Implement SIMD latest opcodes and update LLVM to 13.0 (#758)

Implement the latest SIMD opcodes and update LLVM 13.0,
update the llvm build scripts, update the sample workloads‘ build scripts,
and build customized wasi-sdk to build some workloads.
Also refine the CI rules.

Signed-off-by: Wenyong Huang <wenyong.huang@intel.com>
This commit is contained in:
Wenyong Huang
2021-09-17 19:12:57 +08:00
committed by GitHub
parent 7e60a5db8d
commit 7be0d385a6
82 changed files with 5266 additions and 4698 deletions

View File

@ -1,5 +1,5 @@
All workloads have similar requirment of software dependencies, including
**wasi-sdk**, **emsdk**, **wabt** and **binaryen**
**emsdk**, **wabt** and **binaryen**
> There might be slight differences when using MacOS and other Linux distro than Ubuntu. This document only target
Ubuntu 18.04 as example.
@ -10,16 +10,6 @@ use [preparation.sh](./preparation.sh) to install all dependencies before compil
for details, the script includes below steps:
- **wasi-sdk**. Install
[latest release](https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-12/wasi-sdk-12.0-linux.tar.gz)
to */opt/wasi-sdk*
``` bash
$ wget https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-${WASI_SDK_VER}/${WASI_SDK_FILE}
$ tar zxf ${WASI_SDK_FILE} -C /opt
$ ln -sf /opt/wasi-sdk-${WASI_SDK_VER}.0 /opt/wasi-sdk
```
- **wabt**. Install
[latest release](https://github.com/WebAssembly/wabt/releases/download/1.0.23/wabt-1.0.23-ubuntu.tar.gz)
to */opt/wabt*
@ -32,15 +22,15 @@ $ ln -sf /opt/wabt-${WABT_VER} /opt/wabt
- **emsdk**. Refer to [the guide](https://emscripten.org/docs/getting_started/downloads.html). Don't forget to activate
emsdk and set up environment variables. Verify it with `echo ${EMSDK}`. Please be sure to install and activate the building
of 2.0.12
of 2.0.26
``` bash
$ cd /opt
$ git clone https://github.com/emscripten-core/emsdk.git
$ cd emsdk
$ git pull
$ ./emsdk install 2.0.12
$ ./emsdk activate 2.0.12
$ ./emsdk install 2.0.26
$ ./emsdk activate 2.0.26
$ echo "source /opt/emsdk/emsdk_env.sh" >> "${HOME}"/.bashrc
```

View File

@ -6,85 +6,87 @@ cmake_minimum_required (VERSION 3.0)
project(xnnpack_wasm)
################ EMCC ################
if(NOT DEFINED ENV{EMSDK})
message(SEND_ERROR
"can not find emsdk. "
"please refer to https://emscripten.org/docs/getting_started/downloads.html "
"and install it, "
"or active emsdk by 'source ./emsdk_env.sh'"
)
endif()
include(ExternalProject)
ExternalProject_Add(xnnpack
PREFIX xnnpack
GIT_REPOSITORY https://github.com/google/XNNPACK.git
GIT_TAG 90f520b6482bb99ac1bbfb71be1382f6c9b83241
GIT_TAG master
GIT_PROGRESS ON
SOURCE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/xnnpack
UPDATE_COMMAND git checkout .
&& cmake -E copy ${CMAKE_CURRENT_SOURCE_DIR}/benchmark.patch ${CMAKE_CURRENT_SOURCE_DIR}/xnnpack/third_party
&& git apply ${CMAKE_CURRENT_SOURCE_DIR}/xnnpack.patch
CONFIGURE_COMMAND ""
# grep xnnpack_benchmark -A 1 BUILD.bazel \
# | grep "name =" \
# | awk '{print $3}' \
# | sed -e 's/\"//g' -e 's/,//g' -e 's/^/\/\/:/g'
BUILD_COMMAND cd ${CMAKE_CURRENT_SOURCE_DIR}/xnnpack
&& bazel --output_user_root=build_user_output build -c opt --config=wasm
//:qs8_gemm_bench.wasm
//:qs8_requantization_bench.wasm
//:qu8_gemm_bench.wasm
//:qu8_requantization_bench.wasm
//:f16_igemm_bench.wasm
//:f16_gemm_bench.wasm
//:f16_spmm_bench.wasm
//:f32_igemm_bench.wasm
//:f16_relu_bench.wasm
//:f32_conv_hwc_bench.wasm
//:f32_conv_hwc2chw_bench.wasm
//:f16_dwconv_bench.wasm
//:f32_dwconv_bench.wasm
//:f32_dwconv2d_chw_bench.wasm
//:f32_gemm_bench.wasm
//:f32_hswish_bench.wasm
//:f32_raddexpminusmax_bench.wasm
//:f32_raddextexp_bench.wasm
//:f32_raddstoreexpminusmax_bench.wasm
//:f32_relu_bench.wasm
//:f32_rmax_bench.wasm
//:f32_sigmoid_bench.wasm
//:f32_spmm_bench.wasm
//:f32_softmax_bench.wasm
//:f32_velu_bench.wasm
//:f32_vscaleexpminusmax_bench.wasm
//:f32_vscaleextexp_bench.wasm
//:f32_vsqrt_bench.wasm
//:f32_im2col_gemm_bench.wasm
//:rounding_bench.wasm
//:average_pooling_bench.wasm
//:bankers_rounding_bench.wasm
//:ceiling_bench.wasm
//:channel_shuffle_bench.wasm
//:convolution_bench.wasm
//:deconvolution_bench.wasm
//:elu_bench.wasm
//:floor_bench.wasm
//:global_average_pooling_bench.wasm
//:hardswish_bench.wasm
//:max_pooling_bench.wasm
//:sigmoid_bench.wasm
//:prelu_bench.wasm
//:softmax_bench.wasm
//:square_root_bench.wasm
//:truncation_bench.wasm
//:f32_dwconv_e2e_bench.wasm
//:f32_gemm_e2e_bench.wasm
//:qs8_gemm_e2e_bench.wasm
//:end2end_bench.wasm
//:f32_exp_ulp_eval.wasm
//:f32_expminus_ulp_eval.wasm
//:f32_expm1minus_ulp_eval.wasm
//:f32_extexp_ulp_eval.wasm
//:f32_sigmoid_ulp_eval.wasm
//:f32_sqrt_ulp_eval.wasm
#--sandbox_debug
//:qs8_dwconv_bench.wasm
//:qs8_gemm_bench.wasm
//:qs8_requantization_bench.wasm
//:qs8_vadd_bench.wasm
//:qs8_vaddc_bench.wasm
//:qu8_gemm_bench.wasm
//:qu8_requantization_bench.wasm
//:qu8_vadd_bench.wasm
//:qu8_vaddc_bench.wasm
//:f16_igemm_bench.wasm
//:f16_gemm_bench.wasm
//:f16_spmm_bench.wasm
//:f16_vrelu_bench.wasm
//:f32_igemm_bench.wasm
//:f32_conv_hwc_bench.wasm
//:f32_conv_hwc2chw_bench.wasm
//:f16_dwconv_bench.wasm
//:f32_dwconv_bench.wasm
//:f32_dwconv2d_chw_bench.wasm
//:f32_gemm_bench.wasm
//:f32_raddexpminusmax_bench.wasm
//:f32_raddextexp_bench.wasm
//:f32_raddstoreexpminusmax_bench.wasm
//:f32_rmax_bench.wasm
//:f32_spmm_bench.wasm
//:f32_softmax_bench.wasm
//:f32_velu_bench.wasm
//:f32_vhswish_bench.wasm
//:f32_vrelu_bench.wasm
//:f32_vscaleexpminusmax_bench.wasm
//:f32_vscaleextexp_bench.wasm
//:f32_vsigmoid_bench.wasm
//:f32_vsqrt_bench.wasm
//:f32_im2col_gemm_bench.wasm
//:rounding_bench.wasm
//:average_pooling_bench.wasm
//:bankers_rounding_bench.wasm
//:ceiling_bench.wasm
//:channel_shuffle_bench.wasm
//:convolution_bench.wasm
//:deconvolution_bench.wasm
//:elu_bench.wasm
//:floor_bench.wasm
//:global_average_pooling_bench.wasm
//:hardswish_bench.wasm
//:max_pooling_bench.wasm
//:sigmoid_bench.wasm
//:prelu_bench.wasm
//:softmax_bench.wasm
//:square_root_bench.wasm
//:truncation_bench.wasm
//:f32_dwconv_e2e_bench.wasm
//:f32_gemm_e2e_bench.wasm
//:qs8_dwconv_e2e_bench.wasm
//:qs8_gemm_e2e_bench.wasm
//:qu8_dwconv_e2e_bench.wasm
//:end2end_bench.wasm
//:f32_exp_ulp_eval.wasm
//:f32_expminus_ulp_eval.wasm
//:f32_expm1minus_ulp_eval.wasm
//:f32_extexp_ulp_eval.wasm
//:f32_sigmoid_ulp_eval.wasm
//:f32_sqrt_ulp_eval.wasm
INSTALL_COMMAND ${CMAKE_COMMAND} -E copy_directory
${CMAKE_CURRENT_SOURCE_DIR}/xnnpack/bazel-out/wasm-opt/bin/
${CMAKE_CURRENT_SOURCE_DIR}/build/wasm-opt

View File

@ -5,20 +5,7 @@ This sample demonstrates how to build [XNNPACK](https://github.com/google/XNNPAC
## Installation toolchains
- **bazel**. Please install bazel from [latest release](https://github.com/bazelbuild/bazel/releases)
- **emsdk**. Please install [emsdk](https://github.com/emscripten-core/emsdk) to /opt/emsdk:
```bash
cd /opt
git clone https://github.com/emscripten-core/emsdk.git
cd emsdk
./emsdk install latest
./emsdk activate latest
```
And set up ensdk environment:
```bash
source /opt/emsdk/emsdk_env.sh
```
please refer to [installation instructions](../README.md).
## Build XNNPACK

View File

@ -0,0 +1,14 @@
diff --git include/benchmark/benchmark.h include/benchmark/benchmark.h
index 9b54802..baa5938 100755
--- include/benchmark/benchmark.h
+++ include/benchmark/benchmark.h
@@ -364,7 +364,9 @@ template <class Tp>
inline BENCHMARK_ALWAYS_INLINE void DoNotOptimize(Tp const& value) {
internal::UseCharPointer(&reinterpret_cast<char const volatile&>(value));
}
+
// FIXME Add ClobberMemory() for non-gnu and non-msvc compilers
+inline BENCHMARK_ALWAYS_INLINE void ClobberMemory() { }
#endif
// This class is used for user-defined counters.

View File

@ -0,0 +1 @@
../docker/build_workload.sh

View File

@ -1 +0,0 @@
../docker/docker_build.sh

View File

@ -1,8 +1,8 @@
diff --git a/.bazelrc b/.bazelrc
index ec740f38..2c193244 100644
index ec740f38..29f9d56e 100644
--- a/.bazelrc
+++ b/.bazelrc
@@ -49,4 +49,10 @@ build:ios_fat --watchos_cpus=armv7k
@@ -49,4 +49,9 @@ build:ios_fat --watchos_cpus=armv7k
build:macos --apple_platform_type=macos
build:macos_arm64 --config=macos
@ -10,558 +10,31 @@ index ec740f38..2c193244 100644
\ No newline at end of file
+build:macos_arm64 --cpu=darwin_arm64
+
+build:wasm --copt=-msimd128
+build:wasm --cpu=wasm
+build:wasm --copt=-msimd128
+build:wasm --crosstool_top=@emsdk//emscripten_toolchain:everything
+build:wasm --host_crosstool_top=@bazel_tools//tools/cpp:toolchain
+
diff --git a/BUILD.bazel b/BUILD.bazel
index 1f2b15a8..e7abf838 100644
--- a/BUILD.bazel
+++ b/BUILD.bazel
@@ -4996,7 +4996,7 @@ xnnpack_cc_library(
######################### Benchmarks for micro-kernels #########################
xnnpack_benchmark(
- name = "qs8_gemm_bench",
+ name = "qs8_gemm_bench.wasm",
srcs = [
"bench/gemm.h",
"bench/qs8-gemm.cc",
@@ -5007,7 +5007,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "qs8_requantization_bench",
+ name = "qs8_requantization_bench.wasm",
srcs = [
"bench/qs8-requantization.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5017,7 +5017,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "qu8_gemm_bench",
+ name = "qu8_gemm_bench.wasm",
srcs = [
"bench/gemm.h",
"bench/qu8-gemm.cc",
@@ -5028,7 +5028,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "qu8_requantization_bench",
+ name = "qu8_requantization_bench.wasm",
srcs = [
"bench/qu8-requantization.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5038,11 +5038,10 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f16_igemm_bench",
+ name = "f16_igemm_bench.wasm",
srcs = [
"bench/f16-igemm.cc",
"bench/conv.h",
- "bench/google/conv.h",
"src/xnnpack/AlignedAllocator.h",
] + MICROKERNEL_BENCHMARK_HDRS,
deps = MICROKERNEL_BENCHMARK_DEPS + [
@@ -5052,7 +5051,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f16_gemm_bench",
+ name = "f16_gemm_bench.wasm",
srcs = [
"bench/f16-gemm.cc",
"bench/gemm.h",
@@ -5064,7 +5063,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f16_spmm_bench",
+ name = "f16_spmm_bench.wasm",
srcs = [
"bench/f16-spmm.cc",
"bench/spmm.h",
@@ -5074,7 +5073,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_igemm_bench",
+ name = "f32_igemm_bench.wasm",
srcs = [
"bench/f32-igemm.cc",
"bench/conv.h",
@@ -5087,7 +5086,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f16_relu_bench",
+ name = "f16_relu_bench.wasm",
srcs = [
"bench/f16-relu.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5096,7 +5095,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_conv_hwc_bench",
+ name = "f32_conv_hwc_bench.wasm",
srcs = [
"bench/f32-conv-hwc.cc",
"bench/dconv.h",
@@ -5108,7 +5107,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_conv_hwc2chw_bench",
+ name = "f32_conv_hwc2chw_bench.wasm",
srcs = [
"bench/f32-conv-hwc2chw.cc",
"bench/dconv.h",
@@ -5120,11 +5119,10 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f16_dwconv_bench",
+ name = "f16_dwconv_bench.wasm",
srcs = [
"bench/f16-dwconv.cc",
"bench/dwconv.h",
- "bench/google/dwconv.h",
"src/xnnpack/AlignedAllocator.h",
] + MICROKERNEL_BENCHMARK_HDRS,
deps = MICROKERNEL_BENCHMARK_DEPS + [
@@ -5134,7 +5132,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_dwconv_bench",
+ name = "f32_dwconv_bench.wasm",
srcs = [
"bench/f32-dwconv.cc",
"bench/dwconv.h",
@@ -5147,7 +5145,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_dwconv2d_chw_bench",
+ name = "f32_dwconv2d_chw_bench.wasm",
srcs = [
"bench/f32-dwconv2d-chw.cc",
"bench/dwconv.h",
@@ -5160,7 +5158,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_gemm_bench",
+ name = "f32_gemm_bench.wasm",
srcs = [
"bench/f32-gemm.cc",
"bench/gemm.h",
@@ -5171,7 +5169,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_hswish_bench",
+ name = "f32_hswish_bench.wasm",
srcs = [
"bench/f32-hswish.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5180,7 +5178,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_raddexpminusmax_bench",
+ name = "f32_raddexpminusmax_bench.wasm",
srcs = [
"bench/f32-raddexpminusmax.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5189,7 +5187,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_raddextexp_bench",
+ name = "f32_raddextexp_bench.wasm",
srcs = [
"bench/f32-raddextexp.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5198,7 +5196,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_raddstoreexpminusmax_bench",
+ name = "f32_raddstoreexpminusmax_bench.wasm",
srcs = [
"bench/f32-raddstoreexpminusmax.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5207,7 +5205,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_relu_bench",
+ name = "f32_relu_bench.wasm",
srcs = [
"bench/f32-relu.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5216,7 +5214,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_rmax_bench",
+ name = "f32_rmax_bench.wasm",
srcs = [
"bench/f32-rmax.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5225,7 +5223,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_sigmoid_bench",
+ name = "f32_sigmoid_bench.wasm",
srcs = [
"bench/f32-sigmoid.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5234,7 +5232,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_spmm_bench",
+ name = "f32_spmm_bench.wasm",
srcs = [
"bench/f32-spmm.cc",
"bench/spmm.h",
@@ -5244,7 +5242,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_softmax_bench",
+ name = "f32_softmax_bench.wasm",
srcs = [
"bench/f32-softmax.cc",
] + MICROKERNEL_BENCHMARK_HDRS,
@@ -5253,7 +5251,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_velu_bench",
+ name = "f32_velu_bench.wasm",
srcs = [
"bench/f32-velu.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5262,7 +5260,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_vscaleexpminusmax_bench",
+ name = "f32_vscaleexpminusmax_bench.wasm",
srcs = [
"bench/f32-vscaleexpminusmax.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5271,7 +5269,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_vscaleextexp_bench",
+ name = "f32_vscaleextexp_bench.wasm",
srcs = [
"bench/f32-vscaleextexp.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5280,7 +5278,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_vsqrt_bench",
+ name = "f32_vsqrt_bench.wasm",
srcs = [
"bench/f32-vsqrt.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5289,7 +5287,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_im2col_gemm_bench",
+ name = "f32_im2col_gemm_bench.wasm",
srcs = [
"bench/f32-im2col-gemm.cc",
"bench/conv.h",
@@ -5302,7 +5300,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "rounding_bench",
+ name = "rounding_bench.wasm",
srcs = [
"bench/rounding.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5314,7 +5312,7 @@ xnnpack_benchmark(
########################### Benchmarks for operators ###########################
xnnpack_benchmark(
- name = "average_pooling_bench",
+ name = "average_pooling_bench.wasm",
srcs = ["bench/average-pooling.cc"],
copts = xnnpack_optional_tflite_copts(),
tags = ["nowin32"],
@@ -5322,7 +5320,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "bankers_rounding_bench",
+ name = "bankers_rounding_bench.wasm",
srcs = ["bench/bankers-rounding.cc"],
copts = xnnpack_optional_tflite_copts(),
tags = ["nowin32"],
@@ -5330,7 +5328,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "ceiling_bench",
+ name = "ceiling_bench.wasm",
srcs = ["bench/ceiling.cc"],
copts = xnnpack_optional_tflite_copts(),
tags = ["nowin32"],
@@ -5338,13 +5336,13 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "channel_shuffle_bench",
+ name = "channel_shuffle_bench.wasm",
srcs = ["bench/channel-shuffle.cc"],
deps = OPERATOR_BENCHMARK_DEPS,
)
xnnpack_benchmark(
- name = "convolution_bench",
+ name = "convolution_bench.wasm",
srcs = ["bench/convolution.cc"],
copts = xnnpack_optional_tflite_copts() + xnnpack_optional_armcl_copts(),
tags = ["nowin32"],
@@ -5352,7 +5350,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "deconvolution_bench",
+ name = "deconvolution_bench.wasm",
srcs = ["bench/deconvolution.cc"],
copts = xnnpack_optional_tflite_copts(),
tags = ["nowin32"],
@@ -5360,7 +5358,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "elu_bench",
+ name = "elu_bench.wasm",
srcs = ["bench/elu.cc"],
copts = xnnpack_optional_tflite_copts(),
tags = ["nowin32"],
@@ -5368,7 +5366,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "floor_bench",
+ name = "floor_bench.wasm",
srcs = ["bench/floor.cc"],
copts = xnnpack_optional_tflite_copts(),
tags = ["nowin32"],
@@ -5376,13 +5374,13 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "global_average_pooling_bench",
+ name = "global_average_pooling_bench.wasm",
srcs = ["bench/global-average-pooling.cc"],
deps = OPERATOR_BENCHMARK_DEPS,
)
xnnpack_benchmark(
- name = "hardswish_bench",
+ name = "hardswish_bench.wasm",
srcs = ["bench/hardswish.cc"],
copts = xnnpack_optional_tflite_copts(),
tags = ["nowin32"],
@@ -5390,13 +5388,13 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "max_pooling_bench",
+ name = "max_pooling_bench.wasm",
srcs = ["bench/max-pooling.cc"],
deps = OPERATOR_BENCHMARK_DEPS,
)
xnnpack_benchmark(
- name = "sigmoid_bench",
+ name = "sigmoid_bench.wasm",
srcs = ["bench/sigmoid.cc"],
copts = xnnpack_optional_tflite_copts(),
tags = ["nowin32"],
@@ -5404,7 +5402,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "prelu_bench",
+ name = "prelu_bench.wasm",
srcs = ["bench/prelu.cc"],
copts = xnnpack_optional_tflite_copts(),
tags = ["nowin32"],
@@ -5412,7 +5410,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "softmax_bench",
+ name = "softmax_bench.wasm",
srcs = ["bench/softmax.cc"],
copts = xnnpack_optional_tflite_copts(),
tags = ["nowin32"],
@@ -5420,7 +5418,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "square_root_bench",
+ name = "square_root_bench.wasm",
srcs = ["bench/square-root.cc"],
copts = xnnpack_optional_tflite_copts(),
tags = ["nowin32"],
@@ -5428,7 +5426,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "truncation_bench",
+ name = "truncation_bench.wasm",
srcs = ["bench/truncation.cc"],
deps = OPERATOR_BENCHMARK_DEPS,
)
@@ -5620,7 +5618,7 @@ cc_library(
)
xnnpack_benchmark(
- name = "f32_dwconv_e2e_bench",
+ name = "f32_dwconv_e2e_bench.wasm",
srcs = [
"bench/f32-dwconv-e2e.cc",
"bench/end2end.h",
@@ -5635,7 +5633,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_gemm_e2e_bench",
+ name = "f32_gemm_e2e_bench.wasm",
srcs = [
"bench/f32-gemm-e2e.cc",
"bench/end2end.h",
@@ -5650,7 +5648,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "qs8_gemm_e2e_bench",
+ name = "qs8_gemm_e2e_bench.wasm",
srcs = [
"bench/qs8-gemm-e2e.cc",
"bench/end2end.h",
@@ -5663,7 +5661,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "end2end_bench",
+ name = "end2end_bench.wasm",
srcs = ["bench/end2end.cc"],
deps = [
":XNNPACK",
@@ -5690,7 +5688,7 @@ xnnpack_benchmark(
#################### Accuracy evaluation for math functions ####################
xnnpack_benchmark(
- name = "f32_exp_ulp_eval",
+ name = "f32_exp_ulp_eval.wasm",
srcs = [
"eval/f32-exp-ulp.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5702,7 +5700,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_expminus_ulp_eval",
+ name = "f32_expminus_ulp_eval.wasm",
srcs = [
"eval/f32-expminus-ulp.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5714,7 +5712,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_expm1minus_ulp_eval",
+ name = "f32_expm1minus_ulp_eval.wasm",
srcs = [
"eval/f32-expm1minus-ulp.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5726,7 +5724,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_extexp_ulp_eval",
+ name = "f32_extexp_ulp_eval.wasm",
srcs = [
"eval/f32-extexp-ulp.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5738,7 +5736,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_sigmoid_ulp_eval",
+ name = "f32_sigmoid_ulp_eval.wasm",
srcs = [
"eval/f32-sigmoid-ulp.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5750,7 +5748,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_sqrt_ulp_eval",
+ name = "f32_sqrt_ulp_eval.wasm",
srcs = [
"eval/f32-sqrt-ulp.cc",
"src/xnnpack/AlignedAllocator.h",
diff --git a/WORKSPACE b/WORKSPACE
index 4fa1aa2f..6181aab2 100644
index c58e76b6..30934678 100644
--- a/WORKSPACE
+++ b/WORKSPACE
@@ -5,8 +5,8 @@ load("@bazel_tools//tools/build_defs/repo:http.bzl", "http_archive")
# Bazel rule definitions
http_archive(
name = "rules_cc",
- strip_prefix = "rules_cc-master",
- urls = ["https://github.com/bazelbuild/rules_cc/archive/master.zip"],
+ strip_prefix = "rules_cc-main",
+ urls = ["https://github.com/bazelbuild/rules_cc/archive/main.zip"],
)
# Google Test framework, used by most unit-tests.
@@ -19,8 +19,8 @@ http_archive(
# Google Benchmark library, used in micro-benchmarks.
http_archive(
@@ -21,6 +21,7 @@ http_archive(
name = "com_google_benchmark",
- strip_prefix = "benchmark-master",
- urls = ["https://github.com/google/benchmark/archive/master.zip"],
+ strip_prefix = "benchmark-1.5.3",
+ urls = ["https://github.com/google/benchmark/archive/refs/tags/v1.5.3.zip"],
strip_prefix = "benchmark-master",
urls = ["https://github.com/google/benchmark/archive/master.zip"],
+ patches = ["@//third_party:benchmark.patch"],
)
# FP16 library, used for half-precision conversions
@@ -89,3 +89,18 @@ android_ndk_repository(name = "androidndk")
@@ -84,6 +85,19 @@ http_archive(
],
)
# Android SDK location and API is auto-detected from $ANDROID_HOME environment variable
android_sdk_repository(name = "androidsdk")
+
+# emscripten library
+http_archive(
+ name = "emsdk",
+ strip_prefix = "emsdk-c1589b55641787d55d53e883852035beea9aec3f/bazel",
+ url = "https://github.com/emscripten-core/emsdk/archive/c1589b55641787d55d53e883852035beea9aec3f.tar.gz",
+ sha256 = "7a58a9996b113d3e0675df30b5f17e28aa47de2e684a844f05394fe2f6f12e8e",
+ name = "emsdk",
+ strip_prefix = "emsdk-2.0.26/bazel",
+ url = "https://github.com/emscripten-core/emsdk/archive/refs/tags/2.0.26.tar.gz",
+ sha256 = "79e7166aa8eaae6e52cef1363b2d8db795d03684846066bc51f9dcf905dd58ad",
+)
+
+load("@emsdk//:deps.bzl", emsdk_deps = "deps")
@ -570,67 +43,42 @@ index 4fa1aa2f..6181aab2 100644
+load("@emsdk//:emscripten_deps.bzl", emsdk_emscripten_deps = "emscripten_deps")
+emsdk_emscripten_deps()
+
# Android NDK location and version is auto-detected from $ANDROID_NDK_HOME environment variable
android_ndk_repository(name = "androidndk")
diff --git a/build_defs.bzl b/build_defs.bzl
index 10345032..0e926fca 100644
index 2442bed1..b860dfef 100644
--- a/build_defs.bzl
+++ b/build_defs.bzl
@@ -1,6 +1,6 @@
"""Build definitions and rules for XNNPACK."""
-load(":emscripten.bzl", "xnnpack_emscripten_benchmark_linkopts", "xnnpack_emscripten_deps", "xnnpack_emscripten_minimal_linkopts", "xnnpack_emscripten_test_linkopts")
+load(":emscripten.bzl", "xnnpack_emscripten_benchmark_linkopts", "xnnpack_emscripten_deps", "xnnpack_emscripten_minimal_linkopts", "xnnpack_emscripten_test_linkopts", "xnnpack_emscripten_benchmark_copts")
def xnnpack_visibility():
"""Visibility of :XNNPACK target.
@@ -424,10 +424,15 @@ def xnnpack_benchmark(name, srcs, copts = [], deps = [], tags = []):
":windows_x86_64_mingw": ["-Wno-unused-function"],
":windows_x86_64_msys": ["-Wno-unused-function"],
":windows_x86_64": [],
+ ":emscripten": xnnpack_emscripten_benchmark_copts(),
+ ":emscripten_wasm": xnnpack_emscripten_benchmark_copts(),
+ ":emscripten_wasmsimd": xnnpack_emscripten_benchmark_copts(),
"//conditions:default": ["-Wno-unused-function"],
}) + copts,
linkopts = select({
":emscripten": xnnpack_emscripten_benchmark_linkopts(),
+ ":emscripten_wasm": xnnpack_emscripten_benchmark_linkopts(),
+ ":emscripten_wasmsimd": xnnpack_emscripten_benchmark_linkopts(),
":windows_x86_64_mingw": ["-lshlwapi"],
":windows_x86_64_msys": ["-lshlwapi"],
"//conditions:default": [],
@@ -414,7 +414,7 @@ def xnnpack_benchmark(name, srcs, copts = [], deps = [], tags = []):
explicitly specified.
"""
native.cc_binary(
- name = name,
+ name = name + ".wasm",
srcs = srcs,
copts = xnnpack_std_cxxopts() + [
"-Iinclude",
diff --git a/emscripten.bzl b/emscripten.bzl
index 0a0caedf..aafe3199 100644
index 130d5f16..2696ad54 100644
--- a/emscripten.bzl
+++ b/emscripten.bzl
@@ -6,6 +6,7 @@ def xnnpack_emscripten_minimal_linkopts():
"-s ASSERTIONS=0",
"-s ERROR_ON_UNDEFINED_SYMBOLS=1",
"-s EXIT_RUNTIME=1",
+ "--oformat=wasm",
]
def xnnpack_emscripten_test_linkopts():
@@ -17,21 +18,36 @@ def xnnpack_emscripten_test_linkopts():
"-s EXIT_RUNTIME=1",
"-s ALLOW_MEMORY_GROWTH=1",
"--pre-js $(location :preamble.js.lds)",
+ "--oformat=wasm",
]
def xnnpack_emscripten_benchmark_linkopts():
@@ -25,12 +25,19 @@ def xnnpack_emscripten_benchmark_linkopts():
"""Emscripten-specific linkopts for benchmarks."""
return [
"-s ASSERTIONS=1",
- "-s ENVIRONMENT=node,shell,web",
- "-s ERROR_ON_UNDEFINED_SYMBOLS=1",
- "-s EXIT_RUNTIME=1",
- "-s ALLOW_MEMORY_GROWTH=1",
+ "-s ERROR_ON_UNDEFINED_SYMBOLS=0",
+ "-s ALLOW_MEMORY_GROWTH=0",
"-s TOTAL_MEMORY=436207616", # 416M
"-s ALLOW_MEMORY_GROWTH=1",
"-s TOTAL_MEMORY=445644800", # 425M
- "--pre-js $(location :preamble.js.lds)",
+ "-s USE_PTHREADS=0",
+ "-s STANDALONE_WASM=1",
+ "-Wno-unused",
+ "-Wno-unused-variable",
+ "-Wno-unused-command-line-argument",
+ "-Wl,--export=__heap_base",
+ "-Wl,--export=__data_end",
+ "-Wl,--export=malloc",
@ -639,19 +87,6 @@ index 0a0caedf..aafe3199 100644
]
def xnnpack_emscripten_deps():
"""Emscripten-specific dependencies for unit tests and benchmarks."""
+ return []
+
+def xnnpack_emscripten_benchmark_copts():
return [
- ":preamble.js.lds",
+ "-s ASSERTIONS=1",
+ "-s ERROR_ON_UNDEFINED_SYMBOLS=0",
+ "-s ALLOW_MEMORY_GROWTH=0",
+ "-s USE_PTHREADS=0",
+ "-s STANDALONE_WASM=1",
+ "-Wno-unused",
]
diff --git a/third_party/cpuinfo.BUILD b/third_party/cpuinfo.BUILD
index 128d683e..f6c287c4 100644
--- a/third_party/cpuinfo.BUILD

View File

@ -5,6 +5,8 @@ cmake_minimum_required (VERSION 3.0)
project(bwa_wasm C)
include(${CMAKE_CURRENT_SOURCE_DIR}/../../cmake/preparation.cmake)
################ LIBZ ################
set(LIBZ_SRC_DIR ${CMAKE_CURRENT_SOURCE_DIR}/../libz)
add_library(z_wasm STATIC
@ -86,16 +88,15 @@ set_target_properties(${PROJECT_NAME} PROPERTIES OUTPUT_NAME bwa.wasm)
target_include_directories(${PROJECT_NAME}
PRIVATE
${CMAKE_CURRENT_SOURCE_DIR}/../include
${CMAKE_CURRENT_SOURCE_DIR}/../include/SSE
${CMAKE_CURRENT_SOURCE_DIR}/../include/pthread
${WASI_SDK_HOME}/share/wasi-sysroot/include/libc/musl
${WASI_SDK_HOME}/share/wasi-sysroot/include/sse
)
target_compile_definitions(${PROJECT_NAME}
PRIVATE
USE_MALLOC_WRAPPERS
__SSE__ __SSE2__ __SSE4_1__
_WASI_EMULATED_MMAN _WASI_EMULATED_SIGNAL
_WASI_EMULATED_MMAN _WASI_EMULATED_SIGNAL _WASI_EMULATED_PROCESS_CLOCKS
)
target_compile_options(${PROJECT_NAME}
@ -112,16 +113,7 @@ target_link_options(${PROJECT_NAME}
LINKER:-z,stack-size=1048576
)
target_link_libraries(${PROJECT_NAME} z_wasm)
find_program(WASM_OPT
NAMES wasm-opt
PATHS /opt/binaryen-version_97/bin /opt/binaryen/bin
)
if (NOT WASM_OPT)
message(WARNING "can not find wasm-opt and will not optimize any wasm module")
endif()
target_link_libraries(${PROJECT_NAME} z_wasm wasi-emulated-process-clocks)
add_custom_target(bwa_wasm_opt ALL
COMMAND

View File

@ -10,67 +10,37 @@ include(${CMAKE_CURRENT_SOURCE_DIR}/../cmake/preparation.cmake)
#######################################
include(ExternalProject)
################ HEADERS ################
set(EMSDK_SYSTEM_HEADERS "$ENV{EMSDK}/upstream/emscripten/system/include")
set(EMSDK_SSE_HEADERS "${EMSDK_SYSTEM_HEADERS}/SSE")
set(EMSDK_LIBC_HEADERS "${EMSDK_SYSTEM_HEADERS}/libc")
ExternalProject_Add(headers_from_emcc
PREFIX headers
SOURCE_DIR ${EMSDK_SYSTEM_HEADERS}
CONFIGURE_COMMAND ""
BUILD_COMMAND ""
INSTALL_COMMAND mkdir -p ${CMAKE_CURRENT_SOURCE_DIR}/include/SSE
&& ${CMAKE_COMMAND} -E make_directory ${CMAKE_CURRENT_SOURCE_DIR}/include/pthread/sys
&& ${CMAKE_COMMAND} -E make_directory ${CMAKE_CURRENT_SOURCE_DIR}/include/emscripten
# copy emscripten SSE header files
&& ${CMAKE_COMMAND} -E copy ${EMSDK_SYSTEM_HEADERS}/wasm_simd128.h ${CMAKE_CURRENT_SOURCE_DIR}/include/SSE/
&& ${CMAKE_COMMAND} -E copy ${EMSDK_SSE_HEADERS}/immintrin.h ${CMAKE_CURRENT_SOURCE_DIR}/include/SSE/
# SSE
&& ${CMAKE_COMMAND} -E copy ${EMSDK_SSE_HEADERS}/xmmintrin.h ${CMAKE_CURRENT_SOURCE_DIR}/include/SSE/
# SSE2
&& ${CMAKE_COMMAND} -E copy ${EMSDK_SSE_HEADERS}/emmintrin.h ${CMAKE_CURRENT_SOURCE_DIR}/include/SSE/
# SSE4.1
&& ${CMAKE_COMMAND} -E copy ${EMSDK_SSE_HEADERS}/smmintrin.h ${CMAKE_CURRENT_SOURCE_DIR}/include/SSE/
# a fake empty header to aovid further depenency
&& ${CMAKE_COMMAND} -E touch ${CMAKE_CURRENT_SOURCE_DIR}/include/emscripten/emscripten.h
# copy emscripten pthread related header files
&& ${CMAKE_COMMAND} -E copy ${EMSDK_LIBC_HEADERS}/pthread.h ${CMAKE_CURRENT_SOURCE_DIR}/include/pthread/
&& ${CMAKE_COMMAND} -E copy ${EMSDK_LIBC_HEADERS}/signal.h ${CMAKE_CURRENT_SOURCE_DIR}/include/pthread/
&& ${CMAKE_COMMAND} -E copy ${EMSDK_LIBC_HEADERS}/netdb.h ${CMAKE_CURRENT_SOURCE_DIR}/include/pthread/
&& ${CMAKE_COMMAND} -E copy ${EMSDK_LIBC_HEADERS}/sys/wait.h ${CMAKE_CURRENT_SOURCE_DIR}/include/pthread/sys/
&& ${CMAKE_COMMAND} -E copy ${EMSDK_LIBC_HEADERS}/sys/socket.h ${CMAKE_CURRENT_SOURCE_DIR}/include/pthread/sys/
)
################ libz ################
ExternalProject_Add(libz_src
PREFIX libz
GIT_REPOSITORY https://github.com/madler/zlib.git
GIT_TAG master
GIT_PROGRESS ON
GIT_SHALLOW ON
SOURCE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/libz
UPDATE_COMMAND ""
PATCH_COMMAND ""
CONFIGURE_COMMAND ""
BUILD_COMMAND ""
INSTALL_COMMAND ""
BUILD_COMMAND ""
INSTALL_COMMAND ""
)
################ bwa ################
ExternalProject_Add(bwa
PREFIX bwa
GIT_REPOSITORY https://github.com/lh3/bwa.git
GIT_TAG master
GIT_PROGRESS ON
GIT_SHALLOW ON
SOURCE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/bwa
DEPENDS libz_src headers_from_emcc
UPDATE_COMMAND git clean -fd && git checkout -- *
DEPENDS libz_src
UPDATE_COMMAND git clean -ffdx && git checkout -- *
&& ${CMAKE_COMMAND} -E echo "Copying pre-installed CMakeLists.txt"
&& ${CMAKE_COMMAND} -E copy ${CMAKE_CURRENT_SOURCE_DIR}/CMakeLists.bwa_wasm.txt CMakeLists.txt
&& git apply ../bwa.patch
CONFIGURE_COMMAND ${CMAKE_COMMAND}
-DWASI_SDK_PREFIX=${WASI_SDK_HOME}/wasi-sdk
-DCMAKE_TOOLCHAIN_FILE=${WASI_SDK_HOME}/wasi-sdk/share/cmake/wasi-sdk.cmake
${CMAKE_CURRENT_SOURCE_DIR}/bwa
-DWASI_SDK_PREFIX=${WASI_SDK_HOME}
-DCMAKE_TOOLCHAIN_FILE=${WASI_SDK_HOME}/share/cmake/wasi-sdk.cmake
-DCMAKE_SYSROOT=${WASI_SDK_HOME}/share/wasi-sysroot
${CMAKE_CURRENT_SOURCE_DIR}/bwa
BUILD_COMMAND make bwa_wasm_opt
INSTALL_COMMAND ${CMAKE_COMMAND} -E copy ./bwa.opt.wasm ${CMAKE_CURRENT_SOURCE_DIR}/build/bwa.wasm
)

View File

@ -0,0 +1 @@
../docker/build_workload.sh

View File

@ -1 +0,0 @@
../docker/docker_build.sh

View File

@ -1,76 +1,49 @@
# Copyright (C) 2019 Intel Corporation. All rights reserved.
# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
################ WASI-SDK ################
find_path(WASI_SDK_HOME
NAMES wasi-sdk
PATHS /opt/
REQUIRED
#######################################
include(ExternalProject)
file(REAL_PATH ../../.. WAMR_ROOT
BASE_DIRECTORY ${CMAKE_CURRENT_LIST_DIR}
)
if (NOT WASI_SDK_HOME)
find_path(WASI_SDK_PARENT
name wasi-sdk
PATHS ${WAMR_ROOT}/test-tools/
NO_DEFAULT_PATH
NO_CMAKE_FIND_ROOT_PATH
)
if(NOT WASI_SDK_PARENT)
message(FATAL_ERROR
"can not find wasi-sdk. "
"please download it from "
"https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-12/wasi-sdk-12.0-linux.tar.gz "
"and install it under /opt/wasi-sdk"
"can not find 'wasi-sdk' under ${WAMR_ROOT}/test-tools, "
"please run ${WAMR_ROOT}/test-tools/build-wasi-sdk/build_wasi_sdk.py "
"to build wasi-sdk and try again"
)
endif()
set(WASI_SDK_HOME ${WASI_SDK_PARENT}/wasi-sdk)
message(CHECK_START "Detecting WASI-SDK at ${WASI_SDK_HOME}")
if(EXISTS "${WASI_SDK_HOME}/share/cmake/wasi-sdk.cmake")
message(CHECK_PASS "found")
else()
message(STATUS
"Detecting wasi-sdk info: ${WASI_SDK_HOME}/wasi-sdk"
)
endif()
#
# check clang version
execute_process(COMMAND
${WASI_SDK_HOME}/wasi-sdk/bin/clang --version
OUTPUT_VARIABLE clang_full_version_string
)
string(REGEX REPLACE ".*clang version ([0-9]+\\.[0-9]+).*" "\\1"
CLANG_VERSION_STRING ${clang_full_version_string}
)
message(STATUS "Detecting clang versoin: ${CLANG_VERSION_STRING}")
if(CLANG_VERSION_STRING VERSION_LESS 11.0)
message(FATAL_ERROR
"please install latest wai-sdk to get a clang-11 at least"
)
endif()
################ EMCC ################
if(NOT DEFINED ENV{EMSDK})
message(FATAL_ERROR
"can not find emsdk. "
"please refer to https://emscripten.org/docs/getting_started/downloads.html "
"and install it, "
"or active emsdk by 'source ./emsdk_env.sh'"
)
endif()
message(STATUS "Detecting EMSDK info: $ENV{EMSDK}")
### check if the emsdk is 2.0.12
### upstream/.emsdk_version should be releases-upstream-dcf819a7821f8db0c8f15ac336fea8960ec204f5-64bit
file(STRINGS "$ENV{EMSDK}/upstream/.emsdk_version" EMSDK_VERSION)
if(NOT (${EMSDK_VERSION} STREQUAL "releases-upstream-dcf819a7821f8db0c8f15ac336fea8960ec204f5-64bit"))
message(FATAL_ERROR "please install emsdk 2.0.12")
message(CHECK_FAIL "not found")
endif()
################ BINARYEN ################
find_program(WASM_OPT
NAMES wasm-opt
PATHS /opt/binaryen-version_101/bin /opt/binaryen/bin
NAMES wasm-opt
PATHS /opt/binaryen-version_101/bin /opt/binaryen/bin
NO_DEFAULT_PATH
NO_CMAKE_FIND_ROOT_PATH
)
if (NOT WASM_OPT)
if(NOT WASM_OPT)
message(FATAL_ERROR
"can not find wasm-opt. "
"please download it from "
"https://github.com/WebAssembly/binaryen/releases/download/version_101/binaryen-version_101-x86_64-linux.tar.gz "
"and install it under /opt"
)
else()
message(STATUS
"Detecting EMSDK info: $ENV{EMSDK}"
)
endif()

View File

@ -1 +0,0 @@
build_scripts

View File

@ -1,93 +0,0 @@
FROM ubuntu:18.04 as builder
RUN apt update \
&& apt install -y lsb-release software-properties-common build-essential \
wget curl git tree zip unzip
ARG WASI_SDK_VER=12
ARG WABT_VER=1.0.23
ARG CMAKE_VER=3.16.2
ARG BINARYEN_VER=version_101
#
# install wasi-sdk
ARG WASI_SDK_FILE="wasi-sdk-${WASI_SDK_VER}.0-linux.tar.gz"
COPY ${WASI_SDK_FILE} /opt
RUN cd /opt \
&& tar zxf ${WASI_SDK_FILE} \
&& rm ${WASI_SDK_FILE} \
&& ln -sf /opt/wasi-sdk-${WASI_SDK_VER}.0 /opt/wasi-sdk
#
# install wabt
ARG WABT_FILE="wabt-${WABT_VER}-ubuntu.tar.gz"
COPY ${WABT_FILE} /opt
RUN cd /opt \
&& tar zxf ${WABT_FILE} \
&& rm ${WABT_FILE} \
&& ln -sf /opt/wabt-${WABT_VER} /opt/wabt
#
# install cmake
ARG CMAKE_FILE="cmake-${CMAKE_VER}-Linux-x86_64.sh"
COPY ${CMAKE_FILE} /tmp
RUN cd /tmp \
&& chmod a+x ${CMAKE_FILE} \
&& mkdir /opt/cmake \
&& ./${CMAKE_FILE} --prefix=/opt/cmake --skip-license \
&& ln -sf /opt/cmake/bin/cmake /usr/local/bin/cmake
#
# install emsdk
RUN cd /opt \
&& git clone https://github.com/emscripten-core/emsdk.git \
&& cd emsdk \
&& git pull \
&& ./emsdk install 2.0.12 \
&& ./emsdk activate 2.0.12 \
&& echo "source /opt/emsdk/emsdk_env.sh" >> /root/.bashrc
#
# install binaryen
ARG BINARYEN_FILE="binaryen-${BINARYEN_VER}-x86_64-linux.tar.gz"
COPY ${BINARYEN_FILE} /opt
RUN cd /opt \
&& tar zxf ${BINARYEN_FILE} \
&& rm ${BINARYEN_FILE} \
&& ln -sf /opt/binaryen-${BINARYEN_VER} /opt/binaryen
#
# install bazelisk
ARG BAZEL_FILE="bazelisk-linux-amd64"
COPY ${BAZEL_FILE} /opt/bazelisk/bin/bazelisk
RUN cd /opt/bazelisk/bin/ \
&& chmod a+x bazelisk \
&& ln -sf /opt/bazelisk/bin/bazelisk /usr/local/bin/bazel
RUN apt update \
&& apt install -y python2.7-minimal
#
# Clean up
RUN apt-get autoremove -y \
&& apt-get clean -y \
&& rm -rf /var/lib/apt/lists/* \
&& rm -rf /tmp/*
VOLUME /data
#
#
RUN touch /build.sh \
&& echo "\
#!/bin/bash \n\
if [[ -d /data/project/build ]]; then \n\
rm -r /data/project/build \n\
fi \n\
mkdir /data/project/build \n\
cd /data/project/build \n\
source /opt/emsdk/emsdk_env.sh \n\
cmake .. \n\
make \n\
cd - > /dev/null" > /build.sh \
&& chmod a+x /build.sh

View File

@ -0,0 +1,33 @@
#!/usr/bin/env bash
#
# Copyright (C) 2019 Intel Corporation. All rights reserved.
# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
#
readonly SCRIPT_PATH=$(dirname "$(realpath "$0")")
readonly ROOT=$(realpath "${SCRIPT_PATH}"/../../../)
readonly CURRENT_PATH=$(pwd)
readonly CURRENT_RELATIVE_ROOT=$(realpath --relative-base ${ROOT} ${CURRENT_PATH})
readonly VARIANT=$(lsb_release -c | awk '{print $2}')
docker build \
--build-arg VARIANT=${VARIANT} \
--memory 4G --cpu-quota 50000 \
-t wamr_dev_${VARIANT}:0.1 -f "${ROOT}"/ci/Dockerfile "${ROOT}"/ci &&
docker run --rm -it \
--memory 4G \
--cpus ".5" \
--name workload_build_env \
--mount type=bind,source="${ROOT}",target=/workspace \
wamr_dev_${VARIANT}:0.1 \
/bin/bash -c "\
pwd \
&& pushd ${CURRENT_RELATIVE_ROOT} \
&& rm -rf build \
&& mkdir build \
&& pushd build \
&& cmake .. \
&& cmake --build . --config Release \
&& popd \
&& popd \
&& echo 'Go and find out results under ${CURRENT_RELATIVE_ROOT}/build' "

View File

@ -1,53 +0,0 @@
#!/usr/bin/env bash
#
# Copyright (C) 2019 Intel Corporation. All rights reserved.
# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
#
readonly BUILD_CONTENT="/tmp/build_content"
if [[ ! -d ${BUILD_CONTENT} ]]; then
mkdir ${BUILD_CONTENT}
fi
readonly WASI_SDK_VER=12
readonly WABT_VER=1.0.23
readonly CMAKE_VER=3.16.2
readonly BINARYEN_VER=version_101
readonly BAZELISK_VER=1.7.5
cd ${BUILD_CONTENT} || exit
if [[ ! -f wasi-sdk-${WASI_SDK_VER}.0-linux.tar.gz ]]; then
wget https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-${WASI_SDK_VER}/wasi-sdk-${WASI_SDK_VER}.0-linux.tar.gz
fi
if [[ ! -f wabt-${WABT_VER}-ubuntu.tar.gz ]]; then
wget https://github.com/WebAssembly/wabt/releases/download/${WABT_VER}/wabt-${WABT_VER}-ubuntu.tar.gz
fi
if [[ ! -f cmake-${CMAKE_VER}-Linux-x86_64.sh ]]; then
wget https://github.com/Kitware/CMake/releases/download/v${CMAKE_VER}/cmake-${CMAKE_VER}-Linux-x86_64.sh
fi
if [[ ! -f binaryen-${BINARYEN_VER}-x86_64-linux.tar.gz ]]; then
wget https://github.com/WebAssembly/binaryen/releases/download/${BINARYEN_VER}/binaryen-${BINARYEN_VER}-x86_64-linux.tar.gz
fi
if [[ ! -f bazelisk-linux-amd64 ]]; then
wget https://github.com/bazelbuild/bazelisk/releases/download/v${BAZELISK_VER}/bazelisk-linux-amd64
fi
cd - > /dev/null || exit
DOCKERFILE_PATH=$(dirname "$(realpath "$0")")
docker build \
--build-arg WASI_SDK_VER=${WASI_SDK_VER} \
--build-arg WABT_VER=${WABT_VER} \
--build-arg CMAKE_VER=${CMAKE_VER} \
--build-arg BINARYEN_VER=${BINARYEN_VER} \
-t wamr_workload_env:0.1 -f "${DOCKERFILE_PATH}"/Dockerfile ${BUILD_CONTENT} \
&& docker run --rm \
--name workload_w_clang \
--mount type=bind,source="$(pwd)",target=/data/project \
-w /data/project \
wamr_workload_env:0.1 \
/bin/bash -c /build.sh

View File

@ -21,9 +21,10 @@ ExternalProject_Add(codecbench
&& ${CMAKE_COMMAND} -E echo "Applying patch"
&& git apply ${CMAKE_CURRENT_SOURCE_DIR}/codecbench.patch
CONFIGURE_COMMAND ${CMAKE_COMMAND}
-DWASI_SDK_PREFIX=${WASI_SDK_HOME}/wasi-sdk
-DCMAKE_TOOLCHAIN_FILE=${WASI_SDK_HOME}/wasi-sdk/share/cmake/wasi-sdk.cmake
${CMAKE_CURRENT_SOURCE_DIR}/meshoptimizer
BUILD_COMMAND make codecbench.opt
INSTALL_COMMAND ${CMAKE_COMMAND} -E copy ./codecbench.opt.wasm ${CMAKE_CURRENT_SOURCE_DIR}/build/codecbench.wasm
-DWASI_SDK_PREFIX=${WASI_SDK_HOME}
-DCMAKE_TOOLCHAIN_FILE=${WASI_SDK_HOME}/share/cmake/wasi-sdk.cmake
-DCMAKE_SYSROOT=${WASI_SDK_HOME}/share/wasi-sysroot
${CMAKE_CURRENT_SOURCE_DIR}/meshoptimizer
BUILD_COMMAND make codecbench
INSTALL_COMMAND ${CMAKE_COMMAND} -E copy ./codecbench.wasm ${CMAKE_CURRENT_SOURCE_DIR}/build/codecbench.wasm
)

View File

@ -0,0 +1 @@
../docker/build_workload.sh

View File

@ -1,8 +1,8 @@
diff --git a/CMakeLists.txt b/CMakeLists.txt
index ffdb4da..a397427 100644
index f4378ce..9bc104b 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -127,3 +127,43 @@ install(FILES
@@ -129,3 +129,43 @@ install(FILES
${CMAKE_CURRENT_BINARY_DIR}/meshoptimizerConfig.cmake
${CMAKE_CURRENT_BINARY_DIR}/meshoptimizerConfigVersion.cmake
DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/meshoptimizer)
@ -46,3 +46,74 @@ index ffdb4da..a397427 100644
+)
+
+add_dependencies(codecbench.opt codecbench)
diff --git a/src/vertexcodec.cpp b/src/vertexcodec.cpp
index 5f3ec20..b79bfad 100644
--- a/src/vertexcodec.cpp
+++ b/src/vertexcodec.cpp
@@ -81,13 +81,13 @@
#endif
#ifdef SIMD_WASM
-#define wasmx_splat_v32x4(v, i) wasm_v32x4_shuffle(v, v, i, i, i, i)
-#define wasmx_unpacklo_v8x16(a, b) wasm_v8x16_shuffle(a, b, 0, 16, 1, 17, 2, 18, 3, 19, 4, 20, 5, 21, 6, 22, 7, 23)
-#define wasmx_unpackhi_v8x16(a, b) wasm_v8x16_shuffle(a, b, 8, 24, 9, 25, 10, 26, 11, 27, 12, 28, 13, 29, 14, 30, 15, 31)
-#define wasmx_unpacklo_v16x8(a, b) wasm_v16x8_shuffle(a, b, 0, 8, 1, 9, 2, 10, 3, 11)
-#define wasmx_unpackhi_v16x8(a, b) wasm_v16x8_shuffle(a, b, 4, 12, 5, 13, 6, 14, 7, 15)
-#define wasmx_unpacklo_v64x2(a, b) wasm_v64x2_shuffle(a, b, 0, 2)
-#define wasmx_unpackhi_v64x2(a, b) wasm_v64x2_shuffle(a, b, 1, 3)
+#define wasmx_splat_v32x4(v, i) wasm_i32x4_shuffle(v, v, i, i, i, i)
+#define wasmx_unpacklo_v8x16(a, b) wasm_i8x16_shuffle(a, b, 0, 16, 1, 17, 2, 18, 3, 19, 4, 20, 5, 21, 6, 22, 7, 23)
+#define wasmx_unpackhi_v8x16(a, b) wasm_i8x16_shuffle(a, b, 8, 24, 9, 25, 10, 26, 11, 27, 12, 28, 13, 29, 14, 30, 15, 31)
+#define wasmx_unpacklo_v16x8(a, b) wasm_i16x8_shuffle(a, b, 0, 8, 1, 9, 2, 10, 3, 11)
+#define wasmx_unpackhi_v16x8(a, b) wasm_i16x8_shuffle(a, b, 4, 12, 5, 13, 6, 14, 7, 15)
+#define wasmx_unpacklo_v64x2(a, b) wasm_i64x2_shuffle(a, b, 0, 2)
+#define wasmx_unpackhi_v64x2(a, b) wasm_i64x2_shuffle(a, b, 1, 3)
#endif
namespace meshopt
@@ -700,7 +700,7 @@ static v128_t decodeShuffleMask(unsigned char mask0, unsigned char mask1)
v128_t sm1 = wasm_v128_load(&kDecodeBytesGroupShuffle[mask1]);
v128_t sm1off = wasm_v128_load(&kDecodeBytesGroupCount[mask0]);
- sm1off = wasm_v8x16_shuffle(sm1off, sm1off, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0);
+ sm1off = wasm_i8x16_shuffle(sm1off, sm1off, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0);
v128_t sm1r = wasm_i8x16_add(sm1, sm1off);
@@ -751,7 +751,7 @@ static const unsigned char* decodeBytesGroupSimd(const unsigned char* data, unsi
v128_t shuf = decodeShuffleMask(mask0, mask1);
- v128_t result = wasm_v128_bitselect(wasm_v8x16_swizzle(rest, shuf), sel, mask);
+ v128_t result = wasm_v128_bitselect(wasm_i8x16_swizzle(rest, shuf), sel, mask);
wasm_v128_store(buffer, result);
@@ -773,7 +773,7 @@ static const unsigned char* decodeBytesGroupSimd(const unsigned char* data, unsi
v128_t shuf = decodeShuffleMask(mask0, mask1);
- v128_t result = wasm_v128_bitselect(wasm_v8x16_swizzle(rest, shuf), sel, mask);
+ v128_t result = wasm_v128_bitselect(wasm_i8x16_swizzle(rest, shuf), sel, mask);
wasm_v128_store(buffer, result);
diff --git a/src/vertexfilter.cpp b/src/vertexfilter.cpp
index 023452c..2374cf7 100644
--- a/src/vertexfilter.cpp
+++ b/src/vertexfilter.cpp
@@ -56,10 +56,10 @@
#endif
#ifdef SIMD_WASM
-#define wasmx_unpacklo_v16x8(a, b) wasm_v16x8_shuffle(a, b, 0, 8, 1, 9, 2, 10, 3, 11)
-#define wasmx_unpackhi_v16x8(a, b) wasm_v16x8_shuffle(a, b, 4, 12, 5, 13, 6, 14, 7, 15)
-#define wasmx_unziplo_v32x4(a, b) wasm_v32x4_shuffle(a, b, 0, 2, 4, 6)
-#define wasmx_unziphi_v32x4(a, b) wasm_v32x4_shuffle(a, b, 1, 3, 5, 7)
+#define wasmx_unpacklo_v16x8(a, b) wasm_i16x8_shuffle(a, b, 0, 8, 1, 9, 2, 10, 3, 11)
+#define wasmx_unpackhi_v16x8(a, b) wasm_i16x8_shuffle(a, b, 4, 12, 5, 13, 6, 14, 7, 15)
+#define wasmx_unziplo_v32x4(a, b) wasm_i32x4_shuffle(a, b, 0, 2, 4, 6)
+#define wasmx_unziphi_v32x4(a, b) wasm_i32x4_shuffle(a, b, 1, 3, 5, 7)
#endif
namespace meshopt

View File

@ -1 +0,0 @@
../docker/docker_build.sh

View File

@ -5,8 +5,6 @@
#
readonly BUILD_CONTENT="/tmp/build_content"
readonly WASI_SDK_VER=12
readonly WASI_SDK_FILE="wasi-sdk-${WASI_SDK_VER}.0-linux.tar.gz"
readonly WABT_VER=1.0.23
readonly WABT_FILE="wabt-${WABT_VER}-ubuntu.tar.gz"
readonly CMAKE_VER=3.16.2
@ -28,17 +26,6 @@ function install_deps() {
build-essential git tree zip unzip
}
#
# install wasi-sdk
function install_wasi-sdk() {
if [[ ! -f ${WASI_SDK_FILE} ]]; then
wget https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-${WASI_SDK_VER}/${WASI_SDK_FILE}
fi
tar zxf ${WASI_SDK_FILE} -C /opt
ln -sf /opt/wasi-sdk-${WASI_SDK_VER}.0 /opt/wasi-sdk
}
#
# install wabt
function install_wabt() {
@ -70,8 +57,8 @@ function install_emsdk() {
git clone https://github.com/emscripten-core/emsdk.git
cd emsdk
git pull
./emsdk install 2.0.12
./emsdk activate 2.0.12
./emsdk install 2.0.26
./emsdk activate 2.0.26
echo "source /opt/emsdk/emsdk_env.sh" >> "${HOME}"/.bashrc
}

View File

@ -5,8 +5,8 @@ This sample demonstrates how to build [tensorflow](https://github.com/tensorflow
```bash
git clone https://github.com/emscripten-core/emsdk.git
cd emsdk
./emsdk install 2.0.12
./emsdk activate 2.0.12
./emsdk install 2.0.26
./emsdk activate 2.0.26
```
And set up ensdk environment:
```bash

View File

@ -17,7 +17,7 @@ fi
set -xe
EMSDK_WASM_DIR="${EMSDK}/upstream/emscripten/cache/wasm"
EMSDK_WASM_DIR="${EMSDK}/upstream/emscripten/cache/sysroot/lib/wasm32-emscripten"
BUILD_SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
OUT_DIR="${BUILD_SCRIPT_DIR}/out"
TENSORFLOW_DIR="${BUILD_SCRIPT_DIR}/tensorflow"

View File

@ -5,8 +5,10 @@ cmake_minimum_required (VERSION 2.8...3.16)
project(testavx)
include(${CMAKE_CURRENT_SOURCE_DIR}/../../cmake/preparation.cmake)
# a workaround to let aom find our non-public headers
include_directories(${CMAKE_CURRENT_SOURCE_DIR}/../include/libc)
include_directories(${WASI_SDK_HOME}/share/wasi-sysroot/include/libc/musl)
################ AOM ################
set(ENABLE_CCACHE ON)
@ -58,15 +60,6 @@ target_link_libraries(${PROJECT_NAME}
add_dependencies(${PROJECT_NAME} aom)
find_program(WASM_OPT
NAMES wasm-opt
PATHS /opt/binaryen-version_97/bin /opt/binaryen/bin
)
if (NOT WASM_OPT)
message(WARNING "can not find wasm-opt and will not optimize any wasm module")
endif()
add_custom_target(${PROJECT_NAME}_opt ALL
COMMAND
${WASM_OPT} -Oz --enable-simd -o ${PROJECT_NAME}.opt.wasm ${PROJECT_NAME}.wasm

View File

@ -10,27 +10,6 @@ include(${CMAKE_CURRENT_SOURCE_DIR}/../cmake/preparation.cmake)
#######################################
include(ExternalProject)
################ HEADERS ################
set(EMSDK_SYSTEM_HEADERS "$ENV{EMSDK}/upstream/emscripten/system/include")
set(EMSDK_LIBC_HEADERS "${EMSDK_SYSTEM_HEADERS}/libc")
ExternalProject_Add(headers_from_emcc
PREFIX headers
SOURCE_DIR "${EMSDK_SYSTEM_HEADERS}"
CONFIGURE_COMMAND ""
BUILD_COMMAND ""
INSTALL_COMMAND ${CMAKE_COMMAND} -E make_directory ${CMAKE_CURRENT_SOURCE_DIR}/include/pthread/sys
&& ${CMAKE_COMMAND} -E make_directory ${CMAKE_CURRENT_SOURCE_DIR}/include/libc/bits
# copy emscripten pthread related header files
&& ${CMAKE_COMMAND} -E copy ${EMSDK_LIBC_HEADERS}/pthread.h ${CMAKE_CURRENT_SOURCE_DIR}/include/pthread/
&& ${CMAKE_COMMAND} -E copy ${EMSDK_LIBC_HEADERS}/signal.h ${CMAKE_CURRENT_SOURCE_DIR}/include/pthread/
&& ${CMAKE_COMMAND} -E copy ${EMSDK_LIBC_HEADERS}/netdb.h ${CMAKE_CURRENT_SOURCE_DIR}/include/pthread/
&& ${CMAKE_COMMAND} -E copy ${EMSDK_LIBC_HEADERS}/sys/wait.h ${CMAKE_CURRENT_SOURCE_DIR}/include/pthread/sys/
&& ${CMAKE_COMMAND} -E copy ${EMSDK_LIBC_HEADERS}/sys/socket.h ${CMAKE_CURRENT_SOURCE_DIR}/include/pthread/sys/
# copy emscripten setjmp headers
&& ${CMAKE_COMMAND} -E copy ${EMSDK_LIBC_HEADERS}/setjmp.h ${CMAKE_CURRENT_SOURCE_DIR}/include/libc/setjmp.h
&& ${CMAKE_COMMAND} -E copy ${EMSDK_SYSTEM_HEADERS}/../lib/libc/musl/arch/emscripten/bits/setjmp.h ${CMAKE_CURRENT_SOURCE_DIR}/include/libc/bits/setjmp.h
)
################ av1 ################
ExternalProject_Add(av1
PREFIX av1
@ -39,15 +18,15 @@ ExternalProject_Add(av1
GIT_PROGRESS ON
GIT_SHALLOW ON
SOURCE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/av1
DEPENDS headers_from_emcc
UPDATE_COMMAND git clean -fd && git checkout -- *
&& ${CMAKE_COMMAND} -E echo "Copying pre-installed CMakeLists.txt"
&& ${CMAKE_COMMAND} -E copy ${CMAKE_CURRENT_SOURCE_DIR}/CMakeLists.avx_wasm.txt CMakeLists.txt
&& git apply ../av1-clang.patch
CONFIGURE_COMMAND ${CMAKE_COMMAND}
-DWASI_SDK_PREFIX=${WASI_SDK_HOME}/wasi-sdk
-DCMAKE_TOOLCHAIN_FILE=${WASI_SDK_HOME}/wasi-sdk/share/cmake/wasi-sdk.cmake
${CMAKE_CURRENT_SOURCE_DIR}/av1
-DWASI_SDK_PREFIX=${WASI_SDK_HOME}
-DCMAKE_TOOLCHAIN_FILE=${WASI_SDK_HOME}/share/cmake/wasi-sdk.cmake
-DCMAKE_SYSROOT=${WASI_SDK_HOME}/share/wasi-sysroot
${CMAKE_CURRENT_SOURCE_DIR}/av1
BUILD_COMMAND make testavx_opt
INSTALL_COMMAND ${CMAKE_COMMAND} -E copy testavx.opt.wasm ${CMAKE_CURRENT_BINARY_DIR}/testavx.wasm
)

View File

@ -17,7 +17,7 @@ fi
set -xe
EMSDK_WASM_DIR="${EMSDK}/upstream/emscripten/cache/wasm"
EMSDK_WASM_DIR="${EMSDK}/upstream/emscripten/cache/sysroot/lib/wasm32-emscripten"
BUILD_SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
OUT_DIR="${BUILD_SCRIPT_DIR}/out"
WASM_AV1_DIR="${BUILD_SCRIPT_DIR}/wasm-av1"

View File

@ -0,0 +1 @@
../docker/build_workload.sh

View File

@ -1 +0,0 @@
../docker/docker_build.sh