diff --git a/paddle/fluid/framework/ir/CMakeLists.txt b/paddle/fluid/framework/ir/CMakeLists.txt index a79a53867d85e91250ac4810caa5806c25f35fee..8c7d4cf0e3ef618b7321345064836d8043a83811 100644 --- a/paddle/fluid/framework/ir/CMakeLists.txt +++ b/paddle/fluid/framework/ir/CMakeLists.txt @@ -46,9 +46,6 @@ cc_library(fuse_pass_base SRCS fuse_pass_base.cc DEPS pass) pass_library(graph_to_program_pass base) pass_library(graph_viz_pass base) pass_library(lock_free_optimize_pass base) -pass_library(cpu_quantize_placement_pass base) -pass_library(cpu_quantize_pass inference) -pass_library(cpu_quantize_squash_pass inference) pass_library(fc_fuse_pass inference) pass_library(attention_lstm_fuse_pass inference) pass_library(infer_clean_graph_pass inference) @@ -87,6 +84,9 @@ if(WITH_MKLDNN) pass_library(conv_bias_mkldnn_fuse_pass inference mkldnn) pass_library(conv_relu_mkldnn_fuse_pass inference mkldnn) pass_library(conv_elementwise_add_mkldnn_fuse_pass inference mkldnn) + pass_library(cpu_quantize_placement_pass base mkldnn) + pass_library(cpu_quantize_pass inference mkldnn) + pass_library(cpu_quantize_squash_pass inference mkldnn) endif() cc_library(fuse_elewise_add_act_pass SRCS fuse_elewise_add_act_pass.cc DEPS pass graph_pattern_detector ) @@ -105,9 +105,6 @@ cc_test(test_graph_pattern_detector SRCS graph_pattern_detector_tester.cc DEPS g cc_test(test_fc_fuse_pass SRCS fc_fuse_pass_tester.cc DEPS fc_fuse_pass framework_proto) cc_test(test_seqpool_concat_fuse_pass SRCS seqpool_concat_fuse_pass_tester.cc DEPS seqpool_concat_fuse_pass framework_proto) cc_test(test_is_test_pass SRCS is_test_pass_tester.cc DEPS is_test_pass) -cc_test(test_cpu_quantize_placement_pass SRCS cpu_quantize_placement_pass_tester.cc DEPS cpu_quantize_placement_pass) -cc_test(test_cpu_quantize_pass SRCS cpu_quantize_pass_tester.cc DEPS cpu_quantize_pass naive_executor) -cc_test(test_cpu_quantize_squash_pass SRCS cpu_quantize_squash_pass_tester.cc DEPS cpu_quantize_squash_pass naive_executor) if(NOT WIN32) cc_test(test_sync_batch_norm_pass SRCS sync_batch_norm_pass_tester.cc DEPS sync_batch_norm_pass) endif() @@ -117,4 +114,7 @@ if (WITH_MKLDNN) cc_test(test_conv_relu_mkldnn_fuse_pass SRCS mkldnn/conv_relu_mkldnn_fuse_pass_tester.cc DEPS conv_relu_mkldnn_fuse_pass) cc_test(test_conv_elementwise_add_mkldnn_fuse_pass SRCS mkldnn/conv_elementwise_add_mkldnn_fuse_pass_tester.cc DEPS conv_elementwise_add_mkldnn_fuse_pass) cc_test(test_mkldnn_placement_pass SRCS mkldnn/mkldnn_placement_pass_tester.cc DEPS mkldnn_placement_pass) + cc_test(test_cpu_quantize_placement_pass SRCS mkldnn/cpu_quantize_placement_pass_tester.cc DEPS cpu_quantize_placement_pass) + cc_test(test_cpu_quantize_pass SRCS mkldnn/cpu_quantize_pass_tester.cc DEPS cpu_quantize_pass naive_executor) + cc_test(test_cpu_quantize_squash_pass SRCS mkldnn/cpu_quantize_squash_pass_tester.cc DEPS cpu_quantize_squash_pass naive_executor) endif () diff --git a/paddle/fluid/framework/ir/cpu_quantize_pass.cc b/paddle/fluid/framework/ir/mkldnn/cpu_quantize_pass.cc similarity index 99% rename from paddle/fluid/framework/ir/cpu_quantize_pass.cc rename to paddle/fluid/framework/ir/mkldnn/cpu_quantize_pass.cc index ed80f9cae347cfb2bf23859daea2f1f47dba599b..b3a8c208916f699dc032496c6d0fa5bf86227903 100644 --- a/paddle/fluid/framework/ir/cpu_quantize_pass.cc +++ b/paddle/fluid/framework/ir/mkldnn/cpu_quantize_pass.cc @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -#include "paddle/fluid/framework/ir/cpu_quantize_pass.h" +#include "paddle/fluid/framework/ir/mkldnn/cpu_quantize_pass.h" #include #include #include "paddle/fluid/framework/eigen.h" diff --git a/paddle/fluid/framework/ir/cpu_quantize_pass.h b/paddle/fluid/framework/ir/mkldnn/cpu_quantize_pass.h similarity index 100% rename from paddle/fluid/framework/ir/cpu_quantize_pass.h rename to paddle/fluid/framework/ir/mkldnn/cpu_quantize_pass.h diff --git a/paddle/fluid/framework/ir/cpu_quantize_pass_tester.cc b/paddle/fluid/framework/ir/mkldnn/cpu_quantize_pass_tester.cc similarity index 99% rename from paddle/fluid/framework/ir/cpu_quantize_pass_tester.cc rename to paddle/fluid/framework/ir/mkldnn/cpu_quantize_pass_tester.cc index 89601be7d1c0f5c9d3c3dcefa4327be7c20a7d65..0d0ed989012fced7f639c2bc12a3bafa6edf27f6 100644 --- a/paddle/fluid/framework/ir/cpu_quantize_pass_tester.cc +++ b/paddle/fluid/framework/ir/mkldnn/cpu_quantize_pass_tester.cc @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -#include "paddle/fluid/framework/ir/cpu_quantize_pass.h" +#include "paddle/fluid/framework/ir/mkldnn/cpu_quantize_pass.h" #include #include "paddle/fluid/framework/naive_executor.h" #include "paddle/fluid/platform/place.h" diff --git a/paddle/fluid/framework/ir/cpu_quantize_placement_pass.cc b/paddle/fluid/framework/ir/mkldnn/cpu_quantize_placement_pass.cc similarity index 96% rename from paddle/fluid/framework/ir/cpu_quantize_placement_pass.cc rename to paddle/fluid/framework/ir/mkldnn/cpu_quantize_placement_pass.cc index 50bbe4915b3502a867be397ae0922d982108d12c..511003dce59f91272802766544577e9c473a3a1d 100644 --- a/paddle/fluid/framework/ir/cpu_quantize_placement_pass.cc +++ b/paddle/fluid/framework/ir/mkldnn/cpu_quantize_placement_pass.cc @@ -12,7 +12,7 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. */ -#include "paddle/fluid/framework/ir/cpu_quantize_placement_pass.h" +#include "paddle/fluid/framework/ir/mkldnn/cpu_quantize_placement_pass.h" #include #include diff --git a/paddle/fluid/framework/ir/cpu_quantize_placement_pass.h b/paddle/fluid/framework/ir/mkldnn/cpu_quantize_placement_pass.h similarity index 100% rename from paddle/fluid/framework/ir/cpu_quantize_placement_pass.h rename to paddle/fluid/framework/ir/mkldnn/cpu_quantize_placement_pass.h diff --git a/paddle/fluid/framework/ir/cpu_quantize_placement_pass_tester.cc b/paddle/fluid/framework/ir/mkldnn/cpu_quantize_placement_pass_tester.cc similarity index 98% rename from paddle/fluid/framework/ir/cpu_quantize_placement_pass_tester.cc rename to paddle/fluid/framework/ir/mkldnn/cpu_quantize_placement_pass_tester.cc index 5a4d622645a4377526351bbf4acbcea95a780d22..11d72a56bd66792ff3ed5cc8184f5b242d9cdba5 100644 --- a/paddle/fluid/framework/ir/cpu_quantize_placement_pass_tester.cc +++ b/paddle/fluid/framework/ir/mkldnn/cpu_quantize_placement_pass_tester.cc @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -#include "paddle/fluid/framework/ir/cpu_quantize_placement_pass.h" +#include "paddle/fluid/framework/ir/mkldnn/cpu_quantize_placement_pass.h" #include #include diff --git a/paddle/fluid/framework/ir/cpu_quantize_squash_pass.cc b/paddle/fluid/framework/ir/mkldnn/cpu_quantize_squash_pass.cc similarity index 98% rename from paddle/fluid/framework/ir/cpu_quantize_squash_pass.cc rename to paddle/fluid/framework/ir/mkldnn/cpu_quantize_squash_pass.cc index de62a69de4f25912c5f56973de0dca5343bbe906..6e74cc7787b73d06b1093ed4e846ab83b1234803 100644 --- a/paddle/fluid/framework/ir/cpu_quantize_squash_pass.cc +++ b/paddle/fluid/framework/ir/mkldnn/cpu_quantize_squash_pass.cc @@ -13,7 +13,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -#include "paddle/fluid/framework/ir/cpu_quantize_squash_pass.h" +#include "paddle/fluid/framework/ir/mkldnn/cpu_quantize_squash_pass.h" #include #include #include "paddle/fluid/platform/enforce.h" diff --git a/paddle/fluid/framework/ir/cpu_quantize_squash_pass.h b/paddle/fluid/framework/ir/mkldnn/cpu_quantize_squash_pass.h similarity index 100% rename from paddle/fluid/framework/ir/cpu_quantize_squash_pass.h rename to paddle/fluid/framework/ir/mkldnn/cpu_quantize_squash_pass.h diff --git a/paddle/fluid/framework/ir/cpu_quantize_squash_pass_tester.cc b/paddle/fluid/framework/ir/mkldnn/cpu_quantize_squash_pass_tester.cc similarity index 98% rename from paddle/fluid/framework/ir/cpu_quantize_squash_pass_tester.cc rename to paddle/fluid/framework/ir/mkldnn/cpu_quantize_squash_pass_tester.cc index 3a3eb53f79955b37f5f9af6a09b2f9c8e934aa3e..3cf51d97aa4b8be468b8c2a78dd17aafbbf0e15b 100644 --- a/paddle/fluid/framework/ir/cpu_quantize_squash_pass_tester.cc +++ b/paddle/fluid/framework/ir/mkldnn/cpu_quantize_squash_pass_tester.cc @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -#include "paddle/fluid/framework/ir/cpu_quantize_squash_pass.h" +#include "paddle/fluid/framework/ir/mkldnn/cpu_quantize_squash_pass.h" #include #include "paddle/fluid/framework/naive_executor.h" #include "paddle/fluid/platform/place.h" diff --git a/paddle/fluid/inference/analysis/argument.h b/paddle/fluid/inference/analysis/argument.h index 997f3575f457b67d4df5000705724b46cd8b951d..78cc1e9688359cb82d710ea9e9dba972bf9a7516 100644 --- a/paddle/fluid/inference/analysis/argument.h +++ b/paddle/fluid/inference/analysis/argument.h @@ -41,8 +41,11 @@ namespace inference { namespace analysis { using framework::ir::Graph; + +#ifdef PADDLE_WITH_MKLDNN using VarQuantScale = std::unordered_map>; +#endif /* * The argument definition of both Pass and PassManagers. @@ -132,6 +135,7 @@ struct Argument { DECL_ARGUMENT_FIELD(mkldnn_enabled_op_types, MKLDNNEnabledOpTypes, std::unordered_set); +#ifdef PADDLE_WITH_MKLDNN // A set of op types to enable their quantized kernels DECL_ARGUMENT_FIELD(quantize_enabled_op_types, QuantizeEnabledOpTypes, std::unordered_set); @@ -142,6 +146,7 @@ struct Argument { // Scales for variables to be quantized DECL_ARGUMENT_FIELD(quant_var_scales, QuantVarScales, VarQuantScale); +#endif // Passed from config. DECL_ARGUMENT_FIELD(use_gpu, UseGPU, bool); diff --git a/paddle/fluid/inference/analysis/ir_pass_manager.cc b/paddle/fluid/inference/analysis/ir_pass_manager.cc index 1556caa46412c8a2dacd44f2187666c6a1fda6bf..c7960004665381aa916a08f57fcaa7c56781a1af 100644 --- a/paddle/fluid/inference/analysis/ir_pass_manager.cc +++ b/paddle/fluid/inference/analysis/ir_pass_manager.cc @@ -61,6 +61,7 @@ void IRPassManager::CreatePasses(Argument *argument, pass->Set("mkldnn_enabled_op_types", new std::unordered_set( argument->mkldnn_enabled_op_types())); +#ifdef PADDLE_WITH_MKLDNN } else if (pass_name == "cpu_quantize_placement_pass") { pass->Set("quantize_enabled_op_types", new std::unordered_set( @@ -71,6 +72,7 @@ void IRPassManager::CreatePasses(Argument *argument, } else if (pass_name == "cpu_quantize_pass") { pass->Set("quant_var_scales", new VarQuantScale(argument->quant_var_scales())); +#endif } else if (pass_name == "tensorrt_subgraph_pass") { pass->Set("workspace_size", new int(argument->tensorrt_workspace_size())); pass->Set("max_batch_size", new int(argument->tensorrt_max_batch_size()));