Merge pull request #460 from feizheng10/doc_warp_efficency

Fix PoP of VALU Active Threads with wave_size

[ROCm/rocprofiler-compute commit: b4f9b38e99]
Этот коммит содержится в:
Fei Zheng
2024-11-05 09:00:59 -07:00
коммит произвёл GitHub
родитель 5a30a05800 f1c9c686ff
Коммит 03ffa7a756
6 изменённых файлов: 11 добавлений и 15 удалений
@@ -102,9 +102,8 @@ Panel Config:
value: AVG(((SQ_THREAD_CYCLES_VALU / SQ_ACTIVE_INST_VALU) if (SQ_ACTIVE_INST_VALU
!= 0) else None))
unit: Threads
peak: 64
pop: (AVG(((SQ_THREAD_CYCLES_VALU / SQ_ACTIVE_INST_VALU) if (SQ_ACTIVE_INST_VALU
!= 0) else None)) * 1.5625)
peak: $wave_size
pop: (100 * AVG((SQ_THREAD_CYCLES_VALU / SQ_ACTIVE_INST_VALU / $wave_size) if (SQ_ACTIVE_INST_VALU != 0) else None))
tips:
IPC:
value: AVG((SQ_INSTS / SQ_BUSY_CU_CYCLES))
@@ -102,9 +102,8 @@ Panel Config:
value: AVG(((SQ_THREAD_CYCLES_VALU / SQ_ACTIVE_INST_VALU) if (SQ_ACTIVE_INST_VALU
!= 0) else None))
unit: Threads
peak: 64
pop: (AVG(((SQ_THREAD_CYCLES_VALU / SQ_ACTIVE_INST_VALU) if (SQ_ACTIVE_INST_VALU
!= 0) else None)) * 1.5625)
peak: $wave_size
pop: (100 * AVG((SQ_THREAD_CYCLES_VALU / SQ_ACTIVE_INST_VALU / $wave_size) if (SQ_ACTIVE_INST_VALU != 0) else None))
tips:
IPC:
value: AVG((SQ_INSTS / SQ_BUSY_CU_CYCLES))
@@ -119,9 +119,8 @@ Panel Config:
value: AVG(((SQ_THREAD_CYCLES_VALU / SQ_ACTIVE_INST_VALU) if (SQ_ACTIVE_INST_VALU
!= 0) else None))
unit: Threads
peak: 64
pop: (AVG(((SQ_THREAD_CYCLES_VALU / SQ_ACTIVE_INST_VALU) if (SQ_ACTIVE_INST_VALU
!= 0) else None)) * 1.5625)
peak: $wave_size
pop: (100 * AVG((SQ_THREAD_CYCLES_VALU / SQ_ACTIVE_INST_VALU / $wave_size) if (SQ_ACTIVE_INST_VALU != 0) else None))
tips:
IPC:
value: AVG((SQ_INSTS / SQ_BUSY_CU_CYCLES))
@@ -119,9 +119,8 @@ Panel Config:
value: AVG(((SQ_THREAD_CYCLES_VALU / SQ_ACTIVE_INST_VALU) if (SQ_ACTIVE_INST_VALU
!= 0) else None))
unit: Threads
peak: 64
pop: (AVG(((SQ_THREAD_CYCLES_VALU / SQ_ACTIVE_INST_VALU) if (SQ_ACTIVE_INST_VALU
!= 0) else None)) * 1.5625)
peak: $wave_size
pop: (100 * AVG((SQ_THREAD_CYCLES_VALU / SQ_ACTIVE_INST_VALU / $wave_size) if (SQ_ACTIVE_INST_VALU != 0) else None))
tips:
IPC:
value: AVG((SQ_INSTS / SQ_BUSY_CU_CYCLES))
@@ -119,9 +119,8 @@ Panel Config:
value: AVG(((SQ_THREAD_CYCLES_VALU / SQ_ACTIVE_INST_VALU) if (SQ_ACTIVE_INST_VALU
!= 0) else None))
unit: Threads
peak: 64
pop: (AVG(((SQ_THREAD_CYCLES_VALU / SQ_ACTIVE_INST_VALU) if (SQ_ACTIVE_INST_VALU
!= 0) else None)) * 1.5625)
peak: $wave_size
pop: (100 * AVG((SQ_THREAD_CYCLES_VALU / SQ_ACTIVE_INST_VALU / $wave_size) if (SQ_ACTIVE_INST_VALU != 0) else None))
tips:
IPC:
value: AVG((SQ_INSTS / SQ_BUSY_CU_CYCLES))
+1
Просмотреть файл
@@ -698,6 +698,7 @@ def eval_metric(dfs, dfs_type, sys_info, raw_pmc_df, debug):
ammolite__hbm_bw = sys_info.hbm_bw
ammolite__total_l2_chan = calc_builtin_var("$total_l2_chan", sys_info)
ammolite__num_xcd = sys_info.num_xcd
ammolite__wave_size = sys_info.wave_size
# TODO: fix all $normUnit in Unit column or title