void GSRasterizer::DrawTriangleEdge(const GSVertexSW* v, const GSVector4i& scissor)
{
GSVertexSW dv[3];
dv[0] = v[1] - v[0];
dv[1] = v[2] - v[0];
dv[2] = v[2] - v[1];
GSVector4 dx = dv[0].p.upl(dv[1].p).xyxy(dv[2].p);
GSVector4 dy = dv[0].p.upl(dv[1].p).zwyx(dv[2].p);
GSVector4 a = dx.abs() < dy.abs(); // |x| <= |y|
GSVector4 b = dx < GSVector4::zero(); // x < 0
GSVector4 c = dv[1].p * (dv[0].p / dv[1].p).yyyy() < dv[0].p; // longest.p.x < 0
int i = a.mask();
int j = ((a | b) ^ c.xxxx()).mask() ^ 2; // evil
GSVertexSW dscan;
dscan.p = GSVector4::zero();
dscan.t = GSVector4::zero();
dscan.c = GSVector4::zero();
m_dsf.ssp(v, dscan); // TODO: don't call it twice (can't be sure about the second call if the triangle is too small)
DrawEdge(v[0], v[1], dv[0], scissor, i & 1, j & 1);
DrawEdge(v[0], v[2], dv[1], scissor, i & 2, j & 2);
DrawEdge(v[1], v[2], dv[2], scissor, i & 4, j & 4);
}
void GSRasterizer::DrawTriangleBottom(GSVertexSW* v, const GSVector4i& scissor)
{
GSVertexSW longest;
longest.p = v[1].p - v[0].p;
int i = (longest.p > GSVector4::zero()).upl(longest.p == GSVector4::zero()).mask();
if(i & 2) return;
i &= 1;
GSVertexSW& l = v[1 - i];
GSVector4& r = v[i].p;
GSVector4i tb(l.p.xyxy(v[2].p).ceil());
int top = tb.extract32<1>();
int bottom = tb.extract32<3>();
if(top < scissor.y) top = scissor.y;
if(bottom > scissor.w) bottom = scissor.w;
if(top >= bottom) return;
longest.t = v[1].t - v[0].t;
longest.c = v[1].c - v[0].c;
GSVertexSW dscan = longest * longest.p.xxxx().rcp();
GSVertexSW vl = v[2] - l;
GSVector4 vr = v[2].p - r;
GSVertexSW dl = vl / vl.p.yyyy();
GSVector4 dr = vr / vr.yyyy();
float py = (float)top - l.p.y;
l.p = l.p.upl(r).xyzw(l.p); // r.x => l.y
dl.p = dl.p.upl(dr).xyzw(dl.p); // dr.x => dl.y
if(py > 0) l += dl * py;
(m_ds->*m_dsf.sp)(v, dscan);
// TODO: (m_dsf.ssp)(v, dscan);
DrawTriangleSection(top, bottom, l, dl, dscan, scissor);
}
void GSRasterizer::DrawTriangleBottom(GSVertexSW* v, const GSVector4i& scissor)
{
GSVertexSW longest;
longest.p = v[1].p - v[0].p;
int i = longest.p.upl(longest.p == GSVector4::zero()).mask();
if(i & 2) return;
i &= 1;
GSVertexSW& l = v[i];
GSVector4& r = v[1 - i].p;
GSVector4 fscissor(scissor);
GSVector4 tb = l.p.upl(v[2].p).ceil();
GSVector4 tbmax = tb.max(fscissor.yyyy());
GSVector4 tbmin = tb.min(fscissor.wwww());
GSVector4i tbi = GSVector4i(tbmax.zzww(tbmin));
int top = tbi.extract32<0>();
int bottom = tbi.extract32<2>();
if(top >= bottom) return;
longest.t = v[1].t - v[0].t;
longest.c = v[1].c - v[0].c;
GSVertexSW dscan = longest * longest.p.xxxx().rcp();
GSVertexSW vl = v[2] - l;
GSVector4 vr = v[2].p - r;
GSVertexSW dl = vl / vl.p.yyyy();
GSVector4 dr = vr / vr.yyyy();
GSVector4 dy = tbmax.zzzz() - l.p.yyyy();
l.p = l.p.upl(r).xyzw(l.p); // r.x => l.y
dl.p = dl.p.upl(dr).xyzw(dl.p); // dr.x => dl.y
l += dl * dy;
m_dsf.ssp(v, dscan);
DrawTriangleSection(top, bottom, l, dl, dscan, fscissor);
}
void GSRendererDX9::UpdateFBA(GSTexture* rt)
{
if (!rt)
return;
GSDevice9* dev = (GSDevice9*)m_dev;
dev->BeginScene();
// om
dev->OMSetDepthStencilState(&m_fba.dss);
dev->OMSetBlendState(&m_fba.bs, 0);
// ia
GSVector4 s = GSVector4(rt->GetScale().x / rt->GetWidth(), rt->GetScale().y / rt->GetHeight());
GSVector4 off = GSVector4(-1.0f, 1.0f);
GSVector4 src = ((m_vt.m_min.p.xyxy(m_vt.m_max.p) + off.xxyy()) * s.xyxy()).sat(off.zzyy());
GSVector4 dst = src * 2.0f + off.xxxx();
GSVertexPT1 vertices[] =
{
{GSVector4(dst.x, -dst.y, 0.5f, 1.0f), GSVector2(0)},
{GSVector4(dst.z, -dst.y, 0.5f, 1.0f), GSVector2(0)},
{GSVector4(dst.x, -dst.w, 0.5f, 1.0f), GSVector2(0)},
{GSVector4(dst.z, -dst.w, 0.5f, 1.0f), GSVector2(0)},
};
dev->IASetVertexBuffer(vertices, sizeof(vertices[0]), countof(vertices));
dev->IASetInputLayout(dev->m_convert.il);
dev->IASetPrimitiveTopology(D3DPT_TRIANGLESTRIP);
// vs
dev->VSSetShader(dev->m_convert.vs, NULL, 0);
// ps
dev->PSSetShader(dev->m_convert.ps[4], NULL, 0);
//
