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]
Esse commit está contido em:
Nick Curtis
2024-03-14 11:41:35 -04:00
commit de GitHub
commit c5408fb22b
6 arquivos alterados com 24 adições e 24 exclusões
@@ -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: