void GSDrawScanlineCodeGenerator::modulate16(const Xmm& a, const Operand& f, int shift)
{
#if _M_SSE >= 0x500
if(shift == 0)
{
vpmulhrsw(a, f);
}
else
{
vpsllw(a, shift + 1);
vpmulhw(a, f);
}
#else
if(shift == 0 && m_cpu.has(util::Cpu::tSSSE3))
{
pmulhrsw(a, f);
}
else
{
psllw(a, shift + 1);
pmulhw(a, f);
}
#endif
}
void GSDrawScanlineCodeGenerator::modulate16(const Ymm& a, const Operand& f, int shift)
{
if(shift == 0)
{
vpmulhrsw(a, f);
}
else
{
vpsllw(a, (uint8)(shift + 1));
vpmulhw(a, f);
}
}
void GSDrawScanlineCodeGenerator::SampleTexture()
{
if(!m_sel.fb || m_sel.tfx == TFX_NONE)
{
return;
}
mov(rbx, ptr[r12 + offsetof(GSScanlineGlobalData, tex)]);
// ebx = tex
if(!m_sel.fst)
{
vrcpps(xmm0, xmm12);
vmulps(xmm4, xmm10, xmm0);
vmulps(xmm5, xmm11, xmm0);
vcvttps2dq(xmm4, xmm4);
vcvttps2dq(xmm5, xmm5);
if(m_sel.ltf)
{
// u -= 0x8000;
// v -= 0x8000;
mov(eax, 0x8000);
vmovd(xmm0, eax);
vpshufd(xmm0, xmm0, _MM_SHUFFLE(0, 0, 0, 0));
vpsubd(xmm4, xmm0);
vpsubd(xmm5, xmm0);
}
}
else
{
vmovdqa(xmm4, xmm10);
vmovdqa(xmm5, xmm11);
}
if(m_sel.ltf)
{
// GSVector4i uf = u.xxzzlh().srl16(1);
vpshuflw(xmm6, xmm4, _MM_SHUFFLE(2, 2, 0, 0));
vpshufhw(xmm6, xmm6, _MM_SHUFFLE(2, 2, 0, 0));
vpsrlw(xmm6, 1);
if(!m_sel.sprite)
{
// GSVector4i vf = v.xxzzlh().srl16(1);
vpshuflw(xmm7, xmm5, _MM_SHUFFLE(2, 2, 0, 0));
vpshufhw(xmm7, xmm7, _MM_SHUFFLE(2, 2, 0, 0));
vpsrlw(xmm7, 1);
}
}
// GSVector4i uv0 = u.sra32(16).ps32(v.sra32(16));
vpsrad(xmm4, 16);
vpsrad(xmm5, 16);
vpackssdw(xmm4, xmm5);
if(m_sel.ltf)
{
// GSVector4i uv1 = uv0.add16(GSVector4i::x0001());
vpcmpeqd(xmm0, xmm0);
vpsrlw(xmm0, 15);
vpaddw(xmm5, xmm4, xmm0);
// uv0 = Wrap(uv0);
// uv1 = Wrap(uv1);
Wrap(xmm4, xmm5);
}
else
{
// uv0 = Wrap(uv0);
Wrap(xmm4);
}
// xmm4 = uv0
// xmm5 = uv1 (ltf)
// xmm6 = uf
// xmm7 = vf
// GSVector4i x0 = uv0.upl16();
// GSVector4i y0 = uv0.uph16() << tw;
vpxor(xmm0, xmm0);
vpunpcklwd(xmm2, xmm4, xmm0);
vpunpckhwd(xmm3, xmm4, xmm0);
vpslld(xmm3, m_sel.tw + 3);
// xmm0 = 0
// xmm2 = x0
// xmm3 = y0
// xmm5 = uv1 (ltf)
// xmm6 = uf
// xmm7 = vf
if(m_sel.ltf)
{
// GSVector4i x1 = uv1.upl16();
// GSVector4i y1 = uv1.uph16() << tw;
vpunpcklwd(xmm4, xmm5, xmm0);
vpunpckhwd(xmm5, xmm5, xmm0);
vpslld(xmm5, m_sel.tw + 3);
// xmm2 = x0
// xmm3 = y0
// xmm4 = x1
// xmm5 = y1
// xmm6 = uf
// xmm7 = vf
// GSVector4i addr00 = y0 + x0;
// GSVector4i addr01 = y0 + x1;
// GSVector4i addr10 = y1 + x0;
// GSVector4i addr11 = y1 + x1;
vpaddd(xmm0, xmm3, xmm2);
vpaddd(xmm1, xmm3, xmm4);
vpaddd(xmm2, xmm5, xmm2);
vpaddd(xmm3, xmm5, xmm4);
// xmm0 = addr00
// xmm1 = addr01
// xmm2 = addr10
// xmm3 = addr11
// xmm6 = uf
// xmm7 = vf
// c00 = addr00.gather32_32((const uint32/uint8*)tex[, clut]);
// c01 = addr01.gather32_32((const uint32/uint8*)tex[, clut]);
// c10 = addr10.gather32_32((const uint32/uint8*)tex[, clut]);
// c11 = addr11.gather32_32((const uint32/uint8*)tex[, clut]);
ReadTexel(4, 0);
// xmm0 = c00
// xmm1 = c01
// xmm2 = c10
// xmm3 = c11
// xmm6 = uf
// xmm7 = vf
// GSVector4i rb00 = c00 & mask;
// GSVector4i ga00 = (c00 >> 8) & mask;
vpsllw(xmm4, xmm0, 8);
vpsrlw(xmm4, 8);
vpsrlw(xmm5, xmm0, 8);
// GSVector4i rb01 = c01 & mask;
// GSVector4i ga01 = (c01 >> 8) & mask;
vpsllw(xmm0, xmm1, 8);
vpsrlw(xmm0, 8);
vpsrlw(xmm1, 8);
// xmm0 = rb01
// xmm1 = ga01
// xmm2 = c10
// xmm3 = c11
// xmm4 = rb00
// xmm5 = ga00
// xmm6 = uf
// xmm7 = vf
// rb00 = rb00.lerp16<0>(rb01, uf);
// ga00 = ga00.lerp16<0>(ga01, uf);
lerp16(xmm0, xmm4, xmm6, 0);
lerp16(xmm1, xmm5, xmm6, 0);
// xmm0 = rb00
// xmm1 = ga00
// xmm2 = c10
// xmm3 = c11
// xmm6 = uf
// xmm7 = vf
// GSVector4i rb10 = c10 & mask;
// GSVector4i ga10 = (c10 >> 8) & mask;
vpsrlw(xmm5, xmm2, 8);
vpsllw(xmm2, 8);
vpsrlw(xmm4, xmm2, 8);
// GSVector4i rb11 = c11 & mask;
// GSVector4i ga11 = (c11 >> 8) & mask;
vpsrlw(xmm2, xmm3, 8);
vpsllw(xmm3, 8);
vpsrlw(xmm3, 8);
// xmm0 = rb00
// xmm1 = ga00
// xmm2 = rb11
// xmm3 = ga11
// xmm4 = rb10
// xmm5 = ga10
// xmm6 = uf
// xmm7 = vf
// rb10 = rb10.lerp16<0>(rb11, uf);
// ga10 = ga10.lerp16<0>(ga11, uf);
lerp16(xmm2, xmm4, xmm6, 0);
lerp16(xmm3, xmm5, xmm6, 0);
// xmm0 = rb00
// xmm1 = ga00
// xmm2 = rb10
// xmm3 = ga10
// xmm7 = vf
// rb00 = rb00.lerp16<0>(rb10, vf);
// ga00 = ga00.lerp16<0>(ga10, vf);
lerp16(xmm2, xmm0, xmm7, 0);
lerp16(xmm3, xmm1, xmm7, 0);
}
else
{
// GSVector4i addr00 = y0 + x0;
vpaddd(xmm3, xmm2);
// c00 = addr00.gather32_32((const uint32/uint8*)tex[, clut]);
ReadTexel(1, 0);
// GSVector4i mask = GSVector4i::x00ff();
// c[0] = c00 & mask;
// c[1] = (c00 >> 8) & mask;
vpsrlw(xmm3, xmm2, 8);
vpsllw(xmm2, 8);
vpsrlw(xmm2, 8);
}
// xmm2 = rb
// xmm3 = ga
}
void GSDrawScanlineCodeGenerator::AlphaBlend()
{
if(!m_sel.fwrite)
{
return;
}
if(m_sel.abe == 0 && m_sel.aa1 == 0)
{
return;
}
if((m_sel.aba != m_sel.abb) && (m_sel.aba == 1 || m_sel.abb == 1 || m_sel.abc == 1) || m_sel.abd == 1)
{
switch(m_sel.fpsm)
{
case 0:
case 1:
// c[2] = fd & mask;
// c[3] = (fd >> 8) & mask;
vpsllw(xmm0, xmm6, 8);
vpsrlw(xmm0, 8);
vpsrlw(xmm1, xmm6, 8);
break;
case 2:
// c[2] = ((fd & 0x7c00) << 9) | ((fd & 0x001f) << 3);
// c[3] = ((fd & 0x8000) << 8) | ((fd & 0x03e0) >> 2);
vpcmpeqd(xmm15, xmm15);
vpsrld(xmm15, 27); // 0x0000001f
vpand(xmm0, xmm6, xmm15);
vpslld(xmm0, 3);
vpslld(xmm15, 10); // 0x00007c00
vpand(xmm5, xmm6, xmm15);
vpslld(xmm5, 9);
vpor(xmm0, xmm1);
vpsrld(xmm15, 5); // 0x000003e0
vpand(xmm1, xmm6, xmm15);
vpsrld(xmm1, 2);
vpsllw(xmm15, 10); // 0x00008000
vpand(xmm5, xmm6, xmm15);
vpslld(xmm5, 8);
vpor(xmm1, xmm5);
break;
}
}
// xmm2, xmm3 = src rb, ga
// xmm0, xmm1 = dst rb, ga
// xmm5, xmm15 = free
if(m_sel.pabe || (m_sel.aba != m_sel.abb) && (m_sel.abb == 0 || m_sel.abd == 0))
{
vmovdqa(xmm5, xmm2);
}
if(m_sel.aba != m_sel.abb)
{
// rb = c[aba * 2 + 0];
switch(m_sel.aba)
{
case 0: break;
case 1: vmovdqa(xmm2, xmm0); break;
case 2: vpxor(xmm2, xmm2); break;
}
// rb = rb.sub16(c[abb * 2 + 0]);
switch(m_sel.abb)
{
case 0: vpsubw(xmm2, xmm5); break;
case 1: vpsubw(xmm2, xmm0); break;
case 2: break;
}
if(!(m_sel.fpsm == 1 && m_sel.abc == 1))
{
// GSVector4i a = abc < 2 ? c[abc * 2 + 1].yywwlh().sll16(7) : m_local.gd->afix;
switch(m_sel.abc)
{
case 0:
case 1:
vpshuflw(xmm15, m_sel.abc ? xmm1 : xmm3, _MM_SHUFFLE(3, 3, 1, 1));
vpshufhw(xmm15, xmm15, _MM_SHUFFLE(3, 3, 1, 1));
vpsllw(xmm15, 7);
break;
case 2:
vmovdqa(xmm15, ptr[r12 + offsetof(GSScanlineGlobalData, afix)]);
break;
}
// rb = rb.modulate16<1>(a);
modulate16(xmm2, xmm15, 1);
}
// rb = rb.add16(c[abd * 2 + 0]);
switch(m_sel.abd)
{
case 0: vpaddw(xmm2, xmm5); break;
case 1: vpaddw(xmm2, xmm0); break;
case 2: break;
}
}
else
{
// rb = c[abd * 2 + 0];
switch(m_sel.abd)
{
case 0: break;
case 1: vmovdqa(xmm2, xmm0); break;
case 2: vpxor(xmm2, xmm2); break;
}
}
if(m_sel.pabe)
{
// mask = (c[1] << 8).sra32(31);
vpslld(xmm0, xmm3, 8);
vpsrad(xmm0, 31);
// rb = c[0].blend8(rb, mask);
vpblendvb(xmm2, xmm5, xmm2, xmm0);
}
// xmm0 = pabe mask
// xmm3 = src ga
// xmm1 = dst ga
// xmm2 = rb
// xmm15 = a
// xmm5 = free
vmovdqa(xmm5, xmm3);
if(m_sel.aba != m_sel.abb)
{
// ga = c[aba * 2 + 1];
switch(m_sel.aba)
{
case 0: break;
case 1: vmovdqa(xmm3, xmm1); break;
case 2: vpxor(xmm3, xmm3); break;
}
// ga = ga.sub16(c[abeb * 2 + 1]);
switch(m_sel.abb)
{
case 0: vpsubw(xmm3, xmm5); break;
case 1: vpsubw(xmm3, xmm1); break;
case 2: break;
}
if(!(m_sel.fpsm == 1 && m_sel.abc == 1))
{
// ga = ga.modulate16<1>(a);
modulate16(xmm3, xmm15, 1);
}
// ga = ga.add16(c[abd * 2 + 1]);
switch(m_sel.abd)
{
case 0: vpaddw(xmm3, xmm5); break;
case 1: vpaddw(xmm3, xmm1); break;
case 2: break;
}
}
else
{
// ga = c[abd * 2 + 1];
switch(m_sel.abd)
{
case 0: break;
case 1: vmovdqa(xmm3, xmm1); break;
case 2: vpxor(xmm3, xmm3); break;
}
}
// xmm0 = pabe mask
// xmm5 = src ga
// xmm2 = rb
// xmm3 = ga
// xmm1, xmm15 = free
if(m_sel.pabe)
{
vpsrld(xmm0, 16); // zero out high words to select the source alpha in blend (so it also does mix16)
// ga = c[1].blend8(ga, mask).mix16(c[1]);
vpblendvb(xmm3, xmm5, xmm3, xmm0);
}
else
{
if(m_sel.fpsm != 1) // TODO: fm == 0xffxxxxxx
{
mix16(xmm3, xmm5, xmm15);
}
}
}
