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

Daily update on 20250312-074647 by Nitload Youmu. Additional Message : Change: Temp not want to compile RULE and PROCEDURE

Browse Source 
Add: Add factory method to generate more kind of binary operator node
Add: more SIMD intrinsic C function

Signed-off-by: Nitload <YoumuLiu_NSI@whv0t.onmicrosoft.com>
This commit is contained in:
Nitload
2025-03-12 07:46:48 +08:00
parent 16b813646d
commit 9647c2c68a
29 changed files with 645 additions and 244 deletions
Download Patch File Download Diff File Expand all files Collapse all files
StgSharp.C/intrinsic.c
+4
View File
@@ -84,6 +84,10 @@ void load_intrinsic_sse(ssc_intrinsic *context)
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;
}
void load_intrinsic_avx(ssc_intrinsic *context)
StgSharp.C/mat2.c
+15
View File
@@ -40,3 +40,18 @@ INTERNAL void __cdecl addmatrix2_avx(__2_columnset *left, __2_columnset *right,
__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.C/mat3.c
+26
View File
@@ -101,4 +101,30 @@ INTERNAL void __cdecl addmatrix3_512(__3_columnset *left, __3_columnset *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.C/mat4.c
+19
View File
@@ -82,3 +82,22 @@ INTERNAL void SSCDECL addmatrix4_512(__4_columnset *left, __4_columnset *right,
{
ans->stream = _mm512_add_ps(ALIGN512(left->stream), ALIGN512(right->stream));
}
INTERNAL void SSCDECL 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 SSCDECL 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 SSCDECL submatrix4_512(__4_columnset *left, __4_columnset *right, __4_columnset *ans)
{
ans->stream = _mm512_sub_ps(ALIGN512(left->stream), ALIGN512(right->stream));
}
StgSharp.C/ssc_intrinsic.h
+13 -2
View File
@@ -30,7 +30,7 @@
typedef union uint64_u {
uint64_t u64;
uint32_t u32[2];
}uint64_u;
} uint64_u;
typedef union column_2 {
__m256 stream;
@@ -64,12 +64,17 @@ INTERNAL void normalize(__m128 *source, __m128 *target);
INTERNAL void SSCDECL addmatrix2_sse(__2_columnset *left, __2_columnset *right, __2_columnset *ans);
INTERNAL void SSCDECL addmatrix2_avx(__2_columnset *left, __2_columnset *right, __2_columnset *ans);
INTERNAL void SSCDECL submatrix2_sse(__2_columnset *left, __2_columnset *right, __2_columnset *ans);
INTERNAL void SSCDECL addmatrix2_avx(__2_columnset *left, __2_columnset *right, __2_columnset *ans);
INTERNAL void SSCDECL transpose23(__2_columnset const *source, __3_columnset *target);
INTERNAL void SSCDECL transpose24(__2_columnset const *source, __4_columnset *target);
INTERNAL void SSCDECL addmatrix3_sse(__3_columnset *left, __3_columnset *right, __3_columnset *ans);
INTERNAL void SSCDECL addmatrix3_avx(__3_columnset *left, __3_columnset *right, __3_columnset *ans);
INTERNAL void SSCDECL addmatrix3_512(__3_columnset *left, __3_columnset *right, __3_columnset *ans);
INTERNAL void SSCDECL submatrix3_sse(__3_columnset *left, __3_columnset *right, __3_columnset *ans);
INTERNAL void SSCDECL submatrix3_avx(__3_columnset *left, __3_columnset *right, __3_columnset *ans);
INTERNAL void SSCDECL submatrix3_512(__3_columnset *left, __3_columnset *right, __3_columnset *ans);
INTERNAL void SSCDECL transpose32(__3_columnset const *source, __2_columnset *target);
INTERNAL void SSCDECL transpose33(__3_columnset const *source, __3_columnset *target);
INTERNAL void SSCDECL transpose34(__3_columnset const *source, __4_columnset *target);
@@ -78,6 +83,9 @@ INTERNAL float SSCDECL det_mat3(__3_columnset *matrix, __3_columnset *tranpose);
INTERNAL void SSCDECL addmatrix4_sse(__4_columnset *left, __4_columnset *right, __4_columnset *ans);
INTERNAL void SSCDECL addmatrix4_avx(__4_columnset *left, __4_columnset *right, __4_columnset *ans);
INTERNAL void SSCDECL addmatrix4_512(__4_columnset *left, __4_columnset *right, __4_columnset *ans);
INTERNAL void SSCDECL submatrix4_sse(__4_columnset *left, __4_columnset *right, __4_columnset *ans);
INTERNAL void SSCDECL submatrix4_avx(__4_columnset *left, __4_columnset *right, __4_columnset *ans);
INTERNAL void SSCDECL submatrix4_512(__4_columnset *left, __4_columnset *right, __4_columnset *ans);
INTERNAL void SSCDECL transpose42(__4_columnset const *source, __2_columnset *target);
INTERNAL void SSCDECL transpose43(__4_columnset const *source, __3_columnset *target);
INTERNAL void SSCDECL transpose44(__4_columnset const *source, __4_columnset *target);
@@ -99,7 +107,7 @@ INTERNAL void SSCDECL dot_42_avx(__4_columnset *transpose, __m128 *vector, __m12
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);
//INTERNAL void SSCDECL dot_43_512(__4_columnset *transpose, __m128 *vector, __m128 *ans);
#pragma endregion
@@ -124,6 +132,9 @@ typedef struct ssc_intrinsic {
DOTPROC dot_43;
int (*index_pair)(short const *str, uint32_t target, int length);
VECTORNORMALIZEPROC normalize_3;
MATARITHMETIC sub_m2;
MATARITHMETIC sub_m3;
MATARITHMETIC sub_m4;
MATRIXTRANSPOSEPROC transpose_23;
MATRIXTRANSPOSEPROC transpose_24;
MATRIXTRANSPOSEPROC transpose_32;