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

arrange of c lib

Browse Source
This commit is contained in:
Youmu Liu
2025-08-27 21:57:14 +08:00
parent 9f6e3fe13a
commit a83c5e9246
25 changed files with 312 additions and 2840 deletions
Download Patch File Download Diff File Expand all files Collapse all files
StgSharp.Native/StgSharp.Native.vcxproj
+7 -9
View File
@@ -159,18 +159,16 @@
</Link>
</ItemDefinitionGroup>
<ItemGroup>
<ClCompile Include="fma_sse.c" />
<ClCompile Include="sn-src\k_basicop.c" />
<ClCompile Include="sn-src\dllmain.c" />
<ClCompile Include="sn-src\FactorialSIMD.c" />
<ClCompile Include="sn-src\hash.c" />
<ClCompile Include="sn-src\intrinsic.c" />
<ClCompile Include="sn-src\str_factorial.c" />
<ClCompile Include="sn-src\k_fma.c" />
<ClCompile Include="sn-src\str_hash.c" />
<ClCompile Include="sn-src\load_intrinsic.c" />
<ClCompile Include="sn-src\load_glfw.c" />
<ClCompile Include="sn-src\mat2.c" />
<ClCompile Include="sn-src\mat3.c" />
<ClCompile Include="sn-src\mat4.c" />
<ClCompile Include="sn-src\stb_image.c" />
<ClCompile Include="sn-src\stgsharp_native.c" />
<ClCompile Include="sn-src\string.c" />
<ClCompile Include="sn-src\sn_native.c" />
<ClCompile Include="sn-src\str_string.c" />
<ClCompile Include="sn-src\vector.c" />
<ClCompile Include="sn-src\vector_dot.c" />
</ItemGroup>
StgSharp.Native/sn-head/sn_intrinsic.h
+25 -75
View File
@@ -32,12 +32,12 @@ typedef union uint64_u {
uint32_t u32[2];
} uint64_u;
typedef union mat_kernel{
typedef union mat_kernel {
__m128 xmm[4];
__m256 ymm[2];
__m512 zmm[1];
float m[4][4];
}mat_kernel;
} mat_kernel;
typedef union column_2 {
__m256 stream;
@@ -64,66 +64,33 @@ typedef void(SN_DECL *MATRIXTRANSPOSEPROC)(void *source, void *target);
typedef void(SN_DECL *MATRIXDETPROC)(void *mat, void *transpose);
typedef void(SN_DECL *MATARITHMETIC)(void *left, void *right, void *ans);
typedef void(SN_DECL *SCALERMATARITHMETIC)(void *left, void *right, void *ans);
typedef void(SN_DECL* MATFMAPROC)(mat_kernel* left, mat_kernel* right, mat_kernel* ans);
typedef void(SN_DECL *MATFMAPROC)(mat_kernel *left, mat_kernel *right, mat_kernel *ans);
typedef void(SN_DECL *SCALARMATARITHMETIC)(mat_kernel *matrix, float scalar, mat_kernel *ans);
typedef void(SN_DECL *MATTRANSPOSEPROC)(mat_kernel *source, mat_kernel *target);
typedef uint64_t(SN_DECL *FACTORIALROC)(int n);
#pragma region matix function
INTERNAL void normalize(__m128 *source, __m128 *target);
INTERNAL void SN_DECL addmatrix2_sse(__2_columnset *left, __2_columnset *right, __2_columnset *ans);
INTERNAL void SN_DECL addmatrix2_avx(__2_columnset *left, __2_columnset *right, __2_columnset *ans);
INTERNAL void SN_DECL submatrix2_sse(__2_columnset *left, __2_columnset *right, __2_columnset *ans);
INTERNAL void SN_DECL addmatrix2_avx(__2_columnset *left, __2_columnset *right, __2_columnset *ans);
INTERNAL void SN_DECL transpose23(__2_columnset const *source, __3_columnset *target);
INTERNAL void SN_DECL transpose24(__2_columnset const *source, __4_columnset *target);
INTERNAL void SN_DECL addmatrix3_sse(__3_columnset *left, __3_columnset *right, __3_columnset *ans);
INTERNAL void SN_DECL addmatrix3_avx(__3_columnset *left, __3_columnset *right, __3_columnset *ans);
INTERNAL void SN_DECL addmatrix3_512(__3_columnset *left, __3_columnset *right, __3_columnset *ans);
INTERNAL void SN_DECL submatrix3_sse(__3_columnset *left, __3_columnset *right, __3_columnset *ans);
INTERNAL void SN_DECL submatrix3_avx(__3_columnset *left, __3_columnset *right, __3_columnset *ans);
INTERNAL void SN_DECL submatrix3_512(__3_columnset *left, __3_columnset *right, __3_columnset *ans);
INTERNAL void SN_DECL transpose32(__3_columnset const *source, __2_columnset *target);
INTERNAL void SN_DECL transpose33(__3_columnset const *source, __3_columnset *target);
INTERNAL void SN_DECL transpose34(__3_columnset const *source, __4_columnset *target);
INTERNAL float SN_DECL det_mat3(__3_columnset *matrix, __3_columnset *tranpose);
INTERNAL void SN_DECL addmatrix4_sse(__4_columnset *left, __4_columnset *right, __4_columnset *ans);
INTERNAL void SN_DECL addmatrix4_avx(__4_columnset *left, __4_columnset *right, __4_columnset *ans);
INTERNAL void SN_DECL addmatrix4_512(__4_columnset *left, __4_columnset *right, __4_columnset *ans);
INTERNAL void SN_DECL submatrix4_sse(__4_columnset *left, __4_columnset *right, __4_columnset *ans);
INTERNAL void SN_DECL submatrix4_avx(__4_columnset *left, __4_columnset *right, __4_columnset *ans);
INTERNAL void SN_DECL submatrix4_512(__4_columnset *left, __4_columnset *right, __4_columnset *ans);
INTERNAL void SN_DECL transpose42(__4_columnset const *source, __2_columnset *target);
INTERNAL void SN_DECL transpose43(__4_columnset const *source, __3_columnset *target);
INTERNAL void SN_DECL transpose44(__4_columnset const *source, __4_columnset *target);
INTERNAL float SN_DECL det_mat4(__4_columnset *matrix, __4_columnset *transpose);
INTERNAL void SN_DECL kernel_add_sse(mat_kernel const *left, mat_kernel const *right, mat_kernel *ans);
INTERNAL void SN_DECL kernel_add_avx(mat_kernel const *left, mat_kernel const *right, mat_kernel *ans);
INTERNAL void SN_DECL kernel_add_512(mat_kernel const *left, mat_kernel const *right, mat_kernel *ans);
INTERNAL void SN_DECL kernel_sub_sse(mat_kernel const *left, mat_kernel const *right, mat_kernel *ans);
INTERNAL void SN_DECL kernel_sub_avx(mat_kernel const *left, mat_kernel const *right, mat_kernel *ans);
INTERNAL void SN_DECL kernel_sub_512(mat_kernel const *left, mat_kernel const *right, mat_kernel *ans);
INTERNAL void SN_DECL kernel_transpose_sse(mat_kernel const *source, mat_kernel *target);
INTERNAL void SN_DECL kernel_transpose_avx(mat_kernel const *source, mat_kernel *target);
INTERNAL void SN_DECL kernel_transpose_512(mat_kernel const *source, mat_kernel *target);
INTERNAL void SN_DECL kernel_scalar_mul_sse(mat_kernel const *matrix, float const scalar, mat_kernel *ans);
INTERNAL void SN_DECL kernel_scalar_mul_avx(mat_kernel const *matrix, float const scalar, mat_kernel *ans);
INTERNAL void SN_DECL kernel_scalar_mul_512(mat_kernel const *matrix, float const scalar, mat_kernel *ans);
#pragma endregion
#pragma region fma
INTERNAL void SN_DECL fma_44_sse(mat_kernel* left, mat_kernel* right);
#pragma endregion
#pragma region dot
INTERNAL void SN_DECL dot_41_sse(__4_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SN_DECL dot_31_sse(__3_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SN_DECL dot_41_avx(__4_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SN_DECL dot_31_avx(__3_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SN_DECL dot_32_sse(__3_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SN_DECL dot_32_avx(__3_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SN_DECL dot_42_sse(__4_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SN_DECL dot_42_avx(__4_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SN_DECL dot_43_sse(__4_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SN_DECL dot_43_avx(__4_columnset *transpose, __m128 *vector, __m128 *ans);
//INTERNAL void SN_DECL dot_43_512(__4_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SN_DECL kernel_fma_sse(mat_kernel *left, mat_kernel *right);
#pragma endregion
@@ -138,34 +105,17 @@ INTERNAL int SN_DECL index_pair_sse(short const *str, uint32_t target, int lengt
#pragma endregion
typedef struct ssc_intrinsic {
MATARITHMETIC add_m2;
MATARITHMETIC add_m3;
MATARITHMETIC add_m4;
typedef struct sn_intrinsic {
int (*city_hash_simplify)(byte *str, int length);
MATRIXDETPROC det_matrix33;
MATRIXDETPROC det_matrix44;
DOTPROC dot_31;
DOTPROC dot_32;
DOTPROC dot_41;
DOTPROC dot_42;
DOTPROC dot_43;
FACTORIALROC factorial_simd;
void (*fma_44)(void* left, void* right, void* ans);
int (*index_pair)(short const *str, uint32_t target, int length);
MATARITHMETIC kernel_add;
void (*kernel_fma)(void *left, void *right, void *ans);
SCALARMATARITHMETIC kernel_scalar_mul;
MATARITHMETIC kernel_sub;
MATTRANSPOSEPROC kernel_transpose;
VECTORNORMALIZEPROC normalize_3;
MATARITHMETIC sub_m2;
MATARITHMETIC sub_m3;
MATARITHMETIC sub_m4;
MATRIXTRANSPOSEPROC transpose_23;
MATRIXTRANSPOSEPROC transpose_24;
MATRIXTRANSPOSEPROC transpose_32;
MATRIXTRANSPOSEPROC transpose_33;
MATRIXTRANSPOSEPROC transpose_34;
MATRIXTRANSPOSEPROC transpose_42;
MATRIXTRANSPOSEPROC transpose_43;
MATRIXTRANSPOSEPROC transpose_44;
} ssc_intrinsic;
} sn_intrinsic;
typedef enum most_advanced_instruction {
#ifdef _MSC_VER
StgSharp.Native/sn-src/FactorialSIMD.c
-23
View File
@@ -1,23 +0,0 @@
#include "sn_internal.h"
#include "sn_intrinsic.h"
INTERNAL __m128i facIncrement = { 4, 4, 4, 4 };
INTERNAL __m128i facBase = { 1, 2, 3, 4 };
INTERNAL uint64_t SN_DECL factorial_simd_sse(int n)
{
__m128i baseResult = facBase;
__m128i facBaseLocal = facBase;
const int cycle = (n - 3) >> 2;
const uint32_t rest = (cycle << 2) + 1;
for (size_t i = 0; i < cycle; i++) {
facBaseLocal = _mm_add_epi32(facBaseLocal, facIncrement);
baseResult = _mm_mullo_epi32(baseResult, facBaseLocal);
}
uint64_t result = (uint64_t)baseResult.m128i_i32[0] * (uint64_t)baseResult.m128i_i32[1] *
(uint64_t)baseResult.m128i_i32[2] * (uint64_t)baseResult.m128i_i32[3];
for (uint64_t i = rest; i < n; i++) {
result *= i ;
}
return result;
}
StgSharp.Native/sn-src/hash.c
-45
View File
@@ -1,45 +0,0 @@
#include "sn_intrinsic.h"
#include "StgSharpNative.h"
INTERNAL int SN_DECL city_hash_simplify_sse(byte const *str, int const length)
{
uint32_t hash = length, size = length << 1;
byte *end = str + size, *ptr = str;
if (size < 4) {
for (; ptr < end; ptr += 2) {
hash = _mm_crc32_u16(hash, *ptr);
}
} else if (size < 16) {
byte *last = end - 4;
for (; ptr < end; ptr += 4) {
hash = _mm_crc32_u32(hash, *(uint32_t *)ptr);
}
for (; ptr < end; ptr++) {
hash = _mm_crc32_u16(hash, *ptr);
}
} else {
byte *last_128 = end - 16;
byte *last_32 = end - 4;
uint64_t hash64 = size;
for (; ptr < last_128; ptr += 16) {
__m128i vec = _mm_loadu_si32((int *)ptr);
hash64 = _mm_crc32_u64(hash64, _mm_extract_epi64(vec, 0));
hash64 = _mm_crc32_u64(hash64, _mm_extract_epi64(vec, 1));
}
hash = _mm_crc32_u8(hash, hash64);
hash = _mm_crc32_u8(hash, hash64 >> 32);
for (; ptr < last_32; ptr += 4) {
hash = _mm_crc32_u32(hash,*(uint32_t *)ptr);
}
for (; ptr < end; ptr += 2) {
hash = _mm_crc32_u8(hash,*ptr);
}
}
hash ^= hash >> 16;
hash *= 0x85ebca6b;
hash ^= hash >> 13;
hash *= 0xc2b2ae35;
hash ^= hash >> 16;
return hash;
}
StgSharp.Native/sn-src/intrinsic.c
-146
View File
@@ -1,146 +0,0 @@
#include "StgSharpNative.h"
#include "sn_intrinsic.h"
#if defined(_MSC_VER) && !defined(__clang__)
#include <intrin.h>
#include <isa_availability.h>
#else
#include <cpuid.h>
#endif // _MSC_VER
PRIVATE void load_intrinsic_sse(ssc_intrinsic *context);
PRIVATE void load_intrinsic_avx(ssc_intrinsic *context);
PRIVATE void load_intrinsic_512(ssc_intrinsic *context);
PRIVATE int check_isa(most_advanced_instruction instruction_family)
{
if (instruction_family != NEON) {
#if defined(_MSC_VER) && !defined(__clang__)
return __check_isa_support(instruction_family, 0);
#else
switch (instruction_family) {
case SSE:
return __builtin_cpu_supports("sse4.2");
case AVX256:
return __builtin_cpu_supports("avx2");
case AVX512:
return __builtin_cpu_supports("avx512f");
default:
return 0;
}
#endif // _MSC_VER
}
#if defined(__ARM_NEON) || defined(__ARM_NEON__)
return 1;
#else
return 0;
#endif
}
SN_API int load_intrinsic_function(void *intrinsic_context)
{
sn_null_assert(intrinsic_context);
ssc_intrinsic *context = (ssc_intrinsic *)intrinsic_context;
if (check_isa(AVX512)) {
load_intrinsic_512(context);
return 3;
} else if (check_isa(AVX256)) {
load_intrinsic_avx(context);
return 2;
} else if (check_isa(SSE)) {
load_intrinsic_sse(context);
return 1;
} else if (check_isa(NEON)) {
return -1;
}
return 0;
}
void load_intrinsic_sse(ssc_intrinsic *context)
{
context->add_m2 = addmatrix2_sse;
context->add_m3 = addmatrix3_sse;
context->add_m4 = addmatrix4_sse;
context->transpose_23 = transpose23;
context->transpose_24 = transpose24;
context->transpose_32 = transpose32;
context->transpose_33 = transpose33;
context->transpose_34 = transpose34;
context->transpose_42 = transpose42;
context->transpose_43 = transpose43;
context->transpose_44 = transpose44;
context->normalize_3 = normalize;
context->det_matrix33 = det_mat3;
context->det_matrix44 = det_mat4;
context->dot_31 = dot_31_sse;
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->city_hash_simplify = city_hash_simplify_sse;
context->index_pair = index_pair_sse;
context->sub_m2 = submatrix2_sse;
context->sub_m3 = submatrix3_sse;
context->sub_m4 = submatrix4_sse;
context->factorial_simd = factorial_simd_sse;
context->fma_44 = fma_44_sse;
}
void load_intrinsic_avx(ssc_intrinsic *context)
{
context->add_m2 = addmatrix2_avx;
context->add_m3 = addmatrix3_avx;
context->add_m4 = addmatrix4_avx;
context->transpose_23 = transpose23;
context->transpose_24 = transpose24;
context->transpose_32 = transpose32;
context->transpose_33 = transpose33;
context->transpose_34 = transpose34;
context->transpose_42 = transpose42;
context->transpose_43 = transpose43;
context->transpose_44 = transpose44;
context->normalize_3 = normalize;
context->det_matrix33 = det_mat3;
context->det_matrix44 = det_mat4;
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->city_hash_simplify = city_hash_simplify_sse;
context->index_pair = index_pair_sse;
context->factorial_simd = factorial_simd_sse;
context->fma_44 = fma_44_sse;
}
void load_intrinsic_512(ssc_intrinsic *context)
{
context->add_m2 = addmatrix2_avx;
context->add_m3 = addmatrix3_512;
context->add_m4 = addmatrix4_512;
context->transpose_23 = transpose23;
context->transpose_24 = transpose24;
context->transpose_32 = transpose32;
context->transpose_33 = transpose33;
context->transpose_34 = transpose34;
context->transpose_42 = transpose42;
context->transpose_43 = transpose43;
context->transpose_44 = transpose44;
context->normalize_3 = normalize;
context->det_matrix33 = det_mat3;
context->det_matrix44 = det_mat4;
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->city_hash_simplify = city_hash_simplify_sse;
context->index_pair = index_pair_sse;
context->factorial_simd = factorial_simd_sse;
context->fma_44 = fma_44_sse;
}
StgSharp.Native/sn-src/mat2.c
-57
View File
@@ -1,57 +0,0 @@
#include "StgSharpNative.h"
#include "sn_intrinsic.h"
INTERNAL void __cdecl transpose23(__2_columnset *source, __3_columnset *target)
{
__m128 t0 = _mm_shuffle_ps(ALIGN(source->column[0]), ALIGN(source->column[1]),
_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));
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)
{
__m128 t0 = _mm_shuffle_ps(ALIGN(source->column[0]), ALIGN(source->column[1]),
_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));
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)
{
__m128 temp0 = _mm_add_ps(ALIGN(left->column[0]), ALIGN(right->column[0]));
__m128 temp1 = _mm_add_ps(ALIGN(left->column[1]), ALIGN(right->column[1]));
_mm_storeu_ps(&ans->m[0][0], temp0);
_mm_storeu_ps(&ans->m[0][1], temp1);
}
INTERNAL void __cdecl addmatrix2_avx(__2_columnset *left, __2_columnset *right, __2_columnset *ans)
{
__m256 temp = _mm256_add_ps(ALIGN256(left->column[0]), ALIGN256(right->column[0]));
_mm256_storeu_ps(&ans->m[0][0], temp);
}
INTERNAL void __cdecl submatrix2_sse(__2_columnset *left, __2_columnset *right, __2_columnset *ans)
{
__m128 temp0 = _mm_sub_ps(ALIGN(left->column[0]), ALIGN(right->column[0]));
__m128 temp1 = _mm_sub_ps(ALIGN(left->column[1]), ALIGN(right->column[1]));
_mm_storeu_ps(&ans->m[0][0], temp0);
_mm_storeu_ps(&ans->m[0][1], temp1);
}
INTERNAL void __cdecl submatrix2_avx(__2_columnset *left, __2_columnset *right, __2_columnset *ans)
{
__m256 temp = _mm256_sub_ps(ALIGN256(left->column[0]), ALIGN256(right->column[0]));
_mm256_storeu_ps(&ans->m[0][0], temp);
}
StgSharp.Native/sn-src/mat3.c
-130
View File
@@ -1,130 +0,0 @@
#include "StgSharpNative.h"
#include "sn_intrinsic.h"
#include <stdio.h>
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]), 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));
}
INTERNAL void _cdecl transpose33(__3_columnset *source, __3_columnset *target)
{
__m128 t0 = _mm_shuffle_ps(ALIGN(source->column[0]), ALIGN(source->column[1]),
_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]), 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));
target->column[2] = _mm_shuffle_ps(t1, t3, _MM_SHUFFLE(2, 0, 2, 0));
}
INTERNAL void _cdecl transpose34(__3_columnset *source, __4_columnset *target)
{
__m128 t0 = _mm_shuffle_ps(ALIGN(source->column[0]), ALIGN(source->column[1]),
_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]), 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));
target->column[2] = _mm_shuffle_ps(t1, t3, _MM_SHUFFLE(2, 0, 2, 0));
target->column[3] = _mm_shuffle_ps(t1, t3, _MM_SHUFFLE(3, 1, 3, 1));
}
INTERNAL float _cdecl det_mat3_internal(__m128 const c0, __m128 const c1, __m128 const c2)
{
//let adding row aligned
__m128 addc1 = _mm_shuffle_ps(c1, c1, _MM_SHUFFLE(3, 0, 2, 1));
__m128 addc2 = _mm_shuffle_ps(c2, c2, _MM_SHUFFLE(3, 1, 0, 2));
//let sub row aligned
__m128 subc1 = _mm_shuffle_ps(c1, c1, _MM_SHUFFLE(3, 1, 0, 2));
__m128 subc2 = _mm_shuffle_ps(c2, c2, _MM_SHUFFLE(3, 0, 2, 1));
//multple all add row
__m128 cache = _mm_mul_ps(c0, addc1);
cache = _mm_mul_ps(cache, addc2);
//multiple all sub row
__m128 sub = _mm_mul_ps(c0, subc1);
sub = _mm_mul_ps(sub, subc2);
//main-corner sub sub-corner
cache = _mm_sub_ps(cache, sub);
__m128 ret = _mm_hadd_ps(cache, cache); //element 1 add element 2
ret = _mm_add_ps(ret, cache); //(e1 add e2) add element 3
return ret.m128_f32[2];
}
INTERNAL float _cdecl det_mat3(__3_columnset *source, __3_columnset *transpose)
{
return det_mat3_internal(source->column[0], source->column[1], source->column[2]);
}
#pragma region add
INTERNAL void __cdecl addmatrix3_sse(__3_columnset *left, __3_columnset *right, __3_columnset *ans)
{
__m128 temp0 = _mm_add_ps(ALIGN(left->column[0]), ALIGN(right->column[0]));
__m128 temp1 = _mm_add_ps(ALIGN(left->column[1]), ALIGN(right->column[1]));
__m128 temp2 = _mm_add_ps(ALIGN(left->column[2]), ALIGN(right->column[2]));
_mm_storeu_ps(&ans->m[0][0], temp0);
_mm_storeu_ps(&ans->m[0][1], temp1);
_mm_storeu_ps(&ans->m[0][2], temp2);
}
INTERNAL void __cdecl addmatrix3_avx(__3_columnset *left, __3_columnset *right, __3_columnset *ans)
{
__m256 temp0 = _mm256_add_ps(ALIGN256(left->part), ALIGN256(right->part));
__m128 temp1 = _mm_add_ps(ALIGN(left->column[2]), ALIGN(right->column[2]));
_mm256_storeu_ps(&ans->m[0][0], temp0);
_mm_storeu_ps(&ans->m[0][2], temp1);
}
INTERNAL void __cdecl addmatrix3_512(__3_columnset *left, __3_columnset *right, __3_columnset *ans)
{
__m512 temp = _mm512_add_ps(_mm512_loadu_ps(left), _mm512_loadu_ps(right));
_mm512_mask_store_ps(ans, 0x0fff, temp);
}
INTERNAL void __cdecl submatrix3_sse(__3_columnset *left, __3_columnset *right, __3_columnset *ans)
{
__m128 temp0 = _mm_sub_ps(ALIGN(left->column[0]), ALIGN(right->column[0]));
__m128 temp1 = _mm_sub_ps(ALIGN(left->column[1]), ALIGN(right->column[1]));
__m128 temp2 = _mm_sub_ps(ALIGN(left->column[2]), ALIGN(right->column[2]));
_mm_storeu_ps(&ans->m[0][0], temp0);
_mm_storeu_ps(&ans->m[0][1], temp1);
_mm_storeu_ps(&ans->m[0][2], temp2);
}
INTERNAL void __cdecl submatrix3_avx(__3_columnset *left, __3_columnset *right, __3_columnset *ans)
{
__m256 temp0 = _mm256_sub_ps(ALIGN256(left->part), ALIGN256(right->part));
__m128 temp1 = _mm_sub_ps(ALIGN(left->column[2]), ALIGN(right->column[2]));
_mm256_storeu_ps(&ans->m[0][0], temp0);
_mm_storeu_ps(&ans->m[0][2], temp1);
}
INTERNAL void __cdecl submatrix3_512(__3_columnset *left, __3_columnset *right, __3_columnset *ans)
