// Here we must take all the bits of the gstate that determine what the fragment shader will
// look like, and concatenate them together into an ID.
void ComputeFragmentShaderID(FragmentShaderID *id) {
memset(&id->d[0], 0, sizeof(id->d));
if (gstate.isModeClear()) {
// We only need one clear shader, so let's ignore the rest of the bits.
id->d[0] = 1;
} else {
bool lmode = gstate.isUsingSecondaryColor() && gstate.isLightingEnabled();
bool enableFog = gstate.isFogEnabled() && !gstate.isModeThrough();
bool enableAlphaTest = gstate.isAlphaTestEnabled() && !IsAlphaTestTriviallyTrue() && !g_Config.bDisableAlphaTest;
bool enableColorTest = gstate.isColorTestEnabled() && !IsColorTestTriviallyTrue();
bool enableColorDoubling = gstate.isColorDoublingEnabled();
// This isn't really correct, but it's a hack to get doubled blend modes to work more correctly.
bool enableAlphaDoubling = CanDoubleSrcBlendMode();
bool doTextureProjection = gstate.getUVGenMode() == GE_TEXMAP_TEXTURE_MATRIX;
bool doTextureAlpha = gstate.isTextureAlphaUsed();
ReplaceAlphaType stencilToAlpha = ReplaceAlphaWithStencil();
// All texfuncs except replace are the same for RGB as for RGBA with full alpha.
if (gstate_c.textureFullAlpha && gstate.getTextureFunction() != GE_TEXFUNC_REPLACE)
doTextureAlpha = false;
// id->d[0] |= (gstate.isModeClear() & 1);
if (gstate.isTextureMapEnabled()) {
id->d[0] |= 1 << 1;
id->d[0] |= gstate.getTextureFunction() << 2;
id->d[0] |= (doTextureAlpha & 1) << 5; // rgb or rgba
}
id->d[0] |= (lmode & 1) << 7;
id->d[0] |= enableAlphaTest << 8;
if (enableAlphaTest)
id->d[0] |= gstate.getAlphaTestFunction() << 9;
id->d[0] |= enableColorTest << 12;
if (enableColorTest)
id->d[0] |= gstate.getColorTestFunction() << 13; // color test func
id->d[0] |= (enableFog & 1) << 15;
id->d[0] |= (doTextureProjection & 1) << 16;
id->d[0] |= (enableColorDoubling & 1) << 17;
id->d[0] |= (enableAlphaDoubling & 1) << 18;
id->d[0] |= (stencilToAlpha) << 19;
if (stencilToAlpha != REPLACE_ALPHA_NO) {
// 3 bits
id->d[0] |= ReplaceAlphaWithStencilType() << 21;
}
if (enableAlphaTest)
gpuStats.numAlphaTestedDraws++;
else
gpuStats.numNonAlphaTestedDraws++;
}
}
void TransformDrawEngine::ApplyDrawState(int prim) {
// TODO: All this setup is soon so expensive that we'll need dirty flags, or simply do it in the command writes where we detect dirty by xoring. Silly to do all this work on every drawcall.
if (gstate_c.textureChanged != TEXCHANGE_UNCHANGED && !gstate.isModeClear() && gstate.isTextureMapEnabled()) {
textureCache_->SetTexture();
gstate_c.textureChanged = TEXCHANGE_UNCHANGED;
}
// TODO: The top bit of the alpha channel should be written to the stencil bit somehow. This appears to require very expensive multipass rendering :( Alternatively, one could do a
// single fullscreen pass that converts alpha to stencil (or 2 passes, to set both the 0 and 1 values) very easily.
// Set blend
bool wantBlend = !gstate.isModeClear() && gstate.isAlphaBlendEnabled();
glstate.blend.set(wantBlend);
if (wantBlend) {
// This can't be done exactly as there are several PSP blend modes that are impossible to do on OpenGL ES 2.0, and some even on regular OpenGL for desktop.
// HOWEVER - we should be able to approximate the 2x modes in the shader, although they will clip wrongly.
// Examples of seen unimplementable blend states:
// Mortal Kombat Unchained: FixA=0000ff FixB=000080 FuncA=10 FuncB=10
int blendFuncA = gstate.getBlendFuncA();
int blendFuncB = gstate.getBlendFuncB();
GEBlendMode blendFuncEq = gstate.getBlendEq();
if (blendFuncA > GE_SRCBLEND_FIXA) blendFuncA = GE_SRCBLEND_FIXA;
if (blendFuncB > GE_DSTBLEND_FIXB) blendFuncB = GE_DSTBLEND_FIXB;
float constantAlpha = 1.0f;
ReplaceAlphaType replaceAlphaWithStencil = ReplaceAlphaWithStencil();
if (gstate.isStencilTestEnabled() && replaceAlphaWithStencil == REPLACE_ALPHA_NO) {
if (ReplaceAlphaWithStencilType() == STENCIL_VALUE_UNIFORM) {
constantAlpha = (float) gstate.getStencilTestRef() * (1.0f / 255.0f);
}
}
// Shortcut by using GL_ONE where possible, no need to set blendcolor
GLuint glBlendFuncA = blendFuncA == GE_SRCBLEND_FIXA ? blendColor2Func(gstate.getFixA()) : aLookup[blendFuncA];
GLuint glBlendFuncB = blendFuncB == GE_DSTBLEND_FIXB ? blendColor2Func(gstate.getFixB()) : bLookup[blendFuncB];
if (replaceAlphaWithStencil == REPLACE_ALPHA_DUALSOURCE) {
glBlendFuncA = toDualSource(glBlendFuncA);
glBlendFuncB = toDualSource(glBlendFuncB);
}
if (blendFuncA == GE_SRCBLEND_FIXA || blendFuncB == GE_DSTBLEND_FIXB) {
Vec3f fixA = Vec3f::FromRGB(gstate.getFixA());
Vec3f fixB = Vec3f::FromRGB(gstate.getFixB());
if (glBlendFuncA == GL_INVALID_ENUM && glBlendFuncB != GL_INVALID_ENUM) {
// Can use blendcolor trivially.
const float blendColor[4] = {fixA.x, fixA.y, fixA.z, constantAlpha};
glstate.blendColor.set(blendColor);
glBlendFuncA = GL_CONSTANT_COLOR;
} else if (glBlendFuncA != GL_INVALID_ENUM && glBlendFuncB == GL_INVALID_ENUM) {
// Can use blendcolor trivially.
const float blendColor[4] = {fixB.x, fixB.y, fixB.z, constantAlpha};
glstate.blendColor.set(blendColor);
glBlendFuncB = GL_CONSTANT_COLOR;
} else if (glBlendFuncA == GL_INVALID_ENUM && glBlendFuncB == GL_INVALID_ENUM) {
if (blendColorSimilar(fixA, Vec3f::AssignToAll(constantAlpha) - fixB)) {
glBlendFuncA = GL_CONSTANT_COLOR;
glBlendFuncB = GL_ONE_MINUS_CONSTANT_COLOR;
const float blendColor[4] = {fixA.x, fixA.y, fixA.z, constantAlpha};
glstate.blendColor.set(blendColor);
} else if (blendColorSimilar(fixA, fixB)) {
glBlendFuncA = GL_CONSTANT_COLOR;
glBlendFuncB = GL_CONSTANT_COLOR;
const float blendColor[4] = {fixA.x, fixA.y, fixA.z, constantAlpha};
glstate.blendColor.set(blendColor);
} else {
static bool didReportBlend = false;
if (!didReportBlend)
Reporting::ReportMessage("ERROR INVALID blendcolorstate: FixA=%06x FixB=%06x FuncA=%i FuncB=%i", gstate.getFixA(), gstate.getFixB(), gstate.getBlendFuncA(), gstate.getBlendFuncB());
didReportBlend = true;
DEBUG_LOG(G3D, "ERROR INVALID blendcolorstate: FixA=%06x FixB=%06x FuncA=%i FuncB=%i", gstate.getFixA(), gstate.getFixB(), gstate.getBlendFuncA(), gstate.getBlendFuncB());
// Let's approximate, at least. Close is better than totally off.
const bool nearZeroA = blendColorSimilar(fixA, Vec3f::AssignToAll(0.0f), 0.25f);
const bool nearZeroB = blendColorSimilar(fixB, Vec3f::AssignToAll(0.0f), 0.25f);
if (nearZeroA || blendColorSimilar(fixA, Vec3f::AssignToAll(1.0f), 0.25f)) {
glBlendFuncA = nearZeroA ? GL_ZERO : GL_ONE;
glBlendFuncB = GL_CONSTANT_COLOR;
const float blendColor[4] = {fixB.x, fixB.y, fixB.z, constantAlpha};
glstate.blendColor.set(blendColor);
// We need to pick something. Let's go with A as the fixed color.
} else {
glBlendFuncA = GL_CONSTANT_COLOR;
glBlendFuncB = nearZeroB ? GL_ZERO : GL_ONE;
const float blendColor[4] = {fixA.x, fixA.y, fixA.z, constantAlpha};
glstate.blendColor.set(blendColor);
}
}
} else {
// We optimized both, but that's probably not necessary, so let's pick one to be constant.
// For now let's just pick whichever was fixed instead of checking error.
if (blendFuncA == GE_SRCBLEND_FIXA) {
glBlendFuncA = GL_CONSTANT_COLOR;
const float blendColor[4] = {fixA.x, fixA.y, fixA.z, constantAlpha};
glstate.blendColor.set(blendColor);
} else {
glBlendFuncB = GL_CONSTANT_COLOR;
const float blendColor[4] = {fixB.x, fixB.y, fixB.z, constantAlpha};
glstate.blendColor.set(blendColor);
}
}
} else if (constantAlpha < 1.0f) {
const float blendColor[4] = {1.0f, 1.0f, 1.0f, constantAlpha};
glstate.blendColor.set(blendColor);
}
// Some Android devices (especially Mali, it seems) composite badly if there's alpha in the backbuffer.
// So in non-buffered rendering, we will simply consider the dest alpha to be zero in blending equations.
#ifdef ANDROID
if (g_Config.iRenderingMode == FB_NON_BUFFERED_MODE) {
if (glBlendFuncA == GL_DST_ALPHA) glBlendFuncA = GL_ZERO;
if (glBlendFuncB == GL_DST_ALPHA) glBlendFuncB = GL_ZERO;
if (glBlendFuncA == GL_ONE_MINUS_DST_ALPHA) glBlendFuncA = GL_ONE;
if (glBlendFuncB == GL_ONE_MINUS_DST_ALPHA) glBlendFuncB = GL_ONE;
}
#endif
// At this point, through all paths above, glBlendFuncA and glBlendFuncB will be set right somehow.
// The stencil-to-alpha in fragment shader doesn't apply here (blending is enabled), and we shouldn't
// do any blending in the alpha channel as that doesn't seem to happen on PSP. So lacking a better option,
// the only value we can set alpha to here without multipass and dual source alpha is zero (by setting
// the factors to zero). So let's do that.
if (replaceAlphaWithStencil != REPLACE_ALPHA_NO) {
// Let the fragment shader take care of it.
glstate.blendFuncSeparate.set(glBlendFuncA, glBlendFuncB, GL_ONE, GL_ZERO);
} else if (gstate.isStencilTestEnabled()) {
switch (ReplaceAlphaWithStencilType()) {
case STENCIL_VALUE_KEEP:
glstate.blendFuncSeparate.set(glBlendFuncA, glBlendFuncB, GL_ZERO, GL_ONE);
break;
case STENCIL_VALUE_ONE:
// This won't give one but it's our best shot...
glstate.blendFuncSeparate.set(glBlendFuncA, glBlendFuncB, GL_ONE, GL_ONE);
break;
case STENCIL_VALUE_ZERO:
glstate.blendFuncSeparate.set(glBlendFuncA, glBlendFuncB, GL_ZERO, GL_ZERO);
break;
case STENCIL_VALUE_UNIFORM:
// This won't give a correct value (it multiplies) but it may be better than random values.
glstate.blendFuncSeparate.set(glBlendFuncA, glBlendFuncB, GL_CONSTANT_ALPHA, GL_ZERO);
break;
case STENCIL_VALUE_UNKNOWN:
// For now, let's err at zero. This is INVERT or INCR/DECR.
glstate.blendFuncSeparate.set(glBlendFuncA, glBlendFuncB, GL_ZERO, GL_ZERO);
break;
}
} else {
// Retain the existing value when stencil testing is off.
glstate.blendFuncSeparate.set(glBlendFuncA, glBlendFuncB, GL_ZERO, GL_ONE);
}
if (blendFuncEq == GE_BLENDMODE_ABSDIFF) {
WARN_LOG_REPORT_ONCE(blendAbsdiff, G3D, "Unsupported absdiff blend mode");
}
if (((blendFuncEq >= GE_BLENDMODE_MIN) && gl_extensions.EXT_blend_minmax) || gl_extensions.GLES3) {
if (blendFuncEq == GE_BLENDMODE_ABSDIFF && gl_extensions.EXT_shader_framebuffer_fetch) {
// Handle GE_BLENDMODE_ABSDIFF in fragment shader and turn off regular alpha blending here.
glstate.blend.set(false);
} else {
glstate.blendEquation.set(eqLookup[blendFuncEq]);
}
} else {
glstate.blendEquation.set(eqLookupNoMinMax[blendFuncEq]);
}
}
bool alwaysDepthWrite = g_Config.bAlwaysDepthWrite;
bool enableStencilTest = !g_Config.bDisableStencilTest;
// Dither
if (gstate.isDitherEnabled()) {
glstate.dither.enable();
glstate.dither.set(GL_TRUE);
} else
glstate.dither.disable();
if (gstate.isModeClear()) {
#if !defined(USING_GLES2)
// Logic Ops
glstate.colorLogicOp.disable();
#endif
// Culling
glstate.cullFace.disable();
// Depth Test
glstate.depthTest.enable();
glstate.depthFunc.set(GL_ALWAYS);
glstate.depthWrite.set(gstate.isClearModeDepthMask() || alwaysDepthWrite ? GL_TRUE : GL_FALSE);
if (gstate.isClearModeDepthMask() || alwaysDepthWrite) {
framebufferManager_->SetDepthUpdated();
}
// Color Test
bool colorMask = gstate.isClearModeColorMask();
bool alphaMask = gstate.isClearModeAlphaMask();
glstate.colorMask.set(colorMask, colorMask, colorMask, alphaMask);
// Stencil Test
if (alphaMask && enableStencilTest) {
glstate.stencilTest.enable();
glstate.stencilOp.set(GL_REPLACE, GL_REPLACE, GL_REPLACE);
// TODO: In clear mode, the stencil value is set to the alpha value of the vertex.
// A normal clear will be 2 points, the second point has the color.
// We should set "ref" to that value instead of 0.
// In case of clear rectangles, we set it again once we know what the color is.
glstate.stencilFunc.set(GL_ALWAYS, 255, 0xFF);
} else {
glstate.stencilTest.disable();
}
} else {
#if !defined(USING_GLES2)
// Logic Ops
if (gstate.isLogicOpEnabled() && gstate.getLogicOp() != GE_LOGIC_COPY) {
glstate.colorLogicOp.enable();
glstate.logicOp.set(logicOps[gstate.getLogicOp()]);
} else {
glstate.colorLogicOp.disable();
}
#endif
// Set cull
bool cullEnabled = !gstate.isModeThrough() && prim != GE_PRIM_RECTANGLES && gstate.isCullEnabled();
if (cullEnabled) {
glstate.cullFace.enable();
glstate.cullFaceMode.set(cullingMode[gstate.getCullMode()]);
} else {
glstate.cullFace.disable();
}
// Depth Test
if (gstate.isDepthTestEnabled()) {
glstate.depthTest.enable();
glstate.depthFunc.set(ztests[gstate.getDepthTestFunction()]);
glstate.depthWrite.set(gstate.isDepthWriteEnabled() || alwaysDepthWrite ? GL_TRUE : GL_FALSE);
framebufferManager_->SetDepthUpdated();
} else {
glstate.depthTest.disable();
}
// PSP color/alpha mask is per bit but we can only support per byte.
// But let's do that, at least. And let's try a threshold.
bool rmask = (gstate.pmskc & 0xFF) < 128;
bool gmask = ((gstate.pmskc >> 8) & 0xFF) < 128;
bool bmask = ((gstate.pmskc >> 16) & 0xFF) < 128;
bool amask = (gstate.pmska & 0xFF) < 128;
// Let's not write to alpha if stencil isn't enabled.
if (!gstate.isStencilTestEnabled()) {
amask = false;
} else {
// If the stencil type is set to KEEP, we shouldn't write to the stencil/alpha channel.
if (ReplaceAlphaWithStencilType() == STENCIL_VALUE_KEEP) {
amask = false;
}
}
if (g_Config.bAlphaMaskHack) {
amask = true; // Yes, this makes no sense, but it "fixes" the 3rd Birthday by popular demand.
}
glstate.colorMask.set(rmask, gmask, bmask, amask);
// Stencil Test
if (gstate.isStencilTestEnabled() && enableStencilTest) {
glstate.stencilTest.enable();
glstate.stencilFunc.set(ztests[gstate.getStencilTestFunction()],
gstate.getStencilTestRef(),
gstate.getStencilTestMask());
glstate.stencilOp.set(stencilOps[gstate.getStencilOpSFail()], // stencil fail
stencilOps[gstate.getStencilOpZFail()], // depth fail
stencilOps[gstate.getStencilOpZPass()]); // depth pass
} else {
glstate.stencilTest.disable();
}
}
float renderWidthFactor, renderHeightFactor;
float renderWidth, renderHeight;
float renderX, renderY;
bool useBufferedRendering = g_Config.iRenderingMode != FB_NON_BUFFERED_MODE;
if (useBufferedRendering) {
renderX = 0.0f;
renderY = 0.0f;
renderWidth = framebufferManager_->GetRenderWidth();
renderHeight = framebufferManager_->GetRenderHeight();
} else {
// TODO: Aspect-ratio aware and centered
float pixelW = PSP_CoreParameter().pixelWidth;
float pixelH = PSP_CoreParameter().pixelHeight;
CenterRect(&renderX, &renderY, &renderWidth, &renderHeight, 480, 272, pixelW, pixelH);
}
renderWidthFactor = (float)renderWidth / framebufferManager_->GetTargetWidth();
renderHeightFactor = (float)renderHeight / framebufferManager_->GetTargetHeight();
bool throughmode = gstate.isModeThrough();
// Scissor
int scissorX1 = gstate.getScissorX1();
int scissorY1 = gstate.getScissorY1();
int scissorX2 = gstate.getScissorX2() + 1;
int scissorY2 = gstate.getScissorY2() + 1;
// This is a bit of a hack as the render buffer isn't always that size
if (scissorX1 == 0 && scissorY1 == 0
&& scissorX2 >= (int) gstate_c.curRTWidth
&& scissorY2 >= (int) gstate_c.curRTHeight) {
glstate.scissorTest.disable();
} else {
glstate.scissorTest.enable();
glstate.scissorRect.set(
renderX + scissorX1 * renderWidthFactor,
renderY + renderHeight - (scissorY2 * renderHeightFactor),
(scissorX2 - scissorX1) * renderWidthFactor,
(scissorY2 - scissorY1) * renderHeightFactor);
}
/*
int regionX1 = gstate.region1 & 0x3FF;
int regionY1 = (gstate.region1 >> 10) & 0x3FF;
int regionX2 = (gstate.region2 & 0x3FF) + 1;
int regionY2 = ((gstate.region2 >> 10) & 0x3FF) + 1;
*/
int regionX1 = 0;
int regionY1 = 0;
int regionX2 = gstate_c.curRTWidth;
int regionY2 = gstate_c.curRTHeight;
float offsetX = gstate.getOffsetX();
float offsetY = gstate.getOffsetY();
if (throughmode) {
// No viewport transform here. Let's experiment with using region.
glstate.viewport.set(
renderX + (0 + regionX1) * renderWidthFactor,
renderY + (0 - regionY1) * renderHeightFactor,
(regionX2 - regionX1) * renderWidthFactor,
(regionY2 - regionY1) * renderHeightFactor);
glstate.depthRange.set(0.0f, 1.0f);
} else {
// These we can turn into a glViewport call, offset by offsetX and offsetY. Math after.
float vpXa = getFloat24(gstate.viewportx1);
float vpXb = getFloat24(gstate.viewportx2);
float vpYa = getFloat24(gstate.viewporty1);
float vpYb = getFloat24(gstate.viewporty2);
// The viewport transform appears to go like this:
// Xscreen = -offsetX + vpXb + vpXa * Xview
// Yscreen = -offsetY + vpYb + vpYa * Yview
// Zscreen = vpZb + vpZa * Zview
// This means that to get the analogue glViewport we must:
float vpX0 = vpXb - offsetX - vpXa;
float vpY0 = vpYb - offsetY + vpYa; // Need to account for sign of Y
gstate_c.vpWidth = vpXa * 2.0f;
gstate_c.vpHeight = -vpYa * 2.0f;
float vpWidth = fabsf(gstate_c.vpWidth);
float vpHeight = fabsf(gstate_c.vpHeight);
vpX0 *= renderWidthFactor;
vpY0 *= renderHeightFactor;
vpWidth *= renderWidthFactor;
vpHeight *= renderHeightFactor;
vpX0 = (vpXb - offsetX - fabsf(vpXa)) * renderWidthFactor;
// Flip vpY0 to match the OpenGL coordinate system.
vpY0 = renderHeight - (vpYb - offsetY + fabsf(vpYa)) * renderHeightFactor;
glstate.viewport.set(vpX0 + renderX, vpY0 + renderY, vpWidth, vpHeight);
// Sadly, as glViewport takes integers, we will not be able to support sub pixel offsets this way. But meh.
// shaderManager_->DirtyUniform(DIRTY_PROJMATRIX);
float zScale = getFloat24(gstate.viewportz1) / 65535.0f;
float zOff = getFloat24(gstate.viewportz2) / 65535.0f;
float depthRangeMin = zOff - zScale;
float depthRangeMax = zOff + zScale;
glstate.depthRange.set(depthRangeMin, depthRangeMax);
}
}
// Missing: Z depth range
void GenerateFragmentShader(char *buffer) {
char *p = buffer;
// In GLSL ES 3.0, you use "in" variables instead of varying.
bool glslES30 = false;
const char *varying = "varying";
const char *fragColor0 = "gl_FragColor";
const char *texture = "texture2D";
const char *texelFetch = NULL;
bool highpFog = false;
bool highpTexcoord = false;
bool bitwiseOps = false;
#if defined(USING_GLES2)
// Let's wait until we have a real use for this.
