savepoint

tarekziade · Jan 10, 2025 · 5cdceec · 5cdceec
1 parent 0a8d674
commit 5cdceec
Show file tree

Hide file tree

Showing 2 changed files with 10 additions and 3 deletions.
diff --git a/onnxruntime/contrib_ops/cpu/quantization/dynamic_quantize_matmul.cc b/onnxruntime/contrib_ops/cpu/quantization/dynamic_quantize_matmul.cc
@@ -266,7 +266,10 @@ Status MatMulIntegerToFloatBase::ComputeCommon(OpKernelContext* ctx,
   // rowsA = M
   // width = K
   // colsB = N
+#if 0
   size_t rowsA = static_cast<size_t>(helper.M());
+
+  if (rowsA > 1) {
   size_t width = static_cast<size_t>(helper.K());
   size_t colsB = static_cast<size_t>(helper.N());
 
@@ -283,8 +286,14 @@ Status MatMulIntegerToFloatBase::ComputeCommon(OpKernelContext* ctx,
               is_b_scale_per_column,
               y_data
   );
+  }
+  else {
+#endif
+    MlasGemmBatch(gemm_shape, gemm_data_vec.data(), num_gemms, ctx->GetOperatorThreadPool());
 
- // MlasGemmBatch(gemm_shape, gemm_data_vec.data(), num_gemms, ctx->GetOperatorThreadPool());
+  //}
+
+ //
   /* 
   auto end = std::chrono::steady_clock::now();
   auto duration = std::chrono::duration_cast<std::chrono::nanoseconds>(end - start).count();

diff --git a/onnxruntime/core/providers/cpu/quantization/quantize_linear.cc b/onnxruntime/core/providers/cpu/quantization/quantize_linear.cc
@@ -534,8 +534,6 @@ Status DequantizeLinear<T>::Compute(OpKernelContext* ctx) const {
     } else {
       //auto start = std::chrono::high_resolution_clock::now();
     if (process_block_size > 1000) {
-
-
       GeckoDequantizeLinear(static_cast<float>(process_block_count),
                         static_cast<float>(broadcast_dim),
                         static_cast<float>(process_block_size),