Skip to contentSkip to conversation

drm/i915/selftests/perf: measure memcpy bw between regions

PATCH Superseded Matthew Auld 3 participants 5 messages Jan 27, 2020

Subsystems (2)

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drm/i915/selftests/perf: measure memcpy bw between regions

Measure the memcpy bw between our CPU accessible regions, trying all supported mapping combinations(WC, WB) across various sizes.

---
 .../drm/i915/selftests/i915_perf_selftests.h  |   1 +
 .../drm/i915/selftests/intel_memory_region.c  | 164 ++++++++++++++++++
 2 files changed, 165 insertions(+)
drivers/gpu/drm/i915/selftests/i915_perf_selftests.h +1 -0
index 5a577a1332f5..3bf7f53e9924 100644
@@ -17,3 +17,4 @@
  */
 selftest(engine_cs, intel_engine_cs_perf_selftests)
 selftest(blt, i915_gem_object_blt_perf_selftests)
+selftest(region, intel_memory_region_perf_selftests)
drivers/gpu/drm/i915/selftests/intel_memory_region.c +164 -0

179 lines not on this page · raw message

Signature · 6 lines
2.20.1

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

Quoting Matthew Auld (2020-01-27 12:56:26)

Quoted 32 lines

Measure the memcpy bw between our CPU accessible regions, trying all supported mapping combinations(WC, WB) across various sizes.

---
 .../drm/i915/selftests/i915_perf_selftests.h  |   1 +
 .../drm/i915/selftests/intel_memory_region.c  | 164 ++++++++++++++++++
 2 files changed, 165 insertions(+)
drivers/gpu/drm/i915/selftests/i915_perf_selftests.h +1 -0
index 5a577a1332f5..3bf7f53e9924 100644
@@ -17,3 +17,4 @@
  */
 selftest(engine_cs, intel_engine_cs_perf_selftests)
 selftest(blt, i915_gem_object_blt_perf_selftests)
+selftest(region, intel_memory_region_perf_selftests)
drivers/gpu/drm/i915/selftests/intel_memory_region.c +1 -0
index 3ef3620e0da5..6d493a198eb1 100644
@@ -4,6 +4,7 @@
  */
 
 #include <linux/prime_numbers.h>
+#include <linux/sort.h>
 
 #include "../i915_selftest.h"
 

drivers/gpu/drm/i915/selftests/intel_memory_region.c static int igt_lmem_write_cpu(void *arg) · line 576

Quoted 4 lines
drivers/gpu/drm/i915/selftests/intel_memory_region.c +1 -0
@@ -572,6 +573,157 @@ static int igt_lmem_write_cpu(void *arg)
        return err;
 }
 
+static const char *stringify_map_type(u32 type)

Fwiw, I've begun using repr_foo(foo). Which do we prefer?

Quoted 11 lines
drivers/gpu/drm/i915/selftests/intel_memory_region.c +11 -0
+{
+       switch (type) {
+       case I915_MAP_WB:
+               return "WB";
+       case I915_MAP_WC:
+               return "WC";
+       }
+
+       return "";
+}
+

drivers/gpu/drm/i915/selftests/intel_memory_region.c static int igt_lmem_write_cpu(void *arg) · line 597

Quoted 10 lines
drivers/gpu/drm/i915/selftests/intel_memory_region.c +10 -0
+static inline struct drm_i915_gem_object *
+create_region(struct intel_memory_region *mr, u64 size)
+{
+       struct drm_i915_gem_object *obj;
+
+       obj = i915_gem_object_create_region(mr, size, 0);
+       if (!IS_ERR(obj)) {
+               if (!i915_gem_object_type_has(obj,
+                                             I915_GEM_OBJECT_HAS_STRUCT_PAGE |
+                                             I915_GEM_OBJECT_HAS_IOMEM)) {

Can we not tell from the mr what properties we would end up with?

I would suggest calling it something like create_region_for_mapping() as I had to look around to see why the type_has() was justified.

Hmm. Perhaps better (more agnostic to the pin_map) would be to try and fail to create the map and squash the -ENODEV error there.

Quoted 45 lines
drivers/gpu/drm/i915/selftests/intel_memory_region.c +45 -0
+                       i915_gem_object_put(obj);
+                       return ERR_PTR(-ENODEV);
+               }
+       }
+
+       return obj;
+}
+
+static int wrap_ktime_compare(const void *A, const void *B)
+{
+       const ktime_t *a = A, *b = B;
+
+       return ktime_compare(*a, *b);
+}
+
+int _perf_memcpy(struct intel_memory_region *src_mr,
+                struct intel_memory_region *dst_mr,
+                u64 size, u32 src_type, u32 dst_type)
+{
+       struct drm_i915_gem_object *src, *dst;
+       void *src_addr, *dst_addr;
+       ktime_t t[5];
+       int pass;
+       int ret = 0;
+
+       src = create_region(src_mr, size);
+       if (IS_ERR(src)) {
+               ret = PTR_ERR(src);
+               goto out;
+       }
+
+       src_addr = i915_gem_object_pin_map(src, src_type);
+       if (IS_ERR(src_addr)) {
+               ret = PTR_ERR(src_addr);
+               goto out_put_src;
+       }
+
+       dst = create_region(dst_mr, size);
+       if (IS_ERR(dst)) {
+               ret = PTR_ERR(dst);
+               goto out_unpin_src;
+       }
+
+       dst_addr = i915_gem_object_pin_map(dst, dst_type);
+       if (IS_ERR(dst_addr)) {

drivers/gpu/drm/i915/selftests/intel_memory_region.c static int igt_lmem_write_cpu(void *arg) · line 652

Quoted 10 lines
drivers/gpu/drm/i915/selftests/intel_memory_region.c +10 -0
+               ret = PTR_ERR(dst_addr);
+               goto out_put_dst;
+       }
+
+       for (pass = 0; pass < ARRAY_SIZE(t); pass++) {
+               ktime_t t0, t1;
+
+               t0 = ktime_get();
+
+               memcpy(dst_addr, src_addr, size);

Fwiw, different sizes of memcpy() may prove important, (memcpy32/memcpy64). As would memcpy_from_wc.

Quoted 53 lines
drivers/gpu/drm/i915/selftests/intel_memory_region.c +53 -0
+
+               t1 = ktime_get();
+               t[pass] = ktime_sub(t1, t0);
+       }
+
+       sort(t, ARRAY_SIZE(t), sizeof(*t), wrap_ktime_compare, NULL);
+       pr_info("%s src(%s, %s) -> dst(%s, %s) memcpy %llu KiB copy: %lld MiB/s\n",
+               __func__,
+               src_mr->name,
+               stringify_map_type(src_type),
+               dst_mr->name,
+               stringify_map_type(dst_type),
+               size >> 10,
+               div64_u64(mul_u32_u32(4 * size,
+                                     1000 * 1000 * 1000),
+                         t[1] + 2 * t[2] + t[3]) >> 20);
+
+       i915_gem_object_unpin_map(dst);
+       __i915_gem_object_put_pages(dst);
+out_put_dst:
+       i915_gem_object_put(dst);
+out_unpin_src:
+       i915_gem_object_unpin_map(src);
+       __i915_gem_object_put_pages(src);
+out_put_src:
+       i915_gem_object_put(src);
+out:
+       if (ret == -ENODEV || ret == -ENOMEM)
+               ret = 0;
+
+       return ret;
+}
+
+int perf_memcpy(void *arg)
+{
+       struct drm_i915_private *i915 = arg;
+       static const u32 types[] = {
+               I915_MAP_WB,
+               I915_MAP_WC,
+       };
+       static const u32 sizes[] = {
+               SZ_4K,
+               SZ_64K,
+               SZ_2M,
+               SZ_64M,
+       };
+       struct intel_memory_region *src_mr, *dst_mr;
+       int src_id, dst_id;
+       int i, j, k;
+       int ret;
+
+       for_each_memory_region(src_mr, i915, src_id) {
+               for_each_memory_region(dst_mr, i915, dst_id) {

drivers/gpu/drm/i915/selftests/intel_memory_region.c static int igt_lmem_write_cpu(void *arg) · line 715

Quoted 10 lines
drivers/gpu/drm/i915/selftests/intel_memory_region.c +10 -0
+                       for (i = 0; i < ARRAY_SIZE(sizes); ++i) {
+                               for (j = 0; j < ARRAY_SIZE(types); ++j) {
+                                       for (k = 0; k < ARRAY_SIZE(types); ++k) {
+                                               ret = _perf_memcpy(src_mr,
+                                                                  dst_mr,
+                                                                  sizes[i],
+                                                                  types[j],
+                                                                  types[k]);
+                                               if (ret)
+                                                       return ret;

64M from wc could be quite slow. I might put a caveat here to stop measuring larger sizes if we exceed a time limit. -Chris
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#2 lore raw

[Intel-gfx] ✗ Fi.CI.CHECKPATCH: warning for drm/i915/selftests/perf: measure memcpy bw between regions

In reply to Matthew Auld
== Series Details ==

== Series Details ==

Series: drm/i915/selftests/perf: measure memcpy bw between regions
URL : https://patchwork.freedesktop.org/series/72621/
State : warning

== Summary ==

$ dim checkpatch origin/drm-tip
a5ec4dca083a drm/i915/selftests/perf: measure memcpy bw between regions
-:175: WARNING:DEEP_INDENTATION: Too many leading tabs - consider code refactoring
#175: FILE: drivers/gpu/drm/i915/selftests/intel_memory_region.c:714:
+						if (ret)

total: 0 errors, 1 warnings, 0 checks, 183 lines checked

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

#3 lore raw

[Intel-gfx] ✗ Fi.CI.SPARSE: warning for drm/i915/selftests/perf: measure memcpy bw between regions

In reply to Matthew Auld
== Series Details ==

== Series Details ==

Series: drm/i915/selftests/perf: measure memcpy bw between regions
URL : https://patchwork.freedesktop.org/series/72621/
State : warning

== Summary ==

$ dim sparse origin/drm-tip
Sparse version: v0.6.0
Commit: drm/i915/selftests/perf: measure memcpy bw between regions
+drivers/gpu/drm/i915/selftests/intel_memory_region.c:613:5: warning: symbol '_perf_memcpy' was not declared. Should it be static?
+drivers/gpu/drm/i915/selftests/intel_memory_region.c:686:5: warning: symbol 'perf_memcpy' was not declared. Should it be static?

