From d5974aba788e418136afb97017b9cbacb2223dce Mon Sep 17 00:00:00 2001
From: Kiumars Sabeti <Kiumars.Sabeti@amd.com>
Date: Tue, 8 Feb 2022 00:59:47 +0000
Subject: [PATCH] SWDEV-302380: [ROCm QA][Mainline][Navi21] 6 tests are failing
 in rocprofiler-stg2

This is an attempt to support basic and derived counters for navi21.  This code will not work correctly unless we add navi counters to metrics.xml and gfx_metrics.xml

Change-Id: Ied06a81345a6fbb02fa0fde1889d94bbe64e9a03


[ROCm/rocprofiler commit: b53fd84ade9a1686ac41bf3a0ad0b1d2c9107707]
---
 projects/rocprofiler/src/core/metrics.h       |   7 +-
 .../rocprofiler/test/tool/gfx_metrics.xml     |  58 ++++++
 projects/rocprofiler/test/tool/metrics.xml    | 185 ++++++++----------
 3 files changed, 143 insertions(+), 107 deletions(-)
 mode change 100644 => 100755 projects/rocprofiler/src/core/metrics.h
 mode change 100644 => 100755 projects/rocprofiler/test/tool/gfx_metrics.xml
 mode change 100644 => 100755 projects/rocprofiler/test/tool/metrics.xml

diff --git a/projects/rocprofiler/src/core/metrics.h b/projects/rocprofiler/src/core/metrics.h
old mode 100644
new mode 100755
index 6eeebee3ac..7ba9e35bbb
--- a/projects/rocprofiler/src/core/metrics.h
+++ b/projects/rocprofiler/src/core/metrics.h
@@ -198,8 +198,11 @@ class MetricsDict {
       agent_name_ = agent_info->name;
       if (std::string("gfx906") == agent_info->name ||
           std::string("gfx908") == agent_info->name ||
-          std::string("gfx90a") == agent_info->name) {
-        ImportMetrics(agent_info, agent_info->name);
+          std::string("gfx90a") == agent_info->name ||
+          std::string("gfx1032") == agent_info->name ||
+          std::string("gfx1031") == agent_info->name || 
+          std::string("gfx1030") == agent_info->name){ 
+         ImportMetrics(agent_info, agent_info->name);
       } else {
         agent_name_ = agent_info->gfxip;
         ImportMetrics(agent_info, agent_info->gfxip);
diff --git a/projects/rocprofiler/test/tool/gfx_metrics.xml b/projects/rocprofiler/test/tool/gfx_metrics.xml
old mode 100644
new mode 100755
index 628a629a55..46dcfbb4fd
--- a/projects/rocprofiler/test/tool/gfx_metrics.xml
+++ b/projects/rocprofiler/test/tool/gfx_metrics.xml
@@ -365,3 +365,61 @@
   <metric name="TCC_EA_RDREQ_DRAM" block=TCC event=102 descr="Number of TCC/EA read requests (either 32-byte or 64-byte) destined for DRAM (MC)."></metric>
   <metric name="TCC_EA_WRREQ_DRAM" block=TCC event=103 descr="Number of TCC/EA write requests (either 32-byte of 64-byte) destined for DRAM (MC)."></metric>
 </gfx90a>
+
+<gfx10>
+  <metric name="GRBM_COUNT" block=GRBM event=0 descr="Tie High - Count Number of Clocks"></metric>
+  <metric name="GRBM_GUI_ACTIVE" block=GRBM event=2 descr="The GUI is Active"></metric>
+  <metric name="GRBM_CP_BUSY" block=GRBM event=3 descr="Any of the Command Processor (CPG/CPC/CPF) blocks are busy."></metric>
+  <metric name="GRBM_SPI_BUSY" block=GRBM event=11 descr="Any of the Shader Pipe Interpolators (SPI) are busy in the shader engine(s)."></metric>
+  <metric name="GRBM_TA_BUSY" block=GRBM event=13 descr="Any of the Texture Pipes (TA) are busy in the shader engine(s)."></metric>
+  <metric name="GRBM_GDS_BUSY" block=GRBM event=25 descr="The Global Data Share (GDS) is busy."></metric>
+  <metric name="GRBM_EA_BUSY" block=GRBM event=35 descr="The Efficiency Arbiter (EA) block is busy."></metric>
+  <metric name="GRBM_GL2CC_BUSY" block=GRBM event=40 descr="The GL2CC block is busy."></metric>
+
+  <metric name="GL2C_HIT" block=GL2C event=42 descr="Number of cache hits"></metric>
+  <metric name="GL2C_MISS" block=GL2C event=43 descr="Number of cache misses.  UC reads count as misses."></metric>
+  <metric name="GL2C_MC_WRREQ" block=GL2C event=83 descr="Number of transactions (either 32-byte or 64-byte) going over the GL2C_EA_wrreq interface. Atomics may travel over the same interface and are generally classified as write requests. This does not include probe commands"></metric>
+  <metric name="GL2C_EA_WRREQ_64B" block=GL2C event=85 descr="Number of 64-byte transactions going (64-byte write or CMPSWAP) over the TC_EA_wrreq interface."></metric>
+  <metric name="GL2C_MC_WRREQ_STALL" block=GL2C event=88 descr="Number of cycles a write request was stalled."></metric>
+  <metric name="GL2C_MC_RDREQ" block=GL2C event=96 descr="Number of GL2C/EA read requests (either 32-byte or 64-byte or 128-byte)."></metric>
+  <metric name="GL2C_EA_RDREQ_32B" block=GL2C event=99 descr="Number of 32-byte GL2C/EA read requests"></metric>
+  <metric name="GL2C_EA_RDREQ_64B" block=GL2C event=100 descr="Number of 64-byte GL2C/EA read requests"></metric>
+  <metric name="GL2C_EA_RDREQ_96B" block=GL2C event=101 descr="Number of 96-byte GL2C/EA read requests"></metric>
+  <metric name="GL2C_EA_RDREQ_128B" block=GL2C event=102 descr="Number of 128-byte GL2C/EA read requests"></metric>
+
+  <metric name="SQ_WAVES" block=SQ event=4 descr="Count number of waves sent to SQs. {emulated, global, C1}"></metric>
+  <metric name="SQ_WAVE_CYCLES" block=SQ event=26 descr="Number of clock cycles spent by waves in the SQs. Incremented by # of living (valid) waves each cycle. {nondeterministic, C1}"></metric>
+  <metric name="SQ_WAIT_INST_ANY" block=SQ event=28 descr="Number of clock cycles spent waiting for any instruction issue. In units of cycles. {nondeterministic}"></metric>
+  <metric name="SQ_WAIT_ANY" block=SQ event=37 descr="Number of clock cycles spent waiting for anything. {nondeterministic, C1}"></metric>
+  <metric name="SQ_INSTS_WAVE32" block=SQ event=71 descr="Number of wave32 instructions issued, for flat, lds, valu, tex. {emulated, C1}"></metric>
+  <metric name="SQ_INSTS_WAVE32_LDS" block=SQ event=74 descr="Number of wave32 LDS indexed instructions issued. Wave64 may count 1 or 2, depending on what gets issued. {emulated, C1}"></metric>
+  <metric name="SQ_INSTS_WAVE32_VALU" block=SQ event=75 descr="Number of wave32 valu instructions issued. Wave64 may count 1 or 2, depending on what gets issued. {emulated, C1}"></metric>
+  <metric name="SQ_WAVE32_INSTS" block=SQ event=84 descr="Number of instructions issued by wave32 waves. Skipped instructions are not counted. {emulated}"></metric>
+  <metric name="SQ_WAVE64_INSTS" block=SQ event=85 descr="Number of instructions issued by wave64 waves. Skipped instructions are not counted. {emulated}"></metric>
+  <metric name="SQ_INST_LEVEL_GDS" block=SQ event=98 descr="Number of in-flight GDS instructions. Set next counter to ACCUM_PREV and divide by INSTS_GDS for average latency. {level, nondeterministic, C1}"></metric>
+  <metric name="SQ_INST_LEVEL_LDS" block=SQ event=99 descr="Number of in-flight LDS instructions. Set next counter to ACCUM_PREV and divide by INSTS_LDS for average latency. Includes FLAT instructions. {level, nondeterministic, C1}"></metric>
+  <metric name="SQ_INST_CYCLES_VMEM" block=SQ event=120 descr="Number of cycles needed to send addr and data for VMEM (lds, buffer, image, flat, scratch, global) instructions, windowed by perf_en. {emulated, C1}"></metric>
+  <metric name="SQC_LDS_BANK_CONFLICT" block=SQ event=285 descr="Number of cycles LDS is stalled by bank conflicts. (emulated, C1)"></metric>
+  <metric name="SQC_LDS_IDX_ACTIVE" block=SQ event=290 descr="Number of cycles LDS is used for indexed (non-direct,non-interpolation) operations. {per-simd, emulated, C1}"></metric>
+  <metric name="SQ_INSTS_VALU" block=SQ event=64 descr="Number of VALU instructions issued excluding skipped instructions. {emulated, C1}"></metric>
+  <metric name="SQ_INSTS_SALU" block=SQ event=60 descr="Number of SALU instructions issued. {emulated, C1}"></metric>
+  <metric name="SQ_INSTS_SMEM" block=SQ event=61 descr="Number of SMEM instructions issued. {emulated, C1}"></metric>
+  <metric name="SQ_INSTS_FLAT" block=SQ event=57 descr="Number of FLAT instructions issued. {emulated, C2}"></metric>
+  <metric name="SQ_INSTS_LDS" block=SQ event=59 descr="Number of LDS indexed instructions issued. {emulated, C1}"></metric>
+  <metric name="SQ_INSTS_GDS" block=SQ event=55 descr="Number of GDS instructions issued. {emulated, C1}"></metric>
+  <metric name="SQ_WAIT_INST_LDS" block=SQ event=31 descr="Number of clock cycles spent waiting for LDS (indexed) instruction issue. In units of cycles. {nondeterministic, C1}"></metric>
+  
+  <metric name="TA_TA_BUSY" block=TA event=15 descr="TA block is busy. Perf_Windowing not supported for this counter."></metric>
+  <metric name="TA_FLAT_LOAD_WAVEFRONTS" block=TA event=101 descr=" Number of flat load vec32 packets processed by TA, same as flat_read_wavefronts in earlier IP"></metric>
+  <metric name="TA_FLAT_STORE_WAVEFRONTS" block=TA event=102 descr="Number of flat store vec32 packets processed by TA, same as flat_write_wavefronts in earlier IP"></metric>
+</gfx10>
+
+<gfx1030 base="gfx10">
+</gfx1030>
+
+<gfx1031 base="gfx10">
+</gfx1031>
+
+<gfx1032 base="gfx10">
+</gfx1032>
+
diff --git a/projects/rocprofiler/test/tool/metrics.xml b/projects/rocprofiler/test/tool/metrics.xml
old mode 100644
new mode 100755
index 2f8e10dd17..85f3e6fb1b
--- a/projects/rocprofiler/test/tool/metrics.xml
+++ b/projects/rocprofiler/test/tool/metrics.xml
@@ -16,6 +16,22 @@
   <metric name="FETCH_SIZE" expr=(TCC_MC_RDREQ_sum*32)/1024 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="WRITE_SIZE" expr=(TCC_MC_WRREQ_sum*32)/1024 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="WRITE_REQ_32B" expr=TCC_MC_WRREQ_sum descr="The total number of 32-byte effective memory writes."></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="FlatVMemInsts" expr=(SQ_INSTS_FLAT-SQ_INSTS_FLAT_LDS_ONLY)/SQ_WAVES descr="The average number of FLAT instructions that read from or write to the video memory executed per work item (affected by flow control). Includes FLAT instructions that read from or write to scratch."></metric>
+  <metric name="LDSInsts" expr=(SQ_INSTS_LDS-SQ_INSTS_FLAT_LDS_ONLY)/SQ_WAVES descr="The average number of LDS read or LDS write instructions executed per work item (affected by flow control).  Excludes FLAT instructions that read from or write to LDS."></metric>
+  <metric name="FlatLDSInsts" expr=SQ_INSTS_FLAT_LDS_ONLY/SQ_WAVES descr="The average number of FLAT instructions that read or write to LDS executed per work item (affected by flow control)."></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="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="L2CacheHit" expr=100*sum(TCC_HIT,16)/(sum(TCC_HIT,16)+sum(TCC_MISS,16)) descr="The percentage of fetch, write, atomic, and other instructions that hit the data in L2 cache. Value range: 0% (no hit) to 100% (optimal)."></metric>
+  <metric name="MemUnitStalled" expr=100*max(TCP_TCP_TA_DATA_STALL_CYCLES,16)/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="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>
+  # LDSBankConflict The percentage of GPUTime LDS is stalled by bank conflicts. Value range: 0% (optimal) to 100% (bad).
+  <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>
 </gfx8_expr>
 
 <gfx9_expr>
@@ -36,6 +52,22 @@
   <metric name="FETCH_SIZE" expr=(TCC_EA_RDREQ_32B_sum*32+(TCC_EA_RDREQ_sum-TCC_EA_RDREQ_32B_sum)*64)/1024 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="WRITE_SIZE" expr=((TCC_EA_WRREQ_sum-TCC_EA_WRREQ_64B_sum)*32+TCC_EA_WRREQ_64B_sum*64)/1024 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="WRITE_REQ_32B" expr=TCC_EA_WRREQ_64B_sum*2+(TCC_EA_WRREQ_sum-TCC_EA_WRREQ_64B_sum) descr="The total number of 32-byte effective memory writes."></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="FlatVMemInsts" expr=(SQ_INSTS_FLAT-SQ_INSTS_FLAT_LDS_ONLY)/SQ_WAVES descr="The average number of FLAT instructions that read from or write to the video memory executed per work item (affected by flow control). Includes FLAT instructions that read from or write to scratch."></metric>
+  <metric name="LDSInsts" expr=(SQ_INSTS_LDS-SQ_INSTS_FLAT_LDS_ONLY)/SQ_WAVES descr="The average number of LDS read or LDS write instructions executed per work item (affected by flow control).  Excludes FLAT instructions that read from or write to LDS."></metric>
+  <metric name="FlatLDSInsts" expr=SQ_INSTS_FLAT_LDS_ONLY/SQ_WAVES descr="The average number of FLAT instructions that read or write to LDS executed per work item (affected by flow control)."></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="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="L2CacheHit" expr=100*sum(TCC_HIT,16)/(sum(TCC_HIT,16)+sum(TCC_MISS,16)) descr="The percentage of fetch, write, atomic, and other instructions that hit the data in L2 cache. Value range: 0% (no hit) to 100% (optimal)."></metric>
+  <metric name="MemUnitStalled" expr=100*max(TCP_TCP_TA_DATA_STALL_CYCLES,16)/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="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>
+  # LDSBankConflict The percentage of GPUTime LDS is stalled by bank conflicts. Value range: 0% (optimal) to 100% (bad).
+  <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>
 </gfx9_expr>
 
