Mirrored_Repos/LookingGlass
1
0
Fork 0
mirror of https://github.com/gnif/LookingGlass.git synced 2025-11-04 06:31:54 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

[common] adjust framebuffer read/write strategy for better cache usage

Browse Source
This commit is contained in:
Geoffrey McRae
2020-08-03 12:33:08 +10:00
parent 7af053497e
commit 85b8c12abf
2 changed files with 22 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
VERSION
View File

@@ -1 +1 @@
B2-rc2-17-gaa32c5ffad+1
B2-rc2-19-g7af053497e+1

31
common/src/framebuffer.c
View File

@@ -118,10 +118,15 @@ bool framebuffer_write(FrameBuffer * frame, const void * src, size_t size)
/* copy in chunks */
while(size > 63)
{
_mm_stream_si128(d++, _mm_stream_load_si128(s++));
_mm_stream_si128(d++, _mm_stream_load_si128(s++));
_mm_stream_si128(d++, _mm_stream_load_si128(s++));
_mm_stream_si128(d++, _mm_stream_load_si128(s++));
const __m128i v1 = _mm_stream_load_si128(s++);
const __m128i v2 = _mm_stream_load_si128(s++);
const __m128i v3 = _mm_stream_load_si128(s++);
const __m128i v4 = _mm_stream_load_si128(s++);
_mm_stream_si128(d++, v1);
_mm_stream_si128(d++, v2);
_mm_stream_si128(d++, v3);
_mm_stream_si128(d++, v4);
size -= 64;
wp += 64;
@@ -131,24 +136,30 @@ bool framebuffer_write(FrameBuffer * frame, const void * src, size_t size)
if (size > 47)
{
_mm_stream_si128(d++, _mm_stream_load_si128(s++));
_mm_stream_si128(d++, _mm_stream_load_si128(s++));
_mm_stream_si128(d++, _mm_stream_load_si128(s++));
const __m128i v1 = _mm_stream_load_si128(s++);
const __m128i v2 = _mm_stream_load_si128(s++);
const __m128i v3 = _mm_stream_load_si128(s++);
_mm_stream_si128(d++, v1);
_mm_stream_si128(d++, v2);
_mm_stream_si128(d++, v3);
size -= 48;
wp += 48;
}
if (size > 31)
{
_mm_stream_si128(d++, _mm_stream_load_si128(s++));
_mm_stream_si128(d++, _mm_stream_load_si128(s++));
const __m128i v1 = _mm_stream_load_si128(s++);
const __m128i v2 = _mm_stream_load_si128(s++);
_mm_stream_si128(d++, v1);
_mm_stream_si128(d++, v2);
size -= 32;
wp += 32;
}
if (size > 15)
{
_mm_stream_si128(d++, _mm_stream_load_si128(s++));
const __m128i v1 = _mm_stream_load_si128(s++);
_mm_stream_si128(d++, v1);
size -= 16;
wp += 16;
}