From 3ce140d04bdc09bcb4572bcba68a899d65e60533 Mon Sep 17 00:00:00 2001 From: colramos-amd Date: Wed, 21 Feb 2024 16:47:32 -0600 Subject: [PATCH] Including remaining metric fixes from #251 Co-authored-by: Nick Curtis Signed-off-by: colramos-amd --- .../gfx940/0200_system-speed-of-light.yaml | 8 ++--- .../1000_compute-unit-instruction-mix.yaml | 10 ++++-- .../gfx940/1600_L1_cache.yaml | 25 ++++++++------- .../gfx941/0200_system-speed-of-light.yaml | 8 ++--- .../1000_compute-unit-instruction-mix.yaml | 10 ++++-- .../gfx941/1600_L1_cache.yaml | 25 ++++++++------- .../gfx942/0200_system-speed-of-light.yaml | 18 ++++++----- .../1000_compute-unit-instruction-mix.yaml | 10 ++++-- .../gfx942/1600_L1_cache.yaml | 25 ++++++++------- .../gfx942/1700_L2_cache.yaml | 32 ++++++++++++------- .../profile_configs/gfx940/pmc_tcc_perf.txt | 3 +- src/utils/parser.py | 4 +-- 12 files changed, 106 insertions(+), 72 deletions(-) diff --git a/src/omniperf_soc/analysis_configs/gfx940/0200_system-speed-of-light.yaml b/src/omniperf_soc/analysis_configs/gfx940/0200_system-speed-of-light.yaml index 1e091e84fc..019aa968f1 100644 --- a/src/omniperf_soc/analysis_configs/gfx940/0200_system-speed-of-light.yaml +++ b/src/omniperf_soc/analysis_configs/gfx940/0200_system-speed-of-light.yaml @@ -166,11 +166,11 @@ Panel Config: None)) tips: vL1D Cache BW: - value: AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 64) / (End_Timestamp - Start_Timestamp))) + value: AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 128) / (End_Timestamp - Start_Timestamp))) unit: GB/s - peak: ((($max_sclk / 1000) * 64) * $cu_per_gpu) - pop: ((100 * AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 64) / (End_Timestamp - Start_Timestamp)))) - / ((($max_sclk / 1000) * 64) * $cu_per_gpu)) + peak: ((($max_sclk / 1000) * 128) * $cu_per_gpu) + pop: ((100 * AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 128) / (End_Timestamp - Start_Timestamp)))) + / ((($max_sclk / 1000) * 128) * $cu_per_gpu)) tips: L2 Cache Hit Rate: value: AVG((((100 * TCC_HIT_sum) / (TCC_HIT_sum + TCC_MISS_sum)) if ((TCC_HIT_sum diff --git a/src/omniperf_soc/analysis_configs/gfx940/1000_compute-unit-instruction-mix.yaml b/src/omniperf_soc/analysis_configs/gfx940/1000_compute-unit-instruction-mix.yaml index 64769ecd81..a988498249 100644 --- a/src/omniperf_soc/analysis_configs/gfx940/1000_compute-unit-instruction-mix.yaml +++ b/src/omniperf_soc/analysis_configs/gfx940/1000_compute-unit-instruction-mix.yaml @@ -26,12 +26,16 @@ Panel Config: unit: (instr + $normUnit) tips: VMEM: - avg: AVG(((SQ_INSTS_VMEM - SQ_INSTS_FLAT_LDS_ONLY) / $denom)) - min: MIN(((SQ_INSTS_VMEM - SQ_INSTS_FLAT_LDS_ONLY) / $denom)) - max: MAX(((SQ_INSTS_VMEM - SQ_INSTS_FLAT_LDS_ONLY) / $denom)) + # TODO: need to fix this when the new FLAT/LDS counts + # are present in ROCm + avg: AVG(((SQ_INSTS_VMEM) / $denom)) + min: MIN(((SQ_INSTS_VMEM) / $denom)) + max: MAX(((SQ_INSTS_VMEM) / $denom)) unit: (instr + $normUnit) tips: LDS: + # TODO: need to fix this when the new FLAT/LDS counts + # are present in ROCm avg: AVG((SQ_INSTS_LDS / $denom)) min: MIN((SQ_INSTS_LDS / $denom)) max: MAX((SQ_INSTS_LDS / $denom)) diff --git a/src/omniperf_soc/analysis_configs/gfx940/1600_L1_cache.yaml b/src/omniperf_soc/analysis_configs/gfx940/1600_L1_cache.yaml index 7453647314..a7de6d0e52 100644 --- a/src/omniperf_soc/analysis_configs/gfx940/1600_L1_cache.yaml +++ b/src/omniperf_soc/analysis_configs/gfx940/1600_L1_cache.yaml @@ -25,8 +25,8 @@ Panel Config: unit: Pct of Peak tips: Bandwidth: - value: ((100 * AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 64) / (End_Timestamp - Start_Timestamp)))) - / ((($max_sclk / 1000) * 64) * $cu_per_gpu)) + value: ((100 * AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 128) / (End_Timestamp - Start_Timestamp)))) + / ((($max_sclk / 1000) * 128) * $cu_per_gpu)) unit: Pct of Peak tips: Utilization: @@ -116,9 +116,9 @@ Panel Config: unit: (Req + $normUnit) tips: Cache BW: - avg: AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 64) / $denom)) - min: MIN(((TCP_TOTAL_CACHE_ACCESSES_sum * 64) / $denom)) - max: MAX(((TCP_TOTAL_CACHE_ACCESSES_sum * 64) / $denom)) + avg: AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 128) / $denom)) + min: MIN(((TCP_TOTAL_CACHE_ACCESSES_sum * 128) / $denom)) + max: MAX(((TCP_TOTAL_CACHE_ACCESSES_sum * 128) / $denom)) unit: (Bytes + $normUnit) tips: Cache Hit Rate: @@ -161,12 +161,15 @@ Panel Config: unit: (Req + $normUnit) tips: L1-L2 BW: - avg: AVG(((64 * (((TCP_TCC_READ_REQ_sum + TCP_TCC_WRITE_REQ_sum) + TCP_TCC_ATOMIC_WITH_RET_REQ_sum) - + TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / $denom)) - min: MIN(((64 * (((TCP_TCC_READ_REQ_sum + TCP_TCC_WRITE_REQ_sum) + TCP_TCC_ATOMIC_WITH_RET_REQ_sum) - + TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / $denom)) - max: MAX(((64 * (((TCP_TCC_READ_REQ_sum + TCP_TCC_WRITE_REQ_sum) + TCP_TCC_ATOMIC_WITH_RET_REQ_sum) - + TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / $denom)) + avg: AVG(((128 * TCP_TCC_READ_REQ_sum + 64 * + (TCP_TCC_WRITE_REQ_sum + TCP_TCC_ATOMIC_WITH_RET_REQ_sum + + TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / $denom)) + min: MIN(((128 * TCP_TCC_READ_REQ_sum + 64 * + (TCP_TCC_WRITE_REQ_sum + TCP_TCC_ATOMIC_WITH_RET_REQ_sum + + TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / $denom)) + max: MAX(((128 * TCP_TCC_READ_REQ_sum + 64 * + (TCP_TCC_WRITE_REQ_sum + TCP_TCC_ATOMIC_WITH_RET_REQ_sum + + TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / $denom)) unit: (Bytes + $normUnit) tips: L1-L2 Read: diff --git a/src/omniperf_soc/analysis_configs/gfx941/0200_system-speed-of-light.yaml b/src/omniperf_soc/analysis_configs/gfx941/0200_system-speed-of-light.yaml index 1e091e84fc..019aa968f1 100644 --- a/src/omniperf_soc/analysis_configs/gfx941/0200_system-speed-of-light.yaml +++ b/src/omniperf_soc/analysis_configs/gfx941/0200_system-speed-of-light.yaml @@ -166,11 +166,11 @@ Panel Config: None)) tips: vL1D Cache BW: - value: AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 64) / (End_Timestamp - Start_Timestamp))) + value: AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 128) / (End_Timestamp - Start_Timestamp))) unit: GB/s - peak: ((($max_sclk / 1000) * 64) * $cu_per_gpu) - pop: ((100 * AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 64) / (End_Timestamp - Start_Timestamp)))) - / ((($max_sclk / 1000) * 64) * $cu_per_gpu)) + peak: ((($max_sclk / 1000) * 128) * $cu_per_gpu) + pop: ((100 * AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 128) / (End_Timestamp - Start_Timestamp)))) + / ((($max_sclk / 1000) * 128) * $cu_per_gpu)) tips: L2 Cache Hit Rate: value: AVG((((100 * TCC_HIT_sum) / (TCC_HIT_sum + TCC_MISS_sum)) if ((TCC_HIT_sum diff --git a/src/omniperf_soc/analysis_configs/gfx941/1000_compute-unit-instruction-mix.yaml b/src/omniperf_soc/analysis_configs/gfx941/1000_compute-unit-instruction-mix.yaml index 64769ecd81..a988498249 100644 --- a/src/omniperf_soc/analysis_configs/gfx941/1000_compute-unit-instruction-mix.yaml +++ b/src/omniperf_soc/analysis_configs/gfx941/1000_compute-unit-instruction-mix.yaml @@ -26,12 +26,16 @@ Panel Config: unit: (instr + $normUnit) tips: VMEM: - avg: AVG(((SQ_INSTS_VMEM - SQ_INSTS_FLAT_LDS_ONLY) / $denom)) - min: MIN(((SQ_INSTS_VMEM - SQ_INSTS_FLAT_LDS_ONLY) / $denom)) - max: MAX(((SQ_INSTS_VMEM - SQ_INSTS_FLAT_LDS_ONLY) / $denom)) + # TODO: need to fix this when the new FLAT/LDS counts + # are present in ROCm + avg: AVG(((SQ_INSTS_VMEM) / $denom)) + min: MIN(((SQ_INSTS_VMEM) / $denom)) + max: MAX(((SQ_INSTS_VMEM) / $denom)) unit: (instr + $normUnit) tips: LDS: + # TODO: need to fix this when the new FLAT/LDS counts + # are present in ROCm avg: AVG((SQ_INSTS_LDS / $denom)) min: MIN((SQ_INSTS_LDS / $denom)) max: MAX((SQ_INSTS_LDS / $denom)) diff --git a/src/omniperf_soc/analysis_configs/gfx941/1600_L1_cache.yaml b/src/omniperf_soc/analysis_configs/gfx941/1600_L1_cache.yaml index 7453647314..a7de6d0e52 100644 --- a/src/omniperf_soc/analysis_configs/gfx941/1600_L1_cache.yaml +++ b/src/omniperf_soc/analysis_configs/gfx941/1600_L1_cache.yaml @@ -25,8 +25,8 @@ Panel Config: unit: Pct of Peak tips: Bandwidth: - value: ((100 * AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 64) / (End_Timestamp - Start_Timestamp)))) - / ((($max_sclk / 1000) * 64) * $cu_per_gpu)) + value: ((100 * AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 128) / (End_Timestamp - Start_Timestamp)))) + / ((($max_sclk / 1000) * 128) * $cu_per_gpu)) unit: Pct of Peak tips: Utilization: @@ -116,9 +116,9 @@ Panel Config: unit: (Req + $normUnit) tips: Cache BW: - avg: AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 