saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-07-30 20:18:08 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Make "--cuda_version" build arg optional (#15758)

Browse source 
### Description
This change will allow us building CUDA EP without installing CUDA SDK
on Windows.

### Motivation and Context
Nvidia's CUDA installer comes with a VS extension. In the past, we
require installing the extension. It is a little bit inconvenient since:
1. Visual Studio must be installed before CUDA SDK. CUDA's installer
will not install the extension if your machine doesn't have Visual
Studio.
2. We need to install CUDA SDK on our build machines, instead of just
downloading it and using it.

After this change, we will not need to install CUDA SDK on our build
machines. So it will be easier to add a support for a different CUDA
version.

Also, fix two PreFast warnings.
This commit is contained in:
Changming Sun 2023-05-01 18:00:47 -07:00 committed by GitHub
parent 7f293d065a
commit 5352f6d9b0
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
5 changed files with 19 additions and 10 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

10
cmake/onnxruntime_python.cmake
View file

@ -286,7 +286,13 @@ if (WIN32)
endif()
get_filename_component(CUDNN_DLL_NAME ${CUDNN_DLL_PATH} NAME_WE)
string(REPLACE "cudnn64_" "" CUDNN_VERSION "${CUDNN_DLL_NAME}")
if(NOT onnxruntime_CUDA_VERSION)
message("Reading json file ${onnxruntime_CUDA_HOME}/version.json")
set(CUDA_SDK_JSON_FILE_PATH "${onnxruntime_CUDA_HOME}/version.json")
file(READ ${CUDA_SDK_JSON_FILE_PATH} CUDA_SDK_JSON_CONTENT)
string(JSON onnxruntime_CUDA_VERSION GET ${CUDA_SDK_JSON_CONTENT} "cuda" "version")
message("onnxruntime_CUDA_VERSION=${onnxruntime_CUDA_VERSION}")
endif()
file(APPEND "${VERSION_INFO_FILE}"
"cuda_version = \"${onnxruntime_CUDA_VERSION}\"\n"
"cudnn_version = \"${CUDNN_VERSION}\"\n"
@ -648,7 +654,7 @@ if (NOT onnxruntime_MINIMAL_BUILD AND NOT onnxruntime_EXTENDED_MINIMAL_BUILD
AND NOT ${CMAKE_SYSTEM_NAME} MATCHES "Darwin|iOS"
AND NOT (CMAKE_SYSTEM_NAME STREQUAL "Android")
AND NOT onnxruntime_USE_ROCM
AND NOT onnxruntime_BUILD_WEBASSEMBLY)
AND NOT onnxruntime_BUILD_WEBASSEMBLY)
add_custom_command(
TARGET onnxruntime_pybind11_state POST_BUILD
COMMAND ${CMAKE_COMMAND} -E copy

8
onnxruntime/contrib_ops/cuda/bert/tensorrt_fused_multihead_attention/fused_multihead_attention_v2.h
View file

@ -18,6 +18,11 @@
// Modified based on FasterTransformer 5.2
#pragma once
#if defined(_MSC_VER)
#pragma warning(push)
// Global initializer accesses extern object cubin_fmha_v2_fp16_Causal_64_40_sm86_cu_cubin_len
#pragma warning(disable : 26427)
#endif
#include "contrib_ops/cuda/bert/tensorrt_fused_multihead_attention/fused_multihead_attention.h"
#include <cstdint>
@ -3233,3 +3238,6 @@ inline const FusedMultiHeadAttentionXMMAKernelV2* getXMMAKernelsV2(Data_type typ
} // namespace cuda
} // namespace contrib
} // namespace onnxruntime
#if defined(_MSC_VER)
#pragma warning(pop)
#endif

1
onnxruntime/contrib_ops/cuda/transformers/generation_device_helper.cc
View file

@ -1071,7 +1071,6 @@ Status UpdateDecoderFeeds(
CUDA_RETURN_IF_ERROR(cudaMemcpyAsync(input_ids_data, beam_next_tokens.data(), beam_next_tokens.size_bytes(),
cudaMemcpyHostToDevice, cuda_stream));
} else {
int32_t* input_ids_data = input_ids.GetMutable<Tensor>()->MutableData<int32_t>();
for (int i = 0; i < batch_beam_size; i++) {
gsl::span<const int32_t> sequence = sequences.GetSequence(i);
const int32_t* sequence_data = sequence.data();

2
onnxruntime/test/providers/cpu/signal/signal_ops_test.cc
View file

@ -116,7 +116,7 @@ static void TestDFTInvertible(bool complex) {
for (int64_t signal_dim1 = 2; signal_dim1 <= 5; signal_dim1 += 1) {
for (int64_t signal_dim2 = 2; signal_dim2 <= 5; signal_dim2 += 1) {
DFTInvertibleTester test(axis);
vector<int64_t> input_shape{num_batches, signal_dim1, signal_dim2, 1 + complex};
vector<int64_t> input_shape{num_batches, signal_dim1, signal_dim2, 1 + (complex ? 1 : 0)};
vector<float> input_data = random.Uniform<float>(input_shape, -100.f, 100.f);
test.AddInput("input", input_shape, input_data);

8
tools/ci_build/build.py
View file

@ -2482,6 +2482,8 @@ def main():
toolset = "host=" + host_arch
if args.cuda_version:
toolset += ",cuda=" + args.cuda_version
elif args.cuda_home:
toolset += ",cuda=" + args.cuda_home
cmake_extra_args = ["-A", target_arch, "-T", toolset, "-G", args.cmake_generator]
if args.enable_wcos:
cmake_extra_defines.append("CMAKE_USER_MAKE_RULES_OVERRIDE=wcos_rules_override.cmake")
@ -2520,12 +2522,6 @@ def main():
if args.enable_pybind and is_windows():
install_python_deps(args.numpy_version)
if args.use_cuda and args.cuda_version is None:
if is_windows():
# cuda_version is used while generating version_info.py on Windows.
raise BuildError("cuda_version must be specified on Windows.")
else:
args.cuda_version = ""
if args.use_rocm and args.rocm_version is None:
args.rocm_version = ""