void ShaderManagerVulkan::BoneUpdateUniforms(int dirtyUniforms) {
for (int i = 0; i < 8; i++) {
if (dirtyUniforms & (DIRTY_BONEMATRIX0 << i)) {
ConvertMatrix4x3To4x4(ub_bones.bones[i], gstate.boneMatrix + 12 * i);
}
}
}
// TODO: This probably is not the best interface.
// Also, we should try to merge this into the similar function in DrawEngineCommon.
bool TransformUnit::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 ((gstate.vertType & GE_VTYPE_IDX_MASK) != GE_VTYPE_IDX_NONE) {
const u8 *inds = Memory::GetPointer(gstate_c.indexAddr);
const u16 *inds16 = (const u16 *)inds;
const u32 *inds32 = (const u32 *)inds;
if (inds) {
GetIndexBounds(inds, count, gstate.vertType, &indexLowerBound, &indexUpperBound);
indices.resize(count);
switch (gstate.vertType & GE_VTYPE_IDX_MASK) {
case GE_VTYPE_IDX_8BIT:
for (int i = 0; i < count; ++i) {
indices[i] = inds[i];
}
break;
case GE_VTYPE_IDX_16BIT:
for (int i = 0; i < count; ++i) {
indices[i] = inds16[i];
}
break;
case GE_VTYPE_IDX_32BIT:
WARN_LOG_REPORT_ONCE(simpleIndexes32, G3D, "SimpleVertices: Decoding 32-bit indexes");
for (int i = 0; i < count; ++i) {
// These aren't documented and should be rare. Let's bounds check each one.
if (inds32[i] != (u16)inds32[i]) {
ERROR_LOG_REPORT_ONCE(simpleIndexes32Bounds, G3D, "SimpleVertices: Index outside 16-bit range");
}
indices[i] = (u16)inds32[i];
}
break;
}
} else {
indices.clear();
}
} else {
indices.clear();
}
static std::vector<u32> temp_buffer;
static std::vector<SimpleVertex> simpleVertices;
temp_buffer.resize(65536 * 24 / sizeof(u32));
simpleVertices.resize(indexUpperBound + 1);
VertexDecoder vdecoder;
VertexDecoderOptions options{};
vdecoder.SetVertexType(gstate.vertType, options);
DrawEngineCommon::NormalizeVertices((u8 *)(&simpleVertices[0]), (u8 *)(&temp_buffer[0]), Memory::GetPointer(gstate_c.vertexAddr), &vdecoder, 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);
ScreenCoords screenPos = ClipToScreen(clipPos);
DrawingCoords 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));
}
}
}
// The GE debugger expects these to be set.
gstate_c.curTextureWidth = gstate.getTextureWidth(0);
gstate_c.curTextureHeight = gstate.getTextureHeight(0);
return true;
}
// 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 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]);
}
}
}
// TODO: This probably is not the best interface.
bool TransformUnit::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 ((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(65536 * 24 / sizeof(u32));
simpleVertices.resize(indexUpperBound + 1);
VertexDecoder vdecoder;
VertexDecoderOptions options;
memset(&options, 0, sizeof(options));
options.expandAllUVtoFloat = false; // TODO: True should be fine here
vdecoder.SetVertexType(gstate.vertType, options);
DrawEngineCommon::NormalizeVertices((u8 *)(&simpleVertices[0]), (u8 *)(&temp_buffer[0]), Memory::GetPointer(gstate_c.vertexAddr), &vdecoder, 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);
ScreenCoords screenPos = ClipToScreen(clipPos);
DrawingCoords drawPos = ScreenToDrawing(screenPos);
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 = drawPos.x;
vertices[i].y = drawPos.y;
vertices[i].z = 1.0;
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));
}
}
}
return true;
}
// 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;
}