Skip to content

Commit 2cc3e04

Browse files
co63occo63oc
authored
paddle/fluid/pybind/inference_api.cc add onednn [fluid_ops] (#74436)
1 parent 4614b93 commit 2cc3e04

10 files changed

+71
-31
lines changed

‎paddle/fluid/pybind/inference_api.cc

Lines changed: 53 additions & 13 deletions
Original file line numberDiff line numberDiff line change
@@ -1050,26 +1050,35 @@ void BindAnalysisConfig(py::module *m) {
10501050
&AnalysisConfig::SwitchIrDebug,
10511051
py::arg("x") = true,
10521052
py::arg("passes") = std::vector<std::string>())
1053-
.def("enable_mkldnn", &AnalysisConfig::EnableONEDNN)
1054-
.def("disable_mkldnn", &AnalysisConfig::DisableONEDNN)
1055-
.def("mkldnn_enabled", &AnalysisConfig::onednn_enabled)
1053+
.def("enable_mkldnn", &AnalysisConfig::EnableONEDNN) // deprecated
1054+
.def("disable_mkldnn", &AnalysisConfig::DisableONEDNN) // deprecated
1055+
.def("mkldnn_enabled", &AnalysisConfig::onednn_enabled) // deprecated
1056+
.def("enable_onednn", &AnalysisConfig::EnableONEDNN)
1057+
.def("disable_onednn", &AnalysisConfig::DisableONEDNN)
1058+
.def("onednn_enabled", &AnalysisConfig::onednn_enabled)
10561059
.def("enable_cinn", &AnalysisConfig::EnableCINN)
10571060
.def("set_cpu_math_library_num_threads",
10581061
&AnalysisConfig::SetCpuMathLibraryNumThreads)
10591062
.def("cpu_math_library_num_threads",
10601063
&AnalysisConfig::cpu_math_library_num_threads)
10611064
.def("to_native_config", &AnalysisConfig::ToNativeConfig)
1062-
.def("enable_mkldnn_bfloat16", &AnalysisConfig::EnableOnednnBfloat16)
1065+
.def("enable_mkldnn_bfloat16",
1066+
&AnalysisConfig::EnableOnednnBfloat16) // deprecated
1067+
.def("enable_onednn_bfloat16", &AnalysisConfig::EnableOnednnBfloat16)
10631068
#ifdef PADDLE_WITH_DNNL
10641069
.def("set_mkldnn_cache_capacity",
1070+
&AnalysisConfig::SetOnednnCacheCapacity,
1071+
py::arg("capacity") = 0) // deprecated
1072+
.def("set_onednn_cache_capacity",
10651073
&AnalysisConfig::SetOnednnCacheCapacity,
10661074
py::arg("capacity") = 0)
10671075
.def("set_bfloat16_op", &AnalysisConfig::SetBfloat16Op)
10681076
.def("enable_mkldnn_int8",
10691077
&AnalysisConfig::EnableOnednnInt8,
10701078
py::arg("mkldnn_int8_enabled_op_types") =
1071-
std::unordered_set<std::string>({}))
1072-
.def("mkldnn_int8_enabled", &AnalysisConfig::onednn_int8_enabled)
1079+
std::unordered_set<std::string>({})) // deprecated
1080+
.def("mkldnn_int8_enabled",
1081+
&AnalysisConfig::onednn_int8_enabled) // deprecated
10731082
.def("disable_mkldnn_fc_passes",
10741083
&AnalysisConfig::DisableOnednnFcPasses,
10751084
R"DOC(
@@ -1085,9 +1094,31 @@ void BindAnalysisConfig(py::module *m) {
10851094
>>> config = Config("")
10861095
>>> config.enable_mkldnn()
10871096
>>> config.disable_mkldnn_fc_passes()
1097+
)DOC") // deprecated
1098+
.def("enable_onednn_int8",
1099+
&AnalysisConfig::EnableOnednnInt8,
1100+
py::arg("onednn_int8_enabled_op_types") =
1101+
std::unordered_set<std::string>({}))
1102+
.def("onednn_int8_enabled", &AnalysisConfig::onednn_int8_enabled)
1103+
.def("disable_onednn_fc_passes",
1104+
&AnalysisConfig::DisableOnednnFcPasses,
1105+
R"DOC(
1106+
Disable Onednn FC
1107+
Returns:
1108+
None.
1109+
1110+
Examples:
1111+
.. code-block:: python
1112+
1113+
>>> from paddle.inference import Config
1114+
1115+
>>> config = Config("")
1116+
>>> config.enable_onednn()
1117+
>>> config.disable_onednn_fc_passes()
10881118
)DOC")
10891119
#endif
1090-
.def("set_mkldnn_op", &AnalysisConfig::SetONEDNNOp)
1120+
.def("set_mkldnn_op", &AnalysisConfig::SetONEDNNOp) // deprecated
1121+
.def("set_onednn_op", &AnalysisConfig::SetONEDNNOp)
10911122
.def("set_model_buffer", &AnalysisConfig::SetModelBuffer)
10921123
.def("model_from_memory", &AnalysisConfig::model_from_memory)
10931124
.def("delete_pass", &AnalysisConfig::DeletePass)
@@ -1329,23 +1360,32 @@ void BindPaddlePassBuilder(py::module *m) {
13291360
py::class_<PassStrategy, PaddlePassBuilder>(*m, "PassStrategy")
13301361
.def(py::init<const std::vector<std::string> &>())
13311362
.def("enable_cudnn", &PassStrategy::EnableCUDNN)
1332-
.def("enable_mkldnn", &PassStrategy::EnableONEDNN)
1333-
.def("enable_mkldnn_bfloat16", &PassStrategy::EnableMkldnnBfloat16)
1363+
.def("enable_mkldnn", &PassStrategy::EnableONEDNN) // deprecated
1364+
.def("enable_mkldnn_bfloat16",
1365+
&PassStrategy::EnableMkldnnBfloat16) // deprecated
1366+
.def("enable_onednn", &PassStrategy::EnableONEDNN)
1367+
.def("enable_onednn_bfloat16", &PassStrategy::EnableOnednnBfloat16)
13341368
.def("use_gpu", &PassStrategy::use_gpu);
13351369

13361370
py::class_<CpuPassStrategy, PassStrategy>(*m, "CpuPassStrategy")
13371371
.def(py::init<>())
13381372
.def(py::init<const CpuPassStrategy &>())
13391373
.def("enable_cudnn", &CpuPassStrategy::EnableCUDNN)
1340-
.def("enable_mkldnn", &CpuPassStrategy::EnableONEDNN)
1341-
.def("enable_mkldnn_bfloat16", &CpuPassStrategy::EnableMkldnnBfloat16);
1374+
.def("enable_mkldnn", &CpuPassStrategy::EnableONEDNN) // deprecated
1375+
.def("enable_mkldnn_bfloat16",
1376+
&CpuPassStrategy::EnableMkldnnBfloat16) // deprecated
1377+
.def("enable_onednn", &CpuPassStrategy::EnableONEDNN)
1378+
.def("enable_onednn_bfloat16", &CpuPassStrategy::EnableOnednnBfloat16);
13421379

13431380
py::class_<GpuPassStrategy, PassStrategy>(*m, "GpuPassStrategy")
13441381
.def(py::init<>())
13451382
.def(py::init<const GpuPassStrategy &>())
13461383
.def("enable_cudnn", &GpuPassStrategy::EnableCUDNN)
1347-
.def("enable_mkldnn", &GpuPassStrategy::EnableONEDNN)
1348-
.def("enable_mkldnn_bfloat16", &GpuPassStrategy::EnableMkldnnBfloat16);
1384+
.def("enable_mkldnn", &GpuPassStrategy::EnableONEDNN) // deprecated
1385+
.def("enable_mkldnn_bfloat16",
1386+
&GpuPassStrategy::EnableMkldnnBfloat16) // deprecated
1387+
.def("enable_onednn", &GpuPassStrategy::EnableONEDNN)
1388+
.def("enable_onednn_bfloat16", &GpuPassStrategy::EnableOnednnBfloat16);
13491389
}
13501390

13511391
void BindInternalUtils(py::module *m) {

‎test/deprecated/ir/inference/inference_pass_test.py

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -178,9 +178,9 @@ def _get_analysis_config(
178178
config.enable_tensorrt_varseqlen()
179179

180180
elif use_mkldnn:
181-
config.enable_mkldnn()
181+
config.enable_onednn()
182182
if self.enable_mkldnn_bfloat16:
183-
config.enable_mkldnn_bfloat16()
183+
config.enable_onednn_bfloat16()
184184
return config
185185

186186
def check_output(self, atol=1e-3):

‎test/deprecated/ir/inference/quant_dequant_test.py

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -231,9 +231,9 @@ def _get_analysis_config(
231231
config.enable_tensorrt_varseqlen()
232232

233233
elif use_mkldnn:
234-
config.enable_mkldnn()
234+
config.enable_onednn()
235235
if self.enable_mkldnn_bfloat16:
236-
config.enable_mkldnn_bfloat16()
236+
config.enable_onednn_bfloat16()
237237
return config
238238

239239
def check_output_with_option(

‎test/deprecated/ir/inference/test_trt_inference_predictor_deprecated.py

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -98,14 +98,14 @@ def load(self, config_arg, inputs=None, outputs=None):
9898

9999
if self.args.enable_mkldnn and not self.args.enable_gpu:
100100
config.disable_gpu()
101-
config.enable_mkldnn()
101+
config.enable_onednn()
102102
if self.args.precision == 'int8':
103-
config.enable_mkldnn_int8(
103+
config.enable_onednn_int8(
104104
{"conv2d", "depthwise_conv2d", "transpose2", "pool2d"}
105105
)
106106
if not self.args.enable_mkldnn and not self.args.enable_gpu:
107107
config.disable_gpu()
108-
# config.enable_mkldnn()
108+
# config.enable_onednn()
109109
if self.args.enable_profile:
110110
config.enable_profile()
111111
shape_range_file = os.path.join(

‎test/ir/inference/inference_pass_test.py

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -179,9 +179,9 @@ def _get_analysis_config(
179179
config.enable_tensorrt_varseqlen()
180180

181181
elif use_mkldnn:
182-
config.enable_mkldnn()
182+
config.enable_onednn()
183183
if self.enable_mkldnn_bfloat16:
184-
config.enable_mkldnn_bfloat16()
184+
config.enable_onednn_bfloat16()
185185
print('config summary:', config.summary())
186186
return config
187187

‎test/ir/inference/quant_dequant_test.py

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -231,9 +231,9 @@ def _get_analysis_config(
231231
config.enable_tensorrt_varseqlen()
232232

233233
elif use_mkldnn:
234-
config.enable_mkldnn()
234+
config.enable_onednn()
235235
if self.enable_mkldnn_bfloat16:
236-
config.enable_mkldnn_bfloat16()
236+
config.enable_onednn_bfloat16()
237237
print('config summary:', config.summary())
238238
return config
239239

‎test/ir/inference/test_conv_eltwiseadd_bn_fuse_pass.py

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -47,7 +47,7 @@ def sample_predictor_configs(self, program_config):
4747

4848
# MKLDNN
4949
config = self.create_inference_config(use_gpu=False)
50-
config.enable_mkldnn()
50+
config.enable_onednn()
5151
yield config, ["conv2d", "elementwise_add"], (1e-4, 1e-5)
5252

5353
# for gpu

‎test/mkldnn/test_onnx_format_quantization_mobilenetv1.py

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -208,12 +208,12 @@ def run_program(
208208
image_shape = [3, 224, 224]
209209
config = paddle.inference.Config(model_path)
210210
config.disable_gpu()
211-
config.enable_mkldnn()
211+
config.enable_onednn()
212212
config.switch_ir_optim()
213213
config.set_cpu_math_library_num_threads(1)
214214
config.disable_glog_info()
215215
if is_quantized_model:
216-
config.enable_mkldnn_int8()
216+
config.enable_onednn_int8()
217217
predictor = paddle.inference.create_predictor(config)
218218

219219
input_names = predictor.get_input_names()

‎test/quantization/quant2_int8_lstm_model.py

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -112,7 +112,7 @@ def set_config(
112112
config.disable_gpu()
113113
config.switch_use_feed_fetch_ops(True)
114114
config.switch_ir_optim(True)
115-
config.enable_mkldnn()
115+
config.enable_onednn()
116116
config.disable_mkldnn_fc_passes() # fc passes caused dnnl error
117117
config.pass_builder().insert_pass(5, "fc_lstm_fuse_pass")
118118
config.set_mkldnn_cache_capacity(mkldnn_cache_capacity)
@@ -121,7 +121,7 @@ def set_config(
121121
config.quantizer_config().set_quant_data(warmup_data)
122122
config.quantizer_config().set_quant_batch_size(1)
123123
elif mode == "qat":
124-
config.enable_mkldnn_int8()
124+
config.enable_onednn_int8()
125125

126126
return config
127127

‎test/quantization/quant2_int8_nlp_comparison.py

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -158,9 +158,9 @@ def set_config(
158158
config.switch_specify_input_names(True)
159159
config.switch_ir_optim(True)
160160
config.switch_use_feed_fetch_ops(True)
161-
config.enable_mkldnn()
161+
config.enable_onednn()
162162
if target == 'int8':
163-
config.enable_mkldnn_int8(self._quantized_ops)
163+
config.enable_onednn_int8(self._quantized_ops)
164164
config.delete_pass(
165165
"constant_folding_pass"
166166
) # same reason as in analyzer_ernie_int8_tester.cc

0 commit comments

Comments
 (0)