void Renderer::MapToPlaneQuadrant(deque<Point> *vertices, ProjectionPlane plane, float min, float max) { deque<float> normX, normY; NormalizeVertices(*vertices, &normX, &normY, min, max); //Define quadrant Vector2i minQuad, maxQuad; minQuad.mX = 0; minQuad.mY = 0; maxQuad.mX = sSubwindowSize.mX - 1; maxQuad.mY = sSubwindowSize.mY - 1; //Map normalized values to quadrant long n = vertices->size(); for(int i = 0; i < n; i++) { float zX = normX[i], zY = normY[i]; int x, y; x = minQuad.mX + (zX * (maxQuad.mX - minQuad.mX)); y = minQuad.mY + (zY * (maxQuad.mY - minQuad.mY)); vertices->at(i) = Point(x,y); } }
// This code is HIGHLY unoptimized! // // It does the simplest and safest test possible: If all points of a bbox is outside a single of // our clipping planes, we reject the box. Tighter bounds would be desirable but would take more calculations. bool DrawEngineCommon::TestBoundingBox(void* control_points, int vertexCount, u32 vertType) { SimpleVertex *corners = (SimpleVertex *)(decoded + 65536 * 12); float *verts = (float *)(decoded + 65536 * 18); // Try to skip NormalizeVertices if it's pure positions. No need to bother with a vertex decoder // and a large vertex format. if ((vertType & 0xFFFFFF) == GE_VTYPE_POS_FLOAT) { // memcpy(verts, control_points, 12 * vertexCount); verts = (float *)control_points; } else if ((vertType & 0xFFFFFF) == GE_VTYPE_POS_8BIT) { const s8 *vtx = (const s8 *)control_points; for (int i = 0; i < vertexCount * 3; i++) { verts[i] = vtx[i] * (1.0f / 128.0f); } } else if ((vertType & 0xFFFFFF) == GE_VTYPE_POS_16BIT) { const s16 *vtx = (const s16*)control_points; for (int i = 0; i < vertexCount * 3; i++) { verts[i] = vtx[i] * (1.0f / 32768.0f); } } else { // Simplify away bones and morph before proceeding u8 *temp_buffer = decoded + 65536 * 24; NormalizeVertices((u8 *)corners, temp_buffer, (u8 *)control_points, 0, vertexCount, vertType); // Special case for float positions only. const float *ctrl = (const float *)control_points; for (int i = 0; i < vertexCount; i++) { verts[i * 3] = corners[i].pos.x; verts[i * 3 + 1] = corners[i].pos.y; verts[i * 3 + 2] = corners[i].pos.z; } } Plane planes[6]; float world[16]; float view[16]; float worldview[16]; float worldviewproj[16]; ConvertMatrix4x3To4x4(world, gstate.worldMatrix); ConvertMatrix4x3To4x4(view, gstate.viewMatrix); Matrix4ByMatrix4(worldview, world, view); Matrix4ByMatrix4(worldviewproj, worldview, gstate.projMatrix); PlanesFromMatrix(worldviewproj, planes); for (int plane = 0; plane < 6; plane++) { int inside = 0; int out = 0; for (int i = 0; i < vertexCount; i++) { // Here we can test against the frustum planes! float value = planes[plane].Test(verts + i * 3); if (value < 0) out++; else inside++; } if (inside == 0) { // All out return false; } // Any out. For testing that the planes are in the right locations. // if (out != 0) return false; } return true; }
void DrawEngineCommon::SubmitBezier(const void *control_points, const void *indices, int tess_u, int tess_v, int count_u, int count_v, GEPatchPrimType prim_type, bool computeNormals, bool patchFacing, u32 vertType) { PROFILE_THIS_SCOPE("bezier"); DispatchFlush(); // TODO: Verify correct functionality with < 4. if (count_u < 4 || count_v < 4) return; u16 index_lower_bound = 0; u16 index_upper_bound = count_u * count_v - 1; bool indices_16bit = (vertType & GE_VTYPE_IDX_MASK) == GE_VTYPE_IDX_16BIT; const u8* indices8 = (const u8*)indices; const u16* indices16 = (const u16*)indices; if (indices) GetIndexBounds(indices, count_u*count_v, vertType, &index_lower_bound, &index_upper_bound); // Simplify away bones and morph before proceeding // There are normally not a lot of control points so just splitting decoded should be reasonably safe, although not great. SimpleVertex *simplified_control_points = (SimpleVertex *)(decoded + 65536 * 12); u8 *temp_buffer = decoded + 65536 * 18; u32 origVertType = vertType; vertType = NormalizeVertices((u8 *)simplified_control_points, temp_buffer, (u8 *)control_points, index_lower_bound, index_upper_bound, vertType); VertexDecoder *vdecoder = GetVertexDecoder(vertType); int vertexSize = vdecoder->VertexSize(); if (vertexSize != sizeof(SimpleVertex)) { ERROR_LOG(G3D, "Something went really wrong, vertex size: %i vs %i", vertexSize, (int)sizeof(SimpleVertex)); } // Bezier patches share less control points than spline patches. Otherwise they are pretty much the same (except bezier don't support the open/close thing) int num_patches_u = (count_u - 1) / 3; int num_patches_v = (count_v - 1) / 3; BezierPatch* patches = new BezierPatch[num_patches_u * num_patches_v]; for (int patch_u = 0; patch_u < num_patches_u; patch_u++) { for (int patch_v = 0; patch_v < num_patches_v; patch_v++) { BezierPatch& patch = patches[patch_u + patch_v * num_patches_u]; for (int point = 0; point < 16; ++point) { int idx = (patch_u * 3 + point % 4) + (patch_v * 3 + point / 4) * count_u; if (indices) patch.points[point] = simplified_control_points + (indices_16bit ? indices16[idx] : indices8[idx]); else patch.points[point] = simplified_control_points + idx; } patch.u_index = patch_u * 3; patch.v_index = patch_v * 3; patch.index = patch_v * num_patches_u + patch_u; patch.primType = prim_type; patch.computeNormals = computeNormals; patch.patchFacing = patchFacing; } } int count = 0; u8 *dest = splineBuffer; // Simple approximation of the real tesselation factor. // We shouldn't really split up into separate 4x4 patches, instead we should do something that works // like the splines, so we subdivide across the whole "mega-patch". if (num_patches_u == 0) num_patches_u = 1; if (num_patches_v == 0) num_patches_v = 1; if (tess_u < 4) tess_u = 4; if (tess_v < 4) tess_v = 4; u16 *inds = quadIndices_; int maxVertices = SPLINE_BUFFER_SIZE / vertexSize; for (int patch_idx = 0; patch_idx < num_patches_u*num_patches_v; ++patch_idx) { BezierPatch& patch = patches[patch_idx]; TesselateBezierPatch(dest, inds, count, tess_u, tess_v, patch, origVertType, maxVertices); } delete[] patches; u32 vertTypeWithIndex16 = (vertType & ~GE_VTYPE_IDX_MASK) | GE_VTYPE_IDX_16BIT; UVScale prevUVScale; if (g_Config.bPrescaleUV) { // We scaled during Normalize already so let's turn it off when drawing. prevUVScale = gstate_c.uv; gstate_c.uv.uScale = 1.0f; gstate_c.uv.vScale = 1.0f; gstate_c.uv.uOff = 0; gstate_c.uv.vOff = 0; } int bytesRead; DispatchSubmitPrim(splineBuffer, quadIndices_, primType[prim_type], count, vertTypeWithIndex16, &bytesRead); DispatchFlush(); if (g_Config.bPrescaleUV) { gstate_c.uv = prevUVScale; } }
void TransformDrawEngine::SubmitBezier(void* control_points, void* indices, int count_u, int count_v, GEPatchPrimType prim_type, u32 vertType) { Flush(); if (prim_type != GE_PATCHPRIM_TRIANGLES) { // Only triangles supported! return; } u16 index_lower_bound = 0; u16 index_upper_bound = count_u * count_v - 1; bool indices_16bit = (vertType & GE_VTYPE_IDX_MASK) == GE_VTYPE_IDX_16BIT; const u8* indices8 = (const u8*)indices; const u16* indices16 = (const u16*)indices; if (indices) GetIndexBounds(indices, count_u*count_v, vertType, &index_lower_bound, &index_upper_bound); // Simplify away bones and morph before proceeding SimpleVertex *simplified_control_points = (SimpleVertex *)(decoded + 65536 * 12); u8 *temp_buffer = decoded + 65536 * 24; u32 origVertType = vertType; vertType = NormalizeVertices((u8 *)simplified_control_points, temp_buffer, (u8 *)control_points, index_lower_bound, index_upper_bound, vertType); VertexDecoder *vdecoder = GetVertexDecoder(vertType); int vertexSize = vdecoder->VertexSize(); if (vertexSize != sizeof(SimpleVertex)) { ERROR_LOG(G3D, "Something went really wrong, vertex size: %i vs %i", vertexSize, (int)sizeof(SimpleVertex)); } const DecVtxFormat& vtxfmt = vdecoder->GetDecVtxFmt(); // Bezier patches share less control points than spline patches. Otherwise they are pretty much the same (except bezier don't support the open/close thing) int num_patches_u = (count_u - 1) / 3; int num_patches_v = (count_v - 1) / 3; BezierPatch* patches = new BezierPatch[num_patches_u * num_patches_v]; for (int patch_u = 0; patch_u < num_patches_u; patch_u++) { for (int patch_v = 0; patch_v < num_patches_v; patch_v++) { BezierPatch& patch = patches[patch_u + patch_v * num_patches_u]; for (int point = 0; point < 16; ++point) { int idx = (patch_u * 3 + point%4) + (patch_v * 3 + point/4) * count_u; if (indices) patch.points[point] = simplified_control_points + (indices_16bit ? indices16[idx] : indices8[idx]); else patch.points[point] = simplified_control_points + idx; } patch.u_index = patch_u * 3; patch.v_index = patch_v * 3; } } u8 *decoded2 = decoded + 65536 * 36; int count = 0; u8 *dest = decoded2; // Simple approximation of the real tesselation factor. // We shouldn't really split up into separate 4x4 patches, instead we should do something that works // like the splines, so we subdivide across the whole "mega-patch". if (num_patches_u == 0) num_patches_u = 1; if (num_patches_v == 0) num_patches_v = 1; int tess_u = gstate.getPatchDivisionU() / num_patches_u; int tess_v = gstate.getPatchDivisionV() / num_patches_v; if (tess_u < 4) tess_u = 4; if (tess_v < 4) tess_v = 4; for (int patch_idx = 0; patch_idx < num_patches_u*num_patches_v; ++patch_idx) { BezierPatch& patch = patches[patch_idx]; TesselateBezierPatch(dest, count, tess_u, tess_v, patch, origVertType); } delete[] patches; u32 vertTypeWithIndex16 = (vertType & ~GE_VTYPE_IDX_MASK) | GE_VTYPE_IDX_16BIT; UVScale prevUVScale; if (g_Config.bPrescaleUV) { // We scaled during Normalize already so let's turn it off when drawing. prevUVScale = gstate_c.uv; gstate_c.uv.uScale = 1.0f; gstate_c.uv.vScale = 1.0f; gstate_c.uv.uOff = 0; gstate_c.uv.vOff = 0; } SubmitPrim(decoded2, quadIndices_, GE_PRIM_TRIANGLES, count, vertTypeWithIndex16, 0); Flush(); if (g_Config.bPrescaleUV) { gstate_c.uv = prevUVScale; } }
void DrawEngineCommon::SubmitSpline(const void *control_points, const void *indices, int tess_u, int tess_v, int count_u, int count_v, int type_u, int type_v, GEPatchPrimType prim_type, bool computeNormals, bool patchFacing, u32 vertType) { PROFILE_THIS_SCOPE("spline"); DispatchFlush(); // TODO: Verify correct functionality with < 4. if (count_u < 4 || count_v < 4) return; u16 index_lower_bound = 0; u16 index_upper_bound = count_u * count_v - 1; bool indices_16bit = (vertType & GE_VTYPE_IDX_MASK) == GE_VTYPE_IDX_16BIT; const u8* indices8 = (const u8*)indices; const u16* indices16 = (const u16*)indices; if (indices) GetIndexBounds(indices, count_u*count_v, vertType, &index_lower_bound, &index_upper_bound); // Simplify away bones and morph before proceeding SimpleVertex *simplified_control_points = (SimpleVertex *)(decoded + 65536 * 12); u8 *temp_buffer = decoded + 65536 * 18; u32 origVertType = vertType; vertType = NormalizeVertices((u8 *)simplified_control_points, temp_buffer, (u8 *)control_points, index_lower_bound, index_upper_bound, vertType); VertexDecoder *vdecoder = GetVertexDecoder(vertType); int vertexSize = vdecoder->VertexSize(); if (vertexSize != sizeof(SimpleVertex)) { ERROR_LOG(G3D, "Something went really wrong, vertex size: %i vs %i", vertexSize, (int)sizeof(SimpleVertex)); } // TODO: Do something less idiotic to manage this buffer SimpleVertex **points = new SimpleVertex *[count_u * count_v]; // Make an array of pointers to the control points, to get rid of indices. for (int idx = 0; idx < count_u * count_v; idx++) { if (indices) points[idx] = simplified_control_points + (indices_16bit ? indices16[idx] : indices8[idx]); else points[idx] = simplified_control_points + idx; } int count = 0; u8 *dest = splineBuffer; SplinePatchLocal patch; patch.tess_u = tess_u; patch.tess_v = tess_v; patch.type_u = type_u; patch.type_v = type_v; patch.count_u = count_u; patch.count_v = count_v; patch.points = points; patch.computeNormals = computeNormals; patch.primType = prim_type; patch.patchFacing = patchFacing; int maxVertexCount = SPLINE_BUFFER_SIZE / vertexSize; TesselateSplinePatch(dest, quadIndices_, count, patch, origVertType, maxVertexCount); delete[] points; u32 vertTypeWithIndex16 = (vertType & ~GE_VTYPE_IDX_MASK) | GE_VTYPE_IDX_16BIT; UVScale prevUVScale; if (g_Config.bPrescaleUV) { // We scaled during Normalize already so let's turn it off when drawing. prevUVScale = gstate_c.uv; gstate_c.uv.uScale = 1.0f; gstate_c.uv.vScale = 1.0f; gstate_c.uv.uOff = 0; gstate_c.uv.vOff = 0; } int bytesRead; DispatchSubmitPrim(splineBuffer, quadIndices_, primType[prim_type], count, vertTypeWithIndex16, &bytesRead); DispatchFlush(); if (g_Config.bPrescaleUV) { gstate_c.uv = prevUVScale; } }
u32 TransformDrawEngine::NormalizeVertices(u8 *outPtr, u8 *bufPtr, const u8 *inPtr, int lowerBound, int upperBound, u32 vertType) { const u32 vertTypeID = (vertType & 0xFFFFFF) | (gstate.getUVGenMode() << 24); VertexDecoder *dec = GetVertexDecoder(vertTypeID); return NormalizeVertices(outPtr, bufPtr, inPtr, dec, lowerBound, upperBound, vertType); }
void TransformDrawEngine::SubmitSpline(void* control_points, void* indices, int count_u, int count_v, int type_u, int type_v, GEPatchPrimType prim_type, u32 vertType) { Flush(); if (prim_type != GE_PATCHPRIM_TRIANGLES) { // Only triangles supported! return; } u16 index_lower_bound = 0; u16 index_upper_bound = count_u * count_v - 1; bool indices_16bit = (vertType & GE_VTYPE_IDX_MASK) == GE_VTYPE_IDX_16BIT; const u8* indices8 = (const u8*)indices; const u16* indices16 = (const u16*)indices; if (indices) GetIndexBounds(indices, count_u*count_v, vertType, &index_lower_bound, &index_upper_bound); // Simplify away bones and morph before proceeding SimpleVertex *simplified_control_points = (SimpleVertex *)(decoded + 65536 * 12); u8 *temp_buffer = decoded + 65536 * 24; u32 origVertType = vertType; vertType = NormalizeVertices((u8 *)simplified_control_points, temp_buffer, (u8 *)control_points, index_lower_bound, index_upper_bound, vertType); VertexDecoder *vdecoder = GetVertexDecoder(vertType); int vertexSize = vdecoder->VertexSize(); if (vertexSize != sizeof(SimpleVertex)) { ERROR_LOG(G3D, "Something went really wrong, vertex size: %i vs %i", vertexSize, (int)sizeof(SimpleVertex)); } const DecVtxFormat& vtxfmt = vdecoder->GetDecVtxFmt(); // TODO: Do something less idiotic to manage this buffer SimpleVertex **points = new SimpleVertex *[count_u * count_v]; // Make an array of pointers to the control points, to get rid of indices. for (int idx = 0; idx < count_u * count_v; idx++) { if (indices) points[idx] = simplified_control_points + (indices_16bit ? indices16[idx] : indices8[idx]); else points[idx] = simplified_control_points + idx; } u8 *decoded2 = decoded + 65536 * 36; int count = 0; u8 *dest = decoded2; SplinePatchLocal patch; patch.type_u = type_u; patch.type_v = type_v; patch.count_u = count_u; patch.count_v = count_v; patch.points = points; TesselateSplinePatch(dest, count, patch, origVertType); delete[] points; u32 vertTypeWithIndex16 = (vertType & ~GE_VTYPE_IDX_MASK) | GE_VTYPE_IDX_16BIT; UVScale prevUVScale; if (g_Config.bPrescaleUV) { // We scaled during Normalize already so let's turn it off when drawing. prevUVScale = gstate_c.uv; gstate_c.uv.uScale = 1.0f; gstate_c.uv.vScale = 1.0f; gstate_c.uv.uOff = 0; gstate_c.uv.vOff = 0; } SubmitPrim(decoded2, quadIndices_, GE_PRIM_TRIANGLES, count, vertTypeWithIndex16, 0); Flush(); if (g_Config.bPrescaleUV) { gstate_c.uv = prevUVScale; } }
// TODO: This probably is not the best interface. bool DrawEngineCommon::GetCurrentSimpleVertices(int count, std::vector<GPUDebugVertex> &vertices, std::vector<u16> &indices) { // This is always for the current vertices. u16 indexLowerBound = 0; u16 indexUpperBound = count - 1; if (!Memory::IsValidAddress(gstate_c.vertexAddr)) return false; bool savedVertexFullAlpha = gstate_c.vertexFullAlpha; if ((gstate.vertType & GE_VTYPE_IDX_MASK) != GE_VTYPE_IDX_NONE) { const u8 *inds = Memory::GetPointer(gstate_c.indexAddr); const u16 *inds16 = (const u16 *)inds; if (inds) { GetIndexBounds(inds, count, gstate.vertType, &indexLowerBound, &indexUpperBound); indices.resize(count); switch (gstate.vertType & GE_VTYPE_IDX_MASK) { case GE_VTYPE_IDX_16BIT: for (int i = 0; i < count; ++i) { indices[i] = inds16[i]; } break; case GE_VTYPE_IDX_8BIT: for (int i = 0; i < count; ++i) { indices[i] = inds[i]; } break; default: return false; } } else { indices.clear(); } } else { indices.clear(); } static std::vector<u32> temp_buffer; static std::vector<SimpleVertex> simpleVertices; temp_buffer.resize(std::max((int)indexUpperBound, 8192) * 128 / sizeof(u32)); simpleVertices.resize(indexUpperBound + 1); NormalizeVertices((u8 *)(&simpleVertices[0]), (u8 *)(&temp_buffer[0]), Memory::GetPointer(gstate_c.vertexAddr), indexLowerBound, indexUpperBound, gstate.vertType); float world[16]; float view[16]; float worldview[16]; float worldviewproj[16]; ConvertMatrix4x3To4x4(world, gstate.worldMatrix); ConvertMatrix4x3To4x4(view, gstate.viewMatrix); Matrix4ByMatrix4(worldview, world, view); Matrix4ByMatrix4(worldviewproj, worldview, gstate.projMatrix); vertices.resize(indexUpperBound + 1); for (int i = indexLowerBound; i <= indexUpperBound; ++i) { const SimpleVertex &vert = simpleVertices[i]; if (gstate.isModeThrough()) { if (gstate.vertType & GE_VTYPE_TC_MASK) { vertices[i].u = vert.uv[0]; vertices[i].v = vert.uv[1]; } else { vertices[i].u = 0.0f; vertices[i].v = 0.0f; } vertices[i].x = vert.pos.x; vertices[i].y = vert.pos.y; vertices[i].z = vert.pos.z; if (gstate.vertType & GE_VTYPE_COL_MASK) { memcpy(vertices[i].c, vert.color, sizeof(vertices[i].c)); } else { memset(vertices[i].c, 0, sizeof(vertices[i].c)); } } else { float clipPos[4]; Vec3ByMatrix44(clipPos, vert.pos.AsArray(), worldviewproj); Vec3f screenPos = ClipToScreen(clipPos); Vec3f drawPos = ScreenToDrawing(screenPos); if (gstate.vertType & GE_VTYPE_TC_MASK) { vertices[i].u = vert.uv[0] * (float)gstate.getTextureWidth(0); vertices[i].v = vert.uv[1] * (float)gstate.getTextureHeight(0); } else { vertices[i].u = 0.0f; vertices[i].v = 0.0f; } vertices[i].x = drawPos.x; vertices[i].y = drawPos.y; vertices[i].z = drawPos.z; if (gstate.vertType & GE_VTYPE_COL_MASK) { memcpy(vertices[i].c, vert.color, sizeof(vertices[i].c)); } else { memset(vertices[i].c, 0, sizeof(vertices[i].c)); } } } gstate_c.vertexFullAlpha = savedVertexFullAlpha; return true; }
void DrawEngineCommon::SubmitBezier(const void *control_points, const void *indices, int tess_u, int tess_v, int count_u, int count_v, GEPatchPrimType prim_type, bool computeNormals, bool patchFacing, u32 vertType, int *bytesRead) { PROFILE_THIS_SCOPE("bezier"); DispatchFlush(); u16 index_lower_bound = 0; u16 index_upper_bound = count_u * count_v - 1; IndexConverter idxConv(vertType, indices); if (indices) GetIndexBounds(indices, count_u*count_v, vertType, &index_lower_bound, &index_upper_bound); VertexDecoder *origVDecoder = GetVertexDecoder((vertType & 0xFFFFFF) | (gstate.getUVGenMode() << 24)); *bytesRead = count_u * count_v * origVDecoder->VertexSize(); // Real hardware seems to draw nothing when given < 4 either U or V. // This would result in num_patches_u / num_patches_v being 0. if (count_u < 4 || count_v < 4) { return; } // Simplify away bones and morph before proceeding // There are normally not a lot of control points so just splitting decoded should be reasonably safe, although not great. SimpleVertex *simplified_control_points = (SimpleVertex *)(decoded + 65536 * 12); u8 *temp_buffer = decoded + 65536 * 18; u32 origVertType = vertType; vertType = NormalizeVertices((u8 *)simplified_control_points, temp_buffer, (u8 *)control_points, index_lower_bound, index_upper_bound, vertType); VertexDecoder *vdecoder = GetVertexDecoder(vertType); int vertexSize = vdecoder->VertexSize(); if (vertexSize != sizeof(SimpleVertex)) { ERROR_LOG(G3D, "Something went really wrong, vertex size: %i vs %i", vertexSize, (int)sizeof(SimpleVertex)); } float *pos = (float*)(decoded + 65536 * 18); // Size 4 float float *tex = pos + count_u * count_v * 4; // Size 4 float float *col = tex + count_u * count_v * 4; // Size 4 float const bool hasColor = (origVertType & GE_VTYPE_COL_MASK) != 0; const bool hasTexCoords = (origVertType & GE_VTYPE_TC_MASK) != 0; // Bezier patches share less control points than spline patches. Otherwise they are pretty much the same (except bezier don't support the open/close thing) int num_patches_u = (count_u - 1) / 3; int num_patches_v = (count_v - 1) / 3; BezierPatch *patches = nullptr; if (g_Config.bHardwareTessellation && g_Config.bHardwareTransform && !g_Config.bSoftwareRendering) { int posStride, texStride, colStride; tessDataTransfer->PrepareBuffers(pos, tex, col, posStride, texStride, colStride, count_u * count_v, hasColor, hasTexCoords); float *p = pos; float *t = tex; float *c = col; for (int idx = 0; idx < count_u * count_v; idx++) { SimpleVertex *point = simplified_control_points + (indices ? idxConv.convert(idx) : idx); memcpy(p, point->pos.AsArray(), 3 * sizeof(float)); p += posStride; if (hasTexCoords) { memcpy(t, point->uv, 2 * sizeof(float)); t += texStride; } if (hasColor) { memcpy(c, Vec4f::FromRGBA(point->color_32).AsArray(), 4 * sizeof(float)); c += colStride; } } if (!hasColor) { SimpleVertex *point = simplified_control_points + (indices ? idxConv.convert(0) : 0); memcpy(col, Vec4f::FromRGBA(point->color_32).AsArray(), 4 * sizeof(float)); } } else { patches = new BezierPatch[num_patches_u * num_patches_v]; for (int patch_u = 0; patch_u < num_patches_u; patch_u++) { for (int patch_v = 0; patch_v < num_patches_v; patch_v++) { BezierPatch& patch = patches[patch_u + patch_v * num_patches_u]; for (int point = 0; point < 16; ++point) { int idx = (patch_u * 3 + point % 4) + (patch_v * 3 + point / 4) * count_u; patch.points[point] = simplified_control_points + (indices ? idxConv.convert(idx) : idx); } patch.u_index = patch_u * 3; patch.v_index = patch_v * 3; patch.index = patch_v * num_patches_u + patch_u; patch.primType = prim_type; patch.computeNormals = computeNormals; patch.patchFacing = patchFacing; } } } int count = 0; u8 *dest = splineBuffer; // We shouldn't really split up into separate 4x4 patches, instead we should do something that works // like the splines, so we subdivide across the whole "mega-patch". // If specified as 0, uses 1. if (tess_u < 1) { tess_u = 1; } if (tess_v < 1) { tess_v = 1; } u16 *inds = quadIndices_; if (g_Config.bHardwareTessellation && g_Config.bHardwareTransform && !g_Config.bSoftwareRendering) { tessDataTransfer->SendDataToShader(pos, tex, col, count_u * count_v, hasColor, hasTexCoords); TessellateBezierPatchHardware(dest, inds, count, tess_u, tess_v, prim_type); numPatches = num_patches_u * num_patches_v; } else { int maxVertices = SPLINE_BUFFER_SIZE / vertexSize; // Downsample until it fits, in case crazy tessellation factors are sent. while ((tess_u + 1) * (tess_v + 1) * num_patches_u * num_patches_v > maxVertices) { tess_u /= 2; tess_v /= 2; } for (int patch_idx = 0; patch_idx < num_patches_u*num_patches_v; ++patch_idx) { const BezierPatch &patch = patches[patch_idx]; TessellateBezierPatch(dest, inds, count, tess_u, tess_v, patch, origVertType); } delete[] patches; } u32 vertTypeWithIndex16 = (vertType & ~GE_VTYPE_IDX_MASK) | GE_VTYPE_IDX_16BIT; UVScale prevUVScale; if (origVertType & GE_VTYPE_TC_MASK) { // We scaled during Normalize already so let's turn it off when drawing. prevUVScale = gstate_c.uv; gstate_c.uv.uScale = 1.0f; gstate_c.uv.vScale = 1.0f; gstate_c.uv.uOff = 0; gstate_c.uv.vOff = 0; } uint32_t vertTypeID = GetVertTypeID(vertTypeWithIndex16, gstate.getUVGenMode()); int generatedBytesRead; DispatchSubmitPrim(splineBuffer, quadIndices_, primType[prim_type], count, vertTypeID, &generatedBytesRead); DispatchFlush(); if (origVertType & GE_VTYPE_TC_MASK) { gstate_c.uv = prevUVScale; } }
void DrawEngineCommon::SubmitSpline(const void *control_points, const void *indices, int tess_u, int tess_v, int count_u, int count_v, int type_u, int type_v, GEPatchPrimType prim_type, bool computeNormals, bool patchFacing, u32 vertType, int *bytesRead) { PROFILE_THIS_SCOPE("spline"); DispatchFlush(); u16 index_lower_bound = 0; u16 index_upper_bound = count_u * count_v - 1; IndexConverter idxConv(vertType, indices); if (indices) GetIndexBounds(indices, count_u * count_v, vertType, &index_lower_bound, &index_upper_bound); VertexDecoder *origVDecoder = GetVertexDecoder((vertType & 0xFFFFFF) | (gstate.getUVGenMode() << 24)); *bytesRead = count_u * count_v * origVDecoder->VertexSize(); // Real hardware seems to draw nothing when given < 4 either U or V. if (count_u < 4 || count_v < 4) { return; } // Simplify away bones and morph before proceeding SimpleVertex *simplified_control_points = (SimpleVertex *)(decoded + 65536 * 12); u8 *temp_buffer = decoded + 65536 * 18; u32 origVertType = vertType; vertType = NormalizeVertices((u8 *)simplified_control_points, temp_buffer, (u8 *)control_points, index_lower_bound, index_upper_bound, vertType); VertexDecoder *vdecoder = GetVertexDecoder(vertType); int vertexSize = vdecoder->VertexSize(); if (vertexSize != sizeof(SimpleVertex)) { ERROR_LOG(G3D, "Something went really wrong, vertex size: %i vs %i", vertexSize, (int)sizeof(SimpleVertex)); } // TODO: Do something less idiotic to manage this buffer SimpleVertex **points = new SimpleVertex *[count_u * count_v]; // Make an array of pointers to the control points, to get rid of indices. for (int idx = 0; idx < count_u * count_v; idx++) { points[idx] = simplified_control_points + (indices ? idxConv.convert(idx) : idx); } int count = 0; u8 *dest = splineBuffer; SplinePatchLocal patch; patch.tess_u = tess_u; patch.tess_v = tess_v; patch.type_u = type_u; patch.type_v = type_v; patch.count_u = count_u; patch.count_v = count_v; patch.points = points; patch.computeNormals = computeNormals; patch.primType = prim_type; patch.patchFacing = patchFacing; if (g_Config.bHardwareTessellation && g_Config.bHardwareTransform && !g_Config.bSoftwareRendering) { float *pos = (float*)(decoded + 65536 * 18); // Size 4 float float *tex = pos + count_u * count_v * 4; // Size 4 float float *col = tex + count_u * count_v * 4; // Size 4 float const bool hasColor = (origVertType & GE_VTYPE_COL_MASK) != 0; const bool hasTexCoords = (origVertType & GE_VTYPE_TC_MASK) != 0; int posStride, texStride, colStride; tessDataTransfer->PrepareBuffers(pos, tex, col, posStride, texStride, colStride, count_u * count_v, hasColor, hasTexCoords); float *p = pos; float *t = tex; float *c = col; for (int idx = 0; idx < count_u * count_v; idx++) { memcpy(p, points[idx]->pos.AsArray(), 3 * sizeof(float)); p += posStride; if (hasTexCoords) { memcpy(t, points[idx]->uv, 2 * sizeof(float)); t += texStride; } if (hasColor) { memcpy(c, Vec4f::FromRGBA(points[idx]->color_32).AsArray(), 4 * sizeof(float)); c += colStride; } } if (!hasColor) memcpy(col, Vec4f::FromRGBA(points[0]->color_32).AsArray(), 4 * sizeof(float)); tessDataTransfer->SendDataToShader(pos, tex, col, count_u * count_v, hasColor, hasTexCoords); TessellateSplinePatchHardware(dest, quadIndices_, count, patch); numPatches = (count_u - 3) * (count_v - 3); } else { int maxVertexCount = SPLINE_BUFFER_SIZE / vertexSize; TessellateSplinePatch(dest, quadIndices_, count, patch, origVertType, maxVertexCount); } delete[] points; u32 vertTypeWithIndex16 = (vertType & ~GE_VTYPE_IDX_MASK) | GE_VTYPE_IDX_16BIT; UVScale prevUVScale; if ((origVertType & GE_VTYPE_TC_MASK) != 0) { // We scaled during Normalize already so let's turn it off when drawing. prevUVScale = gstate_c.uv; gstate_c.uv.uScale = 1.0f; gstate_c.uv.vScale = 1.0f; gstate_c.uv.uOff = 0.0f; gstate_c.uv.vOff = 0.0f; } uint32_t vertTypeID = GetVertTypeID(vertTypeWithIndex16, gstate.getUVGenMode()); int generatedBytesRead; DispatchSubmitPrim(splineBuffer, quadIndices_, primType[prim_type], count, vertTypeID, &generatedBytesRead); DispatchFlush(); if ((origVertType & GE_VTYPE_TC_MASK) != 0) { gstate_c.uv = prevUVScale; } }
void TransformDrawEngine::SubmitBezier(void* control_points, void* indices, int count_u, int count_v, GEPatchPrimType prim_type, u32 vertType) { Flush(); if (prim_type != GE_PATCHPRIM_TRIANGLES) { // Only triangles supported! return; } u16 index_lower_bound = 0; u16 index_upper_bound = count_u * count_v - 1; bool indices_16bit = (vertType & GE_VTYPE_IDX_MASK) == GE_VTYPE_IDX_16BIT; const u8* indices8 = (const u8*)indices; const u16* indices16 = (const u16*)indices; if (indices) GetIndexBounds(indices, count_u*count_v, vertType, &index_lower_bound, &index_upper_bound); // Simplify away bones and morph before proceeding SimpleVertex *simplified_control_points = (SimpleVertex *)(decoded + 65536 * 12); u8 *temp_buffer = decoded + 65536 * 24; u32 origVertType = vertType; vertType = NormalizeVertices((u8 *)simplified_control_points, temp_buffer, (u8 *)control_points, index_lower_bound, index_upper_bound, vertType); VertexDecoder *vdecoder = GetVertexDecoder(vertType); int vertexSize = vdecoder->VertexSize(); if (vertexSize != sizeof(SimpleVertex)) { ERROR_LOG(G3D, "Something went really wrong, vertex size: %i vs %i", vertexSize, (int)sizeof(SimpleVertex)); } const DecVtxFormat& vtxfmt = vdecoder->GetDecVtxFmt(); // Bezier patches share less control points than spline patches. Otherwise they are pretty much the same (except bezier don't support the open/close thing) int num_patches_u = (count_u - 1) / 3; int num_patches_v = (count_v - 1) / 3; BezierPatch* patches = new BezierPatch[num_patches_u * num_patches_v]; for (int patch_u = 0; patch_u < num_patches_u; patch_u++) { for (int patch_v = 0; patch_v < num_patches_v; patch_v++) { BezierPatch& patch = patches[patch_u + patch_v * num_patches_u]; for (int point = 0; point < 16; ++point) { int idx = (patch_u * 3 + point%4) + (patch_v * 3 + point/4) * count_u; if (indices) patch.points[point] = simplified_control_points + (indices_16bit ? indices16[idx] : indices8[idx]); else patch.points[point] = simplified_control_points + idx; } patch.u_index = patch_u * 3; patch.v_index = patch_v * 3; } } u8 *decoded2 = decoded + 65536 * 36; int count = 0; u8 *dest = decoded2; for (int patch_idx = 0; patch_idx < num_patches_u*num_patches_v; ++patch_idx) { BezierPatch& patch = patches[patch_idx]; TesselateBezierPatch(dest, count, patch, origVertType); } delete[] patches; u32 vertTypeWithIndex16 = (vertType & ~GE_VTYPE_IDX_MASK) | GE_VTYPE_IDX_16BIT; SubmitPrim(decoded2, quadIndices_, GE_PRIM_TRIANGLES, count, vertTypeWithIndex16, -1, 0); Flush(); }
u32 TransformDrawEngine::NormalizeVertices(u8 *outPtr, u8 *bufPtr, const u8 *inPtr, int lowerBound, int upperBound, u32 vertType) { VertexDecoder *dec = GetVertexDecoder(vertType); return NormalizeVertices(outPtr, bufPtr, inPtr, dec, lowerBound, upperBound, vertType); }