From 6cf9df4ff0089c0eb586403281992d47320ca658 Mon Sep 17 00:00:00 2001
From: Giovanni  LB <gbaraldi@amd.com>
Date: Thu, 25 Apr 2024 17:47:51 -0300
Subject: [PATCH] Fixing occupancy metrics for MI300

Change-Id: I25a150dbcd3582cd397a658aa9a991956577cfa8


[ROCm/rocprofiler commit: 604738fd8de8393ef45d55b1790e9f4b59b7584a]
---
 projects/rocprofiler/src/core/counters/derived/metrics.xml    | 4 ++--
 .../src/core/counters/metrics/derived_counters.xml            | 4 ++--
 projects/rocprofiler/src/core/counters/metrics/metrics.h      | 3 ++-
 projects/rocprofiler/src/core/metrics.h                       | 1 +
 4 files changed, 7 insertions(+), 5 deletions(-)

diff --git a/projects/rocprofiler/src/core/counters/derived/metrics.xml b/projects/rocprofiler/src/core/counters/derived/metrics.xml
index 33793b0f7e..edc12850ad 100755
--- a/projects/rocprofiler/src/core/counters/derived/metrics.xml
+++ b/projects/rocprofiler/src/core/counters/derived/metrics.xml
@@ -266,8 +266,8 @@
   </gfx90a>
 
   <gfx940>
-    <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="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="MeanOccupancyPerCU" expr="SQ_LEVEL_WAVES*0+SQ_ACCUM_PREV_HIRES*XCC_NUM/GRBM_GUI_ACTIVE/CU_NUM" descr="Mean occupancy per compute unit."></metric>
+    <metric name="MeanOccupancyPerActiveCU" expr="SQ_LEVEL_WAVES*0+SQ_ACCUM_PREV_HIRES*4*XCC_NUM/SQ_BUSY_CYCLES/CU_NUM" descr="Mean occupancy per active compute unit."></metric>
 
     <metric name="GPUBusy" expr="100*GRBM_GUI_ACTIVE/GRBM_COUNT" descr="The percentage of time GPU was busy."></metric>
     <metric name="Wavefronts" expr="SQ_WAVES" descr="Total wavefronts."></metric>
diff --git a/projects/rocprofiler/src/core/counters/metrics/derived_counters.xml b/projects/rocprofiler/src/core/counters/metrics/derived_counters.xml
index 892a0bcc6f..5ab53a0c28 100755
--- a/projects/rocprofiler/src/core/counters/metrics/derived_counters.xml
+++ b/projects/rocprofiler/src/core/counters/metrics/derived_counters.xml
@@ -356,8 +356,8 @@
   <metric name="TCP_TCP_TA_DATA_STALL_CYCLES_sum" expr=sum(TCP_TCP_TA_DATA_STALL_CYCLES,16) 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,16) descr="Maximum number of TCP stalls TA data interface."></metric>
 
-  <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="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="MeanOccupancyPerCU" expr=SQ_LEVEL_WAVES*0+SQ_ACCUM_PREV_HIRES*XCC_NUM/GRBM_GUI_ACTIVE/CU_NUM descr="Mean occupancy per compute unit."></metric>
+  <metric name="MeanOccupancyPerActiveCU" expr=SQ_LEVEL_WAVES*0+SQ_ACCUM_PREV_HIRES*4*XCC_NUM/SQ_BUSY_CYCLES/CU_NUM descr="Mean occupancy per active compute unit."></metric>
   <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>
diff --git a/projects/rocprofiler/src/core/counters/metrics/metrics.h b/projects/rocprofiler/src/core/counters/metrics/metrics.h
index 08c28ff5bf..e446b6285b 100644
--- a/projects/rocprofiler/src/core/counters/metrics/metrics.h
+++ b/projects/rocprofiler/src/core/counters/metrics/metrics.h
@@ -211,8 +211,9 @@ class MetricsDict {
     }
     xml_ = xml::Xml::Create(xml_name);
     if (xml_ == NULL) EXC_RAISING(HSA_STATUS_ERROR, "metrics .xml open error '" << xml_name << "'");
-    xml_->AddConst("top.const.metric", "MAX_WAVE_SIZE", agent_info->GetDeviceInfo().getMaxQueueSize());
+    xml_->AddConst("top.const.metric", "MAX_WAVE_SIZE", agent_info->GetDeviceInfo().getMaxWaveSize());
     xml_->AddConst("top.const.metric", "CU_NUM", agent_info->GetDeviceInfo().getCUCount());
+    xml_->AddConst("top.const.metric", "XCC_NUM", agent_info->GetDeviceInfo().getXccCount());
     xml_->AddConst("top.const.metric", "SIMD_NUM",
                    agent_info->GetDeviceInfo().getSimdCountPerCU() * agent_info->GetDeviceInfo().getCUCount());
     xml_->AddConst("top.const.metric", "SE_NUM", agent_info->GetDeviceInfo().getShaderEngineCount());
diff --git a/projects/rocprofiler/src/core/metrics.h b/projects/rocprofiler/src/core/metrics.h
index 26bdeef807..2a5b8f37ef 100644
--- a/projects/rocprofiler/src/core/metrics.h
+++ b/projects/rocprofiler/src/core/metrics.h
@@ -198,6 +198,7 @@ class MetricsDict {
         EXC_RAISING(HSA_STATUS_ERROR, "metrics .xml open error '" << xml_name << "'");
       xml_->AddConst("top.const.metric", "MAX_WAVE_SIZE", agent_info->max_wave_size);
       xml_->AddConst("top.const.metric", "CU_NUM", agent_info->cu_num);
+      xml_->AddConst("top.const.metric", "XCC_NUM", agent_info->xcc_num);
       xml_->AddConst("top.const.metric", "SIMD_NUM", agent_info->simds_per_cu * agent_info->cu_num);
       xml_->AddConst("top.const.metric", "SE_NUM", agent_info->se_num);
       ImportMetrics(agent_info, "const");