Skip to contentSkip to conversation

gem_wsim improvements

PATCH In review Tvrtko Ursulin 2 participants 33 messages Jun 17, 2020 – Jun 18, 2020 Arm/Amlogic Meson SoC support

lore
Share & tools
lore
#0 lore raw

cover gem_wsim improvements

From: Tvrtko Ursulin <[email protected]>

Ripping out of legacy code and adding some new features, plus bugfixes.

First bit is important so the tool can be refactored to support Gen11 and 12 properly. Working sets support is useful for simulating buffer contention and similar.

Tvrtko Ursulin (10):
gem_wsim: Rip out userspace balancing
gem_wsim: Buffer objects working sets and complex dependencies
gem_wsim: Show workload timing stats
gem_wsim: Move BO allocation to a helper
gem_wsim: Support random buffer sizes
gem_wsim: Support scaling workload batch durations
gem_wsim: Log max and active working set sizes in verbose mode
gem_wsim: Snippet of a workload extracted from carchase
gem_wsim: Implement device selection
gem_wsim: Fix calibration handling

 benchmarks/Makefile.am                        |    2 +-
 benchmarks/Makefile.sources                   |    6 -
 benchmarks/ewma.h                             |   71 -
 benchmarks/gem_wsim.c                         | 2054 ++++++-----------
 benchmarks/ilog2.h                            |  104 -
 benchmarks/meson.build                        |    6 +-
 benchmarks/wsim/README                        |   63 +
 benchmarks/wsim/carchasepart.wsim             |  184 ++
 benchmarks/wsim/cloud-gaming-60fps.wsim       |   11 +
 benchmarks/wsim/composited-ui.wsim            |    7 +
 benchmarks/wsim/media-1080p-player.wsim       |    2 +
 benchmarks/wsim/media_1n2_480p.wsim           |   12 +-
 benchmarks/wsim/media_1n2_asy.wsim            |    8 +-
 benchmarks/wsim/media_1n3_480p.wsim           |   16 +-
 benchmarks/wsim/media_1n3_asy.wsim            |    8 +
 benchmarks/wsim/media_1n4_480p.wsim           |   20 +-
 benchmarks/wsim/media_1n4_asy.wsim            |   10 +
 benchmarks/wsim/media_1n5_480p.wsim           |   24 +-
 benchmarks/wsim/media_1n5_asy.wsim            |   12 +
 benchmarks/wsim/media_load_balance_17i7.wsim  |   10 +-
 benchmarks/wsim/media_load_balance_19.wsim    |    4 +-
 .../wsim/media_load_balance_4k12u7.wsim       |    2 +
 .../wsim/media_load_balance_fhd26u7.wsim      |   16 +-
 benchmarks/wsim/media_load_balance_hd01.wsim  |   34 +-
 .../wsim/media_load_balance_hd06mp2.wsim      |    6 +-
 benchmarks/wsim/media_load_balance_hd12.wsim  |    6 +-
 .../wsim/media_load_balance_hd17i4.wsim       |    8 +-
 benchmarks/wsim/media_mfe2_480p.wsim          |   12 +-
 benchmarks/wsim/media_mfe3_480p.wsim          |   18 +-
 benchmarks/wsim/media_mfe4_480p.wsim          |   24 +-
 benchmarks/wsim/media_nn_1080p.wsim           |    4 +
 benchmarks/wsim/media_nn_1080p_s1.wsim        |    4 +-
 benchmarks/wsim/media_nn_1080p_s2.wsim        |    2 +
 benchmarks/wsim/media_nn_1080p_s3.wsim        |    2 +
 benchmarks/wsim/media_nn_480p.wsim            |    4 +
 benchmarks/wsim/vcs_balanced.wsim             |   52 +-
 scripts/media-bench.pl                        |  736 ------
 37 files changed, 1171 insertions(+), 2393 deletions(-)
 delete mode 100644 benchmarks/ewma.h
 delete mode 100644 benchmarks/ilog2.h
 create mode 100644 benchmarks/wsim/carchasepart.wsim
 create mode 100644 benchmarks/wsim/cloud-gaming-60fps.wsim
 create mode 100644 benchmarks/wsim/composited-ui.wsim
 delete mode 100755 scripts/media-bench.pl
Signature · 6 lines
2.20.1

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

PATCH 1/10 gem_wsim: Rip out userspace balancing

From: Tvrtko Ursulin <[email protected]>

Evaluation of userspace load balancing options was how this tool started but since we have settled on doing it in the kernel.

Tomorrow we will want to update the tool for new engine interfaces and all this legacy code will just be a distraction.

Rip out everything not related to explicit load balancing implemented via context engine maps and adjust the workloads to use it.

---
 benchmarks/Makefile.am                        |    2 +-
 benchmarks/Makefile.sources                   |    6 -
 benchmarks/ewma.h                             |   71 -
 benchmarks/gem_wsim.c                         | 1362 +----------------
 benchmarks/ilog2.h                            |  104 --
 benchmarks/meson.build                        |    6 +-
 benchmarks/wsim/media-1080p-player.wsim       |    2 +
 benchmarks/wsim/media_1n2_480p.wsim           |   12 +-
 benchmarks/wsim/media_1n2_asy.wsim            |    8 +-
 benchmarks/wsim/media_1n3_480p.wsim           |   16 +-
 benchmarks/wsim/media_1n3_asy.wsim            |    8 +
 benchmarks/wsim/media_1n4_480p.wsim           |   20 +-
 benchmarks/wsim/media_1n4_asy.wsim            |   10 +
 benchmarks/wsim/media_1n5_480p.wsim           |   24 +-
 benchmarks/wsim/media_1n5_asy.wsim            |   12 +
 benchmarks/wsim/media_load_balance_17i7.wsim  |   10 +-
 benchmarks/wsim/media_load_balance_19.wsim    |    4 +-
 .../wsim/media_load_balance_4k12u7.wsim       |    2 +
 .../wsim/media_load_balance_fhd26u7.wsim      |   16 +-
 benchmarks/wsim/media_load_balance_hd01.wsim  |   34 +-
 .../wsim/media_load_balance_hd06mp2.wsim      |    6 +-
 benchmarks/wsim/media_load_balance_hd12.wsim  |    6 +-
 .../wsim/media_load_balance_hd17i4.wsim       |    8 +-
 benchmarks/wsim/media_mfe2_480p.wsim          |   12 +-
 benchmarks/wsim/media_mfe3_480p.wsim          |   18 +-
 benchmarks/wsim/media_mfe4_480p.wsim          |   24 +-
 benchmarks/wsim/media_nn_1080p.wsim           |    4 +
 benchmarks/wsim/media_nn_1080p_s1.wsim        |    4 +-
 benchmarks/wsim/media_nn_1080p_s2.wsim        |    2 +
 benchmarks/wsim/media_nn_1080p_s3.wsim        |    2 +
 benchmarks/wsim/media_nn_480p.wsim            |    4 +
 benchmarks/wsim/vcs_balanced.wsim             |   52 +-
 scripts/media-bench.pl                        |  736 ---------
 33 files changed, 300 insertions(+), 2307 deletions(-)
 delete mode 100644 benchmarks/ewma.h
 delete mode 100644 benchmarks/ilog2.h
 delete mode 100755 scripts/media-bench.pl
benchmarks/Makefile.am +1 -1
index 1f05adf31527..45b923ebbae3 100644
@@ -25,4 +25,4 @@ gem_latency_CFLAGS = $(AM_CFLAGS) $(THREAD_CFLAGS)
 gem_latency_LDADD = $(LDADD) -lpthread
 gem_syslatency_CFLAGS = $(AM_CFLAGS) $(THREAD_CFLAGS)
 gem_syslatency_LDADD = $(LDADD) -lpthread
-gem_wsim_LDADD = $(LDADD) $(top_builddir)/lib/libigt_perf.la -lpthread
+gem_wsim_LDADD = $(LDADD) -lpthread
benchmarks/Makefile.sources +0 -6
index ee045fb309ad..dae3cdda4cf7 100644
@@ -19,12 +19,6 @@ benchmarks_prog_list =			\
 	vgem_mmap			\
 	$(NULL)
 
-gem_wsim_SOURCES =                      \
-	gem_wsim.c                      \
-	ewma.h                          \
-	ilog2.h                         \
-	$(NULL)
-
 LIBDRM_INTEL_BENCHMARKS =		\
 	intel_upload_blit_large		\
 	intel_upload_blit_large_gtt	\
benchmarks/ewma.h deleted +0 -71

71 lines not on this page · raw message

benchmarks/gem_wsim.c +65 -1297

1,766 lines not on this page · raw message

benchmarks/ilog2.h deleted +0 -104

104 lines not on this page · raw message

benchmarks/meson.build +1 -5
index ef93193b70dd..c70e1aac79c6 100644
@@ -11,6 +11,7 @@ benchmark_progs = [
 	'gem_prw',
 	'gem_set_domain',
 	'gem_syslatency',
+	'gem_wsim',
 	'kms_vblank',
 	'prime_lookup',
 	'vgem_mmap',
@@ -34,8 +35,3 @@ foreach prog : benchmark_progs
 		   install_dir : benchmarksdir,
 		   dependencies : igt_deps)
 endforeach
-
-executable('gem_wsim', 'gem_wsim.c',
-	   install : true,
-	   install_dir : benchmarksdir,
-	   dependencies : igt_deps + [ lib_igt_perf ])
benchmarks/wsim/media-1080p-player.wsim +2 -0
index bcbb0cfd2ad3..c87e1aee4f5d 100644
@@ -1,3 +1,5 @@
+M.1.VCS
+B.1
 1.VCS.5000-10000.0.0
 2.RCS.1000-2000.-1.0
 P.3.1
benchmarks/wsim/media_1n2_480p.wsim +9 -3
index 11a4da6bfae8..3ce15ebc3d71 100644
@@ -1,9 +1,15 @@
-1.VCS.12000-15000.0.0
+M.10.VCS
+B.10
+M.11.VCS
+B.11
+M.12.VCS
+B.12
+10.VCS.12000-15000.0.0
 2.RCS.1000-2200.-1.0
 3.RCS.1000-1400.-1.0
 3.RCS.10000-12000.0.0
-3.VCS.2500-3500.-1.0
+11.VCS.2500-3500.-1.0
 4.RCS.1000-2200.-5.0
 5.RCS.1000-1400.-1.0
 5.RCS.10000-12000.0.0
-5.VCS.2500-3500.-1.1
+12.VCS.2500-3500.-1.1
benchmarks/wsim/media_1n2_asy.wsim +5 -3
index 58c99ca1122c..f9943eb62e8a 100644
@@ -1,9 +1,11 @@
-1.VCS.12000-15000.0.0
+M.10.VCS
+B.10
+10.VCS.12000-15000.0.0
 2.RCS.1000-2200.-1.0
 3.RCS.1000-1400.-1.0
 3.RCS.10000-12000.0.0
-3.VCS.2500-3500.-1.0
+11.VCS.2500-3500.-1.0
 4.RCS.400-800.-5.0
 5.RCS.500-700.-1.0
 5.RCS.5000-6000.0.0
-5.VCS.1200-1500.-1.1
+12.VCS.1200-1500.-1.1
benchmarks/wsim/media_1n3_480p.wsim +12 -4
index c724ab28a1f4..4f585fa8a8e0 100644
@@ -1,13 +1,21 @@
-1.VCS.12000-15000.0.0
+M.10.VCS
+B.10
+M.11.VCS
+B.11
+M.12.VCS
+B.12
+M.13.VCS
+B.13
+10.VCS.12000-15000.0.0
 2.RCS.1000-2200.-1.0
 3.RCS.1000-1400.-1.0
 3.RCS.10000-12000.0.0
-3.VCS.2500-3500.-1.0
+11.VCS.2500-3500.-1.0
 4.RCS.1000-2200.-5.0
 5.RCS.1000-1400.-1.0
 5.RCS.10000-12000.0.0
-5.VCS.2500-3500.-1.0
+12.VCS.2500-3500.-1.0
 6.RCS.1000-2200.-9.0
 7.RCS.1000-1400.-1.0
 7.RCS.10000-12000.0.0
-7.VCS.2500-3500.-1.1
+13.VCS.2500-3500.-1.1
benchmarks/wsim/media_1n3_asy.wsim +8 -0
index c7588328e3f1..dce7789ec1d8 100644
@@ -1,3 +1,11 @@
+M.10.VCS
+B.10
+M.11.VCS
+B.11
+M.12.VCS
+B.12
+M.13.VCS
+B.13
 1.VCS.12000-15000.0.0
 2.RCS.1000-2200.-1.0
 3.RCS.1000-1400.-1.0
benchmarks/wsim/media_1n4_480p.wsim +15 -5
index e67fefc3bf17..06fa9adef5eb 100644
@@ -1,17 +1,27 @@
-1.VCS.12000-15000.0.0
+M.10.VCS
+B.10
+M.11.VCS
+B.11
+M.12.VCS
+B.12
+M.13.VCS
+B.13
+M.14.VCS
+B.14
+10.VCS.12000-15000.0.0
 2.RCS.1000-2200.-1.0
 3.RCS.1000-1400.-1.0
 3.RCS.10000-12000.0.0
-3.VCS.2500-3500.-1.0
+11.VCS.2500-3500.-1.0
 4.RCS.1000-2200.-5.0
 5.RCS.1000-1400.-1.0
 5.RCS.10000-12000.0.0
-5.VCS.2500-3500.-1.0
+12.VCS.2500-3500.-1.0
 6.RCS.1000-2200.-9.0
 7.RCS.1000-1400.-1.0
 7.RCS.10000-12000.0.0
-7.VCS.2500-3500.-1.0
+13.VCS.2500-3500.-1.0
 8.RCS.1000-2200.-13.0
 9.RCS.1000-1400.-1.0
 9.RCS.10000-12000.0.0
-9.VCS.2500-3500.-1.1
+14.VCS.2500-3500.-1.1
benchmarks/wsim/media_1n4_asy.wsim +10 -0
index ede4fd7a2205..6dc6b652e903 100644
@@ -1,3 +1,13 @@
+M.10.VCS
+B.10
+M.11.VCS
+B.11
+M.12.VCS
+B.12
+M.13.VCS
+B.13
+M.14.VCS
+B.14
 1.VCS.12000-15000.0.0
 2.RCS.1000-2200.-1.0
 3.RCS.1000-1400.-1.0
benchmarks/wsim/media_1n5_480p.wsim +18 -6
index 9e43b9845430..3467a386887a 100644
@@ -1,21 +1,33 @@
-1.VCS.12000-15000.0.0
+M.10.VCS
+B.10
+M.11.VCS
+B.11
+M.12.VCS
+B.12
+M.13.VCS
+B.13
+M.14.VCS
+B.14
+M.15.VCS
+B.15
+10.VCS.12000-15000.0.0
 2.RCS.1000-2200.-1.0
 3.RCS.1000-1400.-1.0
 3.RCS.10000-12000.0.0
-3.VCS.2500-3500.-1.0
+11.VCS.2500-3500.-1.0
 4.RCS.1000-2200.-5.0
 5.RCS.1000-1400.-1.0
 5.RCS.10000-12000.0.0
-5.VCS.2500-3500.-1.0
+12.VCS.2500-3500.-1.0
 6.RCS.1000-2200.-9.0
 7.RCS.1000-1400.-1.0
 7.RCS.10000-12000.0.0
-7.VCS.2500-3500.-1.0
+13.VCS.2500-3500.-1.0
 8.RCS.1000-2200.-13.0
 9.RCS.1000-1400.-1.0
 9.RCS.10000-12000.0.0
-9.VCS.2500-3500.-1.0
+14.VCS.2500-3500.-1.0
 10.RCS.1000-2200.-17.0
 11.RCS.1000-1400.-1.0
 11.RCS.10000-12000.0.0
-11.VCS.2500-3500.-1.1
+15.VCS.2500-3500.-1.1
benchmarks/wsim/media_1n5_asy.wsim +12 -0
index 78bb4a86dbca..4b205457a8d4 100644
@@ -1,3 +1,15 @@
+M.10.VCS
+B.10
+M.11.VCS
+B.11
+M.12.VCS
+B.12
+M.13.VCS
+B.13
+M.14.VCS
+B.14
+M.15.VCS
+B.15
 1.VCS.12000-15000.0.0
 2.RCS.2000-3000.-1.0
 3.RCS.500-900.-1.0
benchmarks/wsim/media_load_balance_17i7.wsim +6 -4
index 0830a3231ea9..bcb1ab2f04fa 100644
@@ -1,7 +1,9 @@
+M.1.VCS
+B.1
 1.VCS.2800-3200.0.1
-1.RCS.900-1100.-1.0
-1.RCS.3600-3800.0.0
-1.RCS.900-1100.-2.0
+2.RCS.900-1100.-1.0
+2.RCS.3600-3800.0.0
+2.RCS.900-1100.-2.0
 1.VCS.2200-2400.-2.0
-1.RCS.4500-4900.-1.0
+2.RCS.4500-4900.-1.0
 1.VCS.500-700.-1.1
benchmarks/wsim/media_load_balance_19.wsim +3 -1
index 03890776fda3..88cd34fb6898 100644
@@ -1,3 +1,5 @@
+M.1.VCS
+B.1
 0.VECS.1400-1500.0.0
 0.RCS.1000-1500.-1.0
 s.-2
@@ -5,6 +7,6 @@ s.-2
 1.VCS.1300-1400.0.1
 0.VECS.1400-1500.0.0
 0.RCS.100-300.-1.1
-1.RCS.1300-1500.0.0
+2.RCS.1300-1500.-3.0
 1.VCS.100-300.-1.1
 1.VCS.900-1400.0.1
benchmarks/wsim/media_load_balance_4k12u7.wsim +2 -0
index ff10425b6bec..a417bb18e121 100644
@@ -1,3 +1,5 @@
+M.1.VCS
+B.1
 1.VCS.4000-6000.0.0
 2.RCS.400-800.-1.0
 3.RCS.1900-2200.-1.0
benchmarks/wsim/media_load_balance_fhd26u7.wsim +9 -7
index 56114ddc48c2..4c8225e1fe13 100644
@@ -1,25 +1,27 @@
+M.3.VCS
+B.3
 1.VCS1.1200-1800.0.0
 1.VCS1.1900-2100.0.0
 2.RCS.1500-2000.-1.0
-2.VCS.1400-1800.-1.1
+3.VCS.1400-1800.-1.1
 1.VCS1.1900-2100.-1.0
 2.RCS.1500-2000.-1.0
-2.VCS.1400-1800.-1.1
+3.VCS.1400-1800.-1.1
 1.VCS1.1900-2100.-1.0
 2.RCS.200-400.-1.0
 2.RCS.1500-2000.0.0
-2.VCS.1400-1800.-1.1
+3.VCS.1400-1800.-1.1
 1.VCS1.1900-2100.-1.0
 2.RCS.1500-2000.-1.0
-2.VCS.1400-1800.-1.1
+3.VCS.1400-1800.-1.1
 1.VCS1.1900-2100.-1.0
 2.RCS.200-400.-1.0
 2.RCS.1500-2000.0.0
-2.VCS.1400-1800.-1.1
+3.VCS.1400-1800.-1.1
 1.VCS1.1900-2100.-1.0
 2.RCS.1500-2000.-1.0
-2.VCS.1400-1800.-1.1
+3.VCS.1400-1800.-1.1
 1.VCS1.1900-2100.-1.0
 2.RCS.1500-2000.-1.0
 2.RCS.1500-2000.0.0
-2.VCS.1400-1800.-1.1
+3.VCS.1400-1800.-1.1
benchmarks/wsim/media_load_balance_hd01.wsim +19 -15
index 862931521c90..8e7e9d90e435 100644
@@ -1,23 +1,27 @@
+M.1.VCS
+B.1
+M.3.VCS
+B.3
 1.VCS.1400-1900.0.0
-1.RCS.1200-1600.-1.0
-1.RCS.1000-1400.-1.0
-2.VCS.800-1000.-1.0
+2.RCS.1200-1600.-1.0
+2.RCS.1000-1400.-1.0
+3.VCS.800-1000.-1.0
 1.VCS.1400-1900.-4.0
-1.RCS.1200-1600.-1.0
-1.RCS.1000-1400.-1.0
-2.VCS.800-1000.-1.0
+2.RCS.1200-1600.-1.0
+2.RCS.1000-1400.-1.0
+3.VCS.800-1000.-1.0
 1.VCS.1400-1900.-4.0
-1.RCS.1200-1600.-1.0
-1.RCS.1000-1400.-1.0
-2.VCS.800-1000.-1.0
+2.RCS.1200-1600.-1.0
+2.RCS.1000-1400.-1.0
+3.VCS.800-1000.-1.0
 1.VCS.1400-1900.-4.0
-1.RCS.1200-1600.-1.0
-1.RCS.1000-1400.-1.0
-2.VCS.800-1000.-1.0
+2.RCS.1200-1600.-1.0
+2.RCS.1000-1400.-1.0
+3.VCS.800-1000.-1.0
 1.VCS.1400-1900.-4.0
-1.RCS.1200-1600.-1.0
-1.RCS.1000-1400.-1.0
-2.VCS.800-1000.-1.0
+2.RCS.1200-1600.-1.0
+2.RCS.1000-1400.-1.0
+3.VCS.800-1000.-1.0
 s.-17
 s.-14
 s.-11
benchmarks/wsim/media_load_balance_hd06mp2.wsim +5 -1
index 1e1fc003c755..cfe985019a7b 100644
@@ -1,4 +1,8 @@
+M.1.VCS
+B.1
+M.4.VCS
+B.4
 1.VCS.900-1700.0.0
 2.RCS.100-400.-1.0
 3.RCS.800-900.-1.0
-3.VCS.100-200.-1.1
+4.VCS.100-200.-1.1
benchmarks/wsim/media_load_balance_hd12.wsim +5 -1
index 8f3b41ca5ab6..684e6b511762 100644
@@ -1,4 +1,8 @@
+M.1.VCS
+B.1
+M.4.VCS
+B.4
 1.VCS.850-1300.0.0
 2.RCS.50-250.-1.0
 3.RCS.400-800.-1.0
-3.VCS.100-200.-1.1
+4.VCS.100-200.-1.1
benchmarks/wsim/media_load_balance_hd17i4.wsim +6 -2
index b6195b605bf7..1430f18df033 100644
@@ -1,7 +1,11 @@
+M.1.VCS
+B.1
+M.3.VCS
+B.3
 1.VCS.900-1400.0.0
 2.RCS.200-300.-1.0
 2.RCS.1000-2000.0.0
 2.RCS.1000-2000.0.0
-2.VCS.800-1000.-1.0
-1.RCS.2800-3100.-1.0
+3.VCS.800-1000.-1.0
+4.RCS.2800-3100.-1.0
 1.VCS.800-1000.-1.1
benchmarks/wsim/media_mfe2_480p.wsim +10 -2
index 18bc756f1b55..00ef5c3a7574 100644
@@ -1,3 +1,11 @@
+M.1.VCS
+B.1
+M.4.VCS
+B.4
+M.7.VCS
+B.7
+M.8.VCS
+B.8
 1.VCS.12000-15000.0.0
 2.RCS.1000-2200.-1.0
 3.RCS.800-1600.-1.0
@@ -5,5 +13,5 @@
 5.RCS.1000-2200.-1.0
 6.RCS.800-1600.-1.0
 6.RCS.10000-12000.-4.0
-6.VCS.2500-3500.-1.0
-3.VCS.2500-3500.-2.1
+7.VCS.2500-3500.-1.0
+8.VCS.2500-3500.-2.1
benchmarks/wsim/media_mfe3_480p.wsim +15 -3
index e12a2e6ac29d..3ac4db0eb8ec 100644
@@ -1,3 +1,15 @@
+M.1.VCS
+B.1
+M.4.VCS
+B.4
+M.7.VCS
+B.7
+M.10.VCS
+B.10
+M.11.VCS
+B.11
+M.12.VCS
+B.12
 1.VCS.12000-15000.0.0
 2.RCS.1000-2200.-1.0
 3.RCS.800-1600.-1.0
@@ -8,6 +20,6 @@
 8.RCS.1000-2200.-1.0
 9.RCS.800-1600.-1.0
 9.RCS.10000-12000.-7/-4.0
-9.VCS.2500-3500.-1.0
-3.VCS.2500-3500.-2.0
-6.VCS.2500-3500.-3.1
+10.VCS.2500-3500.-1.0
+11.VCS.2500-3500.-2.0
+12.VCS.2500-3500.-3.1
benchmarks/wsim/media_mfe4_480p.wsim +20 -4
index 75d4f67ea4fb..7f6831569908 100644
@@ -1,3 +1,19 @@
+M.1.VCS
+B.1
+M.4.VCS
+B.4
+M.7.VCS
+B.7
+M.10.VCS
+B.10
+M.13.VCS
+B.13
+M.14.VCS
+B.14
+M.15.VCS
+B.15
+M.16.VCS
+B.16
 1.VCS.12000-15000.0.0
 2.RCS.1000-2200.-1.0
 3.RCS.800-1600.-1.0
@@ -11,7 +27,7 @@
 11.RCS.1000-2200.-1.0
 12.RCS.800-1600.-1.0
 12.RCS.10000-12000.-4/-7/-10.0
-12.VCS.2500-3500.-1.0
-3.VCS.2500-3500.-2.0
-6.VCS.2500-3500.-3.0
-9.VCS.2500-3500.-4.1
+13.VCS.2500-3500.-1.0
+14.VCS.2500-3500.-2.0
+15.VCS.2500-3500.-3.0
+16.VCS.2500-3500.-4.1
benchmarks/wsim/media_nn_1080p.wsim +4 -0
index f9a3ca1b9963..88c5c772202c 100644
@@ -1,3 +1,7 @@
+M.1.VCS
+B.1
+M.3.VCS
+B.3
 1.VCS.13000-17000.0.0
 2.RCS.2000-4000.-1.0
 3.RCS.3000-5000.-1.0
benchmarks/wsim/media_nn_1080p_s1.wsim +3 -1
index 4fa6ca653000..5b47d2a3c7ec 100644
@@ -1,3 +1,5 @@
+M.4.VCS
+B.4
 f
 1.VCS1.6500-8000.f-1.0
 1.VCS2.6500-8000.f-2.0
@@ -5,4 +7,4 @@ a.-3
 2.RCS.2000-4000.-2/-3.0
 3.RCS.3000-5000.-1.0
 3.RCS.23000-27000.0.0
-3.VCS.16000-20000.-1.1
+4.VCS.16000-20000.-1.1
benchmarks/wsim/media_nn_1080p_s2.wsim +2 -0
index 68f0acdfb842..e3678b396b42 100644
@@ -1,3 +1,5 @@
+M.1.VCS
+B.1
 1.VCS.13000-17000.0.0
 2.RCS.2000-4000.-1.0
 3.RCS.3000-5000.-1.0
benchmarks/wsim/media_nn_1080p_s3.wsim +2 -0
index 12368da83dca..ee3b675de9e5 100644
@@ -1,3 +1,5 @@
+M.1.VCS
+B.1
 1.VCS.13000-17000.0.0
 2.RCS.2000-4000.-1.0
 3.RCS.3000-5000.-1.0
benchmarks/wsim/media_nn_480p.wsim +4 -0
index ab64a4569d71..73fc643dc9e5 100644
@@ -1,3 +1,7 @@
+M.1.VCS
+B.1
+M.3.VCS
+B.3
 1.VCS.12000-15000.0.0
 2.RCS.1000-2200.-1.0
 3.RCS.1000-1400.-1.0
benchmarks/wsim/vcs_balanced.wsim +27 -25

53 lines not on this page · raw message

scripts/media-bench.pl deleted +0 -736

559 lines not on this page · raw message

The rest of this diff is not shown. View the raw message.

Signature · 6 lines
2.20.1

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

Part of the diff is not shown here. View the raw message.

#2 lore raw

Re: [Intel-gfx] [PATCH i-g-t 01/10] gem_wsim: Rip out userspace balancing

Quoting Tvrtko Ursulin (2020-06-17 17:01:11)

Quoted 10 lines

From: Tvrtko Ursulin <[email protected]>

Evaluation of userspace load balancing options was how this tool started but since we have settled on doing it in the kernel.

Tomorrow we will want to update the tool for new engine interfaces and all this legacy code will just be a distraction.

Rip out everything not related to explicit load balancing implemented via context engine maps and adjust the workloads to use it.

I'm still using busy & contexts to ground i915. Any chance of a reprieve? At least for context with -R?

Or shall I just keep a pristine copy?
-Chris
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#3 lore raw

Re: [Intel-gfx] [PATCH i-g-t 01/10] gem_wsim: Rip out userspace balancing

In reply to Tvrtko Ursulin

Quoting Tvrtko Ursulin (2020-06-17 17:01:11)

Quoted 10 lines

From: Tvrtko Ursulin <[email protected]>

Evaluation of userspace load balancing options was how this tool started but since we have settled on doing it in the kernel.

Tomorrow we will want to update the tool for new engine interfaces and all this legacy code will just be a distraction.

Rip out everything not related to explicit load balancing implemented via context engine maps and adjust the workloads to use it.

Hmm, if this is on the table, should we also then restrict load-balancing wsim to gen11+ so that we can use the timed loops rather nop batches? That would be a huge selling point, and I'll just keep an old checkout around for nop load balancing with all the trimmings. -Chris
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#4 lore raw

Re: [Intel-gfx] [PATCH i-g-t 01/10] gem_wsim: Rip out userspace balancing

Quoted 9 lines · Chris Wilson

Quoting Tvrtko Ursulin (2020-06-17 17:01:11)

Quoted 8 lines

From: Tvrtko Ursulin <[email protected]>

Evaluation of userspace load balancing options was how this tool started but since we have settled on doing it in the kernel.

Tomorrow we will want to update the tool for new engine interfaces and all this legacy code will just be a distraction.

Quoted 7 lines
Quoted 2 lines

Rip out everything not related to explicit load balancing implemented via context engine maps and adjust the workloads to use it.

Hmm, if this is on the table, should we also then restrict load-balancing wsim to gen11+ so that we can use the timed loops rather nop batches? That would be a huge selling point, and I'll just keep an old checkout around for nop load balancing with all the trimmings.

That was my plan for the next step yes. Just taking your patch without further changes would already make it work I think. But also at some point I want to convert the engine selection (and engine naming in descriptors) to class:instance.

Why do you need the nop/old balancing stuff? I would hope going forward we only need to compare current balancing against any changes. So I'd really like to remoev the userspace balancing stuff.

Regards,

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

#5 lore raw

Re: [Intel-gfx] [PATCH i-g-t 01/10] gem_wsim: Rip out userspace balancing

Quoting Tvrtko Ursulin (2020-06-18 08:40:25)

Quoted 19 lines · Chris Wilson
Quoted 9 lines

Quoting Tvrtko Ursulin (2020-06-17 17:01:11)

Quoted 8 lines

From: Tvrtko Ursulin <[email protected]>

Evaluation of userspace load balancing options was how this tool started but since we have settled on doing it in the kernel.

Tomorrow we will want to update the tool for new engine interfaces and all this legacy code will just be a distraction.

Quoted 7 lines
Quoted 2 lines

Rip out everything not related to explicit load balancing implemented via context engine maps and adjust the workloads to use it.

Hmm, if this is on the table, should we also then restrict load-balancing wsim to gen11+ so that we can use the timed loops rather nop batches? That would be a huge selling point, and I'll just keep an old checkout around for nop load balancing with all the trimmings.

Quoted 8 lines

That was my plan for the next step yes. Just taking your patch without further changes would already make it work I think. But also at some point I want to convert the engine selection (and engine naming in descriptors) to class:instance.

Why do you need the nop/old balancing stuff? I would hope going forward we only need to compare current balancing against any changes. So I'd really like to remoev the userspace balancing stuff.

There are still some cases where i915 is beaten by plain old contexts, usually that is a combination of semaphores and interrupt latency, but some I just don't understand. There is still an uncomfortably large variation between kernel releases, and comparing the regressions in different balancers is useful to narrow down the problem. -Chris
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#6 lore raw

Re: [Intel-gfx] [PATCH i-g-t 01/10] gem_wsim: Rip out userspace balancing

Quoted 29 lines · Chris Wilson

Quoting Tvrtko Ursulin (2020-06-18 08:40:25)

Quoted 19 lines · Chris Wilson
Quoted 9 lines

Quoting Tvrtko Ursulin (2020-06-17 17:01:11)

Quoted 8 lines

From: Tvrtko Ursulin <[email protected]>

Evaluation of userspace load balancing options was how this tool started but since we have settled on doing it in the kernel.

Tomorrow we will want to update the tool for new engine interfaces and all this legacy code will just be a distraction.

Quoted 7 lines
Quoted 2 lines

Rip out everything not related to explicit load balancing implemented via context engine maps and adjust the workloads to use it.

Hmm, if this is on the table, should we also then restrict load-balancing wsim to gen11+ so that we can use the timed loops rather nop batches? That would be a huge selling point, and I'll just keep an old checkout around for nop load balancing with all the trimmings.

Quoted 8 lines

That was my plan for the next step yes. Just taking your patch without further changes would already make it work I think. But also at some point I want to convert the engine selection (and engine naming in descriptors) to class:instance.

Why do you need the nop/old balancing stuff? I would hope going forward we only need to compare current balancing against any changes. So I'd really like to remoev the userspace balancing stuff.

Quoted 5 lines

There are still some cases where i915 is beaten by plain old contexts, usually that is a combination of semaphores and interrupt latency, but some I just don't understand. There is still an uncomfortably large variation between kernel releases, and comparing the regressions in different balancers is useful to narrow down the problem.

You could create separate workloads to simulate "-b context" to a degree? I really want to rip this out. Can you cut your losses and forget it existed? :)

Regards,

Tvrtko

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

#7 lore raw

Re: [Intel-gfx] [PATCH i-g-t 01/10] gem_wsim: Rip out userspace balancing

Quoting Tvrtko Ursulin (2020-06-18 11:03:11)

Quoted 37 lines · Chris Wilson
Quoted 29 lines

Quoting Tvrtko Ursulin (2020-06-18 08:40:25)

Quoted 19 lines · Chris Wilson
Quoted 9 lines

Quoting Tvrtko Ursulin (2020-06-17 17:01:11)

Quoted 8 lines

From: Tvrtko Ursulin <[email protected]>

Evaluation of userspace load balancing options was how this tool started but since we have settled on doing it in the kernel.

Tomorrow we will want to update the tool for new engine interfaces and all this legacy code will just be a distraction.

Quoted 7 lines
Quoted 2 lines

Rip out everything not related to explicit load balancing implemented via context engine maps and adjust the workloads to use it.

Hmm, if this is on the table, should we also then restrict load-balancing wsim to gen11+ so that we can use the timed loops rather nop batches? That would be a huge selling point, and I'll just keep an old checkout around for nop load balancing with all the trimmings.

Quoted 8 lines

That was my plan for the next step yes. Just taking your patch without further changes would already make it work I think. But also at some point I want to convert the engine selection (and engine naming in descriptors) to class:instance.

Why do you need the nop/old balancing stuff? I would hope going forward we only need to compare current balancing against any changes. So I'd really like to remoev the userspace balancing stuff.

Quoted 5 lines

There are still some cases where i915 is beaten by plain old contexts, usually that is a combination of semaphores and interrupt latency, but some I just don't understand. There is still an uncomfortably large variation between kernel releases, and comparing the regressions in different balancers is useful to narrow down the problem.

Quoted 3 lines

You could create separate workloads to simulate "-b context" to a degree? I really want to rip this out. Can you cut your losses and forget it existed? :)

It's fine; I can keep a stable benchmark around of a known checkout. -Chris
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#8 lore raw

PATCH 2/10 gem_wsim: Buffer objects working sets and complex dependencies

In reply to Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

Add support for defining buffer object working sets and targetting them as data dependencies. For more information please see the README file.

---
 benchmarks/gem_wsim.c                   | 453 +++++++++++++++++++++---
 benchmarks/wsim/README                  |  59 +++
 benchmarks/wsim/cloud-gaming-60fps.wsim |  11 +
 benchmarks/wsim/composited-ui.wsim      |   7 +
 4 files changed, 476 insertions(+), 54 deletions(-)
 create mode 100644 benchmarks/wsim/cloud-gaming-60fps.wsim
 create mode 100644 benchmarks/wsim/composited-ui.wsim
benchmarks/gem_wsim.c +399 -54
index 02fe8f5a5e69..9e5bfe6a36d4 100644
@@ -88,14 +88,21 @@ enum w_type
 	LOAD_BALANCE,
 	BOND,
 	TERMINATE,
-	SSEU
+	SSEU,
+	WORKINGSET,
+};
+
+struct dep_entry {
+	int target;
+	bool write;
+	int working_set; /* -1 = step dependecy, >= 0 working set id */
 };
 
 struct deps
 {
 	int nr;
 	bool submit_fence;
-	int *list;
+	struct dep_entry *list;
 };
 
 struct w_arg {
@@ -110,6 +117,14 @@ struct bond {
 	enum intel_engine_id master;
 };
 
+struct working_set {
+	int id;
+	bool shared;
+	unsigned int nr;
+	uint32_t *handles;
+	unsigned long *sizes;
+};
+
 struct workload;
 
 struct w_step
@@ -143,6 +158,7 @@ struct w_step
 			enum intel_engine_id bond_master;
 		};
 		int sseu;
+		struct working_set working_set;
 	};
 
 	/* Implementation details */
@@ -193,6 +209,9 @@ struct workload
 	unsigned int nr_ctxs;
 	struct ctx *ctx_list;
 
+	struct working_set **working_sets; /* array indexed by set id */
+	int max_working_set_id;
+
 	int sync_timeline;
 	uint32_t sync_seqno;
 
@@ -281,11 +300,120 @@ print_engine_calibrations(void)
 	printf("\n");
 }
 
+static void add_dep(struct deps *deps, struct dep_entry entry)
+{
+	deps->list = realloc(deps->list, sizeof(*deps->list) * (deps->nr + 1));
+	igt_assert(deps->list);
+
+	deps->list[deps->nr++] = entry;
+}
+
+static int
+parse_working_set_deps(struct workload *wrk,
+		       struct deps *deps,
+		       struct dep_entry _entry,
+		       char *str)
+{
+	/*
+	 * 1 - target handle index in the specified working set.
+	 * 2-4 - range
+	 */
+	struct dep_entry entry = _entry;
+	char *s;
+
+	s = index(str, '-');
+	if (s) {
+		int from, to;
+
+		from = atoi(str);
+		if (from < 0)
+			return -1;
+
+		to = atoi(++s);
+		if (to <= 0)
+			return -1;
+
+		for (entry.target = from; entry.target <= to; entry.target++)
+			add_dep(deps, entry);
+	} else {
+		entry.target = atoi(str);
+		if (entry.target < 0)
+			return -1;
+
+		add_dep(deps, entry);
+	}
+
+	return 0;
+}
+
+static int
+parse_dependency(unsigned int nr_steps, struct w_step *w, char *str)
+{
+	struct dep_entry entry = { .working_set = -1 };
+	bool submit_fence = false;
+	char *s;
+
+	switch (str[0]) {
+	case '-':
+		if (str[1] < '0' || str[1] > '9')
+			return -1;
+
+		entry.target = atoi(str);
+		if (entry.target > 0 || ((int)nr_steps + entry.target) < 0)
+			return -1;
+
+		add_dep(&w->data_deps, entry);
+
+		break;
+	case 's':
+		submit_fence = true;
+		/* Fall-through. */
+	case 'f':
+		/* Multiple fences not yet supported. */
+		igt_assert_eq(w->fence_deps.nr, 0);
+
+		entry.target = atoi(++str);
+		if (entry.target > 0 || ((int)nr_steps + entry.target) < 0)
+			return -1;
+
+		add_dep(&w->fence_deps, entry);
+
+		w->fence_deps.submit_fence = submit_fence;
+		break;
+	case 'w':
+		entry.write = true;
+		/* Fall-through. */
+	case 'r':
+		/*
+		 * [rw]N-<str>
+		 * r1-<str> or w2-<str>, where N is working set id.
+		 */
+		s = index(++str, '-');
+		if (!s)
+			return -1;
+
+		entry.working_set = atoi(str);
+
+		if (parse_working_set_deps(w->wrk, &w->data_deps, entry, ++s))
+			return -1;
+
+		break;
+	default:
+		return -1;
+	};
+
+	return 0;
+}
+
 static int
 parse_dependencies(unsigned int nr_steps, struct w_step *w, char *_desc)
 {
 	char *desc = strdup(_desc);
 	char *token, *tctx = NULL, *tstart = desc;
+	int ret = 0;
+
+	if (!strcmp(_desc, "0"))
+		goto out;
 
 	igt_assert(desc);
 	igt_assert(!w->data_deps.nr && w->data_deps.nr == w->fence_deps.nr);
@@ -293,47 +421,17 @@ parse_dependencies(unsigned int nr_steps, struct w_step *w, char *_desc)
 		   w->data_deps.list == w->fence_deps.list);
 
 	while ((token = strtok_r(tstart, "/", &tctx)) != NULL) {
-		bool submit_fence = false;
-		char *str = token;
-		struct deps *deps;
-		int dep;
-
 		tstart = NULL;
 
-		if (str[0] == '-' || (str[0] >= '0' && str[0] <= '9')) {
-			deps = &w->data_deps;
-		} else {
-			if (str[0] == 's')
-				submit_fence = true;
-			else if (str[0] != 'f')
-				return -1;
-
-			deps = &w->fence_deps;
-			str++;
-		}
-
-		dep = atoi(str);
-		if (dep > 0 || ((int)nr_steps + dep) < 0) {
-			if (deps->list)
-				free(deps->list);
-			return -1;
-		}
-
-		if (dep < 0) {
-			deps->nr++;
-			/* Multiple fences not yet supported. */
-			igt_assert(deps->nr == 1 || deps != &w->fence_deps);
-			deps->list = realloc(deps->list,
-					     sizeof(*deps->list) * deps->nr);
-			igt_assert(deps->list);
-			deps->list[deps->nr - 1] = dep;
-			deps->submit_fence = submit_fence;
-		}
+		ret = parse_dependency(nr_steps, w, token);
+		if (ret)
+			break;
 	}
 
+out:
 	free(desc);
 
-	return 0;
+	return ret;
 }
 
 static void __attribute__((format(printf, 1, 2)))
@@ -624,6 +722,88 @@ static int parse_engine_map(struct w_step *step, const char *_str)
 	return 0;
 }
 
+static unsigned long parse_size(char *str)
+{
+	const unsigned int len = strlen(str);
+	unsigned int mult = 1;
+
+	if (len == 0)
+		return 0;
+
+	switch (str[len - 1]) {
+	case 'g':
+	case 'G':
+		mult *= 1024;
+		/* Fall-throuogh. */
+	case 'm':
+	case 'M':
+		mult *= 1024;
+		/* Fall-throuogh. */
+	case 'k':
+	case 'K':
+		mult *= 1024;
+
+		str[len - 1] = 0;
+	}
+
+	return atol(str) * mult;
+}
+
+static int add_buffers(struct working_set *set, char *str)
+{
+	/*
+	 * 4096
+	 * 4k
+	 * 4m
+	 * 4g
+	 * 10n4k - 10 4k batches
+	 */
+	unsigned long *sizes, size;
+	unsigned int add, i;
+	char *n;
+
+	n = index(str, 'n');
+	if (n) {
+		*n = 0;
+		add = atoi(str);
+		if (!add)
+			return -1;
+		str = ++n;
+	} else {
+		add = 1;
+	}
+
+	size = parse_size(str);
+	if (!size)
+		return -1;
+
+	sizes = realloc(set->sizes, (set->nr + add) * sizeof(*sizes));
+	if (!sizes)
+		return -1;
+
+	for (i = 0; i < add; i++)
+		sizes[set->nr + i] = size;
+
+	set->nr += add;
+	set->sizes = sizes;
+
+	return 0;
+}
+
+static int parse_working_set(struct working_set *set, char *str)
+{
+	char *token, *tctx = NULL, *tstart = str;
+
+	while ((token = strtok_r(tstart, "/", &tctx))) {
+		tstart = NULL;
+
+		if (add_buffers(set, token))
+			return -1;
+	}
+
+	return 0;
+}
+
 static uint64_t engine_list_mask(const char *_str)
 {
 	uint64_t mask = 0;
@@ -644,6 +824,8 @@ static uint64_t engine_list_mask(const char *_str)
 	return mask;
 }
 
+static void allocate_working_set(struct working_set *set);
+
 #define int_field(_STEP_, _FIELD_, _COND_, _ERR_) \
 	if ((field = strtok_r(fstart, ".", &fctx))) { \
 		tmp = atoi(field); \
@@ -661,7 +843,7 @@ parse_workload(struct w_arg *arg, unsigned int flags, struct workload *app_w)
 	char *desc = strdup(arg->desc);
 	char *_token, *token, *tctx = NULL, *tstart = desc;
 	char *field, *fctx = NULL, *fstart;
-	struct w_step step, *steps = NULL;
+	struct w_step step, *w, *steps = NULL;
 	unsigned int valid;
 	int i, j, tmp;
 
@@ -851,6 +1033,28 @@ parse_workload(struct w_arg *arg, unsigned int flags, struct workload *app_w)
 
 				step.type = BOND;
 				goto add_step;
+			} else if (!strcmp(field, "w") || !strcmp(field, "W")) {
+				unsigned int nr = 0;
+
+				step.working_set.shared = field[0] == 'W';
+
+				while ((field = strtok_r(fstart, ".", &fctx))) {
+					tmp = atoi(field);
+					if (nr == 0) {
+						step.working_set.id = tmp;
+					} else {
+						tmp = parse_working_set(&step.working_set,
+									field);
+						check_arg(tmp < 0,
+							  "Invalid working set at step %u!\n",
+							  nr_steps);
+					}
+
+					nr++;
+				}
+
+				step.type = WORKINGSET;
+				goto add_step;
 			}
 
 			if (!field) {
@@ -975,6 +1179,8 @@ add_step:
 	wrk->steps = steps;
 	wrk->prio = arg->prio;
 	wrk->sseu = arg->sseu;
+	wrk->max_working_set_id = -1;
+	wrk->working_sets = NULL;
 
 	free(desc);
 
@@ -984,7 +1190,7 @@ add_step:
 	 */
 	for (i = 0; i < nr_steps; i++) {
 		for (j = 0; j < steps[i].fence_deps.nr; j++) {
-			tmp = steps[i].idx + steps[i].fence_deps.list[j];
+			tmp = steps[i].idx + steps[i].fence_deps.list[j].target;
 			check_arg(tmp < 0 || tmp >= i ||
 				  (steps[tmp].type != BATCH &&
 				   steps[tmp].type != SW_FENCE),
@@ -1003,6 +1209,51 @@ add_step:
 		}
 	}
 
+	/*
+	 * Check no duplicate working set ids.
+	 */
+	for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
+		struct w_step *w2;
+
+		if (w->type != WORKINGSET)
+			continue;
+
+		for (j = 0, w2 = wrk->steps; j < wrk->nr_steps; w2++, j++) {
+			if (j == i)
+				continue;
+			if (w2->type != WORKINGSET)
+				continue;
+
+			check_arg(w->working_set.id == w2->working_set.id,
+				  "Duplicate working set id at %u!\n", j);
+		}
+	}
+
+	/*
+	 * Allocate shared working sets.
+	 */
+	for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
+		if (w->type == WORKINGSET && w->working_set.shared)
+			allocate_working_set(&w->working_set);
+	}
+
+	wrk->max_working_set_id = -1;
+	for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
+		if (w->type == WORKINGSET &&
+		    w->working_set.shared &&
+		    w->working_set.id > wrk->max_working_set_id)
+			wrk->max_working_set_id = w->working_set.id;
+	}
+
+	wrk->working_sets = calloc(wrk->max_working_set_id + 1,
+				   sizeof(*wrk->working_sets));
+	igt_assert(wrk->working_sets);
+
+	for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
+		if (w->type == WORKINGSET && w->working_set.shared)
+			wrk->working_sets[w->working_set.id] = &w->working_set;
+	}
+
 	return wrk;
 }
 
@@ -1024,6 +1275,18 @@ clone_workload(struct workload *_wrk)
 
 	memcpy(wrk->steps, _wrk->steps, sizeof(struct w_step) * wrk->nr_steps);
 
+	wrk->max_working_set_id = _wrk->max_working_set_id;
+	if (wrk->max_working_set_id >= 0) {
+		wrk->working_sets = calloc(wrk->max_working_set_id + 1,
+					sizeof(*wrk->working_sets));
+		igt_assert(wrk->working_sets);
+
+		memcpy(wrk->working_sets,
+		       _wrk->working_sets,
+		       (wrk->max_working_set_id + 1) *
+		       sizeof(*wrk->working_sets));
+	}
+
 	/* Check if we need a sw sync timeline. */
 	for (i = 0; i < wrk->nr_steps; i++) {
 		if (wrk->steps[i].type == SW_FENCE) {
@@ -1226,17 +1489,36 @@ alloc_step_batch(struct workload *wrk, struct w_step *w, unsigned int flags)
 	igt_assert(j < nr_obj);
 
 	for (i = 0; i < w->data_deps.nr; i++) {
-		igt_assert(w->data_deps.list[i] <= 0);
-		if (w->data_deps.list[i]) {
-			int dep_idx = w->idx + w->data_deps.list[i];
+		struct dep_entry *entry = &w->data_deps.list[i];
+		uint32_t dep_handle;
+
+		if (entry->working_set == -1) {
+			int dep_idx = w->idx + entry->target;
 
+			igt_assert(entry->target <= 0);
 			igt_assert(dep_idx >= 0 && dep_idx < w->idx);
 			igt_assert(wrk->steps[dep_idx].type == BATCH);
 
-			w->obj[j].handle = wrk->steps[dep_idx].obj[0].handle;
-			j++;
-			igt_assert(j < nr_obj);
+			dep_handle = wrk->steps[dep_idx].obj[0].handle;
+		} else {
+			struct working_set *set;
+
+			igt_assert(entry->working_set <=
+				   wrk->max_working_set_id);
+
+			set = wrk->working_sets[entry->working_set];
+
+			igt_assert(set->nr);
+			igt_assert(entry->target < set->nr);
+			igt_assert(set->sizes[entry->target]);
+
+			dep_handle = set->handles[entry->target];
 		}
+
+		w->obj[j].flags = entry->write ? EXEC_OBJECT_WRITE : 0;
+		w->obj[j].handle = dep_handle;
+		j++;
+		igt_assert(j < nr_obj);
 	}
 
 	if (w->unbound_duration)
@@ -1395,11 +1677,23 @@ static size_t sizeof_engines_bond(int count)
 			engines[count]);
 }
 
+static void allocate_working_set(struct working_set *set)
+{
+	unsigned int i;
+
+	set->handles = calloc(set->nr, sizeof(*set->handles));
+	igt_assert(set->handles);
+
+	for (i = 0; i < set->nr; i++)
+		set->handles[i] = gem_create(fd, set->sizes[i]);
+}
+
 #define alloca0(sz) ({ size_t sz__ = (sz); memset(alloca(sz__), 0, sz__); })
 
 static int
 prepare_workload(unsigned int id, struct workload *wrk, unsigned int flags)
 {
+	struct working_set **sets;
 	uint32_t share_vm = 0;
 	int max_ctx = -1;
 	struct w_step *w;
@@ -1634,6 +1928,51 @@ prepare_workload(unsigned int id, struct workload *wrk, unsigned int flags)
 		}
 	}
 
+	/*
+	 * Allocate working sets.
+	 */
+	for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
+		if (w->type == WORKINGSET && !w->working_set.shared)
+			allocate_working_set(&w->working_set);
+	}
+
+	/*
+	 * Map of working set ids.
+	 */
+	wrk->max_working_set_id = -1;
+	for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
+		if (w->type == WORKINGSET &&
+		    w->working_set.id > wrk->max_working_set_id)
+			wrk->max_working_set_id = w->working_set.id;
+	}
+
+	sets = wrk->working_sets;
+	wrk->working_sets = calloc(wrk->max_working_set_id + 1,
+				   sizeof(*wrk->working_sets));
+	igt_assert(wrk->working_sets);
+
+	for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
+		struct working_set *set;
+
+		if (w->type != WORKINGSET)
+			continue;
+
+		if (!w->working_set.shared) {
+			set = &w->working_set;
+		} else {
+			igt_assert(sets);
+
+			set = sets[w->working_set.id];
+			igt_assert(set->shared);
+			igt_assert(set->sizes);
+		}
+
+		wrk->working_sets[w->working_set.id] = set;
+	}
+
+	if (sets)
+		free(sets);
+
 	/*
 	 * Allocate batch buffers.
 	 */
@@ -1704,7 +2043,7 @@ do_eb(struct workload *wrk, struct w_step *w, enum intel_engine_id engine,
 		      2 * sizeof(uint32_t));
 
 	for (i = 0; i < w->fence_deps.nr; i++) {
-		int tgt = w->idx + w->fence_deps.list[i];
+		int tgt = w->idx + w->fence_deps.list[i].target;
 
 		/* TODO: fence merging needed to support multiple inputs */
 		igt_assert(i == 0);
@@ -1735,14 +2074,18 @@ static void sync_deps(struct workload *wrk, struct w_step *w)
 	unsigned int i;
 
 	for (i = 0; i < w->data_deps.nr; i++) {
+		struct dep_entry *entry = &w->data_deps.list[i];
 		int dep_idx;
 
-		igt_assert(w->data_deps.list[i] <= 0);
+		if (entry->working_set == -1)
+			continue;
+
+		igt_assert(entry->target <= 0);
 
-		if (!w->data_deps.list[i])
+		if (!entry->target)
 			continue;
 
-		dep_idx = w->idx + w->data_deps.list[i];
+		dep_idx = w->idx + entry->target;
 
 		igt_assert(dep_idx >= 0 && dep_idx < w->idx);
 		igt_assert(wrk->steps[dep_idx].type == BATCH);
@@ -1842,11 +2185,6 @@ static void *run_workload(void *data)
 					MI_BATCH_BUFFER_END;
 				__sync_synchronize();
 				continue;
-			} else if (w->type == PREEMPTION ||
-				   w->type == ENGINE_MAP ||
-				   w->type == LOAD_BALANCE ||
-				   w->type == BOND) {
-				continue;
 			} else if (w->type == SSEU) {
 				if (w->sseu != wrk->ctx_list[w->context * 2].sseu) {
 					wrk->ctx_list[w->context * 2].sseu =
@@ -1854,6 +2192,13 @@ static void *run_workload(void *data)
 							     w->sseu);
 				}
 				continue;
+			} else if (w->type == PREEMPTION ||
+				   w->type == ENGINE_MAP ||
+				   w->type == LOAD_BALANCE ||
+				   w->type == BOND ||
+				   w->type == WORKINGSET) {
+				   /* No action for these at execution time. */
+				continue;
 			}
 
 			if (do_sleep || w->type == PERIOD) {
benchmarks/wsim/README +59 -0

74 lines not on this page · raw message

benchmarks/wsim/cloud-gaming-60fps.wsim new file +11 -0
new file mode 100644
index 000000000000..9e48bbc2f617
@@ -0,0 +1,11 @@
+w.1.10n8m
+w.2.3n16m
+1.RCS.500-1500.r1-0-4/w2-0.0
+1.RCS.500-1500.r1-5-9/w2-1.0
+1.RCS.500-1500.r2-0-1/w2-2.0
+M.2.VCS
+B.2
+3.RCS.500-1500.r2-2.0
+2.DEFAULT.2000-4000.-1.0
+4.VCS1.250-750.-1.1
+p.16667
benchmarks/wsim/composited-ui.wsim new file +7 -0
new file mode 100644
index 000000000000..4164f8bf7393
@@ -0,0 +1,7 @@
+w.1.10n8m/3n16m
+W.2.16m
+1.RCS.200-600.r1-0-4/w1-10.0
+1.RCS.200-600.r1-5-9/w1-11.0
+1.RCS.400-800.r1-10-11/w1-12.0
+3.BCS.200-800.r1-12/w2-0.1
+p.16667
Signature · 6 lines
2.20.1

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

Re: [Intel-gfx] [PATCH i-g-t 02/10] gem_wsim: Buffer objects working sets and complex dependencies

Quoting Tvrtko Ursulin (2020-06-17 17:01:12)

Quoted 94 lines

From: Tvrtko Ursulin <[email protected]>

Add support for defining buffer object working sets and targetting them as data dependencies. For more information please see the README file.

---
 benchmarks/gem_wsim.c                   | 453 +++++++++++++++++++++---
 benchmarks/wsim/README                  |  59 +++
 benchmarks/wsim/cloud-gaming-60fps.wsim |  11 +
 benchmarks/wsim/composited-ui.wsim      |   7 +
 4 files changed, 476 insertions(+), 54 deletions(-)
 create mode 100644 benchmarks/wsim/cloud-gaming-60fps.wsim
 create mode 100644 benchmarks/wsim/composited-ui.wsim
benchmarks/gem_wsim.c +35 -2
index 02fe8f5a5e69..9e5bfe6a36d4 100644
@@ -88,14 +88,21 @@ enum w_type
        LOAD_BALANCE,
        BOND,
        TERMINATE,
-       SSEU
+       SSEU,
+       WORKINGSET,
+};
+
+struct dep_entry {
+       int target;
+       bool write;
+       int working_set; /* -1 = step dependecy, >= 0 working set id */
 };
 
 struct deps
 {
        int nr;
        bool submit_fence;
-       int *list;
+       struct dep_entry *list;
 };
 
 struct w_arg {
@@ -110,6 +117,14 @@ struct bond {
        enum intel_engine_id master;
 };
 
+struct working_set {
+       int id;
+       bool shared;
+       unsigned int nr;
+       uint32_t *handles;
+       unsigned long *sizes;
+};
+
 struct workload;
 
 struct w_step
@@ -143,6 +158,7 @@ struct w_step
                        enum intel_engine_id bond_master;
                };
                int sseu;
+               struct working_set working_set;
        };
 
        /* Implementation details */
@@ -193,6 +209,9 @@ struct workload
        unsigned int nr_ctxs;
        struct ctx *ctx_list;
 
+       struct working_set **working_sets; /* array indexed by set id */
+       int max_working_set_id;
+
        int sync_timeline;
        uint32_t sync_seqno;
 
@@ -281,11 +300,120 @@ print_engine_calibrations(void)
        printf("\n");
 }
 
+static void add_dep(struct deps *deps, struct dep_entry entry)
+{
+       deps->list = realloc(deps->list, sizeof(*deps->list) * (deps->nr + 1));
+       igt_assert(deps->list);
+
+       deps->list[deps->nr++] = entry;
+}
+
+
+static int
+parse_dependency(unsigned int nr_steps, struct w_step *w, char *str)
+{
+       struct dep_entry entry = { .working_set = -1 };
+       bool submit_fence = false;

benchmarks/gem_wsim.c print_engine_calibrations(void) · line 326

Quoted 10 lines
benchmarks/gem_wsim.c +10 -0
+       char *s;
+
+       switch (str[0]) {
+       case '-':
+               if (str[1] < '0' || str[1] > '9')
+                       return -1;
+
+               entry.target = atoi(str);
+               if (entry.target > 0 || ((int)nr_steps + entry.target) < 0)
+                       return -1;

add_dep for N steps ago, using a handle.

Quoted 8 lines
benchmarks/gem_wsim.c +8 -0
+
+               add_dep(&w->data_deps, entry);
+
+               break;
+       case 's':
+               submit_fence = true;
+               /* Fall-through. */
+       case 'f':

benchmarks/gem_wsim.c print_engine_calibrations(void) · line 381

Quoted 10 lines
benchmarks/gem_wsim.c +10 -0
+               /* Multiple fences not yet supported. */
+               igt_assert_eq(w->fence_deps.nr, 0);
+
+               entry.target = atoi(++str);
+               if (entry.target > 0 || ((int)nr_steps + entry.target) < 0)
+                       return -1;
+
+               add_dep(&w->fence_deps, entry);
+
+               w->fence_deps.submit_fence = submit_fence;

add_dep for N steps ago, using the out-fence from that step [A post processing steps adds emit_fence to the earlier steps.]

benchmarks/gem_wsim.c print_engine_calibrations(void) · line 384

Quoted 3 lines
benchmarks/gem_wsim.c +3 -0
+               break;
+       case 'w':
+               entry.write = true;

Got confused for a moment as I was expecting the submit_fence fallthrough pattern.

benchmarks/gem_wsim.c print_engine_calibrations(void) · line 395

Quoted 11 lines
benchmarks/gem_wsim.c +11 -0
+               /* Fall-through. */
+       case 'r':
+               /*
+                * [rw]N-<str>
+                * r1-<str> or w2-<str>, where N is working set id.
+                */
+               s = index(++str, '-');
+               if (!s)
+                       return -1;
+
+               entry.working_set = atoi(str);

if (entry.working_set < 0)

	return -1;

benchmarks/gem_wsim.c print_engine_calibrations(void) · line 398

Quoted 3 lines
benchmarks/gem_wsim.c +3 -0
+
+               if (parse_working_set_deps(w->wrk, &w->data_deps, entry, ++s))
+                       return -1;

The new one...

Quoted 14 lines
benchmarks/gem_wsim.c +14 -0
+static int
+parse_working_set_deps(struct workload *wrk,
+                      struct deps *deps,
+                      struct dep_entry _entry,
+                      char *str)
+{
+       /*
+        * 1 - target handle index in the specified working set.
+        * 2-4 - range
+        */
+       struct dep_entry entry = _entry;
+       char *s;
+
+       s = index(str, '-');

benchmarks/gem_wsim.c print_engine_calibrations(void) · line 334

Quoted 10 lines
benchmarks/gem_wsim.c +10 -0
+       if (s) {
+               int from, to;
+
+               from = atoi(str);
+               if (from < 0)
+                       return -1;
+
+               to = atoi(++s);
+               if (to <= 0)
+                       return -1;

if to < from, we add nothing. Is that worth the error?

Quoted 23 lines
benchmarks/gem_wsim.c +21 -0
+
+               for (entry.target = from; entry.target <= to; entry.target++)
+                       add_dep(deps, entry);
+       } else {
+               entry.target = atoi(str);
+               if (entry.target < 0)
+                       return -1;
+
+               add_dep(deps, entry);
 
 
+       }
+
+       return 0;
+}
+
+               break;
+       default:
+               return -1;
+       };
+
+       return 0;
+}

benchmarks/gem_wsim.c print_engine_calibrations(void) · line 416

Quoted 10 lines
benchmarks/gem_wsim.c +5 -0
+
 static int
 parse_dependencies(unsigned int nr_steps, struct w_step *w, char *_desc)
 {
        char *desc = strdup(_desc);
        char *token, *tctx = NULL, *tstart = desc;
+       int ret = 0;
+
+       if (!strcmp(_desc, "0"))
+               goto out;

Hang on, what this special case?

benchmarks/gem_wsim.c static int parse_engine_map(struct w_step *step, const char *_str) · line 726

Quoted 10 lines
benchmarks/gem_wsim.c +0 -0
        igt_assert(desc);
        igt_assert(!w->data_deps.nr && w->data_deps.nr == w->fence_deps.nr);

static void __attribute__((format(printf, 1, 2)))

benchmarks/gem_wsim.c +2 -0
@@ -624,6 +722,88 @@ static int parse_engine_map(struct w_step *step, const char *_str)
        return 0;
 }
 
+static unsigned long parse_size(char *str)
+{

/* "1234567890[gGmMkK]" */

benchmarks/gem_wsim.c static int parse_engine_map(struct w_step *step, const char *_str) · line 733

Quoted 7 lines
benchmarks/gem_wsim.c +7 -0
+       const unsigned int len = strlen(str);
+       unsigned int mult = 1;
+
+       if (len == 0)
+               return 0;
+
+       switch (str[len - 1]) {

T? P? E? Let's plan ahead! :)

Quoted 6 lines
benchmarks/gem_wsim.c +6 -0
+       case 'g':
+       case 'G':
+               mult *= 1024;
+               /* Fall-throuogh. */
+       case 'm':
+       case 'M':

benchmarks/gem_wsim.c static int parse_engine_map(struct w_step *step, const char *_str) · line 749

Quoted 10 lines
benchmarks/gem_wsim.c +10 -0
+               mult *= 1024;
+               /* Fall-throuogh. */
+       case 'k':
+       case 'K':
+               mult *= 1024;
+
+               str[len - 1] = 0;
+       }
+
+       return atol(str) * mult;

Negatives?

Quoted 11 lines
benchmarks/gem_wsim.c +11 -0
+}
+
+static int add_buffers(struct working_set *set, char *str)
+{
+       /*
+        * 4096
+        * 4k
+        * 4m
+        * 4g
+        * 10n4k - 10 4k batches
+        */

benchmarks/gem_wsim.c static int parse_engine_map(struct w_step *step, const char *_str) · line 770

Quoted 10 lines
benchmarks/gem_wsim.c +10 -0
+       unsigned long *sizes, size;
+       unsigned int add, i;
+       char *n;
+
+       n = index(str, 'n');
+       if (n) {
+               *n = 0;
+               add = atoi(str);
+               if (!add)
+                       return -1;

if (add <= 0) [int add goes without saying then]

Quoted 60 lines
benchmarks/gem_wsim.c +55 -0
+               str = ++n;
+       } else {
+               add = 1;
+       }
+
+       size = parse_size(str);
+       if (!size)
+               return -1;
+
+       sizes = realloc(set->sizes, (set->nr + add) * sizeof(*sizes));
+       if (!sizes)
+               return -1;
+
+       for (i = 0; i < add; i++)
+               sizes[set->nr + i] = size;
+
+       set->nr += add;
+       set->sizes = sizes;
+
+       return 0;
+}
 
@@ -1003,6 +1209,51 @@ add_step:
                }
        }
 
+       /*
+        * Check no duplicate working set ids.
+        */
+       for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
+               struct w_step *w2;
+
+               if (w->type != WORKINGSET)
+                       continue;
+
+               for (j = 0, w2 = wrk->steps; j < wrk->nr_steps; w2++, j++) {
+                       if (j == i)
+                               continue;
+                       if (w2->type != WORKINGSET)
+                               continue;
+
+                       check_arg(w->working_set.id == w2->working_set.id,
+                                 "Duplicate working set id at %u!\n", j);
+               }
+       }
+
+       /*
+        * Allocate shared working sets.
+        */
+       for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
+               if (w->type == WORKINGSET && w->working_set.shared)
+                       allocate_working_set(&w->working_set);
+       }
+
+       wrk->max_working_set_id = -1;
+       for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
+               if (w->type == WORKINGSET &&
+                   w->working_set.shared &&
+                   w->working_set.id > wrk->max_working_set_id)
+                       wrk->max_working_set_id = w->working_set.id;

benchmarks/gem_wsim.c add_step: · line 1255

Quoted 10 lines
benchmarks/gem_wsim.c +10 -0
+       }
+
+       wrk->working_sets = calloc(wrk->max_working_set_id + 1,
+                                  sizeof(*wrk->working_sets));
+       igt_assert(wrk->working_sets);
+
+       for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
+               if (w->type == WORKINGSET && w->working_set.shared)
+                       wrk->working_sets[w->working_set.id] = &w->working_set;
+       }

Ok, sharing works by reusing the same set of handles within the process.

Is there room in the parser namespace for dmabuf sharing? -Chris
_______________________________________________
Intel-gfx mailing list [email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#10 lore raw

Re: [Intel-gfx] [PATCH i-g-t 02/10] gem_wsim: Buffer objects working sets and complex dependencies

Quoted 148 lines · Chris Wilson

Quoting Tvrtko Ursulin (2020-06-17 17:01:12)

Quoted 94 lines

From: Tvrtko Ursulin <[email protected]>

Add support for defining buffer object working sets and targetting them as data dependencies. For more information please see the README file.

---
  benchmarks/gem_wsim.c                   | 453 +++++++++++++++++++++---
  benchmarks/wsim/README                  |  59 +++
  benchmarks/wsim/cloud-gaming-60fps.wsim |  11 +
  benchmarks/wsim/composited-ui.wsim      |   7 +
  4 files changed, 476 insertions(+), 54 deletions(-)

create mode 100644 benchmarks/wsim/cloud-gaming-60fps.wsim create mode 100644 benchmarks/wsim/composited-ui.wsim

benchmarks/gem_wsim.c +35 -2
index 02fe8f5a5e69..9e5bfe6a36d4 100644
@@ -88,14 +88,21 @@ enum w_type
         LOAD_BALANCE,
         BOND,
         TERMINATE,
-       SSEU
+       SSEU,
+       WORKINGSET,
+};
+
+struct dep_entry {
+       int target;
+       bool write;
+       int working_set; /* -1 = step dependecy, >= 0 working set id */
  };
 
  struct deps
  {
         int nr;
         bool submit_fence;
-       int *list;
+       struct dep_entry *list;
  };
 
  struct w_arg {
@@ -110,6 +117,14 @@ struct bond {
         enum intel_engine_id master;
  };
 
+struct working_set {
+       int id;
+       bool shared;
+       unsigned int nr;
+       uint32_t *handles;
+       unsigned long *sizes;
+};
+
  struct workload;
 
  struct w_step
@@ -143,6 +158,7 @@ struct w_step
                         enum intel_engine_id bond_master;
                 };
                 int sseu;
+               struct working_set working_set;
         };
 
         /* Implementation details */
@@ -193,6 +209,9 @@ struct workload
         unsigned int nr_ctxs;
         struct ctx *ctx_list;
 
+       struct working_set **working_sets; /* array indexed by set id */
+       int max_working_set_id;
+
         int sync_timeline;
         uint32_t sync_seqno;
 
@@ -281,11 +300,120 @@ print_engine_calibrations(void)
         printf("\n");
  }
 
+static void add_dep(struct deps *deps, struct dep_entry entry)
+{
+       deps->list = realloc(deps->list, sizeof(*deps->list) * (deps->nr + 1));
+       igt_assert(deps->list);
+
+       deps->list[deps->nr++] = entry;
+}
+
+
+static int
+parse_dependency(unsigned int nr_steps, struct w_step *w, char *str)
+{
+       struct dep_entry entry = { .working_set = -1 };
+       bool submit_fence = false;

benchmarks/gem_wsim.c print_engine_calibrations(void) · line 326

Quoted 10 lines
benchmarks/gem_wsim.c +10 -0
+       char *s;
+
+       switch (str[0]) {
+       case '-':
+               if (str[1] < '0' || str[1] > '9')
+                       return -1;
+
+               entry.target = atoi(str);
+               if (entry.target > 0 || ((int)nr_steps + entry.target) < 0)
+                       return -1;

add_dep for N steps ago, using a handle.

Quoted 8 lines
benchmarks/gem_wsim.c +8 -0
+
+               add_dep(&w->data_deps, entry);
+
+               break;
+       case 's':
+               submit_fence = true;
+               /* Fall-through. */
+       case 'f':

benchmarks/gem_wsim.c print_engine_calibrations(void) · line 381

Quoted 10 lines
benchmarks/gem_wsim.c +10 -0
+               /* Multiple fences not yet supported. */
+               igt_assert_eq(w->fence_deps.nr, 0);
+
+               entry.target = atoi(++str);
+               if (entry.target > 0 || ((int)nr_steps + entry.target) < 0)
+                       return -1;
+
+               add_dep(&w->fence_deps, entry);
+
+               w->fence_deps.submit_fence = submit_fence;

add_dep for N steps ago, using the out-fence from that step [A post processing steps adds emit_fence to the earlier steps.]

benchmarks/gem_wsim.c print_engine_calibrations(void) · line 384

Quoted 3 lines
benchmarks/gem_wsim.c +3 -0
+               break;
+       case 'w':
+               entry.write = true;

Got confused for a moment as I was expecting the submit_fence fallthrough pattern.

benchmarks/gem_wsim.c print_engine_calibrations(void) · line 395

Quoted 11 lines
benchmarks/gem_wsim.c +11 -0
+               /* Fall-through. */
+       case 'r':
+               /*
+                * [rw]N-<str>
+                * r1-<str> or w2-<str>, where N is working set id.
+                */
+               s = index(++str, '-');
+               if (!s)
+                       return -1;
+
+               entry.working_set = atoi(str);
Quoted 2 lines

if (entry.working_set < 0)

	return -1;

Yep.

Quoted 25 lines

benchmarks/gem_wsim.c print_engine_calibrations(void) · line 398

Quoted 4 lines
benchmarks/gem_wsim.c +3 -0
+
+               if (parse_working_set_deps(w->wrk, &w->data_deps, entry, ++s))
+                       return -1;

The new one...

Quoted 18 lines
benchmarks/gem_wsim.c +18 -0
+static int
+parse_working_set_deps(struct workload *wrk,
+                      struct deps *deps,
+                      struct dep_entry _entry,
+                      char *str)
+{
+       /*
+        * 1 - target handle index in the specified working set.
+        * 2-4 - range
+        */
+       struct dep_entry entry = _entry;
+       char *s;
+
+       s = index(str, '-');
+       if (s) {
+               int from, to;
+
+               from = atoi(str);
Quoted 8 lines

benchmarks/gem_wsim.c print_engine_calibrations(void) · line 334

Quoted 6 lines
benchmarks/gem_wsim.c +6 -0
+               if (from < 0)
+                       return -1;
+
+               to = atoi(++s);
+               if (to <= 0)
+                       return -1;

if to < from, we add nothing. Is that worth the error?

Yep.

Quoted 28 lines
Quoted 28 lines
benchmarks/gem_wsim.c +22 -0
+
+               for (entry.target = from; entry.target <= to; entry.target++)
+                       add_dep(deps, entry);
+       } else {
+               entry.target = atoi(str);
+               if (entry.target < 0)
+                       return -1;
+
+               add_dep(deps, entry);
 
 
+       }
+
+       return 0;
+}
+
+               break;
+       default:
+               return -1;
+       };
+
+       return 0;
+}
+
  static int
  parse_dependencies(unsigned int nr_steps, struct w_step *w, char *_desc)
  {
Quoted 8 lines

benchmarks/gem_wsim.c print_engine_calibrations(void) · line 416

Quoted 6 lines
benchmarks/gem_wsim.c +4 -0
         char *desc = strdup(_desc);
         char *token, *tctx = NULL, *tstart = desc;
+       int ret = 0;
+
+       if (!strcmp(_desc, "0"))
+               goto out;

Hang on, what this special case?

For no dependencies.

If I move the check to parse_dependency then dependency of "0/0/0/0" would be silently accepted. It wouldn't be a big deal, who cares, but I thought it is better to be more strict.

Quoted 12 lines

benchmarks/gem_wsim.c static int parse_engine_map(struct w_step *step, const char *_str) · line 726

Quoted 11 lines
benchmarks/gem_wsim.c +0 -0
         igt_assert(desc);
         igt_assert(!w->data_deps.nr && w->data_deps.nr == w->fence_deps.nr);

static void __attribute__((format(printf, 1, 2)))

benchmarks/gem_wsim.c +2 -0
@@ -624,6 +722,88 @@ static int parse_engine_map(struct w_step *step, const char *_str)
         return 0;
  }
 
+static unsigned long parse_size(char *str)
+{

/* "1234567890[gGmMkK]" */

Quoted 9 lines

benchmarks/gem_wsim.c static int parse_engine_map(struct w_step *step, const char *_str) · line 733

Quoted 7 lines
benchmarks/gem_wsim.c +7 -0
+       const unsigned int len = strlen(str);
+       unsigned int mult = 1;
+
+       if (len == 0)
+               return 0;
+
+       switch (str[len - 1]) {

T? P? E? Let's plan ahead! :)

Error on unrecognized non-digit? Ok.

Quoted 11 lines
Quoted 11 lines
benchmarks/gem_wsim.c +10 -0
+       case 'g':
+       case 'G':
+               mult *= 1024;
+               /* Fall-throuogh. */
+       case 'm':
+       case 'M':
+               mult *= 1024;
+               /* Fall-throuogh. */
+       case 'k':
+       case 'K':
Quoted 8 lines

benchmarks/gem_wsim.c static int parse_engine_map(struct w_step *step, const char *_str) · line 749

Quoted 6 lines
benchmarks/gem_wsim.c +6 -0
+               mult *= 1024;
+
+               str[len - 1] = 0;
+       }
+
+       return atol(str) * mult;

Negatives?

Ok.

Quoted 16 lines
Quoted 16 lines
benchmarks/gem_wsim.c +15 -0
+}
+
+static int add_buffers(struct working_set *set, char *str)
+{
+       /*
+        * 4096
+        * 4k
+        * 4m
+        * 4g
+        * 10n4k - 10 4k batches
+        */
+       unsigned long *sizes, size;
+       unsigned int add, i;
+       char *n;
+
Quoted 8 lines

benchmarks/gem_wsim.c static int parse_engine_map(struct w_step *step, const char *_str) · line 770

Quoted 6 lines
benchmarks/gem_wsim.c +6 -0
+       n = index(str, 'n');
+       if (n) {
+               *n = 0;
+               add = atoi(str);
+               if (!add)
+                       return -1;

if (add <= 0) [int add goes without saying then]

Yep.

Quoted 74 lines
Quoted 61 lines
benchmarks/gem_wsim.c +55 -0
+               str = ++n;
+       } else {
+               add = 1;
+       }
+
+       size = parse_size(str);
+       if (!size)
+               return -1;
+
+       sizes = realloc(set->sizes, (set->nr + add) * sizeof(*sizes));
+       if (!sizes)
+               return -1;
+
+       for (i = 0; i < add; i++)
+               sizes[set->nr + i] = size;
+
+       set->nr += add;
+       set->sizes = sizes;
+
+       return 0;
+}
 
@@ -1003,6 +1209,51 @@ add_step:
                 }
         }
 
+       /*
+        * Check no duplicate working set ids.
+        */
+       for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
+               struct w_step *w2;
+
+               if (w->type != WORKINGSET)
+                       continue;
+
+               for (j = 0, w2 = wrk->steps; j < wrk->nr_steps; w2++, j++) {
+                       if (j == i)
+                               continue;
+                       if (w2->type != WORKINGSET)
+                               continue;
+
+                       check_arg(w->working_set.id == w2->working_set.id,
+                                 "Duplicate working set id at %u!\n", j);
+               }
+       }
+
+       /*
+        * Allocate shared working sets.
+        */
+       for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
+               if (w->type == WORKINGSET && w->working_set.shared)
+                       allocate_working_set(&w->working_set);
+       }
+
+       wrk->max_working_set_id = -1;
+       for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
+               if (w->type == WORKINGSET &&
+                   w->working_set.shared &&
+                   w->working_set.id > wrk->max_working_set_id)
+                       wrk->max_working_set_id = w->working_set.id;

benchmarks/gem_wsim.c add_step: · line 1255

Quoted 10 lines
benchmarks/gem_wsim.c +10 -0
+       }
+
+       wrk->working_sets = calloc(wrk->max_working_set_id + 1,
+                                  sizeof(*wrk->working_sets));
+       igt_assert(wrk->working_sets);
+
+       for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
+               if (w->type == WORKINGSET && w->working_set.shared)
+                       wrk->working_sets[w->working_set.id] = &w->working_set;
+       }

Ok, sharing works by reusing the same set of handles within the process.

Quoted 1 line

Is there room in the parser namespace for dmabuf sharing?

Plenty of unused characters. :)

Regards,

Tvrtko

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

#11 lore raw

Re: [Intel-gfx] [PATCH i-g-t 02/10] gem_wsim: Buffer objects working sets and complex dependencies

Quoting Tvrtko Ursulin (2020-06-18 10:05:56)

Quoted 228 lines · Chris Wilson
Quoted 148 lines

Quoting Tvrtko Ursulin (2020-06-17 17:01:12)

Quoted 94 lines

From: Tvrtko Ursulin <[email protected]>

Add support for defining buffer object working sets and targetting them as data dependencies. For more information please see the README file.

---
  benchmarks/gem_wsim.c                   | 453 +++++++++++++++++++++---
  benchmarks/wsim/README                  |  59 +++
  benchmarks/wsim/cloud-gaming-60fps.wsim |  11 +
  benchmarks/wsim/composited-ui.wsim      |   7 +
  4 files changed, 476 insertions(+), 54 deletions(-)

create mode 100644 benchmarks/wsim/cloud-gaming-60fps.wsim create mode 100644 benchmarks/wsim/composited-ui.wsim

benchmarks/gem_wsim.c +35 -2
index 02fe8f5a5e69..9e5bfe6a36d4 100644
@@ -88,14 +88,21 @@ enum w_type
         LOAD_BALANCE,
         BOND,
         TERMINATE,
-       SSEU
+       SSEU,
+       WORKINGSET,
+};
+
+struct dep_entry {
+       int target;
+       bool write;
+       int working_set; /* -1 = step dependecy, >= 0 working set id */
  };
 
  struct deps
  {
         int nr;
         bool submit_fence;
-       int *list;
+       struct dep_entry *list;
  };
 
  struct w_arg {
@@ -110,6 +117,14 @@ struct bond {
         enum intel_engine_id master;
  };
 
+struct working_set {
+       int id;
+       bool shared;
+       unsigned int nr;
+       uint32_t *handles;
+       unsigned long *sizes;
+};
+
  struct workload;
 
  struct w_step
@@ -143,6 +158,7 @@ struct w_step
                         enum intel_engine_id bond_master;
                 };
                 int sseu;
+               struct working_set working_set;
         };
 
         /* Implementation details */
@@ -193,6 +209,9 @@ struct workload
         unsigned int nr_ctxs;
         struct ctx *ctx_list;
 
+       struct working_set **working_sets; /* array indexed by set id */
+       int max_working_set_id;
+
         int sync_timeline;
         uint32_t sync_seqno;
 
@@ -281,11 +300,120 @@ print_engine_calibrations(void)
         printf("\n");
  }
 
+static void add_dep(struct deps *deps, struct dep_entry entry)
+{
+       deps->list = realloc(deps->list, sizeof(*deps->list) * (deps->nr + 1));
+       igt_assert(deps->list);
+
+       deps->list[deps->nr++] = entry;
+}
+
+
+static int
+parse_dependency(unsigned int nr_steps, struct w_step *w, char *str)
+{
+       struct dep_entry entry = { .working_set = -1 };
+       bool submit_fence = false;

benchmarks/gem_wsim.c print_engine_calibrations(void) · line 326

Quoted 10 lines
benchmarks/gem_wsim.c +10 -0
+       char *s;
+
+       switch (str[0]) {
+       case '-':
+               if (str[1] < '0' || str[1] > '9')
+                       return -1;
+
+               entry.target = atoi(str);
+               if (entry.target > 0 || ((int)nr_steps + entry.target) < 0)
+                       return -1;

add_dep for N steps ago, using a handle.

Quoted 8 lines
benchmarks/gem_wsim.c +8 -0
+
+               add_dep(&w->data_deps, entry);
+
+               break;
+       case 's':
+               submit_fence = true;
+               /* Fall-through. */
+       case 'f':

benchmarks/gem_wsim.c print_engine_calibrations(void) · line 381

Quoted 10 lines
benchmarks/gem_wsim.c +10 -0
+               /* Multiple fences not yet supported. */
+               igt_assert_eq(w->fence_deps.nr, 0);
+
+               entry.target = atoi(++str);
+               if (entry.target > 0 || ((int)nr_steps + entry.target) < 0)
+                       return -1;
+
+               add_dep(&w->fence_deps, entry);
+
+               w->fence_deps.submit_fence = submit_fence;

add_dep for N steps ago, using the out-fence from that step [A post processing steps adds emit_fence to the earlier steps.]

benchmarks/gem_wsim.c print_engine_calibrations(void) · line 384

Quoted 3 lines
benchmarks/gem_wsim.c +3 -0
+               break;
+       case 'w':
+               entry.write = true;

Got confused for a moment as I was expecting the submit_fence fallthrough pattern.

benchmarks/gem_wsim.c print_engine_calibrations(void) · line 395

Quoted 11 lines
benchmarks/gem_wsim.c +11 -0
+               /* Fall-through. */
+       case 'r':
+               /*
+                * [rw]N-<str>
+                * r1-<str> or w2-<str>, where N is working set id.
+                */
+               s = index(++str, '-');
+               if (!s)
+                       return -1;
+
+               entry.working_set = atoi(str);
Quoted 2 lines

if (entry.working_set < 0)

      return -1;

Yep.

Quoted 25 lines

benchmarks/gem_wsim.c print_engine_calibrations(void) · line 398

Quoted 4 lines
benchmarks/gem_wsim.c +3 -0
+
+               if (parse_working_set_deps(w->wrk, &w->data_deps, entry, ++s))
+                       return -1;

The new one...

Quoted 18 lines
benchmarks/gem_wsim.c +18 -0
+static int
+parse_working_set_deps(struct workload *wrk,
+                      struct deps *deps,
+                      struct dep_entry _entry,
+                      char *str)
+{
+       /*
+        * 1 - target handle index in the specified working set.
+        * 2-4 - range
+        */
+       struct dep_entry entry = _entry;
+       char *s;
+
+       s = index(str, '-');
+       if (s) {
+               int from, to;
+
+               from = atoi(str);
Quoted 8 lines

benchmarks/gem_wsim.c print_engine_calibrations(void) · line 334

Quoted 6 lines
benchmarks/gem_wsim.c +6 -0
+               if (from < 0)
+                       return -1;
+
+               to = atoi(++s);
+               if (to <= 0)
+                       return -1;

if to < from, we add nothing. Is that worth the error?

Yep.

Quoted 28 lines
Quoted 28 lines
benchmarks/gem_wsim.c +22 -0
+
+               for (entry.target = from; entry.target <= to; entry.target++)
+                       add_dep(deps, entry);
+       } else {
+               entry.target = atoi(str);
+               if (entry.target < 0)
+                       return -1;
+
+               add_dep(deps, entry);
 
 
+       }
+
+       return 0;
+}
+
+               break;
+       default:
+               return -1;
+       };
+
+       return 0;
+}
+
  static int
  parse_dependencies(unsigned int nr_steps, struct w_step *w, char *_desc)
  {
Quoted 8 lines

benchmarks/gem_wsim.c print_engine_calibrations(void) · line 416

Quoted 6 lines
benchmarks/gem_wsim.c +4 -0
         char *desc = strdup(_desc);
         char *token, *tctx = NULL, *tstart = desc;
+       int ret = 0;
+
+       if (!strcmp(_desc, "0"))
+               goto out;

Hang on, what this special case?

Quoted 5 lines

For no dependencies.

If I move the check to parse_dependency then dependency of "0/0/0/0" would be silently accepted. It wouldn't be a big deal, who cares, but I thought it is better to be more strict.

It was just not clear at this point what is being matched, what the meaning of 0 is.

/* 0 refers to self, a degenerate dependency */
?
-Chris
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#12 lore raw

PATCH 3/10 gem_wsim: Show workload timing stats

In reply to Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

Show average/min/max workload iteration and dropped period stats when 'p' command is used.

---
 benchmarks/gem_wsim.c | 19 +++++++++++++++----
 1 file changed, 15 insertions(+), 4 deletions(-)
benchmarks/gem_wsim.c +15 -4
index 9e5bfe6a36d4..60982cb73ba7 100644
@@ -2101,7 +2101,8 @@ static void *run_workload(void *data)
 	struct w_step *w;
 	int throttle = -1;
 	int qd_throttle = -1;
-	int count;
+	int count, missed = 0;
+	unsigned long time_tot = 0, time_min = ULONG_MAX, time_max = 0;
 	int i;
 
 	clock_gettime(CLOCK_MONOTONIC, &t_start);
@@ -2121,12 +2122,19 @@ static void *run_workload(void *data)
 				do_sleep = w->delay;
 			} else if (w->type == PERIOD) {
 				struct timespec now;
+				int elapsed;
 
 				clock_gettime(CLOCK_MONOTONIC, &now);
-				do_sleep = w->period -
-					   elapsed_us(&wrk->repeat_start, &now);
+				elapsed = elapsed_us(&wrk->repeat_start, &now);
+				do_sleep = w->period - elapsed;
+				time_tot += elapsed;
+				if (elapsed < time_min)
+					time_min = elapsed;
+				if (elapsed > time_max)
+					time_max = elapsed;
 				if (do_sleep < 0) {
-					if (verbose > 1)
+					missed++;
+					if (verbose > 2)
 						printf("%u: Dropped period @ %u/%u (%dus late)!\n",
 						       wrk->id, count, i, do_sleep);
 					continue;
@@ -2280,6 +2288,9 @@ static void *run_workload(void *data)
 		printf("%c%u: %.3fs elapsed (%d cycles, %.3f workloads/s).",
 		       wrk->background ? ' ' : '*', wrk->id,
 		       t, count, count / t);
+		if (time_tot)
+			printf(" Time avg/min/max=%lu/%lu/%luus; %u missed.",
+			       time_tot / count, time_min, time_max, missed);
 		putchar('\n');
 	}
 
Signature · 6 lines
2.20.1

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

Re: [Intel-gfx] [PATCH i-g-t 03/10] gem_wsim: Show workload timing stats

Quoting Tvrtko Ursulin (2020-06-17 17:01:13)

Quoted 30 lines

From: Tvrtko Ursulin <[email protected]>

Show average/min/max workload iteration and dropped period stats when 'p' command is used.

---
 benchmarks/gem_wsim.c | 19 +++++++++++++++----
 1 file changed, 15 insertions(+), 4 deletions(-)
benchmarks/gem_wsim.c +3 -1
index 9e5bfe6a36d4..60982cb73ba7 100644
@@ -2101,7 +2101,8 @@ static void *run_workload(void *data)
        struct w_step *w;
        int throttle = -1;
        int qd_throttle = -1;
-       int count;
+       int count, missed = 0;
+       unsigned long time_tot = 0, time_min = ULONG_MAX, time_max = 0;
        int i;
 
        clock_gettime(CLOCK_MONOTONIC, &t_start);
@@ -2121,12 +2122,19 @@ static void *run_workload(void *data)
                                do_sleep = w->delay;
                        } else if (w->type == PERIOD) {
                                struct timespec now;
+                               int elapsed;
 

benchmarks/gem_wsim.c static void *run_workload(void *data) · line 2134

Quoted 10 lines
benchmarks/gem_wsim.c +7 -2
                                clock_gettime(CLOCK_MONOTONIC, &now);
-                               do_sleep = w->period -
-                                          elapsed_us(&wrk->repeat_start, &now);
+                               elapsed = elapsed_us(&wrk->repeat_start, &now);
+                               do_sleep = w->period - elapsed;
+                               time_tot += elapsed;
+                               if (elapsed < time_min)
+                                       time_min = elapsed;
+                               if (elapsed > time_max)
+                                       time_max = elapsed;

Keep the running average?

Quoted 25 lines
benchmarks/gem_wsim.c +5 -1
                                if (do_sleep < 0) {
-                                       if (verbose > 1)
+                                       missed++;
+                                       if (verbose > 2)
                                                printf("%u: Dropped period @ %u/%u (%dus late)!\n",
                                                       wrk->id, count, i, do_sleep);
                                        continue;
@@ -2280,6 +2288,9 @@ static void *run_workload(void *data)
                printf("%c%u: %.3fs elapsed (%d cycles, %.3f workloads/s).",
                       wrk->background ? ' ' : '*', wrk->id,
                       t, count, count / t);
+               if (time_tot)
+                       printf(" Time avg/min/max=%lu/%lu/%luus; %u missed.",
+                              time_tot / count, time_min, time_max, missed);
                putchar('\n');
        }
 
Signature · 6 lines
2.20.1

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

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

#14 lore raw

Re: [Intel-gfx] [PATCH i-g-t 03/10] gem_wsim: Show workload timing stats

Quoted 31 lines · Chris Wilson

Quoting Tvrtko Ursulin (2020-06-17 17:01:13)

Quoted 30 lines

From: Tvrtko Ursulin <[email protected]>

Show average/min/max workload iteration and dropped period stats when 'p' command is used.

---
  benchmarks/gem_wsim.c | 19 +++++++++++++++----
  1 file changed, 15 insertions(+), 4 deletions(-)
benchmarks/gem_wsim.c +3 -1
index 9e5bfe6a36d4..60982cb73ba7 100644
@@ -2101,7 +2101,8 @@ static void *run_workload(void *data)
         struct w_step *w;
         int throttle = -1;
         int qd_throttle = -1;
-       int count;
+       int count, missed = 0;
+       unsigned long time_tot = 0, time_min = ULONG_MAX, time_max = 0;
         int i;
 
         clock_gettime(CLOCK_MONOTONIC, &t_start);
@@ -2121,12 +2122,19 @@ static void *run_workload(void *data)
                                 do_sleep = w->delay;
                         } else if (w->type == PERIOD) {
                                 struct timespec now;
+                               int elapsed;
 
Quoted 12 lines

benchmarks/gem_wsim.c static void *run_workload(void *data) · line 2134

Quoted 10 lines
benchmarks/gem_wsim.c +7 -2
                                 clock_gettime(CLOCK_MONOTONIC, &now);
-                               do_sleep = w->period -
-                                          elapsed_us(&wrk->repeat_start, &now);
+                               elapsed = elapsed_us(&wrk->repeat_start, &now);
+                               do_sleep = w->period - elapsed;
+                               time_tot += elapsed;
+                               if (elapsed < time_min)
+                                       time_min = elapsed;
+                               if (elapsed > time_max)
+                                       time_max = elapsed;

Keep the running average?

Could do but why? I already have the count so adding up total elapsed frame time sound easiest.

Regards,

Tvrtko

Quoted 18 lines
Quoted 18 lines
benchmarks/gem_wsim.c +5 -1
                                 if (do_sleep < 0) {
-                                       if (verbose > 1)
+                                       missed++;
+                                       if (verbose > 2)
                                                 printf("%u: Dropped period @ %u/%u (%dus late)!\n",
                                                        wrk->id, count, i, do_sleep);
                                         continue;
@@ -2280,6 +2288,9 @@ static void *run_workload(void *data)
                 printf("%c%u: %.3fs elapsed (%d cycles, %.3f workloads/s).",
                        wrk->background ? ' ' : '*', wrk->id,
                        t, count, count / t);
+               if (time_tot)
+                       printf(" Time avg/min/max=%lu/%lu/%luus; %u missed.",
+                              time_tot / count, time_min, time_max, missed);
                 putchar('\n');
         }
 
Quoted 8 lines
Quoted 8 lines
Signature · 6 lines
2.20.1

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

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

#15 lore raw

Re: [Intel-gfx] [PATCH i-g-t 03/10] gem_wsim: Show workload timing stats

Quoting Tvrtko Ursulin (2020-06-18 08:46:18)

Quoted 46 lines · Chris Wilson
Quoted 31 lines

Quoting Tvrtko Ursulin (2020-06-17 17:01:13)

Quoted 30 lines

From: Tvrtko Ursulin <[email protected]>

Show average/min/max workload iteration and dropped period stats when 'p' command is used.

---
  benchmarks/gem_wsim.c | 19 +++++++++++++++----
  1 file changed, 15 insertions(+), 4 deletions(-)
benchmarks/gem_wsim.c +3 -1
index 9e5bfe6a36d4..60982cb73ba7 100644
@@ -2101,7 +2101,8 @@ static void *run_workload(void *data)
         struct w_step *w;
         int throttle = -1;
         int qd_throttle = -1;
-       int count;
+       int count, missed = 0;
+       unsigned long time_tot = 0, time_min = ULONG_MAX, time_max = 0;
         int i;
 
         clock_gettime(CLOCK_MONOTONIC, &t_start);
@@ -2121,12 +2122,19 @@ static void *run_workload(void *data)
                                 do_sleep = w->delay;
                         } else if (w->type == PERIOD) {
                                 struct timespec now;
+                               int elapsed;
 
Quoted 12 lines

benchmarks/gem_wsim.c static void *run_workload(void *data) · line 2134

Quoted 10 lines
benchmarks/gem_wsim.c +7 -2
                                 clock_gettime(CLOCK_MONOTONIC, &now);
-                               do_sleep = w->period -
-                                          elapsed_us(&wrk->repeat_start, &now);
+                               elapsed = elapsed_us(&wrk->repeat_start, &now);
+                               do_sleep = w->period - elapsed;
+                               time_tot += elapsed;
+                               if (elapsed < time_min)
+                                       time_min = elapsed;
+                               if (elapsed > time_max)
+                                       time_max = elapsed;

Keep the running average?

Quoted 2 lines

Could do but why? I already have the count so adding up total elapsed frame time sound easiest.

Because I was blind and didn't see it in the printf. -Chris
_______________________________________________
Intel-gfx mailing list [email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#16 lore raw

PATCH 4/10 gem_wsim: Move BO allocation to a helper

In reply to Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

---
 benchmarks/gem_wsim.c | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)
benchmarks/gem_wsim.c +10 -4
index 60982cb73ba7..5893de38a98e 100644
@@ -1472,6 +1472,11 @@ get_ctxid(struct workload *wrk, struct w_step *w)
 	return wrk->ctx_list[w->context].id;
 }
 
+static uint32_t alloc_bo(int i915, unsigned long size)
+{
+	return gem_create(i915, size);
+}
+
 static void
 alloc_step_batch(struct workload *wrk, struct w_step *w, unsigned int flags)
 {
@@ -1483,7 +1488,7 @@ alloc_step_batch(struct workload *wrk, struct w_step *w, unsigned int flags)
 	w->obj = calloc(nr_obj, sizeof(*w->obj));
 	igt_assert(w->obj);
 
-	w->obj[j].handle = gem_create(fd, 4096);
+	w->obj[j].handle = alloc_bo(fd, 4096);
 	w->obj[j].flags = EXEC_OBJECT_WRITE;
 	j++;
 	igt_assert(j < nr_obj);
@@ -1528,7 +1533,8 @@ alloc_step_batch(struct workload *wrk, struct w_step *w, unsigned int flags)
 	else
 		w->bb_sz = get_bb_sz(w, w->duration.max);
 
-	w->bb_handle = w->obj[j].handle = gem_create(fd, w->bb_sz + (w->unbound_duration ? 4096 : 0));
+	w->bb_handle = w->obj[j].handle =
+		alloc_bo(fd, w->bb_sz + (w->unbound_duration ? 4096 : 0));
 	init_bb(w, flags);
 	w->obj[j].relocation_count = terminate_bb(w, flags);
 
@@ -1685,7 +1691,7 @@ static void allocate_working_set(struct working_set *set)
 	igt_assert(set->handles);
 
 	for (i = 0; i < set->nr; i++)
-		set->handles[i] = gem_create(fd, set->sizes[i]);
+		set->handles[i] = alloc_bo(fd, set->sizes[i]);
 }
 
 #define alloca0(sz) ({ size_t sz__ = (sz); memset(alloca(sz__), 0, sz__); })
@@ -2323,7 +2329,7 @@ static unsigned long calibrate_nop(unsigned int tolerance_pct, struct intel_exec
 	do {
 		struct timespec t_start;
 
-		obj.handle = gem_create(fd, size);
+		obj.handle = alloc_bo(fd, size);
 		gem_write(fd, obj.handle, size - sizeof(bbe), &bbe,
 			  sizeof(bbe));
 		gem_execbuf(fd, &eb);
Signature · 6 lines
2.20.1

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

PATCH 5/10 gem_wsim: Support random buffer sizes

In reply to Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

See README for more details.