64) / $denom)) - min: MIN(((TCP_TOTAL_CACHE_ACCESSES_sum * 64) / $denom)) - max: MAX(((TCP_TOTAL_CACHE_ACCESSES_sum * 64) / $denom)) + avg: AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 128) / $denom)) + min: MIN(((TCP_TOTAL_CACHE_ACCESSES_sum * 128) / $denom)) + max: MAX(((TCP_TOTAL_CACHE_ACCESSES_sum * 128) / $denom)) unit: (Bytes + $normUnit) tips: Cache Hit Rate: @@ -161,12 +161,15 @@ Panel Config: unit: (Req + $normUnit) tips: L1-L2 BW: - avg: AVG(((64 * (((TCP_TCC_READ_REQ_sum + TCP_TCC_WRITE_REQ_sum) + TCP_TCC_ATOMIC_WITH_RET_REQ_sum) - + TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / $denom)) - min: MIN(((64 * (((TCP_TCC_READ_REQ_sum + TCP_TCC_WRITE_REQ_sum) + TCP_TCC_ATOMIC_WITH_RET_REQ_sum) - + TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / $denom)) - max: MAX(((64 * (((TCP_TCC_READ_REQ_sum + TCP_TCC_WRITE_REQ_sum) + TCP_TCC_ATOMIC_WITH_RET_REQ_sum) - + TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / $denom)) + avg: AVG(((128 * TCP_TCC_READ_REQ_sum + 64 * + (TCP_TCC_WRITE_REQ_sum + TCP_TCC_ATOMIC_WITH_RET_REQ_sum + + TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / $denom)) + min: MIN(((128 * TCP_TCC_READ_REQ_sum + 64 * + (TCP_TCC_WRITE_REQ_sum + TCP_TCC_ATOMIC_WITH_RET_REQ_sum + + TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / $denom)) + max: MAX(((128 * TCP_TCC_READ_REQ_sum + 64 * + (TCP_TCC_WRITE_REQ_sum + TCP_TCC_ATOMIC_WITH_RET_REQ_sum + + TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / $denom)) unit: (Bytes + $normUnit) tips: L1-L2 Read: diff --git a/src/omniperf_soc/analysis_configs/gfx942/0200_system-speed-of-light.yaml b/src/omniperf_soc/analysis_configs/gfx942/0200_system-speed-of-light.yaml index 1e091e84fc..b28bc6d9f9 100644 --- a/src/omniperf_soc/analysis_configs/gfx942/0200_system-speed-of-light.yaml +++ b/src/omniperf_soc/analysis_configs/gfx942/0200_system-speed-of-light.yaml @@ -166,11 +166,11 @@ Panel Config: None)) tips: vL1D Cache BW: - value: AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 64) / (End_Timestamp - Start_Timestamp))) + value: AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 128) / (End_Timestamp - Start_Timestamp))) unit: GB/s - peak: ((($max_sclk / 1000) * 64) * $cu_per_gpu) - pop: ((100 * AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 64) / (End_Timestamp - Start_Timestamp)))) - / ((($max_sclk / 1000) * 64) * $cu_per_gpu)) + peak: ((($max_sclk / 1000) * 128) * $cu_per_gpu) + pop: ((100 * AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 128) / (End_Timestamp - Start_Timestamp)))) + / ((($max_sclk / 1000) * 128) * $cu_per_gpu)) tips: L2 Cache Hit Rate: value: AVG((((100 * TCC_HIT_sum) / (TCC_HIT_sum + TCC_MISS_sum)) if ((TCC_HIT_sum @@ -188,12 +188,14 @@ Panel Config: / ((($max_sclk / 1000) * 128) * TO_INT($total_l2_chan))) tips: L2-Fabric Read BW: - value: AVG((((TCC_EA0_RDREQ_32B_sum * 