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 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]]);
}
float TfirFilter::vec_inner_prod_sse(const float *eax, const float *edi, int ecx)
{
__m128 xmm3,xmm4,xmm0,xmm1,xmm5,xmm6;
xorps (xmm3, xmm3);
xorps (xmm4, xmm4);
ecx-=8;// sub $8, %%ecx
if (ecx<0) goto //jb
mul8_skip;
mul8_loop:
movups (xmm0,eax);
movups (xmm1,edi);
movups (xmm5,16/sizeof(float)+eax);
movups (xmm6,16/sizeof(float)+edi);
eax+=32/sizeof(float);
edi+=32/sizeof(float);
mulps (xmm1,xmm0);
mulps (xmm6,xmm5);
addps (xmm3,xmm1);
addps (xmm4,xmm6);
ecx-=8;
if (ecx>=0) //jae
goto mul8_loop;
mul8_skip:
addps (xmm3,xmm4);
ecx+=4;
if (ecx<0) //jl
goto mul4_skip;
movups (xmm0,eax);
movups (xmm1,edi);
eax+=16/sizeof(float);
edi+=16/sizeof(float);
mulps (xmm1, xmm0);
addps (xmm3, xmm1);
ecx-=4;
mul4_skip:
ecx+=4;
goto cond1;
mul1_loop:
movss (xmm0,eax);
movss (xmm1,edi);
eax+=4/sizeof(float);
edi+=4/sizeof(float);
mulss (xmm1,xmm0);
addss (xmm3,xmm1);
cond1:
ecx-=1;
if (ecx>=0) // jae
goto mul1_loop;
movhlps (xmm4,xmm3);
addps (xmm3,xmm4);
movaps (xmm4,xmm3);
//FIXME: which one?
xmm4=_mm_shuffle_ps(xmm4,xmm4,0x55);// shufps $0x55, xmm4, xmm4
// shufps $33, xmm4, xmm4
addss (xmm3, xmm4);
float sum;
movss (&sum , xmm3);
return sum;
}
