AI/rocm-systems
2
0
Vork 0
Code Kwesties Pull-aanvragen Actions Paketten Projecten Publicaties Wiki Activiteit

CMake and Python format

Bladeren bron 
Signed-off-by: coleramos425 <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 1dc4553091]
This commit is contained in:
coleramos425
2024-03-01 14:28:11 -06:00
gecommit door Cole Ramos
bovenliggende 985e97845f
commit 5868974f74
7 gewijzigde bestanden met toevoegingen van 20 en 54 verwijderingen
Download Patch-bestand Download Diff-bestand Expand all files Collapse all files
projects/rocprofiler-compute/src/utils/roofline_calc.py
+12 -46
Bestand weergeven
@@ -195,21 +195,11 @@ 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 = ""
@@ -390,23 +380,11 @@ 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(
@@ -428,23 +406,11 @@ 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)