[webgpu] Always use tile matmulnbits for block_size = 32 (#23140)

### Description After the optimization of prefill time with #23102, it seems that always using the tile matmulnibits with block_size = 32 can bring better performance even for discrete gpu for phi3 model. Phi3 becomes 42.64 tokens/sec from 32.82 tokens/sec in easy mode on my NV RTX 2000 GPU.
microsoft · Dec 20, 2024 · 7c782f6 · 7c782f6
1 parent b4a6a0d
commit 7c782f6
Show file tree

Hide file tree

Showing 2 changed files with 9 additions and 13 deletions.
diff --git a/onnxruntime/contrib_ops/webgpu/quantization/matmul_nbits.cc b/onnxruntime/contrib_ops/webgpu/quantization/matmul_nbits.cc
@@ -60,7 +60,7 @@ Status MatMulNBitsProgram::GenerateShaderCode(ShaderHelper& shader) const {
   const auto& scales = shader.AddInput("scales", ShaderUsage::UseUniform);
   const auto& y = shader.AddOutput("output", ShaderUsage::UseUniform | ShaderUsage::UseValueTypeAlias | ShaderUsage::UseElementTypeAlias | ShaderUsage::UseIndicesTypeAlias);
 
-  if ((is_intel_ || tile_m_ > 1) && block_size_ == 32) {
+  if (block_size_ == 32) {
     const uint32_t workgroup_size = WorkgroupSizeX() * WorkgroupSizeY();
     const uint32_t tile_size = WorkgroupSizeX() * components_b_ * 8;  // each uint32 has 8 data.
     const uint32_t a_length_per_tile = tile_size / a.NumComponents();
@@ -408,14 +408,12 @@ Status MatMulNBits::ComputeInternal(onnxruntime::webgpu::ComputeContext& context
   const uint32_t components_b = GetMaxComponents(blob_size_in_words);
   uint32_t components = GetMaxComponents(N);
 
-  const bool is_intel = context.AdapterInfo().vendor == std::string_view{"intel"} &&
-                        context.AdapterInfo().architecture == std::string_view{"gen-12lp"};
   const bool has_zero_points = zero_points != nullptr;
 
   // TODO: Support output_number > 1. Some cases are failed when output_number > 1.
   constexpr uint32_t output_number = 1;
   const uint32_t tile_m = M > kMinMForTileOptimization ? 4 : 1;
-  MatMulNBitsProgram program{output_number, block_size, tile_m, gsl::narrow<int>(components_b), has_zero_points, is_intel};
+  MatMulNBitsProgram program{output_number, block_size, tile_m, gsl::narrow<int>(components_b), has_zero_points};
   if (M > kMinMForTileOptimization && block_size == 32) {
     components = 1;
     constexpr uint32_t workgroup_size = 64;
@@ -426,7 +424,7 @@ Status MatMulNBits::ComputeInternal(onnxruntime::webgpu::ComputeContext& context
                                  (M + tile_m - 1) / tile_m,
                                  batch_count);
     program.CacheHint("T_M" + std::to_string(tile_m));
-  } else if (is_intel && block_size == 32) {
+  } else if (block_size == 32) {
     components = 1;
     constexpr uint32_t workgroup_size = 128;
     const uint32_t workgroup_y = N % 8 == 0 ? 8 : N % 4 == 0 ? 4

diff --git a/onnxruntime/contrib_ops/webgpu/quantization/matmul_nbits.h b/onnxruntime/contrib_ops/webgpu/quantization/matmul_nbits.h
@@ -14,13 +14,12 @@ using namespace onnxruntime::webgpu;
 
 class MatMulNBitsProgram final : public Program<MatMulNBitsProgram> {
  public:
-  MatMulNBitsProgram(uint32_t output_number, uint32_t block_size, uint32_t tile_m, int components_b, bool has_zero_points, bool is_intel) : Program{"MatMulNBits"},
-                                                                                                                                            output_number_{output_number},
-                                                                                                                                            block_size_{block_size},
-                                                                                                                                            tile_m_{tile_m},
-                                                                                                                                            components_b_{components_b},
-                                                                                                                                            has_zero_points_{has_zero_points},
-                                                                                                                                            is_intel_{is_intel} {
+  MatMulNBitsProgram(uint32_t output_number, uint32_t block_size, uint32_t tile_m, int components_b, bool has_zero_points) : Program{"MatMulNBits"},
+                                                                                                                             output_number_{output_number},
+                                                                                                                             block_size_{block_size},
+                                                                                                                             tile_m_{tile_m},
+                                                                                                                             components_b_{components_b},
+                                                                                                                             has_zero_points_{has_zero_points} {
   }
 
   Status GenerateShaderCode(ShaderHelper& sh) const override;
@@ -32,7 +31,6 @@ class MatMulNBitsProgram final : public Program<MatMulNBitsProgram> {
   uint32_t tile_m_;
   int components_b_;
   bool has_zero_points_;
-  bool is_intel_;
 };
 
 class MatMulNBits final : public WebGpuKernel {