Fix wasm build with WebGPU EP

This commit is contained in:
Yulong Wang 2025-01-10 20:41:59 -08:00
parent a74817ab10
commit 22bcd9a985
22 changed files with 701 additions and 146 deletions

View file

@ -631,78 +631,92 @@ if (onnxruntime_USE_WEBGPU)
URL_HASH SHA1=${DEP_SHA1_dawn}
# All previous patches are merged into the upstream dawn project. We don't need to apply any patches right now.
# if we need to apply patches in the future, we can uncomment the following line.
# PATCH_COMMAND ${Patch_EXECUTABLE} --binary --ignore-whitespace -p1 < ${PROJECT_SOURCE_DIR}/patches/dawn/dawn.patch
PATCH_COMMAND ${Patch_EXECUTABLE} --binary --ignore-whitespace -p1 < ${PROJECT_SOURCE_DIR}/patches/dawn/dawn.patch
)
endif()
if (onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY)
set(DAWN_BUILD_MONOLITHIC_LIBRARY ON CACHE BOOL "" FORCE)
set(DAWN_ENABLE_INSTALL ON CACHE BOOL "" FORCE)
if (onnxruntime_USE_EXTERNAL_DAWN)
message(FATAL_ERROR "onnxruntime_USE_EXTERNAL_DAWN and onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY cannot be enabled at the same time.")
endif()
else()
# use dawn::dawn_native and dawn::dawn_proc instead of the monolithic dawn::webgpu_dawn to minimize binary size
set(DAWN_BUILD_MONOLITHIC_LIBRARY OFF CACHE BOOL "" FORCE)
set(DAWN_ENABLE_INSTALL OFF CACHE BOOL "" FORCE)
endif()
set(DAWN_BUILD_SAMPLES OFF CACHE BOOL "" FORCE)
set(DAWN_ENABLE_NULL OFF CACHE BOOL "" FORCE)
set(DAWN_FETCH_DEPENDENCIES ON CACHE BOOL "" FORCE)
# disable things we don't use
set(DAWN_DXC_ENABLE_ASSERTS_IN_NDEBUG OFF)
set(DAWN_ENABLE_DESKTOP_GL OFF CACHE BOOL "" FORCE)
set(DAWN_ENABLE_OPENGLES OFF CACHE BOOL "" FORCE)
set(DAWN_SUPPORTS_GLFW_FOR_WINDOWING OFF CACHE BOOL "" FORCE)
set(DAWN_USE_GLFW OFF CACHE BOOL "" FORCE)
set(DAWN_USE_WINDOWS_UI OFF CACHE BOOL "" FORCE)
set(DAWN_USE_X11 OFF CACHE BOOL "" FORCE)
if (CMAKE_SYSTEM_NAME STREQUAL "Emscripten")
set(DAWN_EMSCRIPTEN_TOOLCHAIN "${REPO_ROOT}/cmake/external/emsdk/upstream/emscripten" CACHE STRING "" FORCE)
set(TINT_BUILD_TESTS OFF CACHE BOOL "" FORCE)
set(TINT_BUILD_CMD_TOOLS OFF CACHE BOOL "" FORCE)
set(TINT_BUILD_GLSL_WRITER OFF CACHE BOOL "" FORCE)
set(TINT_BUILD_GLSL_VALIDATOR OFF CACHE BOOL "" FORCE)
set(TINT_BUILD_IR_BINARY OFF CACHE BOOL "" FORCE)
set(TINT_BUILD_SPV_READER OFF CACHE BOOL "" FORCE) # don't need. disabling is a large binary size saving
set(TINT_BUILD_WGSL_WRITER ON CACHE BOOL "" FORCE) # needed to create cache key. runtime error if not enabled.
# file "${DAWN_EMSCRIPTEN_TOOLCHAIN}/tools/maint/gen_struct_info.py" is missing from emsdk installation.
# we should copy if from ${PROJECT_SOURCE_DIR}/patches/
file(COPY_FILE
"${PROJECT_SOURCE_DIR}/patches/emscripten/tools/maint/gen_struct_info.py"
"${DAWN_EMSCRIPTEN_TOOLCHAIN}/tools/maint/gen_struct_info.py"
ONLY_IF_DIFFERENT)
else()
if (onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY)
set(DAWN_BUILD_MONOLITHIC_LIBRARY ON CACHE BOOL "" FORCE)
set(DAWN_ENABLE_INSTALL ON CACHE BOOL "" FORCE)
# SPIR-V validation shouldn't be required given we're using Tint to create the SPIR-V.
set(DAWN_ENABLE_SPIRV_VALIDATION OFF CACHE BOOL "" FORCE)
if (WIN32)
# building this requires the HLSL writer to be enabled in Tint. TBD if that we need either of these to be ON.
set(DAWN_USE_BUILT_DXC ON CACHE BOOL "" FORCE)
set(TINT_BUILD_HLSL_WRITER ON CACHE BOOL "" FORCE)
if ((NOT onnxruntime_ENABLE_DAWN_BACKEND_VULKAN) AND (NOT onnxruntime_ENABLE_DAWN_BACKEND_D3D12))
message(FATAL_ERROR "At least one of onnxruntime_ENABLE_DAWN_BACKEND_VULKAN or onnxruntime_ENABLE_DAWN_BACKEND_D3D12 must be enabled when using Dawn on Windows.")
endif()
if (onnxruntime_ENABLE_DAWN_BACKEND_VULKAN)
set(DAWN_ENABLE_VULKAN ON CACHE BOOL "" FORCE)
set(TINT_BUILD_SPV_WRITER ON CACHE BOOL "" FORCE)
if (onnxruntime_USE_EXTERNAL_DAWN)
message(FATAL_ERROR "onnxruntime_USE_EXTERNAL_DAWN and onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY cannot be enabled at the same time.")
endif()
else()
set(DAWN_ENABLE_VULKAN OFF CACHE BOOL "" FORCE)
# use dawn::dawn_native and dawn::dawn_proc instead of the monolithic dawn::webgpu_dawn to minimize binary size
set(DAWN_BUILD_MONOLITHIC_LIBRARY OFF CACHE BOOL "" FORCE)
set(DAWN_ENABLE_INSTALL OFF CACHE BOOL "" FORCE)
endif()
if (onnxruntime_ENABLE_DAWN_BACKEND_D3D12)
set(DAWN_ENABLE_D3D12 ON CACHE BOOL "" FORCE)
else()
set(DAWN_ENABLE_D3D12 OFF CACHE BOOL "" FORCE)
# disable things we don't use
set(DAWN_DXC_ENABLE_ASSERTS_IN_NDEBUG OFF)
set(DAWN_ENABLE_DESKTOP_GL OFF CACHE BOOL "" FORCE)
set(DAWN_ENABLE_OPENGLES OFF CACHE BOOL "" FORCE)
set(DAWN_SUPPORTS_GLFW_FOR_WINDOWING OFF CACHE BOOL "" FORCE)
set(DAWN_USE_GLFW OFF CACHE BOOL "" FORCE)
set(DAWN_USE_WINDOWS_UI OFF CACHE BOOL "" FORCE)
set(DAWN_USE_X11 OFF CACHE BOOL "" FORCE)
set(TINT_BUILD_TESTS OFF CACHE BOOL "" FORCE)
set(TINT_BUILD_CMD_TOOLS OFF CACHE BOOL "" FORCE)
set(TINT_BUILD_GLSL_WRITER OFF CACHE BOOL "" FORCE)
set(TINT_BUILD_GLSL_VALIDATOR OFF CACHE BOOL "" FORCE)
set(TINT_BUILD_IR_BINARY OFF CACHE BOOL "" FORCE)
set(TINT_BUILD_SPV_READER OFF CACHE BOOL "" FORCE) # don't need. disabling is a large binary size saving
set(TINT_BUILD_WGSL_WRITER ON CACHE BOOL "" FORCE) # needed to create cache key. runtime error if not enabled.
# SPIR-V validation shouldn't be required given we're using Tint to create the SPIR-V.
set(DAWN_ENABLE_SPIRV_VALIDATION OFF CACHE BOOL "" FORCE)
if (WIN32)
# building this requires the HLSL writer to be enabled in Tint. TBD if that we need either of these to be ON.
set(DAWN_USE_BUILT_DXC ON CACHE BOOL "" FORCE)
set(TINT_BUILD_HLSL_WRITER ON CACHE BOOL "" FORCE)
if ((NOT onnxruntime_ENABLE_DAWN_BACKEND_VULKAN) AND (NOT onnxruntime_ENABLE_DAWN_BACKEND_D3D12))
message(FATAL_ERROR "At least one of onnxruntime_ENABLE_DAWN_BACKEND_VULKAN or onnxruntime_ENABLE_DAWN_BACKEND_D3D12 must be enabled when using Dawn on Windows.")
endif()
if (onnxruntime_ENABLE_DAWN_BACKEND_VULKAN)
set(DAWN_ENABLE_VULKAN ON CACHE BOOL "" FORCE)
set(TINT_BUILD_SPV_WRITER ON CACHE BOOL "" FORCE)
else()
set(DAWN_ENABLE_VULKAN OFF CACHE BOOL "" FORCE)
endif()
if (onnxruntime_ENABLE_DAWN_BACKEND_D3D12)
set(DAWN_ENABLE_D3D12 ON CACHE BOOL "" FORCE)
else()
set(DAWN_ENABLE_D3D12 OFF CACHE BOOL "" FORCE)
endif()
# We are currently always using the D3D12 backend.
set(DAWN_ENABLE_D3D11 OFF CACHE BOOL "" FORCE)
endif()
# We are currently always using the D3D12 backend.
set(DAWN_ENABLE_D3D11 OFF CACHE BOOL "" FORCE)
endif()
onnxruntime_fetchcontent_makeavailable(dawn)
if (onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY)
list(APPEND onnxruntime_EXTERNAL_LIBRARIES dawn::webgpu_dawn)
else()
if (NOT onnxruntime_USE_EXTERNAL_DAWN)
list(APPEND onnxruntime_EXTERNAL_LIBRARIES dawn::dawn_native)
if (NOT CMAKE_SYSTEM_NAME STREQUAL "Emscripten")
if (onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY)
list(APPEND onnxruntime_EXTERNAL_LIBRARIES dawn::webgpu_dawn)
else()
if (NOT onnxruntime_USE_EXTERNAL_DAWN)
list(APPEND onnxruntime_EXTERNAL_LIBRARIES dawn::dawn_native)
endif()
list(APPEND onnxruntime_EXTERNAL_LIBRARIES dawn::dawn_proc)
endif()
list(APPEND onnxruntime_EXTERNAL_LIBRARIES dawn::dawn_proc)
endif()
endif()

View file

@ -21,44 +21,57 @@
source_group(TREE ${REPO_ROOT} FILES ${onnxruntime_providers_webgpu_cc_srcs})
onnxruntime_add_static_library(onnxruntime_providers_webgpu ${onnxruntime_providers_webgpu_cc_srcs})
onnxruntime_add_include_to_target(onnxruntime_providers_webgpu
onnxruntime_common dawn::dawncpp_headers dawn::dawn_headers onnx onnx_proto flatbuffers::flatbuffers Boost::mp11 safeint_interface)
onnxruntime_common onnx onnx_proto flatbuffers::flatbuffers Boost::mp11 safeint_interface)
set(onnxruntime_providers_webgpu_dll_deps)
if (onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY)
target_link_libraries(onnxruntime_providers_webgpu dawn::webgpu_dawn)
if (WIN32)
if (onnxruntime_ENABLE_DELAY_LOADING_WIN_DLLS)
list(APPEND onnxruntime_DELAYLOAD_FLAGS "/DELAYLOAD:webgpu_dawn.dll")
endif()
list(APPEND onnxruntime_providers_webgpu_dll_deps "$<TARGET_FILE:dawn::webgpu_dawn>")
if (CMAKE_SYSTEM_NAME STREQUAL "Emscripten")
# if "-s DISABLE_EXCEPTION_CATCHING=0" is globally set, we need to remove "-fno-exceptions" from emdawnwebgpu_c
if (CMAKE_CXX_FLAGS MATCHES "DISABLE_EXCEPTION_CATCHING=0")
get_property(EM_DAWN_WEBGPU_C_COMPILE_OPTIONS TARGET emdawnwebgpu_c PROPERTY COMPILE_OPTIONS)
list(REMOVE_ITEM EM_DAWN_WEBGPU_C_COMPILE_OPTIONS "-fno-exceptions")
set_property(TARGET emdawnwebgpu_c PROPERTY COMPILE_OPTIONS ${EM_DAWN_WEBGPU_C_COMPILE_OPTIONS})
endif()
target_link_libraries(onnxruntime_providers_webgpu PUBLIC emdawnwebgpu_cpp)
else()
if (NOT onnxruntime_USE_EXTERNAL_DAWN)
target_link_libraries(onnxruntime_providers_webgpu dawn::dawn_native)
onnxruntime_add_include_to_target(onnxruntime_providers_webgpu dawn::dawncpp_headers dawn::dawn_headers)
set(onnxruntime_providers_webgpu_dll_deps)
if (onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY)
target_link_libraries(onnxruntime_providers_webgpu dawn::webgpu_dawn)
if (WIN32)
if (onnxruntime_ENABLE_DELAY_LOADING_WIN_DLLS)
list(APPEND onnxruntime_DELAYLOAD_FLAGS "/DELAYLOAD:webgpu_dawn.dll")
endif()
list(APPEND onnxruntime_providers_webgpu_dll_deps "$<TARGET_FILE:dawn::webgpu_dawn>")
endif()
else()
if (NOT onnxruntime_USE_EXTERNAL_DAWN)
target_link_libraries(onnxruntime_providers_webgpu dawn::dawn_native)
endif()
target_link_libraries(onnxruntime_providers_webgpu dawn::dawn_proc)
endif()
target_link_libraries(onnxruntime_providers_webgpu dawn::dawn_proc)
endif()
if (WIN32 AND onnxruntime_ENABLE_DAWN_BACKEND_D3D12)
# Ensure dxil.dll and dxcompiler.dll exist in the output directory $<TARGET_FILE_DIR:dxcompiler>
add_dependencies(onnxruntime_providers_webgpu copy_dxil_dll)
add_dependencies(onnxruntime_providers_webgpu dxcompiler)
if (WIN32 AND onnxruntime_ENABLE_DAWN_BACKEND_D3D12)
# Ensure dxil.dll and dxcompiler.dll exist in the output directory $<TARGET_FILE_DIR:dxcompiler>
add_dependencies(onnxruntime_providers_webgpu copy_dxil_dll)
add_dependencies(onnxruntime_providers_webgpu dxcompiler)
list(APPEND onnxruntime_providers_webgpu_dll_deps "$<TARGET_FILE_DIR:dxcompiler>/dxil.dll")
list(APPEND onnxruntime_providers_webgpu_dll_deps "$<TARGET_FILE_DIR:dxcompiler>/dxcompiler.dll")
endif()
list(APPEND onnxruntime_providers_webgpu_dll_deps "$<TARGET_FILE_DIR:dxcompiler>/dxil.dll")
list(APPEND onnxruntime_providers_webgpu_dll_deps "$<TARGET_FILE_DIR:dxcompiler>/dxcompiler.dll")
endif()
if (onnxruntime_providers_webgpu_dll_deps)
# Copy dependency DLLs to the output directory
add_custom_command(
TARGET onnxruntime_providers_webgpu
POST_BUILD
COMMAND ${CMAKE_COMMAND} -E copy_if_different "${onnxruntime_providers_webgpu_dll_deps}" "$<TARGET_FILE_DIR:onnxruntime_providers_webgpu>"
COMMAND_EXPAND_LISTS
VERBATIM )
if (onnxruntime_providers_webgpu_dll_deps)
# Copy dependency DLLs to the output directory
add_custom_command(
TARGET onnxruntime_providers_webgpu
POST_BUILD
COMMAND ${CMAKE_COMMAND} -E copy_if_different "${onnxruntime_providers_webgpu_dll_deps}" "$<TARGET_FILE_DIR:onnxruntime_providers_webgpu>"
COMMAND_EXPAND_LISTS
VERBATIM )
endif()
endif()
add_dependencies(onnxruntime_providers_webgpu ${onnxruntime_EXTERNAL_DEPENDENCIES})

View file

@ -111,6 +111,7 @@ if (onnxruntime_BUILD_WEBASSEMBLY_STATIC_LIB)
${PROVIDERS_JS}
${PROVIDERS_XNNPACK}
${PROVIDERS_WEBNN}
${PROVIDERS_WEBGPU}
onnxruntime_session
onnxruntime_util
re2::re2
@ -188,6 +189,7 @@ else()
${PROVIDERS_JS}
${PROVIDERS_XNNPACK}
${PROVIDERS_WEBNN}
${PROVIDERS_WEBGPU}
onnxruntime_session
onnxruntime_util
re2::re2
@ -386,6 +388,11 @@ jsepDownload:_pp_")
set_target_properties(onnxruntime_webassembly PROPERTIES LINK_DEPENDS ${ONNXRUNTIME_ROOT}/wasm/pre-jsep.js)
endif()
if (onnxruntime_USE_WEBGPU)
target_compile_definitions(onnxruntime_webassembly PRIVATE USE_WEBGPU=1)
# target_link_libraries(onnxruntime_webassembly PUBLIC emdawnwebgpu_cpp)
endif()
if (onnxruntime_EMSCRIPTEN_SETTINGS)
foreach(setting IN LISTS onnxruntime_EMSCRIPTEN_SETTINGS)
target_link_options(onnxruntime_webassembly PRIVATE "SHELL:-s ${setting}")

View file

@ -0,0 +1,12 @@
diff --git a/src/emdawnwebgpu/CMakeLists.txt b/src/emdawnwebgpu/CMakeLists.txt
index de673537d3..c98dc46de7 100644
--- a/src/emdawnwebgpu/CMakeLists.txt
+++ b/src/emdawnwebgpu/CMakeLists.txt
@@ -78,6 +78,7 @@ if (${DAWN_ENABLE_EMSCRIPTEN})
endif()
set(ARGS
+ ${Python3_EXECUTABLE}
"${DAWN_EMSCRIPTEN_TOOLCHAIN}/tools/maint/gen_struct_info.py"
-q
"${EM_BUILD_GEN_DIR}/struct_info_webgpu.json"

View file

@ -0,0 +1,448 @@
#!/usr/bin/env python3
# coding=utf-8
# Copyright 2013 The Emscripten Authors. All rights reserved.
# Emscripten is available under two separate licenses, the MIT license and the
# University of Illinois/NCSA Open Source License. Both these licenses can be
# found in the LICENSE file.
"""This tool extracts information about structs and defines from the C headers.
The JSON input format is as follows:
[
{
'file': 'some/header.h',
'structs': {
'struct_name': [
'field1',
'field2',
'field3',
{
'field4': [
'nested1',
'nested2',
{
'nested3': [
'deep_nested1',
...
]
}
...
]
},
'field5'
],
'other_struct': [
'field1',
'field2',
...
]
},
'defines': [
'DEFINE_1',
'DEFINE_2',
['f', 'FLOAT_DEFINE'],
'DEFINE_3',
...
]
},
{
'file': 'some/other/header.h',
...
}
]
Please note that the 'f' for 'FLOAT_DEFINE' is just the format passed to printf(), you can put
anything printf() understands.
"""
import sys
import os
import re
import json
import argparse
import tempfile
import subprocess
__scriptdir__ = os.path.dirname(os.path.abspath(__file__))
__rootdir__ = os.path.dirname(os.path.dirname(__scriptdir__))
sys.path.insert(0, __rootdir__)
from tools import building
from tools import config
from tools import shared
from tools import system_libs
from tools import utils
QUIET = (__name__ != '__main__')
DEBUG = False
CFLAGS = [
# Avoid parsing problems due to gcc specific syntax.
'-D_GNU_SOURCE',
]
INTERNAL_CFLAGS = [
'-I' + utils.path_from_root('system/lib/libc/musl/src/internal'),
'-I' + utils.path_from_root('system/lib/libc/musl/src/include'),
'-I' + utils.path_from_root('system/lib/pthread/'),
]
CXXFLAGS = [
'-I' + utils.path_from_root('system/lib/libcxxabi/src'),
'-D__EMSCRIPTEN_EXCEPTIONS__',
'-I' + utils.path_from_root('system/lib/wasmfs/'),
'-std=c++17',
]
DEFAULT_JSON_FILES = [
utils.path_from_root('src/struct_info.json'),
utils.path_from_root('src/struct_info_internal.json'),
utils.path_from_root('src/struct_info_cxx.json'),
utils.path_from_root('src/struct_info_webgpu.json'),
]
def show(msg):
if shared.DEBUG or not QUIET:
sys.stderr.write('gen_struct_info: %s\n' % msg)
# The following three functions generate C code. The output of the compiled code will be
# parsed later on and then put back together into a dict structure by parse_c_output().
#
# Example:
# c_descent('test1', code)
# c_set('item', 'i%i', '111', code)
# c_set('item2', 'i%i', '9', code)
# c_set('item3', 's%s', '"Hello"', code)
# c_ascent(code)
# c_set('outer', 'f%f', '0.999', code)
#
# Will result in:
# {
# 'test1': {
# 'item': 111,
# 'item2': 9,
# 'item3': 'Hello',
# },
# 'outer': 0.999
# }
def c_set(name, type_, value, code):
code.append('printf("K' + name + '\\n");')
code.append('printf("V' + type_ + '\\n", ' + value + ');')
def c_descent(name, code):
code.append('printf("D' + name + '\\n");')
def c_ascent(code):
code.append('printf("A\\n");')
def parse_c_output(lines):
result = {}
cur_level = result
parent = []
key = None
for line in lines:
arg = line[1:].strip()
if '::' in arg:
arg = arg.split('::', 1)[1]
if line[0] == 'K':
# This is a key
key = arg
elif line[0] == 'V':
# A value
if arg[0] == 'i':
arg = int(arg[1:])
elif arg[0] == 'f':
arg = float(arg[1:])
elif arg[0] == 's':
arg = arg[1:]
cur_level[key] = arg
elif line[0] == 'D':
# Remember the current level as the last parent.
parent.append(cur_level)
# We descend one level.
cur_level[arg] = {}
cur_level = cur_level[arg]
elif line[0] == 'A':
# We return to the parent dict. (One level up.)
cur_level = parent.pop()
return result
def gen_inspect_code(path, struct, code):
if path[0][-1] == '#':
path[0] = path[0].rstrip('#')
prefix = ''
else:
prefix = 'struct '
c_descent(path[-1], code)
if len(path) == 1:
c_set('__size__', 'i%zu', 'sizeof (' + prefix + path[0] + ')', code)
else:
c_set('__size__', 'i%zu', 'sizeof ((' + prefix + path[0] + ' *)0)->' + '.'.join(path[1:]), code)
# c_set('__offset__', 'i%zu', 'offsetof(' + prefix + path[0] + ', ' + '.'.join(path[1:]) + ')', code)
for field in struct:
if isinstance(field, dict):
# We have to recurse to inspect the nested dict.
fname = list(field.keys())[0]
gen_inspect_code(path + [fname], field[fname], code)
else:
c_set(field, 'i%zu', 'offsetof(' + prefix + path[0] + ', ' + '.'.join(path[1:] + [field]) + ')', code)
c_ascent(code)
def generate_c_code(headers):
code = ['#include <stdio.h>', '#include <stddef.h>']
code.extend(f'''#include "{header['name']}"''' for header in headers)
code.append('int main() {')
c_descent('structs', code)
for header in headers:
for name, struct in header['structs'].items():
gen_inspect_code([name], struct, code)
c_ascent(code)
c_descent('defines', code)
for header in headers:
for name, type_ in header['defines'].items():
# Add the necessary python type, if missing.
if '%' not in type_:
if type_[-1] in {'d', 'i', 'u'}:
# integer
type_ = 'i%' + type_
elif type_[-1] in {'f', 'F', 'e', 'E', 'g', 'G'}:
# float
type_ = 'f%' + type_
elif type_[-1] in {'x', 'X', 'a', 'A', 'c', 's'}:
# hexadecimal or string
type_ = 's%' + type_
c_set(name, type_, name, code)
code.append('return 0;')
code.append('}')
return code
def generate_cmd(js_file_path, src_file_path, cflags):
# Compile the program.
show('Compiling generated code...')
if any('libcxxabi' in f for f in cflags):
compiler = shared.EMXX
else:
compiler = shared.EMCC
node_flags = building.get_emcc_node_flags(shared.check_node_version())
# -O1+ produces calls to iprintf, which libcompiler_rt doesn't support
cmd = [compiler] + cflags + ['-o', js_file_path, src_file_path,
'-O0',
'-Werror',
'-Wno-format',
'-nolibc',
'-sBOOTSTRAPPING_STRUCT_INFO',
'-sINCOMING_MODULE_JS_API=',
'-sSTRICT',
'-sSUPPORT_LONGJMP=0',
'-sASSERTIONS=0'] + node_flags
# Default behavior for emcc is to warn for binaryen version check mismatches
# so we should try to match that behavior.
cmd += ['-Wno-error=version-check']
# TODO(sbc): Remove this one we remove the test_em_config_env_var test
cmd += ['-Wno-deprecated']
show(shared.shlex_join(cmd))
return cmd
def inspect_headers(headers, cflags):
# Write the source code to a temporary file.
src_file_fd, src_file_path = tempfile.mkstemp('.c', text=True)
show('Generating C code... ' + src_file_path)
code = generate_c_code(headers)
os.write(src_file_fd, '\n'.join(code).encode())
os.close(src_file_fd)
js_file_fd, js_file_path = tempfile.mkstemp('.js')
# Close the unneeded FD.
os.close(js_file_fd)
cmd = generate_cmd(js_file_path, src_file_path, cflags)
try:
subprocess.check_call(cmd, env=system_libs.clean_env())
except subprocess.CalledProcessError as e:
sys.stderr.write('FAIL: Compilation failed!: %s\n' % e.cmd)
sys.exit(1)
# Run the compiled program.
show('Calling generated program... ' + js_file_path)
node_args = shared.node_bigint_flags(config.NODE_JS)
info = shared.run_js_tool(js_file_path, node_args=node_args, stdout=shared.PIPE).splitlines()
if not DEBUG:
# Remove all temporary files.
os.unlink(src_file_path)
if os.path.exists(js_file_path):
os.unlink(js_file_path)
wasm_file_path = shared.replace_suffix(js_file_path, '.wasm')
os.unlink(wasm_file_path)
# Parse the output of the program into a dict.
return parse_c_output(info)
def merge_info(target, src):
for key, value in src['defines'].items():
if key in target['defines']:
raise Exception('duplicate define: %s' % key)
target['defines'][key] = value
for key, value in src['structs'].items():
if key in target['structs']:
raise Exception('duplicate struct: %s' % key)
target['structs'][key] = value
def inspect_code(headers, cflags):
if not DEBUG:
info = inspect_headers(headers, cflags)
else:
info = {'defines': {}, 'structs': {}}
for header in headers:
merge_info(info, inspect_headers([header], cflags))
return info
def parse_json(path):
header_files = []
with open(path, 'r') as stream:
# Remove comments before loading the JSON.
data = json.loads(re.sub(r'//.*\n', '', stream.read()))
if not isinstance(data, list):
data = [data]
for item in data:
for key in item:
if key not in ['file', 'defines', 'structs']:
raise 'Unexpected key in json file: %s' % key
header = {'name': item['file'], 'structs': {}, 'defines': {}}
for name, data in item.get('structs', {}).items():
if name in header['structs']:
show('WARN: Description of struct "' + name + '" in file "' + item['file'] + '" replaces an existing description!')
header['structs'][name] = data
for part in item.get('defines', []):
if not isinstance(part, list):
# If no type is specified, assume integer.
part = ['i', part]
if part[1] in header['defines']:
show('WARN: Description of define "' + part[1] + '" in file "' + item['file'] + '" replaces an existing description!')
header['defines'][part[1]] = part[0]
header_files.append(header)
return header_files
def output_json(obj, stream):
json.dump(obj, stream, indent=4, sort_keys=True)
stream.write('\n')
stream.close()
def main(args):
global QUIET
parser = argparse.ArgumentParser(description='Generate JSON infos for structs.')
parser.add_argument('json', nargs='*',
help='JSON file with a list of structs and their fields (defaults to src/struct_info.json)',
default=DEFAULT_JSON_FILES)
parser.add_argument('-q', dest='quiet', action='store_true', default=False,
help='Don\'t output anything besides error messages.')
parser.add_argument('-o', dest='output', metavar='path', default=None,
help='Path to the JSON file that will be written. If omitted, the default location under `src` will be used.')
parser.add_argument('-I', dest='includes', metavar='dir', action='append', default=[],
help='Add directory to include search path')
parser.add_argument('-D', dest='defines', metavar='define', action='append', default=[],
help='Pass a define to the preprocessor')
parser.add_argument('-U', dest='undefines', metavar='undefine', action='append', default=[],
help='Pass an undefine to the preprocessor')
parser.add_argument('--wasm64', action='store_true',
help='use wasm64 architecture')
args = parser.parse_args(args)
QUIET = args.quiet
extra_cflags = []
if args.wasm64:
# Always use =2 here so that we don't generate a binary that actually requires
# memory64 to run. All we care about is that the output is correct.
extra_cflags += ['-sMEMORY64=2', '-Wno-experimental']
# Add the user options to the list as well.
for path in args.includes:
extra_cflags.append('-I' + path)
for arg in args.defines:
extra_cflags.append('-D' + arg)
for arg in args.undefines:
extra_cflags.append('-U' + arg)
# Look for structs in all passed headers.
info = {'defines': {}, 'structs': {}}
for f in args.json:
# This is a JSON file, parse it.
header_files = parse_json(f)
# Inspect all collected structs.
if 'internal' in f:
use_cflags = CFLAGS + extra_cflags + INTERNAL_CFLAGS
elif 'cxx' in f:
use_cflags = CFLAGS + extra_cflags + CXXFLAGS
else:
use_cflags = CFLAGS + extra_cflags
info_fragment = inspect_code(header_files, use_cflags)
merge_info(info, info_fragment)
if args.output:
output_file = args.output
elif args.wasm64:
output_file = utils.path_from_root('src/struct_info_generated_wasm64.json')
else:
output_file = utils.path_from_root('src/struct_info_generated.json')
with open(output_file, 'w') as f:
output_json(info, f)
return 0
if __name__ == '__main__':
sys.exit(main(sys.argv[1:]))

78
js/build_webgpu.bat Normal file
View file

@ -0,0 +1,78 @@
@echo off
rem build_webgpu.bat --- build onnxruntime-web with WebGPU EP
rem
rem Usage:
rem build_webgpu.bat config [clean]
rem
rem Options:
rem config Build configuration, "d" or "r"
rem clean Perform a clean build, "clean" or empty
setlocal enabledelayedexpansion
set ROOT=%~dp0..\
set BUILD_DIR=%ROOT%build_webgpu
:arg1
if ["%~1"]==["d"] (
set CONFIG=Debug
set CONFIG_EXTRA_FLAG=--enable_wasm_profiling --cmake_extra_defines onnxruntime_ENABLE_WEBASSEMBLY_OUTPUT_OPTIMIZED_MODEL=1
@rem --enable_wasm_debug_info
goto :arg2
)
if ["%~1"]==["r"] (
set CONFIG=Release
set CONFIG_EXTRA_FLAG=--enable_wasm_api_exception_catching --disable_rtti
goto :arg2
)
echo Invalid configuration "%~1", must be "d"(Debug) or "r"(Release)
exit /b 1
:arg2
if ["%~2"]==["clean"] (
goto :clean
)
if not exist "%ROOT%js\web\dist" (
goto :npm_ci
)
goto :build_wasm
:clean
if exist "%BUILD_DIR%" (
rd /s /q %BUILD_DIR%
)
pushd %ROOT%
git submodule sync --recursive
git submodule update --init --recursive
popd
:npm_ci
pushd %ROOT%js
call npm ci
popd
pushd %ROOT%js\common
call npm ci
popd
pushd %ROOT%js\web
call npm ci
call npm run pull:wasm
popd
:build_wasm
set PATH=C:\Program Files\Git\usr\bin;%PATH%
call %ROOT%build.bat --config %CONFIG% %CONFIG_EXTRA_FLAG% --skip_submodule_sync --build_wasm^
--enable_wasm_simd --enable_wasm_threads --use_webgpu --build_dir %BUILD_DIR%
@REM --target onnxruntime_webassembly
IF NOT "%ERRORLEVEL%" == "0" (
exit /b %ERRORLEVEL%
)
@REM copy /Y %BUILD_DIR%\%CONFIG%\ort-wasm-simd-threaded.jsep.wasm %ROOT%js\web\dist\
@REM copy /Y %BUILD_DIR%\%CONFIG%\ort-wasm-simd-threaded.jsep.mjs %ROOT%js\web\dist\

View file

@ -116,7 +116,7 @@ Status SkipLayerNorm<simplified>::ComputeInternal(onnxruntime::webgpu::ComputeCo
auto* output = context.Output(0, x_shape);
auto* input_skip_bias_sum = context.Output(3, x_shape);
size_t data_size = x_shape.Size();
int64_t data_size = x_shape.Size();
if (data_size == 0) {
return Status::OK();
}

View file

@ -1,10 +1,6 @@
// Copyright (c) Microsoft Corporation. All rights reserved.
// Licensed under the MIT License.
#ifdef __EMSCRIPTEN__
#include <emscripten.h>
#endif
#include "core/framework/session_state.h"
#include "core/providers/webgpu/allocator.h"
#include "core/providers/webgpu/webgpu_context.h"

View file

@ -85,7 +85,7 @@ class SimpleCacheManager : public IBufferCacheManager {
void OnRefresh() override {
for (auto& buffer : pending_buffers_) {
buffers_[wgpuBufferGetSize(buffer)].push_back(buffer);
buffers_[static_cast<size_t>(wgpuBufferGetSize(buffer))].push_back(buffer);
}
pending_buffers_.clear();
}
@ -167,7 +167,7 @@ class BucketCacheManager : public IBufferCacheManager {
// TODO: consider graph capture. currently not supported
for (auto& buffer : pending_buffers_) {
auto buffer_size = wgpuBufferGetSize(buffer);
auto buffer_size = static_cast<size_t>(wgpuBufferGetSize(buffer));
auto it = buckets_.find(buffer_size);
if (it != buckets_.end() && it->second.size() < buckets_limit_[buffer_size]) {

View file

@ -5,10 +5,6 @@
#include <iosfwd>
#ifdef __EMSCRIPTEN__
#include <emscripten/emscripten.h>
#endif
#include <webgpu/webgpu_cpp.h>
#include "core/framework/execution_provider.h"

View file

@ -3,10 +3,6 @@
#pragma once
#ifdef __EMSCRIPTEN__
#include <emscripten/emscripten.h>
#endif
#include <webgpu/webgpu_cpp.h>
#include <utility>

View file

@ -1,10 +1,6 @@
// Copyright (c) Microsoft Corporation. All rights reserved.
// Licensed under the MIT License.
#ifdef __EMSCRIPTEN__
#include <emscripten.h>
#endif
#include "core/providers/webgpu/data_transfer.h"
#include "core/providers/webgpu/webgpu_context.h"

View file

@ -123,10 +123,10 @@ Status BinaryElementwise::ComputeInternal(ComputeContext& context) const {
if (a_last_dim_divisible_by_4 || b_last_dim_divisible_by_4) {
vectorize = true;
} else {
size_t shared_dimension = 1;
int64_t shared_dimension = 1;
for (size_t i = 1; i < output_shape.NumDimensions(); i++) {
size_t dimA = lhs_shape.NumDimensions() >= i ? lhs_shape[lhs_shape.NumDimensions() - i] : 1;
size_t dimB = rhs_shape.NumDimensions() >= i ? rhs_shape[rhs_shape.NumDimensions() - i] : 1;
int64_t dimA = lhs_shape.NumDimensions() >= i ? lhs_shape[lhs_shape.NumDimensions() - i] : 1;
int64_t dimB = rhs_shape.NumDimensions() >= i ? rhs_shape[rhs_shape.NumDimensions() - i] : 1;
if (dimA == dimB) {
shared_dimension *= dimA;
num_shared_dimension++;

View file

@ -85,8 +85,7 @@ Status LayerNorm<simplified>::ComputeInternal(onnxruntime::webgpu::ComputeContex
auto* output = context.Output(0, x_shape);
size_t data_size = x_shape.Size();
if (data_size == 0) {
if (x_shape.Size() == 0) {
return Status::OK();
}

View file

@ -3,10 +3,6 @@
#pragma once
#ifdef __EMSCRIPTEN__
#include <emscripten/emscripten.h>
#endif
#include <string>
#include <unordered_map>

View file

@ -5,10 +5,6 @@
#include <sstream>
#ifdef __EMSCRIPTEN__
#include <emscripten/emscripten.h>
#endif
#include <webgpu/webgpu_cpp.h>
#include "core/framework/tensor_shape.h"

View file

@ -106,7 +106,8 @@ Status Concat::ComputeInternal(ComputeContext& context) const {
uint32_t output_size = gsl::narrow_cast<int32_t>(prepare.output_tensor->Shape().Size());
ConcatProgram program{prepare.axis};
size_t axis = static_cast<size_t>(prepare.axis);
ConcatProgram program{axis};
std::vector<uint32_t> sizes_in_concat_axis;
sizes_in_concat_axis.reserve(input_count);
@ -118,7 +119,7 @@ Status Concat::ComputeInternal(ComputeContext& context) const {
}
program.AddInput({input.tensor, ProgramTensorMetadataDependency::TypeAndRank});
auto axis_size = input.tensor->Shape()[prepare.axis];
auto axis_size = input.tensor->Shape()[axis];
sum += static_cast<uint32_t>(axis_size);
sizes_in_concat_axis.push_back(sum);
}

View file

@ -31,15 +31,8 @@ class Flatten final : public OpKernel {
return Status(common::ONNXRUNTIME, common::FAIL, "Invalid value for axis, must be less than or equal to input_rank");
}
int64_t first_dim = 1;
for (int64_t i = 0; i < axis; i++) {
first_dim *= input_shape[i];
}
int64_t second_dim = 1;
for (int64_t i = axis; i < input_rank; i++) {
second_dim *= input_shape[i];
}
int64_t first_dim = input_shape.SizeToDimension(static_cast<size_t>(axis));
int64_t second_dim = input_shape.SizeFromDimension(static_cast<size_t>(axis));
TensorShape output_shape({first_dim, second_dim});
Tensor* output_tensor = context->Output(0, output_shape);
@ -59,4 +52,4 @@ class Flatten final : public OpKernel {
};
} // namespace webgpu
} // namespace onnxruntime
} // namespace onnxruntime

View file

@ -59,7 +59,7 @@ Status GatherElements::ComputeInternal(ComputeContext& context) const {
axis += input_rank;
}
auto axis_dim_limit = input_shape[axis];
auto axis_dim_limit = input_shape[static_cast<size_t>(axis)];
auto output_dims = indices_shape.AsShapeVector();
TensorShape output_shape(output_dims);
@ -83,4 +83,4 @@ Status GatherElements::ComputeInternal(ComputeContext& context) const {
}
} // namespace webgpu
} // namespace onnxruntime
} // namespace onnxruntime

View file

@ -4,10 +4,12 @@
#include <memory>
#include <cmath>
#if !defined(__wasm__)
#include "dawn/dawn_proc.h"
#if !defined(USE_EXTERNAL_DAWN)
#include "dawn/native/DawnNative.h"
#endif
#endif
#include "core/common/common.h"
#include "core/common/path_string.h"
@ -29,7 +31,7 @@ void WebGpuContext::Initialize(const WebGpuBufferCacheConfig& buffer_cache_confi
std::call_once(init_flag_, [this, &buffer_cache_config, backend_type]() {
// Create wgpu::Adapter
if (adapter_ == nullptr) {
#if !defined(__EMSCRIPTEN__) && defined(_MSC_VER) && defined(DAWN_ENABLE_D3D12) && !defined(USE_EXTERNAL_DAWN)
#if !defined(__wasm__) && defined(_MSC_VER) && defined(DAWN_ENABLE_D3D12) && !defined(USE_EXTERNAL_DAWN)
// If we are using the D3D12 backend on Windows and the build does not use external Dawn, dxil.dll and dxcompiler.dll are required.
//
// Dawn will try to load them later, but if they are in the different directory to the executable, it may fail to find them.
@ -54,15 +56,19 @@ void WebGpuContext::Initialize(const WebGpuBufferCacheConfig& buffer_cache_confi
#endif
wgpu::RequestAdapterOptions req_adapter_options = {};
wgpu::DawnTogglesDescriptor adapter_toggles_desc = {};
req_adapter_options.nextInChain = &adapter_toggles_desc;
req_adapter_options.backendType = static_cast<wgpu::BackendType>(backend_type);
req_adapter_options.powerPreference = wgpu::PowerPreference::HighPerformance;
#if !defined(__wasm__)
auto enabled_adapter_toggles = GetEnabledAdapterToggles();
wgpu::DawnTogglesDescriptor adapter_toggles_desc = {};
adapter_toggles_desc.enabledToggleCount = enabled_adapter_toggles.size();
adapter_toggles_desc.enabledToggles = enabled_adapter_toggles.data();
req_adapter_options.nextInChain = &adapter_toggles_desc;
#endif
ORT_ENFORCE(wgpu::WaitStatus::Success == instance_.WaitAny(instance_.RequestAdapter(
&req_adapter_options,
wgpu::CallbackMode::WaitAnyOnly,
@ -78,6 +84,8 @@ void WebGpuContext::Initialize(const WebGpuBufferCacheConfig& buffer_cache_confi
// Create wgpu::Device
if (device_ == nullptr) {
wgpu::DeviceDescriptor device_desc = {};
#if !defined(__wasm__)
wgpu::DawnTogglesDescriptor device_toggles_desc = {};
device_desc.nextInChain = &device_toggles_desc;
@ -88,6 +96,7 @@ void WebGpuContext::Initialize(const WebGpuBufferCacheConfig& buffer_cache_confi
auto disabled_device_toggles = GetDisabledDeviceToggles();
device_toggles_desc.disabledToggleCount = disabled_device_toggles.size();
device_toggles_desc.disabledToggles = disabled_device_toggles.data();
#endif
std::vector<wgpu::FeatureName> required_features = GetAvailableRequiredFeatures(adapter_);
if (required_features.size() > 0) {
@ -136,9 +145,12 @@ void WebGpuContext::Initialize(const WebGpuBufferCacheConfig& buffer_cache_confi
program_mgr_ = std::make_unique<ProgramManager>(Device(), DeviceLimits());
// set query type
#if !defined(__wasm__)
if (device_.HasFeature(wgpu::FeatureName::ChromiumExperimentalTimestampQueryInsidePasses)) {
query_type_ = TimestampQueryType::InsidePasses;
} else if (device_.HasFeature(wgpu::FeatureName::TimestampQuery)) {
} else
#endif
if (device_.HasFeature(wgpu::FeatureName::TimestampQuery)) {
query_type_ = TimestampQueryType::AtPasses;
} else {
query_type_ = TimestampQueryType::None;
@ -456,7 +468,9 @@ std::vector<const char*> WebGpuContext::GetDisabledDeviceToggles() const {
std::vector<wgpu::FeatureName> WebGpuContext::GetAvailableRequiredFeatures(const wgpu::Adapter& adapter) const {
std::vector<wgpu::FeatureName> required_features;
constexpr wgpu::FeatureName features[]{
#if !defined(__wasm__)
wgpu::FeatureName::ChromiumExperimentalTimestampQueryInsidePasses,
#endif
wgpu::FeatureName::TimestampQuery,
wgpu::FeatureName::ShaderF16,
wgpu::FeatureName::Subgroups,
@ -531,7 +545,7 @@ void WebGpuContext::CollectProfilingData(profiling::Events& events) {
ORT_ENFORCE(Wait(query_read_buffer.MapAsync(wgpu::MapMode::Read,
0,
query_read_buffer.GetSize(),
static_cast<size_t>(query_read_buffer.GetSize()),
wgpu::CallbackMode::WaitAnyOnly,
[](wgpu::MapAsyncStatus status, wgpu::StringView message) {
ORT_ENFORCE(status == wgpu::MapAsyncStatus::Success, "Failed to download data from buffer: ", std::string_view{message});
@ -658,8 +672,14 @@ WebGpuContext& WebGpuContextFactory::CreateContext(const WebGpuContextConfig& co
ORT_ENFORCE(instance == nullptr && adapter == nullptr && device == nullptr,
"WebGPU EP default context (contextId=0) must not have custom WebGPU instance, adapter or device.");
std::call_once(init_default_flag_, [dawn_proc_table = config.dawn_proc_table]() {
// Step.1 - setup dawn proc table
std::call_once(init_default_flag_, [
#if !defined(__wasm__)
dawn_proc_table = config.dawn_proc_table
#endif
]() {
// Step.1 - setup dawn proc table (only for non-WASM build)
#if !defined(__wasm__)
const DawnProcTable* dawn_procs = reinterpret_cast<const DawnProcTable*>(dawn_proc_table);
#if defined(BUILD_DAWN_MONOLITHIC_LIBRARY)
ORT_ENFORCE(dawn_procs == nullptr, "setting DawnProcTable is not allowed when dynamically linked to webgpu_dawn.");
@ -672,6 +692,7 @@ WebGpuContext& WebGpuContextFactory::CreateContext(const WebGpuContextConfig& co
ORT_ENFORCE(dawn_procs != nullptr, "DawnProcTable must be provided.");
#endif
dawnProcSetProcs(dawn_procs);
#endif
#endif
// Step.2 - Create wgpu::Instance

View file

@ -3,10 +3,6 @@
#pragma once
#ifdef __EMSCRIPTEN__
#include <emscripten/emscripten.h>
#endif
#include <memory>
#include <mutex>

View file

@ -3,9 +3,6 @@
#include "core/providers/webgpu/webgpu_execution_provider.h"
#ifdef __EMSCRIPTEN__
#include <emscripten.h>
#endif
#include <string_view>
#include <unordered_map>
#include <unordered_set>