saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-05-14 20:48:00 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

update mkldnn to 0.17.1 and address assumptions related to tensor padding that come with new mkldnn version. (#79)

Browse source
This commit is contained in:
jywu-msft 2018-12-03 17:07:44 -08:00 committed by GitHub
parent 900e69ceae
commit 47a6992e1b
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
6 changed files with 78 additions and 162 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

11
cmake/external/mkldnn.cmake vendored
View file

@ -1,17 +1,14 @@
include (ExternalProject)
set(MKLDNN_URL https://github.com/intel/mkl-dnn.git)
# If MKLDNN_TAG is updated, check if platform.cmake.patch or mkldnn_sgemm.patch needs to be updated.
set(MKLDNN_TAG v0.15)
# If MKLDNN_TAG is updated, check if platform.cmake.patch needs to be updated.
set(MKLDNN_TAG v0.17.1)
set(MKLDNN_SOURCE ${CMAKE_CURRENT_BINARY_DIR}/mkl-dnn/src/mkl-dnn/src)
set(MKLDNN_INSTALL ${CMAKE_CURRENT_BINARY_DIR}/mkl-dnn/install)
set(MKLDNN_LIB_DIR ${MKLDNN_INSTALL}/lib)
set(MKLDNN_INCLUDE_DIR ${MKLDNN_INSTALL}/include)
# patch for mkldnn_sgemm thread safety bug.
# it can be removed once a fix is available in a validated mkldnn release version.
set(MKLDNN_PATCH_COMMAND1 git apply ${CMAKE_SOURCE_DIR}/patches/mkldnn/mkldnn_sgemm.patch)
set(MKLDNN_PATCH_COMMAND2 git apply ${CMAKE_SOURCE_DIR}/patches/mkldnn/platform.cmake.patch)
set(MKLDNN_PATCH_COMMAND1 git apply ${CMAKE_SOURCE_DIR}/patches/mkldnn/platform.cmake.patch)
# discard prior changes due to patching in mkldnn source to unblock incremental builds.
set(MKLDNN_PATCH_DISCARD_COMMAND cd ${MKLDNN_SOURCE} && git checkout -- .)
@ -44,7 +41,7 @@ ExternalProject_Add(project_mkldnn
PREFIX mkl-dnn
GIT_REPOSITORY ${MKLDNN_URL}
GIT_TAG ${MKLDNN_TAG}
PATCH_COMMAND ${DOWNLOAD_MKLML} COMMAND ${MKLDNN_PATCH_DISCARD_COMMAND} COMMAND ${MKLDNN_PATCH_COMMAND1} COMMAND ${MKLDNN_PATCH_COMMAND2}
PATCH_COMMAND ${DOWNLOAD_MKLML} COMMAND ${MKLDNN_PATCH_DISCARD_COMMAND} COMMAND ${MKLDNN_PATCH_COMMAND1}
SOURCE_DIR ${MKLDNN_SOURCE}
CMAKE_ARGS -DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE} -DCMAKE_INSTALL_PREFIX=${MKLDNN_INSTALL}
)

136
cmake/patches/mkldnn/mkldnn_sgemm.patch
View file

@ -1,136 +0,0 @@
diff --git a/src/cpu/gemm/jit_avx2_gemm_f32.cpp b/src/cpu/gemm/jit_avx2_gemm_f32.cpp
index bf03c57..46793e7 100644
--- a/src/cpu/gemm/jit_avx2_gemm_f32.cpp
+++ b/src/cpu/gemm/jit_avx2_gemm_f32.cpp
@@ -2349,13 +2349,18 @@ void jit_avx2_gemm_f32::sgemm(const char *transa, const char *transb,
nthr_mn = nthr_m * nthr_n;
- unsigned int volatile *ompstatus = (unsigned int volatile *)ompstatus_;
- if (!ompstatus) return;
+ unsigned int *ompstatus_ = nullptr;
+ unsigned int volatile *ompstatus = nullptr;
float *c_buffers = NULL;
float *ws_buffers = NULL;
if (nthr_k > 1) {
+ ompstatus_ = (unsigned int *)malloc(
+ sizeof(unsigned int *) * nthrs_ * CACHE_LINE_SIZE, 64);
+ ompstatus = (unsigned int volatile *)ompstatus_;
+ assert(ompstatus);
+
for (int i = 0; i < nthr; i++)
ompstatus[i * CACHE_LINE_SIZE] = 0;
@@ -2486,8 +2491,10 @@ void jit_avx2_gemm_f32::sgemm(const char *transa, const char *transb,
}
}
- if (nthr_k > 1)
+ if (nthr_k > 1) {
free(c_buffers);
+ free(ompstatus_);
+ }
free(ws_buffers);
}
@@ -2513,9 +2520,6 @@ jit_avx2_gemm_f32::jit_avx2_gemm_f32(
ker_b0_ = ker_bn_;
}
nthrs_ = omp_get_max_threads();
- ompstatus_ = (unsigned int *)malloc(
- sizeof(unsigned int *) * nthrs_ * CACHE_LINE_SIZE, 64);
- assert(ompstatus_);
}
jit_avx2_gemm_f32::~jit_avx2_gemm_f32()
@@ -2525,7 +2529,6 @@ jit_avx2_gemm_f32::~jit_avx2_gemm_f32()
delete ker_b1_;
if (beta_ != 0.0 || (beta_ == 0.0 && hasBias_))
delete ker_b0_;
- free(ompstatus_);
}
}
diff --git a/src/cpu/gemm/jit_avx2_gemm_f32.hpp b/src/cpu/gemm/jit_avx2_gemm_f32.hpp
index 7adb2a2..ebbbde0 100644
--- a/src/cpu/gemm/jit_avx2_gemm_f32.hpp
+++ b/src/cpu/gemm/jit_avx2_gemm_f32.hpp
@@ -49,7 +49,6 @@ private:
bool hasBias_;
struct xbyak_gemm;
xbyak_gemm *ker_bn_, *ker_b1_, *ker_b0_;
- unsigned int *ompstatus_;
int nthrs_;
};
}
diff --git a/src/cpu/gemm/jit_avx512_common_gemm_f32.cpp b/src/cpu/gemm/jit_avx512_common_gemm_f32.cpp
index 7959195..fca14f4 100644
--- a/src/cpu/gemm/jit_avx512_common_gemm_f32.cpp
+++ b/src/cpu/gemm/jit_avx512_common_gemm_f32.cpp
@@ -1866,14 +1866,18 @@ void jit_avx512_common_gemm_f32::sgemm(const char *transa, const char *transb,
nthr = nthr_m * nthr_n * nthr_k;
nthr_mn = nthr_m * nthr_n;
-
- unsigned int volatile *ompstatus = (unsigned int volatile *)ompstatus_;
- if (!ompstatus) return;
+
+ unsigned int *ompstatus_ = nullptr;
+ unsigned int volatile *ompstatus = nullptr;
float *c_buffers = NULL;
float *ws_buffers = NULL;
if (nthr_k > 1) {
+ ompstatus_ = (unsigned int *)malloc(
+ sizeof(unsigned int *) * nthrs_ * CACHE_LINE_SIZE, 64);
+ ompstatus = (unsigned int volatile *)ompstatus_;
+ assert(ompstatus);
for (int i = 0; i < nthr; i++)
ompstatus[i * CACHE_LINE_SIZE] = 0;
@@ -2004,8 +2008,10 @@ void jit_avx512_common_gemm_f32::sgemm(const char *transa, const char *transb,
}
}
- if (nthr_k > 1)
+ if (nthr_k > 1) {
free(c_buffers);
+ free(ompstatus_);
+ }
free(ws_buffers);
}
@@ -2032,10 +2038,6 @@ jit_avx512_common_gemm_f32::jit_avx512_common_gemm_f32(
}
nthrs_ = omp_get_max_threads();
- ompstatus_ = (unsigned int *)malloc(
- sizeof(unsigned int *) * nthrs_ * CACHE_LINE_SIZE, 64);
- assert(ompstatus_);
-
}
jit_avx512_common_gemm_f32::~jit_avx512_common_gemm_f32()
@@ -2045,7 +2047,6 @@ jit_avx512_common_gemm_f32::~jit_avx512_common_gemm_f32()
delete ker_b1_;
if (beta_ != 0.0 || (beta_ == 0.0 && hasBias_))
delete ker_b0_;
- free(ompstatus_);
}
}
}
diff --git a/src/cpu/gemm/jit_avx512_common_gemm_f32.hpp b/src/cpu/gemm/jit_avx512_common_gemm_f32.hpp
index ede1cf9..c057335 100644
--- a/src/cpu/gemm/jit_avx512_common_gemm_f32.hpp
+++ b/src/cpu/gemm/jit_avx512_common_gemm_f32.hpp
@@ -49,7 +49,6 @@ private:
bool hasBias_;
struct xbyak_gemm;
xbyak_gemm *ker_bn_, *ker_b1_, *ker_b0_;
- unsigned int *ompstatus_;
int nthrs_;
};
}

14
cmake/patches/mkldnn/platform.cmake.patch
View file

@ -1,14 +1,14 @@
diff --git a/cmake/platform.cmake b/cmake/platform.cmake
index fa51aa7..3d24fdc 100644
index 3597970a..805ce63e 100644
--- a/cmake/platform.cmake
+++ b/cmake/platform.cmake
@@ -64,9 +64,6 @@ elseif(UNIX OR APPLE OR MINGW)
# unconditionnaly.
set(CMAKE_CCXX_FLAGS "${CMAKE_CCXX_FLAGS} -Wno-pass-failed")
@@ -107,9 +107,6 @@ elseif(UNIX OR MINGW)
append(CMAKE_CCXX_SANITIZER_FLAGS "-g -fno-omit-frame-pointer")
endif()
elseif("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU")
- if(NOT CMAKE_CXX_COMPILER_VERSION VERSION_LESS 5.0)
- set(DEF_ARCH_OPT_FLAGS "-march=native -mtune=native")
- endif()
if(CMAKE_CXX_COMPILER_VERSION VERSION_LESS 6.0)
# suppress warning on assumptions made regarding overflow (#146)
set(CMAKE_CCXX_FLAGS "${CMAKE_CCXX_FLAGS} -Wno-strict-overflow")
# suppress warning on assumptions made regarding overflow (#146)
append(CMAKE_CCXX_NOWARN_FLAGS "-Wno-strict-overflow")
elseif(CMAKE_CXX_COMPILER_ID STREQUAL "Intel")

38
onnxruntime/core/providers/mkldnn/nn/conv.cc
View file

@ -96,6 +96,12 @@ class Conv2dPrimitive : public PrimitiveBase {
mkldnn::memory::format GetDstMemoryFormat() const { return context_.dst_fmt; }
size_t GetSrcSize() const { return context_.src_size; }
size_t GetFilterSize() const { return context_.filter_size; }
size_t GetDstSize() const { return context_.dst_size; }
mkldnn::convolution_forward::primitive_desc* GetPrimitiveDesc() const {
return context_.conv_fwd_pd.get();
}
@ -106,6 +112,10 @@ class Conv2dPrimitive : public PrimitiveBase {
mkldnn::memory::format filter_fmt;
mkldnn::memory::format dst_fmt;
size_t src_size;
size_t filter_size;
size_t dst_size;
std::unique_ptr<mkldnn::memory> src_mem;
std::unique_ptr<mkldnn::memory> filter_mem;
std::unique_ptr<mkldnn::memory> bias_mem;
@ -128,6 +138,9 @@ class Conv2dPrimitive : public PrimitiveBase {
: src_fmt(mkldnn::memory::format::any),
filter_fmt(mkldnn::memory::format::any),
dst_fmt(mkldnn::memory::format::any),
src_size(0),
filter_size(0),
dst_size(0),
src_mem(nullptr),
filter_mem(nullptr),
bias_mem(nullptr),
@ -180,6 +193,12 @@ class Conv2dPrimitive : public PrimitiveBase {
context_.dst_fmt = static_cast<mkldnn::memory::format>(
context_.conv_fwd_pd.get()->dst_primitive_desc().desc().data.format);
context_.src_size = context_.conv_fwd_pd.get()->src_primitive_desc().get_size();
context_.filter_size = context_.conv_fwd_pd.get()->weights_primitive_desc().get_size();
context_.dst_size = context_.conv_fwd_pd.get()->dst_primitive_desc().get_size();
context_.src_mem.reset(
new mkldnn::memory(context_.conv_fwd_pd.get()->src_primitive_desc(), nullptr));
context_.filter_mem.reset(
@ -188,10 +207,9 @@ class Conv2dPrimitive : public PrimitiveBase {
new mkldnn::memory(context_.conv_fwd_pd.get()->dst_primitive_desc(), nullptr));
if (!params.bias_dims.empty()) {
context_.bias_mem.reset(new mkldnn::memory(
{{{params.bias_dims}, MklDnnType<T>(), mkldnn::memory::format::x},
cpu_engine_},
nullptr));
context_.bias_mem.reset(
new mkldnn::memory(context_.conv_fwd_pd.get()->bias_primitive_desc(), nullptr));
context_.conv_fwd.reset(new mkldnn::convolution_forward(
*context_.conv_fwd_pd, *context_.src_mem, *context_.filter_mem,
*context_.bias_mem, *context_.dst_mem));
@ -350,7 +368,9 @@ Status Conv<T>::Compute(OpKernelContext* context) const {
if (src_md.data.format != conv2d_primitive->GetSrcMemoryFormat()) {
auto pd = mkldnn::memory::primitive_desc(src_md, cpu_engine);
mkldnn::memory src = mkldnn::memory(pd, (void*)src_data);
src_reorder_buffer = IAllocator::MakeUniquePtr<void>(alloc, sizeof(T) * X->Shape().Size());
// allocate the size queried from memory primitive desc. it may not match tensor logical size due to
// mkldnn using padding to allow use of blocked format.
src_reorder_buffer = IAllocator::MakeUniquePtr<void>(alloc, conv2d_primitive->GetSrcSize());
mkldnn::memory dst = mkldnn::memory(conv_fwd_pd->src_primitive_desc(), src_reorder_buffer.get());
MemoryReorderParams params(src, dst);
DoReorder<T>(params);
@ -364,7 +384,9 @@ Status Conv<T>::Compute(OpKernelContext* context) const {
filter_format),
cpu_engine);
mkldnn::memory src = mkldnn::memory(pd, (void*)filter_data);
filter_reorder_buffer = IAllocator::MakeUniquePtr<void>(alloc, sizeof(T) * W->Shape().Size());
// allocate the size queried from memory primitive desc. it may not match tensor logical size due to
// mkldnn using padding to allow use of blocked format.
filter_reorder_buffer = IAllocator::MakeUniquePtr<void>(alloc, conv2d_primitive->GetFilterSize());
mkldnn::memory dst = mkldnn::memory(conv_fwd_pd->weights_primitive_desc(), filter_reorder_buffer.get());
MemoryReorderParams params(src, dst);
DoReorder<T>(params);
@ -373,7 +395,9 @@ Status Conv<T>::Compute(OpKernelContext* context) const {
// Allocate dst buffer if reorder is necessary
if (dst_md.data.format != conv2d_primitive->GetDstMemoryFormat()) {
dst_reorder_buffer = IAllocator::MakeUniquePtr<void>(alloc, sizeof(T) * Y->Shape().Size());
// allocate the size queried from memory primitive desc. it may not match tensor logical size due to
// mkldnn using padding to allow use of blocked format.
dst_reorder_buffer = IAllocator::MakeUniquePtr<void>(alloc, conv2d_primitive->GetDstSize());
dst_data = static_cast<T*>(dst_reorder_buffer.get());
}

22
onnxruntime/core/providers/mkldnn/nn/lrn.cc
View file

@ -74,6 +74,10 @@ class LRNPrimitive : public PrimitiveBase {
mkldnn::memory::format GetSrcMemoryFormat() const { return context_.src_fmt; }
size_t GetSrcSize() const { return context_.src_size; }
size_t GetDstSize() const { return context_.dst_size; }
mkldnn::lrn_forward::primitive_desc* GetPrimitiveDesc() const {
return context_.fwd_primitive_desc.get();
}
@ -83,6 +87,9 @@ class LRNPrimitive : public PrimitiveBase {
mkldnn::memory::format src_fmt;
std::unique_ptr<mkldnn::memory::desc> src_md;
size_t src_size;
size_t dst_size;
std::unique_ptr<mkldnn::memory> src_mem;
std::unique_ptr<mkldnn::memory> dst_mem;
@ -96,6 +103,8 @@ class LRNPrimitive : public PrimitiveBase {
LRNContext()
: src_fmt(mkldnn::memory::format::any),
src_md(nullptr),
src_size(0),
dst_size(0),
src_mem(nullptr),
dst_mem(nullptr),
fwd_desc(nullptr),
@ -118,6 +127,9 @@ class LRNPrimitive : public PrimitiveBase {
context_.src_fmt = static_cast<mkldnn::memory::format>(
context_.fwd_primitive_desc.get()->src_primitive_desc().desc().data.format);
context_.src_size = context_.fwd_primitive_desc.get()->src_primitive_desc().get_size();
context_.dst_size = context_.fwd_primitive_desc.get()->dst_primitive_desc().get_size();
context_.src_mem.reset(new mkldnn::memory(context_.fwd_primitive_desc.get()->src_primitive_desc(), nullptr));
context_.dst_mem.reset(new mkldnn::memory(context_.fwd_primitive_desc.get()->dst_primitive_desc(), nullptr));
context_.lrn_fwd.reset(
@ -192,7 +204,9 @@ Status LRN<T>::Compute(OpKernelContext* context) const {
if (src_md.data.format != lrn_primitive->GetSrcMemoryFormat()) {
auto pd = mkldnn::memory::primitive_desc(src_md, cpu_engine);
mkldnn::memory src = mkldnn::memory(pd, (void*)src_data);
src_reorder_buffer = IAllocator::MakeUniquePtr<void>(alloc, sizeof(T) * X->Shape().Size());
// allocate the size queried from memory primitive desc. it may not match tensor logical size due to
// mkldnn using padding to allow use of blocked format.
src_reorder_buffer = IAllocator::MakeUniquePtr<void>(alloc, lrn_primitive->GetSrcSize());
mkldnn::memory dst = mkldnn::memory(fwd_primitive_desc->src_primitive_desc(), src_reorder_buffer.get());
MemoryReorderParams params(src, dst);
DoReorder<T>(params);
@ -201,7 +215,9 @@ Status LRN<T>::Compute(OpKernelContext* context) const {
// Allocate dst buffer if reorder is necessary
if (src_md.data.format != lrn_primitive->GetSrcMemoryFormat()) {
dst_reorder_buffer = IAllocator::MakeUniquePtr<void>(alloc, sizeof(T) * Y->Shape().Size());
// allocate the size queried from memory primitive desc. it may not match tensor logical size due to
// mkldnn using padding to allow use of blocked format.
dst_reorder_buffer = IAllocator::MakeUniquePtr<void>(alloc, lrn_primitive->GetDstSize());
dst_data = static_cast<T*>(dst_reorder_buffer.get());
}
@ -222,5 +238,5 @@ Status LRN<T>::Compute(OpKernelContext* context) const {
return Status::OK();
}
} // namespace mkl_dnn
} // namespace onnxruntime
}

19
onnxruntime/core/providers/mkldnn/nn/pool.cc
View file

@ -103,6 +103,9 @@ class PoolPrimitive : public PrimitiveBase {
mkldnn::memory::format GetSrcMemoryFormat() const { return context_.src_fmt; }
mkldnn::memory::format GetDstMemoryFormat() const { return context_.dst_fmt; }
size_t GetSrcSize() const { return context_.src_size; }
size_t GetDstSize() const { return context_.dst_size; }
// std::unique_ptr<mkldnn::memory::desc> GetDstMemoryDesc() const { return context_.dst_md; }
mkldnn::pooling_forward::primitive_desc* GetPrimitiveDesc() const {
return context_.fwd_primitive_desc.get();
@ -113,6 +116,9 @@ class PoolPrimitive : public PrimitiveBase {
mkldnn::memory::format src_fmt;
mkldnn::memory::format dst_fmt;
size_t src_size;
size_t dst_size;
std::unique_ptr<mkldnn::memory> src_mem;
std::unique_ptr<mkldnn::memory> dst_mem;
@ -131,6 +137,8 @@ class PoolPrimitive : public PrimitiveBase {
PoolContext()
: src_fmt(mkldnn::memory::format::any),
dst_fmt(mkldnn::memory::format::any),
src_size(0),
dst_size(0),
src_mem(nullptr),
dst_mem(nullptr),
fwd_desc(nullptr),
@ -178,6 +186,9 @@ class PoolPrimitive : public PrimitiveBase {
context_.dst_fmt = static_cast<mkldnn::memory::format>(
context_.fwd_primitive_desc.get()->dst_primitive_desc().desc().data.format);
context_.src_size = context_.fwd_primitive_desc.get()->src_primitive_desc().get_size();
context_.dst_size = context_.fwd_primitive_desc.get()->dst_primitive_desc().get_size();
context_.src_mem.reset(
new mkldnn::memory(context_.fwd_primitive_desc.get()->src_primitive_desc(), nullptr));
context_.dst_mem.reset(
@ -288,7 +299,9 @@ Status Pool<T, PoolType>::Compute(OpKernelContext* context) const {
if (src_md.data.format != pool_primitive->GetSrcMemoryFormat()) {
auto pd = mkldnn::memory::primitive_desc(src_md, cpu_engine);
mkldnn::memory src = mkldnn::memory(pd, (void*)src_data);
src_reorder_buffer = IAllocator::MakeUniquePtr<void>(alloc, sizeof(T) * X->Shape().Size());
// allocate the size queried from memory primitive desc. it may not match tensor logical size due to
// mkldnn using padding to allow use of blocked format.
src_reorder_buffer = IAllocator::MakeUniquePtr<void>(alloc, pool_primitive->GetSrcSize());
mkldnn::memory dst = mkldnn::memory(fwd_primitive_desc->src_primitive_desc(), src_reorder_buffer.get());
MemoryReorderParams params(src, dst);
DoReorder<T>(params);
@ -297,7 +310,9 @@ Status Pool<T, PoolType>::Compute(OpKernelContext* context) const {
// Allocate dst buffer if reorder is necessary
if (dst_md.data.format != pool_primitive->GetDstMemoryFormat()) {
dst_reorder_buffer = IAllocator::MakeUniquePtr<void>(alloc, sizeof(T) * Y->Shape().Size());
// allocate the size queried from memory primitive desc. it may not match tensor logical size due to
// mkldnn using padding to allow use of blocked format.
dst_reorder_buffer = IAllocator::MakeUniquePtr<void>(alloc, pool_primitive->GetDstSize());
dst_data = static_cast<T*>(dst_reorder_buffer.get());
}