+1
-1
@@ -32,4 +32,4 @@ HIDDEN_COLUMNS = ["coll_level"]
|
||||
HIDDEN_COLUMNS_CLI = ["Description", "coll_level"]
|
||||
HIDDEN_SECTIONS = [400, 1900, 2000]
|
||||
|
||||
TIME_UNITS = {"s": 10**9, "ms": 10**6, "us": 10**3, "ns": 1}
|
||||
TIME_UNITS = {"s": 10 ** 9, "ms": 10 ** 6, "us": 10 ** 3, "ns": 1}
|
||||
|
||||
@@ -405,9 +405,10 @@ class OmniSoC_Base:
|
||||
subvariable_matches = set()
|
||||
for var in variable_matches:
|
||||
if var in build_in_vars:
|
||||
hw_counter_matches_vars, variable_matches_vars = (
|
||||
self.parse_counters_text(build_in_vars[var])
|
||||
)
|
||||
(
|
||||
hw_counter_matches_vars,
|
||||
variable_matches_vars,
|
||||
) = self.parse_counters_text(build_in_vars[var])
|
||||
hw_counter_matches.update(hw_counter_matches_vars)
|
||||
subvariable_matches.update(variable_matches_vars)
|
||||
# process new found variables
|
||||
|
||||
+56
-14
@@ -259,11 +259,25 @@ def calc_ai(mspec, sort_type, ret_df):
|
||||
df = df.sort_values(by=["Kernel_Name"])
|
||||
df = df.reset_index(drop=True)
|
||||
|
||||
total_flops = valu_flops = mfma_flops_f6f4 = mfma_flops_f8 = mfma_flops_bf16 = (
|
||||
total_flops = (
|
||||
valu_flops
|
||||
) = (
|
||||
mfma_flops_f6f4
|
||||
) = (
|
||||
mfma_flops_f8
|
||||
) = (
|
||||
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_iops_i8
|
||||
) = (
|
||||
mfma_flops_f32
|
||||
) = (
|
||||
mfma_flops_f64
|
||||
) = (
|
||||
lds_data
|
||||
) = L1cache_data = L2cache_data = hbm_data = calls = totalDuration = avgDuration = 0.0
|
||||
|
||||
kernelName = ""
|
||||
|
||||
@@ -484,13 +498,27 @@ def calc_ai(mspec, sort_type, ret_df):
|
||||
kernelName, idx, calls
|
||||
)
|
||||
)
|
||||
total_flops = valu_flops = mfma_flops_f6f4 = mfma_flops_f8 = (
|
||||
total_flops = (
|
||||
valu_flops
|
||||
) = (
|
||||
mfma_flops_f6f4
|
||||
) = (
|
||||
mfma_flops_f8
|
||||
) = (
|
||||
mfma_flops_bf16
|
||||
) = mfma_flops_f16 = mfma_iops_i8 = mfma_flops_f32 = mfma_flops_f64 = (
|
||||
) = (
|
||||
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
|
||||
) = (
|
||||
L1cache_data
|
||||
) = L2cache_data = hbm_data = calls = totalDuration = avgDuration = 0.0
|
||||
|
||||
if sort_type == "dispatches":
|
||||
myList.append(
|
||||
@@ -514,13 +542,27 @@ def calc_ai(mspec, sort_type, ret_df):
|
||||
avgDuration,
|
||||
)
|
||||
)
|
||||
total_flops = valu_flops = mfma_flops_f6f4 = mfma_flops_f8 = (
|
||||
total_flops = (
|
||||
valu_flops
|
||||
) = (
|
||||
mfma_flops_f6f4
|
||||
) = (
|
||||
mfma_flops_f8
|
||||
) = (
|
||||
mfma_flops_bf16
|
||||
) = mfma_flops_f16 = mfma_iops_i8 = mfma_flops_f32 = mfma_flops_f64 = (
|
||||
) = (
|
||||
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
|
||||
) = (
|
||||
L1cache_data
|
||||
) = L2cache_data = hbm_data = calls = totalDuration = avgDuration = 0.0
|
||||
|
||||
myList.sort(key=lambda x: x.totalDuration, reverse=True)
|
||||
|
||||
|
||||
Reference in New Issue
Block a user