Skip to contentSkip to conversation

Add GPU_POWER (v3) + refactored cleanup

UMR New Tom St Denis 3 participants 6 messages Feb 11, 2017 3 messages not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#2 lore raw

PATCH 1/2 Remove unused bits from --top

---
 src/app/top.c | 5 -----
 1 file changed, 5 deletions(-)
src/app/top.c +0 -5
index b081515a5b40..d9a474d78cf7 100644
@@ -171,19 +171,14 @@ static struct umr_bitfield stat_uvd_pgfsm7_bits[] = {
 static struct umr_bitfield stat_mc_hub_bits[] = {
 	 { "OUTSTANDING_READ", 255, 255, &umr_bitfield_default },
 	 { "OUTSTANDING_WRITE", 255, 255, &umr_bitfield_default },
-//	 { "OUTSTANDING_ATOMIC", 255, 255, &umr_bitfield_default },
 	 { "OUTSTANDING_HUB_RDREQ", 255, 255, &umr_bitfield_default },
 	 { "OUTSTANDING_HUB_RDRET", 255, 255, &umr_bitfield_default },
 	 { "OUTSTANDING_HUB_WRREQ", 255, 255, &umr_bitfield_default },
 	 { "OUTSTANDING_HUB_WRRET", 255, 255, &umr_bitfield_default },
-//	 { "OUTSTANDING_HUB_ATOMIC_REQ", 255, 255, &umr_bitfield_default },
-//	 { "OUTSTANDING_HUB_ATOMIC_RET", 255, 255, &umr_bitfield_default },
 	 { "OUTSTANDING_RPB_READ", 255, 255, &umr_bitfield_default },
 	 { "OUTSTANDING_RPB_WRITE", 255, 255, &umr_bitfield_default },
-//	 { "OUTSTANDING_RPB_ATOMIC", 255, 255, &umr_bitfield_default },
 	 { "OUTSTANDING_MCD_READ", 255, 255, &umr_bitfield_default },
 	 { "OUTSTANDING_MCD_WRITE", 255, 255, &umr_bitfield_default },
-//	 { "OUTSTANDING_MCD_ATOMIC", 255, 255, &umr_bitfield_default },
 	 { NULL, 0, 0, NULL },
 };
 
Signature · 6 lines
2.11.0

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#3 lore raw

PATCH 2/2 Add GPU_POWER sensors (v3)

In reply to a message not in the archive

Add the ability to sample GPU_POWER sensors. Because the sensors have a high latency we read them from a background thread which means we've added the requirement for pthreads.

(v2) Cleaned up CMake around pthreads
(v3) Update readme, factor out some deletions, cleanup switch statement

---
 CMakeLists.txt         |  4 +++
 README                 |  5 ++-
 src/app/top.c          | 90 ++++++++++++++++++++++++++++++++++++++++++++------
 src/lib/CMakeLists.txt |  1 +
 src/lib/read_sensor.c  | 37 +++++++++++++++++++++
 src/umr.h              |  5 +++
 6 files changed, 129 insertions(+), 13 deletions(-)
 create mode 100644 src/lib/read_sensor.c
CMakeLists.txt +4 -0
index ef78c97ad763..8d89445c39e3 100644
@@ -19,6 +19,9 @@ add_definitions(-DUMR_BUILD_REV=\"${GIT_REV}\")
 # Add local repository for FindXXX.cmake modules.
 SET(CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/cmake_modules/" ${CMAKE_MODULE_PATH})
 
+set(CMAKE_THREAD_PREFER_PTHREAD TRUE)
+find_package(Threads REQUIRED)
+
 find_package(Curses REQUIRED)
 include_directories(${CURSES_INCLUDE_DIRS})
 
@@ -31,6 +34,7 @@ include_directories(${LIBDRM_INCLUDE_DIR})
 set(REQUIRED_EXTERNAL_LIBS
   ${CURSES_LIBRARIES}
   ${PCIACCESS_LIBRARIES}
+  Threads::Threads
 )
 
 # Global setting: build everything position independent
README +2 -3
index 13cdac663d20..8b75f05a13b5 100644
@@ -28,9 +28,8 @@ mailing list at:
 Building
 ---------
 
-To build umr you will need pciaccess, ncurses, and libdrm headers and
-libraries.  Which are available in both Fedora and Ubuntu (as well as
-other distributions).  To build umr:
+To build umr you will need pciaccess, ncurses, libdrm, and pthread
+headers and libraries.  To build umr:
 
