/* * Copyright (C) 2007-2009 Gabest * http://www.gabest.org * * This Program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by * the Free Software Foundation; either version 2, or (at your option) * any later version. * * This Program is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License for more details. * * You should have received a copy of the GNU General Public License * along with GNU Make; see the file COPYING. If not, write to * the Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA USA. * http://www.gnu.org/copyleft/gpl.html * */ #include "stdafx.h" #include "GSRendererDX.h" #include "GSDeviceDX.h" GSRendererDX::GSRendererDX(GSTextureCache* tc, const GSVector2& pixelcenter) : GSRendererHW(tc) , m_pixelcenter(pixelcenter) { m_logz = !!theApp.GetConfig("logz", 0); m_fba = !!theApp.GetConfig("fba", 1); UserHacks_AlphaHack = !!theApp.GetConfig("UserHacks_AlphaHack", 0) && !!theApp.GetConfig("UserHacks", 0); UserHacks_AlphaStencil = !!theApp.GetConfig("UserHacks_AlphaStencil", 0) && !!theApp.GetConfig("UserHacks", 0); UserHacks_TCOffset = !!theApp.GetConfig("UserHacks", 0) ? theApp.GetConfig("UserHacks_TCOffset", 0) : 0; UserHacks_TCO_x = (UserHacks_TCOffset & 0xFFFF) / -1000.0f; UserHacks_TCO_y = ((UserHacks_TCOffset >> 16) & 0xFFFF) / -1000.0f; } GSRendererDX::~GSRendererDX() { } void GSRendererDX::DrawPrims(GSTexture* rt, GSTexture* ds, GSTextureCache::Source* tex) { GSDrawingEnvironment& env = m_env; GSDrawingContext* context = m_context; const GSVector2i& rtsize = ds ? ds->GetSize() : rt->GetSize(); const GSVector2& rtscale = ds ? ds->GetScale() : 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 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(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 && 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; // Gregory: code is not yet ready so let's only enable it when // CRC is below the FULL level if (m_texture_shuffle && (m_crc_hack_level < 3)) { ps_sel.shuffle = 1; ps_sel.fmt = 0; const GIFRegXYOFFSET& o = m_context->XYOFFSET; GSVertex* v = &m_vertex.buff[0]; size_t count = m_vertex.next; // vertex position is 8 to 16 pixels, therefore it is the 16-31 bits of the colors int pos = (v[0].XYZ.X - o.OFX) & 0xFF; bool write_ba = (pos > 112 && pos < 136); // Read texture is 8 to 16 pixels (same as above) int tex_pos = v[0].U & 0xFF; ps_sel.read_ba = (tex_pos > 112 && tex_pos < 144); GL_INS("Color shuffle %s => %s", ps_sel.read_ba ? "BA" : "RG", write_ba ? "BA" : "RG"); // Convert the vertex info to a 32 bits color format equivalent for (size_t i = 0; i < count; i += 2) { if (write_ba) v[i].XYZ.X -= 128u; else v[i + 1].XYZ.X += 128u; if (ps_sel.read_ba) v[i].U -= 128u; else v[i + 1].U += 128u; // Height is too big (2x). int tex_offset = v[i].V & 0xF; GSVector4i offset(o.OFY, tex_offset, o.OFY, tex_offset); GSVector4i tmp(v[i].XYZ.Y, v[i].V, v[i + 1].XYZ.Y, v[i + 1].V); tmp = GSVector4i(tmp - offset).srl32(1) + offset; v[i].XYZ.Y = (uint16)tmp.x; v[i].V = (uint16)tmp.y; v[i + 1].XYZ.Y = (uint16)tmp.z; v[i + 1].V = (uint16)tmp.w; } // Please bang my head against the wall! // 1/ Reduce the frame mask to a 16 bit format const uint32& m = context->FRAME.FBMSK; uint32 fbmask = ((m >> 3) & 0x1F) | ((m >> 6) & 0x3E0) | ((m >> 9) & 0x7C00) | ((m >> 31) & 0x8000); om_bsel.wrgba = 0; // 2 Select the new mask (Please someone put SSE here) if ((fbmask & 0xFF) == 0) { if (write_ba) om_bsel.wb = 1; else om_bsel.wr = 1; } else if ((fbmask & 0xFF) != 0xFF) { #ifdef _DEBUG fprintf(stderr, "Please fix me! wb %d wr %d\n", om_bsel.wb, om_bsel.wr); #endif //ASSERT(0); } fbmask >>= 8; if ((fbmask & 0xFF) == 0) { if (write_ba) om_bsel.wa = 1; else om_bsel.wg = 1; } else if ((fbmask & 0xFF) != 0xFF) { #ifdef _DEBUG fprintf(stderr, "Please fix me! wa %d wg %d\n", om_bsel.wa, om_bsel.wg); #endif //ASSERT(0); } } else { //ps_sel.fmt = GSLocalMemory::m_psm[context->FRAME.PSM].fmt; om_bsel.wrgba = ~GSVector4i::load((int)context->FRAME.FBMSK).eq8(GSVector4i::xffffffff()).mask(); } 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; ps_sel.aout &= !ps_sel.shuffle; 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; else ps_sel.atst = ATST_ALWAYS; if (context->TEST.ATE && context->TEST.ATST > 1) ps_cb.FogColor_AREF.a = (float)context->TEST.AREF; // Destination alpha pseudo stencil hack: use a stencil operation combined with an alpha test // to only draw pixels which would cause the destination alpha test to fail in the future once. // Unfortunately this also means only drawing those pixels at all, which is why this is a hack. // The interaction with FBA in D3D9 is probably less than ideal. if (UserHacks_AlphaStencil && DATE && dev->HasStencil() && om_bsel.wa && (!context->TEST.ATE || context->TEST.ATST == 1)) { if (!context->FBA.FBA) { if (context->TEST.DATM == 0) ps_sel.atst = 5; // >= else ps_sel.atst = 2; // < ps_cb.FogColor_AREF.a = (float)0x80; } if (!(context->FBA.FBA && context->TEST.DATM == 1)) om_dssel.alpha_stencil = 1; } if(tex) { const GSLocalMemory::psm_t &psm = GSLocalMemory::m_psm[context->TEX0.PSM]; const GSLocalMemory::psm_t &cpsm = psm.pal > 0 ? GSLocalMemory::m_psm[context->TEX0.CPSM] : psm; bool bilinear = m_filter == 2 ? m_vt.IsLinear() : m_filter != 0; bool simple_sample = !tex->m_palette && cpsm.fmt == 0 && context->CLAMP.WMS < 3 && context->CLAMP.WMT < 3; // Don't force extra filtering on sprite (it creates various upscaling issue) bilinear &= !((m_vt.m_primclass == GS_SPRITE_CLASS) && m_userhacks_round_sprite_offset && !m_vt.IsLinear()); ps_sel.wms = context->CLAMP.WMS; ps_sel.wmt = context->CLAMP.WMT; if (ps_sel.shuffle) { ps_sel.fmt = 0; } else { ps_sel.fmt = tex->m_palette ? cpsm.fmt | 4 : cpsm.fmt; } ps_sel.aem = env.TEXA.AEM; ps_sel.tfx = context->TEX0.TFX; ps_sel.tcc = context->TEX0.TCC; ps_sel.ltf = bilinear && !simple_sample; ps_sel.rt = tex->m_target; ps_sel.spritehack = tex->m_spritehack_t; ps_sel.point_sampler = !(bilinear && simple_sample); 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); // TC Offset Hack ps_sel.tcoffsethack = !!UserHacks_TCOffset; ps_cb.TC_OffsetHack = GSVector4(UserHacks_TCO_x, UserHacks_TCO_y).xyxy() / WH.xyxy(); 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 = bilinear && simple_sample; } else { ps_sel.tfx = 4; } // rs GSVector4i scissor = GSVector4i(GSVector4(rtscale).xyxy() * context->scissor.in).rintersect(GSVector4i(rtsize).zwxy()); dev->OMSetRenderTargets(rt, ds, &scissor); dev->PSSetShaderResource(0, tex ? tex->m_texture : NULL); dev->PSSetShaderResource(1, tex ? tex->m_palette : NULL); dev->PSSetShaderResource(2, rtcopy); uint8 afix = context->ALPHA.FIX; SetupIA(); dev->SetupOM(om_dssel, om_bsel, afix); dev->SetupVS(vs_sel, &vs_cb); dev->SetupGS(gs_sel); dev->SetupPS(ps_sel, &ps_cb, ps_ssel); // draw if(context->TEST.DoFirstPass()) { dev->DrawIndexedPrimitive(); if (env.COLCLAMP.CLAMP == 0 && /* hack */ !tex && PRIM->PRIM != GS_POINTLIST) { GSDeviceDX::OMBlendSelector om_bselneg(om_bsel); GSDeviceDX::PSSelector ps_selneg(ps_sel); om_bselneg.negative = 1; ps_selneg.colclip = 2; dev->SetupOM(om_dssel, om_bselneg, afix); dev->SetupPS(ps_selneg, &ps_cb, ps_ssel); dev->DrawIndexedPrimitive(); dev->SetupOM(om_dssel, om_bsel, afix); } } if(context->TEST.DoSecondPass()) { ASSERT(!env.PABE.PABE); static const uint32 iatst[] = {1, 0, 5, 6, 7, 2, 3, 4}; ps_sel.atst = iatst[ps_sel.atst]; dev->SetupPS(ps_sel, &ps_cb, ps_ssel); bool z = om_dssel.zwe; bool r = om_bsel.wr; bool g = om_bsel.wg; bool b = om_bsel.wb; bool a = om_bsel.wa; switch(context->TEST.AFAIL) { case 0: z = r = g = b = a = false; break; // none case 1: z = false; break; // rgba case 2: r = g = b = a = false; break; // z case 3: z = a = false; break; // rgb default: __assume(0); } if(z || r || g || b || a) { om_dssel.zwe = z; om_bsel.wr = r; om_bsel.wg = g; om_bsel.wb = b; om_bsel.wa = a; dev->SetupOM(om_dssel, om_bsel, afix); dev->DrawIndexedPrimitive(); if (env.COLCLAMP.CLAMP == 0 && /* hack */ !tex && PRIM->PRIM != GS_POINTLIST) { GSDeviceDX::OMBlendSelector om_bselneg(om_bsel); GSDeviceDX::PSSelector ps_selneg(ps_sel); om_bselneg.negative = 1; ps_selneg.colclip = 2; dev->SetupOM(om_dssel, om_bselneg, afix); dev->SetupPS(ps_selneg, &ps_cb, ps_ssel); dev->DrawIndexedPrimitive(); } } } dev->EndScene(); dev->Recycle(rtcopy); if(om_dssel.fba) UpdateFBA(rt); }