32) + ((TCC_EA0_RDREQ_sum - TCC_EA0_RDREQ_32B_sum) - * 64)) / (End_Timestamp - Start_Timestamp))) + value: AVG((128 * TCC_BUBBLE_sum + + 64 * (TCC_EA0_RDREQ_sum - TCC_BUBBLE_sum - TCC_EA0_RDREQ_32B_sum) + + 32 * TCC_EA0_RDREQ_32B_sum) / (End_Timestamp - Start_Timestamp)) unit: GB/s peak: $hbm_bw - pop: ((100 * AVG((((TCC_EA0_RDREQ_32B_sum * 32) + ((TCC_EA0_RDREQ_sum - TCC_EA0_RDREQ_32B_sum) - * 64)) / (End_Timestamp - Start_Timestamp)))) / $hbm_bw) + pop: ((100 * (AVG((128 * TCC_BUBBLE_sum + + 64 * (TCC_EA0_RDREQ_sum - TCC_BUBBLE_sum - TCC_EA0_RDREQ_32B_sum) + + 32 * TCC_EA0_RDREQ_32B_sum) / (End_Timestamp - Start_Timestamp)))) / $hbmBW) tips: L2-Fabric Write BW: value: AVG((((TCC_EA0_WRREQ_64B_sum * 64) + ((TCC_EA0_WRREQ_sum - TCC_EA0_WRREQ_64B_sum) diff --git a/src/omniperf_soc/analysis_configs/gfx942/1000_compute-unit-instruction-mix.yaml b/src/omniperf_soc/analysis_configs/gfx942/1000_compute-unit-instruction-mix.yaml index 64769ecd81..a988498249 100644 --- a/src/omniperf_soc/analysis_configs/gfx942/1000_compute-unit-instruction-mix.yaml +++ b/src/omniperf_soc/analysis_configs/gfx942/1000_compute-unit-instruction-mix.yaml @@ -26,12 +26,16 @@ Panel Config: unit: (instr + $normUnit) tips: VMEM: - avg: AVG(((SQ_INSTS_VMEM - SQ_INSTS_FLAT_LDS_ONLY) / $denom)) - min: MIN(((SQ_INSTS_VMEM - SQ_INSTS_FLAT_LDS_ONLY) / $denom)) - max: MAX(((SQ_INSTS_VMEM - SQ_INSTS_FLAT_LDS_ONLY) / $denom)) + # TODO: need to fix this when the new FLAT/LDS counts + # are present in ROCm + avg: AVG(((SQ_INSTS_VMEM) / $denom)) + min: MIN(((SQ_INSTS_VMEM) / $denom)) + max: MAX(((SQ_INSTS_VMEM) / $denom)) unit: (instr + $normUnit) tips: LDS: + # TODO: need to fix this when the new FLAT/LDS counts + # are present in ROCm avg: AVG((SQ_INSTS_LDS / $denom)) min: MIN((SQ_INSTS_LDS / $denom)) max: MAX((SQ_INSTS_LDS / $denom)) diff --git a/src/omniperf_soc/analysis_configs/gfx942/1600_L1_cache.yaml b/src/omniperf_soc/analysis_configs/gfx942/1600_L1_cache.yaml index 7453647314..a7de6d0e52 100644 --- a/src/omniperf_soc/analysis_configs/gfx942/1600_L1_cache.yaml +++ b/src/omniperf_soc/analysis_configs/gfx942/1600_L1_cache.yaml @@ -25,8 +25,8 @@ Panel Config: unit: Pct of Peak tips: Bandwidth: - value: ((100 * AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 64) / (End_Timestamp - Start_Timestamp)))) - / ((($max_sclk / 1000) * 64) * $cu_per_gpu)) + value: ((100 * AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 128) / (End_Timestamp - Start_Timestamp)))) + / ((($max_sclk / 1000) * 128) * $cu_per_gpu)) unit: Pct of Peak tips: Utilization: @@ -116,9 +116,9 @@ Panel Config: unit: (Req + $normUnit) tips: Cache BW: - avg: AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 64) / $denom)) - min: MIN(((TCP_TOTAL_CACHE_ACCESSES_sum * 64) / $denom)) - max: MAX(((TCP_TOTAL_CACHE_ACCESSES_sum * 64) / $denom)) + avg: AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 