update username repo

CMakeLists.txt

@@ -74,25 +74,30 @@ if(WIN32)
         endforeach(flag_var)
     endif()
 
-    # windows build turn off warnings.
+    # windows build turn off warnings, use parallel compiling.
     foreach(flag_var
         CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_DEBUG CMAKE_CXX_FLAGS_RELEASE
         CMAKE_CXX_FLAGS_MINSIZEREL CMAKE_CXX_FLAGS_RELWITHDEBINFO
         CMAKE_C_FLAGS CMAKE_C_FLAGS_DEBUG CMAKE_C_FLAGS_RELEASE
         CMAKE_C_FLAGS_MINSIZEREL CMAKE_C_FLAGS_RELWITHDEBINFO)
         string(REGEX REPLACE "/W[1-4]" " /W0 " ${flag_var} "${${flag_var}}")
+        set(${flag_var} "${${flag_var}} /MP")
     endforeach(flag_var)
     foreach(flag_var CMAKE_CXX_FLAGS CMAKE_C_FLAGS)
         set(${flag_var} "${${flag_var}} /w")
     endforeach(flag_var)
 
-    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /wd4068 /wd4129 /wd4244 /wd4267 /wd4297 /wd4530 /wd4577 /wd4819 /wd4838 /MP")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /wd4068 /wd4129 /wd4244 /wd4267 /wd4297 /wd4530 /wd4577 /wd4819 /wd4838 /MP")
-    message(STATUS "Using parallel compiling (/MP)")
-    set(PADDLE_LINK_FLAGS "/IGNORE:4006 /IGNORE:4098 /IGNORE:4217 /IGNORE:4221")
-    set(CMAKE_STATIC_LINKER_FLAGS  "${CMAKE_STATIC_LINKER_FLAGS} ${PADDLE_LINK_FLAGS}")
-    set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} ${PADDLE_LINK_FLAGS}")
-    set(CMAKE_EXE_LINKER_FLAGS  "${CMAKE_EXE_LINKER_FLAGS} ${PADDLE_LINK_FLAGS}")
+    # Windows Remove /Zi, /ZI for Release, MinSizeRel builds
+    foreach(flag_var
+        CMAKE_C_FLAGS CMAKE_C_FLAGS_RELEASE CMAKE_C_FLAGS_MINSIZEREL
+        CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_RELEASE CMAKE_CXX_FLAGS_MINSIZEREL)
+        if(${flag_var} MATCHES "/Z[iI]")
+            string(REGEX REPLACE "/Z[iI]" "" ${flag_var} "${${flag_var}}")
+        endif()
+    endforeach(flag_var)
+
+    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /wd4068 /wd4129 /wd4244 /wd4267 /wd4297 /wd4530 /wd4577 /wd4819 /wd4838")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /wd4068 /wd4129 /wd4244 /wd4267 /wd4297 /wd4530 /wd4577 /wd4819 /wd4838")
 else(WIN32)
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-error=deprecated-declarations -Wno-deprecated-declarations")
 endif(WIN32)

cmake/init.cmake

@@ -1,7 +1,7 @@
 # Attention: cmake will append these flags to compile command automatically.
 # So if you want to add global option, change this file rather than flags.cmake
 
-# NOT WIN32
+# Linux
 # DEBUG:  default: "-g"
 # RELEASE:  default: "-O3 -DNDEBUG"
 # RELWITHDEBINFO: default: "-O2 -g -DNDEBUG"
@@ -17,6 +17,8 @@ if(NOT WIN32)
     set(CMAKE_CXX_FLAGS_RELEASE "-O3 -DNDEBUG")
     set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "-O2 -g -DNDEBUG")
     set(CMAKE_CXX_FLAGS_MINSIZEREL "-Os -DNDEBUG")
+else()
+    set(WIN_PROPS ${CMAKE_SOURCE_DIR}/cmake/paddle_win.props)
 endif()
 
 if(WITH_GPU)
@@ -25,9 +27,3 @@ if(WITH_GPU)
     set(CMAKE_CUDA_FLAGS_RELWITHDEBINFO "-O2 -g -DNDEBUG")
     set(CMAKE_CUDA_FLAGS_MINSIZEREL "-O1 -DNDEBUG")
 endif()
-
-if(WIN32)
-    set(WIN_PROPS ${CMAKE_SOURCE_DIR}/cmake/paddle_win.props)
-    set(CMAKE_CXX_FLAGS_RELEASE "-O3 -Os -DNDEBUG")
-endif()
-

paddle/fluid/framework/ir/conv_affine_channel_fuse_pass.cc

@@ -238,11 +238,11 @@ REGISTER_PASS(conv_eltwiseadd_affine_channel_fuse_pass,
 REGISTER_PASS_CAPABILITY(conv_affine_channel_fuse_pass)
     .AddCombination(
         paddle::framework::compatible::OpVersionComparatorCombination()
-            .EQ("conv2d", 0)
+            .LE("conv2d", 1)
             .EQ("affine_channel", 0));
 REGISTER_PASS_CAPABILITY(conv_eltwiseadd_affine_channel_fuse_pass)
     .AddCombination(
         paddle::framework::compatible::OpVersionComparatorCombination()
-            .EQ("conv2d", 0)
+            .LE("conv2d", 1)
             .EQ("elementwise_add", 0)
             .EQ("affine_channel", 0));

paddle/fluid/framework/ir/conv_bn_fuse_pass.cc

@@ -383,11 +383,11 @@ REGISTER_PASS(depthwise_conv_eltwiseadd_bn_fuse_pass,
 REGISTER_PASS_CAPABILITY(conv_bn_fuse_pass)
     .AddCombination(
         paddle::framework::compatible::OpVersionComparatorCombination()
-            .EQ("conv2d", 0)
+            .LE("conv2d", 1)
             .EQ("batch_norm", 0));
 REGISTER_PASS_CAPABILITY(conv_eltwiseadd_bn_fuse_pass)
     .AddCombination(
         paddle::framework::compatible::OpVersionComparatorCombination()
-            .EQ("conv2d", 0)
+            .LE("conv2d", 1)
             .EQ("elementwise_add", 0)
             .EQ("batch_norm", 0));

paddle/fluid/framework/ir/conv_elementwise_add2_act_fuse_pass.cc

@@ -12,7 +12,9 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 #include "paddle/fluid/framework/ir/conv_elementwise_add2_act_fuse_pass.h"
+
 #include <string>
+
 #include "paddle/fluid/framework/op_version_registry.h"
 
 namespace paddle {
@@ -119,7 +121,7 @@ REGISTER_PASS(conv_elementwise_add2_act_fuse_pass,
 REGISTER_PASS_CAPABILITY(conv_elementwise_add2_act_fuse_pass)
     .AddCombination(
         paddle::framework::compatible::OpVersionComparatorCombination()
-            .EQ("conv2d", 0)
+            .LE("conv2d", 1)
             .EQ("elementwise_add", 0)
             .EQ("relu", 0)
             .EQ("identity", 0));

paddle/fluid/framework/ir/conv_elementwise_add_act_fuse_pass.cc

@@ -13,6 +13,7 @@
 // limitations under the License.
 
 #include "paddle/fluid/framework/ir/conv_elementwise_add_act_fuse_pass.h"
+
 #include <string>
 
 #include "paddle/fluid/framework/ir/graph_viz_pass.h"
@@ -107,7 +108,7 @@ REGISTER_PASS(conv_elementwise_add_act_fuse_pass,
 REGISTER_PASS_CAPABILITY(conv_elementwise_add_act_fuse_pass)
     .AddCombination(
         paddle::framework::compatible::OpVersionComparatorCombination()
-            .EQ("conv2d", 0)
+            .LE("conv2d", 1)
             .EQ("elementwise_add", 0)
             .EQ("relu", 0)
             .EQ("identity", 0));

paddle/fluid/framework/ir/conv_elementwise_add_fuse_pass.cc

@@ -13,6 +13,7 @@
 // limitations under the License.
 
 #include "paddle/fluid/framework/ir/conv_elementwise_add_fuse_pass.h"
+
 #include <string>
 
 #include "paddle/fluid/framework/ir/graph_viz_pass.h"
@@ -93,5 +94,5 @@ REGISTER_PASS(conv_elementwise_add_fuse_pass,
 REGISTER_PASS_CAPABILITY(conv_elementwise_add_fuse_pass)
     .AddCombination(
         paddle::framework::compatible::OpVersionComparatorCombination()
-            .EQ("conv2d", 0)
+            .LE("conv2d", 1)
             .EQ("elementwise_add", 0));

paddle/fluid/framework/ir/graph_pattern_detector.cc

@@ -2102,8 +2102,8 @@ PDNode *patterns::Bfloat16Placement::operator()(
     const std::unordered_set<std::string> &bfloat16_enabled_op_types) {
   std::unordered_set<std::string> supported_op_types =
       std::unordered_set<std::string>({"concat", "conv2d", "fusion_gru", "gelu",
-                                       "reshape2", "softmax", "sum",
-                                       "transpose2"});
+                                       "layer_norm", "reshape2", "softmax",
+                                       "sum", "transpose2"});
   if (!bfloat16_enabled_op_types.empty()) {
     supported_op_types = bfloat16_enabled_op_types;
   }

paddle/fluid/framework/ir/mkldnn/conv_activation_mkldnn_fuse_pass.cc

@@ -13,7 +13,9 @@
 // limitations under the License.
 
 #include "paddle/fluid/framework/ir/mkldnn/conv_activation_mkldnn_fuse_pass.h"
+
 #include <vector>
+
 #include "paddle/fluid/framework/op_version_registry.h"
 #include "paddle/fluid/platform/enforce.h"
 
@@ -107,29 +109,29 @@ REGISTER_PASS(conv_relu_mkldnn_fuse_pass,
 REGISTER_PASS_CAPABILITY(conv_relu_mkldnn_fuse_pass)
     .AddCombination(
         paddle::framework::compatible::OpVersionComparatorCombination()
-            .EQ("conv2d", 0)
+            .LE("conv2d", 1)
             .EQ("relu", 0));
 
 REGISTER_PASS(conv_leaky_relu_mkldnn_fuse_pass,
               paddle::framework::ir::Conv2DLeakyReLUFusePass);
 REGISTER_PASS_CAPABILITY(conv_leaky_relu_mkldnn_fuse_pass)
     .AddCombination(
         paddle::framework::compatible::OpVersionComparatorCombination()
-            .EQ("conv2d", 0)
+            .LE("conv2d", 1)
             .LE("leaky_relu", 1));
 
 REGISTER_PASS(conv_relu6_mkldnn_fuse_pass,
               paddle::framework::ir::Conv2DReLU6FusePass);
 REGISTER_PASS_CAPABILITY(conv_relu6_mkldnn_fuse_pass)
     .AddCombination(
         paddle::framework::compatible::OpVersionComparatorCombination()
-            .EQ("conv2d", 0)
+            .LE("conv2d", 1)
             .EQ("relu6", 0));
 
 REGISTER_PASS(conv_swish_mkldnn_fuse_pass,
               paddle::framework::ir::Conv2DSwishFusePass);
 REGISTER_PASS_CAPABILITY(conv_swish_mkldnn_fuse_pass)
     .AddCombination(
         paddle::framework::compatible::OpVersionComparatorCombination()
-            .EQ("conv2d", 0)
+            .LE("conv2d", 1)
             .EQ("swish", 0));

paddle/fluid/framework/ir/mkldnn/conv_bias_mkldnn_fuse_pass.cc

@@ -13,8 +13,10 @@
 // limitations under the License.
 
 #include "paddle/fluid/framework/ir/mkldnn/conv_bias_mkldnn_fuse_pass.h"
+
 #include <functional>
 #include <vector>
+
 #include "paddle/fluid/framework/lod_tensor.h"
 #include "paddle/fluid/framework/op_version_registry.h"
 #include "paddle/fluid/platform/enforce.h"
@@ -150,7 +152,7 @@ REGISTER_PASS(conv_bias_mkldnn_fuse_pass,
 REGISTER_PASS_CAPABILITY(conv_bias_mkldnn_fuse_pass)
     .AddCombination(
         paddle::framework::compatible::OpVersionComparatorCombination()
-            .EQ("conv2d", 0)
+            .LE("conv2d", 1)
             .EQ("elementwise_add", 0));
 
 REGISTER_PASS(conv_transpose_bias_mkldnn_fuse_pass,

paddle/fluid/framework/ir/mkldnn/conv_concat_relu_mkldnn_fuse_pass.cc

@@ -13,7 +13,9 @@
 // limitations under the License.
 
 #include "paddle/fluid/framework/ir/mkldnn/conv_concat_relu_mkldnn_fuse_pass.h"
+
 #include <vector>
+
 #include "paddle/fluid/framework/op_version_registry.h"
 #include "paddle/fluid/platform/enforce.h"
 
@@ -128,6 +130,6 @@ REGISTER_PASS(conv_concat_relu_mkldnn_fuse_pass,
 REGISTER_PASS_CAPABILITY(conv_concat_relu_mkldnn_fuse_pass)
     .AddCombination(
         paddle::framework::compatible::OpVersionComparatorCombination()
-            .EQ("conv2d", 0)
+            .LE("conv2d", 1)
             .EQ("concat", 0)
             .EQ("relu", 0));

paddle/fluid/framework/ir/mkldnn/conv_elementwise_add_mkldnn_fuse_pass.cc

@@ -13,11 +13,13 @@
 // limitations under the License.
 
 #include "paddle/fluid/framework/ir/mkldnn/conv_elementwise_add_mkldnn_fuse_pass.h"
+
 #include <functional>
 #include <list>
 #include <map>
 #include <memory>
 #include <tuple>
+
 #include "paddle/fluid/framework/ir/graph_traits.h"
 #include "paddle/fluid/framework/op_version_registry.h"
 
@@ -226,19 +228,20 @@ GraphWithStats ResidualConnectionMKLDNNFusePass::FuseConvAsX(
       pattern->NewNode(elementwise_add_pattern.elementwise_add_y_repr()));
   conv_output->AsIntermediate();
 
-  auto get_node_from_elementwise_add = [&elementwise_add_pattern](
-      const GraphPatternDetector::subgraph_t& subgraph)
+  auto get_node_from_elementwise_add =
+      [&elementwise_add_pattern](
+          const GraphPatternDetector::subgraph_t& subgraph)
       -> std::tuple<Node*, Node*, Node*> {
-        GET_IR_NODE_FROM_SUBGRAPH(elementwise_add_op, elementwise_add_op,
-                                  elementwise_add_pattern);
-        GET_IR_NODE_FROM_SUBGRAPH(elementwise_add_y, elementwise_add_y,
-                                  elementwise_add_pattern);
-        GET_IR_NODE_FROM_SUBGRAPH(elementwise_add_out, elementwise_add_out,
-                                  elementwise_add_pattern);
-
-        return std::make_tuple(elementwise_add_op, elementwise_add_y,
-                               elementwise_add_out);
-      };
+    GET_IR_NODE_FROM_SUBGRAPH(elementwise_add_op, elementwise_add_op,
+                              elementwise_add_pattern);
+    GET_IR_NODE_FROM_SUBGRAPH(elementwise_add_y, elementwise_add_y,
+                              elementwise_add_pattern);
+    GET_IR_NODE_FROM_SUBGRAPH(elementwise_add_out, elementwise_add_out,
+                              elementwise_add_pattern);
+
+    return std::make_tuple(elementwise_add_op, elementwise_add_y,
+                           elementwise_add_out);
+  };
 
   return ExecuteHandleOnGraph<IdentityFuseHandle>(
       &gpd, graph_with_stats,
@@ -263,19 +266,20 @@ GraphWithStats ResidualConnectionMKLDNNFusePass::FuseConvAsY(
       conv_output);
   conv_output->AsIntermediate();
 
-  auto get_node_from_elementwise_add = [&elementwise_add_pattern](
-      const GraphPatternDetector::subgraph_t& subgraph)
+  auto get_node_from_elementwise_add =
+      [&elementwise_add_pattern](
+          const GraphPatternDetector::subgraph_t& subgraph)
       -> std::tuple<Node*, Node*, Node*> {
-        GET_IR_NODE_FROM_SUBGRAPH(elementwise_add_op, elementwise_add_op,
-                                  elementwise_add_pattern);
-        GET_IR_NODE_FROM_SUBGRAPH(elementwise_add_x, elementwise_add_x,
-                                  elementwise_add_pattern);
-        GET_IR_NODE_FROM_SUBGRAPH(elementwise_add_out, elementwise_add_out,
-                                  elementwise_add_pattern);
-
-        return std::make_tuple(elementwise_add_op, elementwise_add_x,
-                               elementwise_add_out);
-      };
+    GET_IR_NODE_FROM_SUBGRAPH(elementwise_add_op, elementwise_add_op,
+                              elementwise_add_pattern);
+    GET_IR_NODE_FROM_SUBGRAPH(elementwise_add_x, elementwise_add_x,
+                              elementwise_add_pattern);
+    GET_IR_NODE_FROM_SUBGRAPH(elementwise_add_out, elementwise_add_out,
+                              elementwise_add_pattern);
+
+    return std::make_tuple(elementwise_add_op, elementwise_add_x,
+                           elementwise_add_out);
+  };
 
   return ExecuteHandleOnGraph<IdentityFuseHandle>(
       &gpd, graph_with_stats,
@@ -302,16 +306,17 @@ GraphWithStats ResidualConnectionMKLDNNFusePass::FuseProjectionConv(
   conv_x_output->AsIntermediate();
   conv_y_output->AsIntermediate();
 
-  auto get_node_from_elementwise_add = [&elementwise_add_pattern](
-      const GraphPatternDetector::subgraph_t& subgraph)
+  auto get_node_from_elementwise_add =
+      [&elementwise_add_pattern](
+          const GraphPatternDetector::subgraph_t& subgraph)
       -> std::tuple<Node*, Node*> {
-        GET_IR_NODE_FROM_SUBGRAPH(elementwise_add_op, elementwise_add_op,
-                                  elementwise_add_pattern);
-        GET_IR_NODE_FROM_SUBGRAPH(elementwise_add_out, elementwise_add_out,
-                                  elementwise_add_pattern);
+    GET_IR_NODE_FROM_SUBGRAPH(elementwise_add_op, elementwise_add_op,
+                              elementwise_add_pattern);
+    GET_IR_NODE_FROM_SUBGRAPH(elementwise_add_out, elementwise_add_out,
+                              elementwise_add_pattern);
 
-        return std::make_tuple(elementwise_add_op, elementwise_add_out);
-      };
+    return std::make_tuple(elementwise_add_op, elementwise_add_out);
+  };
 
   return ExecuteHandleOnGraph<ProjectionFuseHandle>(
       &gpd, graph_with_stats,
@@ -345,5 +350,5 @@ REGISTER_PASS(conv_elementwise_add_mkldnn_fuse_pass,
 REGISTER_PASS_CAPABILITY(conv_elementwise_add_mkldnn_fuse_pass)
     .AddCombination(
         paddle::framework::compatible::OpVersionComparatorCombination()
-            .EQ("conv2d", 0)
+            .LE("conv2d", 1)
             .EQ("elementwise_add", 0));

paddle/fluid/framework/ir/mkldnn/cpu_bfloat16_pass.cc

@@ -16,6 +16,7 @@ limitations under the License. */
 #include <vector>
 
 #include "paddle/fluid/framework/ir/graph_pattern_detector.h"
+#include "paddle/fluid/framework/op_version_registry.h"
 #include "paddle/fluid/platform/mkldnn_helper.h"
 #include "paddle/fluid/string/pretty_log.h"
 
@@ -157,3 +158,8 @@ void CPUBFloat16Pass::ApplyImpl(ir::Graph* graph) const {
 }  // namespace paddle
 
 REGISTER_PASS(cpu_bfloat16_pass, paddle::framework::ir::CPUBFloat16Pass);
+
+REGISTER_PASS_CAPABILITY(cpu_bfloat16_pass)
+    .AddCombination(
+        paddle::framework::compatible::OpVersionComparatorCombination().GE(
+            "quantize", 1));

paddle/fluid/framework/ir/mkldnn/depthwise_conv_mkldnn_pass.cc

@@ -63,5 +63,5 @@ REGISTER_PASS(depthwise_conv_mkldnn_pass,
               paddle::framework::ir::DepthwiseConvMKLDNNPass);
 REGISTER_PASS_CAPABILITY(depthwise_conv_mkldnn_pass)
     .AddCombination(
-        paddle::framework::compatible::OpVersionComparatorCombination().EQ(
-            "depthwise_conv2d", 0));
+        paddle::framework::compatible::OpVersionComparatorCombination().LE(
+            "depthwise_conv2d", 1));

paddle/fluid/framework/ir/mkldnn/mkldnn_inplace_pass.cc

@@ -17,10 +17,12 @@
 #include <memory>
 #include <string>
 #include <unordered_map>
+#include <unordered_set>
 #include <vector>
 #include "paddle/fluid/framework/eigen.h"
 #include "paddle/fluid/framework/lod_tensor.h"
 #include "paddle/fluid/framework/op_info.h"
+#include "paddle/fluid/framework/op_version_registry.h"
 #include "paddle/fluid/platform/enforce.h"
 
 namespace paddle {
@@ -215,3 +217,9 @@ void MKLDNNInPlacePass::ApplyImpl(ir::Graph* graph) const {
 }  // namespace paddle
 
 REGISTER_PASS(mkldnn_inplace_pass, paddle::framework::ir::MKLDNNInPlacePass);
+REGISTER_PASS_CAPABILITY(mkldnn_inplace_pass)
+    .AddCombination(
+        paddle::framework::compatible::OpVersionComparatorCombination()
+            .EQ("softmax", 0)
+            .EQ("elementwise_add", 0)
+            .EQ("tanh", 0));