işlemeyi yapan:
Cole Ramos
ebeveyn
04e8395a2a
işleme
1dc4553091
@@ -360,9 +360,7 @@ install(
|
||||
PATTERN "rooflines*" EXCLUDE)
|
||||
# src/utils/rooflines
|
||||
file(GLOB rooflinebins src/utils/rooflines/roofline-*)
|
||||
install(
|
||||
PROGRAMS ${rooflinebins}
|
||||
DESTINATION ${CMAKE_INSTALL_BINDIR}/utils/rooflines)
|
||||
install(PROGRAMS ${rooflinebins} DESTINATION ${CMAKE_INSTALL_BINDIR}/utils/rooflines)
|
||||
# src/omniperf_soc
|
||||
install(
|
||||
DIRECTORY src/omniperf_soc
|
||||
|
||||
@@ -400,7 +400,7 @@ class OmniProfiler_Base:
|
||||
skip_roof=self.__args.no_roof,
|
||||
roof_only=self.__args.roof_only,
|
||||
mspec=self._soc._mspec,
|
||||
soc=self._soc
|
||||
soc=self._soc,
|
||||
)
|
||||
|
||||
|
||||
|
||||
@@ -67,7 +67,7 @@ class gfx940_soc(OmniSoC_Base):
|
||||
"TCC_channels": 32,
|
||||
}
|
||||
)
|
||||
#self.roofline_obj = Roofline(args, self._mspec)
|
||||
# self.roofline_obj = Roofline(args, self._mspec)
|
||||
|
||||
# Set arch specific specs
|
||||
self._mspec._l2_banks = 16
|
||||
|
||||
@@ -67,7 +67,7 @@ class gfx941_soc(OmniSoC_Base):
|
||||
"TCC_channels": 32,
|
||||
}
|
||||
)
|
||||
#self.roofline_obj = Roofline(args, self._mspec)
|
||||
# self.roofline_obj = Roofline(args, self._mspec)
|
||||
|
||||
# Set arch specific specs
|
||||
self._mspec._l2_banks = 16
|
||||
|
||||
@@ -67,7 +67,7 @@ class gfx942_soc(OmniSoC_Base):
|
||||
"TCC_channels": 32,
|
||||
}
|
||||
)
|
||||
#self.roofline_obj = Roofline(args, self._mspec)
|
||||
# self.roofline_obj = Roofline(args, self._mspec)
|
||||
|
||||
# Set arch specific specs
|
||||
self._mspec._l2_banks = 16
|
||||
|
||||
@@ -376,8 +376,10 @@ class Roofline:
|
||||
sysinfo_path = os.path.join(self.__args.path, "sysinfo.csv")
|
||||
if not os.path.isfile(sysinfo_path):
|
||||
logging.info("[roofline] sysinfo.csv not found. Generating...")
|
||||
|
||||
class Dummy_SoC:
|
||||
roofline_obj = True
|
||||
|
||||
gen_sysinfo(
|
||||
workload_name=self.__args.name,
|
||||
workload_dir=self.__workload_dir,
|
||||
@@ -386,7 +388,7 @@ class Roofline:
|
||||
skip_roof=self.__args.no_roof,
|
||||
roof_only=self.__args.roof_only,
|
||||
mspec=self.__mspec,
|
||||
soc=Dummy_SoC
|
||||
soc=Dummy_SoC,
|
||||
)
|
||||
|
||||
@abstractmethod
|
||||
|
||||
@@ -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)
|
||||
|
||||
|
||||
Yeni konuda referans
Bir kullanıcı engelle