fixup MI300 MFMA rate disagreement between SOL and pipeline (#315)

Signed-off-by: Nicholas Curtis <nicurtis@amd.com>
Co-authored-by: Nicholas Curtis <nicurtis@amd.com>

[ROCm/rocprofiler-compute commit: b610e1c27b]
이 커밋은 다음에 포함됨:
Nick Curtis그리고Nicholas Curtis 작성하고 GitHub가 2024-03-14 10:41:35 -05:00를 커밋함
1 부모 4be73b88c6
커밋 c5408fb22b
6 파일이 변경됨
+24 -24

검색과 일치하는 파일 없음

@@ -73,9 +73,9 @@ Panel Config:
MFMA IOPs (Int8):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp)))
unit: GIOP
peak: ((($max_sclk * $cu_per_gpu) * 4096) / 1000)
peak: ((($max_sclk * $cu_per_gpu) * 8192) / 1000)
pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp))))
/ ((($max_sclk * $cu_per_gpu) * 4096) / 1000))
/ ((($max_sclk * $cu_per_gpu) * 8192) / 1000))
tips:
Active CUs:
value: $numActiveCUs
@@ -44,16 +44,16 @@ Panel Config:
MFMA FLOPs (BF16):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_BF16 * 512) / (End_Timestamp - Start_Timestamp)))
unit: GFLOP
peak: ((($max_sclk * $cu_per_gpu) * 1024) / 1000)
peak: ((($max_sclk * $cu_per_gpu) * 4096) / 1000)
pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_BF16 * 512) / (End_Timestamp - Start_Timestamp))))
/ ((($max_sclk * $cu_per_gpu) * 1024) / 1000))
/ ((($max_sclk * $cu_per_gpu) * 4096) / 1000))
tips:
MFMA FLOPs (F16):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_F16 * 512) / (End_Timestamp - Start_Timestamp)))
unit: GFLOP
peak: ((($max_sclk * $cu_per_gpu) * 1024) / 1000)
peak: ((($max_sclk * $cu_per_gpu) * 4096) / 1000)
pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_F16 * 512) / (End_Timestamp - Start_Timestamp))))
/ ((($max_sclk * $cu_per_gpu) * 1024) / 1000))
/ ((($max_sclk * $cu_per_gpu) * 4096) / 1000))
tips:
MFMA FLOPs (F32):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_F32 * 512) / (End_Timestamp - Start_Timestamp)))
@@ -72,9 +72,9 @@ Panel Config:
MFMA IOPs (INT8):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp)))
unit: GIOP
peak: ((($max_sclk * $cu_per_gpu) * 1024) / 1000)
peak: ((($max_sclk * $cu_per_gpu) * 8192) / 1000)
pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp))))
/ ((($max_sclk * $cu_per_gpu) * 1024) / 1000))
/ ((($max_sclk * $cu_per_gpu) * 8192) / 1000))
tips:
- metric_table:
@@ -73,9 +73,9 @@ Panel Config:
MFMA IOPs (Int8):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp)))
unit: GIOP
peak: ((($max_sclk * $cu_per_gpu) * 4096) / 1000)
peak: ((($max_sclk * $cu_per_gpu) * 8192) / 1000)
pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp))))
/ ((($max_sclk * $cu_per_gpu) * 4096) / 1000))
/ ((($max_sclk * $cu_per_gpu) * 8192) / 1000))
tips:
Active CUs:
value: $numActiveCUs
@@ -44,16 +44,16 @@ Panel Config:
MFMA FLOPs (BF16):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_BF16 * 512) / (End_Timestamp - Start_Timestamp)))
unit: GFLOP
peak: ((($max_sclk * $cu_per_gpu) * 1024) / 1000)
peak: ((($max_sclk * $cu_per_gpu) * 4096) / 1000)
pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_BF16 * 512) / (End_Timestamp - Start_Timestamp))))
/ ((($max_sclk * $cu_per_gpu) * 1024) / 1000))
/ ((($max_sclk * $cu_per_gpu) * 4096) / 1000))
tips:
MFMA FLOPs (F16):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_F16 * 512) / (End_Timestamp - Start_Timestamp)))
unit: GFLOP
peak: ((($max_sclk * $cu_per_gpu) * 1024) / 1000)
peak: ((($max_sclk * $cu_per_gpu) * 4096) / 1000)
pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_F16 * 512) / (End_Timestamp - Start_Timestamp))))
/ ((($max_sclk * $cu_per_gpu) * 1024) / 1000))
/ ((($max_sclk * $cu_per_gpu) * 4096) / 1000))
tips:
MFMA FLOPs (F32):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_F32 * 512) / (End_Timestamp - Start_Timestamp)))
@@ -72,9 +72,9 @@ Panel Config:
MFMA IOPs (INT8):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp)))
unit: GIOP
peak: ((($max_sclk * $cu_per_gpu) * 1024) / 1000)
peak: ((($max_sclk * $cu_per_gpu) * 8192) / 1000)
pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp))))
/ ((($max_sclk * $cu_per_gpu) * 1024) / 1000))
/ ((($max_sclk * $cu_per_gpu) * 8192) / 1000))
tips:
- metric_table:
@@ -73,9 +73,9 @@ Panel Config:
MFMA IOPs (Int8):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp)))
unit: GIOP
peak: ((($max_sclk * $cu_per_gpu) * 4096) / 1000)
peak: ((($max_sclk * $cu_per_gpu) * 8192) / 1000)
pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp))))
/ ((($max_sclk * $cu_per_gpu) * 4096) / 1000))
/ ((($max_sclk * $cu_per_gpu) * 8192) / 1000))
tips:
Active CUs:
value: $numActiveCUs
@@ -44,16 +44,16 @@ Panel Config:
MFMA FLOPs (BF16):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_BF16 * 512) / (End_Timestamp - Start_Timestamp)))
unit: GFLOP
peak: ((($max_sclk * $cu_per_gpu) * 1024) / 1000)
peak: ((($max_sclk * $cu_per_gpu) * 4096) / 1000)
pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_BF16 * 512) / (End_Timestamp - Start_Timestamp))))
/ ((($max_sclk * $cu_per_gpu) * 1024) / 1000))
/ ((($max_sclk * $cu_per_gpu) * 4096) / 1000))
tips:
MFMA FLOPs (F16):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_F16 * 512) / (End_Timestamp - Start_Timestamp)))
unit: GFLOP
peak: ((($max_sclk * $cu_per_gpu) * 1024) / 1000)
peak: ((($max_sclk * $cu_per_gpu) * 4096) / 1000)
pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_F16 * 512) / (End_Timestamp - Start_Timestamp))))
/ ((($max_sclk * $cu_per_gpu) * 1024) / 1000))
/ ((($max_sclk * $cu_per_gpu) * 4096) / 1000))
tips:
MFMA FLOPs (F32):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_F32 * 512) / (End_Timestamp - Start_Timestamp)))
@@ -72,9 +72,9 @@ Panel Config:
MFMA IOPs (INT8):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp)))
unit: GIOP
peak: ((($max_sclk * $cu_per_gpu) * 1024) / 1000)
peak: ((($max_sclk * $cu_per_gpu) * 8192) / 1000)
pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_I8 * 512) / (End_Timestamp - Start_Timestamp))))
/ ((($max_sclk * $cu_per_gpu) * 1024) / 1000))
/ ((($max_sclk * $cu_per_gpu) * 8192) / 1000))
tips:
- metric_table: