AI/rocm-systems
2
0
Garpu 0
Kode Masalah Tarik Permintaan Actions Packages Projects Rilis Wiki Kegiatan

Fix Mi300 clk timings using GRBM_GUI_ACTIVE from #255

Telusuri Sumber 
Co-authored-by: Nick Curtis <nicholas.curtis@amd.com>
Signed-off-by: colramos-amd <colramos@amd.com>
This commit is contained in:
colramos-amd
2024-02-22 13:02:31 -06:00
committed by Cole Ramos
orang tua 2001908fd8
melakukan bd794abb12
52 mengubah file dengan 654 tambahan dan 614 penghapusan
Download Patch File Download Diff File Expand all files Collapse all files
src/omniperf_soc/analysis_configs/gfx906/0200_system-speed-of-light.yaml
+6 -6
Melihat File
@@ -69,16 +69,16 @@ Panel Config:
pop: ((100 * $numActiveCUs) / $cu_per_gpu)
tips:
SALU Utilization:
value: AVG(((100 * SQ_ACTIVE_INST_SCA) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
value: AVG(((100 * SQ_ACTIVE_INST_SCA) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
peak: 100
pop: AVG(((100 * SQ_ACTIVE_INST_SCA) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
pop: AVG(((100 * SQ_ACTIVE_INST_SCA) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
tips:
VALU Utilization:
value: AVG(((100 * SQ_ACTIVE_INST_VALU) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
value: AVG(((100 * SQ_ACTIVE_INST_VALU) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
peak: 100
pop: AVG(((100 * SQ_ACTIVE_INST_VALU) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
pop: AVG(((100 * SQ_ACTIVE_INST_VALU) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
tips:
MFMA Utilization:
value: None # No HW module
@@ -113,10 +113,10 @@ Panel Config:
pop: ((100 * AVG((SQ_INSTS / SQ_BUSY_CU_CYCLES))) / 5)
tips:
Wavefront Occupancy:
value: AVG((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE))
value: AVG((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD))
unit: Wavefronts
peak: ($max_waves_per_cu * $cu_per_gpu)
pop: (100 * AVG(((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE) / ($max_waves_per_cu
pop: (100 * AVG(((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD) / ($max_waves_per_cu
* $cu_per_gpu))))
coll_level: SQ_LEVEL_WAVES
tips:
src/omniperf_soc/analysis_configs/gfx906/0300_mem_chart.yaml
+2 -2
Melihat File
@@ -23,7 +23,7 @@ Panel Config:
#TODO: double check wave_occupancy
Wavefront Occupancy:
#alias: wave_occ_
value: ROUND(AVG((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE) / $numActiveCUs), 0)
value: ROUND(AVG((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD) / $numActiveCUs), 0)
coll_level: SQ_LEVEL_WAVES
tips:
Wave Life:
@@ -110,7 +110,7 @@ Panel Config:
LDS Util:
#alias: lds_util_
value:
ROUND(AVG(((100 * SQ_LDS_IDX_ACTIVE) / (GRBM_GUI_ACTIVE * $cu_per_gpu))),
ROUND(AVG(((100 * SQ_LDS_IDX_ACTIVE) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))),
0)
tips:
LDS Latency:
src/omniperf_soc/analysis_configs/gfx906/0600_shader-processor-input.yaml
+48 -48
Melihat File
@@ -20,33 +20,33 @@ Panel Config:
tips: Tips
metric:
Accelerator Utilization:
avg: AVG(100 * GRBM_GUI_ACTIVE / GRBM_COUNT)
min: MIN(100 * GRBM_GUI_ACTIVE / GRBM_COUNT)
max: MAX(100 * GRBM_GUI_ACTIVE / GRBM_COUNT)
avg: AVG(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
min: MIN(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
max: MAX(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
unit: Pct
tips:
Scheduler-Pipe Utilization:
avg: AVG(100 * SPI_CSN_BUSY / (GRBM_GUI_ACTIVE * $pipes_per_gpu * $se_per_gpu))
min: MIN(100 * SPI_CSN_BUSY / (GRBM_GUI_ACTIVE * $pipes_per_gpu * $se_per_gpu))
max: MAX(100 * SPI_CSN_BUSY / (GRBM_GUI_ACTIVE * $pipes_per_gpu * $se_per_gpu))
avg: AVG(100 * SPI_CSN_BUSY / ($GRBM_GUI_ACTIVE_PER_XCD * $pipes_per_gpu * $se_per_gpu))
min: MIN(100 * SPI_CSN_BUSY / ($GRBM_GUI_ACTIVE_PER_XCD * $pipes_per_gpu * $se_per_gpu))
max: MAX(100 * SPI_CSN_BUSY / ($GRBM_GUI_ACTIVE_PER_XCD * $pipes_per_gpu * $se_per_gpu))
unit: Pct
tips:
Workgroup Manager Utilization:
avg: AVG(100 * GRBM_SPI_BUSY / GRBM_GUI_ACTIVE)
min: MIN(100 * GRBM_SPI_BUSY / GRBM_GUI_ACTIVE)
max: MAX(100 * GRBM_SPI_BUSY / GRBM_GUI_ACTIVE)
avg: AVG(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
min: MIN(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
max: MAX(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
unit: Pct
tips:
Shader Engine Utilization:
avg: AVG(100 * SQ_BUSY_CYCLES / (GRBM_GUI_ACTIVE * $se_per_gpu))
min: MIN(100 * SQ_BUSY_CYCLES / (GRBM_GUI_ACTIVE * $se_per_gpu))
max: MAX(100 * SQ_BUSY_CYCLES / (GRBM_GUI_ACTIVE * $se_per_gpu))
avg: AVG(100 * SQ_BUSY_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $se_per_gpu))
min: MIN(100 * SQ_BUSY_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $se_per_gpu))
max: MAX(100 * SQ_BUSY_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $se_per_gpu))
unit: Pct
tips:
SIMD Utilization:
avg: AVG(100 * SQ_BUSY_CU_CYCLES / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(100 * SQ_BUSY_CU_CYCLES / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(100 * SQ_BUSY_CU_CYCLES / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(100 * SQ_BUSY_CU_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(100 * SQ_BUSY_CU_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(100 * SQ_BUSY_CU_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Dispatched Workgroups:
@@ -91,77 +91,77 @@ Panel Config:
tips: Tips
metric:
Not-scheduled Rate (Workgroup Manager):
avg: AVG((100 * SPI_RA_REQ_NO_ALLOC_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
avg: AVG((100 * SPI_RA_REQ_NO_ALLOC_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
min: MIN((100 * SPI_RA_REQ_NO_ALLOC_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
min: MIN((100 * SPI_RA_REQ_NO_ALLOC_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
max: MAX((100 * SPI_RA_REQ_NO_ALLOC_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
max: MAX((100 * SPI_RA_REQ_NO_ALLOC_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
unit: Pct
tips:
Not-scheduled Rate (Scheduler-Pipe):
avg: AVG((100 * SPI_RA_REQ_NO_ALLOC / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
avg: AVG((100 * SPI_RA_REQ_NO_ALLOC / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
min: MIN((100 * SPI_RA_REQ_NO_ALLOC / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
min: MIN((100 * SPI_RA_REQ_NO_ALLOC / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
max: MAX((100 * SPI_RA_REQ_NO_ALLOC / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
max: MAX((100 * SPI_RA_REQ_NO_ALLOC / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
unit: Pct
tips:
Scheduler-Pipe Stall Rate:
avg: AVG((((100 * SPI_RA_RES_STALL_CSN) / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
avg: AVG((((100 * SPI_RA_RES_STALL_CSN) / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None))
min: MIN((((100 * SPI_RA_RES_STALL_CSN) / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
min: MIN((((100 * SPI_RA_RES_STALL_CSN) / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None))
max: MAX((((100 * SPI_RA_RES_STALL_CSN) / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
max: MAX((((100 * SPI_RA_RES_STALL_CSN) / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None))
unit: Pct
tips:
Scratch Stall Rate:
avg: AVG((100 * SPI_RA_TMP_STALL_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY != 0) else None)
min: MIN((100 * SPI_RA_TMP_STALL_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY != 0) else None)
max: MAX((100 * SPI_RA_TMP_STALL_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY != 0) else None)
avg: AVG((100 * SPI_RA_TMP_STALL_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD != 0) else None)
min: MIN((100 * SPI_RA_TMP_STALL_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD != 0) else None)
max: MAX((100 * SPI_RA_TMP_STALL_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD != 0) else None)
unit: Pct
tips:
Insufficient SIMD Waveslots:
avg: AVG(100 * SPI_RA_WAVE_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(100 * SPI_RA_WAVE_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(100 * SPI_RA_WAVE_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(100 * SPI_RA_WAVE_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(100 * SPI_RA_WAVE_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(100 * SPI_RA_WAVE_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Insufficient SIMD VGPRs:
avg: AVG(100 * SPI_RA_VGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(100 * SPI_RA_VGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(100 * SPI_RA_VGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(100 * SPI_RA_VGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(100 * SPI_RA_VGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(100 * SPI_RA_VGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Insufficient SIMD SGPRs:
avg: AVG(100 * SPI_RA_SGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(100 * SPI_RA_SGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(100 * SPI_RA_SGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(100 * SPI_RA_SGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(100 * SPI_RA_SGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(100 * SPI_RA_SGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Insufficient CU LDS:
avg: AVG(400 * SPI_RA_LDS_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(400 * SPI_RA_LDS_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(400 * SPI_RA_LDS_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(400 * SPI_RA_LDS_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(400 * SPI_RA_LDS_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(400 * SPI_RA_LDS_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Insufficient CU Barriers:
avg: AVG(400 * SPI_RA_BAR_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(400 * SPI_RA_BAR_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(400 * SPI_RA_BAR_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(400 * SPI_RA_BAR_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(400 * SPI_RA_BAR_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(400 * SPI_RA_BAR_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Reached CU Workgroup Limit:
avg: AVG(400 * SPI_RA_TGLIM_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(400 * SPI_RA_TGLIM_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(400 * SPI_RA_TGLIM_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(400 * SPI_RA_TGLIM_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(400 * SPI_RA_TGLIM_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(400 * SPI_RA_TGLIM_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Reached CU Wavefront Limit:
avg: AVG(400 * SPI_RA_WVLIM_STALL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(400 * SPI_RA_WVLIM_STALL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(400 * SPI_RA_WVLIM_STALL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(400 * SPI_RA_WVLIM_STALL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(400 * SPI_RA_WVLIM_STALL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(400 * SPI_RA_WVLIM_STALL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
src/omniperf_soc/analysis_configs/gfx906/0700_wavefront-launch.yaml
+6 -6
Melihat File
@@ -98,9 +98,9 @@ Panel Config:
unit: ns
tips:
Kernel Time (Cycles):
avg: AVG(GRBM_GUI_ACTIVE)
min: MIN(GRBM_GUI_ACTIVE)
max: MAX(GRBM_GUI_ACTIVE)
avg: AVG($GRBM_GUI_ACTIVE_PER_XCD)
min: MIN($GRBM_GUI_ACTIVE_PER_XCD)
max: MAX($GRBM_GUI_ACTIVE_PER_XCD)
unit: Cycle
tips:
Instructions per wavefront:
@@ -134,9 +134,9 @@ Panel Config:
unit: (Cycles + $normUnit)
tips:
Wavefront Occupancy:
avg: AVG((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE))
min: MIN((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE))
max: MAX((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE))
avg: AVG((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD))
min: MIN((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD))
max: MAX((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD))
unit: Wavefronts
coll_level: SQ_LEVEL_WAVES
tips:
src/omniperf_soc/analysis_configs/gfx906/1100_compute-unit-compute-pipeline.yaml
+6 -6
Melihat File
@@ -92,15 +92,15 @@ Panel Config:
unit: Instr/cycle
tips:
SALU Utilization:
avg: AVG((((100 * SQ_ACTIVE_INST_SCA) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_SCA) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_SCA) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
avg: AVG((((100 * SQ_ACTIVE_INST_SCA) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_SCA) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_SCA) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
VALU Utilization:
avg: AVG((((100 * SQ_ACTIVE_INST_VALU) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_VALU) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_VALU) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
avg: AVG((((100 * SQ_ACTIVE_INST_VALU) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_VALU) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_VALU) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
VMEM Utilization:
src/omniperf_soc/analysis_configs/gfx906/1200_lds.yaml
+2 -2
Melihat File
@@ -18,11 +18,11 @@ Panel Config:
tips: Tips
metric:
Utilization:
value: AVG(((100 * SQ_LDS_IDX_ACTIVE) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
value: AVG(((100 * SQ_LDS_IDX_ACTIVE) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: Pct of Peak
tips:
Access Rate:
value: AVG(((200 * SQ_ACTIVE_INST_LDS) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
value: AVG(((200 * SQ_ACTIVE_INST_LDS) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: Pct of Peak
tips:
Theoretical Bandwidth (% of Peak):
src/omniperf_soc/analysis_configs/gfx906/1500_TA_and_TD.yaml
+18 -18
Melihat File
@@ -20,27 +20,27 @@ Panel Config:
tips: Tips
metric:
Address Processing Unit Busy:
avg: AVG(((100 * TA_TA_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TA_TA_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TA_TA_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TA_TA_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TA_TA_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TA_TA_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Address Stall:
avg: AVG(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Data Stall:
avg: AVG(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Data-Processor → Address Stall:
avg: AVG(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Total Instructions:
@@ -128,15 +128,15 @@ Panel Config:
tips: Tips
metric:
Data-Return Busy:
avg: AVG(((100 * TD_TD_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TD_TD_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TD_TD_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TD_TD_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TD_TD_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TD_TD_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Cache RAM → Data-Return Stall:
avg: AVG(((100 * TD_TC_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TD_TC_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TD_TC_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TD_TC_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TD_TC_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TD_TC_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Workgroup manager → Data-Return Stall:
src/omniperf_soc/analysis_configs/gfx906/1700_L2_cache.yaml
+1 -1
Melihat File
@@ -18,7 +18,7 @@ Panel Config:
tips: Tips
metric:
Utilization:
value: AVG(((TCC_BUSY_sum * 100) / (TO_INT($total_l2_chan) * GRBM_GUI_ACTIVE)))
value: AVG(((TCC_BUSY_sum * 100) / (TO_INT($total_l2_chan) * $GRBM_GUI_ACTIVE_PER_XCD)))
unit: pct
tips:
Bandwidth:
src/omniperf_soc/analysis_configs/gfx908/0200_system-speed-of-light.yaml
+6 -6
Melihat File
@@ -69,16 +69,16 @@ Panel Config:
pop: ((100 * $numActiveCUs) / $cu_per_gpu)
tips:
SALU Utilization:
value: AVG(((100 * SQ_ACTIVE_INST_SCA) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
value: AVG(((100 * SQ_ACTIVE_INST_SCA) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
peak: 100
pop: AVG(((100 * SQ_ACTIVE_INST_SCA) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
pop: AVG(((100 * SQ_ACTIVE_INST_SCA) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
tips:
VALU Utilization:
value: AVG(((100 * SQ_ACTIVE_INST_VALU) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
value: AVG(((100 * SQ_ACTIVE_INST_VALU) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
peak: 100
pop: AVG(((100 * SQ_ACTIVE_INST_VALU) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
pop: AVG(((100 * SQ_ACTIVE_INST_VALU) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
tips:
MFMA Utilization:
value: None # No HW module
@@ -113,10 +113,10 @@ Panel Config:
pop: ((100 * AVG((SQ_INSTS / SQ_BUSY_CU_CYCLES))) / 5)
tips:
Wavefront Occupancy:
value: AVG((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE))
value: AVG((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD))
unit: Wavefronts
peak: ($max_waves_per_cu * $cu_per_gpu)
pop: (100 * AVG(((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE) / ($max_waves_per_cu
pop: (100 * AVG(((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD) / ($max_waves_per_cu
* $cu_per_gpu))))
coll_level: SQ_LEVEL_WAVES
tips:
src/omniperf_soc/analysis_configs/gfx908/0300_mem_chart.yaml
+2 -2
Melihat File
@@ -23,7 +23,7 @@ Panel Config:
#TODO: double check wave_occupancy
Wavefront Occupancy:
#alias: wave_occ_
value: ROUND(AVG((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE) / $numActiveCUs), 0)
value: ROUND(AVG((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD) / $numActiveCUs), 0)
coll_level: SQ_LEVEL_WAVES
tips:
Wave Life:
@@ -110,7 +110,7 @@ Panel Config:
LDS Util:
#alias: lds_util_
value:
ROUND(AVG(((100 * SQ_LDS_IDX_ACTIVE) / (GRBM_GUI_ACTIVE * $cu_per_gpu))),
ROUND(AVG(((100 * SQ_LDS_IDX_ACTIVE) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))),
0)
tips:
LDS Latency:
src/omniperf_soc/analysis_configs/gfx908/0600_shader-processor-input.yaml
+48 -48
Melihat File
@@ -20,33 +20,33 @@ Panel Config:
tips: Tips
metric:
Accelerator Utilization:
avg: AVG(100 * GRBM_GUI_ACTIVE / GRBM_COUNT)
min: MIN(100 * GRBM_GUI_ACTIVE / GRBM_COUNT)
max: MAX(100 * GRBM_GUI_ACTIVE / GRBM_COUNT)
avg: AVG(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
min: MIN(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
max: MAX(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
unit: Pct
tips:
Scheduler-Pipe Utilization:
avg: AVG(100 * SPI_CSN_BUSY / (GRBM_GUI_ACTIVE * $pipes_per_gpu * $se_per_gpu))
min: MIN(100 * SPI_CSN_BUSY / (GRBM_GUI_ACTIVE * $pipes_per_gpu * $se_per_gpu))
max: MAX(100 * SPI_CSN_BUSY / (GRBM_GUI_ACTIVE * $pipes_per_gpu * $se_per_gpu))
avg: AVG(100 * SPI_CSN_BUSY / ($GRBM_GUI_ACTIVE_PER_XCD * $pipes_per_gpu * $se_per_gpu))
min: MIN(100 * SPI_CSN_BUSY / ($GRBM_GUI_ACTIVE_PER_XCD * $pipes_per_gpu * $se_per_gpu))
max: MAX(100 * SPI_CSN_BUSY / ($GRBM_GUI_ACTIVE_PER_XCD * $pipes_per_gpu * $se_per_gpu))
unit: Pct
tips:
Workgroup Manager Utilization:
avg: AVG(100 * GRBM_SPI_BUSY / GRBM_GUI_ACTIVE)
min: MIN(100 * GRBM_SPI_BUSY / GRBM_GUI_ACTIVE)
max: MAX(100 * GRBM_SPI_BUSY / GRBM_GUI_ACTIVE)
avg: AVG(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
min: MIN(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
max: MAX(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
unit: Pct
tips:
Shader Engine Utilization:
avg: AVG(100 * SQ_BUSY_CYCLES / (GRBM_GUI_ACTIVE * $se_per_gpu))
min: MIN(100 * SQ_BUSY_CYCLES / (GRBM_GUI_ACTIVE * $se_per_gpu))
max: MAX(100 * SQ_BUSY_CYCLES / (GRBM_GUI_ACTIVE * $se_per_gpu))
avg: AVG(100 * SQ_BUSY_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $se_per_gpu))
min: MIN(100 * SQ_BUSY_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $se_per_gpu))
max: MAX(100 * SQ_BUSY_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $se_per_gpu))
unit: Pct
tips:
SIMD Utilization:
avg: AVG(100 * SQ_BUSY_CU_CYCLES / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(100 * SQ_BUSY_CU_CYCLES / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(100 * SQ_BUSY_CU_CYCLES / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(100 * SQ_BUSY_CU_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(100 * SQ_BUSY_CU_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(100 * SQ_BUSY_CU_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Dispatched Workgroups:
@@ -91,77 +91,77 @@ Panel Config:
tips: Tips
metric:
Not-scheduled Rate (Workgroup Manager):
avg: AVG((100 * SPI_RA_REQ_NO_ALLOC_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
avg: AVG((100 * SPI_RA_REQ_NO_ALLOC_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
min: MIN((100 * SPI_RA_REQ_NO_ALLOC_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
min: MIN((100 * SPI_RA_REQ_NO_ALLOC_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
max: MAX((100 * SPI_RA_REQ_NO_ALLOC_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
max: MAX((100 * SPI_RA_REQ_NO_ALLOC_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
unit: Pct
tips:
Not-scheduled Rate (Scheduler-Pipe):
avg: AVG((100 * SPI_RA_REQ_NO_ALLOC / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
avg: AVG((100 * SPI_RA_REQ_NO_ALLOC / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
min: MIN((100 * SPI_RA_REQ_NO_ALLOC / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
min: MIN((100 * SPI_RA_REQ_NO_ALLOC / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
max: MAX((100 * SPI_RA_REQ_NO_ALLOC / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
max: MAX((100 * SPI_RA_REQ_NO_ALLOC / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
unit: Pct
tips:
Scheduler-Pipe Stall Rate:
avg: AVG((((100 * SPI_RA_RES_STALL_CSN) / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
avg: AVG((((100 * SPI_RA_RES_STALL_CSN) / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None))
min: MIN((((100 * SPI_RA_RES_STALL_CSN) / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
min: MIN((((100 * SPI_RA_RES_STALL_CSN) / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None))
max: MAX((((100 * SPI_RA_RES_STALL_CSN) / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
max: MAX((((100 * SPI_RA_RES_STALL_CSN) / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None))
unit: Pct
tips:
Scratch Stall Rate:
avg: AVG((100 * SPI_RA_TMP_STALL_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY != 0) else None)
min: MIN((100 * SPI_RA_TMP_STALL_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY != 0) else None)
max: MAX((100 * SPI_RA_TMP_STALL_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY != 0) else None)
avg: AVG((100 * SPI_RA_TMP_STALL_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD != 0) else None)
min: MIN((100 * SPI_RA_TMP_STALL_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD != 0) else None)
max: MAX((100 * SPI_RA_TMP_STALL_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD != 0) else None)
unit: Pct
tips:
Insufficient SIMD Waveslots:
avg: AVG(100 * SPI_RA_WAVE_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(100 * SPI_RA_WAVE_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(100 * SPI_RA_WAVE_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(100 * SPI_RA_WAVE_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(100 * SPI_RA_WAVE_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(100 * SPI_RA_WAVE_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Insufficient SIMD VGPRs:
avg: AVG(100 * SPI_RA_VGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(100 * SPI_RA_VGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(100 * SPI_RA_VGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(100 * SPI_RA_VGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(100 * SPI_RA_VGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(100 * SPI_RA_VGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Insufficient SIMD SGPRs:
avg: AVG(100 * SPI_RA_SGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(100 * SPI_RA_SGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(100 * SPI_RA_SGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(100 * SPI_RA_SGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(100 * SPI_RA_SGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(100 * SPI_RA_SGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Insufficient CU LDS:
avg: AVG(400 * SPI_RA_LDS_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(400 * SPI_RA_LDS_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(400 * SPI_RA_LDS_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(400 * SPI_RA_LDS_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(400 * SPI_RA_LDS_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(400 * SPI_RA_LDS_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Insufficient CU Barriers:
avg: AVG(400 * SPI_RA_BAR_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(400 * SPI_RA_BAR_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(400 * SPI_RA_BAR_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(400 * SPI_RA_BAR_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(400 * SPI_RA_BAR_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(400 * SPI_RA_BAR_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Reached CU Workgroup Limit:
avg: AVG(400 * SPI_RA_TGLIM_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(400 * SPI_RA_TGLIM_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(400 * SPI_RA_TGLIM_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(400 * SPI_RA_TGLIM_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(400 * SPI_RA_TGLIM_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(400 * SPI_RA_TGLIM_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Reached CU Wavefront Limit:
avg: AVG(400 * SPI_RA_WVLIM_STALL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(400 * SPI_RA_WVLIM_STALL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(400 * SPI_RA_WVLIM_STALL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(400 * SPI_RA_WVLIM_STALL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(400 * SPI_RA_WVLIM_STALL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(400 * SPI_RA_WVLIM_STALL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
src/omniperf_soc/analysis_configs/gfx908/0700_wavefront-launch.yaml
+6 -6
Melihat File
@@ -98,9 +98,9 @@ Panel Config:
unit: ns
tips:
Kernel Time (Cycles):
avg: AVG(GRBM_GUI_ACTIVE)
min: MIN(GRBM_GUI_ACTIVE)
max: MAX(GRBM_GUI_ACTIVE)
avg: AVG($GRBM_GUI_ACTIVE_PER_XCD)
min: MIN($GRBM_GUI_ACTIVE_PER_XCD)
max: MAX($GRBM_GUI_ACTIVE_PER_XCD)
unit: Cycle
tips:
Instructions per wavefront:
@@ -134,9 +134,9 @@ Panel Config:
unit: (Cycles + $normUnit)
tips:
Wavefront Occupancy:
avg: AVG((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE))
min: MIN((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE))
max: MAX((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE))
avg: AVG((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD))
min: MIN((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD))
max: MAX((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD))
unit: Wavefronts
coll_level: SQ_LEVEL_WAVES
tips:
src/omniperf_soc/analysis_configs/gfx908/1100_compute-unit-compute-pipeline.yaml
+6 -6
Melihat File
@@ -92,15 +92,15 @@ Panel Config:
unit: Instr/cycle
tips:
SALU Utilization:
avg: AVG((((100 * SQ_ACTIVE_INST_SCA) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_SCA) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_SCA) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
avg: AVG((((100 * SQ_ACTIVE_INST_SCA) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_SCA) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_SCA) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
VALU Utilization:
avg: AVG((((100 * SQ_ACTIVE_INST_VALU) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_VALU) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_VALU) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
avg: AVG((((100 * SQ_ACTIVE_INST_VALU) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_VALU) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_VALU) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
VMEM Utilization:
src/omniperf_soc/analysis_configs/gfx908/1200_lds.yaml
+2 -2
Melihat File
@@ -18,11 +18,11 @@ Panel Config:
tips: Tips
metric:
Utilization:
value: AVG(((100 * SQ_LDS_IDX_ACTIVE) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
value: AVG(((100 * SQ_LDS_IDX_ACTIVE) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: Pct of Peak
tips:
Access Rate:
value: AVG(((200 * SQ_ACTIVE_INST_LDS) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
value: AVG(((200 * SQ_ACTIVE_INST_LDS) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: Pct of Peak
tips:
unit: pct
src/omniperf_soc/analysis_configs/gfx908/1500_TA_and_TD.yaml
+18 -18
Melihat File
@@ -20,27 +20,27 @@ Panel Config:
tips: Tips
metric:
Address Processing Unit Busy:
avg: AVG(((100 * TA_TA_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TA_TA_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TA_TA_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TA_TA_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TA_TA_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TA_TA_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Address Stall:
avg: AVG(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Data Stall:
avg: AVG(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Data-Processor → Address Stall:
avg: AVG(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Total Instructions:
@@ -128,15 +128,15 @@ Panel Config:
tips: Tips
metric:
Data-Return Busy:
avg: AVG(((100 * TD_TD_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TD_TD_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TD_TD_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TD_TD_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TD_TD_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TD_TD_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Cache RAM → Data-Return Stall:
avg: AVG(((100 * TD_TC_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TD_TC_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TD_TC_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TD_TC_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TD_TC_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TD_TC_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Workgroup manager → Data-Return Stall:
src/omniperf_soc/analysis_configs/gfx908/1700_L2_cache.yaml
+1 -1
Melihat File
@@ -18,7 +18,7 @@ Panel Config:
tips: Tips
metric:
Utilization:
value: AVG(((TCC_BUSY_sum * 100) / (TO_INT($total_l2_chan) * GRBM_GUI_ACTIVE)))
value: AVG(((TCC_BUSY_sum * 100) / (TO_INT($total_l2_chan) * $GRBM_GUI_ACTIVE_PER_XCD)))
unit: pct
tips:
Bandwidth:
src/omniperf_soc/analysis_configs/gfx90a/0200_system-speed-of-light.yaml
+12 -12
Melihat File
@@ -84,36 +84,36 @@ Panel Config:
pop: ((100 * $numActiveCUs) / $cu_per_gpu)
tips:
SALU Utilization:
value: AVG(((100 * SQ_ACTIVE_INST_SCA) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
value: AVG(((100 * SQ_ACTIVE_INST_SCA) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
peak: 100
pop: AVG(((100 * SQ_ACTIVE_INST_SCA) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
pop: AVG(((100 * SQ_ACTIVE_INST_SCA) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
tips:
VALU Utilization:
value: AVG(((100 * SQ_ACTIVE_INST_VALU) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
value: AVG(((100 * SQ_ACTIVE_INST_VALU) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
peak: 100
pop: AVG(((100 * SQ_ACTIVE_INST_VALU) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
pop: AVG(((100 * SQ_ACTIVE_INST_VALU) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
tips:
MFMA Utilization:
value: AVG(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((GRBM_GUI_ACTIVE * $cu_per_gpu)
value: AVG(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / (($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)
* 4)))
unit: pct
peak: 100
pop: AVG(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((GRBM_GUI_ACTIVE * $cu_per_gpu)
pop: AVG(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / (($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)
* 4)))
tips:
VMEM Utilization:
value: AVG((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
value: AVG((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
peak: 100
pop: AVG((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
pop: AVG((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
tips:
Branch Utilization:
value: AVG((((100 * SQ_ACTIVE_INST_MISC) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
value: AVG((((100 * SQ_ACTIVE_INST_MISC) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
peak: 100
pop: AVG((((100 * SQ_ACTIVE_INST_MISC) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
pop: AVG((((100 * SQ_ACTIVE_INST_MISC) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
tips:
VALU Active Threads:
value: AVG(((SQ_THREAD_CYCLES_VALU / SQ_ACTIVE_INST_VALU) if (SQ_ACTIVE_INST_VALU
@@ -130,10 +130,10 @@ Panel Config:
pop: ((100 * AVG((SQ_INSTS / SQ_BUSY_CU_CYCLES))) / 5)
tips:
Wavefront Occupancy:
value: AVG((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE))
value: AVG((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD))
unit: Wavefronts
peak: ($max_waves_per_cu * $cu_per_gpu)
pop: (100 * AVG(((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE) / ($max_waves_per_cu
pop: (100 * AVG(((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD) / ($max_waves_per_cu
* $cu_per_gpu))))
coll_level: SQ_LEVEL_WAVES
tips:
src/omniperf_soc/analysis_configs/gfx90a/0300_mem_chart.yaml
+2 -2
Melihat File
@@ -23,7 +23,7 @@ Panel Config:
#TODO: double check wave_occupancy
Wavefront Occupancy:
#alias: wave_occ_
value: ROUND(AVG((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE) / $numActiveCUs), 0)
value: ROUND(AVG((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD) / $numActiveCUs), 0)
coll_level: SQ_LEVEL_WAVES
tips:
Wave Life:
@@ -111,7 +111,7 @@ Panel Config:
LDS Util:
#alias: lds_util_
value:
ROUND(AVG(((100 * SQ_LDS_IDX_ACTIVE) / (GRBM_GUI_ACTIVE * $cu_per_gpu))),
ROUND(AVG(((100 * SQ_LDS_IDX_ACTIVE) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))),
0)
tips:
LDS Latency:
src/omniperf_soc/analysis_configs/gfx90a/0600_shader-processor-input.yaml
+48 -48
Melihat File
@@ -20,33 +20,33 @@ Panel Config:
tips: Tips
metric:
Accelerator Utilization:
avg: AVG(100 * GRBM_GUI_ACTIVE / GRBM_COUNT)
min: MIN(100 * GRBM_GUI_ACTIVE / GRBM_COUNT)
max: MAX(100 * GRBM_GUI_ACTIVE / GRBM_COUNT)
avg: AVG(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
min: MIN(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
max: MAX(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
unit: Pct
tips:
Scheduler-Pipe Utilization:
avg: AVG(100 * SPI_CSN_BUSY / (GRBM_GUI_ACTIVE * $pipes_per_gpu * $se_per_gpu))
min: MIN(100 * SPI_CSN_BUSY / (GRBM_GUI_ACTIVE * $pipes_per_gpu * $se_per_gpu))
max: MAX(100 * SPI_CSN_BUSY / (GRBM_GUI_ACTIVE * $pipes_per_gpu * $se_per_gpu))
avg: AVG(100 * SPI_CSN_BUSY / ($GRBM_GUI_ACTIVE_PER_XCD * $pipes_per_gpu * $se_per_gpu))
min: MIN(100 * SPI_CSN_BUSY / ($GRBM_GUI_ACTIVE_PER_XCD * $pipes_per_gpu * $se_per_gpu))
max: MAX(100 * SPI_CSN_BUSY / ($GRBM_GUI_ACTIVE_PER_XCD * $pipes_per_gpu * $se_per_gpu))
unit: Pct
tips:
Workgroup Manager Utilization:
avg: AVG(100 * GRBM_SPI_BUSY / GRBM_GUI_ACTIVE)
min: MIN(100 * GRBM_SPI_BUSY / GRBM_GUI_ACTIVE)
max: MAX(100 * GRBM_SPI_BUSY / GRBM_GUI_ACTIVE)
avg: AVG(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
min: MIN(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
max: MAX(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
unit: Pct
tips:
Shader Engine Utilization:
avg: AVG(100 * SQ_BUSY_CYCLES / (GRBM_GUI_ACTIVE * $se_per_gpu))
min: MIN(100 * SQ_BUSY_CYCLES / (GRBM_GUI_ACTIVE * $se_per_gpu))
max: MAX(100 * SQ_BUSY_CYCLES / (GRBM_GUI_ACTIVE * $se_per_gpu))
avg: AVG(100 * SQ_BUSY_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $se_per_gpu))
min: MIN(100 * SQ_BUSY_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $se_per_gpu))
max: MAX(100 * SQ_BUSY_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $se_per_gpu))
unit: Pct
tips:
SIMD Utilization:
avg: AVG(100 * SQ_BUSY_CU_CYCLES / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(100 * SQ_BUSY_CU_CYCLES / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(100 * SQ_BUSY_CU_CYCLES / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(100 * SQ_BUSY_CU_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(100 * SQ_BUSY_CU_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(100 * SQ_BUSY_CU_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Dispatched Workgroups:
@@ -91,77 +91,77 @@ Panel Config:
tips: Tips
metric:
Not-scheduled Rate (Workgroup Manager):
avg: AVG((100 * SPI_RA_REQ_NO_ALLOC_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
avg: AVG((100 * SPI_RA_REQ_NO_ALLOC_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
min: MIN((100 * SPI_RA_REQ_NO_ALLOC_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
min: MIN((100 * SPI_RA_REQ_NO_ALLOC_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
max: MAX((100 * SPI_RA_REQ_NO_ALLOC_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
max: MAX((100 * SPI_RA_REQ_NO_ALLOC_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
unit: Pct
tips:
Not-scheduled Rate (Scheduler-Pipe):
avg: AVG((100 * SPI_RA_REQ_NO_ALLOC / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
avg: AVG((100 * SPI_RA_REQ_NO_ALLOC / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
min: MIN((100 * SPI_RA_REQ_NO_ALLOC / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
min: MIN((100 * SPI_RA_REQ_NO_ALLOC / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
max: MAX((100 * SPI_RA_REQ_NO_ALLOC / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
max: MAX((100 * SPI_RA_REQ_NO_ALLOC / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
unit: Pct
tips:
Scheduler-Pipe Stall Rate:
avg: AVG((((100 * SPI_RA_RES_STALL_CSN) / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
avg: AVG((((100 * SPI_RA_RES_STALL_CSN) / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None))
min: MIN((((100 * SPI_RA_RES_STALL_CSN) / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
min: MIN((((100 * SPI_RA_RES_STALL_CSN) / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None))
max: MAX((((100 * SPI_RA_RES_STALL_CSN) / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
max: MAX((((100 * SPI_RA_RES_STALL_CSN) / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None))
unit: Pct
tips:
Scratch Stall Rate:
avg: AVG((100 * SPI_RA_TMP_STALL_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY != 0) else None)
min: MIN((100 * SPI_RA_TMP_STALL_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY != 0) else None)
max: MAX((100 * SPI_RA_TMP_STALL_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY != 0) else None)
avg: AVG((100 * SPI_RA_TMP_STALL_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD != 0) else None)
min: MIN((100 * SPI_RA_TMP_STALL_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD != 0) else None)
max: MAX((100 * SPI_RA_TMP_STALL_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD != 0) else None)
unit: Pct
tips:
Insufficient SIMD Waveslots:
avg: AVG(100 * SPI_RA_WAVE_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(100 * SPI_RA_WAVE_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(100 * SPI_RA_WAVE_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(100 * SPI_RA_WAVE_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(100 * SPI_RA_WAVE_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(100 * SPI_RA_WAVE_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Insufficient SIMD VGPRs:
avg: AVG(100 * SPI_RA_VGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(100 * SPI_RA_VGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(100 * SPI_RA_VGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(100 * SPI_RA_VGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(100 * SPI_RA_VGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(100 * SPI_RA_VGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Insufficient SIMD SGPRs:
avg: AVG(100 * SPI_RA_SGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(100 * SPI_RA_SGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(100 * SPI_RA_SGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(100 * SPI_RA_SGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(100 * SPI_RA_SGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(100 * SPI_RA_SGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Insufficient CU LDS:
avg: AVG(400 * SPI_RA_LDS_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(400 * SPI_RA_LDS_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(400 * SPI_RA_LDS_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(400 * SPI_RA_LDS_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(400 * SPI_RA_LDS_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(400 * SPI_RA_LDS_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Insufficient CU Barriers:
avg: AVG(400 * SPI_RA_BAR_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(400 * SPI_RA_BAR_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(400 * SPI_RA_BAR_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(400 * SPI_RA_BAR_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(400 * SPI_RA_BAR_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(400 * SPI_RA_BAR_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Reached CU Workgroup Limit:
avg: AVG(400 * SPI_RA_TGLIM_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(400 * SPI_RA_TGLIM_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(400 * SPI_RA_TGLIM_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(400 * SPI_RA_TGLIM_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(400 * SPI_RA_TGLIM_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(400 * SPI_RA_TGLIM_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Reached CU Wavefront Limit:
avg: AVG(400 * SPI_RA_WVLIM_STALL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(400 * SPI_RA_WVLIM_STALL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(400 * SPI_RA_WVLIM_STALL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(400 * SPI_RA_WVLIM_STALL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(400 * SPI_RA_WVLIM_STALL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(400 * SPI_RA_WVLIM_STALL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
src/omniperf_soc/analysis_configs/gfx90a/0700_wavefront-launch.yaml
+6 -6
Melihat File
@@ -98,9 +98,9 @@ Panel Config:
unit: ns
tips:
Kernel Time (Cycles):
avg: AVG(GRBM_GUI_ACTIVE)
min: MIN(GRBM_GUI_ACTIVE)
max: MAX(GRBM_GUI_ACTIVE)
avg: AVG($GRBM_GUI_ACTIVE_PER_XCD)
min: MIN($GRBM_GUI_ACTIVE_PER_XCD)
max: MAX($GRBM_GUI_ACTIVE_PER_XCD)
unit: Cycle
tips:
Instructions per wavefront:
@@ -134,9 +134,9 @@ Panel Config:
unit: (Cycles + $normUnit)
tips:
Wavefront Occupancy:
avg: AVG((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE))
min: MIN((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE))
max: MAX((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE))
avg: AVG((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD))
min: MIN((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD))
max: MAX((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD))
unit: Wavefronts
coll_level: SQ_LEVEL_WAVES
tips:
src/omniperf_soc/analysis_configs/gfx90a/1100_compute-unit-compute-pipeline.yaml
+15 -15
Melihat File
@@ -107,27 +107,27 @@ Panel Config:
unit: Instr/cycle
tips:
SALU Utilization:
avg: AVG((((100 * SQ_ACTIVE_INST_SCA) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_SCA) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_SCA) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
avg: AVG((((100 * SQ_ACTIVE_INST_SCA) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_SCA) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_SCA) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
VALU Utilization:
avg: AVG((((100 * SQ_ACTIVE_INST_VALU) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_VALU) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_VALU) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
avg: AVG((((100 * SQ_ACTIVE_INST_VALU) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_VALU) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_VALU) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
VMEM Utilization:
avg: AVG((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
min: MIN((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
max: MAX((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
avg: AVG((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
min: MIN((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
max: MAX((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
Branch Utilization:
avg: AVG((((100 * SQ_ACTIVE_INST_MISC) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_MISC) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_MISC) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
avg: AVG((((100 * SQ_ACTIVE_INST_MISC) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_MISC) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_MISC) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
VALU Active Threads:
@@ -140,9 +140,9 @@ Panel Config:
unit: Threads
tips:
MFMA Utilization:
avg: AVG(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((4 * $cu_per_gpu) * GRBM_GUI_ACTIVE)))
min: MIN(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((4 * $cu_per_gpu) * GRBM_GUI_ACTIVE)))
max: MAX(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((4 * $cu_per_gpu) * GRBM_GUI_ACTIVE)))
avg: AVG(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((4 * $cu_per_gpu) * $GRBM_GUI_ACTIVE_PER_XCD)))
min: MIN(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((4 * $cu_per_gpu) * $GRBM_GUI_ACTIVE_PER_XCD)))
max: MAX(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((4 * $cu_per_gpu) * $GRBM_GUI_ACTIVE_PER_XCD)))
unit: pct
tips:
MFMA Instr Cycles:
src/omniperf_soc/analysis_configs/gfx90a/1200_lds.yaml
+2 -2
Melihat File
@@ -18,11 +18,11 @@ Panel Config:
tips: Tips
metric:
Utilization:
value: AVG(((100 * SQ_LDS_IDX_ACTIVE) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
value: AVG(((100 * SQ_LDS_IDX_ACTIVE) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: Pct of Peak
tips:
Access Rate:
value: AVG(((200 * SQ_ACTIVE_INST_LDS) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
value: AVG(((200 * SQ_ACTIVE_INST_LDS) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: Pct of Peak
tips:
Theoretical Bandwidth:
src/omniperf_soc/analysis_configs/gfx90a/1500_TA_and_TD.yaml
+21 -21
Melihat File
@@ -20,27 +20,27 @@ Panel Config:
tips: Tips
metric:
Address Processing Unit Busy:
avg: AVG(((100 * TA_TA_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TA_TA_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TA_TA_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TA_TA_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TA_TA_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TA_TA_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Address Stall:
avg: AVG(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Data Stall:
avg: AVG(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Data-Processor → Address Stall:
avg: AVG(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Total Instructions:
@@ -128,21 +128,21 @@ Panel Config:
tips: Tips
metric:
Data-Return Busy:
avg: AVG(((100 * TD_TD_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TD_TD_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TD_TD_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TD_TD_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TD_TD_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TD_TD_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Cache RAM → Data-Return Stall:
avg: AVG(((100 * TD_TC_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TD_TC_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TD_TC_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TD_TC_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TD_TC_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TD_TC_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Workgroup manager → Data-Return Stall:
avg: AVG(((100 * TD_SPI_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TD_SPI_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TD_SPI_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TD_SPI_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TD_SPI_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TD_SPI_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Coalescable Instructions:
src/omniperf_soc/analysis_configs/gfx90a/1700_L2_cache.yaml
+1 -1
Melihat File
@@ -18,7 +18,7 @@ Panel Config:
tips: Tips
metric:
Utilization:
value: AVG(((TCC_BUSY_sum * 100) / (TO_INT($total_l2_chan) * GRBM_GUI_ACTIVE)))
value: AVG(((TCC_BUSY_sum * 100) / (TO_INT($total_l2_chan) * $GRBM_GUI_ACTIVE_PER_XCD)))
unit: pct
tips:
Bandwidth:
src/omniperf_soc/analysis_configs/gfx940/0200_system-speed-of-light.yaml
+12 -12
Melihat File
@@ -84,36 +84,36 @@ Panel Config:
pop: ((100 * $numActiveCUs) / $cu_per_gpu)
tips:
SALU Utilization:
value: AVG(((100 * SQ_ACTIVE_INST_SCA) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
value: AVG(((100 * SQ_ACTIVE_INST_SCA) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
peak: 100
pop: AVG(((100 * SQ_ACTIVE_INST_SCA) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
pop: AVG(((100 * SQ_ACTIVE_INST_SCA) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
tips:
VALU Utilization:
value: AVG(((100 * SQ_ACTIVE_INST_VALU) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
value: AVG(((100 * SQ_ACTIVE_INST_VALU) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
peak: 100
pop: AVG(((100 * SQ_ACTIVE_INST_VALU) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
pop: AVG(((100 * SQ_ACTIVE_INST_VALU) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
tips:
MFMA Utilization:
value: AVG(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((GRBM_GUI_ACTIVE * $cu_per_gpu)
value: AVG(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / (($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)
* 4)))
unit: pct
peak: 100
pop: AVG(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((GRBM_GUI_ACTIVE * $cu_per_gpu)
pop: AVG(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / (($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)
* 4)))
tips:
VMEM Utilization:
value: AVG((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
value: AVG((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
peak: 100
pop: AVG((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
pop: AVG((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
tips:
Branch Utilization:
value: AVG((((100 * SQ_ACTIVE_INST_MISC) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
value: AVG((((100 * SQ_ACTIVE_INST_MISC) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
peak: 100
pop: AVG((((100 * SQ_ACTIVE_INST_MISC) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
pop: AVG((((100 * SQ_ACTIVE_INST_MISC) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
tips:
VALU Active Threads:
value: AVG(((SQ_THREAD_CYCLES_VALU / SQ_ACTIVE_INST_VALU) if (SQ_ACTIVE_INST_VALU
@@ -130,10 +130,10 @@ Panel Config:
pop: ((100 * AVG((SQ_INSTS / SQ_BUSY_CU_CYCLES))) / 5)
tips:
Wavefront Occupancy:
value: AVG((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE))
value: AVG((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD))
unit: Wavefronts
peak: ($max_waves_per_cu * $cu_per_gpu)
pop: (100 * AVG(((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE) / ($max_waves_per_cu
pop: (100 * AVG(((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD) / ($max_waves_per_cu
* $cu_per_gpu))))
coll_level: SQ_LEVEL_WAVES
tips:
src/omniperf_soc/analysis_configs/gfx940/0300_mem_chart.yaml
+2 -2
Melihat File
@@ -23,7 +23,7 @@ Panel Config:
#TODO: double check wave_occupancy
Wavefront Occupancy:
#alias: wave_occ_
value: ROUND(AVG((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE) / $numActiveCUs), 0)
value: ROUND(AVG((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD) / $numActiveCUs), 0)
coll_level: SQ_LEVEL_WAVES
tips:
Wave Life:
@@ -111,7 +111,7 @@ Panel Config:
LDS Util:
#alias: lds_util_
value:
ROUND(AVG(((100 * SQ_LDS_IDX_ACTIVE) / (GRBM_GUI_ACTIVE * $cu_per_gpu))),
ROUND(AVG(((100 * SQ_LDS_IDX_ACTIVE) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))),
0)
tips:
LDS Latency:
src/omniperf_soc/analysis_configs/gfx940/0600_shader-processor-input.yaml
+48 -48
Melihat File
@@ -20,33 +20,33 @@ Panel Config:
tips: Tips
metric:
Accelerator Utilization:
avg: AVG(100 * GRBM_GUI_ACTIVE / GRBM_COUNT)
min: MIN(100 * GRBM_GUI_ACTIVE / GRBM_COUNT)
max: MAX(100 * GRBM_GUI_ACTIVE / GRBM_COUNT)
avg: AVG(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
min: MIN(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
max: MAX(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
unit: Pct
tips:
Scheduler-Pipe Utilization:
avg: AVG(100 * SPI_CSN_BUSY / (GRBM_GUI_ACTIVE * $pipes_per_gpu * $se_per_gpu))
min: MIN(100 * SPI_CSN_BUSY / (GRBM_GUI_ACTIVE * $pipes_per_gpu * $se_per_gpu))
max: MAX(100 * SPI_CSN_BUSY / (GRBM_GUI_ACTIVE * $pipes_per_gpu * $se_per_gpu))
avg: AVG(100 * SPI_CSN_BUSY / ($GRBM_GUI_ACTIVE_PER_XCD * $pipes_per_gpu * $se_per_gpu))
min: MIN(100 * SPI_CSN_BUSY / ($GRBM_GUI_ACTIVE_PER_XCD * $pipes_per_gpu * $se_per_gpu))
max: MAX(100 * SPI_CSN_BUSY / ($GRBM_GUI_ACTIVE_PER_XCD * $pipes_per_gpu * $se_per_gpu))
unit: Pct
tips:
Workgroup Manager Utilization:
avg: AVG(100 * GRBM_SPI_BUSY / GRBM_GUI_ACTIVE)
min: MIN(100 * GRBM_SPI_BUSY / GRBM_GUI_ACTIVE)
max: MAX(100 * GRBM_SPI_BUSY / GRBM_GUI_ACTIVE)
avg: AVG(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
min: MIN(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
max: MAX(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
unit: Pct
tips:
Shader Engine Utilization:
avg: AVG(100 * SQ_BUSY_CYCLES / (GRBM_GUI_ACTIVE * $se_per_gpu))
min: MIN(100 * SQ_BUSY_CYCLES / (GRBM_GUI_ACTIVE * $se_per_gpu))
max: MAX(100 * SQ_BUSY_CYCLES / (GRBM_GUI_ACTIVE * $se_per_gpu))
avg: AVG(100 * SQ_BUSY_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $se_per_gpu))
min: MIN(100 * SQ_BUSY_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $se_per_gpu))
max: MAX(100 * SQ_BUSY_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $se_per_gpu))
unit: Pct
tips:
SIMD Utilization:
avg: AVG(100 * SQ_BUSY_CU_CYCLES / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(100 * SQ_BUSY_CU_CYCLES / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(100 * SQ_BUSY_CU_CYCLES / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(100 * SQ_BUSY_CU_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(100 * SQ_BUSY_CU_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(100 * SQ_BUSY_CU_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Dispatched Workgroups:
@@ -91,77 +91,77 @@ Panel Config:
tips: Tips
metric:
Not-scheduled Rate (Workgroup Manager):
avg: AVG((100 * SPI_RA_REQ_NO_ALLOC_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
avg: AVG((100 * SPI_RA_REQ_NO_ALLOC_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
min: MIN((100 * SPI_RA_REQ_NO_ALLOC_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
min: MIN((100 * SPI_RA_REQ_NO_ALLOC_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
max: MAX((100 * SPI_RA_REQ_NO_ALLOC_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
max: MAX((100 * SPI_RA_REQ_NO_ALLOC_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
unit: Pct
tips:
Not-scheduled Rate (Scheduler-Pipe):
avg: AVG((100 * SPI_RA_REQ_NO_ALLOC / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
avg: AVG((100 * SPI_RA_REQ_NO_ALLOC / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
min: MIN((100 * SPI_RA_REQ_NO_ALLOC / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
min: MIN((100 * SPI_RA_REQ_NO_ALLOC / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
max: MAX((100 * SPI_RA_REQ_NO_ALLOC / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
max: MAX((100 * SPI_RA_REQ_NO_ALLOC / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
unit: Pct
tips:
Scheduler-Pipe Stall Rate:
avg: AVG((((100 * SPI_RA_RES_STALL_CSN) / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
avg: AVG((((100 * SPI_RA_RES_STALL_CSN) / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None))
min: MIN((((100 * SPI_RA_RES_STALL_CSN) / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
min: MIN((((100 * SPI_RA_RES_STALL_CSN) / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None))
max: MAX((((100 * SPI_RA_RES_STALL_CSN) / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
max: MAX((((100 * SPI_RA_RES_STALL_CSN) / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None))
unit: Pct
tips:
Scratch Stall Rate:
avg: AVG((100 * SPI_RA_TMP_STALL_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY != 0) else None)
min: MIN((100 * SPI_RA_TMP_STALL_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY != 0) else None)
max: MAX((100 * SPI_RA_TMP_STALL_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY != 0) else None)
avg: AVG((100 * SPI_RA_TMP_STALL_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD != 0) else None)
min: MIN((100 * SPI_RA_TMP_STALL_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD != 0) else None)
max: MAX((100 * SPI_RA_TMP_STALL_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD != 0) else None)
unit: Pct
tips:
Insufficient SIMD Waveslots:
avg: AVG(100 * SPI_RA_WAVE_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(100 * SPI_RA_WAVE_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(100 * SPI_RA_WAVE_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(100 * SPI_RA_WAVE_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(100 * SPI_RA_WAVE_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(100 * SPI_RA_WAVE_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Insufficient SIMD VGPRs:
avg: AVG(100 * SPI_RA_VGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(100 * SPI_RA_VGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(100 * SPI_RA_VGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(100 * SPI_RA_VGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(100 * SPI_RA_VGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(100 * SPI_RA_VGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Insufficient SIMD SGPRs:
avg: AVG(100 * SPI_RA_SGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(100 * SPI_RA_SGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(100 * SPI_RA_SGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(100 * SPI_RA_SGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(100 * SPI_RA_SGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(100 * SPI_RA_SGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Insufficient CU LDS:
avg: AVG(400 * SPI_RA_LDS_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(400 * SPI_RA_LDS_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(400 * SPI_RA_LDS_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(400 * SPI_RA_LDS_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(400 * SPI_RA_LDS_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(400 * SPI_RA_LDS_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Insufficient CU Barriers:
avg: AVG(400 * SPI_RA_BAR_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(400 * SPI_RA_BAR_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(400 * SPI_RA_BAR_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(400 * SPI_RA_BAR_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(400 * SPI_RA_BAR_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(400 * SPI_RA_BAR_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Reached CU Workgroup Limit:
avg: AVG(400 * SPI_RA_TGLIM_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(400 * SPI_RA_TGLIM_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(400 * SPI_RA_TGLIM_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(400 * SPI_RA_TGLIM_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(400 * SPI_RA_TGLIM_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(400 * SPI_RA_TGLIM_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Reached CU Wavefront Limit:
avg: AVG(400 * SPI_RA_WVLIM_STALL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(400 * SPI_RA_WVLIM_STALL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(400 * SPI_RA_WVLIM_STALL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(400 * SPI_RA_WVLIM_STALL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(400 * SPI_RA_WVLIM_STALL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(400 * SPI_RA_WVLIM_STALL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
src/omniperf_soc/analysis_configs/gfx940/0700_wavefront-launch.yaml
+6 -6
Melihat File
@@ -98,9 +98,9 @@ Panel Config:
unit: ns
tips:
Kernel Time (Cycles):
avg: AVG(GRBM_GUI_ACTIVE)
min: MIN(GRBM_GUI_ACTIVE)
max: MAX(GRBM_GUI_ACTIVE)
avg: AVG($GRBM_GUI_ACTIVE_PER_XCD)
min: MIN($GRBM_GUI_ACTIVE_PER_XCD)
max: MAX($GRBM_GUI_ACTIVE_PER_XCD)
unit: Cycle
tips:
Instructions per wavefront:
@@ -134,9 +134,9 @@ Panel Config:
unit: (Cycles + $normUnit)
tips:
Wavefront Occupancy:
avg: AVG((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE))
min: MIN((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE))
max: MAX((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE))
avg: AVG((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD))
min: MIN((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD))
max: MAX((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD))
unit: Wavefronts
coll_level: SQ_LEVEL_WAVES
tips:
src/omniperf_soc/analysis_configs/gfx940/1100_compute-unit-compute-pipeline.yaml
+15 -15
Melihat File
@@ -107,27 +107,27 @@ Panel Config:
unit: Instr/cycle
tips:
SALU Utilization:
avg: AVG((((100 * SQ_ACTIVE_INST_SCA) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_SCA) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_SCA) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
avg: AVG((((100 * SQ_ACTIVE_INST_SCA) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_SCA) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_SCA) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
VALU Utilization:
avg: AVG((((100 * SQ_ACTIVE_INST_VALU) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_VALU) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_VALU) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
avg: AVG((((100 * SQ_ACTIVE_INST_VALU) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_VALU) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_VALU) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
VMEM Utilization:
avg: AVG((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
min: MIN((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
max: MAX((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
avg: AVG((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
min: MIN((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
max: MAX((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
Branch Utilization:
avg: AVG((((100 * SQ_ACTIVE_INST_MISC) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_MISC) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_MISC) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
avg: AVG((((100 * SQ_ACTIVE_INST_MISC) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_MISC) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_MISC) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
VALU Active Threads:
@@ -140,9 +140,9 @@ Panel Config:
unit: Threads
tips:
MFMA Utilization:
avg: AVG(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((4 * $cu_per_gpu) * GRBM_GUI_ACTIVE)))
min: MIN(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((4 * $cu_per_gpu) * GRBM_GUI_ACTIVE)))
max: MAX(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((4 * $cu_per_gpu) * GRBM_GUI_ACTIVE)))
avg: AVG(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((4 * $cu_per_gpu) * $GRBM_GUI_ACTIVE_PER_XCD)))
min: MIN(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((4 * $cu_per_gpu) * $GRBM_GUI_ACTIVE_PER_XCD)))
max: MAX(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((4 * $cu_per_gpu) * $GRBM_GUI_ACTIVE_PER_XCD)))
unit: pct
tips:
MFMA Instr Cycles:
src/omniperf_soc/analysis_configs/gfx940/1200_lds.yaml
+2 -2
Melihat File
@@ -18,11 +18,11 @@ Panel Config:
tips: Tips
metric:
Utilization:
value: AVG(((100 * SQ_LDS_IDX_ACTIVE) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
value: AVG(((100 * SQ_LDS_IDX_ACTIVE) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: Pct of Peak
tips:
Access Rate:
value: AVG(((200 * SQ_ACTIVE_INST_LDS) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
value: AVG(((200 * SQ_ACTIVE_INST_LDS) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: Pct of Peak
tips:
Theoretical Bandwidth (% of Peak):
src/omniperf_soc/analysis_configs/gfx940/1500_TA_and_TD.yaml
+21 -21
Melihat File
@@ -20,27 +20,27 @@ Panel Config:
tips: Tips
metric:
Address Processing Unit Busy:
avg: AVG(((100 * TA_TA_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TA_TA_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TA_TA_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TA_TA_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TA_TA_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TA_TA_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Address Stall:
avg: AVG(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Data Stall:
avg: AVG(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Data-Processor → Address Stall:
avg: AVG(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Total Instructions:
@@ -128,21 +128,21 @@ Panel Config:
tips: Tips
metric:
Data-Return Busy:
avg: AVG(((100 * TD_TD_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TD_TD_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TD_TD_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TD_TD_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TD_TD_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TD_TD_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Cache RAM → Data-Return Stall:
avg: AVG(((100 * TD_TC_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TD_TC_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TD_TC_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TD_TC_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TD_TC_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TD_TC_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Workgroup manager → Data-Return Stall:
avg: AVG(((100 * TD_SPI_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TD_SPI_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TD_SPI_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TD_SPI_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TD_SPI_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TD_SPI_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Coalescable Instructions:
src/omniperf_soc/analysis_configs/gfx940/1700_L2_cache.yaml
+1 -1
Melihat File
@@ -18,7 +18,7 @@ Panel Config:
tips: Tips
metric:
Utilization:
value: AVG(((TCC_BUSY_sum * 100) / (TO_INT($total_l2_chan) * GRBM_GUI_ACTIVE)))
value: AVG(((TCC_BUSY_sum * 100) / (TO_INT($total_l2_chan) * $GRBM_GUI_ACTIVE_PER_XCD)))
unit: pct
tips:
Bandwidth:
src/omniperf_soc/analysis_configs/gfx941/0200_system-speed-of-light.yaml
+12 -12
Melihat File
@@ -84,36 +84,36 @@ Panel Config:
pop: ((100 * $numActiveCUs) / $cu_per_gpu)
tips:
SALU Utilization:
value: AVG(((100 * SQ_ACTIVE_INST_SCA) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
value: AVG(((100 * SQ_ACTIVE_INST_SCA) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
peak: 100
pop: AVG(((100 * SQ_ACTIVE_INST_SCA) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
pop: AVG(((100 * SQ_ACTIVE_INST_SCA) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
tips:
VALU Utilization:
value: AVG(((100 * SQ_ACTIVE_INST_VALU) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
value: AVG(((100 * SQ_ACTIVE_INST_VALU) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
peak: 100
pop: AVG(((100 * SQ_ACTIVE_INST_VALU) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
pop: AVG(((100 * SQ_ACTIVE_INST_VALU) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
tips:
MFMA Utilization:
value: AVG(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((GRBM_GUI_ACTIVE * $cu_per_gpu)
value: AVG(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / (($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)
* 4)))
unit: pct
peak: 100
pop: AVG(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((GRBM_GUI_ACTIVE * $cu_per_gpu)
pop: AVG(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / (($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)
* 4)))
tips:
VMEM Utilization:
value: AVG((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
value: AVG((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
peak: 100
pop: AVG((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
pop: AVG((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
tips:
Branch Utilization:
value: AVG((((100 * SQ_ACTIVE_INST_MISC) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
value: AVG((((100 * SQ_ACTIVE_INST_MISC) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
peak: 100
pop: AVG((((100 * SQ_ACTIVE_INST_MISC) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
pop: AVG((((100 * SQ_ACTIVE_INST_MISC) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
tips:
VALU Active Threads:
value: AVG(((SQ_THREAD_CYCLES_VALU / SQ_ACTIVE_INST_VALU) if (SQ_ACTIVE_INST_VALU
@@ -130,10 +130,10 @@ Panel Config:
pop: ((100 * AVG((SQ_INSTS / SQ_BUSY_CU_CYCLES))) / 5)
tips:
Wavefront Occupancy:
value: AVG((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE))
value: AVG((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD))
unit: Wavefronts
peak: ($max_waves_per_cu * $cu_per_gpu)
pop: (100 * AVG(((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE) / ($max_waves_per_cu
pop: (100 * AVG(((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD) / ($max_waves_per_cu
* $cu_per_gpu))))
coll_level: SQ_LEVEL_WAVES
tips:
src/omniperf_soc/analysis_configs/gfx941/0300_mem_chart.yaml
+2 -2
Melihat File
@@ -23,7 +23,7 @@ Panel Config:
#TODO: double check wave_occupancy
Wavefront Occupancy:
#alias: wave_occ_
value: ROUND(AVG((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE) / $numActiveCUs), 0)
value: ROUND(AVG((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD) / $numActiveCUs), 0)
coll_level: SQ_LEVEL_WAVES
tips:
Wave Life:
@@ -111,7 +111,7 @@ Panel Config:
LDS Util:
#alias: lds_util_
value:
ROUND(AVG(((100 * SQ_LDS_IDX_ACTIVE) / (GRBM_GUI_ACTIVE * $cu_per_gpu))),
ROUND(AVG(((100 * SQ_LDS_IDX_ACTIVE) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))),
0)
tips:
LDS Latency:
src/omniperf_soc/analysis_configs/gfx941/0600_shader-processor-input.yaml
+48 -48
Melihat File
@@ -20,33 +20,33 @@ Panel Config:
tips: Tips
metric:
Accelerator Utilization:
avg: AVG(100 * GRBM_GUI_ACTIVE / GRBM_COUNT)
min: MIN(100 * GRBM_GUI_ACTIVE / GRBM_COUNT)
max: MAX(100 * GRBM_GUI_ACTIVE / GRBM_COUNT)
avg: AVG(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
min: MIN(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
max: MAX(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
unit: Pct
tips:
Scheduler-Pipe Utilization:
avg: AVG(100 * SPI_CSN_BUSY / (GRBM_GUI_ACTIVE * $pipes_per_gpu * $se_per_gpu))
min: MIN(100 * SPI_CSN_BUSY / (GRBM_GUI_ACTIVE * $pipes_per_gpu * $se_per_gpu))
max: MAX(100 * SPI_CSN_BUSY / (GRBM_GUI_ACTIVE * $pipes_per_gpu * $se_per_gpu))
avg: AVG(100 * SPI_CSN_BUSY / ($GRBM_GUI_ACTIVE_PER_XCD * $pipes_per_gpu * $se_per_gpu))
min: MIN(100 * SPI_CSN_BUSY / ($GRBM_GUI_ACTIVE_PER_XCD * $pipes_per_gpu * $se_per_gpu))
max: MAX(100 * SPI_CSN_BUSY / ($GRBM_GUI_ACTIVE_PER_XCD * $pipes_per_gpu * $se_per_gpu))
unit: Pct
tips:
Workgroup Manager Utilization:
avg: AVG(100 * GRBM_SPI_BUSY / GRBM_GUI_ACTIVE)
min: MIN(100 * GRBM_SPI_BUSY / GRBM_GUI_ACTIVE)
max: MAX(100 * GRBM_SPI_BUSY / GRBM_GUI_ACTIVE)
avg: AVG(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
min: MIN(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
max: MAX(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
unit: Pct
tips:
Shader Engine Utilization:
avg: AVG(100 * SQ_BUSY_CYCLES / (GRBM_GUI_ACTIVE * $se_per_gpu))
min: MIN(100 * SQ_BUSY_CYCLES / (GRBM_GUI_ACTIVE * $se_per_gpu))
max: MAX(100 * SQ_BUSY_CYCLES / (GRBM_GUI_ACTIVE * $se_per_gpu))
avg: AVG(100 * SQ_BUSY_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $se_per_gpu))
min: MIN(100 * SQ_BUSY_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $se_per_gpu))
max: MAX(100 * SQ_BUSY_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $se_per_gpu))
unit: Pct
tips:
SIMD Utilization:
avg: AVG(100 * SQ_BUSY_CU_CYCLES / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(100 * SQ_BUSY_CU_CYCLES / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(100 * SQ_BUSY_CU_CYCLES / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(100 * SQ_BUSY_CU_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(100 * SQ_BUSY_CU_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(100 * SQ_BUSY_CU_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Dispatched Workgroups:
@@ -91,77 +91,77 @@ Panel Config:
tips: Tips
metric:
Not-scheduled Rate (Workgroup Manager):
avg: AVG((100 * SPI_RA_REQ_NO_ALLOC_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
avg: AVG((100 * SPI_RA_REQ_NO_ALLOC_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
min: MIN((100 * SPI_RA_REQ_NO_ALLOC_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
min: MIN((100 * SPI_RA_REQ_NO_ALLOC_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
max: MAX((100 * SPI_RA_REQ_NO_ALLOC_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
max: MAX((100 * SPI_RA_REQ_NO_ALLOC_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
unit: Pct
tips:
Not-scheduled Rate (Scheduler-Pipe):
avg: AVG((100 * SPI_RA_REQ_NO_ALLOC / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
avg: AVG((100 * SPI_RA_REQ_NO_ALLOC / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
min: MIN((100 * SPI_RA_REQ_NO_ALLOC / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
min: MIN((100 * SPI_RA_REQ_NO_ALLOC / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
max: MAX((100 * SPI_RA_REQ_NO_ALLOC / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
max: MAX((100 * SPI_RA_REQ_NO_ALLOC / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
unit: Pct
tips:
Scheduler-Pipe Stall Rate:
avg: AVG((((100 * SPI_RA_RES_STALL_CSN) / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
avg: AVG((((100 * SPI_RA_RES_STALL_CSN) / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None))
min: MIN((((100 * SPI_RA_RES_STALL_CSN) / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
min: MIN((((100 * SPI_RA_RES_STALL_CSN) / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None))
max: MAX((((100 * SPI_RA_RES_STALL_CSN) / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
max: MAX((((100 * SPI_RA_RES_STALL_CSN) / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None))
unit: Pct
tips:
Scratch Stall Rate:
avg: AVG((100 * SPI_RA_TMP_STALL_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY != 0) else None)
min: MIN((100 * SPI_RA_TMP_STALL_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY != 0) else None)
max: MAX((100 * SPI_RA_TMP_STALL_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY != 0) else None)
avg: AVG((100 * SPI_RA_TMP_STALL_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD != 0) else None)
min: MIN((100 * SPI_RA_TMP_STALL_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD != 0) else None)
max: MAX((100 * SPI_RA_TMP_STALL_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD != 0) else None)
unit: Pct
tips:
Insufficient SIMD Waveslots:
avg: AVG(100 * SPI_RA_WAVE_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(100 * SPI_RA_WAVE_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(100 * SPI_RA_WAVE_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(100 * SPI_RA_WAVE_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(100 * SPI_RA_WAVE_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(100 * SPI_RA_WAVE_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Insufficient SIMD VGPRs:
avg: AVG(100 * SPI_RA_VGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(100 * SPI_RA_VGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(100 * SPI_RA_VGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(100 * SPI_RA_VGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(100 * SPI_RA_VGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(100 * SPI_RA_VGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Insufficient SIMD SGPRs:
avg: AVG(100 * SPI_RA_SGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(100 * SPI_RA_SGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(100 * SPI_RA_SGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(100 * SPI_RA_SGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(100 * SPI_RA_SGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(100 * SPI_RA_SGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Insufficient CU LDS:
avg: AVG(400 * SPI_RA_LDS_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(400 * SPI_RA_LDS_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(400 * SPI_RA_LDS_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(400 * SPI_RA_LDS_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(400 * SPI_RA_LDS_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(400 * SPI_RA_LDS_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Insufficient CU Barriers:
avg: AVG(400 * SPI_RA_BAR_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(400 * SPI_RA_BAR_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(400 * SPI_RA_BAR_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(400 * SPI_RA_BAR_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(400 * SPI_RA_BAR_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(400 * SPI_RA_BAR_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Reached CU Workgroup Limit:
avg: AVG(400 * SPI_RA_TGLIM_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(400 * SPI_RA_TGLIM_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(400 * SPI_RA_TGLIM_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(400 * SPI_RA_TGLIM_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(400 * SPI_RA_TGLIM_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(400 * SPI_RA_TGLIM_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Reached CU Wavefront Limit:
avg: AVG(400 * SPI_RA_WVLIM_STALL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(400 * SPI_RA_WVLIM_STALL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(400 * SPI_RA_WVLIM_STALL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(400 * SPI_RA_WVLIM_STALL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(400 * SPI_RA_WVLIM_STALL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(400 * SPI_RA_WVLIM_STALL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
src/omniperf_soc/analysis_configs/gfx941/0700_wavefront-launch.yaml
+6 -6
Melihat File
@@ -98,9 +98,9 @@ Panel Config:
unit: ns
tips:
Kernel Time (Cycles):
avg: AVG(GRBM_GUI_ACTIVE)
min: MIN(GRBM_GUI_ACTIVE)
max: MAX(GRBM_GUI_ACTIVE)
avg: AVG($GRBM_GUI_ACTIVE_PER_XCD)
min: MIN($GRBM_GUI_ACTIVE_PER_XCD)
max: MAX($GRBM_GUI_ACTIVE_PER_XCD)
unit: Cycle
tips:
Instructions per wavefront:
@@ -134,9 +134,9 @@ Panel Config:
unit: (Cycles + $normUnit)
tips:
Wavefront Occupancy:
avg: AVG((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE))
min: MIN((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE))
max: MAX((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE))
avg: AVG((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD))
min: MIN((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD))
max: MAX((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD))
unit: Wavefronts
coll_level: SQ_LEVEL_WAVES
tips:
src/omniperf_soc/analysis_configs/gfx941/1100_compute-unit-compute-pipeline.yaml
+15 -15
Melihat File
@@ -107,27 +107,27 @@ Panel Config:
unit: Instr/cycle
tips:
SALU Utilization:
avg: AVG((((100 * SQ_ACTIVE_INST_SCA) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_SCA) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_SCA) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
avg: AVG((((100 * SQ_ACTIVE_INST_SCA) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_SCA) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_SCA) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
VALU Utilization:
avg: AVG((((100 * SQ_ACTIVE_INST_VALU) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_VALU) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_VALU) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
avg: AVG((((100 * SQ_ACTIVE_INST_VALU) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_VALU) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_VALU) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
VMEM Utilization:
avg: AVG((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
min: MIN((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
max: MAX((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
avg: AVG((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
min: MIN((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
max: MAX((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
Branch Utilization:
avg: AVG((((100 * SQ_ACTIVE_INST_MISC) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_MISC) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_MISC) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
avg: AVG((((100 * SQ_ACTIVE_INST_MISC) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_MISC) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_MISC) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
VALU Active Threads:
@@ -140,9 +140,9 @@ Panel Config:
unit: Threads
tips:
MFMA Utilization:
avg: AVG(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((4 * $cu_per_gpu) * GRBM_GUI_ACTIVE)))
min: MIN(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((4 * $cu_per_gpu) * GRBM_GUI_ACTIVE)))
max: MAX(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((4 * $cu_per_gpu) * GRBM_GUI_ACTIVE)))
avg: AVG(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((4 * $cu_per_gpu) * $GRBM_GUI_ACTIVE_PER_XCD)))
min: MIN(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((4 * $cu_per_gpu) * $GRBM_GUI_ACTIVE_PER_XCD)))
max: MAX(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((4 * $cu_per_gpu) * $GRBM_GUI_ACTIVE_PER_XCD)))
unit: pct
tips:
MFMA Instr Cycles:
src/omniperf_soc/analysis_configs/gfx941/1200_lds.yaml
+2 -2
Melihat File
@@ -18,11 +18,11 @@ Panel Config:
tips: Tips
metric:
Utilization:
value: AVG(((100 * SQ_LDS_IDX_ACTIVE) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
value: AVG(((100 * SQ_LDS_IDX_ACTIVE) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: Pct of Peak
tips:
Access Rate:
value: AVG(((200 * SQ_ACTIVE_INST_LDS) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
value: AVG(((200 * SQ_ACTIVE_INST_LDS) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: Pct of Peak
tips:
Theoretical Bandwidth (% of Peak):
src/omniperf_soc/analysis_configs/gfx941/1500_TA_and_TD.yaml
+21 -21
Melihat File
@@ -20,27 +20,27 @@ Panel Config:
tips: Tips
metric:
Address Processing Unit Busy:
avg: AVG(((100 * TA_TA_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TA_TA_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TA_TA_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TA_TA_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TA_TA_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TA_TA_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Address Stall:
avg: AVG(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Data Stall:
avg: AVG(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Data-Processor → Address Stall:
avg: AVG(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Total Instructions:
@@ -128,21 +128,21 @@ Panel Config:
tips: Tips
metric:
Data-Return Busy:
avg: AVG(((100 * TD_TD_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TD_TD_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TD_TD_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TD_TD_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TD_TD_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TD_TD_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Cache RAM → Data-Return Stall:
avg: AVG(((100 * TD_TC_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TD_TC_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TD_TC_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TD_TC_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TD_TC_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TD_TC_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Workgroup manager → Data-Return Stall:
avg: AVG(((100 * TD_SPI_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TD_SPI_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TD_SPI_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TD_SPI_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TD_SPI_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TD_SPI_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Coalescable Instructions:
src/omniperf_soc/analysis_configs/gfx941/1700_L2_cache.yaml
+1 -1
Melihat File
@@ -18,7 +18,7 @@ Panel Config:
tips: Tips
metric:
Utilization:
value: AVG(((TCC_BUSY_sum * 100) / (TO_INT($total_l2_chan) * GRBM_GUI_ACTIVE)))
value: AVG(((TCC_BUSY_sum * 100) / (TO_INT($total_l2_chan) * $GRBM_GUI_ACTIVE_PER_XCD)))
unit: pct
tips:
Bandwidth:
src/omniperf_soc/analysis_configs/gfx942/0200_system-speed-of-light.yaml
+12 -12
Melihat File
@@ -84,36 +84,36 @@ Panel Config:
pop: ((100 * $numActiveCUs) / $cu_per_gpu)
tips:
SALU Utilization:
value: AVG(((100 * SQ_ACTIVE_INST_SCA) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
value: AVG(((100 * SQ_ACTIVE_INST_SCA) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
peak: 100
pop: AVG(((100 * SQ_ACTIVE_INST_SCA) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
pop: AVG(((100 * SQ_ACTIVE_INST_SCA) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
tips:
VALU Utilization:
value: AVG(((100 * SQ_ACTIVE_INST_VALU) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
value: AVG(((100 * SQ_ACTIVE_INST_VALU) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
peak: 100
pop: AVG(((100 * SQ_ACTIVE_INST_VALU) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
pop: AVG(((100 * SQ_ACTIVE_INST_VALU) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
tips:
MFMA Utilization:
value: AVG(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((GRBM_GUI_ACTIVE * $cu_per_gpu)
value: AVG(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / (($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)
* 4)))
unit: pct
peak: 100
pop: AVG(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((GRBM_GUI_ACTIVE * $cu_per_gpu)
pop: AVG(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / (($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)
* 4)))
tips:
VMEM Utilization:
value: AVG((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
value: AVG((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
peak: 100
pop: AVG((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
pop: AVG((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
tips:
Branch Utilization:
value: AVG((((100 * SQ_ACTIVE_INST_MISC) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
value: AVG((((100 * SQ_ACTIVE_INST_MISC) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
peak: 100
pop: AVG((((100 * SQ_ACTIVE_INST_MISC) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
pop: AVG((((100 * SQ_ACTIVE_INST_MISC) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
tips:
VALU Active Threads:
value: AVG(((SQ_THREAD_CYCLES_VALU / SQ_ACTIVE_INST_VALU) if (SQ_ACTIVE_INST_VALU
@@ -130,10 +130,10 @@ Panel Config:
pop: ((100 * AVG((SQ_INSTS / SQ_BUSY_CU_CYCLES))) / 5)
tips:
Wavefront Occupancy:
value: AVG((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE))
value: AVG((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD))
unit: Wavefronts
peak: ($max_waves_per_cu * $cu_per_gpu)
pop: (100 * AVG(((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE) / ($max_waves_per_cu
pop: (100 * AVG(((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD) / ($max_waves_per_cu
* $cu_per_gpu))))
coll_level: SQ_LEVEL_WAVES
tips:
src/omniperf_soc/analysis_configs/gfx942/0300_mem_chart.yaml
+2 -2
Melihat File
@@ -23,7 +23,7 @@ Panel Config:
#TODO: double check wave_occupancy
Wavefront Occupancy:
#alias: wave_occ_
value: ROUND(AVG((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE) / $numActiveCUs), 0)
value: ROUND(AVG((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD) / $numActiveCUs), 0)
coll_level: SQ_LEVEL_WAVES
tips:
Wave Life:
@@ -111,7 +111,7 @@ Panel Config:
LDS Util:
#alias: lds_util_
value:
ROUND(AVG(((100 * SQ_LDS_IDX_ACTIVE) / (GRBM_GUI_ACTIVE * $cu_per_gpu))),
ROUND(AVG(((100 * SQ_LDS_IDX_ACTIVE) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))),
0)
tips:
LDS Latency:
src/omniperf_soc/analysis_configs/gfx942/0600_shader-processor-input.yaml
+48 -48
Melihat File
@@ -20,33 +20,33 @@ Panel Config:
tips: Tips
metric:
Accelerator Utilization:
avg: AVG(100 * GRBM_GUI_ACTIVE / GRBM_COUNT)
min: MIN(100 * GRBM_GUI_ACTIVE / GRBM_COUNT)
max: MAX(100 * GRBM_GUI_ACTIVE / GRBM_COUNT)
avg: AVG(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
min: MIN(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
max: MAX(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
unit: Pct
tips:
Scheduler-Pipe Utilization:
avg: AVG(100 * SPI_CSN_BUSY / (GRBM_GUI_ACTIVE * $pipes_per_gpu * $se_per_gpu))
min: MIN(100 * SPI_CSN_BUSY / (GRBM_GUI_ACTIVE * $pipes_per_gpu * $se_per_gpu))
max: MAX(100 * SPI_CSN_BUSY / (GRBM_GUI_ACTIVE * $pipes_per_gpu * $se_per_gpu))
avg: AVG(100 * SPI_CSN_BUSY / ($GRBM_GUI_ACTIVE_PER_XCD * $pipes_per_gpu * $se_per_gpu))
min: MIN(100 * SPI_CSN_BUSY / ($GRBM_GUI_ACTIVE_PER_XCD * $pipes_per_gpu * $se_per_gpu))
max: MAX(100 * SPI_CSN_BUSY / ($GRBM_GUI_ACTIVE_PER_XCD * $pipes_per_gpu * $se_per_gpu))
unit: Pct
tips:
Workgroup Manager Utilization:
avg: AVG(100 * GRBM_SPI_BUSY / GRBM_GUI_ACTIVE)
min: MIN(100 * GRBM_SPI_BUSY / GRBM_GUI_ACTIVE)
max: MAX(100 * GRBM_SPI_BUSY / GRBM_GUI_ACTIVE)
avg: AVG(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
min: MIN(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
max: MAX(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
unit: Pct
tips:
Shader Engine Utilization:
avg: AVG(100 * SQ_BUSY_CYCLES / (GRBM_GUI_ACTIVE * $se_per_gpu))
min: MIN(100 * SQ_BUSY_CYCLES / (GRBM_GUI_ACTIVE * $se_per_gpu))
max: MAX(100 * SQ_BUSY_CYCLES / (GRBM_GUI_ACTIVE * $se_per_gpu))
avg: AVG(100 * SQ_BUSY_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $se_per_gpu))
min: MIN(100 * SQ_BUSY_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $se_per_gpu))
max: MAX(100 * SQ_BUSY_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $se_per_gpu))
unit: Pct
tips:
SIMD Utilization:
avg: AVG(100 * SQ_BUSY_CU_CYCLES / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(100 * SQ_BUSY_CU_CYCLES / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(100 * SQ_BUSY_CU_CYCLES / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(100 * SQ_BUSY_CU_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(100 * SQ_BUSY_CU_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(100 * SQ_BUSY_CU_CYCLES / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Dispatched Workgroups:
@@ -91,77 +91,77 @@ Panel Config:
tips: Tips
metric:
Not-scheduled Rate (Workgroup Manager):
avg: AVG((100 * SPI_RA_REQ_NO_ALLOC_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
avg: AVG((100 * SPI_RA_REQ_NO_ALLOC_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
min: MIN((100 * SPI_RA_REQ_NO_ALLOC_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
min: MIN((100 * SPI_RA_REQ_NO_ALLOC_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
max: MAX((100 * SPI_RA_REQ_NO_ALLOC_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
max: MAX((100 * SPI_RA_REQ_NO_ALLOC_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
unit: Pct
tips:
Not-scheduled Rate (Scheduler-Pipe):
avg: AVG((100 * SPI_RA_REQ_NO_ALLOC / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
avg: AVG((100 * SPI_RA_REQ_NO_ALLOC / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
min: MIN((100 * SPI_RA_REQ_NO_ALLOC / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
min: MIN((100 * SPI_RA_REQ_NO_ALLOC / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
max: MAX((100 * SPI_RA_REQ_NO_ALLOC / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
max: MAX((100 * SPI_RA_REQ_NO_ALLOC / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None)
unit: Pct
tips:
Scheduler-Pipe Stall Rate:
avg: AVG((((100 * SPI_RA_RES_STALL_CSN) / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
avg: AVG((((100 * SPI_RA_RES_STALL_CSN) / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None))
min: MIN((((100 * SPI_RA_RES_STALL_CSN) / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
min: MIN((((100 * SPI_RA_RES_STALL_CSN) / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None))
max: MAX((((100 * SPI_RA_RES_STALL_CSN) / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY !=
max: MAX((((100 * SPI_RA_RES_STALL_CSN) / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None))
unit: Pct
tips:
Scratch Stall Rate:
avg: AVG((100 * SPI_RA_TMP_STALL_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY != 0) else None)
min: MIN((100 * SPI_RA_TMP_STALL_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY != 0) else None)
max: MAX((100 * SPI_RA_TMP_STALL_CSN / (GRBM_SPI_BUSY * $se_per_gpu)) if (GRBM_SPI_BUSY != 0) else None)
avg: AVG((100 * SPI_RA_TMP_STALL_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD != 0) else None)
min: MIN((100 * SPI_RA_TMP_STALL_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD != 0) else None)
max: MAX((100 * SPI_RA_TMP_STALL_CSN / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD != 0) else None)
unit: Pct
tips:
Insufficient SIMD Waveslots:
avg: AVG(100 * SPI_RA_WAVE_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(100 * SPI_RA_WAVE_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(100 * SPI_RA_WAVE_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(100 * SPI_RA_WAVE_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(100 * SPI_RA_WAVE_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(100 * SPI_RA_WAVE_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Insufficient SIMD VGPRs:
avg: AVG(100 * SPI_RA_VGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(100 * SPI_RA_VGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(100 * SPI_RA_VGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(100 * SPI_RA_VGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(100 * SPI_RA_VGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(100 * SPI_RA_VGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Insufficient SIMD SGPRs:
avg: AVG(100 * SPI_RA_SGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(100 * SPI_RA_SGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(100 * SPI_RA_SGPR_SIMD_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(100 * SPI_RA_SGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(100 * SPI_RA_SGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(100 * SPI_RA_SGPR_SIMD_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Insufficient CU LDS:
avg: AVG(400 * SPI_RA_LDS_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(400 * SPI_RA_LDS_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(400 * SPI_RA_LDS_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(400 * SPI_RA_LDS_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(400 * SPI_RA_LDS_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(400 * SPI_RA_LDS_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Insufficient CU Barriers:
