From 5bd386a4f965331dfd379a71c601750a9d0c1bfd Mon Sep 17 00:00:00 2001 From: agibsonccc Date: Tue, 9 Feb 2021 07:44:23 +0900 Subject: [PATCH] Add new clion rules, fix batch norml --- libnd4j/blas/CMakeLists.txt | 98 +++---- .../declarable/generic/nn/fusedBatchNorm.cpp | 254 +++++++++--------- .../linalg/api/ops/custom/FusedBatchNorm.java | 5 +- .../TFGraphs/TFGraphTestAllSameDiff.java | 9 +- .../definitions/TensorflowOpDeclarations.kt | 6 +- .../tensorflow-mapping-ruleset.pbtxt | 18 ++ .../tensorflow/TestTensorflowIR.kt | 10 +- .../tensorflow-processes.pbtxt | 18 ++ 8 files changed, 232 insertions(+), 186 deletions(-) diff --git a/libnd4j/blas/CMakeLists.txt b/libnd4j/blas/CMakeLists.txt index ea5241e1c..f1cecda83 100755 --- a/libnd4j/blas/CMakeLists.txt +++ b/libnd4j/blas/CMakeLists.txt @@ -51,9 +51,9 @@ endif() if(WIN32 AND NOT ANDROID) get_property(dirs DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR} PROPERTY INCLUDE_DIRECTORIES) - if ("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU") - set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wa,-mbig-obj") - endif() + if ("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU") + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wa,-mbig-obj") + endif() foreach(dir ${dirs}) message(STATUS "dir='${dir}'") endforeach() @@ -161,8 +161,8 @@ if(SD_CUDA) endif() if (CUDA_FOUND) - message("CUDA include directory: ${CUDA_INCLUDE_DIRS}") - include_directories(${CUDA_INCLUDE_DIRS}) + message("CUDA include directory: ${CUDA_INCLUDE_DIRS}") + include_directories(${CUDA_INCLUDE_DIRS}) message("CUDA found!") if ("${SD_EXPERIMENTAL}" STREQUAL "yes") message("Experimental mode ENABLED") @@ -181,7 +181,7 @@ if(SD_CUDA) set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xcompiler=-fPIC") endif() - string( TOLOWER "${COMPUTE}" COMPUTE_CMP ) + string( TOLOWER "${COMPUTE}" COMPUTE_CMP ) if ("${COMPUTE_CMP}" STREQUAL "all") CUDA_SELECT_NVCC_ARCH_FLAGS(CUDA_ARCH_FLAGS "Common") elseif("${COMPUTE_CMP}" STREQUAL "auto") @@ -197,9 +197,9 @@ if(SD_CUDA) endif() # list to spaces string (REPLACE ";" " " CUDA_ARCH_FLAGS "${CUDA_ARCH_FLAGS}") - + set(CMAKE_CUDA_FLAGS " ${CMAKE_CUDA_FLAGS} -DCUDA_VERSION_MAJOR=${CUDA_VERSION_MAJOR} ${EXPM} -w --cudart=static --expt-extended-lambda -Xfatbin -compress-all ${CUDA_ARCH_FLAGS}") - + file(GLOB_RECURSE PERF_SOURCES false ../include/performance/*.cpp ../include/performance/*.h) file(GLOB_RECURSE EXCEPTIONS_SOURCES false ../include/exceptions/*.cpp ../include/exceptions/*.h) file(GLOB_RECURSE EXEC_SOURCES false ../include/execution/impl/*.cpp ../include/execution/*.cu ../include/execution/*.h) @@ -218,23 +218,23 @@ if(SD_CUDA) file(GLOB_RECURSE COMPILATION_UNITS false ../include/loops/cuda/compilation_units/*.cu.in - ../include/ops/impl/compilation_units/*.cpp.in) + ../include/ops/impl/compilation_units/*.cpp.in) - foreach(FL_ITEM ${COMPILATION_UNITS}) + foreach(FL_ITEM ${COMPILATION_UNITS}) genCompilation(FL_ITEM) - endforeach() + endforeach() if (HAVE_CUDNN) message("cuDNN included") file(GLOB_RECURSE CUSTOMOPS_CUDNN_SOURCES false ../include/ops/declarable/platform/cudnn/*.cu) endif() - add_library(samediff_obj OBJECT ${LOOPS_SOURCES_CUDA} ${LEGACY_SOURCES} + add_library(samediff_obj OBJECT ${LOOPS_SOURCES_CUDA} ${LEGACY_SOURCES} ${CUSTOMOPS_HELPERS_SOURCES} ${HELPERS_SOURCES} ${EXEC_SOURCES} ${LOOPS_SOURCES} ${ARRAY_SOURCES} ${TYPES_SOURCES} ${MEMORY_SOURCES} ${GRAPH_SOURCES} ${CUSTOMOPS_SOURCES} ${INDEXING_SOURCES} ${EXCEPTIONS_SOURCES} ${OPS_SOURCES} ${PERF_SOURCES} ${CUSTOMOPS_CUDNN_SOURCES} ${CUSTOMOPS_MKLDNN_SOURCES} - ${CUSTOMOPS_ARMCOMPUTE_SOURCES} ${CUSTOMOPS_GENERIC_SOURCES} - ) + ${CUSTOMOPS_ARMCOMPUTE_SOURCES} ${CUSTOMOPS_GENERIC_SOURCES} + ) if (WIN32) message("MSVC runtime for library: ${MSVC_RT_LIB}") @@ -266,10 +266,10 @@ if(SD_CUDA) SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /EHsc /bigobj /std:c++14") endif() - target_link_libraries(${SD_LIBRARY_NAME} ${CUDA_LIBRARIES} ${CUDA_CUBLAS_LIBRARIES} ${CUDA_cusolver_LIBRARY} ${CUDNN} ${MKLDNN}) - set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${PROJECT_BINARY_DIR}/cuda) + target_link_libraries(${SD_LIBRARY_NAME} ${CUDA_LIBRARIES} ${CUDA_CUBLAS_LIBRARIES} ${CUDA_cusolver_LIBRARY} ${CUDNN} ${MKLDNN}) + set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${PROJECT_BINARY_DIR}/cuda) - install(TARGETS ${SD_LIBRARY_NAME} DESTINATION .) + install(TARGETS ${SD_LIBRARY_NAME} DESTINATION .) endif(CUDA_FOUND) elseif(SD_CPU) @@ -295,13 +295,13 @@ elseif(SD_CPU) file(GLOB_RECURSE LOOPS_SOURCES false ../include/loops/*.cpp ../include/loops/*.h) - file(GLOB_RECURSE COMPILATION_UNITS false ../include/ops/declarable/helpers/cpu/compilation_units/*.cpp.in - ../include/loops/cpu/compilation_units/*.cpp.in ../include/helpers/cpu/loops/*.cpp.in - ../include/ops/impl/compilation_units/*.cpp.in) + file(GLOB_RECURSE COMPILATION_UNITS false ../include/ops/declarable/helpers/cpu/compilation_units/*.cpp.in + ../include/loops/cpu/compilation_units/*.cpp.in ../include/helpers/cpu/loops/*.cpp.in + ../include/ops/impl/compilation_units/*.cpp.in) - foreach(FL_ITEM ${COMPILATION_UNITS}) + foreach(FL_ITEM ${COMPILATION_UNITS}) genCompilation(FL_ITEM) - endforeach() + endforeach() if (SD_X86_BUILD) # we disable platform optimizations for certains files for linux/macos @@ -312,36 +312,36 @@ elseif(SD_CPU) if(SD_CHECK_VECTORIZATION) - set(VECT_FILES cpu/NativeOps.cpp ${OPS_SOURCES} ${HELPERS_SOURCES} ${CUSTOMOPS_GENERIC_SOURCES} ${LOOPS_SOURCES}) - if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU") - - if (CMAKE_COMPILER_IS_GNUCC AND CMAKE_CXX_COMPILER_VERSION VERSION_GREATER 9.0) - set(CHECK_VECT_FLAGS "-ftree-vectorize -fsave-optimization-record") - #to process fsave-optimization-record we will need our cython version code - message("Build Auto vectorization helpers") - execute_process(COMMAND "python3" "${CMAKE_CURRENT_SOURCE_DIR}/../auto_vectorization/cython_setup.py" "build_ext" "--inplace" WORKING_DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}/../auto_vectorization/" RESULT_VARIABLE ret) - message("build='${ret}'") + set(VECT_FILES cpu/NativeOps.cpp ${OPS_SOURCES} ${HELPERS_SOURCES} ${CUSTOMOPS_GENERIC_SOURCES} ${LOOPS_SOURCES}) + if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU") - #remove fail cases that gcc fails produce sometimes - file(GLOB_RECURSE FAILURE_CASES false ../include/loops/cpu/compilation_units/reduce3*.cpp) - #message("*****${FAILURE_CASES}") - foreach(FL_ITEM ${FAILURE_CASES}) - message("Removing failure cases ${FL_ITEM}") - list(REMOVE_ITEM VECT_FILES ${FL_ITEM}) - endforeach() - else() - set(CHECK_VECT_FLAGS "-ftree-vectorize -fopt-info-vec-optimized-missed") + if (CMAKE_COMPILER_IS_GNUCC AND CMAKE_CXX_COMPILER_VERSION VERSION_GREATER 9.0) + set(CHECK_VECT_FLAGS "-ftree-vectorize -fsave-optimization-record") + #to process fsave-optimization-record we will need our cython version code + message("Build Auto vectorization helpers") + execute_process(COMMAND "python3" "${CMAKE_CURRENT_SOURCE_DIR}/../auto_vectorization/cython_setup.py" "build_ext" "--inplace" WORKING_DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}/../auto_vectorization/" RESULT_VARIABLE ret) + message("build='${ret}'") + + #remove fail cases that gcc fails produce sometimes + file(GLOB_RECURSE FAILURE_CASES false ../include/loops/cpu/compilation_units/reduce3*.cpp) + #message("*****${FAILURE_CASES}") + foreach(FL_ITEM ${FAILURE_CASES}) + message("Removing failure cases ${FL_ITEM}") + list(REMOVE_ITEM VECT_FILES ${FL_ITEM}) + endforeach() + else() + set(CHECK_VECT_FLAGS "-ftree-vectorize -fopt-info-vec-optimized-missed") + endif() + message("CHECK VECTORIZATION ${CHECK_VECT_FLAGS}") + set_source_files_properties( ${VECT_FILES} PROPERTIES COMPILE_FLAGS "${CHECK_VECT_FLAGS}" ) endif() - message("CHECK VECTORIZATION ${CHECK_VECT_FLAGS}") - set_source_files_properties( ${VECT_FILES} PROPERTIES COMPILE_FLAGS "${CHECK_VECT_FLAGS}" ) - endif() - endif() + endif() message("CPU BLAS") add_definitions(-D__CPUBLAS__=true) add_library(samediff_obj OBJECT ${LEGACY_SOURCES} ${LOOPS_SOURCES} ${HELPERS_SOURCES} ${EXEC_SOURCES} ${ARRAY_SOURCES} ${TYPES_SOURCES} - ${MEMORY_SOURCES} ${GRAPH_SOURCES} ${CUSTOMOPS_SOURCES} ${EXCEPTIONS_SOURCES} ${INDEXING_SOURCES} ${CUSTOMOPS_MKLDNN_SOURCES} + ${MEMORY_SOURCES} ${GRAPH_SOURCES} ${CUSTOMOPS_SOURCES} ${EXCEPTIONS_SOURCES} ${INDEXING_SOURCES} ${CUSTOMOPS_MKLDNN_SOURCES} ${CUSTOMOPS_ARMCOMPUTE_SOURCES} ${CUSTOMOPS_GENERIC_SOURCES} ${OPS_SOURCES} ${PERF_SOURCES}) if(IOS) add_library(${SD_LIBRARY_NAME} STATIC $) @@ -373,7 +373,11 @@ elseif(SD_CPU) foreach (_variableName ${_variableNames}) message(STATUS "${_variableName}=${${_variableName}}") endforeach() - target_link_libraries(${SD_LIBRARY_NAME} ${MKLDNN} ${MKLDNN_LIBRARIES} ${ARMCOMPUTE_LIBRARIES} ${OPENBLAS_LIBRARIES} ${BLAS_LIBRARIES} ${CPU_FEATURES}) + + #This breaks the build. Normally you want to run tests anyways. + if(NOT "$ENV{CLION_IDE}") + target_link_libraries(${SD_LIBRARY_NAME} ${MKLDNN} ${MKLDNN_LIBRARIES} ${ARMCOMPUTE_LIBRARIES} ${OPENBLAS_LIBRARIES} ${BLAS_LIBRARIES} ${CPU_FEATURES}) + endif() if ("${SD_ALL_OPS}" AND "${SD_BUILD_MINIFIER}") message(STATUS "Building minifier...") @@ -382,7 +386,7 @@ elseif(SD_CPU) endif() if ("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU" AND "${CMAKE_CXX_COMPILER_VERSION}" VERSION_LESS 4.9) - message(FATAL_ERROR "You need at least GCC 4.9") + message(FATAL_ERROR "You need at least GCC 4.9") endif() # OpenMP works well pretty much only with GCC diff --git a/libnd4j/include/ops/declarable/generic/nn/fusedBatchNorm.cpp b/libnd4j/include/ops/declarable/generic/nn/fusedBatchNorm.cpp index 0ce6f3ec4..926ba49ea 100644 --- a/libnd4j/include/ops/declarable/generic/nn/fusedBatchNorm.cpp +++ b/libnd4j/include/ops/declarable/generic/nn/fusedBatchNorm.cpp @@ -26,132 +26,138 @@ #include namespace sd { -namespace ops { + namespace ops { - DECLARE_TYPES(fused_batch_norm) { - getOpDescriptor() - ->setAllowedInputTypes(sd::DataType::ANY) - ->setAllowedOutputTypes({ALL_FLOATS}); + DECLARE_TYPES(fused_batch_norm) { + getOpDescriptor() + ->setAllowedInputTypes(sd::DataType::ANY) + ->setAllowedOutputTypes({ALL_FLOATS}); + } + + CUSTOM_OP_IMPL(fused_batch_norm, 3, 3, false, 0, 2) { + auto x = INPUT_VARIABLE(0); // [bS,iH,iW,iD] (NHWC) or [bS,iD,iH,iW] (NCHW) + auto scale = INPUT_VARIABLE(1); // [iD] + auto offset = INPUT_VARIABLE(2); // [iD] + + auto y = OUTPUT_VARIABLE(0); // [bS,iH,iW,iD] (NHWC) or [bS,iD,iH,iW] (NCHW) + auto batchMean = OUTPUT_VARIABLE(1); // [iD] + auto batchVar = OUTPUT_VARIABLE(2); // [iD] + + const bool dataFormat = (bool)INT_ARG(0); // 0->NHWC, 1->NCHW + const bool isTraining = (bool)INT_ARG(1); + + REQUIRE_TRUE(x->rankOf() == 4, 0, "CUSTOM_OP fused_batch_norm: the rank of input x array must be equal to 4, but got %i instead !", x->rankOf()); + + int bS = x->sizeAt(0); // batch size + int iH, iW, iD; // input height, input width, input depth(number of channels) + if(dataFormat) { + iD = x->sizeAt(1); + iH = x->sizeAt(2); + iW = x->sizeAt(3); + } + else { + iD = x->sizeAt(3); + iH = x->sizeAt(1); + iW = x->sizeAt(2); + } + + auto xCast = x->cast(sd::DataType::FLOAT32); + + + REQUIRE_TRUE(scale->rankOf() == 1 && scale->sizeAt(0) == iD, 0, "CUSTOM_OP fused_batch_norm: wrong shape of input scale array, expected is [%i], but got %s instead", iD, ShapeUtils::shapeAsString(scale).c_str()); + REQUIRE_TRUE(offset->rankOf() == 1 && offset->sizeAt(0) == iD, 0, "CUSTOM_OP fused_batch_norm: wrong shape of input offset array, expected is [%i], but got %s instead", iD, ShapeUtils::shapeAsString(offset).c_str()); + + NDArray *mean(nullptr), *variance(nullptr); + if(!isTraining) { + mean = INPUT_VARIABLE(3); + variance = INPUT_VARIABLE(4); + REQUIRE_TRUE(mean->rankOf() == 1 && mean->sizeAt(0) == iD, 0, "CUSTOM_OP fused_batch_norm: wrong shape of input mean array, expected is [%i], but got %s instead", iD, ShapeUtils::shapeAsString(mean).c_str()); + REQUIRE_TRUE(variance->rankOf() == 1 && variance->sizeAt(0) == iD, 0, "CUSTOM_OP fused_batch_norm: wrong shape of input variance array, expected is [%i], but got %s instead", iD, ShapeUtils::shapeAsString(variance).c_str()); + } + else { + //REQUIRE_TRUE(block.width() == 3, 0, "CUSTOM_OP fused_batch_norm: when isTraining=true then number of input arrays must be equal to 3, but got %i instead !", block.width()); + std::vector shape = {iD}; + mean = NDArrayFactory::create_(scale->ordering(), shape, sd::DataType::FLOAT32, block.launchContext()); + variance = NDArrayFactory::create_(scale->ordering(), shape, sd::DataType::FLOAT32, block.launchContext()); + } + + + float epsilon; + if(block.getTArguments()->size() > 0) { + epsilon = (float) (T_ARG(0) > 1.001e-5 ? T_ARG(0) : 1.001e-5); + } + else { + epsilon = 0.001f; + } + + const int restSize = x->lengthOf() / iD; + + auto xAffected = NDArrayFactory::create(x->ordering(), {restSize, iD}, sd::DataType::FLOAT32, block.launchContext()); + xAffected.assign(xCast); + + const int restSizeMinusOne = (restSize > 1) ? (restSize - 1) : 1; + const float restSizeInv = 1.0f / restSize; + const float restSizeAdjust = (float)restSize / restSizeMinusOne; + + if(isTraining) { + auto sum = xAffected.reduceAlongDimension(reduce::Sum, {0}); + sum *= restSizeInv; + mean->assign(sum); + *batchMean = *mean; + } + else + *batchMean = 0.; + + auto xCentered = xAffected - *mean; + xAffected -= *mean; + + if(isTraining) { + int power = 2; + xAffected.applyScalar(scalar::Pow, power, xAffected); + auto sum = xAffected.reduceAlongDimension(reduce::Sum, {0}); + sum *= restSizeInv; + variance->assign(sum); + auto varOutput = (*variance) * restSizeAdjust; + batchVar->assign(varOutput); + } + else + *batchVar = 0.; + + auto scaledVariance = ((*variance + epsilon).transform(transform::RSqrt) * (*scale)).cast(xAffected.dataType()); + auto xScaled1 = xCentered * scaledVariance; + auto xShifted1 = xScaled1 + *offset; + + y->assign(xShifted1); + + if(isTraining) { + delete mean; + delete variance; + } + + return Status::OK(); + } + + + + DECLARE_SHAPE_FN(fused_batch_norm) { + auto xShapeInfo = inputShape->at(0); + auto scaleShapeInfo = inputShape->at(1); + + const bool dataFormat = (bool)INT_ARG(0); // 0->NHWC, 1->NCHW + const int iD = dataFormat ? xShapeInfo[2] : xShapeInfo[4]; + + REQUIRE_TRUE(scaleShapeInfo[0] == 1 && scaleShapeInfo[1] == iD, 0, "CUSTOM_OP fused_batch_norm: wrong shape of input scale array, expected is [%i], but got %s instead", iD, ShapeUtils::shapeAsString(scaleShapeInfo).c_str()); + + Nd4jLong* outShapeInfo(nullptr), *batchMeanShapeInfo(nullptr), *batchVarShapeInfo(nullptr); + + COPY_SHAPE(xShapeInfo, outShapeInfo); + COPY_SHAPE(scaleShapeInfo, batchMeanShapeInfo); + COPY_SHAPE(scaleShapeInfo, batchVarShapeInfo); + + return SHAPELIST(CONSTANT(outShapeInfo), CONSTANT(batchMeanShapeInfo), CONSTANT(batchVarShapeInfo)); + } + } - -CUSTOM_OP_IMPL(fused_batch_norm, 3, 3, false, 0, 2) { - auto x = INPUT_VARIABLE(0); // [bS,iH,iW,iD] (NHWC) or [bS,iD,iH,iW] (NCHW) - auto scale = INPUT_VARIABLE(1); // [iD] - auto offset = INPUT_VARIABLE(2); // [iD] - - auto y = OUTPUT_VARIABLE(0); // [bS,iH,iW,iD] (NHWC) or [bS,iD,iH,iW] (NCHW) - auto batchMean = OUTPUT_VARIABLE(1); // [iD] - auto batchVar = OUTPUT_VARIABLE(2); // [iD] - - const bool dataFormat = (bool)INT_ARG(0); // 0->NHWC, 1->NCHW - const bool isTraining = (bool)INT_ARG(1); - - REQUIRE_TRUE(x->rankOf() == 4, 0, "CUSTOM_OP fused_batch_norm: the rank of input x array must be equal to 4, but got %i instead !", x->rankOf()); - - int bS = x->sizeAt(0); // batch size - int iH, iW, iD; // input height, input width, input depth(number of channels) - if(dataFormat) { - iD = x->sizeAt(1); - iH = x->sizeAt(2); - iW = x->sizeAt(3); - } - else { - iD = x->sizeAt(3); - iH = x->sizeAt(1); - iW = x->sizeAt(2); - } - - REQUIRE_TRUE(scale->rankOf() == 1 && scale->sizeAt(0) == iD, 0, "CUSTOM_OP fused_batch_norm: wrong shape of input scale array, expected is [%i], but got %s instead", iD, ShapeUtils::shapeAsString(scale).c_str()); - REQUIRE_TRUE(offset->rankOf() == 1 && offset->sizeAt(0) == iD, 0, "CUSTOM_OP fused_batch_norm: wrong shape of input offset array, expected is [%i], but got %s instead", iD, ShapeUtils::shapeAsString(offset).c_str()); - - NDArray *mean(nullptr), *variance(nullptr); - if(!isTraining){ - mean = INPUT_VARIABLE(3); - variance = INPUT_VARIABLE(4); - REQUIRE_TRUE(mean->rankOf() == 1 && mean->sizeAt(0) == iD, 0, "CUSTOM_OP fused_batch_norm: wrong shape of input mean array, expected is [%i], but got %s instead", iD, ShapeUtils::shapeAsString(mean).c_str()); - REQUIRE_TRUE(variance->rankOf() == 1 && variance->sizeAt(0) == iD, 0, "CUSTOM_OP fused_batch_norm: wrong shape of input variance array, expected is [%i], but got %s instead", iD, ShapeUtils::shapeAsString(variance).c_str()); - } - else { - //REQUIRE_TRUE(block.width() == 3, 0, "CUSTOM_OP fused_batch_norm: when isTraining=true then number of input arrays must be equal to 3, but got %i instead !", block.width()); - std::vector shape = {iD}; - mean = NDArrayFactory::create_(scale->ordering(), shape, scale->dataType(), block.launchContext()); - variance = NDArrayFactory::create_(scale->ordering(), shape, scale->dataType(), block.launchContext()); - } - - // FIXME: double? - double epsilon; - if(block.getTArguments()->size() > 0) - epsilon = T_ARG(0) > 1.001e-5 ? T_ARG(0) : 1.001e-5; - else - epsilon = 0.001; - - const int restSize = x->lengthOf() / iD; - auto xAffected = NDArrayFactory::create(x->ordering(), {restSize, iD}, mean->dataType(), block.launchContext()); - xAffected.assign(x); - - const int restSizeMinusOne = (restSize > 1) ? (restSize - 1) : 1; - // FIXME: float? - const double restSizeInv = 1.0 / restSize; - const double restSizeAdjust = (double)restSize / restSizeMinusOne; - - if(isTraining) { - auto sum = xAffected.reduceAlongDimension(reduce::Sum, {0}); - sum *= restSizeInv; - mean->assign(sum); - *batchMean = *mean; - //delete sum; - } - else - *batchMean = 0.; - - xAffected -= *mean; - - if(isTraining) { - int power = 2; - xAffected.applyScalar(scalar::Pow, power, xAffected); - auto sum = xAffected.reduceAlongDimension(reduce::Sum, {0}); - sum *= restSizeInv; - variance->assign(sum); - *batchVar = (*variance) * restSizeAdjust; - //delete sum; - } - else - *batchVar = 0.; - xAffected *= (*variance + epsilon).transform(transform::RSqrt) * (*scale) + (*offset); - y->assign( xAffected ); - - if(isTraining) { - delete mean; - delete variance; - } - - return Status::OK(); -} - - - -DECLARE_SHAPE_FN(fused_batch_norm) { - auto xShapeInfo = inputShape->at(0); - auto scaleShapeInfo = inputShape->at(1); - - const bool dataFormat = (bool)INT_ARG(0); // 0->NHWC, 1->NCHW - const int iD = dataFormat ? xShapeInfo[2] : xShapeInfo[4]; - - REQUIRE_TRUE(scaleShapeInfo[0] == 1 && scaleShapeInfo[1] == iD, 0, "CUSTOM_OP fused_batch_norm: wrong shape of input scale array, expected is [%i], but got %s instead", iD, ShapeUtils::shapeAsString(scaleShapeInfo).c_str()); - - Nd4jLong* outShapeInfo(nullptr), *batchMeanShapeInfo(nullptr), *batchVarShapeInfo(nullptr); - - COPY_SHAPE(xShapeInfo, outShapeInfo); - COPY_SHAPE(scaleShapeInfo, batchMeanShapeInfo); - COPY_SHAPE(scaleShapeInfo, batchVarShapeInfo); - - return SHAPELIST(CONSTANT(outShapeInfo), CONSTANT(batchMeanShapeInfo), CONSTANT(batchVarShapeInfo)); -} - - - - -} } #endif \ No newline at end of file diff --git a/nd4j/nd4j-backends/nd4j-api-parent/nd4j-api/src/main/java/org/nd4j/linalg/api/ops/custom/FusedBatchNorm.java b/nd4j/nd4j-backends/nd4j-api-parent/nd4j-api/src/main/java/org/nd4j/linalg/api/ops/custom/FusedBatchNorm.java index 335b8d3c0..751f37ede 100644 --- a/nd4j/nd4j-backends/nd4j-api-parent/nd4j-api/src/main/java/org/nd4j/linalg/api/ops/custom/FusedBatchNorm.java +++ b/nd4j/nd4j-backends/nd4j-api-parent/nd4j-api/src/main/java/org/nd4j/linalg/api/ops/custom/FusedBatchNorm.java @@ -87,9 +87,12 @@ public class FusedBatchNorm extends DynamicCustomOp { } @Override - public List calculateOutputDataTypes(List inputDataTypes){ + public List calculateOutputDataTypes(List inputDataTypes) { int n = args().length; Preconditions.checkState(inputDataTypes != null && inputDataTypes.size() == n, "Expected %s input data types for %s, got %s", n, getClass(), inputDataTypes); + if(!dArguments.isEmpty()) { + return Arrays.asList(dArguments.get(0),dArguments.get(0),dArguments.get(0)); + } return Arrays.asList(outputDataType == null ? DataType.FLOAT : outputDataType, outputDataType == null ? DataType.FLOAT : outputDataType, outputDataType == null ? DataType.FLOAT : outputDataType); diff --git a/nd4j/nd4j-backends/nd4j-tests/src/test/java/org/nd4j/imports/TFGraphs/TFGraphTestAllSameDiff.java b/nd4j/nd4j-backends/nd4j-tests/src/test/java/org/nd4j/imports/TFGraphs/TFGraphTestAllSameDiff.java index 6c097f0b7..8b8dc90d7 100644 --- a/nd4j/nd4j-backends/nd4j-tests/src/test/java/org/nd4j/imports/TFGraphs/TFGraphTestAllSameDiff.java +++ b/nd4j/nd4j-backends/nd4j-tests/src/test/java/org/nd4j/imports/TFGraphs/TFGraphTestAllSameDiff.java @@ -69,10 +69,8 @@ public class TFGraphTestAllSameDiff { //Note: Can't extend BaseNd4jTest here a * the status of the test failing. No tests will run. */ public final static List EXECUTE_ONLY_MODELS = Arrays.asList( - "max_pool_with_argmax/int32_int64_padding_SAME", - // "fused_batch_norm/float32_nhwc", - "max_pool_with_argmax/int64_int64_padding_SAME" - // "fused_batch_norm/float16_nhwc", + "fused_batch_norm/float32_nhwc" + // , "fused_batch_norm/float16_nhwc" ); @@ -86,9 +84,6 @@ public class TFGraphTestAllSameDiff { //Note: Can't extend BaseNd4jTest here a // Still failing 2020/04/27 java.lang.IllegalStateException: Could not find class for TF Ops: TruncateMod "truncatemod/.*", - //Still failing as of 2019/09/11 - https://github.com/deeplearning4j/deeplearning4j/issues/6464 - not sure if related to: https://github.com/deeplearning4j/deeplearning4j/issues/6447 - "cnn2d_nn/nhwc_b1_k12_s12_d12_SAME", - //2019/09/11 - No tensorflow op found for SparseTensorDenseAdd // 2020/04/27 java.lang.IllegalStateException: Could not find class for TF Ops: SparseTensorDenseAdd "confusion/.*", diff --git a/nd4j/samediff-import/samediff-import-tensorflow/src/main/kotlin/org/nd4j/samediff/frameworkimport/tensorflow/definitions/TensorflowOpDeclarations.kt b/nd4j/samediff-import/samediff-import-tensorflow/src/main/kotlin/org/nd4j/samediff/frameworkimport/tensorflow/definitions/TensorflowOpDeclarations.kt index 9333759e6..5b9692017 100644 --- a/nd4j/samediff-import/samediff-import-tensorflow/src/main/kotlin/org/nd4j/samediff/frameworkimport/tensorflow/definitions/TensorflowOpDeclarations.kt +++ b/nd4j/samediff-import/samediff-import-tensorflow/src/main/kotlin/org/nd4j/samediff/frameworkimport/tensorflow/definitions/TensorflowOpDeclarations.kt @@ -958,7 +958,7 @@ val fusedBatchnormV1 = TensorflowMappingProcess( "offset" to "offset","mean" to "mean","variance" to "variance"))), inputFrameworkOpName = "FusedBatchNorm", opMappingRegistry = tensorflowOpRegistry, - attributeMappingRules = listOf(valueMapping(mutableMapOf("epsilon" to "epsilon")), + attributeMappingRules = listOf(valueMapping(mutableMapOf("epsilon" to "epsilon","dtype" to "T")), invertBooleanNumber(mutableMapOf("isTraining" to "is_training")), stringEqualsRule(outputAttribute = "dataFormat",inputFrameworkAttributeName = "data_format",valueToTest = "NCHW",argumentIndex = 0)) ) @@ -971,7 +971,7 @@ val fusedBatchnormV2 = TensorflowMappingProcess( "offset" to "offset","mean" to "mean","variance" to "variance"))), inputFrameworkOpName = "FusedBatchNormV2", opMappingRegistry = tensorflowOpRegistry, - attributeMappingRules = listOf(valueMapping(mutableMapOf("epsilon" to "epsilon")), + attributeMappingRules = listOf(valueMapping(mutableMapOf("epsilon" to "epsilon","dtype" to "T")), invertBooleanNumber(mutableMapOf("isTraining" to "is_training")), stringEqualsRule(outputAttribute = "dataFormat",inputFrameworkAttributeName = "data_format",valueToTest = "NCHW",argumentIndex = 0)) ) @@ -983,7 +983,7 @@ val fusedBatchnormV3 = TensorflowMappingProcess( "offset" to "offset","mean" to "mean","variance" to "variance"))), inputFrameworkOpName = "FusedBatchNormV3", opMappingRegistry = tensorflowOpRegistry, - attributeMappingRules = listOf(valueMapping(mutableMapOf("epsilon" to "epsilon")), + attributeMappingRules = listOf(valueMapping(mutableMapOf("epsilon" to "epsilon","dtype" to "T")), invertBooleanNumber(mutableMapOf("isTraining" to "is_training")), stringEqualsRule(outputAttribute = "dataFormat",inputFrameworkAttributeName = "data_format",valueToTest = "NCHW",argumentIndex = 0)) ) diff --git a/nd4j/samediff-import/samediff-import-tensorflow/src/main/resources/tensorflow-mapping-ruleset.pbtxt b/nd4j/samediff-import/samediff-import-tensorflow/src/main/resources/tensorflow-mapping-ruleset.pbtxt index b8cc5ef02..4089294ca 100644 --- a/nd4j/samediff-import/samediff-import-tensorflow/src/main/resources/tensorflow-mapping-ruleset.pbtxt +++ b/nd4j/samediff-import/samediff-import-tensorflow/src/main/resources/tensorflow-mapping-ruleset.pbtxt @@ -8367,10 +8367,16 @@ mappings { functionName: "valuemapping" inputFloatName: "epsilon" outputDoubleName: "epsilon" + inputDataTypeName: "T" + outputDataTypeName: "dtype" inputToOutput { key: "epsilon" value: "epsilon" } + inputToOutput { + key: "dtype" + value: "T" + } ruleType: "attribute" inputFrameworkOpName: "FusedBatchNorm" } @@ -12480,10 +12486,16 @@ mappings { functionName: "valuemapping" inputFloatName: "epsilon" outputDoubleName: "epsilon" + inputDataTypeName: "T" + outputDataTypeName: "dtype" inputToOutput { key: "epsilon" value: "epsilon" } + inputToOutput { + key: "dtype" + value: "T" + } ruleType: "attribute" inputFrameworkOpName: "FusedBatchNormV3" } @@ -13056,10 +13068,16 @@ mappings { functionName: "valuemapping" inputFloatName: "epsilon" outputDoubleName: "epsilon" + inputDataTypeName: "T" + outputDataTypeName: "dtype" inputToOutput { key: "epsilon" value: "epsilon" } + inputToOutput { + key: "dtype" + value: "T" + } ruleType: "attribute" inputFrameworkOpName: "FusedBatchNormV2" } diff --git a/nd4j/samediff-import/samediff-import-tensorflow/src/test/kotlin/org/nd4j/samediff/frameworkimport/tensorflow/TestTensorflowIR.kt b/nd4j/samediff-import/samediff-import-tensorflow/src/test/kotlin/org/nd4j/samediff/frameworkimport/tensorflow/TestTensorflowIR.kt index 85e47ec6d..6931c20ca 100644 --- a/nd4j/samediff-import/samediff-import-tensorflow/src/test/kotlin/org/nd4j/samediff/frameworkimport/tensorflow/TestTensorflowIR.kt +++ b/nd4j/samediff-import/samediff-import-tensorflow/src/test/kotlin/org/nd4j/samediff/frameworkimport/tensorflow/TestTensorflowIR.kt @@ -90,7 +90,9 @@ class TestTensorflowIR { //val inputMap = mapOf("image" to Nd4j.ones(1,128,128,4)) val inputMap = emptyMap() val tensorflowIRGraph = TensorflowIRGraph(textGraph,tensorflowOps,tfImporter.registry) - val outputList = tensorflowIRGraph.nodeList().map { input -> input.nodeName() }.toSet() + val outputList = tensorflowIRGraph.nodeList().map { input -> input.nodeName() }.toMutableSet() + outputList.add("FusedBatchNormV3:1") + outputList.add("FusedBatchNormV3:2") val tfGraphRunner = TensorflowIRGraphRunner(tensorflowIRGraph, inputMap.keys.toList(), outputList.toList()) val importedGraph = TFGraphMapper.importGraph(textGraph) val graph = tfImporter.importFromGraph(textGraph,inputMap) @@ -104,7 +106,7 @@ class TestTensorflowIR { val names = tensorflowIRGraph.nodeList().map { input -> input.nodeName() } val skipValidation = setOf("parallel_stack/ExpandDims/dim") //assertEquals(output.keys,output2.keys) - val notEquals = HashSet() + /* val notEquals = HashSet() names.forEach { val value = output[it] val value2 = output2[it] @@ -115,9 +117,9 @@ class TestTensorflowIR { val newVar = graph.variables[it] notEquals.add(it) } - } + }*/ - println(notEquals) + //println(notEquals) // assertEquals(output,output2) //assertEquals(tfOutput,output) diff --git a/nd4j/samediff-import/samediff-import-tensorflow/tensorflow-processes.pbtxt b/nd4j/samediff-import/samediff-import-tensorflow/tensorflow-processes.pbtxt index b8cc5ef02..4089294ca 100644 --- a/nd4j/samediff-import/samediff-import-tensorflow/tensorflow-processes.pbtxt +++ b/nd4j/samediff-import/samediff-import-tensorflow/tensorflow-processes.pbtxt @@ -8367,10 +8367,16 @@ mappings { functionName: "valuemapping" inputFloatName: "epsilon" outputDoubleName: "epsilon" + inputDataTypeName: "T" + outputDataTypeName: "dtype" inputToOutput { key: "epsilon" value: "epsilon" } + inputToOutput { + key: "dtype" + value: "T" + } ruleType: "attribute" inputFrameworkOpName: "FusedBatchNorm" } @@ -12480,10 +12486,16 @@ mappings { functionName: "valuemapping" inputFloatName: "epsilon" outputDoubleName: "epsilon" + inputDataTypeName: "T" + outputDataTypeName: "dtype" inputToOutput { key: "epsilon" value: "epsilon" } + inputToOutput { + key: "dtype" + value: "T" + } ruleType: "attribute" inputFrameworkOpName: "FusedBatchNormV3" } @@ -13056,10 +13068,16 @@ mappings { functionName: "valuemapping" inputFloatName: "epsilon" outputDoubleName: "epsilon" + inputDataTypeName: "T" + outputDataTypeName: "dtype" inputToOutput { key: "epsilon" value: "epsilon" } + inputToOutput { + key: "dtype" + value: "T" + } ruleType: "attribute" inputFrameworkOpName: "FusedBatchNormV2" }