gnh1201/wasm-micro-runtime
1
0
Fork 0
mirror of https://github.com/bytecodealliance/wasm-micro-runtime.git synced 2025-05-09 05:06:17 +00:00
Code Issues Actions 9 Packages Projects Releases Wiki Activity

Upgrade XNNPACK workload (#2394)

Browse Source 
- Sync source code to b9d4073a6913891ce9cbd8965c8d506075d2a45a, which is
  referred by tensorflow
- Upgrade emscripten to 3.1.44
- CMake outputs are .wasm files and .aot files
This commit is contained in:
liang.he 2023-08-01 19:13:00 +08:00 committed by GitHub
parent cb6d85069e
commit ecd4fccc96
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 304 additions and 255 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

315
samples/workload/XNNPACK/CMakeLists.txt
View File

@ -1,147 +1,198 @@
# Copyright (C) 2019 Intel Corporation. All rights reserved. # Copyright (C) 2019 Intel Corporation. All rights reserved.
# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception # SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
cmake_minimum_required (VERSION 3.0) cmake_minimum_required (VERSION 3.14)
project(xnnpack_wasm) project(xnnpack_wasm)
################ EMCC ################ ################ EMCC ################
include(ExternalProject) include(ExternalProject)
ExternalProject_Add(xnnpack # grep xnnpack_benchmark -A 1 BUILD.bazel \
# | grep "name =" \
# | awk '{print $3}' \
# | sed -e 's/\"//g; s/,//g; s/^/\"/g; s/$/\"/g'
list(APPEND NATIVE_BENCHMARKS
"qs8_dwconv_bench"
"qs8_f32_vcvt_bench"
"qs8_gemm_bench"
"qs8_requantization_bench"
"qs8_vadd_bench"
"qs8_vaddc_bench"
"qs8_vcvt_bench"
"qs16_qs8_vcvt_bench"
"qs8_vlrelu_bench"
"qs8_vmul_bench"
"qs8_vmulc_bench"
"qu8_f32_vcvt_bench"
"qu8_gemm_bench"
"qu8_requantization_bench"
"qu8_vadd_bench"
"qu8_vaddc_bench"
"qu8_vcvt_bench"
"qu8_vlrelu_bench"
"qu8_vmul_bench"
"qu8_vmulc_bench"
"bf16_gemm_bench"
"f16_f32acc_igemm_bench"
"f16_igemm_bench"
"f16_f32acc_gemm_bench"
"f16_gemm_bench"
"f16_raddstoreexpminusmax_bench"
"f16_spmm_bench"
"f16_vsigmoid_bench"
"f16_vtanh_bench"
"f16_f32_vcvt_bench"
"f32_igemm_bench"
"f32_conv_hwc_bench"
"f16_conv_hwc2chw_bench"
# "f16_gavgpool_cw_bench"
# "f32_gavgpool_cw_bench"
"f32_conv_hwc2chw_bench"
"f16_dwconv_bench"
"f32_dwconv_bench"
"f32_dwconv2d_chw_bench"
"f16_dwconv2d_chw_bench"
"f32_f16_vcvt_bench"
"xx_transpose_bench"
"x8_transpose_bench"
"x16_transpose_bench"
"x24_transpose_bench"
"x32_transpose_bench"
"x64_transpose_bench"
"f32_bgemm_bench"
"f32_gemm_bench"
"f32_qs8_vcvt_bench"
"f32_qu8_vcvt_bench"
"f32_raddexpminusmax_bench"
"f32_raddextexp_bench"
"f32_raddstoreexpminusmax_bench"
"f32_rmax_bench"
"f32_spmm_bench"
"f32_softmax_bench"
"f16_velu_bench"
"f32_velu_bench"
"f32_vhswish_bench"
"f32_vlrelu_bench"
"f32_vrelu_bench"
"f32_vscaleexpminusmax_bench"
"f32_vscaleextexp_bench"
"f32_vsigmoid_bench"
"f16_vsqrt_bench"
"f32_vsqrt_bench"
"f32_vtanh_bench"
"f32_im2col_gemm_bench"
"rounding_bench"
"s16_rmaxabs_bench"
"s16_window_bench"
"u32_filterbank_accumulate_bench"
"u32_filterbank_subtract_bench"
"u32_vlog_bench"
"u64_u32_vsqrtshift_bench"
"i16_vlshift_bench"
"cs16_vsquareabs_bench"
"cs16_bfly4_bench"
"cs16_fftr_bench"
"x8_lut_bench"
"x32_packw_bench"
"x16_packw_bench"
"abs_bench"
"average_pooling_bench"
"bankers_rounding_bench"
"ceiling_bench"
"channel_shuffle_bench"
"convert_bench"
"convolution_bench"
"deconvolution_bench"
"elu_bench"
"floor_bench"
"global_average_pooling_bench"
"hardswish_bench"
"leaky_relu_bench"
"max_pooling_bench"
"negate_bench"
"prelu_bench"
"sigmoid_bench"
"softmax_bench"
"square_bench"
"square_root_bench"
"tanh_bench"
"truncation_bench"
"f16_dwconv_e2e_bench"
"f16_gemm_e2e_bench"
"f32_dwconv_e2e_bench"
"f32_gemm_e2e_bench"
"qs8_dwconv_e2e_bench"
"qs8_gemm_e2e_bench"
"qu8_gemm_e2e_bench"
"qu8_dwconv_e2e_bench"
# "end2end_bench"
"f16_exp_ulp_eval"
"f16_expminus_ulp_eval"
"f16_expm1minus_ulp_eval"
"f16_sigmoid_ulp_eval"
"f16_sqrt_ulp_eval"
"f16_tanh_ulp_eval"
"f32_exp_ulp_eval"
"f32_expminus_ulp_eval"
"f32_expm1minus_ulp_eval"
"f32_extexp_ulp_eval"
"f32_sigmoid_ulp_eval"
"f32_sqrt_ulp_eval"
"f32_tanh_ulp_eval"
)
# Only Download
ExternalProject_Add(xnnpack-download
PREFIX xnnpack PREFIX xnnpack
GIT_REPOSITORY https://github.com/google/XNNPACK.git GIT_REPOSITORY https://github.com/google/XNNPACK.git
GIT_TAG 4570a7151aa4f3e57eca14a575eeff6bb13e26be GIT_TAG b9d4073a6913891ce9cbd8965c8d506075d2a45a
GIT_PROGRESS ON GIT_PROGRESS ON
SOURCE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/xnnpack SOURCE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/xnnpack
UPDATE_COMMAND git restore . UPDATE_COMMAND ""
&& cmake -E copy ${CMAKE_CURRENT_SOURCE_DIR}/xnnpack/google3/third_party/XNNPACK/microkernels.bzl PATCH_COMMAND git apply ${CMAKE_CURRENT_SOURCE_DIR}/xnnpack.patch
${CMAKE_CURRENT_SOURCE_DIR}/xnnpack/
&& git apply ${CMAKE_CURRENT_SOURCE_DIR}/xnnpack.patch
CONFIGURE_COMMAND "" CONFIGURE_COMMAND ""
# grep xnnpack_benchmark -A 1 BUILD.bazel \ BUILD_COMMAND ""
# | grep "name =" \ INSTALL_COMMAND ""
# | awk '{print $3}' \ TEST_COMMAND ""
# | 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_dwconv_bench.wasm
//:qs8_f32_vcvt_bench.wasm
//:qs8_gemm_bench.wasm
//:qs8_requantization_bench.wasm
//:qs8_vadd_bench.wasm
//:qs8_vaddc_bench.wasm
//:qs8_vcvt_bench.wasm
//:qs8_vlrelu_bench.wasm
//:qs8_vmul_bench.wasm
//:qs8_vmulc_bench.wasm
//:qu8_f32_vcvt_bench.wasm
//:qu8_gemm_bench.wasm
//:qu8_requantization_bench.wasm
//:qu8_vadd_bench.wasm
//:qu8_vaddc_bench.wasm
//:qu8_vcvt_bench.wasm
//:qu8_vlrelu_bench.wasm
//:qu8_vmul_bench.wasm
//:qu8_vmulc_bench.wasm
//:bf16_gemm_bench.wasm
//:f16_igemm_bench.wasm
//:f16_gemm_bench.wasm
//:f16_raddstoreexpminusmax_bench.wasm
//:f16_spmm_bench.wasm
//:f16_vsigmoid_bench.wasm
//:f16_f32_vcvt_bench.wasm
//:f32_igemm_bench.wasm
//:f32_conv_hwc_bench.wasm
//:f16_conv_hwc2chw_bench.wasm
//:f16_gavgpool_cw_bench.wasm
//:f32_gavgpool_cw_bench.wasm
//:f32_conv_hwc2chw_bench.wasm
//:f16_dwconv_bench.wasm
//:f32_dwconv_bench.wasm
//:f32_dwconv2d_chw_bench.wasm
//:f16_dwconv2d_chw_bench.wasm
//:f32_f16_vcvt_bench.wasm
//:xx_transpose_bench.wasm
//:x8_transpose_bench.wasm
//:x16_transpose_bench.wasm
//:x24_transpose_bench.wasm
//:x32_transpose_bench.wasm
//:x64_transpose_bench.wasm
//:f32_gemm_bench.wasm
//:f32_qs8_vcvt_bench.wasm
//:f32_qu8_vcvt_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
//:f16_velu_bench.wasm
//:f32_velu_bench.wasm
//:f32_vhswish_bench.wasm
//:f32_vlrelu_bench.wasm
//:f32_vrelu_bench.wasm
//:f32_vscaleexpminusmax_bench.wasm
//:f32_vscaleextexp_bench.wasm
//:f32_vsigmoid_bench.wasm
//:f16_vsqrt_bench.wasm
//:f32_vsqrt_bench.wasm
//:f32_im2col_gemm_bench.wasm
//:rounding_bench.wasm
//:s16_rmaxabs_bench.wasm
//:s16_window_bench.wasm
//:u32_filterbank_accumulate_bench.wasm
//:u32_filterbank_subtract_bench.wasm
//:u32_vlog_bench.wasm
//:u64_u32_vsqrtshift_bench.wasm
//:i16_vlshift_bench.wasm
//:cs16_vsquareabs_bench.wasm
//:cs16_bfly4_bench.wasm
//:cs16_fftr_bench.wasm
//:x8_lut_bench.wasm
//:abs_bench.wasm
//:average_pooling_bench.wasm
//:bankers_rounding_bench.wasm
//:ceiling_bench.wasm
//:channel_shuffle_bench.wasm
//:convert_bench.wasm
//:convolution_bench.wasm
//:deconvolution_bench.wasm
//:elu_bench.wasm
//:floor_bench.wasm
//:global_average_pooling_bench.wasm
//:hardswish_bench.wasm
//:leaky_relu_bench.wasm
//:max_pooling_bench.wasm
//:negate_bench.wasm
//:sigmoid_bench.wasm
//:prelu_bench.wasm
//:softmax_bench.wasm
//:square_bench.wasm
//:square_root_bench.wasm
//:truncation_bench.wasm
//:f16_gemm_e2e_bench.wasm
//:f32_dwconv_e2e_bench.wasm
//:f32_gemm_e2e_bench.wasm
//:qs8_dwconv_e2e_bench.wasm
//:qs8_gemm_e2e_bench.wasm
//:qu8_gemm_e2e_bench.wasm
//:qu8_dwconv_e2e_bench.wasm
//:end2end_bench.wasm
//:f16_exp_ulp_eval.wasm
//:f16_expminus_ulp_eval.wasm
//:f16_expm1minus_ulp_eval.wasm
//:f16_sigmoid_ulp_eval.wasm
//:f16_sqrt_ulp_eval.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
//:f32_tanh_ulp_eval.wasm
INSTALL_COMMAND ${CMAKE_COMMAND} -E copy_directory
${CMAKE_CURRENT_SOURCE_DIR}/xnnpack/bazel-out/wasm-opt/bin/
${CMAKE_BINARY_DIR}/wasm-opt
) )
set(WAMRC "${CMAKE_CURRENT_SOURCE_DIR}/../../../wamr-compiler/build/wamrc")
if(EXISTS ${WAMRC})
message("-- Will generate .aot")
else()
message("Will generate .wasm")
endif()
foreach(BENCHMARK IN LISTS NATIVE_BENCHMARKS)
string(CONCAT WASM_BENCHMARK "//:" ${BENCHMARK} "-wasm")
string(CONCAT WASM_OUTPUT ${BENCHMARK} ".wasm")
add_custom_command(
OUTPUT ${WASM_OUTPUT}
COMMAND bazel --output_user_root=build-user-output build -c opt --config=wasm ${WASM_BENCHMARK}
&& ${CMAKE_COMMAND} -E copy_if_different ./bazel-bin/${WASM_OUTPUT} ${CMAKE_CURRENT_BINARY_DIR}/${WASM_OUTPUT}
WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}/xnnpack
DEPENDS xnnpack-download
COMMENT "Generating ${WASM_OUTPUT} ..."
