diff --git a/src/core/counters/basic/gfx_metrics.xml b/src/core/counters/basic/gfx_metrics.xml
index 83d0bc50d7..310284f4ab 100755
--- a/src/core/counters/basic/gfx_metrics.xml
+++ b/src/core/counters/basic/gfx_metrics.xml
@@ -670,7 +670,11 @@
     <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_ACCUM_PREV" block="SQ" event="1" descr="For counter N, increment by the value of counter N-1."></metric>
+    <metric name="SQ_BUSY_CYCLES" block="SQ" event="3" descr="Clock cycles while SQ is reporting that it is busy. {nondeterministic, global, C2}"></metric>
     <metric name="SQ_WAVES" block="SQ" event="4" descr="Count number of waves sent to SQs. {emulated, global, C1}"></metric>
+    <metric name="SQ_LEVEL_WAVES" block="SQ" event="7" descr="Track the aggregated number of waves over certain period of time, Set next counter to ACCUM_PREV and divide by SQ_PERF_SEL_WAVES for average wave life."></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>
@@ -709,6 +713,8 @@
     <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_ACCUM_PREV" block="SQ" event="1" descr="For counter N, increment by the value of counter N-1."></metric>
+    <metric name="SQ_BUSY_CYCLES" block="SQ" event="3" descr="Clock cycles while SQ is reporting that it is busy. {nondeterministic, global, C2}"></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="24" descr="Number of clock cycles spent by waves in the SQs. Incremented by number of living (valid) waves each cycle. {nondeterministic, C1}"></metric>
     <metric name="SQ_WAIT_INST_ANY" block="SQ" event="26" descr="Number of clock-cycles spent waiting for any instruction issue. In units of cycles. (nondeterministic)"></metric>
diff --git a/src/core/counters/derived/metrics.xml b/src/core/counters/derived/metrics.xml
index 05a959b7f5..2739575e21 100755
--- a/src/core/counters/derived/metrics.xml
+++ b/src/core/counters/derived/metrics.xml
@@ -129,6 +129,9 @@
   </gfx908>
 
   <gfx90a>
+    <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="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>
     <metric name="VALUInsts" expr="SQ_INSTS_VALU/SQ_WAVES" descr="The average number of vector ALU instructions executed per work-item (affected by flow control)."></metric>
@@ -261,6 +264,9 @@
   </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="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>
     <metric name="VALUInsts" expr="SQ_INSTS_VALU/SQ_WAVES" descr="The average number of vector ALU instructions executed per work-item (affected by flow control)."></metric>
@@ -390,6 +396,9 @@
   <gfx942 base="gfx940"></gfx942>
 
   <gfx10>
+    <metric name="MeanOccupancyPerCU" expr="GRBM_COUNT*0+SQ_LEVEL_WAVES*0+SQ_ACCUM_PREV/GRBM_GUI_ACTIVE/CU_NUM" descr="Mean occupancy per compute unit."></metric>
+    <metric name="MeanOccupancyPerActiveCU" expr="GRBM_COUNT*0+SQ_LEVEL_WAVES*0+SQ_ACCUM_PREV*4/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>
     <metric name="VALUInsts" expr="SQ_INSTS_VALU/SQ_WAVES" descr="The average number of vector ALU instructions executed per work-item (affected by flow control)."></metric>
diff --git a/src/core/counters/metrics/basic_counters.xml b/src/core/counters/metrics/basic_counters.xml
index 6a9bdca6dd..5300ed63b3 100755
--- a/src/core/counters/metrics/basic_counters.xml
+++ b/src/core/counters/metrics/basic_counters.xml
@@ -654,7 +654,10 @@
   <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_ACCUM_PREV" block=SQ event=1 descr="For counter N, increment by the value of counter N-1."></metric>
+  <metric name="SQ_BUSY_CYCLES" block=SQ event=3 descr="Clock cycles while SQ is reporting that it is busy. {nondeterministic, global, C2}"></metric>
   <metric name="SQ_WAVES" block=SQ event=4 descr="Count number of waves sent to SQs. {emulated, global, C1}"></metric>
+  <metric name="SQ_LEVEL_WAVES" block=SQ event=7 descr="Track the aggregated number of waves over certain period of time, Set next counter to ACCUM_PREV and divide by SQ_PERF_SEL_WAVES for average wave life."></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>
@@ -703,6 +706,8 @@
   <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_ACCUM_PREV" block=SQ event=1 descr="For counter N, increment by the value of counter N-1."></metric>
+  <metric name="SQ_BUSY_CYCLES" block=SQ event=3 descr="Clock cycles while SQ is reporting that it is busy. {nondeterministic, global, C2}"></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=24 descr="Number of clock cycles spent by waves in the SQs. Incremented by number of living (valid) waves each cycle. {nondeterministic, C1}"></metric>
   <metric name="SQ_WAIT_INST_ANY" block=SQ event=26 descr="Number of clock-cycles spent waiting for any instruction issue. In units of cycles. (nondeterministic)"></metric>
diff --git a/src/core/counters/metrics/derived_counters.xml b/src/core/counters/metrics/derived_counters.xml
index 6e2b54fb4b..91c817dec7 100755
--- a/src/core/counters/metrics/derived_counters.xml
+++ b/src/core/counters/metrics/derived_counters.xml
@@ -103,6 +103,8 @@
 </gfx908_expr>
 
 <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="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>
   <metric name="TA_BUSY_min"                                            expr=min(TA_TA_BUSY,16) descr="TA block is busy. Min over TA instances."></metric>
@@ -294,6 +296,8 @@
   <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="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>
@@ -422,6 +426,9 @@
 </gfx940_expr>
 
 <gfx10_expr>
+  <metric name="MeanOccupancyPerCU" expr=GRBM_COUNT*0+SQ_LEVEL_WAVES*0+SQ_ACCUM_PREV/GRBM_GUI_ACTIVE/CU_NUM descr="Mean occupancy per compute unit."></metric>
+  <metric name="MeanOccupancyPerActiveCU" expr=GRBM_COUNT*0+SQ_LEVEL_WAVES*0+SQ_ACCUM_PREV*4/SQ_BUSY_CYCLES/CU_NUM descr="Mean occupancy per active compute unit."></metric>
+
   <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>
diff --git a/test/tool/gfx_metrics.xml b/test/tool/gfx_metrics.xml
index 7641b682c4..46137c5287 100755
--- a/test/tool/gfx_metrics.xml
+++ b/test/tool/gfx_metrics.xml
@@ -655,7 +655,10 @@
   <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_ACCUM_PREV" block=SQ event=1 descr="For counter N, increment by the value of counter N-1."></metric>
+  <metric name="SQ_BUSY_CYCLES" block=SQ event=3 descr="Clock cycles while SQ is reporting that it is busy. {nondeterministic, global, C2}"></metric>
   <metric name="SQ_WAVES" block=SQ event=4 descr="Count number of waves sent to SQs. {emulated, global, C1}"></metric>
+  <metric name="SQ_LEVEL_WAVES" block=SQ event=7 descr="Track the aggregated number of waves over certain period of time, Set next counter to ACCUM_PREV and divide by SQ_PERF_SEL_WAVES for average wave life."></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>
@@ -708,6 +711,8 @@
   <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_ACCUM_PREV" block=SQ event=1 descr="For counter N, increment by the value of counter N-1."></metric>
+  <metric name="SQ_BUSY_CYCLES" block=SQ event=3 descr="Clock cycles while SQ is reporting that it is busy. {nondeterministic, global, C2}"></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=24 descr="Number of clock cycles spent by waves in the SQs. Incremented by number of living (valid) waves each cycle. {nondeterministic, C1}"></metric>
   <metric name="SQ_WAIT_INST_ANY" block=SQ event=26 descr="Number of clock-cycles spent waiting for any instruction issue. In units of cycles. (nondeterministic)"></metric>
diff --git a/test/tool/metrics.xml b/test/tool/metrics.xml
index 120b2223e1..8ef59b7a1f 100755
--- a/test/tool/metrics.xml
+++ b/test/tool/metrics.xml
@@ -101,6 +101,9 @@
 </gfx908_expr>
 
 <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="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_TA_BUSY_sum"                                         expr=sum(TA_TA_BUSY,16) descr="TA block is busy. Perf_Windowing not supported for this counter. Sum over TA instances."></metric>
   <metric name="TA_TOTAL_WAVEFRONTS_sum"                                expr=sum(TA_TOTAL_WAVEFRONTS,16) descr="Total number of wavefronts processed by TA. Sum over TA instances."></metric>
   <metric name="TA_ADDR_STALLED_BY_TC_CYCLES_sum"                       expr=sum(TA_ADDR_STALLED_BY_TC_CYCLES,16) descr="Number of cycles addr path stalled by TC. Perf_Windowing not supported for this counter. Sum over TA instances."></metric>
@@ -211,6 +214,9 @@
 </gfx90a_expr>
 
 <gfx940_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="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="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>
 
@@ -342,6 +348,9 @@
 </gfx940_expr>
 
 <gfx10_expr>
+  <metric name="MeanOccupancyPerCU" expr=GRBM_COUNT*0+SQ_LEVEL_WAVES*0+SQ_ACCUM_PREV/GRBM_GUI_ACTIVE/CU_NUM descr="Mean occupancy per compute unit."></metric>
+  <metric name="MeanOccupancyPerActiveCU" expr=GRBM_COUNT*0+SQ_LEVEL_WAVES*0+SQ_ACCUM_PREV*4/SQ_BUSY_CYCLES/CU_NUM descr="Mean occupancy per active compute unit."></metric>
+
   <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>