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

StgSharp.Native/sn-src/load_intrinsic.c

@@ -1,7 +1,6 @@
 #include "StgSharpNative.h"
 #include "sn_intrinsic.h"
 
-
 #if defined(_MSC_VER) && !defined(__clang__)
 #include <intrin.h>
 #include <isa_availability.h>
@@ -71,7 +70,7 @@ void load_intrinsic_sse(sn_intrinsic *context)
         context->f32_scalar_mul = f32_scalar_mul_sse;
         context->f32_sub = (MATARITHMETIC)f32_sub_sse;
         context->f32_transpose = (MATTRANSPOSEPROC)f32_transpose_sse;
-        context->normalize_3 = normalize;
+        context->f32_normalize_3 = f32_normalize;
 }
 
 void load_intrinsic_avx(sn_intrinsic *context)
@@ -80,11 +79,11 @@ void load_intrinsic_avx(sn_intrinsic *context)
         context->factorial_simd = factorial_simd_sse;
         context->index_pair = index_pair_sse;
         context->f32_add = (MATARITHMETIC)f32_add_avx;
-        context->f32_fma = (void(*)(void*, void*, void*))f32_fma_sse;
+        context->f32_fma = (void (*)(void *, void *, void *))f32_fma_sse;
         context->f32_scalar_mul = f32_scalar_mul_avx;
         context->f32_sub = (MATARITHMETIC)f32_sub_avx;
         context->f32_transpose = (MATTRANSPOSEPROC)f32_transpose_avx;
-        context->normalize_3 = normalize;
+        context->f32_normalize_3 = f32_normalize;
 }
 
 void load_intrinsic_512(sn_intrinsic *context)
@@ -93,9 +92,9 @@ void load_intrinsic_512(sn_intrinsic *context)
         context->factorial_simd = factorial_simd_sse;
         context->index_pair = index_pair_sse;
         context->f32_add = (MATARITHMETIC)f32_add_512;
-        context->f32_fma = (void(*)(void*, void*, void*))f32_fma_sse;
+        context->f32_fma = (void (*)(void *, void *, void *))f32_fma_sse;
         context->f32_scalar_mul = f32_scalar_mul_512;
         context->f32_sub = (MATARITHMETIC)f32_sub_512;
         context->f32_transpose = (MATTRANSPOSEPROC)f32_transpose_512;
-        context->normalize_3 = normalize;
+        context->f32_normalize_3 = f32_normalize;
 }

StgSharp.Native/sn-src/vector.c

@@ -1,7 +1,7 @@
 #include "sn_intrinsic.h"
 #include "sn_internal.h"
 
-SN_API void SN_DECL normalize(VEC(float) *restrict source, VEC(float) *restrict target)
+INTERNAL void SN_DECL f32_normalize(VEC(float) *restrict source, VEC(float) *restrict target)
 {
         register __m128 s = _mm_load_ps(&source->xmm);
         register __m128 a_length_sq = _mm_mul_ps(s, s);