + +if (WIFEXITED (wstatus)) +printf ("Case process exited. status %d\n\n" , +WEXITSTATUS (wstatus)); +else if (WIFSIGNALED (wstatus)) +printf ("Case process killed by signal %d\n\n" , +WTERMSIG (wstatus)); +else if (WIFSTOPPED (wstatus)) + printf("Case process stopped bysignal %d\n\n",
app/test-dma-perf/benchmark.c
Quoted 5 lines
app/test-dma-perf/benchmark.c new file +5
-0+ WSTOPSIG(wstatus)); + else if (WIFCONTINUED(wstatus)) + printf("Case process continued.\n\n"); + else + printf("Case process unknownterminated.\n\n");
Quoted 6 lines
app/test-dma-perf/benchmark.c new file +6
-0+ } + } + +printf ("Bye...\n" ); +return 0 ; +}<snip>
Hi,
Thanks for working on the comments. Few more top level comment inline.
Thanks,
Anoob
Quoted 262 lines
-----Original Message-----
From: Cheng Jiang <[email protected]>
Sent: Tuesday, June 13, 2023 10:02 AM
To: [email protected]; [email protected]; [email protected]; [email protected]; Amit Prakash Shukla <[email protected]>; Anoob Joseph <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; Cheng Jiang <[email protected]>
Subject: [EXT] [PATCH v6] app/dma-perf: introduce dma-perf applicationExternal Email
----------------------------------------------------------------------
There are many high-performance DMA devices supported in DPDK now, and these DMA devices can also be integrated into other modules of DPDK as accelerators, such as Vhost. Before integrating DMA into applications, developers need to know the performance of these DMA devices in various scenarios and the performance of CPUs in the same scenario, such as different buffer lengths. Only in this way can we know the target performance of the application accelerated by using them. This patch introduces a high-performance testing tool, which supports comparing the performance of CPU and DMA in different scenarios automatically with a pre- set config file. Memory Copy performance test are supported for now.Signed-off-by Cheng Jiang <[email protected]>Signed-off-by Jiayu Hu <[email protected]>Signed-off-by Yuan Wang <[email protected]>Acked-by Morten Brørup <[email protected]>Acked-by Chenbo Xia <[email protected]>---
v6:
improved code based on Anoob's comments; fixed some code structure issues;
v5:
fixed someLONG_LINEwarnings;
v4:
fixed inaccuracy of the memory footprint display;
v3:
fixed some typos;
v2:
added lcore/dmadev designation;
added error case process;
removedworker_threadsparameter from config.ini; improved the logs;
improved config file;app/meson.build | 1 + app/test-dma-perf/benchmark.c | 477 ++++++++++++++++++++++++++++ app/test-dma-perf/config.ini | 59 ++++ app/test-dma-perf/main.c | 569 ++++++++++++++++++++++++++++++++++ app/test-dma-perf/main.h | 69 +++++ app/test-dma-perf/meson.build | 17 + 6 files changed, 1192 insertions(+) create mode 100644 app/test-dma-perf/benchmark.c create mode 100644app/test-dma-perf/config.ini create mode 100644 app/test-dma-
perf/main.ccreate mode 100644app/test-dma-perf/main.hcreate mode 100644 app/test-dma-perf/meson.buildapp/meson.build index renamed from app/meson.build +0
-0
74d2420f67..4fc1a83eba100644app/meson.build +1
-0@@ -19,6 +19,7 @@ apps = [ 'test-cmdline', 'test-compress-perf', 'test-crypto-perf', + 'test-dma-perf', 'test-eventdev', 'test-fib', 'test-flow-perf',app/test-dma- renamed from app/test-dma-perf/benchmark.c +0
-0
perf/benchmark.cnew file mode 100644 index 0000000000..bc1ca82297app/test-dma-perf/benchmark.c new file +22
-0@@ -0,0 +1,477 @@ +/* SPDX-License-Identifier: BSD-3-Clause +* Copyright(c) 2023 Intel Corporation +*/ + +#include <inttypes.h> +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> + +#include <rte_time.h> +#include <rte_mbuf.h> +#include <rte_dmadev.h> +#include <rte_malloc.h> +#include <rte_lcore.h> + +#include "main.h" + +#define MAX_DMA_CPL_NB 255 + +#define TEST_WAIT_U_SECOND 10000 + +#define CSV_LINE_DMA_FMT "Scenario %u,%u,%s,%u,%u,%.2lf,%" PRIu64",%.3lf,%.3lf\n"
+#defineCSV_LINE_CPU_FMT"Scenario %u,%u,NA,%u,%u,%.2lf,%" PRIu64 ",%.3lf,%.3lf\n"app/test-dma-perf/benchmark.c new file +47
-0+ +struct worker_info { +bool ready_flag; +bool start_flag; +bool stop_flag; +uint32_t total_cpl; +uint32_t test_cpl; +}; + +struct lcore_params { +uint8_t scenario_id; +unsigned int lcore_id; +char *dma_name; +uint16_t worker_id; +uint16_t dev_id; +uint32_t nr_buf; +uint16_t kick_batch; +uint32_t buf_size; +uint16_t test_secs; +struct rte_mbuf **srcs; +struct rte_mbuf **dsts; +struct worker_info worker_info; +}; + +static struct rte_mempool *src_pool; +static struct rte_mempool *dst_pool; + +static volatile struct lcore_params *worker_params[MAX_WORKER_NB ]; + +#define PRINT_ERR (...) print_err(__func__, __LINE__, __VA_ARGS__) + +static inline int +__rte_format_printf(3, 4) +print_err(const char *func, int lineno, const char *format, ...) { + va_list ap; + int ret; + + ret = fprintf(stderr, "In %s:%d - ", func, lineno); + va_start(ap, format); + ret += vfprintf(stderr, format, ap); + va_end(ap); + + return ret; +} + +static inline void +calc_result(uint32_t buf_size, uint32_t nr_buf, uint16_t nb_workers,
uint16_ttest_secs,app/test-dma-perf/benchmark.c new file +1
-0+ uint32_t total_cnt, float *memory, uint32_t
*ave_cycle,app/test-dma-perf/benchmark.c new file +3
-0+ float *bandwidth, float *mops) +{ + *memory = (float)(buf_size * (nr_buf / nb_workers) * 2) / (1024 *1024);
app/test-dma-perf/benchmark.c new file +2
-0+ *ave_cycle = test_secs *rte_get_timer_hz () / total_cnt; + *bandwidth = (buf_size * 8 * (rte_get_timer_hz() /(float)*ave_cycle)) / 1000000000;
app/test-dma-perf/benchmark.c new file +4
-0+ *mops = (float )rte_get_timer_hz () / *ave_cycle /1000000 ; } + +static void +output_result(uint8_t scenario_id, uint32_t lcore_id, char *dma_name,
uint64_tave_cycle,app/test-dma-perf/benchmark.c new file +8
-0+ uint32_t buf_size, uint32_t nr_buf, float memory, + float bandwidth, float mops, bool is_dma) { + if (is_dma) + printf("lcore %u, DMA %s:\n", lcore_id, dma_name); + else + printf("lcore %u\n", lcore_id); + + printf("average cycles/op: %" PRIu64 ", buffer size: %u, nr_buf: %u,memory: %.2lfMB, frequency: %" PRIu64 ".\n",
app/test-dma-perf/benchmark.c new file +1
-0+ ave_cycle, buf_size, nr_buf, memory,
rte_get_timer_hz());app/test-dma-perf/benchmark.c new file +5
-0+printf ("Average bandwidth: %.3lfGbps, MOps: %.3lf\n" , bandwidth, +mops); + +if (is_dma) + snprintf(output_str[lcore_id], MAX_OUTPUT_STR_LEN,
CSV_LINE_DMA_FMT,app/test-dma-perf/benchmark.c new file +4
-0+ scenario_id, lcore_id, dma_name, buf_size, + nr_buf, memory, ave_cycle, bandwidth, mops); + else + snprintf(output_str[lcore_id], MAX_OUTPUT_STR_LEN,
CSV_LINE_CPU_FMT,app/test-dma-perf/benchmark.c new file +41
-0+ scenario_id, lcore_id, buf_size, + nr_buf, memory, ave_cycle, bandwidth, mops); } + +static inline void +cache_flush_buf (__maybe_unused struct rte_mbuf **array, +__maybe_unused uint32_t buf_size, +__maybe_unused uint32_t nr_buf) +{ +#ifdef RTE_ARCH_X86_64 +char *data; +struct rte_mbuf **srcs = array; +uint32_t i, offset; + +for (i =0 ; i < nr_buf; i++) { + data = rte_pktmbuf_mtod(srcs[i], char *); +for (offset =0 ; offset < buf_size; offset +=64 ) +__builtin_ia32_clflush (data + offset); + } +#endif +} + +/* Configuration of device. */ +static void +configure_dmadev_queue (uint32_t dev_id,uint32_t ring_size) { +uint16_t vchan =0 ; +struct rte_dma_info info; +struct rte_dma_conf dev_config = { .nb_vchans =1 }; +struct rte_dma_vchan_conf qconf = { + .direction =RTE_DMA_DIR_MEM_TO_MEM , + .nb_desc = ring_size + }; + +if (rte_dma_configure (dev_id, &dev_config) !=0 ) +rte_exit (EXIT_FAILURE ,"Error with dma configure.\n" ); + +if (rte_dma_vchan_setup (dev_id, vchan, &qconf) !=0 ) +rte_exit (EXIT_FAILURE ,"Error with queue configuration.\n" ); + +rte_dma_info_get (dev_id, &info); +if (info.nb_vchans !=1 ) + rte_exit(EXIT_FAILURE, "Error, no configured queuesreported on device id. %u\n",
app/test-dma-perf/benchmark.c new file +20
-0+ dev_id); + +if (rte_dma_start (dev_id) !=0 ) +rte_exit (EXIT_FAILURE ,"Error with dma start.\n" ); } + +static int +config_dmadevs (struct test_configure *cfg) { +uint32_t ring_size = cfg->ring_size.cur; +struct lcore_dma_map_t *ldm = &cfg->lcore_dma_map; +uint32_t nb_workers = ldm->cnt; +uint32_t i; +int dev_id; +uint16_t nb_dmadevs =0 ; +char *dma_name; + +for (i =0 ; i < ldm->cnt; i++) { + dma_name = ldm->dma_names[i]; + dev_id =rte_dma_get_dev_id_by_name (dma_name); +if (dev_id ==-1 ) { + fprintf(stderr, "Error: Fail to find DMA %s.\n",
dma_name);app/test-dma-perf/benchmark.c new file +11
-0+ goto end; + } + + ldm->dma_ids[i] = dev_id; +configure_dmadev_queue (dev_id, ring_size); + ++nb_dmadevs; + } + +end: +if (nb_dmadevs < nb_workers) { + printf("Not enough dmadevs (%u) for all workers (%u).\n",
nb_dmadevs,nb_workers);app/test-dma-perf/benchmark.c new file +1
-0+return -1 ;
app/test-dma-perf/benchmark.c line 187
Quoted 10 lines
app/test-dma-perf/benchmark.c new file +10
-0+ } + +printf ("Number of used dmadevs: %u.\n" , nb_dmadevs); + +return 0 ; +} + +#define POLL_MAX 1000 + +
[Anoob] Extra blank line. You can consider removing.
Quoted 137 lines
app/test-dma-perf/benchmark.c new file +14
-0+static inline void +do_dma_submit_and_poll (uint16_t dev_id,uint64_t *async_cnt, +volatile struct worker_info *worker_info) { +int ret; +uint16_t nr_cpl; + + ret =rte_dma_submit (dev_id,0 ); +if (ret <0 ) { +rte_dma_stop (dev_id); +rte_dma_close (dev_id); +rte_exit (EXIT_FAILURE ,"Error with dma submit.\n" ); + } + + nr_cpl = rte_dma_completed(dev_id, 0, MAX_DMA_CPL_NB, NULL,NULL);
app/test-dma-perf/benchmark.c new file +32
-0+ *async_cnt -= nr_cpl; + worker_info->total_cpl += nr_cpl; +} + +static inline int +do_dma_mem_copy (void *p) +{ +const uint16_t *para_idx = (uint16_t *)p; +volatile struct lcore_params *para = worker_params[*para_idx]; +volatile struct worker_info *worker_info = &(para->worker_info); +const uint16_t dev_id = para->dev_id; +const uint32_t nr_buf = para->nr_buf; +const uint16_t kick_batch = para->kick_batch; +const uint32_t buf_size = para->buf_size; +struct rte_mbuf **srcs = para->srcs; +struct rte_mbuf **dsts = para->dsts; +uint16_t nr_cpl; +uint64_t async_cnt =0 ; +uint32_t i; +uint32_t poll_cnt =0 ; +int ret; + + worker_info->stop_flag = false; + worker_info->ready_flag = true; + +while (!worker_info->start_flag) + ; + +while (1 ) { +for (i =0 ; i < nr_buf; i++) { +dma_copy: + ret = rte_dma_copy(dev_id, 0,
rte_pktmbuf_iova(srcs[i]),app/test-dma-perf/benchmark.c new file +4
-0+ rte_pktmbuf_iova(dsts[i]), buf_size, 0); +if (unlikely (ret <0 )) { +if (ret == -ENOSPC ) { + do_dma_submit_and_poll(dev_id,
&async_cnt,worker_info);app/test-dma-perf/benchmark.c new file +5
-0+ goto dma_copy; + }else { +/* Error exit */ +rte_dma_stop (dev_id); + rte_exit(EXIT_FAILURE, "DMAenqueue failed\n");
app/test-dma-perf/benchmark.c new file +6
-0+ } + } + async_cnt++; + +if ((async_cnt % kick_batch) ==0 ) + do_dma_submit_and_poll(dev_id,
&async_cnt,worker_info);app/test-dma-perf/benchmark.c new file +9
-0+ } + +if (worker_info->stop_flag) +break ; + } + +rte_dma_submit (dev_id,0 ); +while ((async_cnt >0 ) && (poll_cnt++ <POLL_MAX )) { + nr_cpl = rte_dma_completed(dev_id, 0,
MAX_DMA_CPL_NB, NULL, NULL);app/test-dma-perf/benchmark.c new file +28
-0+ async_cnt -= nr_cpl; + } + +return 0 ; +} + +static inline int +do_cpu_mem_copy (void *p) +{ +const uint16_t *para_idx = (uint16_t *)p; +volatile struct lcore_params *para = worker_params[*para_idx]; +volatile struct worker_info *worker_info = &(para->worker_info); +const uint32_t nr_buf = para->nr_buf; +const uint32_t buf_size = para->buf_size; +struct rte_mbuf **srcs = para->srcs; +struct rte_mbuf **dsts = para->dsts; +uint32_t i; + + worker_info->stop_flag = false; + worker_info->ready_flag = true; + +while (!worker_info->start_flag) + ; + +while (1 ) { +for (i =0 ; i < nr_buf; i++) { +/* copy buffer form src to dst */ + rte_memcpy((void*)(uintptr_t)rte_mbuf_data_iova(dsts[i]),
app/test-dma-perf/benchmark.c new file +1
-0+ (void*)(uintptr_t)rte_mbuf_data_iova(srcs[i]),
app/test-dma-perf/benchmark.c new file +22
-0+ (size_t)buf_size); + worker_info->total_cpl++; + } +if (worker_info->stop_flag) +break ; + } + +return 0 ; +} + +static int +setup_memory_env (struct test_configure *cfg,struct rte_mbuf ***srcs, +struct rte_mbuf ***dsts) +{ +unsigned int buf_size = cfg->buf_size.cur; +unsigned int nr_sockets; +uint32_t nr_buf = cfg->nr_buf; + + nr_sockets =rte_socket_count (); +if (cfg->src_numa_node >= nr_sockets || + cfg->dst_numa_node >= nr_sockets) { + printf("Error: Source or destination numa exceeds the acturenuma nodes.\n");
app/test-dma-perf/benchmark.c new file +7
-0+return -1 ; + } + + src_pool = rte_pktmbuf_pool_create("Benchmark_DMA_SRC", + nr_buf, /* n == num elements */ + 64, /* cache size */ + 0, /* priv size */
app/test-dma-perf/benchmark.c line 326
Quoted 10 lines
app/test-dma-perf/benchmark.c new file +10
-0+ buf_size +RTE_PKTMBUF_HEADROOM , + cfg->src_numa_node); +if (src_pool ==NULL ) { +PRINT_ERR ("Error with source mempool creation.\n" ); +return -1 ; + } + + dst_pool = rte_pktmbuf_pool_create("Benchmark_DMA_DST", + nr_buf, /* n == num elements */ + 64, /* cache size */
[Anoob] We do not alloc or free pointers in the datapath, right? So why bother with cache?
app/test-dma-perf/benchmark.c line 339
Quoted 13 lines
app/test-dma-perf/benchmark.c new file +13
-0+0 ,/* priv size */ + buf_size +RTE_PKTMBUF_HEADROOM , + cfg->dst_numa_node); +if (dst_pool ==NULL ) { +PRINT_ERR ("Error with destination mempool creation.\n" ); +return -1 ; + } + + *srcs =rte_malloc (NULL , nr_buf *sizeof (struct rte_mbuf *),0 ); +if (*srcs ==NULL ) { +printf ("Error: srcs malloc failed.\n" ); +return -1 ; + }
[Anoob] Are we freeing these memory? The ones allocated with rte_malloc.
Quoted 61 lines
app/test-dma-perf/benchmark.c new file +29
-0+ + *dsts =rte_malloc (NULL , nr_buf *sizeof (struct rte_mbuf *),0 ); +if (*dsts ==NULL ) { +printf ("Error: dsts malloc failed.\n" ); +return -1 ; + } + +if (rte_mempool_get_bulk (src_pool, (void **)*srcs, nr_buf) !=0 ) { +printf ("get src mbufs failed.\n" ); +return -1 ; + } +if (rte_mempool_get_bulk (dst_pool, (void **)*dsts, nr_buf) !=0 ) { +printf ("get dst mbufs failed.\n" ); +return -1 ; + } + +return 0 ; +} + +void +mem_copy_benchmark (struct test_configure *cfg,bool is_dma) { +uint16_t i; +uint32_t offset; +unsigned int lcore_id =0 ; +struct rte_mbuf **srcs =NULL , **dsts =NULL ; +struct lcore_dma_map_t *ldm = &cfg->lcore_dma_map; +unsigned int buf_size = cfg->buf_size.cur; +uint16_t kick_batch = cfg->kick_batch.cur; + uint32_t nr_buf = cfg->nr_buf = (cfg->mem_size.cur * 1024 * 1024) /(
cfg->buf_size.cur* 2);app/test-dma-perf/benchmark.c new file +27
-0+uint16_t nb_workers = ldm->cnt; +uint16_t test_secs = cfg->test_secs; +float memory; +uint32_t avg_cycles =0 ; +float mops; +float bandwidth; + +if (setup_memory_env (cfg, &srcs, &dsts) <0 ) + goto out; + +if (is_dma) +if (config_dmadevs (cfg) <0 ) + goto out; + +if (cfg->cache_flush) { +cache_flush_buf (srcs, buf_size, nr_buf); +cache_flush_buf (dsts, buf_size, nr_buf); +rte_mb (); + } + +printf ("Start testing....\n" ); + +for (i =0 ; i < nb_workers; i++) { + lcore_id = ldm->lcores[i]; + offset = nr_buf / nb_workers * i; + + worker_params[i] = rte_malloc(NULL, sizeof(struct
lcore_params), 0);app/test-dma-perf/benchmark.c new file +2
-0+if (!worker_params[i]) { + printf("lcore parameters malloc failure for lcore%d\n",
lcore_id);
app/test-dma-perf/benchmark.c
Quoted 2 lines
app/test-dma-perf/benchmark.c new file +2
-0+break ; + }
[Anoob] Are we freeing the above memory?
Quoted 137 lines
app/test-dma-perf/benchmark.c new file +2
-0+if (is_dma) { + worker_params[i]->dma_name = ldm-Quoted 1 line
dma_names[i];
app/test-dma-perf/benchmark.c new file +5
-0+ worker_params[i]->dev_id = ldm->dma_ids[i]; + worker_params[i]->kick_batch = kick_batch; + } + worker_params[i]->worker_id = i; + worker_params[i]->nr_buf = (uint32_t)(nr_buf /
nb_workers);app/test-dma-perf/benchmark.c new file +9
-0+ worker_params[i]->buf_size = buf_size; + worker_params[i]->test_secs = test_secs; + worker_params[i]->srcs = srcs + offset; + worker_params[i]->dsts = dsts + offset; + worker_params[i]->scenario_id = cfg->scenario_id; + worker_params[i]->lcore_id = lcore_id; + +if (is_dma) + rte_eal_remote_launch(do_dma_mem_copy, (void*)(&i),
lcore_id);app/test-dma-perf/benchmark.c new file +2
-0+else + rte_eal_remote_launch(do_cpu_mem_copy, (void*)(&i),
lcore_id);app/test-dma-perf/benchmark.c new file +6
-0+ } + +while (1 ) { +bool ready = true; +for (i =0 ; i < nb_workers; i++) { + if (worker_params[i]->worker_info.ready_flag ==false) {
app/test-dma-perf/benchmark.c new file +19
-0+ ready =0 ; +break ; + } + } +if (ready) +break ; + } + +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.start_flag = true; + +usleep (TEST_WAIT_U_SECOND ); +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.test_cpl = +worker_params[i]->worker_info.total_cpl; + +usleep (test_secs *1000 *1000 ); +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.test_cpl =worker_params[i]->worker_info.total_cpl -
app/test-dma-perf/benchmark.c new file +1
-0+ worker_params[i]-Quoted 1 line
worker_info.test_cpl;
app/test-dma-perf/benchmark.c new file +12
-0+ +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.stop_flag = true; + +rte_eal_mp_wait_lcore (); + +for (i =0 ; i < nb_workers; i++) { +calc_result (buf_size, nr_buf, nb_workers, test_secs, + worker_params[i]->worker_info.test_cpl, + &memory, &avg_cycles, &bandwidth, &mops); + output_result(cfg->scenario_id, worker_params[i]->lcore_id, + worker_params[i]->dma_name,
avg_cycles,buf_size,app/test-dma-perf/benchmark.c new file +1
-0+ nr_buf / nb_workers, memory,bandwidth, mops,
is_dma);app/test-dma-perf/benchmark.c new file +21
-0+ } + +out: +/* free env */ +if (srcs) +rte_pktmbuf_free_bulk (srcs, nr_buf); +if (dsts) +rte_pktmbuf_free_bulk (dsts, nr_buf); + +if (src_pool) +rte_mempool_free (src_pool); +if (dst_pool) +rte_mempool_free (dst_pool); + +if (is_dma) { +for (i =0 ; i < nb_workers; i++) { +printf ("Stopping dmadev %d\n" , ldm->dma_ids[i]); +rte_dma_stop (ldm->dma_ids[i]); + } + } +}app/test-dma-perf/config.ini new renamed from app/test-dma-perf/config.ini +0
-0file mode 100644 index 0000000000..2fd9c3c387
app/test-dma-perf/config.ini new file +10
-0@@ -0,0 +1,59 @@ + +; This is an example configuration file for dma-perf, which details the +meanings of each parameter ; and instructions on how to use dma-perf. + +; Supported test types are DMA_MEM_COPY and CPU_MEM_COPY. + +; Parameters: +; "mem_size" denotes the size of the memory footprint. +; "buf_size" denotes the memory size of a single operation. +; "dma_ring_size" denotes the dma ring buffer size. It should be greaterthan 64 normally.
+; "kick_batch" denotes the dma operation batch size, and should be greater than 1 normally.app/test-dma-perf/config.ini new file +4
-0+ +; The format for variables is variable=first,last,increment,ADD|MUL. + +; src_numa_node is used to control the numa node where the sourcememory is allocated.
+;dst_numa_nodeis used to control the numa node where the destination memory is allocated.app/test-dma-perf/config.ini new file +9
-0+ +; cache_flush is used to determine whether or not the cache should be +flushed, with 1 indicating to ; flush and 0 indicating to not flush. + +; test_seconds controls the test time of the whole case. + +; To use DMA for a test, please specify the "lcore_dma" parameter. +; If you have already set the "-l" and "-a" parameters using EAL, ; +make sure that the value of "lcore_dma" falls within their range of thevalues.
app/test-dma-perf/config.ini new file +6
-0+ +; To use CPU for a test, please specify the "lcore" parameter. +; If you have already set the "-l" and "-a" parameters using EAL, ; +make sure that the value of "lcore" falls within their range of values. + +; To specify a configuration file, use the "--config" flag followed by the pathto the file.
app/test-dma-perf/config.ini new file +2
-0+ +; To specify a result file, use the "--result" flag followed by the path to thefile.
app/test-dma-perf/config.ini new file +2
-0+; If you do not specify a result file, one will be generated with the +same name as the configuration ; file, with the addition of "_result.csv" atthe end.
app/test-dma-perf/config.ini new file +2
-0+ +[case1]
app/test-dma-perf/config.ini line 47
Quoted 10 lines
app/test-dma-perf/config.ini new file +10
-0+type=DMA_MEM_COPY +mem_size=10 +buf_size=64,8192,2,MUL +dma_ring_size=1024 +kick_batch=32 +src_numa_node=0 +dst_numa_node=0 +cache_flush=0 +test_seconds=2 +lcore_dma=lcore10@0000:00:04.2, lcore11@0000:00:04.3
[Anoob] Isn't it better if we allow user to specify DMA dev ID rather than the PCI DBDF?
In the long run, I would expect config file to provide {core, dma_dev_id, queue_id}
Another thought is why to expose this at all? If we can restrict this perf application to have one thread only use one vchan, then application can easily create this mapping in run time. Unless you want one thread to use 2 different vchans which may not be desirable since this is a standalone perf app.
Quoted 544 lines
app/test-dma-perf/config.ini new file +12
-0+eal_args=--in-memory --file-prefix=test + +[case2] +type=CPU_MEM_COPY +mem_size=10 +buf_size=64,8192,2,MUL +src_numa_node=0 +dst_numa_node=1 +cache_flush=0 +test_seconds=2 +lcore = 3, 4 +eal_args=--in-memory --no-pciapp/test-dma-perf/main.c new file renamed from app/test-dma-perf/main.c +0
-0mode 100644 index 0000000000..d65655b87b
app/test-dma-perf/main.c new file +22
-0@@ -0,0 +1,569 @@ +/* SPDX-License-Identifier: BSD-3-Clause +* Copyright(c) 2023 Intel Corporation +*/ + +#include <stdio.h> +#include <stdlib.h> +#include <getopt.h> +#include <signal.h> +#include <stdbool.h> +#include <unistd.h> +#include <sys/wait.h> +#include <inttypes.h> +#include <libgen.h> + +#include <rte_eal.h> +#include <rte_cfgfile.h> +#include <rte_string_fns.h> +#include <rte_lcore.h> + +#include "main.h" + +#define CSV_HDR_FMT "Case %u : %s,lcore,DMA,buffersize,nr_buf,memory(MB),cycle,bandwidth(Gbps),MOps\n"
app/test-dma-perf/main.c new file +51
-0+ +#define MAX_EAL_PARAM_NB 100 +#define MAX_EAL_PARAM_LEN 1024 + +#define DMA_MEM_COPY "DMA_MEM_COPY" +#define CPU_MEM_COPY "CPU_MEM_COPY" + +#define CMDLINE_CONFIG_ARG "--config" +#define CMDLINE_RESULT_ARG "--result" + +#define MAX_PARAMS_PER_ENTRY 4 + +#define MAX_LONG_OPT_SZ 64 + +enum { +TEST_TYPE_NONE =0 , +TEST_TYPE_DMA_MEM_COPY , +TEST_TYPE_CPU_MEM_COPY +}; + +#define MAX_TEST_CASES 16 +static struct test_configure test_cases[MAX_TEST_CASES ]; + +char output_str[MAX_WORKER_NB ][MAX_OUTPUT_STR_LEN ]; + +static FILE *fd; + +static void +output_csv (bool need_blankline) +{ +uint32_t i; + +if (need_blankline) { +fprintf (fd,",,,,,,,,\n" ); +fprintf (fd,",,,,,,,,\n" ); + } + +for (i =0 ; i <RTE_DIM (output_str); i++) { +if (output_str[i][0 ]) { +fprintf (fd,"%s" , output_str[i]); + output_str[i][0 ] ='\0' ; + } + } + +fflush (fd); +} + +static void +output_env_info (void ) +{ + snprintf(output_str[0], MAX_OUTPUT_STR_LEN, "testenvironment:\n");
app/test-dma-perf/main.c new file +67
-0+snprintf (output_str[1 ],MAX_OUTPUT_STR_LEN ,"CPU frequency,%" + PRIu64"\n" ,rte_get_timer_hz ()); + +output_csv (true); +} + +static void +output_header (uint32_t case_id,struct test_configure *case_cfg) { +snprintf (output_str[0 ],MAX_OUTPUT_STR_LEN , +CSV_HDR_FMT , case_id, case_cfg->test_type_str); + +output_csv (true); +} + +static void +run_test_case (struct test_configure *case_cfg) { +switch (case_cfg->test_type) { +case TEST_TYPE_DMA_MEM_COPY : +mem_copy_benchmark (case_cfg, true); +break ; +case TEST_TYPE_CPU_MEM_COPY : +mem_copy_benchmark (case_cfg, false); +break ; +default : +printf ("Unknown test type. %s\n" , case_cfg->test_type_str); +break ; + } +} + +static void +run_test(uint32_t case_id, struct test_configure *case_cfg) { + uint32_t i; + uint32_t nb_lcores = rte_lcore_count(); + struct test_configure_entry *mem_size = &case_cfg->mem_size; + struct test_configure_entry *buf_size = &case_cfg->buf_size; + struct test_configure_entry *ring_size = &case_cfg->ring_size; + struct test_configure_entry *kick_batch = &case_cfg->kick_batch; + struct test_configure_entry dummy = { 0 }; + struct test_configure_entry *var_entry = &dummy; + + for (i = 0; i < RTE_DIM(output_str); i++) + memset(output_str[i], 0, MAX_OUTPUT_STR_LEN); + + if (nb_lcores <= case_cfg->lcore_dma_map.cnt) { + printf("Case %u: Not enough lcores.\n", case_id); + return; + } + + printf("Number of used lcores: %u.\n", nb_lcores); + + if (mem_size->incr != 0) + var_entry = mem_size; + + if (buf_size->incr != 0) + var_entry = buf_size; + + if (ring_size->incr != 0) + var_entry = ring_size; + + if (kick_batch->incr != 0) + var_entry = kick_batch; + + case_cfg->scenario_id = 0; + + output_header(case_id, case_cfg); + + for (var_entry->cur = var_entry->first; var_entry->cur <= var_entry-Quoted 1 line
last;) {
app/test-dma-perf/main.c new file +21
-0+ case_cfg->scenario_id++; +printf ("\nRunning scenario %d\n" , case_cfg->scenario_id); + +run_test_case (case_cfg); +output_csv (false); + +if (var_entry->op ==OP_ADD ) + var_entry->cur += var_entry->incr; +else if (var_entry->op ==OP_MUL ) + var_entry->cur *= var_entry->incr; +else +break ; + } +} + +static int +parse_lcore (struct test_configure *test_case,const char *value) { +size_t len =strlen (value); +char *input = (char *)malloc ((len +1 ) *sizeof (char )); +strcpy (input, value); + struct lcore_dma_map_t *lcore_dma_map = &(test_case-Quoted 1 line
lcore_dma_map);app/test-dma-perf/main.c new file +49
-0+ +if (test_case ==NULL || value ==NULL ) +return -1 ; + +memset (lcore_dma_map,0 ,sizeof (struct lcore_dma_map_t )); + +char *token =strtok (input,", " ); +while (token !=NULL ) { +if (lcore_dma_map->cnt >=MAX_LCORE_NB ) { +free (input); +return -1 ; + } + +uint16_t lcore_id =atoi (token); + lcore_dma_map->lcores[lcore_dma_map->cnt++] = lcore_id; + + token =strtok (NULL ,", " ); + } + +free (input); +return 0 ; +} + +static int +parse_lcore_dma (struct test_configure *test_case,const char *value) { +struct lcore_dma_map_t *lcore_dma_map; +char *input =strndup (value,strlen (value) +1 ); +char *addrs = input; +char *ptrs[2 ]; +char *start, *end, *substr; +uint16_t lcore_id; +int ret =0 ; + +while (*addrs =='\0' ) + addrs++; +if (*addrs =='\0' ) { +fprintf (stderr,"No input DMA addresses\n" ); + ret =-1 ; + goto out; + } + + substr =strtok (addrs,"," ); +if (substr ==NULL ) { +fprintf (stderr,"No input DMA address\n" ); + ret =-1 ; + goto out; + } + + memset(&test_case->lcore_dma_map, 0, sizeof(struct
lcore_dma_map_t));app/test-dma-perf/main.c new file +15
-0+ + do { + rte_strsplit(substr, strlen(substr), ptrs, 2, '@'); + + start = strstr(ptrs[0], "lcore"); + if (start == NULL) { + fprintf(stderr, "Illegal lcore\n"); + ret = -1; + break; + } + + start += 5; + lcore_id = strtol(start, &end, 0); + if (end == start) { + fprintf(stderr, "No input lcore ID or ID %d iswrong\n",
lcore_id);app/test-dma-perf/main.c new file +7
-0+ ret =-1 ; +break ; + } + + lcore_dma_map = &test_case->lcore_dma_map; + lcore_dma_map->lcores[lcore_dma_map->cnt] = lcore_id; + strcpy(lcore_dma_map->dma_names[lcore_dma_map-Quoted 1 line
cnt], ptrs[1]);
app/test-dma-perf/main.c new file +23
-0+ lcore_dma_map->cnt++; + substr =strtok (NULL ,"," ); + }while (substr !=NULL ); + +out: +free (input); +return ret; +} + +static int +parse_entry (const char *value,struct test_configure_entry *entry) { +char input[255 ] = {0 }; +char *args[MAX_PARAMS_PER_ENTRY ]; +int args_nr =-1 ; + +if (value ==NULL || entry ==NULL ) + goto out; + +strncpy (input, value,254 ); +if (*input =='\0' ) + goto out; + + args_nr = rte_strsplit(input, strlen(input), args,
MAX_PARAMS_PER_ENTRY, ',');app/test-dma-perf/main.c new file +35
-0+if (args_nr !=1 && args_nr !=4 ) + goto out; + + entry->cur = entry->first = (uint32_t )atoi (args[0 ]); + +if (args_nr ==4 ) { + entry->last = (uint32_t )atoi (args[1 ]); + entry->incr = (uint32_t )atoi (args[2 ]); +if (!strcmp (args[3 ],"MUL" )) + entry->op =OP_MUL ; +else if (!strcmp (args[3 ],"ADD" )) + entry->op =OP_ADD ; +else { +printf ("Invalid op %s.\n" , args[3 ]); + args_nr =-1 ; + } + }else { + entry->op =OP_NONE ; + entry->last =0 ; + entry->incr =0 ; + } +out: +return args_nr; +} + +static uint16_t +load_configs (const char *path) +{ +struct rte_cfgfile *cfgfile; +int nb_sections, i; +struct test_configure *test_case; +char section_name[CFG_NAME_LEN ]; +const char *case_type; +const char *lcore_dma; + const char *mem_size_str, *buf_size_str, *ring_size_str,
*kick_batch_str;app/test-dma-perf/main.c new file +13
-0+int args_nr, nb_vp; +bool is_dma; + +printf ("config file parsing...\n" ); + cfgfile =rte_cfgfile_load (path,0 ); +if (!cfgfile) { +printf ("Open configure file error.\n" ); +exit (1 ); + } + + nb_sections =rte_cfgfile_num_sections (cfgfile,NULL ,0 ); +if (nb_sections >MAX_TEST_CASES ) { + printf("Error: The maximum number of cases is %d.\n",
MAX_TEST_CASES);app/test-dma-perf/main.c new file +7
-0+exit (1 ); + } + +for (i =0 ; i < nb_sections; i++) { +snprintf (section_name,CFG_NAME_LEN ,"case%d" , i +1 ); + test_case = &test_cases[i]; + case_type = rte_cfgfile_get_entry(cfgfile, section_name,"type");
app/test-dma-perf/main.c new file +2
-0+if (!case_type) { + printf("Error: No case type in case %d, the test will befinished here.\n",
app/test-dma-perf/main.c new file +7
-0+ i + 1); + test_case->is_valid = false; +continue ; + } + +if (strcmp (case_type,DMA_MEM_COPY ) ==0 ) { + test_case->test_type =
TEST_TYPE_DMA_MEM_COPY;app/test-dma-perf/main.c new file +4
-0+ test_case->test_type_str =DMA_MEM_COPY ; + is_dma = true; + }else if (strcmp (case_type,CPU_MEM_COPY ) ==0 ) { + test_case->test_type =
TEST_TYPE_CPU_MEM_COPY;app/test-dma-perf/main.c new file +4
-0+ test_case->test_type_str =CPU_MEM_COPY ; + is_dma = false; + }else { + printf("Error: Cannot find case type %s in case%d.\n",
case_type, i + 1);app/test-dma-perf/main.c new file +7
-0+ test_case->is_valid = false; +continue ; + } + + nb_vp =0 ; + + test_case->src_numa_node =(int)atoi(rte_cfgfile_get_entry(cfgfile,
+section_name, "src_numa_node"));app/test-dma-perf/main.c new file +1
-0+ test_case->dst_numa_node =(int)atoi(rte_cfgfile_get_entry(cfgfile,
+section_name, "dst_numa_node"));app/test-dma-perf/main.c new file +2
-0+ + mem_size_str = rte_cfgfile_get_entry(cfgfile, section_name,"
mem_size");app/test-dma-perf/main.c new file +1
-0+ args_nr = parse_entry(mem_size_str, &test_case-Quoted 1 line
mem_size);app/test-dma-perf/main.c new file +8
-0+if (args_nr <0 ) { +printf ("parse error in case %d.\n" , i +1 ); + test_case->is_valid = false; +continue ; + }else if (args_nr >1 ) + nb_vp++; + + buf_size_str = rte_cfgfile_get_entry(cfgfile, section_name,"
buf_size");app/test-dma-perf/main.c new file +10
-0+ args_nr =parse_entry (buf_size_str, &test_case->buf_size); +if (args_nr <0 ) { +printf ("parse error in case %d.\n" , i +1 ); + test_case->is_valid = false; +continue ; + }else if (args_nr >1 ) + nb_vp++; + +if (is_dma) { + ring_size_str = rte_cfgfile_get_entry(cfgfile,
section_name,
+"dma_ring_size");app/test-dma-perf/main.c new file +1
-0+ args_nr = parse_entry(ring_size_str, &test_case-Quoted 1 line
ring_size);app/test-dma-perf/main.c new file +8
-0+if (args_nr <0 ) { +printf ("parse error in case %d.\n" , i +1 ); + test_case->is_valid = false; +continue ; + }else if (args_nr >1 ) + nb_vp++; + + kick_batch_str = rte_cfgfile_get_entry(cfgfile,
section_name, "kick_batch");app/test-dma-perf/main.c new file +1
-0+ args_nr = parse_entry(kick_batch_str, &test_case-Quoted 1 line
kick_batch);app/test-dma-perf/main.c new file +8
-0+if (args_nr <0 ) { +printf ("parse error in case %d.\n" , i +1 ); + test_case->is_valid = false; +continue ; + }else if (args_nr >1 ) + nb_vp++; + + lcore_dma = rte_cfgfile_get_entry(cfgfile,
section_name, "lcore_dma");app/test-dma-perf/main.c new file +1
-0+ int lcore_ret = parse_lcore_dma(test_case,
lcore_dma);app/test-dma-perf/main.c new file +8
-0+if (lcore_ret <0 ) { +printf ("parse lcore dma error in case %d.\n" , i + 1); + test_case->is_valid = false; +continue ; + } + }else { + lcore_dma = rte_cfgfile_get_entry(cfgfile,
section_name, "lcore");app/test-dma-perf/main.c new file +10
-0+int lcore_ret =parse_lcore (test_case, lcore_dma); +if (lcore_ret <0 ) { +printf ("parse lcore error in case %d.\n" , i +1 ); + test_case->is_valid = false; +continue ; + } + } + +if (nb_vp >1 ) { + printf("Error, each section can only have a singlevariable parameter.\n");
app/test-dma-perf/main.c new file +6
-0+ test_case->is_valid = false; +continue ; + } + + test_case->cache_flush = + (int)atoi(rte_cfgfile_get_entry(cfgfile, section_name,"
cache_flush"));app/test-dma-perf/main.c new file +1
-0+ test_case->test_secs =(uint16_t)atoi(rte_cfgfile_get_entry(cfgfile,
app/test-dma-perf/main.c new file +3
-0+ section_name, "test_seconds")); + + test_case->eal_args = rte_cfgfile_get_entry(cfgfile,
section_name, "eal_args");app/test-dma-perf/main.c new file +19
-0+ test_case->is_valid = true; + } + +rte_cfgfile_close (cfgfile); +printf ("config file parsing complete.\n\n" ); +return i; +} + +/* Parse the argument given in the command line of the application */ +static int append_eal_args (int argc,char **argv,const char *eal_args, +char **new_argv) { +int i; +char *tokens[MAX_EAL_PARAM_NB ]; +char args[MAX_EAL_PARAM_LEN ] = {0 }; +int token_nb, new_argc =0 ; + +for (i =0 ; i < argc; i++) { + if ((strcmp(argv[i], CMDLINE_CONFIG_ARG) == 0) || + (strcmp(argv[i], CMDLINE_RESULT_ARG) ==0)) {
app/test-dma-perf/main.c new file +4
-0+ i++; +continue ; + } + strlcpy(new_argv[new_argc], argv[i],
sizeof(new_argv[new_argc]));app/test-dma-perf/main.c new file +37
-0+ new_argc++; + } + +if (eal_args) { +strlcpy (args, eal_args,sizeof (args)); + token_nb =rte_strsplit (args,strlen (args), + tokens,MAX_EAL_PARAM_NB ,' ' ); +for (i =0 ; i < token_nb; i++) +strcpy (new_argv[new_argc++], tokens[i]); + } + +return new_argc; +} + +int +main (int argc,char *argv[]) +{ +int ret; +uint16_t case_nb; +uint32_t i, nb_lcores; +pid_t cpid, wpid; +int wstatus; +char args[MAX_EAL_PARAM_NB ][MAX_EAL_PARAM_LEN ]; +char *pargs[MAX_EAL_PARAM_NB ]; +char *cfg_path_ptr =NULL ; +char *rst_path_ptr =NULL ; +char rst_path[PATH_MAX ]; +int new_argc; +bool is_first_case = true; + +memset (args,0 ,sizeof (args)); + +for (i =0 ; i <RTE_DIM (pargs); i++) + pargs[i] = args[i]; + +for (i =0 ; i < (uint32_t )argc; i++) { + if (strncmp(argv[i], CMDLINE_CONFIG_ARG,
MAX_LONG_OPT_SZ) == 0)app/test-dma-perf/main.c new file +2
-0+ cfg_path_ptr = argv[i +1 ]; + if (strncmp(argv[i], CMDLINE_RESULT_ARG,
MAX_LONG_OPT_SZ) == 0)app/test-dma-perf/main.c new file +21
-0+ rst_path_ptr = argv[i +1 ]; + } +if (cfg_path_ptr ==NULL ) { +printf ("Config file not assigned.\n" ); +return -1 ; + } +if (rst_path_ptr ==NULL ) { +strlcpy (rst_path, cfg_path_ptr,PATH_MAX ); +strcat (strtok (basename (rst_path),"." ),"_result.csv" ); + rst_path_ptr = rst_path; + } + + case_nb =load_configs (cfg_path_ptr); + fd =fopen (rst_path_ptr,"w" ); +if (fd ==NULL ) { +printf ("Open output CSV file error.\n" ); +return -1 ; + } +fclose (fd); + +for (i =0 ; i < case_nb; i++) {
app/test-dma-perf/main.c line 505
Quoted 10 lines
app/test-dma-perf/main.c new file +10
-0+if (test_cases[i].test_type ==TEST_TYPE_NONE ) { +printf ("No test type in test case %d.\n\n" , i +1 ); +continue ; + } +if (!test_cases[i].is_valid) { +printf ("Invalid test case %d.\n\n" , i +1 ); +continue ; + } + + cpid =fork ();
[Anoob] Do we really need fork()? Can't we use code like,
RTE_LCORE_FOREACH_WORKER(lcore_id) {
ret |= rte_eal_wait_lcore(lcore_id);
}to wait for all threads to exit?
