diff --git a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx906/1200_lds.yaml b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx906/1200_lds.yaml index 218ad2cdad..3fd52c3b1b 100644 --- a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx906/1200_lds.yaml +++ b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx906/1200_lds.yaml @@ -14,6 +14,7 @@ Panel Config: header: metric: Metric value: Value + unit: Unit tips: Tips style: type: simple_bar @@ -23,17 +24,21 @@ Panel Config: metric: Utilization: value: AVG(((100 * SQ_LDS_IDX_ACTIVE) / (GRBM_GUI_ACTIVE * $numCU))) + unit: Pct of Peak tips: Access Rate: value: AVG(((200 * SQ_ACTIVE_INST_LDS) / (GRBM_GUI_ACTIVE * $numCU))) + unit: Pct of Peak tips: Bandwidth (Pct-of-Peak): value: AVG((((((SQ_LDS_IDX_ACTIVE - SQ_LDS_BANK_CONFLICT) * 4) * TO_INT($LDSBanks)) / (EndNs - BeginNs)) / (($sclk * $numCU) * 0.00128))) + unit: Pct of Peak tips: Bank Conflict Rate: value: AVG((((SQ_LDS_BANK_CONFLICT * 3.125) / (SQ_LDS_IDX_ACTIVE - SQ_LDS_BANK_CONFLICT)) if ((SQ_LDS_IDX_ACTIVE - SQ_LDS_BANK_CONFLICT) != 0) else None)) + unit: Pct of Peak tips: - metric_table: diff --git a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx906/1300_instruction-cache.yaml b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx906/1300_instruction-cache.yaml index fb9f384e11..361cb9ae21 100644 --- a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx906/1300_instruction-cache.yaml +++ b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx906/1300_instruction-cache.yaml @@ -14,6 +14,7 @@ Panel Config: header: metric: Metric value: Value + unit: Unit tips: Tips style: type: simple_bar @@ -24,10 +25,12 @@ Panel Config: Bandwidth: value: AVG(((SQC_ICACHE_REQ * 100000) / (($sclk * $numSQC) * (EndNs - BeginNs)))) + unit: Pct of Peak tips: Cache Hit: value: AVG(((SQC_ICACHE_HITS * 100) / ((SQC_ICACHE_HITS + SQC_ICACHE_MISSES) + SQC_ICACHE_MISSES_DUPLICATE))) + unit: Pct of Peak tips: - metric_table: diff --git a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx906/1400_constant-cache.yaml b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx906/1400_constant-cache.yaml index 91a2d6c9f4..eec636b385 100644 --- a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx906/1400_constant-cache.yaml +++ b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx906/1400_constant-cache.yaml @@ -14,6 +14,7 @@ Panel Config: header: mertic: Metric value: Value + unit: Unit tips: Tips style: type: simple_bar @@ -24,11 +25,13 @@ Panel Config: Bandwidth: value: AVG(((SQC_DCACHE_REQ * 100000) / (($sclk * $numSQC) * (EndNs - BeginNs)))) + unit: Pct of Peak tips: Cache Hit: value: AVG((((SQC_DCACHE_HITS * 100) / (SQC_DCACHE_HITS + SQC_DCACHE_MISSES + SQC_DCACHE_MISSES_DUPLICATE)) if ((SQC_DCACHE_HITS + SQC_DCACHE_MISSES + SQC_DCACHE_MISSES_DUPLICATE) != 0) else None)) + unit: Pct of Peak tips: - metric_table: diff --git a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx906/1600_L1_cache.yaml b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx906/1600_L1_cache.yaml index c734e21c8b..e1a7e29ccc 100644 --- a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx906/1600_L1_cache.yaml +++ b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx906/1600_L1_cache.yaml @@ -14,6 +14,7 @@ Panel Config: header: metric: Metric value: Value + unit: Unit tips: Tips style: type: simple_bar @@ -24,20 +25,24 @@ Panel Config: Buffer Coalescing: value: AVG(((((TA_TOTAL_WAVEFRONTS_sum * 64) * 100) / (TCP_TOTAL_ACCESSES_sum * 4)) if (TCP_TOTAL_ACCESSES_sum != 0) else None)) + unit: Pct of Peak tips: Cache Util: value: AVG((((TCP_GATE_EN2_sum * 100) / TCP_GATE_EN1_sum) if (TCP_GATE_EN1_sum != 0) else None)) + unit: Pct of Peak tips: Cache BW: value: ((100 * AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 64) / (EndNs - BeginNs)))) / ((($sclk / 1000) * 64) * $numCU)) + unit: Pct of Peak tips: Cache Hit: value: AVG(((100 - ((100 * (((TCP_TCC_READ_REQ_sum + TCP_TCC_WRITE_REQ_sum) + TCP_TCC_ATOMIC_WITH_RET_REQ_sum) + TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / TCP_TOTAL_CACHE_ACCESSES_sum)) if (TCP_TOTAL_CACHE_ACCESSES_sum != 0) else None)) + unit: Pct of Peak tips: - metric_table: diff --git a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx908/1200_lds.yaml b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx908/1200_lds.yaml index 218ad2cdad..3fd52c3b1b 100644 --- a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx908/1200_lds.yaml +++ b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx908/1200_lds.yaml @@ -14,6 +14,7 @@ Panel Config: header: metric: Metric value: Value + unit: Unit tips: Tips style: type: simple_bar @@ -23,17 +24,21 @@ Panel Config: metric: Utilization: value: AVG(((100 * SQ_LDS_IDX_ACTIVE) / (GRBM_GUI_ACTIVE * $numCU))) + unit: Pct of Peak tips: Access Rate: value: AVG(((200 * SQ_ACTIVE_INST_LDS) / (GRBM_GUI_ACTIVE * $numCU))) + unit: Pct of Peak tips: Bandwidth (Pct-of-Peak): value: AVG((((((SQ_LDS_IDX_ACTIVE - SQ_LDS_BANK_CONFLICT) * 4) * TO_INT($LDSBanks)) / (EndNs - BeginNs)) / (($sclk * $numCU) * 0.00128))) + unit: Pct of Peak tips: Bank Conflict Rate: value: AVG((((SQ_LDS_BANK_CONFLICT * 3.125) / (SQ_LDS_IDX_ACTIVE - SQ_LDS_BANK_CONFLICT)) if ((SQ_LDS_IDX_ACTIVE - SQ_LDS_BANK_CONFLICT) != 0) else None)) + unit: Pct of Peak tips: - metric_table: diff --git a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx908/1300_instruction-cache.yaml b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx908/1300_instruction-cache.yaml index fb9f384e11..361cb9ae21 100644 --- a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx908/1300_instruction-cache.yaml +++ b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx908/1300_instruction-cache.yaml @@ -14,6 +14,7 @@ Panel Config: header: metric: Metric value: Value + unit: Unit tips: Tips style: type: simple_bar @@ -24,10 +25,12 @@ Panel Config: Bandwidth: value: AVG(((SQC_ICACHE_REQ * 100000) / (($sclk * $numSQC) * (EndNs - BeginNs)))) + unit: Pct of Peak tips: Cache Hit: value: AVG(((SQC_ICACHE_HITS * 100) / ((SQC_ICACHE_HITS + SQC_ICACHE_MISSES) + SQC_ICACHE_MISSES_DUPLICATE))) + unit: Pct of Peak tips: - metric_table: diff --git a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx908/1400_constant-cache.yaml b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx908/1400_constant-cache.yaml index 91a2d6c9f4..eec636b385 100644 --- a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx908/1400_constant-cache.yaml +++ b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx908/1400_constant-cache.yaml @@ -14,6 +14,7 @@ Panel Config: header: mertic: Metric value: Value + unit: Unit tips: Tips style: type: simple_bar @@ -24,11 +25,13 @@ Panel Config: Bandwidth: value: AVG(((SQC_DCACHE_REQ * 100000) / (($sclk * $numSQC) * (EndNs - BeginNs)))) + unit: Pct of Peak tips: Cache Hit: value: AVG((((SQC_DCACHE_HITS * 100) / (SQC_DCACHE_HITS + SQC_DCACHE_MISSES + SQC_DCACHE_MISSES_DUPLICATE)) if ((SQC_DCACHE_HITS + SQC_DCACHE_MISSES + SQC_DCACHE_MISSES_DUPLICATE) != 0) else None)) + unit: Pct of Peak tips: - metric_table: diff --git a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx908/1600_L1_cache.yaml