WebGPU: Remove deprecated subgroups-f16 from WebGPU native and JS EP (#23898)

This PR removes the deprecated subgroups-f16 from WebGPU native and JS
EP, and also remove the unused deviceInfo in WebGPU JS EP.

Author: jiangzhaoming

js/web/lib/wasm/jsep/backend-webgpu.ts

@@ -13,7 +13,6 @@ import { ProgramManager } from './webgpu/program-manager';
 import {
   AdapterInfo,
   ComputeContext,
-  DeviceInfo,
   GpuArchitecture,
   GpuData,
   GpuVendor,
@@ -135,34 +134,13 @@ class AdapterInfoImpl implements AdapterInfo {
   }
 }
 
-class DeviceInfoImpl implements DeviceInfo {
-  readonly subgroupsSupported: boolean;
-  readonly subgroupsF16Supported: boolean;
-  readonly subgroupSizeRange?: readonly [number, number];
-
-  constructor(device: GPUDevice) {
-    this.subgroupsSupported = device.features.has('subgroups' as GPUFeatureName);
-    this.subgroupsF16Supported = device.features.has('subgroups' as GPUFeatureName);
-    // Currently subgroups feature is still experimental and size attributes are not in the WebGPU IDL, so we have to
-    // workaround the IDL type checks.
-    // TODO: clean this after subgroups feature is settled in IDL.
-    const deviceSubgroupsLimits = device.limits as { minSubgroupSize?: number; maxSubgroupSize?: number };
-    if (!this.subgroupsSupported || !deviceSubgroupsLimits.minSubgroupSize || !deviceSubgroupsLimits.maxSubgroupSize) {
-      this.subgroupSizeRange = undefined;
-    } else {
-      this.subgroupSizeRange = [deviceSubgroupsLimits.minSubgroupSize, deviceSubgroupsLimits.maxSubgroupSize];
-    }
-  }
-}
-
 /**
  * this class is designed to store status and being used as a singleton for JSEP. It will be passed to jsepInit() as
  * the first parameter so that it is stored for future use.
  */
 export class WebGpuBackend {
   adapterInfo: AdapterInfoImpl;
   device: GPUDevice;
-  deviceInfo: DeviceInfoImpl;
   /**
    * an instance of GpuDataManager to manage a GpuDataId -> GpuBuffer mapping
    */
@@ -274,13 +252,9 @@ export class WebGpuBackend {
     }
     requireFeatureIfAvailable('shader-f16');
     // Try subgroups
-    if (requireFeatureIfAvailable('subgroups' as GPUFeatureName)) {
-      // If subgroups feature is available, also try subgroups-f16
-      requireFeatureIfAvailable('subgroups-f16' as GPUFeatureName);
-    }
+    requireFeatureIfAvailable('subgroups' as GPUFeatureName);
 
     this.device = await adapter.requestDevice(deviceDescriptor);
-    this.deviceInfo = new DeviceInfoImpl(this.device);
     this.adapterInfo = new AdapterInfoImpl(adapter.info || (await adapter.requestAdapterInfo()));
     this.gpuDataManager = createGpuDataManager(this);
     this.programManager = new ProgramManager(this);

js/web/lib/wasm/jsep/init.ts

@@ -11,13 +11,7 @@ import { WebGpuBackend } from './backend-webgpu';
 import { LOG_DEBUG } from './log';
 import { TensorView } from './tensor-view';
 import { ShapeUtil } from './util';
-import {
-  AdapterInfo,
-  ComputeContext,
-  ComputeContextInputsOutputsMapping,
-  DeviceInfo,
-  ProgramInfo,
-} from './webgpu/types';
+import { AdapterInfo, ComputeContext, ComputeContextInputsOutputsMapping, ProgramInfo } from './webgpu/types';
 import { WebNNBackend } from './backend-webnn';
 
 /* eslint-disable no-bitwise */
@@ -76,7 +70,6 @@ class TensorViewImpl implements TensorView {
 
 class ComputeContextImpl implements ComputeContext {
   readonly adapterInfo: AdapterInfo;
-  readonly deviceInfo: DeviceInfo;
   readonly opKernelContext: number;
   readonly inputs: readonly TensorView[];
   readonly outputCount: number;
@@ -94,7 +87,6 @@ class ComputeContextImpl implements ComputeContext {
     contextDataOffset: number,
   ) {
     this.adapterInfo = backend.adapterInfo;
-    this.deviceInfo = backend.deviceInfo;
 
     // extract context data
     const ptrSize = module.PTR_SIZE;

js/web/lib/wasm/jsep/webgpu/program-manager.ts

@@ -99,7 +99,6 @@ export class ProgramManager {
     const extensionsInfo: Array<{ feature: GPUFeatureName; extension: string }> = [
       { feature: 'shader-f16', extension: 'f16' },
       { feature: 'subgroups' as GPUFeatureName, extension: 'subgroups' },
-      { feature: 'subgroups-f16' as GPUFeatureName, extension: 'subgroups_f16' },
     ];
     extensionsInfo.forEach((info) => {
       if (device.features.has(info.feature)) {

js/web/lib/wasm/jsep/webgpu/types.ts

@@ -21,11 +21,6 @@ export interface AdapterInfo {
   isArchitecture: (architecture: GpuArchitecture) => boolean;
   isVendor: (vendor: GpuVendor) => boolean;
 }
-export interface DeviceInfo {
-  readonly subgroupsSupported: boolean;
-  readonly subgroupsF16Supported: boolean;
-  readonly subgroupSizeRange?: readonly [number, number];
-}
 
 export interface GpuData {
   type: GpuDataType;
@@ -165,11 +160,6 @@ export interface ComputeContext {
    */
   readonly adapterInfo: AdapterInfo;
 
-  /**
-   * gpu device info
-   */
-  readonly deviceInfo: DeviceInfo;
-
   /**
    * stores the pointer to OpKernelContext
    */

onnxruntime/core/providers/webgpu/shader_helper.cc

@@ -345,9 +345,6 @@ Status ShaderHelper::GenerateSourceCode(std::string& code, std::vector<int>& sha
                   })) {
     ORT_RETURN_IF_NOT(device_.HasFeature(wgpu::FeatureName::ShaderF16), "Program ", program_.Name(), " requires f16 but the device does not support it.");
     ss << "enable f16;\n";
-    if (device_.HasFeature(wgpu::FeatureName::SubgroupsF16)) {
-      ss << "enable subgroups_f16;\n";
-    }
   }
   if (device_.HasFeature(wgpu::FeatureName::Subgroups)) {
     ss << "enable subgroups;\n";

onnxruntime/core/providers/webgpu/webgpu_context.cc

@@ -489,8 +489,7 @@ std::vector<wgpu::FeatureName> WebGpuContext::GetAvailableRequiredFeatures(const
 #endif
       wgpu::FeatureName::TimestampQuery,
       wgpu::FeatureName::ShaderF16,
-      wgpu::FeatureName::Subgroups,
-      wgpu::FeatureName::SubgroupsF16};
+      wgpu::FeatureName::Subgroups};
   for (auto feature : features) {
     if (adapter.HasFeature(feature)) {
       required_features.push_back(feature);