Quoted 167 lines
app/test-dma-perf/main.c new file +7
-0+if (cpid <0 ) { +printf ("Fork case %d failed.\n" , i +1 ); +exit (EXIT_FAILURE ); + }else if (cpid ==0 ) { +printf ("\nRunning case %u\n\n" , i +1 ); + + new_argc = append_eal_args(argc, argv,test_cases[i].eal_args, pargs);
app/test-dma-perf/main.c new file +3
-0+ ret =rte_eal_init (new_argc, pargs); +if (ret <0 ) + rte_exit(EXIT_FAILURE, "Invalid EALarguments\n");
app/test-dma-perf/main.c new file +6
-0+ +/* Check lcores. */ + nb_lcores =rte_lcore_count (); +if (nb_lcores <2 ) + rte_exit(EXIT_FAILURE, + "There should be at least 2 workerlcores.\n");
app/test-dma-perf/main.c new file +36
-0+ + fd =fopen (rst_path_ptr,"a" ); +if (!fd) { +printf ("Open output CSV file error.\n" ); +return 0 ; + } + +if (is_first_case) { +output_env_info (); + is_first_case = false; + } +run_test (i +1 , &test_cases[i]); + +/* clean up the EAL */ +rte_eal_cleanup (); + +fclose (fd); + +printf ("\nCase %u completed.\n\n" , i +1 ); + +exit (EXIT_SUCCESS ); + }else { + wpid =waitpid (cpid, &wstatus,0 ); +if (wpid ==-1 ) { +printf ("waitpid error.\n" ); +exit (EXIT_FAILURE ); + } + +if (WIFEXITED (wstatus)) +printf ("Case process exited. status %d\n\n" , +WEXITSTATUS (wstatus)); +else if (WIFSIGNALED (wstatus)) +printf ("Case process killed by signal %d\n\n" , +WTERMSIG (wstatus)); +else if (WIFSTOPPED (wstatus)) + printf("Case process stopped by signal%d\n\n",
app/test-dma-perf/main.c new file +5
-0+ WSTOPSIG(wstatus)); + else if (WIFCONTINUED(wstatus)) + printf("Case process continued.\n\n"); + else + printf("Case process unknownterminated.\n\n");
app/test-dma-perf/main.c new file +7
-0+ } + } + +printf ("Bye...\n" ); +return 0 ; +} +app/test-dma-perf/main.h new file renamed from app/test-dma-perf/main.h +0
-0mode 100644 index 0000000000..215ac42673
app/test-dma-perf/main.h new file +69
-0@@ -0,0 +1,69 @@ +/* SPDX-License-Identifier: BSD-3-Clause +* Copyright(c) 2023 Intel Corporation +*/ + +#ifndef _MAIN_H_ +#define _MAIN_H_ + + +#include <rte_common.h> +#include <rte_cycles.h> +#include <rte_dev.h> +#include <rte_dmadev.h> + +#ifndef __maybe_unused +#define __maybe_unused __rte_unused +#endif + +#define MAX_WORKER_NB 128 +#define MAX_OUTPUT_STR_LEN 512 + +#define MAX_DMA_NB 128 +#define MAX_LCORE_NB 256 + +extern char output_str[MAX_WORKER_NB ][MAX_OUTPUT_STR_LEN ]; + +typedef enum { +OP_NONE =0 , +OP_ADD , +OP_MUL +}alg_op_type ; + +struct test_configure_entry { +uint32_t first; +uint32_t last; +uint32_t incr; +alg_op_type op; +uint32_t cur; +}; + +struct lcore_dma_map_t { +uint32_t lcores[MAX_WORKER_NB ]; +char dma_names[MAX_WORKER_NB ][RTE_DEV_NAME_MAX_LEN ]; +int16_t dma_ids[MAX_WORKER_NB ]; +uint16_t cnt; +}; + +struct test_configure { +bool is_valid; +uint8_t test_type; +const char *test_type_str; +uint16_t src_numa_node; +uint16_t dst_numa_node; +uint16_t opcode; +bool is_dma; +struct lcore_dma_map_t lcore_dma_map; +struct test_configure_entry mem_size; +struct test_configure_entry buf_size; +struct test_configure_entry ring_size; +struct test_configure_entry kick_batch; +uint32_t cache_flush; +uint32_t nr_buf; +uint16_t test_secs; +const char *eal_args; +uint8_t scenario_id; +}; + +void mem_copy_benchmark (struct test_configure *cfg,bool is_dma); + +#endif /* _MAIN_H_ */ app/test-dma- renamed from app/test-dma-perf/meson.build +0
-0perf/meson.build new file mode 100644 index 0000000000..bd6c264002
app/test-dma-perf/meson.build new file +17
-1@@ -0,0 +1,17 @@ +# SPDX-License-Identifier: BSD-3-Clause # Copyright(c) 2019-2023 Intel +Corporation + +# meson file, for building this app as part of a main DPDK build. + +if is_windows + build = false + reason = 'not supported on Windows' + subdir_done() +endif + +deps += ['dmadev', 'mbuf', 'cfgfile'] + +sources = files( + 'main.c', + 'benchmark.c', +) --2.40.1
Hi,
Thanks for your comments, the replies are inline.
Thanks,
Cheng
Quoted 285 lines
-----Original Message-----
From: Anoob Joseph <[email protected]>
Sent: Thursday, June 15, 2023 1:22 PM
To: Jiang, Cheng1 <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>; Amit Prakash Shukla <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; Ma, WenwuX <[email protected]>; Wang, YuanX <[email protected]>; He, Xingguang <[email protected]>
Subject: RE: [EXT] [PATCH v6] app/dma-perf: introduce dma-perf applicationHi,
Thanks for working on the comments. Few more top level comment inline.
Thanks,
AnoobQuoted 13 lines
-----Original Message-----
From: Cheng Jiang <[email protected]>
Sent: Tuesday, June 13, 2023 10:02 AM
To: [email protected]; [email protected]; [email protected]; [email protected]; Amit Prakash Shukla <[email protected]>; Anoob Joseph <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; Cheng Jiang <[email protected]>
Subject: [EXT] [PATCH v6] app/dma-perf: introduce dma-perf applicationExternal Email
Quoted 11 lines
----------------------------------------------------------------------
There are many high-performance DMA devices supported in DPDK now, and these DMA devices can also be integrated into other modules of DPDK as accelerators, such as Vhost. Before integrating DMA into applications, developers need to know the performance of these DMA devices in various scenarios and the performance of CPUs in the same scenario, such as different buffer lengths. Only in this way can we know the target performance of the application accelerated by using them. This patch introduces a high-performance testing tool, which supports comparing the performance of CPU and DMA in different scenarios automatically with a pre- set config file. Memory Copy performance test aresupported for now.
Quoted 241 lines
Signed-off-by Cheng Jiang <[email protected]>Signed-off-by Jiayu Hu <[email protected]>Signed-off-by Yuan Wang <[email protected]>Acked-by Morten Brørup <[email protected]>Acked-by Chenbo Xia <[email protected]>---
v6:
improved code based on Anoob's comments; fixed some code structure issues;
v5:
fixed someLONG_LINEwarnings;
v4:
fixed inaccuracy of the memory footprint display;
v3:
fixed some typos;
v2:
added lcore/dmadev designation;
added error case process;
removedworker_threadsparameter from config.ini; improved the logs;
improved config file;app/meson.build | 1 + app/test-dma-perf/benchmark.c | 477 ++++++++++++++++++++++++++++ app/test-dma-perf/config.ini | 59 ++++ app/test-dma-perf/main.c | 569 ++++++++++++++++++++++++++++++++++ app/test-dma-perf/main.h | 69 +++++ app/test-dma-perf/meson.build | 17 + 6 files changed, 1192 insertions(+) create mode 100644 app/test-dma-perf/benchmark.c create mode 100644app/test-dma-perf/config.ini create mode 100644 app/test-dma-
perf/main.ccreate mode 100644app/test-dma-perf/main.hcreate mode 100644 app/test-dma-perf/meson.buildapp/meson.build index renamed from app/meson.build +0
-0
74d2420f67..4fc1a83eba100644app/meson.build +1
-0@@ -19,6 +19,7 @@ apps = [ 'test-cmdline', 'test-compress-perf', 'test-crypto-perf', + 'test-dma-perf', 'test-eventdev', 'test-fib', 'test-flow-perf',app/test-dma- renamed from app/test-dma-perf/benchmark.c +0
-0
perf/benchmark.cnew file mode 100644 index 0000000000..bc1ca82297app/test-dma-perf/benchmark.c new file +21
-0@@ -0,0 +1,477 @@ +/* SPDX-License-Identifier: BSD-3-Clause +* Copyright(c) 2023 Intel Corporation */ + +#include <inttypes.h> +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> + +#include <rte_time.h> +#include <rte_mbuf.h> +#include <rte_dmadev.h> +#include <rte_malloc.h> +#include <rte_lcore.h> + +#include "main.h" + +#define MAX_DMA_CPL_NB 255 + +#define TEST_WAIT_U_SECOND 10000 + +#define CSV_LINE_DMA_FMT "Scenario %u,%u,%s,%u,%u,%.2lf,%" PRIu64",%.3lf,%.3lf\n"
+#defineCSV_LINE_CPU_FMT"Scenario %u,%u,NA,%u,%u,%.2lf,%" PRIu64 ",%.3lf,%.3lf\n"app/test-dma-perf/benchmark.c new file +47
-0+ +struct worker_info { +bool ready_flag; +bool start_flag; +bool stop_flag; +uint32_t total_cpl; +uint32_t test_cpl; +}; + +struct lcore_params { +uint8_t scenario_id; +unsigned int lcore_id; +char *dma_name; +uint16_t worker_id; +uint16_t dev_id; +uint32_t nr_buf; +uint16_t kick_batch; +uint32_t buf_size; +uint16_t test_secs; +struct rte_mbuf **srcs; +struct rte_mbuf **dsts; +struct worker_info worker_info; +}; + +static struct rte_mempool *src_pool; +static struct rte_mempool *dst_pool; + +static volatile struct lcore_params *worker_params[MAX_WORKER_NB ]; + +#define PRINT_ERR (...) print_err(__func__, __LINE__, __VA_ARGS__) + +static inline int +__rte_format_printf(3, 4) +print_err(const char *func, int lineno, const char *format, ...) { + va_list ap; + int ret; + + ret = fprintf(stderr, "In %s:%d - ", func, lineno); + va_start(ap, format); + ret += vfprintf(stderr, format, ap); + va_end(ap); + + return ret; +} + +static inline void +calc_result(uint32_t buf_size, uint32_t nr_buf, uint16_t nb_workers,
uint16_ttest_secs,app/test-dma-perf/benchmark.c new file +1
-0+ uint32_t total_cnt, float *memory, uint32_t
*ave_cycle,app/test-dma-perf/benchmark.c new file +3
-0+ float *bandwidth, float *mops) +{ + *memory = (float)(buf_size * (nr_buf / nb_workers) * 2) / (1024 *1024);
app/test-dma-perf/benchmark.c new file +2
-0+ *ave_cycle = test_secs *rte_get_timer_hz () / total_cnt; + *bandwidth = (buf_size * 8 * (rte_get_timer_hz() /(float)*ave_cycle)) / 1000000000;
app/test-dma-perf/benchmark.c new file +4
-0+ *mops = (float )rte_get_timer_hz () / *ave_cycle /1000000 ; } + +static void +output_result(uint8_t scenario_id, uint32_t lcore_id, char *dma_name,
uint64_tave_cycle,app/test-dma-perf/benchmark.c new file +8
-0+ uint32_t buf_size, uint32_t nr_buf, float memory, + float bandwidth, float mops, bool is_dma) { + if (is_dma) + printf("lcore %u, DMA %s:\n", lcore_id, dma_name); + else + printf("lcore %u\n", lcore_id); + + printf("average cycles/op: %" PRIu64 ", buffer size: %u, nr_buf: %u,memory: %.2lfMB, frequency: %" PRIu64 ".\n",
app/test-dma-perf/benchmark.c new file +1
-0+ ave_cycle, buf_size, nr_buf, memory,
rte_get_timer_hz());app/test-dma-perf/benchmark.c new file +5
-0+printf ("Average bandwidth: %.3lfGbps, MOps: %.3lf\n" , bandwidth, +mops); + +if (is_dma) + snprintf(output_str[lcore_id], MAX_OUTPUT_STR_LEN,
CSV_LINE_DMA_FMT,app/test-dma-perf/benchmark.c new file +4
-0+ scenario_id, lcore_id, dma_name, buf_size, + nr_buf, memory, ave_cycle, bandwidth, mops); + else + snprintf(output_str[lcore_id], MAX_OUTPUT_STR_LEN,
CSV_LINE_CPU_FMT,app/test-dma-perf/benchmark.c new file +41
-0+ scenario_id, lcore_id, buf_size, + nr_buf, memory, ave_cycle, bandwidth, mops); } + +static inline void +cache_flush_buf (__maybe_unused struct rte_mbuf **array, +__maybe_unused uint32_t buf_size, +__maybe_unused uint32_t nr_buf) +{ +#ifdef RTE_ARCH_X86_64 +char *data; +struct rte_mbuf **srcs = array; +uint32_t i, offset; + +for (i =0 ; i < nr_buf; i++) { + data = rte_pktmbuf_mtod(srcs[i], char *); +for (offset =0 ; offset < buf_size; offset +=64 ) +__builtin_ia32_clflush (data + offset); + } +#endif +} + +/* Configuration of device. */ +static void +configure_dmadev_queue (uint32_t dev_id,uint32_t ring_size) { +uint16_t vchan =0 ; +struct rte_dma_info info; +struct rte_dma_conf dev_config = { .nb_vchans =1 }; +struct rte_dma_vchan_conf qconf = { + .direction =RTE_DMA_DIR_MEM_TO_MEM , + .nb_desc = ring_size + }; + +if (rte_dma_configure (dev_id, &dev_config) !=0 ) +rte_exit (EXIT_FAILURE ,"Error with dma configure.\n" ); + +if (rte_dma_vchan_setup (dev_id, vchan, &qconf) !=0 ) +rte_exit (EXIT_FAILURE ,"Error with queue configuration.\n" ); + +rte_dma_info_get (dev_id, &info); +if (info.nb_vchans !=1 ) + rte_exit(EXIT_FAILURE, "Error, no configured queuesreported on device id. %u\n",
app/test-dma-perf/benchmark.c new file +20
-0+ dev_id); + +if (rte_dma_start (dev_id) !=0 ) +rte_exit (EXIT_FAILURE ,"Error with dma start.\n" ); } + +static int +config_dmadevs (struct test_configure *cfg) { +uint32_t ring_size = cfg->ring_size.cur; +struct lcore_dma_map_t *ldm = &cfg->lcore_dma_map; +uint32_t nb_workers = ldm->cnt; +uint32_t i; +int dev_id; +uint16_t nb_dmadevs =0 ; +char *dma_name; + +for (i =0 ; i < ldm->cnt; i++) { + dma_name = ldm->dma_names[i]; + dev_id =rte_dma_get_dev_id_by_name (dma_name); +if (dev_id ==-1 ) { + fprintf(stderr, "Error: Fail to find DMA %s.\n",
dma_name);app/test-dma-perf/benchmark.c new file +11
-0+ goto end; + } + + ldm->dma_ids[i] = dev_id; +configure_dmadev_queue (dev_id, ring_size); + ++nb_dmadevs; + } + +end: +if (nb_dmadevs < nb_workers) { + printf("Not enough dmadevs (%u) for all workers (%u).\n",
nb_dmadevs,nb_workers);app/test-dma-perf/benchmark.c new file +5
-0+return -1 ; + } + +printf ("Number of used dmadevs: %u.\n" , nb_dmadevs); +
Quoted 8 lines
app/test-dma-perf/benchmark.c line 187
Quoted 6 lines
app/test-dma-perf/benchmark.c new file +6
-0+return 0 ; +} + +#define POLL_MAX 1000 + +[Anoob] Extra blank line. You can consider removing.
[Cheng] sure, sorry for the miss.
Quoted 143 lines
Quoted 143 lines
app/test-dma-perf/benchmark.c new file +14
-0+static inline void +do_dma_submit_and_poll (uint16_t dev_id,uint64_t *async_cnt, +volatile struct worker_info *worker_info) { +int ret; +uint16_t nr_cpl; + + ret =rte_dma_submit (dev_id,0 ); +if (ret <0 ) { +rte_dma_stop (dev_id); +rte_dma_close (dev_id); +rte_exit (EXIT_FAILURE ,"Error with dma submit.\n" ); + } + + nr_cpl = rte_dma_completed(dev_id, 0, MAX_DMA_CPL_NB, NULL,NULL);
app/test-dma-perf/benchmark.c new file +32
-0+ *async_cnt -= nr_cpl; + worker_info->total_cpl += nr_cpl; +} + +static inline int +do_dma_mem_copy (void *p) +{ +const uint16_t *para_idx = (uint16_t *)p; +volatile struct lcore_params *para = worker_params[*para_idx]; +volatile struct worker_info *worker_info = &(para->worker_info); +const uint16_t dev_id = para->dev_id; +const uint32_t nr_buf = para->nr_buf; +const uint16_t kick_batch = para->kick_batch; +const uint32_t buf_size = para->buf_size; +struct rte_mbuf **srcs = para->srcs; +struct rte_mbuf **dsts = para->dsts; +uint16_t nr_cpl; +uint64_t async_cnt =0 ; +uint32_t i; +uint32_t poll_cnt =0 ; +int ret; + + worker_info->stop_flag = false; + worker_info->ready_flag = true; + +while (!worker_info->start_flag) + ; + +while (1 ) { +for (i =0 ; i < nr_buf; i++) { +dma_copy: + ret = rte_dma_copy(dev_id, 0,
rte_pktmbuf_iova(srcs[i]),app/test-dma-perf/benchmark.c new file +4
-0+ rte_pktmbuf_iova(dsts[i]), buf_size, 0); +if (unlikely (ret <0 )) { +if (ret == -ENOSPC ) { + do_dma_submit_and_poll(dev_id,
&async_cnt,worker_info);app/test-dma-perf/benchmark.c new file +5
-0+ goto dma_copy; + }else { +/* Error exit */ +rte_dma_stop (dev_id); + rte_exit(EXIT_FAILURE, "DMAenqueue failed\n");
app/test-dma-perf/benchmark.c new file +6
-0+ } + } + async_cnt++; + +if ((async_cnt % kick_batch) ==0 ) + do_dma_submit_and_poll(dev_id,
&async_cnt,worker_info);app/test-dma-perf/benchmark.c new file +9
-0+ } + +if (worker_info->stop_flag) +break ; + } + +rte_dma_submit (dev_id,0 ); +while ((async_cnt >0 ) && (poll_cnt++ <POLL_MAX )) { + nr_cpl = rte_dma_completed(dev_id, 0,
MAX_DMA_CPL_NB, NULL, NULL);app/test-dma-perf/benchmark.c new file +28
-0+ async_cnt -= nr_cpl; + } + +return 0 ; +} + +static inline int +do_cpu_mem_copy (void *p) +{ +const uint16_t *para_idx = (uint16_t *)p; +volatile struct lcore_params *para = worker_params[*para_idx]; +volatile struct worker_info *worker_info = &(para->worker_info); +const uint32_t nr_buf = para->nr_buf; +const uint32_t buf_size = para->buf_size; +struct rte_mbuf **srcs = para->srcs; +struct rte_mbuf **dsts = para->dsts; +uint32_t i; + + worker_info->stop_flag = false; + worker_info->ready_flag = true; + +while (!worker_info->start_flag) + ; + +while (1 ) { +for (i =0 ; i < nr_buf; i++) { +/* copy buffer form src to dst */ + rte_memcpy((void*)(uintptr_t)rte_mbuf_data_iova(dsts[i]),
app/test-dma-perf/benchmark.c new file +1
-0+ (void*)(uintptr_t)rte_mbuf_data_iova(srcs[i]),
app/test-dma-perf/benchmark.c new file +22
-0+ (size_t)buf_size); + worker_info->total_cpl++; + } +if (worker_info->stop_flag) +break ; + } + +return 0 ; +} + +static int +setup_memory_env (struct test_configure *cfg,struct rte_mbuf ***srcs, +struct rte_mbuf ***dsts) +{ +unsigned int buf_size = cfg->buf_size.cur; +unsigned int nr_sockets; +uint32_t nr_buf = cfg->nr_buf; + + nr_sockets =rte_socket_count (); +if (cfg->src_numa_node >= nr_sockets || + cfg->dst_numa_node >= nr_sockets) { + printf("Error: Source or destination numa exceeds the acturenuma nodes.\n");
app/test-dma-perf/benchmark.c new file +12
-0+return -1 ; + } + + src_pool =rte_pktmbuf_pool_create ("Benchmark_DMA_SRC" , + nr_buf,/* n == num elements */ +64 ,/* cache size */ +0 ,/* priv size */ + buf_size +RTE_PKTMBUF_HEADROOM , + cfg->src_numa_node); +if (src_pool ==NULL ) { +PRINT_ERR ("Error with source mempool creation.\n" ); +return -1 ;
Quoted 8 lines
app/test-dma-perf/benchmark.c line 326
Quoted 5 lines
app/test-dma-perf/benchmark.c new file +5
-0+ } + + dst_pool = rte_pktmbuf_pool_create("Benchmark_DMA_DST", + nr_buf, /* n == num elements */ + 64, /* cache size */[Anoob] We do not alloc or free pointers in the datapath, right? So why bother with cache?
[Cheng] Yes, you are right, the cache size is not necessary here, I'll fix it in the next version.
Quoted 16 lines
Quoted 4 lines
app/test-dma-perf/benchmark.c new file +3
-0+0 ,/* priv size */ + buf_size +RTE_PKTMBUF_HEADROOM , + cfg->dst_numa_node);app/test-dma-perf/benchmark.c line 339
Quoted 10 lines
app/test-dma-perf/benchmark.c new file +10
-0+if (dst_pool ==NULL ) { +PRINT_ERR ("Error with destination mempool creation.\n" ); +return -1 ; + } + + *srcs =rte_malloc (NULL , nr_buf *sizeof (struct rte_mbuf *),0 ); +if (*srcs ==NULL ) { +printf ("Error: srcs malloc failed.\n" ); +return -1 ; + }[Anoob] Are we freeing these memory? The ones allocated with
rte_malloc.
[Cheng] yes, we freed the memory in the end of mem_copy_benchmark() when we finished the test.
Quoted 58 lines
Quoted 58 lines
app/test-dma-perf/benchmark.c new file +29
-0+ + *dsts =rte_malloc (NULL , nr_buf *sizeof (struct rte_mbuf *),0 ); +if (*dsts ==NULL ) { +printf ("Error: dsts malloc failed.\n" ); +return -1 ; + } + +if (rte_mempool_get_bulk (src_pool, (void **)*srcs, nr_buf) !=0 ) { +printf ("get src mbufs failed.\n" ); +return -1 ; + } +if (rte_mempool_get_bulk (dst_pool, (void **)*dsts, nr_buf) !=0 ) { +printf ("get dst mbufs failed.\n" ); +return -1 ; + } + +return 0 ; +} + +void +mem_copy_benchmark (struct test_configure *cfg,bool is_dma) { +uint16_t i; +uint32_t offset; +unsigned int lcore_id =0 ; +struct rte_mbuf **srcs =NULL , **dsts =NULL ; +struct lcore_dma_map_t *ldm = &cfg->lcore_dma_map; +unsigned int buf_size = cfg->buf_size.cur; +uint16_t kick_batch = cfg->kick_batch.cur; + uint32_t nr_buf = cfg->nr_buf = (cfg->mem_size.cur * 1024 * 1024) /(
cfg->buf_size.cur* 2);app/test-dma-perf/benchmark.c new file +27
-0+uint16_t nb_workers = ldm->cnt; +uint16_t test_secs = cfg->test_secs; +float memory; +uint32_t avg_cycles =0 ; +float mops; +float bandwidth; + +if (setup_memory_env (cfg, &srcs, &dsts) <0 ) + goto out; + +if (is_dma) +if (config_dmadevs (cfg) <0 ) + goto out; + +if (cfg->cache_flush) { +cache_flush_buf (srcs, buf_size, nr_buf); +cache_flush_buf (dsts, buf_size, nr_buf); +rte_mb (); + } + +printf ("Start testing....\n" ); + +for (i =0 ; i < nb_workers; i++) { + lcore_id = ldm->lcores[i]; + offset = nr_buf / nb_workers * i; + + worker_params[i] = rte_malloc(NULL, sizeof(struct
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 6 lines
lcore_params), 0);app/test-dma-perf/benchmark.c new file +2
-0+if (!worker_params[i]) { + printf("lcore parameters malloc failure for lcore%d\n",
lcore_id);app/test-dma-perf/benchmark.c new file +2
-0+break ; + }[Anoob] Are we freeing the above memory?
[Cheng] sorry, I missed that, I'll add worker_params memory free in the next version, thanks.
Quoted 160 lines
Quoted 146 lines
app/test-dma-perf/benchmark.c new file +2
-0+if (is_dma) { + worker_params[i]->dma_name = ldm-Quoted 1 line
dma_names[i];
app/test-dma-perf/benchmark.c new file +5
-0+ worker_params[i]->dev_id = ldm->dma_ids[i]; + worker_params[i]->kick_batch = kick_batch; + } + worker_params[i]->worker_id = i; + worker_params[i]->nr_buf = (uint32_t)(nr_buf /
nb_workers);app/test-dma-perf/benchmark.c new file +9
-0+ worker_params[i]->buf_size = buf_size; + worker_params[i]->test_secs = test_secs; + worker_params[i]->srcs = srcs + offset; + worker_params[i]->dsts = dsts + offset; + worker_params[i]->scenario_id = cfg->scenario_id; + worker_params[i]->lcore_id = lcore_id; + +if (is_dma) + rte_eal_remote_launch(do_dma_mem_copy, (void*)(&i),
lcore_id);app/test-dma-perf/benchmark.c new file +2
-0+else + rte_eal_remote_launch(do_cpu_mem_copy, (void*)(&i),
lcore_id);app/test-dma-perf/benchmark.c new file +6
-0+ } + +while (1 ) { +bool ready = true; +for (i =0 ; i < nb_workers; i++) { + if (worker_params[i]->worker_info.ready_flag ==false) {
app/test-dma-perf/benchmark.c new file +19
-0+ ready =0 ; +break ; + } + } +if (ready) +break ; + } + +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.start_flag = true; + +usleep (TEST_WAIT_U_SECOND ); +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.test_cpl = +worker_params[i]->worker_info.total_cpl; + +usleep (test_secs *1000 *1000 ); +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.test_cpl =worker_params[i]->worker_info.total_cpl -
app/test-dma-perf/benchmark.c new file +1
-0+ worker_params[i]-Quoted 1 line
worker_info.test_cpl;
app/test-dma-perf/benchmark.c new file +12
-0+ +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.stop_flag = true; + +rte_eal_mp_wait_lcore (); + +for (i =0 ; i < nb_workers; i++) { +calc_result (buf_size, nr_buf, nb_workers, test_secs, + worker_params[i]->worker_info.test_cpl, + &memory, &avg_cycles, &bandwidth, &mops); + output_result(cfg->scenario_id, worker_params[i]->lcore_id, + worker_params[i]->dma_name,
avg_cycles,buf_size,app/test-dma-perf/benchmark.c new file +1
-0+ nr_buf / nb_workers, memory,bandwidth, mops,
is_dma);app/test-dma-perf/benchmark.c new file +21
-0+ } + +out: +/* free env */ +if (srcs) +rte_pktmbuf_free_bulk (srcs, nr_buf); +if (dsts) +rte_pktmbuf_free_bulk (dsts, nr_buf); + +if (src_pool) +rte_mempool_free (src_pool); +if (dst_pool) +rte_mempool_free (dst_pool); + +if (is_dma) { +for (i =0 ; i < nb_workers; i++) { +printf ("Stopping dmadev %d\n" , ldm->dma_ids[i]); +rte_dma_stop (ldm->dma_ids[i]); + } + } +}(quoted diff) +0
-0b/app/test-dma-perf/config.ini new file mode 100644 index 0000000000..2fd9c3c387
app/test-dma-perf/config.ini new file +11
-0@@ -0,0 +1,59 @@ + +; This is an example configuration file for dma-perf, which details +the meanings of each parameter ; and instructions on how to use dma-perf. + +; Supported test types are DMA_MEM_COPY and CPU_MEM_COPY. + +; Parameters: +; "mem_size" denotes the size of the memory footprint. +; "buf_size" denotes the memory size of a single operation. +; "dma_ring_size" denotes the dma ring buffer size. It should be +greaterthan 64 normally.
app/test-dma-perf/config.ini new file +2
-0+; "kick_batch" denotes the dma operation batch size, and should be +greaterthan 1 normally.
app/test-dma-perf/config.ini new file +4
-0+ +; The format for variables is variable=first,last,increment,ADD|MUL. + +; src_numa_node is used to control the numa node where the sourcememory is allocated.
app/test-dma-perf/config.ini new file +2
-0+; dst_numa_node is used to control the numa node where the +destinationmemory is allocated.
app/test-dma-perf/config.ini new file +10
-0+ +; cache_flush is used to determine whether or not the cache should be +flushed, with 1 indicating to ; flush and 0 indicating to not flush. + +; test_seconds controls the test time of the whole case. + +; To use DMA for a test, please specify the "lcore_dma" parameter. +; If you have already set the "-l" and "-a" parameters using EAL, ; +make sure that the value of "lcore_dma" falls within their range of +thevalues.
app/test-dma-perf/config.ini new file +7
-0+ +; To use CPU for a test, please specify the "lcore" parameter. +; If you have already set the "-l" and "-a" parameters using EAL, ; +make sure that the value of "lcore" falls within their range of values. + +; To specify a configuration file, use the "--config" flag followed +by the pathto the file.
app/test-dma-perf/config.ini new file +3
-0+ +; To specify a result file, use the "--result" flag followed by the +path to thefile.
app/test-dma-perf/config.ini new file +3
-0+; If you do not specify a result file, one will be generated with the +same name as the configuration ; file, with the addition of +"_result.csv" atthe end.
app/test-dma-perf/config.ini new file +2
-0+ +[case1]app/test-dma-perf/config.ini line 47
Quoted 10 lines
app/test-dma-perf/config.ini new file +10
-0+type=DMA_MEM_COPY +mem_size=10 +buf_size=64,8192,2,MUL +dma_ring_size=1024 +kick_batch=32 +src_numa_node=0 +dst_numa_node=0 +cache_flush=0 +test_seconds=2 +lcore_dma=lcore10@0000:00:04.2, lcore11@0000:00:04.3[Anoob] Isn't it better if we allow user to specify DMA dev ID rather than the PCI DBDF?
Quoted 7 lines
In the long run, I would expect config file to provide {core,
dma_dev_id,queue_id}Another thought is why to expose this at all? If we can restrict this perf application to have one thread only use one vchan, then application can easily create this mapping in run time. Unless you want one thread to use 2 different vchans which may not be desirable since this is a standalone perf app.
[Cheng] Thank you for the feedback.
Here are my thoughts:
Firstly, the user may not know which device the DMA dev ID corresponds to, or which NUMA node it is on. In my example, I used the CBDMA environment, so I did not specify the work queue ID. When using DSA, the configuration would be something like lcore10@0000:00:04.2-q0 which contains core, dma and work queue id. The reason for exposing these options is that we want the user to fully understand which cores and devices are being used so that they know exactly where the performance data is coming from. For example, performance when cores and DMA devices are not on the same NUMA node, etc. This allows the testing scenario to be precise and flexible. If the application handles the mapping itself, the user loses control over the mapping and may not get the performance data they want. We believe control should be given to the user rather than the application.
Quoted 557 lines
Quoted 424 lines
app/test-dma-perf/config.ini new file +12
-0+eal_args=--in-memory --file-prefix=test + +[case2] +type=CPU_MEM_COPY +mem_size=10 +buf_size=64,8192,2,MUL +src_numa_node=0 +dst_numa_node=1 +cache_flush=0 +test_seconds=2 +lcore = 3, 4 +eal_args=--in-memory --no-pciapp/test-dma-perf/main.c new renamed from app/test-dma-perf/main.c +0
-0file mode 100644 index 0000000000..d65655b87b
app/test-dma-perf/main.c new file +21
-0@@ -0,0 +1,569 @@ +/* SPDX-License-Identifier: BSD-3-Clause +* Copyright(c) 2023 Intel Corporation */ + +#include <stdio.h> +#include <stdlib.h> +#include <getopt.h> +#include <signal.h> +#include <stdbool.h> +#include <unistd.h> +#include <sys/wait.h> +#include <inttypes.h> +#include <libgen.h> + +#include <rte_eal.h> +#include <rte_cfgfile.h> +#include <rte_string_fns.h> +#include <rte_lcore.h> + +#include "main.h" + +#define CSV_HDR_FMT "Case %u : %s,lcore,DMA,buffersize,nr_buf,memory(MB),cycle,bandwidth(Gbps),MOps\n"
app/test-dma-perf/main.c new file +51
-0+ +#define MAX_EAL_PARAM_NB 100 +#define MAX_EAL_PARAM_LEN 1024 + +#define DMA_MEM_COPY "DMA_MEM_COPY" +#define CPU_MEM_COPY "CPU_MEM_COPY" + +#define CMDLINE_CONFIG_ARG "--config" +#define CMDLINE_RESULT_ARG "--result" + +#define MAX_PARAMS_PER_ENTRY 4 + +#define MAX_LONG_OPT_SZ 64 + +enum { +TEST_TYPE_NONE =0 , +TEST_TYPE_DMA_MEM_COPY , +TEST_TYPE_CPU_MEM_COPY +}; + +#define MAX_TEST_CASES 16 +static struct test_configure test_cases[MAX_TEST_CASES ]; + +char output_str[MAX_WORKER_NB ][MAX_OUTPUT_STR_LEN ]; + +static FILE *fd; + +static void +output_csv (bool need_blankline) +{ +uint32_t i; + +if (need_blankline) { +fprintf (fd,",,,,,,,,\n" ); +fprintf (fd,",,,,,,,,\n" ); + } + +for (i =0 ; i <RTE_DIM (output_str); i++) { +if (output_str[i][0 ]) { +fprintf (fd,"%s" , output_str[i]); + output_str[i][0 ] ='\0' ; + } + } + +fflush (fd); +} + +static void +output_env_info (void ) +{ + snprintf(output_str[0], MAX_OUTPUT_STR_LEN, "testenvironment:\n");
app/test-dma-perf/main.c new file +67
-0+snprintf (output_str[1 ],MAX_OUTPUT_STR_LEN ,"CPU frequency,%" + PRIu64"\n" ,rte_get_timer_hz ()); + +output_csv (true); +} + +static void +output_header (uint32_t case_id,struct test_configure *case_cfg) { +snprintf (output_str[0 ],MAX_OUTPUT_STR_LEN , +CSV_HDR_FMT , case_id, case_cfg->test_type_str); + +output_csv (true); +} + +static void +run_test_case (struct test_configure *case_cfg) { +switch (case_cfg->test_type) { +case TEST_TYPE_DMA_MEM_COPY : +mem_copy_benchmark (case_cfg, true); +break ; +case TEST_TYPE_CPU_MEM_COPY : +mem_copy_benchmark (case_cfg, false); +break ; +default : +printf ("Unknown test type. %s\n" , case_cfg->test_type_str); +break ; + } +} + +static void +run_test(uint32_t case_id, struct test_configure *case_cfg) { + uint32_t i; + uint32_t nb_lcores = rte_lcore_count(); + struct test_configure_entry *mem_size = &case_cfg->mem_size; + struct test_configure_entry *buf_size = &case_cfg->buf_size; + struct test_configure_entry *ring_size = &case_cfg->ring_size; + struct test_configure_entry *kick_batch = &case_cfg->kick_batch; + struct test_configure_entry dummy = { 0 }; + struct test_configure_entry *var_entry = &dummy; + + for (i = 0; i < RTE_DIM(output_str); i++) + memset(output_str[i], 0, MAX_OUTPUT_STR_LEN); + + if (nb_lcores <= case_cfg->lcore_dma_map.cnt) { + printf("Case %u: Not enough lcores.\n", case_id); + return; + } + + printf("Number of used lcores: %u.\n", nb_lcores); + + if (mem_size->incr != 0) + var_entry = mem_size; + + if (buf_size->incr != 0) + var_entry = buf_size; + + if (ring_size->incr != 0) + var_entry = ring_size; + + if (kick_batch->incr != 0) + var_entry = kick_batch; + + case_cfg->scenario_id = 0; + + output_header(case_id, case_cfg); + + for (var_entry->cur = var_entry->first; var_entry->cur <= var_entry-Quoted 1 line
last;) {
app/test-dma-perf/main.c new file +21
-0+ case_cfg->scenario_id++; +printf ("\nRunning scenario %d\n" , case_cfg->scenario_id); + +run_test_case (case_cfg); +output_csv (false); + +if (var_entry->op ==OP_ADD ) + var_entry->cur += var_entry->incr; +else if (var_entry->op ==OP_MUL ) + var_entry->cur *= var_entry->incr; +else +break ; + } +} + +static int +parse_lcore (struct test_configure *test_case,const char *value) { +size_t len =strlen (value); +char *input = (char *)malloc ((len +1 ) *sizeof (char )); +strcpy (input, value); + struct lcore_dma_map_t *lcore_dma_map = &(test_case-Quoted 1 line
lcore_dma_map);app/test-dma-perf/main.c new file +49
-0+ +if (test_case ==NULL || value ==NULL ) +return -1 ; + +memset (lcore_dma_map,0 ,sizeof (struct lcore_dma_map_t )); + +char *token =strtok (input,", " ); +while (token !=NULL ) { +if (lcore_dma_map->cnt >=MAX_LCORE_NB ) { +free (input); +return -1 ; + } + +uint16_t lcore_id =atoi (token); + lcore_dma_map->lcores[lcore_dma_map->cnt++] = lcore_id; + + token =strtok (NULL ,", " ); + } + +free (input); +return 0 ; +} + +static int +parse_lcore_dma (struct test_configure *test_case,const char *value) { +struct lcore_dma_map_t *lcore_dma_map; +char *input =strndup (value,strlen (value) +1 ); +char *addrs = input; +char *ptrs[2 ]; +char *start, *end, *substr; +uint16_t lcore_id; +int ret =0 ; + +while (*addrs =='\0' ) + addrs++; +if (*addrs =='\0' ) { +fprintf (stderr,"No input DMA addresses\n" ); + ret =-1 ; + goto out; + } + + substr =strtok (addrs,"," ); +if (substr ==NULL ) { +fprintf (stderr,"No input DMA address\n" ); + ret =-1 ; + goto out; + } + + memset(&test_case->lcore_dma_map, 0, sizeof(struct
lcore_dma_map_t));app/test-dma-perf/main.c new file +15
-0+ + do { + rte_strsplit(substr, strlen(substr), ptrs, 2, '@'); + + start = strstr(ptrs[0], "lcore"); + if (start == NULL) { + fprintf(stderr, "Illegal lcore\n"); + ret = -1; + break; + } + + start += 5; + lcore_id = strtol(start, &end, 0); + if (end == start) { + fprintf(stderr, "No input lcore ID or ID %d iswrong\n",
lcore_id);app/test-dma-perf/main.c new file +7
-0+ ret =-1 ; +break ; + } + + lcore_dma_map = &test_case->lcore_dma_map; + lcore_dma_map->lcores[lcore_dma_map->cnt] = lcore_id; + strcpy(lcore_dma_map->dma_names[lcore_dma_map-Quoted 1 line
cnt], ptrs[1]);
app/test-dma-perf/main.c new file +23
-0+ lcore_dma_map->cnt++; + substr =strtok (NULL ,"," ); + }while (substr !=NULL ); + +out: +free (input); +return ret; +} + +static int +parse_entry (const char *value,struct test_configure_entry *entry) { +char input[255 ] = {0 }; +char *args[MAX_PARAMS_PER_ENTRY ]; +int args_nr =-1 ; + +if (value ==NULL || entry ==NULL ) + goto out; + +strncpy (input, value,254 ); +if (*input =='\0' ) + goto out; + + args_nr = rte_strsplit(input, strlen(input), args,
MAX_PARAMS_PER_ENTRY, ',');app/test-dma-perf/main.c new file +35
-0+if (args_nr !=1 && args_nr !=4 ) + goto out; + + entry->cur = entry->first = (uint32_t )atoi (args[0 ]); + +if (args_nr ==4 ) { + entry->last = (uint32_t )atoi (args[1 ]); + entry->incr = (uint32_t )atoi (args[2 ]); +if (!strcmp (args[3 ],"MUL" )) + entry->op =OP_MUL ; +else if (!strcmp (args[3 ],"ADD" )) + entry->op =OP_ADD ; +else { +printf ("Invalid op %s.\n" , args[3 ]); + args_nr =-1 ; + } + }else { + entry->op =OP_NONE ; + entry->last =0 ; + entry->incr =0 ; + } +out: +return args_nr; +} + +static uint16_t +load_configs (const char *path) +{ +struct rte_cfgfile *cfgfile; +int nb_sections, i; +struct test_configure *test_case; +char section_name[CFG_NAME_LEN ]; +const char *case_type; +const char *lcore_dma; + const char *mem_size_str, *buf_size_str, *ring_size_str,
*kick_batch_str;app/test-dma-perf/main.c new file +13
-0+int args_nr, nb_vp; +bool is_dma; + +printf ("config file parsing...\n" ); + cfgfile =rte_cfgfile_load (path,0 ); +if (!cfgfile) { +printf ("Open configure file error.\n" ); +exit (1 ); + } + + nb_sections =rte_cfgfile_num_sections (cfgfile,NULL ,0 ); +if (nb_sections >MAX_TEST_CASES ) { + printf("Error: The maximum number of cases is %d.\n",
MAX_TEST_CASES);app/test-dma-perf/main.c new file +7
-0+exit (1 ); + } + +for (i =0 ; i < nb_sections; i++) { +snprintf (section_name,CFG_NAME_LEN ,"case%d" , i +1 ); + test_case = &test_cases[i]; + case_type = rte_cfgfile_get_entry(cfgfile, section_name,"type");
app/test-dma-perf/main.c new file +2
-0+if (!case_type) { + printf("Error: No case type in case %d, the test will befinished here.\n",
app/test-dma-perf/main.c new file +7
-0+ i + 1); + test_case->is_valid = false; +continue ; + } + +if (strcmp (case_type,DMA_MEM_COPY ) ==0 ) { + test_case->test_type =
TEST_TYPE_DMA_MEM_COPY;app/test-dma-perf/main.c new file +4
-0+ test_case->test_type_str =DMA_MEM_COPY ; + is_dma = true; + }else if (strcmp (case_type,CPU_MEM_COPY ) ==0 ) { + test_case->test_type =
TEST_TYPE_CPU_MEM_COPY;app/test-dma-perf/main.c new file +4
-0+ test_case->test_type_str =CPU_MEM_COPY ; + is_dma = false; + }else { + printf("Error: Cannot find case type %s in case%d.\n",
case_type, i + 1);app/test-dma-perf/main.c new file +7
-0+ test_case->is_valid = false; +continue ; + } + + nb_vp =0 ; + + test_case->src_numa_node =(int)atoi(rte_cfgfile_get_entry(cfgfile,
+section_name, "src_numa_node"));app/test-dma-perf/main.c new file +1
-0+ test_case->dst_numa_node =(int)atoi(rte_cfgfile_get_entry(cfgfile,
+section_name, "dst_numa_node"));app/test-dma-perf/main.c new file +2
-0+ + mem_size_str = rte_cfgfile_get_entry(cfgfile, section_name,"
mem_size");app/test-dma-perf/main.c new file +1
-0+ args_nr = parse_entry(mem_size_str, &test_case-Quoted 1 line
mem_size);app/test-dma-perf/main.c new file +8
-0+if (args_nr <0 ) { +printf ("parse error in case %d.\n" , i +1 ); + test_case->is_valid = false; +continue ; + }else if (args_nr >1 ) + nb_vp++; + + buf_size_str = rte_cfgfile_get_entry(cfgfile, section_name,"
buf_size");app/test-dma-perf/main.c new file +10
-0+ args_nr =parse_entry (buf_size_str, &test_case->buf_size); +if (args_nr <0 ) { +printf ("parse error in case %d.\n" , i +1 ); + test_case->is_valid = false; +continue ; + }else if (args_nr >1 ) + nb_vp++; + +if (is_dma) { + ring_size_str = rte_cfgfile_get_entry(cfgfile,
section_name,
+"dma_ring_size");app/test-dma-perf/main.c new file +1
-0+ args_nr = parse_entry(ring_size_str, &test_case-Quoted 1 line
ring_size);app/test-dma-perf/main.c new file +8
-0+if (args_nr <0 ) { +printf ("parse error in case %d.\n" , i +1 ); + test_case->is_valid = false; +continue ; + }else if (args_nr >1 ) + nb_vp++; + + kick_batch_str = rte_cfgfile_get_entry(cfgfile,
section_name, "kick_batch");app/test-dma-perf/main.c new file +1
-0+ args_nr = parse_entry(kick_batch_str, &test_case-Quoted 1 line
kick_batch);app/test-dma-perf/main.c new file +8
-0+if (args_nr <0 ) { +printf ("parse error in case %d.\n" , i +1 ); + test_case->is_valid = false; +continue ; + }else if (args_nr >1 ) + nb_vp++; + + lcore_dma = rte_cfgfile_get_entry(cfgfile,
section_name, "lcore_dma");app/test-dma-perf/main.c new file +1
-0+ int lcore_ret = parse_lcore_dma(test_case,
lcore_dma);app/test-dma-perf/main.c
Quoted 2 lines
app/test-dma-perf/main.c new file +2
-0+if (lcore_ret <0 ) { + printf("parse lcore dma error in case %d.\n", i1);
Quoted 117 lines
app/test-dma-perf/main.c new file +5
-0+ test_case->is_valid = false; +continue ; + } + }else { + lcore_dma = rte_cfgfile_get_entry(cfgfile,
section_name, "lcore");app/test-dma-perf/main.c new file +10
-0+int lcore_ret =parse_lcore (test_case, lcore_dma); +if (lcore_ret <0 ) { +printf ("parse lcore error in case %d.\n" , i +1 ); + test_case->is_valid = false; +continue ; + } + } + +if (nb_vp >1 ) { + printf("Error, each section can only have a singlevariable parameter.\n");
app/test-dma-perf/main.c new file +6
-0+ test_case->is_valid = false; +continue ; + } + + test_case->cache_flush = + (int)atoi(rte_cfgfile_get_entry(cfgfile, section_name,"
cache_flush"));app/test-dma-perf/main.c new file +1
-0+ test_case->test_secs =(uint16_t)atoi(rte_cfgfile_get_entry(cfgfile,
app/test-dma-perf/main.c new file +3
-0+ section_name, "test_seconds")); + + test_case->eal_args = rte_cfgfile_get_entry(cfgfile,
section_name, "eal_args");app/test-dma-perf/main.c new file +19
-0+ test_case->is_valid = true; + } + +rte_cfgfile_close (cfgfile); +printf ("config file parsing complete.\n\n" ); +return i; +} + +/* Parse the argument given in the command line of the application */ +static int append_eal_args (int argc,char **argv,const char +*eal_args,char **new_argv) { +int i; +char *tokens[MAX_EAL_PARAM_NB ]; +char args[MAX_EAL_PARAM_LEN ] = {0 }; +int token_nb, new_argc =0 ; + +for (i =0 ; i < argc; i++) { + if ((strcmp(argv[i], CMDLINE_CONFIG_ARG) == 0) || + (strcmp(argv[i], CMDLINE_RESULT_ARG) ==0)) {
app/test-dma-perf/main.c new file +4
-0+ i++; +continue ; + } + strlcpy(new_argv[new_argc], argv[i],
sizeof(new_argv[new_argc]));app/test-dma-perf/main.c new file +37
-0+ new_argc++; + } + +if (eal_args) { +strlcpy (args, eal_args,sizeof (args)); + token_nb =rte_strsplit (args,strlen (args), + tokens,MAX_EAL_PARAM_NB ,' ' ); +for (i =0 ; i < token_nb; i++) +strcpy (new_argv[new_argc++], tokens[i]); + } + +return new_argc; +} + +int +main (int argc,char *argv[]) +{ +int ret; +uint16_t case_nb; +uint32_t i, nb_lcores; +pid_t cpid, wpid; +int wstatus; +char args[MAX_EAL_PARAM_NB ][MAX_EAL_PARAM_LEN ]; +char *pargs[MAX_EAL_PARAM_NB ]; +char *cfg_path_ptr =NULL ; +char *rst_path_ptr =NULL ; +char rst_path[PATH_MAX ]; +int new_argc; +bool is_first_case = true; + +memset (args,0 ,sizeof (args)); + +for (i =0 ; i <RTE_DIM (pargs); i++) + pargs[i] = args[i]; + +for (i =0 ; i < (uint32_t )argc; i++) { + if (strncmp(argv[i], CMDLINE_CONFIG_ARG,
MAX_LONG_OPT_SZ) == 0)app/test-dma-perf/main.c new file +2
-0+ cfg_path_ptr = argv[i +1 ]; + if (strncmp(argv[i], CMDLINE_RESULT_ARG,
MAX_LONG_OPT_SZ) == 0)app/test-dma-perf/main.c new file +21
-0+ rst_path_ptr = argv[i +1 ]; + } +if (cfg_path_ptr ==NULL ) { +printf ("Config file not assigned.\n" ); +return -1 ; + } +if (rst_path_ptr ==NULL ) { +strlcpy (rst_path, cfg_path_ptr,PATH_MAX ); +strcat (strtok (basename (rst_path),"." ),"_result.csv" ); + rst_path_ptr = rst_path; + } + + case_nb =load_configs (cfg_path_ptr); + fd =fopen (rst_path_ptr,"w" ); +if (fd ==NULL ) { +printf ("Open output CSV file error.\n" ); +return -1 ; + } +fclose (fd); + +for (i =0 ; i < case_nb; i++) {app/test-dma-perf/main.c line 505
Quoted 10 lines
app/test-dma-perf/main.c new file +10
-0+if (test_cases[i].test_type ==TEST_TYPE_NONE ) { +printf ("No test type in test case %d.\n\n" , i +1 ); +continue ; + } +if (!test_cases[i].is_valid) { +printf ("Invalid test case %d.\n\n" , i +1 ); +continue ; + } + + cpid =fork ();[Anoob] Do we really need
fork()? Can't we use code like,
Quoted 5 lines
RTE_LCORE_FOREACH_WORKER(lcore_id) { ret |= rte_eal_wait_lcore(lcore_id); }to wait for all threads to exit?
