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

matrix thread pool use ParaWaveFront and heavy sync

Browse Source
This commit is contained in:
Youmu Liu
2026-04-26 17:38:10 +08:00
parent 251762791e
commit ecb1f04770
7 changed files with 152 additions and 131 deletions
Download Patch File Download Diff File Expand all files Collapse all files
StgSharp.Native/sn-head/sn_intrinsic.std.h
+27 -22
View File
@@ -17,18 +17,18 @@
#define GET_MK_PROC_STD(T_type, T_arch, T_feature, T_op) \
((UNI_MK_PROC)MK_PROC_NAME_STD(T_type, T_arch, T_feature, T_op))
#define SN_MK_PROC_DECL(T_type, P_name) INTERNAL void SN_DECL P_name(MAT_TASK(T_type) const *task)
#define SN_MK_PROC_DECL_STD(T_type, T_arch, T_feature, T_op) \
INTERNAL NOINLINE SN_STD_TARGET_ATTR(T_arch, T_feature) void SN_DECL \
MK_PROC_NAME_STD(T_type, T_arch, T_feature, T_op)(MAT_TASK(T_type) const *task)
#define SN_MK_PROC_DECL_STD(T_type, T_arch, T_feature, T_op) \
INTERNAL NOINLINE SN_STD_TARGET_ATTR(T_arch, T_feature) void SN_DECL MK_PROC_NAME_STD( \
T_type, T_arch, T_feature, T_op)(MAT_TASK(T_type) const *task)
#define SK_PROC_NAME_STD(T_type, T_arch, T_feature, T_op) \
T_type##_single_kernel_proc_##T_op##_##T_arch##_##T_feature
#define GET_SK_PROC_STD(T_type, T_arch, T_feature, T_op) \
((UNI_SK_PROC)SK_PROC_NAME_STD(T_type, T_arch, T_feature, T_op))
#define SN_SK_PROC_DECL_STD(T_type, T_arch, T_feature, T_op) \
INTERNAL NOINLINE SN_STD_TARGET_ATTR(T_arch, T_feature) void SN_DECL \
SK_PROC_NAME_STD(T_type, T_arch, T_feature, T_op)( \
MAT_KERNEL(T_type) const *left, MAT_KERNEL(T_type) const *right, \
MAT_KERNEL(T_type) *restrict ans, uint64_t scalar)
#define SN_SK_PROC_DECL_STD(T_type, T_arch, T_feature, T_op) \
INTERNAL NOINLINE SN_STD_TARGET_ATTR(T_arch, T_feature) void SN_DECL SK_PROC_NAME_STD( \
T_type, T_arch, T_feature, \
T_op)(MAT_KERNEL(T_type) const *left, MAT_KERNEL(T_type) const *right, \
MAT_KERNEL(T_type) *restrict ans, uint64_t scalar)
#define _SN_STD__PROBE_0 0
#define _SN_STD__PROBE_1 1
@@ -43,27 +43,27 @@
#define _SN_MATPROC_RIGHT_ANS_SCALAR 1
#define _SN_MATPROC_ANS_SCALAR 1
#define _SN_MK_EXPAND_RIGHT_ANS \
#define _SN_MK_EXPAND_RIGHT_ANS \
SN_MAT_KERNEL_LOCAL const *right = (SN_MAT_KERNEL_LOCAL const *)task->mat_1; \
SN_MAT_KERNEL_LOCAL *restrict ans = (SN_MAT_KERNEL_LOCAL *)task->mat_2;
#define _SN_MK_EXPAND_LEFT_RIGHT_ANS \
#define _SN_MK_EXPAND_LEFT_RIGHT_ANS \
SN_MAT_KERNEL_LOCAL const *left = (SN_MAT_KERNEL_LOCAL const *)task->mat_0; \
SN_MAT_KERNEL_LOCAL const *right = (SN_MAT_KERNEL_LOCAL const *)task->mat_1; \
SN_MAT_KERNEL_LOCAL *restrict ans = (SN_MAT_KERNEL_LOCAL *)task->mat_2;
#define _SN_MK_EXPAND_RIGHT_ANS_SCALAR \
SN_MAT_KERNEL_LOCAL const *right = (SN_MAT_KERNEL_LOCAL const *)task->mat_1; \
SN_MAT_KERNEL_LOCAL *restrict ans = (SN_MAT_KERNEL_LOCAL *)task->mat_2; \
#define _SN_MK_EXPAND_RIGHT_ANS_SCALAR \
SN_MAT_KERNEL_LOCAL const *right = (SN_MAT_KERNEL_LOCAL const *)task->mat_1; \
SN_MAT_KERNEL_LOCAL *restrict ans = (SN_MAT_KERNEL_LOCAL *)task->mat_2; \
scalar_pack *restrict scalar = (scalar_pack *restrict)task->scalar;
#define _SN_MK_EXPAND_LEFT_RIGHT_ANS_SCALAR \
SN_MAT_KERNEL_LOCAL const *left = (SN_MAT_KERNEL_LOCAL const *)task->mat_0; \
SN_MAT_KERNEL_LOCAL const *right = (SN_MAT_KERNEL_LOCAL const *)task->mat_1; \
SN_MAT_KERNEL_LOCAL *restrict ans = (SN_MAT_KERNEL_LOCAL *)task->mat_2; \
#define _SN_MK_EXPAND_LEFT_RIGHT_ANS_SCALAR \
SN_MAT_KERNEL_LOCAL const *left = (SN_MAT_KERNEL_LOCAL const *)task->mat_0; \
SN_MAT_KERNEL_LOCAL const *right = (SN_MAT_KERNEL_LOCAL const *)task->mat_1; \
SN_MAT_KERNEL_LOCAL *restrict ans = (SN_MAT_KERNEL_LOCAL *)task->mat_2; \
scalar_pack *restrict scalar = (scalar_pack *restrict)task->scalar;
#define _SN_MK_EXPAND_ANS_SCALAR \
#define _SN_MK_EXPAND_ANS_SCALAR \
SN_MAT_KERNEL_LOCAL *restrict ans = (SN_MAT_KERNEL_LOCAL *)task->mat_2; \
scalar_pack *restrict scalar = (scalar_pack *restrict)&(task->padding);
@@ -74,9 +74,14 @@
#pragma region mem layout
#define _SN_MATPROC_LAYOUT_BUFFER 1
#define _SN_MATPROC_LAYOUT_TILE 1
#define _SN_MATPROC_LAYOUT_KERNEL 1
#define _MK_MKPROC_EXPAND_INDEX_BUFFER \
size_t offset = UNSAFE_AS(size_t, task->padding.data[0]); \
size_t count = UNSAFE_AS(size_t, task->padding.data[1]);
#define _MK_MKPROC_EXPAND_INDEX_BUFFER \
size_t offset = UNSAFE_AS(size_t, task->padding.data[0]); \
size_t count = UNSAFE_AS(size_t, task->padding.data[1]);
@@ -99,13 +104,13 @@
#pragma endregion
#define __mk_param_std(mem, param, T) \
typedef MAT_KERNEL(T) SN_MAT_KERNEL_LOCAL; \
#define __mk_param_std(mem, param, T) \
typedef MAT_KERNEL(T) SN_MAT_KERNEL_LOCAL; \
matrix_task const *_ = (_SN_PARAM_STYLE(param), (_SN_MEM_LAYOUT(mem)), task); \
_SN_MK_EXPAND_PARAM(param); \
_SN_MK_EXPAND_PARAM(param); \
_SN_MK_EXPAND_MEM(mem)
#define __sk_param_std(T) \
#define __sk_param_std(T) \
typedef MAT_KERNEL(T) SN_MAT_KERNEL_LOCAL; \
(void)scalar
#endif // !SN_INTRINSIC_PARAM_EXPAND