Create SSE2 version of BlendColor

This commit is contained in:
Magnus Norddahl 2019-12-08 13:55:28 +01:00
parent 844b203971
commit aa44fa7c77

View file

@ -435,6 +435,7 @@ static const int shiftTable[] = {
24, 16, 8, 0 // STYLEALPHA_InvDstCol
};
#ifdef NO_SSE
template<typename OptT>
static void BlendColor(int y, int x0, int x1, PolyTriangleThreadData* thread)
{
@ -453,20 +454,25 @@ static void BlendColor(int y, int x0, int x1, PolyTriangleThreadData* thread)
uint32_t* fragcolor = thread->scanline.FragColor;
uint8_t* discard = thread->scanline.discard;
int srcSelect = style.SrcAlpha <= STYLEALPHA_One ? 0 : (style.SrcAlpha >= STYLEALPHA_DstCol ? 1 : 2);
int dstSelect = style.DestAlpha <= STYLEALPHA_One ? 0 : (style.DestAlpha >= STYLEALPHA_DstCol ? 1 : 2);
uint32_t inputs[3];
inputs[0] = 0;
for (int x = x0; x < x1; x++)
{
uint32_t fg = fragcolor[x];
if (OptT::Flags & SWBLEND_AlphaTest)
{
if (discard[x])
continue;
}
uint32_t bg = line[x];
inputs[1] = line[x];
inputs[2] = fragcolor[x];
int srcinput = style.SrcAlpha <= STYLEALPHA_One ? 0 : (style.SrcAlpha >= STYLEALPHA_DstCol ? bg : fg);
int dstinput = style.DestAlpha <= STYLEALPHA_One ? 0 : (style.DestAlpha >= STYLEALPHA_DstCol ? bg : fg);
uint32_t srcinput = inputs[srcSelect];
uint32_t dstinput = inputs[dstSelect];
uint32_t out[4];
for (int i = 0; i < 4; i++)
@ -484,8 +490,8 @@ static void BlendColor(int y, int x0, int x1, PolyTriangleThreadData* thread)
dst = dst + (dst >> 7);
// Multiply with input
src = src * ((fg >> (24 - (i << 3))) & 0xff);
dst = dst * ((bg >> (24 - (i << 3))) & 0xff);
src = src * ((inputs[2] >> (24 - (i << 3))) & 0xff);
dst = dst * ((inputs[1] >> (24 - (i << 3))) & 0xff);
// Apply blend operator
int32_t val;
@ -507,6 +513,88 @@ static void BlendColor(int y, int x0, int x1, PolyTriangleThreadData* thread)
line[x] = MAKEARGB(out[0], out[1], out[2], out[3]);
}
}
#else
template<typename OptT>
static void BlendColor(int y, int x0, int x1, PolyTriangleThreadData* thread)
{
using namespace TriScreenDrawerModes;
FRenderStyle style = thread->RenderStyle;
bool invsrc = style.SrcAlpha & 1;
bool invdst = style.DestAlpha & 1;
__m128i shiftsrc = _mm_loadu_si128((const __m128i*)(shiftTable + (style.SrcAlpha << 2)));
__m128i shiftdst = _mm_loadu_si128((const __m128i*)(shiftTable + (style.DestAlpha << 2)));
uint32_t* dest = (uint32_t*)thread->dest;
uint32_t* line = dest + y * (ptrdiff_t)thread->dest_pitch;
uint32_t* fragcolor = thread->scanline.FragColor;
uint8_t* discard = thread->scanline.discard;
int srcSelect = style.SrcAlpha <= STYLEALPHA_One ? 0 : (style.SrcAlpha >= STYLEALPHA_DstCol ? 1 : 2);
int dstSelect = style.DestAlpha <= STYLEALPHA_One ? 0 : (style.DestAlpha >= STYLEALPHA_DstCol ? 1 : 2);
uint32_t inputs[3];
inputs[0] = 0;
__m128i shiftmul = _mm_set_epi32(24, 16, 8, 0);
for (int x = x0; x < x1; x++)
{
if (OptT::Flags & SWBLEND_AlphaTest)
{
if (discard[x])
continue;
}
inputs[1] = line[x];
inputs[2] = fragcolor[x];
__m128i srcinput = _mm_set1_epi32(inputs[srcSelect]);
__m128i dstinput = _mm_set1_epi32(inputs[dstSelect]);
// Grab component for scale factors
__m128i src = _mm_and_si128(_mm_srlv_epi32(srcinput, shiftsrc), _mm_set1_epi32(0xff));
__m128i dst = _mm_and_si128(_mm_srlv_epi32(dstinput, shiftdst), _mm_set1_epi32(0xff));
// Inverse if needed
src = invsrc ? _mm_sub_epi32(_mm_set1_epi32(0xff), src) : src;
dst = invdst ? _mm_sub_epi32(_mm_set1_epi32(0xff), dst) : dst;
// Rescale 0-255 to 0-256
src = _mm_add_epi32(src, _mm_srli_epi32(src, 7));
dst = _mm_add_epi32(dst, _mm_srli_epi32(dst, 7));
// Multiply with input
__m128i mulsrc = _mm_and_si128(_mm_srlv_epi32(_mm_set1_epi32(inputs[2]), shiftmul), _mm_set1_epi32(0xff));
__m128i muldst = _mm_and_si128(_mm_srlv_epi32(_mm_set1_epi32(inputs[1]), shiftmul), _mm_set1_epi32(0xff));
__m128i mulresult = _mm_mullo_epi16(_mm_packs_epi32(src, dst), _mm_packs_epi32(mulsrc, muldst));
src = _mm_unpacklo_epi16(mulresult, _mm_setzero_si128());
dst = _mm_unpackhi_epi16(mulresult, _mm_setzero_si128());
// Apply blend operator
__m128i val;
if (OptT::Flags & SWBLEND_Sub)
{
val = _mm_sub_epi32(src, dst);
}
else if (OptT::Flags & SWBLEND_RevSub)
{
val = _mm_sub_epi32(dst, src);
}
else
{
val = _mm_add_epi32(src, dst);
}
__m128i out = _mm_srli_epi32(_mm_add_epi32(val, _mm_set1_epi32(127)), 8);
out = _mm_packs_epi32(out, out);
out = _mm_packus_epi16(out, out);
line[x] = _mm_cvtsi128_si32(out);
}
}
#endif
static void WriteColor(int y, int x0, int x1, PolyTriangleThreadData* thread)
{
@ -1006,7 +1094,6 @@ static void RunShader(int x0, int x1, PolyTriangleThreadData* thread)
else // func_normal
{
auto constants = thread->PushConstants;
auto& streamdata = thread->mainVertexShader.Data;
switch (constants->uTextureMode)
{
@ -1023,6 +1110,8 @@ static void RunShader(int x0, int x1, PolyTriangleThreadData* thread)
if (constants->uTextureMode != TM_FOGLAYER)
{
auto& streamdata = thread->mainVertexShader.Data;
if (streamdata.uAddColor.r != 0.0f || streamdata.uAddColor.g != 0.0f || streamdata.uAddColor.b != 0.0f)
{
FuncNormal_AddColor(x0, x1, thread);