daily update on 20240226 by nitload youmu
This commit is contained in:
@@ -1,10 +1,8 @@
|
||||
#include "./include/StgSharpC.h"
|
||||
#include "StgSharpC.h"
|
||||
#include "ssgc_internal.h"
|
||||
#include <stdio.h>
|
||||
#include <pmmintrin.h>
|
||||
|
||||
|
||||
|
||||
|
||||
SSCAPI __inline void __cdecl transpose4to2(matrix4map* source, matrix4map* target)
|
||||
{
|
||||
__m128 t0 = _mm_shuffle_ps(ALIGN(source->colum[0]), ALIGN(source->colum[1]), _MM_SHUFFLE(1, 0, 1, 0));
|
||||
@@ -25,7 +23,6 @@ SSCAPI __inline void __cdecl transpose4to4(matrix4map* source, matrix4map* targe
|
||||
target->colum[1] = _mm_shuffle_ps(t0, t2, _MM_SHUFFLE(3, 1, 3, 1));
|
||||
target->colum[2] = _mm_shuffle_ps(t1, t3, _MM_SHUFFLE(2, 0, 2, 0));
|
||||
target->colum[3] = _mm_shuffle_ps(t1, t3, _MM_SHUFFLE(3, 1, 3, 1));
|
||||
|
||||
}
|
||||
|
||||
SSCAPI __inline void __cdecl transpose4to3(matrix4map* source, matrix3map* target)
|
||||
@@ -38,11 +35,8 @@ SSCAPI __inline void __cdecl transpose4to3(matrix4map* source, matrix3map* targe
|
||||
target->colum[0] = _mm_shuffle_ps(t0, t2, _MM_SHUFFLE(2, 0, 2, 0));
|
||||
target->colum[1] = _mm_shuffle_ps(t0, t2, _MM_SHUFFLE(3, 1, 3, 1));
|
||||
target->colum[2] = _mm_shuffle_ps(t1, t3, _MM_SHUFFLE(2, 0, 2, 0));
|
||||
|
||||
}
|
||||
|
||||
|
||||
|
||||
SSCAPI __inline float __cdecl det_mat4(matrix4map* matPtr, matrix4map* transpose)
|
||||
{
|
||||
__m128 mc0 = ALIGN(matPtr->colum[0]);
|
||||
@@ -60,7 +54,6 @@ SSCAPI __inline float __cdecl det_mat4(matrix4map* matPtr, matrix4map* transpose
|
||||
retvec = _mm_hsub_ps(retvec, retvec);
|
||||
retvec = _mm_hadd_ps(retvec, retvec);
|
||||
return -retvec.m128_f32[0];
|
||||
|
||||
}
|
||||
|
||||
SSCAPI __inline void __cdecl deinit_mat4(matrix4map* matPtr)
|
||||
|
||||
Reference in New Issue
Block a user