128) / $denom)) + min: MIN(((TCP_TOTAL_CACHE_ACCESSES_sum * 128) / $denom)) + max: MAX(((TCP_TOTAL_CACHE_ACCESSES_sum * 128) / $denom)) unit: (Bytes + $normUnit) tips: Cache Hit Rate: @@ -161,12 +161,15 @@ Panel Config: unit: (Req + $normUnit) tips: L1-L2 BW: - avg: AVG(((64 * (((TCP_TCC_READ_REQ_sum + TCP_TCC_WRITE_REQ_sum) + TCP_TCC_ATOMIC_WITH_RET_REQ_sum) - + TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / $denom)) - min: MIN(((64 * (((TCP_TCC_READ_REQ_sum + TCP_TCC_WRITE_REQ_sum) + TCP_TCC_ATOMIC_WITH_RET_REQ_sum) - + TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / $denom)) - max: MAX(((64 * (((TCP_TCC_READ_REQ_sum + TCP_TCC_WRITE_REQ_sum) + TCP_TCC_ATOMIC_WITH_RET_REQ_sum) - + TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / $denom)) + avg: AVG(((128 * TCP_TCC_READ_REQ_sum + 64 * + (TCP_TCC_WRITE_REQ_sum + TCP_TCC_ATOMIC_WITH_RET_REQ_sum + + TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / $denom)) + min: MIN(((128 * TCP_TCC_READ_REQ_sum + 64 * + (TCP_TCC_WRITE_REQ_sum + TCP_TCC_ATOMIC_WITH_RET_REQ_sum + + TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / $denom)) + max: MAX(((128 * TCP_TCC_READ_REQ_sum + 64 * + (TCP_TCC_WRITE_REQ_sum + TCP_TCC_ATOMIC_WITH_RET_REQ_sum + + TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / $denom)) unit: (Bytes + $normUnit) tips: L1-L2 Read: diff --git a/src/omniperf_soc/analysis_configs/gfx942/1700_L2_cache.yaml b/src/omniperf_soc/analysis_configs/gfx942/1700_L2_cache.yaml index 6625a668fc..60c5e0388b 100644 --- a/src/omniperf_soc/analysis_configs/gfx942/1700_L2_cache.yaml +++ b/src/omniperf_soc/analysis_configs/gfx942/1700_L2_cache.yaml @@ -31,8 +31,9 @@ Panel Config: unit: pct tips: L2-Fabric Read BW: - value: AVG((((TCC_EA0_RDREQ_32B_sum * 32) + ((TCC_EA0_RDREQ_sum - TCC_EA0_RDREQ_32B_sum) - * 64)) / (End_Timestamp - Start_Timestamp))) + value: AVG((128 * TCC_BUBBLE_sum + + 64 * (TCC_EA0_RDREQ_sum - TCC_BUBBLE_sum - TCC_EA0_RDREQ_32B_sum) + + 32 * TCC_EA0_RDREQ_32B_sum) / (End_Timestamp - Start_Timestamp)) unit: GB/s tips: L2-Fabric Write and Atomic BW: @@ -53,12 +54,15 @@ Panel Config: tips: Tips metric: L2-Fabric Read BW: - avg: AVG((((TCC_EA0_RDREQ_32B_sum * 32) + ((TCC_EA0_RDREQ_sum - TCC_EA0_RDREQ_32B_sum) - * 64)) / $denom)) - min: MIN((((TCC_EA0_RDREQ_32B_sum * 32) + ((TCC_EA0_RDREQ_sum - TCC_EA0_RDREQ_32B_sum) - * 64)) / $denom)) - max: MAX((((TCC_EA0_RDREQ_32B_sum * 32) + ((TCC_EA0_RDREQ_sum - TCC_EA0_RDREQ_32B_sum) - * 64)) / $denom)) + avg: AVG(((128 * TCC_BUBBLE_sum + + 64 * (TCC_EA0_RDREQ_sum - TCC_BUBBLE_sum - TCC_EA0_RDREQ_32B_sum) + + 32 * TCC_EA0_RDREQ_32B_sum) / $denom)) + min: MIN(((128 * TCC_BUBBLE_sum + + 64 * (TCC_EA0_RDREQ_sum - TCC_BUBBLE_sum - TCC_EA0_RDREQ_32B_sum) + + 32 * TCC_EA0_RDREQ_32B_sum) / $denom)) + max: MAX(((128 * TCC_BUBBLE_sum + + 64 * (TCC_EA0_RDREQ_sum - TCC_BUBBLE_sum - TCC_EA0_RDREQ_32B_sum) + + 32 * TCC_EA0_RDREQ_32B_sum) / $denom)) unit: (Bytes + $normUnit) tips: HBM Read Traffic: @@ -380,9 +384,15 @@ Panel Config: unit: (Req + $normUnit) tips: Read (64B): - avg: AVG(((TCC_EA0_RDREQ_sum - TCC_EA0_RDREQ_32B_sum) / $denom)) - min: MIN(((TCC_EA0_RDREQ_sum - TCC_EA0_RDREQ_32B_sum) / $denom)) - max: MAX(((TCC_EA0_RDREQ_sum - TCC_EA0_RDREQ_32B_sum) / $denom)) + avg: AVG(((TCC_EA0_RDREQ_sum - TCC_BUBBLE_sum - TCC_EA0_RDREQ_32B_sum) / $denom)) + min: MIN(((TCC_EA0_RDREQ_sum - TCC_BUBBLE_sum - TCC_EA0_RDREQ_32B_sum) / $denom)) + max: MAX(((TCC_EA0_RDREQ_sum - TCC_BUBBLE_sum - TCC_EA0_RDREQ_32B_sum) / $denom)) + unit: (Req + $normUnit) + tips: + Read (128B): + avg: AVG(((TCC_BUBBLE_sum) / $denom)) + min: MIN(((TCC_BUBBLE_sum) / $denom)) + max: MAX(((TCC_BUBBLE_sum) / $denom)) unit: (Req + $normUnit) tips: HBM Read: diff --git a/src/omniperf_soc/profile_configs/gfx940/pmc_tcc_perf.txt b/src/omniperf_soc/profile_configs/gfx940/pmc_tcc_perf.txt index 3d0fc79d74..72e3bfa4f3 100644 --- a/src/omniperf_soc/profile_configs/gfx940/pmc_tcc_perf.txt +++ b/src/omniperf_soc/profile_configs/gfx940/pmc_tcc_perf.txt @@ -5,7 +5,8 @@ pmc: TCC_NC_REQ_sum TCC_UC_REQ_sum pmc: TCC_REQ_sum TCC_STREAMING_REQ_sum TCC_HIT_sum TCC_MISS_sum pmc: TCC_READ_sum TCC_WRITE_sum TCC_ATOMIC_sum TCC_WRITEBACK_sum pmc: TCC_EA0_WRREQ_sum TCC_EA0_WRREQ_64B_sum TCC_EA0_WR_UNCACHED_32B_sum TCC_EA0_WRREQ_DRAM_sum -pmc: TCC_EA0_RDREQ_sum TCC_EA0_RDREQ_32B_sum TCC_EA0_RD_UNCACHED_32B_sum TCC_EA0_RDREQ_DRAM_sum +pmc: TCC_EA0_RDREQ_sum TCC_EA0_RDREQ_32B_sum TCC_BUBBLE_sum +pmc: TCC_EA0_RD_UNCACHED_32B_sum TCC_EA0_RDREQ_DRAM_sum pmc: TCC_TAG_STALL_sum TCC_NORMAL_WRITEBACK_sum TCC_ALL_TC_OP_WB_WRITEBACK_sum TCC_NORMAL_EVICT_sum pmc: TCC_ALL_TC_OP_INV_EVICT_sum TCC_TOO_MANY_EA_WRREQS_STALL_sum TCC_EA0_ATOMIC_sum pmc: TCC_EA0_RDREQ_LEVEL_sum TCC_EA0_WRREQ_LEVEL_sum TCC_EA0_ATOMIC_LEVEL_sum TCC_EA0_WRREQ_STALL_sum diff --git a/src/utils/parser.py b/src/utils/parser.py index 0284c48930..06d38def78 100644 --- a/src/utils/parser.py +++ b/src/utils/parser.py @@ -290,7 +290,7 @@ def build_eval_string(equation, coll_level): # build-in variable starts with '$', python can not handle it. # replace '$' with 'ammolite__'. # TODO: pre-check there is no "ammolite__" in all config files. - s = re.sub("\$", "ammolite__", s) + s = re.sub(r"\$", "ammolite__", s) # convert equation string to intermediate expression in df array format ast_node = ast.parse(s) @@ -304,7 +304,7 @@ def build_eval_string(equation, coll_level): # the target is df['TCC_HIT[0]'] s = re.sub(r"\'\]\[(\d+)\]", r"[\g<1>]']", s) # use .get() to catch any potential KeyErrors - s = re.sub("raw_pmc_df\['(.*?)']", r'raw_pmc_df.get("\1")', s) + s = re.sub(r"raw_pmc_df\['(.*?)']", r'raw_pmc_df.get("\1")', s) # apply coll_level s = re.sub(r"raw_pmc_df", "raw_pmc_df.get('" + coll_level + "')", s) # print("--- build_eval_string, return: ", s)