Intravision/reactos
1
0
Fork 0
mirror of https://github.com/reactos/reactos.git synced 2024-12-27 17:44:45 +00:00
Code Issues Projects Releases Packages Wiki Activity

speed up 24bpp small fill and fill

Browse source 
svn path=/trunk/; revision=15859
This commit is contained in:
Magnus Olsen 2005-06-11 13:36:00 +00:00
parent 7c9de0d8d6
commit c1643db40b
1 changed files with 87 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

88
reactos/subsys/win32k/dib/dib24bpp.c
View file

@ -392,11 +392,97 @@ DIB_24BPP_ColorFill(SURFOBJ* DestSurface, RECTL* DestRect, ULONG color)
{
ULONG DestY;
#ifdef _M_IX86
PBYTE xaddr = DestSurface->pvScan0 + DestRect->top * DestSurface->lDelta + (DestRect->left << 1) + DestRect->left;
PBYTE addr;
ULONG Count;
ULONG xCount=DestRect->right - DestRect->left;
for (DestY = DestRect->top; DestY< DestRect->bottom; DestY++)
{
Count = xCount;
addr = xaddr;
xaddr = (PBYTE)((ULONG_PTR)addr + DestSurface->lDelta);
if (Count < 8)
{
/* For small fills, don't bother doing anything fancy */
while (Count--)
{
*(PUSHORT)(addr) = color;
addr += 2;
*(addr) = color >> 16;
addr += 1;
}
}
else
{
/* Align to 4-byte address */
while (0 != ((ULONG_PTR) addr & 0x3))
{
*(PUSHORT)(addr) = color;
addr += 2;
*(addr) = color >> 16;
addr += 1;
Count--;
}
/* If the color we need to fill with is 0ABC, then the final mem pattern
* (note little-endianness) would be:
*
* |C.B.A|C.B.A|C.B.A|C.B.A| <- pixel borders
* |C.B.A.C|B.A.C.B|A.C.B.A| <- ULONG borders
*
* So, taking endianness into account again, we need to fill with these
* ULONGs: CABC BCAB ABCA */
/* This is about 30% faster than the generic C code below */
__asm__ __volatile__ (
" movl %1, %%ecx\n"
" andl $0xffffff, %%ecx\n" /* 0ABC */
" movl %%ecx, %%ebx\n" /* Construct BCAB in ebx */
" shrl $8, %%ebx\n"
" movl %%ecx, %%eax\n"
" shll $16, %%eax\n"
" orl %%eax, %%ebx\n"
" movl %%ecx, %%edx\n" /* Construct ABCA in edx */
" shll $8, %%edx\n"
" movl %%ecx, %%eax\n"
" shrl $16, %%eax\n"
" orl %%eax, %%edx\n"
" movl %%ecx, %%eax\n" /* Construct CABC in eax */
" shll $24, %%eax\n"
" orl %%ecx, %%eax\n"
" movl %2, %%ecx\n" /* Load count */
" shr $2, %%ecx\n"
" movl %3, %%edi\n" /* Load dest */
".FL1:\n"
" movl %%eax, (%%edi)\n" /* Store 4 pixels, 12 bytes */
" movl %%ebx, 4(%%edi)\n"
" movl %%edx, 8(%%edi)\n"
" addl $12, %%edi\n"
" dec %%ecx\n"
" jnz .FL1\n"
" movl %%edi, %0\n"
: "=m"(addr)
: "m"(color), "m"(Count), "m"(addr)
: "%eax", "%ebx", "%ecx", "%edx", "%edi");
Count = Count & 0x03;
while (0 != Count--)
{
*(PUSHORT)(addr) = color;
addr += 2;
*(addr) = color >> 16;
addr += 1;
}
}
}
#else
for (DestY = DestRect->top; DestY< DestRect->bottom; DestY++)
{
DIB_24BPP_HLine(DestSurface, DestRect->left, DestRect->right, DestY, color);
}
#endif
return TRUE;
}