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

Daily update on 20251230-224423 by Nitload Youmu. Additional Message : Adding compute support for mat<double>. Import all opengl core profile fucntions

Browse Source 
Signed-off-by: Nitload <YoumuLiu_NSI@whv0t.onmicrosoft.com>
This commit is contained in:
Nitload
2025-12-30 22:44:23 +08:00
parent cf370ca98a 6484307b8a
commit 5decbd390d
83 changed files with 75819 additions and 6024 deletions
Download Patch File Download Diff File Expand all files Collapse all files
StgSharp.Native/sn-head/sn_intrinsic.h
+42 -20
View File
@@ -186,15 +186,10 @@ typedef void(SN_DECL *KER_PROC_LAS)(mat_kernel *matrix, mat_kernel *ans, scalar_
#pragma region special matrix kernel operation
typedef void(SN_DECL *POVITPROC)(mat_kernel const *source, mat_kernel *ans);
#define MAT_POVIT_PROC(T_type, T_arch) povit_##T_type##_##T_arch
#define DECLARE_MAT_POVIT_PROC(T_type, T_arch) \
INTERNAL void SN_DECL MAT_POVIT_PROC(T_type, T_arch)(MAT_KERNEL(T_type) const *source, \
size_t col_begin, size_t col_length, \
size_t *restrict ans)
DECLARE_MAT_POVIT_PROC(float, sse);
DECLARE_MAT_POVIT_PROC(float, avx);
DECLARE_MAT_POVIT_PROC(float, 512);
typedef void(SN_DECL *PIVOTPROC)(mat_kernel const *source, size_t col_begin, size_t col_length,
size_t *restrict ans);
INTERNAL void SN_DECL pivot_float(MAT_KERNEL(float) const *source, size_t col_begin,
size_t col_length, size_t *restrict ans);
#pragma endregion
@@ -210,10 +205,18 @@ DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, sse, add);
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, avx, add);
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, 512, add);
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(double, sse, add);
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(double, avx, add);
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(double, 512, add);
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, sse, sub);
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, avx, sub);
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, 512, sub);
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(double, sse, sub);
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(double, avx, sub);
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(double, 512, sub);
DECLARE_KER_PROC_RIGHT_ANS(float, sse, transpose);
DECLARE_KER_PROC_RIGHT_ANS(float, avx, transpose);
DECLARE_KER_PROC_RIGHT_ANS(float, 512, transpose);
@@ -222,10 +225,18 @@ DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(float, sse, scalar_mul);
DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(float, avx, scalar_mul);
DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(float, 512, scalar_mul);
DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(double, sse, scalar_mul);
DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(double, avx, scalar_mul);
DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(double, 512, scalar_mul);
DECLARE_BUF_PROC_ANS_SCALAR(float, sse, fill);
DECLARE_BUF_PROC_ANS_SCALAR(float, avx, fill);
DECLARE_BUF_PROC_ANS_SCALAR(float, 512, fill);
DECLARE_BUF_PROC_ANS_SCALAR(double, sse, fill);
DECLARE_BUF_PROC_ANS_SCALAR(double, avx, fill);
DECLARE_BUF_PROC_ANS_SCALAR(double, 512, fill);
#pragma endregion
#pragma region fma
@@ -233,6 +244,10 @@ DECLARE_PNL_PROC_ANS(float, sse, clear);
DECLARE_PNL_PROC_ANS(float, avx, clear);
DECLARE_PNL_PROC_ANS(float, 512, clear);
DECLARE_PNL_PROC_ANS(double, sse, clear);
DECLARE_PNL_PROC_ANS(double, avx, clear);
DECLARE_PNL_PROC_ANS(double, 512, clear);
DECLARE_PNL_PROC_LEFT_RIGHT_ANS(float, sse, fma);
DECLARE_PNL_PROC_LEFT_RIGHT_ANS(float, avx, fma);
DECLARE_PNL_PROC_LEFT_RIGHT_ANS(float, avx_fma, fma);
@@ -269,24 +284,31 @@ INTERNAL int SN_DECL index_pair_sse(short const *str, uint32_t target, int lengt
#pragma endregion
typedef struct mat_intrinsic {
BUF_PROC_LRA buffer_add;
BUF_PROC_AS buffer_fill;
BUF_PROC_RAS buffer_scalar_mul;
BUF_PROC_LRA buffer_sub;
KER_PROC_RA buffer_transpose;
PANEL_PROC build_panel;
PANEL_PROC clear_panel;
PNL_PROC_LRA panel_fma;
PIVOTPROC pivot;
PANEL_PROC store_panel;
} mat_intrinsic;
// Mirrors C# IntrinsicContext field order so managed/unmanaged layouts stay in sync.
typedef struct sn_intrinsic {
HASH city_hash_simplify;
BUF_PROC_LRA f32_buffer_add;
BUF_PROC_AS f32_buffer_fill;
BUF_PROC_RAS f32_buffer_scalar_mul;
BUF_PROC_LRA f32_buffer_sub;
KER_PROC_RA f32_buffer_transpose;
PANEL_PROC f32_build_panel;
PANEL_PROC f32_clear_panel;
VECTORNORMALIZEPROC f32_normalize_3;
PNL_PROC_LRA f32_panel_fma;
PANEL_PROC f32_store_panel;
FACTORIALROC factorial_simd;
INDEX_PAIR index_pair;
BUF_PROC_RAS variant;
VECTORNORMALIZEPROC vec_f32_normalize_3;
mat_intrinsic mat[4];
} sn_intrinsic;
typedef enum elment_type { F32 = 0, F64 = 1, I32 = 2, I64 = 3 } element_type;
typedef enum most_advanced_instruction {
#ifdef _MSC_VER
SSE = __IA_SUPPORT_SSE42,