[Cheng] Good question. Fork() is used here to establish a new process for the new test case. In order for each test case to have a new EAL environment (for the flexibility), the EAL must be reinitialized for each case. However, the EAL parameters can only be initialized once per process. Therefore, we use a new process to run each new test case. Moreover, each test case runs sequentially and does not affect the others, ensuring the accuracy of the performance data. Your code would wait for all threads to exit in the same process. However, it would not provide a "clean" environment for each test case like fork() does. Fork() allows us to have a fully reinitialized environment, with no impact or side effects from previous test cases. This results in clean, precise performance data for each case.
Please let me know your thoughts on this. And please let me know if you have any other questions or require any clarification.
Thanks,
Cheng
Quoted 167 lines
Quoted 167 lines
app/test-dma-perf/main.c new file +7
-0+if (cpid <0 ) { +printf ("Fork case %d failed.\n" , i +1 ); +exit (EXIT_FAILURE ); + }else if (cpid ==0 ) { +printf ("\nRunning case %u\n\n" , i +1 ); + + new_argc = append_eal_args(argc, argv,test_cases[i].eal_args, pargs);
app/test-dma-perf/main.c new file +3
-0+ ret =rte_eal_init (new_argc, pargs); +if (ret <0 ) + rte_exit(EXIT_FAILURE, "Invalid EALarguments\n");
app/test-dma-perf/main.c new file +6
-0+ +/* Check lcores. */ + nb_lcores =rte_lcore_count (); +if (nb_lcores <2 ) + rte_exit(EXIT_FAILURE, + "There should be at least 2 workerlcores.\n");
app/test-dma-perf/main.c new file +36
-0+ + fd =fopen (rst_path_ptr,"a" ); +if (!fd) { +printf ("Open output CSV file error.\n" ); +return 0 ; + } + +if (is_first_case) { +output_env_info (); + is_first_case = false; + } +run_test (i +1 , &test_cases[i]); + +/* clean up the EAL */ +rte_eal_cleanup (); + +fclose (fd); + +printf ("\nCase %u completed.\n\n" , i +1 ); + +exit (EXIT_SUCCESS ); + }else { + wpid =waitpid (cpid, &wstatus,0 ); +if (wpid ==-1 ) { +printf ("waitpid error.\n" ); +exit (EXIT_FAILURE ); + } + +if (WIFEXITED (wstatus)) +printf ("Case process exited. status %d\n\n" , +WEXITSTATUS (wstatus)); +else if (WIFSIGNALED (wstatus)) +printf ("Case process killed by signal %d\n\n" , +WTERMSIG (wstatus)); +else if (WIFSTOPPED (wstatus)) + printf("Case process stopped by signal%d\n\n",
app/test-dma-perf/main.c new file +5
-0+ WSTOPSIG(wstatus)); + else if (WIFCONTINUED(wstatus)) + printf("Case process continued.\n\n"); + else + printf("Case process unknownterminated.\n\n");
app/test-dma-perf/main.c new file +7
-0+ } + } + +printf ("Bye...\n" ); +return 0 ; +} +app/test-dma-perf/main.h new renamed from app/test-dma-perf/main.h +0
-0file mode 100644 index 0000000000..215ac42673
app/test-dma-perf/main.h new file +68
-0@@ -0,0 +1,69 @@ +/* SPDX-License-Identifier: BSD-3-Clause +* Copyright(c) 2023 Intel Corporation */ + +#ifndef _MAIN_H_ +#define _MAIN_H_ + + +#include <rte_common.h> +#include <rte_cycles.h> +#include <rte_dev.h> +#include <rte_dmadev.h> + +#ifndef __maybe_unused +#define __maybe_unused __rte_unused +#endif + +#define MAX_WORKER_NB 128 +#define MAX_OUTPUT_STR_LEN 512 + +#define MAX_DMA_NB 128 +#define MAX_LCORE_NB 256 + +extern char output_str[MAX_WORKER_NB ][MAX_OUTPUT_STR_LEN ]; + +typedef enum { +OP_NONE =0 , +OP_ADD , +OP_MUL +}alg_op_type ; + +struct test_configure_entry { +uint32_t first; +uint32_t last; +uint32_t incr; +alg_op_type op; +uint32_t cur; +}; + +struct lcore_dma_map_t { +uint32_t lcores[MAX_WORKER_NB ]; +char dma_names[MAX_WORKER_NB ][RTE_DEV_NAME_MAX_LEN ]; +int16_t dma_ids[MAX_WORKER_NB ]; +uint16_t cnt; +}; + +struct test_configure { +bool is_valid; +uint8_t test_type; +const char *test_type_str; +uint16_t src_numa_node; +uint16_t dst_numa_node; +uint16_t opcode; +bool is_dma; +struct lcore_dma_map_t lcore_dma_map; +struct test_configure_entry mem_size; +struct test_configure_entry buf_size; +struct test_configure_entry ring_size; +struct test_configure_entry kick_batch; +uint32_t cache_flush; +uint32_t nr_buf; +uint16_t test_secs; +const char *eal_args; +uint8_t scenario_id; +}; + +void mem_copy_benchmark (struct test_configure *cfg,bool is_dma); + +#endif /* _MAIN_H_ */ app/test-dma- renamed from app/test-dma-perf/meson.build +0
-0perf/meson.build new file mode 100644 index 0000000000..bd6c264002
app/test-dma-perf/meson.build new file +17
-1@@ -0,0 +1,17 @@ +# SPDX-License-Identifier: BSD-3-Clause # Copyright(c) 2019-2023 +Intel Corporation + +# meson file, for building this app as part of a main DPDK build. + +if is_windows + build = false + reason = 'not supported on Windows' + subdir_done() +endif + +deps += ['dmadev', 'mbuf', 'cfgfile'] + +sources = files( + 'main.c', + 'benchmark.c', +) --2.40.1
Hi Cheng,
Please see inline.
Thanks,
Anoob
Quoted 501 lines
-----Original Message-----
From: Jiang, Cheng1 <[email protected]>
Sent: Thursday, June 15, 2023 1:31 PM
To: Anoob Joseph <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>; Amit Prakash Shukla <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; Ma, WenwuX <[email protected]>; Wang, YuanX <[email protected]>; He, Xingguang <[email protected]>
Subject: RE: [EXT] [PATCH v6] app/dma-perf: introduce dma-perf applicationHi,
Thanks for your comments, the replies are inline.
Thanks,
ChengQuoted 9 lines
-----Original Message-----
From: Anoob Joseph <[email protected]>
Sent: Thursday, June 15, 2023 1:22 PM
To: Jiang, Cheng1 <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>; Amit Prakash Shukla <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; Ma, WenwuX <[email protected]>; Wang,YuanX
Quoted 4 lines
<[email protected]>; He, Xingguang <[email protected]>
Subject: RE: [EXT] [PATCH v6] app/dma-perf: introduce dma-perf applicationQuoted 12 lines
Hi,
Thanks for working on the comments. Few more top level comment inline.
Thanks,
AnoobQuoted 5 lines
-----Original Message-----
From: Cheng Jiang <[email protected]>
Sent: Tuesday, June 13, 2023 10:02 AM
To: [email protected]; [email protected]; [email protected]; [email protected]; Amit PrakashShukla
Quoted 3 lines
Quoted 3 lines
<[email protected]>; Anoob Joseph <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected];Quoted 18 lines
Quoted 6 lines
Cheng Jiang <[email protected]>
Subject: [EXT] [PATCH v6] app/dma-perf: introduce dma-perf applicationExternal Email
Quoted 12 lines
--------------------------------------------------------------------
-- There are many high-performance DMA devices supported in DPDK now, and these DMA devices can also be integrated into other modules of DPDK as accelerators, such as Vhost. Before integrating DMA into applications, developers need to know the performance of these DMA devices in various scenarios and the performance of CPUs in the same scenario, such as different buffer lengths. Only in this way can we know the target performance of the application accelerated by using them. This patch introduces a high-performance testing tool, which supports comparing the performance of CPU and DMA in different scenarios automatically with a pre- set config file. Memory Copy performance test areQuoted 26 lines
supported for now.
Quoted 25 lines
Signed-off-by Cheng Jiang <[email protected]>Signed-off-by Jiayu Hu <[email protected]>Signed-off-by Yuan Wang <[email protected]>Acked-by Morten Brørup <[email protected]>Acked-by Chenbo Xia <[email protected]>---
v6:
improved code based on Anoob's comments; fixed some code structure issues;
v5:
fixed someLONG_LINEwarnings;
v4:
fixed inaccuracy of the memory footprint display;
v3:
fixed some typos;
v2:
added lcore/dmadev designation;
added error case process;
removedworker_threadsparameter from config.ini; improved the logs;
improved config file;app/meson.build | 1 + app/test-dma-perf/benchmark.c | 477++++++++++++++++++++++++++++
Quoted 42 lines
Quoted 42 lines
app/test-dma-perf/config.ini | 59 ++++
app/test-dma-perf/main.c | 569 ++++++++++++++++++++++++++++++++++ app/test-dma-perf/main.h | 69 +++++ app/test-dma-perf/meson.build | 17 + 6 files changed, 1192 insertions(+) create mode 100644
app/test-dma-perf/benchmark.ccreate mode 100644 app/test-dma-perf/config.ini create mode 100644 app/test-dma-perf/main.ccreate mode 100644app/test-dma-perf/main.hcreate mode
100644 app/test-dma-perf/meson.buildapp/meson.build index renamed from app/meson.build +0
-0
74d2420f67..4fc1a83eba100644app/meson.build +1
-0@@ -19,6 +19,7 @@ apps = [ 'test-cmdline', 'test-compress-perf', 'test-crypto-perf', + 'test-dma-perf', 'test-eventdev', 'test-fib', 'test-flow-perf',app/test-dma- renamed from app/test-dma-perf/benchmark.c +0
-0
perf/benchmark.cnew file mode 100644 index 0000000000..bc1ca82297app/test-dma-perf/benchmark.c new file +13
-0@@ -0,0 +1,477 @@ +/* SPDX-License-Identifier: BSD-3-Clause +* Copyright(c) 2023 Intel Corporation */ + +#include <inttypes.h> +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> + +#include <rte_time.h> +#include <rte_mbuf.h> +#include <rte_dmadev.h> +#include <rte_malloc.h> +#include <rte_lcore.h> Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+ +#include "main.h" + +#define MAX_DMA_CPL_NB 255 + +#define TEST_WAIT_U_SECOND 10000 + +#define CSV_LINE_DMA_FMT "Scenario %u,%u,%s,%u,%u,%.2lf,%"PRIu64
Quoted 2 lines
Quoted 2 lines
",%.3lf,%.3lf\n"
+#defineCSV_LINE_CPU_FMT"Scenario %u,%u,NA,%u,%u,%.2lf,%"PRIu64
Quoted 21 lines
Quoted 21 lines
",%.3lf,%.3lf\n"
app/test-dma-perf/benchmark.c new file +20
-0+ +struct worker_info { +bool ready_flag; +bool start_flag; +bool stop_flag; +uint32_t total_cpl; +uint32_t test_cpl; +}; + +struct lcore_params { +uint8_t scenario_id; +unsigned int lcore_id; +char *dma_name; +uint16_t worker_id; +uint16_t dev_id; +uint32_t nr_buf; +uint16_t kick_batch; +uint32_t buf_size; +uint16_t test_secs; +struct rte_mbuf **srcs;Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+struct rte_mbuf **dsts; +struct worker_info worker_info; +}; + +static struct rte_mempool *src_pool;static struct rte_mempool +*dst_pool; + +static volatile struct lcore_params*worker_params[MAX_WORKER_NB];
Quoted 139 lines
Quoted 139 lines
app/test-dma-perf/benchmark.c new file +20
-0+ +#define PRINT_ERR (...) print_err(__func__, __LINE__, __VA_ARGS__) + +static inline int +__rte_format_printf(3, 4) +print_err(const char *func, int lineno, const char *format, ...) { + va_list ap; + int ret; + + ret = fprintf(stderr, "In %s:%d - ", func, lineno); + va_start(ap, format); + ret += vfprintf(stderr, format, ap); + va_end(ap); + + return ret; +} + +static inline void +calc_result(uint32_t buf_size, uint32_t nr_buf, uint16_t +nb_workers,
uint16_ttest_secs,app/test-dma-perf/benchmark.c new file +1
-0+ uint32_t total_cnt, float *memory, uint32_t
*ave_cycle,app/test-dma-perf/benchmark.c new file +3
-0+ float *bandwidth, float *mops) +{ + *memory = (float)(buf_size * (nr_buf / nb_workers) * 2) / (1024 *1024);
app/test-dma-perf/benchmark.c new file +2
-0+ *ave_cycle = test_secs *rte_get_timer_hz () / total_cnt; + *bandwidth = (buf_size * 8 * (rte_get_timer_hz() /(float)*ave_cycle)) / 1000000000;
app/test-dma-perf/benchmark.c new file +5
-0+ *mops = (float )rte_get_timer_hz () / *ave_cycle /1000000 ; } + +static void +output_result(uint8_t scenario_id, uint32_t lcore_id, char +*dma_name,
uint64_tave_cycle,app/test-dma-perf/benchmark.c new file +9
-0+ uint32_t buf_size, uint32_t nr_buf, float memory, + float bandwidth, float mops, bool is_dma) { + if (is_dma) + printf("lcore %u, DMA %s:\n", lcore_id, dma_name); + else + printf("lcore %u\n", lcore_id); + + printf("average cycles/op: %" PRIu64 ", buffer size: %u, nr_buf: +%u,memory: %.2lfMB, frequency: %" PRIu64 ".\n",
app/test-dma-perf/benchmark.c new file +1
-0+ ave_cycle, buf_size, nr_buf, memory,
rte_get_timer_hz());app/test-dma-perf/benchmark.c new file +5
-0+printf ("Average bandwidth: %.3lfGbps, MOps: %.3lf\n" , bandwidth, +mops); + +if (is_dma) + snprintf(output_str[lcore_id], MAX_OUTPUT_STR_LEN,
CSV_LINE_DMA_FMT,app/test-dma-perf/benchmark.c new file +4
-0+ scenario_id, lcore_id, dma_name, buf_size, + nr_buf, memory, ave_cycle, bandwidth, mops); + else + snprintf(output_str[lcore_id], MAX_OUTPUT_STR_LEN,
CSV_LINE_CPU_FMT,app/test-dma-perf/benchmark.c new file +41
-0+ scenario_id, lcore_id, buf_size, + nr_buf, memory, ave_cycle, bandwidth, mops); } + +static inline void +cache_flush_buf (__maybe_unused struct rte_mbuf **array, +__maybe_unused uint32_t buf_size, +__maybe_unused uint32_t nr_buf) +{ +#ifdef RTE_ARCH_X86_64 +char *data; +struct rte_mbuf **srcs = array; +uint32_t i, offset; + +for (i =0 ; i < nr_buf; i++) { + data = rte_pktmbuf_mtod(srcs[i], char *); +for (offset =0 ; offset < buf_size; offset +=64 ) +__builtin_ia32_clflush (data + offset); + } +#endif +} + +/* Configuration of device. */ +static void +configure_dmadev_queue (uint32_t dev_id,uint32_t ring_size) { +uint16_t vchan =0 ; +struct rte_dma_info info; +struct rte_dma_conf dev_config = { .nb_vchans =1 }; +struct rte_dma_vchan_conf qconf = { + .direction =RTE_DMA_DIR_MEM_TO_MEM , + .nb_desc = ring_size + }; + +if (rte_dma_configure (dev_id, &dev_config) !=0 ) +rte_exit (EXIT_FAILURE ,"Error with dma configure.\n" ); + +if (rte_dma_vchan_setup (dev_id, vchan, &qconf) !=0 ) +rte_exit (EXIT_FAILURE ,"Error with queue configuration.\n" ); + +rte_dma_info_get (dev_id, &info); +if (info.nb_vchans !=1 ) + rte_exit(EXIT_FAILURE, "Error, no configured queuesreported on device id. %u\n",
app/test-dma-perf/benchmark.c new file +20
-0+ dev_id); + +if (rte_dma_start (dev_id) !=0 ) +rte_exit (EXIT_FAILURE ,"Error with dma start.\n" ); } + +static int +config_dmadevs (struct test_configure *cfg) { +uint32_t ring_size = cfg->ring_size.cur; +struct lcore_dma_map_t *ldm = &cfg->lcore_dma_map; +uint32_t nb_workers = ldm->cnt; +uint32_t i; +int dev_id; +uint16_t nb_dmadevs =0 ; +char *dma_name; + +for (i =0 ; i < ldm->cnt; i++) { + dma_name = ldm->dma_names[i]; + dev_id =rte_dma_get_dev_id_by_name (dma_name); +if (dev_id ==-1 ) { + fprintf(stderr, "Error: Fail to find DMA %s.\n",
dma_name);app/test-dma-perf/benchmark.c new file +11
-0+ goto end; + } + + ldm->dma_ids[i] = dev_id; +configure_dmadev_queue (dev_id, ring_size); + ++nb_dmadevs; + } + +end: +if (nb_dmadevs < nb_workers) { + printf("Not enough dmadevs (%u) for all workers (%u).\n",
nb_dmadevs,nb_workers);app/test-dma-perf/benchmark.c new file +5
-0+return -1 ; + } + +printf ("Number of used dmadevs: %u.\n" , nb_dmadevs); +Quoted 8 lines
app/test-dma-perf/benchmark.c line 187
Quoted 6 lines
app/test-dma-perf/benchmark.c new file +6
-0+return 0 ; +} + +#define POLL_MAX 1000 + +[Anoob] Extra blank line. You can consider removing.
[Cheng] sure, sorry for the miss.
Quoted 112 lines
Quoted 112 lines
app/test-dma-perf/benchmark.c new file +14
-0+static inline void +do_dma_submit_and_poll (uint16_t dev_id,uint64_t *async_cnt, +volatile struct worker_info *worker_info) { +int ret; +uint16_t nr_cpl; + + ret =rte_dma_submit (dev_id,0 ); +if (ret <0 ) { +rte_dma_stop (dev_id); +rte_dma_close (dev_id); +rte_exit (EXIT_FAILURE ,"Error with dma submit.\n" ); + } + + nr_cpl = rte_dma_completed(dev_id, 0, MAX_DMA_CPL_NB, NULL,NULL);
app/test-dma-perf/benchmark.c new file +32
-0+ *async_cnt -= nr_cpl; + worker_info->total_cpl += nr_cpl; +} + +static inline int +do_dma_mem_copy (void *p) +{ +const uint16_t *para_idx = (uint16_t *)p; +volatile struct lcore_params *para = worker_params[*para_idx]; +volatile struct worker_info *worker_info = &(para->worker_info); +const uint16_t dev_id = para->dev_id; +const uint32_t nr_buf = para->nr_buf; +const uint16_t kick_batch = para->kick_batch; +const uint32_t buf_size = para->buf_size; +struct rte_mbuf **srcs = para->srcs; +struct rte_mbuf **dsts = para->dsts; +uint16_t nr_cpl; +uint64_t async_cnt =0 ; +uint32_t i; +uint32_t poll_cnt =0 ; +int ret; + + worker_info->stop_flag = false; + worker_info->ready_flag = true; + +while (!worker_info->start_flag) + ; + +while (1 ) { +for (i =0 ; i < nr_buf; i++) { +dma_copy: + ret = rte_dma_copy(dev_id, 0,
rte_pktmbuf_iova(srcs[i]),app/test-dma-perf/benchmark.c new file +4
-0+ rte_pktmbuf_iova(dsts[i]), buf_size, 0); +if (unlikely (ret <0 )) { +if (ret == -ENOSPC ) { + do_dma_submit_and_poll(dev_id,
&async_cnt,worker_info);app/test-dma-perf/benchmark.c new file +5
-0+ goto dma_copy; + }else { +/* Error exit */ +rte_dma_stop (dev_id); + rte_exit(EXIT_FAILURE, "DMAenqueue failed\n");
app/test-dma-perf/benchmark.c new file +6
-0+ } + } + async_cnt++; + +if ((async_cnt % kick_batch) ==0 ) + do_dma_submit_and_poll(dev_id,
&async_cnt,worker_info);app/test-dma-perf/benchmark.c new file +9
-0+ } + +if (worker_info->stop_flag) +break ; + } + +rte_dma_submit (dev_id,0 ); +while ((async_cnt >0 ) && (poll_cnt++ <POLL_MAX )) { + nr_cpl = rte_dma_completed(dev_id, 0,
MAX_DMA_CPL_NB, NULL, NULL);app/test-dma-perf/benchmark.c new file +28
-0+ async_cnt -= nr_cpl; + } + +return 0 ; +} + +static inline int +do_cpu_mem_copy (void *p) +{ +const uint16_t *para_idx = (uint16_t *)p; +volatile struct lcore_params *para = worker_params[*para_idx]; +volatile struct worker_info *worker_info = &(para->worker_info); +const uint32_t nr_buf = para->nr_buf; +const uint32_t buf_size = para->buf_size; +struct rte_mbuf **srcs = para->srcs; +struct rte_mbuf **dsts = para->dsts; +uint32_t i; + + worker_info->stop_flag = false; + worker_info->ready_flag = true; + +while (!worker_info->start_flag) + ; + +while (1 ) { +for (i =0 ; i < nr_buf; i++) { +/* copy buffer form src to dst */ + rte_memcpy((void*)(uintptr_t)rte_mbuf_data_iova(dsts[i]),
app/test-dma-perf/benchmark.c new file +1
-0+ (void*)(uintptr_t)rte_mbuf_data_iova(srcs[i]),
app/test-dma-perf/benchmark.c new file +4
-0+ (size_t)buf_size); + worker_info->total_cpl++; + } +if (worker_info->stop_flag)Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+break ; + } + +return 0 ; +} + +static int +setup_memory_env(struct test_configure *cfg, struct rte_mbuf***srcs,
Quoted 23 lines
Quoted 23 lines
app/test-dma-perf/benchmark.c new file +10
-0+ struct rte_mbuf ***dsts) +{ +unsigned int buf_size = cfg->buf_size.cur; +unsigned int nr_sockets; +uint32_t nr_buf = cfg->nr_buf; + + nr_sockets =rte_socket_count (); +if (cfg->src_numa_node >= nr_sockets || + cfg->dst_numa_node >= nr_sockets) { + printf("Error: Source or destination numa exceeds the acturenuma nodes.\n");
app/test-dma-perf/benchmark.c new file +12
-0+return -1 ; + } + + src_pool =rte_pktmbuf_pool_create ("Benchmark_DMA_SRC" , + nr_buf,/* n == num elements */ +64 ,/* cache size */ +0 ,/* priv size */ + buf_size +RTE_PKTMBUF_HEADROOM , + cfg->src_numa_node); +if (src_pool ==NULL ) { +PRINT_ERR ("Error with source mempool creation.\n" ); +return -1 ;Quoted 8 lines
app/test-dma-perf/benchmark.c line 326
Quoted 5 lines
app/test-dma-perf/benchmark.c new file +5
-0+ } + + dst_pool = rte_pktmbuf_pool_create("Benchmark_DMA_DST", + nr_buf, /* n == num elements */ + 64, /* cache size */[Anoob] We do not alloc or free pointers in the datapath, right? So why bother with cache?
[Cheng] Yes, you are right, the cache size is not necessary here, I'll fix it in the next version.
Quoted 16 lines
Quoted 4 lines
app/test-dma-perf/benchmark.c new file +3
-0+0 ,/* priv size */ + buf_size +RTE_PKTMBUF_HEADROOM , + cfg->dst_numa_node);app/test-dma-perf/benchmark.c line 339
Quoted 10 lines
app/test-dma-perf/benchmark.c new file +10
-0+if (dst_pool ==NULL ) { +PRINT_ERR ("Error with destination mempool creation.\n" ); +return -1 ; + } + + *srcs =rte_malloc (NULL , nr_buf *sizeof (struct rte_mbuf *),0 ); +if (*srcs ==NULL ) { +printf ("Error: srcs malloc failed.\n" ); +return -1 ; + }[Anoob] Are we freeing these memory? The ones allocated with
Quoted 4 lines
rte_malloc.[Cheng] yes, we freed the memory in the end of
mem_copy_benchmark()when we finished the test.
[Anoob] I think we are not freeing this mem. In the place where we free all mem, we do free all objects to mempool as well as the mempools. But this memory is to hold the pointers, right? Is that getting freed anywhere?
Also, in the mem clearing paths, do we need to clear the static variables (ie, set srcs, src_pool, dsts, dst_pool to NULL) so that there won't be any scope for any double free.
Quoted 242 lines
Quoted 60 lines
Quoted 60 lines
app/test-dma-perf/benchmark.c new file +30
-0+ + *dsts =rte_malloc (NULL , nr_buf *sizeof (struct rte_mbuf *),0 ); +if (*dsts ==NULL ) { +printf ("Error: dsts malloc failed.\n" ); +return -1 ; + } + +if (rte_mempool_get_bulk (src_pool, (void **)*srcs, nr_buf) !=0 ) { +printf ("get src mbufs failed.\n" ); +return -1 ; + } +if (rte_mempool_get_bulk (dst_pool, (void **)*dsts, nr_buf) !=0 ) { +printf ("get dst mbufs failed.\n" ); +return -1 ; + } + +return 0 ; +} + +void +mem_copy_benchmark (struct test_configure *cfg,bool is_dma) { +uint16_t i; +uint32_t offset; +unsigned int lcore_id =0 ; +struct rte_mbuf **srcs =NULL , **dsts =NULL ; +struct lcore_dma_map_t *ldm = &cfg->lcore_dma_map; +unsigned int buf_size = cfg->buf_size.cur; +uint16_t kick_batch = cfg->kick_batch.cur; +uint32_t nr_buf = cfg->nr_buf = (cfg->mem_size.cur *1024 *1024 ) +/(
cfg->buf_size.cur* 2);app/test-dma-perf/benchmark.c new file +27
-0+uint16_t nb_workers = ldm->cnt; +uint16_t test_secs = cfg->test_secs; +float memory; +uint32_t avg_cycles =0 ; +float mops; +float bandwidth; + +if (setup_memory_env (cfg, &srcs, &dsts) <0 ) + goto out; + +if (is_dma) +if (config_dmadevs (cfg) <0 ) + goto out; + +if (cfg->cache_flush) { +cache_flush_buf (srcs, buf_size, nr_buf); +cache_flush_buf (dsts, buf_size, nr_buf); +rte_mb (); + } + +printf ("Start testing....\n" ); + +for (i =0 ; i < nb_workers; i++) { + lcore_id = ldm->lcores[i]; + offset = nr_buf / nb_workers * i; + + worker_params[i] = rte_malloc(NULL, sizeof(structQuoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 6 lines
lcore_params), 0);app/test-dma-perf/benchmark.c new file +2
-0+if (!worker_params[i]) { + printf("lcore parameters malloc failure for lcore%d\n",
lcore_id);app/test-dma-perf/benchmark.c new file +2
-0+break ; + }[Anoob] Are we freeing the above memory?
[Cheng] sorry, I missed that, I'll add
worker_paramsmemory free in the next version, thanks.Quoted 89 lines
Quoted 89 lines
app/test-dma-perf/benchmark.c new file +2
-0+if (is_dma) { + worker_params[i]->dma_name = ldm-Quoted 1 line
dma_names[i];
app/test-dma-perf/benchmark.c new file +5
-0+ worker_params[i]->dev_id = ldm->dma_ids[i]; + worker_params[i]->kick_batch = kick_batch; + } + worker_params[i]->worker_id = i; + worker_params[i]->nr_buf = (uint32_t)(nr_buf /
nb_workers);app/test-dma-perf/benchmark.c new file +9
-0+ worker_params[i]->buf_size = buf_size; + worker_params[i]->test_secs = test_secs; + worker_params[i]->srcs = srcs + offset; + worker_params[i]->dsts = dsts + offset; + worker_params[i]->scenario_id = cfg->scenario_id; + worker_params[i]->lcore_id = lcore_id; + +if (is_dma) + rte_eal_remote_launch(do_dma_mem_copy, (void*)(&i),
lcore_id);app/test-dma-perf/benchmark.c new file +2
-0+else + rte_eal_remote_launch(do_cpu_mem_copy, (void*)(&i),
lcore_id);app/test-dma-perf/benchmark.c new file +6
-0+ } + +while (1 ) { +bool ready = true; +for (i =0 ; i < nb_workers; i++) { + if (worker_params[i]->worker_info.ready_flag ==false) {
app/test-dma-perf/benchmark.c new file +19
-0+ ready =0 ; +break ; + } + } +if (ready) +break ; + } + +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.start_flag = true; + +usleep (TEST_WAIT_U_SECOND ); +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.test_cpl = +worker_params[i]->worker_info.total_cpl; + +usleep (test_secs *1000 *1000 ); +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.test_cpl =worker_params[i]->worker_info.total_cpl -
app/test-dma-perf/benchmark.c new file +1
-0+ worker_params[i]-Quoted 1 line
worker_info.test_cpl;
app/test-dma-perf/benchmark.c new file +12
-0+ +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.stop_flag = true; + +rte_eal_mp_wait_lcore (); + +for (i =0 ; i < nb_workers; i++) { +calc_result (buf_size, nr_buf, nb_workers, test_secs, + worker_params[i]->worker_info.test_cpl, + &memory, &avg_cycles, &bandwidth, &mops); + output_result(cfg->scenario_id, worker_params[i]->lcore_id, + worker_params[i]->dma_name,
avg_cycles,buf_size,app/test-dma-perf/benchmark.c new file +1
-0+ nr_buf / nb_workers, memory,bandwidth, mops,
is_dma);app/test-dma-perf/benchmark.c new file +21
-0+ } + +out: +/* free env */ +if (srcs) +rte_pktmbuf_free_bulk (srcs, nr_buf); +if (dsts) +rte_pktmbuf_free_bulk (dsts, nr_buf); + +if (src_pool) +rte_mempool_free (src_pool); +if (dst_pool) +rte_mempool_free (dst_pool); + +if (is_dma) { +for (i =0 ; i < nb_workers; i++) { +printf ("Stopping dmadev %d\n" , ldm->dma_ids[i]); +rte_dma_stop (ldm->dma_ids[i]); + } + } +}(quoted diff) +0
-0Quoted 8 lines
app/test-dma-perf/config.ini line 3
Quoted 8 lines
b/app/test-dma-perf/config.ini new file mode 100644 index 0000000000..2fd9c3c387
app/test-dma-perf/config.ini new file +3
-0@@ -0,0 +1,59 @@ + +; This is an example configuration file for dma-perf, which details +the meanings of each parameter ; and instructions on how to use dma-perf.
Quoted 63 lines
Quoted 49 lines
app/test-dma-perf/config.ini new file +8
-0+ +; Supported test types are DMA_MEM_COPY and CPU_MEM_COPY. + +; Parameters: +; "mem_size" denotes the size of the memory footprint. +; "buf_size" denotes the memory size of a single operation. +; "dma_ring_size" denotes the dma ring buffer size. It should be +greaterthan 64 normally.
app/test-dma-perf/config.ini new file +2
-0+; "kick_batch" denotes the dma operation batch size, and should be +greaterthan 1 normally.
app/test-dma-perf/config.ini new file +4
-0+ +; The format for variables is variable=first,last,increment,ADD|MUL. + +; src_numa_node is used to control the numa node where the sourcememory is allocated.
app/test-dma-perf/config.ini new file +2
-0+; dst_numa_node is used to control the numa node where the +destinationmemory is allocated.
app/test-dma-perf/config.ini new file +10
-0+ +; cache_flush is used to determine whether or not the cache should +be flushed, with 1 indicating to ; flush and 0 indicating to not flush. + +; test_seconds controls the test time of the whole case. + +; To use DMA for a test, please specify the "lcore_dma" parameter. +; If you have already set the "-l" and "-a" parameters using EAL, ; +make sure that the value of "lcore_dma" falls within their range of +thevalues.
app/test-dma-perf/config.ini new file +7
-0+ +; To use CPU for a test, please specify the "lcore" parameter. +; If you have already set the "-l" and "-a" parameters using EAL, ; +make sure that the value of "lcore" falls within their range of values. + +; To specify a configuration file, use the "--config" flag followed +by the pathto the file.
app/test-dma-perf/config.ini new file +3
-0+ +; To specify a result file, use the "--result" flag followed by the +path to thefile.
app/test-dma-perf/config.ini new file +3
-0+; If you do not specify a result file, one will be generated with +the same name as the configuration ; file, with the addition of +"_result.csv" atthe end.
app/test-dma-perf/config.ini new file +2
-0+ +[case1]app/test-dma-perf/config.ini line 47
Quoted 10 lines
app/test-dma-perf/config.ini new file +10
-0+type=DMA_MEM_COPY +mem_size=10 +buf_size=64,8192,2,MUL +dma_ring_size=1024 +kick_batch=32 +src_numa_node=0 +dst_numa_node=0 +cache_flush=0 +test_seconds=2 +lcore_dma=lcore10@0000:00:04.2, lcore11@0000:00:04.3[Anoob] Isn't it better if we allow user to specify DMA dev ID rather than the PCI DBDF?
Quoted 7 lines
In the long run, I would expect config file to provide {core,
dma_dev_id,queue_id}Another thought is why to expose this at all? If we can restrict this perf application to have one thread only use one vchan, then application can easily create this mapping in run time. Unless you want one thread to use 2 different vchans which may not be desirable
since this is a standalone perf app.
Quoted 15 lines
[Cheng] Thank you for the feedback.
Here are my thoughts:
Firstly, the user may not know which device the DMA dev ID corresponds to, or which NUMA node it is on. In my example, I used the CBDMA environment, so I did not specify the work queue ID. When using DSA, the configuration would be something like lcore10@0000:00:04.2-q0 which contains core, dma and work queue id. The reason for exposing these options is that we want the user to fully understand which cores and devices are being used so that they know exactly where the performance data is coming from. For example, performance when cores and DMA devices are not on the same NUMA node, etc. This allows the testing scenario to be precise and flexible. If the application handles the mapping itself, the user loses control over the mapping and may not get the performance data they want. We believe control should be given to the user rather than the application.
[Anoob] I understand your view points. Thanks for the explanation.
