[MLAS] DequantizeLinear int8/uint8

cmake/onnxruntime_mlas.cmake

@@ -31,6 +31,7 @@ onnxruntime_add_static_library(onnxruntime_mlas
   ${MLAS_SRC_DIR}/eltwise.cpp
   ${MLAS_SRC_DIR}/erf.cpp
   ${MLAS_SRC_DIR}/compute.cpp
+  ${MLAS_SRC_DIR}/dequantize.cpp
   ${MLAS_SRC_DIR}/quantize.cpp
   ${MLAS_SRC_DIR}/qgemm_kernel_default.cpp
   ${MLAS_SRC_DIR}/qladd.cpp

onnxruntime/core/mlas/inc/mlas.h

@@ -1223,6 +1223,21 @@ MlasQuantizeLinearS4(
     int8_t ZeroPoint
     );
 
+//
+// Linear dequantization routines.
+//
+
+template<typename InputType>
+void
+MLASCALL
+MlasDequantizeLinear(
+    const InputType* Input,
+    float* Output,
+    size_t N,
+    float Scale,
+    InputType ZeroPoint
+    );
+
 /**
  * @brief Requantize a block of the intermediate buffer to the output buffer,
  *        optionally adding the supplied bias