 <gfx906_expr base="gfx9_expr">
@@ -183,14 +215,62 @@
   <metric name="WRITE_REQ_32B"                                          expr=TCC_EA_WRREQ_64B_sum*2+(TCC_EA_WRREQ_sum-TCC_EA_WRREQ_64B_sum) descr="The total number of 32-byte effective memory writes."></metric>
 </gfx90a_expr>
 
+<gfx10_expr>
+  <metric name="GPU_UTIL" expr=100*GRBM_GUI_ACTIVE/GRBM_COUNT descr="Percentage of the time that GUI is active"></metric>
+  <metric name="CP_UTIL" expr=100*GRBM_CP_BUSY/GRBM_GUI_ACTIVE descr="Percentage of the GRBM_GUI_ACTIVE time that any of the Command Processor (CPG/CPC/CPF) blocks are busy"></metric>
+  <metric name="SPI_UTIL" expr=100*GRBM_SPI_BUSY/GRBM_GUI_ACTIVE descr="Percentage of the GRBM_GUI_ACTIVE time that any of the Shader Pipe Interpolators (SPI) are busy in the shader engine(s)"></metric>
+  <metric name="TA_UTIL" expr=100*GRBM_TA_BUSY/GRBM_GUI_ACTIVE descr="Percentage of the GRBM_GUI_ACTIVE time that any of the Texture Pipes (TA) are busy in the shader engine(s)."></metric>
+  <metric name="GDS_UTIL" expr=100*GRBM_GDS_BUSY/GRBM_GUI_ACTIVE descr="Percentage of the GRBM_GUI_ACTIVE time that the Global Data Share (GDS) is busy."></metric>
+  <metric name="EA_UTIL" expr=100*GRBM_EA_BUSY/GRBM_GUI_ACTIVE descr="Percentage of the GRBM_GUI_ACTIVE time that the Efficiency Arbiter (EA) block is busy."></metric>
+  <metric name="WAVE_DEP_WAIT" expr=100*SQ_WAIT_ANY/SQ_WAVE_CYCLES descr="Percentage of the SQ_WAVE_CYCLE time spent waiting for anything."></metric>
+  <metric name="WAVE_ISSUE_WAIT" expr=100*SQ_WAIT_INST_ANY/SQ_WAVE_CYCLES descr="Percentage of the SQ_WAVE_CYCLE time spent waiting for any instruction issue."></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>
+  <metric name="TA_FLAT_LOAD_WAVEFRONTS_sum" expr=sum(TA_FLAT_LOAD_WAVEFRONTS,16) descr="Number of flat load vec32 packets processed by the TA. Sum over TA instances."></metric>
+  <metric name="TA_FLAT_STORE_WAVEFRONTS_sum" expr=sum(TA_FLAT_STORE_WAVEFRONTS,16) descr="Number of flat store vec32 packets processed by the TA. Sum over TA instances."></metric>
+
+  <metric name="GL2C_HIT_sum" expr=sum(GL2C_HIT,16) descr="Number of cache hits. Sum over GL2C instances."></metric>
+  <metric name="GL2C_MISS_sum" expr=sum(GL2C_MISS,16) descr="Number of cache misses. Sum over GL2C instances."></metric>
+  <metric name="GL2C_EA_RDREQ_32B_sum" expr=sum(GL2C_EA_RDREQ_32B,16) descr="Number of 32-byte GL2C/EA read requests. Sum over GL2C instances."></metric>
+  <metric name="GL2C_EA_RDREQ_64B_sum" expr=sum(GL2C_EA_RDREQ_64B,16) descr="Number of 64-byte GL2C/EA read requests. Sum over GL2C instances."></metric>
+  <metric name="GL2C_EA_RDREQ_96B_sum" expr=sum(GL2C_EA_RDREQ_96B,16) descr="Number of 96-byte GL2C/EA read requests. Sum over GL2C instances."></metric>
+  <metric name="GL2C_EA_RDREQ_128B_sum" expr=sum(GL2C_EA_RDREQ_128B,16) descr="Number of 128-byte GL2C/EA read requests. Sum over GL2C instances."></metric>
+  <metric name="GL2C_MC_RDREQ_sum" expr=sum(GL2C_MC_RDREQ,16) descr="Number of GL2C/EA read requests (either 32-byte or 64-byte or 128-byte). Sum over GL2C instances."></metric>
+  <metric name="GL2C_MC_WRREQ_sum" expr=sum(GL2C_MC_WRREQ,16) descr="Number of transactions (either 32-byte or 64-byte) going over the GL2C_MC_wrreq interface. Sum over GL2C instances."></metric>
+  <metric name="GL2C_EA_WRREQ_64B_sum" expr=sum(GL2C_EA_WRREQ_64B,16) descr="Number of 64-byte transactions going (64-byte write or CMPSWAP) over the GL2C_EA_wrreq interface. Sum over GL2C instances."></metric>
+  <metric name="GL2C_WRREQ_STALL_max" expr=max(GL2C_MC_WRREQ_STALL,16) descr="Number of cycles a write request was stalled. Max over GL2C instances."></metric>
+  <metric name="L2CacheHit" expr=100*sum(GL2C_HIT,16)/(sum(GL2C_HIT,16)+sum(GL2C_MISS,16)) descr="The percentage of fetch, write, atomic, and other instructions that hit the data in L2 cache. Value range: 0% (no hit) to 100% (optimal)."></metric>
+  <metric name="FETCH_SIZE" expr=(GL2C_EA_RDREQ_32B_sum*32+GL2C_EA_RDREQ_64B_sum*64+GL2C_EA_RDREQ_96B_sum*96+GL2C_EA_RDREQ_128B_sum*128)/1024 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="WriteUnitStalled" expr=100*GL2C_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="LDSBankConflict" expr=100*SQC_LDS_BANK_CONFLICT/SQC_LDS_IDX_ACTIVE descr="The percentage of GPUTime LDS is stalled by bank conflicts. Value range: 0% (optimal) to 100% (bad)."></metric>
+</gfx10_expr>
+
+<gfx1030_expr base="gfx10_expr">
+</gfx1030_expr>
+
+<gfx1031_expr base="gfx10_expr">
+</gfx1031_expr>
+
+<gfx1032_expr base="gfx10_expr">
+</gfx1032_expr>
+
 <gfx8 base="gfx8_expr"></gfx8>
 <gfx9 base="gfx9_expr"></gfx9>
+<gfx10 base="gfx10_expr"></gfx10>
 # Vega20
 <gfx906 base="gfx906_expr"></gfx906>
 # Arcturus
 <gfx908 base="gfx908_expr"></gfx908>
 # Aldebaran
 <gfx90a base="gfx90a_expr"></gfx90a>
+#Navi21
+<gfx1030 base="gfx1030_expr"></gfx1030>
+
+<gfx1031 base="gfx1031_expr"></gfx1031>
+
+<gfx1032 base="gfx1032_expr"></gfx1032>
 