     $ mkdir build && cd build/ && cmake ../
     $ make
src/app/top.c +80 -10

153 lines not on this page · raw message

src/lib/CMakeLists.txt +1 -0
index 46c75d61acb4..a5f0bda38689 100644
@@ -14,6 +14,7 @@ add_library(umrcore STATIC
   find_reg.c
   mmio.c
   query_drm.c
+  read_sensor.c
   read_sgpr.c
   read_vram.c
   ring_decode.c
src/lib/read_sensor.c new file +37 -0
new file mode 100644
index 000000000000..c4907f83208f
@@ -0,0 +1,37 @@
+/*
+ * Copyright 2017 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Tom St Denis <[email protected]>
+ *
+ */
+#include "umr.h"
+
+int umr_read_sensor(struct umr_asic *asic, int sensor, void *dst, int *size)
+{
+	int r;
+	lseek(asic->fd.sensors, sensor*4, SEEK_SET);
+	r = read(asic->fd.sensors, dst, *size);
+	if (r != *size) {
+		return -1;
+	}
+	*size = r;
+	return 0;
+}
src/umr.h +5 -0
index f2bce13d104b..34e8809eb415 100644
@@ -31,6 +31,7 @@
 #include <sys/stat.h>
 #include <fcntl.h>
 #include <pciaccess.h>
+#include <pthread.h>
 
 /* sourced from amd_powerplay.h from the kernel */
 enum amd_pp_sensors {
@@ -43,6 +44,9 @@ enum amd_pp_sensors {
 	AMDGPU_PP_SENSOR_GPU_LOAD,
 	AMDGPU_PP_SENSOR_GFX_MCLK,
 	AMDGPU_PP_SENSOR_GPU_TEMP,
+	AMDGPU_PP_SENSOR_VCE_POWER,
+	AMDGPU_PP_SENSOR_UVD_POWER,
+	AMDGPU_PP_SENSOR_GPU_POWER,
 };
 
 enum chipfamily {
@@ -409,6 +413,7 @@ void umr_enumerate_devices(void);
 int umr_get_wave_status(struct umr_asic *asic, unsigned se, unsigned sh, unsigned cu, unsigned simd, unsigned wave, struct umr_wave_status *ws);
 int umr_get_wave_sq_info(struct umr_asic *asic, unsigned se, unsigned sh, unsigned cu, struct umr_wave_status *ws);
 int umr_read_sgprs(struct umr_asic *asic, struct umr_wave_status *ws, uint32_t *dst);
+int umr_read_sensor(struct umr_asic *asic, int sensor, void *dst, int *size);
 
 /* mmio helpers */
 uint32_t umr_find_reg(struct umr_asic *asic, char *regname);
Signature · 6 lines
2.11.0

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): 3627d968-1817-5ca3-024e-1d91291680b5-1ZKVMVCtJ2dx9oSEVPI0kiST3g8Odh+X@public.gmane.org

#6 lore raw
Quoted 10 lines · Kai Wasserbäch

Hey Tom,
Tom St Denis wrote on 11.02.2017 18:52:

Quoted 1 line

I split out the deletion of the unused bits in --top.

thank you so much for bearing with me (and Edward, though I can't speak for him, obviously)! You can still have my R-b. ;-)

Cheers,
Kai

No worries. It's no secret I'm no fan of NFC re-submits but hard to get upset in the face of genuinely good feedback. And it is a community project in the long run so it's not like community pushback/feedback is not welcomed (my rumblings notwithstanding).

Thanks to you and Edward for the feedback.

I've pushed them out. You'll need a staging kernel to use GPU_POWER (with my and Eric's patches) to see any of this on a VI dGPU.

It's kinda interesting to see the power usage graphs (using umr's --top 'logger' mode) comparing Polaris10 and fiji while running say unigine-heaven. I think the sensors will help debug/optimize power usage going forward.

Cheers,
Tom
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): 2b15805a-3e56-71b7-0242-98e3db103add-5C7GfCeVMHo@public.gmane.org

#8 lore raw

Hey Tom,
Tom St Denis wrote on 11.02.2017 20:05:

Quoted 7 lines · Kai Wasserbäch
Quoted 6 lines

Tom St Denis wrote on 11.02.2017 18:52:

Quoted 1 line

I split out the deletion of the unused bits in --top.

thank you so much for bearing with me (and Edward, though I can't speak for him, obviously)! You can still have my R-b. ;-)

Quoted 8 lines
Quoted 3 lines

Cheers,
Kai

No worries. It's no secret I'm no fan of NFC re-submits but hard to get upset in the face of genuinely good feedback. And it is a community project in the long run so it's not like community pushback/feedback is not welcomed (my rumblings notwithstanding).

I'm not sure what AMD's internal policy might be on this, but NFC re-submits are usually covered by the "with this/these fixed, you can have my R-b". Which is the reason I had that kind of language in there. I didn't expect a resend to the list, to be honest. (And I would guess the same applied to Edward's R-b.)

Quoted 8 lines

Thanks to you and Edward for the feedback.

I've pushed them out. You'll need a staging kernel to use GPU_POWER (with my and Eric's patches) to see any of this on a VI dGPU.

It's kinda interesting to see the power usage graphs (using umr's --top 'logger' mode) comparing Polaris10 and fiji while running say unigine-heaven. I think the sensors will help debug/optimize power usage going forward.

Oh, should there have been a DRM version check in there then? Like Mesa guards some features?

In any case I will have to wait for the Vega-upgrade I'm planning for this year to actually test these features out.

Cheers,
Kai

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx