christoph/wamr
1
Fork 0
Code Releases Activity

Implement more wasm-c-apis and enable Envoy integration (#622)

Browse Source 
Implement more wasm-c-api APIs to support Envoy integration:
- sync up with latest c-api definition
- change CMakeLists to export necessary headers and install the static library of iwasm
- enable to export tables and memories
- support memorytype and tabletype APIs
- update wasm-c-api sampels
- enable to export importtype APIs

And refine bazel scripts for sample XNNPACK workload, add license headers for sample simple.

Signed-off-by: Wenyong Huang <wenyong.huang@intel.com>
This commit is contained in:
Wenyong Huang
2021-04-27 17:05:40 +08:00
committed by GitHub
parent a332a49a0d
commit eb29385963
36 changed files with 3190 additions and 1939 deletions
Download Patch File Download Diff File Expand all files Collapse all files

133
samples/workload/XNNPACK/CMakeLists.txt
View File

@ -20,85 +20,72 @@ include(ExternalProject)
ExternalProject_Add(xnnpack
PREFIX xnnpack
GIT_REPOSITORY https://github.com/google/XNNPACK.git
GIT_TAG 2da0de89960b829c6fae74204a102db524e73047
GIT_TAG master
GIT_PROGRESS ON
SOURCE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/xnnpack
UPDATE_COMMAND git checkout .bazelrc BUILD.bazel emscripten.bzl
UPDATE_COMMAND git checkout .
&& git apply ${CMAKE_CURRENT_SOURCE_DIR}/xnnpack.patch
&& cmake -E copy_directory ${CMAKE_CURRENT_SOURCE_DIR}/toolchain ${CMAKE_CURRENT_SOURCE_DIR}/xnnpack/toolchain
# replace string "$ENV{HOME}" with actual home directory
&& sed -i "s|\$ENV{HOME}|$ENV{HOME}|g" ${CMAKE_CURRENT_SOURCE_DIR}/xnnpack/toolchain/emscripten_toolchain_config.bzl
CONFIGURE_COMMAND ""
BUILD_COMMAND cd ${CMAKE_CURRENT_SOURCE_DIR}/xnnpack
&& bazel build -c opt --sandbox_writable_path=$ENV{HOME} --config=emscripten_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_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
//: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
//:fp32_mobilenet_v1.wasm
//:fp16_mobilenet_v1.wasm
//:qs8_mobilenet_v1.wasm
//:qs8_mobilenet_v2.wasm
//:fp32_mobilenet_v2.wasm
//:fp16_mobilenet_v2.wasm
//:fp32_mobilenet_v3_large.wasm
//:fp16_mobilenet_v3_large.wasm
//:fp32_mobilenet_v3_small.wasm
//:fp16_mobilenet_v3_small.wasm
//:f32_dwconv_e2e_bench.wasm
//:f32_gemm_e2e_bench.wasm
//:end2end_bench.wasm
//:f32_exp_eval.wasm
//:f32_expminus_eval.wasm
//:f32_extexp_eval.wasm
//:f32_roundne_eval.wasm
//:f32_roundd_eval.wasm
//:f32_roundu_eval.wasm
//:f32_roundz_eval.wasm
//:f32_sigmoid_eval.wasm
//:f32_sqrt_eval.wasm
&& bazel --output_user_root=build_user_output build -c opt --config=wasm
//:qs8_gemm_bench
//:qs8_requantization_bench
//:qu8_gemm_bench
//:qu8_requantization_bench
//:f16_igemm_bench
//:f16_gemm_bench
//:f16_spmm_bench
//:f32_igemm_bench
//:f16_relu_bench
//:f32_conv_hwc_bench
//:f32_conv_hwc2chw_bench
//:f16_dwconv_bench
//:f32_dwconv_bench
//:f32_dwconv2d_chw_bench
//:f32_gemm_bench
//:f32_hswish_bench
//:f32_raddexpminusmax_bench
//:f32_raddextexp_bench
//:f32_raddstoreexpminusmax_bench
//:f32_relu_bench
//:f32_rmax_bench
//:f32_sigmoid_bench
//:f32_spmm_bench
//:f32_softmax_bench
//:f32_velu_bench
//:f32_vscaleexpminusmax_bench
//:f32_vscaleextexp_bench
//:f32_vsqrt_bench
//:f32_im2col_gemm_bench
//:rounding_bench
//:average_pooling_bench
//:bankers_rounding_bench
//:ceiling_bench
//:channel_shuffle_bench
//:convolution_bench
//:deconvolution_bench
//:elu_bench
//:floor_bench
//:global_average_pooling_bench
//:hardswish_bench
//:max_pooling_bench
//:sigmoid_bench
//:prelu_bench
//:softmax_bench
//:square_root_bench
//:truncation_bench
//:f32_dwconv_e2e_bench
//:f32_gemm_e2e_bench
//:qs8_gemm_e2e_bench
//:end2end_bench
//:f32_exp_ulp_eval
//:f32_expminus_ulp_eval
//:f32_expm1minus_ulp_eval
//:f32_extexp_ulp_eval
//:f32_sigmoid_ulp_eval
//:f32_sqrt_ulp_eval
#--sandbox_debug
INSTALL_COMMAND ${CMAKE_COMMAND} -E create_symlink
INSTALL_COMMAND ${CMAKE_COMMAND} -E copy_directory
${CMAKE_CURRENT_SOURCE_DIR}/xnnpack/bazel-out/wasm-opt/bin/
${CMAKE_CURRENT_SOURCE_DIR}/build/wasm-opt
)

30
samples/workload/XNNPACK/toolchain/BUILD.bazel
View File

@ -1,30 +0,0 @@
# Copyright (C) 2019 Intel Corporation. All rights reserved.
# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
package(default_visibility = ['//visibility:public'])
load(":emscripten_toolchain_config.bzl", "emsdk_toolchain_config")
cc_toolchain_suite(
name = "emscripten",
toolchains = {
"wasm": ":emsdk_toolchain",
},
)
filegroup(name = "empty")
emsdk_toolchain_config(name = "emsdk_toolchain_config")
cc_toolchain(
name = "emsdk_toolchain",
toolchain_identifier = "emsdk-toolchain",
toolchain_config = ":emsdk_toolchain_config",
all_files = ":empty",
compiler_files = ":empty",
dwp_files = ":empty",
linker_files = ":empty",
objcopy_files = ":empty",
strip_files = ":empty",
supports_param_files = 0,
)

142
samples/workload/XNNPACK/toolchain/emscripten_toolchain_config.bzl
View File

@ -1,142 +0,0 @@
# Copyright (C) 2019 Intel Corporation. All rights reserved.
# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
load("@bazel_tools//tools/build_defs/cc:action_names.bzl", "ACTION_NAMES")
load(
"@bazel_tools//tools/cpp:cc_toolchain_config_lib.bzl",
"feature",
"flag_group",
"flag_set",
"tool_path",
)
all_compile_actions = [
ACTION_NAMES.c_compile,
ACTION_NAMES.cpp_compile,
]
all_link_actions = [
ACTION_NAMES.cpp_link_executable,
ACTION_NAMES.cpp_link_dynamic_library,
ACTION_NAMES.cpp_link_nodeps_dynamic_library,
]
def _impl(ctx):
tool_paths = [
tool_path(
name = "gcc",
path = "/opt/emsdk/upstream/emscripten/emcc",
),
tool_path(
name = "ld",
path = "/opt/emsdk/upstream/emscripten/emcc",
),
tool_path(
name = "ar",
path = "/opt/emsdk/upstream/emscripten/emar",
),
tool_path(
name = "cpp",
path = "/opt/emsdk/upstream/emscripten/em++",
),
tool_path(
name = "gcov",
path = "/bin/false",
),
tool_path(
name = "nm",
path = "/bin/false",
),
tool_path(
name = "objdump",
path = "/bin/false",
),
tool_path(
name = "strip",
path = "/bin/false",
),
]
features = [ # NEW
feature(
name = "default_compile_flags",
enabled = True,
flag_sets = [
flag_set(
actions = all_compile_actions,
flag_groups = ([
flag_group(
flags = [
"-O3",
"-msimd128",
"-s",
"USE_PTHREADS=0",
"-s",
"ERROR_ON_UNDEFINED_SYMBOLS=0",
"-s",
"STANDALONE_WASM=1",
],
),
]),
),
],
),
feature(
name = "default_linker_flags",
enabled = True,
flag_sets = [
flag_set(
actions = all_link_actions,
flag_groups = ([
flag_group(
flags = [
"-O3",
"-msimd128",
"-s",
"USE_PTHREADS=0",
"-s",
"ERROR_ON_UNDEFINED_SYMBOLS=0",
"-s",
"STANDALONE_WASM=1",
"-Wl,--export=__heap_base",
"-Wl,--export=__data_end",
"-Wl,--export=malloc",
"-Wl,--export=free",
],
),
]),
),
],
),
]
return cc_common.create_cc_toolchain_config_info(
ctx = ctx,
features = features, # NEW
cxx_builtin_include_directories = [
"/opt/emsdk/upstream/emscripten/system/include",
"/opt/emsdk/upstream/emscripten/system/lib/libc/include",
"/opt/emsdk/upstream/emscripten/system/lib/libcxx/include",
"/opt/emsdk/upstream/emscripten/system/lib/libcxxabi/include",
"/opt/emsdk/upstream/emscripten/system/lib/libc/musl/arch/emscripten",
"/opt/emsdk/upstream/emscripten/system/include/compat",
"/opt/emsdk/upstream/emscripten/cache/sysroot/include",
"/opt/emsdk/upstream/lib/clang/13.0.0/include",
"$ENV{HOME}/.emscripten_cache/sysroot/include",
],
toolchain_identifier = "wasm-emsdk",
host_system_name = "i686-unknown-linux-gnu",
target_system_name = "wasm32-unknown-emscripten",
target_cpu = "wasm32",
target_libc = "unknown",
compiler = "emsdk",
abi_version = "unknown",
abi_libc_version = "unknown",
tool_paths = tool_paths,
)
emsdk_toolchain_config = rule(
implementation = _impl,
attrs = {},
provides = [CcToolchainConfigInfo],
)

805
samples/workload/XNNPACK/xnnpack.patch
View File

@ -1,742 +1,137 @@
diff --git a/.bazelrc b/.bazelrc
index ea28201..ffd4ed4 100644
index ec740f38..2c193244 100644
--- a/.bazelrc
+++ b/.bazelrc
@@ -44,3 +44,7 @@ build:ios_arm64e --watchos_cpus=armv7k
build:ios_fat --config=ios
build:ios_fat --ios_multi_cpus=armv7,arm64
build:ios_fat --watchos_cpus=armv7k
@@ -49,4 +49,10 @@ build:ios_fat --watchos_cpus=armv7k
build:macos --apple_platform_type=macos
build:macos_arm64 --config=macos
-build:macos_arm64 --cpu=darwin_arm64
\ No newline at end of file
+build:macos_arm64 --cpu=darwin_arm64
+
+build:wasm --copt=-msimd128
+build:wasm --cpu=wasm
+build:wasm --crosstool_top=@emsdk//emscripten_toolchain:everything
+build:wasm --host_crosstool_top=@bazel_tools//tools/cpp:toolchain
+
+# WASM configs
+build:emscripten_wasm --cpu=wasm
+build:emscripten_wasm --crosstool_top=//toolchain:emscripten
diff --git a/BUILD.bazel b/BUILD.bazel
index d38ef1e..f261eb5 100644
index ae4108bc..1c11fac2 100644
--- a/BUILD.bazel
+++ b/BUILD.bazel
@@ -3228,13 +3228,19 @@ xnnpack_cc_library(
hdrs = INTERNAL_HDRS,
gcc_copts = xnnpack_gcc_std_copts(),
msvc_copts = xnnpack_msvc_std_copts(),
- wasm_srcs = WASM_UKERNELS,
- wasmsimd_srcs = WASM_UKERNELS + WASMSIMD_UKERNELS,
+ optimized_copts = [
+ "-ffast-math",
+ ],
+ wasm_srcs = WASM_UKERNELS + WASMSIMD_UKERNELS +
+ PSIMD_FASTMATH_UKERNELS + PSIMD_ACCMATH_UKERNELS,
+ wasmsimd_srcs = WASM_UKERNELS + WASMSIMD_UKERNELS +
+ PSIMD_FASTMATH_UKERNELS + PSIMD_ACCMATH_UKERNELS,
deps = [
":tables",
"@FP16",
"@FXdiv",
"@pthreadpool",
+ "@psimd",
],
)
@@ -3247,13 +3253,19 @@ xnnpack_cc_library(
],
gcc_copts = xnnpack_gcc_std_copts(),
msvc_copts = xnnpack_msvc_std_copts(),
- wasm_srcs = WASM_UKERNELS,
- wasmsimd_srcs = WASM_UKERNELS + WASMSIMD_UKERNELS,
+ optimized_copts = [
+ "-ffast-math",
+ ],
+ wasm_srcs = WASM_UKERNELS + WASMSIMD_UKERNELS +
+ PSIMD_FASTMATH_UKERNELS + PSIMD_ACCMATH_UKERNELS,
+ wasmsimd_srcs = WASM_UKERNELS + WASMSIMD_UKERNELS +
+ PSIMD_FASTMATH_UKERNELS + PSIMD_ACCMATH_UKERNELS,
deps = [
":tables",
"@FP16",
"@FXdiv",
"@pthreadpool",
+ "@psimd",
],
)
@@ -4495,7 +4507,7 @@ 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",
@@ -4506,7 +4518,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "qs8_requantization_bench",
+ name = "qs8_requantization_bench.wasm",
srcs = [
"bench/qs8-requantization.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -4516,7 +4528,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "qu8_gemm_bench",
+ name = "qu8_gemm_bench.wasm",
srcs = [
"bench/gemm.h",
"bench/qu8-gemm.cc",
@@ -4527,7 +4539,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "qu8_requantization_bench",
+ name = "qu8_requantization_bench.wasm",
srcs = [
"bench/qu8-requantization.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -4537,11 +4549,11 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f16_igemm_bench",
+ name = "f16_igemm_bench.wasm",
@@ -5038,7 +5038,6 @@ xnnpack_benchmark(
srcs = [
"bench/f16-igemm.cc",
"bench/conv.h",
- "bench/google/conv.h",
+ #"bench/google/conv.h",
"src/xnnpack/AlignedAllocator.h",
] + MICROKERNEL_BENCHMARK_HDRS,
deps = MICROKERNEL_BENCHMARK_DEPS + [
@@ -4551,7 +4563,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f16_gemm_bench",
+ name = "f16_gemm_bench.wasm",
srcs = [
"bench/f16-gemm.cc",
"bench/gemm.h",
@@ -4563,7 +4575,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f16_spmm_bench",
+ name = "f16_spmm_bench.wasm",
srcs = [
"bench/f16-spmm.cc",
"bench/gemm.h",
@@ -4573,7 +4585,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_igemm_bench",
+ name = "f32_igemm_bench.wasm",
srcs = [
"bench/f32-igemm.cc",
"bench/conv.h",
@@ -4586,7 +4598,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f16_relu_bench",
+ name = "f16_relu_bench.wasm",
srcs = [
"bench/f16-relu.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -4595,7 +4607,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_conv_hwc_bench",
+ name = "f32_conv_hwc_bench.wasm",
srcs = [
"bench/f32-conv-hwc.cc",
"bench/dconv.h",
@@ -4607,7 +4619,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_conv_hwc2chw_bench",
+ name = "f32_conv_hwc2chw_bench.wasm",
srcs = [
"bench/f32-conv-hwc2chw.cc",
"bench/dconv.h",
@@ -4619,11 +4631,11 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f16_dwconv_bench",
+ name = "f16_dwconv_bench.wasm",
@@ -5120,7 +5119,6 @@ xnnpack_benchmark(
srcs = [
"bench/f16-dwconv.cc",
"bench/dwconv.h",
- "bench/google/dwconv.h",
+ #"bench/google/dwconv.h",
"src/xnnpack/AlignedAllocator.h",
] + MICROKERNEL_BENCHMARK_HDRS,
deps = MICROKERNEL_BENCHMARK_DEPS + [
@@ -4633,7 +4645,7 @@ xnnpack_benchmark(
)
diff --git a/WORKSPACE b/WORKSPACE
index 4fa1aa2f..86040d42 100644
--- a/WORKSPACE
+++ b/WORKSPACE
@@ -89,3 +89,18 @@ android_ndk_repository(name = "androidndk")
xnnpack_benchmark(
- name = "f32_dwconv_bench",
+ name = "f32_dwconv_bench.wasm",
srcs = [
"bench/f32-dwconv.cc",
"bench/dwconv.h",
@@ -4646,7 +4658,7 @@ xnnpack_benchmark(
)
# 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",
+)
+
+load("@emsdk//:deps.bzl", emsdk_deps = "deps")
+emsdk_deps()
+
+load("@emsdk//:emscripten_deps.bzl", emsdk_emscripten_deps = "emscripten_deps")
+emsdk_emscripten_deps()
+
diff --git a/build_defs.bzl b/build_defs.bzl
index 10345032..0e926fca 100644
--- a/build_defs.bzl
+++ b/build_defs.bzl
@@ -1,6 +1,6 @@
"""Build definitions and rules for XNNPACK."""
xnnpack_benchmark(
- name = "f32_dwconv2d_chw_bench",
+ name = "f32_dwconv2d_chw_bench.wasm",
srcs = [
"bench/f32-dwconv2d-chw.cc",
"bench/dwconv.h",
@@ -4659,7 +4671,7 @@ xnnpack_benchmark(
)
-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")
xnnpack_benchmark(
- name = "f32_gemm_bench",
+ name = "f32_gemm_bench.wasm",
srcs = [
"bench/f32-gemm.cc",
"bench/gemm.h",
@@ -4670,7 +4682,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_hswish_bench",
+ name = "f32_hswish_bench.wasm",
srcs = [
"bench/f32-hswish.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -4679,7 +4691,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_raddexpminusmax_bench",
+ name = "f32_raddexpminusmax_bench.wasm",
srcs = [
"bench/f32-raddexpminusmax.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -4688,7 +4700,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_raddextexp_bench",
+ name = "f32_raddextexp_bench.wasm",
srcs = [
"bench/f32-raddextexp.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -4697,7 +4709,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_raddstoreexpminusmax_bench",
+ name = "f32_raddstoreexpminusmax_bench.wasm",
srcs = [
"bench/f32-raddstoreexpminusmax.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -4706,7 +4718,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_relu_bench",
+ name = "f32_relu_bench.wasm",
srcs = [
"bench/f32-relu.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -4715,7 +4727,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_rmax_bench",
+ name = "f32_rmax_bench.wasm",
srcs = [
"bench/f32-rmax.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -4724,7 +4736,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_sigmoid_bench",
+ name = "f32_sigmoid_bench.wasm",
srcs = [
"bench/f32-sigmoid.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -4733,7 +4745,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_spmm_bench",
+ name = "f32_spmm_bench.wasm",
srcs = [
"bench/f32-spmm.cc",
"bench/gemm.h",
@@ -4743,7 +4755,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_softmax_bench",
+ name = "f32_softmax_bench.wasm",
srcs = [
"bench/f32-softmax.cc",
] + MICROKERNEL_BENCHMARK_HDRS,
@@ -4752,7 +4764,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_vscaleexpminusmax_bench",
+ name = "f32_vscaleexpminusmax_bench.wasm",
srcs = [
"bench/f32-vscaleexpminusmax.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -4761,7 +4773,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_vscaleextexp_bench",
+ name = "f32_vscaleextexp_bench.wasm",
srcs = [
"bench/f32-vscaleextexp.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -4770,7 +4782,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_vsqrt_bench",
+ name = "f32_vsqrt_bench.wasm",
srcs = [
"bench/f32-vsqrt.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -4779,7 +4791,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_im2col_gemm_bench",
+ name = "f32_im2col_gemm_bench.wasm",
srcs = [
"bench/f32-im2col-gemm.cc",
"bench/conv.h",
@@ -4792,7 +4804,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "rounding_bench",
+ name = "rounding_bench.wasm",
srcs = [
"bench/rounding.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -4804,7 +4816,7 @@ 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"],
@@ -4812,7 +4824,7 @@ 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"],
@@ -4820,7 +4832,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "ceiling_bench",
+ name = "ceiling_bench.wasm",
srcs = ["bench/ceiling.cc"],
copts = xnnpack_optional_tflite_copts(),
tags = ["nowin32"],
@@ -4828,13 +4840,13 @@ 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"],
@@ -4842,7 +4854,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "deconvolution_bench",
+ name = "deconvolution_bench.wasm",
srcs = ["bench/deconvolution.cc"],
copts = xnnpack_optional_tflite_copts(),
tags = ["nowin32"],
@@ -4850,7 +4862,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "floor_bench",
+ name = "floor_bench.wasm",
srcs = ["bench/floor.cc"],
copts = xnnpack_optional_tflite_copts(),
tags = ["nowin32"],
@@ -4858,13 +4870,13 @@ 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"],
@@ -4872,13 +4884,13 @@ 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"],
@@ -4886,7 +4898,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "prelu_bench",
+ name = "prelu_bench.wasm",
srcs = ["bench/prelu.cc"],
copts = xnnpack_optional_tflite_copts(),
tags = ["nowin32"],
@@ -4894,7 +4906,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "softmax_bench",
+ name = "softmax_bench.wasm",
srcs = ["bench/softmax.cc"],
copts = xnnpack_optional_tflite_copts(),
tags = ["nowin32"],
@@ -4902,7 +4914,7 @@ 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"],
@@ -4910,7 +4922,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "truncation_bench",
+ name = "truncation_bench.wasm",
srcs = ["bench/truncation.cc"],
deps = OPERATOR_BENCHMARK_DEPS,
)
@@ -4918,7 +4930,7 @@ xnnpack_benchmark(
############################# End-to-end benchmarks ############################
cc_library(
- name = "fp32_mobilenet_v1",
+ name = "fp32_mobilenet_v1.wasm",
srcs = ["models/fp32-mobilenet-v1.cc"],
hdrs = ["models/models.h"],
copts = xnnpack_std_cxxopts(),
@@ -4930,7 +4942,7 @@ cc_library(
)
cc_library(
- name = "fp16_mobilenet_v1",
+ name = "fp16_mobilenet_v1.wasm",
srcs = ["models/fp16-mobilenet-v1.cc"],
hdrs = ["models/models.h"],
copts = xnnpack_std_cxxopts(),
@@ -4943,7 +4955,7 @@ cc_library(
)
cc_library(
- name = "qs8_mobilenet_v1",
+ name = "qs8_mobilenet_v1.wasm",
srcs = ["models/qs8-mobilenet-v1.cc"],
hdrs = ["models/models.h"],
copts = xnnpack_std_cxxopts(),
@@ -4955,7 +4967,7 @@ cc_library(
)
cc_library(
- name = "qs8_mobilenet_v2",
+ name = "qs8_mobilenet_v2.wasm",
srcs = ["models/qs8-mobilenet-v2.cc"],
hdrs = ["models/models.h"],
copts = xnnpack_std_cxxopts(),
@@ -4967,7 +4979,7 @@ cc_library(
)
cc_library(
- name = "fp32_mobilenet_v2",
+ name = "fp32_mobilenet_v2.wasm",
srcs = ["models/fp32-mobilenet-v2.cc"],
hdrs = ["models/models.h"],
copts = xnnpack_std_cxxopts(),
@@ -4979,7 +4991,7 @@ cc_library(
)
cc_library(
- name = "fp16_mobilenet_v2",
+ name = "fp16_mobilenet_v2.wasm",
srcs = ["models/fp16-mobilenet-v2.cc"],
hdrs = ["models/models.h"],
copts = xnnpack_std_cxxopts(),
@@ -4992,7 +5004,7 @@ cc_library(
)
cc_library(
- name = "fp32_mobilenet_v3_large",
+ name = "fp32_mobilenet_v3_large.wasm",
srcs = ["models/fp32-mobilenet-v3-large.cc"],
hdrs = ["models/models.h"],
copts = xnnpack_std_cxxopts(),
@@ -5004,7 +5016,7 @@ cc_library(
)
cc_library(
- name = "fp16_mobilenet_v3_large",
+ name = "fp16_mobilenet_v3_large.wasm",
srcs = ["models/fp16-mobilenet-v3-large.cc"],
hdrs = ["models/models.h"],
copts = xnnpack_std_cxxopts(),
@@ -5017,7 +5029,7 @@ cc_library(
)
cc_library(
- name = "fp32_mobilenet_v3_small",
+ name = "fp32_mobilenet_v3_small.wasm",
srcs = ["models/fp32-mobilenet-v3-small.cc"],
hdrs = ["models/models.h"],
copts = xnnpack_std_cxxopts(),
@@ -5029,7 +5041,7 @@ cc_library(
)
cc_library(
- name = "fp16_mobilenet_v3_small",
+ name = "fp16_mobilenet_v3_small.wasm",
srcs = ["models/fp16-mobilenet-v3-small.cc"],
hdrs = ["models/models.h"],
copts = xnnpack_std_cxxopts(),
@@ -5042,51 +5054,51 @@ cc_library(
)
xnnpack_benchmark(
- name = "f32_dwconv_e2e_bench",
+ name = "f32_dwconv_e2e_bench.wasm",
srcs = [
"bench/f32-dwconv-e2e.cc",
"bench/end2end.h",
] + MICROKERNEL_BENCHMARK_HDRS,
deps = MICROKERNEL_BENCHMARK_DEPS + [
":XNNPACK",
- ":fp32_mobilenet_v1",
- ":fp32_mobilenet_v2",
- ":fp32_mobilenet_v3_large",
- ":fp32_mobilenet_v3_small",
+ ":fp32_mobilenet_v1.wasm",
+ ":fp32_mobilenet_v2.wasm",
+ ":fp32_mobilenet_v3_large.wasm",
+ ":fp32_mobilenet_v3_small.wasm",
],
)
xnnpack_benchmark(
- name = "f32_gemm_e2e_bench",
+ name = "f32_gemm_e2e_bench.wasm",
srcs = [
"bench/f32-gemm-e2e.cc",
"bench/end2end.h",
] + MICROKERNEL_BENCHMARK_HDRS,
deps = MICROKERNEL_BENCHMARK_DEPS + [
":XNNPACK",
- ":fp32_mobilenet_v1",
- ":fp32_mobilenet_v2",
- ":fp32_mobilenet_v3_large",
- ":fp32_mobilenet_v3_small",
+ ":fp32_mobilenet_v1.wasm",
+ ":fp32_mobilenet_v2.wasm",
+ ":fp32_mobilenet_v3_large.wasm",
+ ":fp32_mobilenet_v3_small.wasm",
],
)
xnnpack_benchmark(
- name = "end2end_bench",
+ name = "end2end_bench.wasm",
srcs = ["bench/end2end.cc"],
deps = [
":XNNPACK",
":bench_utils",
- ":fp16_mobilenet_v1",
- ":fp16_mobilenet_v2",
- ":fp16_mobilenet_v3_large",
- ":fp16_mobilenet_v3_small",
- ":fp32_mobilenet_v1",
- ":fp32_mobilenet_v2",
- ":fp32_mobilenet_v3_large",
- ":fp32_mobilenet_v3_small",
- ":qs8_mobilenet_v1",
- ":qs8_mobilenet_v2",
+ ":fp16_mobilenet_v1.wasm",
+ ":fp16_mobilenet_v2.wasm",
+ ":fp16_mobilenet_v3_large.wasm",
+ ":fp16_mobilenet_v3_small.wasm",
+ ":fp32_mobilenet_v1.wasm",
+ ":fp32_mobilenet_v2.wasm",
+ ":fp32_mobilenet_v3_large.wasm",
+ ":fp32_mobilenet_v3_small.wasm",
+ ":qs8_mobilenet_v1.wasm",
+ ":qs8_mobilenet_v2.wasm",
"@pthreadpool",
],
)
@@ -5094,7 +5106,7 @@ xnnpack_benchmark(
#################### Accuracy evaluation for math functions ####################
xnnpack_benchmark(
- name = "f32_exp_eval",
+ name = "f32_exp_eval.wasm",
srcs = [
"eval/f32-exp.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5103,7 +5115,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_expminus_eval",
