void GSDrawScanlineCodeGenerator::Wrap(const Xmm& uv)
{
// xmm0, xmm1, xmm2, xmm3 = free
int wms_clamp = ((m_sel.wms + 1) >> 1) & 1;
int wmt_clamp = ((m_sel.wmt + 1) >> 1) & 1;
int region = ((m_sel.wms | m_sel.wmt) >> 1) & 1;
if(wms_clamp == wmt_clamp)
{
if(wms_clamp)
{
if(region)
{
vpmaxsw(uv, ptr[r12 + offsetof(GSScanlineGlobalData, t.min)]);
}
else
{
vpxor(xmm0, xmm0);
vpmaxsw(uv, xmm0);
}
vpminsw(uv, ptr[r12 + offsetof(GSScanlineGlobalData, t.max)]);
}
else
{
vpand(uv, ptr[r12 + offsetof(GSScanlineGlobalData, t.min)]);
if(region)
{
vpor(uv, ptr[r12 + offsetof(GSScanlineGlobalData, t.max)]);
}
}
}
else
{
vmovdqa(xmm2, ptr[r12 + offsetof(GSScanlineGlobalData, t.min)]);
vmovdqa(xmm3, ptr[r12 + offsetof(GSScanlineGlobalData, t.max)]);
vmovdqa(xmm0, ptr[r12 + offsetof(GSScanlineGlobalData, t.mask)]);
// GSVector4i repeat = (t & m_local.gd->t.min) | m_local.gd->t.max;
vpand(xmm1, uv, xmm2);
if(region)
{
vpor(xmm1, xmm3);
}
// GSVector4i clamp = t.sat_i16(m_local.gd->t.min, m_local.gd->t.max);
vpmaxsw(uv, xmm2);
vpminsw(uv, xmm3);
// clamp.blend8(repeat, m_local.gd->t.mask);
vpblendvb(uv, xmm1, xmm0);
}
}
void GSDrawScanlineCodeGenerator::blendr(const Xmm& b, const Xmm& a, const Xmm& mask)
{
#if _M_SSE >= 0x500
vpand(b, mask);
vpandn(mask, a);
vpor(b, mask);
#else
pand(b, mask);
pandn(mask, a);
por(b, mask);
#endif
}
void GSDrawScanlineCodeGenerator::TestZ(const Xmm& temp1, const Xmm& temp2)
{
if(!m_sel.zb)
{
return;
}
// int za = fza_base.y + fza_offset->y;
movsxd(rbp, dword[rsi + 4]);
movsxd(rax, dword[rdi + 4]);
add(rbp, rax);
// GSVector4i zs = zi;
if(!m_sel.sprite)
{
if(m_sel.zoverflow)
{
// zs = (GSVector4i(z * 0.5f) << 1) | (GSVector4i(z) & GSVector4i::x00000001());
mov(rax, (size_t)&GSVector4::m_half);
vbroadcastss(xmm0, ptr[rax]);
vmulps(xmm0, xmm8);
vcvttps2dq(xmm0, xmm0);
vpslld(xmm0, 1);
vcvttps2dq(xmm1, xmm8);
vpcmpeqd(xmm2, xmm2);
vpsrld(xmm2, 31);
vpand(xmm1, xmm2);
vpor(xmm0, xmm1);
}
else
{
// zs = GSVector4i(z);
vcvttps2dq(xmm0, xmm8);
}
if(m_sel.zwrite)
{
vmovdqa(ptr[r11 + offsetof(GSScanlineLocalData, temp.zs)], xmm0);
}
}
if(m_sel.ztest)
{
ReadPixel(xmm1, rbp);
if(m_sel.zwrite && m_sel.zpsm < 2)
{
vmovdqa(ptr[r11 + offsetof(GSScanlineLocalData, temp.zd)], xmm1);
}
// zd &= 0xffffffff >> m_sel.zpsm * 8;
if(m_sel.zpsm)
{
vpslld(xmm1, m_sel.zpsm * 8);
vpsrld(xmm1, m_sel.zpsm * 8);
}
if(m_sel.zoverflow || m_sel.zpsm == 0)
{
// GSVector4i off = GSVector4i::x80000000();
vpcmpeqd(xmm2, xmm2);
vpslld(xmm2, 31);
// GSVector4i zso = zs - off;
// GSVector4i zdo = zd - off;
vpsubd(xmm0, xmm2);
vpsubd(xmm1, xmm2);
}
switch(m_sel.ztst)
{
case ZTST_GEQUAL:
// test |= zso < zdo; // ~(zso >= zdo)
vpcmpgtd(xmm1, xmm0);
vpor(xmm15, xmm1);
break;
case ZTST_GREATER: // TODO: tidus hair and chocobo wings only appear fully when this is tested as ZTST_GEQUAL
// test |= zso <= zdo; // ~(zso > zdo)
vpcmpgtd(xmm0, xmm1);
vpcmpeqd(xmm2, xmm2);
vpxor(xmm0, xmm2);
vpor(xmm15, xmm0);
break;
}
alltrue();
}
}
void GSDrawScanlineCodeGenerator::WriteMask()
{
// fm |= test;
// zm |= test;
if(m_sel.fwrite)
{
vpor(xmm4, xmm15);
}
if(m_sel.zwrite)
{
vpor(xmm5, xmm15);
}
// int fzm = ~(fm == GSVector4i::xffffffff()).ps32(zm == GSVector4i::xffffffff()).mask();
vpcmpeqd(xmm1, xmm1);
if(m_sel.fwrite && m_sel.zwrite)
{
vpcmpeqd(xmm0, xmm1, xmm5);
vpcmpeqd(xmm1, xmm4);
vpackssdw(xmm1, xmm0);
}
else if(m_sel.fwrite)
{
vpcmpeqd(xmm1, xmm4);
vpackssdw(xmm1, xmm1);
}
else if(m_sel.zwrite)
{
vpcmpeqd(xmm1, xmm5);
vpackssdw(xmm1, xmm1);
}
vpmovmskb(edx, xmm1);
not(edx);
}
void GSDrawScanlineCodeGenerator::TestDestAlpha()
{
if(!m_sel.date || m_sel.fpsm != 0 && m_sel.fpsm != 2)
{
return;
}
// test |= ((fd [<< 16]) ^ m_local.gd->datm).sra32(31);
if(m_sel.datm)
{
if(m_sel.fpsm == 2)
{
vpxor(xmm0, xmm0);
//vpsrld(xmm1, xmm6, 15);
vpslld(xmm1, xmm6, 16);
vpsrad(xmm1, 31);
vpcmpeqd(xmm1, xmm0);
}
else
{
vpcmpeqd(xmm0, xmm0);
vpxor(xmm1, xmm6, xmm0);
vpsrad(xmm1, 31);
}
}
else
{
if(m_sel.fpsm == 2)
{
vpslld(xmm1, xmm6, 16);
vpsrad(xmm1, 31);
}
else
{
vpsrad(xmm1, xmm6, 31);
}
}
vpor(xmm15, xmm1);
alltrue();
}
void GSDrawScanlineCodeGenerator::Init()
{
// int skip = left & 3;
mov(rbx, rdx);
and(rdx, 3);
// left -= skip;
sub(rbx, rdx);
// int steps = pixels + skip - 4;
lea(rcx, ptr[rcx + rdx - 4]);
// GSVector4i test = m_test[skip] | m_test[7 + (steps & (steps >> 31))];
shl(rdx, 4);
vmovdqa(xmm15, ptr[rdx + r10]);
mov(rax, rcx);
sar(rax, 63);
and(rax, rcx);
shl(rax, 4);
vpor(xmm15, ptr[rax + r10 + 7 * 16]);
// GSVector2i* fza_base = &m_local.gd->fzbr[top];
mov(rax, (size_t)m_local.gd->fzbr);
lea(rsi, ptr[rax + r8 * 8]);
// GSVector2i* fza_offset = &m_local.gd->fzbc[left >> 2];
mov(rax, (size_t)m_local.gd->fzbc);
lea(rdi, ptr[rax + rbx * 2]);
if(!m_sel.sprite && (m_sel.fwrite && m_sel.fge || m_sel.zb) || m_sel.fb && (m_sel.edge || m_sel.tfx != TFX_NONE || m_sel.iip))
{
// edx = &m_local.d[skip]
lea(rdx, ptr[rdx * 8 + r11 + offsetof(GSScanlineLocalData, d)]);
}
if(!m_sel.sprite)
{
if(m_sel.fwrite && m_sel.fge || m_sel.zb)
{
vmovaps(xmm0, ptr[r9 + offsetof(GSVertexSW, p)]); // v.p
if(m_sel.fwrite && m_sel.fge)
{
// f = GSVector4i(vp).zzzzh().zzzz().add16(m_local.d[skip].f);
vcvttps2dq(xmm9, xmm0);
vpshufhw(xmm9, xmm9, _MM_SHUFFLE(2, 2, 2, 2));
vpshufd(xmm9, xmm9, _MM_SHUFFLE(2, 2, 2, 2));
vpaddw(xmm9, ptr[rdx + 16 * 6]);
}
if(m_sel.zb)
{
// z = vp.zzzz() + m_local.d[skip].z;
vshufps(xmm8, xmm0, xmm0, _MM_SHUFFLE(2, 2, 2, 2));
vaddps(xmm8, ptr[rdx]);
}
}
}
else
{
if(m_sel.ztest)
{
vmovdqa(xmm8, ptr[r11 + offsetof(GSScanlineLocalData, p.z)]);
}
}
if(m_sel.fb)
{
if(m_sel.edge || m_sel.tfx != TFX_NONE)
{
vmovaps(xmm0, ptr[r9 + offsetof(GSVertexSW, t)]); // v.t
}
if(m_sel.edge)
{
vpshufhw(xmm1, xmm0, _MM_SHUFFLE(2, 2, 2, 2));
vpshufd(xmm1, xmm1, _MM_SHUFFLE(3, 3, 3, 3));
vpsrlw(xmm1, 9);
vmovdqa(ptr[r11 + offsetof(GSScanlineLocalData, temp.cov)], xmm1);
}
if(m_sel.tfx != TFX_NONE)
{
if(m_sel.fst)
{
// GSVector4i vti(vt);
vcvttps2dq(xmm0, xmm0);
// s = vti.xxxx() + m_local.d[skip].s;
// t = vti.yyyy(); if(!sprite) t += m_local.d[skip].t;
vpshufd(xmm10, xmm0, _MM_SHUFFLE(0, 0, 0, 0));
vpshufd(xmm11, xmm0, _MM_SHUFFLE(1, 1, 1, 1));
vpaddd(xmm10, ptr[rdx + offsetof(GSScanlineLocalData::skip, s)]);
if(!m_sel.sprite || m_sel.mmin)
{
vpaddd(xmm11, ptr[rdx + offsetof(GSScanlineLocalData::skip, t)]);
}
else
{
if(m_sel.ltf)
{
vpshuflw(xmm6, xmm11, _MM_SHUFFLE(2, 2, 0, 0));
vpshufhw(xmm6, xmm6, _MM_SHUFFLE(2, 2, 0, 0));
vpsrlw(xmm6, 1);
}
}
}
else
{
// s = vt.xxxx() + m_local.d[skip].s;
// t = vt.yyyy() + m_local.d[skip].t;
// q = vt.zzzz() + m_local.d[skip].q;
vshufps(xmm10, xmm0, xmm0, _MM_SHUFFLE(0, 0, 0, 0));
vshufps(xmm11, xmm0, xmm0, _MM_SHUFFLE(1, 1, 1, 1));
vshufps(xmm12, xmm0, xmm0, _MM_SHUFFLE(2, 2, 2, 2));
vaddps(xmm10, ptr[rdx + offsetof(GSScanlineLocalData::skip, s)]);
vaddps(xmm11, ptr[rdx + offsetof(GSScanlineLocalData::skip, t)]);
vaddps(xmm12, ptr[rdx + offsetof(GSScanlineLocalData::skip, q)]);
}
}
if(!(m_sel.tfx == TFX_DECAL && m_sel.tcc))
{
if(m_sel.iip)
{
// GSVector4i vc = GSVector4i(v.c);
vcvttps2dq(xmm0, ptr[r9 + offsetof(GSVertexSW, c)]); // v.c
// vc = vc.upl16(vc.zwxy());
vpshufd(xmm1, xmm0, _MM_SHUFFLE(1, 0, 3, 2));
vpunpcklwd(xmm0, xmm1);
// rb = vc.xxxx().add16(m_local.d[skip].rb);
// ga = vc.zzzz().add16(m_local.d[skip].ga);
vpshufd(xmm13, xmm0, _MM_SHUFFLE(0, 0, 0, 0));
vpshufd(xmm14, xmm0, _MM_SHUFFLE(2, 2, 2, 2));
vpaddw(xmm13, ptr[rdx + offsetof(GSScanlineLocalData::skip, rb)]);
vpaddw(xmm14, ptr[rdx + offsetof(GSScanlineLocalData::skip, ga)]);
}
else
{
vmovdqa(xmm13, ptr[r11 + offsetof(GSScanlineLocalData, c.rb)]);
vmovdqa(xmm14, ptr[r11 + offsetof(GSScanlineLocalData, c.ga)]);
}
}
}
}
void GSDrawScanlineCodeGenerator::WriteFrame()
{
if(!m_sel.fwrite)
{
return;
}
if(m_sel.colclamp == 0)
{
// c[0] &= 0x00ff00ff;
// c[1] &= 0x00ff00ff;
vpcmpeqd(xmm15, xmm15);
vpsrlw(xmm15, 8);
vpand(xmm2, xmm15);
vpand(xmm3, xmm15);
}
if(m_sel.fpsm == 2 && m_sel.dthe)
{
mov(rax, r8);
and(rax, 3);
shl(rax, 5);
vpaddw(xmm2, ptr[r12 + rax + offsetof(GSScanlineGlobalData, dimx[0])]);
vpaddw(xmm3, ptr[r12 + rax + offsetof(GSScanlineGlobalData, dimx[1])]);
}
// GSVector4i fs = c[0].upl16(c[1]).pu16(c[0].uph16(c[1]));
vpunpckhwd(xmm15, xmm2, xmm3);
vpunpcklwd(xmm2, xmm3);
vpackuswb(xmm2, xmm15);
if(m_sel.fba && m_sel.fpsm != 1)
{
// fs |= 0x80000000;
vpcmpeqd(xmm15, xmm15);
vpslld(xmm15, 31);
vpor(xmm2, xmm15);
}
// xmm2 = fs
// xmm4 = fm
// xmm6 = fd
if(m_sel.fpsm == 2)
{
// GSVector4i rb = fs & 0x00f800f8;
// GSVector4i ga = fs & 0x8000f800;
mov(eax, 0x00f800f8);
vmovd(xmm0, eax);
vpshufd(xmm0, xmm0, _MM_SHUFFLE(0, 0, 0, 0));
mov(eax, 0x8000f800);
vmovd(xmm1, eax);
vpshufd(xmm1, xmm1, _MM_SHUFFLE(0, 0, 0, 0));
vpand(xmm0, xmm2);
vpand(xmm1, xmm2);
// fs = (ga >> 16) | (rb >> 9) | (ga >> 6) | (rb >> 3);
vpsrld(xmm2, xmm0, 9);
vpsrld(xmm0, 3);
vpsrld(xmm3, xmm1, 16);
vpsrld(xmm1, 6);
vpor(xmm0, xmm1);
vpor(xmm2, xmm3);
vpor(xmm2, xmm0);
}
if(m_sel.rfb)
{
// fs = fs.blend(fd, fm);
blend(xmm2, xmm6, xmm4); // TODO: could be skipped in certain cases, depending on fpsm and fm
}
bool fast = m_sel.rfb && m_sel.fpsm < 2;
WritePixel(xmm2, rbx, dl, fast, m_sel.fpsm, 0);
}
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);
}
}
}
void GSDrawScanlineCodeGenerator::TestAlpha()
{
switch(m_sel.afail)
{
case AFAIL_FB_ONLY:
if(!m_sel.zwrite) return;
break;
case AFAIL_ZB_ONLY:
if(!m_sel.fwrite) return;
break;
case AFAIL_RGB_ONLY:
if(!m_sel.zwrite && m_sel.fpsm == 1) return;
break;
}
switch(m_sel.atst)
{
case ATST_NEVER:
// t = GSVector4i::xffffffff();
vpcmpeqd(xmm1, xmm1);
break;
case ATST_ALWAYS:
return;
case ATST_LESS:
case ATST_LEQUAL:
// t = (ga >> 16) > m_local.gd->aref;
vpsrld(xmm1, xmm3, 16);
vpcmpgtd(xmm1, ptr[r12 + offsetof(GSScanlineGlobalData, aref)]);
break;
case ATST_EQUAL:
// t = (ga >> 16) != m_local.gd->aref;
vpsrld(xmm1, xmm3, 16);
vpcmpeqd(xmm1, ptr[r12 + offsetof(GSScanlineGlobalData, aref)]);
vpcmpeqd(xmm0, xmm0);
vpxor(xmm1, xmm0);
break;
case ATST_GEQUAL:
case ATST_GREATER:
// t = (ga >> 16) < m_local.gd->aref;
vpsrld(xmm0, xmm3, 16);
vmovdqa(xmm1, ptr[r12 + offsetof(GSScanlineGlobalData, aref)]);
vpcmpgtd(xmm1, xmm0);
break;
case ATST_NOTEQUAL:
// t = (ga >> 16) == m_local.gd->aref;
vpsrld(xmm1, xmm3, 16);
vpcmpeqd(xmm1, ptr[r12 + offsetof(GSScanlineGlobalData, aref)]);
break;
}
switch(m_sel.afail)
{
case AFAIL_KEEP:
// test |= t;
vpor(xmm15, xmm1);
alltrue();
break;
case AFAIL_FB_ONLY:
// zm |= t;
vpor(xmm5, xmm1);
break;
case AFAIL_ZB_ONLY:
// fm |= t;
vpor(xmm4, xmm1);
break;
case AFAIL_RGB_ONLY:
// zm |= t;
vpor(xmm5, xmm1);
// fm |= t & GSVector4i::xff000000();
vpsrld(xmm1, 24);
vpslld(xmm1, 24);
vpor(xmm4, xmm1);
break;
}
}
void GSDrawScanlineCodeGenerator::blendr(const Ymm& b, const Ymm& a, const Ymm& mask)
{
vpand(b, mask);
vpandn(mask, a);
vpor(b, mask);
}
