From c5408fb22b9acec2b31a085a9b8c67bf01001fa6 Mon Sep 17 00:00:00 2001 From: Nick Curtis <6463881+skyreflectedinmirrors@users.noreply.github.com> Date: Thu, 14 Mar 2024 11:41:35 -0400 Subject: [PATCH] fixup MI300 MFMA rate disagreement between SOL and pipeline (#315) Signed-off-by: Nicholas Curtis Co-authored-by: Nicholas Curtis [ROCm/rocprofiler-compute commit: b610e1c27b9778854222d153bd45d6b503f68d18] --- .../gfx940/0200_system-speed-of-light.yaml | 4 ++-- .../gfx940/1100_compute-unit-compute-pipeline.yaml | 12 ++++++------ .../gfx941/0200_system-speed-of-light.yaml | 4 ++-- .../gfx941/1100_compute-unit-compute-pipeline.yaml | 12 ++++++------ .../gfx942/0200_system-speed-of-light.yaml | 4 ++-- .../gfx942/1100_compute-unit-compute-pipeline.yaml | 12 ++++++------ 6 files changed, 24 insertions(+), 24 deletions(-) diff --git a/projects/rocprofiler-compute/src/omniperf_soc/analysis_configs/gfx940/0200_system-speed-of-light.yaml b/projects/rocprofiler-compute/src/omniperf_soc/analysis_configs/gfx940/0200_system-speed-of-light.yaml index e14a52c602..7752ee879e 100644 --- a/projects/rocprofiler-compute/src/omniperf_soc/analysis_configs/gfx940/0200_system-speed-of-light.yaml +++ b/projects/rocprofiler-compute/src/omniperf_soc/analysis_configs/gfx940/0200_system-speed-of-light.yaml @@ -73,9 +73,9 @@ Panel Config: MFMA IOPs (Int8): value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp))) unit: GIOP - peak: ((($max_sclk * $cu_per_gpu) * 4096) / 1000) + peak: ((($max_sclk * $cu_per_gpu) * 8192) / 1000) pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp)))) - / ((($max_sclk * $cu_per_gpu) * 4096) / 1000)) + / ((($max_sclk * $cu_per_gpu) * 8192) / 1000)) tips: Active CUs: value: $numActiveCUs diff --git a/projects/rocprofiler-compute/src/omniperf_soc/analysis_configs/gfx940/1100_compute-unit-compute-pipeline.yaml b/projects/rocprofiler-compute/src/omniperf_soc/analysis_configs/gfx940/1100_compute-unit-compute-pipeline.yaml index bda6f5cc9e..4ff260eb6d 100644 --- a/projects/rocprofiler-compute/src/omniperf_soc/analysis_configs/gfx940/1100_compute-unit-compute-pipeline.yaml +++ b/projects/rocprofiler-compute/src/omniperf_soc/analysis_configs/gfx940/1100_compute-unit-compute-pipeline.yaml @@ -44,16 +44,16 @@ Panel Config: MFMA FLOPs (BF16): value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_BF16 * 512) / (End_Timestamp - Start_Timestamp))) unit: GFLOP - peak: ((($max_sclk * $cu_per_gpu) * 1024) / 1000) + peak: ((($max_sclk * $cu_per_gpu) * 4096) / 1000) pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_BF16 * 512) / (End_Timestamp - Start_Timestamp)))) - / ((($max_sclk * $cu_per_gpu) * 1024) / 1000)) + / ((($max_sclk * $cu_per_gpu) * 4096) / 1000)) tips: MFMA FLOPs (F16): value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_F16 * 512) / (End_Timestamp - Start_Timestamp))) unit: GFLOP - peak: ((($max_sclk * $cu_per_gpu) * 1024) / 1000) + peak: ((($max_sclk * $cu_per_gpu) * 4096) / 1000) pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_F16 * 512) / (End_Timestamp - Start_Timestamp)))) - / ((($max_sclk * $cu_per_gpu) * 1024) / 1000)) + / ((($max_sclk * $cu_per_gpu) * 4096) / 