dev->DrawPrimitive();
//
dev->EndScene();
}
void GSRasterizer::DrawTriangleSection(int top, int bottom, GSVertexSW& l, const GSVertexSW& dl, GSVector4& r, const GSVector4& dr, const GSVertexSW& dscan, const GSVector4& fscissor)
{
ASSERT(top < bottom);
while(1)
{
do
{
if(IsOneOfMyScanlines(top))
{
GSVector4 lr = l.p.xyxy(r).ceil();
GSVector4 lrmax = lr.max(fscissor.xxxx());
GSVector4 lrmin = lr.min(fscissor.zzzz());
GSVector4i lri = GSVector4i(lrmax.xxzz(lrmin));
int left = lri.extract32<0>();
int right = lri.extract32<2>();
int pixels = right - left;
if(pixels > 0)
{
m_stats.pixels += pixels;
GSVertexSW scan = l + dscan * (lrmax - l.p).xxxx();
m_dsf.ssl(right, left, top, scan);
}
}
}
while(0);
if(++top >= bottom) break;
l += dl;
r += dr;
}
}
void GSRendererDX9::VertexKick(bool skip)
{
GSVector4 p = GSVector4(((GSVector4i)m_v.XYZ).upl16());
if(tme && !fst)
{
p = p.xyxy(GSVector4((float)m_v.XYZ.Z, m_v.RGBAQ.Q));
}
else
{
p = p.xyxy(GSVector4::load((float)m_v.XYZ.Z));
}
GSVertexHW9& dst = m_vl.AddTail();
dst.p = p;
int Uadjust = 0;
int Vadjust = 0;
if(tme)
{
if(fst)
{
dst.t = m_v.GetUV();
#ifdef USE_UPSCALE_HACKS
int Udiff = 0;
int Vdiff = 0;
int multiplier = GetUpscaleMultiplier();
if(multiplier > 1)
{
Udiff = m_v.UV.U & 4095;
Vdiff = m_v.UV.V & 4095;
if(Udiff != 0)
{
if (Udiff >= 4080) {/*printf("U+ %d %d\n", Udiff, m_v.UV.U);*/ Uadjust = -1; }
else if (Udiff <= 16) {/*printf("U- %d %d\n", Udiff, m_v.UV.U);*/ Uadjust = 1; }
}
if(Vdiff != 0)
{
if (Vdiff >= 4080) {/*printf("V+ %d %d\n", Vdiff, m_v.UV.V);*/ Vadjust = -1; }
else if (Vdiff <= 16) {/*printf("V- %d %d\n", Vdiff, m_v.UV.V);*/ Vadjust = 1; }
}
Udiff = m_v.UV.U & 255;
Vdiff = m_v.UV.V & 255;
if(Udiff != 0)
{
if (Udiff >= 248) { Uadjust = -1; }
else if (Udiff <= 8) { Uadjust = 1; }
}
if(Vdiff != 0)
{
if (Vdiff >= 248) { Vadjust = -1; }
else if (Vdiff <= 8) { Vadjust = 1; }
}
Udiff = m_v.UV.U & 15;
Vdiff = m_v.UV.V & 15;
if(Udiff != 0)
{
if (Udiff >= 15) { Uadjust = -1; }
else if (Udiff <= 1) { Uadjust = 1; }
}
if(Vdiff != 0)
{
if (Vdiff >= 15) { Vadjust = -1; }
else if (Vdiff <= 1) { Vadjust = 1; }
}
}
dst.t.x -= (float) Uadjust;
dst.t.y -= (float) Vadjust;
#endif
}
else
{
dst.t = GSVector4::loadl(&m_v.ST);
}
}
dst._c0() = m_v.RGBAQ.u32[0];
dst._c1() = m_v.FOG.u32[1];
//
// BaseDrawingKick can never return NULL here because the DrawingKick function
// tables route to DrawingKickNull for GS_INVLALID prim types (and that's the only
// condition where this function would return NULL).
int count = 0;
if(GSVertexHW9* v = DrawingKick<prim>(skip, count))
{
GSVector4 scissor = m_context->scissor.dx9;
GSVector4 pmin, pmax;
switch(prim)
{
case GS_POINTLIST:
pmin = v[0].p;
pmax = v[0].p;
break;
case GS_LINELIST:
case GS_LINESTRIP:
case GS_SPRITE:
pmin = v[0].p.min(v[1].p);
pmax = v[0].p.max(v[1].p);
break;
case GS_TRIANGLELIST:
case GS_TRIANGLESTRIP:
case GS_TRIANGLEFAN:
pmin = v[0].p.min(v[1].p).min(v[2].p);
pmax = v[0].p.max(v[1].p).max(v[2].p);
break;
}
GSVector4 test = (pmax < scissor) | (pmin > scissor.zwxy());
switch(prim)
{
case GS_TRIANGLELIST:
case GS_TRIANGLESTRIP:
case GS_TRIANGLEFAN:
case GS_SPRITE:
test |= pmin == pmax;
break;
}
if(test.mask() & 3)
{
return;
}
switch(prim)
{
case GS_POINTLIST:
break;
case GS_LINELIST:
case GS_LINESTRIP:
if(PRIM->IIP == 0) {v[0]._c0() = v[1]._c0();}
break;
case GS_TRIANGLELIST:
case GS_TRIANGLESTRIP:
case GS_TRIANGLEFAN:
if(PRIM->IIP == 0) {v[0]._c0() = v[1]._c0() = v[2]._c0();}
break;
case GS_SPRITE:
if(PRIM->IIP == 0) {v[0]._c0() = v[1]._c0();}
v[0].p.z = v[1].p.z;
v[0].p.w = v[1].p.w;
v[0]._c1() = v[1]._c1();
v[2] = v[1];
v[3] = v[1];
v[1].p.y = v[0].p.y;
v[1].t.y = v[0].t.y;
v[2].p.x = v[0].p.x;
v[2].t.x = v[0].t.x;
v[4] = v[1];
v[5] = v[2];
count += 4;
break;
}
m_count += count;
}
}
void GSVertexTrace::FindMinMax(const void* vertex, const uint32* index, int count)
{
const GSDrawingContext* context = m_state->m_context;
int n = 1;
switch(primclass)
{
case GS_POINT_CLASS:
n = 1;
break;
case GS_LINE_CLASS:
case GS_SPRITE_CLASS:
n = 2;
break;
case GS_TRIANGLE_CLASS:
n = 3;
break;
}
GSVector4 tmin = s_minmax.xxxx();
GSVector4 tmax = s_minmax.yyyy();
GSVector4i cmin = GSVector4i::xffffffff();
GSVector4i cmax = GSVector4i::zero();
#if _M_SSE >= 0x401
GSVector4i pmin = GSVector4i::xffffffff();
GSVector4i pmax = GSVector4i::zero();
#else
GSVector4 pmin = s_minmax.xxxx();
GSVector4 pmax = s_minmax.yyyy();
#endif
const GSVertex* RESTRICT v = (GSVertex*)vertex;
for(int i = 0; i < count; i += n)
{
if(primclass == GS_POINT_CLASS)
{
GSVector4i c(v[index[i]].m[0]);
if(color)
{
cmin = cmin.min_u8(c);
cmax = cmax.max_u8(c);
}
if(tme)
{
if(!fst)
{
GSVector4 stq = GSVector4::cast(c);
GSVector4 q = stq.wwww();
stq = (stq.xyww() * q.rcpnr()).xyww(q);
tmin = tmin.min(stq);
tmax = tmax.max(stq);
}
else
{
GSVector4i uv(v[index[i]].m[1]);
GSVector4 st = GSVector4(uv.uph16()).xyxy();
tmin = tmin.min(st);
tmax = tmax.max(st);
}
}
GSVector4i xyzf(v[index[i]].m[1]);
GSVector4i xy = xyzf.upl16();
GSVector4i z = xyzf.yyyy();
#if _M_SSE >= 0x401
GSVector4i p = xy.blend16<0xf0>(z.uph32(xyzf));
pmin = pmin.min_u32(p);
pmax = pmax.max_u32(p);
#else
GSVector4 p = GSVector4(xy.upl64(z.srl32(1).upl32(xyzf.wwww())));
pmin = pmin.min(p);
pmax = pmax.max(p);
#endif
}
else if(primclass == GS_LINE_CLASS)
{
GSVector4i c0(v[index[i + 0]].m[0]);
GSVector4i c1(v[index[i + 1]].m[0]);
if(color)
{
if(iip)
{
cmin = cmin.min_u8(c0.min_u8(c1));
cmax = cmax.max_u8(c0.max_u8(c1));
}
else
{
cmin = cmin.min_u8(c1);
cmax = cmax.max_u8(c1);
}
}
if(tme)
{
if(!fst)
{
GSVector4 stq0 = GSVector4::cast(c0);
GSVector4 stq1 = GSVector4::cast(c1);
GSVector4 q = stq0.wwww(stq1).rcpnr();
stq0 = (stq0.xyww() * q.xxxx()).xyww(stq0);
stq1 = (stq1.xyww() * q.zzzz()).xyww(stq1);
tmin = tmin.min(stq0.min(stq1));
tmax = tmax.max(stq0.max(stq1));
}
else
{
GSVector4i uv0(v[index[i + 0]].m[1]);
GSVector4i uv1(v[index[i + 1]].m[1]);
GSVector4 st0 = GSVector4(uv0.uph16()).xyxy();
GSVector4 st1 = GSVector4(uv1.uph16()).xyxy();
tmin = tmin.min(st0.min(st1));
tmax = tmax.max(st0.max(st1));
}
}
GSVector4i xyzf0(v[index[i + 0]].m[1]);
GSVector4i xyzf1(v[index[i + 1]].m[1]);
GSVector4i xy0 = xyzf0.upl16();
GSVector4i z0 = xyzf0.yyyy();
GSVector4i xy1 = xyzf1.upl16();
GSVector4i z1 = xyzf1.yyyy();
#if _M_SSE >= 0x401
GSVector4i p0 = xy0.blend16<0xf0>(z0.uph32(xyzf0));
GSVector4i p1 = xy1.blend16<0xf0>(z1.uph32(xyzf1));
pmin = pmin.min_u32(p0.min_u32(p1));
pmax = pmax.max_u32(p0.max_u32(p1));
#else
GSVector4 p0 = GSVector4(xy0.upl64(z0.srl32(1).upl32(xyzf0.wwww())));
GSVector4 p1 = GSVector4(xy1.upl64(z1.srl32(1).upl32(xyzf1.wwww())));
pmin = pmin.min(p0.min(p1));
pmax = pmax.max(p0.max(p1));
#endif
}
else if(primclass == GS_TRIANGLE_CLASS)
{
GSVector4i c0(v[index[i + 0]].m[0]);
GSVector4i c1(v[index[i + 1]].m[0]);
GSVector4i c2(v[index[i + 2]].m[0]);
if(color)
{
if(iip)
{
cmin = cmin.min_u8(c2).min_u8(c0.min_u8(c1));
cmax = cmax.max_u8(c2).max_u8(c0.max_u8(c1));
}
else
{
cmin = cmin.min_u8(c2);
cmax = cmax.max_u8(c2);
}
}
if(tme)
{
if(!fst)
{
GSVector4 stq0 = GSVector4::cast(c0);
GSVector4 stq1 = GSVector4::cast(c1);
GSVector4 stq2 = GSVector4::cast(c2);
GSVector4 q = stq0.wwww(stq1).xzww(stq2).rcpnr();
stq0 = (stq0.xyww() * q.xxxx()).xyww(stq0);
stq1 = (stq1.xyww() * q.yyyy()).xyww(stq1);
stq2 = (stq2.xyww() * q.zzzz()).xyww(stq2);
tmin = tmin.min(stq2).min(stq0.min(stq1));
tmax = tmax.max(stq2).max(stq0.max(stq1));
}
else
{
GSVector4i uv0(v[index[i + 0]].m[1]);
GSVector4i uv1(v[index[i + 1]].m[1]);
GSVector4i uv2(v[index[i + 2]].m[1]);
GSVector4 st0 = GSVector4(uv0.uph16()).xyxy();
GSVector4 st1 = GSVector4(uv1.uph16()).xyxy();
GSVector4 st2 = GSVector4(uv2.uph16()).xyxy();
tmin = tmin.min(st2).min(st0.min(st1));
tmax = tmax.max(st2).max(st0.max(st1));
}
}
GSVector4i xyzf0(v[index[i + 0]].m[1]);
GSVector4i xyzf1(v[index[i + 1]].m[1]);
GSVector4i xyzf2(v[index[i + 2]].m[1]);
GSVector4i xy0 = xyzf0.upl16();
GSVector4i z0 = xyzf0.yyyy();
GSVector4i xy1 = xyzf1.upl16();
GSVector4i z1 = xyzf1.yyyy();
GSVector4i xy2 = xyzf2.upl16();
GSVector4i z2 = xyzf2.yyyy();
#if _M_SSE >= 0x401
GSVector4i p0 = xy0.blend16<0xf0>(z0.uph32(xyzf0));
GSVector4i p1 = xy1.blend16<0xf0>(z1.uph32(xyzf1));
GSVector4i p2 = xy2.blend16<0xf0>(z2.uph32(xyzf2));
pmin = pmin.min_u32(p2).min_u32(p0.min_u32(p1));
pmax = pmax.max_u32(p2).max_u32(p0.max_u32(p1));
#else
GSVector4 p0 = GSVector4(xy0.upl64(z0.srl32(1).upl32(xyzf0.wwww())));
GSVector4 p1 = GSVector4(xy1.upl64(z1.srl32(1).upl32(xyzf1.wwww())));
GSVector4 p2 = GSVector4(xy2.upl64(z2.srl32(1).upl32(xyzf2.wwww())));
pmin = pmin.min(p2).min(p0.min(p1));
pmax = pmax.max(p2).max(p0.max(p1));
#endif
}
else if(primclass == GS_SPRITE_CLASS)
{
GSVector4i c0(v[index[i + 0]].m[0]);
GSVector4i c1(v[index[i + 1]].m[0]);
if(color)
{
if(iip)
{
cmin = cmin.min_u8(c0.min_u8(c1));
cmax = cmax.max_u8(c0.max_u8(c1));
}
else
{
cmin = cmin.min_u8(c1);
cmax = cmax.max_u8(c1);
}
}
if(tme)
{
if(!fst)
{
GSVector4 stq0 = GSVector4::cast(c0);
GSVector4 stq1 = GSVector4::cast(c1);
GSVector4 q = stq1.wwww().rcpnr();
stq0 = (stq0.xyww() * q).xyww(stq1);
stq1 = (stq1.xyww() * q).xyww(stq1);
tmin = tmin.min(stq0.min(stq1));
tmax = tmax.max(stq0.max(stq1));
}
else
{
GSVector4i uv0(v[index[i + 0]].m[1]);
GSVector4i uv1(v[index[i + 1]].m[1]);
GSVector4 st0 = GSVector4(uv0.uph16()).xyxy();
GSVector4 st1 = GSVector4(uv1.uph16()).xyxy();
tmin = tmin.min(st0.min(st1));
tmax = tmax.max(st0.max(st1));
}
}
GSVector4i xyzf0(v[index[i + 0]].m[1]);
GSVector4i xyzf1(v[index[i + 1]].m[1]);
GSVector4i xy0 = xyzf0.upl16();
GSVector4i z0 = xyzf0.yyyy();
GSVector4i xy1 = xyzf1.upl16();
GSVector4i z1 = xyzf1.yyyy();
#if _M_SSE >= 0x401
GSVector4i p0 = xy0.blend16<0xf0>(z0.uph32(xyzf1));
GSVector4i p1 = xy1.blend16<0xf0>(z1.uph32(xyzf1));
pmin = pmin.min_u32(p0.min_u32(p1));
pmax = pmax.max_u32(p0.max_u32(p1));
#else
GSVector4 p0 = GSVector4(xy0.upl64(z0.srl32(1).upl32(xyzf1.wwww())));
GSVector4 p1 = GSVector4(xy1.upl64(z1.srl32(1).upl32(xyzf1.wwww())));
pmin = pmin.min(p0.min(p1));
pmax = pmax.max(p0.max(p1));
#endif
}
}
#if _M_SSE >= 0x401
pmin = pmin.blend16<0x30>(pmin.srl32(1));
pmax = pmax.blend16<0x30>(pmax.srl32(1));
#endif
GSVector4 o(context->XYOFFSET);
GSVector4 s(1.0f / 16, 1.0f / 16, 2.0f, 1.0f);
m_min.p = (GSVector4(pmin) - o) * s;
m_max.p = (GSVector4(pmax) - o) * s;
if(tme)
{
if(fst)
{
s = GSVector4(1.0f / 16, 1.0f).xxyy();
}
else
{
s = GSVector4(1 << context->TEX0.TW, 1 << context->TEX0.TH, 1, 1);
}
m_min.t = tmin * s;
m_max.t = tmax * s;
}
else
{
m_min.t = GSVector4::zero();
m_max.t = GSVector4::zero();
}
if(color)
{
m_min.c = cmin.zzzz().u8to32();
m_max.c = cmax.zzzz().u8to32();
}
else
{
m_min.c = GSVector4i::zero();
m_max.c = GSVector4i::zero();
}
}
void GSRendererDX9::SetupIA()
{
D3DPRIMITIVETYPE topology;
switch(m_vt.m_primclass)
{
case GS_POINT_CLASS:
topology = D3DPT_POINTLIST;
break;
case GS_LINE_CLASS:
topology = D3DPT_LINELIST;
if(PRIM->IIP == 0)
{
for(size_t i = 0, j = m_index.tail; i < j; i += 2)
{
uint32 tmp = m_index.buff[i + 0];
m_index.buff[i + 0] = m_index.buff[i + 1];
m_index.buff[i + 1] = tmp;
}
}
break;
case GS_TRIANGLE_CLASS:
topology = D3DPT_TRIANGLELIST;
if(PRIM->IIP == 0)
{
for(size_t i = 0, j = m_index.tail; i < j; i += 3)
{
uint32 tmp = m_index.buff[i + 0];
m_index.buff[i + 0] = m_index.buff[i + 2];
m_index.buff[i + 2] = tmp;
}
}
break;
case GS_SPRITE_CLASS:
topology = D3DPT_TRIANGLELIST;
// each sprite converted to quad needs twice the space
while(m_vertex.tail * 2 > m_vertex.maxcount)
{
GrowVertexBuffer();
}
// assume vertices are tightly packed and sequentially indexed (it should be the case)
if(m_vertex.next >= 2)
{
size_t count = m_vertex.next;
int i = (int)count * 2 - 4;
GSVertex* s = &m_vertex.buff[count - 2];
GSVertex* q = &m_vertex.buff[count * 2 - 4];
uint32* RESTRICT index = &m_index.buff[count * 3 - 6];
for(; i >= 0; i -= 4, s -= 2, q -= 4, index -= 6)
{
GSVertex v0 = s[0];
GSVertex v1 = s[1];
v0.RGBAQ = v1.RGBAQ;
v0.XYZ.Z = v1.XYZ.Z;
v0.FOG = v1.FOG;
q[0] = v0;
q[3] = v1;
// swap x, s, u
uint16 x = v0.XYZ.X;
v0.XYZ.X = v1.XYZ.X;
v1.XYZ.X = x;
float s = v0.ST.S;
v0.ST.S = v1.ST.S;
v1.ST.S = s;
uint16 u = v0.U;
v0.U = v1.U;
v1.U = u;
q[1] = v0;
q[2] = v1;
index[0] = i + 0;
index[1] = i + 1;
index[2] = i + 2;
index[3] = i + 1;
index[4] = i + 2;
index[5] = i + 3;
}
m_vertex.head = m_vertex.tail = m_vertex.next = count * 2;
m_index.tail = count * 3;
}
break;
default:
__assume(0);
}
GSDevice9* dev = (GSDevice9*)m_dev;
(*dev)->SetRenderState(D3DRS_SHADEMODE, PRIM->IIP ? D3DSHADE_GOURAUD : D3DSHADE_FLAT); // TODO
void* ptr = NULL;
if(dev->IAMapVertexBuffer(&ptr, sizeof(GSVertexHW9), m_vertex.next))
{
GSVertex* RESTRICT s = (GSVertex*)m_vertex.buff;
GSVertexHW9* RESTRICT d = (GSVertexHW9*)ptr;
for(uint32 i = 0; i < m_vertex.next; i++, s++, d++)
{
GSVector4 p = GSVector4(GSVector4i::load(s->XYZ.u32[0]).upl16());
if(PRIM->TME && !PRIM->FST)
{
p = p.xyxy(GSVector4((float)s->XYZ.Z, s->RGBAQ.Q));
}
else
{
p = p.xyxy(GSVector4::load((float)s->XYZ.Z));
}
GSVector4 t = GSVector4::zero();
if(PRIM->TME)
{
if(PRIM->FST)
{
if(UserHacks_WildHack && !isPackedUV_HackFlag)
{
t = GSVector4(GSVector4i::load(s->UV & 0x3FEF3FEF).upl16());
//printf("GSDX: %08X | D3D9(%d) %s\n", s->UV & 0x3FEF3FEF, m_vertex.next, i == 0 ? "*" : "");
}
else
{
t = GSVector4(GSVector4i::load(s->UV).upl16());
}
}
else
{
t = GSVector4::loadl(&s->ST);
}
}
t = t.xyxy(GSVector4::cast(GSVector4i(s->RGBAQ.u32[0], s->FOG)));
d->p = p;
d->t = t;
}
dev->IAUnmapVertexBuffer();
}
dev->IASetIndexBuffer(m_index.buff, m_index.tail);
dev->IASetPrimitiveTopology(topology);
}
void GSRendererDX::DrawPrims(GSTexture* rt, GSTexture* ds, GSTextureCache::Source* tex)
{
GSDrawingEnvironment& env = m_env;
GSDrawingContext* context = m_context;
const GSVector2i& rtsize = rt->GetSize();
const GSVector2& rtscale = rt->GetScale();
bool DATE = m_context->TEST.DATE && context->FRAME.PSM != PSM_PSMCT24;
GSTexture* rtcopy = NULL;
ASSERT(m_dev != NULL);
GSDeviceDX* dev = (GSDeviceDX*)m_dev;
if(DATE)
{
if(dev->HasStencil())
{
GSVector4 s = GSVector4(rtscale.x / rtsize.x, rtscale.y / rtsize.y);
GSVector4 o = GSVector4(-1.0f, 1.0f);
GSVector4 src = ((m_vt.m_min.p.xyxy(m_vt.m_max.p) + o.xxyy()) * s.xyxy()).sat(o.zzyy());
GSVector4 dst = src * 2.0f + o.xxxx();
GSVertexPT1 vertices[] =
{
{GSVector4(dst.x, -dst.y, 0.5f, 1.0f), GSVector2(src.x, src.y)},
{GSVector4(dst.z, -dst.y, 0.5f, 1.0f), GSVector2(src.z, src.y)},
{GSVector4(dst.x, -dst.w, 0.5f, 1.0f), GSVector2(src.x, src.w)},
{GSVector4(dst.z, -dst.w, 0.5f, 1.0f), GSVector2(src.z, src.w)},
};
dev->SetupDATE(rt, ds, vertices, m_context->TEST.DATM);
}
else
{
rtcopy = dev->CreateRenderTarget(rtsize.x, rtsize.y, false, rt->GetFormat());
// I'll use VertexTrace when I consider it more trustworthy
dev->CopyRect(rt, rtcopy, GSVector4i(rtsize).zwxy());
}
}
//
dev->BeginScene();
// om
GSDeviceDX::OMDepthStencilSelector om_dssel;
if(context->TEST.ZTE)
{
om_dssel.ztst = context->TEST.ZTST;
om_dssel.zwe = !context->ZBUF.ZMSK;
}
else
{
om_dssel.ztst = ZTST_ALWAYS;
}
if(m_fba)
{
om_dssel.fba = context->FBA.FBA;
}
GSDeviceDX::OMBlendSelector om_bsel;
if(!IsOpaque())
{
om_bsel.abe = PRIM->ABE || PRIM->AA1 && m_vt.m_primclass == GS_LINE_CLASS;
om_bsel.a = context->ALPHA.A;
om_bsel.b = context->ALPHA.B;
om_bsel.c = context->ALPHA.C;
om_bsel.d = context->ALPHA.D;
if(env.PABE.PABE)
{
if(om_bsel.a == 0 && om_bsel.b == 1 && om_bsel.c == 0 && om_bsel.d == 1)
{
// this works because with PABE alpha blending is on when alpha >= 0x80, but since the pixel shader
// cannot output anything over 0x80 (== 1.0) blending with 0x80 or turning it off gives the same result
om_bsel.abe = 0;
}
else
{
//Breath of Fire Dragon Quarter triggers this in battles. Graphics are fine though.
//ASSERT(0);
}
}
}
om_bsel.wrgba = ~GSVector4i::load((int)context->FRAME.FBMSK).eq8(GSVector4i::xffffffff()).mask();
// vs
GSDeviceDX::VSSelector vs_sel;
vs_sel.tme = PRIM->TME;
vs_sel.fst = PRIM->FST;
vs_sel.logz = dev->HasDepth32() ? 0 : m_logz ? 1 : 0;
vs_sel.rtcopy = !!rtcopy;
// The real GS appears to do no masking based on the Z buffer format and writing larger Z values
// than the buffer supports seems to be an error condition on the real GS, causing it to crash.
// We are probably receiving bad coordinates from VU1 in these cases.
if(om_dssel.ztst >= ZTST_ALWAYS && om_dssel.zwe)
{
if(context->ZBUF.PSM == PSM_PSMZ24)
{
if(m_vt.m_max.p.z > 0xffffff)
{
ASSERT(m_vt.m_min.p.z > 0xffffff);
// Fixme :Following conditional fixes some dialog frame in Wild Arms 3, but may not be what was intended.
if (m_vt.m_min.p.z > 0xffffff)
{
vs_sel.bppz = 1;
om_dssel.ztst = ZTST_ALWAYS;
}
}
}
else if(context->ZBUF.PSM == PSM_PSMZ16 || context->ZBUF.PSM == PSM_PSMZ16S)
{
if(m_vt.m_max.p.z > 0xffff)
{
ASSERT(m_vt.m_min.p.z > 0xffff); // sfex capcom logo
// Fixme : Same as above, I guess.
if (m_vt.m_min.p.z > 0xffff)
{
vs_sel.bppz = 2;
om_dssel.ztst = ZTST_ALWAYS;
}
}
}
}
GSDeviceDX::VSConstantBuffer vs_cb;
float sx = 2.0f * rtscale.x / (rtsize.x << 4);
float sy = 2.0f * rtscale.y / (rtsize.y << 4);
float ox = (float)(int)context->XYOFFSET.OFX;
float oy = (float)(int)context->XYOFFSET.OFY;
float ox2 = 2.0f * m_pixelcenter.x / rtsize.x;
float oy2 = 2.0f * m_pixelcenter.y / rtsize.y;
//This hack subtracts around half a pixel from OFX and OFY. (Cannot do this directly,
//because DX10 and DX9 have a different pixel center.)
//
//The resulting shifted output aligns better with common blending / corona / blurring effects,
//but introduces a few bad pixels on the edges.
if(rt->LikelyOffset)
{
// DX9 has pixelcenter set to 0.0, so give it some value here
if(m_pixelcenter.x == 0 && m_pixelcenter.y == 0) { ox2 = -0.0003f; oy2 = -0.0003f; }
ox2 *= rt->OffsetHack_modx;
oy2 *= rt->OffsetHack_mody;
}
vs_cb.VertexScale = GSVector4(sx, -sy, ldexpf(1, -32), 0.0f);
vs_cb.VertexOffset = GSVector4(ox * sx + ox2 + 1, -(oy * sy + oy2 + 1), 0.0f, -1.0f);
// gs
GSDeviceDX::GSSelector gs_sel;
gs_sel.iip = PRIM->IIP;
gs_sel.prim = m_vt.m_primclass;
// ps
GSDeviceDX::PSSelector ps_sel;
GSDeviceDX::PSSamplerSelector ps_ssel;
GSDeviceDX::PSConstantBuffer ps_cb;
if(DATE)
{
if(dev->HasStencil())
{
om_dssel.date = 1;
}
else
{
ps_sel.date = 1 + context->TEST.DATM;
}
}
if(env.COLCLAMP.CLAMP == 0 && /* hack */ !tex && PRIM->PRIM != GS_POINTLIST)
{
ps_sel.colclip = 1;
}
ps_sel.clr1 = om_bsel.IsCLR1();
ps_sel.fba = context->FBA.FBA;
ps_sel.aout = context->FRAME.PSM == PSM_PSMCT16 || context->FRAME.PSM == PSM_PSMCT16S || (context->FRAME.FBMSK & 0xff000000) == 0x7f000000 ? 1 : 0;
if(UserHacks_AlphaHack) ps_sel.aout = 1;
if(PRIM->FGE)
{
ps_sel.fog = 1;
ps_cb.FogColor_AREF = GSVector4::rgba32(env.FOGCOL.u32[0]) / 255;
}
if(context->TEST.ATE)
{
ps_sel.atst = context->TEST.ATST;
switch(ps_sel.atst)
{
case ATST_LESS:
ps_cb.FogColor_AREF.a = (float)((int)context->TEST.AREF - 1);
break;
case ATST_GREATER:
ps_cb.FogColor_AREF.a = (float)((int)context->TEST.AREF + 1);
break;
default:
ps_cb.FogColor_AREF.a = (float)(int)context->TEST.AREF;
break;
}
}
else
{
ps_sel.atst = ATST_ALWAYS;
}
if(tex)
{
ps_sel.wms = context->CLAMP.WMS;
ps_sel.wmt = context->CLAMP.WMT;
ps_sel.fmt = tex->m_fmt;
ps_sel.aem = env.TEXA.AEM;
ps_sel.tfx = context->TEX0.TFX;
ps_sel.tcc = context->TEX0.TCC;
ps_sel.ltf = m_filter == 2 ? m_vt.IsLinear() : m_filter;
ps_sel.rt = tex->m_target;
int w = tex->m_texture->GetWidth();
int h = tex->m_texture->GetHeight();
int tw = (int)(1 << context->TEX0.TW);
int th = (int)(1 << context->TEX0.TH);
GSVector4 WH(tw, th, w, h);
if(PRIM->FST)
{
vs_cb.TextureScale = GSVector4(1.0f / 16) / WH.xyxy();
//Maybe better?
//vs_cb.TextureScale = GSVector4(1.0f / 16) * GSVector4(tex->m_texture->GetScale()).xyxy() / WH.zwzw();
ps_sel.fst = 1;
}
ps_cb.WH = WH;
ps_cb.HalfTexel = GSVector4(-0.5f, 0.5f).xxyy() / WH.zwzw();
ps_cb.MskFix = GSVector4i(context->CLAMP.MINU, context->CLAMP.MINV, context->CLAMP.MAXU, context->CLAMP.MAXV);
GSVector4 clamp(ps_cb.MskFix);
GSVector4 ta(env.TEXA & GSVector4i::x000000ff());
ps_cb.MinMax = clamp / WH.xyxy();
ps_cb.MinF_TA = (clamp + 0.5f).xyxy(ta) / WH.xyxy(GSVector4(255, 255));
ps_ssel.tau = (context->CLAMP.WMS + 3) >> 1;
ps_ssel.tav = (context->CLAMP.WMT + 3) >> 1;
ps_ssel.ltf = ps_sel.ltf;
}
else
{
void GSRendererDX9::SetupIA(const float& sx, const float& sy)
{
D3DPRIMITIVETYPE topology;
switch(m_vt.m_primclass)
{
case GS_POINT_CLASS:
topology = D3DPT_POINTLIST;
break;
case GS_LINE_CLASS:
topology = D3DPT_LINELIST;
if(PRIM->IIP == 0)
{
for(size_t i = 0, j = m_index.tail; i < j; i += 2)
{
uint32 tmp = m_index.buff[i + 0];
m_index.buff[i + 0] = m_index.buff[i + 1];
m_index.buff[i + 1] = tmp;
}
}
break;
case GS_TRIANGLE_CLASS:
topology = D3DPT_TRIANGLELIST;
if(PRIM->IIP == 0)
{
for(size_t i = 0, j = m_index.tail; i < j; i += 3)
{
uint32 tmp = m_index.buff[i + 0];
m_index.buff[i + 0] = m_index.buff[i + 2];
m_index.buff[i + 2] = tmp;
}
}
break;
case GS_SPRITE_CLASS:
topology = D3DPT_TRIANGLELIST;
// each sprite converted to quad needs twice the space
Lines2Sprites();
break;
default:
__assume(0);
}
GSDevice9* dev = (GSDevice9*)m_dev;
(*dev)->SetRenderState(D3DRS_SHADEMODE, PRIM->IIP ? D3DSHADE_GOURAUD : D3DSHADE_FLAT); // TODO
void* ptr = NULL;
if(dev->IAMapVertexBuffer(&ptr, sizeof(GSVertexHW9), m_vertex.next))
{
GSVertex* RESTRICT s = (GSVertex*)m_vertex.buff;
GSVertexHW9* RESTRICT d = (GSVertexHW9*)ptr;
for(uint32 i = 0; i < m_vertex.next; i++, s++, d++)
{
GSVector4 p = GSVector4(GSVector4i::load(s->XYZ.u32[0]).upl16());
if(PRIM->TME && !PRIM->FST)
{
p = p.xyxy(GSVector4((float)s->XYZ.Z, s->RGBAQ.Q));
}
else
{
p = p.xyxy(GSVector4::load((float)s->XYZ.Z));
}
GSVector4 t = GSVector4::zero();
if(PRIM->TME)
{
if(PRIM->FST)
{
if(UserHacks_WildHack && !isPackedUV_HackFlag)
{
t = GSVector4(GSVector4i::load(s->UV & 0x3FEF3FEF).upl16());
//printf("GSDX: %08X | D3D9(%d) %s\n", s->UV & 0x3FEF3FEF, m_vertex.next, i == 0 ? "*" : "");
}
else
{
t = GSVector4(GSVector4i::load(s->UV).upl16());
}
}
else
{
t = GSVector4::loadl(&s->ST);
}
}
t = t.xyxy(GSVector4::cast(GSVector4i(s->RGBAQ.u32[0], s->FOG)));
d->p = p;
d->t = t;
}
dev->IAUnmapVertexBuffer();
}
dev->IASetIndexBuffer(m_index.buff, m_index.tail);
dev->IASetPrimitiveTopology(topology);
}
void GSRasterizer::DrawEdge(const GSVertexSW& v0, const GSVertexSW& v1, const GSVertexSW& dv, const GSVector4i& scissor, int orientation, int side)
{
// orientation:
// - true: |dv.p.y| > |dv.p.x|
// - false |dv.p.x| > |dv.p.y|
// side:
// - true: top/left edge
// - false: bottom/right edge
// TODO: bit slow and too much duplicated code
// TODO: inner pre-step is still missing (hardly noticable)
GSVector4 fscissor(scissor);
GSVector4 lrtb = v0.p.upl(v1.p).ceil();
if(orientation)
{
GSVector4 tbmax = lrtb.max(fscissor.yyyy());
GSVector4 tbmin = lrtb.min(fscissor.wwww());
GSVector4i tbi = GSVector4i(tbmax.zwzw(tbmin));
int top, bottom;
GSVertexSW edge, dedge;
if((dv.p >= GSVector4::zero()).mask() & 2)
{
top = tbi.extract32<0>();
bottom = tbi.extract32<3>();
if(top >= bottom) return;
edge = v0;
dedge = dv / dv.p.yyyy();
edge += dedge * (tbmax.zzzz() - edge.p.yyyy());
}
else
{
top = tbi.extract32<1>();
bottom = tbi.extract32<2>();
if(top >= bottom) return;
edge = v1;
dedge = dv / dv.p.yyyy();
edge += dedge * (tbmax.wwww() - edge.p.yyyy());
}
GSVector4i p = GSVector4i(edge.p.upl(dedge.p) * 0x10000);
int x = p.extract32<0>();
int dx = p.extract32<1>();
if(side)
{
while(1)
{
do
{
int xi = x >> 16;
int xf = x & 0xffff;
if(scissor.left <= xi && xi < scissor.right && IsOneOfMyScanlines(xi))
{
m_stats.pixels++;
edge.t.u32[3] = (0x10000 - xf) & 0xffff;
m_dsf.ssle(xi + 1, xi, top, edge);
edge.t.u32[3] = 0;
}
}
while(0);
if(++top >= bottom) break;
edge += dedge;
x += dx;
}
}
else
{
while(1)
{
do
{
int xi = (x >> 16) + 1;
int xf = x & 0xffff;
if(scissor.left <= xi && xi < scissor.right && IsOneOfMyScanlines(xi))
{
m_stats.pixels++;
edge.t.u32[3] = xf;
m_dsf.ssle(xi + 1, xi, top, edge);
edge.t.u32[3] = 0;
}
}
while(0);
if(++top >= bottom) break;
edge += dedge;
x += dx;
}
}
}
void GSRasterizer::DrawTriangleTopBottom(GSVertexSW* v, const GSVector4i& scissor)
{
GSVertexSW dv[3];
dv[0] = v[1] - v[0];
dv[1] = v[2] - v[0];
GSVertexSW longest = dv[1] * (dv[0].p / dv[1].p).yyyy() - dv[0];
int i = longest.p.upl(longest.p == GSVector4::zero()).mask();
if(i & 2) return;
i &= 1;
GSVertexSW dscan = longest * longest.p.xxxx().rcp();
m_dsf.ssp(v, dscan);
GSVector4 fscissor(scissor);
GSVector4 tb = v[0].p.upl(v[1].p).zwzw(v[1].p.upl(v[2].p)).ceil();
GSVector4 tbmax = tb.max(fscissor.yyyy());
GSVector4 tbmin = tb.min(fscissor.wwww());
GSVector4i tbi = GSVector4i(tbmax.xzyw(tbmin));
int top = tbi.extract32<0>();
int bottom = tbi.extract32<2>();
GSVertexSW& l = v[0];
GSVector4 r = v[0].p;
GSVertexSW dl = dv[i] / dv[i].p.yyyy();
GSVector4 dr = dv[1 - i].p / dv[1 - i].p.yyyy();
GSVector4 dy = tbmax.xxxx() - l.p.yyyy();
l += dl * dy;
r += dr * dy;
if(top < bottom)
{
DrawTriangleSection(top, bottom, l, dl, r, dr, dscan, fscissor);
}
top = tbi.y;
bottom = tbi.w;
if(top < bottom)
{
if(i == 0)
{
l = v[1];
dv[2] = v[2] - v[1];
dl = dv[2] / dv[2].p.yyyy();
}
else
{
r = v[1].p;
dv[2].p = v[2].p - v[1].p;
dr = dv[2].p / dv[2].p.yyyy();
}
l += dl * (tbmax.zzzz() - l.p.yyyy());
r += dr * (tbmax.zzzz() - r.yyyy());
l.p = l.p.upl(r).xyzw(l.p); // r.x => l.y
dl.p = dl.p.upl(dr).xyzw(dl.p); // dr.x => dl.y
DrawTriangleSection(top, bottom, l, dl, dscan, fscissor);
}
}
void GSRasterizer::DrawLine(const GSVertexSW* v, const GSVector4i& scissor)
{
GSVertexSW dv = v[1] - v[0];
GSVector4 dp = dv.p.abs();
if(m_dsf.ssle)
{
int i = (dp < dp.yxwz()).mask() & 1; // |x| <= |y|
GSVertexSW dscan;
dscan.p = GSVector4::zero();
dscan.t = GSVector4::zero();
dscan.c = GSVector4::zero();
m_dsf.ssp(v, dscan);
DrawEdge(v[0], v[1], dv, scissor, i, 0);
DrawEdge(v[0], v[1], dv, scissor, i, 1);
return;
}
GSVector4i dpi(dp);
if(dpi.y == 0)
{
if(dpi.x > 0)
{
// shortcut for horizontal lines
GSVector4 mask = (v[0].p > v[1].p).xxxx();
GSVertexSW l, dl;
l.p = v[0].p.blend8(v[1].p, mask);
l.t = v[0].t.blend8(v[1].t, mask);
l.c = v[0].c.blend8(v[1].c, mask);
GSVector4 r;
r = v[1].p.blend8(v[0].p, mask);
GSVector4i p(l.p);
if(scissor.top <= p.y && p.y < scissor.bottom)
{
GSVertexSW dscan = dv / dv.p.xxxx();
m_dsf.ssp(v, dscan);
l.p = l.p.upl(r).xyzw(l.p); // r.x => l.y
GSVector4 fscissor(scissor);
DrawTriangleSection(p.y, p.y + 1, l, dl, dscan, fscissor);
}
}
return;
}
int i = dpi.x > dpi.y ? 0 : 1;
GSVertexSW edge = v[0];
GSVertexSW dedge = dv / dp.v[i];
// TODO: prestep + clip with the scissor
int steps = dpi.v[i];
while(steps-- > 0)
{
DrawPoint(&edge, scissor);
edge += dedge;
}
}
