VertexData TransformUnit::ReadVertex(VertexReader& vreader)
{
VertexData vertex;
float pos[3];
// VertexDecoder normally scales z, but we want it unscaled.
vreader.ReadPosThroughZ16(pos);
if (!gstate.isModeClear() && gstate.isTextureMapEnabled() && vreader.hasUV()) {
float uv[2];
vreader.ReadUV(uv);
vertex.texturecoords = Vec2<float>(uv[0], uv[1]);
}
if (vreader.hasNormal()) {
float normal[3];
vreader.ReadNrm(normal);
vertex.normal = Vec3<float>(normal[0], normal[1], normal[2]);
if (gstate.areNormalsReversed())
vertex.normal = -vertex.normal;
}
if (vertTypeIsSkinningEnabled(gstate.vertType) && !gstate.isModeThrough()) {
float W[8] = { 1.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f };
vreader.ReadWeights(W);
Vec3<float> tmppos(0.f, 0.f, 0.f);
Vec3<float> tmpnrm(0.f, 0.f, 0.f);
for (int i = 0; i < vertTypeGetNumBoneWeights(gstate.vertType); ++i) {
Mat3x3<float> bone(&gstate.boneMatrix[12*i]);
tmppos += (bone * ModelCoords(pos[0], pos[1], pos[2]) + Vec3<float>(gstate.boneMatrix[12*i+9], gstate.boneMatrix[12*i+10], gstate.boneMatrix[12*i+11])) * W[i];
if (vreader.hasNormal())
tmpnrm += (bone * vertex.normal) * W[i];
}
pos[0] = tmppos.x;
pos[1] = tmppos.y;
pos[2] = tmppos.z;
if (vreader.hasNormal())
vertex.normal = tmpnrm;
}
if (vreader.hasColor0()) {
float col[4];
vreader.ReadColor0(col);
vertex.color0 = Vec4<int>(col[0]*255, col[1]*255, col[2]*255, col[3]*255);
} else {
vertex.color0 = Vec4<int>(gstate.getMaterialAmbientR(), gstate.getMaterialAmbientG(), gstate.getMaterialAmbientB(), gstate.getMaterialAmbientA());
}
if (vreader.hasColor1()) {
float col[3];
vreader.ReadColor1(col);
vertex.color1 = Vec3<int>(col[0]*255, col[1]*255, col[2]*255);
} else {
vertex.color1 = Vec3<int>(0, 0, 0);
}
if (!gstate.isModeThrough()) {
vertex.modelpos = ModelCoords(pos[0], pos[1], pos[2]);
vertex.worldpos = WorldCoords(TransformUnit::ModelToWorld(vertex.modelpos));
ModelCoords viewpos = TransformUnit::WorldToView(vertex.worldpos);
vertex.clippos = ClipCoords(TransformUnit::ViewToClip(viewpos));
if (gstate.isFogEnabled()) {
float fog_end = getFloat24(gstate.fog1);
float fog_slope = getFloat24(gstate.fog2);
// Same fixup as in ShaderManagerGLES.cpp
if (my_isnanorinf(fog_end)) {
// Not really sure what a sensible value might be, but let's try 64k.
fog_end = std::signbit(fog_end) ? -65535.0f : 65535.0f;
}
if (my_isnanorinf(fog_slope)) {
fog_slope = std::signbit(fog_slope) ? -65535.0f : 65535.0f;
}
vertex.fogdepth = (viewpos.z + fog_end) * fog_slope;
} else {
vertex.fogdepth = 1.0f;
}
vertex.screenpos = ClipToScreenInternal(vertex.clippos, &outside_range_flag);
if (vreader.hasNormal()) {
vertex.worldnormal = TransformUnit::ModelToWorldNormal(vertex.normal);
// TODO: Isn't there a flag that controls whether to normalize the normal?
vertex.worldnormal /= vertex.worldnormal.Length();
} else {
vertex.worldnormal = Vec3<float>(0.0f, 0.0f, 1.0f);
}
Lighting::Process(vertex, vreader.hasColor0());
} else {
vertex.screenpos.x = (int)(pos[0] * 16) + gstate.getOffsetX16();
vertex.screenpos.y = (int)(pos[1] * 16) + gstate.getOffsetY16();
vertex.screenpos.z = pos[2];
vertex.clippos.w = 1.f;
vertex.fogdepth = 1.f;
}
return vertex;
}
void BaseUpdateUniforms(UB_VS_FS_Base *ub, uint64_t dirtyUniforms, bool flipViewport) {
if (dirtyUniforms & DIRTY_TEXENV) {
Uint8x3ToFloat4(ub->texEnvColor, gstate.texenvcolor);
}
if (dirtyUniforms & DIRTY_ALPHACOLORREF) {
Uint8x3ToInt4_Alpha(ub->alphaColorRef, gstate.getColorTestRef(), gstate.getAlphaTestRef() & gstate.getAlphaTestMask());
}
if (dirtyUniforms & DIRTY_ALPHACOLORMASK) {
Uint8x3ToInt4_Alpha(ub->colorTestMask, gstate.getColorTestMask(), gstate.getAlphaTestMask());
}
if (dirtyUniforms & DIRTY_FOGCOLOR) {
Uint8x3ToFloat4(ub->fogColor, gstate.fogcolor);
}
if (dirtyUniforms & DIRTY_SHADERBLEND) {
Uint8x3ToFloat4(ub->blendFixA, gstate.getFixA());
Uint8x3ToFloat4(ub->blendFixB, gstate.getFixB());
}
if (dirtyUniforms & DIRTY_TEXCLAMP) {
const float invW = 1.0f / (float)gstate_c.curTextureWidth;
const float invH = 1.0f / (float)gstate_c.curTextureHeight;
const int w = gstate.getTextureWidth(0);
const int h = gstate.getTextureHeight(0);
const float widthFactor = (float)w * invW;
const float heightFactor = (float)h * invH;
// First wrap xy, then half texel xy (for clamp.)
