fix

microsoft · Nov 1, 2024 · 4233379 · 4233379
1 parent 6df1676
commit 4233379
Show file tree

Hide file tree

Showing 4 changed files with 25 additions and 22 deletions.
diff --git a/onnxruntime/core/mlas/lib/mlasi.h b/onnxruntime/core/mlas/lib/mlasi.h
@@ -713,24 +713,26 @@ void
     int8_t ZeroPoint);
 
 typedef
-void (MLASCALL MLAS_QUANTIZE_SOFTMAX_I8_KERNEL)(
-                 size_t D,
-                 const int8_t* Xdata,
-                 int8_t* Ydata,
-                 const float* LookupTable,
-                 float Yscale,
-                 int8_t YZeroPoint,
-                 float* Buff);
+void
+(MLASCALL MLAS_QUANTIZE_SOFTMAX_I8_KERNEL)(
+    size_t D,
+    const int8_t* Xdata,
+    int8_t* Ydata,
+    const float* LookupTable,
+    float Yscale,
+    int8_t YZeroPoint,
+    float* Buff);
 
 typedef
-void (MLASCALL MLAS_QUANTIZE_SOFTMAX_U8_KERNEL)(
-                 size_t D,
-                 const uint8_t* Xdata,
-                 uint8_t* Ydata,
-                 const float* LookupTable,
-                 float Yscale,
-                 uint8_t YZeroPoint,
-                 float* Buff);
+void
+(MLASCALL MLAS_QUANTIZE_SOFTMAX_U8_KERNEL)(
+    size_t D,
+    const uint8_t* Xdata,
+    uint8_t* Ydata,
+    const float* LookupTable,
+    float Yscale,
+    uint8_t YZeroPoint,
+    float* Buff);
 
 template<typename InputType, typename FilterType>
 struct MLAS_QUANT_KERNEL

diff --git a/onnxruntime/core/mlas/lib/qsoftmax.cpp b/onnxruntime/core/mlas/lib/qsoftmax.cpp
@@ -34,8 +34,8 @@ struct MLAS_QSOFTMAX_WORK_BLOCK {
 };
 
 static void BuildLookupTable(gsl::span<float> table,
-                                   const float x_scale,
-                                   size_t reduce_len, bool is_signed) {
+                             const float x_scale,
+                             size_t reduce_len, bool is_signed) {
   // make sure sum(exp(x)) < max<T>()
   double bit_shift = log(std::numeric_limits<float>::max() / reduce_len);
   double reserve_bit = std::is_same_v<float, float> ? 5 : 3;

diff --git a/onnxruntime/core/mlas/lib/qsoftmax_kernel_naive.cpp b/onnxruntime/core/mlas/lib/qsoftmax_kernel_naive.cpp
@@ -21,8 +21,8 @@ Module Name:
 
 #include "mlasi.h"
 
-void MlasQuantizeSoftmaxU8KernelNaive(size_t D, const uint8_t* x_data, uint8_t* y_data, const float* lookup_table,
-                                      float y_scale, uint8_t yzp, float*) {
+void MLASCALL MlasQuantizeSoftmaxU8KernelNaive(size_t D, const uint8_t* x_data, uint8_t* y_data,
+                                               const float* lookup_table, float y_scale, uint8_t yzp, float*) {
   constexpr size_t N = 1;
   const auto c_y_scale = y_scale;
   const auto c_y_zp = yzp;
@@ -66,8 +66,8 @@ void MlasQuantizeSoftmaxU8KernelNaive(size_t D, const uint8_t* x_data, uint8_t*
   }
 }
 
-void MlasQuantizeSoftmaxI8KernelNaive(size_t D, const int8_t* x_data, int8_t* y_data, const float* lookup_table,
-                                      float y_scale, int8_t yzp, float*) {
+void MLASCALL MlasQuantizeSoftmaxI8KernelNaive(size_t D, const int8_t* x_data, int8_t* y_data,
+                                               const float* lookup_table, float y_scale, int8_t yzp, float*) {
   constexpr size_t N = 1;
   const auto c_y_scale = y_scale;
   const auto c_y_zp = yzp;

diff --git a/onnxruntime/test/mlas/bench/bench_qsoftmax.cpp b/onnxruntime/test/mlas/bench/bench_qsoftmax.cpp
@@ -1,5 +1,6 @@
 #include "bench_util.h"
 #include "core/mlas/lib/mlasi.h"
+#include "core/util/thread_utils.h"
 
 static const std::vector<std::string> qsoftmax_bench_arg_names = {"N", "D", "is_signed"};
 //(const void* Input, void* Output, size_t N, size_t D, const float* LoopupTable,float Scale, int ZeroPoint, bool is_signed, MLAS_THREADPOOL* ThreadPool);