aleajactaest/khanat-code-old
1
0
Fork 0
mirror of https://port.numenaute.org/aleajactaest/khanat-code-old.git synced 2024-11-10 09:19:01 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

SSE2: Replace prefetch

Browse source
This commit is contained in:
kaetemi 2014-06-14 01:09:05 +02:00
parent 92d00936aa
commit 2a87923ced
1 changed files with 36 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

38
code/nel/src/3d/mesh_mrm_skin_template.cpp
View file

@ -39,7 +39,23 @@ static void applyArraySkinNormalT(uint numMatrixes, uint32 *infPtr, CMesh::CSkin
{
/* Prefetch all vertex/normal before, it is to be faster.
*/
#if defined(NL_OS_WINDOWS) && !defined(NL_NO_ASM)
#ifdef NL_HAS_SSE2
{
uint nInfTmp= nInf;
uint32 *infTmpPtr= infPtr;
for(;nInfTmp>0;nInfTmp--, infTmpPtr++)
{
uint index= *infTmpPtr;
CMesh::CSkinWeight *srcSkin= srcSkinPtr + index;
CVector *srcVertex= srcVertexPtr + index;
CVector *srcNormal= srcNormalPtr + index;
_mm_prefetch((const char *)(void *)srcSkin, _MM_HINT_T1);
_mm_prefetch((const char *)(void *)srcVertex, _MM_HINT_T1);
_mm_prefetch((const char *)(void *)srcNormal, _MM_HINT_T1);
}
}
#elif defined(NL_OS_WINDOWS) && !defined(NL_NO_ASM)
{
uint nInfTmp= nInf;
uint32 *infTmpPtr= infPtr;
@ -176,7 +192,25 @@ static void applyArraySkinTangentSpaceT(uint numMatrixes, uint32 *infPtr, CMesh:
{
/* Prefetch all vertex/normal/tgSpace before, it is faster.
*/
#if defined(NL_OS_WINDOWS) && !defined(NL_NO_ASM)
#ifdef NL_HAS_SSE2
{
uint nInfTmp= nInf;
uint32 *infTmpPtr= infPtr;
for(;nInfTmp>0;nInfTmp--, infTmpPtr++)
{
uint index= *infTmpPtr;
CMesh::CSkinWeight *srcSkin= srcSkinPtr + index;
CVector *srcVertex= srcVertexPtr + index;
CVector *srcNormal= srcNormalPtr + index;
CVector *srcTgSpace= tgSpacePtr + index;
_mm_prefetch((const char *)(void *)srcSkin, _MM_HINT_T1);
_mm_prefetch((const char *)(void *)srcVertex, _MM_HINT_T1);
_mm_prefetch((const char *)(void *)srcNormal, _MM_HINT_T1);
_mm_prefetch((const char *)(void *)srcTgSpace, _MM_HINT_T1);
}
}
#elif defined(NL_OS_WINDOWS) && !defined(NL_NO_ASM)
{
uint nInfTmp= nInf;
uint32 *infTmpPtr= infPtr;