piwalker/dolphin
1
0
Fork 0
mirror of https://github.com/dolphin-emu/dolphin.git synced 2025-07-24 06:39:46 -06:00
Code Issues Packages Projects Releases Wiki Activity

Use alignment for ReadDataU32XN. Revert james temp fix.

Browse Source 
Should provide some form of a speedup.

git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@6812 8ced0084-cf51-0410-be5f-012b33b47a6e
This commit is contained in:
xsacha
2011-01-11 04:54:35 +00:00
parent d78be40be9
commit 3f5d1afc6a
4 changed files with 24 additions and 29 deletions
Download Patch File Download Diff File Expand all files Collapse all files

42
Source/Core/VideoCommon/Src/DataReader.h
View File

@ -83,35 +83,29 @@ const __m128i mask4 = _mm_set_epi8(12,13,14,15,8,9,10,11,4,5,6,7,0,1,2,3);
template<unsigned int N>
void DataReadU32xN_SSSE3(u32 *bufx16)
{
__m128i* store = (__m128i *)bufx16;
__m128i* load = (__m128i *)g_pVideoData;
memcpy(bufx16, g_pVideoData, sizeof(u32) * N);
__m128i* buf = (__m128i *)bufx16;
switch(N)
{
case 13: _mm_storeu_si128(store++, _mm_shuffle_epi8(_mm_loadu_si128(load++), mask4));
case 9: _mm_storeu_si128(store++, _mm_shuffle_epi8(_mm_loadu_si128(load++), mask4));
case 5: _mm_storeu_si128(store++, _mm_shuffle_epi8(_mm_loadu_si128(load++), mask4));
case 1: // 1 U32 left:
((u32 *)store)[0] = Common::swap32(((u32 *)load)[0]);
case 13: _mm_store_si128(buf, _mm_shuffle_epi8(_mm_load_si128(buf), mask4)); buf++;
case 9: _mm_store_si128(buf, _mm_shuffle_epi8(_mm_load_si128(buf), mask4)); buf++;
case 5: _mm_store_si128(buf, _mm_shuffle_epi8(_mm_load_si128(buf), mask4)); buf++;
case 1: _mm_store_si128(buf, _mm_shuffle_epi8(_mm_load_si128(buf), mask1));
break;
case 14: _mm_storeu_si128(store++, _mm_shuffle_epi8(_mm_loadu_si128(load++), mask4));
case 10: _mm_storeu_si128(store++, _mm_shuffle_epi8(_mm_loadu_si128(load++), mask4));
case 6: _mm_storeu_si128(store++, _mm_shuffle_epi8(_mm_loadu_si128(load++), mask4));
case 2: // 2 U32s left:
((u32 *)store)[0] = Common::swap32(((u32 *)load)[0]);
((u32 *)store)[1] = Common::swap32(((u32 *)load)[1]);
case 14: _mm_store_si128(buf, _mm_shuffle_epi8(_mm_load_si128(buf), mask4)); buf++;
case 10: _mm_store_si128(buf, _mm_shuffle_epi8(_mm_load_si128(buf), mask4)); buf++;
case 6: _mm_store_si128(buf, _mm_shuffle_epi8(_mm_load_si128(buf), mask4)); buf++;
case 2: _mm_store_si128(buf, _mm_shuffle_epi8(_mm_load_si128(buf), mask2));
break;
case 15: _mm_storeu_si128(store++, _mm_shuffle_epi8(_mm_loadu_si128(load++), mask4));
case 11: _mm_storeu_si128(store++, _mm_shuffle_epi8(_mm_loadu_si128(load++), mask4));
case 7: _mm_storeu_si128(store++, _mm_shuffle_epi8(_mm_loadu_si128(load++), mask4));
case 3: // 3 U32s left:
((u32 *)store)[0] = Common::swap32(((u32 *)load)[0]);
((u32 *)store)[1] = Common::swap32(((u32 *)load)[1]);
((u32 *)store)[2] = Common::swap32(((u32 *)load)[2]);
case 15: _mm_store_si128(buf, _mm_shuffle_epi8(_mm_load_si128(buf), mask4)); buf++;
case 11: _mm_store_si128(buf, _mm_shuffle_epi8(_mm_load_si128(buf), mask4)); buf++;
case 7: _mm_store_si128(buf, _mm_shuffle_epi8(_mm_load_si128(buf), mask4)); buf++;
case 3: _mm_store_si128(buf, _mm_shuffle_epi8(_mm_load_si128(buf), mask3));
break;
case 16: _mm_storeu_si128(store++, _mm_shuffle_epi8(_mm_loadu_si128(load++), mask4));
case 12: _mm_storeu_si128(store++, _mm_shuffle_epi8(_mm_loadu_si128(load++), mask4));
case 8: _mm_storeu_si128(store++, _mm_shuffle_epi8(_mm_loadu_si128(load++), mask4));
case 4: _mm_storeu_si128(store, _mm_shuffle_epi8(_mm_loadu_si128(load), mask4));
case 16: _mm_store_si128(buf, _mm_shuffle_epi8(_mm_load_si128(buf), mask4)); buf++;
case 12: _mm_store_si128(buf, _mm_shuffle_epi8(_mm_load_si128(buf), mask4)); buf++;
case 8: _mm_store_si128(buf, _mm_shuffle_epi8(_mm_load_si128(buf), mask4)); buf++;
case 4: _mm_store_si128(buf, _mm_shuffle_epi8(_mm_load_si128(buf), mask4));
break;
}
g_pVideoData += (sizeof(u32) * N);