Quoted 577 lines
Quoted 559 lines
Quoted 426 lines
app/test-dma-perf/config.ini new file +12
-0+eal_args=--in-memory --file-prefix=test + +[case2] +type=CPU_MEM_COPY +mem_size=10 +buf_size=64,8192,2,MUL +src_numa_node=0 +dst_numa_node=1 +cache_flush=0 +test_seconds=2 +lcore = 3, 4 +eal_args=--in-memory --no-pciapp/test-dma-perf/main.c new renamed from app/test-dma-perf/main.c +0
-0file mode 100644 index 0000000000..d65655b87b
app/test-dma-perf/main.c new file +21
-0@@ -0,0 +1,569 @@ +/* SPDX-License-Identifier: BSD-3-Clause +* Copyright(c) 2023 Intel Corporation */ + +#include <stdio.h> +#include <stdlib.h> +#include <getopt.h> +#include <signal.h> +#include <stdbool.h> +#include <unistd.h> +#include <sys/wait.h> +#include <inttypes.h> +#include <libgen.h> + +#include <rte_eal.h> +#include <rte_cfgfile.h> +#include <rte_string_fns.h> +#include <rte_lcore.h> + +#include "main.h" + +#define CSV_HDR_FMT "Case %u : %s,lcore,DMA,buffersize,nr_buf,memory(MB),cycle,bandwidth(Gbps),MOps\n"
app/test-dma-perf/main.c new file +51
-0+ +#define MAX_EAL_PARAM_NB 100 +#define MAX_EAL_PARAM_LEN 1024 + +#define DMA_MEM_COPY "DMA_MEM_COPY" +#define CPU_MEM_COPY "CPU_MEM_COPY" + +#define CMDLINE_CONFIG_ARG "--config" +#define CMDLINE_RESULT_ARG "--result" + +#define MAX_PARAMS_PER_ENTRY 4 + +#define MAX_LONG_OPT_SZ 64 + +enum { +TEST_TYPE_NONE =0 , +TEST_TYPE_DMA_MEM_COPY , +TEST_TYPE_CPU_MEM_COPY +}; + +#define MAX_TEST_CASES 16 +static struct test_configure test_cases[MAX_TEST_CASES ]; + +char output_str[MAX_WORKER_NB ][MAX_OUTPUT_STR_LEN ]; + +static FILE *fd; + +static void +output_csv (bool need_blankline) +{ +uint32_t i; + +if (need_blankline) { +fprintf (fd,",,,,,,,,\n" ); +fprintf (fd,",,,,,,,,\n" ); + } + +for (i =0 ; i <RTE_DIM (output_str); i++) { +if (output_str[i][0 ]) { +fprintf (fd,"%s" , output_str[i]); + output_str[i][0 ] ='\0' ; + } + } + +fflush (fd); +} + +static void +output_env_info (void ) +{ + snprintf(output_str[0], MAX_OUTPUT_STR_LEN, "testenvironment:\n");
app/test-dma-perf/main.c new file +68
-0+snprintf (output_str[1 ],MAX_OUTPUT_STR_LEN ,"CPU frequency,%" + PRIu64"\n" ,rte_get_timer_hz ()); + +output_csv (true); +} + +static void +output_header (uint32_t case_id,struct test_configure *case_cfg) { +snprintf (output_str[0 ],MAX_OUTPUT_STR_LEN , +CSV_HDR_FMT , case_id, case_cfg->test_type_str); + +output_csv (true); +} + +static void +run_test_case (struct test_configure *case_cfg) { +switch (case_cfg->test_type) { +case TEST_TYPE_DMA_MEM_COPY : +mem_copy_benchmark (case_cfg, true); +break ; +case TEST_TYPE_CPU_MEM_COPY : +mem_copy_benchmark (case_cfg, false); +break ; +default : +printf ("Unknown test type. %s\n" , case_cfg->test_type_str); +break ; + } +} + +static void +run_test(uint32_t case_id, struct test_configure *case_cfg) { + uint32_t i; + uint32_t nb_lcores = rte_lcore_count(); + struct test_configure_entry *mem_size = &case_cfg->mem_size; + struct test_configure_entry *buf_size = &case_cfg->buf_size; + struct test_configure_entry *ring_size = &case_cfg->ring_size; + struct test_configure_entry *kick_batch = &case_cfg->kick_batch; + struct test_configure_entry dummy = { 0 }; + struct test_configure_entry *var_entry = &dummy; + + for (i = 0; i < RTE_DIM(output_str); i++) + memset(output_str[i], 0, MAX_OUTPUT_STR_LEN); + + if (nb_lcores <= case_cfg->lcore_dma_map.cnt) { + printf("Case %u: Not enough lcores.\n", case_id); + return; + } + + printf("Number of used lcores: %u.\n", nb_lcores); + + if (mem_size->incr != 0) + var_entry = mem_size; + + if (buf_size->incr != 0) + var_entry = buf_size; + + if (ring_size->incr != 0) + var_entry = ring_size; + + if (kick_batch->incr != 0) + var_entry = kick_batch; + + case_cfg->scenario_id = 0; + + output_header(case_id, case_cfg); + + for (var_entry->cur = var_entry->first; var_entry->cur <= +var_entry-Quoted 1 line
last;) {
app/test-dma-perf/main.c new file +21
-0+ case_cfg->scenario_id++; +printf ("\nRunning scenario %d\n" , case_cfg->scenario_id); + +run_test_case (case_cfg); +output_csv (false); + +if (var_entry->op ==OP_ADD ) + var_entry->cur += var_entry->incr; +else if (var_entry->op ==OP_MUL ) + var_entry->cur *= var_entry->incr; +else +break ; + } +} + +static int +parse_lcore (struct test_configure *test_case,const char *value) { +size_t len =strlen (value); +char *input = (char *)malloc ((len +1 ) *sizeof (char )); +strcpy (input, value); + struct lcore_dma_map_t *lcore_dma_map = &(test_case-Quoted 1 line
lcore_dma_map);app/test-dma-perf/main.c new file +49
-0+ +if (test_case ==NULL || value ==NULL ) +return -1 ; + +memset (lcore_dma_map,0 ,sizeof (struct lcore_dma_map_t )); + +char *token =strtok (input,", " ); +while (token !=NULL ) { +if (lcore_dma_map->cnt >=MAX_LCORE_NB ) { +free (input); +return -1 ; + } + +uint16_t lcore_id =atoi (token); + lcore_dma_map->lcores[lcore_dma_map->cnt++] = lcore_id; + + token =strtok (NULL ,", " ); + } + +free (input); +return 0 ; +} + +static int +parse_lcore_dma (struct test_configure *test_case,const char *value) { +struct lcore_dma_map_t *lcore_dma_map; +char *input =strndup (value,strlen (value) +1 ); +char *addrs = input; +char *ptrs[2 ]; +char *start, *end, *substr; +uint16_t lcore_id; +int ret =0 ; + +while (*addrs =='\0' ) + addrs++; +if (*addrs =='\0' ) { +fprintf (stderr,"No input DMA addresses\n" ); + ret =-1 ; + goto out; + } + + substr =strtok (addrs,"," ); +if (substr ==NULL ) { +fprintf (stderr,"No input DMA address\n" ); + ret =-1 ; + goto out; + } + + memset(&test_case->lcore_dma_map, 0, sizeof(struct
lcore_dma_map_t));app/test-dma-perf/main.c new file +15
-0+ + do { + rte_strsplit(substr, strlen(substr), ptrs, 2, '@'); + + start = strstr(ptrs[0], "lcore"); + if (start == NULL) { + fprintf(stderr, "Illegal lcore\n"); + ret = -1; + break; + } + + start += 5; + lcore_id = strtol(start, &end, 0); + if (end == start) { + fprintf(stderr, "No input lcore ID or ID %d iswrong\n",
lcore_id);app/test-dma-perf/main.c new file +7
-0+ ret =-1 ; +break ; + } + + lcore_dma_map = &test_case->lcore_dma_map; + lcore_dma_map->lcores[lcore_dma_map->cnt] = lcore_id; + strcpy(lcore_dma_map->dma_names[lcore_dma_map-Quoted 1 line
cnt], ptrs[1]);
app/test-dma-perf/main.c new file +23
-0+ lcore_dma_map->cnt++; + substr =strtok (NULL ,"," ); + }while (substr !=NULL ); + +out: +free (input); +return ret; +} + +static int +parse_entry (const char *value,struct test_configure_entry *entry) { +char input[255 ] = {0 }; +char *args[MAX_PARAMS_PER_ENTRY ]; +int args_nr =-1 ; + +if (value ==NULL || entry ==NULL ) + goto out; + +strncpy (input, value,254 ); +if (*input =='\0' ) + goto out; + + args_nr = rte_strsplit(input, strlen(input), args,
MAX_PARAMS_PER_ENTRY, ',');app/test-dma-perf/main.c new file +35
-0+if (args_nr !=1 && args_nr !=4 ) + goto out; + + entry->cur = entry->first = (uint32_t )atoi (args[0 ]); + +if (args_nr ==4 ) { + entry->last = (uint32_t )atoi (args[1 ]); + entry->incr = (uint32_t )atoi (args[2 ]); +if (!strcmp (args[3 ],"MUL" )) + entry->op =OP_MUL ; +else if (!strcmp (args[3 ],"ADD" )) + entry->op =OP_ADD ; +else { +printf ("Invalid op %s.\n" , args[3 ]); + args_nr =-1 ; + } + }else { + entry->op =OP_NONE ; + entry->last =0 ; + entry->incr =0 ; + } +out: +return args_nr; +} + +static uint16_t +load_configs (const char *path) +{ +struct rte_cfgfile *cfgfile; +int nb_sections, i; +struct test_configure *test_case; +char section_name[CFG_NAME_LEN ]; +const char *case_type; +const char *lcore_dma; + const char *mem_size_str, *buf_size_str, *ring_size_str,
*kick_batch_str;app/test-dma-perf/main.c new file +13
-0+int args_nr, nb_vp; +bool is_dma; + +printf ("config file parsing...\n" ); + cfgfile =rte_cfgfile_load (path,0 ); +if (!cfgfile) { +printf ("Open configure file error.\n" ); +exit (1 ); + } + + nb_sections =rte_cfgfile_num_sections (cfgfile,NULL ,0 ); +if (nb_sections >MAX_TEST_CASES ) { + printf("Error: The maximum number of cases is %d.\n",
MAX_TEST_CASES);app/test-dma-perf/main.c new file +7
-0+exit (1 ); + } + +for (i =0 ; i < nb_sections; i++) { +snprintf (section_name,CFG_NAME_LEN ,"case%d" , i +1 ); + test_case = &test_cases[i]; + case_type = rte_cfgfile_get_entry(cfgfile, section_name,"type");
app/test-dma-perf/main.c new file +2
-0+if (!case_type) { + printf("Error: No case type in case %d, the test will befinished here.\n",
app/test-dma-perf/main.c new file +7
-0+ i + 1); + test_case->is_valid = false; +continue ; + } + +if (strcmp (case_type,DMA_MEM_COPY ) ==0 ) { + test_case->test_type =
TEST_TYPE_DMA_MEM_COPY;app/test-dma-perf/main.c new file +4
-0+ test_case->test_type_str =DMA_MEM_COPY ; + is_dma = true; + }else if (strcmp (case_type,CPU_MEM_COPY ) ==0 ) { + test_case->test_type =
TEST_TYPE_CPU_MEM_COPY;app/test-dma-perf/main.c new file +4
-0+ test_case->test_type_str =CPU_MEM_COPY ; + is_dma = false; + }else { + printf("Error: Cannot find case type %s in case%d.\n",
case_type, i + 1);app/test-dma-perf/main.c new file +7
-0+ test_case->is_valid = false; +continue ; + } + + nb_vp =0 ; + + test_case->src_numa_node =(int)atoi(rte_cfgfile_get_entry(cfgfile,
+section_name, "src_numa_node"));app/test-dma-perf/main.c new file +1
-0+ test_case->dst_numa_node =(int)atoi(rte_cfgfile_get_entry(cfgfile,
+section_name, "dst_numa_node"));app/test-dma-perf/main.c new file +2
-0+ + mem_size_str = rte_cfgfile_get_entry(cfgfile, section_name,"
mem_size");app/test-dma-perf/main.c new file +1
-0+ args_nr = parse_entry(mem_size_str, &test_case-Quoted 1 line
mem_size);app/test-dma-perf/main.c new file +8
-0+if (args_nr <0 ) { +printf ("parse error in case %d.\n" , i +1 ); + test_case->is_valid = false; +continue ; + }else if (args_nr >1 ) + nb_vp++; + + buf_size_str = rte_cfgfile_get_entry(cfgfile, section_name,"
buf_size");app/test-dma-perf/main.c new file +10
-0+ args_nr =parse_entry (buf_size_str, &test_case->buf_size); +if (args_nr <0 ) { +printf ("parse error in case %d.\n" , i +1 ); + test_case->is_valid = false; +continue ; + }else if (args_nr >1 ) + nb_vp++; + +if (is_dma) { + ring_size_str = rte_cfgfile_get_entry(cfgfile,
section_name,
+"dma_ring_size");app/test-dma-perf/main.c new file +1
-0+ args_nr = parse_entry(ring_size_str, &test_case-Quoted 1 line
ring_size);app/test-dma-perf/main.c new file +8
-0+if (args_nr <0 ) { +printf ("parse error in case %d.\n" , i +1 ); + test_case->is_valid = false; +continue ; + }else if (args_nr >1 ) + nb_vp++; + + kick_batch_str = rte_cfgfile_get_entry(cfgfile,
section_name, "kick_batch");app/test-dma-perf/main.c new file +1
-0+ args_nr = parse_entry(kick_batch_str, &test_case-Quoted 1 line
kick_batch);app/test-dma-perf/main.c new file +8
-0+if (args_nr <0 ) { +printf ("parse error in case %d.\n" , i +1 ); + test_case->is_valid = false; +continue ; + }else if (args_nr >1 ) + nb_vp++; + + lcore_dma = rte_cfgfile_get_entry(cfgfile,
section_name, "lcore_dma");app/test-dma-perf/main.c new file +1
-0+ int lcore_ret = parse_lcore_dma(test_case,
lcore_dma);app/test-dma-perf/main.c
Quoted 2 lines
app/test-dma-perf/main.c new file +2
-0+if (lcore_ret <0 ) { + printf("parse lcore dma error in case %d.\n", i1);
Quoted 117 lines
app/test-dma-perf/main.c new file +5
-0+ test_case->is_valid = false; +continue ; + } + }else { + lcore_dma = rte_cfgfile_get_entry(cfgfile,
section_name, "lcore");app/test-dma-perf/main.c new file +10
-0+int lcore_ret =parse_lcore (test_case, lcore_dma); +if (lcore_ret <0 ) { +printf ("parse lcore error in case %d.\n" , i +1 ); + test_case->is_valid = false; +continue ; + } + } + +if (nb_vp >1 ) { + printf("Error, each section can only have a singlevariable parameter.\n");
app/test-dma-perf/main.c new file +6
-0+ test_case->is_valid = false; +continue ; + } + + test_case->cache_flush = + (int)atoi(rte_cfgfile_get_entry(cfgfile, section_name,"
cache_flush"));app/test-dma-perf/main.c new file +1
-0+ test_case->test_secs =(uint16_t)atoi(rte_cfgfile_get_entry(cfgfile,
app/test-dma-perf/main.c new file +3
-0+ section_name, "test_seconds")); + + test_case->eal_args = rte_cfgfile_get_entry(cfgfile,
section_name, "eal_args");app/test-dma-perf/main.c new file +19
-0+ test_case->is_valid = true; + } + +rte_cfgfile_close (cfgfile); +printf ("config file parsing complete.\n\n" ); +return i; +} + +/* Parse the argument given in the command line of the application +*/ static int append_eal_args (int argc,char **argv,const char +*eal_args,char **new_argv) { +int i; +char *tokens[MAX_EAL_PARAM_NB ]; +char args[MAX_EAL_PARAM_LEN ] = {0 }; +int token_nb, new_argc =0 ; + +for (i =0 ; i < argc; i++) { + if ((strcmp(argv[i], CMDLINE_CONFIG_ARG) == 0) || + (strcmp(argv[i], CMDLINE_RESULT_ARG) ==0)) {
app/test-dma-perf/main.c new file +4
-0+ i++; +continue ; + } + strlcpy(new_argv[new_argc], argv[i],
sizeof(new_argv[new_argc]));app/test-dma-perf/main.c new file +37
-0+ new_argc++; + } + +if (eal_args) { +strlcpy (args, eal_args,sizeof (args)); + token_nb =rte_strsplit (args,strlen (args), + tokens,MAX_EAL_PARAM_NB ,' ' ); +for (i =0 ; i < token_nb; i++) +strcpy (new_argv[new_argc++], tokens[i]); + } + +return new_argc; +} + +int +main (int argc,char *argv[]) +{ +int ret; +uint16_t case_nb; +uint32_t i, nb_lcores; +pid_t cpid, wpid; +int wstatus; +char args[MAX_EAL_PARAM_NB ][MAX_EAL_PARAM_LEN ]; +char *pargs[MAX_EAL_PARAM_NB ]; +char *cfg_path_ptr =NULL ; +char *rst_path_ptr =NULL ; +char rst_path[PATH_MAX ]; +int new_argc; +bool is_first_case = true; + +memset (args,0 ,sizeof (args)); + +for (i =0 ; i <RTE_DIM (pargs); i++) + pargs[i] = args[i]; + +for (i =0 ; i < (uint32_t )argc; i++) { + if (strncmp(argv[i], CMDLINE_CONFIG_ARG,
MAX_LONG_OPT_SZ) == 0)app/test-dma-perf/main.c new file +2
-0+ cfg_path_ptr = argv[i +1 ]; + if (strncmp(argv[i], CMDLINE_RESULT_ARG,
MAX_LONG_OPT_SZ) == 0)app/test-dma-perf/main.c new file +21
-0+ rst_path_ptr = argv[i +1 ]; + } +if (cfg_path_ptr ==NULL ) { +printf ("Config file not assigned.\n" ); +return -1 ; + } +if (rst_path_ptr ==NULL ) { +strlcpy (rst_path, cfg_path_ptr,PATH_MAX ); +strcat (strtok (basename (rst_path),"." ),"_result.csv" ); + rst_path_ptr = rst_path; + } + + case_nb =load_configs (cfg_path_ptr); + fd =fopen (rst_path_ptr,"w" ); +if (fd ==NULL ) { +printf ("Open output CSV file error.\n" ); +return -1 ; + } +fclose (fd); + +for (i =0 ; i < case_nb; i++) {app/test-dma-perf/main.c line 505
Quoted 10 lines
app/test-dma-perf/main.c new file +10
-0+if (test_cases[i].test_type ==TEST_TYPE_NONE ) { +printf ("No test type in test case %d.\n\n" , i +1 ); +continue ; + } +if (!test_cases[i].is_valid) { +printf ("Invalid test case %d.\n\n" , i +1 ); +continue ; + } + + cpid =fork ();[Anoob] Do we really need
fork()? Can't we use code like,Quoted 5 lines
RTE_LCORE_FOREACH_WORKER(lcore_id) { ret |= rte_eal_wait_lcore(lcore_id); }to wait for all threads to exit?
[Cheng] Good question.
Fork()is used here to establish a new process for the new test case. In order for each test case to have a new EAL environment (for the flexibility), the EAL must be reinitialized for each case. However, the EAL parameters can only be initialized once per process. Therefore, we use a new process to run each new test case. Moreover, each test case runs sequentially and does not affect the others, ensuring the accuracy of the performance data. Your code would wait for all threads to exit in the same process. However, it would not provide a "clean" environment for each test case likefork()does.Fork()allows us to have a fully reinitialized environment, with no impact or side effects from previous test cases. This results in clean, precise performance data for each case.
Quoted 2 lines
Please let me know your thoughts on this. And please let me know if you have any other questions or require any clarification.
[Anoob] This was just a generic observation. I do not have a strong opinion either way.
Quoted 171 lines
Thanks,
ChengQuoted 167 lines
Quoted 167 lines
app/test-dma-perf/main.c new file +7
-0+if (cpid <0 ) { +printf ("Fork case %d failed.\n" , i +1 ); +exit (EXIT_FAILURE ); + }else if (cpid ==0 ) { +printf ("\nRunning case %u\n\n" , i +1 ); + + new_argc = append_eal_args(argc, argv,test_cases[i].eal_args, pargs);
app/test-dma-perf/main.c new file +3
-0+ ret =rte_eal_init (new_argc, pargs); +if (ret <0 ) + rte_exit(EXIT_FAILURE, "Invalid EALarguments\n");
app/test-dma-perf/main.c new file +6
-0+ +/* Check lcores. */ + nb_lcores =rte_lcore_count (); +if (nb_lcores <2 ) + rte_exit(EXIT_FAILURE, + "There should be at least 2 workerlcores.\n");
app/test-dma-perf/main.c new file +36
-0+ + fd =fopen (rst_path_ptr,"a" ); +if (!fd) { +printf ("Open output CSV file error.\n" ); +return 0 ; + } + +if (is_first_case) { +output_env_info (); + is_first_case = false; + } +run_test (i +1 , &test_cases[i]); + +/* clean up the EAL */ +rte_eal_cleanup (); + +fclose (fd); + +printf ("\nCase %u completed.\n\n" , i +1 ); + +exit (EXIT_SUCCESS ); + }else { + wpid =waitpid (cpid, &wstatus,0 ); +if (wpid ==-1 ) { +printf ("waitpid error.\n" ); +exit (EXIT_FAILURE ); + } + +if (WIFEXITED (wstatus)) +printf ("Case process exited. status %d\n\n" , +WEXITSTATUS (wstatus)); +else if (WIFSIGNALED (wstatus)) +printf ("Case process killed by signal %d\n\n" , +WTERMSIG (wstatus)); +else if (WIFSTOPPED (wstatus)) + printf("Case process stopped by signal%d\n\n",
app/test-dma-perf/main.c new file +5
-0+ WSTOPSIG(wstatus)); + else if (WIFCONTINUED(wstatus)) + printf("Case process continued.\n\n"); + else + printf("Case process unknownterminated.\n\n");
app/test-dma-perf/main.c new file +7
-0+ } + } + +printf ("Bye...\n" ); +return 0 ; +} +app/test-dma-perf/main.h new renamed from app/test-dma-perf/main.h +0
-0file mode 100644 index 0000000000..215ac42673
app/test-dma-perf/main.h new file +68
-0@@ -0,0 +1,69 @@ +/* SPDX-License-Identifier: BSD-3-Clause +* Copyright(c) 2023 Intel Corporation */ + +#ifndef _MAIN_H_ +#define _MAIN_H_ + + +#include <rte_common.h> +#include <rte_cycles.h> +#include <rte_dev.h> +#include <rte_dmadev.h> + +#ifndef __maybe_unused +#define __maybe_unused __rte_unused +#endif + +#define MAX_WORKER_NB 128 +#define MAX_OUTPUT_STR_LEN 512 + +#define MAX_DMA_NB 128 +#define MAX_LCORE_NB 256 + +extern char output_str[MAX_WORKER_NB ][MAX_OUTPUT_STR_LEN ]; + +typedef enum { +OP_NONE =0 , +OP_ADD , +OP_MUL +}alg_op_type ; + +struct test_configure_entry { +uint32_t first; +uint32_t last; +uint32_t incr; +alg_op_type op; +uint32_t cur; +}; + +struct lcore_dma_map_t { +uint32_t lcores[MAX_WORKER_NB ]; +char dma_names[MAX_WORKER_NB ][RTE_DEV_NAME_MAX_LEN ]; +int16_t dma_ids[MAX_WORKER_NB ]; +uint16_t cnt; +}; + +struct test_configure { +bool is_valid; +uint8_t test_type; +const char *test_type_str; +uint16_t src_numa_node; +uint16_t dst_numa_node; +uint16_t opcode; +bool is_dma; +struct lcore_dma_map_t lcore_dma_map; +struct test_configure_entry mem_size; +struct test_configure_entry buf_size; +struct test_configure_entry ring_size; +struct test_configure_entry kick_batch; +uint32_t cache_flush; +uint32_t nr_buf; +uint16_t test_secs; +const char *eal_args; +uint8_t scenario_id; +}; + +void mem_copy_benchmark (struct test_configure *cfg,bool is_dma); + +#endif /* _MAIN_H_ */ app/test-dma- renamed from app/test-dma-perf/meson.build +0
-0perf/meson.build new file mode 100644 index 0000000000..bd6c264002
app/test-dma-perf/meson.build new file +17
-1@@ -0,0 +1,17 @@ +# SPDX-License-Identifier: BSD-3-Clause # Copyright(c) 2019-2023 +Intel Corporation + +# meson file, for building this app as part of a main DPDK build. + +if is_windows + build = false + reason = 'not supported on Windows' + subdir_done() +endif + +deps += ['dmadev', 'mbuf', 'cfgfile'] + +sources = files( + 'main.c', + 'benchmark.c', +) --2.40.1
Hi Anoob,
Replies are inline.
Thanks,
Cheng
Quoted 529 lines
-----Original Message-----
From: Anoob Joseph <[email protected]>
Sent: Thursday, June 15, 2023 4:45 PM
To: Jiang, Cheng1 <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>; Amit Prakash Shukla <[email protected]>; Ma, WenwuX <[email protected]>; Wang, YuanX <[email protected]>; He, Xingguang <[email protected]>
Subject: RE: [EXT] [PATCH v6] app/dma-perf: introduce dma-perf applicationHi Cheng,
Please see inline.
Thanks,
AnoobQuoted 9 lines
-----Original Message-----
From: Jiang, Cheng1 <[email protected]>
Sent: Thursday, June 15, 2023 1:31 PM
To: Anoob Joseph <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>; Amit Prakash Shukla <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; Ma, WenwuX <[email protected]>; Wang,YuanX
Quoted 4 lines
<[email protected]>; He, Xingguang <[email protected]>
Subject: RE: [EXT] [PATCH v6] app/dma-perf: introduce dma-perf applicationQuoted 13 lines
Hi,
Thanks for your comments, the replies are inline.
Thanks,
ChengQuoted 6 lines
-----Original Message-----
From: Anoob Joseph <[email protected]>
Sent: Thursday, June 15, 2023 1:22 PM
To: Jiang, Cheng1 <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>;Amit
Quoted 11 lines
Quoted 3 lines
Prakash Shukla <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; Ma, WenwuX <[email protected]>; Wang,YuanX
Quoted 7 lines
<[email protected]>; He, Xingguang <[email protected]>
Subject: RE: [EXT] [PATCH v6] app/dma-perf: introduce dma-perf applicationHi,
Thanks for working on the comments. Few more top level comment
inline.
Quoted 202 lines
Quoted 9 lines
Thanks,
AnoobQuoted 5 lines
-----Original Message-----
From: Cheng Jiang <[email protected]>
Sent: Tuesday, June 13, 2023 10:02 AM
To: [email protected]; [email protected]; [email protected]; [email protected]; Amit PrakashShukla
Quoted 3 lines
Quoted 3 lines
<[email protected]>; Anoob Joseph <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected];Quoted 19 lines
Quoted 6 lines
Cheng Jiang <[email protected]>
Subject: [EXT] [PATCH v6] app/dma-perf: introduce dma-perf applicationExternal Email
Quoted 13 lines
------------------------------------------------------------------
--
-- There are many high-performance DMA devices supported in DPDK now, and these DMA devices can also be integrated into other modules of DPDK as accelerators, such as Vhost. Before integrating DMA into applications, developers need to know the performance of these DMA devices in various scenarios and the performance of CPUs in the same scenario, such as different buffer lengths. Only in this way can we know the target performance of the application accelerated by using them. This patch introduces a high-performance testing tool, which supports comparing the performance of CPU and DMA in different scenarios automatically with a pre- set config file. Memory Copy performance test areQuoted 26 lines
supported for now.
Quoted 25 lines
Signed-off-by Cheng Jiang <[email protected]>Signed-off-by Jiayu Hu <[email protected]>Signed-off-by Yuan Wang <[email protected]>Acked-by Morten Brørup <[email protected]>Acked-by Chenbo Xia <[email protected]>---
v6:
improved code based on Anoob's comments; fixed some code structure issues;
v5:
fixed someLONG_LINEwarnings;
v4:
fixed inaccuracy of the memory footprint display;
v3:
fixed some typos;
v2:
added lcore/dmadev designation;
added error case process;
removedworker_threadsparameter from config.ini; improved the logs;
improved config file;app/meson.build | 1 + app/test-dma-perf/benchmark.c | 477++++++++++++++++++++++++++++
Quoted 42 lines
Quoted 42 lines
app/test-dma-perf/config.ini | 59 ++++
app/test-dma-perf/main.c | 569 ++++++++++++++++++++++++++++++++++ app/test-dma-perf/main.h | 69 +++++ app/test-dma-perf/meson.build | 17 + 6 files changed, 1192 insertions(+) create mode 100644
app/test-dma-perf/benchmark.ccreate mode 100644 app/test-dma-perf/config.ini create mode 100644 app/test-dma-perf/main.ccreate mode 100644app/test-dma-perf/main.hcreate mode
100644 app/test-dma-perf/meson.buildapp/meson.build index renamed from app/meson.build +0
-0
74d2420f67..4fc1a83eba100644app/meson.build +1
-0@@ -19,6 +19,7 @@ apps = [ 'test-cmdline', 'test-compress-perf', 'test-crypto-perf', + 'test-dma-perf', 'test-eventdev', 'test-fib', 'test-flow-perf',app/test-dma- renamed from app/test-dma-perf/benchmark.c +0
-0
perf/benchmark.cnew file mode 100644 index 0000000000..bc1ca82297app/test-dma-perf/benchmark.c new file +13
-0@@ -0,0 +1,477 @@ +/* SPDX-License-Identifier: BSD-3-Clause +* Copyright(c) 2023 Intel Corporation */ + +#include <inttypes.h> +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> + +#include <rte_time.h> +#include <rte_mbuf.h> +#include <rte_dmadev.h> +#include <rte_malloc.h> +#include <rte_lcore.h> Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+ +#include "main.h" + +#define MAX_DMA_CPL_NB 255 + +#define TEST_WAIT_U_SECOND 10000 + +#define CSV_LINE_DMA_FMT "Scenario %u,%u,%s,%u,%u,%.2lf,%"PRIu64
Quoted 2 lines
Quoted 2 lines
",%.3lf,%.3lf\n"
+#defineCSV_LINE_CPU_FMT"Scenario %u,%u,NA,%u,%u,%.2lf,%"PRIu64
Quoted 21 lines
Quoted 21 lines
",%.3lf,%.3lf\n"
app/test-dma-perf/benchmark.c new file +20
-0+ +struct worker_info { +bool ready_flag; +bool start_flag; +bool stop_flag; +uint32_t total_cpl; +uint32_t test_cpl; +}; + +struct lcore_params { +uint8_t scenario_id; +unsigned int lcore_id; +char *dma_name; +uint16_t worker_id; +uint16_t dev_id; +uint32_t nr_buf; +uint16_t kick_batch; +uint32_t buf_size; +uint16_t test_secs; +struct rte_mbuf **srcs;Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+struct rte_mbuf **dsts; +struct worker_info worker_info; +}; + +static struct rte_mempool *src_pool;static struct rte_mempool +*dst_pool; + +static volatile struct lcore_params*worker_params[MAX_WORKER_NB];
Quoted 58 lines
Quoted 58 lines
app/test-dma-perf/benchmark.c new file +20
-0+ +#define PRINT_ERR (...) print_err(__func__, __LINE__, __VA_ARGS__) + +static inline int +__rte_format_printf(3, 4) +print_err(const char *func, int lineno, const char *format, ...) { + va_list ap; + int ret; + + ret = fprintf(stderr, "In %s:%d - ", func, lineno); + va_start(ap, format); + ret += vfprintf(stderr, format, ap); + va_end(ap); + + return ret; +} + +static inline void +calc_result(uint32_t buf_size, uint32_t nr_buf, uint16_t +nb_workers,
uint16_ttest_secs,app/test-dma-perf/benchmark.c new file +1
-0+ uint32_t total_cnt, float *memory, uint32_t
*ave_cycle,app/test-dma-perf/benchmark.c new file +3
-0+ float *bandwidth, float *mops) { + *memory = (float)(buf_size * (nr_buf / nb_workers) * 2) / (1024 +*1024);
app/test-dma-perf/benchmark.c new file +2
-0+ *ave_cycle = test_secs *rte_get_timer_hz () / total_cnt; + *bandwidth = (buf_size * 8 * (rte_get_timer_hz() /(float)*ave_cycle)) / 1000000000;
app/test-dma-perf/benchmark.c new file +5
-0+ *mops = (float )rte_get_timer_hz () / *ave_cycle /1000000 ; } + +static void +output_result(uint8_t scenario_id, uint32_t lcore_id, char +*dma_name,
uint64_tave_cycle,app/test-dma-perf/benchmark.c new file +9
-0+ uint32_t buf_size, uint32_t nr_buf, float memory, + float bandwidth, float mops, bool is_dma) { + if (is_dma) + printf("lcore %u, DMA %s:\n", lcore_id, dma_name); + else + printf("lcore %u\n", lcore_id); + + printf("average cycles/op: %" PRIu64 ", buffer size: %u, nr_buf: +%u,memory: %.2lfMB, frequency: %" PRIu64 ".\n",
app/test-dma-perf/benchmark.c new file +1
-0+ ave_cycle, buf_size, nr_buf, memory,
rte_get_timer_hz());app/test-dma-perf/benchmark.c new file +5
-0+printf ("Average bandwidth: %.3lfGbps, MOps: %.3lf\n" , bandwidth, +mops); + +if (is_dma) + snprintf(output_str[lcore_id], MAX_OUTPUT_STR_LEN,
CSV_LINE_DMA_FMT,app/test-dma-perf/benchmark.c new file +4
-0+ scenario_id, lcore_id, dma_name, buf_size, + nr_buf, memory, ave_cycle, bandwidth, mops); + else + snprintf(output_str[lcore_id], MAX_OUTPUT_STR_LEN,Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
CSV_LINE_CPU_FMT,app/test-dma-perf/benchmark.c new file +7
-0+ scenario_id, lcore_id, buf_size, + nr_buf, memory, ave_cycle, bandwidth, mops); } + +static inline void +cache_flush_buf (__maybe_unused struct rte_mbuf **array, +__maybe_unused uint32_t buf_size, + __maybe_unused uint32_t nr_buf) { #ifdef
RTE_ARCH_X86_64Quoted 253 lines
Quoted 71 lines
Quoted 71 lines
app/test-dma-perf/benchmark.c new file +32
-0+char *data; +struct rte_mbuf **srcs = array; +uint32_t i, offset; + +for (i =0 ; i < nr_buf; i++) { + data = rte_pktmbuf_mtod(srcs[i], char *); +for (offset =0 ; offset < buf_size; offset +=64 ) +__builtin_ia32_clflush (data + offset); + } +#endif +} + +/* Configuration of device. */ +static void +configure_dmadev_queue (uint32_t dev_id,uint32_t ring_size) { +uint16_t vchan =0 ; +struct rte_dma_info info; +struct rte_dma_conf dev_config = { .nb_vchans =1 }; +struct rte_dma_vchan_conf qconf = { + .direction =RTE_DMA_DIR_MEM_TO_MEM , + .nb_desc = ring_size + }; + +if (rte_dma_configure (dev_id, &dev_config) !=0 ) +rte_exit (EXIT_FAILURE ,"Error with dma configure.\n" ); + +if (rte_dma_vchan_setup (dev_id, vchan, &qconf) !=0 ) +rte_exit (EXIT_FAILURE ,"Error with queue configuration.\n" ); + +rte_dma_info_get (dev_id, &info); +if (info.nb_vchans !=1 ) + rte_exit(EXIT_FAILURE, "Error, no configured queuesreported on device id. %u\n",
app/test-dma-perf/benchmark.c new file +20
-0+ dev_id); + +if (rte_dma_start (dev_id) !=0 ) +rte_exit (EXIT_FAILURE ,"Error with dma start.\n" ); } + +static int +config_dmadevs (struct test_configure *cfg) { +uint32_t ring_size = cfg->ring_size.cur; +struct lcore_dma_map_t *ldm = &cfg->lcore_dma_map; +uint32_t nb_workers = ldm->cnt; +uint32_t i; +int dev_id; +uint16_t nb_dmadevs =0 ; +char *dma_name; + +for (i =0 ; i < ldm->cnt; i++) { + dma_name = ldm->dma_names[i]; + dev_id =rte_dma_get_dev_id_by_name (dma_name); +if (dev_id ==-1 ) { + fprintf(stderr, "Error: Fail to find DMA %s.\n",
dma_name);app/test-dma-perf/benchmark.c new file +11
-0+ goto end; + } + + ldm->dma_ids[i] = dev_id; +configure_dmadev_queue (dev_id, ring_size); + ++nb_dmadevs; + } + +end: +if (nb_dmadevs < nb_workers) { + printf("Not enough dmadevs (%u) for all workers (%u).\n",
nb_dmadevs,nb_workers);app/test-dma-perf/benchmark.c new file +5
-0+return -1 ; + } + +printf ("Number of used dmadevs: %u.\n" , nb_dmadevs); +Quoted 8 lines
app/test-dma-perf/benchmark.c line 187
Quoted 6 lines
app/test-dma-perf/benchmark.c new file +6
-0+return 0 ; +} + +#define POLL_MAX 1000 + +[Anoob] Extra blank line. You can consider removing.
[Cheng] sure, sorry for the miss.
Quoted 111 lines
Quoted 111 lines
app/test-dma-perf/benchmark.c new file +14
-0+static inline void +do_dma_submit_and_poll (uint16_t dev_id,uint64_t *async_cnt, +volatile struct worker_info *worker_info) { +int ret; +uint16_t nr_cpl; + + ret =rte_dma_submit (dev_id,0 ); +if (ret <0 ) { +rte_dma_stop (dev_id); +rte_dma_close (dev_id); +rte_exit (EXIT_FAILURE ,"Error with dma submit.\n" ); + } + + nr_cpl = rte_dma_completed(dev_id, 0, MAX_DMA_CPL_NB, NULL,NULL);
app/test-dma-perf/benchmark.c new file +31
-0+ *async_cnt -= nr_cpl; + worker_info->total_cpl += nr_cpl; } + +static inline int +do_dma_mem_copy (void *p) +{ +const uint16_t *para_idx = (uint16_t *)p; +volatile struct lcore_params *para = worker_params[*para_idx]; +volatile struct worker_info *worker_info = &(para->worker_info); +const uint16_t dev_id = para->dev_id; +const uint32_t nr_buf = para->nr_buf; +const uint16_t kick_batch = para->kick_batch; +const uint32_t buf_size = para->buf_size; +struct rte_mbuf **srcs = para->srcs; +struct rte_mbuf **dsts = para->dsts; +uint16_t nr_cpl; +uint64_t async_cnt =0 ; +uint32_t i; +uint32_t poll_cnt =0 ; +int ret; + + worker_info->stop_flag = false; + worker_info->ready_flag = true; + +while (!worker_info->start_flag) + ; + +while (1 ) { +for (i =0 ; i < nr_buf; i++) { +dma_copy: + ret = rte_dma_copy(dev_id, 0,
rte_pktmbuf_iova(srcs[i]),app/test-dma-perf/benchmark.c new file +4
-0+ rte_pktmbuf_iova(dsts[i]), buf_size, 0); +if (unlikely (ret <0 )) { +if (ret == -ENOSPC ) { + do_dma_submit_and_poll(dev_id,
&async_cnt,worker_info);app/test-dma-perf/benchmark.c new file +5
-0+ goto dma_copy; + }else { +/* Error exit */ +rte_dma_stop (dev_id); + rte_exit(EXIT_FAILURE, "DMAenqueue failed\n");
app/test-dma-perf/benchmark.c new file +6
-0+ } + } + async_cnt++; + +if ((async_cnt % kick_batch) ==0 ) + do_dma_submit_and_poll(dev_id,
&async_cnt,worker_info);app/test-dma-perf/benchmark.c new file +9
-0+ } + +if (worker_info->stop_flag) +break ; + } + +rte_dma_submit (dev_id,0 ); +while ((async_cnt >0 ) && (poll_cnt++ <POLL_MAX )) { + nr_cpl = rte_dma_completed(dev_id, 0,
MAX_DMA_CPL_NB, NULL, NULL);app/test-dma-perf/benchmark.c new file +28
-0+ async_cnt -= nr_cpl; + } + +return 0 ; +} + +static inline int +do_cpu_mem_copy (void *p) +{ +const uint16_t *para_idx = (uint16_t *)p; +volatile struct lcore_params *para = worker_params[*para_idx]; +volatile struct worker_info *worker_info = &(para->worker_info); +const uint32_t nr_buf = para->nr_buf; +const uint32_t buf_size = para->buf_size; +struct rte_mbuf **srcs = para->srcs; +struct rte_mbuf **dsts = para->dsts; +uint32_t i; + + worker_info->stop_flag = false; + worker_info->ready_flag = true; + +while (!worker_info->start_flag) + ; + +while (1 ) { +for (i =0 ; i < nr_buf; i++) { +/* copy buffer form src to dst */ + rte_memcpy((void*)(uintptr_t)rte_mbuf_data_iova(dsts[i]),
app/test-dma-perf/benchmark.c new file +1
-0+ (void*)(uintptr_t)rte_mbuf_data_iova(srcs[i]),
app/test-dma-perf/benchmark.c new file +4
-0+ (size_t)buf_size); + worker_info->total_cpl++; + } +if (worker_info->stop_flag)Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+break ; + } + +return 0 ; +} + +static int +setup_memory_env(struct test_configure *cfg, struct rte_mbuf***srcs,
Quoted 23 lines
Quoted 23 lines
app/test-dma-perf/benchmark.c new file +10
-0+ struct rte_mbuf ***dsts) +{ +unsigned int buf_size = cfg->buf_size.cur; +unsigned int nr_sockets; +uint32_t nr_buf = cfg->nr_buf; + + nr_sockets =rte_socket_count (); +if (cfg->src_numa_node >= nr_sockets || + cfg->dst_numa_node >= nr_sockets) { + printf("Error: Source or destination numa exceeds the acturenuma nodes.\n");
app/test-dma-perf/benchmark.c new file +12
-0+return -1 ; + } + + src_pool =rte_pktmbuf_pool_create ("Benchmark_DMA_SRC" , + nr_buf,/* n == num elements */ +64 ,/* cache size */ +0 ,/* priv size */ + buf_size +RTE_PKTMBUF_HEADROOM , + cfg->src_numa_node); +if (src_pool ==NULL ) { +PRINT_ERR ("Error with source mempool creation.\n" ); +return -1 ;Quoted 8 lines
app/test-dma-perf/benchmark.c line 326
Quoted 5 lines
app/test-dma-perf/benchmark.c new file +5
-0+ } + + dst_pool = rte_pktmbuf_pool_create("Benchmark_DMA_DST", + nr_buf, /* n == num elements */ + 64, /* cache size */[Anoob] We do not alloc or free pointers in the datapath, right? So why bother with cache?
[Cheng] Yes, you are right, the cache size is not necessary here, I'll fix it in the next version.
Quoted 16 lines
Quoted 4 lines
app/test-dma-perf/benchmark.c new file +3
-0+0 ,/* priv size */ + buf_size +RTE_PKTMBUF_HEADROOM , + cfg->dst_numa_node);app/test-dma-perf/benchmark.c line 339
Quoted 10 lines
app/test-dma-perf/benchmark.c new file +10
-0+if (dst_pool ==NULL ) { +PRINT_ERR ("Error with destination mempool creation.\n" ); +return -1 ; + } + + *srcs =rte_malloc (NULL , nr_buf *sizeof (struct rte_mbuf *),0 ); +if (*srcs ==NULL ) { +printf ("Error: srcs malloc failed.\n" ); +return -1 ; + }[Anoob] Are we freeing these memory? The ones allocated with
Quoted 4 lines
rte_malloc.[Cheng] yes, we freed the memory in the end of
mem_copy_benchmark()when we finished the test.
Quoted 8 lines
[Anoob] I think we are not freeing this mem. In the place where we free all mem, we do free all objects to mempool as well as the mempools. But this memory is to hold the pointers, right? Is that getting freed anywhere?
Also, in the mem clearing paths, do we need to clear the static variables (ie, set srcs,
src_pool, dsts,dst_poolto NULL) so that there won't be any scope for any double free.
[Cheng] My apologies for the misunderstanding earlier. I now understand your point that you are right, the memory used to store the pointers is not being freed. I will fix this issue in the next version. Regarding the static variables you mentioned, I agree with your view that they should be cleared. I will address this in the upcoming version as well. Thank you very much for the feedback. It is greatly appreciated.
In addition, I think we also need to nullify these variables when initializing them to ensure safety and standardization of use. What do you think?
Thanks!
Quoted 260 lines
Quoted 186 lines
Quoted 60 lines
Quoted 60 lines
app/test-dma-perf/benchmark.c new file +30
-0+ + *dsts =rte_malloc (NULL , nr_buf *sizeof (struct rte_mbuf *),0 ); +if (*dsts ==NULL ) { +printf ("Error: dsts malloc failed.\n" ); +return -1 ; + } + +if (rte_mempool_get_bulk (src_pool, (void **)*srcs, nr_buf) !=0 ) { +printf ("get src mbufs failed.\n" ); +return -1 ; + } +if (rte_mempool_get_bulk (dst_pool, (void **)*dsts, nr_buf) !=0 ) { +printf ("get dst mbufs failed.\n" ); +return -1 ; + } + +return 0 ; +} + +void +mem_copy_benchmark (struct test_configure *cfg,bool is_dma) { +uint16_t i; +uint32_t offset; +unsigned int lcore_id =0 ; +struct rte_mbuf **srcs =NULL , **dsts =NULL ; +struct lcore_dma_map_t *ldm = &cfg->lcore_dma_map; +unsigned int buf_size = cfg->buf_size.cur; +uint16_t kick_batch = cfg->kick_batch.cur; +uint32_t nr_buf = cfg->nr_buf = (cfg->mem_size.cur *1024 * +1024) /(
cfg->buf_size.cur* 2);app/test-dma-perf/benchmark.c new file +27
-0+uint16_t nb_workers = ldm->cnt; +uint16_t test_secs = cfg->test_secs; +float memory; +uint32_t avg_cycles =0 ; +float mops; +float bandwidth; + +if (setup_memory_env (cfg, &srcs, &dsts) <0 ) + goto out; + +if (is_dma) +if (config_dmadevs (cfg) <0 ) + goto out; + +if (cfg->cache_flush) { +cache_flush_buf (srcs, buf_size, nr_buf); +cache_flush_buf (dsts, buf_size, nr_buf); +rte_mb (); + } + +printf ("Start testing....\n" ); + +for (i =0 ; i < nb_workers; i++) { + lcore_id = ldm->lcores[i]; + offset = nr_buf / nb_workers * i; + + worker_params[i] = rte_malloc(NULL, sizeof(structQuoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 6 lines
lcore_params), 0);app/test-dma-perf/benchmark.c new file +2
-0+if (!worker_params[i]) { + printf("lcore parameters malloc failure for lcore%d\n",
lcore_id);app/test-dma-perf/benchmark.c new file +2
-0+break ; + }[Anoob] Are we freeing the above memory?
[Cheng] sorry, I missed that, I'll add
worker_paramsmemory free in the next version, thanks.Quoted 90 lines
Quoted 90 lines
app/test-dma-perf/benchmark.c new file +2
-0+if (is_dma) { + worker_params[i]->dma_name = ldm-Quoted 1 line
dma_names[i];
app/test-dma-perf/benchmark.c new file +5
-0+ worker_params[i]->dev_id = ldm->dma_ids[i]; + worker_params[i]->kick_batch = kick_batch; + } + worker_params[i]->worker_id = i; + worker_params[i]->nr_buf = (uint32_t)(nr_buf /
nb_workers);app/test-dma-perf/benchmark.c new file +9
-0+ worker_params[i]->buf_size = buf_size; + worker_params[i]->test_secs = test_secs; + worker_params[i]->srcs = srcs + offset; + worker_params[i]->dsts = dsts + offset; + worker_params[i]->scenario_id = cfg->scenario_id; + worker_params[i]->lcore_id = lcore_id; + +if (is_dma) + rte_eal_remote_launch(do_dma_mem_copy, (void*)(&i),
lcore_id);app/test-dma-perf/benchmark.c new file +2
-0+else + rte_eal_remote_launch(do_cpu_mem_copy, (void*)(&i),
lcore_id);app/test-dma-perf/benchmark.c new file +6
-0+ } + +while (1 ) { +bool ready = true; +for (i =0 ; i < nb_workers; i++) { + if (worker_params[i]->worker_info.ready_flag ==false) {
app/test-dma-perf/benchmark.c new file +19
-0+ ready =0 ; +break ; + } + } +if (ready) +break ; + } + +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.start_flag = true; + +usleep (TEST_WAIT_U_SECOND ); +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.test_cpl = +worker_params[i]->worker_info.total_cpl; + +usleep (test_secs *1000 *1000 ); +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.test_cpl =worker_params[i]->worker_info.total_cpl -
app/test-dma-perf/benchmark.c new file +1
-0+ worker_params[i]-Quoted 1 line
worker_info.test_cpl;
app/test-dma-perf/benchmark.c new file +12
-0+ +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.stop_flag = true; + +rte_eal_mp_wait_lcore (); + +for (i =0 ; i < nb_workers; i++) { +calc_result (buf_size, nr_buf, nb_workers, test_secs, + worker_params[i]->worker_info.test_cpl, + &memory, &avg_cycles, &bandwidth, &mops); + output_result(cfg->scenario_id, worker_params[i]->lcore_id, + worker_params[i]->dma_name,
avg_cycles,buf_size,app/test-dma-perf/benchmark.c new file +1
-0+ nr_buf / nb_workers, memory,bandwidth, mops,
is_dma);app/test-dma-perf/benchmark.c new file +21
-0+ } + +out: +/* free env */ +if (srcs) +rte_pktmbuf_free_bulk (srcs, nr_buf); +if (dsts) +rte_pktmbuf_free_bulk (dsts, nr_buf); + +if (src_pool) +rte_mempool_free (src_pool); +if (dst_pool) +rte_mempool_free (dst_pool); + +if (is_dma) { +for (i =0 ; i < nb_workers; i++) { +printf ("Stopping dmadev %d\n" , ldm->dma_ids[i]); +rte_dma_stop (ldm->dma_ids[i]); + } + } +}(quoted diff) +0
-0b/app/test-dma-perf/config.ini new file mode 100644 index
Quoted 8 lines
app/test-dma-perf/config.ini line 4
Quoted 8 lines
0000000000..2fd9c3c387
app/test-dma-perf/config.ini new file +4
-0@@ -0,0 +1,59 @@ + +; This is an example configuration file for dma-perf, which +details the meanings of each parameter ; and instructions on how +to use dma-perf.
Quoted 15 lines
Quoted 15 lines
app/test-dma-perf/config.ini new file +8
-0+ +; Supported test types are DMA_MEM_COPY and CPU_MEM_COPY. + +; Parameters: +; "mem_size" denotes the size of the memory footprint. +; "buf_size" denotes the memory size of a single operation. +; "dma_ring_size" denotes the dma ring buffer size. It should be +greaterthan 64 normally.
app/test-dma-perf/config.ini new file +2
-0+; "kick_batch" denotes the dma operation batch size, and should +be greaterthan 1 normally.
app/test-dma-perf/config.ini new file +3
-0+ +; The format for variables is variable=first,last,increment,ADD|MUL. +Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/config.ini
Quoted 8 lines
+;
src_numa_nodeis used to control the numa node where the source memory is allocated.app/test-dma-perf/config.ini new file +2
-0+; dst_numa_node is used to control the numa node where the +destinationmemory is allocated.
app/test-dma-perf/config.ini new file +3
-0+ +; cache_flush is used to determine whether or not the cache +should be flushed, with 1 indicating to ; flush and 0 indicating to notflush.
Quoted 49 lines
Quoted 40 lines
Quoted 26 lines
app/test-dma-perf/config.ini new file +7
-0+ +; test_seconds controls the test time of the whole case. + +; To use DMA for a test, please specify the "lcore_dma" parameter. +; If you have already set the "-l" and "-a" parameters using EAL, +; make sure that the value of "lcore_dma" falls within their +range of thevalues.
app/test-dma-perf/config.ini new file +7
-0+ +; To use CPU for a test, please specify the "lcore" parameter. +; If you have already set the "-l" and "-a" parameters using EAL, +; make sure that the value of "lcore" falls within their range of values. + +; To specify a configuration file, use the "--config" flag +followed by the pathto the file.
app/test-dma-perf/config.ini new file +3
-0+ +; To specify a result file, use the "--result" flag followed by +the path to thefile.
app/test-dma-perf/config.ini new file +3
-0+; If you do not specify a result file, one will be generated with +the same name as the configuration ; file, with the addition of +"_result.csv" atthe end.
app/test-dma-perf/config.ini new file +2
-0+ +[case1]app/test-dma-perf/config.ini line 47
Quoted 10 lines
app/test-dma-perf/config.ini new file +10
-0+type=DMA_MEM_COPY +mem_size=10 +buf_size=64,8192,2,MUL +dma_ring_size=1024 +kick_batch=32 +src_numa_node=0 +dst_numa_node=0 +cache_flush=0 +test_seconds=2 +lcore_dma=lcore10@0000:00:04.2, lcore11@0000:00:04.3[Anoob] Isn't it better if we allow user to specify DMA dev ID rather than the PCI DBDF?
Quoted 7 lines
In the long run, I would expect config file to provide {core,
dma_dev_id,queue_id}Another thought is why to expose this at all? If we can restrict this perf application to have one thread only use one vchan, then application can easily create this mapping in run time. Unless you want one thread to use 2 different vchans which may not be desirable
since this is a standalone perf app.
Quoted 15 lines
[Cheng] Thank you for the feedback.
Here are my thoughts:
Firstly, the user may not know which device the DMA dev ID corresponds to, or which NUMA node it is on. In my example, I used the CBDMA environment, so I did not specify the work queue ID. When using DSA, the configuration would be something like lcore10@0000:00:04.2-q0 which contains core, dma and work queue id. The reason for exposing these options is that we want the user to fully understand which cores and devices are being used so that they know exactly where the performance data is coming from. For example, performance when cores and DMA devices are not on the same NUMA node, etc. This allows the testing scenario to be precise and flexible. If the application handles the mapping itself, the user loses control over the mapping and may not get the performance data they want. We believe control should be given to the user rather than the application.
Quoted 2 lines
[Anoob] I understand your view points. Thanks for the explanation.
[Cheng] sure, no problem.
Quoted 578 lines
Quoted 201 lines
Quoted 201 lines
Quoted 201 lines
app/test-dma-perf/config.ini new file +12
-0+eal_args=--in-memory --file-prefix=test + +[case2] +type=CPU_MEM_COPY +mem_size=10 +buf_size=64,8192,2,MUL +src_numa_node=0 +dst_numa_node=1 +cache_flush=0 +test_seconds=2 +lcore = 3, 4 +eal_args=--in-memory --no-pciapp/test-dma-perf/main.c new file +21
-0@@ -0,0 +1,569 @@ +/* SPDX-License-Identifier: BSD-3-Clause +* Copyright(c) 2023 Intel Corporation */ + +#include <stdio.h> +#include <stdlib.h> +#include <getopt.h> +#include <signal.h> +#include <stdbool.h> +#include <unistd.h> +#include <sys/wait.h> +#include <inttypes.h> +#include <libgen.h> + +#include <rte_eal.h> +#include <rte_cfgfile.h> +#include <rte_string_fns.h> +#include <rte_lcore.h> + +#include "main.h" + +#define CSV_HDR_FMT "Case %u : %s,lcore,DMA,buffersize,nr_buf,memory(MB),cycle,bandwidth(Gbps),MOps\n"
app/test-dma-perf/main.c new file +51
-0+ +#define MAX_EAL_PARAM_NB 100 +#define MAX_EAL_PARAM_LEN 1024 + +#define DMA_MEM_COPY "DMA_MEM_COPY" +#define CPU_MEM_COPY "CPU_MEM_COPY" + +#define CMDLINE_CONFIG_ARG "--config" +#define CMDLINE_RESULT_ARG "--result" + +#define MAX_PARAMS_PER_ENTRY 4 + +#define MAX_LONG_OPT_SZ 64 + +enum { +TEST_TYPE_NONE =0 , +TEST_TYPE_DMA_MEM_COPY , +TEST_TYPE_CPU_MEM_COPY +}; + +#define MAX_TEST_CASES 16 +static struct test_configure test_cases[MAX_TEST_CASES ]; + +char output_str[MAX_WORKER_NB ][MAX_OUTPUT_STR_LEN ]; + +static FILE *fd; + +static void +output_csv (bool need_blankline) +{ +uint32_t i; + +if (need_blankline) { +fprintf (fd,",,,,,,,,\n" ); +fprintf (fd,",,,,,,,,\n" ); + } + +for (i =0 ; i <RTE_DIM (output_str); i++) { +if (output_str[i][0 ]) { +fprintf (fd,"%s" , output_str[i]); + output_str[i][0 ] ='\0' ; + } + } + +fflush (fd); +} + +static void +output_env_info (void ) +{ + snprintf(output_str[0], MAX_OUTPUT_STR_LEN, "testenvironment:\n");
app/test-dma-perf/main.c new file +68
-0+snprintf (output_str[1 ],MAX_OUTPUT_STR_LEN ,"CPU frequency,%" + PRIu64"\n" ,rte_get_timer_hz ()); + +output_csv (true); +} + +static void +output_header (uint32_t case_id,struct test_configure *case_cfg) { +snprintf (output_str[0 ],MAX_OUTPUT_STR_LEN , +CSV_HDR_FMT , case_id, case_cfg->test_type_str); + +output_csv (true); +} + +static void +run_test_case (struct test_configure *case_cfg) { +switch (case_cfg->test_type) { +case TEST_TYPE_DMA_MEM_COPY : +mem_copy_benchmark (case_cfg, true); +break ; +case TEST_TYPE_CPU_MEM_COPY : +mem_copy_benchmark (case_cfg, false); +break ; +default : +printf ("Unknown test type. %s\n" , case_cfg->test_type_str); +break ; + } +} + +static void +run_test(uint32_t case_id, struct test_configure *case_cfg) { + uint32_t i; + uint32_t nb_lcores = rte_lcore_count(); + struct test_configure_entry *mem_size = &case_cfg->mem_size; + struct test_configure_entry *buf_size = &case_cfg->buf_size; + struct test_configure_entry *ring_size = &case_cfg->ring_size; + struct test_configure_entry *kick_batch = &case_cfg->kick_batch; + struct test_configure_entry dummy = { 0 }; + struct test_configure_entry *var_entry = &dummy; + + for (i = 0; i < RTE_DIM(output_str); i++) + memset(output_str[i], 0, MAX_OUTPUT_STR_LEN); + + if (nb_lcores <= case_cfg->lcore_dma_map.cnt) { + printf("Case %u: Not enough lcores.\n", case_id); + return; + } + + printf("Number of used lcores: %u.\n", nb_lcores); + + if (mem_size->incr != 0) + var_entry = mem_size; + + if (buf_size->incr != 0) + var_entry = buf_size; + + if (ring_size->incr != 0) + var_entry = ring_size; + + if (kick_batch->incr != 0) + var_entry = kick_batch; + + case_cfg->scenario_id = 0; + + output_header(case_id, case_cfg); + + for (var_entry->cur = var_entry->first; var_entry->cur <= +var_entry-Quoted 1 line
last;) {
app/test-dma-perf/main.c new file +21
-0+ case_cfg->scenario_id++; +printf ("\nRunning scenario %d\n" , case_cfg->scenario_id); + +run_test_case (case_cfg); +output_csv (false); + +if (var_entry->op ==OP_ADD ) + var_entry->cur += var_entry->incr; +else if (var_entry->op ==OP_MUL ) + var_entry->cur *= var_entry->incr; +else +break ; + } +} + +static int +parse_lcore (struct test_configure *test_case,const char *value) { +size_t len =strlen (value); +char *input = (char *)malloc ((len +1 ) *sizeof (char )); +strcpy (input, value); + struct lcore_dma_map_t *lcore_dma_map = &(test_case-Quoted 1 line
lcore_dma_map);app/test-dma-perf/main.c new file +17
-0+ +if (test_case ==NULL || value ==NULL ) +return -1 ; + +memset (lcore_dma_map,0 ,sizeof (struct lcore_dma_map_t )); + +char *token =strtok (input,", " ); +while (token !=NULL ) { +if (lcore_dma_map->cnt >=MAX_LCORE_NB ) { +free (input); +return -1 ; + } + +uint16_t lcore_id =atoi (token); + lcore_dma_map->lcores[lcore_dma_map->cnt++] = lcore_id; + + token =strtok (NULL ,", " );Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/main.c line 190
Quoted 8 lines
app/test-dma-perf/main.c new file +8
-0+ } + +free (input); +return 0 ; +} + +static int +parse_lcore_dma (struct test_configure *test_case,const char *value){
Quoted 356 lines
Quoted 350 lines
Quoted 217 lines
app/test-dma-perf/main.c new file +24
-0+struct lcore_dma_map_t *lcore_dma_map; +char *input =strndup (value,strlen (value) +1 ); +char *addrs = input; +char *ptrs[2 ]; +char *start, *end, *substr; +uint16_t lcore_id; +int ret =0 ; + +while (*addrs =='\0' ) + addrs++; +if (*addrs =='\0' ) { +fprintf (stderr,"No input DMA addresses\n" ); + ret =-1 ; + goto out; + } + + substr =strtok (addrs,"," ); +if (substr ==NULL ) { +fprintf (stderr,"No input DMA address\n" ); + ret =-1 ; + goto out; + } + + memset(&test_case->lcore_dma_map, 0, sizeof(struct
lcore_dma_map_t));app/test-dma-perf/main.c new file +15
-0+ + do { + rte_strsplit(substr, strlen(substr), ptrs, 2, '@'); + + start = strstr(ptrs[0], "lcore"); + if (start == NULL) { + fprintf(stderr, "Illegal lcore\n"); + ret = -1; + break; + } + + start += 5; + lcore_id = strtol(start, &end, 0); + if (end == start) { + fprintf(stderr, "No input lcore ID or ID %d iswrong\n",
lcore_id);app/test-dma-perf/main.c new file +7
-0+ ret =-1 ; +break ; + } + + lcore_dma_map = &test_case->lcore_dma_map; + lcore_dma_map->lcores[lcore_dma_map->cnt] = lcore_id; + strcpy(lcore_dma_map->dma_names[lcore_dma_map-Quoted 1 line
cnt], ptrs[1]);
app/test-dma-perf/main.c new file +23
-0+ lcore_dma_map->cnt++; + substr =strtok (NULL ,"," ); + }while (substr !=NULL ); + +out: +free (input); +return ret; +} + +static int +parse_entry (const char *value,struct test_configure_entry *entry) { +char input[255 ] = {0 }; +char *args[MAX_PARAMS_PER_ENTRY ]; +int args_nr =-1 ; + +if (value ==NULL || entry ==NULL ) + goto out; + +strncpy (input, value,254 ); +if (*input =='\0' ) + goto out; + + args_nr = rte_strsplit(input, strlen(input), args,
MAX_PARAMS_PER_ENTRY, ',');app/test-dma-perf/main.c new file +35
-0+if (args_nr !=1 && args_nr !=4 ) + goto out; + + entry->cur = entry->first = (uint32_t )atoi (args[0 ]); + +if (args_nr ==4 ) { + entry->last = (uint32_t )atoi (args[1 ]); + entry->incr = (uint32_t )atoi (args[2 ]); +if (!strcmp (args[3 ],"MUL" )) + entry->op =OP_MUL ; +else if (!strcmp (args[3 ],"ADD" )) + entry->op =OP_ADD ; +else { +printf ("Invalid op %s.\n" , args[3 ]); + args_nr =-1 ; + } + }else { + entry->op =OP_NONE ; + entry->last =0 ; + entry->incr =0 ; + } +out: +return args_nr; +} + +static uint16_t +load_configs (const char *path) +{ +struct rte_cfgfile *cfgfile; +int nb_sections, i; +struct test_configure *test_case; +char section_name[CFG_NAME_LEN ]; +const char *case_type; +const char *lcore_dma; + const char *mem_size_str, *buf_size_str, *ring_size_str,
*kick_batch_str;app/test-dma-perf/main.c new file +13
-0+int args_nr, nb_vp; +bool is_dma; + +printf ("config file parsing...\n" ); + cfgfile =rte_cfgfile_load (path,0 ); +if (!cfgfile) { +printf ("Open configure file error.\n" ); +exit (1 ); + } + + nb_sections =rte_cfgfile_num_sections (cfgfile,NULL ,0 ); +if (nb_sections >MAX_TEST_CASES ) { + printf("Error: The maximum number of cases is %d.\n",
MAX_TEST_CASES);app/test-dma-perf/main.c new file +7
-0+exit (1 ); + } + +for (i =0 ; i < nb_sections; i++) { +snprintf (section_name,CFG_NAME_LEN ,"case%d" , i +1 ); + test_case = &test_cases[i]; + case_type = rte_cfgfile_get_entry(cfgfile, section_name,"type");
app/test-dma-perf/main.c new file +2
-0+if (!case_type) { + printf("Error: No case type in case %d, the test will befinished here.\n",
app/test-dma-perf/main.c new file +7
-0+ i + 1); + test_case->is_valid = false; +continue ; + } + +if (strcmp (case_type,DMA_MEM_COPY ) ==0 ) { + test_case->test_type =
TEST_TYPE_DMA_MEM_COPY;app/test-dma-perf/main.c new file +4
-0+ test_case->test_type_str =DMA_MEM_COPY ; + is_dma = true; + }else if (strcmp (case_type,CPU_MEM_COPY ) ==0 ) { + test_case->test_type =
TEST_TYPE_CPU_MEM_COPY;app/test-dma-perf/main.c new file +4
-0+ test_case->test_type_str =CPU_MEM_COPY ; + is_dma = false; + }else { + printf("Error: Cannot find case type %s in case%d.\n",
case_type, i + 1);app/test-dma-perf/main.c new file +7
-0+ test_case->is_valid = false; +continue ; + } + + nb_vp =0 ; + + test_case->src_numa_node =(int)atoi(rte_cfgfile_get_entry(cfgfile,
+section_name, "src_numa_node"));app/test-dma-perf/main.c new file +1
-0+ test_case->dst_numa_node =(int)atoi(rte_cfgfile_get_entry(cfgfile,
+section_name, "dst_numa_node"));app/test-dma-perf/main.c new file +2
-0+ + mem_size_str = rte_cfgfile_get_entry(cfgfile, section_name,"
mem_size");app/test-dma-perf/main.c new file +1
-0+ args_nr = parse_entry(mem_size_str, &test_case-Quoted 1 line
mem_size);app/test-dma-perf/main.c new file +8
-0+if (args_nr <0 ) { +printf ("parse error in case %d.\n" , i +1 ); + test_case->is_valid = false; +continue ; + }else if (args_nr >1 ) + nb_vp++; + + buf_size_str = rte_cfgfile_get_entry(cfgfile, section_name,"
buf_size");app/test-dma-perf/main.c new file +10
-0+ args_nr =parse_entry (buf_size_str, &test_case->buf_size); +if (args_nr <0 ) { +printf ("parse error in case %d.\n" , i +1 ); + test_case->is_valid = false; +continue ; + }else if (args_nr >1 ) + nb_vp++; + +if (is_dma) { + ring_size_str = rte_cfgfile_get_entry(cfgfile,
section_name,
+"dma_ring_size");app/test-dma-perf/main.c new file +1
-0+ args_nr = parse_entry(ring_size_str, &test_case-Quoted 1 line
ring_size);app/test-dma-perf/main.c new file +8
-0+if (args_nr <0 ) { +printf ("parse error in case %d.\n" , i +1 ); + test_case->is_valid = false; +continue ; + }else if (args_nr >1 ) + nb_vp++; + + kick_batch_str = rte_cfgfile_get_entry(cfgfile,
section_name, "kick_batch");app/test-dma-perf/main.c new file +1
-0+ args_nr = parse_entry(kick_batch_str, &test_case-Quoted 1 line
kick_batch);app/test-dma-perf/main.c new file +8
-0+if (args_nr <0 ) { +printf ("parse error in case %d.\n" , i +1 ); + test_case->is_valid = false; +continue ; + }else if (args_nr >1 ) + nb_vp++; + + lcore_dma = rte_cfgfile_get_entry(cfgfile,
section_name, "lcore_dma");app/test-dma-perf/main.c new file +1
-0+ int lcore_ret = parse_lcore_dma(test_case,
lcore_dma);app/test-dma-perf/main.c
Quoted 2 lines
app/test-dma-perf/main.c new file +2
-0+if (lcore_ret <0 ) { + printf("parse lcore dma error in case %d.\n", i1);
Quoted 117 lines
app/test-dma-perf/main.c new file +5
-0+ test_case->is_valid = false; +continue ; + } + }else { + lcore_dma = rte_cfgfile_get_entry(cfgfile,
section_name, "lcore");app/test-dma-perf/main.c new file +10
-0+int lcore_ret =parse_lcore (test_case, lcore_dma); +if (lcore_ret <0 ) { +printf ("parse lcore error in case %d.\n" , i +1 ); + test_case->is_valid = false; +continue ; + } + } + +if (nb_vp >1 ) { + printf("Error, each section can only have a singlevariable parameter.\n");
app/test-dma-perf/main.c new file +6
-0+ test_case->is_valid = false; +continue ; + } + + test_case->cache_flush = + (int)atoi(rte_cfgfile_get_entry(cfgfile, section_name,"
cache_flush"));app/test-dma-perf/main.c new file +1
-0+ test_case->test_secs =(uint16_t)atoi(rte_cfgfile_get_entry(cfgfile,
app/test-dma-perf/main.c new file +3
-0+ section_name, "test_seconds")); + + test_case->eal_args = rte_cfgfile_get_entry(cfgfile,
section_name, "eal_args");app/test-dma-perf/main.c new file +19
-0+ test_case->is_valid = true; + } + +rte_cfgfile_close (cfgfile); +printf ("config file parsing complete.\n\n" ); +return i; +} + +/* Parse the argument given in the command line of the +application */ static int append_eal_args (int argc,char **argv, +const char *eal_args,char **new_argv) { +int i; +char *tokens[MAX_EAL_PARAM_NB ]; +char args[MAX_EAL_PARAM_LEN ] = {0 }; +int token_nb, new_argc =0 ; + +for (i =0 ; i < argc; i++) { + if ((strcmp(argv[i], CMDLINE_CONFIG_ARG) == 0) || + (strcmp(argv[i], CMDLINE_RESULT_ARG) ==0)) {
app/test-dma-perf/main.c new file +4
-0+ i++; +continue ; + } + strlcpy(new_argv[new_argc], argv[i],
sizeof(new_argv[new_argc]));app/test-dma-perf/main.c new file +37
-0+ new_argc++; + } + +if (eal_args) { +strlcpy (args, eal_args,sizeof (args)); + token_nb =rte_strsplit (args,strlen (args), + tokens,MAX_EAL_PARAM_NB ,' ' ); +for (i =0 ; i < token_nb; i++) +strcpy (new_argv[new_argc++], tokens[i]); + } + +return new_argc; +} + +int +main (int argc,char *argv[]) +{ +int ret; +uint16_t case_nb; +uint32_t i, nb_lcores; +pid_t cpid, wpid; +int wstatus; +char args[MAX_EAL_PARAM_NB ][MAX_EAL_PARAM_LEN ]; +char *pargs[MAX_EAL_PARAM_NB ]; +char *cfg_path_ptr =NULL ; +char *rst_path_ptr =NULL ; +char rst_path[PATH_MAX ]; +int new_argc; +bool is_first_case = true; + +memset (args,0 ,sizeof (args)); + +for (i =0 ; i <RTE_DIM (pargs); i++) + pargs[i] = args[i]; + +for (i =0 ; i < (uint32_t )argc; i++) { + if (strncmp(argv[i], CMDLINE_CONFIG_ARG,
MAX_LONG_OPT_SZ) == 0)app/test-dma-perf/main.c new file +2
-0+ cfg_path_ptr = argv[i +1 ]; + if (strncmp(argv[i], CMDLINE_RESULT_ARG,
MAX_LONG_OPT_SZ) == 0)app/test-dma-perf/main.c new file +21
-0+ rst_path_ptr = argv[i +1 ]; + } +if (cfg_path_ptr ==NULL ) { +printf ("Config file not assigned.\n" ); +return -1 ; + } +if (rst_path_ptr ==NULL ) { +strlcpy (rst_path, cfg_path_ptr,PATH_MAX ); +strcat (strtok (basename (rst_path),"." ),"_result.csv" ); + rst_path_ptr = rst_path; + } + + case_nb =load_configs (cfg_path_ptr); + fd =fopen (rst_path_ptr,"w" ); +if (fd ==NULL ) { +printf ("Open output CSV file error.\n" ); +return -1 ; + } +fclose (fd); + +for (i =0 ; i < case_nb; i++) {app/test-dma-perf/main.c line 505
Quoted 10 lines
app/test-dma-perf/main.c new file +10
-0+if (test_cases[i].test_type ==TEST_TYPE_NONE ) { +printf ("No test type in test case %d.\n\n" , i +1 ); +continue ; + } +if (!test_cases[i].is_valid) { +printf ("Invalid test case %d.\n\n" , i +1 ); +continue ; + } + + cpid =fork ();[Anoob] Do we really need
fork()? Can't we use code like,Quoted 5 lines
RTE_LCORE_FOREACH_WORKER(lcore_id) { ret |= rte_eal_wait_lcore(lcore_id); }to wait for all threads to exit?
Quoted 8 lines
[Cheng] Good question.
Fork()is used here to establish a new process for the new test case. In order for each test case to have a new EAL environment (for the flexibility), the EAL must be reinitialized for each case. However, the EAL parameters can only be initialized once per process. Therefore, we use a new process to run each new test case. Moreover, each test case runs sequentially and does not affect the others, ensuring the accuracy of the performance data. Your code would wait for all threads to exit in the same process. However, it would not provide a"clean"
Quoted 3 lines
environment for each test case like
fork()does.Fork()allows us to have a fully reinitialized environment, with no impact or side effects from previous test cases. This results in clean, precise performance data for
Quoted 8 lines
each case.
Quoted 3 lines
Please let me know your thoughts on this. And please let me know if you have any other questions or require any clarification.
[Anoob] This was just a generic observation. I do not have a strong opinion either way.
[Cheng] sure, got it.
Quoted 171 lines
Quoted 171 lines
Thanks,
ChengQuoted 167 lines
Quoted 167 lines
app/test-dma-perf/main.c new file +7
-0+if (cpid <0 ) { +printf ("Fork case %d failed.\n" , i +1 ); +exit (EXIT_FAILURE ); + }else if (cpid ==0 ) { +printf ("\nRunning case %u\n\n" , i +1 ); + + new_argc = append_eal_args(argc, argv,test_cases[i].eal_args, pargs);
app/test-dma-perf/main.c new file +3
-0+ ret =rte_eal_init (new_argc, pargs); +if (ret <0 ) + rte_exit(EXIT_FAILURE, "Invalid EALarguments\n");
app/test-dma-perf/main.c new file +6
-0+ +/* Check lcores. */ + nb_lcores =rte_lcore_count (); +if (nb_lcores <2 ) + rte_exit(EXIT_FAILURE, + "There should be at least 2 workerlcores.\n");
app/test-dma-perf/main.c new file +36
-0+ + fd =fopen (rst_path_ptr,"a" ); +if (!fd) { +printf ("Open output CSV file error.\n" ); +return 0 ; + } + +if (is_first_case) { +output_env_info (); + is_first_case = false; + } +run_test (i +1 , &test_cases[i]); + +/* clean up the EAL */ +rte_eal_cleanup (); + +fclose (fd); + +printf ("\nCase %u completed.\n\n" , i +1 ); + +exit (EXIT_SUCCESS ); + }else { + wpid =waitpid (cpid, &wstatus,0 ); +if (wpid ==-1 ) { +printf ("waitpid error.\n" ); +exit (EXIT_FAILURE ); + } + +if (WIFEXITED (wstatus)) +printf ("Case process exited. status %d\n\n" , +WEXITSTATUS (wstatus)); +else if (WIFSIGNALED (wstatus)) +printf ("Case process killed by signal %d\n\n" , +WTERMSIG (wstatus)); +else if (WIFSTOPPED (wstatus)) + printf("Case process stopped by signal%d\n\n",
app/test-dma-perf/main.c new file +5
-0+ WSTOPSIG(wstatus)); + else if (WIFCONTINUED(wstatus)) + printf("Case process continued.\n\n"); + else + printf("Case process unknownterminated.\n\n");
app/test-dma-perf/main.c new file +7
-0+ } + } + +printf ("Bye...\n" ); +return 0 ; +} +app/test-dma-perf/main.h new file +68
-0@@ -0,0 +1,69 @@ +/* SPDX-License-Identifier: BSD-3-Clause +* Copyright(c) 2023 Intel Corporation */ + +#ifndef _MAIN_H_ +#define _MAIN_H_ + + +#include <rte_common.h> +#include <rte_cycles.h> +#include <rte_dev.h> +#include <rte_dmadev.h> + +#ifndef __maybe_unused +#define __maybe_unused __rte_unused +#endif + +#define MAX_WORKER_NB 128 +#define MAX_OUTPUT_STR_LEN 512 + +#define MAX_DMA_NB 128 +#define MAX_LCORE_NB 256 + +extern char output_str[MAX_WORKER_NB ][MAX_OUTPUT_STR_LEN ]; + +typedef enum { +OP_NONE =0 , +OP_ADD , +OP_MUL +}alg_op_type ; + +struct test_configure_entry { +uint32_t first; +uint32_t last; +uint32_t incr; +alg_op_type op; +uint32_t cur; +}; + +struct lcore_dma_map_t { +uint32_t lcores[MAX_WORKER_NB ]; +char dma_names[MAX_WORKER_NB ][RTE_DEV_NAME_MAX_LEN ]; +int16_t dma_ids[MAX_WORKER_NB ]; +uint16_t cnt; +}; + +struct test_configure { +bool is_valid; +uint8_t test_type; +const char *test_type_str; +uint16_t src_numa_node; +uint16_t dst_numa_node; +uint16_t opcode; +bool is_dma; +struct lcore_dma_map_t lcore_dma_map; +struct test_configure_entry mem_size; +struct test_configure_entry buf_size; +struct test_configure_entry ring_size; +struct test_configure_entry kick_batch; +uint32_t cache_flush; +uint32_t nr_buf; +uint16_t test_secs; +const char *eal_args; +uint8_t scenario_id; +}; + +void mem_copy_benchmark (struct test_configure *cfg,bool is_dma); + +#endif /* _MAIN_H_ */ app/test-dma- renamed from app/test-dma-perf/meson.build +0
-0perf/meson.build new file mode 100644 index 0000000000..bd6c264002
app/test-dma-perf/meson.build new file +17
-1@@ -0,0 +1,17 @@ +# SPDX-License-Identifier: BSD-3-Clause # Copyright(c) 2019-2023 +Intel Corporation + +# meson file, for building this app as part of a main DPDK build. + +if is_windows + build = false + reason = 'not supported on Windows' + subdir_done() +endif + +deps += ['dmadev', 'mbuf', 'cfgfile'] + +sources = files( + 'main.c', + 'benchmark.c', +) --2.40.1
Hi Cheng,
Please see inline.
Thanks,
Anoob
Quoted 257 lines
-----Original Message-----
From: Jiang, Cheng1 <[email protected]>
Sent: Thursday, June 15, 2023 7:36 PM
To: Anoob Joseph <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>; Amit Prakash Shukla <[email protected]>; Ma, WenwuX <[email protected]>; Wang, YuanX <[email protected]>; He, Xingguang <[email protected]>
Subject: RE: [EXT] [PATCH v6] app/dma-perf: introduce dma-perf applicationHi Anoob,
Replies are inline.
Thanks,
ChengQuoted 4 lines
-----Original Message-----
From: Anoob Joseph <[email protected]>
Sent: Thursday, June 15, 2023 4:45 PM
To: Jiang, Cheng1 <[email protected]>Quoted 8 lines
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>; Amit Prakash Shukla <[email protected]>; Ma, WenwuX <[email protected]>; Wang, YuanX <[email protected]>; He, Xingguang <[email protected]>
Subject: RE: [EXT] [PATCH v6] app/dma-perf: introduce dma-perfapplication
Quoted 14 lines
Hi Cheng,
Please see inline.
Thanks,
AnoobQuoted 6 lines
-----Original Message-----
From: Jiang, Cheng1 <[email protected]>
Sent: Thursday, June 15, 2023 1:31 PM
To: Anoob Joseph <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>;Amit
Quoted 15 lines
Quoted 3 lines
Prakash Shukla <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; Ma, WenwuX <[email protected]>; Wang,YuanX
Quoted 11 lines
<[email protected]>; He, Xingguang <[email protected]>
Subject: RE: [EXT] [PATCH v6] app/dma-perf: introduce dma-perf applicationHi,
Thanks for your comments, the replies are inline.
Thanks,
ChengQuoted 10 lines
Quoted 6 lines
Quoted 6 lines
-----Original Message-----
From: Anoob Joseph <[email protected]>
Sent: Thursday, June 15, 2023 1:22 PM
To: Jiang, Cheng1 <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>;Amit
Quoted 3 lines
Quoted 3 lines
Prakash Shukla <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; Ma, WenwuX <[email protected]>;Wang,
Quoted 20 lines
Quoted 8 lines
YuanX
Quoted 7 lines
<[email protected]>; He, Xingguang <[email protected]>
Subject: RE: [EXT] [PATCH v6] app/dma-perf: introduce dma-perf applicationHi,
Thanks for working on the comments. Few more top level comment
inline.
Quoted 11 lines
Quoted 9 lines
Thanks,
AnoobQuoted 5 lines
-----Original Message-----
From: Cheng Jiang <[email protected]>
Sent: Tuesday, June 13, 2023 10:02 AM
To: [email protected]; [email protected]; [email protected]; [email protected]; Amit PrakashShukla
Quoted 1 line
Quoted 1 line
<[email protected]>; Anoob Joseph
Quoted 61 lines
Quoted 61 lines
Quoted 2 lines
Quoted 2 lines
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected];
Quoted 19 lines
Quoted 6 lines
Cheng Jiang <[email protected]>
Subject: [EXT] [PATCH v6] app/dma-perf: introduce dma-perf applicationExternal Email
Quoted 13 lines
------------------------------------------------------------------
--
-- There are many high-performance DMA devices supported in DPDK now, and these DMA devices can also be integrated into other modules of DPDK as accelerators, such as Vhost. Before integrating DMA into applications, developers need to know the performance of these DMA devices in various scenarios and the performance of CPUs in the same scenario, such as different buffer lengths. Only in this way can we know the target performance of the application accelerated by using them. This patch introduces a high-performance testing tool, which supports comparing the performance of CPU and DMA in different scenarios automatically with a pre- set config file. Memory Copy performance test areQuoted 26 lines
supported for now.
Quoted 25 lines
Signed-off-by Cheng Jiang <[email protected]>Signed-off-by Jiayu Hu <[email protected]>Signed-off-by Yuan Wang <[email protected]>Acked-by Morten Brørup <[email protected]>Acked-by Chenbo Xia <[email protected]>---
v6:
improved code based on Anoob's comments; fixed some code structure issues;
v5:
fixed someLONG_LINEwarnings;
v4:
fixed inaccuracy of the memory footprint display;
v3:
fixed some typos;
v2:
added lcore/dmadev designation;
added error case process;
removedworker_threadsparameter from config.ini; improved the logs;
improved config file;app/meson.build | 1 + app/test-dma-perf/benchmark.c | 477++++++++++++++++++++++++++++
Quoted 12 lines
Quoted 12 lines
app/test-dma-perf/config.ini | 59 ++++
app/test-dma-perf/main.c | 569 ++++++++++++++++++++++++++++++++++ app/test-dma-perf/main.h | 69 +++++ app/test-dma-perf/meson.build | 17 + 6 files changed, 1192 insertions(+) create mode 100644
app/test-dma-perf/benchmark.ccreate mode 100644 app/test-dma-perf/config.ini create mode 100644 app/test-dma-perf/main.ccreate mode 100644app/test-dma-perf/main.hcreate mode
100644 app/test-dma-perf/meson.buildQuoted 14 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
app/meson.build index renamed from app/meson.build +0
-0
74d2420f67..4fc1a83eba100644app/meson.build +0
-0@@ -19,6 +19,7 @@ apps = [ 'test-cmdline',Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/meson.build +1
-0'test-compress-perf', 'test-crypto-perf', + 'test-dma-perf', 'test-eventdev', 'test-fib', 'test-flow-perf',app/test-dma- renamed from app/test-dma-perf/benchmark.c +0
-0
perf/benchmark.cnew file mode 100644 index0000000000..bc1ca82297
Quoted 58 lines
Quoted 58 lines
Quoted 16 lines
Quoted 16 lines
app/test-dma-perf/benchmark.c new file +13
-0@@ -0,0 +1,477 @@ +/* SPDX-License-Identifier: BSD-3-Clause +* Copyright(c) 2023 Intel Corporation */ + +#include <inttypes.h> +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> + +#include <rte_time.h> +#include <rte_mbuf.h> +#include <rte_dmadev.h> +#include <rte_malloc.h> +#include <rte_lcore.h> Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+ +#include "main.h" + +#define MAX_DMA_CPL_NB 255 + +#define TEST_WAIT_U_SECOND 10000 + +#define CSV_LINE_DMA_FMT "Scenario %u,%u,%s,%u,%u,%.2lf,%"PRIu64
Quoted 2 lines
Quoted 2 lines
",%.3lf,%.3lf\n"
+#defineCSV_LINE_CPU_FMT"Scenario %u,%u,NA,%u,%u,%.2lf,%"PRIu64
Quoted 21 lines
Quoted 21 lines
",%.3lf,%.3lf\n"
app/test-dma-perf/benchmark.c new file +20
-0+ +struct worker_info { +bool ready_flag; +bool start_flag; +bool stop_flag; +uint32_t total_cpl; +uint32_t test_cpl; +}; + +struct lcore_params { +uint8_t scenario_id; +unsigned int lcore_id; +char *dma_name; +uint16_t worker_id; +uint16_t dev_id; +uint32_t nr_buf; +uint16_t kick_batch; +uint32_t buf_size; +uint16_t test_secs; +struct rte_mbuf **srcs;Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+struct rte_mbuf **dsts; +struct worker_info worker_info; +}; + +static struct rte_mempool *src_pool;static struct rte_mempool +*dst_pool; + +static volatile struct lcore_params*worker_params[MAX_WORKER_NB];
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
app/test-dma-perf/benchmark.c
Quoted 2 lines
app/test-dma-perf/benchmark.c new file +2
-0+ +#define PRINT_ERR (...) print_err(__func__, __LINE__,
__VA_ARGS__)Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
app/test-dma-perf/benchmark.c new file +12
-0+ +static inline int +__rte_format_printf(3, 4) +print_err(const char *func, int lineno, const char *format, ...) { + va_list ap; + int ret; + + ret = fprintf(stderr, "In %s:%d - ", func, lineno); + va_start(ap, format); + ret += vfprintf(stderr, format, ap); + va_end(ap); +Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +6
-0+return ret; +} + +static inline void +calc_result(uint32_t buf_size, uint32_t nr_buf, uint16_t +nb_workers,
uint16_ttest_secs,app/test-dma-perf/benchmark.c new file +1
-0+ uint32_t total_cnt, float *memory,
uint32_tQuoted 3 lines
Quoted 3 lines
Quoted 3 lines
app/test-dma-perf/benchmark.c
Quoted 3 lines
*ave_cycle,app/test-dma-perf/benchmark.c new file +2
-0+ float *bandwidth, float *mops) { + *memory = (float)(buf_size * (nr_buf / nb_workers) * 2) /(1024
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
+*
1024);app/test-dma-perf/benchmark.c new file +2
-0+ *ave_cycle = test_secs *rte_get_timer_hz () / total_cnt; + *bandwidth = (buf_size * 8 * (rte_get_timer_hz() /(float)*ave_cycle)) / 1000000000;
[Anoob] The above calculation may not yield actual results. 'ave_cycle' would get converted to integer and then bandwidth would be allowed to report only very few values. Instead, we can do the calculation directly like,
*bandwidth = ((float)buf_size * 8 * total_cnt / test_secs) / 1000000000; *mops = (float)total_cnt / test_secs / 1000000;
Same issue is there with below calculation as well. Please check.
