Remove build time dependency on webgpu (#1201)

Author: baijumeswani

CMakeLists.txt

@@ -41,8 +41,6 @@ include(cmake/check_cuda.cmake)
 include(cmake/check_rocm.cmake)
 # Checking if DML is supported
 include(cmake/check_dml.cmake)
-# Checking if WebGpu is supported
-include(cmake/check_webgpu.cmake)
 
 include(cmake/cxx_standard.cmake)
 

build.py

@@ -126,8 +126,6 @@ class HelpFormatter(argparse.ArgumentDefaultsHelpFormatter, argparse.RawDescript
 
     parser.add_argument("--use_rocm", action="store_true", help="Whether to use ROCm. Default is to not use rocm.")
 
-    parser.add_argument("--use_webgpu", action="store_true", help="Whether to use WebGpu. Default is to not use WebGpu.")
-
     parser.add_argument("--use_dml", action="store_true", help="Whether to use DML. Default is to not use DML.")
 
     # The following options are mutually exclusive (cross compiling options such as android, ios, etc.)
@@ -473,7 +471,6 @@ def update(args: argparse.Namespace, env: dict[str, str]):
         "-DCMAKE_POSITION_INDEPENDENT_CODE=ON",
         f"-DUSE_CUDA={'ON' if args.use_cuda else 'OFF'}",
         f"-DUSE_ROCM={'ON' if args.use_rocm else 'OFF'}",
-        f"-DUSE_WEBGPU={'ON' if args.use_webgpu else 'OFF'}",
         f"-DUSE_DML={'ON' if args.use_dml else 'OFF'}",
         f"-DENABLE_JAVA={'ON' if args.build_java else 'OFF'}",
         f"-DBUILD_WHEEL={build_wheel}",

cmake/check_webgpu.cmake

@@ -4,7 +4,6 @@ include(CMakeDependentOption)
 option(USE_CUDA "Build with CUDA support" ON)
 option(USE_ROCM "Build with ROCm support" ON)
 option(USE_DML "Build with DML support" OFF)
-option(USE_WEBGPU "Build with WEBGPU support" ON)
 
 # bindings
 option(ENABLE_JAVA "Build the Java API." OFF)

cmake/options.cmake

@@ -317,14 +317,12 @@ void Model::InitDeviceAllocator(OrtSession& session) {
   }
 #endif
 
-#if USE_WEBGPU
   if (device_type_ == DeviceType::WEBGPU) {
     // for webgpu we only use device memory for kv_cache
     memory_info_device_ = OrtMemoryInfo::Create("WebGPU_Buffer", OrtAllocatorType::OrtDeviceAllocator, 0, OrtMemType::OrtMemTypeDefault);
     owned_allocator_device_ = Ort::Allocator::Create(session, *memory_info_device_);
     allocator_kvcache_ = owned_allocator_device_.get();
   }
-#endif
 
   if (device_type_ == DeviceType::QNN) {
     memory_info_device_ = OrtMemoryInfo::Create("QnnHtpShared", OrtAllocatorType::OrtDeviceAllocator, 0,

src/models/model.cpp

@@ -554,7 +554,8 @@ PYBIND11_MODULE(onnxruntime_genai, m) {
   m.def("is_cuda_available", []() { return USE_CUDA != 0; });
   m.def("is_dml_available", []() { return USE_DML != 0; });
   m.def("is_rocm_available", []() { return USE_ROCM != 0; });
-  m.def("is_webgpu_available", []() { return USE_WEBGPU != 0; });
+  m.def("is_webgpu_available", []() { return true; });
+  m.def("is_qnn_available", []() { return true; });
 
   m.def("set_current_gpu_device_id", [](int device_id) { Ort::SetCurrentGpuDeviceId(device_id); });
   m.def("get_current_gpu_device_id", []() { return Ort::GetCurrentGpuDeviceId(); });

src/python/python.cpp

@@ -7,7 +7,6 @@ std::unique_ptr<RuntimeSettings> CreateRuntimeSettings() {
 }
 
 std::string RuntimeSettings::GenerateConfigOverlay() const {
-  // #if USE_WEBGPU
   constexpr std::string_view webgpu_overlay_pre = R"({
   "model": {
     "decoder": {