Update sample workload wasm-av1 and add workload XNNPACK (#443)

README.md

@@ -115,6 +115,7 @@ The WAMR [samples](./samples) integrate the iwasm VM core, application manager a
 - **[spawn-thread](./samples/spawn-thread)**: Demonstrating how to execute wasm functions of the same wasm application concurrently, in threads created by host embedder or runtime, but not the wasm application itself.
 - **[multi-module](./samples/multi-module)**: Demonstrating the [multiple modules as dependencies](./doc/multi_module.md) feature which implements the [load-time dynamic linking](https://webassembly.org/docs/dynamic-linking/).
 - **[wasm-c-api](./samples/wasm-c-api/README.md)**: Demonstrating how to run some samples from [wasm-c-api proposal](https://github.com/WebAssembly/wasm-c-api) and showing the supported API's.
+- **[workload](./samples/workload/README.md)**: Demonstrating how to build and run some complex workloads, e.g. tensorflow-lite, XNNPACK, wasm-av1, meshoptimizer and bwa.
 
 
 License

core/iwasm/aot/aot_reloc.h

@@ -36,6 +36,8 @@ typedef struct {
     REG_SYM(aot_call_indirect),           \
     REG_SYM(wasm_runtime_enlarge_memory), \
     REG_SYM(wasm_runtime_set_exception),  \
+    REG_SYM(memset),                      \
+    REG_SYM(memmove),                     \
     REG_BULK_MEMORY_SYM()                 \
     REG_ATOMIC_WAIT_SYM()
 #else /* else of (defined(_WIN32) || defined(_WIN32_)) && defined(NDEBUG) */
@@ -45,6 +47,8 @@ typedef struct {
     REG_SYM(aot_call_indirect),           \
     REG_SYM(wasm_runtime_enlarge_memory), \
     REG_SYM(wasm_runtime_set_exception),  \
+    REG_SYM(memset),                      \
+    REG_SYM(memmove),                     \
     REG_SYM(fmin),                        \
     REG_SYM(fminf),                       \
     REG_SYM(fmax),                        \

core/iwasm/libraries/libc-builtin/libc_builtin_wrapper.c

@@ -1007,6 +1007,21 @@ __cxa_throw_wrapper(wasm_exec_env_t exec_env,
     wasm_runtime_set_exception(module_inst, buf);
 }
 
+static int
+setjmp_wrapper(wasm_exec_env_t exec_env,
+               void *jmp_buf)
+{
+    os_printf("in setjmp()\n");
+    return 0;
+}
+
+static void
+longjmp_wrapper(wasm_exec_env_t exec_env,
+               void *jmp_buf, int val)
+{
+    os_printf("in longjmp()\n");
+}
+
 #if WASM_ENABLE_SPEC_TEST != 0
 static void
 print_wrapper(wasm_exec_env_t exec_env)
@@ -1104,7 +1119,9 @@ static NativeSymbol native_symbols_libc_builtin[] = {
     REG_NATIVE_FUNC(nullFunc_X, "(i)"),
     REG_NATIVE_FUNC(__cxa_allocate_exception, "(i)i"),
     REG_NATIVE_FUNC(__cxa_begin_catch, "(*)"),
-    REG_NATIVE_FUNC(__cxa_throw, "(**i)")
+    REG_NATIVE_FUNC(__cxa_throw, "(**i)"),
+    REG_NATIVE_FUNC(setjmp, "(*)i"),
+    REG_NATIVE_FUNC(longjmp, "(*i)"),
 };
 
 #if WASM_ENABLE_SPEC_TEST != 0

core/iwasm/libraries/libc-emcc/libc_emcc_wrapper.c

@@ -11,18 +11,27 @@
 #define get_module_inst(exec_env) \
     wasm_runtime_get_module_inst(exec_env)
 
+#define validate_app_addr(offset, size) \
+    wasm_runtime_validate_app_addr(module_inst, offset, size)
+
+#define validate_app_str_addr(offset) \
+    wasm_runtime_validate_app_str_addr(module_inst, offset)
+
 #define validate_native_addr(addr, size) \
     wasm_runtime_validate_native_addr(module_inst, addr, size)
 
+#define addr_app_to_native(offset) \
+    wasm_runtime_addr_app_to_native(module_inst, offset)
+
+#define addr_native_to_app(ptr) \
+    wasm_runtime_addr_native_to_app(module_inst, ptr)
+
 #define module_malloc(size, p_native_addr) \
     wasm_runtime_module_malloc(module_inst, size, p_native_addr)
 
 #define module_free(offset) \
     wasm_runtime_module_free(module_inst, offset)
 
-#define REG_NATIVE_FUNC(func_name, signature)  \
-    { #func_name, func_name##_wrapper, signature, NULL }
-
 extern bool
 wasm_runtime_call_indirect(wasm_exec_env_t exec_env,
                            uint32 element_idx,
@@ -282,12 +291,15 @@ fopen_wrapper(wasm_exec_env_t exec_env,
     int file_id;
 
     if (pathname == NULL || mode == NULL)
-        return -1;
+        return 0;
 
     if ((file_id = get_free_file_slot()) == -1)
-        return -1;
+        return 0;
 
     file = fopen(pathname, mode);
+    if (!file)
+        return 0;
+
     file_list[file_id] = file;
     return file_id + 1;
 }
@@ -308,6 +320,22 @@ fread_wrapper(wasm_exec_env_t exec_env,
     return (uint32)fread(ptr, size, nmemb, file);
 }
 
+static int
+fseeko_wrapper(wasm_exec_env_t exec_env,
+               int file_id, int64 offset, int whence)
+{
+    FILE *file;
+
+    file_id = file_id - 1;
+    if ((unsigned)file_id >= sizeof(file_list) / sizeof(FILE *)) {
+        return -1;
+    }
+    if ((file = file_list[file_id]) == NULL) {
+        return -1;
+    }
+    return (uint32)fseek(file, offset, whence);
+}
+
 static uint32
 emcc_fwrite_wrapper(wasm_exec_env_t exec_env,
                     const void *ptr, uint32 size, uint32 nmemb,
@@ -351,6 +379,113 @@ fclose_wrapper(wasm_exec_env_t exec_env, int file_id)
     file_list[file_id] = NULL;
     return fclose(file);
 }
+
+static int
+__sys_mkdir_wrapper(wasm_exec_env_t exec_env,
+                    const char *pathname, int mode)
+{
+    if (!pathname)
+        return -1;
+    return mkdir(pathname, mode);
+}
+
+static int
+__sys_rmdir_wrapper(wasm_exec_env_t exec_env, const char *pathname)
+{
+    if (!pathname)
+        return -1;
+    return rmdir(pathname);
+}
+
+static int
+__sys_unlink_wrapper(wasm_exec_env_t exec_env, const char *pathname)
+{
+    if (!pathname)
+        return -1;
+    return unlink(pathname);
+}
+
+static uint32
+__sys_getcwd_wrapper(wasm_exec_env_t exec_env, char *buf, uint32 size)
+{
+    wasm_module_inst_t module_inst = get_module_inst(exec_env);
+    char *ret;
+
+    if (!buf)
+        return -1;
+
+    ret = getcwd(buf, size);
+    return ret ? addr_native_to_app(ret) : 0;
+}
+
+#include <sys/utsname.h>
+
+struct utsname_app {
+    char sysname[64];
+    char nodename[64];
+    char release[64];
+    char version[64];
+    char machine[64];
+    char domainname[64];
+};
+
+static int
+__sys_uname_wrapper(wasm_exec_env_t exec_env, struct utsname_app *uname_app)
+{
+    wasm_module_inst_t module_inst = get_module_inst(exec_env);
+    struct utsname uname_native = { 0 };
+    uint32 length;
+
+    if (!validate_native_addr(uname_app, sizeof(struct utsname_app)))
+        return -1;
+
+    if (uname(&uname_native) != 0) {
+        return -1;
+    }
+
+    memset(uname_app, 0, sizeof(struct utsname_app));
+
+    length = strlen(uname_native.sysname);
+    if (length > sizeof(uname_app->sysname) - 1)
+        length = sizeof(uname_app->sysname) - 1;
+    bh_memcpy_s(uname_app->sysname, sizeof(uname_app->sysname),
+                uname_native.sysname, length);
+
+    length = strlen(uname_native.nodename);
+    if (length > sizeof(uname_app->nodename) - 1)
+        length = sizeof(uname_app->nodename) - 1;
+    bh_memcpy_s(uname_app->nodename, sizeof(uname_app->nodename),
+                uname_native.nodename, length);
+
+    length = strlen(uname_native.release);
+    if (length > sizeof(uname_app->release) - 1)
+        length = sizeof(uname_app->release) - 1;
+    bh_memcpy_s(uname_app->release, sizeof(uname_app->release),
+                uname_native.release, length);
+
+    length = strlen(uname_native.version);
+    if (length > sizeof(uname_app->version) - 1)
+        length = sizeof(uname_app->version) - 1;
+    bh_memcpy_s(uname_app->version, sizeof(uname_app->version),
+                uname_native.version, length);
+
+#ifdef _GNU_SOURCE
+    length = strlen(uname_native.domainname);
+    if (length > sizeof(uname_app->domainname) - 1)
+        length = sizeof(uname_app->domainname) - 1;
+    bh_memcpy_s(uname_app->domainname, sizeof(uname_app->domainname),
+                uname_native.domainname, length);
+#endif
+
+    return 0;
+}
+
+static void
+emscripten_notify_memory_growth_wrapper(wasm_exec_env_t exec_env, int i)
+{
+    (void)i;
+}
+
 #endif /* end of BH_PLATFORM_LINUX_SGX */
 
 #define REG_NATIVE_FUNC(func_name, signature)  \
@@ -374,9 +509,16 @@ static NativeSymbol native_symbols_libc_emcc[] = {
 #if !defined(BH_PLATFORM_LINUX_SGX)
     REG_NATIVE_FUNC(fopen, "($$)i"),
     REG_NATIVE_FUNC(fread, "(*iii)i"),
+    REG_NATIVE_FUNC(fseeko, "(iIi)i"),
     REG_NATIVE_FUNC(emcc_fwrite, "(*iii)i"),
     REG_NATIVE_FUNC(feof, "(i)i"),
     REG_NATIVE_FUNC(fclose, "(i)i"),
+    REG_NATIVE_FUNC(__sys_mkdir, "($i)i"),
+    REG_NATIVE_FUNC(__sys_rmdir, "($)i"),
+    REG_NATIVE_FUNC(__sys_unlink, "($)i"),
+    REG_NATIVE_FUNC(__sys_getcwd, "(*~)i"),
+    REG_NATIVE_FUNC(__sys_uname, "(*)i"),
+    REG_NATIVE_FUNC(emscripten_notify_memory_growth, "(i)"),
 #endif /* end of BH_PLATFORM_LINUX_SGX */
 };
 

samples/workload/XNNPACK/CMakeLists.txt

@@ -0,0 +1,102 @@
+# Copyright (C) 2019 Intel Corporation. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+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        2da0de89960b829c6fae74204a102db524e73047
+    GIT_PROGRESS   ON
+    SOURCE_DIR     ${CMAKE_CURRENT_SOURCE_DIR}/xnnpack
+    UPDATE_COMMAND git checkout .bazelrc BUILD.bazel emscripten.bzl
+                   && git apply ${CMAKE_CURRENT_SOURCE_DIR}/xnnpack.patch
+                   && cmake -E copy_directory ${CMAKE_CURRENT_SOURCE_DIR}/toolchain ${CMAKE_CURRENT_SOURCE_DIR}/xnnpack/toolchain
+    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
+                        #--sandbox_debug
+    INSTALL_COMMAND ${CMAKE_COMMAND} -E create_symlink
+                      ${CMAKE_CURRENT_SOURCE_DIR}/xnnpack/bazel-out/wasm-opt/bin/
+                      ${CMAKE_CURRENT_SOURCE_DIR}/build/wasm-opt
+)

samples/workload/XNNPACK/README.md

@@ -0,0 +1,60 @@
+"XNNPACK" sample introduction
+==============
+This sample demonstrates how to build [XNNPACK](https://github.com/google/XNNPACK) benchmarks into WebAssembly with emcc toolchain and run them with iwasm.
+
+## 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
+```
+
+## Build XNNPACK
+
+```bash
+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_SIMD=1 -DWAMR_BUILD_LIBC_EMCC=1 -DWAMR_BUILD_LIB_PTHREAD=1
+$ make
+```
+
+And please build wamrc:
+
+``` bash
+cd <wamr-dir>/wamr-compiler
+./build_llvm.sh
+mkdir build && cd build
+cmake ..
+make
+```
+
+Then compile wasm file to aot file and run:
+
+``` shell
+$ cd <wamr-dir>/samples/workload/XNNPACK/xnnpack/bazel-bin
+$ wamrc --enable-simd -o average_pooling_bench.aot average_pooling_bench.wasm  (or other wasm files)
+$ iwasm average_pooling_bench.aot
+```
+

samples/workload/XNNPACK/toolchain/BUILD.bazel

@@ -0,0 +1,30 @@
+# 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,
+)

samples/workload/XNNPACK/toolchain/emscripten_toolchain_config.bzl

@@ -0,0 +1,137 @@
+# 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",
+                            ],
+                        ),
+                    ]),
+                ),
+            ],
+        ),
+    ]
+
+    return cc_common.create_cc_toolchain_config_info(
+        ctx = ctx,
+        features = features, # NEW
+        cxx_builtin_include_directories = [
+            "/opt/emsdk/upstream/emscripten/system/include/libcxx",
+            "/opt/emsdk/upstream/emscripten/system/lib/libcxxabi/include",
+            "/opt/emsdk/upstream/emscripten/system/include",
+            "/opt/emsdk/upstream/emscripten/system/include/libc",
+            "/opt/emsdk/upstream/emscripten/system/lib/libc/musl/arch/emscripten",
+            "/opt/emsdk/upstream/lib/clang/12.0.0/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],
+)

samples/workload/XNNPACK/xnnpack.patch

@@ -0,0 +1,742 @@
+diff --git a/.bazelrc b/.bazelrc
+index ea28201..ffd4ed4 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
++
++# 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
+--- 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",
+     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",
+     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(
+ )
+ 
+ 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(
+ )
+ 
+ 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(
+ )
+ 
+ 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",
+diff --git a/emscripten.bzl b/emscripten.bzl
+index faad087..2b4763f 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():
+     """Emscripten-specific linkopts for benchmarks."""
+     return [
+         "-s ASSERTIONS=1",
+-        "-s ERROR_ON_UNDEFINED_SYMBOLS=1",
+-        "-s EXIT_RUNTIME=1",
++        "-s ERROR_ON_UNDEFINED_SYMBOLS=0",
+         "-s ALLOW_MEMORY_GROWTH=1",
+         "-s TOTAL_MEMORY=268435456",  # 256M
+-        "--pre-js $(location :preamble.js.lds)",
+     ]
+ 
+ def xnnpack_emscripten_deps():

samples/workload/cmake/toolchain.cmake

@@ -59,6 +59,8 @@ add_compile_options(
   $<$<CONFIG:Debug>:-v>
 )
 
+# need users to create their own additional include files
+
 ################  AR  ################
 find_program(LLVM_AR NAMES llvm-ar llvm-ar-11 REQUIRED)
 

samples/workload/docker/Dockerfile

@@ -13,6 +13,7 @@ ARG WASI_SDK_VER=11.0
 ARG WABT_VER=1.0.19
 ARG CMAKE_VER=3.16.2
 ARG BINARYEN_VER=version_97
+ARG BAZEL_VER=3.7.0
 
 #
 # install wasi-sdk
@@ -66,6 +67,16 @@ RUN cd /opt \
     && rm ${BINARYEN_FILE}  \
     && ln -sf /opt/binaryen-${BINARYEN_VER} /opt/binaryen
 
+RUN apt install -y unzip zip
+
+#
+# install bazel
+ARG BAZEL_FILE=bazel-${BAZEL_VER}-installer-linux-x86_64.sh
+COPY ${BAZEL_FILE} /tmp
+RUN cd /tmp \
+      && chmod a+x ${BAZEL_FILE} \
+      && ./${BAZEL_FILE}
+
 #
 # Clean up
 RUN apt-get autoremove -y \

samples/workload/docker/build.sh

@@ -13,6 +13,7 @@ WASI_SDK_VER=11.0
 WABT_VER=1.0.19
 CMAKE_VER=3.16.2
 BINARYEN_VER=version_97
+BAZEL_VER=3.7.0
 
 cd build_scripts
 if [[ ! -f wasi-sdk-${WASI_SDK_VER}-linux.tar.gz ]]; then
@@ -34,6 +35,10 @@ fi
 if [[ ! -f binaryen-${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 bazel-${BAZEL_VER}-installer-linux-x86_64.sh ]]; then
+  wget https://github.com/bazelbuild/bazel/releases/download/${BAZEL_VER}/bazel-${BAZEL_VER}-installer-linux-x86_64.sh
+fi
 cd -
 
 docker build \
@@ -45,4 +50,5 @@ docker build \
   --build-arg WABT_VER=${WABT_VER} \
   --build-arg CMAKE_VER=${CMAKE_VER} \
   --build-arg BINARYEN_VER=${BINARYEN_VER} \
+  --build-arg BAZEL_VER=${BAZEL_VER} \
   -t clang_env:0.1 -f Dockerfile build_scripts

samples/workload/wasm-av1/.gitignore

@@ -0,0 +1,8 @@
+# from CMakeLists
+av1
+build
+include
+
+# from build.sh
+wasm-av1
+out

samples/workload/wasm-av1/CMakeLists.avx_wasm.txt

@@ -0,0 +1,79 @@
+# Copyright (C) 2019 Intel Corporation. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+cmake_minimum_required (VERSION 3.0)
+
+project(testavx)
+
+# a workaround to let aom find our non-public headers
+include_directories(${CMAKE_CURRENT_SOURCE_DIR}/../include/libc)
+
+################  AOM  ################
+set(ENABLE_CCACHE ON)
+set(ENABLE_DOCS OFF CACHE BOOL "ENABLE_DOCS" FORCE)
+set(ENABLE_EXAMPLES OFF CACHE BOOL "ENABLE_EXAMPLES" FORCE)
+set(ENABLE_NEON OFF CACHE BOOL "ENABLE_EXAMPLES" FORCE)
+set(ENABLE_NEON_ASM OFF CACHE BOOL "ENABLE_EXAMPLES" FORCE)
+set(ENABLE_VSX OFF CACHE BOOL "ENABLE_EXAMPLES" FORCE)
+set(ENABLE_MMX OFF CACHE BOOL "ENABLE_EXAMPLES" FORCE)
+set(AOM_TARGET_CPU generic)
+set(CONFIG_ACCOUNTING 1 CACHE NUMBER "" FORCE)
+set(CONFIG_INSPECTION 1 CACHE NUMBER "" FORCE)
+set(CONFIG_MULTITHREAD 0 CACHE NUMBER "" FORCE)
+set(CONFIG_RUNTIME_CPU_DETECT 0 CACHE NUMBER "" FORCE)
+set(CONFIG_UNIT_TESTS 0 CACHE NUMBER "" FORCE)
+set(CONFIG_WEBM_IO 0 CACHE NUMBER "" FORCE)
+add_subdirectory(third_party/aom third_party/aom/bin EXCLUDE_FROM_ALL)
+
+################  AV ################
+add_executable(${PROJECT_NAME}
+  test.c
+  decode-av1.c
+)
+
+target_include_directories(${PROJECT_NAME}
+  PRIVATE
+    third_party/aom/
+    ${CMAKE_CURRENT_BINARY_DIR}/third_party/aom/bin
+)
+
+set_target_properties(${PROJECT_NAME}
+  PROPERTIES
+    OUTPUT_NAME ${PROJECT_NAME}.wasm
+)
+
+target_link_options(${PROJECT_NAME}
+  PRIVATE
+    LINKER:--allow-undefined
+    LINKER:--export=__heap_base
+    LINKER:--export=__data_end
+    LINKER:--initial-memory=33554432
+    LINKER:-z,stack-size=25165824
+)
+
+target_link_libraries(${PROJECT_NAME}
+  PRIVATE
+    aom
+)
+
+add_dependencies(${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
+  BYPRODUCTS
+    ${CMAKE_CURRENT_BINARY_DIR}/${PROJECT_NAME}.opt.wasm
+  WORKING_DIRECTORY
+    ${CMAKE_CURRENT_BINARY_DIR}
+)
+
+add_dependencies(${PROJECT_NAME}_opt ${PROJECT_NAME})

samples/workload/wasm-av1/CMakeLists.txt

@@ -0,0 +1,61 @@
+# Copyright (C) 2019 Intel Corporation. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+cmake_minimum_required (VERSION 3.0)
+
+project(av1_wasm)
+
+################  BINARYEN ################
+find_program(WASM_OPT
+    NAMES wasm-opt
+    PATHS /opt/binaryen-version_97/bin /opt/binaryen/bin
+)
+
+if (NOT WASM_OPT)
+  message(FATAL_ERROR
+    "can not find wasm-opt. "
+    "please download it from "
+    "https://github.com/WebAssembly/binaryen/releases/download/version_97/binaryen-version_97-x86_64-linux.tar.gz "
+    "and install it under /opt"
+  )
+endif()
+
+#######################################
+include(ExternalProject)
+
+################ HEADERS ################
+ExternalProject_Add(headers_from_emcc
+   PREFIX headers
+   SOURCE_DIR "$ENV{EMSDK}/upstream/emscripten/system/"
+   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 $ENV{EMSDK}/upstream/emscripten/system/include/libc/pthread.h ${CMAKE_CURRENT_SOURCE_DIR}/include/pthread/
+                     && ${CMAKE_COMMAND} -E copy $ENV{EMSDK}/upstream/emscripten/system/include/libc/signal.h ${CMAKE_CURRENT_SOURCE_DIR}/include/pthread/
+                     && ${CMAKE_COMMAND} -E copy $ENV{EMSDK}/upstream/emscripten/system/include/libc/netdb.h ${CMAKE_CURRENT_SOURCE_DIR}/include/pthread/
+                     && ${CMAKE_COMMAND} -E copy $ENV{EMSDK}/upstream/emscripten/system/include/libc/sys/wait.h ${CMAKE_CURRENT_SOURCE_DIR}/include/pthread/sys/
+                     && ${CMAKE_COMMAND} -E copy $ENV{EMSDK}/upstream/emscripten/system/include/libc/sys/socket.h ${CMAKE_CURRENT_SOURCE_DIR}/include/pthread/sys/
+                     # copy emscripten setjmp headers
+                     && ${CMAKE_COMMAND} -E copy $ENV{EMSDK}/upstream/emscripten/system/include/libc/setjmp.h ${CMAKE_CURRENT_SOURCE_DIR}/include/libc/setjmp.h
+                     && ${CMAKE_COMMAND} -E copy $ENV{EMSDK}/upstream/emscripten/system/lib/libc/musl/arch/emscripten/bits/setjmp.h ${CMAKE_CURRENT_SOURCE_DIR}/include/libc/bits/setjmp.h
+)
+
+################ av1 ################
+ExternalProject_Add(av1
+  PREFIX av1
+  GIT_REPOSITORY https://github.com/GoogleChromeLabs/wasm-av1.git
+  GIT_TAG        master
+  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 ${CMAKE_COMMAND} -DCMAKE_TOOLCHAIN_FILE=${CMAKE_CURRENT_SOURCE_DIR}/../cmake/toolchain.cmake ${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
+)

samples/workload/wasm-av1/README.md

@@ -1,22 +1,56 @@
 "wasm-av1" sample introduction
 ==============
-This sample demonstrates how to build [wasm-av1](https://github.com/GoogleChromeLabs/wasm-av1) into WebAssembly with emcc toolchain and run it with iwasm. Please first install [emsdk](https://github.com/emscripten-core/emsdk):
+
-```bash
+This sample demonstrates how to build [wasm-av1](https://github.com/GoogleChromeLabs/wasm-av1) into
-git clone https://github.com/emscripten-core/emsdk.git
+WebAssembly with simd support and run it with iwasm.
-cd emsdk
+
-./emsdk install latest
+## Preparation
-./emsdk activate latest
+
-```
+please refer to [installation instructions](../README.md).
-And set up ensdk environment:
+
-```bash
+## Build with EMSDK
-source emsdk_env.sh
+
-```
+just run the convenience script:
-Then run
+
 ```bash
 ./build.sh
 ```
-to build wasm-av1 and run it with iwasm, which basically contains the following steps:
+
+it is going to build wasm-av1 and run it with iwasm, which basically contains the following steps:
 - hack emcc to delete some objects in libc.a
 - patch wasm-av1 and build it with emcc compiler
 - build iwasm with simd and libc-emcc support
 - run testav1.aot with iwasm
+
+## Or build with clang-11 and wasi-sdk
+
+``` shell
+$ mkdir build && cd build
+$ cmake ..
+$ make
+# to verify
+$ ls testavx.wasm
+```
+
+### Run workload
+
+Firstly please build iwasm with simd support:
+
+``` shell
+$ cd <wamr dir>/product-mini/platforms/linux/
+$ mkdir build && cd build
+$ cmake .. -DWAMR_BUILD_SIMD=1
+$ make
+```
+
+Then compile wasm file to aot file and run:
+
+``` shell
+$ cd <wamr dir>/wamr-compiler/build
+$ ./wamrc --enable-simd -o testavx.aot testavx.wasm
+$ cd <wamr dir>/product-mini/platforms/linux/
+$ # copy sample data like <wamr dir>/samples/workload/wasm-av1/av1/third_party/samples/elephants_dream_480p24.ivf
+$ # copy testavx.aot
+$ # make sure you declare the access priority of the directory in which the sample data is
+$ ./iwasm --dir=. ./testavx.aot ./elephants_dream_480p24.ivf
+```

samples/workload/wasm-av1/av1-clang.patch

@@ -0,0 +1,19 @@
+diff --git a/test.c b/test.c
+index df2d44b..520bf13 100644
+--- a/test.c
++++ b/test.c
+@@ -63,9 +63,14 @@ main(int argc, char *argv[]) {
+                     static int     i = 0;
+                     
+                     ++i;
++                    printf("Decoding frame #%d\n", i);
+                     if (30 <= i && i < 40) {
++                        printf("Dumping frame #%d\n", i);
+                         dump_raw_frame(af, i);
+                     }
++                    if (i >= 1000) {
++                        break;
++                    }
+                 }
+                 /*
+                  * Run the decoder every time, so that we keep