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

#4 lore raw

[Intel-gfx] ✗ Fi.CI.BAT: failure for drm/i915/selftests/perf: measure memcpy bw between regions

In reply to Matthew Auld
== Series Details ==

== Series Details ==

Series: drm/i915/selftests/perf: measure memcpy bw between regions
URL : https://patchwork.freedesktop.org/series/72621/
State : failure

== Summary ==

CI Bug Log - changes from CI_DRM_7827 -> Patchwork_16278
====================================================

Summary
-------

**FAILURE**

Serious unknown changes coming with Patchwork_16278 absolutely need to be verified manually.

If you think the reported changes have nothing to do with the changes introduced in Patchwork_16278, please notify your bug team to allow them to document this new failure mode, which will reduce false positives in CI.

External URL: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_16278/index.html

Possible new issues
-------------------

Here are the unknown changes that may have been introduced in Patchwork_16278:

### IGT changes ###

#### Possible regressions ####

* igt@i915_selftest@live_gem_contexts:
- fi-byt-j1900: [PASS][1] -> [DMESG-FAIL][2]
[1]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_7827/fi-byt-j1900/igt@i915_selftest@live_gem_contexts.html
[2]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_16278/fi-byt-j1900/igt@i915_selftest@live_gem_contexts.html

Known issues
------------

Here are the changes found in Patchwork_16278 that come from known issues:

### IGT changes ###

#### Issues hit ####

* igt@gem_close_race@basic-threads:
- fi-byt-j1900: [PASS][3] -> [TIMEOUT][4] ([fdo#112271] / [i915#816])
[3]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_7827/fi-byt-j1900/igt@[email protected]
[4]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_16278/fi-byt-j1900/igt@[email protected]

* igt@kms_chamelium@hdmi-hpd-fast:
- fi-kbl-7500u: [PASS][5] -> [FAIL][6] ([fdo#111407])
[5]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_7827/fi-kbl-7500u/igt@[email protected]
[6]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_16278/fi-kbl-7500u/igt@[email protected]

#### Possible fixes ####

* igt@i915_selftest@live_blt:
- fi-hsw-4770: [DMESG-FAIL][7] ([i915#553] / [i915#725]) -> [PASS][8]
[7]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_7827/fi-hsw-4770/igt@i915_selftest@live_blt.html
[8]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_16278/fi-hsw-4770/igt@i915_selftest@live_blt.html

* igt@i915_selftest@live_hangcheck:
- fi-icl-u3: [INCOMPLETE][9] ([fdo#108569] / [i915#140]) -> [PASS][10]
[9]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_7827/fi-icl-u3/igt@i915_selftest@live_hangcheck.html
[10]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_16278/fi-icl-u3/igt@i915_selftest@live_hangcheck.html

[fdo#108569]: https://bugs.freedesktop.org/show_bug.cgi?id=108569
[fdo#111407]: https://bugs.freedesktop.org/show_bug.cgi?id=111407
[fdo#112271]: https://bugs.freedesktop.org/show_bug.cgi?id=112271
[i915#140]: https://gitlab.freedesktop.org/drm/intel/issues/140
[i915#553]: https://gitlab.freedesktop.org/drm/intel/issues/553
[i915#725]: https://gitlab.freedesktop.org/drm/intel/issues/725
[i915#816]: https://gitlab.freedesktop.org/drm/intel/issues/816

Participating hosts (46 -> 43)
------------------------------

Additional (5): fi-bsw-n3050 fi-gdg-551 fi-cfl-8109u fi-ivb-3770 fi-skl-6600u
Missing (8): fi-ilk-m540 fi-byt-squawks fi-bsw-cyan fi-ctg-p8600 fi-kbl-x1275 fi-kbl-7560u fi-byt-clapper fi-bdw-samus

Build changes
-------------

* CI: CI-20190529 -> None
* Linux: CI_DRM_7827 -> Patchwork_16278

CI-20190529: 20190529
CI_DRM_7827: c8969aeacfff681c83a800e82b0f18a6ab3e77ea @ git://anongit.freedesktop.org/gfx-ci/linux
IGT_5389: 966c58649dee31bb5bf2fad92f75ffd365968b81 @ git://anongit.freedesktop.org/xorg/app/intel-gpu-tools
Patchwork_16278: a5ec4dca083a83834ed122d8a96655ad4d05c205 @ git://anongit.freedesktop.org/gfx-ci/linux

== Linux commits ==

a5ec4dca083a drm/i915/selftests/perf: measure memcpy bw between regions

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_16278/index.html
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx