Merge branch 'giga' of ssh://ssh.github.com:443/Nitload-NSI/CCD_Github into local

2025-12-17 22:00:44 +08:00
parent b13adc1a1b 688a420574
commit cffe44772e
8 changed files with 181 additions and 14 deletions
@@ -7,6 +7,7 @@
 #include "sn_matkernel.h"
 #include <immintrin.h>
 #include <smmintrin.h>
+#include <stdint.h>

 #ifdef _MSC_VER
 #include <intrin.h>
@@ -203,6 +204,20 @@ typedef void(SN_DECL *KER_PROC_LAS)(mat_kernel *matrix, mat_kernel *ans, scalar_

 #pragma endregion

+#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);
+
+#pragma endregion
+
 typedef void(SN_DECL *VECTORNORMALIZEPROC)(__m128 *source, __m128 *target);
 typedef void(SN_DECL *DOTPROC)(void *transpose, __m128 *vector, __m128 *ans);
 typedef uint64_t(SN_DECL *FACTORIALROC)(int n);