From 6d80088c84e9f03b76bd77a908737a4da5ca5455 Mon Sep 17 00:00:00 2001
From: Benjamin Welton <bewelton@amd.com>
Date: Thu, 10 Oct 2024 12:04:03 -0700
Subject: [PATCH] Added FP64_ACTIVE and ENGINE_ACTIVE

Should replicate DCGM_FI_PROF_EVAL_FLOPS_64 and
DCGM_FI_PROF_GR_ENGINE_ACTIVE respectively. See
https://ontrack-internal.amd.com/browse/SWDEV-490046
and
https://ontrack-internal.amd.com/browse/SWDEV-490031

Change-Id: Ia79f6a1601beac48a350493f2e83ce322c1d8d33
---
 src/core/counters/derived/metrics.xml          | 5 +++++
 src/core/counters/metrics/derived_counters.xml | 5 +++++
 test/tool/metrics.xml                          | 5 +++++
 3 files changed, 15 insertions(+)

diff --git a/src/core/counters/derived/metrics.xml b/src/core/counters/derived/metrics.xml
index 90e643275f..13f47b298d 100755
--- a/src/core/counters/derived/metrics.xml
+++ b/src/core/counters/derived/metrics.xml
@@ -395,6 +395,11 @@
     <metric name="WRITE_REQ_32B" expr="TCC_EA0_WRREQ_64B_sum*2+(TCC_EA0_WRREQ_sum-TCC_EA0_WRREQ_64B_sum)" descr="The total number of 32-byte effective memory writes."></metric>
     <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>  
   </gfx940>
 
   <gfx941 base="gfx940"></gfx941>
diff --git a/src/core/counters/metrics/derived_counters.xml b/src/core/counters/metrics/derived_counters.xml
index 1bfb9e7f85..8ac94080d4 100755
--- a/src/core/counters/metrics/derived_counters.xml
+++ b/src/core/counters/metrics/derived_counters.xml
@@ -492,6 +492,11 @@
   <metric name="ACTIVE_WAVES"                                           expr=SQ_WAVES descr="Active Waves"></metric>
   <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>  
 </gfx940_expr>
 
 <gfx10_expr>
diff --git a/test/tool/metrics.xml b/test/tool/metrics.xml
index f5efe5c32c..3f3fa0cd93 100755
--- a/test/tool/metrics.xml
+++ b/test/tool/metrics.xml
@@ -362,6 +362,11 @@
   <metric name="ACTIVE_WAVES"                                           expr=SQ_WAVES descr="Active Waves"></metric>
   <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>  
 </gfx940_expr>
 
 <gfx10_expr>