 <global>
   # GPUBusy         The percentage of time GPU was busy.
@@ -221,13 +301,6 @@
     expr=SQ_INSTS_SALU/SQ_WAVES
   ></metric>
 
-  # VFetchInsts     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
-    name="VFetchInsts"
-    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."
-    expr=(SQ_INSTS_VMEM_RD-TA_FLAT_READ_WAVEFRONTS_sum)/SQ_WAVES
-  ></metric>
-
   # SFetchInsts     The average number of scalar fetch instructions from the video memory executed per work-item (affected by flow control).
   <metric
     name="SFetchInsts"
@@ -235,34 +308,6 @@
     expr=SQ_INSTS_SMEM/SQ_WAVES
   ></metric>
 
-  # VWriteInsts     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
-    name="VWriteInsts"
-    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."
-    expr=(SQ_INSTS_VMEM_WR-TA_FLAT_WRITE_WAVEFRONTS_sum)/SQ_WAVES
-  ></metric>
-
-  # FlatVMemInsts   The average number of FLAT instructions that read from or write to the video memory executed per work item (affected by flow control). Includes FLAT instructions that read from or write to scratch.
-  <metric
-    name="FlatVMemInsts"
-    descr="The average number of FLAT instructions that read from or write to the video memory executed per work item (affected by flow control). Includes FLAT instructions that read from or write to scratch."
-    expr=(SQ_INSTS_FLAT-SQ_INSTS_FLAT_LDS_ONLY)/SQ_WAVES
-  ></metric>
-
-  # LDSInsts        The average number of LDS read or LDS write instructions executed per work item (affected by flow control).  Excludes FLAT instructions that read from or write to LDS.
-  <metric
-    name="LDSInsts"
-    descr="The average number of LDS read or LDS write instructions executed per work item (affected by flow control).  Excludes FLAT instructions that read from or write to LDS."
-    expr=(SQ_INSTS_LDS-SQ_INSTS_FLAT_LDS_ONLY)/SQ_WAVES
-  ></metric>
-
-  # FlatLDSInsts    The average number of FLAT instructions that read or write to LDS executed per work item (affected by flow control).
-  <metric
-    name="FlatLDSInsts"
-    descr="The average number of FLAT instructions that read or write to LDS executed per work item (affected by flow control)."
-    expr=SQ_INSTS_FLAT_LDS_ONLY/SQ_WAVES
-  ></metric>
-
   # GDSInsts        The average number of GDS read or GDS write instructions executed per work item (affected by flow control).
   <metric
     name="GDSInsts"
@@ -270,55 +315,6 @@
     expr=SQ_INSTS_GDS/SQ_WAVES
   ></metric>
 
-  # VALUUtilization 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
-    name="VALUUtilization"
-    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)."
-    expr=100*SQ_THREAD_CYCLES_VALU/(SQ_ACTIVE_INST_VALU*MAX_WAVE_SIZE)
-  ></metric>
-
-  # VALUBusy        The percentage of GPUTime vector ALU instructions are processed. Value range: 0% (bad) to 100% (optimal).
-  <metric
-    name="VALUBusy"
-    descr="The percentage of GPUTime vector ALU instructions are processed. Value range: 0% (bad) to 100% (optimal)."
-    expr=100*SQ_ACTIVE_INST_VALU*4/SIMD_NUM/GRBM_GUI_ACTIVE
-  ></metric>
-
-  # SALUBusy        The percentage of GPUTime scalar ALU instructions are processed. Value range: 0% (bad) to 100% (optimal).
-  <metric
-    name="SALUBusy"
-    descr="The percentage of GPUTime scalar ALU instructions are processed. Value range: 0% (bad) to 100% (optimal)."
-    expr=100*SQ_INST_CYCLES_SALU*4/SIMD_NUM/GRBM_GUI_ACTIVE
-  ></metric>
-
-  # FetchSize       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
-    name="FetchSize"
-    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."
-    expr=FETCH_SIZE
-  ></metric>
-
-  # WriteSize       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
-    name="WriteSize"
-    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."
-    expr=WRITE_SIZE
-  ></metric>
-
-  # MemWrites32B    The total number of effective 32B write transactions to the memory
-  <metric
-    name="MemWrites32B"
-    descr="The total number of effective 32B write transactions to the memory"
-    expr=WRITE_REQ_32B
-  ></metric>
-
-  # L2CacheHit      The percentage of fetch, write, atomic, and other instructions that hit the data in L2 cache. Value range: 0% (no hit) to 100% (optimal).
-  <metric
-    name="L2CacheHit"
-    descr="The percentage of fetch, write, atomic, and other instructions that hit the data in L2 cache. Value range: 0% (no hit) to 100% (optimal)."
-    expr=100*sum(TCC_HIT,16)/(sum(TCC_HIT,16)+sum(TCC_MISS,16))
-  ></metric>
-
   # MemUnitBusy     The percentage of GPUTime the memory unit is active. The result includes the stall time (MemUnitStalled). This is measured with all extra fetches and writes and any cache or memory effects taken into account. Value range: 0% to 100% (fetch-bound).
   <metric
     name="MemUnitBusy"
@@ -326,20 +322,6 @@
     expr=100*max(TA_TA_BUSY,16)/GRBM_GUI_ACTIVE/SE_NUM
   ></metric>
 
-  # MemUnitStalled  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
-    name="MemUnitStalled"
-    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)."
-    expr=100*max(TCP_TCP_TA_DATA_STALL_CYCLES,16)/GRBM_GUI_ACTIVE/SE_NUM
-  ></metric>
-
-  # WriteUnitStalled The percentage of GPUTime the Write unit is stalled. Value range: 0% to 100% (bad).
-  <metric
-    name="WriteUnitStalled"
-    descr="The percentage of GPUTime the Write unit is stalled. Value range: 0% to 100% (bad)."
-    expr=100*TCC_WRREQ_STALL_max/GRBM_GUI_ACTIVE
-  ></metric>
-
   # ALUStalledByLDS The percentage of GPUTime ALU units are stalled by the LDS input queue being full or the output queue being not ready. If there are LDS bank conflicts, reduce them. Otherwise, try reducing the number of LDS accesses if possible. Value range: 0% (optimal) to 100% (bad).
   <metric
     name="ALUStalledByLDS"
@@ -347,11 +329,4 @@
     expr=100*SQ_WAIT_INST_LDS*4/SQ_WAVES/GRBM_GUI_ACTIVE
   ></metric>
 
-  # LDSBankConflict The percentage of GPUTime LDS is stalled by bank conflicts. Value range: 0% (optimal) to 100% (bad).
-  <metric
-    name="LDSBankConflict"
-    descr="The percentage of GPUTime LDS is stalled by bank conflicts. Value range: 0% (optimal) to 100% (bad)."
-    expr=100*SQ_LDS_BANK_CONFLICT/GRBM_GUI_ACTIVE/CU_NUM
-  ></metric>
-
 </global>