void GSDrawScanlineCodeGenerator::SampleTexture()
{
if(!m_sel.fb || m_sel.tfx == TFX_NONE)
{
return;
}
mov(ebx, dword[&m_env.tex]);
// ebx = tex
if(!m_sel.fst)
{
// TODO: move these into Init/Step too?
cvttps2dq(xmm2, xmm2);
cvttps2dq(xmm3, xmm3);
if(m_sel.ltf)
{
// u -= 0x8000;
// v -= 0x8000;
mov(eax, 0x8000);
movd(xmm4, eax);
pshufd(xmm4, xmm4, _MM_SHUFFLE(0, 0, 0, 0));
psubd(xmm2, xmm4);
psubd(xmm3, xmm4);
}
}
// xmm2 = u
// xmm3 = v
if(m_sel.ltf)
{
// GSVector4i uf = u.xxzzlh().srl16(1);
movdqa(xmm0, xmm2);
pshuflw(xmm0, xmm0, _MM_SHUFFLE(2, 2, 0, 0));
pshufhw(xmm0, xmm0, _MM_SHUFFLE(2, 2, 0, 0));
psrlw(xmm0, 1);
movdqa(xmmword[&m_env.temp.uf], xmm0);
if(!m_sel.sprite)
{
// GSVector4i vf = v.xxzzlh().srl16(1);
movdqa(xmm1, xmm3);
pshuflw(xmm1, xmm1, _MM_SHUFFLE(2, 2, 0, 0));
pshufhw(xmm1, xmm1, _MM_SHUFFLE(2, 2, 0, 0));
psrlw(xmm1, 1);
movdqa(xmmword[&m_env.temp.vf], xmm1);
}
}
// GSVector4i uv0 = u.sra32(16).ps32(v.sra32(16));
psrad(xmm2, 16);
psrad(xmm3, 16);
packssdw(xmm2, xmm3);
if(m_sel.ltf)
{
// GSVector4i uv1 = uv0.add16(GSVector4i::x0001());
movdqa(xmm3, xmm2);
pcmpeqd(xmm1, xmm1);
psrlw(xmm1, 15);
paddw(xmm3, xmm1);
// uv0 = Wrap(uv0);
// uv1 = Wrap(uv1);
Wrap(xmm2, xmm3);
}
else
{
// uv0 = Wrap(uv0);
Wrap(xmm2);
}
// xmm2 = uv0
// xmm3 = uv1 (ltf)
// xmm0, xmm1, xmm4, xmm5, xmm6 = free
// xmm7 = used
// GSVector4i y0 = uv0.uph16() << tw;
// GSVector4i x0 = uv0.upl16();
pxor(xmm0, xmm0);
movd(xmm1, ptr[&m_env.tw]);
movdqa(xmm4, xmm2);
punpckhwd(xmm2, xmm0);
punpcklwd(xmm4, xmm0);
pslld(xmm2, xmm1);
// xmm0 = 0
// xmm1 = tw
// xmm2 = y0
// xmm3 = uv1 (ltf)
// xmm4 = x0
// xmm5, xmm6 = free
// xmm7 = used
if(m_sel.ltf)
{
// GSVector4i y1 = uv1.uph16() << tw;
// GSVector4i x1 = uv1.upl16();
movdqa(xmm6, xmm3);
punpckhwd(xmm3, xmm0);
punpcklwd(xmm6, xmm0);
pslld(xmm3, xmm1);
// xmm2 = y0
// xmm3 = y1
// xmm4 = x0
// xmm6 = x1
// xmm0, xmm5, xmm6 = free
// xmm7 = used
// GSVector4i addr00 = y0 + x0;
// GSVector4i addr01 = y0 + x1;
// GSVector4i addr10 = y1 + x0;
// GSVector4i addr11 = y1 + x1;
movdqa(xmm5, xmm2);
paddd(xmm5, xmm4);
paddd(xmm2, xmm6);
movdqa(xmm0, xmm3);
paddd(xmm0, xmm4);
paddd(xmm3, xmm6);
// xmm5 = addr00
// xmm2 = addr01
// xmm0 = addr10
// xmm3 = addr11
// xmm1, xmm4, xmm6 = free
// xmm7 = used
// 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(xmm6, xmm5, xmm1, xmm4);
// xmm2, xmm5, xmm1 = free
ReadTexel(xmm4, xmm2, xmm5, xmm1);
// xmm0, xmm2, xmm5 = free
ReadTexel(xmm1, xmm0, xmm2, xmm5);
// xmm3, xmm0, xmm2 = free
ReadTexel(xmm5, xmm3, xmm0, xmm2);
// xmm6 = c00
// xmm4 = c01
// xmm1 = c10
// xmm5 = c11
// xmm0, xmm2, xmm3 = free
// xmm7 = used
movdqa(xmm0, xmmword[&m_env.temp.uf]);
// GSVector4i rb00 = c00 & mask;
// GSVector4i ga00 = (c00 >> 8) & mask;
movdqa(xmm2, xmm6);
psllw(xmm2, 8);
psrlw(xmm2, 8);
psrlw(xmm6, 8);
// GSVector4i rb01 = c01 & mask;
// GSVector4i ga01 = (c01 >> 8) & mask;
movdqa(xmm3, xmm4);
psllw(xmm3, 8);
psrlw(xmm3, 8);
psrlw(xmm4, 8);
// xmm0 = uf
// xmm2 = rb00
// xmm3 = rb01
// xmm6 = ga00
// xmm4 = ga01
// xmm1 = c10
// xmm5 = c11
// xmm7 = used
// rb00 = rb00.lerp16<0>(rb01, uf);
// ga00 = ga00.lerp16<0>(ga01, uf);
lerp16<0>(xmm3, xmm2, xmm0);
lerp16<0>(xmm4, xmm6, xmm0);
// xmm0 = uf
// xmm3 = rb00
// xmm4 = ga00
// xmm1 = c10
// xmm5 = c11
// xmm2, xmm6 = free
// xmm7 = used
// GSVector4i rb10 = c10 & mask;
// GSVector4i ga10 = (c10 >> 8) & mask;
movdqa(xmm2, xmm1);
psllw(xmm1, 8);
psrlw(xmm1, 8);
psrlw(xmm2, 8);
// GSVector4i rb11 = c11 & mask;
// GSVector4i ga11 = (c11 >> 8) & mask;
movdqa(xmm6, xmm5);
psllw(xmm5, 8);
psrlw(xmm5, 8);
psrlw(xmm6, 8);
// xmm0 = uf
// xmm3 = rb00
// xmm4 = ga00
// xmm1 = rb10
// xmm5 = rb11
// xmm2 = ga10
// xmm6 = ga11
// xmm7 = used
// rb10 = rb10.lerp16<0>(rb11, uf);
// ga10 = ga10.lerp16<0>(ga11, uf);
lerp16<0>(xmm5, xmm1, xmm0);
lerp16<0>(xmm6, xmm2, xmm0);
// xmm3 = rb00
// xmm4 = ga00
// xmm5 = rb10
// xmm6 = ga10
// xmm0, xmm1, xmm2 = free
// xmm7 = used
// rb00 = rb00.lerp16<0>(rb10, vf);
// ga00 = ga00.lerp16<0>(ga10, vf);
movdqa(xmm0, xmmword[&m_env.temp.vf]);
lerp16<0>(xmm5, xmm3, xmm0);
lerp16<0>(xmm6, xmm4, xmm0);
}
else
{
// GSVector4i addr00 = y0 + x0;
paddd(xmm2, xmm4);
// c00 = addr00.gather32_32((const uint32/uint8*)tex[, clut]);
ReadTexel(xmm5, xmm2, xmm0, xmm1);
// GSVector4i mask = GSVector4i::x00ff();
// c[0] = c00 & mask;
// c[1] = (c00 >> 8) & mask;
movdqa(xmm6, xmm5);
psllw(xmm5, 8);
psrlw(xmm5, 8);
psrlw(xmm6, 8);
}
}
void GPUDrawScanlineCodeGenerator::Init()
{
mov(eax, dword[esp + _top]);
// uint16* fb = (uint16*)m_global.vm + (top << (10 + sel.scalex)) + left;
mov(edi, eax);
shl(edi, 10 + m_sel.scalex);
add(edi, edx);
lea(edi, ptr[edi * 2 + (size_t)m_local.gd->vm]);
// int steps = pixels - 8;
sub(ecx, 8);
if(m_sel.dtd)
{
// dither = GSVector4i::load<false>(&m_dither[top & 3][left & 3]);
and(eax, 3);
shl(eax, 5);
and(edx, 3);
shl(edx, 1);
movdqu(xmm0, ptr[eax + edx + (size_t)m_dither]);
movdqa(ptr[&m_local.temp.dither], xmm0);
}
mov(edx, dword[esp + _v]);
if(m_sel.tme)
{
mov(esi, dword[&m_local.gd->tex]);
// GSVector4i vt = GSVector4i(v.t).xxzzl();
cvttps2dq(xmm4, ptr[edx + offsetof(GSVertexSW, t)]);
pshuflw(xmm4, xmm4, _MM_SHUFFLE(2, 2, 0, 0));
// s = vt.xxxx().add16(m_local.d.s);
// t = vt.yyyy().add16(m_local.d.t);
pshufd(xmm2, xmm4, _MM_SHUFFLE(0, 0, 0, 0));
pshufd(xmm3, xmm4, _MM_SHUFFLE(1, 1, 1, 1));
paddw(xmm2, ptr[&m_local.d.s]);
if(!m_sel.sprite)
{
paddw(xmm3, ptr[&m_local.d.t]);
}
else
{
if(m_sel.ltf)
{
movdqa(xmm0, xmm3);
psllw(xmm0, 8);
psrlw(xmm0, 1);
movdqa(ptr[&m_local.temp.vf], xmm0);
}
}
movdqa(ptr[&m_local.temp.s], xmm2);
movdqa(ptr[&m_local.temp.t], xmm3);
}
if(m_sel.tfx != 3) // != decal
{
// GSVector4i vc = GSVector4i(v.c).xxzzlh();
cvttps2dq(xmm6, ptr[edx + offsetof(GSVertexSW, c)]);
pshuflw(xmm6, xmm6, _MM_SHUFFLE(2, 2, 0, 0));
pshufhw(xmm6, xmm6, _MM_SHUFFLE(2, 2, 0, 0));
// r = vc.xxxx();
// g = vc.yyyy();
// b = vc.zzzz();
pshufd(xmm4, xmm6, _MM_SHUFFLE(0, 0, 0, 0));
pshufd(xmm5, xmm6, _MM_SHUFFLE(1, 1, 1, 1));
pshufd(xmm6, xmm6, _MM_SHUFFLE(2, 2, 2, 2));
