Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

Commitf56e5d6

Browse files
yushangdifacebook-github-bot
authored andcommitted
Fix too big to optimize in test, actually use O0 when aot_inductor.compile_wrapper_with_O0 is set (#148714)
Summary:1. the '\0' char cannot be properly recognized. Change to use '0' instead.2. We got the following error when using anything other than O0 flag: `error: Function ZN5torch12aot_inductorL22__check_inputs_outputsEPP16AtenTensorOpaqueS3 is too big to optimize [-Werror,-Wignored-optimization-argument]` So we use O0 flag in wrapper code when `aot_inductor.compile_wrapper_with_O0` is setTest Plan:``` buck run 'fbcode//mode/opt' fbcode//deeplearning/aot_inductor/cpu/test:ads_second_stage_dsnn_models_aoti_lowering_test -- -r AdsSecondStageDSNNModelsAOTILoweringTest```Differential Revision: D70670957
1 parent179b7a0 commitf56e5d6

File tree

5 files changed

+8
-45
lines changed

5 files changed

+8
-45
lines changed

‎test/inductor/test_aot_inductor.py‎

Lines changed: 0 additions & 18 deletions
Original file line numberDiff line numberDiff line change
@@ -139,24 +139,6 @@ def forward(self, x, y):
139139
model,example_inputs,"AOTInductorModelRunMinimalArrayrefInterface(",1
140140
)
141141

142-
deftest_compile_wrapper_with_O0(self):
143-
classModel(torch.nn.Module):
144-
def__init__(self)->None:
145-
super().__init__()
146-
self.linear=torch.nn.Linear(10,10)
147-
148-
defforward(self,x,y):
149-
returnx+self.linear(y)
150-
151-
example_inputs= (
152-
torch.randn(10,10,device=self.device),
153-
torch.randn(10,10,device=self.device),
154-
)
155-
model=Model()
156-
withconfig.patch("aot_inductor.compile_wrapper_with_O0",True):
157-
self.check_model(model,example_inputs)
158-
self.code_check_count(model,example_inputs,"__attribute__((",2)
159-
160142
deftest_small_constant(self):
161143
classModel(torch.nn.Module):
162144
def__init__(self)->None:

‎torch/_inductor/codegen/cpp_wrapper_cpu.py‎

Lines changed: 2 additions & 12 deletions
Original file line numberDiff line numberDiff line change
@@ -412,7 +412,7 @@ def gen_check(handle_kind, idx, name, tensor):
412412
"""
413413
bool _check_aoti_runtime_check_inputs_env() {
414414
const static char* env_var_value = getenv("AOTI_RUNTIME_CHECK_INPUTS");
415-
const static bool result = env_var_value != nullptr && env_var_value[0] != '\0';
415+
const static bool result = env_var_value != nullptr && env_var_value[0] != '0';
416416
return result;
417417
}
418418
@@ -466,17 +466,7 @@ def write_wrapper_decl(self):
466466
"""
467467
)
468468

469-
run_impl_proto=""
470-
ifconfig.aot_inductor.compile_wrapper_with_O0:
471-
run_impl_proto+="""
472-
#ifdef __clang__
473-
__attribute__((optnone))
474-
#else
475-
__attribute__((optimize("O0")))
476-
#endif
477-
"""
478-
479-
run_impl_proto+="""
469+
run_impl_proto="""
480470
void AOTInductorModel::run_impl(
481471
AtenTensorHandle*
482472
input_handles, // array of input AtenTensorHandle; handles

‎torch/_inductor/codegen/cpp_wrapper_cpu_array_ref.py‎

Lines changed: 1 addition & 11 deletions
Original file line numberDiff line numberDiff line change
@@ -195,17 +195,7 @@ def write_wrapper_decl(self):
195195
"""
196196
)
197197

198-
run_impl_proto=""
199-
ifconfig.aot_inductor.compile_wrapper_with_O0:
200-
run_impl_proto+="""
201-
#ifdef __clang__
202-
__attribute__((optnone))
203-
#else
204-
__attribute__((optimize("O0")))
205-
#endif
206-
"""
207-
208-
run_impl_proto+="""
198+
run_impl_proto="""
209199
void AOTInductorModel::run_impl(
210200
AtenTensorHandle*
211201
input_handles, // array of input AtenTensorHandle; handles

‎torch/_inductor/config.py‎

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -1169,9 +1169,9 @@ class aot_inductor:
11691169
debug_compile=os.environ.get("AOT_INDUCTOR_DEBUG_COMPILE","0")=="1"
11701170

11711171
# Annotate generated main wrapper function, i.e. AOTInductorModel::run_impl,
1172-
# toskipcpp compileroptimizations for faster compilation.
1173-
compile_wrapper_with_O0= (
1174-
os.environ.get("AOT_INDUCTOR_COMPILE_WRAPPER_WITH_O0","0")=="1"
1172+
# touse whichcpp compileroptimization level, default to O1
1173+
compile_wrapper_opt_level=os.environ.get(
1174+
"AOT_INDUCTOR_COMPILE_WRAPPER_OPT_LEVEL","O1"
11751175
)
11761176

11771177
# option for debug printing/saving for intermediate tensor values for aot inductor

‎torch/_inductor/cpp_builder.py‎

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -542,10 +542,11 @@ def _get_optimization_cflags(
542542
if_IS_WINDOWS:
543543
return ["O1"ifmin_optimizeelse"O2"]
544544
else:
545+
wrapper_opt_level=config.aot_inductor.compile_wrapper_opt_level
545546
cflags= (
546547
["O0","g"]
547548
ifconfig.aot_inductor.debug_compile
548-
else ["O1"ifmin_optimizeelse"O3","DNDEBUG"]
549+
else [wrapper_opt_levelifmin_optimizeelse"O3","DNDEBUG"]
549550
)
550551
cflags+=_get_ffast_math_flags()
551552
cflags.append("fno-finite-math-only")

0 commit comments

Comments
 (0)
[8]ページ先頭
©2009-2025 Movatter.jp