1000)) tips: MFMA FLOPs (F32): value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_F32 * 512) / (End_Timestamp - Start_Timestamp))) @@ -72,9 +72,9 @@ Panel Config: MFMA IOPs (INT8): value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp))) unit: GIOP - peak: ((($max_sclk * $cu_per_gpu) * 1024) / 1000) + peak: ((($max_sclk * $cu_per_gpu) * 8192) / 1000) pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp)))) - / ((($max_sclk * $cu_per_gpu) * 1024) / 1000)) + / ((($max_sclk * $cu_per_gpu) * 8192) / 1000)) tips: - metric_table: diff --git a/projects/rocprofiler-compute/src/omniperf_soc/analysis_configs/gfx941/0200_system-speed-of-light.yaml b/projects/rocprofiler-compute/src/omniperf_soc/analysis_configs/gfx941/0200_system-speed-of-light.yaml index e14a52c602..7752ee879e 100644 --- a/projects/rocprofiler-compute/src/omniperf_soc/analysis_configs/gfx941/0200_system-speed-of-light.yaml +++ b/projects/rocprofiler-compute/src/omniperf_soc/analysis_configs/gfx941/0200_system-speed-of-light.yaml @@ -73,9 +73,9 @@ Panel Config: MFMA IOPs (Int8): value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp))) unit: GIOP - peak: ((($max_sclk * $cu_per_gpu) * 4096) / 1000) + peak: ((($max_sclk * $cu_per_gpu) * 8192) / 1000) pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp)))) - / ((($max_sclk * $cu_per_gpu) * 4096) / 1000)) + / ((($max_sclk * $cu_per_gpu) * 8192) / 1000)) tips: Active CUs: value: $numActiveCUs diff --git a/projects/rocprofiler-compute/src/omniperf_soc/analysis_configs/gfx941/1100_compute-unit-compute-pipeline.yaml b/projects/rocprofiler-compute/src/omniperf_soc/analysis_configs/gfx941/1100_compute-unit-compute-pipeline.yaml index bda6f5cc9e..4ff260eb6d 100644 --- a/projects/rocprofiler-compute/src/omniperf_soc/analysis_configs/gfx941/1100_compute-unit-compute-pipeline.yaml +++ b/projects/rocprofiler-compute/src/omniperf_soc/analysis_configs/gfx941/1100_compute-unit-compute-pipeline.yaml @@ -44,16 +44,16 @@ Panel Config: MFMA FLOPs (BF16): value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_BF16 * 512) / (End_Timestamp - Start_Timestamp))) unit: GFLOP - peak: ((($max_sclk * $cu_per_gpu) * 1024) / 1000) + peak: ((($max_sclk * $cu_per_gpu) * 4096) / 1000) pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_BF16 * 512) / (End_Timestamp - Start_Timestamp)))) - / ((($max_sclk * $cu_per_gpu) * 1024) / 1000)) + / ((($max_sclk * $cu_per_gpu) * 4096) / 1000)) tips: MFMA FLOPs (F16): value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_F16 * 512) / (End_Timestamp - Start_Timestamp))) unit: GFLOP - peak: ((($max_sclk * $cu_per_gpu) * 1024) / 1000) + peak: ((($max_sclk * $cu_per_gpu) * 4096) / 1000) pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_F16 * 512) / (End_Timestamp - Start_Timestamp)))) - / ((($max_sclk * $cu_per_gpu) * 1024) / 1000)) + / ((($max_sclk * $cu_per_gpu) * 4096) / 1000)) tips: MFMA FLOPs (F32): value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_F32 * 512) / (End_Timestamp - Start_Timestamp))) @@ -72,9 +72,9 @@ Panel Config: MFMA IOPs (INT8): value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp))) unit: GIOP - peak: ((($max_sclk * $cu_per_gpu) * 1024) / 1000) + peak: ((($max_sclk * $cu_per_gpu) * 8192) / 1000) pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp)))) - / ((($max_sclk * $cu_per_gpu) * 1024) / 1000)) + / ((($max_sclk * $cu_per_gpu) * 8192) / 1000)) tips: - metric_table: diff --git a/projects/rocprofiler-compute/src/omniperf_soc/analysis_configs/gfx942/0200_system-speed-of-light.yaml b/projects/rocprofiler-compute/src/omniperf_soc/analysis_configs/gfx942/0200_system-speed-of-light.yaml index fd80a20262..9909e68a71 100644 --- a/projects/rocprofiler-compute/src/omniperf_soc/analysis_configs/gfx942/0200_system-speed-of-light.yaml +++ b/projects/rocprofiler-compute/src/omniperf_soc/analysis_configs/gfx942/0200_system-speed-of-light.yaml @@ -73,9 +73,9 @@ Panel Config: MFMA IOPs (Int8): value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp))) unit: GIOP - peak: ((($max_sclk * $cu_per_gpu) * 4096) / 1000) + peak: ((($max_sclk * $cu_per_gpu) * 8192) / 1000) pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp)))) - / ((($max_sclk * $cu_per_gpu) * 4096) / 1000)) + / ((($max_sclk * $cu_per_gpu) * 8192) / 1000)) tips: Active CUs: value: $numActiveCUs diff --git a/projects/rocprofiler-compute/src/omniperf_soc/analysis_configs/gfx942/1100_compute-unit-compute-pipeline.yaml b/projects/rocprofiler-compute/src/omniperf_soc/analysis_configs/gfx942/1100_compute-unit-compute-pipeline.yaml index bda6f5cc9e..4ff260eb6d 100644 --- a/projects/rocprofiler-compute/src/omniperf_soc/analysis_configs/gfx942/1100_compute-unit-compute-pipeline.yaml +++ b/projects/rocprofiler-compute/src/omniperf_soc/analysis_configs/gfx942/1100_compute-unit-compute-pipeline.yaml @@ -44,16 +44,16 @@ Panel Config: MFMA FLOPs (BF16): value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_BF16 * 512) / (End_Timestamp - Start_Timestamp))) unit: GFLOP - peak: ((($max_sclk * $cu_per_gpu) * 1024) / 1000) + peak: ((($max_sclk * $cu_per_gpu) * 4096) / 1000) pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_BF16 * 512) / (End_Timestamp - Start_Timestamp)))) - / ((($max_sclk * $cu_per_gpu) * 1024) / 1000)) + / ((($max_sclk * $cu_per_gpu) * 4096) / 1000)) tips: MFMA FLOPs (F16): value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_F16 * 512) / (End_Timestamp - Start_Timestamp))) unit: GFLOP - peak: ((($max_sclk * $cu_per_gpu) * 1024) / 1000) + peak: ((($max_sclk * $cu_per_gpu) * 4096) / 1000) pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_F16 * 512) / (End_Timestamp - Start_Timestamp)))) - / ((($max_sclk * $cu_per_gpu) * 1024) / 1000)) + / ((($max_sclk * $cu_per_gpu) * 4096) / 1000)) tips: MFMA FLOPs (F32): value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_F32 * 512) / (End_Timestamp - Start_Timestamp))) @@ -72,9 +72,9 @@ Panel Config: MFMA IOPs (INT8): value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp))) unit: GIOP - peak: ((($max_sclk * $cu_per_gpu) * 1024) / 1000) + peak: ((($max_sclk * $cu_per_gpu) * 8192) / 1000) pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp)))) - / ((($max_sclk * $cu_per_gpu) * 1024) / 1000)) + / ((($max_sclk * $cu_per_gpu) * 8192) / 1000)) tips: - metric_table: