From 59e9e3a0c33a568639a1629e56cb70ffc07789bb Mon Sep 17 00:00:00 2001
From: Giovanni Baraldi <gbaraldi@amd.com>
Date: Thu, 12 Dec 2024 12:53:48 -0600
Subject: [PATCH] SWDEV-490031: Fixing activity metrics

Change-Id: Id4e74b0f3ff35d892de05e044faf399d98199354
---
 src/core/counters/derived/metrics.xml         |  4 +--
 .../counters/metrics/derived_counters.xml     | 25 +++++++++----------
 test/tool/metrics.xml                         |  6 ++---
 3 files changed, 17 insertions(+), 18 deletions(-)
 mode change 100755 => 100644 src/core/counters/metrics/derived_counters.xml
 mode change 100755 => 100644 test/tool/metrics.xml

diff --git a/src/core/counters/derived/metrics.xml b/src/core/counters/derived/metrics.xml
index 1db2c3d1d5..1e8cb3878f 100755
--- a/src/core/counters/derived/metrics.xml
+++ b/src/core/counters/derived/metrics.xml
@@ -284,8 +284,8 @@
     <metric name="VFetchInsts" expr="(SQ_INSTS_VMEM_RD-TA_FLAT_READ_WAVEFRONTS_sum)/SQ_WAVES" descr="The average number of vector fetch instructions from the video memory executed per work-item (affected by flow control). Excludes FLAT instructions that fetch from video memory."></metric>
     <metric name="VWriteInsts" expr="(SQ_INSTS_VMEM_WR-TA_FLAT_WRITE_WAVEFRONTS_sum)/SQ_WAVES" descr="The average number of vector write instructions to the video memory executed per work-item (affected by flow control). Excludes FLAT instructions that write to video memory."></metric>
     <metric name="VALUUtilization" expr="100*SQ_THREAD_CYCLES_VALU/(SQ_ACTIVE_INST_VALU*MAX_WAVE_SIZE)" descr="The percentage of active vector ALU threads in a wave. A lower number can mean either more thread divergence in a wave or that the work-group size is not a multiple of 64. Value range: 0% (bad), 100% (ideal - no thread divergence)."></metric>
-    <metric name="VALUBusy" expr="100*SQ_ACTIVE_INST_VALU*4/SIMD_NUM/GRBM_GUI_ACTIVE" descr="The percentage of GPUTime vector ALU instructions are processed. Value range: 0% (bad) to 100% (optimal)."></metric>
-    <metric name="SALUBusy" expr="100*SQ_INST_CYCLES_SALU*4/SIMD_NUM/GRBM_GUI_ACTIVE" descr="The percentage of GPUTime scalar ALU instructions are processed. Value range: 0% (bad) to 100% (optimal)."></metric>
+    <metric name="VALUBusy" expr="100*SQ_ACTIVE_INST_VALU*4*XCC_NUM/SIMD_NUM/GRBM_GUI_ACTIVE" descr="The percentage of GPUTime vector ALU instructions are processed. Value range: 0% (bad) to 100% (optimal)."></metric>
+    <metric name="SALUBusy" expr="100*SQ_INST_CYCLES_SALU*4*XCC_NUM/SIMD_NUM/GRBM_GUI_ACTIVE" descr="The percentage of GPUTime scalar ALU instructions are processed. Value range: 0% (bad) to 100% (optimal)."></metric>
     <metric name="FetchSize" expr="FETCH_SIZE" descr="The total kilobytes fetched from the video memory. This is measured with all extra fetches and any cache or memory effects taken into account."></metric>
     <metric name="WriteSize" expr="WRITE_SIZE" descr="The total kilobytes written to the video memory. This is measured with all extra fetches and any cache or memory effects taken into account."></metric>
     <metric name="MemWrites32B" expr="WRITE_REQ_32B" descr="The total number of effective 32B write transactions to the memory"></metric>
diff --git a/src/core/counters/metrics/derived_counters.xml b/src/core/counters/metrics/derived_counters.xml
old mode 100755
new mode 100644
index b1ac1df1a4..c3b888a2d6
--- a/src/core/counters/metrics/derived_counters.xml
+++ b/src/core/counters/metrics/derived_counters.xml
@@ -104,7 +104,7 @@
 
 <gfx90a_expr base="gfx9_expr">
   <metric name="MeanOccupancyPerCU" expr=SQ_LEVEL_WAVES*0+SQ_ACCUM_PREV_HIRES/GRBM_GUI_ACTIVE/CU_NUM descr="Mean occupancy per compute unit."></metric>
-  <metric name="SIMD_UTILIZATION" expr="SQ_BUSY_CU_CYCLES*XCC_NUM/CU_NUM/GRBM_COUNT" descr="Percent of time at least one warp was active on a multiprocessor, averaged over all multiprocessors."></metric>
+  <metric name="SIMD_UTILIZATION" expr=SQ_BUSY_CU_CYCLES*XCC_NUM/CU_NUM/GRBM_COUNT descr="Percent of time at least one warp was active on a multiprocessor, averaged over all multiprocessors."></metric>
   <metric name="MeanOccupancyPerActiveCU" expr=SQ_LEVEL_WAVES*0+SQ_ACCUM_PREV_HIRES*4/SQ_BUSY_CYCLES/CU_NUM descr="Mean occupancy per active compute unit."></metric>
   <metric name="TA_BUSY_avr"                                            expr=avr(TA_TA_BUSY,16) descr="TA block is busy. Average over TA instances."></metric>
   <metric name="TA_BUSY_max"                                            expr=max(TA_TA_BUSY,16) descr="TA block is busy. Max over TA instances."></metric>
@@ -320,17 +320,17 @@
   <metric name="MfmaFlopsBF16"                                          expr=SQ_INSTS_VALU_MFMA_MOPS_BF16*512                         descr="Unit: FLOP"></metric>
   <metric name="MfmaFlopsF32"                                           expr=SQ_INSTS_VALU_MFMA_MOPS_F32*512                          descr="Unit: FLOP"></metric>
   <metric name="MfmaFlopsF64"                                           expr=SQ_INSTS_VALU_MFMA_MOPS_F64*512                          descr="Unit: IOP"></metric>
-  <metric name="ScaPipeIssueUtil"                                       expr=100*SQ_ACTIVE_INST_SCA/(GRBM_GUI_ACTIVE*CU_NUM)          descr="Unit: percent"></metric>
-  <metric name="ValuPipeIssueUtil"                                      expr=100*SQ_ACTIVE_INST_VALU/(GRBM_GUI_ACTIVE*CU_NUM)         descr="Unit: percent"></metric>
-  <metric name="VmemPipeIssueUtil"                                      expr=100*4*(SQ_ACTIVE_INST_VMEM+SQ_ACTIVE_INST_FLAT)/(GRBM_GUI_ACTIVE*CU_NUM)      descr="Unit: percent"></metric>
-  <metric name="MfmaUtil"                                               expr=100*XCC_NUM*SQ_VALU_MFMA_BUSY_CYCLES/(GRBM_GUI_ACTIVE*CU_NUM*4)               descr="The percentage of kernel's duration, the MFMA unit was busy executing instructions"></metric>
+  <metric name="ScaPipeIssueUtil"                                       expr=100*SQ_ACTIVE_INST_SCA/(ACTIVE_CYCLES*CU_NUM)          descr="Unit: percent"></metric>
+  <metric name="ValuPipeIssueUtil"                                      expr=100*SQ_ACTIVE_INST_VALU/(ACTIVE_CYCLES*CU_NUM)         descr="Unit: percent"></metric>
+  <metric name="VmemPipeIssueUtil"                                      expr=100*4*(SQ_ACTIVE_INST_VMEM+SQ_ACTIVE_INST_FLAT)/(ACTIVE_CYCLES*CU_NUM)      descr="Unit: percent"></metric>
+  <metric name="MfmaUtil"                                               expr=100*SQ_VALU_MFMA_BUSY_CYCLES/(ACTIVE_CYCLES*CU_NUM*4)               descr="The percentage of kernel's duration, the MFMA unit was busy executing instructions"></metric>
   <metric name="AvgNumActiveThreads"                                    expr=SQ_THREAD_CYCLES_VALU/SQ_ACTIVE_INST_VALU                descr="Unit: percent"></metric>
 
   ## Local Data Share (LDS) Metrics
-  <metric name="LdsUtil"                                                expr=100*SQ_LDS_IDX_ACTIVE/(GRBM_GUI_ACTIVE*CU_NUM)                       descr="Unit: percent"></metric>
-  <metric name="LdsPipeIssueUtil"                                       expr=100*4*SQ_ACTIVE_INST_LDS/(GRBM_GUI_ACTIVE*CU_NUM*2)                  descr="Unit: percent"></metric>
+  <metric name="LdsUtil"                                                expr=100*SQ_LDS_IDX_ACTIVE/(ACTIVE_CYCLES*CU_NUM)                       descr="Unit: percent"></metric>
+  <metric name="LdsPipeIssueUtil"                                       expr=100*4*SQ_ACTIVE_INST_LDS/(ACTIVE_CYCLES*CU_NUM*2)                  descr="Unit: percent"></metric>
   <metric name="LdsBankConflict"                                        expr=SQ_LDS_BANK_CONFLICT/(SQ_LDS_IDX_ACTIVE-SQ_LDS_BANK_CONFLICT)        descr="Unit: conflicts/access"></metric>
-  <metric name="LDSBankConflict"                                        expr=100*SQ_LDS_BANK_CONFLICT/GRBM_GUI_ACTIVE/CU_NUM                      descr="The percentage of GPUTime LDS is stalled by bank conflicts. Value range: 0% (optimal) to 100% (bad)."></metric>
+  <metric name="LDSBankConflict"                                        expr=100*SQ_LDS_BANK_CONFLICT/ACTIVE_CYCLES/CU_NUM                      descr="The percentage of GPUTime LDS is stalled by bank conflicts. Value range: 0% (optimal) to 100% (bad)."></metric>
 
   ## L1I and sL1D Cache Metrics
   <metric name="L1iCacheHitRate"                                        expr=100*SQC_ICACHE_HITS/SQC_ICACHE_REQ                               descr="Unit: percent"></metric>
@@ -354,7 +354,7 @@
   <metric name="EaWrStarveRate"                                         expr=100*TCC_TOO_MANY_EA_WRREQS_STALL_sum/TCC_BUSY_sum                descr="Unit: percent"></metric>
 
   <metric name="KERNEL_DURATION" expr=1 descr="The duration of the kernel dispatch"></metric>
-  <metric name="WriteUnitStalled" expr=100*TCC_WRREQ_STALL_max/GRBM_GUI_ACTIVE descr="The percentage of GPUTime the Write unit is stalled. Value range: 0% to 100% (bad)."></metric>
+  <metric name="WriteUnitStalled" expr=100*TCC_WRREQ_STALL_max/ACTIVE_CYCLES descr="The percentage of GPUTime the Write unit is stalled. Value range: 0% to 100% (bad)."></metric>
   <metric name="TCP_TCP_TA_DATA_STALL_CYCLES_sum" expr=sum(TCP_TCP_TA_DATA_STALL_CYCLES,10) descr="Total number of TCP stalls TA data interface."></metric>
   <metric name="TCP_TCP_TA_DATA_STALL_CYCLES_max" expr=max(TCP_TCP_TA_DATA_STALL_CYCLES,10) descr="Maximum number of TCP stalls TA data interface."></metric>
 
@@ -364,8 +364,8 @@
   <metric name="VFetchInsts" expr=(SQ_INSTS_VMEM_RD-TA_FLAT_READ_WAVEFRONTS_sum)/SQ_WAVES descr="The average number of vector fetch instructions from the video memory executed per work-item (affected by flow control). Excludes FLAT instructions that fetch from video memory."></metric>
   <metric name="VWriteInsts" expr=(SQ_INSTS_VMEM_WR-TA_FLAT_WRITE_WAVEFRONTS_sum)/SQ_WAVES descr="The average number of vector write instructions to the video memory executed per work-item (affected by flow control). Excludes FLAT instructions that write to video memory."></metric>
   <metric name="VALUUtilization" expr=100*SQ_THREAD_CYCLES_VALU/(SQ_ACTIVE_INST_VALU*MAX_WAVE_SIZE) descr="The percentage of active vector ALU threads in a wave. A lower number can mean either more thread divergence in a wave or that the work-group size is not a multiple of 64. Value range: 0% (bad), 100% (ideal - no thread divergence)."></metric>
-  <metric name="VALUBusy" expr=100*SQ_ACTIVE_INST_VALU*4/SIMD_NUM/GRBM_GUI_ACTIVE descr="The percentage of GPUTime vector ALU instructions are processed. Value range: 0% (bad) to 100% (optimal)."></metric>
-  <metric name="SALUBusy" expr=100*SQ_INST_CYCLES_SALU*4/SIMD_NUM/GRBM_GUI_ACTIVE descr="The percentage of GPUTime scalar ALU instructions are processed. Value range: 0% (bad) to 100% (optimal)."></metric>
+  <metric name="VALUBusy" expr=100*SQ_ACTIVE_INST_VALU*4/SIMD_NUM/ACTIVE_CYCLES descr="The percentage of GPUTime vector ALU instructions are processed. Value range: 0% (bad) to 100% (optimal)."></metric>
+  <metric name="SALUBusy" expr=100*SQ_INST_CYCLES_SALU*4/SIMD_NUM/ACTIVE_CYCLES descr="The percentage of GPUTime scalar ALU instructions are processed. Value range: 0% (bad) to 100% (optimal)."></metric>
   <metric name="FetchSize" expr=FETCH_SIZE descr="The total kilobytes fetched from the video memory. This is measured with all extra fetches and any cache or memory effects taken into account."></metric>
   <metric name="WriteSize" expr=WRITE_SIZE descr="The total kilobytes written to the video memory. This is measured with all extra fetches and any cache or memory effects taken into account."></metric>
   <metric name="MemWrites32B" expr=WRITE_REQ_32B descr="The total number of effective 32B write transactions to the memory"></metric>
@@ -489,10 +489,9 @@
   <metric name="BANDWIDTH_EA"                                           expr=1024*(FETCH_SIZE+WRITE_SIZE)*XCC_NUM/GRBM_GUI_ACTIVE descr="Memory Bandwidth measured at the TCC_EA interface. In units of bytes/cycle."></metric>
   <metric name="OccupancyPercent"                                       expr=400*SQ_WAVE_CYCLES*XCC_NUM/GRBM_GUI_ACTIVE/CU_NUM/32 descr="GPU occupancy as % of maximum."></metric>
   <metric name="GPU_UTIL"                                               expr=100*GRBM_GUI_ACTIVE/GRBM_COUNT descr="Percentage of the time that GUI is active"></metric>  
-  <metric name="MfmaUtil"                                               expr=100*SQ_VALU_MFMA_BUSY_CYCLES/(GRBM_GUI_ACTIVE*CU_NUM*4) descr="Unit: percent"></metric>
   <metric name="FP64_ACTIVE"                                            expr=TOTAL_64_OPS/GRBM_GUI_ACTIVE descr="The ratio of total floating point 64 bit ops / total number of cycles across all XCCs."></metric>
   <metric name="ENGINE_ACTIVE"                                          expr=GPU_UTIL/100 descr="Ratio between 0-1 of the time the GPU is active"></metric>
-  <metric name="TENSOR_ACTIVE"                                          expr=MfmaUtil descr="Tensor core active in percent, identical to MfmaUtil"></metric>  
+  <metric name="TENSOR_ACTIVE"                                          expr=MfmaUtil descr="Tensor core active in percent, identical to MfmaUtil"></metric>
 </gfx940_expr>
 
 <gfx10_expr>
diff --git a/test/tool/metrics.xml b/test/tool/metrics.xml
old mode 100755
new mode 100644
index 84774646a7..efae266698
--- a/test/tool/metrics.xml
+++ b/test/tool/metrics.xml
@@ -238,12 +238,12 @@
   <metric name="VFetchInsts" expr=(SQ_INSTS_VMEM_RD-TA_FLAT_READ_WAVEFRONTS_sum)/SQ_WAVES descr="The average number of vector fetch instructions from the video memory executed per work-item (affected by flow control). Excludes FLAT instructions that fetch from video memory."></metric>
   <metric name="VWriteInsts" expr=(SQ_INSTS_VMEM_WR-TA_FLAT_WRITE_WAVEFRONTS_sum)/SQ_WAVES descr="The average number of vector write instructions to the video memory executed per work-item (affected by flow control). Excludes FLAT instructions that write to video memory."></metric>
   <metric name="VALUUtilization" expr=100*SQ_THREAD_CYCLES_VALU/(SQ_ACTIVE_INST_VALU*MAX_WAVE_SIZE) descr="The percentage of active vector ALU threads in a wave. A lower number can mean either more thread divergence in a wave or that the work-group size is not a multiple of 64. Value range: 0% (bad), 100% (ideal - no thread divergence)."></metric>
-  <metric name="VALUBusy" expr=100*SQ_ACTIVE_INST_VALU*4/SIMD_NUM/GRBM_GUI_ACTIVE descr="The percentage of GPUTime vector ALU instructions are processed. Value range: 0% (bad) to 100% (optimal)."></metric>
-  <metric name="SALUBusy" expr=100*SQ_INST_CYCLES_SALU*4/SIMD_NUM/GRBM_GUI_ACTIVE descr="The percentage of GPUTime scalar ALU instructions are processed. Value range: 0% (bad) to 100% (optimal)."></metric>
+  <metric name="VALUBusy" expr=100*SQ_ACTIVE_INST_VALU*4/SIMD_NUM/ACTIVE_CYCLES descr="The percentage of GPUTime vector ALU instructions are processed. Value range: 0% (bad) to 100% (optimal)."></metric>
+  <metric name="SALUBusy" expr=100*SQ_INST_CYCLES_SALU*4/SIMD_NUM/ACTIVE_CYCLES descr="The percentage of GPUTime scalar ALU instructions are processed. Value range: 0% (bad) to 100% (optimal)."></metric>
   <metric name="FetchSize" expr=FETCH_SIZE descr="The total kilobytes fetched from the video memory. This is measured with all extra fetches and any cache or memory effects taken into account."></metric>
   <metric name="WriteSize" expr=WRITE_SIZE descr="The total kilobytes written to the video memory. This is measured with all extra fetches and any cache or memory effects taken into account."></metric>
   <metric name="MemWrites32B" expr=WRITE_REQ_32B descr="The total number of effective 32B write transactions to the memory"></metric>
-  <metric name="MemUnitStalled" expr=100*TCP_TCP_TA_DATA_STALL_CYCLES_max/GRBM_GUI_ACTIVE/SE_NUM descr="The percentage of GPUTime the memory unit is stalled. Try reducing the number or size of fetches and writes if possible. Value range: 0% (optimal) to 100% (bad)."></metric>
+  <metric name="MemUnitStalled" expr=100*TCP_TCP_TA_DATA_STALL_CYCLES_max/ACTIVE_CYCLES/SE_NUM descr="The percentage of GPUTime the memory unit is stalled. Try reducing the number or size of fetches and writes if possible. Value range: 0% (optimal) to 100% (bad)."></metric>
   <metric name="TA_BUSY_avr"                                            expr=avr(TA_TA_BUSY,16) descr="TA block is busy. Average over TA instances."></metric>
   <metric name="TA_BUSY_max"                                            expr=max(TA_TA_BUSY,16) descr="TA block is busy. Max over TA instances."></metric>
   <metric name="TA_BUSY_min"                                            expr=min(TA_TA_BUSY,16) descr="TA block is busy. Min over TA instances."></metric>