---
 benchmarks/gem_wsim.c  | 71 +++++++++++++++++++++++++++++++++---------
 benchmarks/wsim/README |  4 +++
 2 files changed, 61 insertions(+), 14 deletions(-)
benchmarks/gem_wsim.c +57 -14
index 5893de38a98e..c1405596c46a 100644
@@ -117,12 +117,18 @@ struct bond {
 	enum intel_engine_id master;
 };
 
+struct work_buffer_size {
+	unsigned long size;
+	unsigned long min;
+	unsigned long max;
+};
+
 struct working_set {
 	int id;
 	bool shared;
 	unsigned int nr;
 	uint32_t *handles;
-	unsigned long *sizes;
+	struct work_buffer_size *sizes;
 };
 
 struct workload;
@@ -203,6 +209,7 @@ struct workload
 	bool print_stats;
 
 	uint32_t bb_prng;
+	uint32_t bo_prng;
 
 	struct timespec repeat_start;
 
@@ -757,10 +764,12 @@ static int add_buffers(struct working_set *set, char *str)
 	 * 4m
 	 * 4g
 	 * 10n4k - 10 4k batches
+	 * 4096-16k - random size in range
 	 */
-	unsigned long *sizes, size;
+	struct work_buffer_size *sizes;
+	unsigned long min_sz, max_sz;
+	char *n, *max = NULL;
 	unsigned int add, i;
-	char *n;
 
 	n = index(str, 'n');
 	if (n) {
@@ -773,16 +782,34 @@ static int add_buffers(struct working_set *set, char *str)
 		add = 1;
 	}
 
-	size = parse_size(str);
-	if (!size)
+	n = index(str, '-');
+	if (n) {
+		*n = 0;
+		max = ++n;
+	}
+
+	min_sz = parse_size(str);
+	if (!min_sz)
 		return -1;
 
+	if (max) {
+		max_sz = parse_size(max);
+		if (!max_sz)
+			return -1;
+	} else {
+		max_sz = min_sz;
+	}
+
 	sizes = realloc(set->sizes, (set->nr + add) * sizeof(*sizes));
 	if (!sizes)
 		return -1;
 
-	for (i = 0; i < add; i++)
-		sizes[set->nr + i] = size;
+	for (i = 0; i < add; i++) {
+		struct work_buffer_size *sz = &sizes[set->nr + i];
+		sz->min = min_sz;
+		sz->max = max_sz;
+		sz->size = 0;
+	}
 
 	set->nr += add;
 	set->sizes = sizes;
@@ -824,7 +851,7 @@ static uint64_t engine_list_mask(const char *_str)
 	return mask;
 }
 
-static void allocate_working_set(struct working_set *set);
+static void allocate_working_set(struct workload *wrk, struct working_set *set);
 
 #define int_field(_STEP_, _FIELD_, _COND_, _ERR_) \
 	if ((field = strtok_r(fstart, ".", &fctx))) { \
@@ -1177,10 +1204,12 @@ add_step:
 
 	wrk->nr_steps = nr_steps;
 	wrk->steps = steps;
+	wrk->flags = flags;
 	wrk->prio = arg->prio;
 	wrk->sseu = arg->sseu;
 	wrk->max_working_set_id = -1;
 	wrk->working_sets = NULL;
+	wrk->bo_prng = (wrk->flags & SYNCEDCLIENTS) ? master_prng : rand();
 
 	free(desc);
 
@@ -1234,7 +1263,7 @@ add_step:
 	 */
 	for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
 		if (w->type == WORKINGSET && w->working_set.shared)
-			allocate_working_set(&w->working_set);
+			allocate_working_set(wrk, &w->working_set);
 	}
 
 	wrk->max_working_set_id = -1;
@@ -1267,6 +1296,7 @@ clone_workload(struct workload *_wrk)
 	igt_assert(wrk);
 	memset(wrk, 0, sizeof(*wrk));
 
+	wrk->flags = _wrk->flags;
 	wrk->prio = _wrk->prio;
 	wrk->sseu = _wrk->sseu;
 	wrk->nr_steps = _wrk->nr_steps;
@@ -1515,7 +1545,7 @@ alloc_step_batch(struct workload *wrk, struct w_step *w, unsigned int flags)
 
 			igt_assert(set->nr);
 			igt_assert(entry->target < set->nr);
-			igt_assert(set->sizes[entry->target]);
+			igt_assert(set->sizes[entry->target].size);
 
 			dep_handle = set->handles[entry->target];
 		}
@@ -1683,15 +1713,27 @@ static size_t sizeof_engines_bond(int count)
 			engines[count]);
 }
 
-static void allocate_working_set(struct working_set *set)
+static unsigned long
+get_buffer_size(struct workload *wrk, const struct work_buffer_size *sz)
+{
+	if (sz->min == sz->max)
+		return sz->min;
+	else
+		return sz->min + hars_petruska_f54_1_random(&wrk->bo_prng) %
+		       (sz->max + 1 - sz->min);
+}
+
+static void allocate_working_set(struct workload *wrk, struct working_set *set)
 {
 	unsigned int i;
 
 	set->handles = calloc(set->nr, sizeof(*set->handles));
 	igt_assert(set->handles);
 
-	for (i = 0; i < set->nr; i++)
-		set->handles[i] = alloc_bo(fd, set->sizes[i]);
+	for (i = 0; i < set->nr; i++) {
+		set->sizes[i].size = get_buffer_size(wrk, &set->sizes[i]);
+		set->handles[i] = alloc_bo(fd, set->sizes[i].size);
+	}
 }
 
 #define alloca0(sz) ({ size_t sz__ = (sz); memset(alloca(sz__), 0, sz__); })
@@ -1707,6 +1749,7 @@ prepare_workload(unsigned int id, struct workload *wrk, unsigned int flags)
 
 	wrk->id = id;
 	wrk->bb_prng = (wrk->flags & SYNCEDCLIENTS) ? master_prng : rand();
+	wrk->bo_prng = (wrk->flags & SYNCEDCLIENTS) ? master_prng : rand();
 	wrk->run = true;
 
 	/*
@@ -1939,7 +1982,7 @@ prepare_workload(unsigned int id, struct workload *wrk, unsigned int flags)
 	 */
 	for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
 		if (w->type == WORKINGSET && !w->working_set.shared)
-			allocate_working_set(&w->working_set);
+			allocate_working_set(wrk, &w->working_set);
 	}
 
 	/*
benchmarks/wsim/README +4 -0
index 3d9143226740..8c71f2fe6579 100644
@@ -307,11 +307,15 @@ For size a byte size can be given, or suffix 'k', 'm' or 'g' can be used (case
 insensitive). Prefix in the format of "<int>n<size>" can also be given to create
 multiple objects of the same size.
 
+Ranges can also be specified using the <min>-<max> syntax.
+
 Examples:
 
   w.1.4k - Working set 1 with a single 4KiB object in it.
   W.2.2M/32768 - Working set 2 with one 2MiB and one 32768 byte object.
   w.3.10n4k/2n20000 - Working set 3 with ten 4KiB and two 20000 byte objects.
+  w.4.4n4k-1m - Working set 4 with four objects of random size between 4KiB and
+		1MiB.
 
 Working set objects can be referenced as data dependency targets using the new
 'r'/'w' syntax. Simple example:
Signature · 6 lines
2.20.1

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

Re: [Intel-gfx] [PATCH i-g-t 05/10] gem_wsim: Support random buffer sizes

Quoting Tvrtko Ursulin (2020-06-17 17:01:15)

Quoted 128 lines

From: Tvrtko Ursulin <[email protected]>

See README for more details.

---
 benchmarks/gem_wsim.c  | 71 +++++++++++++++++++++++++++++++++---------
 benchmarks/wsim/README |  4 +++
 2 files changed, 61 insertions(+), 14 deletions(-)
benchmarks/gem_wsim.c +38 -9
index 5893de38a98e..c1405596c46a 100644
@@ -117,12 +117,18 @@ struct bond {
        enum intel_engine_id master;
 };
 
+struct work_buffer_size {
+       unsigned long size;
+       unsigned long min;
+       unsigned long max;
+};
+
 struct working_set {
        int id;
        bool shared;
        unsigned int nr;
        uint32_t *handles;
-       unsigned long *sizes;
+       struct work_buffer_size *sizes;
 };
 
 struct workload;
@@ -203,6 +209,7 @@ struct workload
        bool print_stats;
 
        uint32_t bb_prng;
+       uint32_t bo_prng;
 
        struct timespec repeat_start;
 
@@ -757,10 +764,12 @@ static int add_buffers(struct working_set *set, char *str)
         * 4m
         * 4g
         * 10n4k - 10 4k batches
+        * 4096-16k - random size in range
         */
-       unsigned long *sizes, size;
+       struct work_buffer_size *sizes;
+       unsigned long min_sz, max_sz;
+       char *n, *max = NULL;
        unsigned int add, i;
-       char *n;
 
        n = index(str, 'n');
        if (n) {
@@ -773,16 +782,34 @@ static int add_buffers(struct working_set *set, char *str)
                add = 1;
        }
 
-       size = parse_size(str);
-       if (!size)
+       n = index(str, '-');
+       if (n) {
+               *n = 0;
+               max = ++n;
+       }
+
+       min_sz = parse_size(str);
+       if (!min_sz)
                return -1;
 
+       if (max) {
+               max_sz = parse_size(max);
+               if (!max_sz)
+                       return -1;
+       } else {
+               max_sz = min_sz;
+       }
+
        sizes = realloc(set->sizes, (set->nr + add) * sizeof(*sizes));
        if (!sizes)
                return -1;
 
-       for (i = 0; i < add; i++)
-               sizes[set->nr + i] = size;
+       for (i = 0; i < add; i++) {
+               struct work_buffer_size *sz = &sizes[set->nr + i];
+               sz->min = min_sz;
+               sz->max = max_sz;
+               sz->size = 0;
+       }
 
        set->nr += add;
        set->sizes = sizes;
@@ -824,7 +851,7 @@ static uint64_t engine_list_mask(const char *_str)
        return mask;
 }
 
-static void allocate_working_set(struct working_set *set);
+static void allocate_working_set(struct workload *wrk, struct working_set *set);
 
 #define int_field(_STEP_, _FIELD_, _COND_, _ERR_) \
        if ((field = strtok_r(fstart, ".", &fctx))) { \
@@ -1177,10 +1204,12 @@ add_step:
 
        wrk->nr_steps = nr_steps;
        wrk->steps = steps;
+       wrk->flags = flags;
        wrk->prio = arg->prio;
        wrk->sseu = arg->sseu;
        wrk->max_working_set_id = -1;
        wrk->working_sets = NULL;
+       wrk->bo_prng = (wrk->flags & SYNCEDCLIENTS) ? master_prng : rand();
 
        free(desc);
 
@@ -1234,7 +1263,7 @@ add_step:
         */
        for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
                if (w->type == WORKINGSET && w->working_set.shared)
-                       allocate_working_set(&w->working_set);
+                       allocate_working_set(wrk, &w->working_set);
        }
 
        wrk->max_working_set_id = -1;

benchmarks/gem_wsim.c clone_workload(struct workload *_wrk) · line 1302

Quoted 7 lines
benchmarks/gem_wsim.c +1 -0
@@ -1267,6 +1296,7 @@ clone_workload(struct workload *_wrk)
        igt_assert(wrk);
        memset(wrk, 0, sizeof(*wrk));
 
+       wrk->flags = _wrk->flags;
        wrk->prio = _wrk->prio;
        wrk->sseu = _wrk->sseu;
        wrk->nr_steps = _wrk->nr_steps;

wrk->flags wasn't introduced in this patch, why are we needing to copy them now.

I see wrk->bo_prn = flags & SYNC above, but I haven't seem them used again later. They used to carry the balancer info and were setup in main. Am I not mistaken in thinking they still are being set in main() as well?
-Chris
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#19 lore raw

Re: [Intel-gfx] [PATCH i-g-t 05/10] gem_wsim: Support random buffer sizes

Quoted 140 lines · Chris Wilson

Quoting Tvrtko Ursulin (2020-06-17 17:01:15)

Quoted 128 lines

From: Tvrtko Ursulin <[email protected]>

See README for more details.

---
  benchmarks/gem_wsim.c  | 71 +++++++++++++++++++++++++++++++++---------
  benchmarks/wsim/README |  4 +++
  2 files changed, 61 insertions(+), 14 deletions(-)
benchmarks/gem_wsim.c +38 -9
index 5893de38a98e..c1405596c46a 100644
@@ -117,12 +117,18 @@ struct bond {
         enum intel_engine_id master;
  };
 
+struct work_buffer_size {
+       unsigned long size;
+       unsigned long min;
+       unsigned long max;
+};
+
  struct working_set {
         int id;
         bool shared;
         unsigned int nr;
         uint32_t *handles;
-       unsigned long *sizes;
+       struct work_buffer_size *sizes;
  };
 
  struct workload;
@@ -203,6 +209,7 @@ struct workload
         bool print_stats;
 
         uint32_t bb_prng;
+       uint32_t bo_prng;
 
         struct timespec repeat_start;
 
@@ -757,10 +764,12 @@ static int add_buffers(struct working_set *set, char *str)
          * 4m
          * 4g
          * 10n4k - 10 4k batches
+        * 4096-16k - random size in range
          */
-       unsigned long *sizes, size;
+       struct work_buffer_size *sizes;
+       unsigned long min_sz, max_sz;
+       char *n, *max = NULL;
         unsigned int add, i;
-       char *n;
 
         n = index(str, 'n');
         if (n) {
@@ -773,16 +782,34 @@ static int add_buffers(struct working_set *set, char *str)
                 add = 1;
         }
 
-       size = parse_size(str);
-       if (!size)
+       n = index(str, '-');
+       if (n) {
+               *n = 0;
+               max = ++n;
+       }
+
+       min_sz = parse_size(str);
+       if (!min_sz)
                 return -1;
 
+       if (max) {
+               max_sz = parse_size(max);
+               if (!max_sz)
+                       return -1;
+       } else {
+               max_sz = min_sz;
+       }
+
         sizes = realloc(set->sizes, (set->nr + add) * sizeof(*sizes));
         if (!sizes)
                 return -1;
 
-       for (i = 0; i < add; i++)
-               sizes[set->nr + i] = size;
+       for (i = 0; i < add; i++) {
+               struct work_buffer_size *sz = &sizes[set->nr + i];
+               sz->min = min_sz;
+               sz->max = max_sz;
+               sz->size = 0;
+       }
 
         set->nr += add;
         set->sizes = sizes;
@@ -824,7 +851,7 @@ static uint64_t engine_list_mask(const char *_str)
         return mask;
  }
 
-static void allocate_working_set(struct working_set *set);
+static void allocate_working_set(struct workload *wrk, struct working_set *set);
 
  #define int_field(_STEP_, _FIELD_, _COND_, _ERR_) \
         if ((field = strtok_r(fstart, ".", &fctx))) { \
@@ -1177,10 +1204,12 @@ add_step:
 
         wrk->nr_steps = nr_steps;
         wrk->steps = steps;
+       wrk->flags = flags;
         wrk->prio = arg->prio;
         wrk->sseu = arg->sseu;
         wrk->max_working_set_id = -1;
         wrk->working_sets = NULL;
+       wrk->bo_prng = (wrk->flags & SYNCEDCLIENTS) ? master_prng : rand();
 
         free(desc);
 
@@ -1234,7 +1263,7 @@ add_step:
          */
         for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
                 if (w->type == WORKINGSET && w->working_set.shared)
-                       allocate_working_set(&w->working_set);
+                       allocate_working_set(wrk, &w->working_set);
         }
 
         wrk->max_working_set_id = -1;

benchmarks/gem_wsim.c clone_workload(struct workload *_wrk) · line 1302

Quoted 7 lines
benchmarks/gem_wsim.c +1 -0
@@ -1267,6 +1296,7 @@ clone_workload(struct workload *_wrk)
         igt_assert(wrk);
         memset(wrk, 0, sizeof(*wrk));
 
+       wrk->flags = _wrk->flags;
         wrk->prio = _wrk->prio;
         wrk->sseu = _wrk->sseu;
         wrk->nr_steps = _wrk->nr_steps;

wrk->flags wasn't introduced in this patch, why are we needing to copy them now.

Quoted 4 lines

I see wrk->bo_prn = flags & SYNC above, but I haven't seem them used again later. They used to carry the balancer info and were setup in main. Am I not mistaken in thinking they still are being set in main() as well?

I couldn't remember but looking around looks like you are right. I'll do some experiments and remove it if confirmed.

Regards,

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

#20 lore raw

PATCH 6/10 gem_wsim: Support scaling workload batch durations

In reply to Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

-f <float> on the command line can be used to scale batch buffer durations in all parsed workloads.

---
 benchmarks/gem_wsim.c | 27 ++++++++++++++++++++-------
 1 file changed, 20 insertions(+), 7 deletions(-)
benchmarks/gem_wsim.c +20 -7
index c1405596c46a..025385a144b8 100644
@@ -41,6 +41,7 @@
 #include <assert.h>
 #include <limits.h>
 #include <pthread.h>
+#include <math.h>
 
 #include "intel_chipset.h"
 #include "intel_reg.h"
@@ -853,6 +854,11 @@ static uint64_t engine_list_mask(const char *_str)
 
 static void allocate_working_set(struct workload *wrk, struct working_set *set);
 
+static long __duration(long dur, double scale)
+{
+	return round(scale * dur);
+}
+
 #define int_field(_STEP_, _FIELD_, _COND_, _ERR_) \
 	if ((field = strtok_r(fstart, ".", &fctx))) { \
 		tmp = atoi(field); \
@@ -863,7 +869,8 @@ static void allocate_working_set(struct workload *wrk, struct working_set *set);
 	} \
 
 static struct workload *
-parse_workload(struct w_arg *arg, unsigned int flags, struct workload *app_w)
+parse_workload(struct w_arg *arg, unsigned int flags, double scale_dur,
+	       struct workload *app_w)
 {
 	struct workload *wrk;
 	unsigned int nr_steps = 0;
@@ -1129,7 +1136,7 @@ parse_workload(struct w_arg *arg, unsigned int flags, struct workload *app_w)
 					  tmpl == LONG_MAX,
 					  "Invalid duration at step %u!\n",
 					  nr_steps);
-				step.duration.min = tmpl;
+				step.duration.min = __duration(tmpl, scale_dur);
 
 				if (sep && *sep == '-') {
 					tmpl = strtol(sep + 1, NULL, 10);
@@ -1139,7 +1146,8 @@ parse_workload(struct w_arg *arg, unsigned int flags, struct workload *app_w)
 						tmpl == LONG_MAX,
 						"Invalid duration range at step %u!\n",
 						nr_steps);
-					step.duration.max = tmpl;
+					step.duration.max = __duration(tmpl,
+								       scale_dur);
 				} else {
 					step.duration.max = step.duration.min;
 				}
@@ -2494,7 +2502,8 @@ static void print_help(void)
 "                    command line. Subsequent -s switches it off.\n"
 "  -S                Synchronize the sequence of random batch durations between\n"
 "                    clients.\n"
-"  -d                Sync between data dependencies in userspace."
+"  -d                Sync between data dependencies in userspace.\n"
+"  -f <scale>        Scale factor for batch durations."
 	);
 }
 
@@ -2556,6 +2565,7 @@ int main(int argc, char **argv)
 	struct w_arg *w_args = NULL;
 	unsigned int tolerance_pct = 1;
 	int exitcode = EXIT_FAILURE;
+	double scale_arg = 1.0f;
 	int prio = 0;
 	double t;
 	int i, c;
@@ -2576,7 +2586,7 @@ int main(int argc, char **argv)
 	master_prng = time(NULL);
 
 	while ((c = getopt(argc, argv,
-			   "ThqvsSdc:n:r:w:W:a:t:p:I:")) != -1) {
+			   "ThqvsSdc:n:r:w:W:a:t:p:I:f:")) != -1) {
 		switch (c) {
 		case 'W':
 			if (master_workload >= 0) {
@@ -2687,6 +2697,9 @@ int main(int argc, char **argv)
 		case 'I':
 			master_prng = strtol(optarg, NULL, 0);
 			break;
+		case 'f':
+			scale_arg = atof(optarg);
+			break;
 		case 'h':
 			print_help();
 			goto out;
@@ -2744,7 +2757,7 @@ int main(int argc, char **argv)
 
 	if (append_workload_arg) {
 		struct w_arg arg = { NULL, append_workload_arg, 0 };
-		app_w = parse_workload(&arg, flags, NULL);
+		app_w = parse_workload(&arg, flags, scale_arg, NULL);
 		if (!app_w) {
 			wsim_err("Failed to parse append workload!\n");
 			goto err;
@@ -2762,7 +2775,7 @@ int main(int argc, char **argv)
 			goto err;
 		}
 
-		wrk[i] = parse_workload(&w_args[i], flags, app_w);
+		wrk[i] = parse_workload(&w_args[i], flags, scale_arg, app_w);
 		if (!wrk[i]) {
 			wsim_err("Failed to parse workload %u!\n", i);
 			goto err;
Signature · 6 lines
2.20.1

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

Re: [Intel-gfx] [PATCH i-g-t 06/10] gem_wsim: Support scaling workload batch durations

Quoting Tvrtko Ursulin (2020-06-17 17:01:16)

Quoted 4 lines

From: Tvrtko Ursulin <[email protected]>

-f <float> on the command line can be used to scale batch buffer durations in all parsed workloads.

But not the period?
-Chris
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#22 lore raw

Re: [Intel-gfx] [PATCH i-g-t 06/10] gem_wsim: Support scaling workload batch durations

Quoted 7 lines · Chris Wilson

Quoting Tvrtko Ursulin (2020-06-17 17:01:16)

Quoted 4 lines

From: Tvrtko Ursulin <[email protected]>

-f <float> on the command line can be used to scale batch buffer durations in all parsed workloads.

But not the period?

I had it scale both at some point but then it ended up more useful to only do batches. So I could stuff more clients in before saturation. I suppose that's an argument to have both independently controlled.

Regards,

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

#23 lore raw

Re: [Intel-gfx] [PATCH i-g-t 06/10] gem_wsim: Support scaling workload batch durations

Quoting Tvrtko Ursulin (2020-06-18 09:01:10)

Quoted 13 lines · Chris Wilson
Quoted 7 lines

Quoting Tvrtko Ursulin (2020-06-17 17:01:16)

Quoted 4 lines

From: Tvrtko Ursulin <[email protected]>

-f <float> on the command line can be used to scale batch buffer durations in all parsed workloads.

But not the period?

I had it scale both at some point but then it ended up more useful to only do batches. So I could stuff more clients in before saturation. I suppose that's an argument to have both independently controlled.

I was moreover trying to work out why one would want to. I was guessing shrink the duration and add more clients, and there you would the independent control, but if you just shrank the duration for a fixed number of clients, you would also want to shrink the period. -Chris
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#24 lore raw

PATCH 7/10 gem_wsim: Log max and active working set sizes in verbose mode

In reply to Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

It is useful to know how much memory workload is allocating.

---
 benchmarks/gem_wsim.c | 100 +++++++++++++++++++++++++++++++++++++++---
 1 file changed, 95 insertions(+), 5 deletions(-)
benchmarks/gem_wsim.c +95 -5
index 025385a144b8..96ee923fb699 100644
@@ -852,7 +852,8 @@ static uint64_t engine_list_mask(const char *_str)
 	return mask;
 }
 
-static void allocate_working_set(struct workload *wrk, struct working_set *set);
+static unsigned long
+allocate_working_set(struct workload *wrk, struct working_set *set);
 
 static long __duration(long dur, double scale)
 {
@@ -1270,8 +1271,14 @@ add_step:
 	 * Allocate shared working sets.
 	 */
 	for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
-		if (w->type == WORKINGSET && w->working_set.shared)
-			allocate_working_set(wrk, &w->working_set);
+		if (w->type == WORKINGSET && w->working_set.shared) {
+			unsigned long total =
+				allocate_working_set(wrk, &w->working_set);
+
+			if (verbose > 1)
+				printf("%u: %lu bytes in shared working set %u\n",
+				       wrk->id, total, w->working_set.id);
+		}
 	}
 
 	wrk->max_working_set_id = -1;
@@ -1731,8 +1738,10 @@ get_buffer_size(struct workload *wrk, const struct work_buffer_size *sz)
 		       (sz->max + 1 - sz->min);
 }
 
-static void allocate_working_set(struct workload *wrk, struct working_set *set)
+static unsigned long
+allocate_working_set(struct workload *wrk, struct working_set *set)
 {
+	unsigned long total = 0;
 	unsigned int i;
 
 	set->handles = calloc(set->nr, sizeof(*set->handles));
@@ -1741,7 +1750,82 @@ static void allocate_working_set(struct workload *wrk, struct working_set *set)
 	for (i = 0; i < set->nr; i++) {
 		set->sizes[i].size = get_buffer_size(wrk, &set->sizes[i]);
 		set->handles[i] = alloc_bo(fd, set->sizes[i].size);
+		total += set->sizes[i].size;
+	}
+
+	return total;
+}
+
+static bool
+find_dep(struct dep_entry *deps, unsigned int nr, struct dep_entry dep)
+{
+	unsigned int i;
+
+	for (i = 0; i < nr; i++) {
+		if (deps[i].working_set == dep.working_set &&
+		    deps[i].target == dep.target)
+			return true;
 	}
+
+	return false;
+}
+
+static void measure_active_set(struct workload *wrk)
+{
+	unsigned long total = 0, batch_sizes = 0;
+	struct dep_entry *deps = NULL;
+	unsigned int nr = 0, i, j;
+	struct w_step *w;
+
+	if (verbose < 3)
+		return;
+
+	for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
+		if (w->type != BATCH)
+			continue;
+
+		batch_sizes += w->bb_sz;
+
+		for (j = 0; j < w->data_deps.nr; j++) {
+			struct dep_entry *dep = &w->data_deps.list[j];
+			struct dep_entry _dep = *dep;
+
+			if (dep->working_set == -1 && dep->target < 0) {
+				int idx = w->idx + dep->target;
+
+				igt_assert(idx >= 0 && idx < w->idx);
+				igt_assert(wrk->steps[idx].type == BATCH);
+
+				_dep.target = wrk->steps[idx].obj[0].handle;
+			}
+
+			if (!find_dep(deps, nr, _dep)) {
+				if (dep->working_set == -1) {
+					total += 4096;
+				} else {
+					struct working_set *set;
+
+					igt_assert(dep->working_set <=
+						   wrk->max_working_set_id);
+
+					set = wrk->working_sets[dep->working_set];
+					igt_assert(set->nr);
+					igt_assert(dep->target < set->nr);
+					igt_assert(set->sizes[dep->target].size);
+
+					total += set->sizes[dep->target].size;
+				}
+
+				deps = realloc(deps, (nr + 1) * sizeof(*deps));
+				deps[nr++] = *dep;
+			}
+		}
+	}
+
+	free(deps);
+
+	printf("%u: %lu bytes active working set in %u buffers. %lu in batch buffers.\n",
+	       wrk->id, total, nr, batch_sizes);
 }
 
 #define alloca0(sz) ({ size_t sz__ = (sz); memset(alloca(sz__), 0, sz__); })
@@ -1750,6 +1834,7 @@ static int
 prepare_workload(unsigned int id, struct workload *wrk, unsigned int flags)
 {
 	struct working_set **sets;
+	unsigned long total = 0;
 	uint32_t share_vm = 0;
 	int max_ctx = -1;
 	struct w_step *w;
@@ -1990,9 +2075,12 @@ prepare_workload(unsigned int id, struct workload *wrk, unsigned int flags)
 	 */
 	for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
 		if (w->type == WORKINGSET && !w->working_set.shared)
-			allocate_working_set(wrk, &w->working_set);
+			total += allocate_working_set(wrk, &w->working_set);
 	}
 
+	if (verbose > 2)
+		printf("%u: %lu bytes in working sets.\n", wrk->id, total);
+
 	/*
 	 * Map of working set ids.
 	 */
@@ -2040,6 +2128,8 @@ prepare_workload(unsigned int id, struct workload *wrk, unsigned int flags)
 		alloc_step_batch(wrk, w, flags);
 	}
 
+	measure_active_set(wrk);
+
 	return 0;
 }
 
Signature · 6 lines
2.20.1

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Reviewed-by on v1 7/10
#25 lore raw

Re: [Intel-gfx] [PATCH i-g-t 07/10] gem_wsim: Log max and active working set sizes in verbose mode

Quoting Tvrtko Ursulin (2020-06-17 17:01:17)

Quoted 28 lines

From: Tvrtko Ursulin <[email protected]>

It is useful to know how much memory workload is allocating.