{
__m512 temp = _mm512_sub_ps(_mm512_loadu_ps(left), _mm512_loadu_ps(right));
_mm512_mask_store_ps(ans, 0x0fff, temp);
}
#pragma endregion
StgSharp.Native/sn-src/mat4.c
-103
View File
@@ -1,103 +0,0 @@
#include "StgSharpNative.h"
#include "sn_intrinsic.h"
#include <stdio.h>
#include <pmmintrin.h>
INTERNAL void __cdecl transpose42(__4_columnset const *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]), ALIGN(source->column[3]),
_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));
}
INTERNAL void __cdecl transpose43(__4_columnset const *source, __3_columnset *target)
{
__m128 t0 = _mm_shuffle_ps(ALIGN(source->column[0]), ALIGN(source->column[1]),
_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]), ALIGN(source->column[3]),
_MM_SHUFFLE(1, 0, 1, 0));
__m128 t3 = _mm_shuffle_ps(ALIGN(source->column[2]), ALIGN(source->column[3]),
_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));
target->column[2] = _mm_shuffle_ps(t1, t3, _MM_SHUFFLE(2, 0, 2, 0));
}
INTERNAL void __cdecl transpose44(__4_columnset const *source, __4_columnset *target)
{
__m128 t0 = _mm_shuffle_ps(ALIGN(source->column[0]), ALIGN(source->column[1]),
_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]), ALIGN(source->column[3]),
_MM_SHUFFLE(1, 0, 1, 0));
__m128 t3 = _mm_shuffle_ps(ALIGN(source->column[2]), ALIGN(source->column[3]),
_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));
target->column[2] = _mm_shuffle_ps(t1, t3, _MM_SHUFFLE(2, 0, 2, 0));
target->column[3] = _mm_shuffle_ps(t1, t3, _MM_SHUFFLE(3, 1, 3, 1));
}
INTERNAL float __cdecl det_mat4(__4_columnset const *mat_ptr, __4_columnset const *transpose)
{
__m128 mc0 = ALIGN(mat_ptr->column[0]);
__m128 mc1 = ALIGN(mat_ptr->column[1]);
__m128 mc2 = ALIGN(mat_ptr->column[2]);
__m128 mc3 = ALIGN(mat_ptr->column[3]);
__m128 cogeneration =
_mm_set_ps(det_mat3_internal(mc0, mc1, mc2), det_mat3_internal(mc0, mc1, mc3),
det_mat3_internal(mc0, mc2, mc3), det_mat3_internal(mc1, mc2, mc3));
__m128 retvec = _mm_mul_ps(ALIGN(transpose->column[3]), cogeneration);
retvec = _mm_hsub_ps(retvec, retvec);
retvec = _mm_hadd_ps(retvec, retvec);
return -retvec.m128_f32[0];
}
INTERNAL void SN_DECL addmatrix4_sse(__4_columnset *left, __4_columnset *right, __4_columnset *ans)
{
ans->column[0] = _mm_add_ps(ALIGN(left->column[0]), ALIGN(right->column[0]));
ans->column[1] = _mm_add_ps(ALIGN(left->column[1]), ALIGN(right->column[1]));
ans->column[2] = _mm_add_ps(ALIGN(left->column[2]), ALIGN(right->column[2]));
ans->column[3] = _mm_add_ps(ALIGN(left->column[3]), ALIGN(right->column[3]));
}
INTERNAL void SN_DECL addmatrix4_avx(__4_columnset *left, __4_columnset *right, __4_columnset *ans)
{
ans->half[0] = _mm256_add_ps(ALIGN256(left->half[0]), ALIGN256(right->half[0]));
ans->half[1] = _mm256_add_ps(ALIGN256(left->half[1]), ALIGN256(right->half[1]));
}
INTERNAL void SN_DECL addmatrix4_512(__4_columnset *left, __4_columnset *right, __4_columnset *ans)
{
ans->stream = _mm512_add_ps(ALIGN512(left->stream), ALIGN512(right->stream));
}
INTERNAL void SN_DECL submatrix4_sse(__4_columnset *left, __4_columnset *right, __4_columnset *ans)
{
ans->column[0] = _mm_sub_ps(ALIGN(left->column[0]), ALIGN(right->column[0]));
ans->column[1] = _mm_sub_ps(ALIGN(left->column[1]), ALIGN(right->column[1]));
ans->column[2] = _mm_sub_ps(ALIGN(left->column[2]), ALIGN(right->column[2]));
ans->column[3] = _mm_sub_ps(ALIGN(left->column[3]), ALIGN(right->column[3]));
}
INTERNAL void SN_DECL submatrix4_avx(__4_columnset *left, __4_columnset *right, __4_columnset *ans)
{
ans->half[0] = _mm256_sub_ps(ALIGN256(left->half[0]), ALIGN256(right->half[0]));
ans->half[1] = _mm256_sub_ps(ALIGN256(left->half[1]), ALIGN256(right->half[1]));
}
INTERNAL void SN_DECL submatrix4_512(__4_columnset *left, __4_columnset *right, __4_columnset *ans)
{
ans->stream = _mm512_sub_ps(ALIGN512(left->stream), ALIGN512(right->stream));
}
StgSharp.Native/sn-src/stgsharp_native.c
-75
View File
@@ -1,75 +0,0 @@
#include "StgSharpNative.h"
#include "gl.h"
#include "../glfw/glfw3.h"
#include <stdio.h>
#include <string.h>
extern GladGLContext *currentContext;
extern char infolog[512];
SN_API int SN_DECL glCheckShaderStat(GladGLContext *context, uint64_t shaderHandle, int key,
char **logRef)
{
int stat = 0;
context->GetShaderiv(shaderHandle, key, &stat);
if (!stat) {
context->GetShaderInfoLog(shaderHandle, 512, NULL, infolog);
}
*logRef = infolog;
return stat;
}
SN_API void SN_DECL initGL(int majorVersion, int minorVersion)
{
glfwWindowHint(GLFW_CONTEXT_VERSION_MAJOR, majorVersion);
glfwWindowHint(GLFW_CONTEXT_VERSION_MINOR, minorVersion);
glfwWindowHint(GLFW_OPENGL_PROFILE, GLFW_OPENGL_CORE_PROFILE);
glfwWindowHint(GLFW_RESIZABLE, GLFW_TRUE);
}
SN_API GLFWglproc SN_DECL loadGlfuncDefault(char *procName)
{
void *ret = glfwGetProcAddress(procName);
// printf("%llu\n", (uint64_t)ret);
if (ret == NULL) {
strcpy_s(infolog,30, "Failed to load OpenGL api:");
for (size_t i = 0; i < sizeof(procName); i++) {
infolog[30 + i] = procName[i];
}
//printf("%s\n", infolog);
}
return (GLFWglproc)ret;
}
SN_API unsigned int linkShaderProgram(GladGLContext *context, GLuint shaderProgram)
{
context->LinkProgram(shaderProgram);
int success;
context->GetProgramiv(shaderProgram, GL_LINK_STATUS, &success);
if (!success) {
context->GetProgramInfoLog(shaderProgram, 512, NULL, infolog);
return 0;
} else {
return shaderProgram;
}
}
SN_API void SN_DECL loadImageData(char *location, Image *out, imageLoader loader)
{
if (loader == NULL) {
loader = stbi_load;
}
out->pixelPtr = loader(location, &(out->width), &(out->height), &(out->channel), 0);
}
SN_API void SN_DECL unloadImageData(Image *out)
{
free(out->pixelPtr);
}
SN_API char *SN_DECL readLog()
{
infolog[511] = '\0';
return infolog;
}
StgSharp.Native/sn-src/string.c
-53
View File
@@ -1,53 +0,0 @@
#include "sn_intrinsic.h"
int 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 vec, com;
while (index < end) {
vec = _mm_loadu_epi32(str + index);
com = _mm_cmpeq_epi16(vec, s);
result = _mm_movemask_epi8(com);
if (result) {
for (size_t i = 0; i < 7; i++) {
if (((result & 0xF) == 0xF)) {
return index + i;
}
if (result == 0) {
break;
}
result >>= 2;
}
}
index += 1;
vec = _mm_loadu_epi32(str + index);
com = _mm_cmpeq_epi16(vec, s);
result = _mm_movemask_epi8(com);
if (result) {
for (size_t i = 0; i < 7; i++) {
if (((result & 0xF) == 0xF)) {
return index + i;
}
if (result == 0) {
break;
}
result >>= 2;
}
}
index += 7;
}
for (; (index < size - 1) && (!result); index += 2) {
result = *(int *)(str + index) == target;
}
if (result) {
return index;
}
return -1;
}