Side note: in bandwidth calculation, shouldn't we be dividing by 1024*1024*1024? I've just carried the calculation that you used. Feel free to correct as required.
Quoted 337 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
app/test-dma-perf/benchmark.c
Quoted 7 lines
app/test-dma-perf/benchmark.c new file +5
-0+ *mops = (float )rte_get_timer_hz () / *ave_cycle /1000000 ; } + +static void +output_result(uint8_t scenario_id, uint32_t lcore_id, char +*dma_name,
uint64_tave_cycle,app/test-dma-perf/benchmark.c new file +1
-0+ uint32_t buf_size, uint32_t nr_buf, floatmemory,
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
app/test-dma-perf/benchmark.c
Quoted 7 lines
app/test-dma-perf/benchmark.c new file +7
-0+ float bandwidth, float mops, bool is_dma) { + if (is_dma) + printf("lcore %u, DMA %s:\n", lcore_id, dma_name); + else + printf("lcore %u\n", lcore_id); + + printf("average cycles/op: %" PRIu64 ", buffer size: %u,
nr_buf:Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
app/test-dma-perf/benchmark.c
Quoted 5 lines
+%u,
memory: %.2lfMB, frequency: %" PRIu64 ".\n",app/test-dma-perf/benchmark.c new file +1
-0+ ave_cycle, buf_size, nr_buf, memory,
rte_get_timer_hz());app/test-dma-perf/benchmark.c new file +1
-0+ printf("Average bandwidth: %.3lfGbps, MOps: %.3lf\n",bandwidth,
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
app/test-dma-perf/benchmark.c
Quoted 4 lines
app/test-dma-perf/benchmark.c new file +4
-0+mops); + +if (is_dma) + snprintf(output_str[lcore_id],
MAX_OUTPUT_STR_LEN,Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
app/test-dma-perf/benchmark.c
Quoted 3 lines
CSV_LINE_DMA_FMT,app/test-dma-perf/benchmark.c new file +2
-0+ scenario_id, lcore_id, dma_name, buf_size, + nr_buf, memory, ave_cycle, bandwidth,mops);
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
app/test-dma-perf/benchmark.c
Quoted 2 lines
app/test-dma-perf/benchmark.c new file +2
-0+else + snprintf(output_str[lcore_id],
MAX_OUTPUT_STR_LEN,Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
app/test-dma-perf/benchmark.c
Quoted 3 lines
CSV_LINE_CPU_FMT,app/test-dma-perf/benchmark.c new file +2
-0+ scenario_id, lcore_id, buf_size, + nr_buf, memory, ave_cycle, bandwidth,mops); }
Quoted 23 lines
Quoted 5 lines
Quoted 5 lines
app/test-dma-perf/benchmark.c
Quoted 5 lines
app/test-dma-perf/benchmark.c new file +5
-0+ +static inline void +cache_flush_buf (__maybe_unused struct rte_mbuf **array, +__maybe_unused uint32_t buf_size, + __maybe_unused uint32_t nr_buf) { #ifdef
RTE_ARCH_X86_64Quoted 17 lines
Quoted 17 lines
Quoted 17 lines
app/test-dma-perf/benchmark.c new file +17
-0+char *data; +struct rte_mbuf **srcs = array; +uint32_t i, offset; + +for (i =0 ; i < nr_buf; i++) { + data = rte_pktmbuf_mtod(srcs[i], char *); +for (offset =0 ; offset < buf_size; offset +=64 ) +__builtin_ia32_clflush (data + offset); + } +#endif +} + +/* Configuration of device. */ +static void +configure_dmadev_queue (uint32_t dev_id,uint32_t ring_size) { +uint16_t vchan =0 ; +struct rte_dma_info info;Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+struct rte_dma_conf dev_config = { .nb_vchans =1 }; +struct rte_dma_vchan_conf qconf = { + .direction =RTE_DMA_DIR_MEM_TO_MEM , + .nb_desc = ring_size + }; + +if (rte_dma_configure (dev_id, &dev_config) !=0 ) + rte_exit(EXIT_FAILURE, "Error with dmaconfigure.\n");
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
app/test-dma-perf/benchmark.c
Quoted 3 lines
app/test-dma-perf/benchmark.c new file +3
-0+ +if (rte_dma_vchan_setup (dev_id, vchan, &qconf) !=0 ) + rte_exit(EXIT_FAILURE, "Error with queueconfiguration.\n");
Quoted 15 lines
Quoted 15 lines
Quoted 15 lines
Quoted 15 lines
app/test-dma-perf/benchmark.c new file +4
-0+ +rte_dma_info_get (dev_id, &info); +if (info.nb_vchans !=1 ) + rte_exit(EXIT_FAILURE, "Error, no configured queuesreported on device id. %u\n",
app/test-dma-perf/benchmark.c new file +10
-0+ dev_id); + +if (rte_dma_start (dev_id) !=0 ) +rte_exit (EXIT_FAILURE ,"Error with dma start.\n" ); } + +static int +config_dmadevs (struct test_configure *cfg) { +uint32_t ring_size = cfg->ring_size.cur; +struct lcore_dma_map_t *ldm = &cfg->lcore_dma_map; +uint32_t nb_workers = ldm->cnt;Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+uint32_t i; +int dev_id; +uint16_t nb_dmadevs =0 ; +char *dma_name; + +for (i =0 ; i < ldm->cnt; i++) { + dma_name = ldm->dma_names[i]; + dev_id =
rte_dma_get_dev_id_by_name(dma_name);Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
app/test-dma-perf/benchmark.c new file +2
-0+if (dev_id ==-1 ) { + fprintf(stderr, "Error: Fail to find DMA %s.\n",
dma_name);app/test-dma-perf/benchmark.c new file +3
-0+ goto end; + } +Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+ ldm->dma_ids[i] = dev_id; +configure_dmadev_queue (dev_id, ring_size); + ++nb_dmadevs; + } + +end: +if (nb_dmadevs < nb_workers) { + printf("Not enough dmadevs (%u) for all workers(%u).\n",
Quoted 18 lines
Quoted 18 lines
Quoted 6 lines
Quoted 6 lines
nb_dmadevs,nb_workers);app/test-dma-perf/benchmark.c new file +5
-0+return -1 ; + } + +printf ("Number of used dmadevs: %u.\n" , nb_dmadevs); +Quoted 8 lines
app/test-dma-perf/benchmark.c line 187
Quoted 6 lines
app/test-dma-perf/benchmark.c new file +6
-0+return 0 ; +} + +#define POLL_MAX 1000 + +[Anoob] Extra blank line. You can consider removing.
[Cheng] sure, sorry for the miss.
Quoted 1 line
Quoted 14 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
app/test-dma-perf/benchmark.c new file +6
-0+static inline void +do_dma_submit_and_poll (uint16_t dev_id,uint64_t *async_cnt, +volatile struct worker_info *worker_info) { +int ret; +uint16_t nr_cpl; +Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+ ret =rte_dma_submit (dev_id,0 ); +if (ret <0 ) { +rte_dma_stop (dev_id); +rte_dma_close (dev_id); +rte_exit (EXIT_FAILURE ,"Error with dma submit.\n" ); + } + + nr_cpl = rte_dma_completed(dev_id, 0,
MAX_DMA_CPL_NB, NULL,Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
app/test-dma-perf/benchmark.c
Quoted 9 lines
NULL);
app/test-dma-perf/benchmark.c new file +8
-0+ *async_cnt -= nr_cpl; + worker_info->total_cpl += nr_cpl; } + +static inline int +do_dma_mem_copy (void *p) +{ +const uint16_t *para_idx = (uint16_t *)p; + volatile struct lcore_params *para =worker_params[*para_idx];
Quoted 18 lines
Quoted 1 line
Quoted 1 line
app/test-dma-perf/benchmark.c
Quoted 1 line
app/test-dma-perf/benchmark.c new file +1
-0+ volatile struct worker_info *worker_info = &(para-
worker_info);Quoted 16 lines
Quoted 16 lines
Quoted 16 lines
app/test-dma-perf/benchmark.c new file +16
-0+const uint16_t dev_id = para->dev_id; +const uint32_t nr_buf = para->nr_buf; +const uint16_t kick_batch = para->kick_batch; +const uint32_t buf_size = para->buf_size; +struct rte_mbuf **srcs = para->srcs; +struct rte_mbuf **dsts = para->dsts; +uint16_t nr_cpl; +uint64_t async_cnt =0 ; +uint32_t i; +uint32_t poll_cnt =0 ; +int ret; + + worker_info->stop_flag = false; + worker_info->ready_flag = true; + +while (!worker_info->start_flag)Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +6
-0+ ; + +while (1 ) { +for (i =0 ; i < nr_buf; i++) { +dma_copy: + ret = rte_dma_copy(dev_id, 0,
rte_pktmbuf_iova(srcs[i]),app/test-dma-perf/benchmark.c new file +1
-0+ rte_pktmbuf_iova(dsts[i]), buf_size,0);
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
app/test-dma-perf/benchmark.c line 247
Quoted 3 lines
app/test-dma-perf/benchmark.c new file +3
-0+if (unlikely (ret <0 )) { +if (ret == -ENOSPC ) { +do_dma_submit_and_poll(dev_id,
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
app/test-dma-perf/benchmark.c
Quoted 6 lines
&async_cnt,worker_info);app/test-dma-perf/benchmark.c new file +5
-0+ goto dma_copy; + }else { +/* Error exit */ +rte_dma_stop (dev_id); + rte_exit(EXIT_FAILURE,"DMA
Quoted 21 lines
Quoted 21 lines
Quoted 21 lines
Quoted 21 lines
enqueue failed\n");
app/test-dma-perf/benchmark.c new file +6
-0+ } + } + async_cnt++; + +if ((async_cnt % kick_batch) ==0 ) + do_dma_submit_and_poll(dev_id,
&async_cnt,worker_info);app/test-dma-perf/benchmark.c new file +9
-0+ } + +if (worker_info->stop_flag) +break ; + } + +rte_dma_submit (dev_id,0 ); +while ((async_cnt >0 ) && (poll_cnt++ <POLL_MAX )) { + nr_cpl = rte_dma_completed(dev_id, 0,
MAX_DMA_CPL_NB, NULL, NULL);app/test-dma-perf/benchmark.c new file +3
-0+ async_cnt -= nr_cpl; + } +Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+return 0 ; +} + +static inline int +do_cpu_mem_copy (void *p) +{ +const uint16_t *para_idx = (uint16_t *)p; + volatile struct lcore_params *para =worker_params[*para_idx];
Quoted 34 lines
Quoted 1 line
Quoted 1 line
app/test-dma-perf/benchmark.c
Quoted 1 line
app/test-dma-perf/benchmark.c new file +1
-0+ volatile struct worker_info *worker_info = &(para-
worker_info);Quoted 32 lines
Quoted 23 lines
Quoted 23 lines
app/test-dma-perf/benchmark.c new file +16
-0+const uint32_t nr_buf = para->nr_buf; +const uint32_t buf_size = para->buf_size; +struct rte_mbuf **srcs = para->srcs; +struct rte_mbuf **dsts = para->dsts; +uint32_t i; + + worker_info->stop_flag = false; + worker_info->ready_flag = true; + +while (!worker_info->start_flag) + ; + +while (1 ) { +for (i =0 ; i < nr_buf; i++) { +/* copy buffer form src to dst */ + rte_memcpy((void*)(uintptr_t)rte_mbuf_data_iova(dsts[i]),
app/test-dma-perf/benchmark.c new file +1
-0+ (void*)(uintptr_t)rte_mbuf_data_iova(srcs[i]),
app/test-dma-perf/benchmark.c new file +4
-0+ (size_t)buf_size); + worker_info->total_cpl++; + } +if (worker_info->stop_flag)Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+break ; + } + +return 0 ; +} + +static int +setup_memory_env(struct test_configure *cfg, struct rte_mbuf***srcs,
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
app/test-dma-perf/benchmark.c
Quoted 10 lines
app/test-dma-perf/benchmark.c new file +10
-0+ struct rte_mbuf ***dsts) +{ +unsigned int buf_size = cfg->buf_size.cur; +unsigned int nr_sockets; +uint32_t nr_buf = cfg->nr_buf; + + nr_sockets =rte_socket_count (); +if (cfg->src_numa_node >= nr_sockets || + cfg->dst_numa_node >= nr_sockets) { + printf("Error: Source or destination numa exceedsthe acture
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
app/test-dma-perf/benchmark.c
Quoted 5 lines
numa nodes.\n");
app/test-dma-perf/benchmark.c new file +4
-0+return -1 ; + } + + src_pool =rte_pktmbuf_pool_create("Benchmark_DMA_SRC",
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
app/test-dma-perf/benchmark.c
Quoted 7 lines
app/test-dma-perf/benchmark.c new file +7
-0+ nr_buf,/* n == num elements */ +64 ,/* cache size */ +0 ,/* priv size */ + buf_size +RTE_PKTMBUF_HEADROOM , + cfg->src_numa_node); +if (src_pool ==NULL ) { + PRINT_ERR("Error with source mempoolcreation.\n");
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
app/test-dma-perf/benchmark.c
Quoted 4 lines
app/test-dma-perf/benchmark.c new file +4
-0+return -1 ; + } + + dst_pool =rte_pktmbuf_pool_create("Benchmark_DMA_DST",
Quoted 10 lines
Quoted 10 lines
Quoted 5 lines
app/test-dma-perf/benchmark.c
Quoted 2 lines
app/test-dma-perf/benchmark.c new file +2
-0+ nr_buf,/* n == num elements */ + 64, /* cache size */[Anoob] We do not alloc or free pointers in the datapath, right? So why bother with cache?
[Cheng] Yes, you are right, the cache size is not necessary here, I'll fix it in the next version.
Quoted 1 line
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
app/test-dma-perf/benchmark.c
Quoted 5 lines
app/test-dma-perf/benchmark.c new file +5
-0+0 ,/* priv size */ + buf_size +RTE_PKTMBUF_HEADROOM , + cfg->dst_numa_node); +if (dst_pool ==NULL ) { + PRINT_ERR("Error with destination mempoolcreation.\n");
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
app/test-dma-perf/benchmark.c
Quoted 4 lines
app/test-dma-perf/benchmark.c new file +4
-0+return -1 ; + } + + *srcs = rte_malloc(NULL, nr_buf * sizeof(struct rte_mbuf *),0);
Quoted 9 lines
Quoted 9 lines
Quoted 6 lines
app/test-dma-perf/benchmark.c line 339
Quoted 4 lines
app/test-dma-perf/benchmark.c new file +4
-0+if (*srcs ==NULL ) { +printf ("Error: srcs malloc failed.\n" ); +return -1 ; + }[Anoob] Are we freeing these memory? The ones allocated with
rte_malloc.[Cheng] yes, we freed the memory in the end of
mem_copy_benchmark()Quoted 10 lines
Quoted 1 line
when we finished the test.
[Anoob] I think we are not freeing this mem. In the place where we free all mem, we do free all objects to mempool as well as the mempools. But this memory is to hold the pointers, right? Is that getting freed anywhere?
Also, in the mem clearing paths, do we need to clear the static variables (ie, set srcs,
src_pool, dsts,dst_poolto NULL) so that there won't be any scope for any double free.[Cheng] My apologies for the misunderstanding earlier. I now understand your point that you are right, the memory used to store the pointers is not being freed. I will fix this issue in the next version. Regarding the static variables you mentioned, I agree with your view that they should be cleared. I will address this in the upcoming version as well. Thank you very much for the feedback. It is greatly appreciated.
Quoted 2 lines
In addition, I think we also need to nullify these variables when initializing them to ensure safety and standardization of use. What do you think?
[Anoob] Since these are static variables, it is probably okay to skip the init part. But when we use it, we should clear it after use.
Please check above. I've posted one more comment. In case you missed.
Quoted 1,095 lines
Thanks!
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
app/test-dma-perf/benchmark.c
Quoted 4 lines
app/test-dma-perf/benchmark.c new file +2
-0+ + *dsts = rte_malloc(NULL, nr_buf * sizeof(struct rte_mbuf *),0);
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
app/test-dma-perf/benchmark.c
Quoted 6 lines
app/test-dma-perf/benchmark.c new file +6
-0+if (*dsts ==NULL ) { +printf ("Error: dsts malloc failed.\n" ); +return -1 ; + } + + if (rte_mempool_get_bulk(src_pool, (void **)*srcs, nr_buf)!= 0) {
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
app/test-dma-perf/benchmark.c
Quoted 4 lines
app/test-dma-perf/benchmark.c new file +4
-0+printf ("get src mbufs failed.\n" ); +return -1 ; + } + if (rte_mempool_get_bulk(dst_pool, (void **)*dsts, nr_buf)!= 0) {
Quoted 41 lines
Quoted 41 lines
Quoted 41 lines
Quoted 41 lines
app/test-dma-perf/benchmark.c new file +18
-0+printf ("get dst mbufs failed.\n" ); +return -1 ; + } + +return 0 ; +} + +void +mem_copy_benchmark (struct test_configure *cfg,bool is_dma) { +uint16_t i; +uint32_t offset; +unsigned int lcore_id =0 ; +struct rte_mbuf **srcs =NULL , **dsts =NULL ; +struct lcore_dma_map_t *ldm = &cfg->lcore_dma_map; +unsigned int buf_size = cfg->buf_size.cur; +uint16_t kick_batch = cfg->kick_batch.cur; +uint32_t nr_buf = cfg->nr_buf = (cfg->mem_size.cur *1024 * +1024) /(
cfg->buf_size.cur* 2);app/test-dma-perf/benchmark.c new file +22
-0+uint16_t nb_workers = ldm->cnt; +uint16_t test_secs = cfg->test_secs; +float memory; +uint32_t avg_cycles =0 ; +float mops; +float bandwidth; + +if (setup_memory_env (cfg, &srcs, &dsts) <0 ) + goto out; + +if (is_dma) +if (config_dmadevs (cfg) <0 ) + goto out; + +if (cfg->cache_flush) { +cache_flush_buf (srcs, buf_size, nr_buf); +cache_flush_buf (dsts, buf_size, nr_buf); +rte_mb (); + } + +printf ("Start testing....\n" ); +Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +5
-0+for (i =0 ; i < nb_workers; i++) { + lcore_id = ldm->lcores[i]; + offset = nr_buf / nb_workers * i; + + worker_params[i] = rte_malloc(NULL, sizeof(struct
lcore_params), 0);app/test-dma-perf/benchmark.c new file +2
-0+if (!worker_params[i]) { + printf("lcore parameters malloc failure forlcore
Quoted 15 lines
Quoted 6 lines
Quoted 5 lines
app/test-dma-perf/benchmark.c
Quoted 3 lines
%d\n",
lcore_id);app/test-dma-perf/benchmark.c new file +2
-0+break ; + }[Anoob] Are we freeing the above memory?
Quoted 8 lines
[Cheng] sorry, I missed that, I'll add
worker_paramsmemory free in the next version, thanks.Quoted 5 lines
app/test-dma-perf/benchmark.c
Quoted 5 lines
app/test-dma-perf/benchmark.c new file +2
-0+if (is_dma) { + worker_params[i]->dma_name = ldm-Quoted 1 line
dma_names[i];
app/test-dma-perf/benchmark.c new file +1
-0+ worker_params[i]->dev_id = ldm-dma_ids[i];
Quoted 14 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
app/test-dma-perf/benchmark.c new file +4
-0+ worker_params[i]->kick_batch = kick_batch; + } + worker_params[i]->worker_id = i; + worker_params[i]->nr_buf = (uint32_t)(nr_buf /
nb_workers);app/test-dma-perf/benchmark.c new file +1
-0+ worker_params[i]->buf_size = buf_size;Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c line 420
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+ worker_params[i]->test_secs = test_secs; + worker_params[i]->srcs = srcs + offset; + worker_params[i]->dsts = dsts + offset; + worker_params[i]->scenario_id = cfg->scenario_id; + worker_params[i]->lcore_id = lcore_id; + +if (is_dma) +rte_eal_remote_launch(do_dma_mem_copy, (void
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
app/test-dma-perf/benchmark.c
Quoted 3 lines
*)(&i),
lcore_id);app/test-dma-perf/benchmark.c new file +2
-0+else + rte_eal_remote_launch(do_cpu_mem_copy,(void
Quoted 31 lines
Quoted 7 lines
Quoted 7 lines
app/test-dma-perf/benchmark.c
Quoted 7 lines
*)(&i),
lcore_id);app/test-dma-perf/benchmark.c new file +6
-0+ } + +while (1 ) { +bool ready = true; +for (i =0 ; i < nb_workers; i++) { + if (worker_params[i]-worker_info.ready_flag ==
Quoted 23 lines
Quoted 23 lines
Quoted 23 lines
false) {
app/test-dma-perf/benchmark.c new file +19
-0+ ready =0 ; +break ; + } + } +if (ready) +break ; + } + +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.start_flag = true; + +usleep (TEST_WAIT_U_SECOND ); +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.test_cpl = +worker_params[i]->worker_info.total_cpl; + +usleep (test_secs *1000 *1000 ); +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.test_cpl =worker_params[i]->worker_info.total_cpl -
app/test-dma-perf/benchmark.c new file +1
-0+ worker_params[i]-Quoted 1 line
worker_info.test_cpl;
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
app/test-dma-perf/benchmark.c
Quoted 10 lines
app/test-dma-perf/benchmark.c new file +10
-0+ +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.stop_flag = true; + +rte_eal_mp_wait_lcore (); + +for (i =0 ; i < nb_workers; i++) { + calc_result(buf_size, nr_buf, nb_workers, test_secs, + worker_params[i]->worker_info.test_cpl, + &memory, &avg_cycles, &bandwidth,&mops);
Quoted 6 lines
Quoted 1 line
Quoted 1 line
app/test-dma-perf/benchmark.c
Quoted 1 line
app/test-dma-perf/benchmark.c new file +1
-0+ output_result(cfg->scenario_id, worker_params[i]-
lcore_id,Quoted 1 line
Quoted 1 line
app/test-dma-perf/benchmark.c
Quoted 1 line
app/test-dma-perf/benchmark.c new file +1
-0+ worker_params[i]-
dma_name,Quoted 2 lines
Quoted 2 lines
app/test-dma-perf/benchmark.c
Quoted 2 lines
avg_cycles,buf_size,app/test-dma-perf/benchmark.c new file +1
-0+ nr_buf / nb_workers,memory,
Quoted 34 lines
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
bandwidth, mops,
is_dma);app/test-dma-perf/benchmark.c new file +9
-0+ } + +out: +/* free env */ +if (srcs) +rte_pktmbuf_free_bulk (srcs, nr_buf); +if (dsts) +rte_pktmbuf_free_bulk (dsts, nr_buf); +Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+if (src_pool) +rte_mempool_free (src_pool); +if (dst_pool) +rte_mempool_free (dst_pool); + +if (is_dma) { +for (i =0 ; i < nb_workers; i++) { + printf("Stopping dmadev %d\n", ldm-dma_ids[i]);
Quoted 15 lines
Quoted 6 lines
Quoted 6 lines
app/test-dma-perf/benchmark.c new file +4
-0+rte_dma_stop (ldm->dma_ids[i]); + } + } +}(quoted diff) +0
-0b/app/test-dma-perf/config.ini new file mode 100644 index
Quoted 8 lines
app/test-dma-perf/config.ini line 4
Quoted 8 lines
0000000000..2fd9c3c387
app/test-dma-perf/config.ini new file +4
-0@@ -0,0 +1,59 @@ + +; This is an example configuration file for dma-perf, which +details the meanings of each parameter ; and instructions on how +to use dma-perf.
Quoted 16 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/config.ini new file +8
-0+ +; Supported test types are DMA_MEM_COPY and CPU_MEM_COPY. + +; Parameters: +; "mem_size" denotes the size of the memory footprint. +; "buf_size" denotes the memory size of a single operation. +; "dma_ring_size" denotes the dma ring buffer size. It should be +greaterQuoted 8 lines
Quoted 8 lines
app/test-dma-perf/config.ini
Quoted 8 lines
than 64 normally.
app/test-dma-perf/config.ini new file +2
-0+; "kick_batch" denotes the dma operation batch size, and should +be greaterthan 1 normally.
app/test-dma-perf/config.ini new file +4
-0+ +; The format for variables is variable=first,last,increment,ADD|MUL. + +; src_numa_node is used to control the numa node where thesource
Quoted 8 lines
Quoted 7 lines
Quoted 7 lines
app/test-dma-perf/config.ini
Quoted 7 lines
memory is allocated.
app/test-dma-perf/config.ini new file +2
-0+; dst_numa_node is used to control the numa node where the +destinationmemory is allocated.
app/test-dma-perf/config.ini new file +3
-0+ +; cache_flush is used to determine whether or not the cache +should be flushed, with 1 indicating to ; flush and 0 indicating to notflush.
Quoted 12 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
app/test-dma-perf/config.ini new file +4
-0+ +; test_seconds controls the test time of the whole case. + +; To use DMA for a test, please specify the "lcore_dma" parameter.Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/config.ini
Quoted 8 lines
app/test-dma-perf/config.ini new file +3
-0+; If you have already set the "-l" and "-a" parameters using EAL, +; make sure that the value of "lcore_dma" falls within their +range of thevalues.
app/test-dma-perf/config.ini new file +4
-0+ +; To use CPU for a test, please specify the "lcore" parameter. +; If you have already set the "-l" and "-a" parameters using EAL, +; make sure that the value of "lcore" falls within their range ofvalues.
Quoted 53 lines
Quoted 37 lines
Quoted 28 lines
Quoted 14 lines
app/test-dma-perf/config.ini new file +3
-0+ +; To specify a configuration file, use the "--config" flag +followed by the pathto the file.
app/test-dma-perf/config.ini new file +3
-0+ +; To specify a result file, use the "--result" flag followed by +the path to thefile.
app/test-dma-perf/config.ini new file +3
-0+; If you do not specify a result file, one will be generated with +the same name as the configuration ; file, with the addition of +"_result.csv" atthe end.
app/test-dma-perf/config.ini new file +2
-0+ +[case1]app/test-dma-perf/config.ini line 47
Quoted 10 lines
app/test-dma-perf/config.ini new file +10
-0+type=DMA_MEM_COPY +mem_size=10 +buf_size=64,8192,2,MUL +dma_ring_size=1024 +kick_batch=32 +src_numa_node=0 +dst_numa_node=0 +cache_flush=0 +test_seconds=2 +lcore_dma=lcore10@0000:00:04.2, lcore11@0000:00:04.3[Anoob] Isn't it better if we allow user to specify DMA dev ID rather than the PCI DBDF?
Quoted 7 lines
In the long run, I would expect config file to provide {core,
dma_dev_id,queue_id}Another thought is why to expose this at all? If we can restrict this perf application to have one thread only use one vchan, then application can easily create this mapping in run time. Unless you want one thread to use 2 different vchans which may not be desirable
since this is a standalone perf app.
Quoted 15 lines
[Cheng] Thank you for the feedback.
Here are my thoughts:
Firstly, the user may not know which device the DMA dev ID corresponds to, or which NUMA node it is on. In my example, I used the CBDMA environment, so I did not specify the work queue ID. When using DSA, the configuration would be something like lcore10@0000:00:04.2-q0 which contains core, dma and work queue id. The reason for exposing these options is that we want the user to fully understand which cores and devices are being used so that they know exactly where the performance data is coming from. For example, performance when cores and DMA devices are not on the same NUMA node, etc. This allows the testing scenario to be precise and flexible. If the application handles the mapping itself, the user loses control over the mapping and may not get the performance data they want. We believe control should be given to the user rather than the application.Quoted 2 lines
[Anoob] I understand your view points. Thanks for the explanation.
[Cheng] sure, no problem.
Quoted 86 lines
Quoted 86 lines
Quoted 86 lines
Quoted 86 lines
app/test-dma-perf/config.ini new file +12
-0+eal_args=--in-memory --file-prefix=test + +[case2] +type=CPU_MEM_COPY +mem_size=10 +buf_size=64,8192,2,MUL +src_numa_node=0 +dst_numa_node=1 +cache_flush=0 +test_seconds=2 +lcore = 3, 4 +eal_args=--in-memory --no-pciapp/test-dma-perf/main.c new file +21
-0@@ -0,0 +1,569 @@ +/* SPDX-License-Identifier: BSD-3-Clause +* Copyright(c) 2023 Intel Corporation */ + +#include <stdio.h> +#include <stdlib.h> +#include <getopt.h> +#include <signal.h> +#include <stdbool.h> +#include <unistd.h> +#include <sys/wait.h> +#include <inttypes.h> +#include <libgen.h> + +#include <rte_eal.h> +#include <rte_cfgfile.h> +#include <rte_string_fns.h> +#include <rte_lcore.h> + +#include "main.h" + +#define CSV_HDR_FMT "Case %u : %s,lcore,DMA,buffersize,nr_buf,memory(MB),cycle,bandwidth(Gbps),MOps\n"
app/test-dma-perf/main.c new file +45
-0+ +#define MAX_EAL_PARAM_NB 100 +#define MAX_EAL_PARAM_LEN 1024 + +#define DMA_MEM_COPY "DMA_MEM_COPY" +#define CPU_MEM_COPY "CPU_MEM_COPY" + +#define CMDLINE_CONFIG_ARG "--config" +#define CMDLINE_RESULT_ARG "--result" + +#define MAX_PARAMS_PER_ENTRY 4 + +#define MAX_LONG_OPT_SZ 64 + +enum { +TEST_TYPE_NONE =0 , +TEST_TYPE_DMA_MEM_COPY , +TEST_TYPE_CPU_MEM_COPY +}; + +#define MAX_TEST_CASES 16 +static struct test_configure test_cases[MAX_TEST_CASES ]; + +char output_str[MAX_WORKER_NB ][MAX_OUTPUT_STR_LEN ]; + +static FILE *fd; + +static void +output_csv (bool need_blankline) +{ +uint32_t i; + +if (need_blankline) { +fprintf (fd,",,,,,,,,\n" ); +fprintf (fd,",,,,,,,,\n" ); + } + +for (i =0 ; i <RTE_DIM (output_str); i++) { +if (output_str[i][0 ]) { +fprintf (fd,"%s" , output_str[i]); + output_str[i][0 ] ='\0' ; + } + } + +fflush (fd);Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/main.c
Quoted 8 lines
app/test-dma-perf/main.c new file +6
-0+} + +static void +output_env_info (void ) +{ + snprintf(output_str[0], MAX_OUTPUT_STR_LEN, "testenvironment:\n");
app/test-dma-perf/main.c new file +1
-0+ snprintf(output_str[1], MAX_OUTPUT_STR_LEN, "CPUfrequency,%"
Quoted 96 lines
Quoted 9 lines
Quoted 9 lines
app/test-dma-perf/main.c
Quoted 9 lines
app/test-dma-perf/main.c new file +9
-0+ PRIu64 "\n", rte_get_timer_hz()); + +output_csv (true); +} + +static void +output_header (uint32_t case_id,struct test_configure *case_cfg) { + snprintf(output_str[0], MAX_OUTPUT_STR_LEN, + CSV_HDR_FMT, case_id, case_cfg-
test_type_str);Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
app/test-dma-perf/main.c new file +7
-0+ +output_csv (true); +} + +static void +run_test_case (struct test_configure *case_cfg) { +switch (case_cfg->test_type) {Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/main.c
Quoted 8 lines
app/test-dma-perf/main.c new file +8
-0+case TEST_TYPE_DMA_MEM_COPY : +mem_copy_benchmark (case_cfg, true); +break ; +case TEST_TYPE_CPU_MEM_COPY : +mem_copy_benchmark (case_cfg, false); +break ; +default : + printf("Unknown test type. %s\n", case_cfg-
test_type_str);Quoted 9 lines
Quoted 9 lines
app/test-dma-perf/main.c
Quoted 9 lines
app/test-dma-perf/main.c new file +9
-0+break ; + } +} + +static void +run_test (uint32_t case_id,struct test_configure *case_cfg) { +uint32_t i; +uint32_t nb_lcores =rte_lcore_count (); + struct test_configure_entry *mem_size = &case_cfg-
mem_size;Quoted 1 line
Quoted 1 line
app/test-dma-perf/main.c
Quoted 1 line
app/test-dma-perf/main.c new file +1
-0+ struct test_configure_entry *buf_size = &case_cfg-
buf_size;Quoted 1 line
Quoted 1 line
app/test-dma-perf/main.c
Quoted 1 line
app/test-dma-perf/main.c new file +1
-0+ struct test_configure_entry *ring_size = &case_cfg-
ring_size;Quoted 1 line
Quoted 1 line
app/test-dma-perf/main.c
Quoted 1 line
app/test-dma-perf/main.c new file +1
-0+ struct test_configure_entry *kick_batch = &case_cfg-
kick_batch;Quoted 31 lines
Quoted 31 lines
Quoted 31 lines
app/test-dma-perf/main.c new file +31
-0+struct test_configure_entry dummy = {0 }; +struct test_configure_entry *var_entry = &dummy; + +for (i =0 ; i <RTE_DIM (output_str); i++) +memset (output_str[i],0 ,MAX_OUTPUT_STR_LEN ); + +if (nb_lcores <= case_cfg->lcore_dma_map.cnt) { +printf ("Case %u: Not enough lcores.\n" , case_id); +return ; + } + +printf ("Number of used lcores: %u.\n" , nb_lcores); + +if (mem_size->incr !=0 ) + var_entry = mem_size; + +if (buf_size->incr !=0 ) + var_entry = buf_size; + +if (ring_size->incr !=0 ) + var_entry = ring_size; + +if (kick_batch->incr !=0 ) + var_entry = kick_batch; + + case_cfg->scenario_id =0 ; + +output_header (case_id, case_cfg); + + for (var_entry->cur = var_entry->first; var_entry->cur <= +var_entry-Quoted 3 lines
Quoted 3 lines
app/test-dma-perf/main.c
Quoted 3 lines
Quoted 1 line
last;) {
app/test-dma-perf/main.c new file +2
-0+ case_cfg->scenario_id++; + printf("\nRunning scenario %d\n", case_cfg-
scenario_id);Quoted 19 lines
Quoted 19 lines
Quoted 19 lines
app/test-dma-perf/main.c new file +19
-0+ +run_test_case (case_cfg); +output_csv (false); + +if (var_entry->op ==OP_ADD ) + var_entry->cur += var_entry->incr; +else if (var_entry->op ==OP_MUL ) + var_entry->cur *= var_entry->incr; +else +break ; + } +} + +static int +parse_lcore (struct test_configure *test_case,const char *value) { +size_t len =strlen (value); +char *input = (char *)malloc ((len +1 ) *sizeof (char )); +strcpy (input, value); + struct lcore_dma_map_t *lcore_dma_map = &(test_case-Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
app/test-dma-perf/main.c
Quoted 6 lines
Quoted 1 line
lcore_dma_map);app/test-dma-perf/main.c new file +5
-0+ +if (test_case ==NULL || value ==NULL ) +return -1 ; + + memset(lcore_dma_map, 0, sizeof(struct
lcore_dma_map_t));Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
app/test-dma-perf/main.c
Quoted 10 lines
app/test-dma-perf/main.c new file +10
-0+ +char *token =strtok (input,", " ); +while (token !=NULL ) { +if (lcore_dma_map->cnt >=MAX_LCORE_NB ) { +free (input); +return -1 ; + } + +uint16_t lcore_id =atoi (token); + lcore_dma_map->lcores[lcore_dma_map->cnt++] =
lcore_id;Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
app/test-dma-perf/main.c
Quoted 10 lines
app/test-dma-perf/main.c new file +10
-0+ + token =strtok (NULL ,", " ); + } + +free (input); +return 0 ; +} + +static int +parse_lcore_dma(struct test_configure *test_case, const char*value)
Quoted 40 lines
{
Quoted 39 lines
Quoted 39 lines
Quoted 39 lines
app/test-dma-perf/main.c new file +24
-0+struct lcore_dma_map_t *lcore_dma_map; +char *input =strndup (value,strlen (value) +1 ); +char *addrs = input; +char *ptrs[2 ]; +char *start, *end, *substr; +uint16_t lcore_id; +int ret =0 ; + +while (*addrs =='\0' ) + addrs++; +if (*addrs =='\0' ) { +fprintf (stderr,"No input DMA addresses\n" ); + ret =-1 ; + goto out; + } + + substr =strtok (addrs,"," ); +if (substr ==NULL ) { +fprintf (stderr,"No input DMA address\n" ); + ret =-1 ; + goto out; + } + + memset(&test_case->lcore_dma_map, 0, sizeof(struct
lcore_dma_map_t));app/test-dma-perf/main.c new file +14
-0+ + do { + rte_strsplit(substr, strlen(substr), ptrs, 2, '@'); + + start = strstr(ptrs[0], "lcore"); + if (start == NULL) { + fprintf(stderr, "Illegal lcore\n"); + ret = -1; + break; + } + + start += 5; + lcore_id = strtol(start, &end, 0); + if (end == start) {Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/main.c
Quoted 8 lines
app/test-dma-perf/main.c new file +1
-0+ fprintf(stderr, "No input lcore ID or ID %d iswrong\n",
lcore_id);app/test-dma-perf/main.c new file +6
-0+ ret =-1 ; +break ; + } + + lcore_dma_map = &test_case->lcore_dma_map; + lcore_dma_map->lcores[lcore_dma_map->cnt] =
lcore_id;Quoted 14 lines
Quoted 1 line
Quoted 1 line
app/test-dma-perf/main.c
Quoted 1 line
app/test-dma-perf/main.c new file +1
-0+ strcpy(lcore_dma_map-dma_names[lcore_dma_map-
Quoted 12 lines
Quoted 12 lines
app/test-dma-perf/main.c line 248
Quoted 12 lines
Quoted 1 line
cnt], ptrs[1]);
app/test-dma-perf/main.c new file +11
-0+ lcore_dma_map->cnt++; + substr =strtok (NULL ,"," ); + }while (substr !=NULL ); + +out: +free (input); +return ret; +} + +static int +parse_entry (const char *value,struct test_configure_entry *entry){
Quoted 54 lines
Quoted 54 lines
Quoted 54 lines
Quoted 54 lines
app/test-dma-perf/main.c new file +12
-0+char input[255 ] = {0 }; +char *args[MAX_PARAMS_PER_ENTRY ]; +int args_nr =-1 ; + +if (value ==NULL || entry ==NULL ) + goto out; + +strncpy (input, value,254 ); +if (*input =='\0' ) + goto out; + + args_nr = rte_strsplit(input, strlen(input), args,
MAX_PARAMS_PER_ENTRY, ',');app/test-dma-perf/main.c new file +35
-0+if (args_nr !=1 && args_nr !=4 ) + goto out; + + entry->cur = entry->first = (uint32_t )atoi (args[0 ]); + +if (args_nr ==4 ) { + entry->last = (uint32_t )atoi (args[1 ]); + entry->incr = (uint32_t )atoi (args[2 ]); +if (!strcmp (args[3 ],"MUL" )) + entry->op =OP_MUL ; +else if (!strcmp (args[3 ],"ADD" )) + entry->op =OP_ADD ; +else { +printf ("Invalid op %s.\n" , args[3 ]); + args_nr =-1 ; + } + }else { + entry->op =OP_NONE ; + entry->last =0 ; + entry->incr =0 ; + } +out: +return args_nr; +} + +static uint16_t +load_configs (const char *path) +{ +struct rte_cfgfile *cfgfile; +int nb_sections, i; +struct test_configure *test_case; +char section_name[CFG_NAME_LEN ]; +const char *case_type; +const char *lcore_dma; + const char *mem_size_str, *buf_size_str, *ring_size_str,
*kick_batch_str;app/test-dma-perf/main.c new file +5
-0+int args_nr, nb_vp; +bool is_dma; + +printf ("config file parsing...\n" ); + cfgfile =rte_cfgfile_load (path,0 );Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/main.c
Quoted 8 lines
app/test-dma-perf/main.c new file +8
-0+if (!cfgfile) { +printf ("Open configure file error.\n" ); +exit (1 ); + } + + nb_sections =rte_cfgfile_num_sections (cfgfile,NULL ,0 ); +if (nb_sections >MAX_TEST_CASES ) { + printf("Error: The maximum number of cases is%d.\n",
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
app/test-dma-perf/main.c
Quoted 6 lines
MAX_TEST_CASES);app/test-dma-perf/main.c new file +5
-0+exit (1 ); + } + +for (i =0 ; i < nb_sections; i++) { + snprintf(section_name, CFG_NAME_LEN, "case%d", i+ 1);
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
app/test-dma-perf/main.c
Quoted 2 lines
app/test-dma-perf/main.c new file +2
-0+ test_case = &test_cases[i]; + case_type = rte_cfgfile_get_entry(cfgfile,
section_name,Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
app/test-dma-perf/main.c
Quoted 3 lines
"type");
app/test-dma-perf/main.c new file +2
-0+if (!case_type) { + printf("Error: No case type in case %d, thetest will be
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
app/test-dma-perf/main.c
Quoted 10 lines
finished here.\n",
app/test-dma-perf/main.c new file +7
-0+ i + 1); + test_case->is_valid = false; +continue ; + } + +if (strcmp (case_type,DMA_MEM_COPY ) ==0 ) { + test_case->test_type =
TEST_TYPE_DMA_MEM_COPY;app/test-dma-perf/main.c new file +1
-0+ test_case->test_type_str =
DMA_MEM_COPY;Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
app/test-dma-perf/main.c
Quoted 5 lines
app/test-dma-perf/main.c new file +3
-0+ is_dma = true; + }else if (strcmp (case_type,CPU_MEM_COPY ) ==0 ) { + test_case->test_type =
TEST_TYPE_CPU_MEM_COPY;app/test-dma-perf/main.c new file +1
-0+ test_case->test_type_str =
CPU_MEM_COPY;Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
app/test-dma-perf/main.c
Quoted 3 lines
app/test-dma-perf/main.c new file +3
-0+ is_dma = false; + }else { + printf("Error: Cannot find case type %s incase%d.\n",
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
case_type, i + 1);app/test-dma-perf/main.c new file +7
-0+ test_case->is_valid = false; +continue ; + } + + nb_vp =0 ; + + test_case->src_numa_node =(int)atoi(rte_cfgfile_get_entry(cfgfile,
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/main.c
Quoted 8 lines
+
section_name, "src_numa_node"));app/test-dma-perf/main.c new file +1
-0+ test_case->dst_numa_node =(int)atoi(rte_cfgfile_get_entry(cfgfile,
+section_name, "dst_numa_node"));app/test-dma-perf/main.c new file +2
-0+ + mem_size_str = rte_cfgfile_get_entry(cfgfile,
section_name,Quoted 11 lines
Quoted 11 lines
Quoted 11 lines
app/test-dma-perf/main.c
Quoted 11 lines
"
mem_size");app/test-dma-perf/main.c new file +1
-0+ args_nr = parse_entry(mem_size_str, &test_case-Quoted 1 line
mem_size);app/test-dma-perf/main.c new file +8
-0+if (args_nr <0 ) { +printf ("parse error in case %d.\n" , i +1 ); + test_case->is_valid = false; +continue ; + }else if (args_nr >1 ) + nb_vp++; + + buf_size_str = rte_cfgfile_get_entry(cfgfile,
section_name,Quoted 16 lines
Quoted 2 lines
Quoted 2 lines
app/test-dma-perf/main.c
Quoted 2 lines
"
buf_size");app/test-dma-perf/main.c new file +1
-0+ args_nr = parse_entry(buf_size_str, &test_case-
buf_size);Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
app/test-dma-perf/main.c new file +5
-0+if (args_nr <0 ) { +printf ("parse error in case %d.\n" , i +1 ); + test_case->is_valid = false; +continue ; + }else if (args_nr >1 )Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/main.c
Quoted 8 lines
app/test-dma-perf/main.c new file +4
-0+ nb_vp++; + +if (is_dma) { + ring_size_str = rte_cfgfile_get_entry(cfgfile,
section_name,
+"dma_ring_size");app/test-dma-perf/main.c new file +1
-0+ args_nr = parse_entry(ring_size_str,&test_case-
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
app/test-dma-perf/main.c
Quoted 3 lines
Quoted 1 line
ring_size);app/test-dma-perf/main.c new file +2
-0+if (args_nr <0 ) { + printf("parse error in case %d.\n", i +1);
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
app/test-dma-perf/main.c
Quoted 6 lines
app/test-dma-perf/main.c new file +6
-0+ test_case->is_valid = false; +continue ; + }else if (args_nr >1 ) + nb_vp++; + + kick_batch_str =rte_cfgfile_get_entry(cfgfile,
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
app/test-dma-perf/main.c
Quoted 2 lines
section_name, "kick_batch");app/test-dma-perf/main.c new file +1
-0+ args_nr = parse_entry(kick_batch_str,&test_case-
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
app/test-dma-perf/main.c
Quoted 3 lines
Quoted 1 line
kick_batch);app/test-dma-perf/main.c new file +2
-0+if (args_nr <0 ) { + printf("parse error in case %d.\n", i +1);
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines
app/test-dma-perf/main.c
Quoted 11 lines
app/test-dma-perf/main.c new file +6
-0+ test_case->is_valid = false; +continue ; + }else if (args_nr >1 ) + nb_vp++; + + lcore_dma = rte_cfgfile_get_entry(cfgfile,
section_name, "lcore_dma");app/test-dma-perf/main.c new file +1
-0+ int lcore_ret = parse_lcore_dma(test_case,
lcore_dma);app/test-dma-perf/main.c new file +2
-0+if (lcore_ret <0 ) { + printf("parse lcore dma error in case%d.\n", i
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
1);
app/test-dma-perf/main.c
Quoted 7 lines
app/test-dma-perf/main.c new file +5
-0+ test_case->is_valid = false; +continue ; + } + }else { + lcore_dma = rte_cfgfile_get_entry(cfgfile,
section_name, "lcore");app/test-dma-perf/main.c new file +1
-0+ int lcore_ret = parse_lcore(test_case,
lcore_dma);Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
app/test-dma-perf/main.c
Quoted 2 lines
app/test-dma-perf/main.c new file +2
-0+if (lcore_ret <0 ) { + printf("parse lcore error in case%d.\n", i + 1);
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
app/test-dma-perf/main.c
Quoted 7 lines
app/test-dma-perf/main.c new file +7
-0+ test_case->is_valid = false; +continue ; + } + } + +if (nb_vp >1 ) { + printf("Error, each section can only have asingle
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
app/test-dma-perf/main.c
Quoted 7 lines
variable parameter.\n");
app/test-dma-perf/main.c new file +6
-0+ test_case->is_valid = false; +continue ; + } + + test_case->cache_flush = + (int)atoi(rte_cfgfile_get_entry(cfgfile,
section_name,Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
app/test-dma-perf/main.c
Quoted 4 lines
"
cache_flush"));app/test-dma-perf/main.c new file +1
-0+ test_case->test_secs =(uint16_t)atoi(rte_cfgfile_get_entry(cfgfile,
app/test-dma-perf/main.c new file +1
-0+ section_name,"
test_seconds"));Quoted 14 lines
Quoted 14 lines
Quoted 14 lines
Quoted 14 lines
app/test-dma-perf/main.c new file +2
-0+ + test_case->eal_args = rte_cfgfile_get_entry(cfgfile,
section_name, "eal_args");app/test-dma-perf/main.c new file +11
-0+ test_case->is_valid = true; + } + +rte_cfgfile_close (cfgfile); +printf ("config file parsing complete.\n\n" ); +return i; +} + +/* Parse the argument given in the command line of the +application */ static int append_eal_args (int argc,char **argv, +const char *eal_args,char **new_argv) {Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/main.c
Quoted 8 lines
app/test-dma-perf/main.c new file +8
-0+int i; +char *tokens[MAX_EAL_PARAM_NB ]; +char args[MAX_EAL_PARAM_LEN ] = {0 }; +int token_nb, new_argc =0 ; + +for (i =0 ; i < argc; i++) { + if ((strcmp(argv[i], CMDLINE_CONFIG_ARG) == 0) || + (strcmp(argv[i],
CMDLINE_RESULT_ARG) ==Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
0)) {
app/test-dma-perf/main.c new file +4
-0+ i++; +continue ; + } + strlcpy(new_argv[new_argc], argv[i],Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/main.c
Quoted 8 lines
sizeof(new_argv[new_argc]));app/test-dma-perf/main.c new file +7
-0+ new_argc++; + } + +if (eal_args) { +strlcpy (args, eal_args,sizeof (args)); + token_nb = rte_strsplit(args, strlen(args), + tokens,
MAX_EAL_PARAM_NB, ' ');Quoted 35 lines
Quoted 35 lines
Quoted 35 lines
Quoted 35 lines
app/test-dma-perf/main.c new file +30
-0+for (i =0 ; i < token_nb; i++) +strcpy (new_argv[new_argc++], tokens[i]); + } + +return new_argc; +} + +int +main (int argc,char *argv[]) +{ +int ret; +uint16_t case_nb; +uint32_t i, nb_lcores; +pid_t cpid, wpid; +int wstatus; +char args[MAX_EAL_PARAM_NB ][MAX_EAL_PARAM_LEN ]; +char *pargs[MAX_EAL_PARAM_NB ]; +char *cfg_path_ptr =NULL ; +char *rst_path_ptr =NULL ; +char rst_path[PATH_MAX ]; +int new_argc; +bool is_first_case = true; + +memset (args,0 ,sizeof (args)); + +for (i =0 ; i <RTE_DIM (pargs); i++) + pargs[i] = args[i]; + +for (i =0 ; i < (uint32_t )argc; i++) { + if (strncmp(argv[i], CMDLINE_CONFIG_ARG,
MAX_LONG_OPT_SZ) == 0)app/test-dma-perf/main.c new file +2
-0+ cfg_path_ptr = argv[i +1 ]; + if (strncmp(argv[i], CMDLINE_RESULT_ARG,
MAX_LONG_OPT_SZ) == 0)app/test-dma-perf/main.c new file +1
-0+ rst_path_ptr = argv[i +1 ];Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/main.c
Quoted 8 lines
app/test-dma-perf/main.c new file +8
-0+ } +if (cfg_path_ptr ==NULL ) { +printf ("Config file not assigned.\n" ); +return -1 ; + } +if (rst_path_ptr ==NULL ) { +strlcpy (rst_path, cfg_path_ptr,PATH_MAX ); + strcat(strtok(basename(rst_path), "."),"_result.csv");
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
app/test-dma-perf/main.c new file +6
-0+ rst_path_ptr = rst_path; + } + + case_nb =load_configs (cfg_path_ptr); + fd =fopen (rst_path_ptr,"w" ); +if (fd ==NULL ) {Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/main.c
Quoted 8 lines
app/test-dma-perf/main.c new file +8
-0+printf ("Open output CSV file error.\n" ); +return -1 ; + } +fclose (fd); + +for (i =0 ; i < case_nb; i++) { +if (test_cases[i].test_type ==TEST_TYPE_NONE ) { + printf("No test type in test case %d.\n\n", i +1);
Quoted 15 lines
Quoted 15 lines
Quoted 15 lines
app/test-dma-perf/main.c line 505
Quoted 8 lines
app/test-dma-perf/main.c new file +8
-0+continue ; + } +if (!test_cases[i].is_valid) { +printf ("Invalid test case %d.\n\n" , i +1 ); +continue ; + } + + cpid =fork ();[Anoob] Do we really need
fork()? Can't we use code like,RTE_LCORE_FOREACH_WORKER(lcore_id) { ret |= rte_eal_wait_lcore(lcore_id); }Quoted 5 lines
Quoted 5 lines
Quoted 1 line
to wait for all threads to exit?