---
 benchmarks/gem_wsim.c | 100 +++++++++++++++++++++++++++++++++++++++---
 1 file changed, 95 insertions(+), 5 deletions(-)
benchmarks/gem_wsim.c +2 -2
index 025385a144b8..96ee923fb699 100644
@@ -852,7 +852,8 @@ static uint64_t engine_list_mask(const char *_str)
        return mask;
 }
 
-static void allocate_working_set(struct workload *wrk, struct working_set *set);
+static unsigned long
+allocate_working_set(struct workload *wrk, struct working_set *set);
 
 static long __duration(long dur, double scale)
 {
@@ -1270,8 +1271,14 @@ add_step:
         * Allocate shared working sets.
         */
        for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
-               if (w->type == WORKINGSET && w->working_set.shared)

benchmarks/gem_wsim.c add_step: · line 1282

Quoted 10 lines
benchmarks/gem_wsim.c +8 -1
-                       allocate_working_set(wrk, &w->working_set);
+               if (w->type == WORKINGSET && w->working_set.shared) {
+                       unsigned long total =
+                               allocate_working_set(wrk, &w->working_set);
+
+                       if (verbose > 1)
+                               printf("%u: %lu bytes in shared working set %u\n",
+                                      wrk->id, total, w->working_set.id);
+               }
        }

The total total might be nice; although that doesn't reflect usage so might be misleading as to what is the active RSS is at any time.

Quoted 78 lines
benchmarks/gem_wsim.c +63 -1
        wrk->max_working_set_id = -1;
@@ -1731,8 +1738,10 @@ get_buffer_size(struct workload *wrk, const struct work_buffer_size *sz)
                       (sz->max + 1 - sz->min);
 }
 
-static void allocate_working_set(struct workload *wrk, struct working_set *set)
+static unsigned long
+allocate_working_set(struct workload *wrk, struct working_set *set)
 {
+       unsigned long total = 0;
        unsigned int i;
 
        set->handles = calloc(set->nr, sizeof(*set->handles));
@@ -1741,7 +1750,82 @@ static void allocate_working_set(struct workload *wrk, struct working_set *set)
        for (i = 0; i < set->nr; i++) {
                set->sizes[i].size = get_buffer_size(wrk, &set->sizes[i]);
                set->handles[i] = alloc_bo(fd, set->sizes[i].size);
+               total += set->sizes[i].size;
+       }
+
+       return total;
+}
+
+static bool
+find_dep(struct dep_entry *deps, unsigned int nr, struct dep_entry dep)
+{
+       unsigned int i;
+
+       for (i = 0; i < nr; i++) {
+               if (deps[i].working_set == dep.working_set &&
+                   deps[i].target == dep.target)
+                       return true;
        }
+
+       return false;
+}
+
+static void measure_active_set(struct workload *wrk)
+{
+       unsigned long total = 0, batch_sizes = 0;
+       struct dep_entry *deps = NULL;
+       unsigned int nr = 0, i, j;
+       struct w_step *w;
+
+       if (verbose < 3)
+               return;
+
+       for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
+               if (w->type != BATCH)
+                       continue;
+
+               batch_sizes += w->bb_sz;
+
+               for (j = 0; j < w->data_deps.nr; j++) {
+                       struct dep_entry *dep = &w->data_deps.list[j];
+                       struct dep_entry _dep = *dep;
+
+                       if (dep->working_set == -1 && dep->target < 0) {
+                               int idx = w->idx + dep->target;
+
+                               igt_assert(idx >= 0 && idx < w->idx);
+                               igt_assert(wrk->steps[idx].type == BATCH);
+
+                               _dep.target = wrk->steps[idx].obj[0].handle;
+                       }
+
+                       if (!find_dep(deps, nr, _dep)) {
+                               if (dep->working_set == -1) {
+                                       total += 4096;
+                               } else {
+                                       struct working_set *set;
+
+                                       igt_assert(dep->working_set <=
+                                                  wrk->max_working_set_id);
+
+                                       set = wrk->working_sets[dep->working_set];
+                                       igt_assert(set->nr);
+                                       igt_assert(dep->target < set->nr);

benchmarks/gem_wsim.c static void allocate_working_set(struct workload *wrk, struct working_set *set) · line 1823

Quoted 10 lines
benchmarks/gem_wsim.c +10 -0
+                                       igt_assert(set->sizes[dep->target].size);
+
+                                       total += set->sizes[dep->target].size;
+                               }
+
+                               deps = realloc(deps, (nr + 1) * sizeof(*deps));
+                               deps[nr++] = *dep;
+                       }
+               }
+       }

So a sum of all the unique handles used by all the steps.

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

#26 lore raw

PATCH 8/10 gem_wsim: Snippet of a workload extracted from carchase

In reply to Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

Some frames from the middle of a demo with corresponding buffers.

---
 benchmarks/wsim/carchasepart.wsim | 184 ++++++++++++++++++++++++++++++
 1 file changed, 184 insertions(+)
 create mode 100644 benchmarks/wsim/carchasepart.wsim
benchmarks/wsim/carchasepart.wsim new file +45 -0
new file mode 100644
index 000000000000..4407d0ef47dc
@@ -0,0 +1,184 @@
+w.0.118n8192
+w.1.69n12288
+w.10.145n131072
+w.11.1n163840
+w.12.3n196608
+w.13.2n229376
+w.14.2n262144
+w.15.7n327680
+w.16.2n393216
+w.17.9n458752
+w.18.30n524288
+w.19.1n655360
+w.2.74n16384
+w.20.2n917504
+w.21.1n1048576
+w.22.33n1310720
+w.23.1n1572864
+w.24.24n1835008
+w.25.117n2097152
+w.26.1n2621440
+w.27.2n3670016
+w.28.4n4194304
+w.29.3n6291456
+w.3.123n20480
+w.30.1n7340032
+w.31.1n8388608
+w.32.20n10485760
+w.33.4n12582912
+w.34.3n14680064
+w.35.1n25165824
+w.4.19n24576
+w.5.2n32768
+w.6.2n40960
+w.7.4n49152
+w.8.2n65536
+w.9.9n81920
+1.RCS.5736.r0-58/r3-80/r22-31/r3-41/r3-42/r9-4/w27-1/r3-18/r22-1/r9-5/r0-32/r3-6/r32-8/r3-78/r3-113/r0-107/w28-0/r3-98/r3-91/w28-1/r3-93/w32-14/r32-0/r33-2/w32-16.0
+2.RCS.1000.r9-4/r3-41/r3-42/r32-16/r3-93/w4-16/r0-3/w3-122/r3-91/r0-58.0
+2.RCS.5950.r9-4/r33-2/r3-6/r22-31/r3-42/w0-103/w0-106/r3-41/r2-54.0
+1.RCS.14562.r0-58/r3-80/r22-31/r3-41/r3-42/r9-4/w32-18/r3-18/r22-1/r9-5/r3-98/r32-14/r3-93/r33-2/r3-6/r32-16/r3-122/r3-106/r3-78/r2-54/r0-107/w32-7/w32-11/w32-2/r33-0/r25-116/r0-106/r16-0/r28-3/r0-3/r0-111/r22-32/r10-58/r4-13/r4-3/r4-1.0
+1.RCS.6494.r3-80/r22-31/r3-41/r3-42/r9-4/w9-5/r0-71/r0-58/w32-19/r3-18/r22-1/r3-78/r3-88/r22-5/r22-14/r22-6/r22-29/r22-7/r22-12/r22-18/r22-28/r22-3/r22-0/r3-75/r3-104/w3-64/r0-60/r22-20/r25-99/r22-16/r21-0/r22-21/r0-97/r22-27/r0-98/r22-4/r0-84/r3-47/w28-2/r0-107/w8-0/r0-48.0
+d.36394
+2.RCS.2207.r9-4/r28-2/r3-42/r22-31/w8-0/r2-47/r20-0/w22-23/r3-47/r3-95/r0-58/w22-10/r3-80.0
+d.17275
+1.RCS.1000.r3-47/w27-0/r0-58/r3-80/r22-31/r3-42/r9-4/w34-0/r3-18/r3-41/r21-0/r9-5/r3-78/r25-4/r3-104/r3-23/r30-0/r3-88/r22-27/r22-1/r25-45/r3-50/r22-12/r22-22/r22-3/r22-0/r25-56/r3-4/r22-15/r25-113/r3-7/r22-18/r25-60/r3-81/r25-21/r3-89/r18-5/r3-93/r17-8/r25-28/r25-87/r25-9/r25-13/r25-42/r25-90/r22-16/r34-2/r3-15/w3-64/r0-67/r25-99/r25-73/r3-6/r25-40/r3-90/r22-20/r0-45/r3-110/w32-17/w32-16/w32-3/w32-1/w33-2/r28-2/r3-98/r3-85/r25-48/r3-12/r25-104/r24-23/r3-87/r3-108/r3-26/r3-96/r22-5/r22-14/r3-49/r3-103/r22-6/r3-68/r3-112/r22-29/r22-28/r25-14/r25-44/r25-19/r3-67/r25-111/r18-4/r3-66/r18-17/r4-5/r25-68/r25-86/r25-26/r25-67/r3-37/r25-0/r22-7/r25-59/r25-71/r25-101/r25-75/r25-20/r25-91/r3-2/r3-117/r3-33/r22-2/r25-55/r25-66/r25-24/r25-105/r25-61/r25-11/r25-51/r25-64/r25-70/r18-19/r18-26/r18-21/r25-81/r25-78/r25-37/r25-50/r25-102/r25-35/r18-18/r18-13/r18-12/r3-69/r3-19/r3-100/r22-21/r25-22/r3-29/r25-93/r18-2/r18-14/r18-3/r22-10/r18-23/r18-7/r18-11/r3-73/r8-0/r25-92/r25-41/w33-3/r0-107/w19-0.

