Enable baseline comparison between workloads from the same path (#447)
* Enable baseline comparison between workloads from the same path
Signed-off-by: zichguan-amd <zichuan.guan@amd.com>
* Formatting
Signed-off-by: zichguan-amd <zichuan.guan@amd.com>
* Change test to verify baseline works with same path
Signed-off-by: zichguan-amd <zichuan.guan@amd.com>
* Add three-way comparison test case
Signed-off-by: zichguan-amd <zichuan.guan@amd.com>
---------
Signed-off-by: zichguan-amd <zichuan.guan@amd.com>
[ROCm/rocprofiler-compute commit: 13a06f5131]
Dieser Commit ist enthalten in:
zichguan-amd
GitHub
@@ -151,7 +151,7 @@ class OmniAnalyze_Base:
|
||||
|
||||
self.load_options(normalization_filter)
|
||||
|
||||
for d in self.__args.path:
|
||||
for i, d in enumerate(self.__args.path):
|
||||
w = schema.Workload()
|
||||
w.sys_info = file_io.load_sys_info(Path(d[0], "sysinfo.csv"))
|
||||
arch = w.sys_info.iloc[0]["gpu_arch"]
|
||||
@@ -161,7 +161,7 @@ class OmniAnalyze_Base:
|
||||
w.avail_ips = w.sys_info["ip_blocks"].item().split("|")
|
||||
w.dfs = copy.deepcopy(self._arch_configs[arch].dfs)
|
||||
w.dfs_type = self._arch_configs[arch].dfs_type
|
||||
self._runs[d[0]] = w
|
||||
self._runs[i] = w
|
||||
|
||||
return self._runs
|
||||
|
||||
@@ -184,15 +184,6 @@ class OmniAnalyze_Base:
|
||||
# validate profiling data
|
||||
is_workload_empty(dir[0])
|
||||
|
||||
# no using same paths
|
||||
occurances = set()
|
||||
for dir in self.__args.path:
|
||||
dir = dir[0]
|
||||
if dir in occurances:
|
||||
console_error("You cannot provide the same path twice.")
|
||||
else:
|
||||
occurances.add(dir)
|
||||
|
||||
# ----------------------------------------------------
|
||||
# Required methods to be implemented by child classes
|
||||
# ----------------------------------------------------
|
||||
@@ -211,20 +202,20 @@ class OmniAnalyze_Base:
|
||||
|
||||
# set filters
|
||||
if self.__args.gpu_kernel:
|
||||
for d, gk in zip(self.__args.path, self.__args.gpu_kernel):
|
||||
self._runs[d[0]].filter_kernel_ids = gk
|
||||
for i, gk in enumerate(self.__args.gpu_kernel):
|
||||
self._runs[i].filter_kernel_ids = gk
|
||||
if self.__args.gpu_id:
|
||||
if len(self.__args.gpu_id) == 1 and len(self.__args.path) != 1:
|
||||
for i in range(len(self.__args.path) - 1):
|
||||
self.__args.gpu_id.extend(self.__args.gpu_id)
|
||||
for d, gi in zip(self.__args.path, self.__args.gpu_id):
|
||||
self._runs[d[0]].filter_gpu_ids = gi
|
||||
for i, gi in enumerate(self.__args.gpu_id):
|
||||
self._runs[i].filter_gpu_ids = gi
|
||||
if self.__args.gpu_dispatch_id:
|
||||
if len(self.__args.gpu_dispatch_id) == 1 and len(self.__args.path) != 1:
|
||||
for i in range(len(self.__args.path) - 1):
|
||||
self.__args.gpu_dispatch_id.extend(self.__args.gpu_dispatch_id)
|
||||
for d, gd in zip(self.__args.path, self.__args.gpu_dispatch_id):
|
||||
self._runs[d[0]].filter_dispatch_ids = gd
|
||||
for i, gd in enumerate(self.__args.gpu_dispatch_id):
|
||||
self._runs[i].filter_dispatch_ids = gd
|
||||
|
||||
@abstractmethod
|
||||
def run_analysis(self):
|
||||
|
||||
@@ -38,27 +38,25 @@ class cli_analysis(OmniAnalyze_Base):
|
||||
super().pre_processing()
|
||||
if self.get_args().random_port:
|
||||
console_error("--gui flag is required to enable --random-port")
|
||||
for d in self.get_args().path:
|
||||
for i, d in enumerate(self.get_args().path):
|
||||
file_io.create_df_kernel_top_stats(
|
||||
raw_data_dir=d[0],
|
||||
filter_gpu_ids=self._runs[d[0]].filter_gpu_ids,
