zdoom/gzdoom
0
0
Fork 0
mirror of https://github.com/ZDoom/gzdoom.git synced 2024-11-13 16:07:45 +00:00
Code Issues Projects Releases Packages Wiki Activity

A little bit of SSE to speed it slightly more

Browse source
This commit is contained in:
Magnus Norddahl 2016-11-08 17:38:19 +01:00
parent 466274d7f6
commit c1d496c17f
1 changed files with 33 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

33
src/r_poly_triangle.cpp
View file

@ -638,6 +638,10 @@ void ScreenPolyTriangleDrawer::draw32(const ScreenPolyTriangleDrawerArgs *args,
int textureHeight = args->textureHeight;
uint32_t light = args->uniforms->light;
#if !defined(NO_SSE)
__m128i mlight = _mm_set1_epi16(light);
#endif
// 28.4 fixed-point coordinates
const int Y1 = (int)round(16.0f * v1.y);
const int Y2 = (int)round(16.0f * v2.y);
@ -781,6 +785,7 @@ void ScreenPolyTriangleDrawer::draw32(const ScreenPolyTriangleDrawerArgs *args,
varyingStep[i] = (uint32_t)(step * 0x100000000LL);
}
#if NO_SSE
for (int ix = x; ix < x + q; ix++)
{
uint32_t ufrac = varying[0];
@ -802,6 +807,34 @@ void ScreenPolyTriangleDrawer::draw32(const ScreenPolyTriangleDrawerArgs *args,
for (int i = 0; i < TriVertex::NumVarying; i++)
varying[i] += varyingStep[i];
}
#else
for (int sse = 0; sse < q / 4; sse++)
{
uint32_t fg[4];
for (int ix = 0; ix < 4; ix++)
{
uint32_t ufrac = varying[0];
uint32_t vfrac = varying[1];
uint32_t upos = ((ufrac >> 16) * textureWidth) >> 16;
uint32_t vpos = ((vfrac >> 16) * textureHeight) >> 16;
uint32_t uvoffset = upos * textureHeight + vpos;
fg[ix] = texturePixels[uvoffset];
for (int i = 0; i < TriVertex::NumVarying; i++)
varying[i] += varyingStep[i];
}
__m128i mfg = _mm_loadu_si128((const __m128i*)fg);
__m128i mfg0 = _mm_unpacklo_epi8(mfg, _mm_setzero_si128());
__m128i mfg1 = _mm_unpackhi_epi8(mfg, _mm_setzero_si128());
__m128i mout0 = _mm_srli_epi16(_mm_mullo_epi16(mfg0, mlight), 8);
__m128i mout1 = _mm_srli_epi16(_mm_mullo_epi16(mfg1, mlight), 8);
__m128i mout = _mm_packus_epi16(mout0, mout1);
__m128i mmask0 = _mm_shufflehi_epi16(_mm_shufflelo_epi16(mfg0, _MM_SHUFFLE(3, 3, 3, 3)), _MM_SHUFFLE(3, 3, 3, 3));
__m128i mmask1 = _mm_shufflehi_epi16(_mm_shufflelo_epi16(mfg1, _MM_SHUFFLE(3, 3, 3, 3)), _MM_SHUFFLE(3, 3, 3, 3));
__m128i mmask = _mm_cmplt_epi8(_mm_packus_epi16(mmask0, mmask1), _mm_setzero_si128());
_mm_maskmoveu_si128(mout, mmask, (char*)(&buffer[x + sse * 4]));
}
#endif
buffer += pitch;
}