scratchfoundation/bgfx
1
0
Fork 0
mirror of https://github.com/scratchfoundation/bgfx.git synced 2024-11-28 18:45:54 -05:00
Code Issues Projects Releases Packages Wiki Activity Actions

Fixed bad logic for ref/simd path.

Browse source
This commit is contained in:
bkaradzic 2013-08-15 20:14:27 -07:00
parent 7bdae66027
commit 752fa73cec
1 changed files with 25 additions and 30 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

55
src/image.cpp
View file

@ -3,60 +3,55 @@
* License: http://www.opensource.org/licenses/BSD-2-Clause * License: http://www.opensource.org/licenses/BSD-2-Clause
*/ */
#include "bgfx_p.h"
#include <bx/float4_t.h> #include <bx/float4_t.h>
namespace bgfx namespace bgfx
{ {
static void imageSwizzleBGRA8Ref(uint8_t* _rgbaData, uint32_t _width, uint32_t _height) static void imageSwizzleBGRA8Ref(uint8_t* _rgbaData, uint32_t _width, uint32_t _height)
{ {
uint32_t dstpitch = _width*4; const uint32_t dstpitch = _width*4;
for (uint32_t yy = 0; yy < _height; ++yy) uint8_t* ptr = _rgbaData;
{
uint8_t* dst = &_rgbaData[yy*dstpitch];
for (uint32_t xx = 0; xx < _width; ++xx) for (uint32_t xx = 0, num = _width*_height; xx < num; ++xx)
{ {
uint8_t tmp = dst[0]; uint8_t tmp = ptr[0];
dst[0] = dst[2]; ptr[0] = ptr[2];
dst[2] = tmp; ptr[2] = tmp;
dst += 4; ptr += 4;
}
} }
} }
void imageSwizzleBGRA8(uint8_t* _rgbaData, uint32_t _width, uint32_t _height) void imageSwizzleBGRA8(uint8_t* _rgbaData, uint32_t _width, uint32_t _height)
{ {
if (0 != (_width&0xf) // Test can we do four 4-byte pixels at the time.
|| _width < 16) if (0 != (_width&0x3)
|| _width < 4)
{ {
BX_WARN(_width < 4, "Image swizzle is taking slow path. Image width must be multiple of 4 (width %d).", _width);
imageSwizzleBGRA8Ref(_rgbaData, _width, _height); imageSwizzleBGRA8Ref(_rgbaData, _width, _height);
return; return;
} }
uint32_t dstpitch = _width*4; const uint32_t dstpitch = _width*4;
uint32_t num = dstpitch/16;
using namespace bx; using namespace bx;
const float4_t mf0f0 = float4_isplat(0xff00ff00); const float4_t mf0f0 = float4_isplat(0xff00ff00);
const float4_t m0f0f = float4_isplat(0x00ff00ff); const float4_t m0f0f = float4_isplat(0x00ff00ff);
uint8_t* ptr = _rgbaData;
for (uint32_t yy = 0; yy < _height; ++yy) for (uint32_t xx = 0, num = dstpitch/16*_height; xx < num; ++xx)
{ {
uint8_t* ptr = &_rgbaData[yy*dstpitch]; const float4_t tabgr = float4_ld(ptr);
const float4_t t00ab = float4_srl(tabgr, 16);
for (uint32_t xx = 0; xx < num; ++xx) const float4_t tgr00 = float4_sll(tabgr, 16);
{ const float4_t tgrab = float4_or(t00ab, tgr00);
const float4_t tabgr = float4_ld(ptr); const float4_t ta0g0 = float4_and(tabgr, mf0f0);
const float4_t t00ab = float4_srl(tabgr, 16); const float4_t t0r0b = float4_and(tgrab, m0f0f);
const float4_t tgr00 = float4_sll(tabgr, 16); const float4_t targb = float4_or(ta0g0, t0r0b);
const float4_t tgrab = float4_or(t00ab, tgr00); float4_st(ptr, targb);
const float4_t ta0g0 = float4_and(tabgr, mf0f0); ptr += 16;
const float4_t t0g0b = float4_and(tgrab, m0f0f);
const float4_t targb = float4_or(ta0g0, t0g0b);
float4_st(ptr, targb);
ptr += 16;
}
} }
} }