0
+d.2101
+1.RCS.32763.r9-4/r3-18/r3-47/r22-3/r22-31/w19-0/r3-98/r33-3/r3-6/r32-3/r3-78/r3-80/r3-73/r0-107/r0-58/r3-110/w31-0/w33-2/r32-19/r3-114/w32-4/w29-1/w32-8/r9-5/r32-17/r32-1/r30-0/r3-104/r35-0/r34-1/r34-0/r3-92/r34-2/r22-12/r22-1/w32-0/r3-75/r3-118/r3-70/r3-1/r3-113/r22-29/r22-14/r3-109/r3-28/r3-79/r22-7/r3-8/r4-5/r3-85/r3-115/r3-55/r3-13/r3-59/r22-2.0
+d.1788
+2.RCS.33594.r9-4/r0-26/r3-110/w7-2/r3-47/r0-109/r0-75/w6-1/r0-29/w7-0/r0-68/w7-1/r0-58/r32-8/r3-6/r22-31/w0-2/r3-113/r9-5/w0-32.0
+1.RCS.5573.r0-58/r3-80/r22-31/r3-47/r3-110/r9-4/w27-1/r3-18/r22-1/r9-5/r0-32/r3-6/r32-8/r3-78/r3-113/r0-107/w28-0/r3-98/r3-91/w28-1/r3-93/r3-60/w32-14/r32-0/r33-2/w32-16.0
+2.RCS.1000.r9-4/r3-60/r3-110/r32-16/r3-93/w4-16/r0-3/w3-122/r3-91/r0-58.0
+2.RCS.5826.r9-4/r33-2/r3-6/r22-31/r3-110/w0-103/w0-22/r3-60/r2-54.0
+1.RCS.14378.r0-58/r3-80/r22-31/r3-60/r3-110/r9-4/w32-18/r3-18/r3-47/r22-1/r9-5/r3-98/r32-14/r3-93/r33-2/r3-6/r32-16/r3-122/r3-106/r3-78/r2-54/r0-107/w32-7/w32-11/w32-2/r33-0/r25-116/r0-22/r16-0/r28-3/r0-3/r0-111/r22-32/r10-58/r4-13/r4-3/r4-1.0
+1.RCS.6218.r3-80/r22-31/r3-60/r3-110/r9-4/w9-5/r0-53/r0-58/w32-19/r3-18/r3-47/r22-1/r3-78/r3-88/r22-5/r22-14/r22-6/r22-29/r22-7/r22-12/r22-18/r22-28/r22-3/r22-0/r3-75/r3-104/w3-64/r0-115/r22-20/r25-99/r22-16/r21-0/r22-21/r0-35/r22-27/r0-19/r22-4/r0-92/w28-2/r0-107/r3-0/w8-0/r0-76.0
+d.36049
+2.RCS.2085.r9-4/r28-2/r3-0/r22-31/w8-0/r2-9/r20-0/w22-23/r3-60/r3-95/r0-58/w22-10/r3-80.0
+d.12310 +1.RCS.1000.r3-60/w27-0/r0-58/r3-80/r22-31/r3-0/r9-4/w34-0/r3-18/r3-47/r21-0/r9-5/r3-78/r25-4/r3-104/r3-23/r30-0/r3-88/r22-27/r22-1/r25-45/r3-50/r22-12/r22-22/r22-3/r22-0/r25-56/r3-4/r22-15/r25-113/r3-7/r22-18/r25-60/r3-81/r25-21/r3-89/r18-5/r3-93/r17-8/r25-28/r25-87/r25-9/r25-13/r25-42/r25-90/r22-16/r34-2/r3-15/w3-64/r0-88/r25-99/r25-73/r3-6/r25-40/r3-90/r22-20/r0-71/w32-17/w32-16/w32-3/w32-1/w33-2/r28-2/r3-98/r3-85/r25-48/r3-12/r25-104/r24-23/r3-87/r3-108/r3-26/r3-96/r22-5/r22-14/r3-49/r3-103/r22-6/r3-68/r3-112/r22-29/r22-28/r25-14/r25-44/r25-19/r3-67/r25-111/r18-4/r3-66/r18-17/r4-5/r25-68/r25-86/r25-26/r25-67/r3-37/r25-0/r22-7/r25-59/r25-71/r25-101/r25-75/r25-20/r25-91/r3-2/r3-117/r3-33/r22-2/r25-55/r25-66/r25-24/r25-105/r25-61/r25-11/r25-51/r25-64/r25-70/r18-19/r18-26/r18-21/r25-81/r25-78/r25-37/r25-50/r25-102/r25-35/r18-18/r18-13/r18-12/r3-69/r3-19/r3-100/r22-21/r25-22/r3-29/r25-93/r18-2/r18-14/r18-3/r22-10/r18-23/r18-7/r18-11/r3-73/r8-0/r25-92/r25-41/w33-3/r0-107/w19-0.0 +d.3238
+1.RCS.32454.r9-4/r3-18/r3-60/r22-3/r22-31/w19-0/r3-98/r33-3/r3-6/r32-3/r3-78/r3-80/r3-73/r0-107/r0-58/r3-0/w31-0/w33-2/r32-19/r3-114/w32-4/w29-1/r3-74/w32-8/r9-5/r32-17/r32-1/r30-0/r3-104/r35-0/r34-1/r34-0/r3-92/r34-2/r22-12/r22-1/w32-0/r3-75/r3-118/r3-70/r3-1/r3-113/r22-29/r22-14/r3-109/r3-28/r3-79/r22-7/r3-8/r4-5/r3-85/r3-115/r3-55/r3-13/r3-59/r22-2.0
+d.3160
+2.RCS.33268.r9-4/r0-60/r3-0/w7-2/r3-60/r0-109/r3-74/r0-97/w6-1/r0-98/w7-0/r0-84/w7-1/r0-58/r32-8/r3-6/r22-31/w0-2/r3-113/r9-5/w0-32.0
+1.RCS.5675.r0-58/r3-80/r22-31/r3-74/r3-0/r9-4/w27-1/r3-18/r3-60/r22-1/r9-5/r0-32/r3-6/r32-8/r3-78/r3-113/r0-107/w28-0/r3-98/r3-91/w28-1/r3-93/w32-14/r32-0/r33-2/w32-16.0
+d.1148
+2.RCS.1000.r9-4/r3-74/r3-0/r32-16/r3-93/w4-16/r0-3/w3-122/r3-91/r0-58.0
+2.RCS.5911.r9-4/r33-2/r3-6/r22-31/r3-0/w0-103/w0-48/r3-74/r2-54.0
+1.RCS.13981.r0-58/r3-80/r22-31/r3-74/r3-0/r9-4/w32-18/r3-18/r3-60/r22-1/r9-5/r3-98/r32-14/r3-93/r33-2/r3-6/r32-16/r3-122/r3-106/r3-78/r2-54/r0-107/w32-7/w32-11/w32-2/r33-0/r25-116/r0-48/r16-0/r28-3/r0-3/r0-111/r22-32/r10-58/r4-13/r4-3/r4-1.0
+1.RCS.6208.r3-80/r22-31/r3-74/r3-0/r9-4/w9-5/r0-67/r0-58/w32-19/r3-18/r3-60/r22-1/r3-78/r3-88/r22-5/r22-14/r22-6/r22-29/r22-7/r22-12/r22-18/r22-28/r22-3/r22-0/r3-75/r3-104/w3-64/r0-45/r22-20/r25-99/r22-16/r21-0/r0-64/r3-42/r22-21/r22-27/r0-42/r22-4/r0-52/w28-2/r0-107/w8-0/r0-87.0
+d.41148
+2.RCS.2070.r9-4/r28-2/r3-42/r22-31/w8-0/r2-47/r20-0/w22-23/r3-74/r3-95/r0-58/w22-10/r3-80.0
+d.18190 +1.RCS.1000.r3-74/w27-0/r0-58/r3-80/r22-31/r3-42/r9-4/w34-0/r3-18/r3-60/r21-0/r9-5/r3-78/r25-4/r3-104/r3-23/r30-0/r3-88/r22-27/r22-1/r25-45/r3-50/r22-12/r22-22/r22-3/r22-0/r25-56/r3-4/r22-15/r25-113/r3-7/r22-18/r25-60/r3-81/r25-21/r3-89/r18-5/r3-93/r17-8/r25-28/r25-87/r25-9/r25-13/r25-42/r25-90/r22-16/r34-2/r3-15/w3-64/r0-106/r25-99/r25-73/r3-6/r25-40/r3-90/r22-20/r0-53/r3-31/w32-17/w32-16/w32-3/w32-1/w33-2/r28-2/r3-98/r3-85/r25-48/r3-12/r25-104/r24-23/r3-49/r3-103/r3-96/r22-6/r22-14/r3-87/r3-108/r3-26/r22-5/r22-29/r3-68/r3-112/r22-28/r25-14/r25-44/r25-19/r3-67/r25-111/r18-4/r3-66/r18-17/r4-5/r25-68/r25-86/r25-26/r25-67/r3-37/r25-0/r25-59/r25-71/r25-101/r25-75/r25-20/r25-91/r22-7/r3-2/r3-117/r3-33/r22-2/r25-55/r25-66/r25-24/r25-105/r25-61/r25-11/r25-51/r25-64/r25-70/r25-81/r25-78/r25-37/r25-50/r25-102/r25-35/r18-18/r18-13/r18-12/r3-69/r3-19/r3-100/r18-19/r18-26/r18-21/r22-21/r25-22/r3-29/r25-93/r22-10/r18-23/r18-7/r18-11/r3-73/r8-0/r25-92/r25-41/w33-3/r0-107/w19-0.0 +d.1071
+1.RCS.31421.r0-58/r3-80/r22-31/r3-31/r3-42/r9-4/w31-0/r3-18/r22-3/r3-98/r19-0/r3-6/w33-2/r32-19/r3-114/r0-107/w32-4/r3-78/w29-1/w32-8/r9-5/r32-17/r32-3/r32-1/r30-0/r3-104/r35-0/r34-1/r34-0/r3-92/r34-2/r22-12/r22-1/w32-0/r3-75/r3-118/r3-70/r3-1/r3-113/r22-29/r22-14/r3-109/r3-28/r3-79/r22-7/r3-8/r4-5/r3-85/r3-115/r3-55/r3-13/r3-59/r22-2.0
+d.1795
+2.RCS.32222.r9-4/r0-115/r3-42/w7-2/r3-74/r0-109/r3-31/r0-35/w6-1/r0-19/w7-0/r0-92/w7-1/r0-58/r32-8/r3-6/r22-31/w0-2/r3-113/r9-5/w0-32.0
+1.RCS.5663.r0-58/r3-80/r22-31/r3-31/r3-42/r9-4/w27-1/r3-18/r22-1/r9-5/r0-32/r3-6/r32-8/r3-78/r3-113/r0-107/w28-0/r3-98/r3-91/w28-1/r3-93/r3-110/w32-14/r32-0/r33-2/w32-16.0
+2.RCS.1000.r9-4/r3-31/r3-110/r32-16/r3-93/w4-16/r0-3/w3-122/r3-91/r0-58.0
+2.RCS.5878.r9-4/r33-2/r3-6/r22-31/r3-110/w0-103/w0-76/r3-31/r2-54.0
+1.RCS.14059.r0-58/r3-80/r22-31/r3-31/r3-110/r9-4/w32-18/r3-18/r22-1/r9-5/r3-98/r32-14/r3-93/r33-2/r3-6/r32-16/r3-122/r3-106/r3-78/r2-54/r0-107/w32-7/w32-11/w32-2/r33-0/r25-116/r0-76/r16-0/r28-3/r0-3/r0-111/r22-32/r10-58/r4-13/r4-3/r4-1.0
+1.RCS.6168.r3-80/r22-31/r3-31/r3-110/r9-4/w9-5/r0-88/r0-58/w32-19/r3-18/r22-1/r3-78/r3-88/r22-5/r22-14/r22-6/r22-29/r22-7/r22-12/r22-18/r22-28/r22-3/r22-0/r3-75/r3-104/w3-64/r0-71/r22-20/r25-99/r22-16/r21-0/r0-26/r22-21/r0-75/r3-20/r22-4/r22-27/r0-29/w28-2/r0-107/w8-0/r0-68.0
+d.32708
+2.RCS.2145.r9-4/r28-2/r3-110/r22-31/w8-0/r2-9/r20-0/w22-23/r3-20/r3-95/r0-58/w22-10/r3-80.0
+d.11843 +1.RCS.1000.r3-20/w27-0/r0-58/r3-80/r22-31/r3-110/r9-4/w34-0/r3-18/r3-31/r21-0/r9-5/r3-78/r25-4/r3-104/r3-23/r30-0/r3-88/r22-27/r22-1/r25-45/r3-50/r22-12/r22-22/r22-3/r22-0/r25-56/r3-4/r22-15/r25-113/r3-7/r22-18/r25-60/r3-81/r25-21/r3-89/r18-5/r3-93/r17-8/r25-28/r25-87/r25-9/r25-13/r25-42/r25-90/r22-16/r34-2/r3-15/w3-64/r0-22/r25-99/r25-73/r3-6/r25-40/r3-90/r22-20/r0-67/w32-17/w32-16/w32-3/w32-1/w33-2/r28-2/r3-98/r3-85/r25-48/r3-12/r25-104/r3-87/r3-108/r3-26/r3-96/r22-5/r22-14/r22-29/r3-68/r3-112/r3-103/r3-49/r22-6/r22-28/r25-14/r25-44/r25-19/r3-67/r25-111/r18-4/r3-66/r18-17/r4-5/r25-68/r25-86/r25-26/r25-59/r25-71/r25-101/r25-67/r3-37/r25-0/r22-7/r25-75/r25-20/r25-91/r3-2/r3-117/r3-33/r22-2/r25-55/r25-66/r25-24/r25-105/r25-61/r25-11/r25-51/r25-64/r25-70/r18-19/r18-26/r18-21/r25-81/r25-78/r25-37/r25-50/r25-102/r25-35/r9-2/r18-18/r18-13/r18-12/r3-69/r3-19/r3-100/r22-21/r25-82/r25-77/r25-33/r25-22/r3-29/r25-93/r22-10/r18-23/r18-7/r18-11/r3-73/r8-0/r25-92/r25-41/w33-3/r0-107/w19-0.0 +d.3354
+1.RCS.31662.r0-58/r3-80/r22-31/r3-20/r3-110/r9-2/w31-0/r3-18/r22-3/r3-98/r19-0/r3-6/w33-2/r32-19/r3-114/r0-107/w32-4/r3-78/w29-1/w32-8/r9-5/r32-17/r32-3/r32-1/r30-0/r3-104/r35-0/r34-1/r34-0/r3-92/r34-2/r22-12/r22-1/w32-0/r3-75/r3-118/r3-70/r3-1/r3-113/r22-29/r22-14/r3-109/r3-28/r3-79/r22-7/r3-8/r4-5/r3-85/r3-115/r3-55/r3-13/r3-59/r22-2.0
+d.2763
+2.RCS.32465.r0-58/r9-2/r0-45/r3-110/w7-2/r3-20/r0-109/r0-64/w6-1/r0-42/w7-0/r0-52/w7-1/r32-8/r3-6/r22-31/w0-2/r3-113/r9-5/w0-32.0
+1.RCS.5493.r0-58/r3-80/r22-31/r3-20/r3-110/r9-2/w27-1/r3-18/r22-1/r9-5/r0-32/r3-6/r32-8/r3-78/r3-113/r0-107/w28-0/r3-98/r3-91/r3-0/w28-1/r3-93/r3-41/w32-14/r32-0/r33-2/w32-16.0
+2.RCS.1000.r9-2/r3-41/r3-0/r32-16/r3-93/w4-16/r0-3/w3-122/r3-91/r0-58.0
+2.RCS.5680.r9-2/r33-2/r3-6/r22-31/r3-0/w0-103/w0-87/r3-41/r2-54.0
+1.RCS.13742.r0-58/r3-80/r22-31/r3-41/r3-0/r9-2/w32-18/r3-18/r3-20/r22-1/r9-5/r3-98/r32-14/r3-93/r33-2/r3-6/r32-16/r3-122/r3-106/r3-78/r2-54/r0-107/w32-7/w32-11/w32-2/r33-0/r25-116/r0-87/r16-0/r28-3/r0-3/r0-111/r22-32/r10-58/r4-13/r4-3/r4-1.0
+1.RCS.5993.r3-80/r22-31/r3-41/r3-0/r9-2/w9-5/r0-106/r0-58/w32-19/r3-18/r3-20/r22-1/r3-78/r3-88/r22-5/r22-14/r22-6/r22-29/r22-7/r22-12/r22-18/r22-28/r22-3/r22-0/r3-75/r3-104/w3-64/r0-53/r22-20/r25-99/r22-16/r21-0/r0-60/r22-21/r0-97/r22-27/r0-98/r22-4/w28-2/r0-107/w8-0/r0-84.0
+d.37477
+2.RCS.1875.r9-2/r28-2/r3-0/r22-31/w8-0/r2-47/r20-0/w22-23/r3-41/r3-95/r0-58/w22-10/r3-80.0
+d.12907 +1.RCS.1000.r3-41/w27-0/r0-58/r3-80/r22-31/r3-0/r9-2/w34-0/r3-18/r3-20/r21-0/r9-5/r3-78/r25-4/r3-104/r3-23/r30-0/r3-88/r22-27/r22-1/r25-45/r3-50/r22-12/r22-22/r22-3/r22-0/r25-56/r3-4/r22-15/r25-113/r3-7/r22-18/r25-60/r3-81/r25-21/r3-89/r18-5/r3-93/r17-8/r25-28/r25-87/r25-9/r25-13/r25-42/r25-90/r22-16/r34-2/r3-15/w3-64/r0-48/r25-99/r25-73/r3-6/r25-40/r3-90/r22-20/r0-88/w32-17/w32-16/w32-3/w32-1/w33-2/r28-2/r3-98/r3-87/r3-108/r3-26/r3-96/r22-5/r22-14/r22-29/r3-68/r3-112/r3-103/r3-49/r3-12/r22-6/r3-85/r25-104/r22-28/r25-14/r25-44/r25-19/r3-67/r25-111/r18-4/r3-66/r18-17/r4-5/r25-68/r25-86/r25-26/r25-59/r25-71/r25-101/r25-67/r3-37/r25-0/r25-75/r25-20/r25-91/r22-7/r3-2/r3-117/r3-33/r22-2/r25-55/r25-66/r25-24/r25-105/r25-61/r25-11/r25-51/r25-64/r25-70/r25-81/r25-78/r25-37/r25-50/r25-102/r25-35/r18-18/r18-13/r18-12/r3-69/r3-19/r3-100/r18-19/r18-26/r18-21/r22-21/r25-82/r25-77/r25-33/r25-22/r3-29/r25-93/r22-10/r25-110/r25-62/r25-72/r22-9/r8-0/r18-23/r18-7/r18-11/r3-73/r25-92/r25-41/w33-3/r0-10
7/w19-0/w31-0.0
+d.3000
+1.RCS.33546.r9-2/r3-18/r3-41/r22-3/r22-31/w31-0/r3-98/r19-0/r3-6/w33-2/r32-19/r3-0/r3-80/r3-114/r0-107/r0-58/w32-4/r3-47/r3-78/r3-42/w29-1/w32-8/r9-5/r32-17/r32-3/r32-1/r30-0/r3-104/r35-0/r34-1/r34-0/r3-92/r34-2/r22-12/r22-1/w32-0/r3-75/r3-118/r3-70/r3-1/r3-113/r22-29/r22-14/r3-109/r3-28/r3-79/r22-7/r3-8/r4-5/r3-85/r3-115/r3-55/r3-13/r3-59/r22-2.0
+d.3250
+2.RCS.34417.r9-2/r0-71/r3-42/w7-2/r3-41/r0-109/r3-47/r0-26/w6-1/r0-75/w7-0/r0-29/w7-1/r0-58/r32-8/r3-6/r22-31/w0-2/r3-113/r9-5/w0-32.0
+1.RCS.5575.r0-58/r3-80/r22-31/r3-47/r3-42/r9-2/w27-1/r3-18/r3-41/r22-1/r9-5/r0-32/r3-6/r32-8/r3-78/r3-113/r0-107/w28-0/r3-98/r3-91/w28-1/r3-93/w32-14/r32-0/r33-2/w32-16.0
+2.RCS.1000.r9-2/r3-47/r3-42/r32-16/r3-93/w4-16/r0-3/w3-122/r3-91/r0-58.0
+2.RCS.5743.r9-2/r33-2/r3-6/r22-31/r3-42/w0-103/w0-68/r3-47/r2-54.0
+1.RCS.13966.r0-58/r3-80/r22-31/r3-47/r3-42/r9-2/w32-18/r3-18/r3-41/r22-1/r9-5/r3-98/r32-14/r3-93/r33-2/r3-6/r32-16/r3-122/r3-106/r3-78/r2-54/r0-107/w32-7/w32-11/w32-2/r33-0/r25-116/r0-68/r16-0/r28-3/r0-3/r0-111/r22-32/r10-58/r4-13/r4-3/r4-1.0
+1.RCS.6056.r3-80/r22-31/r3-47/r3-42/r9-2/w9-5/r0-22/r0-58/w32-19/r3-18/r3-41/r22-1/r3-78/r3-88/r22-5/r22-14/r22-6/r22-29/r22-7/r22-12/r22-18/r22-28/r22-3/r22-0/r3-75/r3-104/w3-64/r0-67/r22-20/r25-99/r22-16/r21-0/r0-115/r22-21/r0-35/r22-27/r0-19/w28-2/r0-107/w8-0/r0-92.0
+d.36302
+2.RCS.1862.r9-2/r28-2/r3-42/r22-31/w8-0/r2-9/r20-0/w22-23/r3-47/r3-95/r0-58/w22-10/r3-80.0
+d.15703 +1.RCS.1000.r3-47/w27-0/r0-58/r3-80/r22-31/r3-42/r9-2/w34-0/r3-18/r3-41/r21-0/r9-5/r3-78/r25-4/r3-104/r3-23/r30-0/r3-88/r22-27/r22-1/r25-45/r3-50/r22-12/r22-22/r22-3/r22-0/r25-56/r3-4/r22-15/r25-113/r3-7/r22-18/r25-60/r3-81/r25-21/r3-89/r18-5/r3-93/r17-8/r25-28/r25-87/r25-9/r25-13/r25-42/r25-90/r22-16/r34-2/r3-15/w3-64/r0-76/r25-99/r25-73/r3-6/r25-40/r3-90/r22-20/r0-106/r3-60/w32-17/r3-27/w32-16/w32-3/w32-1/w33-2/r28-2/r3-98/r3-87/r3-108/r3-26/r3-96/r22-5/r22-14/r22-29/r3-68/r3-112/r3-103/r3-49/r3-12/r22-6/r3-85/r25-104/r22-28/r25-14/r25-44/r25-19/r3-67/r25-111/r18-4/r3-66/r18-17/r4-5/r25-68/r25-86/r25-26/r25-59/r25-71/r25-101/r25-67/r3-37/r25-0/r25-75/r25-20/r25-91/r22-7/r3-2/r3-117/r3-33/r22-2/r25-55/r25-66/r25-24/r25-105/r25-61/r25-11/r25-51/r25-64/r25-70/r25-81/r25-78/r25-37/r25-50/r25-102/r25-35/r18-18/r18-13/r18-12/r3-69/r3-19/r3-100/r18-19/r18-26/r18-21/r22-21/r25-82/r25-77/r25-33/r25-22/r3-29/r25-93/r22-10/r25-110/r25-62/r25-72/r22-9/r8-0/r18-23/r18-7/r18-11/r3-73/r25-92/r25-
41/w33-3/r0-107/w19-0/w31-0.0
+d.2669
+1.RCS.33805.r9-2/r3-18/r3-60/r22-3/r22-31/w31-0/r3-98/r19-0/r3-6/w33-2/r32-19/r3-27/r3-80/r3-114/r0-107/r0-58/w32-4/r3-78/w29-1/w32-8/r9-5/r32-17/r32-3/r32-1/r30-0/r3-104/r35-0/r34-1/r34-0/r3-92/r34-2/r22-12/r22-1/w32-0/r3-75/r3-118/r3-70/r3-1/r3-113/r22-29/r22-14/r3-109/r3-28/r3-79/r22-7/r3-8/r4-5/r3-85/r3-115/r3-55/r3-13/r3-59/r22-2.0
+d.2564
+2.RCS.34680.r9-2/r0-53/r3-27/w7-2/r3-47/r0-109/r3-60/r0-60/w6-1/r0-97/w7-0/r0-98/w7-1/r0-58/r32-8/r3-6/r22-31/w0-2/r3-113/r9-5/w0-32.0
+1.RCS.5597.r0-58/r3-80/r22-31/r3-60/r3-27/r9-2/w27-1/r3-18/r22-1/r9-5/r0-32/r3-6/r32-8/r3-78/r3-113/r0-107/w28-0/r3-98/r3-91/w28-1/r3-93/w32-14/r32-0/r33-2/w32-16.0
+2.RCS.1000.r9-2/r3-60/r3-27/r32-16/r3-93/w4-16/r0-3/w3-122/r3-91/r0-58.0
+2.RCS.5780.r9-2/r33-2/r3-6/r22-31/r3-27/w0-103/w0-84/r3-60/r2-54.0
+1.RCS.14008.r0-58/r3-80/r22-31/r3-60/r3-27/r9-2/w32-18/r3-18/r22-1/r9-5/r3-98/r32-14/r3-93/r33-2/r3-6/r32-16/r3-122/r3-106/r3-78/r2-54/r0-107/w32-7/w32-11/w32-2/r33-0/r25-116/r0-84/r16-0/r28-3/r0-3/r0-111/r22-32/r10-58/r4-13/r4-3/r4-1.0
+1.RCS.6132.r3-80/r22-31/r3-60/r3-27/r9-2/w9-5/r0-48/r0-58/w32-19/r3-18/r22-1/r3-78/r3-88/r22-5/r22-14/r22-6/r22-29/r22-7/r22-12/r22-18/r22-28/r22-3/r22-0/r3-75/r3-104/w3-64/r0-88/r22-20/r25-99/r22-16/r21-0/r0-45/r22-21/r0-64/r3-74/r22-27/r0-42/w28-2/r0-107/w8-0/r0-52.0
+d.37446
+2.RCS.1900.r9-2/r28-2/r3-27/r22-31/w8-0/r2-47/r20-0/w22-23/r3-74/r3-95/r0-58/w22-10/r3-80.0
+d.18811 +1.RCS.1000.r3-74/w27-0/r0-58/r3-80/r22-31/r3-27/r9-2/w34-0/r3-18/r3-60/r21-0/r9-5/r3-78/r25-4/r3-104/r3-23/r30-0/r3-88/r22-27/r22-1/r25-45/r3-50/r22-12/r22-22/r22-3/r22-0/r25-56/r3-4/r22-15/r25-113/r3-7/r22-18/r25-60/r3-81/r25-21/r3-89/r18-5/r3-93/r17-8/r25-28/r25-87/r25-9/r25-13/r25-42/r25-90/r22-16/r34-2/r3-15/w3-64/r0-87/r25-99/r25-73/r3-6/r25-40/r3-90/r22-20/r3-110/r0-22/w32-17/w32-16/w32-3/w32-1/w33-2/r28-2/r3-98/r3-87/r3-108/r3-26/r3-96/r22-5/r22-14/r22-29/r3-68/r3-112/r3-103/r3-49/r3-12/r22-6/r3-85/r25-104/r22-28/r25-14/r25-44/r25-19/r3-67/r25-111/r18-4/r3-66/r18-17/r4-5/r25-68/r25-86/r25-26/r25-59/r25-71/r25-101/r25-67/r3-37/r25-0/r25-75/r25-20/r25-91/r22-7/r3-2/r3-117/r3-33/r22-2/r25-55/r25-66/r25-24/r25-105/r25-61/r25-11/r25-51/r25-64/r25-70/r25-81/r25-78/r25-37/r25-50/r25-102/r25-35/r18-18/r18-13/r18-12/r3-69/r3-19/r3-100/r18-19/r18-26/r18-21/r22-21/r25-82/r25-77/r25-33/r25-22/r3-29/r25-93/r22-10/r25-110/r25-62/r25-72/r22-9/r8-0/r18-23/r18-7/r18-11/r3-73/r25-92/r25-41/w33
-3/r0-107/w19-0/w31-0.0
+d.1865
+1.RCS.34308.r9-2/r3-18/r3-74/r22-3/r22-31/w31-0/r3-98/r19-0/r3-6/w33-2/r32-19/r3-110/r3-80/r3-114/r0-107/r0-58/w32-4/r3-78/w29-1/w32-8/r9-5/r32-17/r32-3/r32-1/r30-0/r3-104/r35-0/r34-1/r34-0/r3-92/r34-2/r22-12/r22-1/w32-0/r3-75/r3-118/r3-70/r3-1/r3-113/r22-29/r22-14/r3-109/r3-28/r3-79/r22-7/r3-8/r4-5/r3-85/r3-115/r3-55/r3-13/r3-59/r22-2.0
+d.1284
+2.RCS.35212.r9-2/r0-67/r3-110/w7-2/r3-74/r0-109/r0-115/w6-1/r0-35/w7-0/r0-19/w7-1/r0-58/r32-8/r3-6/r22-31/w0-2/r3-113/r9-5/w0-32.0
+1.RCS.5686.r0-58/r3-80/r22-31/r3-74/r3-110/r9-2/w27-1/r3-18/r22-1/r9-5/r0-32/r3-6/r32-8/r3-78/r3-113/r0-107/w28-0/r3-98/r3-91/w28-1/r3-31/r3-93/w32-14/r32-0/r33-2/w32-16.0
+2.RCS.1000.r9-2/r3-31/r3-110/r32-16/r3-93/w4-16/r0-3/w3-122/r3-91/r0-58.0
+2.RCS.5791.r9-2/r33-2/r3-6/r22-31/r3-110/w0-103/w0-92/r3-31/r2-54.0
+1.RCS.14049.r0-58/r3-80/r22-31/r3-31/r3-110/r9-2/w32-18/r3-18/r3-74/r22-1/r9-5/r3-98/r32-14/r3-93/r33-2/r3-6/r32-16/r3-122/r3-106/r3-78/r2-54/r0-107/w32-7/w32-11/w32-2/r33-0/r25-116/r0-92/r16-0/r28-3/r0-3/r0-111/r22-32/r10-58/r4-13/r4-3/r4-1.0
+1.RCS.6085.r3-80/r22-31/r3-31/r3-110/r9-2/w9-5/r0-76/r0-58/w32-19/r3-18/r3-74/r22-1/r3-78/r3-88/r22-5/r22-14/r22-6/r22-29/r22-7/r22-12/r22-18/r22-28/r22-3/r22-0/r3-75/r3-104/w3-64/r0-106/r22-20/r25-99/r22-16/r21-0/r0-71/r0-26/r22-27/r0-75/w28-2/r0-107/r3-0/w8-0/r0-29.0
+d.34977
+2.RCS.1794.r9-2/r28-2/r3-0/r22-31/w8-0/r2-9/r20-0/w22-23/r3-31/r3-95/r0-58/w22-10/r3-80.0
+d.11049 +1.RCS.32894.r3-31/w27-0/r0-58/r3-80/r22-31/r3-0/r9-2/w34-0/r3-18/r3-74/r21-0/r9-5/r3-78/r25-4/r3-104/r3-23/r30-0/r3-88/r22-27/r22-1/r25-45/r3-50/r22-12/r22-22/r22-3/r22-0/r25-56/r3-4/r22-15/r25-113/r3-7/r22-18/r25-60/r3-81/r25-21/r3-89/r18-5/r3-93/r17-8/r25-28/r25-87/r25-9/r25-13/r25-42/r25-90/r22-16/r34-2/r3-15/w3-64/r0-68/r25-99/r25-73/r3-6/r25-40/r3-90/r22-20/r0-48/w32-17/w32-16/w32-3/w32-1/w33-2/r28-2/r3-98/r3-87/r3-108/r3-26/r3-96/r22-5/r22-14/r22-29/r3-68/r3-112/r3-103/r3-49/r3-12/r22-6/r3-85/r25-104/r22-28/r25-14/r25-44/r25-19/r3-67/r18-17/r4-5/r18-4/r3-2/r3-117/r3-33/r22-2/r25-81/r25-78/r25-37/r25-50/r25-102/r25-35/r22-7/r18-18/r18-13/r18-12/r3-69/r3-19/r3-100/r18-19/r18-26/r18-21/r25-105/r25-61/r25-11/r25-75/r25-20/r25-91/r25-51/r25-64/r25-70/r25-55/r25-66/r25-24/r25-111/r3-66/r25-68/r25-86/r25-26/r22-21/r25-82/r25-77/r25-33/r22-10/r25-110/r25-62/r25-72/r22-9/r8-0/r18-23/r18-7/r18-11/r3-73/r25-92/r25-41/w33-3/r0-107/w19-0/w31-0/r32-19/r3-20/r3-114/w32-4/w29-1/w32-8/r35-0/r3
4-1/r3-92/w32-0.0
+d.5753
+2.RCS.33634.r9-2/r0-88/r3-0/w7-2/r3-31/r0-109/r3-20/r0-45/w6-1/r0-64/w7-0/r0-42/w7-1/r0-58/r32-8/r3-6/r22-31/w0-2/r3-113/r9-5/w0-32.0
+d.1325
+1.RCS.6527.r9-2/r3-18/r3-31/r3-20/r22-1/r22-31/r9-5/r3-78/w32-0/r3-98/r3-75/r3-104/r3-118/r3-6/r3-70/r30-0/r35-0/r34-1/r34-0/r32-19/r3-0/r3-1/r3-113/w33-2/r22-29/r22-14/r3-109/r3-28/r3-79/r22-7/r0-58/r3-80/w27-1/r0-32/r32-8/r0-107/w28-0/r3-91/w28-1/r3-93/w32-14/w32-16.0
+d.2261
+2.RCS.1000.r9-2/r3-20/r3-0/r32-16/r3-93/w4-16/r0-3/w3-122/r3-91/r0-58.0
+2.RCS.6788.r9-2/r33-2/r3-6/r22-31/r3-0/w0-103/w0-52/r3-20/r2-54.0
+1.RCS.13974.r0-58/r3-80/r22-31/r3-20/r3-0/r9-2/w32-18/r3-18/r3-31/r22-14/r9-5/r3-98/r32-14/r3-93/r33-2/r3-6/r32-16/r3-122/r3-106/r3-78/r2-54/r0-107/w32-7/w32-11/w32-2/r33-0/r25-116/r0-52/r16-0/r28-3/r0-3/r0-111/r22-32/r10-58/r4-13/r4-3/r4-1.0
+1.RCS.6098.r3-80/r22-31/r3-20/r3-0/r9-2/w9-5/r0-87/r0-58/w32-19/r3-18/r3-31/r22-14/r3-78/r3-42/r3-88/r22-5/r22-6/r22-29/r22-7/r22-1/r22-12/r22-18/r22-28/r22-3/r22-0/r3-75/r3-104/w3-64/r0-22/r22-20/r25-99/r22-16/r21-0/r0-53/r0-60/r0-97/w28-2/r0-107/w8-0/r0-98.0
+d.39039
+2.RCS.1753.r9-2/r28-2/r3-42/r22-31/w8-0/r2-47/r20-0/w22-23/r3-20/r3-95/r0-58/w22-10/r3-80.0
+d.14507 +1.RCS.31161.r3-20/w27-0/r0-58/r3-80/r22-31/r3-42/r9-2/w34-0/r3-18/r3-31/r21-0/r9-5/r3-78/r25-4/r3-104/r3-23/r30-0/r3-88/r22-27/r22-1/r25-45/r3-50/r22-12/r22-22/r22-3/r22-0/r25-56/r3-4/r22-15/r25-113/r3-7/r22-18/r25-60/r3-81/r25-21/r3-89/r18-5/r3-93/r17-8/r25-28/r25-87/r25-9/r25-13/r25-42/r25-90/r22-16/r34-2/r3-15/w3-64/r0-84/r25-40/r3-90/r22-20/r0-76/r3-41/w32-17/w32-16/w32-3/w32-1/w33-2/r28-2/r3-98/r3-87/r3-108/r3-26/r3-96/r22-29/r22-14/r3-68/r3-112/r3-103/r22-5/r3-49/r3-12/r22-6/r3-85/r25-14/r3-6/r25-104/r22-28/r25-44/r25-19/r3-67/r18-17/r4-5/r18-4/r25-111/r3-66/r25-81/r25-78/r25-37/r25-50/r25-102/r25-35/r22-7/r18-18/r18-13/r18-12/r3-69/r3-19/r3-100/r18-19/r18-26/r18-21/r3-2/r3-117/r3-33/r22-2/r25-51/r25-64/r25-70/r25-105/r25-61/r25-11/r25-75/r25-20/r25-91/r25-55/r25-66/r25-24/r25-68/r25-86/r25-26/r22-21/r25-82/r25-77/r25-33/r22-10/r25-110/r25-62/r25-72/r22-9/r8-0/r18-23/r18-7/r18-11/r3-73/r25-99/r25-73/r25-92/r25-41/w33-3/r0-107/w19-0/w31-0/r32-19/r3-114/w32-4/w29-1/w32-8/r35-0/r
34-1/r3-92.0
+d.3932
+2.RCS.31979.r9-2/r0-106/r3-42/w7-2/r3-20/r0-109/r3-41/r0-71/w6-1/r0-26/w7-0/r0-75/w7-1/r0-58/r32-8/r3-6/r22-31/w0-2/r3-113/r9-5/w0-32.0
+d.1568
+1.RCS.6364.r0-58/r3-80/r22-31/r3-41/r3-42/r9-2/w32-0/r3-18/r22-1/r9-5/r3-78/r3-98/r3-75/r3-104/r3-118/r3-6/r3-70/r30-0/r35-0/r34-1/r34-0/r32-19/r3-1/r3-113/w33-2/r22-29/r22-14/r3-109/r3-28/r3-79/r22-7/w27-1/r0-32/r32-8/r0-107/w28-0/r3-91/w28-1/r3-93/r3-27/w32-14/w32-16.0
+2.RCS.1000.r9-2/r3-41/r3-27/r32-16/r3-93/w4-16/r0-3/w3-122/r3-91/r0-58.0
+2.RCS.6617.r9-2/r33-2/r3-6/r22-31/r3-27/w0-103/w0-29/r3-41/r2-54.0
+1.RCS.13317.r0-58/r3-80/r22-31/r3-41/r3-27/r9-2/w32-18/r3-18/r22-14/r9-5/r3-98/r32-14/r3-93/r33-2/r3-6/r32-16/r3-122/r3-106/r3-78/r2-54/r0-107/w32-7/w32-11/w32-2/r33-0/r25-116/r0-29/r16-0/r28-3/r0-3/r0-111/r22-32/r10-58/r4-13/r4-3/r4-1.0
+1.RCS.6237.r3-80/r22-31/r3-41/r3-27/r9-2/w9-5/r0-68/r0-58/w32-19/r3-18/r22-14/r3-78/r3-88/r22-5/r22-6/r22-29/r22-7/r22-1/r22-12/r22-18/r22-28/r22-3/r22-0/r3-75/r3-104/w3-64/r0-48/r22-20/r25-99/r22-16/r21-0/r0-67/r3-47/r0-115/r0-35/w28-2/r0-107/w8-0/r0-19.0
+d.38304
+2.RCS.1901.r9-2/r28-2/r3-27/r22-31/w8-0/r2-9/r20-0/w22-23/r3-47/r3-95/r0-58/w22-10/r3-80.0
+d.20018 +1.RCS.31362.r3-47/w27-0/r0-58/r3-80/r22-31/r3-27/r9-2/w34-0/r3-18/r3-41/r21-0/r9-5/r3-78/r25-4/r3-104/r3-23/r30-0/r3-88/r22-27/r22-1/r25-45/r3-50/r22-12/r22-22/r22-3/r22-0/r25-56/r3-4/r22-15/r25-113/r3-7/r22-18/r25-60/r3-81/r25-21/r3-89/r18-5/r3-93/r17-8/r25-28/r25-87/r25-9/r25-13/r25-42/r25-90/r22-16/r34-2/r3-15/w3-64/r0-92/r25-40/r3-90/r22-20/r0-87/w32-17/w32-16/w32-3/w32-1/w33-2/r28-2/r3-98/r3-87/r3-108/r3-26/r3-96/r22-29/r22-14/r3-68/r3-112/r3-103/r22-5/r3-49/r3-12/r22-6/r3-85/r25-14/r3-6/r25-104/r22-28/r24-23/r25-44/r25-19/r3-67/r18-17/r4-5/r18-4/r25-111/r3-66/r25-81/r25-78/r25-37/r25-50/r25-102/r25-35/r22-7/r18-18/r18-13/r18-12/r3-69/r3-19/r3-100/r18-19/r18-26/r18-21/r3-2/r3-117/r3-33/r22-2/r25-51/r25-64/r25-70/r25-105/r25-61/r25-11/r25-75/r25-20/r25-91/r25-55/r25-66/r25-24/r25-68/r25-86/r25-26/r22-21/r25-82/r25-77/r25-33/r22-10/r25-110/r25-62/r25-72/r22-9/r8-0/r18-23/r18-7/r18-11/r3-73/r25-99/r25-73/r25-92/r25-41/w33-3/r0-107/w19-0/w31-0/r32-19/r3-114/w32-4/w29-1/w32-8/r35-0/
r34-1/r3-92.0
+d.1960
+2.RCS.32169.r9-2/r0-22/r3-27/w7-2/r3-47/r0-109/r0-53/w6-1/r0-60/w7-0/r0-97/w7-1/r0-58/r32-8/r3-6/r22-31/w0-2/r3-113/r9-5/w0-32.0
+1.RCS.6329.r0-58/r3-80/r22-31/r3-47/r3-27/r9-2/w32-0/r3-18/r22-1/r9-5/r3-78/r3-98/r3-75/r3-104/r3-118/r3-6/r3-70/r30-0/r35-0/r34-1/r34-0/r32-19/r3-1/r3-113/w33-2/r22-29/r22-14/r3-109/r3-28/r3-79/r22-7/w27-1/r0-32/r32-8/r0-107/w28-0/r3-91/r3-110/w28-1/r3-60/r3-93/w32-14/w32-16.0
+2.RCS.1000.r9-2/r3-60/r3-110/r32-16/r3-93/w4-16/r0-3/w3-122/r3-91/r0-58.0
+2.RCS.6597.r9-2/r33-2/r3-6/r22-31/r3-110/w0-103/w0-98/r3-60/r2-54.0
+1.RCS.13180.r0-58/r3-80/r22-31/r3-60/r3-110/r9-2/w32-18/r3-18/r3-47/r22-14/r9-5/r3-98/r32-14/r3-93/r33-2/r3-6/r32-16/r3-122/r3-106/r3-78/r2-54/r0-107/w32-7/w32-11/w32-2/r33-0/r25-116/r0-98/r16-0/r28-3/r0-3/r0-111/r22-32/r10-58/r4-13/r4-3/r4-1.0
+1.RCS.6083.r3-80/r22-31/r3-60/r3-110/r9-2/w9-5/r0-84/r0-58/w32-19/r3-18/r3-47/r22-14/r3-78/r3-88/r22-5/r22-6/r22-29/r22-7/r22-1/r22-12/r22-18/r22-28/r22-3/r22-0/r3-75/r3-104/w3-64/r0-76/r22-20/r9-4/r25-99/r22-16/r21-0/r0-88/r0-45/r0-64/w28-2/r0-107/w8-0/r0-42.0
+d.31011
+2.RCS.1752.r0-58/r9-4/r28-2/r3-110/r22-31/w8-0/r2-47/r20-0/w22-23/r3-60/r3-95/w22-10/r3-80.0
+d.8789 +1.RCS.1000.r3-60/w27-0/r0-58/r3-80/r22-31/r3-110/r9-4/w34-0/r3-18/r3-47/r21-0/r9-5/r3-78/r25-4/r3-104/r3-23/r30-0/r3-88/r22-27/r22-1/r25-45/r3-50/r22-12/r22-22/r22-3/r22-0/r25-56/r3-4/r22-15/r25-113/r3-7/r22-18/r25-60/r3-81/r25-21/r3-89/r18-5/r3-93/r17-8/r25-28/r25-87/r25-9/r25-13/r25-42/r25-90/r22-16/r34-2/r3-15/w3-64/r0-52/r25-99/r25-73/r3-6/r25-40/r3-90/r22-20/r0-68/w32-17/w32-16/w32-3/w32-1/w33-2/r28-2/r3-98/r3-85/r25-14/r3-12/r3-87/r3-108/r3-26/r3-96/r22-29/r22-14/r3-68/r3-112/r3-103/r22-5/r3-49/r22-6/r25-104/r22-28/r24-23/r25-44/r25-19/r3-67/r18-17/r4-5/r18-4/r25-111/r3-66/r25-81/r25-78/r25-37/r25-50/r25-102/r25-35/r22-7/r18-18/r18-13/r18-12/r3-69/r3-19/r3-100/r18-19/r18-26/r18-21/r3-2/r3-117/r3-33/r22-2/r25-51/r25-64/r25-70/r25-105/r25-61/r25-11/r25-75/r25-20/r25-91/r25-55/r25-66/r25-24/r25-68/r25-86/r25-26/r22-21/r25-82/r25-77/r25-33/r22-10/r25-110/r25-62/r25-72/r22-9/r8-0/r18-23/r18-7/r18-11/r3-73/r25-92/r25-41/w33-3/r0-107/w19-0/w31-0/r3-74/r32-19/r3-0/r3-114/w32-4/w29-1/w
32-8.0
+d.3470
+1.RCS.32141.r9-4/r3-60/r3-18/r3-74/r22-3/r22-31/r9-5/w32-8/r3-98/r32-17/r3-6/r32-3/r32-1/w33-2/r29-1/r30-0/r3-104/r35-0/r34-1/r34-0/r3-78/r3-80/r3-92/r0-107/r34-2/r22-12/r22-1/r0-58/r3-0/w32-0/r3-75/r3-118/r3-70/r32-19/r3-1/r3-113/r22-29/r22-14/r3-109/r3-28/r3-79/r22-7.0
+d.2303
+2.RCS.33018.r9-4/r0-48/r3-0/w7-2/r3-60/r0-109/r3-74/r0-67/w6-1/r0-115/w7-0/r0-35/w7-1/r0-58/r32-8/r3-6/r22-31/w0-2/r3-113/r9-5/w0-32.0
+1.RCS.5660.r0-58/r3-80/r22-31/r3-74/r3-0/r9-4/w27-1/r3-18/r3-60/r22-14/r9-5/r0-32/r3-6/r32-8/r3-78/r3-113/r0-107/w28-0/r3-98/r3-91/w28-1/r3-93/w32-14/r32-0/r33-2/w32-16.0
+d.1494
+2.RCS.1000.r9-4/r3-74/r3-0/r32-16/r3-93/w4-16/r0-3/w3-122/r3-91/r0-58.0
+2.RCS.5873.r9-4/r33-2/r3-6/r22-31/r3-0/w0-103/w0-19/r3-74/r2-54.0
+d.1151
+1.RCS.12578.r0-58/r3-80/r22-31/r3-74/r3-0/r9-4/w32-18/r3-18/r3-60/r22-14/r9-5/r3-98/r32-14/r3-93/r33-2/r3-6/r32-16/r3-122/r3-106/r3-78/r2-54/r0-107/w32-7/w32-11/w32-2/r33-0/r25-116/r0-19/r16-0/r28-3/r0-3/r0-111/r22-32/r10-58/r4-13/r4-3/r4-1.0
+1.RCS.6159.r3-80/r22-31/r3-74/r3-0/r9-4/w9-5/r0-92/r0-58/w32-19/r3-18/r3-60/r22-14/r3-78/r3-88/r22-5/r22-6/r22-29/r22-7/r22-1/r22-12/r22-18/r22-28/r22-3/r22-0/r3-75/r3-104/w3-64/r0-87/r22-20/r25-99/r22-16/r21-0/r0-106/r0-71/r0-26/w28-2/r0-107/w8-0/r0-75.0

