AI/rocm-systems
2
0
Forkuj 0
Kod Zgłoszenia Oczekujące zmiany Actions Packages Projekty Wydania Wiki Aktywność

More formatting!

Przeglądaj źródła 
Signed-off-by: coleramos425 <colramos@amd.com>


[ROCm/rocprofiler-compute commit: eaf701abf0]
This commit is contained in:
coleramos425
2024-03-01 13:53:12 -06:00
zatwierdzone przez Cole Ramos
rodzic 9ff07b51ff
commit a984ddc9b9
1 zmienionych plików z 46 dodań i 12 usunięć
Ściągnij plik aktualizacji Ściągnij plik porównania Expand all files Collapse all files
projects/rocprofiler-compute/src/utils/roofline_calc.py
+46 -12
Wyświetl plik
@@ -195,11 +195,21 @@ def calc_ai(sort_type, ret_df):
df = df.sort_values(by=["Kernel_Name"])
df = df.reset_index(drop=True)
total_flops = valu_flops = mfma_flops_bf16 = mfma_flops_f16 = mfma_iops_i8 = (
total_flops = (
valu_flops
) = (
mfma_flops_bf16
) = (
mfma_flops_f16
) = (
mfma_iops_i8
) = (
mfma_flops_f32
) = mfma_flops_f64 = lds_data = L1cache_data = L2cache_data = hbm_data = calls = (
totalDuration
) = avgDuration = 0.0
) = (
mfma_flops_f64
) = (
lds_data
) = L1cache_data = L2cache_data = hbm_data = calls = totalDuration = avgDuration = 0.0
kernelName = ""
@@ -380,11 +390,23 @@ def calc_ai(sort_type, ret_df):
kernelName, idx, calls
)
)
total_flops = valu_flops = mfma_flops_bf16 = mfma_flops_f16 = mfma_iops_i8 = (
total_flops = (
valu_flops
) = (
mfma_flops_bf16
) = (
mfma_flops_f16
) = (
mfma_iops_i8
) = (
mfma_flops_f32
) = mfma_flops_f64 = lds_data = L1cache_data = L2cache_data = hbm_data = (
calls
) = totalDuration = avgDuration = 0.0
) = (
mfma_flops_f64
) = (
lds_data
) = (
L1cache_data
) = L2cache_data = hbm_data = calls = totalDuration = avgDuration = 0.0
if sort_type == "dispatches":
myList.append(
@@ -406,11 +428,23 @@ def calc_ai(sort_type, ret_df):
avgDuration,
)
)
total_flops = valu_flops = mfma_flops_bf16 = mfma_flops_f16 = mfma_iops_i8 = (
total_flops = (
valu_flops
) = (
mfma_flops_bf16
) = (
mfma_flops_f16
) = (
mfma_iops_i8
) = (
mfma_flops_f32
) = mfma_flops_f64 = lds_data = L1cache_data = L2cache_data = hbm_data = (
calls
) = totalDuration = avgDuration = 0.0
) = (
mfma_flops_f64
) = (
lds_data
) = (
L1cache_data
) = L2cache_data = hbm_data = calls = totalDuration = avgDuration = 0.0
myList.sort(key=lambda x: x.totalDuration, reverse=True)