Mirrored_Repos/LookingGlass
1
0
Fork 0
mirror of https://github.com/gnif/LookingGlass.git synced 2024-11-25 14:57:20 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

[common] unroll the framebuffer write loop and increase the chunk size

Browse Source
This commit is contained in:
Geoffrey McRae 2020-08-03 12:24:11 +10:00
parent 9e3a42cb62
commit 7af053497e
2 changed files with 35 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
VERSION
View File

@ -1 +1 @@
B2-rc2-16-g62d1bd1ea2+1 B2-rc2-17-gaa32c5ffad+1

40
common/src/framebuffer.c
View File

@ -25,7 +25,7 @@ Place, Suite 330, Boston, MA 02111-1307 USA
#include <emmintrin.h> #include <emmintrin.h>
#include <smmintrin.h> #include <smmintrin.h>
#define FB_CHUNK_SIZE 1024 #define FB_CHUNK_SIZE 1048576
struct stFrameBuffer struct stFrameBuffer
{ {
@ -112,22 +112,50 @@ void framebuffer_prepare(FrameBuffer * frame)
bool framebuffer_write(FrameBuffer * frame, const void * src, size_t size) bool framebuffer_write(FrameBuffer * frame, const void * src, size_t size)
{ {
__m128i * s = (__m128i *)src; __m128i * s = (__m128i *)src;
__m128i * d = (__m128i *)frame->data;
size_t wp = 0; size_t wp = 0;
/* copy in chunks */ /* copy in chunks */
while(size > 15) while(size > 63)
{ {
_mm_stream_si128((__m128i *)(frame->data + wp), _mm_stream_load_si128(s++)); _mm_stream_si128(d++, _mm_stream_load_si128(s++));
size -= 16; _mm_stream_si128(d++, _mm_stream_load_si128(s++));
wp += 16; _mm_stream_si128(d++, _mm_stream_load_si128(s++));
_mm_stream_si128(d++, _mm_stream_load_si128(s++));
size -= 64;
wp += 64;
if (wp % FB_CHUNK_SIZE == 0) if (wp % FB_CHUNK_SIZE == 0)
atomic_store_explicit(&frame->wp, wp, memory_order_release); atomic_store_explicit(&frame->wp, wp, memory_order_release);
} }
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++));
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++));
size -= 32;
wp += 32;
}
if (size > 15)
{
_mm_stream_si128(d++, _mm_stream_load_si128(s++));
size -= 16;
wp += 16;
}
if(size) if(size)
{ {
memcpy(frame->data + frame->wp, s, size); memcpy(frame->data + wp, s, size);
wp += size; wp += size;
} }