vector and matrix optimization

StgSharp.Native/sn-head/sn_intrinsic.h

@@ -1,7 +1,3 @@
-#ifdef _MSC_VER
-#pragma once
-#endif
-
 #ifndef SSC_INTRIN
 #define SSC_INTRIN
 
@@ -128,7 +124,7 @@ typedef struct sn_intrinsic {
         SCALARMATARITHMETIC f32_scalar_mul;
         MATARITHMETIC f32_sub;
         MATTRANSPOSEPROC f32_transpose;
-        VECTORNORMALIZEPROC normalize_3;
+        VECTORNORMALIZEPROC f32_normalize_3;
 } sn_intrinsic;
 
 typedef enum most_advanced_instruction {
@@ -145,4 +141,4 @@ typedef enum most_advanced_instruction {
 #endif
 } most_advanced_instruction;
 
-#endif // !SSC_INTRIN
+#endif

StgSharp.Native/sn-head/sn_matkernel.h

@@ -1,15 +1,15 @@
 #ifndef SN_MATKERNEL
 #define SN_MATKERNEL
 
-#include <immintrin.h>
-
 #define ELEMENTSIZE(T) sizeof(T)
 #define XMMCOUNT(T) sizeof(T)
 #define YMMCOUNT(T) sizeof(T) / 2
 #define ZMMCOUNT(T) sizeof(T) / 4
 
+#include <immintrin.h>
+
 #define MAT_KERNEL(T)                    \
-        union mat_kernel_##T##{          \
+        union {                          \
                 __m128 xmm[XMMCOUNT(T)]; \
                 __m256 ymm[YMMCOUNT(T)]; \
                 __m512 zmm[ZMMCOUNT(T)]; \
@@ -17,18 +17,11 @@
         }
 
 #define VEC(T)                               \
-        union vec_##T##{                     \
+        union {                              \
                 __m128 xmm[XMMCOUNT(T) / 4]; \
                 __m256 ymm[YMMCOUNT(T) / 4]; \
                 __m512 zmm[ZMMCOUNT(T) / 4]; \
                 T v[4];                      \
         }
 
-/*
-#undef ELEMENTSIZE
-#undef XMMCOUNT
-#undef YMMCOUNT
-#undef ZMMCOUNT
-/**/
-
 #endif