b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx908/1600_L1_cache.yaml index 7eeed04774..22efba955c 100644 --- a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx908/1600_L1_cache.yaml +++ b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx908/1600_L1_cache.yaml @@ -14,6 +14,7 @@ Panel Config: header: metric: Metric value: Value + unit: Unit tips: Tips style: type: simple_bar @@ -24,20 +25,24 @@ Panel Config: Buffer Coalescing: value: AVG(((((TA_TOTAL_WAVEFRONTS_sum * 64) * 100) / (TCP_TOTAL_ACCESSES_sum * 4)) if (TCP_TOTAL_ACCESSES_sum != 0) else None)) + unit: Pct of Peak tips: Cache Util: value: AVG((((TCP_GATE_EN2_sum * 100) / TCP_GATE_EN1_sum) if (TCP_GATE_EN1_sum != 0) else None)) + unit: Pct of Peak tips: Cache BW: value: ((100 * AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 64) / (EndNs - BeginNs)))) / ((($sclk / 1000) * 64) * $numCU)) + unit: Pct of Peak tips: Cache Hit: value: AVG(((100 - ((100 * (((TCP_TCC_READ_REQ_sum + TCP_TCC_WRITE_REQ_sum) + TCP_TCC_ATOMIC_WITH_RET_REQ_sum) + TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / TCP_TOTAL_CACHE_ACCESSES_sum)) if (TCP_TOTAL_CACHE_ACCESSES_sum != 0) else None)) + unit: Pct of Peak tips: - metric_table: diff --git a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx90a/1100_compute-unit-compute-pipeline.yaml b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx90a/1100_compute-unit-compute-pipeline.yaml index 3e29bc4a24..769212ed5c 100644 --- a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx90a/1100_compute-unit-compute-pipeline.yaml +++ b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx90a/1100_compute-unit-compute-pipeline.yaml @@ -14,6 +14,7 @@ Panel Config: header: metric: Metric value: Value + unit: Unit tips: Tips style: type: simple_bar @@ -28,26 +29,32 @@ Panel Config: + (64 * (((SQ_INSTS_VALU_ADD_F64 + SQ_INSTS_VALU_MUL_F64) + SQ_INSTS_VALU_TRANS_F64) + (2 * SQ_INSTS_VALU_FMA_F64)))) / (EndNs - BeginNs)))) / (((($sclk * $numCU) * 64) * 2) / 1000)) + unit: Pct of Peak tips: mfma_flops_bf16_pop: value: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_BF16 * 512) / (EndNs - BeginNs)))) / ((($sclk * $numCU) * 512) / 1000)) + unit: Pct of Peak tips: mfma_flops_f16_pop: value: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_F16 * 512) / (EndNs - BeginNs)))) / ((($sclk * $numCU) * 1024) / 1000)) + unit: Pct of Peak tips: mfma_flops_f32_pop: value: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_F32 * 512) / (EndNs - BeginNs)))) / ((($sclk * $numCU) * 256) / 1000)) + unit: Pct of Peak tips: mfma_flops_f64_pop: value: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_F64 * 512) / (EndNs - BeginNs)))) / ((($sclk * $numCU) * 256) / 1000)) + unit: Pct of Peak tips: mfma_flops_i8_pop: value: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (EndNs - BeginNs)))) / ((($sclk * $numCU) * 1024) / 1000)) + unit: Pct of Peak tips: - metric_table: diff --git a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx90a/1200_lds.yaml b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx90a/1200_lds.yaml index 218ad2cdad..3fd52c3b1b 100644 --- a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx90a/1200_lds.yaml +++ b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx90a/1200_lds.yaml @@ -14,6 +14,7 @@ Panel Config: header: metric: Metric value: Value + unit: Unit tips: Tips style: type: simple_bar @@ -23,17 +24,21 @@ Panel Config: metric: Utilization: value: AVG(((100 * SQ_LDS_IDX_ACTIVE) / (GRBM_GUI_ACTIVE * $numCU))) + unit: Pct of Peak tips: Access Rate: value: AVG(((200 * SQ_ACTIVE_INST_LDS) / (GRBM_GUI_ACTIVE * $numCU))) + unit: Pct of Peak tips: Bandwidth (Pct-of-Peak): value: AVG((((((SQ_LDS_IDX_ACTIVE - SQ_LDS_BANK_CONFLICT) * 4) * TO_INT($LDSBanks)) / (EndNs - BeginNs)) / (($sclk * $numCU) * 0.00128))) + unit: Pct of Peak tips: Bank Conflict Rate: value: AVG((((SQ_LDS_BANK_CONFLICT * 3.125) / (SQ_LDS_IDX_ACTIVE - SQ_LDS_BANK_CONFLICT)) if ((SQ_LDS_IDX_ACTIVE - SQ_LDS_BANK_CONFLICT) != 0) else None)) + unit: Pct of Peak tips: - metric_table: diff --git a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx90a/1300_instruction-cache.yaml b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx90a/1300_instruction-cache.yaml index 1a7000e930..c287c13bd7 100644 --- a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx90a/1300_instruction-cache.yaml +++ b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx90a/1300_instruction-cache.yaml @@ -14,6 +14,7 @@ Panel Config: header: metric: Metric value: Value + unit: Unit tips: Tips style: type: simple_bar @@ -24,10 +25,12 @@ Panel Config: Bandwidth: value: AVG(((SQC_ICACHE_REQ * 100000) / (($sclk * $numSQC) * (EndNs - BeginNs)))) + unit: Pct of Peak tips: Cache Hit: value: AVG(((SQC_ICACHE_HITS * 100) / ((SQC_ICACHE_HITS + SQC_ICACHE_MISSES) + SQC_ICACHE_MISSES_DUPLICATE))) + unit: Pct of Peak tips: - metric_table: diff --git a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx90a/1400_constant-cache.yaml b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx90a/1400_constant-cache.yaml index 91a2d6c9f4..eec636b385 100644 --- a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx90a/1400_constant-cache.yaml +++ b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx90a/1400_constant-cache.yaml @@ -14,6 +14,7 @@ Panel Config: header: mertic: Metric value: Value + unit: Unit tips: Tips style: type: simple_bar @@ -24,11 +25,13 @@ Panel Config: Bandwidth: value: AVG(((SQC_DCACHE_REQ * 100000) / (($sclk * $numSQC) * (EndNs - BeginNs)))) + unit: Pct of Peak tips: Cache Hit: value: AVG((((SQC_DCACHE_HITS * 100) / (SQC_DCACHE_HITS + SQC_DCACHE_MISSES + SQC_DCACHE_MISSES_DUPLICATE)) if ((SQC_DCACHE_HITS + SQC_DCACHE_MISSES + SQC_DCACHE_MISSES_DUPLICATE) != 0) else None)) + unit: Pct of Peak tips: - metric_table: diff --git a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx90a/1600_L1_cache.yaml b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx90a/1600_L1_cache.yaml index 7ea26db051..28450f2ae3 100644 --- a/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx90a/1600_L1_cache.yaml +++ b/projects/rocprofiler-compute/src/omniperf_analyze/configs/gfx90a/1600_L1_cache.yaml @@ -14,6 +14,7 @@ Panel Config: header: metric: Metric value: Value + unit: Unit tips: Tips style: type: simple_bar @@ -24,20 +25,24 @@ Panel Config: Buffer Coalescing: value: AVG(((((TA_TOTAL_WAVEFRONTS_sum * 64) * 100) / (TCP_TOTAL_ACCESSES_sum * 4)) if (TCP_TOTAL_ACCESSES_sum != 0) else None)) + unit: Pct of Peak tips: Cache Util: value: AVG((((TCP_GATE_EN2_sum * 100) / TCP_GATE_EN1_sum) if (TCP_GATE_EN1_sum != 0) else None)) + unit: Pct of Peak tips: Cache BW: value: ((100 * AVG(((TCP_TOTAL_CACHE_ACCESSES_sum * 64) / (EndNs - BeginNs)))) / ((($sclk / 1000) * 64) * $numCU)) + unit: Pct of Peak tips: Cache Hit: value: AVG(((100 - ((100 * (((TCP_TCC_READ_REQ_sum + TCP_TCC_WRITE_REQ_sum) + TCP_TCC_ATOMIC_WITH_RET_REQ_sum) + TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / TCP_TOTAL_CACHE_ACCESSES_sum)) if (TCP_TOTAL_CACHE_ACCESSES_sum != 0) else None)) + unit: Pct of Peak tips: - metric_table: