AI/rocm-systems
2
0
انشعاب 0
کد مسائل تقاضاهای واکشی Actions Packages پروژه‌ها انتشارها دانشنامه فعالیت

Improve baseline comparison (#817)

فهرست منبع 
* Do not force unsupported metrics to be specified in older gpu
  architectures as None

* Remove metrics which are explicitly set to None

* Update CHANGELOG

* Fix analysis configuration to fix baseline comparisons across all gpu
  architectures
    * Add missing 1812 section for gfx908
    * Add missing 1812 section for gfx90a

* Baseline comparision will only show common metrics
   * First workload will be used to set Metric ID index column
This commit is contained in:
vedithal-amd
2025-07-24 11:49:02 -04:00
کامیت شده توسط GitHub
والد 137f35e700
کامیت d4c316a730
52فایلهای تغییر یافته به همراه66 افزوده شده و 2480 حذف شده
دانلود پرونده وصله دانلود فایل تغییرات diff Expand all files Collapse all files
CHANGELOG.md
+2
مشاهده پرونده
@@ -83,6 +83,8 @@ Full documentation for ROCm Compute Profiler is available at [https://rocm.docs.
* OS distribution support minimum for roofline feature is now Ubuntu22.04, RHEL9, and SLES15SP6
* Improve analysis block based filtering to accept metric id level filtering
* This can be used to collect individual metrics from various sections of analysis config
* CLI analysis mode baseline comparison will now only compare common metrics across workloads and will not show Metric ID
* Remove metrics from analysis configuration files which are explicitly marked as empty or None
### Optimized
src/rocprof_compute_soc/analysis_configs/gfx908/0200_system-speed-of-light.yaml
-13
مشاهده پرونده
@@ -32,12 +32,6 @@ Panel Config:
peak: (((($max_sclk * $cu_per_gpu) * 64) * 2) / 1000)
pop: None # No perf counter
tips:
MFMA FLOPs (F8):
value: None # No HW module
unit: GFLOP/s
peak: None # No HW module
pop: None # No HW module
tips:
MFMA FLOPs (BF16):
value: None # No perf counter
unit: GFLOP/s
@@ -62,13 +56,6 @@ Panel Config:
peak: ((($max_sclk * $cu_per_gpu) * 256) / 1000)
pop: None # No perf counter
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then remove this
MFMA FLOPs (F6F4):
value: None
unit: GFLOP/s
peak: None
pop: None
tips:
MFMA IOPs (Int8):
value: None # No perf counter
unit: GIOP/s
src/rocprof_compute_soc/analysis_configs/gfx908/0300_mem_chart.yaml
-4
مشاهده پرونده
@@ -45,10 +45,6 @@ Panel Config:
#alias: valu_
value: ROUND(AVG((SQ_INSTS_VALU / $denom)), 0)
tips:
MFMA:
#alias: mfma_
value: None # No perf counter
tips:
VMEM:
#alias: vmem_
value: ROUND(AVG((SQ_INSTS_VMEM / $denom)), 0)
src/rocprof_compute_soc/analysis_configs/gfx908/0500_command-processor.yaml
-21
مشاهده پرونده
@@ -19,27 +19,6 @@ Panel Config:
unit: Unit
tips: Tips
metric:
# TODO: Fix baseline comparision logic to handle non existent metrics, then remove this
CPC SYNC FIFO Full Rate:
avg: None
min: None
max: None
unit: pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then remove this
CPC CANE Stall Rate:
avg: None
min: None
max: None
unit: pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then remove this
CPC ADC Utilization:
avg: None
min: None
max: None
unit: pct
tips:
CPF Utilization:
avg: AVG((((100 * CPF_CPF_STAT_BUSY) / (CPF_CPF_STAT_BUSY + CPF_CPF_STAT_IDLE))
if ((CPF_CPF_STAT_BUSY + CPF_CPF_STAT_IDLE) != 0) else None))
src/rocprof_compute_soc/analysis_configs/gfx908/0600_shader-processor-input.yaml
-21
مشاهده پرونده
@@ -19,13 +19,6 @@ Panel Config:
unit: Unit
tips: Tips
metric:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Schedule-Pipe Wave Occupancy:
avg: None
min: None
max: None
unit: Wave
tips:
Accelerator Utilization:
avg: AVG(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
min: MIN(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
@@ -38,13 +31,6 @@ Panel Config:
max: MAX(100 * SPI_CSN_BUSY / ($GRBM_GUI_ACTIVE_PER_XCD * $pipes_per_gpu * $se_per_gpu))
unit: Pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Scheduler-Pipe Wave Utilization:
avg: None
min: None
max: None
unit: Pct
tips:
Workgroup Manager Utilization:
avg: AVG(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
min: MIN(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
@@ -122,13 +108,6 @@ Panel Config:
0) else None)
unit: Pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Scheduler-Pipe FIFO Full Rate:
avg: None
min: None
max: None
unit: Pct
tips:
Scheduler-Pipe Stall Rate:
avg: AVG((((100 * SPI_RA_RES_STALL_CSN) / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None))
src/rocprof_compute_soc/analysis_configs/gfx908/1000_compute-unit-instruction-mix.yaml
-158
مشاهده پرونده
@@ -19,30 +19,12 @@ Panel Config:
unit: Unit
tips: Tips
metric:
VALU:
avg: None # No HW module
min: None # No HW module
max: None # No HW module
unit: (instr + $normUnit)
tips:
VMEM:
avg: None # No HW module
min: None # No HW module
max: None # No HW module
unit: (instr + $normUnit)
tips:
LDS:
avg: AVG((SQ_INSTS_LDS / $denom))
min: MIN((SQ_INSTS_LDS / $denom))
max: MAX((SQ_INSTS_LDS / $denom))
unit: (instr + $normUnit)
tips:
MFMA:
avg: None # No HW module
min: None # No HW module
max: None # No HW module
unit: (instr + $normUnit)
tips:
SALU:
avg: AVG((SQ_INSTS_SALU / $denom))
min: MIN((SQ_INSTS_SALU / $denom))
@@ -73,96 +55,6 @@ Panel Config:
unit: Unit
tips: Tips
metric:
INT32:
avg: None # No perf counter
min: None # No perf counter
max: None # No perf counter
unit: (instr + $normUnit)
tips:
INT64:
avg: None # No perf counter
min: None # No perf counter
max: None # No perf counter
unit: (instr + $normUnit)
tips:
F16-ADD:
avg: None # No perf counter
min: None # No perf counter
max: None # No perf counter
unit: (instr + $normUnit)
tips:
F16-MUL:
avg: None # No perf counter
min: None # No perf counter
max: None # No perf counter
unit: (instr + $normUnit)
tips:
F16-FMA:
avg: None # No perf counter
min: None # No perf counter
max: None # No perf counter
unit: (instr + $normUnit)
tips:
F16-Trans:
avg: None # No perf counter
min: None # No perf counter
max: None # No perf counter
unit: (instr + $normUnit)
tips:
F32-ADD:
avg: None # No perf counter
min: None # No perf counter
max: None # No perf counter
unit: (instr + $normUnit)
tips:
F32-MUL:
avg: None # No perf counter
min: None # No perf counter
max: None # No perf counter
unit: (instr + $normUnit)
tips:
F32-FMA:
avg: None # No perf counter
min: None # No perf counter
max: None # No perf counter
unit: (instr + $normUnit)
tips:
F32-Trans:
avg: None # No perf counter
min: None # No perf counter
max: None # No perf counter
unit: (instr + $normUnit)
tips:
F64-ADD:
avg: None # No perf counter
min: None # No perf counter
max: None # No perf counter
unit: (instr + $normUnit)
tips:
F64-MUL:
avg: None # No perf counter
min: None # No perf counter
max: None # No perf counter
unit: (instr + $normUnit)
tips:
F64-FMA:
avg: None # No perf counter
min: None # No perf counter
max: None # No perf counter
unit: (instr + $normUnit)
tips:
F64-Trans:
avg: None # No perf counter
min: None # No perf counter
max: None # No perf counter
unit: (instr + $normUnit)
tips:
Conversion:
avg: None # No perf counter
min: None # No perf counter
max: None # No perf counter
unit: (instr + $normUnit)
tips:
- metric_table:
id: 1003
@@ -181,13 +73,6 @@ Panel Config:
max: MAX((TA_FLAT_WAVEFRONTS_sum / $denom))
unit: (instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then remove this
Spill/Stack Coalesceable Instr:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
Global/Generic Read:
avg: AVG((TA_FLAT_READ_WAVEFRONTS_sum / $denom))
min: MIN((TA_FLAT_READ_WAVEFRONTS_sum / $denom))
@@ -242,46 +127,3 @@ Panel Config:
unit: Unit
tips: Tips
metric:
MFMA-I8:
avg: None # No HW module
min: None # No HW module
max: None # No HW module
unit: (instr + $normUnit)
tips:
MFMA-F8:
avg: None # No HW module
min: None # No HW module
max: None # No HW module None # No HW module
unit: (instr + $normUnit)
tips:
MFMA-F16:
avg: None # No HW module
min: None # No HW module
max: None # No HW module
unit: (instr + $normUnit)
tips:
MFMA-BF16:
avg: None # No HW module
min: None # No HW module
max: None # No HW module
unit: (instr + $normUnit)
tips:
MFMA-F32:
avg: None # No HW module
min: None # No HW module
max: None # No HW module
unit: (instr + $normUnit)
tips:
MFMA-F64:
avg: None # No HW module
min: None # No HW module
max: None # No HW module
unit: (instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
MFMA-F6F4:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx908/1100_compute-unit-compute-pipeline.yaml
-155
مشاهده پرونده
@@ -19,61 +19,6 @@ Panel Config:
pop: Pct of Peak
tips: Tips
metric:
VALU FLOPs:
value: None # No perf counter
unit: None
peak: None
pop: None
tips:
VALU IOPs:
value: None # No perf counter
unit: None
peak: None
pop: None
tips:
MFMA FLOPs (F8):
value: None # No perf counter
unit: GFLOP
peak: None # No perf counter
pop: None # No perf counter
tips:
MFMA FLOPs (BF16):
value: None # No perf counter
Unit: None
peak: None
pop: None
tips:
MFMA FLOPs (F16):
value: None # No perf counter
unit: None
peak: None
pop: None
tips:
MFMA FLOPs (F32):
value: None # No perf counter
unit: None
peak: None
pop: None
tips:
MFMA FLOPs (F64):
value: None # No perf counter
unit: None
peak: None
pop: None
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
MFMA FLOPs (F6F4):
value: None
unit: GFLOP
peak: None
pop: None
tips:
MFMA IOPs (INT8):
value: None # No perf counter
unit: None
peak: None
pop: None
tips:
- metric_table:
id: 1102
@@ -116,25 +61,6 @@ Panel Config:
max: MAX((((100 * SQ_ACTIVE_INST_VALU) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
VALU Co-Issue Efficiency:
avg: None
min: None
max: None
unit: pct
tips:
VMEM Utilization:
avg: None # No HW module
min: None # No HW module
max: None # No HW module
unit: pct
tips:
Branch Utilization:
avg: None # No HW module
min: None # No HW module
max: None # No HW module
unit: pct
tips:
VALU Active Threads:
avg: AVG(((SQ_THREAD_CYCLES_VALU / SQ_ACTIVE_INST_VALU) if (SQ_ACTIVE_INST_VALU
!= 0) else None))
@@ -144,32 +70,6 @@ Panel Config:
!= 0) else None))
unit: Threads
tips:
MFMA Utilization:
avg: None # No HW module
min: None # No HW module
max: None # No HW module
unit: pct
tips:
MFMA Instr Cycles:
avg: None # No HW module
min: None # No HW module
max: None # No HW module
unit: cycles/instr
tips:
VMEM Latency:
avg: None # No perf counter
min: None # No perf counter
max: None # No perf counter
unit: Cycles
coll_level: SQ_INST_LEVEL_VMEM
tips:
SMEM Latency:
avg: None # No perf counter
min: None # No perf counter
max: None # No perf counter
unit: Cycles
coll_level: SQ_INST_LEVEL_SMEM
tips:
- metric_table:
id: 1103
@@ -182,58 +82,3 @@ Panel Config:
unit: Unit
tips: Tips
metric:
FLOPs (Total):
avg: None # No perf counter
min: None # No perf counter
max: None # No perf counter
unit: (OPs + $normUnit)
tips:
IOPs (Total):
avg: None # No perf counter
min: None # No perf counter
max: None # No perf counter
unit: (OPs + $normUnit)
tips:
F8 OPs:
avg: None # No HW module
min: None # No HW module
max: None # No HW module
unit: (OPs + $normUnit)
tips:
F16 OPs:
avg: None # No perf counter
min: None # No perf counter
max: None # No perf counter
unit: (OPs + $normUnit)
tips:
BF16 OPs:
avg: None # No perf counter
min: None # No perf counter
max: None # No perf counter
unit: (OPs + $normUnit)
tips:
F32 OPs:
avg: None # No perf counter
min: None # No perf counter
max: None # No perf counter
unit: (OPs + $normUnit)
tips:
F64 OPs:
avg: None # No perf counter
min: None # No perf counter
max: None # No perf counter
unit: (OPs + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
F6F4 OPs:
avg: None
min: None
max: None
unit: (OPs + $normUnit)
tips:
INT8 OPs:
avg: None # No perf counter
min: None # No perf counter
max: None # No perf counter
unit: (OPs + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx908/1200_lds.yaml
-56
مشاهده پرونده
@@ -55,48 +55,6 @@ Panel Config:
max: MAX((SQ_INSTS_LDS / $denom))
unit: (Instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS LOAD:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS STORE:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS ATOMIC:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS LOAD Bandwidth:
avg: None
min: None
max: None
units: Gbps
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS STORE Bandwidth:
avg: None
min: None
max: None
units: Gbps
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS ATOMIC Bandwidth:
avg: None
min: None
max: None
units: Gbps
tips:
Theoretical Bandwidth:
avg: AVG(((((SQ_LDS_IDX_ACTIVE - SQ_LDS_BANK_CONFLICT) * 4) * TO_INT($lds_banks_per_cu))
/ $denom))
@@ -158,17 +116,3 @@ Panel Config:
max: MAX((SQ_LDS_MEM_VIOLATIONS / $denom))
unit: (Accesses + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS Command FIFO Full Rate:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS Data FIFO Full Rate:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx908/1500_TA_and_TD.yaml
-45
مشاهده پرونده
@@ -43,27 +43,6 @@ Panel Config:
max: MAX(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Sequencer → TA Address Stall:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Sequencer → TA Command Stall:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Sequencer → TA Data Stall:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
Total Instructions:
avg: AVG((TA_TOTAL_WAVEFRONTS_sum / $denom))
min: MIN((TA_TOTAL_WAVEFRONTS_sum / $denom))
@@ -82,12 +61,6 @@ Panel Config:
max: MAX((TA_FLAT_READ_WAVEFRONTS_sum / $denom))
unit: (Instructions + $normUnit)
tips:
Global/Generic Read Instructions for LDS:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Instructions + $normUnit)
tips:
Global/Generic Write Instructions:
avg: AVG((TA_FLAT_WRITE_WAVEFRONTS_sum / $denom))
min: MIN((TA_FLAT_WRITE_WAVEFRONTS_sum / $denom))
@@ -112,12 +85,6 @@ Panel Config:
max: MAX((TA_BUFFER_READ_WAVEFRONTS_sum / $denom))
unit: (Instructions + $normUnit)
tips:
Spill/Stack Read Instructions for LDS:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Instructions + $normUnit)
tips:
Spill/Stack Write Instructions:
avg: AVG((TA_BUFFER_WRITE_WAVEFRONTS_sum / $denom))
min: MIN((TA_BUFFER_WRITE_WAVEFRONTS_sum / $denom))
@@ -172,12 +139,6 @@ Panel Config:
max: MAX(((100 * TD_TC_STALL_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
Workgroup manager → Data-Return Stall:
avg: # No perf counter
min: # No perf counter
max: # No perf counter
unit: pct
tips:
Coalescable Instructions:
avg: AVG((TD_COALESCABLE_WAVEFRONT_sum / $denom))
min: MIN((TD_COALESCABLE_WAVEFRONT_sum / $denom))
@@ -205,9 +166,3 @@ Panel Config:
max: MAX((TD_ATOMIC_WAVEFRONT_sum / $denom))
unit: (Instructions + $normUnit)
tips:
Write Ack Instructions:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Instructions + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx908/1600_L1_cache.yaml
-72
مشاهده پرونده
@@ -189,30 +189,6 @@ Panel Config:
+ TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / $denom))
unit: (Bytes + $normUnit)
tips:
Tag RAM 0 Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Tag RAM 1 Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Tag RAM 2 Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Tag RAM 3 Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
L1-L2 Read:
avg: AVG((TCP_TCC_READ_REQ_sum / $denom))
min: MIN((TCP_TCC_READ_REQ_sum / $denom))
@@ -419,12 +395,6 @@ Panel Config:
max: MAX((TCP_UTCL1_TRANSLATION_MISS_sum / $denom))
units: (Req + $normUnit)
tips:
Misses under Translation Miss:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Req + $normUnit)
tips:
Permission Misses:
avg: AVG((TCP_UTCL1_PERMISSION_MISS_sum / $denom))
min: MIN((TCP_UTCL1_PERMISSION_MISS_sum / $denom))
@@ -442,45 +412,3 @@ Panel Config:
units: Units
tips: Tips
metric:
Cache Full Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Cache Miss Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Serialization Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Thrashing Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Latency FIFO Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Resident Page Full Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
UTCL2 Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx908/1700_L2_cache.yaml
-168
مشاهده پرونده
@@ -143,18 +143,6 @@ Panel Config:
!= 0) else None))
unit: Cycles
tips:
Read Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
- metric_table:
id: 1703
@@ -173,24 +161,6 @@ Panel Config:
max: MAX((TCC_REQ_sum * 64) / $denom)
unit: (Bytes + $normUnit)
tips:
Read Bandwidth:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Write Bandwidth:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Atomic Bandwidth:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Req:
avg: AVG((TCC_REQ_sum / $denom))
min: MIN((TCC_REQ_sum / $denom))
@@ -221,24 +191,12 @@ Panel Config:
max: MAX((TCC_STREAMING_REQ_sum / $denom))
unit: (Req + $normUnit)
tips:
Bypasss Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Probe Req:
avg: AVG((TCC_PROBE_sum / $denom))
min: MIN((TCC_PROBE_sum / $denom))
max: MAX((TCC_PROBE_sum / $denom))
unit: (Req + $normUnit)
tips:
Input Buffer Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Cache Hit:
avg: AVG((((100 * TCC_HIT_sum) / (TCC_HIT_sum + TCC_MISS_sum)) if ((TCC_HIT_sum
+ TCC_MISS_sum) != 0) else None))
@@ -326,24 +284,6 @@ Panel Config:
unit: Unit
tips: Tips
metric:
Stalled on Latency FIFO:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Cycles + $normUnit)
tips:
Stalled on Write Data FIFO:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Cycles + $normUnit)
tips:
Input Buffer Stalled on L2:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Cycles + $normUnit)
tips:
- metric_table:
id: 1705
@@ -360,54 +300,6 @@ Panel Config:
style:
type: simple_multi_bar
metric:
Read - PCIe Stall:
type: PCIe Stall
transaction: Read
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Read - Infinity Fabric™ Stall:
type: Infinity Fabric™ Stall
transaction: Read
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Read - HBM Stall:
type: HBM Stall
transaction: Read
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write - PCIe Stall:
type: PCIe Stall
transaction: Write
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write - Infinity Fabric™ Stall:
type: Infinity Fabric™ Stall
transaction: Write
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write - HBM Stall:
type: HBM Stall
transaction: Write
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write - Credit Starvation:
type: Credit Starvation
transaction: Write
@@ -458,24 +350,6 @@ Panel Config:
max: MAX((MAX((TCC_EA_RDREQ_sum - TCC_EA_RDREQ_DRAM_sum), 0) / $denom))
unit: (Req + $normUnit)
tips:
Read Bandwidth - PCIe:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Read Bandwidth - Infinity Fabric™:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Read Bandwidth - HBM:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Write and Atomic (32B):
avg: AVG(((TCC_EA_WRREQ_sum - TCC_EA_WRREQ_64B_sum) / $denom))
min: MIN(((TCC_EA_WRREQ_sum - TCC_EA_WRREQ_64B_sum) / $denom))
@@ -506,51 +380,9 @@ Panel Config:
max: MAX((MAX((TCC_EA_WRREQ_sum - TCC_EA_WRREQ_DRAM_sum), 0) / $denom))
unit: (Req + $normUnit)
tips:
Write Bandwidth - PCIe:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Write Bandwidth - Infinity Fabric™:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Write Bandwidth - HBM:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Atomic:
avg: AVG((TCC_EA_ATOMIC_sum / $denom))
min: MIN((TCC_EA_ATOMIC_sum / $denom))
max: MAX((TCC_EA_ATOMIC_sum / $denom))
unit: (Req + $normUnit)
tips:
Atomic - HBM:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Atomic Bandwidth - PCIe:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Atomic Bandwidth - Infinity Fabric™:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Atomic Bandwidth - HBM:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx908/1800_L2_cache_per_channel.yaml
+15
مشاهده پرونده
@@ -333,3 +333,18 @@ Panel Config:
placeholder_range:
"::_1": $total_l2_chan
cli_style: simple_multiple_bar
- metric_table:
id: 1812
title: L2-Fabric (128B read requests per normUnit)
header:
metric: Channel
expr: Expression
metric:
"::_1":
expr: (TO_INT(TCC_BUBBLE[::_1]) / $denom)
placeholder_range:
"::_1": $total_l2_chan
# tips: Number of 128-byte read requests sent to EA
cli_style: simple_box
tui_style: simple_box
src/rocprof_compute_soc/analysis_configs/gfx90a/0200_system-speed-of-light.yaml
-13
مشاهده پرونده
@@ -42,12 +42,6 @@ Panel Config:
pop: ((100 * AVG(((64 * (SQ_INSTS_VALU_INT32 + SQ_INSTS_VALU_INT64)) / (End_Timestamp
- Start_Timestamp)))) / (((($max_sclk * $cu_per_gpu) * 64) * 2) / 1000))
tips:
MFMA FLOPs (F8):
value: None
unit: GFLOP/s
peak: None
pop: None
tips:
MFMA FLOPs (BF16):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_BF16 * 512) / (End_Timestamp - Start_Timestamp)))
unit: GFLOP/s
@@ -76,13 +70,6 @@ Panel Config:
pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_F64 * 512) / (End_Timestamp - Start_Timestamp))))
/ ((($max_sclk * $cu_per_gpu) * 256) / 1000))
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then remove this
MFMA FLOPs (F6F4):
value: None
unit: GFLOP/s
peak: None
pop: None
tips:
MFMA IOPs (Int8):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp)))
unit: GIOP/s
src/rocprof_compute_soc/analysis_configs/gfx90a/0500_command-processor.yaml
-18
مشاهده پرونده
@@ -19,24 +19,6 @@ Panel Config:
unit: Unit
tips: Tips
metric:
CPC SYNC FIFO Full Rate:
avg: None
min: None
max: None
unit: pct
tips:
CPC CANE Stall Rate:
avg: None
min: None
max: None
unit: pct
tips:
CPC ADC Utilization:
avg: None
min: None
max: None
unit: pct
tips:
CPF Utilization:
avg: AVG((((100 * CPF_CPF_STAT_BUSY) / (CPF_CPF_STAT_BUSY + CPF_CPF_STAT_IDLE))
if ((CPF_CPF_STAT_BUSY + CPF_CPF_STAT_IDLE) != 0) else None))
src/rocprof_compute_soc/analysis_configs/gfx90a/0600_shader-processor-input.yaml
-21
مشاهده پرونده
@@ -19,13 +19,6 @@ Panel Config:
unit: Unit
tips: Tips
metric:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Schedule-Pipe Wave Occupancy:
avg: None
min: None
max: None
unit: Wave
tips:
Accelerator Utilization:
avg: AVG(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
min: MIN(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
@@ -38,13 +31,6 @@ Panel Config:
max: MAX(100 * SPI_CSN_BUSY / ($GRBM_GUI_ACTIVE_PER_XCD * $pipes_per_gpu * $se_per_gpu))
unit: Pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Scheduler-Pipe Wave Utilization:
avg: None
min: None
max: None
unit: Pct
tips:
Workgroup Manager Utilization:
avg: AVG(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
min: MIN(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
@@ -122,13 +108,6 @@ Panel Config:
0) else None)
unit: Pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Scheduler-Pipe FIFO Full Rate:
avg: None
min: None
max: None
unit: Pct
tips:
Scheduler-Pipe Stall Rate:
avg: AVG((((100 * SPI_RA_RES_STALL_CSN) / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None))
src/rocprof_compute_soc/analysis_configs/gfx90a/1000_compute-unit-instruction-mix.yaml
-20
مشاهده پرونده
@@ -181,13 +181,6 @@ Panel Config:
max: MAX((TA_FLAT_WAVEFRONTS_sum / $denom))
unit: (instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then remove this
Spill/Stack Coalesceable Instr:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
Global/Generic Read:
avg: AVG((TA_FLAT_READ_WAVEFRONTS_sum / $denom))
min: MIN((TA_FLAT_READ_WAVEFRONTS_sum / $denom))
@@ -248,12 +241,6 @@ Panel Config:
max: MAX((SQ_INSTS_VALU_MFMA_I8 / $denom))
unit: (instr + $normUnit)
tips:
MFMA-F8:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
MFMA-F16:
avg: AVG((SQ_INSTS_VALU_MFMA_F16 / $denom))
min: MIN((SQ_INSTS_VALU_MFMA_F16 / $denom))
@@ -278,10 +265,3 @@ Panel Config:
max: MAX((SQ_INSTS_VALU_MFMA_F64 / $denom))
unit: (instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
MFMA-F6F4:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx90a/1100_compute-unit-compute-pipeline.yaml
-33
مشاهده پرونده
@@ -41,12 +41,6 @@ Panel Config:
pop: ((100 * AVG(((64 * (SQ_INSTS_VALU_INT32 + SQ_INSTS_VALU_INT64)) / (End_Timestamp
- Start_Timestamp)))) / (((($max_sclk * $cu_per_gpu) * 64) * 2) / 1000))
tips:
MFMA FLOPs (F8):
value: None
unit: GFLOP
peak: None
pop: None
tips:
MFMA FLOPs (BF16):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_BF16 * 512) / (End_Timestamp - Start_Timestamp)))
unit: GFLOP
@@ -75,13 +69,6 @@ Panel Config:
pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_F64 * 512) / (End_Timestamp - Start_Timestamp))))
/ ((($max_sclk * $cu_per_gpu) * 256) / 1000))
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
MFMA FLOPs (F6F4):
value: None
unit: GFLOP
peak: None
pop: None
tips:
MFMA IOPs (INT8):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp)))
unit: GIOP
@@ -131,13 +118,6 @@ Panel Config:
max: MAX((((100 * SQ_ACTIVE_INST_VALU) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
VALU Co-Issue Efficiency:
avg: None
min: None
max: None
unit: pct
tips:
VMEM Utilization:
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))
@@ -236,12 +216,6 @@ Panel Config:
max: MAX(((64 * (SQ_INSTS_VALU_INT32 + SQ_INSTS_VALU_INT64)) + (SQ_INSTS_VALU_MFMA_MOPS_I8 * 512)) / $denom)
unit: (OPs + $normUnit)
tips:
F8 OPs:
avg: None
min: None
max: None
unit: (OPs + $normUnit)
tips:
F16 OPs:
avg: AVG(((((((64 * SQ_INSTS_VALU_ADD_F16) + (64 * SQ_INSTS_VALU_MUL_F16)) +
(64 * SQ_INSTS_VALU_TRANS_F16)) + (128 * SQ_INSTS_VALU_FMA_F16)) + (512 *
@@ -278,13 +252,6 @@ Panel Config:
+ (SQ_INSTS_VALU_FMA_F64 * 2))) + (512 * SQ_INSTS_VALU_MFMA_MOPS_F64)) / $denom))
unit: (OPs + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
F6F4 OPs:
avg: None
min: None
max: None
unit: (OPs + $normUnit)
tips:
INT8 OPs:
avg: AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / $denom))
min: MIN(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / $denom))
src/rocprof_compute_soc/analysis_configs/gfx90a/1200_lds.yaml
-56
مشاهده پرونده
@@ -55,48 +55,6 @@ Panel Config:
max: MAX((SQ_INSTS_LDS / $denom))
unit: (Instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS LOAD:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS STORE:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS ATOMIC:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS LOAD Bandwidth:
avg: None
min: None
max: None
units: Gbps
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS STORE Bandwidth:
avg: None
min: None
max: None
units: Gbps
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS ATOMIC Bandwidth:
avg: None
min: None
max: None
units: Gbps
tips:
Theoretical Bandwidth:
avg: AVG(((((SQ_LDS_IDX_ACTIVE - SQ_LDS_BANK_CONFLICT) * 4) * TO_INT($lds_banks_per_cu))
/ $denom))
@@ -158,17 +116,3 @@ Panel Config:
max: MAX((SQ_LDS_MEM_VIOLATIONS / $denom))
unit: (Accesses + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS Command FIFO Full Rate:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS Data FIFO Full Rate:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx90a/1500_TA_and_TD.yaml
-39
مشاهده پرونده
@@ -43,27 +43,6 @@ Panel Config:
max: MAX(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Sequencer → TA Address Stall:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Sequencer → TA Command Stall:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Sequencer → TA Data Stall:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
Total Instructions:
avg: AVG((TA_TOTAL_WAVEFRONTS_sum / $denom))
min: MIN((TA_TOTAL_WAVEFRONTS_sum / $denom))
@@ -82,12 +61,6 @@ Panel Config:
max: MAX((TA_FLAT_READ_WAVEFRONTS_sum / $denom))
unit: (Instructions + $normUnit)
tips:
Global/Generic Read Instructions for LDS:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Instructions + $normUnit)
tips:
Global/Generic Write Instructions:
avg: AVG((TA_FLAT_WRITE_WAVEFRONTS_sum / $denom))
min: MIN((TA_FLAT_WRITE_WAVEFRONTS_sum / $denom))
@@ -112,12 +85,6 @@ Panel Config:
max: MAX((TA_BUFFER_READ_WAVEFRONTS_sum / $denom))
unit: (Instructions + $normUnit)
tips:
Spill/Stack Read Instructions for LDS:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Instructions + $normUnit)
tips:
Spill/Stack Write Instructions:
avg: AVG((TA_BUFFER_WRITE_WAVEFRONTS_sum / $denom))
min: MIN((TA_BUFFER_WRITE_WAVEFRONTS_sum / $denom))
@@ -205,9 +172,3 @@ Panel Config:
max: MAX((TD_ATOMIC_WAVEFRONT_sum / $denom))
unit: (Instructions + $normUnit)
tips:
Write Ack Instructions:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Instructions + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx90a/1600_L1_cache.yaml
-72
مشاهده پرونده
@@ -189,30 +189,6 @@ Panel Config:
+ TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / $denom))
unit: (Bytes + $normUnit)
tips:
Tag RAM 0 Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Tag RAM 1 Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Tag RAM 2 Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Tag RAM 3 Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
L1-L2 Read:
avg: AVG((TCP_TCC_READ_REQ_sum / $denom))
min: MIN((TCP_TCC_READ_REQ_sum / $denom))
@@ -419,12 +395,6 @@ Panel Config:
max: MAX((TCP_UTCL1_TRANSLATION_MISS_sum / $denom))
units: (Req + $normUnit)
tips:
Misses under Translation Miss:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Req + $normUnit)
tips:
Permission Misses:
avg: AVG((TCP_UTCL1_PERMISSION_MISS_sum / $denom))
min: MIN((TCP_UTCL1_PERMISSION_MISS_sum / $denom))
@@ -442,45 +412,3 @@ Panel Config:
units: Units
tips: Tips
metric:
Cache Full Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Cache Miss Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Serialization Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Thrashing Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Latency FIFO Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Resident Page Full Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
UTCL2 Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx90a/1700_L2_cache.yaml
-168
مشاهده پرونده
@@ -143,18 +143,6 @@ Panel Config:
!= 0) else None))
unit: Cycles
tips:
Read Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
- metric_table:
id: 1703
@@ -173,24 +161,6 @@ Panel Config:
max: MAX((TCC_REQ_sum * 128) / $denom)
unit: (Bytes + $normUnit)
tips:
Read Bandwidth:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Write Bandwidth:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Atomic Bandwidth:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Req:
avg: AVG((TCC_REQ_sum / $denom))
min: MIN((TCC_REQ_sum / $denom))
@@ -221,24 +191,12 @@ Panel Config:
max: MAX((TCC_STREAMING_REQ_sum / $denom))
unit: (Req + $normUnit)
tips:
Bypasss Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Probe Req:
avg: AVG((TCC_PROBE_sum / $denom))
min: MIN((TCC_PROBE_sum / $denom))
max: MAX((TCC_PROBE_sum / $denom))
unit: (Req + $normUnit)
tips:
Input Buffer Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Cache Hit:
avg: AVG((((100 * TCC_HIT_sum) / (TCC_HIT_sum + TCC_MISS_sum)) if ((TCC_HIT_sum
+ TCC_MISS_sum) != 0) else None))
@@ -326,24 +284,6 @@ Panel Config:
unit: Unit
tips: Tips
metric:
Stalled on Latency FIFO:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Cycles + $normUnit)
tips:
Stalled on Write Data FIFO:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Cycles + $normUnit)
tips:
Input Buffer Stalled on L2:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Cycles + $normUnit)
tips:
- metric_table:
id: 1705
@@ -360,54 +300,6 @@ Panel Config:
style:
type: simple_multi_bar
metric:
Read - PCIe Stall:
type: PCIe Stall
transaction: Read
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Read - Infinity Fabric™ Stall:
type: Infinity Fabric™ Stall
transaction: Read
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Read - HBM Stall:
type: HBM Stall
transaction: Read
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write - PCIe Stall:
type: PCIe Stall
transaction: Write
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write - Infinity Fabric™ Stall:
type: Infinity Fabric™ Stall
transaction: Write
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write - HBM Stall:
type: HBM Stall
transaction: Write
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write - Credit Starvation:
type: Credit Starvation
transaction: Write
@@ -458,24 +350,6 @@ Panel Config:
max: MAX((MAX((TCC_EA_RDREQ_sum - TCC_EA_RDREQ_DRAM_sum), 0) / $denom))
unit: (Req + $normUnit)
tips:
Read Bandwidth - PCIe:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Read Bandwidth - Infinity Fabric™:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Read Bandwidth - HBM:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Write and Atomic (32B):
avg: AVG(((TCC_EA_WRREQ_sum - TCC_EA_WRREQ_64B_sum) / $denom))
min: MIN(((TCC_EA_WRREQ_sum - TCC_EA_WRREQ_64B_sum) / $denom))
@@ -506,51 +380,9 @@ Panel Config:
max: MAX((MAX((TCC_EA_WRREQ_sum - TCC_EA_WRREQ_DRAM_sum), 0) / $denom))
unit: (Req + $normUnit)
tips:
Write Bandwidth - PCIe:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Write Bandwidth - Infinity Fabric™:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Write Bandwidth - HBM:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Atomic:
avg: AVG((TCC_EA_ATOMIC_sum / $denom))
min: MIN((TCC_EA_ATOMIC_sum / $denom))
max: MAX((TCC_EA_ATOMIC_sum / $denom))
unit: (Req + $normUnit)
tips:
Atomic - HBM:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Atomic Bandwidth - PCIe:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Atomic Bandwidth - Infinity Fabric™:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Atomic Bandwidth - HBM:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx90a/1800_L2_cache_per_channel.yaml
+15
مشاهده پرونده
@@ -333,3 +333,18 @@ Panel Config:
placeholder_range:
"::_1": $total_l2_chan
cli_style: simple_multiple_bar
- metric_table:
id: 1812
title: L2-Fabric (128B read requests per normUnit)
header:
metric: Channel
expr: Expression
metric:
"::_1":
expr: (TO_INT(TCC_BUBBLE[::_1]) / $denom)
placeholder_range:
"::_1": $total_l2_chan
# tips: Number of 128-byte read requests sent to EA
cli_style: simple_box
tui_style: simple_box
src/rocprof_compute_soc/analysis_configs/gfx940/0200_system-speed-of-light.yaml
-7
مشاهده پرونده
@@ -77,13 +77,6 @@ Panel Config:
pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_F64 * 512) / (End_Timestamp - Start_Timestamp))))
/ ((($max_sclk * $cu_per_gpu) * 256) / 1000))
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then remove this
MFMA FLOPs (F6F4):
value: None
unit: GFLOP/s
peak: None
pop: None
tips:
MFMA IOPs (Int8):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp)))
unit: GIOP/s
src/rocprof_compute_soc/analysis_configs/gfx940/0500_command-processor.yaml
-21
مشاهده پرونده
@@ -19,27 +19,6 @@ Panel Config:
unit: Unit
tips: Tips
metric:
# TODO: Fix baseline comparision logic to handle non existent metrics, then remove this
CPC SYNC FIFO Full Rate:
avg: None
min: None
max: None
unit: pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then remove this
CPC CANE Stall Rate:
avg: None
min: None
max: None
unit: pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then remove this
CPC ADC Utilization:
avg: None
min: None
max: None
unit: pct
tips:
CPF Utilization:
avg: AVG((((100 * CPF_CPF_STAT_BUSY) / (CPF_CPF_STAT_BUSY + CPF_CPF_STAT_IDLE))
if ((CPF_CPF_STAT_BUSY + CPF_CPF_STAT_IDLE) != 0) else None))
src/rocprof_compute_soc/analysis_configs/gfx940/0600_shader-processor-input.yaml
-21
مشاهده پرونده
@@ -19,13 +19,6 @@ Panel Config:
unit: Unit
tips: Tips
metric:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Schedule-Pipe Wave Occupancy:
avg: None
min: None
max: None
unit: Wave
tips:
Accelerator Utilization:
avg: AVG(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
min: MIN(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
@@ -38,13 +31,6 @@ Panel Config:
max: MAX(100 * SPI_CSN_BUSY / ($GRBM_GUI_ACTIVE_PER_XCD * $pipes_per_gpu * $se_per_gpu))
unit: Pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Scheduler-Pipe Wave Utilization:
avg: None
min: None
max: None
unit: Pct
tips:
Workgroup Manager Utilization:
avg: AVG(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
min: MIN(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
@@ -122,13 +108,6 @@ Panel Config:
0) else None)
unit: Pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Scheduler-Pipe FIFO Full Rate:
avg: None
min: None
max: None
unit: Pct
tips:
Scheduler-Pipe Stall Rate:
avg: AVG((((100 * SPI_RA_RES_STALL_CSN) / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None))
src/rocprof_compute_soc/analysis_configs/gfx940/1000_compute-unit-instruction-mix.yaml
-12
مشاهده پرونده
@@ -209,13 +209,6 @@ Panel Config:
max: MAX((TA_BUFFER_WAVEFRONTS_sum / $denom))
unit: (instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then remove this
Spill/Stack Coalesceable Instr:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
Spill/Stack Read:
avg: AVG((TA_BUFFER_READ_WAVEFRONTS_sum / $denom))
min: MIN((TA_BUFFER_READ_WAVEFRONTS_sum / $denom))
@@ -281,9 +274,4 @@ Panel Config:
min: MIN((SQ_INSTS_VALU_MFMA_F64 / $denom))
max: MAX((SQ_INSTS_VALU_MFMA_F64 / $denom))
unit: (instr + $normUnit)
MFMA-F6F4:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx940/1100_compute-unit-compute-pipeline.yaml
-21
مشاهده پرونده
@@ -76,13 +76,6 @@ Panel Config:
pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_F64 * 512) / (End_Timestamp - Start_Timestamp))))
/ ((($max_sclk * $cu_per_gpu) * 256) / 1000))
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
MFMA FLOPs (F6F4):
value: None
unit: GFLOP
peak: None
pop: None
tips:
MFMA IOPs (INT8):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp)))
unit: GIOP
@@ -132,13 +125,6 @@ Panel Config:
max: MAX((((100 * SQ_ACTIVE_INST_VALU) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
VALU Co-Issue Efficiency:
avg: None
min: None
max: None
unit: pct
tips:
VMEM Utilization:
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))
@@ -279,13 +265,6 @@ Panel Config:
+ (SQ_INSTS_VALU_FMA_F64 * 2))) + (512 * SQ_INSTS_VALU_MFMA_MOPS_F64)) / $denom))
unit: (OPs + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
F6F4 OPs:
avg: None
min: None
max: None
unit: (OPs + $normUnit)
tips:
INT8 OPs:
avg: AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / $denom))
min: MIN(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / $denom))
src/rocprof_compute_soc/analysis_configs/gfx940/1200_lds.yaml
-56
مشاهده پرونده
@@ -55,48 +55,6 @@ Panel Config:
max: MAX((SQ_INSTS_LDS / $denom))
unit: (Instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS LOAD:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS STORE:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS ATOMIC:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS LOAD Bandwidth:
avg: None
min: None
max: None
units: Gbps
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS STORE Bandwidth:
avg: None
min: None
max: None
units: Gbps
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS ATOMIC Bandwidth:
avg: None
min: None
max: None
units: Gbps
tips:
Theoretical Bandwidth:
avg: AVG(((((SQ_LDS_IDX_ACTIVE - SQ_LDS_BANK_CONFLICT) * 4) * TO_INT($lds_banks_per_cu))
/ $denom))
@@ -158,17 +116,3 @@ Panel Config:
max: MAX((SQ_LDS_MEM_VIOLATIONS / $denom))
unit: (Accesses + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS Command FIFO Full Rate:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS Data FIFO Full Rate:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx940/1500_TA_and_TD.yaml
-39
مشاهده پرونده
@@ -43,27 +43,6 @@ Panel Config:
max: MAX(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Sequencer → TA Address Stall:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Sequencer → TA Command Stall:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Sequencer → TA Data Stall:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
Total Instructions:
avg: AVG((TA_TOTAL_WAVEFRONTS_sum / $denom))
min: MIN((TA_TOTAL_WAVEFRONTS_sum / $denom))
@@ -82,12 +61,6 @@ Panel Config:
max: MAX((TA_FLAT_READ_WAVEFRONTS_sum / $denom))
unit: (Instructions + $normUnit)
tips:
Global/Generic Read Instructions for LDS:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Instructions + $normUnit)
tips:
Global/Generic Write Instructions:
avg: AVG((TA_FLAT_WRITE_WAVEFRONTS_sum / $denom))
min: MIN((TA_FLAT_WRITE_WAVEFRONTS_sum / $denom))
@@ -112,12 +85,6 @@ Panel Config:
max: MAX((TA_BUFFER_READ_WAVEFRONTS_sum / $denom))
unit: (Instructions + $normUnit)
tips:
Spill/Stack Read Instructions for LDS:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Instructions + $normUnit)
tips:
Spill/Stack Write Instructions:
avg: AVG((TA_BUFFER_WRITE_WAVEFRONTS_sum / $denom))
min: MIN((TA_BUFFER_WRITE_WAVEFRONTS_sum / $denom))
@@ -205,9 +172,3 @@ Panel Config:
max: MAX((TD_ATOMIC_WAVEFRONT_sum / $denom))
unit: (Instructions + $normUnit)
tips:
Write Ack Instructions:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Instructions + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx940/1600_L1_cache.yaml
-90
مشاهده پرونده
@@ -192,30 +192,6 @@ Panel Config:
TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / $denom))
unit: (Bytes + $normUnit)
tips:
Tag RAM 0 Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Tag RAM 1 Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Tag RAM 2 Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Tag RAM 3 Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
L1-L2 Read:
avg: AVG((TCP_TCC_READ_REQ_sum / $denom))
min: MIN((TCP_TCC_READ_REQ_sum / $denom))
@@ -237,24 +213,6 @@ Panel Config:
/ $denom))
unit: (Req + $normUnit)
tips:
L1 Access Latency:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: Cycles
tips:
L1-L2 Read Latency:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: Cycles
tips:
L1-L2 Write Latency:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: Cycles
tips:
- metric_table:
id: 1604
@@ -410,12 +368,6 @@ Panel Config:
max: MAX((TCP_UTCL1_TRANSLATION_MISS_sum / $denom))
units: (Req + $normUnit)
tips:
Misses under Translation Miss:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Req + $normUnit)
tips:
Permission Misses:
avg: AVG((TCP_UTCL1_PERMISSION_MISS_sum / $denom))
min: MIN((TCP_UTCL1_PERMISSION_MISS_sum / $denom))
@@ -433,45 +385,3 @@ Panel Config:
units: Units
tips: Tips
metric:
Cache Full Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Cache Miss Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Serialization Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Thrashing Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Latency FIFO Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Resident Page Full Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
UTCL2 Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx940/1700_L2_cache.yaml
-168
مشاهده پرونده
@@ -146,18 +146,6 @@ Panel Config:
!= 0) else None))
unit: Cycles
tips:
Read Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
- metric_table:
id: 1703
@@ -176,24 +164,6 @@ Panel Config:
max: MAX((TCC_REQ_sum * 128) / $denom)
unit: (Bytes + $normUnit)
tips:
Read Bandwidth:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Write Bandwidth:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Atomic Bandwidth:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Req:
avg: AVG((TCC_REQ_sum / $denom))
min: MIN((TCC_REQ_sum / $denom))
@@ -224,24 +194,12 @@ Panel Config:
max: MAX((TCC_STREAMING_REQ_sum / $denom))
unit: (Req + $normUnit)
tips:
Bypasss Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Probe Req:
avg: AVG((TCC_PROBE_sum / $denom))
min: MIN((TCC_PROBE_sum / $denom))
max: MAX((TCC_PROBE_sum / $denom))
unit: (Req + $normUnit)
tips:
Input Buffer Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Cache Hit:
avg: AVG((((100 * TCC_HIT_sum) / (TCC_HIT_sum + TCC_MISS_sum)) if ((TCC_HIT_sum
+ TCC_MISS_sum) != 0) else None))
@@ -329,24 +287,6 @@ Panel Config:
unit: Unit
tips: Tips
metric:
Stalled on Latency FIFO:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Cycles + $normUnit)
tips:
Stalled on Write Data FIFO:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Cycles + $normUnit)
tips:
Input Buffer Stalled on L2:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Cycles + $normUnit)
tips:
- metric_table:
id: 1705
@@ -363,54 +303,6 @@ Panel Config:
style:
type: simple_multi_bar
metric:
Read - PCIe Stall:
type: PCIe Stall
transaction: Read
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Read - Infinity Fabric™ Stall:
type: Infinity Fabric™ Stall
transaction: Read
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Read - HBM Stall:
type: HBM Stall
transaction: Read
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write - PCIe Stall:
type: PCIe Stall
transaction: Write
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write - Infinity Fabric™ Stall:
type: Infinity Fabric™ Stall
transaction: Write
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write - HBM Stall:
type: HBM Stall
transaction: Write
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write - Credit Starvation:
type: Credit Starvation
transaction: Write
@@ -461,24 +353,6 @@ Panel Config:
max: MAX((MAX((TCC_EA0_RDREQ_sum - TCC_EA0_RDREQ_DRAM_sum), 0) / $denom))
unit: (Req + $normUnit)
tips:
Read Bandwidth - PCIe:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Read Bandwidth - Infinity Fabric™:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Read Bandwidth - HBM:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Write and Atomic (32B):
avg: AVG(((TCC_EA0_WRREQ_sum - TCC_EA0_WRREQ_64B_sum) / $denom))
min: MIN(((TCC_EA0_WRREQ_sum - TCC_EA0_WRREQ_64B_sum) / $denom))
@@ -509,51 +383,9 @@ Panel Config:
max: MAX((MAX((TCC_EA0_WRREQ_sum - TCC_EA0_WRREQ_DRAM_sum), 0) / $denom))
unit: (Req + $normUnit)
tips:
Write Bandwidth - PCIe:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Write Bandwidth - Infinity Fabric™:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Write Bandwidth - HBM:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Atomic:
avg: AVG((TCC_EA0_ATOMIC_sum / $denom))
min: MIN((TCC_EA0_ATOMIC_sum / $denom))
max: MAX((TCC_EA0_ATOMIC_sum / $denom))
unit: (Req + $normUnit)
tips:
Atomic - HBM:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Atomic Bandwidth - PCIe:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Atomic Bandwidth - Infinity Fabric™:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Atomic Bandwidth - HBM:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx941/0200_system-speed-of-light.yaml
-7
مشاهده پرونده
@@ -77,13 +77,6 @@ Panel Config:
pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_F64 * 512) / (End_Timestamp - Start_Timestamp))))
/ ((($max_sclk * $cu_per_gpu) * 256) / 1000))
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then remove this
MFMA FLOPs (F6F4):
value: None
unit: GFLOP/s
peak: None
pop: None
tips:
MFMA IOPs (Int8):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp)))
unit: GIOP/s
src/rocprof_compute_soc/analysis_configs/gfx941/0500_command-processor.yaml
-21
مشاهده پرونده
@@ -19,27 +19,6 @@ Panel Config:
unit: Unit
tips: Tips
metric:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
CPC SYNC FIFO Full Rate:
avg: None
min: None
max: None
unit: pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
CPC CANE Stall Rate:
avg: None
min: None
max: None
unit: pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
CPC ADC Utilization:
avg: None
min: None
max: None
unit: pct
tips:
CPF Utilization:
avg: AVG((((100 * CPF_CPF_STAT_BUSY) / (CPF_CPF_STAT_BUSY + CPF_CPF_STAT_IDLE))
if ((CPF_CPF_STAT_BUSY + CPF_CPF_STAT_IDLE) != 0) else None))
src/rocprof_compute_soc/analysis_configs/gfx941/0600_shader-processor-input.yaml
-21
مشاهده پرونده
@@ -19,13 +19,6 @@ Panel Config:
unit: Unit
tips: Tips
metric:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Schedule-Pipe Wave Occupancy:
avg: None
min: None
max: None
unit: Wave
tips:
Accelerator Utilization:
avg: AVG(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
min: MIN(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
@@ -38,13 +31,6 @@ Panel Config:
max: MAX(100 * SPI_CSN_BUSY / ($GRBM_GUI_ACTIVE_PER_XCD * $pipes_per_gpu * $se_per_gpu))
unit: Pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Scheduler-Pipe Wave Utilization:
avg: None
min: None
max: None
unit: Pct
tips:
Workgroup Manager Utilization:
avg: AVG(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
min: MIN(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
@@ -122,13 +108,6 @@ Panel Config:
0) else None)
unit: Pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Scheduler-Pipe FIFO Full Rate:
avg: None
min: None
max: None
unit: Pct
tips:
Scheduler-Pipe Stall Rate:
avg: AVG((((100 * SPI_RA_RES_STALL_CSN) / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None))
src/rocprof_compute_soc/analysis_configs/gfx941/1000_compute-unit-instruction-mix.yaml
-14
مشاهده پرونده
@@ -209,13 +209,6 @@ Panel Config:
max: MAX((TA_BUFFER_WAVEFRONTS_sum / $denom))
unit: (instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then remove this
Spill/Stack Coalesceable Instr:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
Spill/Stack Read:
avg: AVG((TA_BUFFER_READ_WAVEFRONTS_sum / $denom))
min: MIN((TA_BUFFER_READ_WAVEFRONTS_sum / $denom))
@@ -282,10 +275,3 @@ Panel Config:
max: MAX((SQ_INSTS_VALU_MFMA_F64 / $denom))
unit: (instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
MFMA-F6F4:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx941/1100_compute-unit-compute-pipeline.yaml
-21
مشاهده پرونده
@@ -76,13 +76,6 @@ Panel Config:
pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_F64 * 512) / (End_Timestamp - Start_Timestamp))))
/ ((($max_sclk * $cu_per_gpu) * 256) / 1000))
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
MFMA FLOPs (F6F4):
value: None
unit: GFLOP
peak: None
pop: None
tips:
MFMA IOPs (INT8):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp)))
unit: GIOP
@@ -132,13 +125,6 @@ Panel Config:
max: MAX((((100 * SQ_ACTIVE_INST_VALU) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
VALU Co-Issue Efficiency:
avg: None
min: None
max: None
unit: pct
tips:
VMEM Utilization:
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))
@@ -279,13 +265,6 @@ Panel Config:
+ (SQ_INSTS_VALU_FMA_F64 * 2))) + (512 * SQ_INSTS_VALU_MFMA_MOPS_F64)) / $denom))
unit: (OPs + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
F6F4 OPs:
avg: None
min: None
max: None
unit: (OPs + $normUnit)
tips:
INT8 OPs:
avg: AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / $denom))
min: MIN(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / $denom))
src/rocprof_compute_soc/analysis_configs/gfx941/1200_lds.yaml
-56
مشاهده پرونده
@@ -55,48 +55,6 @@ Panel Config:
max: MAX((SQ_INSTS_LDS / $denom))
unit: (Instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS LOAD:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS STORE:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS ATOMIC:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS LOAD Bandwidth:
avg: None
min: None
max: None
units: Gbps
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS STORE Bandwidth:
avg: None
min: None
max: None
units: Gbps
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS ATOMIC Bandwidth:
avg: None
min: None
max: None
units: Gbps
tips:
Theoretical Bandwidth:
avg: AVG(((((SQ_LDS_IDX_ACTIVE - SQ_LDS_BANK_CONFLICT) * 4) * TO_INT($lds_banks_per_cu))
/ $denom))
@@ -158,17 +116,3 @@ Panel Config:
max: MAX((SQ_LDS_MEM_VIOLATIONS / $denom))
unit: (Accesses + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS Command FIFO Full Rate:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS Data FIFO Full Rate:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx941/1500_TA_and_TD.yaml
-39
مشاهده پرونده
@@ -43,27 +43,6 @@ Panel Config:
max: MAX(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Sequencer → TA Address Stall:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Sequencer → TA Command Stall:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Sequencer → TA Data Stall:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
Total Instructions:
avg: AVG((TA_TOTAL_WAVEFRONTS_sum / $denom))
min: MIN((TA_TOTAL_WAVEFRONTS_sum / $denom))
@@ -82,12 +61,6 @@ Panel Config:
max: MAX((TA_FLAT_READ_WAVEFRONTS_sum / $denom))
unit: (Instructions + $normUnit)
tips:
Global/Generic Read Instructions for LDS:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Instructions + $normUnit)
tips:
Global/Generic Write Instructions:
avg: AVG((TA_FLAT_WRITE_WAVEFRONTS_sum / $denom))
min: MIN((TA_FLAT_WRITE_WAVEFRONTS_sum / $denom))
@@ -112,12 +85,6 @@ Panel Config:
max: MAX((TA_BUFFER_READ_WAVEFRONTS_sum / $denom))
unit: (Instructions + $normUnit)
tips:
Spill/Stack Read Instructions for LDS:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Instructions + $normUnit)
tips:
Spill/Stack Write Instructions:
avg: AVG((TA_BUFFER_WRITE_WAVEFRONTS_sum / $denom))
min: MIN((TA_BUFFER_WRITE_WAVEFRONTS_sum / $denom))
@@ -205,9 +172,3 @@ Panel Config:
max: MAX((TD_ATOMIC_WAVEFRONT_sum / $denom))
unit: (Instructions + $normUnit)
tips:
Write Ack Instructions:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Instructions + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx941/1600_L1_cache.yaml
-90
مشاهده پرونده
@@ -192,30 +192,6 @@ Panel Config:
TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / $denom))
unit: (Bytes + $normUnit)
tips:
Tag RAM 0 Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Tag RAM 1 Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Tag RAM 2 Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Tag RAM 3 Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
L1-L2 Read:
avg: AVG((TCP_TCC_READ_REQ_sum / $denom))
min: MIN((TCP_TCC_READ_REQ_sum / $denom))
@@ -237,24 +213,6 @@ Panel Config:
/ $denom))
unit: (Req + $normUnit)
tips:
L1 Access Latency:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: Cycles
tips:
L1-L2 Read Latency:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: Cycles
tips:
L1-L2 Write Latency:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: Cycles
tips:
- metric_table:
id: 1604
@@ -410,12 +368,6 @@ Panel Config:
max: MAX((TCP_UTCL1_TRANSLATION_MISS_sum / $denom))
units: (Req + $normUnit)
tips:
Misses under Translation Miss:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Req + $normUnit)
tips:
Permission Misses:
avg: AVG((TCP_UTCL1_PERMISSION_MISS_sum / $denom))
min: MIN((TCP_UTCL1_PERMISSION_MISS_sum / $denom))
@@ -433,45 +385,3 @@ Panel Config:
units: Units
tips: Tips
metric:
Cache Full Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Cache Miss Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Serialization Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Thrashing Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Latency FIFO Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Resident Page Full Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
UTCL2 Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx941/1700_L2_cache.yaml
-168
مشاهده پرونده
@@ -146,18 +146,6 @@ Panel Config:
!= 0) else None))
unit: Cycles
tips:
Read Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
- metric_table:
id: 1703
@@ -176,24 +164,6 @@ Panel Config:
max: MAX((TCC_REQ_sum * 128) / $denom)
unit: (Bytes + $normUnit)
tips:
Read Bandwidth:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Write Bandwidth:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Atomic Bandwidth:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Req:
avg: AVG((TCC_REQ_sum / $denom))
min: MIN((TCC_REQ_sum / $denom))
@@ -224,24 +194,12 @@ Panel Config:
max: MAX((TCC_STREAMING_REQ_sum / $denom))
unit: (Req + $normUnit)
tips:
Bypasss Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Probe Req:
avg: AVG((TCC_PROBE_sum / $denom))
min: MIN((TCC_PROBE_sum / $denom))
max: MAX((TCC_PROBE_sum / $denom))
unit: (Req + $normUnit)
tips:
Input Buffer Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Cache Hit:
avg: AVG((((100 * TCC_HIT_sum) / (TCC_HIT_sum + TCC_MISS_sum)) if ((TCC_HIT_sum
+ TCC_MISS_sum) != 0) else None))
@@ -329,24 +287,6 @@ Panel Config:
unit: Unit
tips: Tips
metric:
Stalled on Latency FIFO:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Cycles + $normUnit)
tips:
Stalled on Write Data FIFO:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Cycles + $normUnit)
tips:
Input Buffer Stalled on L2:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Cycles + $normUnit)
tips:
- metric_table:
id: 1705
@@ -363,54 +303,6 @@ Panel Config:
style:
type: simple_multi_bar
metric:
Read - PCIe Stall:
type: PCIe Stall
transaction: Read
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Read - Infinity Fabric™ Stall:
type: Infinity Fabric™ Stall
transaction: Read
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Read - HBM Stall:
type: HBM Stall
transaction: Read
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write - PCIe Stall:
type: PCIe Stall
transaction: Write
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write - Infinity Fabric™ Stall:
type: Infinity Fabric™ Stall
transaction: Write
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write - HBM Stall:
type: HBM Stall
transaction: Write
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write - Credit Starvation:
type: Credit Starvation
transaction: Write
@@ -461,24 +353,6 @@ Panel Config:
max: MAX((MAX((TCC_EA0_RDREQ_sum - TCC_EA0_RDREQ_DRAM_sum), 0) / $denom))
unit: (Req + $normUnit)
tips:
Read Bandwidth - PCIe:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Read Bandwidth - Infinity Fabric™:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Read Bandwidth - HBM:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Write and Atomic (32B):
avg: AVG(((TCC_EA0_WRREQ_sum - TCC_EA0_WRREQ_64B_sum) / $denom))
min: MIN(((TCC_EA0_WRREQ_sum - TCC_EA0_WRREQ_64B_sum) / $denom))
@@ -509,51 +383,9 @@ Panel Config:
max: MAX((MAX((TCC_EA0_WRREQ_sum - TCC_EA0_WRREQ_DRAM_sum), 0) / $denom))
unit: (Req + $normUnit)
tips:
Write Bandwidth - PCIe:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Write Bandwidth - Infinity Fabric™:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Write Bandwidth - HBM:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Atomic:
avg: AVG((TCC_EA0_ATOMIC_sum / $denom))
min: MIN((TCC_EA0_ATOMIC_sum / $denom))
max: MAX((TCC_EA0_ATOMIC_sum / $denom))
unit: (Req + $normUnit)
tips:
Atomic - HBM:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Atomic Bandwidth - PCIe:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Atomic Bandwidth - Infinity Fabric™:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Atomic Bandwidth - HBM:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx942/0200_system-speed-of-light.yaml
-7
مشاهده پرونده
@@ -77,13 +77,6 @@ Panel Config:
pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_F64 * 512) / (End_Timestamp - Start_Timestamp))))
/ ((($max_sclk * $cu_per_gpu) * 256) / 1000))
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then remove this
MFMA FLOPs (F6F4):
value: None
unit: GFLOP/s
peak: None
pop: None
tips:
MFMA IOPs (Int8):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp)))
unit: GIOP/s
src/rocprof_compute_soc/analysis_configs/gfx942/0500_command-processor.yaml
-21
مشاهده پرونده
@@ -76,27 +76,6 @@ Panel Config:
unit: Unit
tips: Tips
metric:
# TODO: Fix baseline comparision logic to handle non existent metrics, then remove this
CPC SYNC FIFO Full Rate:
avg: None
min: None
max: None
unit: pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then remove this
CPC CANE Stall Rate:
avg: None
min: None
max: None
unit: pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then remove this
CPC ADC Utilization:
avg: None
min: None
max: None
unit: pct
tips:
CPC Utilization:
avg: AVG((((100 * CPC_CPC_STAT_BUSY) / (CPC_CPC_STAT_BUSY + CPC_CPC_STAT_IDLE))
if ((CPC_CPC_STAT_BUSY + CPC_CPC_STAT_IDLE) != 0) else None))
src/rocprof_compute_soc/analysis_configs/gfx942/0600_shader-processor-input.yaml
-21
مشاهده پرونده
@@ -19,13 +19,6 @@ Panel Config:
unit: Unit
tips: Tips
metric:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Schedule-Pipe Wave Occupancy:
avg: None
min: None
max: None
unit: Wave
tips:
Accelerator Utilization:
avg: AVG(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
min: MIN(100 * $GRBM_GUI_ACTIVE_PER_XCD / $GRBM_COUNT_PER_XCD)
@@ -38,13 +31,6 @@ Panel Config:
max: MAX(100 * SPI_CSN_BUSY / ($GRBM_GUI_ACTIVE_PER_XCD * $pipes_per_gpu * $se_per_gpu))
unit: Pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Scheduler-Pipe Wave Utilization:
avg: None
min: None
max: None
unit: Pct
tips:
Workgroup Manager Utilization:
avg: AVG(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
min: MIN(100 * $GRBM_SPI_BUSY_PER_XCD / $GRBM_GUI_ACTIVE_PER_XCD)
@@ -122,13 +108,6 @@ Panel Config:
0) else None)
unit: Pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Scheduler-Pipe FIFO Full Rate:
avg: None
min: None
max: None
unit: Pct
tips:
Scheduler-Pipe Stall Rate:
avg: AVG((((100 * SPI_RA_RES_STALL_CSN) / ($GRBM_SPI_BUSY_PER_XCD * $se_per_gpu)) if ($GRBM_SPI_BUSY_PER_XCD !=
0) else None))
src/rocprof_compute_soc/analysis_configs/gfx942/1000_compute-unit-instruction-mix.yaml
-14
مشاهده پرونده
@@ -209,13 +209,6 @@ Panel Config:
max: MAX((TA_BUFFER_WAVEFRONTS_sum / $denom))
unit: (instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then remove this
Spill/Stack Coalesceable Instr:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
Spill/Stack Read:
avg: AVG((TA_BUFFER_READ_WAVEFRONTS_sum / $denom))
min: MIN((TA_BUFFER_READ_WAVEFRONTS_sum / $denom))
@@ -282,10 +275,3 @@ Panel Config:
max: MAX((SQ_INSTS_VALU_MFMA_F64 / $denom))
unit: (instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
MFMA-F6F4:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx942/1100_compute-unit-compute-pipeline.yaml
-21
مشاهده پرونده
@@ -76,13 +76,6 @@ Panel Config:
pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_F64 * 512) / (End_Timestamp - Start_Timestamp))))
/ ((($max_sclk * $cu_per_gpu) * 256) / 1000))
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
MFMA FLOPs (F6F4):
value: None
unit: GFLOP
peak: None
pop: None
tips:
MFMA IOPs (INT8):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp)))
unit: GIOP
@@ -132,13 +125,6 @@ Panel Config:
max: MAX((((100 * SQ_ACTIVE_INST_VALU) / $GRBM_GUI_ACTIVE_PER_XCD) / $cu_per_gpu))
unit: pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
VALU Co-Issue Efficiency:
avg: None
min: None
max: None
unit: pct
tips:
VMEM Utilization:
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))
@@ -279,13 +265,6 @@ Panel Config:
+ (SQ_INSTS_VALU_FMA_F64 * 2))) + (512 * SQ_INSTS_VALU_MFMA_MOPS_F64)) / $denom))
unit: (OPs + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
F6F4 OPs:
avg: None
min: None
max: None
unit: (OPs + $normUnit)
tips:
INT8 OPs:
avg: AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / $denom))
min: MIN(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / $denom))
src/rocprof_compute_soc/analysis_configs/gfx942/1200_lds.yaml
-56
مشاهده پرونده
@@ -56,48 +56,6 @@ Panel Config:
max: MAX((SQ_INSTS_LDS / $denom))
unit: (Instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS LOAD:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS STORE:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS ATOMIC:
avg: None
min: None
max: None
unit: (instr + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS LOAD Bandwidth:
avg: None
min: None
max: None
units: Gbps
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS STORE Bandwidth:
avg: None
min: None
max: None
units: Gbps
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS ATOMIC Bandwidth:
avg: None
min: None
max: None
units: Gbps
tips:
Theoretical Bandwidth:
avg: AVG(((((SQ_LDS_IDX_ACTIVE - SQ_LDS_BANK_CONFLICT) * 4) * TO_INT($lds_banks_per_cu))
/ $denom))
@@ -159,17 +117,3 @@ Panel Config:
max: MAX((SQ_LDS_MEM_VIOLATIONS / $denom))
unit: (Accesses + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS Command FIFO Full Rate:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
LDS Data FIFO Full Rate:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx942/1500_TA_and_TD.yaml
-39
مشاهده پرونده
@@ -43,27 +43,6 @@ Panel Config:
max: MAX(((100 * TA_ADDR_STALLED_BY_TD_CYCLES_sum) / ($GRBM_GUI_ACTIVE_PER_XCD * $cu_per_gpu)))
unit: pct
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Sequencer → TA Address Stall:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Sequencer → TA Command Stall:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
# TODO: Fix baseline comparision logic to handle non existent metrics, then
Sequencer → TA Data Stall:
avg: None
min: None
max: None
unit: (Cycles + $normUnit)
tips:
Total Instructions:
avg: AVG((TA_TOTAL_WAVEFRONTS_sum / $denom))
min: MIN((TA_TOTAL_WAVEFRONTS_sum / $denom))
@@ -82,12 +61,6 @@ Panel Config:
max: MAX((TA_FLAT_READ_WAVEFRONTS_sum / $denom))
unit: (Instructions + $normUnit)
tips:
Global/Generic Read Instructions for LDS:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Instructions + $normUnit)
tips:
Global/Generic Write Instructions:
avg: AVG((TA_FLAT_WRITE_WAVEFRONTS_sum / $denom))
min: MIN((TA_FLAT_WRITE_WAVEFRONTS_sum / $denom))
@@ -112,12 +85,6 @@ Panel Config:
max: MAX((TA_BUFFER_READ_WAVEFRONTS_sum / $denom))
unit: (Instructions + $normUnit)
tips:
Spill/Stack Read Instructions for LDS:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Instructions + $normUnit)
tips:
Spill/Stack Write Instructions:
avg: AVG((TA_BUFFER_WRITE_WAVEFRONTS_sum / $denom))
min: MIN((TA_BUFFER_WRITE_WAVEFRONTS_sum / $denom))
@@ -205,9 +172,3 @@ Panel Config:
max: MAX((TD_ATOMIC_WAVEFRONT_sum / $denom))
unit: (Instructions + $normUnit)
tips:
Write Ack Instructions:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Instructions + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx942/1600_L1_cache.yaml
-90
مشاهده پرونده
@@ -194,30 +194,6 @@ Panel Config:
TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum)) / $denom))
unit: (Bytes + $normUnit)
tips:
Tag RAM 0 Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Tag RAM 1 Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Tag RAM 2 Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Tag RAM 3 Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
L1-L2 Read:
avg: AVG((TCP_TCC_READ_REQ_sum / $denom))
min: MIN((TCP_TCC_READ_REQ_sum / $denom))
@@ -239,24 +215,6 @@ Panel Config:
/ $denom))
unit: (Req + $normUnit)
tips:
L1 Access Latency:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: Cycles
tips:
L1-L2 Read Latency:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: Cycles
tips:
L1-L2 Write Latency:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: Cycles
tips:
- metric_table:
id: 1604
@@ -412,12 +370,6 @@ Panel Config:
max: MAX((TCP_UTCL1_TRANSLATION_MISS_sum / $denom))
units: (Req + $normUnit)
tips:
Misses under Translation Miss:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Req + $normUnit)
tips:
Permission Misses:
avg: AVG((TCP_UTCL1_PERMISSION_MISS_sum / $denom))
min: MIN((TCP_UTCL1_PERMISSION_MISS_sum / $denom))
@@ -435,45 +387,3 @@ Panel Config:
units: Units
tips: Tips
metric:
Cache Full Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Cache Miss Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Serialization Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Thrashing Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Latency FIFO Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
Resident Page Full Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
UTCL2 Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
units: (Cycles + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx942/1700_L2_cache.yaml
-168
مشاهده پرونده
@@ -150,18 +150,6 @@ Panel Config:
!= 0) else None))
unit: Cycles
tips:
Read Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write Stall:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
- metric_table:
id: 1703
@@ -180,24 +168,6 @@ Panel Config:
max: MAX((TCC_REQ_sum * 128) / $denom)
unit: (Bytes + $normUnit)
tips:
Read Bandwidth:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Write Bandwidth:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Atomic Bandwidth:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Req:
avg: AVG((TCC_REQ_sum / $denom))
min: MIN((TCC_REQ_sum / $denom))
@@ -228,24 +198,12 @@ Panel Config:
max: MAX((TCC_STREAMING_REQ_sum / $denom))
unit: (Req + $normUnit)
tips:
Bypasss Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Probe Req:
avg: AVG((TCC_PROBE_sum / $denom))
min: MIN((TCC_PROBE_sum / $denom))
max: MAX((TCC_PROBE_sum / $denom))
unit: (Req + $normUnit)
tips:
Input Buffer Req:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Cache Hit:
avg: AVG((((100 * TCC_HIT_sum) / (TCC_HIT_sum + TCC_MISS_sum)) if ((TCC_HIT_sum
+ TCC_MISS_sum) != 0) else None))
@@ -333,24 +291,6 @@ Panel Config:
unit: Unit
tips: Tips
metric:
Stalled on Latency FIFO:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Cycles + $normUnit)
tips:
Stalled on Write Data FIFO:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Cycles + $normUnit)
tips:
Input Buffer Stalled on L2:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Cycles + $normUnit)
tips:
- metric_table:
id: 1705
@@ -367,54 +307,6 @@ Panel Config:
style:
type: simple_multi_bar
metric:
Read - PCIe Stall:
type: PCIe Stall
transaction: Read
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Read - Infinity Fabric™ Stall:
type: Infinity Fabric™ Stall
transaction: Read
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Read - HBM Stall:
type: HBM Stall
transaction: Read
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write - PCIe Stall:
type: PCIe Stall
transaction: Write
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write - Infinity Fabric™ Stall:
type: Infinity Fabric™ Stall
transaction: Write
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write - HBM Stall:
type: HBM Stall
transaction: Write
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: pct
tips:
Write - Credit Starvation:
type: Credit Starvation
transaction: Write
@@ -471,24 +363,6 @@ Panel Config:
max: MAX((MAX((TCC_EA0_RDREQ_sum - TCC_EA0_RDREQ_DRAM_sum), 0) / $denom))
unit: (Req + $normUnit)
tips:
Read Bandwidth - PCIe:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Read Bandwidth - Infinity Fabric™:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Read Bandwidth - HBM:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Write and Atomic (32B):
avg: AVG(((TCC_EA0_WRREQ_sum - TCC_EA0_WRREQ_64B_sum) / $denom))
min: MIN(((TCC_EA0_WRREQ_sum - TCC_EA0_WRREQ_64B_sum) / $denom))
@@ -519,51 +393,9 @@ Panel Config:
max: MAX((MAX((TCC_EA0_WRREQ_sum - TCC_EA0_WRREQ_DRAM_sum), 0) / $denom))
unit: (Req + $normUnit)
tips:
Write Bandwidth - PCIe:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Write Bandwidth - Infinity Fabric™:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Write Bandwidth - HBM:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Atomic:
avg: AVG((TCC_EA0_ATOMIC_sum / $denom))
min: MIN((TCC_EA0_ATOMIC_sum / $denom))
max: MAX((TCC_EA0_ATOMIC_sum / $denom))
unit: (Req + $normUnit)
tips:
Atomic - HBM:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Req + $normUnit)
tips:
Atomic Bandwidth - PCIe:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Atomic Bandwidth - Infinity Fabric™:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
Atomic Bandwidth - HBM:
avg: None # Missing perfmon
min: None # Missing perfmon
max: None # Missing perfmon
unit: (Bytes + $normUnit)
tips:
src/rocprof_compute_soc/analysis_configs/gfx950/1700_L2_cache.yaml
-18
مشاهده پرونده
@@ -473,24 +473,6 @@ Panel Config:
max: MAX((MAX((TCC_EA0_RDREQ_sum - TCC_EA0_RDREQ_DRAM_sum), 0) / $denom))
unit: (Req + $normUnit)
tips:
Read Bandwidth - PCIe:
avg: None # AVG(TCC_EA0_RDREQ_IO_32B_sum / $denom)
min: None # MIN(TCC_EA0_RDREQ_IO_32B_sum / $denom)
max: None # MAX(TCC_EA0_RDREQ_IO_32B_sum / $denom)
unit: (Bytes + $normUnit)
tips:
Read Bandwidth - Infinity Fabric™:
avg: None # AVG(TCC_EA0_RDREQ_GMI_32B_sum / $denom)
min: None # MIN(TCC_EA0_RDREQ_GMI_32B_sum / $denom)
max: None # MAX(TCC_EA0_RDREQ_GMI_32B_sum / $denom)
unit: (Bytes + $normUnit)
tips:
Read Bandwidth - HBM:
avg: None # AVG(TCC_EA0_RDREQ_DRAM_32B_sum / $denom)
min: None # MIN(TCC_EA0_RDREQ_DRAM_32B_sum / $denom)
max: None # MAX(TCC_EA0_RDREQ_DRAM_32B_sum / $denom)
unit: (Bytes + $normUnit)
tips:
Write and Atomic (32B):
avg: AVG(((TCC_EA0_WRREQ_sum - TCC_EA0_WRREQ_64B_sum) / $denom))
min: MIN(((TCC_EA0_WRREQ_sum - TCC_EA0_WRREQ_64B_sum) / $denom))
src/utils/file_io.py
+6
مشاهده پرونده
@@ -75,6 +75,12 @@ def load_panel_configs(dir):
if f.endswith(".yaml"):
with open(str(Path(root).joinpath(f))) as file:
config = yaml.safe_load(file)
# metric key can be None due to some metric tables not having any metrics
# metric key should be empty dict instead of None
for data_source in config["Panel Config"]["data source"]:
metric_table = data_source.get("metric_table")
if metric_table and metric_table["metric"] is None:
metric_table["metric"] = {}
d[config["Panel Config"]["id"]] = config["Panel Config"]
# TODO: sort metrics as the header order in case they are not defined in the same order
src/utils/tty.py
+28
مشاهده پرونده
@@ -114,6 +114,34 @@ def show_all(args, runs, archConfigs, output, profiling_config, roof_plot=None):
show_roof_plot(roof_plot)
continue
# Metrics baseline comparison mode
# We cannot guarantee that all runs have the same metrics. Only show common metrics.
if (
type == "metric_table"
and "Metric" in table_config["header"].values()
and len(runs) > 1
):
# Common metrics across all runs
common_metrics = set()
for _, data in runs.items():
if not common_metrics:
common_metrics = set(data.dfs[table_config["id"]]["Metric"])
else:
common_metrics &= set(data.dfs[table_config["id"]]["Metric"])
# Apply common metrics across all runs
# Reindex all runs based on first run
initial_index = None
for key in runs.keys():
runs[key].dfs[table_config["id"]] = (
runs[key]
.dfs[table_config["id"]]
.loc[lambda d: d["Metric"].isin(common_metrics)]
)
if initial_index is None:
initial_index= runs[key].dfs[table_config["id"]].index
else:
runs[key].dfs[table_config["id"]].index = initial_index
# take the 1st run as baseline
base_run, base_data = next(iter(runs.items()))
base_df = base_data.dfs[table_config["id"]]