avg: AVG(400 * SPI_RA_BAR_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(400 * SPI_RA_BAR_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(400 * SPI_RA_BAR_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(400 * SPI_RA_BAR_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(400 * SPI_RA_BAR_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(400 * SPI_RA_BAR_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Reached CU Workgroup Limit:
avg: AVG(400 * SPI_RA_TGLIM_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(400 * SPI_RA_TGLIM_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(400 * SPI_RA_TGLIM_CU_FULL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(400 * SPI_RA_TGLIM_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(400 * SPI_RA_TGLIM_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(400 * SPI_RA_TGLIM_CU_FULL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
Reached CU Wavefront Limit:
avg: AVG(400 * SPI_RA_WVLIM_STALL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
min: MIN(400 * SPI_RA_WVLIM_STALL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
max: MAX(400 * SPI_RA_WVLIM_STALL_CSN / (GRBM_GUI_ACTIVE * $cu_per_gpu))
avg: AVG(400 * SPI_RA_WVLIM_STALL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
min: MIN(400 * SPI_RA_WVLIM_STALL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
max: MAX(400 * SPI_RA_WVLIM_STALL_CSN / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: Pct
tips:
src/omniperf_soc/analysis_configs/gfx942/0700_wavefront-launch.yaml
+6 -6
Melihat File
@@ -98,9 +98,9 @@ Panel Config:
unit: ns
tips:
Kernel Time (Cycles):
avg: AVG(GRBM_GUI_ACTIVE)
min: MIN(GRBM_GUI_ACTIVE)
max: MAX(GRBM_GUI_ACTIVE)
avg: AVG($GRBM_GUI_ACTIVE_PER_XCD)
min: MIN($GRBM_GUI_ACTIVE_PER_XCD)
max: MAX($GRBM_GUI_ACTIVE_PER_XCD)
unit: Cycle
tips:
Instructions per wavefront:
@@ -134,9 +134,9 @@ Panel Config:
unit: (Cycles + $normUnit)
tips:
Wavefront Occupancy:
avg: AVG((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE))
min: MIN((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE))
max: MAX((SQ_ACCUM_PREV_HIRES / GRBM_GUI_ACTIVE))
avg: AVG((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD))
min: MIN((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD))
max: MAX((SQ_ACCUM_PREV_HIRES / $GRBM_GUI_ACTIVE_PER_XCD))
unit: Wavefronts
coll_level: SQ_LEVEL_WAVES
tips:
src/omniperf_soc/analysis_configs/gfx942/1100_compute-unit-compute-pipeline.yaml
+15 -15
Melihat File
@@ -107,27 +107,27 @@ Panel Config:
unit: Instr/cycle
tips:
SALU Utilization:
avg: AVG((((100 * SQ_ACTIVE_INST_SCA) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_SCA) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_SCA) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
avg: AVG((((100 * SQ_ACTIVE_INST_SCA) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_SCA) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_SCA) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
VALU Utilization:
avg: AVG((((100 * SQ_ACTIVE_INST_VALU) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_VALU) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_VALU) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
avg: AVG((((100 * SQ_ACTIVE_INST_VALU) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_VALU) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_VALU) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
VMEM Utilization:
avg: AVG((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
min: MIN((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
max: MAX((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
avg: AVG((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
min: MIN((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
max: MAX((((100 * (SQ_ACTIVE_INST_FLAT+SQ_ACTIVE_INST_VMEM)) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
Branch Utilization:
avg: AVG((((100 * SQ_ACTIVE_INST_MISC) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_MISC) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_MISC) / GRBM_GUI_ACTIVE) / $cu_per_gpu))
avg: AVG((((100 * SQ_ACTIVE_INST_MISC) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
min: MIN((((100 * SQ_ACTIVE_INST_MISC) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
max: MAX((((100 * SQ_ACTIVE_INST_MISC) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
VALU Active Threads:
@@ -140,9 +140,9 @@ Panel Config:
unit: Threads
tips:
MFMA Utilization:
avg: AVG(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((4 * $cu_per_gpu) * GRBM_GUI_ACTIVE)))
min: MIN(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((4 * $cu_per_gpu) * GRBM_GUI_ACTIVE)))
max: MAX(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((4 * $cu_per_gpu) * GRBM_GUI_ACTIVE)))
avg: AVG(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((4 * $cu_per_gpu) * $GRBM_GUI_ACTIVE_PER_XCD)))
min: MIN(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((4 * $cu_per_gpu) * $GRBM_GUI_ACTIVE_PER_XCD)))
max: MAX(((100 * SQ_VALU_MFMA_BUSY_CYCLES) / ((4 * $cu_per_gpu) * $GRBM_GUI_ACTIVE_PER_XCD)))
unit: pct
tips:
MFMA Instr Cycles:
src/omniperf_soc/analysis_configs/gfx942/1200_lds.yaml
+2 -2
Melihat File
@@ -18,11 +18,11 @@ Panel Config:
tips: Tips
metric:
Utilization:
value: AVG(((100 * SQ_LDS_IDX_ACTIVE) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
value: AVG(((100 * SQ_LDS_IDX_ACTIVE) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: Pct of Peak
tips:
Access Rate:
value: AVG(((200 * SQ_ACTIVE_INST_LDS) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
value: AVG(((200 * SQ_ACTIVE_INST_LDS) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: Pct of Peak
tips:
Theoretical Bandwidth (% of Peak):
src/omniperf_soc/analysis_configs/gfx942/1500_TA_and_TD.yaml
+21 -21
Melihat File
@@ -20,27 +20,27 @@ Panel Config:
tips: Tips
metric:
Address Processing Unit Busy:
avg: AVG(((100 * TA_TA_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TA_TA_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TA_TA_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TA_TA_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TA_TA_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TA_TA_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Address Stall:
avg: AVG(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TA_ADDR_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Data Stall:
avg: AVG(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TA_DATA_STALLED_BY_TC_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Data-Processor → Address Stall:
avg: AVG(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Total Instructions:
@@ -128,21 +128,21 @@ Panel Config:
tips: Tips
metric:
Data-Return Busy:
avg: AVG(((100 * TD_TD_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TD_TD_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TD_TD_BUSY_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TD_TD_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TD_TD_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TD_TD_BUSY_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Cache RAM → Data-Return Stall:
avg: AVG(((100 * TD_TC_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TD_TC_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TD_TC_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TD_TC_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TD_TC_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TD_TC_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Workgroup manager → Data-Return Stall:
avg: AVG(((100 * TD_SPI_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
min: MIN(((100 * TD_SPI_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
max: MAX(((100 * TD_SPI_STALL_sum) / (GRBM_GUI_ACTIVE * $cu_per_gpu)))
avg: AVG(((100 * TD_SPI_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
min: MIN(((100 * TD_SPI_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
max: MAX(((100 * TD_SPI_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Coalescable Instructions:
src/omniperf_soc/analysis_configs/gfx942/1700_L2_cache.yaml
+1 -1
Melihat File
@@ -18,7 +18,7 @@ Panel Config:
tips: Tips
metric:
Utilization:
value: AVG(((TCC_BUSY_sum * 100) / (TO_INT($total_l2_chan) * GRBM_GUI_ACTIVE)))
value: AVG(((TCC_BUSY_sum * 100) / (TO_INT($total_l2_chan) * $GRBM_GUI_ACTIVE_PER_XCD)))
unit: pct
tips:
Bandwidth:
src/omniperf_soc/analysis_configs/panel_config_template.yaml
+4 -4
Melihat File
@@ -34,16 +34,16 @@ Panel Config:
tips: Tips
metric:
METRIC01:
value: AVG(100 * SQ_ACTIVE_INST_SCA / ( GRBM_GUI_ACTIVE * $cu_per_gpu ))
value: AVG(100 * SQ_ACTIVE_INST_SCA / ( $GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu ))
unit: pct
peak: 100
pop: AVG(100* SQ_ACTIVE_INST_SCA/(GRBM_GUI_ACTIVE*$cu_per_gpu))
pop: AVG(100* SQ_ACTIVE_INST_SCA/($GRBM_GUI_ACTIVE_PER_XCD*$cu_per_gpu))
tips: *METRIC01_anchor
METRIC02:
value: AVG(100 * SQ_ACTIVE_INST_VALU / ( GRBM_GUI_ACTIVE * $cu_per_gpu))
value: AVG(100 * SQ_ACTIVE_INST_VALU / ( $GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu))
unit: pct
peak: 100
pop: AVG(100* SQ_ACTIVE_INST_VALU/(GRBM_GUI_ACTIVE*$cu_per_gpu))
pop: AVG(100* SQ_ACTIVE_INST_VALU/($GRBM_GUI_ACTIVE_PER_XCD*$cu_per_gpu))
tips:
# CSV table sample
- raw_csv_table:
src/omniperf_soc/soc_base.py
+3 -4
Melihat File
@@ -90,7 +90,7 @@ class OmniSoC_Base():
@demarcate
def populate_mspec(self):
from utils.specs import search, run, total_sqc
from utils.specs import search, run, total_sqc, total_xcds
if not hasattr(self._mspec, "_rocminfo"):
return
@@ -160,9 +160,6 @@ class OmniSoC_Base():
self._mspec.cur_sclk = self._mspec.max_sclk
self._mspec.cur_mclk = self._mspec.max_mclk
self._mspec.compute_partition = ""
self._mspec.memory_partition = ""
# specify gpu name for gfx942 hardware
self._mspec.gpu_model = list(SUPPORTED_ARCHS[self._mspec.gpu_arch].keys())[0].upper()
if self._mspec.gpu_model == "MI300":
@@ -172,6 +169,8 @@ class OmniSoC_Base():
if (self._mspec.gpu_arch == "gfx942") and ("MI300A" not in self._mspec._rocminfo):
self._mspec.gpu_model = "MI300X_A1"
self._mspec.num_xcd = str(total_xcds(self._mspec.gpu_model, self._mspec.compute_partition))
@demarcate
def perfmon_filter(self, roofline_perfmon_only: bool):
"""Filter default performance counter set based on user arguments"""
src/utils/parser.py
+8
Melihat File
@@ -81,6 +81,9 @@ build_in_vars = {
0) / $max_waves_per_cu) * 8) + MIN(MOD(ROUND(AVG(((4 * SQ_BUSY_CU_CYCLES) \
/ GRBM_GUI_ACTIVE)), 0), $max_waves_per_cu), 8)), $cu_per_gpu))",
"kernelBusyCycles": "ROUND(AVG((((End_Timestamp - Start_Timestamp) / 1000) * $max_sclk)), 0)",
"GRBM_GUI_ACTIVE_PER_XCD": "(GRBM_GUI_ACTIVE / $num_xcd)",
"GRBM_COUNT_PER_XCD": "(GRBM_COUNT / $num_xcd)",
"GRBM_SPI_BUSY_PER_XCD" : "(GRBM_SPI_BUSY / $num_xcd)"
}
supported_call = {
@@ -691,6 +694,7 @@ def eval_metric(dfs, dfs_type, sys_info, raw_pmc_df, debug):
ammolite__max_waves_per_cu = sys_info.max_waves_per_cu
ammolite__hbm_bw = sys_info.hbm_bw
ammolite__total_l2_chan = calc_builtin_var("$total_l2_chan", sys_info)
ammolite__num_xcd = sys_info.num_xcd
# TODO: fix all $normUnit in Unit column or title
@@ -709,6 +713,9 @@ def eval_metric(dfs, dfs_type, sys_info, raw_pmc_df, debug):
ammolite__numActiveCUs = ammolite__build_in["numActiveCUs"]
ammolite__kernelBusyCycles = ammolite__build_in["kernelBusyCycles"]
ammolite__GRBM_GUI_ACTIVE_PER_XCD = ammolite__build_in["GRBM_GUI_ACTIVE_PER_XCD"]
ammolite__GRBM_COUNT_PER_XCD = ammolite__build_in["GRBM_COUNT_PER_XCD"]
ammolite__GRBM_SPI_BUSY_PER_XCD = ammolite__build_in["GRBM_SPI_BUSY_PER_XCD"]
# Hmmm... apply + lambda should just work
# df['Value'] = df['Value'].apply(lambda s: eval(compile(str(s), '<string>', 'eval')))
@@ -1003,6 +1010,7 @@ def correct_sys_info(df, specs_correction):
"hbmBW": "hbmBW",
"compute_partition": "compute_partition",
"memory_partition": "memory_partition",
"num_xcd": "num_xcd"
}
# todo: more err checking for string specs_correction
src/utils/specs.py
+33
Melihat File
@@ -132,6 +132,7 @@ class MachineSpecs:
self.pipes_per_gpu: str = None
self.total_l2_chan: str = None
self.hbm_bw: str = None
self.num_xcd: str = None
# Load above SoC specs via module import
try:
soc_module = importlib.import_module('omniperf_soc.soc_'+ self.gpu_arch)
@@ -226,6 +227,7 @@ class MachineSpecs:
HBM BW: {self.hbm_bw} MB/s
Compute Partition: {self.compute_partition}
Memory Partition: {self.memory_partition}
Num XCDs: {self.num_xcd}
"""
)
@@ -299,6 +301,37 @@ def total_sqc(archname, numCUs, numSEs):
sq_per_se = ceil(sq_per_se)
return int(sq_per_se) * int(numSEs)
def total_xcds(archname, compute_partition):
# check MI300 has a valid compute partition
mi300a_archs = ["mi300a_a0", "mi300a_a1"]
mi300x_archs = ["mi300x_a0", "mi300x_a1"]
if archname.lower() in mi300a_archs + mi300x_archs \
and compute_partition == "NA":
error("Invalid compute partition found for {}".format(archname))
if archname.lower() not in mi300a_archs + mi300x_archs:
return 1
# from the whitepaper
# https://www.amd.com/content/dam/amd/en/documents/instinct-tech-docs/white-papers/amd-cdna-3-white-paper.pdf
if compute_partition.lower() == "spx":
if archname.lower() in mi300a_archs:
return 6
if archname.lower() in mi300x_archs:
return 8
if compute_partition.lower() == "tpx":
if archname.lower() in mi300a_archs:
return 2
if compute_partition.lower() == "dpx":
if archname.lower() in mi300x_archs:
return 4
if compute_partition.lower() == "qpx":
if archname.lower() in mi300x_archs:
return 2
if compute_partition.lower() == "cpx":
if archname.lower() in mi300x_archs:
return 2
error("Unknown compute partition / arch found for {} / {}".format(
compute_partition, archname))
if __name__ == "__main__":
print(MachineSpecs())