|
||||
filter_dispatch_ids=self._runs[d[0]].filter_dispatch_ids,
|
||||
filter_gpu_ids=self._runs[i].filter_gpu_ids,
|
||||
filter_dispatch_ids=self._runs[i].filter_dispatch_ids,
|
||||
time_unit=self.get_args().time_unit,
|
||||
max_stat_num=self.get_args().max_stat_num,
|
||||
kernel_verbose=self.get_args().kernel_verbose,
|
||||
)
|
||||
# create 'mega dataframe'
|
||||
self._runs[d[0]].raw_pmc = file_io.create_df_pmc(
|
||||
self._runs[i].raw_pmc = file_io.create_df_pmc(
|
||||
d[0], self.get_args().kernel_verbose, self.get_args().verbose
|
||||
)
|
||||
# demangle and overwrite original 'Kernel_Name'
|
||||
kernel_name_shortener(
|
||||
self._runs[d[0]].raw_pmc, self.get_args().kernel_verbose
|
||||
)
|
||||
kernel_name_shortener(self._runs[i].raw_pmc, self.get_args().kernel_verbose)
|
||||
|
||||
# create the loaded table
|
||||
parser.load_table_data(
|
||||
workload=self._runs[d[0]],
|
||||
workload=self._runs[i],
|
||||
dir=d[0],
|
||||
is_gui=False,
|
||||
debug=self.get_args().debug,
|
||||
@@ -73,17 +71,13 @@ class cli_analysis(OmniAnalyze_Base):
|
||||
tty.show_kernel_stats(
|
||||
self.get_args(),
|
||||
self._runs,
|
||||
self._arch_configs[
|
||||
self._runs[self.get_args().path[0][0]].sys_info.iloc[0]["gpu_arch"]
|
||||
],
|
||||
self._arch_configs[self._runs[0].sys_info.iloc[0]["gpu_arch"]],
|
||||
self._output,
|
||||
)
|
||||
else:
|
||||
tty.show_all(
|
||||
self.get_args(),
|
||||
self._runs,
|
||||
self._arch_configs[
|
||||
self._runs[self.get_args().path[0][0]].sys_info.iloc[0]["gpu_arch"]
|
||||
],
|
||||
self._arch_configs[self._runs[0].sys_info.iloc[0]["gpu_arch"]],
|
||||
self._output,
|
||||
)
|
||||
|
||||
@@ -274,20 +274,20 @@ class webui_analysis(OmniAnalyze_Base):
|
||||
args = self.get_args()
|
||||
file_io.create_df_kernel_top_stats(
|
||||
raw_data_dir=self.dest_dir,
|
||||
filter_gpu_ids=self._runs[self.dest_dir].filter_gpu_ids,
|
||||
filter_dispatch_ids=self._runs[self.dest_dir].filter_dispatch_ids,
|
||||
filter_gpu_ids=self._runs[0].filter_gpu_ids,
|
||||
filter_dispatch_ids=self._runs[0].filter_dispatch_ids,
|
||||
time_unit=args.time_unit,
|
||||
max_stat_num=args.max_stat_num,
|
||||
kernel_verbose=self.get_args().kernel_verbose,
|
||||
)
|
||||
# create 'mega dataframe'
|
||||
self._runs[self.dest_dir].raw_pmc = file_io.create_df_pmc(
|
||||
self._runs[0].raw_pmc = file_io.create_df_pmc(
|
||||
self.dest_dir, self.get_args().kernel_verbose, args.verbose
|
||||
)
|
||||
# create the loaded kernel stats
|
||||
parser.load_kernel_top(self._runs[self.dest_dir], self.dest_dir)
|
||||
parser.load_kernel_top(self._runs[0], self.dest_dir)
|
||||
# set architecture
|
||||
self.arch = self._runs[self.dest_dir].sys_info.iloc[0]["gpu_arch"]
|
||||
self.arch = self._runs[0].sys_info.iloc[0]["gpu_arch"]
|
||||
|
||||
else:
|
||||
console_error(
|
||||
@@ -300,9 +300,9 @@ class webui_analysis(OmniAnalyze_Base):
|
||||
super().run_analysis()
|
||||
args = self.get_args()
|
||||
input_filters = {
|
||||
"kernel": self._runs[self.dest_dir].filter_kernel_ids,
|
||||
"gpu": self._runs[self.dest_dir].filter_gpu_ids,
|
||||
"dispatch": self._runs[self.dest_dir].filter_dispatch_ids,
|
||||
"kernel": self._runs[0].filter_kernel_ids,
|
||||
"gpu": self._runs[0].filter_gpu_ids,
|
||||
"dispatch": self._runs[0].filter_dispatch_ids,
|
||||
"normalization": args.normal_unit,
|
||||
"top_n": args.max_stat_num,
|
||||
}
|
||||
|
||||
In neuem Issue referenzieren
Einen Benutzer sperren