other gemm kernel use half-vector-machine style

StgSharp.Native/sn-head/StgSharpNative.h

@@ -1,10 +1,10 @@
-#ifdef _MSC_VER
-#pragma once
-#endif
-
 #ifndef SNative
 #define SNative
 
+#ifdef __cplusplus
+#error "This header is not meant to be included in C++ code. StgSharp.Native is a C library and should be used with C linkage."
+#endif
+
 #include "sn_internal.h"
 #include "sn_target.h"
 
@@ -26,10 +26,6 @@
 #ifndef SN_
 #endif
 
-#define ALIGN(simdVec) _mm_loadu_ps(&simdVec)
-#define ALIGN256(simdVec) _mm256_loadu_ps(&simdVec)
-#define ALIGN512(simdVec) _mm512_loadu_ps(&simdVec)
-
 typedef struct Image {
         int width;
         int height;

StgSharp.Native/sn-head/sn_intrinsic.h

@@ -4,6 +4,7 @@
 #define SSC_INTRIN_HEAD
 
 #include "StgSharpNative.h"
+#include "sn_target.h"
 #include <xmmintrin.h>
 #include <immintrin.h>
 #include <smmintrin.h>
@@ -26,9 +27,31 @@
 #define M256_PERMUTE(s3, s2, s1, s0) \
         ((LASTTWO(s3) << 6) | (LASTTWO(s2) << 4) | (LASTTWO(s1) << 2) | (LASTTWO(s0)))
 #define BIT_CVRT(pack, index, T) *(T *)(pack->data + index)
+
+#if SN_IS_ARCH(x86_64)
+
+#define vec_128 __m128
+#define vec_256 __m256
+#define vec_512 __m512
+
 #define zero_vec _mm_setzero_ps()
 #define zero_vec_256 _mm256_setzero_ps()
 
+#define ALIGN(simdVec) _mm_loadu_ps(&simdVec)
+#define ALIGN256(simdVec) _mm256_loadu_ps(&simdVec)
+#define ALIGN512(simdVec) _mm512_loadu_ps(&simdVec)
+
+#else
+
+#define zero_vec
+#define zero_vec_256
+
+#define ALIGN(simdVec) 
+#define ALIGN256(simdVec) 
+#define ALIGN512(simdVec) 
+
+#endif
+
 #define SPAN(T)                \
         struct SPAN_T {        \
                 T *pointer;    \
@@ -126,6 +149,4 @@ typedef struct sn_intrinsic {
 
 typedef enum elment_mask { F32 = 0, F64 = 1, I32 = 2, I64 = 3 } element_mask;
 
-#include "sn_target.h"
-
 #endif