)
set_property(DIRECTORY APPEND PROPERTY ADDITIONAL_CLEAN_FILES ${CMAKE_CURRENT_BINARY_DIR}/${WASM_OUTPUT})
if(EXISTS ${WAMRC})
string(CONCAT AOT_OUTPUT ${BENCHMARK} ".aot")
add_custom_command(
OUTPUT ${AOT_OUTPUT}
COMMAND ${WAMRC} -o ${AOT_OUTPUT} ${WASM_OUTPUT}
WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
DEPENDS ${WASM_OUTPUT}
COMMENT "Generating ${AOT_OUTPUT} ..."
)
add_custom_target(${BENCHMARK} ALL DEPENDS ${AOT_OUTPUT})
else()
add_custom_target(${BENCHMARK} ALL DEPENDS ${WASM_OUTPUT})
endif()
endforeach()

51
samples/workload/XNNPACK/README.md
View File

@ -9,26 +9,7 @@ please refer to [installation instructions](../README.md).
## Build XNNPACK ## Build XNNPACK
```bash please build wamrc:
cd <wamr-dir>/samples/workload/XNNPACK
mkdir build
cd build
cmake ..
```
The wasm files are generated under folder samples/workload/XNNPACK/xnnpack/bazel-bin.
## Run benchmarks
Firstly please build iwasm with simd, libc-emcc and lib-pthread support:
``` bash
$ cd <wamr-dir>/product-mini/platforms/linux/
$ mkdir build && cd build
$ cmake .. -DWAMR_BUILD_LIBC_EMCC=1 -DWAMR_BUILD_LIB_PTHREAD=1
$ make
```
And please build wamrc:
``` bash ``` bash
cd <wamr-dir>/wamr-compiler cd <wamr-dir>/wamr-compiler
@ -38,11 +19,31 @@ cmake ..
make make
``` ```
Then compile wasm file to aot file and run: And then build xnnpack standalone wasm files
``` shell ```bash
$ cd <wamr-dir>/samples/workload/XNNPACK/xnnpack/bazel-bin $ cd <wamr-dir>/samples/workload/XNNPACK
$ wamrc -o average_pooling_bench.aot average_pooling_bench.wasm (or other wasm files) $ cmake -S . -B build
$ iwasm average_pooling_bench.aot $ cmake --build build
```
Generated .wasm(and .aot) files are under *samples/workload/XNNPACK/build*.
