Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into multihead-fuse-pass-support-matmul

Author: cryoco

CMakeLists.txt

@@ -61,6 +61,7 @@ if(WITH_MUSL)
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-error=deprecated-declarations -Wno-deprecated-declarations -Wno-error=pessimizing-move -Wno-error=deprecated-copy")
 endif()
 
+
 if(WIN32)
     option(MSVC_STATIC_CRT "use static C Runtime library by default" ON)
 
@@ -124,7 +125,7 @@ if(WIN32)
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /wd4068 /wd4129 /wd4244 /wd4267 /wd4297 /wd4530 /wd4577 /wd4819 /wd4838")
 
     foreach(flag_var CMAKE_SHARED_LINKER_FLAGS CMAKE_STATIC_LINKER_FLAGS CMAKE_EXE_LINKER_FLAGS CMAKE_LINKER_FLAGS)
-        string(APPEND ${flag_var} "/ignore:4049 /ignore:4217 /ignore:4006 /ignore:4221")
+        set(${flag_var} "${${flag_var}} /ignore:4049 /ignore:4217 /ignore:4006 /ignore:4221")
     endforeach(flag_var)
 
     if (WITH_WIN_DUMP_DBG)

cmake/external/warpctc.cmake

@@ -18,7 +18,7 @@ SET(WARPCTC_PREFIX_DIR  ${THIRD_PARTY_PATH}/warpctc)
 SET(WARPCTC_SOURCE_DIR  ${THIRD_PARTY_PATH}/warpctc/src/extern_warpctc)
 SET(WARPCTC_INSTALL_DIR ${THIRD_PARTY_PATH}/install/warpctc)
 set(WARPCTC_REPOSITORY  ${GIT_URL}/baidu-research/warp-ctc.git)
-set(WARPCTC_TAG         95a461eddeabd51099ef059dcfada1117eb1bfb8)
+set(WARPCTC_TAG         cd828e5b6c3b953b82af73f7f44cddc393a20efa)
 
 SET(WARPCTC_INCLUDE_DIR "${WARPCTC_INSTALL_DIR}/include"
     CACHE PATH "Warp-ctc Directory" FORCE)

cmake/generic.cmake

@@ -492,10 +492,8 @@ function(nv_library TARGET_NAME)
         message(FATAL "Please specify source file or library in nv_library.")
       endif()
     endif(nv_library_SRCS)
-    if (WIN32 AND ${CMAKE_CUDA_COMPILER_VERSION} LESS 11.0)
-      if(${MSVC_VERSION} LESS_EQUAL 1900)
-        set_target_properties(${TARGET_NAME} PROPERTIES VS_USER_PROPS ${WIN_PROPS})
-      endif()
+    if((CUDA_VERSION GREATER 9.2) AND (CUDA_VERSION LESS 11.0) AND (MSVC_VERSION LESS 1910))
+      set_target_properties(${TARGET_NAME} PROPERTIES VS_USER_PROPS ${WIN_PROPS})
     endif()
   endif()
 endfunction(nv_library)
@@ -512,7 +510,7 @@ function(nv_binary TARGET_NAME)
       add_dependencies(${TARGET_NAME} ${nv_binary_DEPS})
       common_link(${TARGET_NAME})
     endif()
-    if (WIN32 AND ${CMAKE_CUDA_COMPILER_VERSION} LESS 11.0)
+    if((CUDA_VERSION GREATER 9.2) AND (CUDA_VERSION LESS 11.0) AND (MSVC_VERSION LESS 1910))
       set_target_properties(${TARGET_NAME} PROPERTIES VS_USER_PROPS ${WIN_PROPS})
     endif()
   endif()
@@ -539,7 +537,7 @@ function(nv_test TARGET_NAME)
     set_property(TEST ${TARGET_NAME} PROPERTY ENVIRONMENT FLAGS_cpu_deterministic=true)
     set_property(TEST ${TARGET_NAME} PROPERTY ENVIRONMENT FLAGS_init_allocated_mem=true)
     set_property(TEST ${TARGET_NAME} PROPERTY ENVIRONMENT FLAGS_cudnn_deterministic=true)
-    if (WIN32 AND ${CMAKE_CUDA_COMPILER_VERSION} LESS 11.0)
+    if((CUDA_VERSION GREATER 9.2) AND (CUDA_VERSION LESS 11.0) AND (MSVC_VERSION LESS 1910))
       set_target_properties(${TARGET_NAME} PROPERTIES VS_USER_PROPS ${WIN_PROPS})
     endif()
   endif()

cmake/init.cmake

@@ -18,6 +18,10 @@ if(NOT WIN32)
     set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "-O2 -g -DNDEBUG")
     set(CMAKE_CXX_FLAGS_MINSIZEREL "-Os -DNDEBUG")
 else()
+    # It has not been used now, it can specify CUDA compile flag manualy,
+    # its use is to remvoe /Zi to reduce GPU static library size. But it's dangerous
+    # because CUDA will update by nvidia, then error will occur.
+    # Now, it's used in CUDA:[10.0, 10.2]
     set(WIN_PROPS ${CMAKE_SOURCE_DIR}/cmake/paddle_win.props)
 endif()
 

cmake/paddle_win.props

@@ -15,7 +15,7 @@
             <Warning>InheritFromHost</Warning>
 
             <BaseCommandLineTemplate>-ccbin "%(VCBinDir)" -x cu [GenerateRelocatableDeviceCode] [Include] [RequiredIncludes] [InterleaveSourceInPTX] [GPUDebugInfo] [GenerateLineInfo] [Keep] [KeepDir] [MaxRegCount] [PtxAsOptionV] [TargetMachinePlatform] [NvccCompilation] [CudaRuntime] [AdditionalOptions]</BaseCommandLineTemplate>
-            <BuildCommandLineTemplate>--use-local-env --cl-version $(CudaClVersion)</BuildCommandLineTemplate>
+            <BuildCommandLineTemplate>--use-local-env $(CudaClVersion)</BuildCommandLineTemplate>
             <BuildDynamicCommandLineTemplate>[CodeGeneration]</BuildDynamicCommandLineTemplate>
             <CleanCommandLineTemplate>-clean</CleanCommandLineTemplate>
             <!-- <HostCommandLineTemplate>-Xcompiler &quot;/EHsc [Warning] /nologo [Optimization] $(CudaForceSynchronousPdbWrites) /Zi [RuntimeChecks] [Runtime] [TypeInfo]&quot;</HostCommandLineTemplate> -->

go/README_cn.md

@@ -50,6 +50,7 @@ output_data := value.Interface().([][]float32)
 
 运行
 ```bash
+go mod init github.com/paddlepaddle
 export LD_LIBRARY_PATH=`pwd`/paddle_c/paddle/lib:$LD_LIBRARY_PATH
 go run ./demo/mobilenet.go
 ```

go/demo/mobilenet.go

@@ -13,7 +13,7 @@
 // limitations under the License.
 package main
 
-import "../paddle"
+import "github.com/paddlepaddle/paddle"
 import "strings"
 import "io/ioutil"
 import "strconv"

go/paddle/common.go

@@ -15,7 +15,7 @@
 package paddle
 
 // #cgo CFLAGS: -I${SRCDIR}/../paddle_c/paddle/include
-// #cgo LDFLAGS: -L${SRCDIR}/../paddle_c/paddle/lib -lpaddle_fluid_c
+// #cgo LDFLAGS: -L${SRCDIR}/../paddle_c/paddle/lib -lpaddle_inference_c
 // #include <stdbool.h>
 // #include <paddle_c_api.h>
 import "C"

go/paddle/config.go

@@ -15,7 +15,7 @@
 package paddle
 
 // #cgo CFLAGS: -I${SRCDIR}/../paddle_c/paddle/include
-// #cgo LDFLAGS: -L${SRCDIR}/../paddle_c/paddle/lib -lpaddle_fluid_c
+// #cgo LDFLAGS: -L${SRCDIR}/../paddle_c/paddle/lib -lpaddle_inference_c
 // #include <stdbool.h>
 // #include <stdlib.h>
 // #include <paddle_c_api.h>

go/paddle/predictor.go

@@ -15,7 +15,7 @@
 package paddle
 
 // #cgo CFLAGS: -I${SRCDIR}/../paddle_c/paddle/include
-// #cgo LDFLAGS: -L${SRCDIR}/../paddle_c/paddle/lib -lpaddle_fluid_c
+// #cgo LDFLAGS: -L${SRCDIR}/../paddle_c/paddle/lib -lpaddle_inference_c
 // #include <stdbool.h>
 // #include "paddle_c_api.h"
 import "C"
@@ -88,7 +88,7 @@ func (predictor *Predictor) GetInputNames() []string {
 }
 
 func (predictor *Predictor) GetOutputNames() []string {
-	names := make([]string, predictor.GetInputNum())
+	names := make([]string, predictor.GetOutputNum())
 	for i := 0; i < len(names); i++ {
 		names[i] = predictor.GetOutputName(i)
 	}