+ name = "f32_expminus_eval.wasm",
srcs = [
"eval/f32-expminus.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5112,7 +5124,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_extexp_eval",
+ name = "f32_extexp_eval.wasm",
srcs = [
"eval/f32-extexp.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5121,7 +5133,7 @@ xnnpack_benchmark(
)
xnnpack_unit_test(
- name = "f32_roundne_eval",
+ name = "f32_roundne_eval.wasm",
srcs = [
"eval/f32-roundne.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5132,7 +5144,7 @@ xnnpack_unit_test(
)
xnnpack_unit_test(
- name = "f32_roundd_eval",
+ name = "f32_roundd_eval.wasm",
srcs = [
"eval/f32-roundd.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5143,7 +5155,7 @@ xnnpack_unit_test(
)
xnnpack_unit_test(
- name = "f32_roundu_eval",
+ name = "f32_roundu_eval.wasm",
srcs = [
"eval/f32-roundu.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5154,7 +5166,7 @@ xnnpack_unit_test(
)
xnnpack_unit_test(
- name = "f32_roundz_eval",
+ name = "f32_roundz_eval.wasm",
srcs = [
"eval/f32-roundz.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5165,7 +5177,7 @@ xnnpack_unit_test(
)
xnnpack_benchmark(
- name = "f32_sigmoid_eval",
+ name = "f32_sigmoid_eval.wasm",
srcs = [
"eval/f32-sigmoid.cc",
"src/xnnpack/AlignedAllocator.h",
@@ -5174,7 +5186,7 @@ xnnpack_benchmark(
)
xnnpack_benchmark(
- name = "f32_sqrt_eval",
+ name = "f32_sqrt_eval.wasm",
srcs = [
"eval/f32-sqrt.cc",
"src/xnnpack/AlignedAllocator.h",
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": [],
diff --git a/emscripten.bzl b/emscripten.bzl
index faad087..2b4763f 100644
index 0a0caedf..d28afa30 100644
--- a/emscripten.bzl
+++ b/emscripten.bzl
@@ -4,30 +4,25 @@ def xnnpack_emscripten_minimal_linkopts():
"""Minimal Emscripten-specific linkopts for binaries."""
return [
"-s ASSERTIONS=0",
- "-s ERROR_ON_UNDEFINED_SYMBOLS=1",
- "-s EXIT_RUNTIME=1",
+ "-s ERROR_ON_UNDEFINED_SYMBOLS=0",
]
def xnnpack_emscripten_test_linkopts():
"""Emscripten-specific linkopts for unit tests."""
return [
"-s ASSERTIONS=2",
- "-s ERROR_ON_UNDEFINED_SYMBOLS=1",
+ "-s ERROR_ON_UNDEFINED_SYMBOLS=0",
"-s DEMANGLE_SUPPORT=1",
- "-s EXIT_RUNTIME=1",
"-s ALLOW_MEMORY_GROWTH=1",
- "--pre-js $(location :preamble.js.lds)",
]
def xnnpack_emscripten_benchmark_linkopts():
@@ -23,15 +23,28 @@ def xnnpack_emscripten_benchmark_linkopts():
"""Emscripten-specific linkopts for benchmarks."""
return [
"-s ASSERTIONS=1",
- "-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=1",
"-s TOTAL_MEMORY=268435456", # 256M
+ "-s ALLOW_MEMORY_GROWTH=0",
"-s TOTAL_MEMORY=436207616", # 416M
- "--pre-js $(location :preamble.js.lds)",
+ "-s USE_PTHREADS=0",
+ "-s STANDALONE_WASM=1",
+ "-Wno-unused",
+ "-Wl,--export=__heap_base",
+ "-Wl,--export=__data_end",
+ "-Wl,--export=malloc",
+ "-Wl,--export=free",
]
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
+++ b/third_party/cpuinfo.BUILD
@@ -343,5 +343,5 @@ config_setting(
config_setting(
name = "emscripten",
- values = {"crosstool_top": "//toolchain:emscripten"},
+ values = {"crosstool_top": "@emsdk//emscripten_toolchain:everything"},
)