## Run benchmarks
Firstly please build iwasm with simd, libc-emcc and lib-pthread supporting:
``` bash
$ cd <wamr-dir>/product-mini/platforms/linux/
$ mkdir build && cd build
$ cmake .. -DWAMR_BUILD_LIBC_EMCC=1 -DWAMR_BUILD_LIB_PTHREAD=1
$ make
```
Then run:
``` shell
$ cd <wamr-dir>/samples/workload/XNNPACK/build
$ iwasm averag_pooling_bench.aot # (or other aot files)
``` ```

193
samples/workload/XNNPACK/xnnpack.patch
View File

@ -1,141 +1,138 @@
diff --git a/.bazelrc b/.bazelrc diff --git a/.bazelrc b/.bazelrc
index 688279da1..376996885 100644 index fcaff1063..e61d53337 100644
--- a/.bazelrc --- a/.bazelrc
+++ b/.bazelrc +++ b/.bazelrc
@@ -53,4 +53,9 @@ build:ios_fat --watchos_cpus=armv7k @@ -1,6 +1,7 @@
build:macos --apple_platform_type=macos # Basic build settings
build --jobs 128
build --cxxopt='-std=gnu++14'
+build --incompatible_enable_cc_toolchain_resolution
# Sets the default Apple platform to macOS.
build --apple_platform_type=macos
@@ -55,3 +56,10 @@ build:macos --apple_platform_type=macos
build:macos_arm64 --config=macos build:macos_arm64 --config=macos
-build:macos_arm64 --cpu=darwin_arm64 build:macos_arm64 --cpu=darwin_arm64
\ No newline at end of file
+build:macos_arm64 --cpu=darwin_arm64
+ +
+# Emscripten configs
+build:wasm --copt="-Wno-unused"
+build:wasm --copt="-Wno-unused-function"
+build:wasm --copt="-Wno-unused-but-set-variable"
+build:wasm --cpu=wasm +build:wasm --cpu=wasm
+build:wasm --features=wasm_simd +build:wasm --features=wasm_simd
+build:wasm --crosstool_top=@emsdk//emscripten_toolchain:everything
+build:wasm --host_crosstool_top=@bazel_tools//tools/cpp:toolchain
diff --git a/WORKSPACE b/WORKSPACE diff --git a/WORKSPACE b/WORKSPACE
index cd8960ffa..787e03ca8 100644 index 2e568088b..3961371ca 100644
--- a/WORKSPACE --- a/WORKSPACE
+++ b/WORKSPACE +++ b/WORKSPACE
@@ -29,8 +29,9 @@ http_archive( @@ -83,7 +83,23 @@ http_archive(
# Google Benchmark library, used in micro-benchmarks.
http_archive(
name = "com_google_benchmark",
- strip_prefix = "benchmark-main",
- urls = ["https://github.com/google/benchmark/archive/main.zip"],
+ sha256 = "1ba14374fddcd9623f126b1a60945e4deac4cdc4fb25a5f25e7f779e36f2db52",
+ strip_prefix = "benchmark-d2a8a4ee41b923876c034afb939c4fc03598e622",
+ urls = ["https://github.com/google/benchmark/archive/d2a8a4ee41b923876c034afb939c4fc03598e622.zip"],
) )
# FP16 library, used for half-precision conversions # Android NDK location and version is auto-detected from $ANDROID_NDK_HOME environment variable
@@ -92,8 +93,25 @@ http_archive( -android_ndk_repository(name = "androidndk")
], +# android_ndk_repository(name = "androidndk")
)
+load("@bazel_tools//tools/build_defs/repo:http.bzl", "http_archive") # Android SDK location and API is auto-detected from $ANDROID_HOME environment variable
-android_sdk_repository(name = "androidsdk")
+# android_sdk_repository(name = "androidsdk")
+
+http_archive( +http_archive(
+ name = "emsdk", + name = "emsdk",
+ # Use emsdk-3.0.0 since the larger version may: + sha256 = "5fa6f5eb45a4d50264610c4c9e1c155535359b63bfaad69b4e5101d16c1e7e32",
+ # - compress the wasm file into a tar file but not directly generate wasm file + strip_prefix = "emsdk-a896e3d066448b3530dbcaa48869fafefd738f57/bazel",
+ # - generate incomplete implementation of libc API, e.g. throw exception in getentropy + url = "https://github.com/emscripten-core/emsdk/archive/a896e3d066448b3530dbcaa48869fafefd738f57.tar.gz",
+ strip_prefix = "emsdk-3.0.0/bazel",
+ url = "https://github.com/emscripten-core/emsdk/archive/refs/tags/3.0.0.tar.gz",
+ sha256 = "a41dccfd15be9e85f923efaa0ac21943cbab77ec8d39e52f25eca1ec61a9ac9e"
+) +)
+ +
+load("@emsdk//:deps.bzl", emsdk_deps = "deps") +load("@emsdk//:deps.bzl", emsdk_deps = "deps")
+emsdk_deps() +emsdk_deps()
+ +
+load("@emsdk//:emscripten_deps.bzl", emsdk_emscripten_deps = "emscripten_deps") +load("@emsdk//:emscripten_deps.bzl", emsdk_emscripten_deps = "emscripten_deps")
+emsdk_emscripten_deps() +emsdk_emscripten_deps(emscripten_version = "3.1.44")
+ +
# Android NDK location and version is auto-detected from $ANDROID_NDK_HOME environment variable +load("@emsdk//:toolchains.bzl", "register_emscripten_toolchains")
-android_ndk_repository(name = "androidndk") +register_emscripten_toolchains()
+#android_ndk_repository(name = "androidndk") diff --git a/bench/utils.cc b/bench/utils.cc
index 3b32503a7..656845336 100644
--- a/bench/utils.cc
+++ b/bench/utils.cc
@@ -456,3 +456,13 @@ CodeMemoryHelper::~CodeMemoryHelper() {
# Android SDK location and API is auto-detected from $ANDROID_HOME environment variable } // namespace utils
-android_sdk_repository(name = "androidsdk") } // namespace benchmark
+#android_sdk_repository(name = "androidsdk") +
+
+extern "C"
+__attribute__((import_module("env"), import_name("getentropy"))) int import_getentropy(void* buffer, size_t length);
+
+extern "C"
+int getentropy(void* buffer, size_t length)
+{
+ return import_getentropy(buffer, length);
+}
diff --git a/build_defs.bzl b/build_defs.bzl diff --git a/build_defs.bzl b/build_defs.bzl
index b8217a18d..6f2d1675e 100644 index 01b436eb7..2738fd50a 100644
--- a/build_defs.bzl --- a/build_defs.bzl
+++ b/build_defs.bzl +++ b/build_defs.bzl
@@ -380,7 +380,7 @@ def xnnpack_benchmark(name, srcs, copts = [], deps = [], tags = []): @@ -1,6 +1,7 @@
explicitly specified. """Build definitions and rules for XNNPACK."""
"""
native.cc_binary( -load(":emscripten.bzl", "xnnpack_emscripten_benchmark_linkopts", "xnnpack_emscripten_deps", "xnnpack_emscripten_minimal_linkopts", "xnnpack_emscripten_test_linkopts")
- name = name, +load(":emscripten.bzl", "xnnpack_emscripten_benchmark_linkopts", "xnnpack_emscripten_deps", "xnnpack_emscripten_minimal_linkopts", "xnnpack_emscripten_test_linkopts", "xnnpack_emscripten_standalone_benchmark_linkopts")
+ name = name + ".wasm", +load("@emsdk//emscripten_toolchain:wasm_rules.bzl", "wasm_cc_binary")
srcs = srcs,
copts = xnnpack_std_cxxopts() + [ def xnnpack_visibility():
"-Iinclude", """Visibility of :XNNPACK target.
@@ -405,5 +405,5 @@ def xnnpack_benchmark(name, srcs, copts = [], deps = [], tags = []): @@ -393,7 +394,8 @@ def xnnpack_benchmark(name, srcs, copts = [], deps = [], tags = []):
"//conditions:default": ["-Wno-unused-function"],
}) + copts,
linkopts = select({
- ":emscripten": xnnpack_emscripten_benchmark_linkopts(),
+ ":emscripten": xnnpack_emscripten_standalone_benchmark_linkopts(),
+ ":emscripten_wasmsimd": xnnpack_emscripten_standalone_benchmark_linkopts(),
":windows_x86_64_mingw": ["-lshlwapi"],
":windows_x86_64_msys": ["-lshlwapi"],
"//conditions:default": [],
@@ -405,5 +407,16 @@ def xnnpack_benchmark(name, srcs, copts = [], deps = [], tags = []):
":emscripten": xnnpack_emscripten_deps(), ":emscripten": xnnpack_emscripten_deps(),
"//conditions:default": [], "//conditions:default": [],
}), }),
- tags = tags, - tags = tags,
+ tags = tags, + tags = tags,
+ )
+
+ wasm_cc_binary(
+ name = name + "-wasm",
+ cc_target = ":" + name,
+ threads = "off",
+ simd = True,
+ standalone= True,
+ outputs = [
+ name + ".wasm",
+ ]
) )
diff --git a/emscripten.bzl b/emscripten.bzl diff --git a/emscripten.bzl b/emscripten.bzl
index f1557a7b1..7f964a094 100644 index f1557a7b1..a3c4f93b9 100644
--- a/emscripten.bzl --- a/emscripten.bzl
+++ b/emscripten.bzl +++ b/emscripten.bzl
@@ -25,12 +25,19 @@ def xnnpack_emscripten_benchmark_linkopts(): @@ -33,6 +33,21 @@ def xnnpack_emscripten_benchmark_linkopts():
"""Emscripten-specific linkopts for benchmarks.""" "--pre-js $(location :preamble.js.lds)",
return [ ]
"-s ASSERTIONS=1",
- "-s ENVIRONMENT=node,shell,web", +def xnnpack_emscripten_standalone_benchmark_linkopts():
- "-s ERROR_ON_UNDEFINED_SYMBOLS=1", + return [
- "-s EXIT_RUNTIME=1", + "-s ASSERTIONS=1",
+ "-s ERROR_ON_UNDEFINED_SYMBOLS=0", + "-s ERROR_ON_UNDEFINED_SYMBOLS=0",
"-s ALLOW_MEMORY_GROWTH=1", + "-s ALLOW_MEMORY_GROWTH=1",
"-s TOTAL_MEMORY=536870912", # 512M + "-s TOTAL_MEMORY=536870912", # 512M
- "--pre-js $(location :preamble.js.lds)",
+ "-s USE_PTHREADS=0", + "-s USE_PTHREADS=0",
+ "-s STANDALONE_WASM=1", + "-s STANDALONE_WASM=1",
+ "-Wno-unused",
+ "-Wno-unused-variable",
+ "-Wno-unused-command-line-argument",
+ "-Wl,--export=__heap_base", + "-Wl,--export=__heap_base",
+ "-Wl,--export=__data_end", + "-Wl,--export=__data_end",
+ "-Wl,--export=malloc", + "-Wl,--export=malloc",
+ "-Wl,--export=free", + "-Wl,--export=free",
+ "--oformat=wasm", + ]
] +
+
def xnnpack_emscripten_deps(): def xnnpack_emscripten_deps():
diff --git a/src/log.c b/src/log.c """Emscripten-specific dependencies for unit tests and benchmarks."""
index 5715f2f85..4b3e4261b 100644 return [
--- a/src/log.c
+++ b/src/log.c
@@ -55,7 +55,7 @@
#endif
#if XNN_LOG_TO_STDIO
-static void xnn_vlog(int output_handle, const char* prefix, size_t prefix_length, const char* format, va_list args) {
+void xnn_vlog(int output_handle, const char* prefix, size_t prefix_length, const char* format, va_list args) {
char stack_buffer[XNN_LOG_STACK_BUFFER_SIZE];
char* heap_buffer = NULL;
char* out_buffer = &stack_buffer[0];
diff --git a/third_party/cpuinfo.BUILD b/third_party/cpuinfo.BUILD
index 1997f4e3a..5e03c43af 100644
--- a/third_party/cpuinfo.BUILD
+++ b/third_party/cpuinfo.BUILD
@@ -150,7 +150,7 @@ cc_library(
"src/arm/midr.h",
],
deps = [
- "@clog",
+ "//deps/clog"
],
)
@@ -352,5 +352,5 @@ config_setting(
config_setting(
name = "emscripten",
- values = {"crosstool_top": "//toolchain:emscripten"},
+ values = {"crosstool_top": "@emsdk//emscripten_toolchain:everything"},
)