GSdx: When mipmapping is on, LOD is calculated per pixel, it isn't used for anything, but it's there. I cannot really measure any significant slowdown, but rest of the fun is yet to come.

git-svn-id: http://pcsx2.googlecode.com/svn/trunk@4428 96395faa-99c1-11dd-bbfe-3dabce05a288
This commit is contained in:
gabest11 committed 2011-03-14 03:32:28 +00:00
1 parent 5e9930a9dc
commit 1d759c852d
18 files changed
+405 -138

No files matched your search

+107 -46
View File
@@ -49,8 +49,9 @@ L("loop");
// esi = fzbr
// edi = fzbc
// xmm0 = z/zi
// xmm2 = u (tme)
// xmm3 = v (tme)
// xmm2 = s/u (tme)
// xmm3 = t/v (tme)
// xmm4 = q (tme)
// xmm5 = rb (!tme)
// xmm6 = ga (!tme)
// xmm7 = test
@@ -63,8 +64,9 @@ L("loop");
// esi = fzbr
// edi = fzbc
// - xmm0
// xmm2 = u (tme)
// xmm3 = v (tme)
// xmm2 = s/u (tme)
// xmm3 = t/v (tme)
// xmm4 = q (tme)
// xmm5 = rb (!tme)
// xmm6 = ga (!tme)
// xmm7 = test
@@ -281,7 +283,7 @@ void GSDrawScanlineCodeGenerator::Init()
{
// edx = &m_local.d[skip]
shl(edx, 4);
shl(edx, 3);
lea(edx, ptr[edx + (size_t)m_local.d]);
// ebx = &v
@@ -302,7 +304,7 @@ void GSDrawScanlineCodeGenerator::Init()
cvttps2dq(xmm1, xmm0);
pshufhw(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
pshufd(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
paddw(xmm1, ptr[edx + 16 * 6]);
paddw(xmm1, ptr[edx + offsetof(GSScanlineLocalData::skip, f)]);
movdqa(ptr[&m_local.temp.f], xmm1);
}
@@ -312,7 +314,7 @@ void GSDrawScanlineCodeGenerator::Init()
// z = vp.zzzz() + m_local.d[skip].z;
shufps(xmm0, xmm0, _MM_SHUFFLE(2, 2, 2, 2));
addps(xmm0, ptr[edx]);
addps(xmm0, ptr[edx + offsetof(GSScanlineLocalData::skip, z)]);
movaps(ptr[&m_local.temp.z], xmm0);
}
@@ -348,34 +350,41 @@ void GSDrawScanlineCodeGenerator::Init()
{
// GSVector4i vti(vt);
cvttps2dq(xmm4, xmm4);
cvttps2dq(xmm6, xmm4);
// si = vti.xxxx() + m_local.d[skip].si;
// ti = vti.yyyy(); if(!sprite) ti += m_local.d[skip].ti;
// s = vti.xxxx() + m_local.d[skip].s;
// t = vti.yyyy(); if(!sprite) t += m_local.d[skip].t;
pshufd(xmm2, xmm4, _MM_SHUFFLE(0, 0, 0, 0));
pshufd(xmm3, xmm4, _MM_SHUFFLE(1, 1, 1, 1));
pshufd(xmm2, xmm6, _MM_SHUFFLE(0, 0, 0, 0));
pshufd(xmm3, xmm6, _MM_SHUFFLE(1, 1, 1, 1));
paddd(xmm2, ptr[edx + 16 * 7]);
paddd(xmm2, ptr[edx + offsetof(GSScanlineLocalData::skip, s)]);
if(!m_sel.sprite)
{
paddd(xmm3, ptr[edx + 16 * 8]);
paddd(xmm3, ptr[edx + offsetof(GSScanlineLocalData::skip, t)]);
}
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(ptr[&m_local.temp.vf], xmm4);
movdqa(xmm6, xmm3);
pshuflw(xmm6, xmm6, _MM_SHUFFLE(2, 2, 0, 0));
pshufhw(xmm6, xmm6, _MM_SHUFFLE(2, 2, 0, 0));
psrlw(xmm6, 1);
movdqa(ptr[&m_local.temp.vf], xmm6);
}
}
movdqa(ptr[&m_local.temp.s], xmm2);
movdqa(ptr[&m_local.temp.t], xmm3);
if(m_sel.mipmap && !m_sel.lcm)
{
shufps(xmm4, xmm4, _MM_SHUFFLE(2, 2, 2, 2));
addps(xmm4, ptr[edx + offsetof(GSScanlineLocalData::skip, q)]);
movaps(ptr[&m_local.temp.q], xmm4);
}
}
else
{
@@ -390,17 +399,13 @@ void GSDrawScanlineCodeGenerator::Init()
shufps(xmm3, xmm3, _MM_SHUFFLE(1, 1, 1, 1));
shufps(xmm4, xmm4, _MM_SHUFFLE(2, 2, 2, 2));
addps(xmm2, ptr[edx + 16 * 1]);
addps(xmm3, ptr[edx + 16 * 2]);
addps(xmm4, ptr[edx + 16 * 3]);
addps(xmm2, ptr[edx + offsetof(GSScanlineLocalData::skip, s)]);
addps(xmm3, ptr[edx + offsetof(GSScanlineLocalData::skip, t)]);
addps(xmm4, ptr[edx + offsetof(GSScanlineLocalData::skip, q)]);
movaps(ptr[&m_local.temp.s], xmm2);
movaps(ptr[&m_local.temp.t], xmm3);
movaps(ptr[&m_local.temp.q], xmm4);
rcpps(xmm4, xmm4);
mulps(xmm2, xmm4);
mulps(xmm3, xmm4);
}
}
@@ -423,8 +428,8 @@ void GSDrawScanlineCodeGenerator::Init()
pshufd(xmm5, xmm6, _MM_SHUFFLE(0, 0, 0, 0));
pshufd(xmm6, xmm6, _MM_SHUFFLE(2, 2, 2, 2));
paddw(xmm5, ptr[edx + 16 * 4]);
paddw(xmm6, ptr[edx + 16 * 5]);
paddw(xmm5, ptr[edx + offsetof(GSScanlineLocalData::skip, rb)]);
paddw(xmm6, ptr[edx + offsetof(GSScanlineLocalData::skip, ga)]);
movdqa(ptr[&m_local.temp.rb], xmm5);
movdqa(ptr[&m_local.temp.ga], xmm6);
@@ -485,12 +490,12 @@ void GSDrawScanlineCodeGenerator::Step()
{
if(m_sel.fst)
{
// GSVector4i st = m_local.d4.st;
// GSVector4i st = m_local.d4.stq;
// si += st.xxxx();
// if(!sprite) ti += st.yyyy();
movdqa(xmm4, ptr[&m_local.d4.st]);
movdqa(xmm4, ptr[&m_local.d4.stq]);
pshufd(xmm2, xmm4, _MM_SHUFFLE(0, 0, 0, 0));
paddd(xmm2, ptr[&m_local.temp.s]);
@@ -506,6 +511,13 @@ void GSDrawScanlineCodeGenerator::Step()
{
movdqa(xmm3, ptr[&m_local.temp.t]);
}
if(m_sel.mipmap && !m_sel.lcm)
{
shufps(xmm4, xmm4, _MM_SHUFFLE(2, 2, 2, 2));
addps(xmm4, ptr[&m_local.temp.q]);
movaps(ptr[&m_local.temp.q], xmm4);
}
}
else
{
@@ -515,9 +527,9 @@ void GSDrawScanlineCodeGenerator::Step()
// t += stq.yyyy();
// q += stq.zzzz();
movaps(xmm2, ptr[&m_local.d4.stq]);
movaps(xmm3, xmm2);
movaps(xmm4, xmm2);
movaps(xmm4, ptr[&m_local.d4.stq]);
movaps(xmm2, xmm4);
movaps(xmm3, xmm4);
shufps(xmm2, xmm2, _MM_SHUFFLE(0, 0, 0, 0));
shufps(xmm3, xmm3, _MM_SHUFFLE(1, 1, 1, 1));
@@ -530,10 +542,6 @@ void GSDrawScanlineCodeGenerator::Step()
movaps(ptr[&m_local.temp.s], xmm2);
movaps(ptr[&m_local.temp.t], xmm3);
movaps(ptr[&m_local.temp.q], xmm4);
rcpps(xmm4, xmm4);
mulps(xmm2, xmm4);
mulps(xmm3, xmm4);
}
}
@@ -650,16 +658,14 @@ void GSDrawScanlineCodeGenerator::TestZ(const Xmm& temp1, const Xmm& temp2)
{
// GSVector4i o = GSVector4i::x80000000();
pcmpeqd(xmm4, xmm4);
pslld(xmm4, 31);
pcmpeqd(temp1, temp1);
pslld(temp1, 31);
// GSVector4i zso = zs - o;
psubd(xmm0, xmm4);
// GSVector4i zdo = zd - o;
psubd(xmm1, xmm4);
psubd(xmm0, temp1);
psubd(xmm1, temp1);
}
switch(m_sel.ztst)
@@ -673,8 +679,8 @@ void GSDrawScanlineCodeGenerator::TestZ(const Xmm& temp1, const Xmm& temp2)
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);
pcmpeqd(temp1, temp1);
pxor(xmm0, temp1);
por(xmm7, xmm0);
break;
}
@@ -696,11 +702,66 @@ void GSDrawScanlineCodeGenerator::SampleTexture()
if(!m_sel.fst)
{
// TODO: move these into Init/Step too?
rcpps(xmm4, xmm4);
mulps(xmm2, xmm4);
mulps(xmm3, xmm4);
cvttps2dq(xmm2, xmm2);
cvttps2dq(xmm3, xmm3);
}
if(m_sel.mipmap)
{
// TODO: if the fractional part is not needed in round-off mode then there is a faster integer log2 (but can we round it?)
if(!m_sel.lcm)
{
// lod = -log2(Q) * (1 << L) + K
vpcmpeqd(xmm1, xmm1);
vpsrld(xmm1, xmm1, 25);
vpslld(xmm0, xmm4, 1);
vpsrld(xmm0, xmm0, 24);
vpsubd(xmm0, xmm1);
vcvtdq2ps(xmm0, xmm0);
// xmm0 = (float)(exp(e) - 127)
vpslld(xmm4, xmm4, 9);
vpsrld(xmm4, xmm4, 9);
vorps(xmm4, ptr[&GSDrawScanlineCodeGenerator::m_log2_coef[3]]);
// xmm4 = mant(q) | 1.0f
vmulps(xmm5, xmm4, ptr[&GSDrawScanlineCodeGenerator::m_log2_coef[0]]);
vaddps(xmm5, ptr[&GSDrawScanlineCodeGenerator::m_log2_coef[1]]);
vmulps(xmm5, xmm4);
vsubps(xmm4, ptr[&GSDrawScanlineCodeGenerator::m_log2_coef[3]]);
vaddps(xmm5, ptr[&GSDrawScanlineCodeGenerator::m_log2_coef[2]]);
vmulps(xmm4, xmm5);
vaddps(xmm4, xmm0);
// xmm4 = log2(Q) = (((((c0 * xmm4) + c1) * xmm4) + c2) * (xmm4 - 1.0f) + xmm0)
vmulps(xmm4, ptr[&m_local.gd->l]);
vaddps(xmm4, ptr[&m_local.gd->k]);
// xmm4 = (-log2(Q) * (1 << L) + K) * 0x10000
vcvtps2dq(xmm4, xmm4);
vmovdqa(ptr[&m_local.temp.lod], xmm4);
}
else
{
// lod = K (=> use m_local->gd.k later when lod is needed)
}
}
// TODO: if(m_sel.mipmap) ...
if(!m_sel.fst)
{
if(m_sel.ltf)
{
// u -= 0x8000;