// ES doesn't support dual source alpha :(
if (gl_extensions.GLES3) {
WRITE(p, "#version 300 es\n"); // GLSL ES 3.0
fragColor0 = "fragColor0";
texture = "texture";
glslES30 = true;
bitwiseOps = true;
texelFetch = "texelFetch";
} else {
WRITE(p, "#version 100\n"); // GLSL ES 1.0
if (gl_extensions.EXT_gpu_shader4) {
WRITE(p, "#extension GL_EXT_gpu_shader4 : enable\n");
bitwiseOps = true;
texelFetch = "texelFetch2D";
}
}
// PowerVR needs highp to do the fog in MHU correctly.
// Others don't, and some can't handle highp in the fragment shader.
highpFog = (gl_extensions.bugs & BUG_PVR_SHADER_PRECISION_BAD) ? true : false;
highpTexcoord = highpFog;
// GL_NV_shader_framebuffer_fetch available on mobile platform and ES 2.0 only but not desktop
if (gl_extensions.NV_shader_framebuffer_fetch) {
WRITE(p, "#extension GL_NV_shader_framebuffer_fetch : require\n");
}
WRITE(p, "precision lowp float;\n");
#elif !defined(FORCE_OPENGL_2_0)
if (gl_extensions.VersionGEThan(3, 3, 0)) {
fragColor0 = "fragColor0";
texture = "texture";
glslES30 = true;
bitwiseOps = true;
texelFetch = "texelFetch";
WRITE(p, "#version 330\n");
WRITE(p, "#define lowp\n");
WRITE(p, "#define mediump\n");
WRITE(p, "#define highp\n");
} else if (gl_extensions.VersionGEThan(3, 0, 0)) {
fragColor0 = "fragColor0";
bitwiseOps = true;
texelFetch = "texelFetch";
WRITE(p, "#version 130\n");
if (gl_extensions.EXT_gpu_shader4) {
WRITE(p, "#extension GL_EXT_gpu_shader4 : enable\n");
}
// Remove lowp/mediump in non-mobile non-glsl 3 implementations
WRITE(p, "#define lowp\n");
WRITE(p, "#define mediump\n");
WRITE(p, "#define highp\n");
} else {
WRITE(p, "#version 110\n");
if (gl_extensions.EXT_gpu_shader4) {
WRITE(p, "#extension GL_EXT_gpu_shader4 : enable\n");
bitwiseOps = true;
texelFetch = "texelFetch2D";
}
// Remove lowp/mediump in non-mobile non-glsl 3 implementations
WRITE(p, "#define lowp\n");
WRITE(p, "#define mediump\n");
WRITE(p, "#define highp\n");
}
#else
// Need to remove lowp/mediump for Mac
WRITE(p, "#define lowp\n");
WRITE(p, "#define mediump\n");
WRITE(p, "#define highp\n");
#endif
if (glslES30) {
varying = "in";
}
bool lmode = gstate.isUsingSecondaryColor() && gstate.isLightingEnabled();
bool doTexture = gstate.isTextureMapEnabled() && !gstate.isModeClear();
bool enableFog = gstate.isFogEnabled() && !gstate.isModeThrough() && !gstate.isModeClear();
bool enableAlphaTest = gstate.isAlphaTestEnabled() && !IsAlphaTestTriviallyTrue() && !gstate.isModeClear() && !g_Config.bDisableAlphaTest;
bool alphaTestAgainstZero = IsAlphaTestAgainstZero();
bool enableColorTest = gstate.isColorTestEnabled() && !IsColorTestTriviallyTrue() && !gstate.isModeClear();
bool enableColorDoubling = gstate.isColorDoublingEnabled() && gstate.isTextureMapEnabled();
bool doTextureProjection = gstate.getUVGenMode() == GE_TEXMAP_TEXTURE_MATRIX;
bool doTextureAlpha = gstate.isTextureAlphaUsed();
bool doFlatShading = gstate.getShadeMode() == GE_SHADE_FLAT && !gstate.isModeClear();
bool textureAtOffset = gstate_c.curTextureXOffset != 0 || gstate_c.curTextureYOffset != 0;
ReplaceBlendType replaceBlend = ReplaceBlendWithShader();
ReplaceAlphaType stencilToAlpha = ReplaceAlphaWithStencil(replaceBlend);
const char *shading = "";
if (glslES30)
shading = doFlatShading ? "flat" : "";
if (gstate_c.textureFullAlpha && gstate.getTextureFunction() != GE_TEXFUNC_REPLACE)
doTextureAlpha = false;
if (doTexture)
WRITE(p, "uniform sampler2D tex;\n");
if (!gstate.isModeClear() && replaceBlend > REPLACE_BLEND_STANDARD) {
if (!gl_extensions.NV_shader_framebuffer_fetch && replaceBlend == REPLACE_BLEND_COPY_FBO) {
if (!texelFetch) {
WRITE(p, "uniform vec2 u_fbotexSize;\n");
}
WRITE(p, "uniform sampler2D fbotex;\n");
}
if (gstate.getBlendFuncA() == GE_SRCBLEND_FIXA) {
WRITE(p, "uniform vec3 u_blendFixA;\n");
}
if (gstate.getBlendFuncB() == GE_DSTBLEND_FIXB) {
WRITE(p, "uniform vec3 u_blendFixB;\n");
}
}
if (gstate_c.needShaderTexClamp && doTexture) {
WRITE(p, "uniform vec4 u_texclamp;\n");
if (textureAtOffset) {
WRITE(p, "uniform vec2 u_texclampoff;\n");
}
}
if (enableAlphaTest || enableColorTest) {
if (g_Config.bFragmentTestCache) {
WRITE(p, "uniform sampler2D testtex;\n");
} else {
WRITE(p, "uniform vec4 u_alphacolorref;\n");
if (bitwiseOps && (enableColorTest || !alphaTestAgainstZero)) {
WRITE(p, "uniform ivec4 u_alphacolormask;\n");
}
}
}
if (stencilToAlpha && ReplaceAlphaWithStencilType() == STENCIL_VALUE_UNIFORM) {
WRITE(p, "uniform float u_stencilReplaceValue;\n");
}
if (gstate.isTextureMapEnabled() && gstate.getTextureFunction() == GE_TEXFUNC_BLEND)
WRITE(p, "uniform vec3 u_texenv;\n");
WRITE(p, "%s %s vec4 v_color0;\n", shading, varying);
if (lmode)
WRITE(p, "%s %s vec3 v_color1;\n", shading, varying);
if (enableFog) {
WRITE(p, "uniform vec3 u_fogcolor;\n");
WRITE(p, "%s %s float v_fogdepth;\n", varying, highpFog ? "highp" : "mediump");
}
if (doTexture) {
if (doTextureProjection)
WRITE(p, "%s %s vec3 v_texcoord;\n", varying, highpTexcoord ? "highp" : "mediump");
else
WRITE(p, "%s %s vec2 v_texcoord;\n", varying, highpTexcoord ? "highp" : "mediump");
}
if (!g_Config.bFragmentTestCache) {
if (enableAlphaTest && !alphaTestAgainstZero) {
if (bitwiseOps) {
WRITE(p, "int roundAndScaleTo255i(in float x) { return int(floor(x * 255.0 + 0.5)); }\n");
} else if (gl_extensions.gpuVendor == GPU_VENDOR_POWERVR) {
WRITE(p, "float roundTo255thf(in mediump float x) { mediump float y = x + (0.5/255.0); return y - fract(y * 255.0) * (1.0 / 255.0); }\n");
} else {
WRITE(p, "float roundAndScaleTo255f(in float x) { return floor(x * 255.0 + 0.5); }\n");
}
}
if (enableColorTest) {
if (bitwiseOps) {
WRITE(p, "ivec3 roundAndScaleTo255iv(in vec3 x) { return ivec3(floor(x * 255.0 + 0.5)); }\n");
} else if (gl_extensions.gpuVendor == GPU_VENDOR_POWERVR) {
WRITE(p, "vec3 roundTo255thv(in vec3 x) { vec3 y = x + (0.5/255.0); return y - fract(y * 255.0) * (1.0 / 255.0); }\n");
} else {
WRITE(p, "vec3 roundAndScaleTo255v(in vec3 x) { return floor(x * 255.0 + 0.5); }\n");
}
}
}
if (stencilToAlpha == REPLACE_ALPHA_DUALSOURCE) {
WRITE(p, "out vec4 fragColor0;\n");
WRITE(p, "out vec4 fragColor1;\n");
} else if (!strcmp(fragColor0, "fragColor0")) {
WRITE(p, "out vec4 fragColor0;\n");
}
// PowerVR needs a custom modulo function. For some reason, this has far higher precision than the builtin one.
if ((gl_extensions.bugs & BUG_PVR_SHADER_PRECISION_BAD) && gstate_c.needShaderTexClamp) {
WRITE(p, "float mymod(float a, float b) { return a - b * floor(a / b); }\n");
}
WRITE(p, "void main() {\n");
if (gstate.isModeClear()) {
// Clear mode does not allow any fancy shading.
WRITE(p, " vec4 v = v_color0;\n");
} else {
const char *secondary = "";
// Secondary color for specular on top of texture
if (lmode) {
WRITE(p, " vec4 s = vec4(v_color1, 0.0);\n");
secondary = " + s";
} else {
secondary = "";
}
if (gstate.isTextureMapEnabled()) {
const char *texcoord = "v_texcoord";
// TODO: Not sure the right way to do this for projection.
// This path destroys resolution on older PowerVR no matter what I do, so we disable it on SGX 540 and lesser, and live with the consequences.
if (gstate_c.needShaderTexClamp && !(gl_extensions.bugs & BUG_PVR_SHADER_PRECISION_TERRIBLE)) {
// We may be clamping inside a larger surface (tex = 64x64, buffer=480x272).
// We may also be wrapping in such a surface, or either one in a too-small surface.
// Obviously, clamping to a smaller surface won't work. But better to clamp to something.
std::string ucoord = "v_texcoord.x";
std::string vcoord = "v_texcoord.y";
if (doTextureProjection) {
ucoord += " / v_texcoord.z";
vcoord = "(v_texcoord.y / v_texcoord.z)";
// Vertex texcoords are NOT flipped when projecting despite gstate_c.flipTexture.
} else if (gstate_c.flipTexture) {
vcoord = "1.0 - " + vcoord;
}
std::string modulo = (gl_extensions.bugs & BUG_PVR_SHADER_PRECISION_BAD) ? "mymod" : "mod";
if (gstate.isTexCoordClampedS()) {
ucoord = "clamp(" + ucoord + ", u_texclamp.z, u_texclamp.x - u_texclamp.z)";
} else {
ucoord = modulo + "(" + ucoord + ", u_texclamp.x)";
}
if (gstate.isTexCoordClampedT()) {
vcoord = "clamp(" + vcoord + ", u_texclamp.w, u_texclamp.y - u_texclamp.w)";
} else {
vcoord = modulo + "(" + vcoord + ", u_texclamp.y)";
}
if (textureAtOffset) {
ucoord = "(" + ucoord + " + u_texclampoff.x)";
vcoord = "(" + vcoord + " + u_texclampoff.y)";
}
if (gstate_c.flipTexture) {
vcoord = "1.0 - " + vcoord;
}
WRITE(p, " vec2 fixedcoord = vec2(%s, %s);\n", ucoord.c_str(), vcoord.c_str());
texcoord = "fixedcoord";
// We already projected it.
doTextureProjection = false;
} else if (doTextureProjection && gstate_c.flipTexture) {
// Since we need to flip v, we project manually.
WRITE(p, " vec2 fixedcoord = vec2(v_texcoord.x / v_texcoord.z, 1.0 - (v_texcoord.y / v_texcoord.z));\n");
texcoord = "fixedcoord";
doTextureProjection = false;
}
if (doTextureProjection) {
WRITE(p, " vec4 t = %sProj(tex, %s);\n", texture, texcoord);
} else {
WRITE(p, " vec4 t = %s(tex, %s);\n", texture, texcoord);
}
WRITE(p, " vec4 p = v_color0;\n");
if (doTextureAlpha) { // texfmt == RGBA
switch (gstate.getTextureFunction()) {
case GE_TEXFUNC_MODULATE:
WRITE(p, " vec4 v = p * t%s;\n", secondary);
break;
case GE_TEXFUNC_DECAL:
WRITE(p, " vec4 v = vec4(mix(p.rgb, t.rgb, t.a), p.a)%s;\n", secondary);
break;
case GE_TEXFUNC_BLEND:
WRITE(p, " vec4 v = vec4(mix(p.rgb, u_texenv.rgb, t.rgb), p.a * t.a)%s;\n", secondary);
break;
case GE_TEXFUNC_REPLACE:
WRITE(p, " vec4 v = t%s;\n", secondary);
break;
case GE_TEXFUNC_ADD:
case GE_TEXFUNC_UNKNOWN1:
case GE_TEXFUNC_UNKNOWN2:
case GE_TEXFUNC_UNKNOWN3:
WRITE(p, " vec4 v = vec4(p.rgb + t.rgb, p.a * t.a)%s;\n", secondary);
break;
default:
WRITE(p, " vec4 v = p;\n"); break;
}
} else { // texfmt == RGB
switch (gstate.getTextureFunction()) {
case GE_TEXFUNC_MODULATE:
WRITE(p, " vec4 v = vec4(t.rgb * p.rgb, p.a)%s;\n", secondary);
break;
case GE_TEXFUNC_DECAL:
WRITE(p, " vec4 v = vec4(t.rgb, p.a)%s;\n", secondary);
break;
case GE_TEXFUNC_BLEND:
WRITE(p, " vec4 v = vec4(mix(p.rgb, u_texenv.rgb, t.rgb), p.a)%s;\n", secondary);
break;
case GE_TEXFUNC_REPLACE:
WRITE(p, " vec4 v = vec4(t.rgb, p.a)%s;\n", secondary);
break;
case GE_TEXFUNC_ADD:
case GE_TEXFUNC_UNKNOWN1:
case GE_TEXFUNC_UNKNOWN2:
case GE_TEXFUNC_UNKNOWN3:
WRITE(p, " vec4 v = vec4(p.rgb + t.rgb, p.a)%s;\n", secondary); break;
default:
WRITE(p, " vec4 v = p;\n"); break;
}
}
} else {
// No texture mapping
WRITE(p, " vec4 v = v_color0 %s;\n", secondary);
}
// Texture access is at half texels [0.5/256, 255.5/256], but colors are normalized [0, 255].
// So we have to scale to account for the difference.
std::string alphaTestXCoord = "0";
if (g_Config.bFragmentTestCache) {
if (enableColorTest) {
WRITE(p, " vec4 vScale256 = v * %f + %f;\n", 255.0 / 256.0, 0.5 / 256.0);
alphaTestXCoord = "vScale256.a";
} else if (enableAlphaTest && !alphaTestAgainstZero) {
char temp[64];
snprintf(temp, sizeof(temp), "v.a * %f + %f", 255.0 / 256.0, 0.5 / 256.0);
alphaTestXCoord = temp;
}
}
if (enableAlphaTest) {
if (alphaTestAgainstZero) {
GEComparison alphaTestFunc = gstate.getAlphaTestFunction();
// When testing against 0 (extremely common), we can avoid some math.
// 0.002 is approximately half of 1.0 / 255.0.
if (alphaTestFunc == GE_COMP_NOTEQUAL || alphaTestFunc == GE_COMP_GREATER) {
WRITE(p, " if (v.a < 0.002) discard;\n");
} else if (alphaTestFunc != GE_COMP_NEVER) {
// Anything else is a test for == 0. Happens sometimes, actually...
WRITE(p, " if (v.a > 0.002) discard;\n");
} else {
// NEVER has been logged as used by games, although it makes little sense - statically failing.
// Maybe we could discard the drawcall, but it's pretty rare. Let's just statically discard here.
WRITE(p, " discard;\n");
}
} else if (g_Config.bFragmentTestCache) {
WRITE(p, " float aResult = %s(testtex, vec2(%s, 0)).a;\n", texture, alphaTestXCoord.c_str());
WRITE(p, " if (aResult < 0.5) discard;\n");
} else {
GEComparison alphaTestFunc = gstate.getAlphaTestFunction();
const char *alphaTestFuncs[] = { "#", "#", " != ", " == ", " >= ", " > ", " <= ", " < " };
if (alphaTestFuncs[alphaTestFunc][0] != '#') {
if (bitwiseOps) {
WRITE(p, " if ((roundAndScaleTo255i(v.a) & u_alphacolormask.a) %s int(u_alphacolorref.a)) discard;\n", alphaTestFuncs[alphaTestFunc]);
} else if (gl_extensions.gpuVendor == GPU_VENDOR_POWERVR) {
// Work around bad PVR driver problem where equality check + discard just doesn't work.
if (alphaTestFunc != GE_COMP_NOTEQUAL) {
WRITE(p, " if (roundTo255thf(v.a) %s u_alphacolorref.a) discard;\n", alphaTestFuncs[alphaTestFunc]);
}
} else {
WRITE(p, " if (roundAndScaleTo255f(v.a) %s u_alphacolorref.a) discard;\n", alphaTestFuncs[alphaTestFunc]);
}
} else {
// This means NEVER. See above.
WRITE(p, " discard;\n");
}
}
}
if (enableColorTest) {
if (g_Config.bFragmentTestCache) {
WRITE(p, " float rResult = %s(testtex, vec2(vScale256.r, 0)).r;\n", texture);
WRITE(p, " float gResult = %s(testtex, vec2(vScale256.g, 0)).g;\n", texture);
WRITE(p, " float bResult = %s(testtex, vec2(vScale256.b, 0)).b;\n", texture);
GEComparison colorTestFunc = gstate.getColorTestFunction();
if (colorTestFunc == GE_COMP_EQUAL) {
// Equal means all parts must be equal.
WRITE(p, " if (rResult < 0.5 || gResult < 0.5 || bResult < 0.5) discard;\n");
} else {
// Not equal means any part must be not equal.
WRITE(p, " if (rResult < 0.5 && gResult < 0.5 && bResult < 0.5) discard;\n");
}
} else {
GEComparison colorTestFunc = gstate.getColorTestFunction();
const char *colorTestFuncs[] = { "#", "#", " != ", " == " };
if (colorTestFuncs[colorTestFunc][0] != '#') {
if (bitwiseOps) {
// Apparently GLES3 does not support vector bitwise ops.
WRITE(p, " ivec3 v_scaled = roundAndScaleTo255iv(v.rgb);\n");
const char *maskedFragColor = "ivec3(v_scaled.r & u_alphacolormask.r, v_scaled.g & u_alphacolormask.g, v_scaled.b & u_alphacolormask.b)";
const char *maskedColorRef = "ivec3(int(u_alphacolorref.r) & u_alphacolormask.r, int(u_alphacolorref.g) & u_alphacolormask.g, int(u_alphacolorref.b) & u_alphacolormask.b)";
WRITE(p, " if (%s %s %s) discard;\n", maskedFragColor, colorTestFuncs[colorTestFunc], maskedColorRef);
} else if (gl_extensions.gpuVendor == GPU_VENDOR_POWERVR) {
WRITE(p, " if (roundTo255thv(v.rgb) %s u_alphacolorref.rgb) discard;\n", colorTestFuncs[colorTestFunc]);
} else {
WRITE(p, " if (roundAndScaleTo255v(v.rgb) %s u_alphacolorref.rgb) discard;\n", colorTestFuncs[colorTestFunc]);
}
} else {
WRITE(p, " discard;\n");
}
}
}
// Color doubling happens after the color test.
if (enableColorDoubling && replaceBlend == REPLACE_BLEND_2X_SRC) {
WRITE(p, " v.rgb = v.rgb * 4.0;\n");
} else if (enableColorDoubling || replaceBlend == REPLACE_BLEND_2X_SRC) {
WRITE(p, " v.rgb = v.rgb * 2.0;\n");
}
if (enableFog) {
WRITE(p, " float fogCoef = clamp(v_fogdepth, 0.0, 1.0);\n");
WRITE(p, " v = mix(vec4(u_fogcolor, v.a), v, fogCoef);\n");
// WRITE(p, " v.x = v_depth;\n");
}
if (replaceBlend == REPLACE_BLEND_PRE_SRC || replaceBlend == REPLACE_BLEND_PRE_SRC_2X_ALPHA) {
GEBlendSrcFactor funcA = gstate.getBlendFuncA();
const char *srcFactor = "ERROR";
switch (funcA) {
case GE_SRCBLEND_DSTCOLOR: srcFactor = "ERROR"; break;
case GE_SRCBLEND_INVDSTCOLOR: srcFactor = "ERROR"; break;
case GE_SRCBLEND_SRCALPHA: srcFactor = "vec3(v.a)"; break;
case GE_SRCBLEND_INVSRCALPHA: srcFactor = "vec3(1.0 - v.a)"; break;
case GE_SRCBLEND_DSTALPHA: srcFactor = "ERROR"; break;
case GE_SRCBLEND_INVDSTALPHA: srcFactor = "ERROR"; break;
case GE_SRCBLEND_DOUBLESRCALPHA: srcFactor = "vec3(v.a * 2.0)"; break;
// TODO: Double inverse, or inverse double? Following softgpu for now...
case GE_SRCBLEND_DOUBLEINVSRCALPHA: srcFactor = "vec3(1.0 - v.a * 2.0)"; break;
case GE_SRCBLEND_DOUBLEDSTALPHA: srcFactor = "ERROR"; break;
case GE_SRCBLEND_DOUBLEINVDSTALPHA: srcFactor = "ERROR"; break;
case GE_SRCBLEND_FIXA: srcFactor = "u_blendFixA"; break;
}
WRITE(p, " v.rgb = v.rgb * %s;\n", srcFactor);
}
if (replaceBlend == REPLACE_BLEND_COPY_FBO) {
// If we have NV_shader_framebuffer_fetch / EXT_shader_framebuffer_fetch, we skip the blit.
// We can just read the prev value more directly.
// TODO: EXT_shader_framebuffer_fetch on iOS 6, possibly others.
if (gl_extensions.NV_shader_framebuffer_fetch) {
WRITE(p, " lowp vec4 destColor = gl_LastFragData[0];\n");
} else if (!texelFetch) {
WRITE(p, " lowp vec4 destColor = %s(fbotex, gl_FragCoord.xy * u_fbotexSize.xy);\n", texture);
} else {
WRITE(p, " lowp vec4 destColor = %s(fbotex, ivec2(gl_FragCoord.x, gl_FragCoord.y), 0);\n", texelFetch);
}
GEBlendSrcFactor funcA = gstate.getBlendFuncA();
GEBlendDstFactor funcB = gstate.getBlendFuncB();
GEBlendMode eq = gstate.getBlendEq();
const char *srcFactor = "vec3(1.0)";
const char *dstFactor = "vec3(0.0)";
switch (funcA)
{
case GE_SRCBLEND_DSTCOLOR: srcFactor = "destColor.rgb"; break;
case GE_SRCBLEND_INVDSTCOLOR: srcFactor = "(vec3(1.0) - destColor.rgb)"; break;
case GE_SRCBLEND_SRCALPHA: srcFactor = "vec3(v.a)"; break;
case GE_SRCBLEND_INVSRCALPHA: srcFactor = "vec3(1.0 - v.a)"; break;
case GE_SRCBLEND_DSTALPHA: srcFactor = "vec3(destColor.a)"; break;
case GE_SRCBLEND_INVDSTALPHA: srcFactor = "vec3(1.0 - destColor.a)"; break;
case GE_SRCBLEND_DOUBLESRCALPHA: srcFactor = "vec3(v.a * 2.0)"; break;
// TODO: Double inverse, or inverse double? Following softgpu for now...
case GE_SRCBLEND_DOUBLEINVSRCALPHA: srcFactor = "vec3(1.0 - v.a * 2.0)"; break;
case GE_SRCBLEND_DOUBLEDSTALPHA: srcFactor = "vec3(destColor.a * 2.0)"; break;
case GE_SRCBLEND_DOUBLEINVDSTALPHA: srcFactor = "vec3(1.0 - destColor.a * 2.0)"; break;
case GE_SRCBLEND_FIXA: srcFactor = "u_blendFixA"; break;
}
switch (funcB)
{
case GE_DSTBLEND_SRCCOLOR: dstFactor = "v.rgb"; break;
case GE_DSTBLEND_INVSRCCOLOR: dstFactor = "(vec3(1.0) - v.rgb)"; break;
case GE_DSTBLEND_SRCALPHA: dstFactor = "vec3(v.a)"; break;
case GE_DSTBLEND_INVSRCALPHA: dstFactor = "vec3(1.0 - v.a)"; break;
case GE_DSTBLEND_DSTALPHA: dstFactor = "vec3(destColor.a)"; break;
case GE_DSTBLEND_INVDSTALPHA: dstFactor = "vec3(1.0 - destColor.a)"; break;
case GE_DSTBLEND_DOUBLESRCALPHA: dstFactor = "vec3(v.a * 2.0)"; break;
case GE_DSTBLEND_DOUBLEINVSRCALPHA: dstFactor = "vec3(1.0 - v.a * 2.0)"; break;
case GE_DSTBLEND_DOUBLEDSTALPHA: dstFactor = "vec3(destColor.a * 2.0)"; break;
case GE_DSTBLEND_DOUBLEINVDSTALPHA: dstFactor = "vec3(1.0 - destColor.a * 2.0)"; break;
case GE_DSTBLEND_FIXB: dstFactor = "u_blendFixB"; break;
}
switch (eq)
{
case GE_BLENDMODE_MUL_AND_ADD:
WRITE(p, " v.rgb = v.rgb * %s + destColor.rgb * %s;\n", srcFactor, dstFactor);
break;
case GE_BLENDMODE_MUL_AND_SUBTRACT:
WRITE(p, " v.rgb = v.rgb * %s - destColor.rgb * %s;\n", srcFactor, dstFactor);
break;
case GE_BLENDMODE_MUL_AND_SUBTRACT_REVERSE:
WRITE(p, " v.rgb = destColor.rgb * %s - v.rgb * %s;\n", dstFactor, srcFactor);
break;
case GE_BLENDMODE_MIN:
WRITE(p, " v.rgb = min(v.rgb, destColor.rgb);\n");
break;
case GE_BLENDMODE_MAX:
WRITE(p, " v.rgb = max(v.rgb, destColor.rgb);\n");
break;
case GE_BLENDMODE_ABSDIFF:
WRITE(p, " v.rgb = abs(v.rgb - destColor.rgb);\n");
break;
}
}
if (replaceBlend == REPLACE_BLEND_2X_ALPHA || replaceBlend == REPLACE_BLEND_PRE_SRC_2X_ALPHA) {
WRITE(p, " v.a = v.a * 2.0;\n");
}
}
std::string replacedAlpha = "0.0";
char replacedAlphaTemp[64] = "";
if (stencilToAlpha != REPLACE_ALPHA_NO) {
switch (ReplaceAlphaWithStencilType()) {
case STENCIL_VALUE_UNIFORM:
replacedAlpha = "u_stencilReplaceValue";
break;
case STENCIL_VALUE_ZERO:
replacedAlpha = "0.0";
break;
case STENCIL_VALUE_ONE:
case STENCIL_VALUE_INVERT:
// In invert, we subtract by one, but we want to output one here.
replacedAlpha = "1.0";
break;
case STENCIL_VALUE_INCR_4:
case STENCIL_VALUE_DECR_4:
// We're adding/subtracting, just by the smallest value in 4-bit.
snprintf(replacedAlphaTemp, sizeof(replacedAlphaTemp), "%f", 1.0 / 15.0);
replacedAlpha = replacedAlphaTemp;
break;
case STENCIL_VALUE_INCR_8:
case STENCIL_VALUE_DECR_8:
// We're adding/subtracting, just by the smallest value in 8-bit.
snprintf(replacedAlphaTemp, sizeof(replacedAlphaTemp), "%f", 1.0 / 255.0);
replacedAlpha = replacedAlphaTemp;
break;
case STENCIL_VALUE_KEEP:
// Do nothing. We'll mask out the alpha using color mask.
break;
}
}
switch (stencilToAlpha) {
case REPLACE_ALPHA_DUALSOURCE:
WRITE(p, " fragColor0 = vec4(v.rgb, %s);\n", replacedAlpha.c_str());
WRITE(p, " fragColor1 = vec4(0.0, 0.0, 0.0, v.a);\n");
break;
case REPLACE_ALPHA_YES:
WRITE(p, " %s = vec4(v.rgb, %s);\n", fragColor0, replacedAlpha.c_str());
break;
case REPLACE_ALPHA_NO:
WRITE(p, " %s = v;\n", fragColor0);
break;
}
switch (ReplaceLogicOpType()) {
case LOGICOPTYPE_ONE:
WRITE(p, " %s.rgb = vec3(1.0, 1.0, 1.0);\n", fragColor0);
break;
case LOGICOPTYPE_INVERT:
WRITE(p, " %s.rgb = vec3(1.0, 1.0, 1.0) - %s.rgb;\n", fragColor0, fragColor0);
break;
case LOGICOPTYPE_NORMAL:
break;
}
#ifdef DEBUG_SHADER
if (doTexture) {
WRITE(p, " %s = texture2D(tex, v_texcoord.xy);\n", fragColor0);
WRITE(p, " %s += vec4(0.3,0,0.3,0.3);\n", fragColor0);
} else {
WRITE(p, " %s = vec4(1,0,1,1);\n", fragColor0);
}
#endif
WRITE(p, "}\n");
}
// Here we must take all the bits of the gstate that determine what the fragment shader will
// look like, and concatenate them together into an ID.
void ComputeFragmentShaderID(ShaderID *id) {
int id0 = 0;
int id1 = 0;
if (gstate.isModeClear()) {
// We only need one clear shader, so let's ignore the rest of the bits.
id0 = 1;
} else {
bool lmode = gstate.isUsingSecondaryColor() && gstate.isLightingEnabled();
bool enableFog = gstate.isFogEnabled() && !gstate.isModeThrough();
bool enableAlphaTest = gstate.isAlphaTestEnabled() && !IsAlphaTestTriviallyTrue() && !g_Config.bDisableAlphaTest;
bool alphaTestAgainstZero = IsAlphaTestAgainstZero();
bool enableColorTest = gstate.isColorTestEnabled() && !IsColorTestTriviallyTrue();
bool enableColorDoubling = gstate.isColorDoublingEnabled() && gstate.isTextureMapEnabled();
bool doTextureProjection = gstate.getUVGenMode() == GE_TEXMAP_TEXTURE_MATRIX;
bool doTextureAlpha = gstate.isTextureAlphaUsed();
bool doFlatShading = gstate.getShadeMode() == GE_SHADE_FLAT;
ReplaceBlendType replaceBlend = ReplaceBlendWithShader();
ReplaceAlphaType stencilToAlpha = ReplaceAlphaWithStencil(replaceBlend);
// All texfuncs except replace are the same for RGB as for RGBA with full alpha.
if (gstate_c.textureFullAlpha && gstate.getTextureFunction() != GE_TEXFUNC_REPLACE)
doTextureAlpha = false;
// id0 |= (gstate.isModeClear() & 1);
if (gstate.isTextureMapEnabled()) {
id0 |= 1 << 1;
id0 |= gstate.getTextureFunction() << 2;
id0 |= (doTextureAlpha & 1) << 5; // rgb or rgba
id0 |= (gstate_c.flipTexture & 1) << 6;
if (gstate_c.needShaderTexClamp) {
bool textureAtOffset = gstate_c.curTextureXOffset != 0 || gstate_c.curTextureYOffset != 0;
// 3 bits total.
id0 |= 1 << 7;
id0 |= gstate.isTexCoordClampedS() << 8;
id0 |= gstate.isTexCoordClampedT() << 9;
id0 |= (textureAtOffset & 1) << 10;
}
}
id0 |= (lmode & 1) << 11;
#if !defined(DX9_USE_HW_ALPHA_TEST)
if (enableAlphaTest) {
// 4 bits total.
id0 |= 1 << 12;
id0 |= gstate.getAlphaTestFunction() << 13;
}
#endif
if (enableColorTest) {
// 3 bits total.
id0 |= 1 << 16;
id0 |= gstate.getColorTestFunction() << 17;
}
id0 |= (enableFog & 1) << 19;
id0 |= (doTextureProjection & 1) << 20;
id0 |= (enableColorDoubling & 1) << 21;
// 2 bits
id0 |= (stencilToAlpha) << 22;
if (stencilToAlpha != REPLACE_ALPHA_NO) {
// 4 bits
id0 |= ReplaceAlphaWithStencilType() << 24;
}
id0 |= (alphaTestAgainstZero & 1) << 28;
if (enableAlphaTest)
gpuStats.numAlphaTestedDraws++;
else
gpuStats.numNonAlphaTestedDraws++;
// 29 is free.
// 2 bits.
id0 |= ReplaceLogicOpType() << 30;
// 3 bits.
id1 |= replaceBlend << 0;
if (replaceBlend > REPLACE_BLEND_STANDARD) {
// 11 bits total.
id1 |= gstate.getBlendEq() << 3;
id1 |= gstate.getBlendFuncA() << 6;
id1 |= gstate.getBlendFuncB() << 10;
}
id1 |= (doFlatShading & 1) << 14;
}
id->d[0] = id0;
id->d[1] = id1;
}
// Missing: Z depth range
void GenerateFragmentShader(char *buffer) {
char *p = buffer;
// In GLSL ES 3.0, you use "in" variables instead of varying.
bool glslES30 = false;
const char *varying = "varying";
const char *fragColor0 = "gl_FragColor";
const char *texture = "texture2D";
bool highpFog = false;
#if defined(USING_GLES2)
// Let's wait until we have a real use for this.
// ES doesn't support dual source alpha :(
if (gl_extensions.GLES3) {
WRITE(p, "#version 300 es\n"); // GLSL ES 1.0
fragColor0 = "fragColor0";
texture = "texture";
glslES30 = true;
} else {
WRITE(p, "#version 100\n"); // GLSL ES 1.0
}
WRITE(p, "precision lowp float;\n");
// PowerVR needs highp to do the fog in MHU correctly.
// Others don't, and some can't handle highp in the fragment shader.
highpFog = gl_extensions.gpuVendor == GPU_VENDOR_POWERVR;
#elif !defined(FORCE_OPENGL_2_0)
if (gl_extensions.VersionGEThan(3, 3, 0)) {
fragColor0 = "fragColor0";
texture = "texture";
glslES30 = true;
WRITE(p, "#version 330\n");
} else if (gl_extensions.VersionGEThan(3, 0, 0)) {
fragColor0 = "fragColor0";
WRITE(p, "#version 130\n");
// Remove lowp/mediump in non-mobile non-glsl 3 implementations
WRITE(p, "#define lowp\n");
WRITE(p, "#define mediump\n");
WRITE(p, "#define highp\n");
} else {
WRITE(p, "#version 110\n");
// Remove lowp/mediump in non-mobile non-glsl 3 implementations
WRITE(p, "#define lowp\n");
WRITE(p, "#define mediump\n");
WRITE(p, "#define highp\n");
}
#else
// Need to remove lowp/mediump for Mac
WRITE(p, "#define lowp\n");
WRITE(p, "#define mediump\n");
WRITE(p, "#define highp\n");
#endif
if (glslES30) {
varying = "in";
}
bool lmode = gstate.isUsingSecondaryColor() && gstate.isLightingEnabled();
bool doTexture = gstate.isTextureMapEnabled() && !gstate.isModeClear();
bool enableFog = gstate.isFogEnabled() && !gstate.isModeThrough() && !gstate.isModeClear();
bool enableAlphaTest = gstate.isAlphaTestEnabled() && !IsAlphaTestTriviallyTrue() && !gstate.isModeClear() && !g_Config.bDisableAlphaTest;
bool enableColorTest = gstate.isColorTestEnabled() && !IsColorTestTriviallyTrue() && !gstate.isModeClear();
bool enableColorDoubling = gstate.isColorDoublingEnabled();
// This isn't really correct, but it's a hack to get doubled blend modes to work more correctly.
bool enableAlphaDoubling = CanDoubleSrcBlendMode();
bool doTextureProjection = gstate.getUVGenMode() == GE_TEXMAP_TEXTURE_MATRIX;
bool doTextureAlpha = gstate.isTextureAlphaUsed();
ReplaceAlphaType stencilToAlpha = ReplaceAlphaWithStencil();
if (gstate_c.textureFullAlpha && gstate.getTextureFunction() != GE_TEXFUNC_REPLACE)
doTextureAlpha = false;
if (doTexture)
WRITE(p, "uniform sampler2D tex;\n");
if (enableAlphaTest || enableColorTest) {
WRITE(p, "uniform vec4 u_alphacolorref;\n");
}
if (stencilToAlpha && ReplaceAlphaWithStencilType() == STENCIL_VALUE_UNIFORM) {
WRITE(p, "uniform float u_stencilReplaceValue;\n");
}
if (gstate.isTextureMapEnabled() && gstate.getTextureFunction() == GE_TEXFUNC_BLEND)
WRITE(p, "uniform vec3 u_texenv;\n");
WRITE(p, "%s vec4 v_color0;\n", varying);
if (lmode)
WRITE(p, "%s vec3 v_color1;\n", varying);
if (enableFog) {
WRITE(p, "uniform vec3 u_fogcolor;\n");
WRITE(p, "%s %s float v_fogdepth;\n", varying, highpFog ? "highp" : "mediump");
}
if (doTexture) {
if (doTextureProjection)
WRITE(p, "%s mediump vec3 v_texcoord;\n", varying);
else
WRITE(p, "%s mediump vec2 v_texcoord;\n", varying);
}
if (enableAlphaTest) {
if (gl_extensions.gpuVendor == GPU_VENDOR_POWERVR)
WRITE(p, "float roundTo255thf(in mediump float x) { mediump float y = x + (0.5/255.0); return y - fract(y * 255.0) * (1.0 / 255.0); }\n");
else
WRITE(p, "float roundAndScaleTo255f(in float x) { return floor(x * 255.0 + 0.5); }\n");
}
if (enableColorTest) {
if (gl_extensions.gpuVendor == GPU_VENDOR_POWERVR)
WRITE(p, "vec3 roundTo255thv(in vec3 x) { vec3 y = x + (0.5/255.0); return y - fract(y * 255.0) * (1.0 / 255.0); }\n");
else
WRITE(p, "vec3 roundAndScaleTo255v(in vec3 x) { return floor(x * 255.0 + 0.5); }\n");
}
if (stencilToAlpha == REPLACE_ALPHA_DUALSOURCE) {
WRITE(p, "out vec4 fragColor0;\n");
WRITE(p, "out vec4 fragColor1;\n");
} else if (gl_extensions.VersionGEThan(3, 0, 0)) {
WRITE(p, "out vec4 fragColor0;\n");
}
WRITE(p, "void main() {\n");
if (gstate.isModeClear()) {
// Clear mode does not allow any fancy shading.
WRITE(p, " vec4 v = v_color0;\n");
} else {
const char *secondary = "";
// Secondary color for specular on top of texture
if (lmode) {
WRITE(p, " vec4 s = vec4(v_color1, 0.0);\n");
secondary = " + s";
} else {
secondary = "";
}
if (gstate.isTextureMapEnabled()) {
if (doTextureProjection) {
WRITE(p, " vec4 t = %sProj(tex, v_texcoord);\n", texture);
} else {
WRITE(p, " vec4 t = %s(tex, v_texcoord);\n", texture);
}
WRITE(p, " vec4 p = v_color0;\n");
if (doTextureAlpha) { // texfmt == RGBA
switch (gstate.getTextureFunction()) {
case GE_TEXFUNC_MODULATE:
WRITE(p, " vec4 v = p * t%s;\n", secondary);
break;
case GE_TEXFUNC_DECAL:
WRITE(p, " vec4 v = vec4(mix(p.rgb, t.rgb, t.a), p.a)%s;\n", secondary);
break;
case GE_TEXFUNC_BLEND:
WRITE(p, " vec4 v = vec4(mix(p.rgb, u_texenv.rgb, t.rgb), p.a * t.a)%s;\n", secondary);
break;
case GE_TEXFUNC_REPLACE:
WRITE(p, " vec4 v = t%s;\n", secondary);
break;
case GE_TEXFUNC_ADD:
case GE_TEXFUNC_UNKNOWN1:
case GE_TEXFUNC_UNKNOWN2:
case GE_TEXFUNC_UNKNOWN3:
WRITE(p, " vec4 v = vec4(p.rgb + t.rgb, p.a * t.a)%s;\n", secondary);
break;
default:
WRITE(p, " vec4 v = p;\n"); break;
}
} else { // texfmt == RGB
switch (gstate.getTextureFunction()) {
case GE_TEXFUNC_MODULATE:
WRITE(p, " vec4 v = vec4(t.rgb * p.rgb, p.a)%s;\n", secondary);
break;
case GE_TEXFUNC_DECAL:
WRITE(p, " vec4 v = vec4(t.rgb, p.a)%s;\n", secondary);
break;
case GE_TEXFUNC_BLEND:
WRITE(p, " vec4 v = vec4(mix(p.rgb, u_texenv.rgb, t.rgb), p.a)%s;\n", secondary);
break;
case GE_TEXFUNC_REPLACE:
WRITE(p, " vec4 v = vec4(t.rgb, p.a)%s;\n", secondary);
break;
case GE_TEXFUNC_ADD:
case GE_TEXFUNC_UNKNOWN1:
case GE_TEXFUNC_UNKNOWN2:
case GE_TEXFUNC_UNKNOWN3:
WRITE(p, " vec4 v = vec4(p.rgb + t.rgb, p.a)%s;\n", secondary); break;
default:
WRITE(p, " vec4 v = p;\n"); break;
}
}
} else {
// No texture mapping
WRITE(p, " vec4 v = v_color0 %s;\n", secondary);
}
if (enableAlphaTest) {
GEComparison alphaTestFunc = gstate.getAlphaTestFunction();
const char *alphaTestFuncs[] = { "#", "#", " != ", " == ", " >= ", " > ", " <= ", " < " }; // never/always don't make sense
if (alphaTestFuncs[alphaTestFunc][0] != '#') {
if (gl_extensions.gpuVendor == GPU_VENDOR_POWERVR) {
// Work around bad PVR driver problem where equality check + discard just doesn't work.
if (alphaTestFunc != 3)
WRITE(p, " if (roundTo255thf(v.a) %s u_alphacolorref.a) discard;\n", alphaTestFuncs[alphaTestFunc]);
} else {
WRITE(p, " if (roundAndScaleTo255f(v.a) %s u_alphacolorref.a) discard;\n", alphaTestFuncs[alphaTestFunc]);
}
}
}
// TODO: Before or after the color test?
if (enableColorDoubling && enableAlphaDoubling) {
WRITE(p, " v = v * 2.0;\n");
} else if (enableColorDoubling) {
WRITE(p, " v.rgb = v.rgb * 2.0;\n");
} else if (enableAlphaDoubling) {
WRITE(p, " v.a = v.a * 2.0;\n");
}
if (enableColorTest) {
GEComparison colorTestFunc = gstate.getColorTestFunction();
const char *colorTestFuncs[] = { "#", "#", " != ", " == " }; // never/always don't make sense
WARN_LOG_REPORT_ONCE(colortest, G3D, "Color test function : %s", colorTestFuncs[colorTestFunc]);
u32 colorTestMask = gstate.getColorTestMask();
if (colorTestFuncs[colorTestFunc][0] != '#') {
if (gl_extensions.gpuVendor == GPU_VENDOR_POWERVR)
WRITE(p, " if (roundTo255thv(v.rgb) %s u_alphacolorref.rgb) discard;\n", colorTestFuncs[colorTestFunc]);
else
WRITE(p, " if (roundAndScaleTo255v(v.rgb) %s u_alphacolorref.rgb) discard;\n", colorTestFuncs[colorTestFunc]);
}
}
if (enableFog) {
WRITE(p, " float fogCoef = clamp(v_fogdepth, 0.0, 1.0);\n");
WRITE(p, " v = mix(vec4(u_fogcolor, v.a), v, fogCoef);\n");
// WRITE(p, " v.x = v_depth;\n");
}
}
switch (stencilToAlpha) {
case REPLACE_ALPHA_DUALSOURCE:
WRITE(p, " fragColor0 = vec4(v.rgb, 0.0);\n");
WRITE(p, " fragColor1 = vec4(0.0, 0.0, 0.0, v.a);\n");
break;
case REPLACE_ALPHA_YES:
WRITE(p, " %s = vec4(v.rgb, 0.0);\n", fragColor0);
break;
case REPLACE_ALPHA_NO:
WRITE(p, " %s = v;\n", fragColor0);
break;
}
if (stencilToAlpha != REPLACE_ALPHA_NO) {
switch (ReplaceAlphaWithStencilType()) {
case STENCIL_VALUE_UNIFORM:
WRITE(p, " %s.a = u_stencilReplaceValue;\n", fragColor0);
break;
case STENCIL_VALUE_ZERO:
WRITE(p, " %s.a = 0.0;\n", fragColor0);
break;
case STENCIL_VALUE_ONE:
WRITE(p, " %s.a = 1.0;\n", fragColor0);
break;
case STENCIL_VALUE_UNKNOWN:
// Maybe we should even mask away alpha using glColorMask and not change it at all? We do get here
// if the stencil mode is KEEP for example.
WRITE(p, " %s.a = 0.0;\n", fragColor0);
break;
case STENCIL_VALUE_KEEP:
// Do nothing. We'll mask out the alpha using color mask.
break;
}
}
#ifdef DEBUG_SHADER
if (doTexture) {
WRITE(p, " %s = texture2D(tex, v_texcoord.xy);\n", fragColor0);
WRITE(p, " %s += vec4(0.3,0,0.3,0.3);\n", fragColor0);
} else {
WRITE(p, " %s = vec4(1,0,1,1);\n", fragColor0);
}
#endif
WRITE(p, "}\n");
}
// Missing: Z depth range
void GenerateFragmentShader(char *buffer) {
char *p = buffer;
// In GLSL ES 3.0, you use "in" variables instead of varying.
bool glslES30 = false;
const char *varying = "varying";
const char *fragColor0 = "gl_FragColor";
const char *texture = "texture2D";
bool highpFog = false;
#if defined(USING_GLES2)
// Let's wait until we have a real use for this.
// ES doesn't support dual source alpha :(
if (gl_extensions.GLES3) {
WRITE(p, "#version 300 es\n"); // GLSL ES 1.0
fragColor0 = "fragColor0";
texture = "texture";
glslES30 = true;
} else {
WRITE(p, "#version 100\n"); // GLSL ES 1.0
}
WRITE(p, "precision lowp float;\n");
// PowerVR needs highp to do the fog in MHU correctly.
// Others don't, and some can't handle highp in the fragment shader.
highpFog = gl_extensions.gpuVendor == GPU_VENDOR_POWERVR;
// GL_NV_shader_framebuffer_fetch available on mobile platform and ES 2.0 only but not desktop
if (gl_extensions.NV_shader_framebuffer_fetch) {
WRITE(p, " #extension GL_NV_shader_framebuffer_fetch : require\n");
}
#elif !defined(FORCE_OPENGL_2_0)
if (gl_extensions.VersionGEThan(3, 3, 0)) {
fragColor0 = "fragColor0";
texture = "texture";
glslES30 = true;
WRITE(p, "#version 330\n");
} else if (gl_extensions.VersionGEThan(3, 0, 0)) {
fragColor0 = "fragColor0";
WRITE(p, "#version 130\n");
// Remove lowp/mediump in non-mobile non-glsl 3 implementations
WRITE(p, "#define lowp\n");
WRITE(p, "#define mediump\n");
WRITE(p, "#define highp\n");
} else {
WRITE(p, "#version 110\n");
// Remove lowp/mediump in non-mobile non-glsl 3 implementations
WRITE(p, "#define lowp\n");
WRITE(p, "#define mediump\n");
WRITE(p, "#define highp\n");
}
#else
// Need to remove lowp/mediump for Mac
WRITE(p, "#define lowp\n");
WRITE(p, "#define mediump\n");
WRITE(p, "#define highp\n");
#endif
if (glslES30) {
varying = "in";
}
bool lmode = gstate.isUsingSecondaryColor() && gstate.isLightingEnabled();
bool doTexture = gstate.isTextureMapEnabled() && !gstate.isModeClear();
bool enableFog = gstate.isFogEnabled() && !gstate.isModeThrough() && !gstate.isModeClear();
bool enableAlphaTest = gstate.isAlphaTestEnabled() && !IsAlphaTestTriviallyTrue() && !gstate.isModeClear();
bool alphaTestAgainstZero = gstate.getAlphaTestRef() == 0;
bool enableColorTest = gstate.isColorTestEnabled() && !IsColorTestTriviallyTrue() && !gstate.isModeClear();
bool alphaToColorDoubling = AlphaToColorDoubling();
bool enableColorDoubling = (gstate.isColorDoublingEnabled() && gstate.isTextureMapEnabled()) || alphaToColorDoubling;
// This isn't really correct, but it's a hack to get doubled blend modes to work more correctly.
bool enableAlphaDoubling = !alphaToColorDoubling && CanDoubleSrcBlendMode();
bool doTextureProjection = gstate.getUVGenMode() == GE_TEXMAP_TEXTURE_MATRIX;
bool doTextureAlpha = gstate.isTextureAlphaUsed();
ReplaceAlphaType stencilToAlpha = ReplaceAlphaWithStencil();
if (gstate_c.textureFullAlpha && gstate.getTextureFunction() != GE_TEXFUNC_REPLACE)
doTextureAlpha = false;
if (doTexture)
WRITE(p, "uniform sampler2D tex;\n");
if (ShouldUseShaderBlending() && !gstate.isModeClear()) {
if (!gl_extensions.NV_shader_framebuffer_fetch) {
WRITE(p, "uniform sampler2D fbotex;\n");
}
if (gstate.getBlendFuncA() == GE_SRCBLEND_FIXA) {
WRITE(p, "uniform vec3 u_blendFixA;\n");
}
if (gstate.getBlendFuncB() == GE_DSTBLEND_FIXB) {
WRITE(p, "uniform vec3 u_blendFixB;\n");
}
}
if (enableAlphaTest || enableColorTest) {
WRITE(p, "uniform vec4 u_alphacolorref;\n");
}
if (stencilToAlpha && ReplaceAlphaWithStencilType() == STENCIL_VALUE_UNIFORM) {
WRITE(p, "uniform float u_stencilReplaceValue;\n");
}
if (gstate.isTextureMapEnabled() && gstate.getTextureFunction() == GE_TEXFUNC_BLEND)
WRITE(p, "uniform vec3 u_texenv;\n");
WRITE(p, "%s vec4 v_color0;\n", varying);
if (lmode)
WRITE(p, "%s vec3 v_color1;\n", varying);
if (enableFog) {
WRITE(p, "uniform vec3 u_fogcolor;\n");
WRITE(p, "%s %s float v_fogdepth;\n", varying, highpFog ? "highp" : "mediump");
}
if (doTexture) {
if (doTextureProjection)
WRITE(p, "%s mediump vec3 v_texcoord;\n", varying);
else
WRITE(p, "%s mediump vec2 v_texcoord;\n", varying);
}
if (enableAlphaTest && !alphaTestAgainstZero) {
if (gl_extensions.gpuVendor == GPU_VENDOR_POWERVR)
WRITE(p, "float roundTo255thf(in mediump float x) { mediump float y = x + (0.5/255.0); return y - fract(y * 255.0) * (1.0 / 255.0); }\n");
else
WRITE(p, "float roundAndScaleTo255f(in float x) { return floor(x * 255.0 + 0.5); }\n");
}
if (enableColorTest) {
if (gl_extensions.gpuVendor == GPU_VENDOR_POWERVR)
WRITE(p, "vec3 roundTo255thv(in vec3 x) { vec3 y = x + (0.5/255.0); return y - fract(y * 255.0) * (1.0 / 255.0); }\n");
else
WRITE(p, "vec3 roundAndScaleTo255v(in vec3 x) { return floor(x * 255.0 + 0.5); }\n");
}
if (stencilToAlpha == REPLACE_ALPHA_DUALSOURCE) {
WRITE(p, "out vec4 fragColor0;\n");
WRITE(p, "out vec4 fragColor1;\n");
} else if (!strcmp(fragColor0, "fragColor0")) {
WRITE(p, "out vec4 fragColor0;\n");
}
WRITE(p, "void main() {\n");
if (gstate.isModeClear()) {
// Clear mode does not allow any fancy shading.
WRITE(p, " vec4 v = v_color0;\n");
} else {
const char *secondary = "";
// Secondary color for specular on top of texture
if (lmode) {
WRITE(p, " vec4 s = vec4(v_color1, 0.0);\n");
secondary = " + s";
} else {
secondary = "";
}
if (gstate.isTextureMapEnabled()) {
if (doTextureProjection) {
WRITE(p, " vec4 t = %sProj(tex, v_texcoord);\n", texture);
} else {
WRITE(p, " vec4 t = %s(tex, v_texcoord);\n", texture);
}
WRITE(p, " vec4 p = v_color0;\n");
if (doTextureAlpha) { // texfmt == RGBA
switch (gstate.getTextureFunction()) {
case GE_TEXFUNC_MODULATE:
WRITE(p, " vec4 v = p * t%s;\n", secondary);
break;
case GE_TEXFUNC_DECAL:
WRITE(p, " vec4 v = vec4(mix(p.rgb, t.rgb, t.a), p.a)%s;\n", secondary);
break;
case GE_TEXFUNC_BLEND:
WRITE(p, " vec4 v = vec4(mix(p.rgb, u_texenv.rgb, t.rgb), p.a * t.a)%s;\n", secondary);
break;
case GE_TEXFUNC_REPLACE:
WRITE(p, " vec4 v = t%s;\n", secondary);
break;
case GE_TEXFUNC_ADD:
case GE_TEXFUNC_UNKNOWN1:
case GE_TEXFUNC_UNKNOWN2:
case GE_TEXFUNC_UNKNOWN3:
WRITE(p, " vec4 v = vec4(p.rgb + t.rgb, p.a * t.a)%s;\n", secondary);
break;
default:
WRITE(p, " vec4 v = p;\n"); break;
}
} else { // texfmt == RGB
switch (gstate.getTextureFunction()) {
case GE_TEXFUNC_MODULATE:
WRITE(p, " vec4 v = vec4(t.rgb * p.rgb, p.a)%s;\n", secondary);
break;
case GE_TEXFUNC_DECAL:
WRITE(p, " vec4 v = vec4(t.rgb, p.a)%s;\n", secondary);
break;
case GE_TEXFUNC_BLEND:
WRITE(p, " vec4 v = vec4(mix(p.rgb, u_texenv.rgb, t.rgb), p.a)%s;\n", secondary);
break;
case GE_TEXFUNC_REPLACE:
WRITE(p, " vec4 v = vec4(t.rgb, p.a)%s;\n", secondary);
break;
case GE_TEXFUNC_ADD:
case GE_TEXFUNC_UNKNOWN1:
case GE_TEXFUNC_UNKNOWN2:
case GE_TEXFUNC_UNKNOWN3:
WRITE(p, " vec4 v = vec4(p.rgb + t.rgb, p.a)%s;\n", secondary); break;
default:
WRITE(p, " vec4 v = p;\n"); break;
}
}
} else {
// No texture mapping
WRITE(p, " vec4 v = v_color0 %s;\n", secondary);
}
if (enableAlphaTest) {
GEComparison alphaTestFunc = gstate.getAlphaTestFunction();
const char *alphaTestFuncs[] = { "#", "#", " != ", " == ", " >= ", " > ", " <= ", " < " }; // never/always don't make sense
if (alphaTestFuncs[alphaTestFunc][0] != '#') {
if (alphaTestAgainstZero) {
// When testing against 0 (extremely common), we can avoid some math.
// 0.002 is approximately half of 1.0 / 255.0.
if (alphaTestFunc == GE_COMP_NOTEQUAL || alphaTestFunc == GE_COMP_GREATER) {
WRITE(p, " if (v.a < 0.002) discard;\n");
} else {
// Anything else is a test for == 0. Happens sometimes, actually...
WRITE(p, " if (v.a > 0.002) discard;\n");
}
} else if (gl_extensions.gpuVendor == GPU_VENDOR_POWERVR) {
// Work around bad PVR driver problem where equality check + discard just doesn't work.
if (alphaTestFunc != GE_COMP_NOTEQUAL)
WRITE(p, " if (roundTo255thf(v.a) %s u_alphacolorref.a) discard;\n", alphaTestFuncs[alphaTestFunc]);
} else {
WRITE(p, " if (roundAndScaleTo255f(v.a) %s u_alphacolorref.a) discard;\n", alphaTestFuncs[alphaTestFunc]);
}
}
}
if (enableColorTest) {
GEComparison colorTestFunc = gstate.getColorTestFunction();
const char *colorTestFuncs[] = { "#", "#", " != ", " == " }; // never/always don't make sense
WARN_LOG_REPORT_ONCE(colortest, G3D, "Color test function : %s", colorTestFuncs[colorTestFunc]);
u32 colorTestMask = gstate.getColorTestMask();
if (colorTestFuncs[colorTestFunc][0] != '#') {
if (gl_extensions.gpuVendor == GPU_VENDOR_POWERVR)
WRITE(p, " if (roundTo255thv(v.rgb) %s u_alphacolorref.rgb) discard;\n", colorTestFuncs[colorTestFunc]);
else
WRITE(p, " if (roundAndScaleTo255v(v.rgb) %s u_alphacolorref.rgb) discard;\n", colorTestFuncs[colorTestFunc]);
}
}
// Color doubling happens after the color test.
if (enableColorDoubling && enableAlphaDoubling) {
WRITE(p, " v = v * 2.0;\n");
} else if (enableColorDoubling) {
WRITE(p, " v.rgb = v.rgb * 2.0;\n");
} else if (enableAlphaDoubling) {
WRITE(p, " v.a = v.a * 2.0;\n");
}
if (enableFog) {
WRITE(p, " float fogCoef = clamp(v_fogdepth, 0.0, 1.0);\n");
WRITE(p, " v = mix(vec4(u_fogcolor, v.a), v, fogCoef);\n");
// WRITE(p, " v.x = v_depth;\n");
}
if (ShouldUseShaderBlending()) {
// If we have NV_shader_framebuffer_fetch / EXT_shader_framebuffer_fetch, we skip the blit.
// We can just read the prev value more directly.
// TODO: EXT_shader_framebuffer_fetch on iOS 6, possibly others.
if (gl_extensions.NV_shader_framebuffer_fetch) {
WRITE(p, " lowp vec4 destColor = gl_LastFragData[0];\n");
} else {
WRITE(p, " lowp vec4 destColor = texelFetch(fbotex, ivec2(gl_FragCoord.x, gl_FragCoord.y), 0);\n");
}
GEBlendSrcFactor funcA = gstate.getBlendFuncA();
GEBlendDstFactor funcB = gstate.getBlendFuncB();
GEBlendMode eq = gstate.getBlendEq();
const char *srcFactor = "vec3(1.0)";
const char *dstFactor = "vec3(0.0)";
switch (funcA)
{
case GE_SRCBLEND_DSTCOLOR: srcFactor = "destColor.rgb"; break;
case GE_SRCBLEND_INVDSTCOLOR: srcFactor = "(vec3(1.0) - destColor.rgb)"; break;
case GE_SRCBLEND_SRCALPHA: srcFactor = "vec3(v.a)"; break;
case GE_SRCBLEND_INVSRCALPHA: srcFactor = "vec3(1.0 - v.a)"; break;
case GE_SRCBLEND_DSTALPHA: srcFactor = "vec3(destColor.a)"; break;
case GE_SRCBLEND_INVDSTALPHA: srcFactor = "vec3(1.0 - destColor.a)"; break;
case GE_SRCBLEND_DOUBLESRCALPHA: srcFactor = "vec3(v.a * 2.0)"; break;
// TODO: Double inverse, or inverse double? Following softgpu for now...
case GE_SRCBLEND_DOUBLEINVSRCALPHA: srcFactor = "vec3(1.0 - v.a * 2.0)"; break;
case GE_SRCBLEND_DOUBLEDSTALPHA: srcFactor = "vec3(destColor.a * 2.0)"; break;
case GE_SRCBLEND_DOUBLEINVDSTALPHA: srcFactor = "vec3(1.0 - destColor.a * 2.0)"; break;
case GE_SRCBLEND_FIXA: srcFactor = "u_blendFixA"; break;
}
switch (funcB)
{
case GE_DSTBLEND_SRCCOLOR: dstFactor = "v.rgb"; break;
case GE_DSTBLEND_INVSRCCOLOR: dstFactor = "(vec3(1.0) - v.rgb)"; break;
case GE_DSTBLEND_SRCALPHA: dstFactor = "vec3(v.a)"; break;
case GE_DSTBLEND_INVSRCALPHA: dstFactor = "vec3(1.0 - v.a)"; break;
case GE_DSTBLEND_DSTALPHA: dstFactor = "vec3(destColor.a)"; break;
case GE_DSTBLEND_INVDSTALPHA: dstFactor = "vec3(1.0 - destColor.a)"; break;
case GE_DSTBLEND_DOUBLESRCALPHA: dstFactor = "vec3(v.a * 2.0)"; break;
case GE_DSTBLEND_DOUBLEINVSRCALPHA: dstFactor = "vec3(1.0 - v.a * 2.0)"; break;
case GE_DSTBLEND_DOUBLEDSTALPHA: dstFactor = "vec3(destColor.a * 2.0)"; break;
case GE_DSTBLEND_DOUBLEINVDSTALPHA: dstFactor = "vec3(1.0 - destColor.a * 2.0)"; break;
case GE_DSTBLEND_FIXB: dstFactor = "u_blendFixB"; break;
}
switch (eq)
{
case GE_BLENDMODE_MUL_AND_ADD:
WRITE(p, " v.rgb = v.rgb * %s + destColor.rgb * %s;\n", srcFactor, dstFactor);
break;
case GE_BLENDMODE_MUL_AND_SUBTRACT:
WRITE(p, " v.rgb = v.rgb * %s - destColor.rgb * %s;\n", srcFactor, dstFactor);
break;
case GE_BLENDMODE_MUL_AND_SUBTRACT_REVERSE:
WRITE(p, " v.rgb = destColor.rgb * %s - v.rgb * %s;\n", srcFactor, dstFactor);
break;
case GE_BLENDMODE_MIN:
WRITE(p, " v.rgb = min(v.rgb, destColor.rgb);\n");
break;
case GE_BLENDMODE_MAX:
WRITE(p, " v.rgb = max(v.rgb, destColor.rgb);\n");
break;
case GE_BLENDMODE_ABSDIFF:
WRITE(p, " v.rgb = abs(v.rgb - destColor.rgb);\n");
break;
}
}
}
switch (stencilToAlpha) {
case REPLACE_ALPHA_DUALSOURCE:
WRITE(p, " fragColor0 = vec4(v.rgb, 0.0);\n");
WRITE(p, " fragColor1 = vec4(0.0, 0.0, 0.0, v.a);\n");
break;
case REPLACE_ALPHA_YES:
WRITE(p, " %s = vec4(v.rgb, 0.0);\n", fragColor0);
break;
case REPLACE_ALPHA_NO:
WRITE(p, " %s = v;\n", fragColor0);
break;
}
if (stencilToAlpha != REPLACE_ALPHA_NO) {
switch (ReplaceAlphaWithStencilType()) {
case STENCIL_VALUE_UNIFORM:
WRITE(p, " %s.a = u_stencilReplaceValue;\n", fragColor0);
break;
case STENCIL_VALUE_ZERO:
WRITE(p, " %s.a = 0.0;\n", fragColor0);
break;
case STENCIL_VALUE_ONE:
WRITE(p, " %s.a = 1.0;\n", fragColor0);
break;
case STENCIL_VALUE_UNKNOWN:
// Maybe we should even mask away alpha using glColorMask and not change it at all? We do get here
// if the stencil mode is KEEP for example.
WRITE(p, " %s.a = 0.0;\n", fragColor0);
break;
case STENCIL_VALUE_KEEP:
// Do nothing. We'll mask out the alpha using color mask.
break;
}
}
#ifdef DEBUG_SHADER
if (doTexture) {
WRITE(p, " %s = texture2D(tex, v_texcoord.xy);\n", fragColor0);
WRITE(p, " %s += vec4(0.3,0,0.3,0.3);\n", fragColor0);
} else {
WRITE(p, " %s = vec4(1,0,1,1);\n", fragColor0);
}
#endif
WRITE(p, "}\n");
}
// Here we must take all the bits of the gstate that determine what the fragment shader will
// look like, and concatenate them together into an ID.
void ComputeFragmentShaderID(FragmentShaderID *id) {
int id0 = 0;
int id1 = 0;
if (gstate.isModeClear()) {
// We only need one clear shader, so let's ignore the rest of the bits.
id0 = 1;
} else {
bool lmode = gstate.isUsingSecondaryColor() && gstate.isLightingEnabled();
bool enableFog = gstate.isFogEnabled() && !gstate.isModeThrough();
bool enableAlphaTest = gstate.isAlphaTestEnabled() && !IsAlphaTestTriviallyTrue();
bool alphaTestAgainstZero = gstate.getAlphaTestRef() == 0;
bool enableColorTest = gstate.isColorTestEnabled() && !IsColorTestTriviallyTrue();
bool alphaToColorDoubling = AlphaToColorDoubling();
bool enableColorDoubling = (gstate.isColorDoublingEnabled() && gstate.isTextureMapEnabled()) || alphaToColorDoubling;
// This isn't really correct, but it's a hack to get doubled blend modes to work more correctly.
bool enableAlphaDoubling = !alphaToColorDoubling && CanDoubleSrcBlendMode();
bool doTextureProjection = gstate.getUVGenMode() == GE_TEXMAP_TEXTURE_MATRIX;
bool doTextureAlpha = gstate.isTextureAlphaUsed();
ReplaceAlphaType stencilToAlpha = ReplaceAlphaWithStencil();
// All texfuncs except replace are the same for RGB as for RGBA with full alpha.
if (gstate_c.textureFullAlpha && gstate.getTextureFunction() != GE_TEXFUNC_REPLACE)
doTextureAlpha = false;
// id0 |= (gstate.isModeClear() & 1);
if (gstate.isTextureMapEnabled()) {
id0 |= 1 << 1;
id0 |= gstate.getTextureFunction() << 2;
id0 |= (doTextureAlpha & 1) << 5; // rgb or rgba
}
// 6 is free.
id0 |= (lmode & 1) << 7;
if (enableAlphaTest) {
id0 |= 1 << 8;
id0 |= gstate.getAlphaTestFunction() << 9;
}
if (enableColorTest) {
id0 |= 1 << 12;
id0 |= gstate.getColorTestFunction() << 13; // color test func
}
id0 |= (enableFog & 1) << 15;
id0 |= (doTextureProjection & 1) << 16;
id0 |= (enableColorDoubling & 1) << 17;
id0 |= (enableAlphaDoubling & 1) << 18;
id0 |= (stencilToAlpha) << 19;
if (stencilToAlpha != REPLACE_ALPHA_NO) {
// 3 bits
id0 |= ReplaceAlphaWithStencilType() << 21;
}
id0 |= (alphaTestAgainstZero & 1) << 24;
if (enableAlphaTest)
gpuStats.numAlphaTestedDraws++;
else
gpuStats.numNonAlphaTestedDraws++;
if (ShouldUseShaderBlending()) {
// 12 bits total.
id1 |= 1;
id1 |= (gstate.getBlendEq() << 1);
id1 |= (gstate.getBlendFuncA() << 4);
id1 |= (gstate.getBlendFuncB() << 8);
}
}
id->d[0] = id0;
id->d[1] = id1;
}
