mpl going on good process

StgSharp.Native/StgSharp.Native.vcxproj

@@ -88,6 +88,7 @@
     <Link>
       <SubSystem>Console</SubSystem>
       <GenerateDebugInformation>true</GenerateDebugInformation>
+      <AdditionalDependencies>opengl32.lib;$(CoreLibraryDependencies);%(AdditionalDependencies)</AdditionalDependencies>
     </Link>
   </ItemDefinitionGroup>
   <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
@@ -105,6 +106,7 @@
       <EnableCOMDATFolding>true</EnableCOMDATFolding>
       <OptimizeReferences>true</OptimizeReferences>
       <GenerateDebugInformation>true</GenerateDebugInformation>
+      <AdditionalDependencies>opengl32.lib;$(CoreLibraryDependencies);%(AdditionalDependencies)</AdditionalDependencies>
     </Link>
   </ItemDefinitionGroup>
   <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">

StgSharp.Native/sn-head/sn_internal.h

@@ -2,11 +2,11 @@
 #pragma once
 #endif
 
-#include <xmmintrin.h>
-
 #ifndef SN_INTERNAL
 #define SN_INTERNAL
 
+#include <xmmintrin.h>
+
 #ifdef _MSC_VER
 #define _CRT_SECURE_NO_WARNINGS 1
 #define WIN32_LEAN_AND_MEAN

StgSharp.Native/sn-head/sn_intrinsic.h

@@ -59,7 +59,7 @@ typedef uint64_t(SN_DECL *FACTORIALROC)(int n);
 
 #pragma region matix function
 
-INTERNAL void normalize(__m128 *source, __m128 *target);
+INTERNAL void f32_normalize(VEC(float) * source, VEC(float) * target);
 
 INTERNAL void SN_DECL f32_add_sse(MAT_KERNEL(float) const *left, MAT_KERNEL(float) const *right,
                                   MAT_KERNEL(float) *restrict ans);

StgSharp.Native/sn-head/sn_matkernel.h

@@ -9,11 +9,26 @@
 #define ZMMCOUNT(T) sizeof(T) / 4
 
 #define MAT_KERNEL(T)                    \
-        union {                          \
+        union mat_kernel_##T##{          \
                 __m128 xmm[XMMCOUNT(T)]; \
                 __m256 ymm[YMMCOUNT(T)]; \
                 __m512 zmm[ZMMCOUNT(T)]; \
                 T m[4][4];               \
         }
 
+#define VEC(T)                               \
+        union vec_##T##{                     \
+                __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/k_fma.c

@@ -1,8 +1,9 @@
 #include "sn_intrinsic.h"
 
-// ans = left * right(T) + ans
+// ans = left * right + ans
 INTERNAL void SN_DECL f32_fma_sse(MAT_KERNEL(float) const *restrict left,
-                                     MAT_KERNEL(float) const *restrict right, MAT_KERNEL(float) *restrict ans)
+                                  MAT_KERNEL(float) const *restrict right,
+                                  MAT_KERNEL(float) *restrict ans)
 {
         register __m128 source0 = _mm_load_ps(&left->xmm[0]);
         register __m128 source1 = _mm_load_ps(&left->xmm[1]);

StgSharp.Native/sn-src/str_string.c

@@ -3,12 +3,7 @@
 INTERNAL int SN_DECL index_pair_sse(short const *str, uint32_t target, int length)
 {
         int index = 0, result = 0, size = length, end = size - 8;
-        __m128i s = { 
-                                   target,
-                                   target,
-                                   target,
-                                   target,
-                           };
+        __m128i s = _mm_set1_epi32(target);
         __m128i vec, com;
 
         while (index < end) {

StgSharp.Native/sn-src/vector.c

@@ -1,24 +1,25 @@
-#include "StgSharpNative.h"
+#include "sn_intrinsic.h"
 #include "sn_internal.h"
 
-SN_API void SN_DECL normalize(__m128 *restrict source, __m128 *restrict target)
+SN_API void SN_DECL normalize(VEC(float) *restrict source, VEC(float) *restrict target)
 {
-        __m128 a_length_sq = _mm_mul_ps(*source, *source);
+        register __m128 s = _mm_load_ps(&source->xmm);
+        register __m128 a_length_sq = _mm_mul_ps(s, s);
         a_length_sq = _mm_add_ps(a_length_sq,
                                  _mm_shuffle_ps(a_length_sq, a_length_sq, _MM_SHUFFLE(2, 3, 0, 1)));
         a_length_sq = _mm_add_ss(a_length_sq,
                                  _mm_shuffle_ps(a_length_sq, a_length_sq, _MM_SHUFFLE(0, 0, 3, 2)));
 
         __m128 a_inv_length = _mm_rsqrt_ps(a_length_sq);
-        *source = _mm_mul_ps(*source, a_inv_length);
+        s = _mm_mul_ps(s, a_inv_length);
 
-        __m128 dot_product = _mm_mul_ps(*target, *source);
+        __m128 dot_product = _mm_mul_ps(target->xmm[0], s);
         dot_product = _mm_add_ps(dot_product,
                                  _mm_shuffle_ps(dot_product, dot_product, _MM_SHUFFLE(2, 3, 0, 1)));
         dot_product = _mm_add_ss(dot_product,
                                  _mm_shuffle_ps(dot_product, dot_product, _MM_SHUFFLE(0, 0, 3, 2)));
 
-        __m128 b_orthogonal = _mm_sub_ps(*target, _mm_mul_ps(*source, dot_product));
+        __m128 b_orthogonal = _mm_sub_ps(target->xmm[0], _mm_mul_ps(s, dot_product));
 
         __m128 b_length_sq = _mm_mul_ps(b_orthogonal, b_orthogonal);
         b_length_sq = _mm_add_ps(b_length_sq,
@@ -27,5 +28,5 @@ SN_API void SN_DECL normalize(__m128 *restrict source, __m128 *restrict target)
                                  _mm_shuffle_ps(b_length_sq, b_length_sq, _MM_SHUFFLE(0, 0, 3, 2)));
 
         __m128 b_inv_length = _mm_rsqrt_ps(b_length_sq);
-        *target = _mm_mul_ps(b_orthogonal, b_inv_length);
+        target->xmm[0] = _mm_mul_ps(b_orthogonal, b_inv_length);
 }