if(m_sel.iip)
{
// r = r.add16(m_local.d.r);
// g = g.add16(m_local.d.g);
// b = b.add16(m_local.d.b);
paddw(xmm4, ptr[&m_local.d.r]);
paddw(xmm5, ptr[&m_local.d.g]);
paddw(xmm6, ptr[&m_local.d.b]);
}
movdqa(ptr[&m_local.temp.r], xmm4);
movdqa(ptr[&m_local.temp.g], xmm5);
movdqa(ptr[&m_local.temp.b], xmm6);
}
}
static __forceinline void idct_M128ASM(__m128i &src0,__m128i &src1,__m128i &src2,__m128i &src3,__m128i &src4,__m128i &src5,__m128i &src6,__m128i &src7)
{
/*src0=_mm_srai_epi16(src0,4);
src1=_mm_srai_epi16(src1,4);
src2=_mm_srai_epi16(src2,4);
src3=_mm_srai_epi16(src3,4);
src4=_mm_srai_epi16(src4,4);
src5=_mm_srai_epi16(src5,4);
src6=_mm_srai_epi16(src6,4);
src7=_mm_srai_epi16(src7,4); // No idea why they needed to be shifted? ...and it makes the picture all green...*/
__m128i xmm0,xmm1,xmm2,xmm3,xmm4,xmm5,xmm6,xmm7;
movdqa (xmm0, src0);
uint8_t *esi=(uint8_t*)M128_tab_i_04;
movdqa (xmm4, src2);
uint8_t *ecx=(uint8_t*)M128_tab_i_26;
DCT_8_INV_ROW(ecx,esi,xmm0,xmm1,xmm2,xmm3,xmm4,xmm5,xmm6,xmm7);
movdqa (src0, xmm0);
movdqa (src2, xmm4);
movdqa (xmm0, src4);
movdqa (xmm4, src6);
DCT_8_INV_ROW(ecx,esi,xmm0,xmm1,xmm2,xmm3,xmm4,xmm5,xmm6,xmm7);
movdqa (src4, xmm0);
movdqa (src6, xmm4);
movdqa (xmm0, src3);
esi=(uint8_t*)M128_tab_i_35;
movdqa (xmm4, src1);
ecx=(uint8_t*)M128_tab_i_17;
DCT_8_INV_ROW(ecx,esi,xmm0,xmm1,xmm2,xmm3,xmm4,xmm5,xmm6,xmm7);
movdqa (src3, xmm0);
movdqa (src1, xmm4);
movdqa (xmm0, src5);
movdqa (xmm4, src7);
DCT_8_INV_ROW(ecx,esi,xmm0,xmm1,xmm2,xmm3,xmm4,xmm5,xmm6,xmm7);
DCT_8_INV_COL_8(src0,src1,src2,src3,src4,src5,src6,src7,
xmm0,xmm1,xmm2,xmm3,xmm4,xmm5,xmm6,xmm7);
}
void GSDrawScanlineCodeGenerator::TestZ(const Xmm& temp1, const Xmm& temp2)
{
if(!m_sel.zb)
{
return;
}
// int za = fza_base.y + fza_offset->y;
mov(ebp, dword[esi + 4]);
add(ebp, dword[edi + 4]);
// GSVector4i zs = zi;
if(!m_sel.sprite)
{
if(m_sel.zoverflow)
{
// zs = (GSVector4i(z * 0.5f) << 1) | (GSVector4i(z) & GSVector4i::x00000001());
static float half = 0.5f;
movss(temp1, dword[&half]);
shufps(temp1, temp1, _MM_SHUFFLE(0, 0, 0, 0));
mulps(temp1, xmm0);
cvttps2dq(temp1, temp1);
pslld(temp1, 1);
cvttps2dq(xmm0, xmm0);
pcmpeqd(temp2, temp2);
psrld(temp2, 31);
pand(xmm0, temp2);
por(xmm0, temp1);
}
else
{
// zs = GSVector4i(z);
cvttps2dq(xmm0, xmm0);
}
if(m_sel.zwrite)
{
movdqa(xmmword[&m_env.temp.zs], xmm0);
}
}
if(m_sel.ztest)
{
ReadPixel(xmm1, ebp);
if(m_sel.zwrite && m_sel.zpsm < 2)
{
movdqa(xmmword[&m_env.temp.zd], xmm1);
}
// zd &= 0xffffffff >> m_sel.zpsm * 8;
if(m_sel.zpsm)
{
pslld(xmm1, m_sel.zpsm * 8);
psrld(xmm1, m_sel.zpsm * 8);
}
if(m_sel.zoverflow || m_sel.zpsm == 0)
{
// GSVector4i o = GSVector4i::x80000000();
pcmpeqd(xmm4, xmm4);
pslld(xmm4, 31);
// GSVector4i zso = zs - o;
psubd(xmm0, xmm4);
// GSVector4i zdo = zd - o;
psubd(xmm1, xmm4);
}
switch(m_sel.ztst)
{
case ZTST_GEQUAL:
// test |= zso < zdo; // ~(zso >= zdo)
pcmpgtd(xmm1, xmm0);
por(xmm7, 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)
pcmpgtd(xmm0, xmm1);
pcmpeqd(xmm4, xmm4);
pxor(xmm0, xmm4);
por(xmm7, xmm0);
break;
}
alltrue();
}
}
void GPUDrawScanlineCodeGenerator::WriteFrame()
{
// GSVector4i fs = r | g | b | (m_sel.md ? GSVector4i(0x80008000) : m_sel.tme ? a : 0);
pcmpeqd(xmm0, xmm0);
if(m_sel.md || m_sel.tme)
{
movdqa(xmm2, xmm0);
psllw(xmm2, 15);
}
psrlw(xmm0, 11);
psllw(xmm0, 3);
// xmm0 = 0x00f8
// xmm2 = 0x8000 (md)
// GSVector4i r = (c[0] & 0x00f800f8) >> 3;
pand(xmm4, xmm0);
psrlw(xmm4, 3);
// GSVector4i g = (c[1] & 0x00f800f8) << 2;
pand(xmm5, xmm0);
psllw(xmm5, 2);
por(xmm4, xmm5);
// GSVector4i b = (c[2] & 0x00f800f8) << 7;
pand(xmm6, xmm0);
psllw(xmm6, 7);
por(xmm4, xmm6);
if(m_sel.md)
{
// GSVector4i a = GSVector4i(0x80008000);
por(xmm4, xmm2);
}
else if(m_sel.tme)
{
// GSVector4i a = (c[3] << 8) & 0x80008000;
psllw(xmm3, 8);
pand(xmm3, xmm2);
por(xmm4, xmm3);
}
// fs = fs.blend8(fd, test);
movdqa(xmm0, xmm7);
blend8(xmm4, xmm1);
// GSVector4i::store<false>(fb, fs);
// movdqu(ptr[edi], xmm4);
movq(qword[edi], xmm4);
movhps(qword[edi + 8], xmm4);
}
void GSSetupPrimCodeGenerator::Color()
{
if(!m_en.c)
{
return;
}
if(m_env.sel.iip)
{
// GSVector4 c = dscan.c;
movaps(xmm0, xmmword[edx]);
movaps(xmm1, xmm0);
// m_env.d4.c = GSVector4i(c * 4.0f).xzyw().ps32();
movaps(xmm2, xmm0);
mulps(xmm2, xmm3);
cvttps2dq(xmm2, xmm2);
pshufd(xmm2, xmm2, _MM_SHUFFLE(3, 1, 2, 0));
packssdw(xmm2, xmm2);
movdqa(xmmword[&m_env.d4.c], xmm2);
// xmm3 is not needed anymore
// GSVector4 dr = c.xxxx();
// GSVector4 db = c.zzzz();
shufps(xmm0, xmm0, _MM_SHUFFLE(0, 0, 0, 0));
shufps(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
for(int i = 0; i < 4; i++)
{
// GSVector4i r = GSVector4i(dr * m_shift[i]).ps32();
movaps(xmm2, xmm0);
mulps(xmm2, Xmm(4 + i));
cvttps2dq(xmm2, xmm2);
packssdw(xmm2, xmm2);
// GSVector4i b = GSVector4i(db * m_shift[i]).ps32();
movaps(xmm3, xmm1);
mulps(xmm3, Xmm(4 + i));
cvttps2dq(xmm3, xmm3);
packssdw(xmm3, xmm3);
// m_env.d[i].rb = r.upl16(b);
punpcklwd(xmm2, xmm3);
movdqa(xmmword[&m_env.d[i].rb], xmm2);
}
// GSVector4 c = dscan.c;
movaps(xmm0, xmmword[edx]); // not enough regs, have to reload it
movaps(xmm1, xmm0);
// GSVector4 dg = c.yyyy();
// GSVector4 da = c.wwww();
shufps(xmm0, xmm0, _MM_SHUFFLE(1, 1, 1, 1));
shufps(xmm1, xmm1, _MM_SHUFFLE(3, 3, 3, 3));
for(int i = 0; i < 4; i++)
{
// GSVector4i g = GSVector4i(dg * m_shift[i]).ps32();
movaps(xmm2, xmm0);
mulps(xmm2, Xmm(4 + i));
cvttps2dq(xmm2, xmm2);
packssdw(xmm2, xmm2);
// GSVector4i a = GSVector4i(da * m_shift[i]).ps32();
movaps(xmm3, xmm1);
mulps(xmm3, Xmm(4 + i));
cvttps2dq(xmm3, xmm3);
packssdw(xmm3, xmm3);
// m_env.d[i].ga = g.upl16(a);
punpcklwd(xmm2, xmm3);
movdqa(xmmword[&m_env.d[i].ga], xmm2);
}
}
else
{
// GSVector4i c = GSVector4i(vertices[0].c);
movaps(xmm0, xmmword[ecx]);
cvttps2dq(xmm0, xmm0);
// c = c.upl16(c.zwxy());
movdqa(xmm1, xmm0);
pshufd(xmm1, xmm1, _MM_SHUFFLE(1, 0, 3, 2));
punpcklwd(xmm0, xmm1);
// if(!tme) c = c.srl16(7);
if(m_env.sel.tfx == TFX_NONE)
{
psrlw(xmm0, 7);
}
// m_env.c.rb = c.xxxx();
// m_env.c.ga = c.zzzz();
movdqa(xmm1, xmm0);
pshufd(xmm0, xmm0, _MM_SHUFFLE(0, 0, 0, 0));
pshufd(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
movdqa(xmmword[&m_env.c.rb], xmm0);
movdqa(xmmword[&m_env.c.ga], xmm1);
}
}
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;
movdqa(xmm0, xmm2);
movdqa(xmm1, xmm2);
psllw(xmm0, 8);
psrlw(xmm0, 8);
psrlw(xmm1, 8);
break;
case 2:
// c[2] = ((fd & 0x7c00) << 9) | ((fd & 0x001f) << 3);
// c[3] = ((fd & 0x8000) << 8) | ((fd & 0x03e0) >> 2);
movdqa(xmm0, xmm2);
movdqa(xmm1, xmm2);
movdqa(xmm4, xmm2);
pcmpeqd(xmm7, xmm7);
psrld(xmm7, 27); // 0x0000001f
pand(xmm0, xmm7);
pslld(xmm0, 3);
pslld(xmm7, 10); // 0x00007c00
pand(xmm4, xmm7);
pslld(xmm4, 9);
por(xmm0, xmm4);
movdqa(xmm4, xmm1);
psrld(xmm7, 5); // 0x000003e0
pand(xmm1, xmm7);
psrld(xmm1, 2);
psllw(xmm7, 10); // 0x00008000
pand(xmm4, xmm7);
pslld(xmm4, 8);
por(xmm1, xmm4);
break;
}
}
// xmm5, xmm6 = src rb, ga
// xmm0, xmm1 = dst rb, ga
// xmm2, xmm3 = used
// xmm4, xmm7 = free
if(m_sel.pabe || (m_sel.aba != m_sel.abb) && (m_sel.abb == 0 || m_sel.abd == 0))
{
movdqa(xmm4, xmm5);
}
if(m_sel.aba != m_sel.abb)
{
// rb = c[aba * 2 + 0];
switch(m_sel.aba)
{
case 0:
break;
case 1:
movdqa(xmm5, xmm0);
break;
case 2:
pxor(xmm5, xmm5);
break;
}
// rb = rb.sub16(c[abb * 2 + 0]);
switch(m_sel.abb)
{
case 0:
psubw(xmm5, xmm4);
break;
case 1:
psubw(xmm5, 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_env.afix;
switch(m_sel.abc)
{
case 0:
case 1:
movdqa(xmm7, m_sel.abc ? xmm1 : xmm6);
pshuflw(xmm7, xmm7, _MM_SHUFFLE(3, 3, 1, 1));
pshufhw(xmm7, xmm7, _MM_SHUFFLE(3, 3, 1, 1));
psllw(xmm7, 7);
break;
case 2:
movdqa(xmm7, xmmword[&m_env.afix]);
break;
}
// rb = rb.modulate16<1>(a);
modulate16<1>(xmm5, xmm7);
}
// rb = rb.add16(c[abd * 2 + 0]);
switch(m_sel.abd)
{
case 0:
paddw(xmm5, xmm4);
break;
case 1:
paddw(xmm5, xmm0);
break;
case 2:
break;
}
}
else
{
// rb = c[abd * 2 + 0];
switch(m_sel.abd)
{
case 0:
break;
case 1:
movdqa(xmm5, xmm0);
break;
case 2:
pxor(xmm5, xmm5);
break;
}
}
if(m_sel.pabe)
{
// mask = (c[1] << 8).sra32(31);
movdqa(xmm0, xmm6);
pslld(xmm0, 8);
psrad(xmm0, 31);
// rb = c[0].blend8(rb, mask);
blend8r(xmm5, xmm4);
}
// xmm6 = src ga
// xmm1 = dst ga
// xmm5 = rb
// xmm7 = a
// xmm2, xmm3 = used
// xmm0, xmm4 = free
movdqa(xmm4, xmm6);
if(m_sel.aba != m_sel.abb)
{
// ga = c[aba * 2 + 1];
switch(m_sel.aba)
{
case 0:
break;
case 1:
movdqa(xmm6, xmm1);
break;
case 2:
pxor(xmm6, xmm6);
break;
}
// ga = ga.sub16(c[abeb * 2 + 1]);
switch(m_sel.abb)
{
case 0:
psubw(xmm6, xmm4);
break;
case 1:
psubw(xmm6, xmm1);
break;
case 2:
break;
}
if(!(m_sel.fpsm == 1 && m_sel.abc == 1))
{
// ga = ga.modulate16<1>(a);
modulate16<1>(xmm6, xmm7);
}
// ga = ga.add16(c[abd * 2 + 1]);
switch(m_sel.abd)
{
case 0:
paddw(xmm6, xmm4);
break;
case 1:
paddw(xmm6, xmm1);
break;
case 2:
break;
}
}
else
{
// ga = c[abd * 2 + 1];
switch(m_sel.abd)
{
case 0:
break;
case 1:
movdqa(xmm6, xmm1);
break;
case 2:
pxor(xmm6, xmm6);
break;
}
}
// xmm4 = src ga
// xmm5 = rb
// xmm6 = ga
// xmm2, xmm3 = used
// xmm0, xmm1, xmm7 = free
if(m_sel.pabe)
{
if(!m_cpu.has(util::Cpu::tSSE41))
{
// doh, previous blend8r overwrote xmm0 (sse41 uses pblendvb)
movdqa(xmm0, xmm4);
pslld(xmm0, 8);
psrad(xmm0, 31);
}
psrld(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]);
blend8r(xmm6, xmm4);
}
else
{
if(m_sel.fpsm != 1) // TODO: fm == 0xffxxxxxx
{
mix16(xmm6, xmm4, xmm7);
}
}
}
void GPUDrawScanlineCodeGenerator::Generate()
{
push(esi);
push(edi);
Init();
align(16);
L("loop");
// GSVector4i test = m_test[7 + (steps & (steps >> 31))];
mov(edx, ecx);
sar(edx, 31);
and(edx, ecx);
shl(edx, 4);
movdqa(xmm7, ptr[edx + (size_t)&m_test[7]]);
// movdqu(xmm1, ptr[edi]);
movq(xmm1, qword[edi]);
movhps(xmm1, qword[edi + 8]);
// ecx = steps
// esi = tex (tme)
// edi = fb
// xmm1 = fd
// xmm2 = s
// xmm3 = t
// xmm4 = r
// xmm5 = g
// xmm6 = b
// xmm7 = test
TestMask();
SampleTexture();
// xmm1 = fd
// xmm3 = a
// xmm4 = r
// xmm5 = g
// xmm6 = b
// xmm7 = test
// xmm0, xmm2 = free
ColorTFX();
AlphaBlend();
Dither();
WriteFrame();
L("step");
// if(steps <= 0) break;
test(ecx, ecx);
jle("exit", T_NEAR);
Step();
jmp("loop", T_NEAR);
L("exit");
pop(edi);
pop(esi);
ret(8);
}
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();
pcmpeqd(xmm1, xmm1);
break;
case ATST_ALWAYS:
return;
case ATST_LESS:
case ATST_LEQUAL:
// t = (ga >> 16) > m_env.aref;
movdqa(xmm1, xmm6);
psrld(xmm1, 16);
pcmpgtd(xmm1, xmmword[&m_env.aref]);
break;
case ATST_EQUAL:
// t = (ga >> 16) != m_env.aref;
movdqa(xmm1, xmm6);
psrld(xmm1, 16);
pcmpeqd(xmm1, xmmword[&m_env.aref]);
pcmpeqd(xmm0, xmm0);
pxor(xmm1, xmm0);
break;
case ATST_GEQUAL:
case ATST_GREATER:
// t = (ga >> 16) < m_env.aref;
movdqa(xmm0, xmm6);
psrld(xmm0, 16);
movdqa(xmm1, xmmword[&m_env.aref]);
pcmpgtd(xmm1, xmm0);
break;
case ATST_NOTEQUAL:
// t = (ga >> 16) == m_env.aref;
movdqa(xmm1, xmm6);
psrld(xmm1, 16);
pcmpeqd(xmm1, xmmword[&m_env.aref]);
break;
}
switch(m_sel.afail)
{
case AFAIL_KEEP:
// test |= t;
por(xmm7, xmm1);
alltrue();
break;
case AFAIL_FB_ONLY:
// zm |= t;
por(xmm4, xmm1);
break;
case AFAIL_ZB_ONLY:
// fm |= t;
por(xmm3, xmm1);
break;
case AFAIL_RGB_ONLY:
// zm |= t;
por(xmm4, xmm1);
// fm |= t & GSVector4i::xff000000();
psrld(xmm1, 24);
pslld(xmm1, 24);
por(xmm3, xmm1);
break;
}
}
void GSDrawScanlineCodeGenerator::ColorTFX()
{
if(!m_sel.fwrite)
{
return;
}
switch(m_sel.tfx)
{
case TFX_MODULATE:
// GSVector4i rb = iip ? rbf : m_env.c.rb;
// rbt = rbt.modulate16<1>(rb).clamp8();
modulate16<1>(xmm5, xmmword[m_sel.iip ? &m_env.temp.rb : &m_env.c.rb]);
clamp16(xmm5, xmm1);
break;
case TFX_DECAL:
break;
case TFX_HIGHLIGHT:
case TFX_HIGHLIGHT2:
if(m_sel.tfx == TFX_HIGHLIGHT2 && m_sel.tcc)
{
// GSVector4i ga = iip ? gaf : m_env.c.ga;
movdqa(xmm2, xmmword[m_sel.iip ? &m_env.temp.ga : &m_env.c.ga]);
}
// gat = gat.modulate16<1>(ga).add16(af).clamp8().mix16(gat);
movdqa(xmm1, xmm6);
modulate16<1>(xmm6, xmm2);
pshuflw(xmm2, xmm2, _MM_SHUFFLE(3, 3, 1, 1));
pshufhw(xmm2, xmm2, _MM_SHUFFLE(3, 3, 1, 1));
psrlw(xmm2, 7);
paddw(xmm6, xmm2);
clamp16(xmm6, xmm0);
mix16(xmm6, xmm1, xmm0);
// GSVector4i rb = iip ? rbf : m_env.c.rb;
// rbt = rbt.modulate16<1>(rb).add16(af).clamp8();
modulate16<1>(xmm5, xmmword[m_sel.iip ? &m_env.temp.rb : &m_env.c.rb]);
paddw(xmm5, xmm2);
clamp16(xmm5, xmm0);
break;
case TFX_NONE:
// rbt = iip ? rb.srl16(7) : rb;
if(m_sel.iip)
{
psrlw(xmm5, 7);
}
break;
}
}
void GSDrawScanlineCodeGenerator::AlphaTFX()
{
if(!m_sel.fb)
{
return;
}
switch(m_sel.tfx)
{
case TFX_MODULATE:
// GSVector4i ga = iip ? gaf : m_env.c.ga;
movdqa(xmm4, xmmword[m_sel.iip ? &m_env.temp.ga : &m_env.c.ga]);
// gat = gat.modulate16<1>(ga).clamp8();
modulate16<1>(xmm6, xmm4);
clamp16(xmm6, xmm3);
// if(!tcc) gat = gat.mix16(ga.srl16(7));
if(!m_sel.tcc)
{
psrlw(xmm4, 7);
mix16(xmm6, xmm4, xmm3);
}
break;
case TFX_DECAL:
// if(!tcc) gat = gat.mix16(ga.srl16(7));
if(!m_sel.tcc)
{
// GSVector4i ga = iip ? gaf : m_env.c.ga;
movdqa(xmm4, xmmword[m_sel.iip ? &m_env.temp.ga : &m_env.c.ga]);
psrlw(xmm4, 7);
mix16(xmm6, xmm4, xmm3);
}
break;
case TFX_HIGHLIGHT:
// GSVector4i ga = iip ? gaf : m_env.c.ga;
movdqa(xmm4, xmmword[m_sel.iip ? &m_env.temp.ga : &m_env.c.ga]);
movdqa(xmm2, xmm4);
// gat = gat.mix16(!tcc ? ga.srl16(7) : gat.addus8(ga.srl16(7)));
psrlw(xmm4, 7);
if(m_sel.tcc)
{
paddusb(xmm4, xmm6);
}
mix16(xmm6, xmm4, xmm3);
break;
case TFX_HIGHLIGHT2:
// if(!tcc) gat = gat.mix16(ga.srl16(7));
if(!m_sel.tcc)
{
// GSVector4i ga = iip ? gaf : m_env.c.ga;
movdqa(xmm4, xmmword[m_sel.iip ? &m_env.temp.ga : &m_env.c.ga]);
movdqa(xmm2, xmm4);
psrlw(xmm4, 7);
mix16(xmm6, xmm4, xmm3);
}
break;
case TFX_NONE:
// gat = iip ? ga.srl16(7) : ga;
if(m_sel.iip)
{
psrlw(xmm6, 7);
}
break;
}
if(m_sel.aa1)
{
// gs_user figure 3-2: anti-aliasing after tfx, before tests, modifies alpha
// FIXME: bios config screen cubes
if(!m_sel.abe)
{
// a = cov
if(m_sel.edge)
{
movdqa(xmm0, xmmword[&m_env.temp.cov]);
}
else
{
pcmpeqd(xmm0, xmm0);
psllw(xmm0, 15);
psrlw(xmm0, 8);
}
mix16(xmm6, xmm0, xmm1);
}
else
{
// a = a == 0x80 ? cov : a
pcmpeqd(xmm0, xmm0);
psllw(xmm0, 15);
psrlw(xmm0, 8);
if(m_sel.edge)
{
movdqa(xmm1, xmmword[&m_env.temp.cov]);
}
else
{
movdqa(xmm1, xmm0);
}
pcmpeqw(xmm0, xmm6);
psrld(xmm0, 16);
pslld(xmm0, 16);
blend8(xmm6, xmm1);
}
}
}
void GSDrawScanlineCodeGenerator::Wrap(const Xmm& uv0, const Xmm& uv1)
{
// xmm0, xmm1, xmm4, xmm5, xmm6 = 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)
{
movdqa(xmm4, xmmword[&m_env.t.min]);
pmaxsw(uv0, xmm4);
pmaxsw(uv1, xmm4);
}
else
{
pxor(xmm0, xmm0);
pmaxsw(uv0, xmm0);
pmaxsw(uv1, xmm0);
}
movdqa(xmm5, xmmword[&m_env.t.max]);
pminsw(uv0, xmm5);
pminsw(uv1, xmm5);
}
else
{
movdqa(xmm4, xmmword[&m_env.t.min]);
pand(uv0, xmm4);
pand(uv1, xmm4);
if(region)
{
movdqa(xmm5, xmmword[&m_env.t.max]);
por(uv0, xmm5);
por(uv1, xmm5);
}
}
}
else
{
movdqa(xmm1, uv0);
movdqa(xmm6, uv1);
movdqa(xmm4, xmmword[&m_env.t.min]);
movdqa(xmm5, xmmword[&m_env.t.max]);
// GSVector4i clamp = t.sat_i16(m_env.t.min, m_env.t.max);
pmaxsw(uv0, xmm4);
pmaxsw(uv1, xmm4);
pminsw(uv0, xmm5);
pminsw(uv1, xmm5);
// GSVector4i repeat = (t & m_env.t.min) | m_env.t.max;
pand(xmm1, xmm4);
pand(xmm6, xmm4);
if(region)
{
por(xmm1, xmm5);
por(xmm6, xmm5);
}
// clamp.blend8(repeat, m_env.t.mask);
if(m_cpu.has(util::Cpu::tSSE41))
{
movdqa(xmm0, xmmword[&m_env.t.mask]);
pblendvb(uv0, xmm1);
pblendvb(uv1, xmm6);
}
else
{
movdqa(xmm0, xmmword[&m_env.t.invmask]);
movdqa(xmm4, xmm0);
pand(uv0, xmm0);
pandn(xmm0, xmm1);
por(uv0, xmm0);
pand(uv1, xmm4);
pandn(xmm4, xmm6);
por(uv1, xmm4);
}
}
}
void GSDrawScanlineCodeGenerator::Wrap(const Xmm& uv)
{
// xmm0, xmm1, xmm4, xmm5, xmm6 = 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)
{
pmaxsw(uv, xmmword[&m_env.t.min]);
}
else
{
pxor(xmm0, xmm0);
pmaxsw(uv, xmm0);
}
pminsw(uv, xmmword[&m_env.t.max]);
}
else
{
pand(uv, xmmword[&m_env.t.min]);
if(region)
{
por(uv, xmmword[&m_env.t.max]);
}
}
}
else
{
movdqa(xmm1, uv);
movdqa(xmm4, xmmword[&m_env.t.min]);
movdqa(xmm5, xmmword[&m_env.t.max]);
// GSVector4i clamp = t.sat_i16(m_env.t.min, m_env.t.max);
pmaxsw(uv, xmm4);
pminsw(uv, xmm5);
// GSVector4i repeat = (t & m_env.t.min) | m_env.t.max;
pand(xmm1, xmm4);
if(region)
{
por(xmm1, xmm5);
}
// clamp.blend8(repeat, m_env.t.mask);
movdqa(xmm0, xmmword[&m_env.t.mask]);
blend8(uv, xmm1);
}
}
void GPUDrawScanlineCodeGenerator::Step()
{
// steps -= 8;
sub(ecx, 8);
// fb += 8;
add(edi, 8 * sizeof(uint16));
if(m_sel.tme)
{
// GSVector4i st = m_local.d8.st;
movdqa(xmm4, ptr[&m_local.d8.st]);
// s = s.add16(st.xxxx());
// t = t.add16(st.yyyy());
pshufd(xmm2, xmm4, _MM_SHUFFLE(0, 0, 0, 0));
paddw(xmm2, ptr[&m_local.temp.s]);
movdqa(ptr[&m_local.temp.s], xmm2);
// TODO: if(!sprite) ... else reload t
pshufd(xmm3, xmm4, _MM_SHUFFLE(1, 1, 1, 1));
paddw(xmm3, ptr[&m_local.temp.t]);
movdqa(ptr[&m_local.temp.t], xmm3);
}
if(m_sel.tfx != 3) // != decal
{
if(m_sel.iip)
{
// GSVector4i c = m_local.d8.c;
// r = r.add16(c.xxxx());
// g = g.add16(c.yyyy());
// b = b.add16(c.zzzz());
movdqa(xmm6, ptr[&m_local.d8.c]);
pshufd(xmm4, xmm6, _MM_SHUFFLE(0, 0, 0, 0));
pshufd(xmm5, xmm6, _MM_SHUFFLE(1, 1, 1, 1));
pshufd(xmm6, xmm6, _MM_SHUFFLE(2, 2, 2, 2));
paddw(xmm4, ptr[&m_local.temp.r]);
paddw(xmm5, ptr[&m_local.temp.g]);
paddw(xmm6, ptr[&m_local.temp.b]);
movdqa(ptr[&m_local.temp.r], xmm4);
movdqa(ptr[&m_local.temp.g], xmm5);
movdqa(ptr[&m_local.temp.b], xmm6);
}
else
{
movdqa(xmm4, ptr[&m_local.temp.r]);
movdqa(xmm5, ptr[&m_local.temp.g]);
movdqa(xmm6, ptr[&m_local.temp.b]);
}
}
}
void GSDrawScanlineCodeGenerator::WriteFrame(int params)
{
const int _top = params + 4;
if(!m_sel.fwrite)
{
return;
}
if(m_sel.colclamp == 0)
{
// c[0] &= 0x000000ff;
// c[1] &= 0x000000ff;
pcmpeqd(xmm7, xmm7);
psrlw(xmm7, 8);
pand(xmm5, xmm7);
pand(xmm6, xmm7);
}
if(m_sel.fpsm == 2 && m_sel.dthe)
{
mov(eax, dword[esp + _top]);
and(eax, 3);
shl(eax, 5);
paddw(xmm5, xmmword[eax + (size_t)&m_env.dimx[0]]);
paddw(xmm6, xmmword[eax + (size_t)&m_env.dimx[1]]);
}
// GSVector4i fs = c[0].upl16(c[1]).pu16(c[0].uph16(c[1]));
movdqa(xmm7, xmm5);
punpcklwd(xmm5, xmm6);
punpckhwd(xmm7, xmm6);
packuswb(xmm5, xmm7);
if(m_sel.fba && m_sel.fpsm != 1)
{
// fs |= 0x80000000;
pcmpeqd(xmm7, xmm7);
pslld(xmm7, 31);
por(xmm5, xmm7);
}
if(m_sel.fpsm == 2)
{
// GSVector4i rb = fs & 0x00f800f8;
// GSVector4i ga = fs & 0x8000f800;
mov(eax, 0x00f800f8);
movd(xmm6, eax);
pshufd(xmm6, xmm6, _MM_SHUFFLE(0, 0, 0, 0));
mov(eax, 0x8000f800);
movd(xmm7, eax);
pshufd(xmm7, xmm7, _MM_SHUFFLE(0, 0, 0, 0));
movdqa(xmm4, xmm5);
pand(xmm4, xmm6);
pand(xmm5, xmm7);
// fs = (ga >> 16) | (rb >> 9) | (ga >> 6) | (rb >> 3);
movdqa(xmm6, xmm4);
movdqa(xmm7, xmm5);
psrld(xmm4, 3);
psrld(xmm6, 9);
psrld(xmm5, 6);
psrld(xmm7, 16);
por(xmm5, xmm4);
por(xmm7, xmm6);
por(xmm5, xmm7);
}
if(m_sel.rfb)
{
// fs = fs.blend(fd, fm);
blend(xmm5, xmm2, xmm3); // TODO: could be skipped in certain cases, depending on fpsm and fm
}
bool fast = m_sel.rfb && m_sel.fpsm < 2;
WritePixel(xmm5, xmm0, ebx, dl, fast, m_sel.fpsm);
}
void GPUDrawScanlineCodeGenerator::SampleTexture()
{
if(!m_sel.tme)
{
return;
}
if(m_sel.tlu)
{
mov(edx, ptr[&m_local.gd->clut]);
}
// xmm2 = s
// xmm3 = t
// xmm7 = test
// xmm0, xmm4, xmm5, xmm6 = free
// xmm1 = used
if(m_sel.ltf)
{
// GSVector4i u = s.sub16(GSVector4i(0x00200020)); // - 0.125f
// GSVector4i v = t.sub16(GSVector4i(0x00200020)); // - 0.125f
mov(eax, 0x00200020);
movd(xmm0, eax);
pshufd(xmm0, xmm0, _MM_SHUFFLE(0, 0, 0, 0));
psubw(xmm2, xmm0);
psubw(xmm3, xmm0);
// GSVector4i uf = (u & GSVector4i::x00ff()) << 7;
// GSVector4i vf = (v & GSVector4i::x00ff()) << 7;
movdqa(xmm0, xmm2);
psllw(xmm0, 8);
psrlw(xmm0, 1);
movdqa(ptr[&m_local.temp.uf], xmm0);
if(!m_sel.sprite)
{
movdqa(xmm0, xmm3);
psllw(xmm0, 8);
psrlw(xmm0, 1);
movdqa(ptr[&m_local.temp.vf], xmm0);
}
}
// GSVector4i u0 = s.srl16(8);
// GSVector4i v0 = t.srl16(8);
psrlw(xmm2, 8);
psrlw(xmm3, 8);
// xmm2 = u
// xmm3 = v
// xmm7 = test
// xmm0, xmm4, xmm5, xmm6 = free
// xmm1 = used
if(m_sel.ltf)
{
// GSVector4i u1 = u0.add16(GSVector4i::x0001());
// GSVector4i v1 = v0.add16(GSVector4i::x0001());
movdqa(xmm4, xmm2);
movdqa(xmm5, xmm3);
pcmpeqd(xmm0, xmm0);
psrlw(xmm0, 15);
paddw(xmm4, xmm0);
paddw(xmm5, xmm0);
if(m_sel.twin)
{
// u0 = (u0 & m_local.twin[0].u).add16(m_local.twin[1].u);
// v0 = (v0 & m_local.twin[0].v).add16(m_local.twin[1].v);
// u1 = (u1 & m_local.twin[0].u).add16(m_local.twin[1].u);
// v1 = (v1 & m_local.twin[0].v).add16(m_local.twin[1].v);
movdqa(xmm0, ptr[&m_local.twin[0].u]);
movdqa(xmm6, ptr[&m_local.twin[1].u]);
pand(xmm2, xmm0);
paddw(xmm2, xmm6);
pand(xmm4, xmm0);
paddw(xmm4, xmm6);
movdqa(xmm0, ptr[&m_local.twin[0].v]);
movdqa(xmm6, ptr[&m_local.twin[1].v]);
pand(xmm3, xmm0);
paddw(xmm3, xmm6);
pand(xmm5, xmm0);
paddw(xmm5, xmm6);
}
else
{
// u0 = u0.min_i16(m_local.twin[2].u);
// v0 = v0.min_i16(m_local.twin[2].v);
// u1 = u1.min_i16(m_local.twin[2].u);
// v1 = v1.min_i16(m_local.twin[2].v);
// TODO: if(!sprite) clamp16 else:
movdqa(xmm0, ptr[&m_local.twin[2].u]);
movdqa(xmm6, ptr[&m_local.twin[2].v]);
pminsw(xmm2, xmm0);
pminsw(xmm3, xmm6);
pminsw(xmm4, xmm0);
pminsw(xmm5, xmm6);
}
// xmm2 = u0
// xmm3 = v0
// xmm4 = u1
// xmm5 = v1
// xmm7 = test
// xmm0, xmm6 = free
// xmm1 = used
// GSVector4i addr00 = v0.sll16(8) | u0;
// GSVector4i addr01 = v0.sll16(8) | u1;
// GSVector4i addr10 = v1.sll16(8) | u0;
// GSVector4i addr11 = v1.sll16(8) | u1;
psllw(xmm3, 8);
movdqa(xmm0, xmm3);
por(xmm3, xmm2);
por(xmm0, xmm4);
psllw(xmm5, 8);
movdqa(xmm6, xmm5);
por(xmm5, xmm2);
por(xmm6, xmm4);
// xmm3 = addr00
// xmm0 = addr01
// xmm5 = addr10
// xmm6 = addr11
// xmm7 = test
// xmm2, xmm4 = free
// xmm1 = used
ReadTexel(xmm2, xmm3);
ReadTexel(xmm4, xmm0);
ReadTexel(xmm3, xmm5);
ReadTexel(xmm5, xmm6);
// xmm2 = c00
// xmm4 = c01
// xmm3 = c10
// xmm5 = c11
// xmm7 = test
// xmm0, xmm6 = free
// xmm1 = used
// spill (TODO)
movdqa(ptr[&m_local.temp.fd], xmm1);
movdqa(ptr[&m_local.temp.test], xmm7);
// xmm2 = c00
// xmm4 = c01
// xmm3 = c10
// xmm5 = c11
// xmm0, xmm1, xmm6, xmm7 = free
movdqa(xmm1, xmm2);
psllw(xmm1, 11);
psrlw(xmm1, 8);
movdqa(xmm0, xmm4);
psllw(xmm0, 11);
psrlw(xmm0, 8);
lerp16<0>(xmm0, xmm1, ptr[&m_local.temp.uf]);
movdqa(xmm6, xmm2);
psllw(xmm6, 6);
psrlw(xmm6, 11);
psllw(xmm6, 3);
movdqa(xmm1, xmm4);
psllw(xmm1, 6);
psrlw(xmm1, 11);
psllw(xmm1, 3);
lerp16<0>(xmm1, xmm6, ptr[&m_local.temp.uf]);
movdqa(xmm7, xmm2);
psllw(xmm7, 1);
psrlw(xmm7, 11);
psllw(xmm7, 3);
movdqa(xmm6, xmm4);
psllw(xmm6, 1);
psrlw(xmm6, 11);
psllw(xmm6, 3);
lerp16<0>(xmm6, xmm7, ptr[&m_local.temp.uf]);
psraw(xmm2, 15);
psrlw(xmm2, 8);
psraw(xmm4, 15);
psrlw(xmm4, 8);
lerp16<0>(xmm4, xmm2, ptr[&m_local.temp.uf]);
// xmm0 = r00
// xmm1 = g00
// xmm6 = b00
// xmm4 = a00
// xmm3 = c10
// xmm5 = c11
// xmm2, xmm7 = free
movdqa(xmm7, xmm3);
psllw(xmm7, 11);
psrlw(xmm7, 8);
movdqa(xmm2, xmm5);
psllw(xmm2, 11);
psrlw(xmm2, 8);
lerp16<0>(xmm2, xmm7, ptr[&m_local.temp.uf]);
lerp16<0>(xmm2, xmm0, ptr[&m_local.temp.vf]);
// xmm2 = r
// xmm1 = g00
// xmm6 = b00
// xmm4 = a00
// xmm3 = c10
// xmm5 = c11
// xmm0, xmm7 = free
movdqa(xmm7, xmm3);
psllw(xmm7, 6);
psrlw(xmm7, 11);
psllw(xmm7, 3);
movdqa(xmm0, xmm5);
psllw(xmm0, 6);
psrlw(xmm0, 11);
psllw(xmm0, 3);
lerp16<0>(xmm0, xmm7, ptr[&m_local.temp.uf]);
lerp16<0>(xmm0, xmm1, ptr[&m_local.temp.vf]);
// xmm2 = r
// xmm0 = g
// xmm6 = b00
// xmm4 = a00
// xmm3 = c10
// xmm5 = c11
// xmm1, xmm7 = free
movdqa(xmm7, xmm3);
psllw(xmm7, 1);
psrlw(xmm7, 11);
psllw(xmm7, 3);
movdqa(xmm1, xmm5);
psllw(xmm1, 1);
psrlw(xmm1, 11);
psllw(xmm1, 3);
lerp16<0>(xmm1, xmm7, ptr[&m_local.temp.uf]);
lerp16<0>(xmm1, xmm6, ptr[&m_local.temp.vf]);
// xmm2 = r
// xmm0 = g
// xmm1 = b
// xmm4 = a00
// xmm3 = c10
// xmm5 = c11
// xmm6, xmm7 = free
psraw(xmm3, 15);
psrlw(xmm3, 8);
psraw(xmm5, 15);
psrlw(xmm5, 8);
lerp16<0>(xmm5, xmm3, ptr[&m_local.temp.uf]);
lerp16<0>(xmm5, xmm4, ptr[&m_local.temp.vf]);
// xmm2 = r
// xmm0 = g
// xmm1 = b
// xmm5 = a
// xmm3, xmm4, xmm6, xmm7 = free
// TODO
movdqa(xmm3, xmm5); // a
movdqa(xmm4, xmm2); // r
movdqa(xmm6, xmm1); // b
movdqa(xmm5, xmm0); // g
// reload test
movdqa(xmm7, ptr[&m_local.temp.test]);
// xmm4 = r
// xmm5 = g
// xmm6 = b
// xmm3 = a
// xmm7 = test
// xmm0, xmm1, xmm2 = free
// test |= (c[0] | c[1] | c[2] | c[3]).eq16(GSVector4i::zero()); // mask out blank pixels (not perfect)
movdqa(xmm1, xmm3);
por(xmm1, xmm4);
movdqa(xmm2, xmm5);
por(xmm2, xmm6);
por(xmm1, xmm2);
pxor(xmm0, xmm0);
pcmpeqw(xmm1, xmm0);
por(xmm7, xmm1);
// a = a.gt16(GSVector4i::zero());
pcmpgtw(xmm3, xmm0);
// reload fd
movdqa(xmm1, ptr[&m_local.temp.fd]);
}
else
{
if(m_sel.twin)
{
// u = (u & m_local.twin[0].u).add16(m_local.twin[1].u);
// v = (v & m_local.twin[0].v).add16(m_local.twin[1].v);
pand(xmm2, ptr[&m_local.twin[0].u]);
paddw(xmm2, ptr[&m_local.twin[1].u]);
pand(xmm3, ptr[&m_local.twin[0].v]);
paddw(xmm3, ptr[&m_local.twin[1].v]);
}
else
{
// u = u.min_i16(m_local.twin[2].u);
// v = v.min_i16(m_local.twin[2].v);
// TODO: if(!sprite) clamp16 else:
pminsw(xmm2, ptr[&m_local.twin[2].u]);
pminsw(xmm3, ptr[&m_local.twin[2].v]);
}
// xmm2 = u
// xmm3 = v
// xmm7 = test
// xmm0, xmm4, xmm5, xmm6 = free
// xmm1 = used
// GSVector4i addr = v.sll16(8) | u;
psllw(xmm3, 8);
por(xmm3, xmm2);
// xmm3 = addr
// xmm7 = test
// xmm0, xmm2, xmm4, xmm5, xmm6 = free
// xmm1 = used
ReadTexel(xmm6, xmm3);
// xmm3 = c00
// xmm7 = test
// xmm0, xmm2, xmm4, xmm5, xmm6 = free
// xmm1 = used
// test |= c00.eq16(GSVector4i::zero()); // mask out blank pixels
pxor(xmm0, xmm0);
pcmpeqw(xmm0, xmm6);
por(xmm7, xmm0);
// c[0] = (c00 << 3) & 0x00f800f8;
// c[1] = (c00 >> 2) & 0x00f800f8;
// c[2] = (c00 >> 7) & 0x00f800f8;
// c[3] = c00.sra16(15);
movdqa(xmm3, xmm6);
psraw(xmm3, 15); // a
pcmpeqd(xmm0, xmm0);
psrlw(xmm0, 11);
psllw(xmm0, 3); // 0x00f8
movdqa(xmm4, xmm6);
psllw(xmm4, 3);
pand(xmm4, xmm0); // r
movdqa(xmm5, xmm6);
psrlw(xmm5, 2);
pand(xmm5, xmm0); // g
psrlw(xmm6, 7);
pand(xmm6, xmm0); // b
}
}
void GSDrawScanlineCodeGenerator::Init(int params)
{
const int _top = params + 4;
const int _v = params + 8;
// int skip = left & 3;
mov(ebx, edx);
and(edx, 3);
// left -= skip;
sub(ebx, edx);
// int steps = right - left - 4;
sub(ecx, ebx);
sub(ecx, 4);
// GSVector4i test = m_test[skip] | m_test[7 + (steps & (steps >> 31))];
shl(edx, 4);
movdqa(xmm7, xmmword[edx + (size_t)&m_test[0]]);
mov(eax, ecx);
sar(eax, 31);
and(eax, ecx);
shl(eax, 4);
por(xmm7, xmmword[eax + (size_t)&m_test[7]]);
// GSVector2i* fza_base = &m_env.fzbr[top];
mov(esi, dword[esp + _top]);
lea(esi, ptr[esi * 8]);
add(esi, dword[&m_env.fzbr]);
// GSVector2i* fza_offset = &m_env.fzbc[left >> 2];
lea(edi, ptr[ebx * 2]);
add(edi, dword[&m_env.fzbc]);
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_env.d[skip]
shl(edx, 4);
lea(edx, ptr[edx + (size_t)m_env.d]);
// ebx = &v
mov(ebx, dword[esp + _v]);
}
if(!m_sel.sprite)
{
if(m_sel.fwrite && m_sel.fge || m_sel.zb)
{
movaps(xmm0, xmmword[ebx + 16]); // v.p
if(m_sel.fwrite && m_sel.fge)
{
// f = GSVector4i(vp).zzzzh().zzzz().add16(m_env.d[skip].f);
cvttps2dq(xmm1, xmm0);
pshufhw(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
pshufd(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
paddw(xmm1, xmmword[edx + 16 * 6]);
movdqa(xmmword[&m_env.temp.f], xmm1);
}
if(m_sel.zb)
{
// z = vp.zzzz() + m_env.d[skip].z;
shufps(xmm0, xmm0, _MM_SHUFFLE(2, 2, 2, 2));
addps(xmm0, xmmword[edx]);
movaps(xmmword[&m_env.temp.z], xmm0);
}
}
}
else
{
if(m_sel.ztest)
{
movdqa(xmm0, xmmword[&m_env.p.z]);
}
}
if(m_sel.fb)
{
if(m_sel.edge || m_sel.tfx != TFX_NONE)
{
movaps(xmm4, xmmword[ebx + 32]); // v.t
}
if(m_sel.edge)
{
pshufhw(xmm3, xmm4, _MM_SHUFFLE(2, 2, 2, 2));
pshufd(xmm3, xmm3, _MM_SHUFFLE(3, 3, 3, 3));
psrlw(xmm3, 9);
movdqa(xmmword[&m_env.temp.cov], xmm3);
}
if(m_sel.tfx != TFX_NONE)
{
if(m_sel.fst)
{
// GSVector4i vti(vt);
cvttps2dq(xmm4, xmm4);
// si = vti.xxxx() + m_env.d[skip].si;
// ti = vti.yyyy(); if(!sprite) ti += m_env.d[skip].ti;
pshufd(xmm2, xmm4, _MM_SHUFFLE(0, 0, 0, 0));
pshufd(xmm3, xmm4, _MM_SHUFFLE(1, 1, 1, 1));
paddd(xmm2, xmmword[edx + 16 * 7]);
if(!m_sel.sprite)
{
paddd(xmm3, xmmword[edx + 16 * 8]);
}
else
{
if(m_sel.ltf)
{
movdqa(xmm4, xmm3);
pshuflw(xmm4, xmm4, _MM_SHUFFLE(2, 2, 0, 0));
pshufhw(xmm4, xmm4, _MM_SHUFFLE(2, 2, 0, 0));
psrlw(xmm4, 1);
movdqa(xmmword[&m_env.temp.vf], xmm4);
}
}
movdqa(xmmword[&m_env.temp.s], xmm2);
movdqa(xmmword[&m_env.temp.t], xmm3);
}
else
{
// s = vt.xxxx() + m_env.d[skip].s;
// t = vt.yyyy() + m_env.d[skip].t;
// q = vt.zzzz() + m_env.d[skip].q;
movaps(xmm2, xmm4);
movaps(xmm3, xmm4);
shufps(xmm2, xmm2, _MM_SHUFFLE(0, 0, 0, 0));
shufps(xmm3, xmm3, _MM_SHUFFLE(1, 1, 1, 1));
shufps(xmm4, xmm4, _MM_SHUFFLE(2, 2, 2, 2));
addps(xmm2, xmmword[edx + 16 * 1]);
addps(xmm3, xmmword[edx + 16 * 2]);
addps(xmm4, xmmword[edx + 16 * 3]);
movaps(xmmword[&m_env.temp.s], xmm2);
movaps(xmmword[&m_env.temp.t], xmm3);
movaps(xmmword[&m_env.temp.q], xmm4);
rcpps(xmm4, xmm4);
mulps(xmm2, xmm4);
mulps(xmm3, xmm4);
}
}
if(!(m_sel.tfx == TFX_DECAL && m_sel.tcc))
{
if(m_sel.iip)
{
// GSVector4i vc = GSVector4i(v.c);
cvttps2dq(xmm6, xmmword[ebx]); // v.c
// vc = vc.upl16(vc.zwxy());
pshufd(xmm5, xmm6, _MM_SHUFFLE(1, 0, 3, 2));
punpcklwd(xmm6, xmm5);
// rb = vc.xxxx().add16(m_env.d[skip].rb);
// ga = vc.zzzz().add16(m_env.d[skip].ga);
pshufd(xmm5, xmm6, _MM_SHUFFLE(0, 0, 0, 0));
pshufd(xmm6, xmm6, _MM_SHUFFLE(2, 2, 2, 2));
paddw(xmm5, xmmword[edx + 16 * 4]);
paddw(xmm6, xmmword[edx + 16 * 5]);
movdqa(xmmword[&m_env.temp.rb], xmm5);
movdqa(xmmword[&m_env.temp.ga], xmm6);
}
else
{
if(m_sel.tfx == TFX_NONE)
{
movdqa(xmm5, xmmword[&m_env.c.rb]);
movdqa(xmm6, xmmword[&m_env.c.ga]);
}
}
}
}
}
void GPUDrawScanlineCodeGenerator::AlphaBlend()
{
if(!m_sel.abe)
{
return;
}
// xmm1 = fd
// xmm3 = a
// xmm4 = r
// xmm5 = g
// xmm6 = b
// xmm7 = test
// xmm0, xmm2 = free
// GSVector4i r = (fd & 0x001f001f) << 3;
pcmpeqd(xmm0, xmm0);
psrlw(xmm0, 11); // 0x001f
movdqa(xmm2, xmm1);
pand(xmm2, xmm0);
psllw(xmm2, 3);
switch(m_sel.abr)
{
case 0:
// r = r.avg8(c[0]);
pavgb(xmm2, xmm4);
break;
case 1:
// r = r.addus8(c[0]);
paddusb(xmm2, xmm4);
break;
case 2:
// r = r.subus8(c[0]);
psubusb(xmm2, xmm4);
break;
case 3:
// r = r.addus8(c[0].srl16(2));
movdqa(xmm0, xmm4);
psrlw(xmm0, 2);
paddusb(xmm2, xmm0);
break;
}
if(m_sel.tme)
{
movdqa(xmm0, xmm3);
blend8(xmm4, xmm2);
}
else
{
movdqa(xmm4, xmm2);
}
// GSVector4i g = (d & 0x03e003e0) >> 2;
pcmpeqd(xmm0, xmm0);
psrlw(xmm0, 11);
psllw(xmm0, 5); // 0x03e0
movdqa(xmm2, xmm1);
pand(xmm2, xmm0);
psrlw(xmm2, 2);
switch(m_sel.abr)
{
case 0:
// g = g.avg8(c[2]);
pavgb(xmm2, xmm5);
break;
case 1:
// g = g.addus8(c[2]);
paddusb(xmm2, xmm5);
break;
case 2:
// g = g.subus8(c[2]);
psubusb(xmm2, xmm5);
break;
case 3:
// g = g.addus8(c[2].srl16(2));
movdqa(xmm0, xmm5);
psrlw(xmm0, 2);
paddusb(xmm2, xmm0);
break;
}
if(m_sel.tme)
{
movdqa(xmm0, xmm3);
blend8(xmm5, xmm2);
}
else
{
movdqa(xmm5, xmm2);
}
// GSVector4i b = (d & 0x7c007c00) >> 7;
pcmpeqd(xmm0, xmm0);
psrlw(xmm0, 11);
psllw(xmm0, 10); // 0x7c00
movdqa(xmm2, xmm1);
pand(xmm2, xmm0);
psrlw(xmm2, 7);
switch(m_sel.abr)
{
case 0:
// b = b.avg8(c[2]);
pavgb(xmm2, xmm6);
break;
case 1:
// b = b.addus8(c[2]);
paddusb(xmm2, xmm6);
break;
case 2:
// b = b.subus8(c[2]);
psubusb(xmm2, xmm6);
break;
case 3:
// b = b.addus8(c[2].srl16(2));
movdqa(xmm0, xmm6);
psrlw(xmm0, 2);
paddusb(xmm2, xmm0);
break;
}
if(m_sel.tme)
{
movdqa(xmm0, xmm3);
blend8(xmm6, xmm2);
}
else
{
movdqa(xmm6, xmm2);
}
}
void GSDrawScanlineCodeGenerator::Generate()
{
push(ebx);
push(esi);
push(edi);
push(ebp);
const int params = 16;
Init(params);
if(!m_sel.edge)
{
align(16);
}
L("loop");
// ecx = steps
// esi = fzbr
// edi = fzbc
// xmm0 = z/zi
// xmm2 = u (tme)
// xmm3 = v (tme)
// xmm5 = rb (!tme)
// xmm6 = ga (!tme)
// xmm7 = test
bool tme = m_sel.tfx != TFX_NONE;
TestZ(tme ? xmm5 : xmm2, tme ? xmm6 : xmm3);
// ecx = steps
// esi = fzbr
// edi = fzbc
// - xmm0
// xmm2 = u (tme)
// xmm3 = v (tme)
// xmm5 = rb (!tme)
// xmm6 = ga (!tme)
// xmm7 = test
SampleTexture();
// ecx = steps
// esi = fzbr
// edi = fzbc
// ebp = za
// - xmm2
// - xmm3
// - xmm4
// xmm5 = rb
// xmm6 = ga
// xmm7 = test
AlphaTFX();
// ecx = steps
// esi = fzbr
// edi = fzbc
// ebp = za
// xmm2 = gaf (TFX_HIGHLIGHT || TFX_HIGHLIGHT2 && !tcc)
// xmm5 = rb
// xmm6 = ga
// xmm7 = test
if(m_sel.fwrite)
{
movdqa(xmm3, xmmword[&m_env.fm]);
}
if(m_sel.zwrite)
{
movdqa(xmm4, xmmword[&m_env.zm]);
}
// ecx = steps
// esi = fzbr
// edi = fzbc
// ebp = za
// xmm2 = gaf (TFX_HIGHLIGHT || TFX_HIGHLIGHT2 && !tcc)
// xmm3 = fm
// xmm4 = zm
// xmm5 = rb
// xmm6 = ga
// xmm7 = test
TestAlpha();
// ecx = steps
// esi = fzbr
// edi = fzbc
// ebp = za
// xmm2 = gaf (TFX_HIGHLIGHT || TFX_HIGHLIGHT2 && !tcc)
// xmm3 = fm
// xmm4 = zm
// xmm5 = rb
// xmm6 = ga
// xmm7 = test
ColorTFX();
// ecx = steps
// esi = fzbr
// edi = fzbc
// ebp = za
// xmm3 = fm
// xmm4 = zm
// xmm5 = rb
// xmm6 = ga
// xmm7 = test
Fog();
// ecx = steps
// esi = fzbr
// edi = fzbc
// ebp = za
// xmm3 = fm
// xmm4 = zm
// xmm5 = rb
// xmm6 = ga
// xmm7 = test
ReadFrame();
// ecx = steps
// esi = fzbr
// edi = fzbc
// ebp = za
// xmm2 = fd
// xmm3 = fm
// xmm4 = zm
// xmm5 = rb
// xmm6 = ga
// xmm7 = test
TestDestAlpha();
// fm |= test;
// zm |= test;
if(m_sel.fwrite)
{
por(xmm3, xmm7);
}
if(m_sel.zwrite)
{
por(xmm4, xmm7);
}
// int fzm = ~(fm == GSVector4i::xffffffff()).ps32(zm == GSVector4i::xffffffff()).mask();
pcmpeqd(xmm1, xmm1);
if(m_sel.fwrite && m_sel.zwrite)
{
movdqa(xmm0, xmm1);
pcmpeqd(xmm1, xmm3);
pcmpeqd(xmm0, xmm4);
packssdw(xmm1, xmm0);
}
else if(m_sel.fwrite)
{
pcmpeqd(xmm1, xmm3);
packssdw(xmm1, xmm1);
}
else if(m_sel.zwrite)
{
pcmpeqd(xmm1, xmm4);
packssdw(xmm1, xmm1);
}
pmovmskb(edx, xmm1);
not(edx);
// ebx = fa
// ecx = steps
// edx = fzm
// esi = fzbr
// edi = fzbc
// ebp = za
// xmm2 = fd
// xmm3 = fm
// xmm4 = zm
// xmm5 = rb
// xmm6 = ga
WriteZBuf();
// ebx = fa
// ecx = steps
// edx = fzm
// esi = fzbr
// edi = fzbc
// - ebp
// xmm2 = fd
// xmm3 = fm
// - xmm4
// xmm5 = rb
// xmm6 = ga
AlphaBlend();
// ebx = fa
// ecx = steps
// edx = fzm
// esi = fzbr
// edi = fzbc
// xmm2 = fd
// xmm3 = fm
// xmm5 = rb
// xmm6 = ga
WriteFrame(params);
L("step");
// if(steps <= 0) break;
if(!m_sel.edge)
{
test(ecx, ecx);
jle("exit", T_NEAR);
Step();
jmp("loop", T_NEAR);
}
L("exit");
pop(ebp);
pop(edi);
pop(esi);
pop(ebx);
ret(8);
}
void GSSetupPrimCodeGenerator::Texture()
{
if(!m_en.t)
{
return;
}
// GSVector4 t = dscan.t;
movaps(xmm0, xmmword[edx + 32]);
movaps(xmm1, xmm0);
mulps(xmm1, xmm3);
if(m_env.sel.fst)
{
// m_env.d4.st = GSVector4i(t * 4.0f);
cvttps2dq(xmm1, xmm1);
movdqa(xmmword[&m_env.d4.st], xmm1);
}
else
{
// m_env.d4.stq = t * 4.0f;
movaps(xmmword[&m_env.d4.stq], xmm1);
}
for(int j = 0, k = m_env.sel.fst ? 2 : 3; j < k; j++)
{
// GSVector4 ds = t.xxxx();
// GSVector4 dt = t.yyyy();
// GSVector4 dq = t.zzzz();
movaps(xmm1, xmm0);
shufps(xmm1, xmm1, (uint8)_MM_SHUFFLE(j, j, j, j));
for(int i = 0; i < 4; i++)
{
// GSVector4 v = ds/dt * m_shift[i];
movaps(xmm2, xmm1);
mulps(xmm2, Xmm(4 + i));
if(m_env.sel.fst)
{
// m_env.d[i].si/ti = GSVector4i(v);
cvttps2dq(xmm2, xmm2);
switch(j)
{
case 0: movdqa(xmmword[&m_env.d[i].si], xmm2); break;
case 1: movdqa(xmmword[&m_env.d[i].ti], xmm2); break;
}
}
else
{
// m_env.d[i].s/t/q = v;
switch(j)
{
case 0: movaps(xmmword[&m_env.d[i].s], xmm2); break;
case 1: movaps(xmmword[&m_env.d[i].t], xmm2); break;
case 2: movaps(xmmword[&m_env.d[i].q], xmm2); break;
}
}
}
}
}
void GSDrawScanlineCodeGenerator::Step()
{
// steps -= 4;
sub(ecx, 4);
// fza_offset++;
add(edi, 8);
if(!m_sel.sprite)
{
// z += m_env.d4.z;
if(m_sel.zb)
{
movaps(xmm0, xmmword[&m_env.temp.z]);
addps(xmm0, xmmword[&m_env.d4.z]);
movaps(xmmword[&m_env.temp.z], xmm0);
}
// f = f.add16(m_env.d4.f);
if(m_sel.fwrite && m_sel.fge)
{
movdqa(xmm1, xmmword[&m_env.temp.f]);
paddw(xmm1, xmmword[&m_env.d4.f]);
movdqa(xmmword[&m_env.temp.f], xmm1);
}
}
else
{
if(m_sel.ztest)
{
movdqa(xmm0, xmmword[&m_env.p.z]);
}
}
if(m_sel.fb)
{
if(m_sel.tfx != TFX_NONE)
{
if(m_sel.fst)
{
// GSVector4i st = m_env.d4.st;
// si += st.xxxx();
// if(!sprite) ti += st.yyyy();
movdqa(xmm4, xmmword[&m_env.d4.st]);
pshufd(xmm2, xmm4, _MM_SHUFFLE(0, 0, 0, 0));
paddd(xmm2, xmmword[&m_env.temp.s]);
movdqa(xmmword[&m_env.temp.s], xmm2);
if(!m_sel.sprite)
{
pshufd(xmm3, xmm4, _MM_SHUFFLE(1, 1, 1, 1));
paddd(xmm3, xmmword[&m_env.temp.t]);
movdqa(xmmword[&m_env.temp.t], xmm3);
}
else
{
movdqa(xmm3, xmmword[&m_env.temp.t]);
}
}
else
{
// GSVector4 stq = m_env.d4.stq;
// s += stq.xxxx();
// t += stq.yyyy();
// q += stq.zzzz();
movaps(xmm2, xmmword[&m_env.d4.stq]);
movaps(xmm3, xmm2);
movaps(xmm4, xmm2);
shufps(xmm2, xmm2, _MM_SHUFFLE(0, 0, 0, 0));
shufps(xmm3, xmm3, _MM_SHUFFLE(1, 1, 1, 1));
shufps(xmm4, xmm4, _MM_SHUFFLE(2, 2, 2, 2));
addps(xmm2, xmmword[&m_env.temp.s]);
addps(xmm3, xmmword[&m_env.temp.t]);
addps(xmm4, xmmword[&m_env.temp.q]);
movaps(xmmword[&m_env.temp.s], xmm2);
movaps(xmmword[&m_env.temp.t], xmm3);
movaps(xmmword[&m_env.temp.q], xmm4);
rcpps(xmm4, xmm4);
mulps(xmm2, xmm4);
mulps(xmm3, xmm4);
}
}
if(!(m_sel.tfx == TFX_DECAL && m_sel.tcc))
{
if(m_sel.iip)
{
// GSVector4i c = m_env.d4.c;
// rb = rb.add16(c.xxxx());
// ga = ga.add16(c.yyyy());
movdqa(xmm7, xmmword[&m_env.d4.c]);
pshufd(xmm5, xmm7, _MM_SHUFFLE(0, 0, 0, 0));
pshufd(xmm6, xmm7, _MM_SHUFFLE(1, 1, 1, 1));
paddw(xmm5, xmmword[&m_env.temp.rb]);
paddw(xmm6, xmmword[&m_env.temp.ga]);
movdqa(xmmword[&m_env.temp.rb], xmm5);
movdqa(xmmword[&m_env.temp.ga], xmm6);
}
else
{
if(m_sel.tfx == TFX_NONE)
{
movdqa(xmm5, xmmword[&m_env.c.rb]);
movdqa(xmm6, xmmword[&m_env.c.ga]);
}
}
}
}
// test = m_test[7 + (steps & (steps >> 31))];
mov(edx, ecx);
sar(edx, 31);
and(edx, ecx);
shl(edx, 4);
movdqa(xmm7, xmmword[edx + (size_t)&m_test[7]]);
}
void GSSetupPrimCodeGenerator::Depth()
{
if(!m_en.z && !m_en.f)
{
return;
}
if(!m_env.sel.sprite)
{
// GSVector4 t = dscan.p;
movaps(xmm0, xmmword[edx + 16]);
if(m_en.f)
{
// GSVector4 df = p.wwww();
movaps(xmm1, xmm0);
shufps(xmm1, xmm1, _MM_SHUFFLE(3, 3, 3, 3));
// m_env.d4.f = GSVector4i(df * 4.0f).xxzzlh();
movaps(xmm2, xmm1);
mulps(xmm2, xmm3);
cvttps2dq(xmm2, xmm2);
pshuflw(xmm2, xmm2, _MM_SHUFFLE(2, 2, 0, 0));
pshufhw(xmm2, xmm2, _MM_SHUFFLE(2, 2, 0, 0));
movdqa(xmmword[&m_env.d4.f], xmm2);
for(int i = 0; i < 4; i++)
{
// m_env.d[i].f = GSVector4i(df * m_shift[i]).xxzzlh();
movaps(xmm2, xmm1);
mulps(xmm2, Xmm(4 + i));
cvttps2dq(xmm2, xmm2);
pshuflw(xmm2, xmm2, _MM_SHUFFLE(2, 2, 0, 0));
pshufhw(xmm2, xmm2, _MM_SHUFFLE(2, 2, 0, 0));
movdqa(xmmword[&m_env.d[i].f], xmm2);
}
}
if(m_en.z)
{
// GSVector4 dz = p.zzzz();
shufps(xmm0, xmm0, _MM_SHUFFLE(2, 2, 2, 2));
// m_env.d4.z = dz * 4.0f;
movaps(xmm1, xmm0);
mulps(xmm1, xmm3);
movdqa(xmmword[&m_env.d4.z], xmm1);
for(int i = 0; i < 4; i++)
{
// m_env.d[i].z = dz * m_shift[i];
movaps(xmm1, xmm0);
mulps(xmm1, Xmm(4 + i));
movdqa(xmmword[&m_env.d[i].z], xmm1);
}
}
}
else
{
// GSVector4 p = vertices[0].p;
movaps(xmm0, xmmword[ecx + 16]);
if(m_en.f)
{
// m_env.p.f = GSVector4i(p).zzzzh().zzzz();
movaps(xmm1, xmm0);
cvttps2dq(xmm1, xmm1);
pshufhw(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
pshufd(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
movdqa(xmmword[&m_env.p.f], xmm1);
}
if(m_en.z)
{
// GSVector4 z = p.zzzz();
shufps(xmm0, xmm0, _MM_SHUFFLE(2, 2, 2, 2));
if(m_env.sel.zoverflow)
{
// m_env.p.z = (GSVector4i(z * 0.5f) << 1) | (GSVector4i(z) & GSVector4i::x00000001());
static const float half = 0.5f;
movss(xmm1, dword[&half]);
shufps(xmm1, xmm1, _MM_SHUFFLE(0, 0, 0, 0));
mulps(xmm1, xmm0);
cvttps2dq(xmm1, xmm1);
pslld(xmm1, 1);
cvttps2dq(xmm0, xmm0);
pcmpeqd(xmm2, xmm2);
psrld(xmm2, 31);
pand(xmm0, xmm2);
por(xmm0, xmm1);
}
else
{
// m_env.p.z = GSVector4i(z);
cvttps2dq(xmm0, xmm0);
}
movdqa(xmmword[&m_env.p.z], xmm0);
}
}
}
static __forceinline void DCT_8_INV_COL_8(__m128i &src0,__m128i &src1,__m128i &src2,__m128i &src3,__m128i &src4,__m128i &src5,__m128i &src6,__m128i &src7,
__m128i &xmm0,__m128i &xmm1,__m128i &xmm2,__m128i &xmm3,__m128i &xmm4,__m128i &xmm5,__m128i &xmm6,__m128i &xmm7)
{
movdqa( xmm1, M128_tg_3_16 );
movdqa( xmm2, xmm0 );
movdqa( xmm3, src3 );
pmulhw( xmm0, xmm1 );
pmulhw( xmm1, xmm3 );
movdqa( xmm5, M128_tg_1_16 );
movdqa( xmm6, xmm4 );
pmulhw( xmm4, xmm5 );
paddsw( xmm0, xmm2 );
pmulhw( xmm5, src1 );
paddsw( xmm1, xmm3 );
movdqa( xmm7, src6 );
paddsw( xmm0, xmm3 );
movdqa( xmm3, M128_tg_2_16 );
psubsw( xmm2, xmm1 );
pmulhw( xmm7, xmm3 );
movdqa( xmm1, xmm0 );
pmulhw( xmm3, src2 );
psubsw( xmm5, xmm6 );
paddsw( xmm4, src1 );
paddsw( xmm0, xmm4 );
paddsw( xmm0, M128_one_corr );
psubsw( xmm4, xmm1 );
movdqa( xmm6, xmm5 );
psubsw( xmm5, xmm2 );
paddsw( xmm5, M128_one_corr );
paddsw( xmm6, xmm2 );
movdqa( src7, xmm0 );
movdqa( xmm1, xmm4 );
movdqa( xmm0, M128_cos_4_16 );
paddsw( xmm4, xmm5 );
movdqa( xmm2, M128_cos_4_16 );
pmulhw( xmm2, xmm4 );
movdqa( src3, xmm6 );
psubsw( xmm1, xmm5 );
paddsw( xmm7, src2 );
psubsw( xmm3, src6 );
movdqa( xmm6, src0 );
pmulhw( xmm0, xmm1 );
movdqa( xmm5, src4 );
paddsw( xmm5, xmm6 );
psubsw( xmm6, src4 );
paddsw( xmm4, xmm2 );
por ( xmm4, M128_one_corr );
paddsw( xmm0, xmm1 );
por ( xmm0, M128_one_corr );
movdqa( xmm2, xmm5 );
paddsw( xmm5, xmm7 );
movdqa( xmm1, xmm6 );
paddsw( xmm5, M128_round_inv_col );
psubsw( xmm2, xmm7 );
movdqa( xmm7, src7 );
paddsw( xmm6, xmm3 );
paddsw( xmm6, M128_round_inv_col );
paddsw( xmm7, xmm5 );
psraw ( xmm7, SHIFT_INV_COL );
psubsw( xmm1, xmm3 );
paddsw( xmm1, M128_round_inv_corr );
movdqa( xmm3, xmm6 );
paddsw( xmm2, M128_round_inv_corr );
paddsw( xmm6, xmm4 );
movdqa( src0,xmm7 );
psraw (xmm6, SHIFT_INV_COL );
movdqa( xmm7, xmm1 );
paddsw( xmm1, xmm0 );
movdqa( src1, xmm6 );
psraw (xmm1, SHIFT_INV_COL );
movdqa( xmm6, src3 );
psubsw( xmm7, xmm0 );
psraw (xmm7, SHIFT_INV_COL );
movdqa( src2, xmm1 );
psubsw( xmm5, src7 );
psraw (xmm5, SHIFT_INV_COL );
movdqa( src7, xmm5 );
psubsw( xmm3, xmm4 );
paddsw( xmm6, xmm2 );
psubsw( xmm2, src3 );
psraw (xmm6, SHIFT_INV_COL );
psraw (xmm2, SHIFT_INV_COL );
movdqa( src3, xmm6 );
psraw (xmm3, SHIFT_INV_COL );
movdqa( src4, xmm2 );
movdqa( src5, xmm7 );
movdqa( src6, xmm3 );
}