Signature · 6 lines
2.20.1

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

Re: [Intel-gfx] [PATCH i-g-t 08/10] gem_wsim: Snippet of a workload extracted from carchase

Quoting Tvrtko Ursulin (2020-06-17 17:01:18)

Quoted 3 lines

+1.RCS.1000.r3-47/w27-0/r0-58/r3-80/r22-31/r3-42/r9-4/w34-0/r3-18/r3-41/r21-0/r9-5/r3-78/r25-4/r3-104/r3-23/r30-0/r3-88/r22-27/r22-1/r25-45/r3-50/r22-12/r22-22/r22-3/r22-0/r25-56/r3-4/r22-15/r25-113/r3-7/r22-18/r25-60/r3-81/r25-21/r3-89/r18-5/r3-93/r17-8/r25-28/r25-87/r25-9/r25-13/r25-42/r25-90/r22-16/r34-2/r3-15/w3-64/r0-67/r25-99/r25-73/r3-6/r25-40/r3-90/r22-20/r0-45/r3-110/w32-17/w32-16/w32-3/w32-1/w33-2/r28-2/r3-98/r3-85/r25-48/r3-12/r25-104/r24-23/r3-87/r3-108/r3-26/r3-96/r22-5/r22-14/r3-49/r3-103/r22-6/r3-68/r3-112/r22-29/r22-28/r25-14/r25-44/r25-19/r3-67/r25-111/r18-4/r3-66/r18-17/r4-5/r25-68/r25-86/r25-26/r25-67/r3-37/r25-0/r22-7/r25-59/r25-71/r25-101/r25-75/r25-20/r25-91/r3-2/r3-117/r3-33/r22-2/r25-55/r25-66/r25-24/r25-105/r25-61/r25-11/r25-51/r25-64/r25-70/r18-19/r18-26/r18-21/r25-81/r25-78/r25-37/r25-50/r25-102/r25-35/r18-18/r18-13/r18-12/r3-69/r3-19/r3-100/r22-21/r25-22/r3-29/r25-93/r18-2/r18-14/r18-3/r22-10/r18-23/r18-7/r18-11/r3-73/r8-0/r25-92/r25-41/w33-3/r0-1!
07/w19-0.
0

This patch has been mangled.
-Chris
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#28 lore raw

Re: [Intel-gfx] [PATCH i-g-t 08/10] gem_wsim: Snippet of a workload extracted from carchase

Quoted 6 lines · Chris Wilson

Quoting Tvrtko Ursulin (2020-06-17 17:01:18)

Quoted 3 lines

+1.RCS.1000.r3-47/w27-0/r0-58/r3-80/r22-31/r3-42/r9-4/w34-0/r3-18/r3-41/r21-0/r9-5/r3-78/r25-4/r3-104/r3-23/r30-0/r3-88/r22-27/r22-1/r25-45/r3-50/r22-12/r22-22/r22-3/r22-0/r25-56/r3-4/r22-15/r25-113/r3-7/r22-18/r25-60/r3-81/r25-21/r3-89/r18-5/r3-93/r17-8/r25-28/r25-87/r25-9/r25-13/r25-42/r25-90/r22-16/r34-2/r3-15/w3-64/r0-67/r25-99/r25-73/r3-6/r25-40/r3-90/r22-20/r0-45/r3-110/w32-17/w32-16/w32-3/w32-1/w33-2/r28-2/r3-98/r3-85/r25-48/r3-12/r25-104/r24-23/r3-87/r3-108/r3-26/r3-96/r22-5/r22-14/r3-49/r3-103/r22-6/r3-68/r3-112/r22-29/r22-28/r25-14/r25-44/r25-19/r3-67/r25-111/r18-4/r3-66/r18-17/r4-5/r25-68/r25-86/r25-26/r25-67/r3-37/r25-0/r22-7/r25-59/r25-71/r25-101/r25-75/r25-20/r25-91/r3-2/r3-117/r3-33/r22-2/r25-55/r25-66/r25-24/r25-105/r25-61/r25-11/r25-51/r25-64/r25-70/r18-19/r18-26/r18-21/r25-81/r25-78/r25-37/r25-50/r25-102/r25-35/r18-18/r18-13/r18-12/r3-69/r3-19/r3-100/r22-21/r25-22/r3-29/r25-93/r18-2/r18-14/r18-3/r22-10/r18-23/r18-7/r18-11/r3-73/r8-0/r25-92/r25-41/w33-3/r0-1!
07/w19-0.
0

This patch has been mangled.

Could it be your email client? (Very long lines in the patch.) I don't see corruption anywhere on my side, or on the copy I received from the mailing list.

Regards,

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

#29 lore raw

Re: [Intel-gfx] [PATCH i-g-t 08/10] gem_wsim: Snippet of a workload extracted from carchase

Quoting Tvrtko Ursulin (2020-06-18 08:53:47)

Quoted 13 lines · Chris Wilson
Quoted 2 lines

Quoting Tvrtko Ursulin (2020-06-17 17:01:18)

Quoted 1 line

+1.RCS.1000.r3-47/w27-0/r0-58/r3-80/r22-31/r3-42/r9-4/w34-0/r3-18/r3-41/r21-0/r9-5/r3-78/r25-4/r3-104/r3-23/r30-0/r3-88/r22-27/r22-1/r25-45/r3-50/r22-12/r22-22/r22-3/r22-0/r25-56/r3-4/r22-15/r25-113/r3-7/r22-18/r25-60/r3-81/r25-21/r3-89/r18-5/r3-93/r17-8/r25-28/r25-87/r25-9/r25-13/r25-42/r25-90/r22-16/r34-2/r3-15/w3-64/r0-67/r25-99/r25-73/r3-6/r25-40/r3-90/r22-20/r0-45/r3-110/w32-17/w32-16/w32-3/w32-1/w33-2/r28-2/r3-98/r3-85/r25-48/r3-12/r25-104/r24-23/r3-87/r3-108/r3-26/r3-96/r22-5/r22-14/r3-49/r3-103/r22-6/r3-68/r3-112/r22-29/r22-28/r25-14/r25-44/r25-19/r3-67/r25-111/r18-4/r3-66/r18-17/r4-5/r25-68/r25-86/r25-26/r25-67/r3-37/r25-0/r22-7/r25-59/r25-71/r25-101/r25-75/r25-20/r25-91/r3-2/r3-117/r3-33/r22-2/r25-55/r25-66/r25-24/r25-105/r25-61/r25-11/r25-51/r25-64/r25-70/r18-19/r18-26/r18-21/r25-81/r25-78/r25-37/r25-50/r25-102/r25-35/r18-18/r18-13/r18-12/r3-69/r3-19/r3-100/r22-21/r25-22/r3-29/r25-93/r18-2/r18-14/r18-3/r22-10/r18-23/r18-7/r18-11/r3-73/r8-0/r25-92/r25-41/w33-3/r!

0-1!

Quoted 4 lines
Quoted 2 lines

07/w19-0.
0

This patch has been mangled.

Could it be your email client? (Very long lines in the patch.) I don't see corruption anywhere on my side, or on the copy I received from the mailing list.

Somewhere in the chain, it's the same breakage in the file as well. It's not something I've seen before; and extremely strange choice for breaking the subsequent lines.

Anyway, it did lead me to suggest to add printing the _token for the invalid step; and perhaps change that from "step" to "line"? -Chris
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#30 lore raw

PATCH 9/10 gem_wsim: Implement device selection

In reply to Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

New command line options -L and -D <device> can respectively be used to list and select a GPU when more than one is present.

---
 benchmarks/gem_wsim.c | 62 +++++++++++++++++++++++++++++++------------
 1 file changed, 45 insertions(+), 17 deletions(-)
benchmarks/gem_wsim.c +45 -17
index 96ee923fb699..ca07b670bd42 100644
@@ -43,6 +43,7 @@
 #include <pthread.h>
 #include <math.h>
 
+#include "igt_device_scan.h"
 #include "intel_chipset.h"
 #include "intel_reg.h"
 #include "drm.h"
@@ -2593,7 +2594,9 @@ static void print_help(void)
 "  -S                Synchronize the sequence of random batch durations between\n"
 "                    clients.\n"
 "  -d                Sync between data dependencies in userspace.\n"
-"  -f <scale>        Scale factor for batch durations."
+"  -f <scale>        Scale factor for batch durations.\n"
+"  -L                List GPUs.\n"
+"  -D <gpu>          One of the GPUs from -L."
 	);
 }
 
@@ -2654,30 +2657,31 @@ int main(int argc, char **argv)
 	char *append_workload_arg = NULL;
 	struct w_arg *w_args = NULL;
 	unsigned int tolerance_pct = 1;
+	enum igt_devices_print_type printtype = IGT_PRINT_SIMPLE;
+	bool list_devices_arg = false;
 	int exitcode = EXIT_FAILURE;
+	struct igt_device_card card;
 	double scale_arg = 1.0f;
+	char *device_arg = NULL;
 	int prio = 0;
 	double t;
-	int i, c;
+	int i, c, ret;
 	char *subopts, *value;
 	int raw_number = 0;
 	long calib_val;
 	int eng;
 
-	/*
-	 * Open the device via the low-level API so we can do the GPU quiesce
-	 * manually as close as possible in time to the start of the workload.
-	 * This minimizes the gap in engine utilization tracking when observed
-	 * via external tools like trace.pl.
-	 */
-	fd = __drm_open_driver_render(DRIVER_INTEL);
-	igt_require(fd);
-
 	master_prng = time(NULL);
 
 	while ((c = getopt(argc, argv,
-			   "ThqvsSdc:n:r:w:W:a:t:p:I:f:")) != -1) {
+			   "LThqvsSdc:n:r:w:W:a:t:p:I:f:D:")) != -1) {
 		switch (c) {
+		case 'L':
+			list_devices_arg = true;
+			break;
+		case 'D':
+			device_arg = strdup(optarg);
+			break;
 		case 'W':
 			if (master_workload >= 0) {
 				wsim_err("Only one master workload can be given!\n");
@@ -2798,6 +2802,33 @@ int main(int argc, char **argv)
 		}
 	}
 
+
+	igt_devices_scan(false);
+
+	if (list_devices_arg) {
+		igt_devices_print(printtype);
+		return EXIT_SUCCESS;
+	}
+
+	if (device_arg) {
+		ret = igt_device_card_match(device_arg, &card);
+		if (!ret) {
+			wsim_err("Requested device %s not found!\n", device_arg);
+			free(device_arg);
+
+			return EXIT_FAILURE;
+		}
+		free(device_arg);
+	} else {
+		igt_device_find_first_i915_discrete_card(&card);
+	}
+
+	if (card.render[0])
+		fd = igt_open_render(&card);
+	else
+		fd = __drm_open_driver_render(DRIVER_INTEL);
+	igt_require(fd);
+
 	if (!has_nop_calibration) {
 		if (verbose > 1) {
 			printf("Calibrating nop delays with %u%% tolerance...\n",
@@ -2908,15 +2939,12 @@ int main(int argc, char **argv)
 	clock_gettime(CLOCK_MONOTONIC, &t_start);
 
 	for (i = 0; i < clients; i++) {
-		int ret;
-
 		ret = pthread_create(&w[i]->thread, NULL, run_workload, w[i]);
 		igt_assert_eq(ret, 0);
 	}
 
 	if (master_workload >= 0) {
-		int ret = pthread_join(w[master_workload]->thread, NULL);
-
+		ret = pthread_join(w[master_workload]->thread, NULL);
 		igt_assert(ret == 0);
 
 		for (i = 0; i < clients; i++)
@@ -2925,7 +2953,7 @@ int main(int argc, char **argv)
 
 	for (i = 0; i < clients; i++) {
 		if (master_workload != i) {
-			int ret = pthread_join(w[i]->thread, NULL);
+			ret = pthread_join(w[i]->thread, NULL);
 			igt_assert(ret == 0);
 		}
 	}
Signature · 6 lines
2.20.1

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

Re: [Intel-gfx] [PATCH i-g-t 09/10] gem_wsim: Implement device selection

Quoting Tvrtko Ursulin (2020-06-17 17:01:19)

Quoted 23 lines

From: Tvrtko Ursulin <[email protected]>

New command line options -L and -D <device> can respectively be used to list and select a GPU when more than one is present.

---
 benchmarks/gem_wsim.c | 62 +++++++++++++++++++++++++++++++------------
 1 file changed, 45 insertions(+), 17 deletions(-)
benchmarks/gem_wsim.c +1 -0
index 96ee923fb699..ca07b670bd42 100644
@@ -43,6 +43,7 @@
 #include <pthread.h>
 #include <math.h>
 
+#include "igt_device_scan.h"
 #include "intel_chipset.h"
 #include "intel_reg.h"
 #include "drm.h"

benchmarks/gem_wsim.c static void print_help(void) · line 2599

Quoted 7 lines
benchmarks/gem_wsim.c +3 -1
@@ -2593,7 +2594,9 @@ static void print_help(void)
 "  -S                Synchronize the sequence of random batch durations between\n"
 "                    clients.\n"
 "  -d                Sync between data dependencies in userspace.\n"
-"  -f <scale>        Scale factor for batch durations."
+"  -f <scale>        Scale factor for batch durations.\n"
+"  -L                List GPUs.\n"
+"  -D <gpu>          One of the GPUs from -L."

This is unlike you, I was expecting a range!
-Chris
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#32 lore raw

PATCH 10/10 gem_wsim: Fix calibration handling

In reply to Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

Intended use case was that run without arguments prints out the calibrations which can be simply copied and pasted to the -n argument and things should just work.

Two problems we need to solve: If the print out loops shows zero calibrations (engine not present) they are later rejected and also if some calibration is not given it is only an error if it needs to be used (engine present).

---
 benchmarks/gem_wsim.c | 20 ++------------------
 1 file changed, 2 insertions(+), 18 deletions(-)
benchmarks/gem_wsim.c +2 -18
index ca07b670bd42..3bbb8fcbe17c 100644
@@ -296,8 +296,8 @@ print_engine_calibrations(void)
 
 	printf("Nop calibration for %uus delay is: ", nop_calibration_us);
 	for (int i = 0; i < NUM_ENGINES; i++) {
-		/* skip DEFAULT and VCS engines */
-		if (i != DEFAULT && i != VCS) {
+		/* skip engines not present and DEFAULT and VCS */
+		if (i != DEFAULT && i != VCS && engine_calib_map[i]) {
 			if (first_entry) {
 				printf("%s=%lu", ring_str_map[i], engine_calib_map[i]);
 				first_entry = false;
@@ -2840,22 +2840,6 @@ int main(int argc, char **argv)
 		if (verbose)
 			print_engine_calibrations();
 		goto out;
-	} else {
-		bool missing = false;
-
-		for (i = 0; i < NUM_ENGINES; i++) {
-			if (i == VCS)
-				continue;
-
-			if (!engine_calib_map[i]) {
-				wsim_err("Missing calibration for '%s'!\n",
-					 ring_str_map[i]);
-				missing = true;
-			}
-		}
-
-		if (missing)
-			goto err;
 	}
 
 	if (!nr_w_args) {
Signature · 6 lines
2.20.1

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