youmuliu/stgsharp
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

daily update on 20250109 by nitload youmu, building STEP file reader

Browse Source
This commit is contained in:
Nitload
2025-01-09 07:28:07 +08:00
parent 63c3cd17e2
commit 6cfaa42ff3
84 changed files with 6688 additions and 922 deletions
Download Patch File Download Diff File Expand all files Collapse all files
内核源代码/StgSharpC/intrinsic.c
+9 -4
View File
@@ -81,6 +81,7 @@ void load_intrinsic_sse(ssc_intrinsic *context)
context->dot_32 = dot_32_sse;
context->dot_41 = dot_41_sse;
context->dot_42 = dot_42_sse;
context->dot_43 = dot_43_sse;
context->string_quick_hash = quick_string_hash;
}
@@ -100,10 +101,11 @@ void load_intrinsic_avx(ssc_intrinsic *context)
context->normalize_3 = normalize;
context->det_matrix33 = det_mat3;
context->det_matrix44 = det_mat4;
context->dot_31 = dot_31_sse;
context->dot_31 = dot_31_avx;
context->dot_32 = dot_32_avx;
context->dot_41 = dot_41_sse;
context->dot_41 = dot_41_avx;
context->dot_42 = dot_42_avx;
context->dot_43 = dot_43_avx;
context->string_quick_hash = quick_string_hash;
}
@@ -123,7 +125,10 @@ void load_intrinsic_512(ssc_intrinsic *context)
context->normalize_3 = normalize;
context->det_matrix33 = det_mat3;
context->det_matrix44 = det_mat4;
context->dot_31 = dot_31_sse;
context->dot_41 = dot_41_sse;
context->dot_31 = dot_31_avx;
context->dot_32 = dot_32_avx;
context->dot_41 = dot_41_avx;
context->dot_42 = dot_42_avx;
context->dot_43 = dot_43_avx;
context->string_quick_hash = quick_string_hash;
}
内核源代码/StgSharpC/mat2.c
+7 -7
View File
@@ -8,9 +8,9 @@ INTERNAL void __cdecl transpose23(__2_columnset *source, __3_columnset *target)
__m128 t1 = _mm_shuffle_ps(ALIGN(source->column[0]), ALIGN(source->column[1]),
_MM_SHUFFLE(3, 2, 3, 2));
target->column[0] = _mm_shuffle_ps(t0, zeroVec, _MM_SHUFFLE(2, 0, 2, 0));
target->column[1] = _mm_shuffle_ps(t0, zeroVec, _MM_SHUFFLE(3, 1, 3, 1));
target->column[2] = _mm_shuffle_ps(t1, zeroVec, _MM_SHUFFLE(2, 0, 2, 0));
target->column[0] = _mm_shuffle_ps(t0, zero_vec, _MM_SHUFFLE(2, 0, 2, 0));
target->column[1] = _mm_shuffle_ps(t0, zero_vec, _MM_SHUFFLE(3, 1, 3, 1));
target->column[2] = _mm_shuffle_ps(t1, zero_vec, _MM_SHUFFLE(2, 0, 2, 0));
}
INTERNAL void __cdecl transpose24(__2_columnset *source, __4_columnset *target)
@@ -20,10 +20,10 @@ INTERNAL void __cdecl transpose24(__2_columnset *source, __4_columnset *target)
__m128 t1 = _mm_shuffle_ps(ALIGN(source->column[0]), ALIGN(source->column[1]),
_MM_SHUFFLE(3, 2, 3, 2));
target->column[0] = _mm_shuffle_ps(t0, zeroVec, _MM_SHUFFLE(2, 0, 2, 0));
target->column[1] = _mm_shuffle_ps(t0, zeroVec, _MM_SHUFFLE(3, 1, 3, 1));
target->column[2] = _mm_shuffle_ps(t1, zeroVec, _MM_SHUFFLE(2, 0, 2, 0));
target->column[3] = _mm_shuffle_ps(t1, zeroVec, _MM_SHUFFLE(3, 1, 3, 1));
target->column[0] = _mm_shuffle_ps(t0, zero_vec, _MM_SHUFFLE(2, 0, 2, 0));
target->column[1] = _mm_shuffle_ps(t0, zero_vec, _MM_SHUFFLE(3, 1, 3, 1));
target->column[2] = _mm_shuffle_ps(t1, zero_vec, _MM_SHUFFLE(2, 0, 2, 0));
target->column[3] = _mm_shuffle_ps(t1, zero_vec, _MM_SHUFFLE(3, 1, 3, 1));
}
INTERNAL void __cdecl addmatrix2_sse(__2_columnset *left, __2_columnset *right, __2_columnset *ans)
内核源代码/StgSharpC/mat3.c
+5 -5
View File
@@ -6,7 +6,7 @@ INTERNAL void _cdecl transpose32(__3_columnset *source, __2_columnset *target)
{
__m128 t0 = _mm_shuffle_ps(ALIGN(source->column[0]), ALIGN(source->column[1]),
_MM_SHUFFLE(1, 0, 1, 0));
__m128 t2 = _mm_shuffle_ps(ALIGN(source->column[2]), zeroVec, _MM_SHUFFLE(1, 0, 1, 0));
__m128 t2 = _mm_shuffle_ps(ALIGN(source->column[2]), zero_vec, _MM_SHUFFLE(1, 0, 1, 0));
target->column[0] = _mm_shuffle_ps(t0, t2, _MM_SHUFFLE(2, 0, 2, 0));
target->column[1] = _mm_shuffle_ps(t0, t2, _MM_SHUFFLE(3, 1, 3, 1));
@@ -18,8 +18,8 @@ INTERNAL void _cdecl transpose33(__3_columnset *source, __3_columnset *target)
_MM_SHUFFLE(1, 0, 1, 0));
__m128 t1 = _mm_shuffle_ps(ALIGN(source->column[0]), ALIGN(source->column[1]),
_MM_SHUFFLE(3, 2, 3, 2));
__m128 t2 = _mm_shuffle_ps(ALIGN(source->column[2]), zeroVec, _MM_SHUFFLE(1, 0, 1, 0));
__m128 t3 = _mm_shuffle_ps(ALIGN(source->column[2]), zeroVec, _MM_SHUFFLE(3, 2, 3, 2));
__m128 t2 = _mm_shuffle_ps(ALIGN(source->column[2]), zero_vec, _MM_SHUFFLE(1, 0, 1, 0));
__m128 t3 = _mm_shuffle_ps(ALIGN(source->column[2]), zero_vec, _MM_SHUFFLE(3, 2, 3, 2));
target->column[0] = _mm_shuffle_ps(t0, t2, _MM_SHUFFLE(2, 0, 2, 0));
target->column[1] = _mm_shuffle_ps(t0, t2, _MM_SHUFFLE(3, 1, 3, 1));
@@ -32,8 +32,8 @@ INTERNAL void _cdecl transpose34(__3_columnset *source, __4_columnset *target)
_MM_SHUFFLE(1, 0, 1, 0));
__m128 t1 = _mm_shuffle_ps(ALIGN(source->column[0]), ALIGN(source->column[1]),
_MM_SHUFFLE(3, 2, 3, 2));
__m128 t2 = _mm_shuffle_ps(ALIGN(source->column[2]), zeroVec, _MM_SHUFFLE(1, 0, 1, 0));
__m128 t3 = _mm_shuffle_ps(ALIGN(source->column[2]), zeroVec, _MM_SHUFFLE(3, 2, 3, 2));
__m128 t2 = _mm_shuffle_ps(ALIGN(source->column[2]), zero_vec, _MM_SHUFFLE(1, 0, 1, 0));
__m128 t3 = _mm_shuffle_ps(ALIGN(source->column[2]), zero_vec, _MM_SHUFFLE(3, 2, 3, 2));
target->column[0] = _mm_shuffle_ps(t0, t2, _MM_SHUFFLE(2, 0, 2, 0));
target->column[1] = _mm_shuffle_ps(t0, t2, _MM_SHUFFLE(3, 1, 3, 1));
内核源代码/StgSharpC/ssc_internal.h
+9 -12
View File
@@ -12,26 +12,23 @@
#ifndef SSC_INTERNAL
#define SSC_INTERNAL
#ifdef _WIN32
#ifdef _MSC_VER
#define _CRT_SECURE_NO_WARNINGS 1
#define WIN32_LEAN_AND_MEAN
#define SSCDECL __cdecl
#define SSCAPI __declspec(dllexport)
#define FORCEINLINE __forceinline
#include <windows.h>
#elif defined _LINUX_
#elif defined(__clang__) || defined(__GNUC__)
#define SSCDECL __attribute__((cdecl))
#define SSCAPI __attribute__((dllexport))
#elif defined _APPLE_
#define SSCAPI __attribute__((visibility("default")))
#define FORCEINLINE __attribute__((always_inline))
#else
#define SSCAPI
#define SSCDECL
#define FORCEINLINE
#endif
#ifdef _WIN32
#define _CRT_SECURE_NO_WARNINGS 1
#endif
#define SSCDECL __cdecl
#define zeroVec _mm_setzero_ps()
#define SSCAPI_DEFINE
#define PRIVATE static
内核源代码/StgSharpC/ssc_intrinsic.h
+16 -8
View File
@@ -19,18 +19,19 @@
#define SSE_INSERT(source, target, reset3, reset2, reset1, reset0) \
(LASTTWO(source - 1) << 6) | (LASTTWO(target - 1) << 4) | (LAST(reset3) << 3) | \
(LAST(reset2) << 2) | (LAST(reset1) << 1) | LAST(reset0)
#define SSE_INSERT_SIMPLE(source, target) SSE_INSERT(source, target, 0, 0, 0, 0)
#define M256_PERMUTE(s3, s2, s1, s0) (LASTTWO(s3) << 6) | (LASTTWO(s2) << 4) | (LASTTWO(s1) << 2) | (LASTTWO(s0)
#define zero_vec _mm_setzero_ps()
#define zero_vec_256 _mm256_setzero_ps()
typedef union column_2 {
__m256 stream[1];
__m256 stream;
__m128 column[2];
float m[4][2];
} __2_columnset;
typedef struct column_3 {
typedef union column_3 {
__m256 part;
__m128 column[3];
float m[4][3];
@@ -79,13 +80,19 @@ INTERNAL float SSCDECL det_mat4(__4_columnset *matrix, __4_columnset *transpose)
#pragma region dot
INTERNAL void SSCDECL dot_41_sse(__4_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SSCDECL dot_31_sse(__3_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void FORCEINLINE SSCDECL dot_41_sse(__4_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void FORCEINLINE SSCDECL dot_31_sse(__3_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void FORCEINLINE SSCDECL dot_41_avx(__4_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void FORCEINLINE SSCDECL dot_31_avx(__3_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SSCDECL dot_42_sse(__4_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SSCDECL dot_32_sse(__3_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SSCDECL dot_42_avx(__4_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SSCDECL dot_32_avx(__3_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SSCDECL dot_42_sse(__4_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SSCDECL dot_42_avx(__4_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SSCDECL dot_43_sse(__4_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SSCDECL dot_43_avx(__4_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SSCDECL dot_43_512(__4_columnset *transpose, __m128 *vector, __m128 *ans);
#pragma endregion
@@ -105,6 +112,7 @@ typedef struct ssc_intrinsic {
DOTPROC dot_32;
DOTPROC dot_41;
DOTPROC dot_42;
DOTPROC dot_43;
VECTORNORMALIZEPROC normalize_3;
int (*string_quick_hash)(char *str, int length);
MATRIXTRANSPOSEPROC transpose_23;
内核源代码/StgSharpC/vector_dot.c
+96 -42
View File
@@ -1,93 +1,147 @@
#include "ssc_intrinsic.h"
#include "SSC_internal.h"
INTERNAL void __inline SSCDECL dot_41_sse(__4_columnset *transpose, __m128 *vector, __m128 *ans)
static const int avxdot_indices[] = { 0, 4, 1, 5, 2, 3, 6, 7 };
INTERNAL void FORCEINLINE SSCDECL dot_41_sse(__4_columnset *transpose, __m128 *vector, __m128 *ans)
{
*ans = _mm_dp_ps(transpose->column[0], *vector, 0B00011111);
*ans = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[1], *vector, 0B00011111),
*ans = _mm_dp_ps(transpose->column[0], *vector, 0B11110001);
*ans = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[1], *vector, 0B11110001),
SSE_INSERT_SIMPLE(1, 1));
*ans = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[2], *vector, 0B00011111),
*ans = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[2], *vector, 0B11110001),
SSE_INSERT_SIMPLE(1, 3));
*ans = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[3], *vector, 0B00011111),
*ans = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[3], *vector, 0B11110001),
SSE_INSERT_SIMPLE(1, 4));
}
INTERNAL void SSCDECL dot_31_sse(__3_columnset *transpose, __m128 *vector, __m128 *ans)
INTERNAL void SSCDECL dot_41_avx(__4_columnset *transpose, __m128 *vector, __m128 *ans)
{
*ans = _mm_dp_ps(transpose->column[0], *vector, 0B00011111);
*ans = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[1], *vector, 0B00011111),
SSE_INSERT_SIMPLE(1, 1));
*ans = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[2], *vector, 0B00011111),
__m256 _double = _mm256_set_m128(vector[0], vector[0]);
__m256 tmp = _mm256_dp_ps(transpose->half[0], _double, 0B11110001);
tmp = _mm256_blend_ps(tmp, _mm256_dp_ps(transpose->half[0], _double, 0B11110010),
0B00100010);
*ans = _mm256_castps256_ps128(
_mm256_permutevar8x32_ps(tmp, _mm256_loadu_epi32(avxdot_indices)));
}
INTERNAL void FORCEINLINE SSCDECL dot_31_sse(__3_columnset *transpose, __m128 *vector, __m128 *ans)
{
*ans = _mm_dp_ps(transpose->column[0], *vector, 0B11110001);
*ans = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[1], *vector, 0B11110001),
SSE_INSERT_SIMPLE(1, 2));
*ans = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[2], *vector, 0B11110001),
SSE_INSERT_SIMPLE(1, 3));
}
INTERNAL void FORCEINLINE SSCDECL dot_31_avx(__3_columnset *transpose, __m128 *vector, __m128 *ans)
{
__m256 _double = _mm256_set_m128(vector[0], vector[0]);
__m256 tmp = _mm256_dp_ps(transpose->part, _double, 0B11110001);
*ans = _mm256_castps256_ps128(
_mm256_permutevar8x32_ps(tmp, _mm256_loadu_epi32(avxdot_indices)));
*ans = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[2], *vector, 0B11110001),
SSE_INSERT_SIMPLE(1, 3));
}
INTERNAL void SSCDECL dot_42_sse(__4_columnset *transpose, __m128 *vector, __m128 *ans)
{
ans[0] = _mm_dp_ps(transpose->column[0], vector[0], 0B00011111);
ans[0] = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[1], vector[0], 0B00011111),
SSE_INSERT_SIMPLE(1, 1));
ans[0] = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[2], vector[0], 0B00011111),
ans[0] = _mm_dp_ps(transpose->column[0], vector[0], 0B11110001);
ans[0] = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[1], vector[0], 0B11110001),
SSE_INSERT_SIMPLE(1, 2));
ans[0] = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[2], vector[0], 0B11110001),
SSE_INSERT_SIMPLE(1, 3));
ans[0] = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[3], vector[0], 0B00011111),
ans[0] = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[3], vector[0], 0B11110001),
SSE_INSERT_SIMPLE(1, 4));
ans[1] = _mm_dp_ps(transpose->column[0], vector[1], 0B00011111);
ans[1] = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[1], vector[1], 0B00011111),
SSE_INSERT_SIMPLE(1, 1));
ans[1] = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[2], vector[1], 0B00011111),
ans[1] = _mm_dp_ps(transpose->column[0], vector[1], 0B11110001);
ans[1] = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[1], vector[1], 0B11110001),
SSE_INSERT_SIMPLE(1, 2));
ans[1] = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[2], vector[1], 0B11110001),
SSE_INSERT_SIMPLE(1, 3));
ans[1] = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[3], vector[1], 0B00011111),
ans[1] = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[3], vector[1], 0B11110001),
SSE_INSERT_SIMPLE(1, 4));
}
INTERNAL void SSCDECL dot_32_sse(__3_columnset *transpose, __m128 *vector, __m128 *ans)
{
ans[0] = _mm_dp_ps(transpose->column[0], vector[0], 0B00011111);
ans[0] = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[1], vector[0], 0B00011111),
SSE_INSERT_SIMPLE(1, 1));
ans[0] = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[2], vector[0], 0B00011111),
ans[0] = _mm_dp_ps(transpose->column[0], vector[0], 0B11110001);
ans[0] = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[1], vector[0], 0B11110001),
SSE_INSERT_SIMPLE(1, 2));
ans[0] = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[2], vector[0], 0B11110001),
SSE_INSERT_SIMPLE(1, 3));
ans[1] = _mm_dp_ps(transpose->column[0], vector[1], 0B00011111);
ans[1] = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[1], vector[1], 0B00011111),
SSE_INSERT_SIMPLE(1, 1));
ans[1] = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[2], vector[1], 0B00011111),
ans[1] = _mm_dp_ps(transpose->column[0], vector[1], 0B11110001);
ans[1] = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[1], vector[1], 0B11110001),
SSE_INSERT_SIMPLE(1, 2));
ans[1] = _mm_insert_ps(*ans, _mm_dp_ps(transpose->column[2], vector[1], 0B11110001),
SSE_INSERT_SIMPLE(1, 3));
}
INTERNAL void SSCDECL dot_42_avx(__4_columnset *transpose, __m128 *vector, __m128 *ans)
{
__2_columnset *ansPtr = (__2_columnset *)ans;
__2_columnset *ans_ptr = (__2_columnset *)ans, *right_ptr = (__2_columnset *)vector;
__m256 temp = ALIGN256(transpose->half[0]);
ansPtr->stream[0] = _mm256_dp_ps(temp, *(__m256 *)vector, 0B11110001);
ans_ptr->stream = _mm256_dp_ps(temp, right_ptr->stream, 0B11110001);
temp = _mm256_permute2f128_ps(temp, temp, 0B00000001);
ansPtr->stream[0] = _mm256_blend_ps(
ansPtr->stream[0], _mm256_dp_ps(temp, *(__m256 *)vector, 0B11110010), 0B001000010);
ans_ptr->stream = _mm256_blend_ps(
ans_ptr->stream, _mm256_dp_ps(temp, right_ptr->stream, 0B11110010), 0B00100010);
temp = ALIGN256(transpose->half[1]);
ansPtr->stream[0] = _mm256_blend_ps(
ansPtr->stream[0], _mm256_dp_ps(temp, *(__m256 *)vector, 0B11110100), 0B01000100);
ans_ptr->stream = _mm256_blend_ps(
ans_ptr->stream, _mm256_dp_ps(temp, right_ptr->stream, 0B11110100), 0B01000100);
temp = _mm256_permute2f128_ps(temp, temp, 0B00000001);
ansPtr->stream[0] = _mm256_blend_ps(
ansPtr->stream[0], _mm256_dp_ps(temp, *(__m256 *)vector, 0B11111000), 0B10001000);
ans_ptr->stream = _mm256_blend_ps(
ans_ptr->stream, _mm256_dp_ps(temp, right_ptr->stream, 0B11111000), 0B10001000);
ans[1] = _mm_shuffle_ps(ALIGN(ans[1]), ALIGN(ans[1]), _MM_SHUFFLE(2, 3, 0, 1));
}
INTERNAL void SSCDECL dot_32_avx(__3_columnset *transpose, __m128 *vector, __m128 *ans)
{
__2_columnset *ansPtr = (__2_columnset *)ans;
__2_columnset *ans_ptr = (__2_columnset *)ans, *right_ptr = (__2_columnset *)vector;
__m256 temp = transpose->part;
ansPtr->stream[0] = _mm256_dp_ps(temp, *(__m256 *)vector, 0B11110001);
ans_ptr->stream = _mm256_dp_ps(temp, right_ptr->stream, 0B11110001);
temp = _mm256_permute2f128_ps(temp, temp, 0B00000001);
ansPtr->stream[0] = _mm256_blend_ps(
ansPtr->stream[0], _mm256_dp_ps(temp, *(__m256 *)vector, 0B11110010), 0B001000010);
ans_ptr->stream = _mm256_blend_ps(
ans_ptr->stream, _mm256_dp_ps(temp, right_ptr->stream, 0B11110010), 0B00100010);
temp = _mm256_set_m128(transpose->column[2], transpose->column[2]);
ansPtr->stream[0] = _mm256_blend_ps(
ansPtr->stream[0], _mm256_dp_ps(temp, *(__m256 *)vector, 0B11110100), 0B01000100);
ans_ptr->stream = _mm256_blend_ps(
ans_ptr->stream, _mm256_dp_ps(temp, right_ptr->stream, 0B11110100), 0B01000100);
ans[1] = _mm_shuffle_ps(ALIGN(ans[1]), ALIGN(ans[1]), _MM_SHUFFLE(2, 2, 0, 1));
}
INTERNAL void SSCDECL dot_43_sse(__4_columnset *transpose, __m128 *vector, __m128 *ans)
{
dot_41_sse(transpose, vector, ans);
dot_41_sse(transpose, vector + 1, ans + 1);
dot_41_sse(transpose, vector + 2, ans + 2);
}
INTERNAL void SSCDECL dot_43_avx(__4_columnset *transpose, __m128 *vector, __m128 *ans)
{
__3_columnset *ans_ptr = (__3_columnset *)ans, *right_ptr = (__3_columnset *)vector;
__m256 temp = ALIGN256(transpose->half[0]);
__m256 third = _mm256_set_m128(right_ptr->column[2], right_ptr->column[2]);
ans_ptr->part = _mm256_dp_ps(temp, right_ptr->part, 0B11110001);
__m256 ans_third = _mm256_dp_ps(temp, third, 0B11110001);
temp = _mm256_permute2f128_ps(temp, temp, 0B00000001);
ans_ptr->part = _mm256_blend_ps(
ans_ptr->part, _mm256_dp_ps(temp, right_ptr->part, 0B11110010), 0B00100010);
temp = ALIGN256(transpose->half[1]);
ans_ptr->part = _mm256_blend_ps(
ans_ptr->part, _mm256_dp_ps(temp, right_ptr->part, 0B11110100), 0B01000100);
ans_third = _mm256_blend_ps(ans_third, _mm256_dp_ps(temp, third, 0B11110010), 0B00100010);
temp = _mm256_permute2f128_ps(temp, temp, 0B00000001);
ans_ptr->part = _mm256_blend_ps(
ans_ptr->part, _mm256_dp_ps(temp, right_ptr->part, 0B11111000), 0B10001000);
ans[1] = _mm_shuffle_ps(ALIGN(ans[1]), ALIGN(ans[1]), _MM_SHUFFLE(2, 3, 0, 1));
ans[2] = _mm256_castps256_ps128(
_mm256_permutevar8x32_ps(ans_third, _mm256_loadu_epi32(avxdot_indices)));
}