[Cheng] Good question.
Fork()is used here to establish a new process for the new test case. In order for each test case to have a new EAL environment (for the flexibility), the EAL must be reinitialized for eachcase.
Quoted 5 lines
Quoted 5 lines
However, the EAL parameters can only be initialized once per process. Therefore, we use a new process to run each new test case. Moreover, each test case runs sequentially and does not affect the others, ensuring the accuracy of the performance data. Your code would wait for all threads to exit in the same process. However, it would not provide
a
Quoted 4 lines
"clean"
Quoted 3 lines
environment for each test case like
fork()does.Fork()allows us to have a fully reinitialized environment, with no impact or side effects from previous test cases. This results in clean, precise performance datafor
Quoted 8 lines
each case.
Quoted 3 lines
Please let me know your thoughts on this. And please let me know if you have any other questions or require any clarification.
[Anoob] This was just a generic observation. I do not have a strong opinion either way.
[Cheng] sure, got it.
Quoted 15 lines
Quoted 15 lines
Thanks,
ChengQuoted 11 lines
Quoted 11 lines
app/test-dma-perf/main.c new file +7
-0+if (cpid <0 ) { +printf ("Fork case %d failed.\n" , i +1 ); +exit (EXIT_FAILURE ); + }else if (cpid ==0 ) { +printf ("\nRunning case %u\n\n" , i +1 ); + + new_argc = append_eal_args(argc, argv,test_cases[i].eal_args, pargs);
app/test-dma-perf/main.c new file +2
-0+ ret =rte_eal_init (new_argc, pargs); +if (ret <0 )Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/main.c
Quoted 8 lines
app/test-dma-perf/main.c new file +1
-0+ rte_exit(EXIT_FAILURE, "Invalid EALarguments\n");
app/test-dma-perf/main.c new file +6
-0+ +/* Check lcores. */ + nb_lcores =rte_lcore_count (); +if (nb_lcores <2 ) + rte_exit(EXIT_FAILURE, + "There should be at least 2worker
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
app/test-dma-perf/main.c
Quoted 5 lines
lcores.\n");
app/test-dma-perf/main.c new file +4
-0+ + fd =fopen (rst_path_ptr,"a" ); +if (!fd) { + printf("Open output CSV fileerror.\n");
Quoted 18 lines
Quoted 18 lines
Quoted 18 lines
Quoted 18 lines
app/test-dma-perf/main.c new file +18
-0+return 0 ; + } + +if (is_first_case) { +output_env_info (); + is_first_case = false; + } +run_test (i +1 , &test_cases[i]); + +/* clean up the EAL */ +rte_eal_cleanup (); + +fclose (fd); + +printf ("\nCase %u completed.\n\n" , i +1 ); + +exit (EXIT_SUCCESS ); + }else {Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/main.c
Quoted 8 lines
app/test-dma-perf/main.c new file +8
-0+ wpid =waitpid (cpid, &wstatus,0 ); +if (wpid ==-1 ) { +printf ("waitpid error.\n" ); +exit (EXIT_FAILURE ); + } + +if (WIFEXITED (wstatus)) + printf("Case process exited. status%d\n\n",
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
app/test-dma-perf/main.c
Quoted 3 lines
app/test-dma-perf/main.c new file +3
-0+ WEXITSTATUS(wstatus)); + else if (WIFSIGNALED(wstatus)) + printf("Case process killed by signal%d\n\n",
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
app/test-dma-perf/main.c
Quoted 3 lines
app/test-dma-perf/main.c new file +3
-0+ WTERMSIG(wstatus)); + else if (WIFSTOPPED(wstatus)) + printf("Case process stopped bysignal
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
app/test-dma-perf/main.c
Quoted 4 lines
%d\n\n",
app/test-dma-perf/main.c new file +3
-0+ WSTOPSIG(wstatus)); + else if (WIFCONTINUED(wstatus)) + printf("Case processcontinued.\n\n");
Quoted 30 lines
Quoted 30 lines
Quoted 30 lines
Quoted 30 lines
app/test-dma-perf/main.c new file +2
-0+else + printf("Case process unknownterminated.\n\n");
app/test-dma-perf/main.c new file +7
-0+ } + } + +printf ("Bye...\n" ); +return 0 ; +} +app/test-dma-perf/main.h new file +15
-0@@ -0,0 +1,69 @@ +/* SPDX-License-Identifier: BSD-3-Clause +* Copyright(c) 2023 Intel Corporation */ + +#ifndef _MAIN_H_ +#define _MAIN_H_ + + +#include <rte_common.h> +#include <rte_cycles.h> +#include <rte_dev.h> +#include <rte_dmadev.h> + +#ifndef __maybe_unused +#define __maybe_unused __rte_unused +#endif Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/main.h
Quoted 8 lines
app/test-dma-perf/main.h new file +8
-0+ +#define MAX_WORKER_NB 128 +#define MAX_OUTPUT_STR_LEN 512 + +#define MAX_DMA_NB 128 +#define MAX_LCORE_NB 256 + +extern charoutput_str[MAX_WORKER_NB][MAX_OUTPUT_STR_LEN];
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
app/test-dma-perf/main.h new file +10
-0+ +typedef enum { +OP_NONE =0 , +OP_ADD , +OP_MUL +}alg_op_type ; + +struct test_configure_entry { +uint32_t first; +uint32_t last;Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/main.h
Quoted 8 lines
app/test-dma-perf/main.h new file +8
-0+uint32_t incr; +alg_op_type op; +uint32_t cur; +}; + +struct lcore_dma_map_t { +uint32_t lcores[MAX_WORKER_NB ]; +char dma_names[MAX_WORKER_NB][RTE_DEV_NAME_MAX_LEN];
Quoted 17 lines
Quoted 17 lines
Quoted 17 lines
Quoted 17 lines
app/test-dma-perf/main.h new file +17
-0+int16_t dma_ids[MAX_WORKER_NB ]; +uint16_t cnt; +}; + +struct test_configure { +bool is_valid; +uint8_t test_type; +const char *test_type_str; +uint16_t src_numa_node; +uint16_t dst_numa_node; +uint16_t opcode; +bool is_dma; +struct lcore_dma_map_t lcore_dma_map; +struct test_configure_entry mem_size; +struct test_configure_entry buf_size; +struct test_configure_entry ring_size; +struct test_configure_entry kick_batch;Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/main.h
Quoted 8 lines
app/test-dma-perf/main.h new file +8
-0+uint32_t cache_flush; +uint32_t nr_buf; +uint16_t test_secs; +const char *eal_args; +uint8_t scenario_id; +}; + +void mem_copy_benchmark(struct test_configure *cfg, bool
is_dma);Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
app/test-dma-perf/main.h new file +2
-0+ +#endif /* _MAIN_H_ */ app/test-dma- renamed from app/test-dma-perf/meson.build +0
-0perf/meson.build new file mode 100644 index
0000000000..bd6c264002
Quoted 22 lines
Quoted 22 lines
Quoted 22 lines
Quoted 22 lines
app/test-dma-perf/meson.build new file +17
-1@@ -0,0 +1,17 @@ +# SPDX-License-Identifier: BSD-3-Clause # Copyright(c) 2019-2023 +Intel Corporation + +# meson file, for building this app as part of a main DPDK build. + +if is_windows + build = false + reason = 'not supported on Windows' + subdir_done() +endif + +deps += ['dmadev', 'mbuf', 'cfgfile'] + +sources = files( + 'main.c', + 'benchmark.c', +) --2.40.1
Hi Anoob,
Replies are inline.
Thanks,
Cheng
Quoted 293 lines
-----Original Message-----
From: Anoob Joseph <[email protected]>
Sent: Thursday, June 15, 2023 11:48 PM
To: Jiang, Cheng1 <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>; Amit Prakash Shukla <[email protected]>; Ma, WenwuX <[email protected]>; Wang, YuanX <[email protected]>; He, Xingguang <[email protected]>
Subject: RE: [EXT] [PATCH v6] app/dma-perf: introduce dma-perf applicationHi Cheng,
Please see inline.
Thanks,
AnoobQuoted 4 lines
-----Original Message-----
From: Jiang, Cheng1 <[email protected]>
Sent: Thursday, June 15, 2023 7:36 PM
To: Anoob Joseph <[email protected]>Quoted 8 lines
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>; Amit Prakash Shukla <[email protected]>; Ma, WenwuX <[email protected]>; Wang, YuanX <[email protected]>; He, Xingguang <[email protected]>
Subject: RE: [EXT] [PATCH v6] app/dma-perf: introduce dma-perfapplication
Quoted 15 lines
Hi Anoob,
Replies are inline.
Thanks,
ChengQuoted 7 lines
-----Original Message-----
From: Anoob Joseph <[email protected]>
Sent: Thursday, June 15, 2023 4:45 PM
To: Jiang, Cheng1 <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia,Chenbo
Quoted 14 lines
Quoted 5 lines
<[email protected]>; Amit Prakash Shukla <[email protected]>; Ma, WenwuX <[email protected]>; Wang, YuanX <[email protected]>; He, Xingguang <[email protected]>
Subject: RE: [EXT] [PATCH v6] app/dma-perf: introduce dma-perfapplication
Quoted 8 lines
Hi Cheng,
Please see inline.
Thanks,
AnoobQuoted 10 lines
Quoted 6 lines
Quoted 6 lines
-----Original Message-----
From: Jiang, Cheng1 <[email protected]>
Sent: Thursday, June 15, 2023 1:31 PM
To: Anoob Joseph <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>;Amit
Quoted 3 lines
Quoted 3 lines
Prakash Shukla <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; Ma, WenwuX <[email protected]>;Wang,
Quoted 36 lines
Quoted 12 lines
YuanX
Quoted 11 lines
<[email protected]>; He, Xingguang <[email protected]>
Subject: RE: [EXT] [PATCH v6] app/dma-perf: introduce dma-perf applicationHi,
Thanks for your comments, the replies are inline.
Thanks,
ChengQuoted 10 lines
Quoted 6 lines
Quoted 6 lines
-----Original Message-----
From: Anoob Joseph <[email protected]>
Sent: Thursday, June 15, 2023 1:22 PM
To: Jiang, Cheng1 <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>;Amit
Quoted 3 lines
Quoted 3 lines
Prakash Shukla <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; Ma, WenwuX <[email protected]>;Wang,
Quoted 13 lines
Quoted 8 lines
YuanX
Quoted 7 lines
<[email protected]>; He, Xingguang <[email protected]>
Subject: RE: [EXT] [PATCH v6] app/dma-perf: introduce dma-perf applicationHi,
Thanks for working on the comments. Few more top level comment
inline.
Quoted 4 lines
Quoted 4 lines
Thanks,
AnoobQuoted 20 lines
Quoted 7 lines
Quoted 7 lines
Quoted 5 lines
Quoted 5 lines
-----Original Message-----
From: Cheng Jiang <[email protected]>
Sent: Tuesday, June 13, 2023 10:02 AM
To: [email protected]; [email protected]; [email protected]; [email protected]; Amit PrakashShukla
Quoted 1 line
Quoted 1 line
<[email protected]>; Anoob Joseph
Quoted 12 lines
Quoted 12 lines
Quoted 2 lines
Quoted 2 lines
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected];
Quoted 9 lines
Quoted 9 lines
Cheng Jiang <[email protected]>
Subject: [EXT] [PATCH v6] app/dma-perf: introduce dma-perf applicationExternal Email
------------------------------------------------------------------
--
-- There are many high-performance DMA devices supported inDPDK
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
now, and these DMA devices can also be integrated into other modules of DPDK as accelerators, such as Vhost. Before integrating DMA into applications, developers need to know the performance
of
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
these DMA devices in various scenarios and the performance of
CPUs
Quoted 146 lines
Quoted 45 lines
Quoted 45 lines
Quoted 6 lines
Quoted 6 lines
in the same scenario, such as different buffer lengths. Only in this way can we know the target performance of the application accelerated by using them. This patch introduces a high-performance testing tool, which supports comparing the performance of CPU and DMA in different scenarios automatically with a pre- set config file. Memory Copy performance test are
Quoted 26 lines
supported for now.
Quoted 25 lines
Signed-off-by Cheng Jiang <[email protected]>Signed-off-by Jiayu Hu <[email protected]>Signed-off-by Yuan Wang <[email protected]>Acked-by Morten Brørup <[email protected]>Acked-by Chenbo Xia <[email protected]>---
v6:
improved code based on Anoob's comments; fixed some code structure issues;
v5:
fixed someLONG_LINEwarnings;
v4:
fixed inaccuracy of the memory footprint display;
v3:
fixed some typos;
v2:
added lcore/dmadev designation;
added error case process;
removedworker_threadsparameter from config.ini; improved the logs;
improved config file;app/meson.build | 1 + app/test-dma-perf/benchmark.c | 477++++++++++++++++++++++++++++
Quoted 12 lines
Quoted 12 lines
app/test-dma-perf/config.ini | 59 ++++
app/test-dma-perf/main.c | 569 ++++++++++++++++++++++++++++++++++ app/test-dma-perf/main.h | 69 +++++ app/test-dma-perf/meson.build | 17 + 6 files changed, 1192 insertions(+) create mode 100644
app/test-dma-perf/benchmark.ccreate mode 100644 app/test-dma-perf/config.ini create mode 100644 app/test-dma-perf/main.ccreate mode 100644app/test-dma-perf/main.hcreate mode
100644 app/test-dma-perf/meson.buildQuoted 14 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
app/meson.build index renamed from app/meson.build +0
-0
74d2420f67..4fc1a83eba100644app/meson.build +0
-0@@ -19,6 +19,7 @@ apps = [ 'test-cmdline',Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/meson.build +1
-0'test-compress-perf', 'test-crypto-perf', + 'test-dma-perf', 'test-eventdev', 'test-fib', 'test-flow-perf',app/test-dma- renamed from app/test-dma-perf/benchmark.c +0
-0
perf/benchmark.cnew file mode 100644 index0000000000..bc1ca82297
Quoted 58 lines
Quoted 58 lines
Quoted 16 lines
Quoted 16 lines
app/test-dma-perf/benchmark.c new file +13
-0@@ -0,0 +1,477 @@ +/* SPDX-License-Identifier: BSD-3-Clause +* Copyright(c) 2023 Intel Corporation */ + +#include <inttypes.h> +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> + +#include <rte_time.h> +#include <rte_mbuf.h> +#include <rte_dmadev.h> +#include <rte_malloc.h> +#include <rte_lcore.h> Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+ +#include "main.h" + +#define MAX_DMA_CPL_NB 255 + +#define TEST_WAIT_U_SECOND 10000 + +#define CSV_LINE_DMA_FMT "Scenario %u,%u,%s,%u,%u,%.2lf,%"PRIu64
Quoted 2 lines
Quoted 2 lines
",%.3lf,%.3lf\n"
+#defineCSV_LINE_CPU_FMT"Scenario %u,%u,NA,%u,%u,%.2lf,%"PRIu64
Quoted 21 lines
Quoted 21 lines
",%.3lf,%.3lf\n"
app/test-dma-perf/benchmark.c new file +20
-0+ +struct worker_info { +bool ready_flag; +bool start_flag; +bool stop_flag; +uint32_t total_cpl; +uint32_t test_cpl; +}; + +struct lcore_params { +uint8_t scenario_id; +unsigned int lcore_id; +char *dma_name; +uint16_t worker_id; +uint16_t dev_id; +uint32_t nr_buf; +uint16_t kick_batch; +uint32_t buf_size; +uint16_t test_secs; +struct rte_mbuf **srcs;Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+struct rte_mbuf **dsts; +struct worker_info worker_info; +}; + +static struct rte_mempool *src_pool;static struct rte_mempool +*dst_pool; + +static volatile struct lcore_params*worker_params[MAX_WORKER_NB];
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
app/test-dma-perf/benchmark.c
Quoted 2 lines
app/test-dma-perf/benchmark.c new file +2
-0+ +#define PRINT_ERR (...) print_err(__func__, __LINE__,
__VA_ARGS__)Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
app/test-dma-perf/benchmark.c new file +12
-0+ +static inline int +__rte_format_printf(3, 4) +print_err(const char *func, int lineno, const char *format, ...) { + va_list ap; + int ret; + + ret = fprintf(stderr, "In %s:%d - ", func, lineno); + va_start(ap, format); + ret += vfprintf(stderr, format, ap); + va_end(ap); +Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +6
-0+return ret; +} + +static inline void +calc_result(uint32_t buf_size, uint32_t nr_buf, uint16_t +nb_workers,
uint16_ttest_secs,app/test-dma-perf/benchmark.c new file +1
-0+ uint32_t total_cnt, float *memory,
uint32_tQuoted 3 lines
Quoted 3 lines
Quoted 3 lines
app/test-dma-perf/benchmark.c
Quoted 3 lines
*ave_cycle,app/test-dma-perf/benchmark.c new file +2
-0+ float *bandwidth, float *mops) { + *memory = (float)(buf_size * (nr_buf / nb_workers) * 2) /(1024
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
+*
1024);app/test-dma-perf/benchmark.c new file +2
-0+ *ave_cycle = test_secs *rte_get_timer_hz () / total_cnt; + *bandwidth = (buf_size * 8 * (rte_get_timer_hz() /(float)*ave_cycle)) / 1000000000;
[Anoob] The above calculation may not yield actual results. '
ave_cycle' would get converted to integer and then bandwidth would be allowed to report only very few values. Instead, we can do the calculation directly like,
Quoted 5 lines
*bandwidth = ((float)buf_size * 8 *
total_cnt/test_secs) /1000000000; *mops = (float)total_cnt / test_secs / 1000000;Same issue is there with below calculation as well. Please check.
[Cheng] Yes, I've noticed as well. Dengdui also mentioned this in his comments. I will address this issue in v7. Thank you very much.
Quoted 4 lines
Side note: in bandwidth calculation, shouldn't we be dividing by 1024*1024*1024? I've just carried the calculation that you used. Feel free to correct as required.
[Cheng] The unit I'm using in my calculations is Gb/s (Gigabits per second), which is based on the decimal system. Therefore, I use the factor of 1000^3 (or 1,000,000,000). The method you mentioned, dividing by 1024^3, is typically used when calculating GiB/s (Gibibits per second), a binary-based unit.
I think both methods are acceptable as long as the units and calculation methods correspond.
What do you think?
Quoted 343 lines
Quoted 316 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +5
-0+ *mops = (float )rte_get_timer_hz () / *ave_cycle /1000000 ; } + +static void +output_result(uint8_t scenario_id, uint32_t lcore_id, char +*dma_name,
uint64_tave_cycle,app/test-dma-perf/benchmark.c new file +1
-0+ uint32_t buf_size, uint32_t nr_buf, floatmemory,
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
app/test-dma-perf/benchmark.c
Quoted 7 lines
app/test-dma-perf/benchmark.c new file +7
-0+ float bandwidth, float mops, bool is_dma) { + if (is_dma) + printf("lcore %u, DMA %s:\n", lcore_id, dma_name); + else + printf("lcore %u\n", lcore_id); + + printf("average cycles/op: %" PRIu64 ", buffer size: %u,
nr_buf:Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
app/test-dma-perf/benchmark.c
Quoted 5 lines
+%u,
memory: %.2lfMB, frequency: %" PRIu64 ".\n",app/test-dma-perf/benchmark.c new file +1
-0+ ave_cycle, buf_size, nr_buf, memory,
rte_get_timer_hz());app/test-dma-perf/benchmark.c new file +1
-0+ printf("Average bandwidth: %.3lfGbps, MOps: %.3lf\n",bandwidth,
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
app/test-dma-perf/benchmark.c
Quoted 4 lines
app/test-dma-perf/benchmark.c new file +4
-0+mops); + +if (is_dma) + snprintf(output_str[lcore_id],
MAX_OUTPUT_STR_LEN,Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
app/test-dma-perf/benchmark.c
Quoted 3 lines
CSV_LINE_DMA_FMT,app/test-dma-perf/benchmark.c new file +2
-0+ scenario_id, lcore_id, dma_name, buf_size, + nr_buf, memory, ave_cycle, bandwidth,mops);
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
app/test-dma-perf/benchmark.c
Quoted 2 lines
app/test-dma-perf/benchmark.c new file +2
-0+else + snprintf(output_str[lcore_id],
MAX_OUTPUT_STR_LEN,Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
app/test-dma-perf/benchmark.c
Quoted 3 lines
CSV_LINE_CPU_FMT,app/test-dma-perf/benchmark.c new file +2
-0+ scenario_id, lcore_id, buf_size, + nr_buf, memory, ave_cycle, bandwidth,mops); }
Quoted 23 lines
Quoted 5 lines
Quoted 5 lines
app/test-dma-perf/benchmark.c
Quoted 5 lines
app/test-dma-perf/benchmark.c new file +5
-0+ +static inline void +cache_flush_buf (__maybe_unused struct rte_mbuf **array, +__maybe_unused uint32_t buf_size, + __maybe_unused uint32_t nr_buf) { #ifdef
RTE_ARCH_X86_64Quoted 17 lines
Quoted 17 lines
Quoted 17 lines
app/test-dma-perf/benchmark.c new file +17
-0+char *data; +struct rte_mbuf **srcs = array; +uint32_t i, offset; + +for (i =0 ; i < nr_buf; i++) { + data = rte_pktmbuf_mtod(srcs[i], char *); +for (offset =0 ; offset < buf_size; offset +=64 ) +__builtin_ia32_clflush (data + offset); + } +#endif +} + +/* Configuration of device. */ +static void +configure_dmadev_queue (uint32_t dev_id,uint32_t ring_size) { +uint16_t vchan =0 ; +struct rte_dma_info info;Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+struct rte_dma_conf dev_config = { .nb_vchans =1 }; +struct rte_dma_vchan_conf qconf = { + .direction =RTE_DMA_DIR_MEM_TO_MEM , + .nb_desc = ring_size + }; + +if (rte_dma_configure (dev_id, &dev_config) !=0 ) + rte_exit(EXIT_FAILURE, "Error with dmaconfigure.\n");
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
app/test-dma-perf/benchmark.c
Quoted 3 lines
app/test-dma-perf/benchmark.c new file +3
-0+ +if (rte_dma_vchan_setup (dev_id, vchan, &qconf) !=0 ) + rte_exit(EXIT_FAILURE, "Error with queueconfiguration.\n");
Quoted 15 lines
Quoted 15 lines
Quoted 15 lines
Quoted 15 lines
app/test-dma-perf/benchmark.c new file +4
-0+ +rte_dma_info_get (dev_id, &info); +if (info.nb_vchans !=1 ) + rte_exit(EXIT_FAILURE, "Error, no configured queuesreported on device id. %u\n",
app/test-dma-perf/benchmark.c new file +10
-0+ dev_id); + +if (rte_dma_start (dev_id) !=0 ) +rte_exit (EXIT_FAILURE ,"Error with dma start.\n" ); } + +static int +config_dmadevs (struct test_configure *cfg) { +uint32_t ring_size = cfg->ring_size.cur; +struct lcore_dma_map_t *ldm = &cfg->lcore_dma_map; +uint32_t nb_workers = ldm->cnt;Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+uint32_t i; +int dev_id; +uint16_t nb_dmadevs =0 ; +char *dma_name; + +for (i =0 ; i < ldm->cnt; i++) { + dma_name = ldm->dma_names[i]; + dev_id =
rte_dma_get_dev_id_by_name(dma_name);Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
app/test-dma-perf/benchmark.c new file +2
-0+if (dev_id ==-1 ) { + fprintf(stderr, "Error: Fail to find DMA %s.\n",
dma_name);app/test-dma-perf/benchmark.c new file +3
-0+ goto end; + } +Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+ ldm->dma_ids[i] = dev_id; +configure_dmadev_queue (dev_id, ring_size); + ++nb_dmadevs; + } + +end: +if (nb_dmadevs < nb_workers) { + printf("Not enough dmadevs (%u) for all workers(%u).\n",
Quoted 18 lines
Quoted 18 lines
Quoted 6 lines
Quoted 6 lines
nb_dmadevs,nb_workers);app/test-dma-perf/benchmark.c new file +5
-0+return -1 ; + } + +printf ("Number of used dmadevs: %u.\n" , nb_dmadevs); +Quoted 8 lines
app/test-dma-perf/benchmark.c line 187
Quoted 6 lines
app/test-dma-perf/benchmark.c new file +6
-0+return 0 ; +} + +#define POLL_MAX 1000 + +[Anoob] Extra blank line. You can consider removing.
[Cheng] sure, sorry for the miss.
Quoted 1 line
Quoted 14 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
app/test-dma-perf/benchmark.c new file +6
-0+static inline void +do_dma_submit_and_poll (uint16_t dev_id,uint64_t *async_cnt, +volatile struct worker_info *worker_info) { +int ret; +uint16_t nr_cpl; +Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+ ret =rte_dma_submit (dev_id,0 ); +if (ret <0 ) { +rte_dma_stop (dev_id); +rte_dma_close (dev_id); +rte_exit (EXIT_FAILURE ,"Error with dma submit.\n" ); + } + + nr_cpl = rte_dma_completed(dev_id, 0,
MAX_DMA_CPL_NB, NULL,Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
app/test-dma-perf/benchmark.c
Quoted 9 lines
NULL);
app/test-dma-perf/benchmark.c new file +8
-0+ *async_cnt -= nr_cpl; + worker_info->total_cpl += nr_cpl; } + +static inline int +do_dma_mem_copy (void *p) +{ +const uint16_t *para_idx = (uint16_t *)p; + volatile struct lcore_params *para =worker_params[*para_idx];
Quoted 18 lines
Quoted 1 line
Quoted 1 line
app/test-dma-perf/benchmark.c
Quoted 1 line
app/test-dma-perf/benchmark.c new file +1
-0+ volatile struct worker_info *worker_info = &(para-
worker_info);Quoted 16 lines
Quoted 16 lines
Quoted 16 lines
app/test-dma-perf/benchmark.c new file +16
-0+const uint16_t dev_id = para->dev_id; +const uint32_t nr_buf = para->nr_buf; +const uint16_t kick_batch = para->kick_batch; +const uint32_t buf_size = para->buf_size; +struct rte_mbuf **srcs = para->srcs; +struct rte_mbuf **dsts = para->dsts; +uint16_t nr_cpl; +uint64_t async_cnt =0 ; +uint32_t i; +uint32_t poll_cnt =0 ; +int ret; + + worker_info->stop_flag = false; + worker_info->ready_flag = true; + +while (!worker_info->start_flag)Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +6
-0+ ; + +while (1 ) { +for (i =0 ; i < nr_buf; i++) { +dma_copy: + ret = rte_dma_copy(dev_id, 0,
rte_pktmbuf_iova(srcs[i]),app/test-dma-perf/benchmark.c new file +1
-0+ rte_pktmbuf_iova(dsts[i]), buf_size,0);
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
app/test-dma-perf/benchmark.c line 247
Quoted 3 lines
app/test-dma-perf/benchmark.c new file +3
-0+if (unlikely (ret <0 )) { +if (ret == -ENOSPC ) { +do_dma_submit_and_poll(dev_id,
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
app/test-dma-perf/benchmark.c
Quoted 6 lines
&async_cnt,worker_info);app/test-dma-perf/benchmark.c new file +5
-0+ goto dma_copy; + }else { +/* Error exit */ +rte_dma_stop (dev_id); + rte_exit(EXIT_FAILURE,"DMA
Quoted 21 lines
Quoted 21 lines
Quoted 21 lines
Quoted 21 lines
enqueue failed\n");
app/test-dma-perf/benchmark.c new file +6
-0+ } + } + async_cnt++; + +if ((async_cnt % kick_batch) ==0 ) + do_dma_submit_and_poll(dev_id,
&async_cnt,worker_info);app/test-dma-perf/benchmark.c new file +9
-0+ } + +if (worker_info->stop_flag) +break ; + } + +rte_dma_submit (dev_id,0 ); +while ((async_cnt >0 ) && (poll_cnt++ <POLL_MAX )) { + nr_cpl = rte_dma_completed(dev_id, 0,
MAX_DMA_CPL_NB, NULL, NULL);app/test-dma-perf/benchmark.c new file +3
-0+ async_cnt -= nr_cpl; + } +Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+return 0 ; +} + +static inline int +do_cpu_mem_copy (void *p) +{ +const uint16_t *para_idx = (uint16_t *)p; + volatile struct lcore_params *para =worker_params[*para_idx];
Quoted 34 lines
Quoted 1 line
Quoted 1 line
app/test-dma-perf/benchmark.c
Quoted 1 line
app/test-dma-perf/benchmark.c new file +1
-0+ volatile struct worker_info *worker_info = &(para-
worker_info);Quoted 32 lines
Quoted 23 lines
Quoted 23 lines
app/test-dma-perf/benchmark.c new file +16
-0+const uint32_t nr_buf = para->nr_buf; +const uint32_t buf_size = para->buf_size; +struct rte_mbuf **srcs = para->srcs; +struct rte_mbuf **dsts = para->dsts; +uint32_t i; + + worker_info->stop_flag = false; + worker_info->ready_flag = true; + +while (!worker_info->start_flag) + ; + +while (1 ) { +for (i =0 ; i < nr_buf; i++) { +/* copy buffer form src to dst */ + rte_memcpy((void*)(uintptr_t)rte_mbuf_data_iova(dsts[i]),
app/test-dma-perf/benchmark.c new file +1
-0+ (void*)(uintptr_t)rte_mbuf_data_iova(srcs[i]),
app/test-dma-perf/benchmark.c new file +4
-0+ (size_t)buf_size); + worker_info->total_cpl++; + } +if (worker_info->stop_flag)Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+break ; + } + +return 0 ; +} + +static int +setup_memory_env(struct test_configure *cfg, struct rte_mbuf***srcs,
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
app/test-dma-perf/benchmark.c
Quoted 10 lines
app/test-dma-perf/benchmark.c new file +10
-0+ struct rte_mbuf ***dsts) +{ +unsigned int buf_size = cfg->buf_size.cur; +unsigned int nr_sockets; +uint32_t nr_buf = cfg->nr_buf; + + nr_sockets =rte_socket_count (); +if (cfg->src_numa_node >= nr_sockets || + cfg->dst_numa_node >= nr_sockets) { + printf("Error: Source or destination numa exceedsthe acture
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
app/test-dma-perf/benchmark.c
Quoted 5 lines
numa nodes.\n");
app/test-dma-perf/benchmark.c new file +4
-0+return -1 ; + } + + src_pool =rte_pktmbuf_pool_create("Benchmark_DMA_SRC",
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
app/test-dma-perf/benchmark.c
Quoted 7 lines
app/test-dma-perf/benchmark.c new file +7
-0+ nr_buf,/* n == num elements */ +64 ,/* cache size */ +0 ,/* priv size */ + buf_size +RTE_PKTMBUF_HEADROOM , + cfg->src_numa_node); +if (src_pool ==NULL ) { + PRINT_ERR("Error with source mempoolcreation.\n");
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
app/test-dma-perf/benchmark.c
Quoted 4 lines
app/test-dma-perf/benchmark.c new file +4
-0+return -1 ; + } + + dst_pool =rte_pktmbuf_pool_create("Benchmark_DMA_DST",
Quoted 10 lines
Quoted 10 lines
Quoted 5 lines
app/test-dma-perf/benchmark.c
Quoted 2 lines
app/test-dma-perf/benchmark.c new file +2
-0+ nr_buf,/* n == num elements */ + 64, /* cache size */[Anoob] We do not alloc or free pointers in the datapath, right? So why bother with cache?
[Cheng] Yes, you are right, the cache size is not necessary here, I'll fix it in the next version.
Quoted 1 line
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
app/test-dma-perf/benchmark.c
Quoted 5 lines
app/test-dma-perf/benchmark.c new file +5
-0+0 ,/* priv size */ + buf_size +RTE_PKTMBUF_HEADROOM , + cfg->dst_numa_node); +if (dst_pool ==NULL ) { + PRINT_ERR("Error with destination mempoolcreation.\n");
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
app/test-dma-perf/benchmark.c
Quoted 4 lines
app/test-dma-perf/benchmark.c new file +4
-0+return -1 ; + } + + *srcs = rte_malloc(NULL, nr_buf * sizeof(struct rte_mbuf *),0);
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
app/test-dma-perf/benchmark.c line 339
Quoted 4 lines
app/test-dma-perf/benchmark.c new file +4
-0+if (*srcs ==NULL ) { +printf ("Error: srcs malloc failed.\n" ); +return -1 ; + }[Anoob] Are we freeing these memory? The ones allocated with
Quoted 7 lines
Quoted 3 lines
Quoted 3 lines
rte_malloc.[Cheng] yes, we freed the memory in the end of
mem_copy_benchmark()Quoted 3 lines
Quoted 1 line
when we finished the test.
[Anoob] I think we are not freeing this mem. In the place where we free
all
Quoted 1 line
Quoted 1 line
mem, we do free all objects to mempool as well as the mempools. But
this
Quoted 3 lines
Quoted 3 lines
memory is to hold the pointers, right? Is that getting freed anywhere?
Also, in the mem clearing paths, do we need to clear the static variables
(ie,
Quoted 1 line
Quoted 1 line
set srcs,
src_pool, dsts,dst_poolto NULL) so that there won't be anyscope
Quoted 7 lines
Quoted 2 lines
for any double free.
[Cheng] My apologies for the misunderstanding earlier. I now understand your point that you are right, the memory used to store the pointers is not being freed. I will fix this issue in the next version. Regarding the static variables you mentioned, I agree with your view that they should be
cleared.
Quoted 3 lines
I will address this in the upcoming version as well. Thank you very much for the feedback. It is greatly appreciated.
Quoted 8 lines
Quoted 2 lines
In addition, I think we also need to nullify these variables when initializing them to ensure safety and standardization of use. What do you think?
[Anoob] Since these are static variables, it is probably okay to skip the init part. But when we use it, we should clear it after use.
Please check above. I've posted one more comment. In case you missed.
[Cheng] sure, thanks for your advice, I'll clear it after use in the next version, thanks.
Quoted 1,100 lines
Quoted 204 lines
Thanks!
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
app/test-dma-perf/benchmark.c
Quoted 4 lines
app/test-dma-perf/benchmark.c new file +2
-0+ + *dsts = rte_malloc(NULL, nr_buf * sizeof(struct rte_mbuf *),0);
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
app/test-dma-perf/benchmark.c
Quoted 6 lines
app/test-dma-perf/benchmark.c new file +6
-0+if (*dsts ==NULL ) { +printf ("Error: dsts malloc failed.\n" ); +return -1 ; + } + + if (rte_mempool_get_bulk(src_pool, (void **)*srcs, nr_buf)!= 0) {
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
app/test-dma-perf/benchmark.c
Quoted 4 lines
app/test-dma-perf/benchmark.c new file +4
-0+printf ("get src mbufs failed.\n" ); +return -1 ; + } + if (rte_mempool_get_bulk(dst_pool, (void **)*dsts, nr_buf)!= 0) {
Quoted 41 lines
Quoted 41 lines
Quoted 41 lines
Quoted 41 lines
app/test-dma-perf/benchmark.c new file +18
-0+printf ("get dst mbufs failed.\n" ); +return -1 ; + } + +return 0 ; +} + +void +mem_copy_benchmark (struct test_configure *cfg,bool is_dma) { +uint16_t i; +uint32_t offset; +unsigned int lcore_id =0 ; +struct rte_mbuf **srcs =NULL , **dsts =NULL ; +struct lcore_dma_map_t *ldm = &cfg->lcore_dma_map; +unsigned int buf_size = cfg->buf_size.cur; +uint16_t kick_batch = cfg->kick_batch.cur; +uint32_t nr_buf = cfg->nr_buf = (cfg->mem_size.cur *1024 * +1024) /(
cfg->buf_size.cur* 2);app/test-dma-perf/benchmark.c new file +22
-0+uint16_t nb_workers = ldm->cnt; +uint16_t test_secs = cfg->test_secs; +float memory; +uint32_t avg_cycles =0 ; +float mops; +float bandwidth; + +if (setup_memory_env (cfg, &srcs, &dsts) <0 ) + goto out; + +if (is_dma) +if (config_dmadevs (cfg) <0 ) + goto out; + +if (cfg->cache_flush) { +cache_flush_buf (srcs, buf_size, nr_buf); +cache_flush_buf (dsts, buf_size, nr_buf); +rte_mb (); + } + +printf ("Start testing....\n" ); +Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +5
-0+for (i =0 ; i < nb_workers; i++) { + lcore_id = ldm->lcores[i]; + offset = nr_buf / nb_workers * i; + + worker_params[i] = rte_malloc(NULL, sizeof(struct
lcore_params), 0);app/test-dma-perf/benchmark.c new file +2
-0+if (!worker_params[i]) { + printf("lcore parameters malloc failure forlcore
Quoted 15 lines
Quoted 6 lines
Quoted 5 lines
app/test-dma-perf/benchmark.c
Quoted 3 lines
%d\n",
lcore_id);app/test-dma-perf/benchmark.c new file +2
-0+break ; + }[Anoob] Are we freeing the above memory?
Quoted 8 lines
[Cheng] sorry, I missed that, I'll add
worker_paramsmemory free in the next version, thanks.Quoted 5 lines
app/test-dma-perf/benchmark.c
Quoted 5 lines
app/test-dma-perf/benchmark.c new file +2
-0+if (is_dma) { + worker_params[i]->dma_name = ldm-Quoted 1 line
dma_names[i];
app/test-dma-perf/benchmark.c new file +1
-0+ worker_params[i]->dev_id = ldm-dma_ids[i];
Quoted 14 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
app/test-dma-perf/benchmark.c new file +4
-0+ worker_params[i]->kick_batch = kick_batch; + } + worker_params[i]->worker_id = i; + worker_params[i]->nr_buf = (uint32_t)(nr_buf /
nb_workers);app/test-dma-perf/benchmark.c new file +1
-0+ worker_params[i]->buf_size = buf_size;Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c line 420
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+ worker_params[i]->test_secs = test_secs; + worker_params[i]->srcs = srcs + offset; + worker_params[i]->dsts = dsts + offset; + worker_params[i]->scenario_id = cfg->scenario_id; + worker_params[i]->lcore_id = lcore_id; + +if (is_dma) +rte_eal_remote_launch(do_dma_mem_copy, (void
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
app/test-dma-perf/benchmark.c
Quoted 3 lines
*)(&i),
lcore_id);app/test-dma-perf/benchmark.c new file +2
-0+else + rte_eal_remote_launch(do_cpu_mem_copy,(void
Quoted 31 lines
Quoted 7 lines
Quoted 7 lines
app/test-dma-perf/benchmark.c
Quoted 7 lines
*)(&i),
lcore_id);app/test-dma-perf/benchmark.c new file +6
-0+ } + +while (1 ) { +bool ready = true; +for (i =0 ; i < nb_workers; i++) { + if (worker_params[i]-worker_info.ready_flag ==
Quoted 23 lines
Quoted 23 lines
Quoted 23 lines
false) {
app/test-dma-perf/benchmark.c new file +19
-0+ ready =0 ; +break ; + } + } +if (ready) +break ; + } + +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.start_flag = true; + +usleep (TEST_WAIT_U_SECOND ); +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.test_cpl = +worker_params[i]->worker_info.total_cpl; + +usleep (test_secs *1000 *1000 ); +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.test_cpl =worker_params[i]->worker_info.total_cpl -
app/test-dma-perf/benchmark.c new file +1
-0+ worker_params[i]-Quoted 1 line
worker_info.test_cpl;
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
app/test-dma-perf/benchmark.c
Quoted 10 lines
app/test-dma-perf/benchmark.c new file +10
-0+ +for (i =0 ; i < nb_workers; i++) + worker_params[i]->worker_info.stop_flag = true; + +rte_eal_mp_wait_lcore (); + +for (i =0 ; i < nb_workers; i++) { + calc_result(buf_size, nr_buf, nb_workers, test_secs, + worker_params[i]->worker_info.test_cpl, + &memory, &avg_cycles, &bandwidth,&mops);
Quoted 6 lines
Quoted 1 line
Quoted 1 line
app/test-dma-perf/benchmark.c
Quoted 1 line
app/test-dma-perf/benchmark.c new file +1
-0+ output_result(cfg->scenario_id, worker_params[i]-
lcore_id,Quoted 1 line
Quoted 1 line
app/test-dma-perf/benchmark.c
Quoted 1 line
app/test-dma-perf/benchmark.c new file +1
-0+ worker_params[i]-
dma_name,Quoted 2 lines
Quoted 2 lines
app/test-dma-perf/benchmark.c
Quoted 2 lines
avg_cycles,buf_size,app/test-dma-perf/benchmark.c new file +1
-0+ nr_buf / nb_workers,memory,
Quoted 34 lines
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
bandwidth, mops,
is_dma);app/test-dma-perf/benchmark.c new file +9
-0+ } + +out: +/* free env */ +if (srcs) +rte_pktmbuf_free_bulk (srcs, nr_buf); +if (dsts) +rte_pktmbuf_free_bulk (dsts, nr_buf); +Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+if (src_pool) +rte_mempool_free (src_pool); +if (dst_pool) +rte_mempool_free (dst_pool); + +if (is_dma) { +for (i =0 ; i < nb_workers; i++) { + printf("Stopping dmadev %d\n", ldm-dma_ids[i]);
Quoted 15 lines
Quoted 6 lines
Quoted 6 lines
app/test-dma-perf/benchmark.c new file +4
-0+rte_dma_stop (ldm->dma_ids[i]); + } + } +}(quoted diff) +0
-0b/app/test-dma-perf/config.ini new file mode 100644 index
Quoted 8 lines
app/test-dma-perf/config.ini line 4
Quoted 8 lines
0000000000..2fd9c3c387
app/test-dma-perf/config.ini new file +4
-0@@ -0,0 +1,59 @@ + +; This is an example configuration file for dma-perf, which +details the meanings of each parameter ; and instructions on how +to use dma-perf.
Quoted 16 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/config.ini new file +8
-0+ +; Supported test types are DMA_MEM_COPY and CPU_MEM_COPY. + +; Parameters: +; "mem_size" denotes the size of the memory footprint. +; "buf_size" denotes the memory size of a single operation. +; "dma_ring_size" denotes the dma ring buffer size. It should be +greaterQuoted 8 lines
Quoted 8 lines
app/test-dma-perf/config.ini
Quoted 8 lines
than 64 normally.
app/test-dma-perf/config.ini new file +2
-0+; "kick_batch" denotes the dma operation batch size, and should +be greaterthan 1 normally.
app/test-dma-perf/config.ini new file +4
-0+ +; The format for variables is variable=first,last,increment,ADD|MUL. + +; src_numa_node is used to control the numa node where thesource
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
app/test-dma-perf/config.ini
Quoted 7 lines
memory is allocated.
app/test-dma-perf/config.ini new file +2
-0+; dst_numa_node is used to control the numa node where the +destinationmemory is allocated.
app/test-dma-perf/config.ini new file +3
-0+ +; cache_flush is used to determine whether or not the cache +should be flushed, with 1 indicating to ; flush and 0 indicating tonot
Quoted 54 lines
Quoted 13 lines
flush.
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
app/test-dma-perf/config.ini new file +4
-0+ +; test_seconds controls the test time of the whole case. + +; To use DMA for a test, please specify the "lcore_dma" parameter.Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/config.ini
Quoted 8 lines
app/test-dma-perf/config.ini new file +3
-0+; If you have already set the "-l" and "-a" parameters using EAL, +; make sure that the value of "lcore_dma" falls within their +range of thevalues.
app/test-dma-perf/config.ini new file +4
-0+ +; To use CPU for a test, please specify the "lcore" parameter. +; If you have already set the "-l" and "-a" parameters using EAL, +; make sure that the value of "lcore" falls within their range ofvalues.
Quoted 40 lines
Quoted 40 lines
Quoted 28 lines
Quoted 14 lines
app/test-dma-perf/config.ini new file +3
-0+ +; To specify a configuration file, use the "--config" flag +followed by the pathto the file.
app/test-dma-perf/config.ini new file +3
-0+ +; To specify a result file, use the "--result" flag followed by +the path to the1 line not shown · raw message
app/test-dma-perf/config.ini new file +3
-0+; If you do not specify a result file, one will be generated with +the same name as the configuration ; file, with the addition of +"_result.csv" at1 line not shown · raw message
app/test-dma-perf/config.ini new file +2
-0+ +[case1]app/test-dma-perf/config.ini line 47
Quoted 10 lines
app/test-dma-perf/config.ini new file +10
-0+type=DMA_MEM_COPY +mem_size=10 +buf_size=64,8192,2,MUL +dma_ring_size=1024 +kick_batch=32 +src_numa_node=0 +dst_numa_node=0 +cache_flush=0 +test_seconds=2 +lcore_dma=lcore10@0000:00:04.2, lcore11@0000:00:04.32 lines not shown · raw message
Quoted 7 lines
6 lines not shown · raw message
4 lines not shown · raw message
834 lines not shown · raw message
Part of this message is not shown here. View the raw message.
Hi Cheng,
Please see inline.
Thanks,
Anoob
Quoted 339 lines
-----Original Message-----
From: Jiang, Cheng1 <[email protected]>
Sent: Friday, June 16, 2023 8:26 AM
To: Anoob Joseph <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>; Amit Prakash Shukla <[email protected]>; Ma, WenwuX <[email protected]>; Wang, YuanX <[email protected]>; He, Xingguang <[email protected]>
Subject: RE: [EXT] [PATCH v6] app/dma-perf: introduce dma-perf applicationHi Anoob,
Replies are inline.
Thanks,
ChengQuoted 4 lines
-----Original Message-----
From: Anoob Joseph <[email protected]>
Sent: Thursday, June 15, 2023 11:48 PM
To: Jiang, Cheng1 <[email protected]>Quoted 8 lines
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>; Amit Prakash Shukla <[email protected]>; Ma, WenwuX <[email protected]>; Wang, YuanX <[email protected]>; He, Xingguang <[email protected]>
Subject: RE: [EXT] [PATCH v6] app/dma-perf: introduce dma-perfapplication
Quoted 15 lines
Hi Cheng,
Please see inline.
Thanks,
AnoobQuoted 7 lines
-----Original Message-----
From: Jiang, Cheng1 <[email protected]>
Sent: Thursday, June 15, 2023 7:36 PM
To: Anoob Joseph <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia,Chenbo
Quoted 14 lines
Quoted 5 lines
<[email protected]>; Amit Prakash Shukla <[email protected]>; Ma, WenwuX <[email protected]>; Wang, YuanX <[email protected]>; He, Xingguang <[email protected]>
Subject: RE: [EXT] [PATCH v6] app/dma-perf: introduce dma-perfapplication
Quoted 8 lines
Hi Anoob,
Replies are inline.
Thanks,
ChengQuoted 10 lines
Quoted 7 lines
Quoted 7 lines
-----Original Message-----
From: Anoob Joseph <[email protected]>
Sent: Thursday, June 15, 2023 4:45 PM
To: Jiang, Cheng1 <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia,Chenbo
Quoted 2 lines
Quoted 2 lines
<[email protected]>; Amit Prakash Shukla <[email protected]>; Ma, WenwuX
Quoted 18 lines
Quoted 18 lines
Quoted 3 lines
Wang, YuanX <[email protected]>; He, Xingguang <[email protected]>
Subject: RE: [EXT] [PATCH v6] app/dma-perf: introduce dma-perfapplication
Quoted 14 lines
Hi Cheng,
Please see inline.
Thanks,
AnoobQuoted 6 lines
-----Original Message-----
From: Jiang, Cheng1 <[email protected]>
Sent: Thursday, June 15, 2023 1:31 PM
To: Anoob Joseph <[email protected]>; [email protected]; Richardson, Bruce <[email protected]>; [email protected]; Xia, Chenbo <[email protected]>;Quoted 21 lines
Quoted 4 lines
Amit
Quoted 3 lines
Quoted 3 lines
Prakash Shukla <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; Ma, WenwuX <[email protected]>;Wang,
Quoted 16 lines
Quoted 16 lines
YuanX
Quoted 15 lines
<[email protected]>; He, Xingguang <[email protected]>
Subject: RE: [EXT] [PATCH v6] app/dma-perf: introduce dma-perf applicationHi,
Thanks for your comments, the replies are inline.
Thanks,
ChengQuoted 4 lines
-----Original Message-----
From: Anoob Joseph <[email protected]>
Sent: Thursday, June 15, 2023 1:22 PM
To: Jiang, Cheng1 <[email protected]>;Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Richardson, Bruce <[email protected]>;
[email protected]; Xia, ChenboQuoted 7 lines
Quoted 7 lines
Quoted 4 lines
Amit
Quoted 3 lines
Quoted 3 lines
Prakash Shukla <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Ding, Xuan <[email protected]>; Ma, WenwuX <[email protected]>;Wang,
Quoted 2 lines
Quoted 2 lines
YuanX
Quoted 1 line
<[email protected]>; He, Xingguang
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 6 lines
Quoted 6 lines
Subject: RE: [EXT] [PATCH v6] app/dma-perf: introduce dma-perf application
Hi,
Thanks for working on the comments. Few more top level comment
inline.
Quoted 1 line
Quoted 1 line
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Thanks,
AnoobQuoted 5 lines
-----Original Message-----
From: Cheng Jiang <[email protected]>
Sent: Tuesday, June 13, 2023 10:02 AM
To: [email protected]; [email protected]; [email protected]; [email protected]; AmitPrakash
Quoted 10 lines
Quoted 10 lines
Quoted 2 lines
Quoted 2 lines
Shukla
Quoted 1 line
Quoted 1 line
<[email protected]>; Anoob Joseph
Quoted 7 lines
Quoted 7 lines
Quoted 2 lines
Quoted 2 lines
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected];
Quoted 4 lines
Quoted 4 lines
Cheng Jiang <[email protected]>
Subject: [EXT] [PATCH v6] app/dma-perf: introduce dma-perf applicationQuoted 8 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
External Email
------------------------------------------------------------------
--
-- There are many high-performance DMA devices supported inDPDK
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
now, and these DMA devices can also be integrated into other modules of DPDK as accelerators, such as Vhost. Before
integrating
Quoted 46 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
DMA into applications, developers need to know the performance
of
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
these DMA devices in various scenarios and the performance of
CPUs
Quoted 42 lines
Quoted 42 lines
Quoted 42 lines
Quoted 6 lines
Quoted 6 lines
in the same scenario, such as different buffer lengths. Only in this way can we know the target performance of the application accelerated by using them. This patch introduces a high-performance testing tool, which supports comparing the performance of CPU and DMA in different scenarios automatically with a pre- set config file. Memory Copy performance test are
Quoted 26 lines
supported for now.
Quoted 25 lines
Signed-off-by Cheng Jiang <[email protected]>Signed-off-by Jiayu Hu <[email protected]>Signed-off-by Yuan Wang <[email protected]>Acked-by Morten Brørup <[email protected]>Acked-by Chenbo Xia <[email protected]>---
v6:
improved code based on Anoob's comments; fixed some code structure issues;
v5:
fixed someLONG_LINEwarnings;
v4:
fixed inaccuracy of the memory footprint display;
v3:
fixed some typos;
v2:
added lcore/dmadev designation;
added error case process;
removedworker_threadsparameter from config.ini; improved the logs;
improved config file;app/meson.build | 1 + app/test-dma-perf/benchmark.c | 477++++++++++++++++++++++++++++
Quoted 9 lines
Quoted 9 lines
app/test-dma-perf/config.ini | 59 ++++
app/test-dma-perf/main.c | 569 ++++++++++++++++++++++++++++++++++ app/test-dma-perf/main.h | 69 +++++ app/test-dma-perf/meson.build | 17 + 6 files changed, 1192 insertions(+) create mode 100644
app/test-dma-perf/benchmark.ccreate mode 100644 app/test-dma-perf/config.ini create mode 100644 app/test-dma-perf/main.ccreate mode 100644app/test-dma-perf/main.hcreate
Quoted 34 lines
Quoted 34 lines
Quoted 17 lines
Quoted 17 lines
Quoted 17 lines
Quoted 16 lines
mode
100644 app/test-dma-perf/meson.buildapp/meson.build index renamed from app/meson.build +0
-0
74d2420f67..4fc1a83eba100644app/meson.build +1
-0@@ -19,6 +19,7 @@ apps = [ 'test-cmdline', 'test-compress-perf', 'test-crypto-perf', + 'test-dma-perf', 'test-eventdev', 'test-fib', 'test-flow-perf',app/test-dma- renamed from app/test-dma-perf/benchmark.c +0
-0Quoted 1 line
perf/benchmark.cnew file mode 100644 index0000000000..bc1ca82297
Quoted 16 lines
Quoted 16 lines
Quoted 16 lines
Quoted 16 lines
app/test-dma-perf/benchmark.c new file +13
-0@@ -0,0 +1,477 @@ +/* SPDX-License-Identifier: BSD-3-Clause +* Copyright(c) 2023 Intel Corporation */ + +#include <inttypes.h> +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> + +#include <rte_time.h> +#include <rte_mbuf.h> +#include <rte_dmadev.h> +#include <rte_malloc.h> +#include <rte_lcore.h> Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c line 21
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+ +#include "main.h" + +#define MAX_DMA_CPL_NB 255 + +#define TEST_WAIT_U_SECOND 10000 + +#define CSV_LINE_DMA_FMT "Scenario%u,%u,%s,%u,%u,%.2lf,%"
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
PRIu64
Quoted 2 lines
Quoted 2 lines
",%.3lf,%.3lf\n"
+#defineCSV_LINE_CPU_FMT"Scenario%u,%u,NA,%u,%u,%.2lf,%"
Quoted 59 lines
Quoted 59 lines
Quoted 31 lines
Quoted 31 lines
PRIu64
Quoted 21 lines
Quoted 21 lines
",%.3lf,%.3lf\n"
app/test-dma-perf/benchmark.c new file +20
-0+ +struct worker_info { +bool ready_flag; +bool start_flag; +bool stop_flag; +uint32_t total_cpl; +uint32_t test_cpl; +}; + +struct lcore_params { +uint8_t scenario_id; +unsigned int lcore_id; +char *dma_name; +uint16_t worker_id; +uint16_t dev_id; +uint32_t nr_buf; +uint16_t kick_batch; +uint32_t buf_size; +uint16_t test_secs; +struct rte_mbuf **srcs;Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+struct rte_mbuf **dsts; +struct worker_info worker_info; +}; + +static struct rte_mempool *src_pool;static struct rte_mempool +*dst_pool; + +static volatile struct lcore_params*worker_params[MAX_WORKER_NB];
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
app/test-dma-perf/benchmark.c
Quoted 2 lines
app/test-dma-perf/benchmark.c new file +2
-0+ +#define PRINT_ERR (...) print_err(__func__, __LINE__,
__VA_ARGS__)Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
app/test-dma-perf/benchmark.c new file +12
-0+ +static inline int +__rte_format_printf(3, 4) +print_err(const char *func, int lineno, const char *format, ...) { + va_list ap; + int ret; + + ret = fprintf(stderr, "In %s:%d - ", func, lineno); + va_start(ap, format); + ret += vfprintf(stderr, format, ap); + va_end(ap); +Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +6
-0+return ret; +} + +static inline void +calc_result(uint32_t buf_size, uint32_t nr_buf, uint16_t +nb_workers,
uint16_ttest_secs,app/test-dma-perf/benchmark.c new file +1
-0+ uint32_t total_cnt, float *memory,
uint32_tQuoted 3 lines
Quoted 3 lines
Quoted 3 lines
app/test-dma-perf/benchmark.c
Quoted 3 lines
*ave_cycle,app/test-dma-perf/benchmark.c new file +2
-0+ float *bandwidth, float *mops) { + *memory = (float)(buf_size * (nr_buf / nb_workers) * 2) /(1024
Quoted 7 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
+*
1024);app/test-dma-perf/benchmark.c new file +2
-0+ *ave_cycle = test_secs *rte_get_timer_hz () / total_cnt; + *bandwidth = (buf_size * 8 * (rte_get_timer_hz() /(float)*ave_cycle)) / 1000000000;
[Anoob] The above calculation may not yield actual results. '
ave_cycle'would
Quoted 8 lines
get converted to integer and then bandwidth would be allowed to report only very few values. Instead, we can do the calculation directly like,
*bandwidth = ((float)buf_size * 8 *
total_cnt/test_secs) /1000000000; *mops = (float)total_cnt / test_secs / 1000000;Same issue is there with below calculation as well. Please check.
[Cheng] Yes, I've noticed as well. Dengdui also mentioned this in his comments. I will address this issue in v7. Thank you very much.
Quoted 4 lines
Side note: in bandwidth calculation, shouldn't we be dividing by 1024*1024*1024? I've just carried the calculation that you used. Feel free to correct as required.
Quoted 8 lines
[Cheng] The unit I'm using in my calculations is Gb/s (Gigabits per second), which is based on the decimal system. Therefore, I use the factor of 1000^3 (or 1,000,000,000).
The method you mentioned, dividing by 1024^3, is typically used when calculating GiB/s (Gibibits per second), a binary-based unit.
I think both methods are acceptable as long as the units and calculation methods correspond.
What do you think?
[Anoob] You are right. Existing logic is correct. Thanks for the explanation.
Quoted 1,477 lines
Quoted 130 lines
Quoted 130 lines
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
app/test-dma-perf/benchmark.c
Quoted 9 lines
app/test-dma-perf/benchmark.c new file +5
-0+ *mops = (float )rte_get_timer_hz () / *ave_cycle /1000000 ; } + +static void +output_result(uint8_t scenario_id, uint32_t lcore_id, char +*dma_name,
uint64_tave_cycle,app/test-dma-perf/benchmark.c new file +1
-0+ uint32_t buf_size, uint32_t nr_buf, floatmemory,
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
app/test-dma-perf/benchmark.c
Quoted 7 lines
app/test-dma-perf/benchmark.c new file +7
-0+ float bandwidth, float mops, bool is_dma) { + if (is_dma) + printf("lcore %u, DMA %s:\n", lcore_id, dma_name); + else + printf("lcore %u\n", lcore_id); + + printf("average cycles/op: %" PRIu64 ", buffer size: %u,
nr_buf:Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
app/test-dma-perf/benchmark.c
Quoted 5 lines
+%u,
memory: %.2lfMB, frequency: %" PRIu64 ".\n",app/test-dma-perf/benchmark.c new file +1
-0+ ave_cycle, buf_size, nr_buf, memory,
rte_get_timer_hz());app/test-dma-perf/benchmark.c new file +1
-0+ printf("Average bandwidth: %.3lfGbps, MOps: %.3lf\n",bandwidth,
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
app/test-dma-perf/benchmark.c
Quoted 4 lines
app/test-dma-perf/benchmark.c new file +4
-0+mops); + +if (is_dma) + snprintf(output_str[lcore_id],
MAX_OUTPUT_STR_LEN,Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
app/test-dma-perf/benchmark.c
Quoted 3 lines
CSV_LINE_DMA_FMT,app/test-dma-perf/benchmark.c new file +2
-0+ scenario_id, lcore_id, dma_name, buf_size, + nr_buf, memory, ave_cycle, bandwidth,mops);
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
app/test-dma-perf/benchmark.c
Quoted 2 lines
app/test-dma-perf/benchmark.c new file +2
-0+else + snprintf(output_str[lcore_id],
MAX_OUTPUT_STR_LEN,Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
app/test-dma-perf/benchmark.c
Quoted 3 lines
CSV_LINE_CPU_FMT,app/test-dma-perf/benchmark.c new file +2
-0+ scenario_id, lcore_id, buf_size, + nr_buf, memory, ave_cycle, bandwidth,mops); }
Quoted 23 lines
Quoted 5 lines
Quoted 5 lines
app/test-dma-perf/benchmark.c
Quoted 5 lines
app/test-dma-perf/benchmark.c new file +5
-0+ +static inline void +cache_flush_buf (__maybe_unused struct rte_mbuf **array, +__maybe_unused uint32_t buf_size, + __maybe_unused uint32_t nr_buf) { #ifdef
RTE_ARCH_X86_64Quoted 17 lines
Quoted 17 lines
Quoted 17 lines
app/test-dma-perf/benchmark.c new file +17
-0+char *data; +struct rte_mbuf **srcs = array; +uint32_t i, offset; + +for (i =0 ; i < nr_buf; i++) { + data = rte_pktmbuf_mtod(srcs[i], char *); +for (offset =0 ; offset < buf_size; offset +=64 ) +__builtin_ia32_clflush (data + offset); + } +#endif +} + +/* Configuration of device. */ +static void +configure_dmadev_queue (uint32_t dev_id,uint32_t ring_size) { +uint16_t vchan =0 ; +struct rte_dma_info info;Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+struct rte_dma_conf dev_config = { .nb_vchans =1 }; +struct rte_dma_vchan_conf qconf = { + .direction =RTE_DMA_DIR_MEM_TO_MEM , + .nb_desc = ring_size + }; + +if (rte_dma_configure (dev_id, &dev_config) !=0 ) + rte_exit(EXIT_FAILURE, "Error with dmaconfigure.\n");
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
app/test-dma-perf/benchmark.c
Quoted 3 lines
app/test-dma-perf/benchmark.c new file +3
-0+ +if (rte_dma_vchan_setup (dev_id, vchan, &qconf) !=0 ) + rte_exit(EXIT_FAILURE, "Error with queueconfiguration.\n");
Quoted 15 lines
Quoted 15 lines
Quoted 15 lines
Quoted 15 lines
app/test-dma-perf/benchmark.c new file +4
-0+ +rte_dma_info_get (dev_id, &info); +if (info.nb_vchans !=1 ) + rte_exit(EXIT_FAILURE, "Error, no configured queuesreported on device id. %u\n",
app/test-dma-perf/benchmark.c new file +10
-0+ dev_id); + +if (rte_dma_start (dev_id) !=0 ) +rte_exit (EXIT_FAILURE ,"Error with dma start.\n" ); } + +static int +config_dmadevs (struct test_configure *cfg) { +uint32_t ring_size = cfg->ring_size.cur; +struct lcore_dma_map_t *ldm = &cfg->lcore_dma_map; +uint32_t nb_workers = ldm->cnt;Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+uint32_t i; +int dev_id; +uint16_t nb_dmadevs =0 ; +char *dma_name; + +for (i =0 ; i < ldm->cnt; i++) { + dma_name = ldm->dma_names[i]; + dev_id =
rte_dma_get_dev_id_by_name(dma_name);Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
app/test-dma-perf/benchmark.c new file +2
-0+if (dev_id ==-1 ) { + fprintf(stderr, "Error: Fail to find DMA %s.\n",
dma_name);app/test-dma-perf/benchmark.c new file +3
-0+ goto end; + } +Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+ ldm->dma_ids[i] = dev_id; +configure_dmadev_queue (dev_id, ring_size); + ++nb_dmadevs; + } + +end: +if (nb_dmadevs < nb_workers) { + printf("Not enough dmadevs (%u) for all workers(%u).\n",
Quoted 15 lines
Quoted 15 lines
Quoted 14 lines
app/test-dma-perf/benchmark.c
Quoted 12 lines
nb_dmadevs,nb_workers);app/test-dma-perf/benchmark.c new file +11
-0+return -1 ; + } + +printf ("Number of used dmadevs: %u.\n" , nb_dmadevs); + +return 0 ; +} + +#define POLL_MAX 1000 + +[Anoob] Extra blank line. You can consider removing.
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
[Cheng] sure, sorry for the miss.
Quoted 3 lines
app/test-dma-perf/benchmark.c
Quoted 3 lines
app/test-dma-perf/benchmark.c new file +2
-0+static inline void +do_dma_submit_and_poll(uint16_t dev_id, uint64_t
*async_cnt,Quoted 182 lines
Quoted 182 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
app/test-dma-perf/benchmark.c new file +4
-0+ volatile struct worker_info *worker_info) { +int ret; +uint16_t nr_cpl; +Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+ ret =rte_dma_submit (dev_id,0 ); +if (ret <0 ) { +rte_dma_stop (dev_id); +rte_dma_close (dev_id); +rte_exit (EXIT_FAILURE ,"Error with dma submit.\n" ); + } + + nr_cpl = rte_dma_completed(dev_id, 0,
MAX_DMA_CPL_NB, NULL,Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
app/test-dma-perf/benchmark.c
Quoted 9 lines
NULL);
app/test-dma-perf/benchmark.c new file +8
-0+ *async_cnt -= nr_cpl; + worker_info->total_cpl += nr_cpl; } + +static inline int +do_dma_mem_copy (void *p) +{ +const uint16_t *para_idx = (uint16_t *)p; + volatile struct lcore_params *para =worker_params[*para_idx];
Quoted 18 lines
Quoted 1 line
Quoted 1 line
app/test-dma-perf/benchmark.c
Quoted 1 line
app/test-dma-perf/benchmark.c new file +1
-0+ volatile struct worker_info *worker_info = &(para-
worker_info);Quoted 16 lines
Quoted 16 lines
Quoted 16 lines
app/test-dma-perf/benchmark.c new file +16
-0+const uint16_t dev_id = para->dev_id; +const uint32_t nr_buf = para->nr_buf; +const uint16_t kick_batch = para->kick_batch; +const uint32_t buf_size = para->buf_size; +struct rte_mbuf **srcs = para->srcs; +struct rte_mbuf **dsts = para->dsts; +uint16_t nr_cpl; +uint64_t async_cnt =0 ; +uint32_t i; +uint32_t poll_cnt =0 ; +int ret; + + worker_info->stop_flag = false; + worker_info->ready_flag = true; + +while (!worker_info->start_flag)Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +6
-0+ ; + +while (1 ) { +for (i =0 ; i < nr_buf; i++) { +dma_copy: + ret = rte_dma_copy(dev_id, 0,
rte_pktmbuf_iova(srcs[i]),app/test-dma-perf/benchmark.c new file +1
-0+ rte_pktmbuf_iova(dsts[i]), buf_size,0);
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
app/test-dma-perf/benchmark.c
Quoted 3 lines
app/test-dma-perf/benchmark.c new file +3
-0+if (unlikely (ret <0 )) { +if (ret == -ENOSPC ) { +do_dma_submit_and_poll(dev_id,
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
app/test-dma-perf/benchmark.c
Quoted 6 lines
&async_cnt,worker_info);app/test-dma-perf/benchmark.c new file +5
-0+ goto dma_copy; + }else { +/* Error exit */ +rte_dma_stop (dev_id); + rte_exit(EXIT_FAILURE,"DMA
Quoted 21 lines
Quoted 21 lines
Quoted 21 lines
Quoted 21 lines
enqueue failed\n");
app/test-dma-perf/benchmark.c new file +6
-0+ } + } + async_cnt++; + +if ((async_cnt % kick_batch) ==0 ) + do_dma_submit_and_poll(dev_id,
&async_cnt,worker_info);app/test-dma-perf/benchmark.c new file +9
-0+ } + +if (worker_info->stop_flag) +break ; + } + +rte_dma_submit (dev_id,0 ); +while ((async_cnt >0 ) && (poll_cnt++ <POLL_MAX )) { + nr_cpl = rte_dma_completed(dev_id, 0,
MAX_DMA_CPL_NB, NULL, NULL);app/test-dma-perf/benchmark.c new file +3
-0+ async_cnt -= nr_cpl; + } +Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+return 0 ; +} + +static inline int +do_cpu_mem_copy (void *p) +{ +const uint16_t *para_idx = (uint16_t *)p; + volatile struct lcore_params *para =worker_params[*para_idx];
Quoted 34 lines
Quoted 1 line
Quoted 1 line
app/test-dma-perf/benchmark.c
Quoted 1 line
app/test-dma-perf/benchmark.c new file +1
-0+ volatile struct worker_info *worker_info = &(para-
worker_info);Quoted 32 lines
Quoted 23 lines
Quoted 23 lines
app/test-dma-perf/benchmark.c new file +16
-0+const uint32_t nr_buf = para->nr_buf; +const uint32_t buf_size = para->buf_size; +struct rte_mbuf **srcs = para->srcs; +struct rte_mbuf **dsts = para->dsts; +uint32_t i; + + worker_info->stop_flag = false; + worker_info->ready_flag = true; + +while (!worker_info->start_flag) + ; + +while (1 ) { +for (i =0 ; i < nr_buf; i++) { +/* copy buffer form src to dst */ + rte_memcpy((void*)(uintptr_t)rte_mbuf_data_iova(dsts[i]),
app/test-dma-perf/benchmark.c new file +1
-0+ (void*)(uintptr_t)rte_mbuf_data_iova(srcs[i]),
app/test-dma-perf/benchmark.c new file +4
-0+ (size_t)buf_size); + worker_info->total_cpl++; + } +if (worker_info->stop_flag)Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +8
-0+break ; + } + +return 0 ; +} + +static int +setup_memory_env(struct test_configure *cfg, struct rte_mbuf***srcs,
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
app/test-dma-perf/benchmark.c
Quoted 10 lines
app/test-dma-perf/benchmark.c new file +10
-0+ struct rte_mbuf ***dsts) +{ +unsigned int buf_size = cfg->buf_size.cur; +unsigned int nr_sockets; +uint32_t nr_buf = cfg->nr_buf; + + nr_sockets =rte_socket_count (); +if (cfg->src_numa_node >= nr_sockets || + cfg->dst_numa_node >= nr_sockets) { + printf("Error: Source or destination numa exceedsthe acture
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
app/test-dma-perf/benchmark.c
Quoted 5 lines
numa nodes.\n");
app/test-dma-perf/benchmark.c new file +4
-0+return -1 ; + } + + src_pool =rte_pktmbuf_pool_create("Benchmark_DMA_SRC",
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
app/test-dma-perf/benchmark.c
Quoted 7 lines
app/test-dma-perf/benchmark.c new file +7
-0+ nr_buf,/* n == num elements */ +64 ,/* cache size */ +0 ,/* priv size */ + buf_size +RTE_PKTMBUF_HEADROOM , + cfg->src_numa_node); +if (src_pool ==NULL ) { + PRINT_ERR("Error with source mempoolcreation.\n");
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
app/test-dma-perf/benchmark.c
Quoted 4 lines
app/test-dma-perf/benchmark.c new file +4
-0+return -1 ; + } + + dst_pool =rte_pktmbuf_pool_create("Benchmark_DMA_DST",
Quoted 10 lines
Quoted 10 lines
Quoted 5 lines
app/test-dma-perf/benchmark.c
Quoted 2 lines
app/test-dma-perf/benchmark.c new file +2
-0+ nr_buf,/* n == num elements */ + 64, /* cache size */[Anoob] We do not alloc or free pointers in the datapath, right? So why bother with cache?
[Cheng] Yes, you are right, the cache size is not necessary here, I'll fix it in the next version.
Quoted 1 line
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
app/test-dma-perf/benchmark.c
Quoted 5 lines
app/test-dma-perf/benchmark.c new file +5
-0+0 ,/* priv size */ + buf_size +RTE_PKTMBUF_HEADROOM , + cfg->dst_numa_node); +if (dst_pool ==NULL ) { + PRINT_ERR("Error with destination mempoolcreation.\n");
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
app/test-dma-perf/benchmark.c
Quoted 4 lines
app/test-dma-perf/benchmark.c new file +4
-0+return -1 ; + } + + *srcs = rte_malloc(NULL, nr_buf * sizeof(struct rte_mbuf *),0);
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
app/test-dma-perf/benchmark.c
Quoted 4 lines
app/test-dma-perf/benchmark.c new file +4
-0+if (*srcs ==NULL ) { +printf ("Error: srcs malloc failed.\n" ); +return -1 ; + }[Anoob] Are we freeing these memory? The ones allocated with
Quoted 7 lines
Quoted 7 lines
Quoted 3 lines
Quoted 3 lines
rte_malloc.[Cheng] yes, we freed the memory in the end of
mem_copy_benchmark()Quoted 3 lines
Quoted 1 line
when we finished the test.
[Anoob] I think we are not freeing this mem. In the place where we
free
Quoted 14 lines
all
Quoted 1 line
Quoted 1 line
mem, we do free all objects to mempool as well as the mempools. But
this
Quoted 3 lines
Quoted 3 lines
memory is to hold the pointers, right? Is that getting freed anywhere?
Also, in the mem clearing paths, do we need to clear the static variables
(ie,
Quoted 1 line
Quoted 1 line
set srcs,
src_pool, dsts,dst_poolto NULL) so that there won't be anyscope
Quoted 5 lines
Quoted 2 lines
for any double free.
[Cheng] My apologies for the misunderstanding earlier. I now understand your point that you are right, the memory used to store the pointers is
not
Quoted 4 lines
Quoted 2 lines
being freed. I will fix this issue in the next version. Regarding the static variables you mentioned, I agree with your view that they should be
cleared.
Quoted 1 line
I will address this in the upcoming version as well. Thank you very much
for
Quoted 10 lines
Quoted 4 lines
the feedback. It is greatly appreciated.
In addition, I think we also need to nullify these variables when initializing them to ensure safety and standardization of use. What do you think?
[Anoob] Since these are static variables, it is probably okay to skip the init part. But when we use it, we should clear it after use.
Please check above. I've posted one more comment. In case you missed.
[Cheng] sure, thanks for your advice, I'll clear it after use in the next version, thanks.
Quoted 20 lines
Quoted 20 lines
Thanks!
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
app/test-dma-perf/benchmark.c
Quoted 4 lines
app/test-dma-perf/benchmark.c new file +2
-0+ + *dsts = rte_malloc(NULL, nr_buf * sizeof(struct rte_mbuf *),0);
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
app/test-dma-perf/benchmark.c
Quoted 6 lines
app/test-dma-perf/benchmark.c new file +6
-0+if (*dsts ==NULL ) { +printf ("Error: dsts malloc failed.\n" ); +return -1 ; + } + + if (rte_mempool_get_bulk(src_pool, (void **)*srcs, nr_buf)!= 0) {
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
app/test-dma-perf/benchmark.c
Quoted 4 lines
app/test-dma-perf/benchmark.c new file +4
-0+printf ("get src mbufs failed.\n" ); +return -1 ; + } + if (rte_mempool_get_bulk(dst_pool, (void **)*dsts, nr_buf)!= 0) {
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
app/test-dma-perf/benchmark.c
Quoted 9 lines
app/test-dma-perf/benchmark.c new file +9
-0+printf ("get dst mbufs failed.\n" ); +return -1 ; + } + +return 0 ; +} + +void +mem_copy_benchmark (struct test_configure *cfg,bool is_dma){
Quoted 143 lines
Quoted 143 lines
Quoted 32 lines
Quoted 32 lines
Quoted 32 lines
Quoted 32 lines
app/test-dma-perf/benchmark.c new file +9
-0+uint16_t i; +uint32_t offset; +unsigned int lcore_id =0 ; +struct rte_mbuf **srcs =NULL , **dsts =NULL ; +struct lcore_dma_map_t *ldm = &cfg->lcore_dma_map; +unsigned int buf_size = cfg->buf_size.cur; +uint16_t kick_batch = cfg->kick_batch.cur; +uint32_t nr_buf = cfg->nr_buf = (cfg->mem_size.cur *1024 * +1024) /(
cfg->buf_size.cur* 2);app/test-dma-perf/benchmark.c new file +22
-0+uint16_t nb_workers = ldm->cnt; +uint16_t test_secs = cfg->test_secs; +float memory; +uint32_t avg_cycles =0 ; +float mops; +float bandwidth; + +if (setup_memory_env (cfg, &srcs, &dsts) <0 ) + goto out; + +if (is_dma) +if (config_dmadevs (cfg) <0 ) + goto out; + +if (cfg->cache_flush) { +cache_flush_buf (srcs, buf_size, nr_buf); +cache_flush_buf (dsts, buf_size, nr_buf); +rte_mb (); + } + +printf ("Start testing....\n" ); +Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
app/test-dma-perf/benchmark.c
Quoted 8 lines
app/test-dma-perf/benchmark.c new file +5
-0+for (i =0 ; i < nb_workers; i++) { + lcore_id = ldm->lcores[i]; + offset = nr_buf / nb_workers * i; + + worker_params[i] = rte_malloc(NULL, sizeof(struct
lcore_params), 0);app/test-dma-perf/benchmark.c new file +2
-0+if (!worker_params[i]) { + printf("lcore parameters malloc failure forlcore
Quoted 15 lines
Quoted 6 lines
Quoted 5 lines
app/test-dma-perf/benchmark.c
Quoted 3 lines
%d\n",
lcore_id);app/test-dma-perf/benchmark.c new file +2
-0+break ; + }1 line not shown · raw message
Quoted 8 lines
7 lines not shown · raw message
1 line not shown · raw message
Quoted 14 lines
14 lines not shown · raw message
73 lines not shown · raw message
Quoted 13 lines
13 lines not shown · raw message
931 lines not shown · raw message
Part of this message is not shown here. View the raw message.