ub->texClamp[0] = widthFactor;
ub->texClamp[1] = heightFactor;
ub->texClamp[2] = invW * 0.5f;
ub->texClamp[3] = invH * 0.5f;
ub->texClampOffset[0] = gstate_c.curTextureXOffset * invW;
ub->texClampOffset[1] = gstate_c.curTextureYOffset * invH;
}
if (dirtyUniforms & DIRTY_PROJMATRIX) {
Matrix4x4 flippedMatrix;
memcpy(&flippedMatrix, gstate.projMatrix, 16 * sizeof(float));
const bool invertedY = gstate_c.vpHeight < 0;
if (invertedY) {
flippedMatrix[1] = -flippedMatrix[1];
flippedMatrix[5] = -flippedMatrix[5];
flippedMatrix[9] = -flippedMatrix[9];
flippedMatrix[13] = -flippedMatrix[13];
}
const bool invertedX = gstate_c.vpWidth < 0;
if (invertedX) {
flippedMatrix[0] = -flippedMatrix[0];
flippedMatrix[4] = -flippedMatrix[4];
flippedMatrix[8] = -flippedMatrix[8];
flippedMatrix[12] = -flippedMatrix[12];
}
if (flipViewport) {
ConvertProjMatrixToD3D11(flippedMatrix);
} else {
ConvertProjMatrixToVulkan(flippedMatrix);
}
if (g_Config.iRenderingMode == 0 && g_display_rotation != DisplayRotation::ROTATE_0) {
flippedMatrix = flippedMatrix * g_display_rot_matrix;
}
CopyMatrix4x4(ub->proj, flippedMatrix.getReadPtr());
}
if (dirtyUniforms & DIRTY_PROJTHROUGHMATRIX) {
Matrix4x4 proj_through;
if (flipViewport) {
proj_through.setOrthoD3D(0.0f, gstate_c.curRTWidth, gstate_c.curRTHeight, 0, 0, 1);
} else {
proj_through.setOrthoVulkan(0.0f, gstate_c.curRTWidth, 0, gstate_c.curRTHeight, 0, 1);
}
if (g_Config.iRenderingMode == 0 && g_display_rotation != DisplayRotation::ROTATE_0) {
proj_through = proj_through * g_display_rot_matrix;
}
CopyMatrix4x4(ub->proj_through, proj_through.getReadPtr());
}
// Transform
if (dirtyUniforms & DIRTY_WORLDMATRIX) {
ConvertMatrix4x3To3x4Transposed(ub->world, gstate.worldMatrix);
}
if (dirtyUniforms & DIRTY_VIEWMATRIX) {
ConvertMatrix4x3To3x4Transposed(ub->view, gstate.viewMatrix);
}
if (dirtyUniforms & DIRTY_TEXMATRIX) {
ConvertMatrix4x3To3x4Transposed(ub->tex, gstate.tgenMatrix);
}
// Combined two small uniforms
if (dirtyUniforms & (DIRTY_FOGCOEF | DIRTY_STENCILREPLACEVALUE)) {
float fogcoef_stencil[3] = {
getFloat24(gstate.fog1),
getFloat24(gstate.fog2),
(float)gstate.getStencilTestRef()/255.0f
};
if (my_isinf(fogcoef_stencil[1])) {
// not really sure what a sensible value might be.
fogcoef_stencil[1] = fogcoef_stencil[1] < 0.0f ? -10000.0f : 10000.0f;
} else if (my_isnan(fogcoef_stencil[1])) {
// Workaround for https://github.com/hrydgard/ppsspp/issues/5384#issuecomment-38365988
// Just put the fog far away at a large finite distance.
// Infinities and NaNs are rather unpredictable in shaders on many GPUs
// so it's best to just make it a sane calculation.
fogcoef_stencil[0] = 100000.0f;
fogcoef_stencil[1] = 1.0f;
}
#ifndef MOBILE_DEVICE
else if (my_isnanorinf(fogcoef_stencil[1]) || my_isnanorinf(fogcoef_stencil[0])) {
ERROR_LOG_REPORT_ONCE(fognan, G3D, "Unhandled fog NaN/INF combo: %f %f", fogcoef_stencil[0], fogcoef_stencil[1]);
}
#endif
CopyFloat3(ub->fogCoef_stencil, fogcoef_stencil);
}
// Note - this one is not in lighting but in transformCommon as it has uses beyond lighting
if (dirtyUniforms & DIRTY_MATAMBIENTALPHA) {
Uint8x3ToFloat4_AlphaUint8(ub->matAmbient, gstate.materialambient, gstate.getMaterialAmbientA());
}
// Texturing
if (dirtyUniforms & DIRTY_UVSCALEOFFSET) {
const float invW = 1.0f / (float)gstate_c.curTextureWidth;
const float invH = 1.0f / (float)gstate_c.curTextureHeight;
const int w = gstate.getTextureWidth(0);
const int h = gstate.getTextureHeight(0);
const float widthFactor = (float)w * invW;
const float heightFactor = (float)h * invH;
if (gstate_c.bezier || gstate_c.spline) {
// When we are generating UV coordinates through the bezier/spline, we need to apply the scaling.
// However, this is missing a check that we're not getting our UV:s supplied for us in the vertices.
ub->uvScaleOffset[0] = gstate_c.uv.uScale * widthFactor;
ub->uvScaleOffset[1] = gstate_c.uv.vScale * heightFactor;
ub->uvScaleOffset[2] = gstate_c.uv.uOff * widthFactor;
ub->uvScaleOffset[3] = gstate_c.uv.vOff * heightFactor;
} else {
ub->uvScaleOffset[0] = widthFactor;
ub->uvScaleOffset[1] = heightFactor;
ub->uvScaleOffset[2] = 0.0f;
ub->uvScaleOffset[3] = 0.0f;
}
}
if (dirtyUniforms & DIRTY_DEPTHRANGE) {
float viewZScale = gstate.getViewportZScale();
float viewZCenter = gstate.getViewportZCenter();
// We had to scale and translate Z to account for our clamped Z range.
// Therefore, we also need to reverse this to round properly.
//
// Example: scale = 65535.0, center = 0.0
// Resulting range = -65535 to 65535, clamped to [0, 65535]
// gstate_c.vpDepthScale = 2.0f
// gstate_c.vpZOffset = -1.0f
//
// The projection already accounts for those, so we need to reverse them.
//
// Additionally, D3D9 uses a range from [0, 1]. We double and move the center.
viewZScale *= (1.0f / gstate_c.vpDepthScale) * 2.0f;
viewZCenter -= 65535.0f * gstate_c.vpZOffset + 32768.5f;
float viewZInvScale;
if (viewZScale != 0.0) {
viewZInvScale = 1.0f / viewZScale;
} else {
viewZInvScale = 0.0;
}
ub->depthRange[0] = viewZScale;
ub->depthRange[1] = viewZCenter;
ub->depthRange[2] = viewZCenter;
ub->depthRange[3] = viewZInvScale;
}
if (dirtyUniforms & DIRTY_BEZIERSPLINE) {
ub->spline_count_u = gstate_c.spline_count_u;
ub->spline_count_v = gstate_c.spline_count_v;
ub->spline_type_u = gstate_c.spline_type_u;
ub->spline_type_v = gstate_c.spline_type_v;
}
}
void ShaderManagerVulkan::BaseUpdateUniforms(int dirtyUniforms) {
if (dirtyUniforms & DIRTY_TEXENV) {
Uint8x3ToFloat4(ub_base.texEnvColor, gstate.texenvcolor);
}
if (dirtyUniforms & DIRTY_ALPHACOLORREF) {
Uint8x3ToInt4_Alpha(ub_base.alphaColorRef, gstate.getColorTestRef(), gstate.getAlphaTestRef() & gstate.getAlphaTestMask());
}
if (dirtyUniforms & DIRTY_ALPHACOLORMASK) {
Uint8x3ToInt4_Alpha(ub_base.colorTestMask, gstate.getColorTestMask(), gstate.getAlphaTestMask());
}
if (dirtyUniforms & DIRTY_FOGCOLOR) {
Uint8x3ToFloat4(ub_base.fogColor, gstate.fogcolor);
}
if (dirtyUniforms & DIRTY_SHADERBLEND) {
Uint8x3ToFloat4(ub_base.blendFixA, gstate.getFixA());
Uint8x3ToFloat4(ub_base.blendFixB, gstate.getFixB());
}
if (dirtyUniforms & DIRTY_TEXCLAMP) {
const float invW = 1.0f / (float)gstate_c.curTextureWidth;
const float invH = 1.0f / (float)gstate_c.curTextureHeight;
const int w = gstate.getTextureWidth(0);
const int h = gstate.getTextureHeight(0);
const float widthFactor = (float)w * invW;
const float heightFactor = (float)h * invH;
// First wrap xy, then half texel xy (for clamp.)
ub_base.texClamp[0] = widthFactor;
ub_base.texClamp[1] = heightFactor;
ub_base.texClamp[2] = invW * 0.5f;
ub_base.texClamp[3] = invH * 0.5f;
ub_base.texClampOffset[0] = gstate_c.curTextureXOffset * invW;
ub_base.texClampOffset[1] = gstate_c.curTextureYOffset * invH;
}
if (dirtyUniforms & DIRTY_PROJMATRIX) {
Matrix4x4 flippedMatrix;
memcpy(&flippedMatrix, gstate.projMatrix, 16 * sizeof(float));
const bool invertedY = gstate_c.vpHeight < 0;
if (invertedY) {
flippedMatrix[1] = -flippedMatrix[1];
flippedMatrix[5] = -flippedMatrix[5];
flippedMatrix[9] = -flippedMatrix[9];
flippedMatrix[13] = -flippedMatrix[13];
}
const bool invertedX = gstate_c.vpWidth < 0;
if (invertedX) {
flippedMatrix[0] = -flippedMatrix[0];
flippedMatrix[4] = -flippedMatrix[4];
flippedMatrix[8] = -flippedMatrix[8];
flippedMatrix[12] = -flippedMatrix[12];
}
ConvertProjMatrixToVulkan(flippedMatrix, invertedX, invertedY);
CopyMatrix4x4(ub_base.proj, flippedMatrix.getReadPtr());
}
if (dirtyUniforms & DIRTY_PROJTHROUGHMATRIX) {
Matrix4x4 proj_through;
proj_through.setOrthoVulkan(0.0f, gstate_c.curRTWidth, 0, gstate_c.curRTHeight, 0, 1);
CopyMatrix4x4(ub_base.proj_through, proj_through.getReadPtr());
}
// Transform
if (dirtyUniforms & DIRTY_WORLDMATRIX) {
ConvertMatrix4x3To4x4(ub_base.world, gstate.worldMatrix);
}
if (dirtyUniforms & DIRTY_VIEWMATRIX) {
ConvertMatrix4x3To4x4(ub_base.view, gstate.viewMatrix);
}
if (dirtyUniforms & DIRTY_TEXMATRIX) {
ConvertMatrix4x3To4x4(ub_base.tex, gstate.tgenMatrix);
}
// Combined two small uniforms
if (dirtyUniforms & (DIRTY_FOGCOEF | DIRTY_STENCILREPLACEVALUE)) {
float fogcoef_stencil[3] = {
getFloat24(gstate.fog1),
getFloat24(gstate.fog2),
(float)gstate.getStencilTestRef()
};
if (my_isinf(fogcoef_stencil[1])) {
// not really sure what a sensible value might be.
fogcoef_stencil[1] = fogcoef_stencil[1] < 0.0f ? -10000.0f : 10000.0f;
} else if (my_isnan(fogcoef_stencil[1])) {
// Workaround for https://github.com/hrydgard/ppsspp/issues/5384#issuecomment-38365988
// Just put the fog far away at a large finite distance.
// Infinities and NaNs are rather unpredictable in shaders on many GPUs
// so it's best to just make it a sane calculation.
fogcoef_stencil[0] = 100000.0f;
fogcoef_stencil[1] = 1.0f;
}
#ifndef MOBILE_DEVICE
else if (my_isnanorinf(fogcoef_stencil[1]) || my_isnanorinf(fogcoef_stencil[0])) {
ERROR_LOG_REPORT_ONCE(fognan, G3D, "Unhandled fog NaN/INF combo: %f %f", fogcoef_stencil[0], fogcoef_stencil[1]);
}
#endif
CopyFloat3(ub_base.fogCoef_stencil, fogcoef_stencil);
}
// Texturing
if (dirtyUniforms & DIRTY_UVSCALEOFFSET) {
const float invW = 1.0f / (float)gstate_c.curTextureWidth;
const float invH = 1.0f / (float)gstate_c.curTextureHeight;
const int w = gstate.getTextureWidth(0);
const int h = gstate.getTextureHeight(0);
const float widthFactor = (float)w * invW;
const float heightFactor = (float)h * invH;
ub_base.uvScaleOffset[0] = widthFactor;
ub_base.uvScaleOffset[1] = heightFactor;
ub_base.uvScaleOffset[2] = 0.0f;
ub_base.uvScaleOffset[3] = 0.0f;
}
if (dirtyUniforms & DIRTY_DEPTHRANGE) {
float viewZScale = gstate.getViewportZScale();
float viewZCenter = gstate.getViewportZCenter();
float viewZInvScale;
// We had to scale and translate Z to account for our clamped Z range.
// Therefore, we also need to reverse this to round properly.
//
// Example: scale = 65535.0, center = 0.0
// Resulting range = -65535 to 65535, clamped to [0, 65535]
// gstate_c.vpDepthScale = 2.0f
// gstate_c.vpZOffset = -1.0f
//
// The projection already accounts for those, so we need to reverse them.
//
// Additionally, D3D9 uses a range from [0, 1]. We double and move the center.
viewZScale *= (1.0f / gstate_c.vpDepthScale) * 2.0f;
viewZCenter -= 65535.0f * gstate_c.vpZOffset + 32768.5f;
if (viewZScale != 0.0) {
viewZInvScale = 1.0f / viewZScale;
} else {
viewZInvScale = 0.0;
}
ub_base.depthRange[0] = viewZScale;
ub_base.depthRange[1] = viewZCenter;
ub_base.depthRange[2] = viewZCenter;
ub_base.depthRange[3] = viewZInvScale;
}
}
void ShaderManagerVulkan::BaseUpdateUniforms(int dirtyUniforms) {
if (dirtyUniforms & DIRTY_TEXENV) {
Uint8x3ToFloat4(ub_base.texEnvColor, gstate.texenvcolor);
}
if (dirtyUniforms & DIRTY_ALPHACOLORREF) {
Uint8x3ToInt4_Alpha(ub_base.alphaColorRef, gstate.getColorTestRef(), gstate.getAlphaTestRef() & gstate.getAlphaTestMask());
}
if (dirtyUniforms & DIRTY_ALPHACOLORMASK) {
Uint8x3ToInt4_Alpha(ub_base.colorTestMask, gstate.getColorTestMask(), gstate.getAlphaTestMask());
}
if (dirtyUniforms & DIRTY_FOGCOLOR) {
Uint8x3ToFloat4(ub_base.fogColor, gstate.fogcolor);
}
if (dirtyUniforms & DIRTY_SHADERBLEND) {
Uint8x3ToFloat4(ub_base.blendFixA, gstate.getFixA());
Uint8x3ToFloat4(ub_base.blendFixB, gstate.getFixB());
}
if (dirtyUniforms & DIRTY_TEXCLAMP) {
const float invW = 1.0f / (float)gstate_c.curTextureWidth;
const float invH = 1.0f / (float)gstate_c.curTextureHeight;
const int w = gstate.getTextureWidth(0);
const int h = gstate.getTextureHeight(0);
const float widthFactor = (float)w * invW;
const float heightFactor = (float)h * invH;
// First wrap xy, then half texel xy (for clamp.)
const float texclamp[4] = {
widthFactor,
heightFactor,
invW * 0.5f,
invH * 0.5f,
};
const float texclampoff[2] = {
gstate_c.curTextureXOffset * invW,
gstate_c.curTextureYOffset * invH,
};
CopyFloat4(ub_base.texClamp, texclamp);
CopyFloat2(ub_base.texClampOffset, texclampoff);
}
if (dirtyUniforms & DIRTY_PROJMATRIX) {
Matrix4x4 flippedMatrix;
memcpy(&flippedMatrix, gstate.projMatrix, 16 * sizeof(float));
const bool invertedY = gstate_c.vpHeight < 0;
if (invertedY) {
flippedMatrix[1] = -flippedMatrix[1];
flippedMatrix[5] = -flippedMatrix[5];
flippedMatrix[9] = -flippedMatrix[9];
flippedMatrix[13] = -flippedMatrix[13];
}
const bool invertedX = gstate_c.vpWidth < 0;
if (invertedX) {
flippedMatrix[0] = -flippedMatrix[0];
flippedMatrix[4] = -flippedMatrix[4];
flippedMatrix[8] = -flippedMatrix[8];
flippedMatrix[12] = -flippedMatrix[12];
}
ConvertProjMatrixToVulkan(flippedMatrix, invertedX, invertedY);
CopyMatrix4x4(ub_base.proj, flippedMatrix.getReadPtr());
}
if (dirtyUniforms & DIRTY_PROJTHROUGHMATRIX) {
Matrix4x4 proj_through;
proj_through.setOrthoVulkan(0.0f, gstate_c.curRTWidth, 0, gstate_c.curRTHeight, 0, 1);
CopyMatrix4x4(ub_base.proj_through, proj_through.getReadPtr());
}
// Transform
if (dirtyUniforms & DIRTY_WORLDMATRIX) {
ConvertMatrix4x3To4x4(ub_base.world, gstate.worldMatrix);
}
if (dirtyUniforms & DIRTY_VIEWMATRIX) {
ConvertMatrix4x3To4x4(ub_base.view, gstate.viewMatrix);
}
if (dirtyUniforms & DIRTY_TEXMATRIX) {
ConvertMatrix4x3To4x4(ub_base.tex, gstate.tgenMatrix);
}
// Combined two small uniforms
if (dirtyUniforms & (DIRTY_FOGCOEF | DIRTY_STENCILREPLACEVALUE)) {
float fogcoef_stencil[3] = {
getFloat24(gstate.fog1),
getFloat24(gstate.fog2),
(float)gstate.getStencilTestRef()
};
if (my_isinf(fogcoef_stencil[1])) {
// not really sure what a sensible value might be.
fogcoef_stencil[1] = fogcoef_stencil[1] < 0.0f ? -10000.0f : 10000.0f;
} else if (my_isnan(fogcoef_stencil[1])) {
// Workaround for https://github.com/hrydgard/ppsspp/issues/5384#issuecomment-38365988
// Just put the fog far away at a large finite distance.
// Infinities and NaNs are rather unpredictable in shaders on many GPUs
// so it's best to just make it a sane calculation.
fogcoef_stencil[0] = 100000.0f;
fogcoef_stencil[1] = 1.0f;
}
#ifndef MOBILE_DEVICE
else if (my_isnanorinf(fogcoef_stencil[1]) || my_isnanorinf(fogcoef_stencil[0])) {
ERROR_LOG_REPORT_ONCE(fognan, G3D, "Unhandled fog NaN/INF combo: %f %f", fogcoef_stencil[0], fogcoef_stencil[1]);
}
#endif
CopyFloat3(ub_base.fogCoef_stencil, fogcoef_stencil);
}
// Texturing
if (dirtyUniforms & DIRTY_UVSCALEOFFSET) {
const float invW = 1.0f / (float)gstate_c.curTextureWidth;
const float invH = 1.0f / (float)gstate_c.curTextureHeight;
const int w = gstate.getTextureWidth(0);
const int h = gstate.getTextureHeight(0);
const float widthFactor = (float)w * invW;
const float heightFactor = (float)h * invH;
static const float rescale[4] = { 1.0f, 2 * 127.5f / 128.f, 2 * 32767.5f / 32768.f, 1.0f };
const float factor = rescale[(gstate.vertType & GE_VTYPE_TC_MASK) >> GE_VTYPE_TC_SHIFT];
float uvscaleoff[4];
switch (gstate.getUVGenMode()) {
case GE_TEXMAP_TEXTURE_COORDS:
// Not sure what GE_TEXMAP_UNKNOWN is, but seen in Riviera. Treating the same as GE_TEXMAP_TEXTURE_COORDS works.
case GE_TEXMAP_UNKNOWN:
if (g_Config.bPrescaleUV) {
// We are here but are prescaling UV in the decoder? Let's do the same as in the other case
// except consider *Scale and *Off to be 1 and 0.
uvscaleoff[0] = widthFactor;
uvscaleoff[1] = heightFactor;
uvscaleoff[2] = 0.0f;
uvscaleoff[3] = 0.0f;
} else {
uvscaleoff[0] = gstate_c.uv.uScale * factor * widthFactor;
uvscaleoff[1] = gstate_c.uv.vScale * factor * heightFactor;
uvscaleoff[2] = gstate_c.uv.uOff * widthFactor;
uvscaleoff[3] = gstate_c.uv.vOff * heightFactor;
}
break;
// These two work the same whether or not we prescale UV.
case GE_TEXMAP_TEXTURE_MATRIX:
// We cannot bake the UV coord scale factor in here, as we apply a matrix multiplication
// before this is applied, and the matrix multiplication may contain translation. In this case
// the translation will be scaled which breaks faces in Hexyz Force for example.
// So I've gone back to applying the scale factor in the shader.
uvscaleoff[0] = widthFactor;
uvscaleoff[1] = heightFactor;
uvscaleoff[2] = 0.0f;
uvscaleoff[3] = 0.0f;
break;
case GE_TEXMAP_ENVIRONMENT_MAP:
// In this mode we only use uvscaleoff to scale to the texture size.
uvscaleoff[0] = widthFactor;
uvscaleoff[1] = heightFactor;
uvscaleoff[2] = 0.0f;
uvscaleoff[3] = 0.0f;
break;
default:
ERROR_LOG_REPORT(G3D, "Unexpected UV gen mode: %d", gstate.getUVGenMode());
}
CopyFloat4(ub_base.uvScaleOffset, uvscaleoff);
}
if (dirtyUniforms & DIRTY_DEPTHRANGE) {
float viewZScale = gstate.getViewportZScale();
float viewZCenter = gstate.getViewportZCenter();
float viewZInvScale;
// We had to scale and translate Z to account for our clamped Z range.
// Therefore, we also need to reverse this to round properly.
//
// Example: scale = 65535.0, center = 0.0
// Resulting range = -65535 to 65535, clamped to [0, 65535]
// gstate_c.vpDepthScale = 2.0f
// gstate_c.vpZOffset = -1.0f
//
// The projection already accounts for those, so we need to reverse them.
//
// Additionally, D3D9 uses a range from [0, 1]. We double and move the center.
viewZScale *= (1.0f / gstate_c.vpDepthScale) * 2.0f;
viewZCenter -= 65535.0f * gstate_c.vpZOffset + 32768.5f;
if (viewZScale != 0.0) {
viewZInvScale = 1.0f / viewZScale;
} else {
viewZInvScale = 0.0;
}
float data[4] = { viewZScale, viewZCenter, viewZCenter, viewZInvScale };
CopyFloat4(ub_base.depthRange, data);
}
}
void ShaderManagerDX9::VSUpdateUniforms(int dirtyUniforms) {
// Update any dirty uniforms before we draw
if (dirtyUniforms & DIRTY_PROJMATRIX) {
Matrix4x4 flippedMatrix;
memcpy(&flippedMatrix, gstate.projMatrix, 16 * sizeof(float));
const bool invertedY = gstate_c.vpHeight < 0;
if (!invertedY) {
flippedMatrix[1] = -flippedMatrix[1];
flippedMatrix[5] = -flippedMatrix[5];
flippedMatrix[9] = -flippedMatrix[9];
flippedMatrix[13] = -flippedMatrix[13];
}
const bool invertedX = gstate_c.vpWidth < 0;
if (invertedX) {
flippedMatrix[0] = -flippedMatrix[0];
flippedMatrix[4] = -flippedMatrix[4];
flippedMatrix[8] = -flippedMatrix[8];
flippedMatrix[12] = -flippedMatrix[12];
}
ConvertProjMatrixToD3D(flippedMatrix, invertedX, invertedY);
VSSetMatrix(CONST_VS_PROJ, flippedMatrix.getReadPtr());
}
if (dirtyUniforms & DIRTY_PROJTHROUGHMATRIX) {
Matrix4x4 proj_through;
proj_through.setOrtho(0.0f, gstate_c.curRTWidth, gstate_c.curRTHeight, 0, 0, 1);
ConvertProjMatrixToD3DThrough(proj_through);
VSSetMatrix(CONST_VS_PROJ_THROUGH, proj_through.getReadPtr());
}
// Transform
if (dirtyUniforms & DIRTY_WORLDMATRIX) {
VSSetMatrix4x3_3(CONST_VS_WORLD, gstate.worldMatrix);
}
if (dirtyUniforms & DIRTY_VIEWMATRIX) {
VSSetMatrix4x3_3(CONST_VS_VIEW, gstate.viewMatrix);
}
if (dirtyUniforms & DIRTY_TEXMATRIX) {
VSSetMatrix4x3_3(CONST_VS_TEXMTX, gstate.tgenMatrix);
}
if (dirtyUniforms & DIRTY_FOGCOEF) {
float fogcoef[2] = {
getFloat24(gstate.fog1),
getFloat24(gstate.fog2),
};
if (my_isinf(fogcoef[1])) {
// not really sure what a sensible value might be.
fogcoef[1] = fogcoef[1] < 0.0f ? -10000.0f : 10000.0f;
} else if (my_isnan(fogcoef[1])) {
// Workaround for https://github.com/hrydgard/ppsspp/issues/5384#issuecomment-38365988
// Just put the fog far away at a large finite distance.
// Infinities and NaNs are rather unpredictable in shaders on many GPUs
// so it's best to just make it a sane calculation.
fogcoef[0] = 100000.0f;
fogcoef[1] = 1.0f;
}
#ifndef MOBILE_DEVICE
else if (my_isnanorinf(fogcoef[1]) || my_isnanorinf(fogcoef[0])) {
ERROR_LOG_REPORT_ONCE(fognan, G3D, "Unhandled fog NaN/INF combo: %f %f", fogcoef[0], fogcoef[1]);
}
#endif
VSSetFloatArray(CONST_VS_FOGCOEF, fogcoef, 2);
}
// TODO: Could even set all bones in one go if they're all dirty.
#ifdef USE_BONE_ARRAY
if (u_bone != 0) {
float allBones[8 * 16];
bool allDirty = true;
for (int i = 0; i < numBones; i++) {
if (dirtyUniforms & (DIRTY_BONEMATRIX0 << i)) {
ConvertMatrix4x3To4x4(allBones + 16 * i, gstate.boneMatrix + 12 * i);
} else {
allDirty = false;
}
}
if (allDirty) {
// Set them all with one call
//glUniformMatrix4fv(u_bone, numBones, GL_FALSE, allBones);
} else {
// Set them one by one. Could try to coalesce two in a row etc but too lazy.
for (int i = 0; i < numBones; i++) {
if (dirtyUniforms & (DIRTY_BONEMATRIX0 << i)) {
//glUniformMatrix4fv(u_bone + i, 1, GL_FALSE, allBones + 16 * i);
}
}
}
}
#else
for (int i = 0; i < 8; i++) {
if (dirtyUniforms & (DIRTY_BONEMATRIX0 << i)) {
VSSetMatrix4x3_3(CONST_VS_BONE0 + 3 * i, gstate.boneMatrix + 12 * i);
}
}
#endif
// Texturing
if (dirtyUniforms & DIRTY_UVSCALEOFFSET) {
const float invW = 1.0f / (float)gstate_c.curTextureWidth;
const float invH = 1.0f / (float)gstate_c.curTextureHeight;
const int w = gstate.getTextureWidth(0);
const int h = gstate.getTextureHeight(0);
const float widthFactor = (float)w * invW;
const float heightFactor = (float)h * invH;
float uvscaleoff[4];
switch (gstate.getUVGenMode()) {
case GE_TEXMAP_TEXTURE_COORDS:
// Not sure what GE_TEXMAP_UNKNOWN is, but seen in Riviera. Treating the same as GE_TEXMAP_TEXTURE_COORDS works.
case GE_TEXMAP_UNKNOWN:
if (g_Config.bPrescaleUV) {
// We are here but are prescaling UV in the decoder? Let's do the same as in the other case
// except consider *Scale and *Off to be 1 and 0.
uvscaleoff[0] = widthFactor;
uvscaleoff[1] = heightFactor;
uvscaleoff[2] = 0.0f;
uvscaleoff[3] = 0.0f;
} else {
uvscaleoff[0] = gstate_c.uv.uScale * widthFactor;
uvscaleoff[1] = gstate_c.uv.vScale * heightFactor;
uvscaleoff[2] = gstate_c.uv.uOff * widthFactor;
uvscaleoff[3] = gstate_c.uv.vOff * heightFactor;
}
break;
// These two work the same whether or not we prescale UV.
case GE_TEXMAP_TEXTURE_MATRIX:
// We cannot bake the UV coord scale factor in here, as we apply a matrix multiplication
// before this is applied, and the matrix multiplication may contain translation. In this case
// the translation will be scaled which breaks faces in Hexyz Force for example.
// So I've gone back to applying the scale factor in the shader.
uvscaleoff[0] = widthFactor;
uvscaleoff[1] = heightFactor;
uvscaleoff[2] = 0.0f;
uvscaleoff[3] = 0.0f;
break;
case GE_TEXMAP_ENVIRONMENT_MAP:
// In this mode we only use uvscaleoff to scale to the texture size.
uvscaleoff[0] = widthFactor;
uvscaleoff[1] = heightFactor;
uvscaleoff[2] = 0.0f;
uvscaleoff[3] = 0.0f;
break;
default:
ERROR_LOG_REPORT(G3D, "Unexpected UV gen mode: %d", gstate.getUVGenMode());
}
VSSetFloatArray(CONST_VS_UVSCALEOFFSET, uvscaleoff, 4);
}
if (dirtyUniforms & DIRTY_DEPTHRANGE) {
// Depth is [0, 1] mapping to [minz, maxz], not too hard.
float vpZScale = gstate.getViewportZScale();
float vpZCenter = gstate.getViewportZCenter();
// These are just the reverse of the formulas in GPUStateUtils.
float halfActualZRange = vpZScale / gstate_c.vpDepthScale;
float minz = -((gstate_c.vpZOffset * halfActualZRange) - vpZCenter) - halfActualZRange;
float viewZScale = halfActualZRange * 2.0f;
// Account for the half pixel offset.
float viewZCenter = minz + (DepthSliceFactor() / 256.0f) * 0.5f;
float viewZInvScale;
if (viewZScale != 0.0) {
viewZInvScale = 1.0f / viewZScale;
} else {
viewZInvScale = 0.0;
}
float data[4] = { viewZScale, viewZCenter, viewZCenter, viewZInvScale };
VSSetFloatUniform4(CONST_VS_DEPTHRANGE, data);
}
// Lighting
if (dirtyUniforms & DIRTY_AMBIENT) {
VSSetColorUniform3Alpha(CONST_VS_AMBIENT, gstate.ambientcolor, gstate.getAmbientA());
}
if (dirtyUniforms & DIRTY_MATAMBIENTALPHA) {
VSSetColorUniform3Alpha(CONST_VS_MATAMBIENTALPHA, gstate.materialambient, gstate.getMaterialAmbientA());
}
if (dirtyUniforms & DIRTY_MATDIFFUSE) {
VSSetColorUniform3(CONST_VS_MATDIFFUSE, gstate.materialdiffuse);
}
if (dirtyUniforms & DIRTY_MATEMISSIVE) {
VSSetColorUniform3(CONST_VS_MATEMISSIVE, gstate.materialemissive);
}
if (dirtyUniforms & DIRTY_MATSPECULAR) {
VSSetColorUniform3ExtraFloat(CONST_VS_MATSPECULAR, gstate.materialspecular, getFloat24(gstate.materialspecularcoef));
}
for (int i = 0; i < 4; i++) {
if (dirtyUniforms & (DIRTY_LIGHT0 << i)) {
if (gstate.isDirectionalLight(i)) {
// Prenormalize
float x = getFloat24(gstate.lpos[i * 3 + 0]);
float y = getFloat24(gstate.lpos[i * 3 + 1]);
float z = getFloat24(gstate.lpos[i * 3 + 2]);
float len = sqrtf(x*x + y*y + z*z);
if (len == 0.0f)
len = 1.0f;
else
len = 1.0f / len;
float vec[3] = { x * len, y * len, z * len };
VSSetFloatArray(CONST_VS_LIGHTPOS + i, vec, 3);
} else {
VSSetFloat24Uniform3(CONST_VS_LIGHTPOS + i, &gstate.lpos[i * 3]);
}
VSSetFloat24Uniform3(CONST_VS_LIGHTDIR + i, &gstate.ldir[i * 3]);
VSSetFloat24Uniform3(CONST_VS_LIGHTATT + i, &gstate.latt[i * 3]);
VSSetFloat(CONST_VS_LIGHTANGLE + i, getFloat24(gstate.lcutoff[i]));
VSSetFloat(CONST_VS_LIGHTSPOTCOEF + i, getFloat24(gstate.lconv[i]));
VSSetColorUniform3(CONST_VS_LIGHTAMBIENT + i, gstate.lcolor[i * 3]);
VSSetColorUniform3(CONST_VS_LIGHTDIFFUSE + i, gstate.lcolor[i * 3 + 1]);
VSSetColorUniform3(CONST_VS_LIGHTSPECULAR + i, gstate.lcolor[i * 3 + 2]);
}
}
}
void SoftwareTransform(
int prim, int vertexCount, u32 vertType, u16 *&inds, int indexType,
const DecVtxFormat &decVtxFormat, int &maxIndex, TransformedVertex *&drawBuffer, int &numTrans, bool &drawIndexed, const SoftwareTransformParams *params, SoftwareTransformResult *result) {
u8 *decoded = params->decoded;
FramebufferManagerCommon *fbman = params->fbman;
TextureCacheCommon *texCache = params->texCache;
TransformedVertex *transformed = params->transformed;
TransformedVertex *transformedExpanded = params->transformedExpanded;
float ySign = 1.0f;
bool throughmode = (vertType & GE_VTYPE_THROUGH_MASK) != 0;
bool lmode = gstate.isUsingSecondaryColor() && gstate.isLightingEnabled();
float uscale = 1.0f;
float vscale = 1.0f;
if (throughmode) {
uscale /= gstate_c.curTextureWidth;
vscale /= gstate_c.curTextureHeight;
}
bool skinningEnabled = vertTypeIsSkinningEnabled(vertType);
const int w = gstate.getTextureWidth(0);
const int h = gstate.getTextureHeight(0);
float widthFactor = (float) w / (float) gstate_c.curTextureWidth;
float heightFactor = (float) h / (float) gstate_c.curTextureHeight;
Lighter lighter(vertType);
float fog_end = getFloat24(gstate.fog1);
float fog_slope = getFloat24(gstate.fog2);
// Same fixup as in ShaderManagerGLES.cpp
if (my_isnanorinf(fog_end)) {
// Not really sure what a sensible value might be, but let's try 64k.
fog_end = std::signbit(fog_end) ? -65535.0f : 65535.0f;
}
if (my_isnanorinf(fog_slope)) {
fog_slope = std::signbit(fog_slope) ? -65535.0f : 65535.0f;
}
int provokeIndOffset = 0;
if (params->provokeFlatFirst) {
provokeIndOffset = ColorIndexOffset(prim, gstate.getShadeMode(), gstate.isModeClear());
}
VertexReader reader(decoded, decVtxFormat, vertType);
if (throughmode) {
for (int index = 0; index < maxIndex; index++) {
// Do not touch the coordinates or the colors. No lighting.
reader.Goto(index);
// TODO: Write to a flexible buffer, we don't always need all four components.
TransformedVertex &vert = transformed[index];
reader.ReadPos(vert.pos);
if (reader.hasColor0()) {
if (provokeIndOffset != 0 && index + provokeIndOffset < maxIndex) {
reader.Goto(index + provokeIndOffset);
reader.ReadColor0_8888(vert.color0);
reader.Goto(index);
} else {
reader.ReadColor0_8888(vert.color0);
}
} else {
vert.color0_32 = gstate.getMaterialAmbientRGBA();
}
if (reader.hasUV()) {
reader.ReadUV(vert.uv);
vert.u *= uscale;
vert.v *= vscale;
} else {
vert.u = 0.0f;
vert.v = 0.0f;
}
// Ignore color1 and fog, never used in throughmode anyway.
// The w of uv is also never used (hardcoded to 1.0.)
}
} else {
// Okay, need to actually perform the full transform.
for (int index = 0; index < maxIndex; index++) {
reader.Goto(index);
float v[3] = {0, 0, 0};
Vec4f c0 = Vec4f(1, 1, 1, 1);
Vec4f c1 = Vec4f(0, 0, 0, 0);
float uv[3] = {0, 0, 1};
float fogCoef = 1.0f;
float out[3];
float pos[3];
Vec3f normal(0, 0, 1);
Vec3f worldnormal(0, 0, 1);
reader.ReadPos(pos);
float ruv[2] = { 0.0f, 0.0f };
if (reader.hasUV())
reader.ReadUV(ruv);
// Read all the provoking vertex values here.
Vec4f unlitColor;
if (provokeIndOffset != 0 && index + provokeIndOffset < maxIndex)
reader.Goto(index + provokeIndOffset);
if (reader.hasColor0())
reader.ReadColor0(unlitColor.AsArray());
else
unlitColor = Vec4f::FromRGBA(gstate.getMaterialAmbientRGBA());
if (reader.hasNormal())
reader.ReadNrm(normal.AsArray());
if (!skinningEnabled) {
Vec3ByMatrix43(out, pos, gstate.worldMatrix);
if (reader.hasNormal()) {
if (gstate.areNormalsReversed()) {
normal = -normal;
}
Norm3ByMatrix43(worldnormal.AsArray(), normal.AsArray(), gstate.worldMatrix);
worldnormal = worldnormal.Normalized();
}
} else {
float weights[8];
// TODO: For flat, are weights from the provoking used for color/normal?
reader.Goto(index);
reader.ReadWeights(weights);
// Skinning
Vec3f psum(0, 0, 0);
Vec3f nsum(0, 0, 0);
for (int i = 0; i < vertTypeGetNumBoneWeights(vertType); i++) {
if (weights[i] != 0.0f) {
Vec3ByMatrix43(out, pos, gstate.boneMatrix+i*12);
Vec3f tpos(out);
psum += tpos * weights[i];
if (reader.hasNormal()) {
Vec3f norm;
Norm3ByMatrix43(norm.AsArray(), normal.AsArray(), gstate.boneMatrix+i*12);
nsum += norm * weights[i];
}
}
}
// Yes, we really must multiply by the world matrix too.
Vec3ByMatrix43(out, psum.AsArray(), gstate.worldMatrix);
if (reader.hasNormal()) {
normal = nsum;
if (gstate.areNormalsReversed()) {
normal = -normal;
}
Norm3ByMatrix43(worldnormal.AsArray(), normal.AsArray(), gstate.worldMatrix);
worldnormal = worldnormal.Normalized();
}
}
// Perform lighting here if enabled.
if (gstate.isLightingEnabled()) {
float litColor0[4];
float litColor1[4];
lighter.Light(litColor0, litColor1, unlitColor.AsArray(), out, worldnormal);
// Don't ignore gstate.lmode - we should send two colors in that case
for (int j = 0; j < 4; j++) {
c0[j] = litColor0[j];
}
if (lmode) {
// Separate colors
for (int j = 0; j < 4; j++) {
c1[j] = litColor1[j];
}
} else {
// Summed color into c0 (will clamp in ToRGBA().)
for (int j = 0; j < 4; j++) {
c0[j] += litColor1[j];
}
}
} else {
if (reader.hasColor0()) {
for (int j = 0; j < 4; j++) {
c0[j] = unlitColor[j];
}
} else {
c0 = Vec4f::FromRGBA(gstate.getMaterialAmbientRGBA());
}
if (lmode) {
// c1 is already 0.
}
}
// Perform texture coordinate generation after the transform and lighting - one style of UV depends on lights.
switch (gstate.getUVGenMode()) {
case GE_TEXMAP_TEXTURE_COORDS: // UV mapping
case GE_TEXMAP_UNKNOWN: // Seen in Riviera. Unsure of meaning, but this works.
// We always prescale in the vertex decoder now.
uv[0] = ruv[0];
uv[1] = ruv[1];
uv[2] = 1.0f;
break;
case GE_TEXMAP_TEXTURE_MATRIX:
{
// TODO: What's the correct behavior with flat shading? Provoked normal or real normal?
// Projection mapping
Vec3f source;
switch (gstate.getUVProjMode()) {
case GE_PROJMAP_POSITION: // Use model space XYZ as source
source = pos;
break;
case GE_PROJMAP_UV: // Use unscaled UV as source
source = Vec3f(ruv[0], ruv[1], 0.0f);
break;
case GE_PROJMAP_NORMALIZED_NORMAL: // Use normalized normal as source
source = normal.Normalized();
if (!reader.hasNormal()) {
ERROR_LOG_REPORT(G3D, "Normal projection mapping without normal?");
}
break;
case GE_PROJMAP_NORMAL: // Use non-normalized normal as source!
source = normal;
if (!reader.hasNormal()) {
ERROR_LOG_REPORT(G3D, "Normal projection mapping without normal?");
}
break;
}
float uvw[3];
Vec3ByMatrix43(uvw, &source.x, gstate.tgenMatrix);
uv[0] = uvw[0];
uv[1] = uvw[1];
uv[2] = uvw[2];
}
break;
case GE_TEXMAP_ENVIRONMENT_MAP:
// Shade mapping - use two light sources to generate U and V.
{
auto getLPosFloat = [&](int l, int i) {
return getFloat24(gstate.lpos[l * 3 + i]);
};
auto getLPos = [&](int l) {
return Vec3f(getLPosFloat(l, 0), getLPosFloat(l, 1), getLPosFloat(l, 2));
};
auto calcShadingLPos = [&](int l) {
Vec3f pos = getLPos(l);
if (pos.Length2() == 0.0f) {
return Vec3f(0.0f, 0.0f, 1.0f);
} else {
return pos.Normalized();
}
};
// Might not have lighting enabled, so don't use lighter.
Vec3f lightpos0 = calcShadingLPos(gstate.getUVLS0());
Vec3f lightpos1 = calcShadingLPos(gstate.getUVLS1());
uv[0] = (1.0f + Dot(lightpos0, worldnormal))/2.0f;
uv[1] = (1.0f + Dot(lightpos1, worldnormal))/2.0f;
uv[2] = 1.0f;
}
break;
default:
// Illegal
ERROR_LOG_REPORT(G3D, "Impossible UV gen mode? %d", gstate.getUVGenMode());
break;
}
uv[0] = uv[0] * widthFactor;
uv[1] = uv[1] * heightFactor;
// Transform the coord by the view matrix.
Vec3ByMatrix43(v, out, gstate.viewMatrix);
fogCoef = (v[2] + fog_end) * fog_slope;
// TODO: Write to a flexible buffer, we don't always need all four components.
memcpy(&transformed[index].x, v, 3 * sizeof(float));
transformed[index].fog = fogCoef;
memcpy(&transformed[index].u, uv, 3 * sizeof(float));
transformed[index].color0_32 = c0.ToRGBA();
transformed[index].color1_32 = c1.ToRGBA();
// The multiplication by the projection matrix is still performed in the vertex shader.
// So is vertex depth rounding, to simulate the 16-bit depth buffer.
}
}
// Here's the best opportunity to try to detect rectangles used to clear the screen, and
// replace them with real clears. This can provide a speedup on certain mobile chips.
//
// An alternative option is to simply ditch all the verts except the first and last to create a single
// rectangle out of many. Quite a small optimization though.
// Experiment: Disable on PowerVR (see issue #6290)
// TODO: This bleeds outside the play area in non-buffered mode. Big deal? Probably not.
// TODO: Allow creating a depth clear and a color draw.
bool reallyAClear = false;
if (maxIndex > 1 && prim == GE_PRIM_RECTANGLES && gstate.isModeClear() && params->allowClear) {
int scissorX2 = gstate.getScissorX2() + 1;
int scissorY2 = gstate.getScissorY2() + 1;
reallyAClear = IsReallyAClear(transformed, maxIndex, scissorX2, scissorY2);
}
if (reallyAClear && gl_extensions.gpuVendor != GPU_VENDOR_IMGTEC) {
// If alpha is not allowed to be separate, it must match for both depth/stencil and color. Vulkan requires this.
bool alphaMatchesColor = gstate.isClearModeColorMask() == gstate.isClearModeAlphaMask();
bool depthMatchesStencil = gstate.isClearModeAlphaMask() == gstate.isClearModeDepthMask();
bool matchingComponents = params->allowSeparateAlphaClear || (alphaMatchesColor && depthMatchesStencil);
bool stencilNotMasked = !gstate.isClearModeAlphaMask() || gstate.getStencilWriteMask() == 0x00;
if (matchingComponents && stencilNotMasked) {
result->color = transformed[1].color0_32;
// Need to rescale from a [0, 1] float. This is the final transformed value.
result->depth = ToScaledDepthFromIntegerScale((int)(transformed[1].z * 65535.0f));
result->action = SW_CLEAR;
gpuStats.numClears++;
return;
}
}
// This means we're using a framebuffer (and one that isn't big enough.)
if (gstate_c.curTextureHeight < (u32)h && maxIndex >= 2) {
// Even if not rectangles, this will detect if either of the first two are outside the framebuffer.
// HACK: Adding one pixel margin to this detection fixes issues in Assassin's Creed : Bloodlines,
// while still keeping BOF working (see below).
const float invTexH = 1.0f / gstate_c.curTextureHeight; // size of one texel.
bool tlOutside;
bool tlAlmostOutside;
bool brOutside;
// If we're outside heightFactor, then v must be wrapping or clamping. Avoid this workaround.
// If we're <= 1.0f, we're inside the framebuffer (workaround not needed.)
// We buffer that 1.0f a little more with a texel to avoid some false positives.
tlOutside = transformed[0].v <= heightFactor && transformed[0].v > 1.0f + invTexH;
brOutside = transformed[1].v <= heightFactor && transformed[1].v > 1.0f + invTexH;
// Careful: if br is outside, but tl is well inside, this workaround still doesn't make sense.
// We go with halfway, since we overestimate framebuffer heights sometimes but not by much.
tlAlmostOutside = transformed[0].v <= heightFactor && transformed[0].v >= 0.5f;
if (tlOutside || (brOutside && tlAlmostOutside)) {
// Okay, so we're texturing from outside the framebuffer, but inside the texture height.
// Breath of Fire 3 does this to access a render surface at an offset.
const u32 bpp = fbman->GetTargetFormat() == GE_FORMAT_8888 ? 4 : 2;
const u32 prevH = texCache->AttachedDrawingHeight();
const u32 fb_size = bpp * fbman->GetTargetStride() * prevH;
const u32 prevYOffset = gstate_c.curTextureYOffset;
if (texCache->SetOffsetTexture(fb_size)) {
const float oldWidthFactor = widthFactor;
const float oldHeightFactor = heightFactor;
widthFactor = (float) w / (float) gstate_c.curTextureWidth;
heightFactor = (float) h / (float) gstate_c.curTextureHeight;
// We've already baked in the old gstate_c.curTextureYOffset, so correct.
const float yDiff = (float) (prevH + prevYOffset - gstate_c.curTextureYOffset) / (float) h;
for (int index = 0; index < maxIndex; ++index) {
transformed[index].u *= widthFactor / oldWidthFactor;
// Inverse it back to scale to the new FBO, and add 1.0f to account for old FBO.
transformed[index].v = (transformed[index].v / oldHeightFactor - yDiff) * heightFactor;
}
}
}
}
// Step 2: expand rectangles.
drawBuffer = transformed;
numTrans = 0;
drawIndexed = false;
if (prim != GE_PRIM_RECTANGLES) {
// We can simply draw the unexpanded buffer.
numTrans = vertexCount;
drawIndexed = true;
} else {
bool useBufferedRendering = g_Config.iRenderingMode != FB_NON_BUFFERED_MODE;
if (useBufferedRendering)
ySign = -ySign;
float flippedMatrix[16];
if (!throughmode) {
memcpy(&flippedMatrix, gstate.projMatrix, 16 * sizeof(float));
const bool invertedY = useBufferedRendering ? (gstate_c.vpHeight < 0) : (gstate_c.vpHeight > 0);
if (invertedY) {
flippedMatrix[1] = -flippedMatrix[1];
flippedMatrix[5] = -flippedMatrix[5];
flippedMatrix[9] = -flippedMatrix[9];
flippedMatrix[13] = -flippedMatrix[13];
}
const bool invertedX = gstate_c.vpWidth < 0;
if (invertedX) {
flippedMatrix[0] = -flippedMatrix[0];
flippedMatrix[4] = -flippedMatrix[4];
flippedMatrix[8] = -flippedMatrix[8];
flippedMatrix[12] = -flippedMatrix[12];
}
}
//rectangles always need 2 vertices, disregard the last one if there's an odd number
vertexCount = vertexCount & ~1;
numTrans = 0;
drawBuffer = transformedExpanded;
TransformedVertex *trans = &transformedExpanded[0];
const u16 *indsIn = (const u16 *)inds;
u16 *newInds = inds + vertexCount;
u16 *indsOut = newInds;
maxIndex = 4 * (vertexCount / 2);
for (int i = 0; i < vertexCount; i += 2) {
const TransformedVertex &transVtxTL = transformed[indsIn[i + 0]];
const TransformedVertex &transVtxBR = transformed[indsIn[i + 1]];
// We have to turn the rectangle into two triangles, so 6 points.
// This is 4 verts + 6 indices.
// bottom right
trans[0] = transVtxBR;
// top right
trans[1] = transVtxBR;
trans[1].y = transVtxTL.y;
trans[1].v = transVtxTL.v;
// top left
trans[2] = transVtxBR;
trans[2].x = transVtxTL.x;
trans[2].y = transVtxTL.y;
trans[2].u = transVtxTL.u;
trans[2].v = transVtxTL.v;
// bottom left
trans[3] = transVtxBR;
trans[3].x = transVtxTL.x;
trans[3].u = transVtxTL.u;
// That's the four corners. Now process UV rotation.
if (throughmode)
RotateUVThrough(trans);
else
RotateUV(trans, flippedMatrix, ySign);
// Triangle: BR-TR-TL
indsOut[0] = i * 2 + 0;
indsOut[1] = i * 2 + 1;
indsOut[2] = i * 2 + 2;
// Triangle: BL-BR-TL
indsOut[3] = i * 2 + 3;
indsOut[4] = i * 2 + 0;
indsOut[5] = i * 2 + 2;
trans += 4;
indsOut += 6;
numTrans += 6;
}
inds = newInds;
drawIndexed = true;
// We don't know the color until here, so we have to do it now, instead of in StateMapping.
// Might want to reconsider the order of things later...
if (gstate.isModeClear() && gstate.isClearModeAlphaMask()) {
result->setStencil = true;
if (vertexCount > 1) {
// Take the bottom right alpha value of the first rect as the stencil value.
// Technically, each rect could individually fill its stencil, but most of the
// time they use the same one.
result->stencilValue = transformed[indsIn[1]].color0[3];
} else {
result->stencilValue = 0;
}
}
}
if (gstate.isModeClear()) {
gpuStats.numClears++;
}
result->action = SW_DRAW_PRIMITIVES;
}
