LinkedShader::LinkedShader(ShaderID VSID, Shader *vs, ShaderID FSID, Shader *fs, bool useHWTransform)
: useHWTransform_(useHWTransform), program(0), dirtyUniforms(0) {
PROFILE_THIS_SCOPE("shaderlink");
program = glCreateProgram();
vs_ = vs;
glAttachShader(program, vs->shader);
glAttachShader(program, fs->shader);
// Bind attribute locations to fixed locations so that they're
// the same in all shaders. We use this later to minimize the calls to
// glEnableVertexAttribArray and glDisableVertexAttribArray.
glBindAttribLocation(program, ATTR_POSITION, "position");
glBindAttribLocation(program, ATTR_TEXCOORD, "texcoord");
glBindAttribLocation(program, ATTR_NORMAL, "normal");
glBindAttribLocation(program, ATTR_W1, "w1");
glBindAttribLocation(program, ATTR_W2, "w2");
glBindAttribLocation(program, ATTR_COLOR0, "color0");
glBindAttribLocation(program, ATTR_COLOR1, "color1");
#if !defined(USING_GLES2)
if (gstate_c.featureFlags & GPU_SUPPORTS_DUALSOURCE_BLEND) {
// Dual source alpha
glBindFragDataLocationIndexed(program, 0, 0, "fragColor0");
glBindFragDataLocationIndexed(program, 0, 1, "fragColor1");
} else if (gl_extensions.VersionGEThan(3, 3, 0)) {
glBindFragDataLocation(program, 0, "fragColor0");
}
#elif !defined(IOS)
if (gl_extensions.GLES3) {
if (gstate_c.featureFlags & GPU_SUPPORTS_DUALSOURCE_BLEND) {
glBindFragDataLocationIndexedEXT(program, 0, 0, "fragColor0");
glBindFragDataLocationIndexedEXT(program, 0, 1, "fragColor1");
}
}
#endif
glLinkProgram(program);
GLint linkStatus = GL_FALSE;
glGetProgramiv(program, GL_LINK_STATUS, &linkStatus);
if (linkStatus != GL_TRUE) {
GLint bufLength = 0;
glGetProgramiv(program, GL_INFO_LOG_LENGTH, &bufLength);
if (bufLength) {
char* buf = new char[bufLength];
glGetProgramInfoLog(program, bufLength, NULL, buf);
#ifdef ANDROID
ELOG("Could not link program:\n %s", buf);
#endif
ERROR_LOG(G3D, "Could not link program:\n %s", buf);
ERROR_LOG(G3D, "VS desc:\n%s\n", vs->GetShaderString(SHADER_STRING_SHORT_DESC, VSID).c_str());
ERROR_LOG(G3D, "FS desc:\n%s\n", fs->GetShaderString(SHADER_STRING_SHORT_DESC, FSID).c_str());
std::string vs_source = vs->GetShaderString(SHADER_STRING_SOURCE_CODE, VSID);
std::string fs_source = fs->GetShaderString(SHADER_STRING_SOURCE_CODE, FSID);
ERROR_LOG(G3D, "VS:\n%s\n", vs_source.c_str());
ERROR_LOG(G3D, "FS:\n%s\n", fs_source.c_str());
Reporting::ReportMessage("Error in shader program link: info: %s / fs: %s / vs: %s", buf, fs_source.c_str(), vs_source.c_str());
#ifdef SHADERLOG
OutputDebugStringUTF8(buf);
OutputDebugStringUTF8(vs_source.c_str());
OutputDebugStringUTF8(fs_source.c_str());
#endif
delete [] buf; // we're dead!
}
// Prevent a buffer overflow.
numBones = 0;
return;
}
INFO_LOG(G3D, "Linked shader: vs %i fs %i", (int)vs->shader, (int)fs->shader);
u_tex = glGetUniformLocation(program, "tex");
u_proj = glGetUniformLocation(program, "u_proj");
u_proj_through = glGetUniformLocation(program, "u_proj_through");
u_texenv = glGetUniformLocation(program, "u_texenv");
u_fogcolor = glGetUniformLocation(program, "u_fogcolor");
u_fogcoef = glGetUniformLocation(program, "u_fogcoef");
u_alphacolorref = glGetUniformLocation(program, "u_alphacolorref");
u_alphacolormask = glGetUniformLocation(program, "u_alphacolormask");
u_stencilReplaceValue = glGetUniformLocation(program, "u_stencilReplaceValue");
u_testtex = glGetUniformLocation(program, "testtex");
u_fbotex = glGetUniformLocation(program, "fbotex");
u_blendFixA = glGetUniformLocation(program, "u_blendFixA");
u_blendFixB = glGetUniformLocation(program, "u_blendFixB");
u_fbotexSize = glGetUniformLocation(program, "u_fbotexSize");
// Transform
u_view = glGetUniformLocation(program, "u_view");
u_world = glGetUniformLocation(program, "u_world");
u_texmtx = glGetUniformLocation(program, "u_texmtx");
if (VSID.Bit(VS_BIT_ENABLE_BONES))
numBones = TranslateNumBones(VSID.Bits(VS_BIT_BONES, 3) + 1);
else
numBones = 0;
u_depthRange = glGetUniformLocation(program, "u_depthRange");
#ifdef USE_BONE_ARRAY
u_bone = glGetUniformLocation(program, "u_bone");
#else
for (int i = 0; i < 8; i++) {
char name[10];
sprintf(name, "u_bone%i", i);
u_bone[i] = glGetUniformLocation(program, name);
}
#endif
// Lighting, texturing
u_ambient = glGetUniformLocation(program, "u_ambient");
u_matambientalpha = glGetUniformLocation(program, "u_matambientalpha");
u_matdiffuse = glGetUniformLocation(program, "u_matdiffuse");
u_matspecular = glGetUniformLocation(program, "u_matspecular");
u_matemissive = glGetUniformLocation(program, "u_matemissive");
u_uvscaleoffset = glGetUniformLocation(program, "u_uvscaleoffset");
u_texclamp = glGetUniformLocation(program, "u_texclamp");
u_texclampoff = glGetUniformLocation(program, "u_texclampoff");
for (int i = 0; i < 4; i++) {
char temp[64];
sprintf(temp, "u_lightpos%i", i);
u_lightpos[i] = glGetUniformLocation(program, temp);
sprintf(temp, "u_lightdir%i", i);
u_lightdir[i] = glGetUniformLocation(program, temp);
sprintf(temp, "u_lightatt%i", i);
u_lightatt[i] = glGetUniformLocation(program, temp);
sprintf(temp, "u_lightangle%i", i);
u_lightangle[i] = glGetUniformLocation(program, temp);
sprintf(temp, "u_lightspotCoef%i", i);
u_lightspotCoef[i] = glGetUniformLocation(program, temp);
sprintf(temp, "u_lightambient%i", i);
u_lightambient[i] = glGetUniformLocation(program, temp);
sprintf(temp, "u_lightdiffuse%i", i);
u_lightdiffuse[i] = glGetUniformLocation(program, temp);
sprintf(temp, "u_lightspecular%i", i);
u_lightspecular[i] = glGetUniformLocation(program, temp);
}
attrMask = 0;
if (-1 != glGetAttribLocation(program, "position")) attrMask |= 1 << ATTR_POSITION;
if (-1 != glGetAttribLocation(program, "texcoord")) attrMask |= 1 << ATTR_TEXCOORD;
if (-1 != glGetAttribLocation(program, "normal")) attrMask |= 1 << ATTR_NORMAL;
if (-1 != glGetAttribLocation(program, "w1")) attrMask |= 1 << ATTR_W1;
if (-1 != glGetAttribLocation(program, "w2")) attrMask |= 1 << ATTR_W2;
if (-1 != glGetAttribLocation(program, "color0")) attrMask |= 1 << ATTR_COLOR0;
if (-1 != glGetAttribLocation(program, "color1")) attrMask |= 1 << ATTR_COLOR1;
availableUniforms = 0;
if (u_proj != -1) availableUniforms |= DIRTY_PROJMATRIX;
if (u_proj_through != -1) availableUniforms |= DIRTY_PROJTHROUGHMATRIX;
if (u_texenv != -1) availableUniforms |= DIRTY_TEXENV;
if (u_alphacolorref != -1) availableUniforms |= DIRTY_ALPHACOLORREF;
if (u_alphacolormask != -1) availableUniforms |= DIRTY_ALPHACOLORMASK;
if (u_fogcolor != -1) availableUniforms |= DIRTY_FOGCOLOR;
if (u_fogcoef != -1) availableUniforms |= DIRTY_FOGCOEF;
if (u_texenv != -1) availableUniforms |= DIRTY_TEXENV;
if (u_uvscaleoffset != -1) availableUniforms |= DIRTY_UVSCALEOFFSET;
if (u_texclamp != -1) availableUniforms |= DIRTY_TEXCLAMP;
if (u_world != -1) availableUniforms |= DIRTY_WORLDMATRIX;
if (u_view != -1) availableUniforms |= DIRTY_VIEWMATRIX;
if (u_texmtx != -1) availableUniforms |= DIRTY_TEXMATRIX;
if (u_stencilReplaceValue != -1) availableUniforms |= DIRTY_STENCILREPLACEVALUE;
if (u_blendFixA != -1 || u_blendFixB != -1 || u_fbotexSize != -1) availableUniforms |= DIRTY_SHADERBLEND;
if (u_depthRange != -1)
availableUniforms |= DIRTY_DEPTHRANGE;
// Looping up to numBones lets us avoid checking u_bone[i]
#ifdef USE_BONE_ARRAY
if (u_bone != -1) {
for (int i = 0; i < numBones; i++) {
availableUniforms |= DIRTY_BONEMATRIX0 << i;
}
}
#else
for (int i = 0; i < numBones; i++) {
if (u_bone[i] != -1)
availableUniforms |= DIRTY_BONEMATRIX0 << i;
}
#endif
if (u_ambient != -1) availableUniforms |= DIRTY_AMBIENT;
if (u_matambientalpha != -1) availableUniforms |= DIRTY_MATAMBIENTALPHA;
if (u_matdiffuse != -1) availableUniforms |= DIRTY_MATDIFFUSE;
if (u_matemissive != -1) availableUniforms |= DIRTY_MATEMISSIVE;
if (u_matspecular != -1) availableUniforms |= DIRTY_MATSPECULAR;
for (int i = 0; i < 4; i++) {
if (u_lightdir[i] != -1 ||
u_lightspecular[i] != -1 ||
u_lightpos[i] != -1)
availableUniforms |= DIRTY_LIGHT0 << i;
}
glUseProgram(program);
// Default uniform values
glUniform1i(u_tex, 0);
glUniform1i(u_fbotex, 1);
glUniform1i(u_testtex, 2);
// The rest, use the "dirty" mechanism.
dirtyUniforms = DIRTY_ALL;
}
bool GenerateVulkanGLSLVertexShader(const ShaderID &id, char *buffer, bool *usesLighting) {
char *p = buffer;
WRITE(p, "%s", vulkan_glsl_preamble);
bool highpFog = false;
bool highpTexcoord = false;
bool isModeThrough = id.Bit(VS_BIT_IS_THROUGH);
bool lmode = id.Bit(VS_BIT_LMODE) && !isModeThrough; // TODO: Different expression than in shaderIDgen
bool doTexture = id.Bit(VS_BIT_DO_TEXTURE);
bool doTextureTransform = id.Bit(VS_BIT_DO_TEXTURE_TRANSFORM);
GETexMapMode uvGenMode = static_cast<GETexMapMode>(id.Bits(VS_BIT_UVGEN_MODE, 2));
// this is only valid for some settings of uvGenMode
GETexProjMapMode uvProjMode = static_cast<GETexProjMapMode>(id.Bits(VS_BIT_UVPROJ_MODE, 2));
bool doShadeMapping = uvGenMode == GE_TEXMAP_ENVIRONMENT_MAP;
bool doFlatShading = id.Bit(VS_BIT_FLATSHADE);
bool useHWTransform = id.Bit(VS_BIT_USE_HW_TRANSFORM);
bool hasColor = id.Bit(VS_BIT_HAS_COLOR) || !useHWTransform;
bool hasNormal = id.Bit(VS_BIT_HAS_NORMAL) && useHWTransform;
bool hasTexcoord = id.Bit(VS_BIT_HAS_TEXCOORD) || !useHWTransform;
bool enableFog = id.Bit(VS_BIT_ENABLE_FOG);
bool throughmode = id.Bit(VS_BIT_IS_THROUGH);
bool flipNormal = id.Bit(VS_BIT_NORM_REVERSE);
int ls0 = id.Bits(VS_BIT_LS0, 2);
int ls1 = id.Bits(VS_BIT_LS1, 2);
bool enableBones = id.Bit(VS_BIT_ENABLE_BONES);
bool enableLighting = id.Bit(VS_BIT_LIGHTING_ENABLE);
int matUpdate = id.Bits(VS_BIT_MATERIAL_UPDATE, 3);
bool doBezier = id.Bit(VS_BIT_BEZIER);
bool doSpline = id.Bit(VS_BIT_SPLINE);
bool hasColorTess = id.Bit(VS_BIT_HAS_COLOR_TESS);
bool hasTexcoordTess = id.Bit(VS_BIT_HAS_TEXCOORD_TESS);
bool flipNormalTess = id.Bit(VS_BIT_NORM_REVERSE_TESS);
// The uniforms are passed in as three "clumps" that may or may not be present.
// We will memcpy the parts into place in a big buffer so we can be quite dynamic about what parts
// are present and what parts aren't, but we will not be ultra detailed about it.
*usesLighting = enableLighting || doShadeMapping;
WRITE(p, "\n");
WRITE(p, "layout (std140, set = 0, binding = 2) uniform baseVars {\n%s} base;\n", ub_baseStr);
if (enableLighting || doShadeMapping)
WRITE(p, "layout (std140, set = 0, binding = 3) uniform lightVars {\n%s} light;\n", ub_vs_lightsStr);
if (enableBones)
WRITE(p, "layout (std140, set = 0, binding = 4) uniform boneVars {\n%s} bone;\n", ub_vs_bonesStr);
const char *shading = doFlatShading ? "flat " : "";
DoLightComputation doLight[4] = { LIGHT_OFF, LIGHT_OFF, LIGHT_OFF, LIGHT_OFF };
if (useHWTransform) {
int shadeLight0 = doShadeMapping ? ls0 : -1;
int shadeLight1 = doShadeMapping ? ls1 : -1;
for (int i = 0; i < 4; i++) {
if (i == shadeLight0 || i == shadeLight1)
doLight[i] = LIGHT_SHADE;
if (id.Bit(VS_BIT_LIGHTING_ENABLE) && id.Bit(VS_BIT_LIGHT0_ENABLE + i))
doLight[i] = LIGHT_FULL;
}
}
int numBoneWeights = 0;
int boneWeightScale = id.Bits(VS_BIT_WEIGHT_FMTSCALE, 2);
if (enableBones) {
numBoneWeights = 1 + id.Bits(VS_BIT_BONES, 3);
WRITE(p, "%s", boneWeightDecl[numBoneWeights]);
}
if (useHWTransform)
WRITE(p, "layout (location = %d) in vec3 position;\n", PspAttributeLocation::POSITION);
else
// we pass the fog coord in w
WRITE(p, "layout (location = %d) in vec4 position;\n", PspAttributeLocation::POSITION);
if (useHWTransform && hasNormal)
WRITE(p, "layout (location = %d) in vec3 normal;\n", PspAttributeLocation::NORMAL);
bool texcoordInVec3 = false;
if (doTexture && hasTexcoord) {
if (!useHWTransform && doTextureTransform && !throughmode) {
WRITE(p, "layout (location = %d) in vec3 texcoord;\n", PspAttributeLocation::TEXCOORD);
texcoordInVec3 = true;
}
else
WRITE(p, "layout (location = %d) in vec2 texcoord;\n", PspAttributeLocation::TEXCOORD);
}
if (hasColor) {
WRITE(p, "layout (location = %d) in vec4 color0;\n", PspAttributeLocation::COLOR0);
if (lmode && !useHWTransform) // only software transform supplies color1 as vertex data
WRITE(p, "layout (location = %d) in vec3 color1;\n", PspAttributeLocation::COLOR1);
}
WRITE(p, "layout (location = 1) %sout vec4 v_color0;\n", shading);
if (lmode) {
WRITE(p, "layout (location = 2) %sout vec3 v_color1;\n", shading);
}
if (doTexture) {
WRITE(p, "layout (location = 0) out vec3 v_texcoord;\n");
}
if (enableFog) {
// See the fragment shader generator
WRITE(p, "layout (location = 3) out float v_fogdepth;\n");
}
// See comment above this function (GenerateVertexShader).
if (!isModeThrough && gstate_c.Supports(GPU_ROUND_DEPTH_TO_16BIT)) {
// Apply the projection and viewport to get the Z buffer value, floor to integer, undo the viewport and projection.
WRITE(p, "\nvec4 depthRoundZVP(vec4 v) {\n");
WRITE(p, " float z = v.z / v.w;\n");
WRITE(p, " z = z * base.depthRange.x + base.depthRange.y;\n");
WRITE(p, " z = floor(z);\n");
WRITE(p, " z = (z - base.depthRange.z) * base.depthRange.w;\n");
WRITE(p, " return vec4(v.x, v.y, z * v.w, v.w);\n");
WRITE(p, "}\n\n");
}
WRITE(p, "out gl_PerVertex { vec4 gl_Position; };\n");
if (doBezier || doSpline) {
WRITE(p, "layout (binding = 5) uniform sampler2D u_tess_pos_tex;\n");
WRITE(p, "layout (binding = 6) uniform sampler2D u_tess_tex_tex;\n");
WRITE(p, "layout (binding = 7) uniform sampler2D u_tess_col_tex;\n");
for (int i = 2; i <= 4; i++) {
// Define 3 types vec2, vec3, vec4
WRITE(p, "vec%d tess_sample(in vec%d points[16], in vec2 weights[4]) {\n", i, i);
WRITE(p, " vec%d pos = vec%d(0);\n", i, i);
WRITE(p, " for (int i = 0; i < 4; ++i) {\n");
WRITE(p, " for (int j = 0; j < 4; ++j) {\n");
WRITE(p, " float f = weights[j].x * weights[i].y;\n");
WRITE(p, " if (f != 0)\n");
WRITE(p, " pos = pos + f * points[i * 4 + j];\n");
WRITE(p, " }\n");
WRITE(p, " }\n");
WRITE(p, " return pos;\n");
WRITE(p, "}\n");
}
if (doSpline) {
WRITE(p, "void spline_knot(ivec2 num_patches, ivec2 type, out vec2 knot[6], ivec2 patch_pos) {\n");
WRITE(p, " for (int i = 0; i < 6; ++i) {\n");
WRITE(p, " knot[i] = vec2(i + patch_pos.x - 2, i + patch_pos.y - 2);\n");
WRITE(p, " }\n");
WRITE(p, " if ((type.x & 1) != 0) {\n");
WRITE(p, " if (patch_pos.x <= 2)\n");
WRITE(p, " knot[0].x = 0;\n");
WRITE(p, " if (patch_pos.x <= 1)\n");
WRITE(p, " knot[1].x = 0;\n");
WRITE(p, " }\n");
WRITE(p, " if ((type.x & 2) != 0) {\n");
WRITE(p, " if (patch_pos.x >= (num_patches.x - 2))\n");
WRITE(p, " knot[5].x = num_patches.x;\n");
WRITE(p, " if (patch_pos.x == (num_patches.x - 1))\n");
WRITE(p, " knot[4].x = num_patches.x;\n");
WRITE(p, " }\n");
WRITE(p, " if ((type.y & 1) != 0) {\n");
WRITE(p, " if (patch_pos.y <= 2)\n");
WRITE(p, " knot[0].y = 0;\n");
WRITE(p, " if (patch_pos.y <= 1)\n");
WRITE(p, " knot[1].y = 0;\n");
WRITE(p, " }\n");
WRITE(p, " if ((type.y & 2) != 0) {\n");
WRITE(p, " if (patch_pos.y >= (num_patches.y - 2))\n");
WRITE(p, " knot[5].y = num_patches.y;\n");
WRITE(p, " if (patch_pos.y == (num_patches.y - 1))\n");
WRITE(p, " knot[4].y = num_patches.y;\n");
WRITE(p, " }\n");
WRITE(p, "}\n");
WRITE(p, "void spline_weight(vec2 t, in vec2 knot[6], out vec2 weights[4]) {\n");
// TODO: Maybe compilers could be coaxed into vectorizing this code without the above explicitly...
WRITE(p, " vec2 t0 = (t - knot[0]);\n");
WRITE(p, " vec2 t1 = (t - knot[1]);\n");
WRITE(p, " vec2 t2 = (t - knot[2]);\n");
// TODO: All our knots are integers so we should be able to get rid of these divisions (How?)
WRITE(p, " vec2 f30 = t0 / (knot[3] - knot[0]);\n");
WRITE(p, " vec2 f41 = t1 / (knot[4] - knot[1]);\n");
WRITE(p, " vec2 f52 = t2 / (knot[5] - knot[2]);\n");
WRITE(p, " vec2 f31 = t1 / (knot[3] - knot[1]);\n");
WRITE(p, " vec2 f42 = t2 / (knot[4] - knot[2]);\n");
WRITE(p, " vec2 f32 = t2 / (knot[3] - knot[2]);\n");
WRITE(p, " vec2 a = (1 - f30)*(1 - f31);\n");
WRITE(p, " vec2 b = (f31*f41);\n");
WRITE(p, " vec2 c = (1 - f41)*(1 - f42);\n");
WRITE(p, " vec2 d = (f42*f52);\n");
WRITE(p, " weights[0] = a - (a*f32);\n");
WRITE(p, " weights[1] = 1 - a - b + ((a + b + c - 1)*f32);\n");
WRITE(p, " weights[2] = b + ((1 - b - c - d)*f32);\n");
WRITE(p, " weights[3] = d*f32;\n");
WRITE(p, "}\n");
}
}
WRITE(p, "void main() {\n");
if (!useHWTransform) {
// Simple pass-through of vertex data to fragment shader
if (doTexture) {
if (texcoordInVec3) {
WRITE(p, " v_texcoord = texcoord;\n");
} else {
WRITE(p, " v_texcoord = vec3(texcoord, 1.0);\n");
}
}
if (hasColor) {
WRITE(p, " v_color0 = color0;\n");
if (lmode)
WRITE(p, " v_color1 = color1;\n");
} else {
WRITE(p, " v_color0 = base.matambientalpha;\n");
if (lmode)
WRITE(p, " v_color1 = vec3(0.0);\n");
}
if (enableFog) {
WRITE(p, " v_fogdepth = position.w;\n");
}
if (isModeThrough) {
WRITE(p, " gl_Position = base.proj_through_mtx * vec4(position.xyz, 1.0);\n");
} else {
// The viewport is used in this case, so need to compensate for that.
if (gstate_c.Supports(GPU_ROUND_DEPTH_TO_16BIT)) {
WRITE(p, " gl_Position = depthRoundZVP(base.proj_mtx * vec4(position.xyz, 1.0));\n");
} else {
WRITE(p, " gl_Position = base.proj_mtx * vec4(position.xyz, 1.0);\n");
}
}
} else {
// Step 1: World Transform / Skinning
if (!enableBones) {
if (doBezier || doSpline) {
WRITE(p, " vec3 _pos[16];\n");
WRITE(p, " vec2 _tex[16];\n");
WRITE(p, " vec4 _col[16];\n");
WRITE(p, " int num_patches_u = %s;\n", doBezier ? "(base.spline_count_u - 1) / 3" : "base.spline_count_u - 3");
WRITE(p, " int u = int(mod(gl_InstanceIndex, num_patches_u));\n");
WRITE(p, " int v = gl_InstanceIndex / num_patches_u;\n");
WRITE(p, " ivec2 patch_pos = ivec2(u, v);\n");
WRITE(p, " for (int i = 0; i < 4; i++) {\n");
WRITE(p, " for (int j = 0; j < 4; j++) {\n");
WRITE(p, " int idx = (i + v%s) * base.spline_count_u + (j + u%s);\n", doBezier ? " * 3" : "", doBezier ? " * 3" : "");
WRITE(p, " ivec2 index = ivec2(idx, 0);\n");
WRITE(p, " _pos[i * 4 + j] = texelFetch(u_tess_pos_tex, index, 0).xyz;\n");
if (doTexture && hasTexcoord && hasTexcoordTess)
WRITE(p, " _tex[i * 4 + j] = texelFetch(u_tess_tex_tex, index, 0).xy;\n");
if (hasColor && hasColorTess)
WRITE(p, " _col[i * 4 + j] = texelFetch(u_tess_col_tex, index, 0).rgba;\n");
WRITE(p, " }\n");
WRITE(p, " }\n");
WRITE(p, " vec2 tess_pos = position.xy;\n");
WRITE(p, " vec2 weights[4];\n");
if (doBezier) {
// Bernstein 3D
WRITE(p, " weights[0] = (1 - tess_pos) * (1 - tess_pos) * (1 - tess_pos);\n");
WRITE(p, " weights[1] = 3 * tess_pos * (1 - tess_pos) * (1 - tess_pos);\n");
WRITE(p, " weights[2] = 3 * tess_pos * tess_pos * (1 - tess_pos);\n");
WRITE(p, " weights[3] = tess_pos * tess_pos * tess_pos;\n");
} else { // Spline
WRITE(p, " ivec2 spline_num_patches = ivec2(base.spline_count_u - 3, base.spline_count_v - 3);\n");
WRITE(p, " ivec2 spline_type = ivec2(base.spline_type_u, base.spline_type_v);\n");
WRITE(p, " vec2 knots[6];\n");
WRITE(p, " spline_knot(spline_num_patches, spline_type, knots, patch_pos);\n");
WRITE(p, " spline_weight(tess_pos + patch_pos, knots, weights);\n");
}
WRITE(p, " vec3 pos = tess_sample(_pos, weights);\n");
if (doTexture && hasTexcoord) {
if (hasTexcoordTess)
WRITE(p, " vec2 tex = tess_sample(_tex, weights);\n");
else
WRITE(p, " vec2 tex = tess_pos + patch_pos;\n");
}
if (hasColor) {
if (hasColorTess)
WRITE(p, " vec4 col = tess_sample(_col, weights);\n");
else
WRITE(p, " vec4 col = texelFetch(u_tess_col_tex, ivec2(0, 0), 0).rgba;\n");
}
if (hasNormal) {
// Curved surface is probably always need to compute normal(not sampling from control points)
if (doBezier) {
// Bernstein derivative
WRITE(p, " vec2 bernderiv[4];\n");
WRITE(p, " bernderiv[0] = -3 * (tess_pos - 1) * (tess_pos - 1); \n");
WRITE(p, " bernderiv[1] = 9 * tess_pos * tess_pos - 12 * tess_pos + 3; \n");
WRITE(p, " bernderiv[2] = 3 * (2 - 3 * tess_pos) * tess_pos; \n");
WRITE(p, " bernderiv[3] = 3 * tess_pos * tess_pos; \n");
WRITE(p, " vec2 bernderiv_u[4];\n");
WRITE(p, " vec2 bernderiv_v[4];\n");
WRITE(p, " for (int i = 0; i < 4; i++) {\n");
WRITE(p, " bernderiv_u[i] = vec2(bernderiv[i].x, weights[i].y);\n");
WRITE(p, " bernderiv_v[i] = vec2(weights[i].x, bernderiv[i].y);\n");
WRITE(p, " }\n");
WRITE(p, " vec3 du = tess_sample(_pos, bernderiv_u);\n");
WRITE(p, " vec3 dv = tess_sample(_pos, bernderiv_v);\n");
} else { // Spline
WRITE(p, " vec2 tess_next_u = vec2(normal.x, 0);\n");
WRITE(p, " vec2 tess_next_v = vec2(0, normal.y);\n");
// Right
WRITE(p, " vec2 tess_pos_r = tess_pos + tess_next_u;\n");
WRITE(p, " spline_weight(tess_pos_r + patch_pos, knots, weights);\n");
WRITE(p, " vec3 pos_r = tess_sample(_pos, weights);\n");
// Left
WRITE(p, " vec2 tess_pos_l = tess_pos - tess_next_u;\n");
WRITE(p, " spline_weight(tess_pos_l + patch_pos, knots, weights);\n");
WRITE(p, " vec3 pos_l = tess_sample(_pos, weights);\n");
// Down
WRITE(p, " vec2 tess_pos_d = tess_pos + tess_next_v;\n");
WRITE(p, " spline_weight(tess_pos_d + patch_pos, knots, weights);\n");
WRITE(p, " vec3 pos_d = tess_sample(_pos, weights);\n");
// Up
WRITE(p, " vec2 tess_pos_u = tess_pos - tess_next_v;\n");
WRITE(p, " spline_weight(tess_pos_u + patch_pos, knots, weights);\n");
WRITE(p, " vec3 pos_u = tess_sample(_pos, weights);\n");
WRITE(p, " vec3 du = pos_r - pos_l;\n");
WRITE(p, " vec3 dv = pos_d - pos_u;\n");
}
WRITE(p, " vec3 nrm = cross(du, dv);\n");
WRITE(p, " nrm = normalize(nrm);\n");
}
WRITE(p, " vec3 worldpos = vec4(pos.xyz, 1.0) * base.world_mtx;\n");
if (hasNormal) {
WRITE(p, " mediump vec3 worldnormal = normalize(vec4(%snrm, 0.0) * base.world_mtx);\n", flipNormalTess ? "-" : "");
} else {
WRITE(p, " mediump vec3 worldnormal = vec3(0.0, 0.0, 1.0);\n");
}
} else {
// No skinning, just standard T&L.
WRITE(p, " vec3 worldpos = vec4(position.xyz, 1.0) * base.world_mtx;\n");
if (hasNormal)
WRITE(p, " mediump vec3 worldnormal = normalize(vec4(%snormal, 0.0) * base.world_mtx);\n", flipNormal ? "-" : "");
else
WRITE(p, " mediump vec3 worldnormal = vec3(0.0, 0.0, 1.0);\n");
}
} else {
static const char *rescale[4] = { "", " * 1.9921875", " * 1.999969482421875", "" }; // 2*127.5f/128.f, 2*32767.5f/32768.f, 1.0f};
const char *factor = rescale[boneWeightScale];
static const char * const boneWeightAttr[8] = {
"w1.x", "w1.y", "w1.z", "w1.w",
"w2.x", "w2.y", "w2.z", "w2.w",
};
WRITE(p, " mat3x4 skinMatrix = w1.x * bone.m[0];\n");
if (numBoneWeights > 1) {
for (int i = 1; i < numBoneWeights; i++) {
WRITE(p, " skinMatrix += %s * bone.m[%i];\n", boneWeightAttr[i], i);
}
}
WRITE(p, ";\n");
// Trying to simplify this results in bugs in LBP...
WRITE(p, " vec3 skinnedpos = (vec4(position, 1.0) * skinMatrix) %s;\n", factor);
WRITE(p, " vec3 worldpos = vec4(skinnedpos, 1.0) * base.world_mtx;\n");
if (hasNormal) {
WRITE(p, " mediump vec3 skinnednormal = vec4(%snormal, 0.0) * skinMatrix %s;\n", flipNormal ? "-" : "", factor);
} else {
WRITE(p, " mediump vec3 skinnednormal = vec4(0.0, 0.0, %s1.0, 0.0) * skinMatrix %s;\n", flipNormal ? "-" : "", factor);
}
WRITE(p, " mediump vec3 worldnormal = normalize(vec4(skinnednormal, 0.0) * base.world_mtx);\n");
}
WRITE(p, " vec4 viewPos = vec4(vec4(worldpos, 1.0) * base.view_mtx, 1.0);\n");
// Final view and projection transforms.
if (gstate_c.Supports(GPU_ROUND_DEPTH_TO_16BIT)) {
WRITE(p, " gl_Position = depthRoundZVP(base.proj_mtx * viewPos);\n");
} else {
WRITE(p, " gl_Position = base.proj_mtx * viewPos;\n");
}
// TODO: Declare variables for dots for shade mapping if needed.
const char *ambientStr = ((matUpdate & 1) && hasColor) ? "color0" : "base.matambientalpha";
const char *diffuseStr = ((matUpdate & 2) && hasColor) ? "color0.rgb" : "light.matdiffuse";
const char *specularStr = ((matUpdate & 4) && hasColor) ? "color0.rgb" : "light.matspecular.rgb";
if (doBezier || doSpline) {
ambientStr = (matUpdate & 1) && hasColor ? "col" : "base.matambientalpha";
diffuseStr = (matUpdate & 2) && hasColor ? "col.rgb" : "light.matdiffuse";
specularStr = (matUpdate & 4) && hasColor ? "col.rgb" : "light.matspecular.rgb";
}
bool diffuseIsZero = true;
bool specularIsZero = true;
bool distanceNeeded = false;
if (enableLighting) {
WRITE(p, " vec4 lightSum0 = light.u_ambient * %s + vec4(light.matemissive, 0.0);\n", ambientStr);
for (int i = 0; i < 4; i++) {
GELightType type = static_cast<GELightType>(id.Bits(VS_BIT_LIGHT0_TYPE + 4 * i, 2));
GELightComputation comp = static_cast<GELightComputation>(id.Bits(VS_BIT_LIGHT0_COMP + 4 * i, 2));
if (doLight[i] != LIGHT_FULL)
continue;
diffuseIsZero = false;
if (comp != GE_LIGHTCOMP_ONLYDIFFUSE)
specularIsZero = false;
if (type != GE_LIGHTTYPE_DIRECTIONAL)
distanceNeeded = true;
}
if (!specularIsZero) {
WRITE(p, " vec3 lightSum1 = vec3(0.0);\n");
}
if (!diffuseIsZero) {
WRITE(p, " vec3 toLight;\n");
WRITE(p, " vec3 diffuse;\n");
}
if (distanceNeeded) {
WRITE(p, " float distance;\n");
WRITE(p, " float lightScale;\n");
}
}
// Calculate lights if needed. If shade mapping is enabled, lights may need to be
// at least partially calculated.
for (int i = 0; i < 4; i++) {
if (doLight[i] != LIGHT_FULL)
continue;
GELightType type = static_cast<GELightType>(id.Bits(VS_BIT_LIGHT0_TYPE + 4 * i, 2));
GELightComputation comp = static_cast<GELightComputation>(id.Bits(VS_BIT_LIGHT0_COMP + 4 * i, 2));
if (type == GE_LIGHTTYPE_DIRECTIONAL) {
// We prenormalize light positions for directional lights.
WRITE(p, " toLight = light.pos[%i];\n", i);
} else {
WRITE(p, " toLight = light.pos[%i] - worldpos;\n", i);
WRITE(p, " distance = length(toLight);\n");
WRITE(p, " toLight /= distance;\n");
}
bool doSpecular = comp != GE_LIGHTCOMP_ONLYDIFFUSE;
bool poweredDiffuse = comp == GE_LIGHTCOMP_BOTHWITHPOWDIFFUSE;
WRITE(p, " mediump float dot%i = max(dot(toLight, worldnormal), 0.0);\n", i);
if (poweredDiffuse) {
// pow(0.0, 0.0) may be undefined, but the PSP seems to treat it as 1.0.
// Seen in Tales of the World: Radiant Mythology (#2424.)
WRITE(p, " if (dot%i == 0.0 && light.matspecular.a == 0.0) {\n", i);
WRITE(p, " dot%i = 1.0;\n", i);
WRITE(p, " } else {\n");
WRITE(p, " dot%i = pow(dot%i, light.matspecular.a);\n", i, i);
WRITE(p, " }\n");
}
const char *timesLightScale = " * lightScale";
// Attenuation
switch (type) {
case GE_LIGHTTYPE_DIRECTIONAL:
timesLightScale = "";
break;
case GE_LIGHTTYPE_POINT:
WRITE(p, " lightScale = clamp(1.0 / dot(light.att[%i], vec3(1.0, distance, distance*distance)), 0.0, 1.0);\n", i);
break;
case GE_LIGHTTYPE_SPOT:
case GE_LIGHTTYPE_UNKNOWN:
WRITE(p, " float angle%i = dot(normalize(light.dir[%i]), toLight);\n", i, i);
WRITE(p, " if (angle%i >= light.angle[%i]) {\n", i, i);
WRITE(p, " lightScale = clamp(1.0 / dot(light.att[%i], vec3(1.0, distance, distance*distance)), 0.0, 1.0) * pow(angle%i, light.spotCoef[%i]);\n", i, i, i);
WRITE(p, " } else {\n");
WRITE(p, " lightScale = 0.0;\n");
WRITE(p, " }\n");
break;
default:
// ILLEGAL
break;
}
WRITE(p, " diffuse = (light.diffuse[%i] * %s) * dot%i;\n", i, diffuseStr, i);
if (doSpecular) {
WRITE(p, " dot%i = dot(normalize(toLight + vec3(0.0, 0.0, 1.0)), worldnormal);\n", i);
WRITE(p, " if (dot%i > 0.0)\n", i);
WRITE(p, " lightSum1 += light.specular[%i] * %s * (pow(dot%i, light.matspecular.a) %s);\n", i, specularStr, i, timesLightScale);
}
WRITE(p, " lightSum0.rgb += (light.ambient[%i] * %s.rgb + diffuse)%s;\n", i, ambientStr, timesLightScale);
}
if (enableLighting) {
// Sum up ambient, emissive here.
if (lmode) {
WRITE(p, " v_color0 = clamp(lightSum0, 0.0, 1.0);\n");
// v_color1 only exists when lmode = 1.
if (specularIsZero) {
WRITE(p, " v_color1 = vec3(0.0);\n");
} else {
WRITE(p, " v_color1 = clamp(lightSum1, 0.0, 1.0);\n");
}
} else {
if (specularIsZero) {
WRITE(p, " v_color0 = clamp(lightSum0, 0.0, 1.0);\n");
} else {
WRITE(p, " v_color0 = clamp(clamp(lightSum0, 0.0, 1.0) + vec4(lightSum1, 0.0), 0.0, 1.0);\n");
}
}
} else {
// Lighting doesn't affect color.
if (hasColor) {
if (doBezier || doSpline)
WRITE(p, " v_color0 = col;\n");
else
WRITE(p, " v_color0 = color0;\n");
} else {
WRITE(p, " v_color0 = base.matambientalpha;\n");
}
if (lmode) {
WRITE(p, " v_color1 = vec3(0.0);\n");
}
}
bool scaleUV = !throughmode && (uvGenMode == GE_TEXMAP_TEXTURE_COORDS || uvGenMode == GE_TEXMAP_UNKNOWN);
// Step 3: UV generation
if (doTexture) {
switch (uvGenMode) {
case GE_TEXMAP_TEXTURE_COORDS: // Scale-offset. Easy.
case GE_TEXMAP_UNKNOWN: // Not sure what this is, but Riviera uses it. Treating as coords works.
if (scaleUV) {
if (hasTexcoord) {
if (doBezier || doSpline)
WRITE(p, " v_texcoord = vec3(tex.xy * base.uvscaleoffset.xy + base.uvscaleoffset.zw, 0.0);\n");
else
WRITE(p, " v_texcoord = vec3(texcoord.xy, 0.0);\n");
} else {
WRITE(p, " v_texcoord = vec3(0.0);\n");
}
} else {
if (hasTexcoord) {
if (doBezier || doSpline)
WRITE(p, " v_texcoord = vec3(tex.xy * base.uvscaleoffset.xy + base.uvscaleoffset.zw, 0.0);\n");
else
WRITE(p, " v_texcoord = vec3(texcoord.xy * base.uvscaleoffset.xy + base.uvscaleoffset.zw, 0.0);\n");
} else {
WRITE(p, " v_texcoord = vec3(base.uvscaleoffset.zw, 0.0);\n");
}
}
break;
case GE_TEXMAP_TEXTURE_MATRIX: // Projection mapping.
{
std::string temp_tc;
switch (uvProjMode) {
case GE_PROJMAP_POSITION: // Use model space XYZ as source
temp_tc = "vec4(position.xyz, 1.0)";
break;
case GE_PROJMAP_UV: // Use unscaled UV as source
{
// scaleUV is false here.
if (hasTexcoord) {
temp_tc = "vec4(texcoord.xy, 0.0, 1.0)";
} else {
temp_tc = "vec4(0.0, 0.0, 0.0, 1.0)";
}
}
break;
case GE_PROJMAP_NORMALIZED_NORMAL: // Use normalized transformed normal as source
if (hasNormal)
temp_tc = flipNormal ? "vec4(normalize(-normal), 1.0)" : "vec4(normalize(normal), 1.0)";
else
temp_tc = "vec4(0.0, 0.0, 1.0, 1.0)";
break;
case GE_PROJMAP_NORMAL: // Use non-normalized transformed normal as source
if (hasNormal)
temp_tc = flipNormal ? "vec4(-normal, 1.0)" : "vec4(normal, 1.0)";
else
temp_tc = "vec4(0.0, 0.0, 1.0, 1.0)";
break;
}
// Transform by texture matrix. XYZ as we are doing projection mapping.
WRITE(p, " v_texcoord = (%s * base.tex_mtx).xyz * vec3(base.uvscaleoffset.xy, 1.0);\n", temp_tc.c_str());
}
break;
case GE_TEXMAP_ENVIRONMENT_MAP: // Shade mapping - use dots from light sources.
WRITE(p, " v_texcoord = vec3(base.uvscaleoffset.xy * vec2(1.0 + dot(normalize(light.pos[%i]), worldnormal), 1.0 + dot(normalize(light.pos[%i]), worldnormal)) * 0.5, 1.0);\n", ls0, ls1);
break;
default:
// ILLEGAL
break;
}
}
// Compute fogdepth
if (enableFog)
WRITE(p, " v_fogdepth = (viewPos.z + base.fogcoef_stencilreplace.x) * base.fogcoef_stencilreplace.y;\n");
}
WRITE(p, "}\n");
return true;
}
bool GenerateVulkanGLSLVertexShader(const ShaderID &id, char *buffer, bool *usesLighting) {
char *p = buffer;
WRITE(p, "%s", vulkan_glsl_preamble);
bool highpFog = false;
bool highpTexcoord = false;
bool isModeThrough = id.Bit(VS_BIT_IS_THROUGH);
bool lmode = id.Bit(VS_BIT_LMODE) && !isModeThrough; // TODO: Different expression than in shaderIDgen
bool doTexture = id.Bit(VS_BIT_DO_TEXTURE);
bool doTextureTransform = id.Bit(VS_BIT_DO_TEXTURE_TRANSFORM);
GETexMapMode uvGenMode = static_cast<GETexMapMode>(id.Bits(VS_BIT_UVGEN_MODE, 2));
// this is only valid for some settings of uvGenMode
GETexProjMapMode uvProjMode = static_cast<GETexProjMapMode>(id.Bits(VS_BIT_UVPROJ_MODE, 2));
bool doShadeMapping = uvGenMode == GE_TEXMAP_ENVIRONMENT_MAP;
bool doFlatShading = id.Bit(VS_BIT_FLATSHADE);
bool useHWTransform = id.Bit(VS_BIT_USE_HW_TRANSFORM);
bool hasColor = id.Bit(VS_BIT_HAS_COLOR) || !useHWTransform;
bool hasNormal = id.Bit(VS_BIT_HAS_NORMAL) && useHWTransform;
bool hasTexcoord = id.Bit(VS_BIT_HAS_TEXCOORD) || !useHWTransform;
bool enableFog = id.Bit(VS_BIT_ENABLE_FOG);
bool throughmode = id.Bit(VS_BIT_IS_THROUGH);
bool flipNormal = id.Bit(VS_BIT_NORM_REVERSE);
int ls0 = id.Bits(VS_BIT_LS0, 2);
int ls1 = id.Bits(VS_BIT_LS1, 2);
bool enableBones = id.Bit(VS_BIT_ENABLE_BONES);
bool enableLighting = id.Bit(VS_BIT_LIGHTING_ENABLE);
int matUpdate = id.Bits(VS_BIT_MATERIAL_UPDATE, 3);
// The uniforms are passed in as three "clumps" that may or may not be present.
// We will memcpy the parts into place in a big buffer so we can be quite dynamic about what parts
// are present and what parts aren't, but we will not be ultra detailed about it.
*usesLighting = enableLighting || doShadeMapping;
WRITE(p, "\n");
WRITE(p, "layout (std140, set = 0, binding = 2) uniform baseVars {\n%s} base;\n", ub_baseStr);
if (enableLighting || doShadeMapping)
WRITE(p, "layout (std140, set = 0, binding = 3) uniform lightVars {\n%s} light;\n", ub_vs_lightsStr);
if (enableBones)
WRITE(p, "layout (std140, set = 0, binding = 4) uniform boneVars {\n%s} bone;\n", ub_vs_bonesStr);
const char *shading = doFlatShading ? "flat " : "";
DoLightComputation doLight[4] = { LIGHT_OFF, LIGHT_OFF, LIGHT_OFF, LIGHT_OFF };
if (useHWTransform) {
int shadeLight0 = doShadeMapping ? ls0 : -1;
int shadeLight1 = doShadeMapping ? ls1 : -1;
for (int i = 0; i < 4; i++) {
if (i == shadeLight0 || i == shadeLight1)
doLight[i] = LIGHT_SHADE;
if (id.Bit(VS_BIT_LIGHTING_ENABLE) && id.Bit(VS_BIT_LIGHT0_ENABLE + i))
doLight[i] = LIGHT_FULL;
}
}
int numBoneWeights = 0;
int boneWeightScale = id.Bits(VS_BIT_WEIGHT_FMTSCALE, 2);
if (enableBones) {
numBoneWeights = 1 + id.Bits(VS_BIT_BONES, 3);
WRITE(p, "%s", boneWeightDecl[numBoneWeights]);
}
if (useHWTransform)
WRITE(p, "layout (location = %d) in vec3 position;\n", PspAttributeLocation::POSITION);
else
// we pass the fog coord in w
WRITE(p, "layout (location = %d) in vec4 position;\n", PspAttributeLocation::POSITION);
if (useHWTransform && hasNormal)
WRITE(p, "layout (location = %d) in vec3 normal;\n", PspAttributeLocation::NORMAL);
bool texcoordInVec3 = false;
if (doTexture && hasTexcoord) {
if (!useHWTransform && doTextureTransform && !throughmode) {
WRITE(p, "layout (location = %d) in vec3 texcoord;\n", PspAttributeLocation::TEXCOORD);
texcoordInVec3 = true;
}
else
WRITE(p, "layout (location = %d) in vec2 texcoord;\n", PspAttributeLocation::TEXCOORD);
}
if (hasColor) {
WRITE(p, "layout (location = %d) in vec4 color0;\n", PspAttributeLocation::COLOR0);
if (lmode && !useHWTransform) // only software transform supplies color1 as vertex data
WRITE(p, "layout (location = %d) in vec3 color1;\n", PspAttributeLocation::COLOR1);
}
WRITE(p, "layout (location = 1) %sout vec4 v_color0;\n", shading);
if (lmode) {
WRITE(p, "layout (location = 2) %sout vec3 v_color1;\n", shading);
}
if (doTexture) {
WRITE(p, "layout (location = 0) out vec3 v_texcoord;\n");
}
if (enableFog) {
// See the fragment shader generator
WRITE(p, "layout (location = 3) out float v_fogdepth;\n");
}
// See comment above this function (GenerateVertexShader).
if (!isModeThrough && gstate_c.Supports(GPU_ROUND_DEPTH_TO_16BIT)) {
// Apply the projection and viewport to get the Z buffer value, floor to integer, undo the viewport and projection.
WRITE(p, "\nvec4 depthRoundZVP(vec4 v) {\n");
WRITE(p, " float z = v.z / v.w;\n");
WRITE(p, " z = z * base.depthRange.x + base.depthRange.y;\n");
WRITE(p, " z = floor(z);\n");
WRITE(p, " z = (z - base.depthRange.z) * base.depthRange.w;\n");
WRITE(p, " return vec4(v.x, v.y, z * v.w, v.w);\n");
WRITE(p, "}\n\n");
}
WRITE(p, "out gl_PerVertex { vec4 gl_Position; };\n");
WRITE(p, "void main() {\n");
if (!useHWTransform) {
// Simple pass-through of vertex data to fragment shader
if (doTexture) {
if (texcoordInVec3) {
WRITE(p, " v_texcoord = texcoord;\n");
} else {
WRITE(p, " v_texcoord = vec3(texcoord, 1.0);\n");
}
}
if (hasColor) {
WRITE(p, " v_color0 = color0;\n");
if (lmode)
WRITE(p, " v_color1 = color1;\n");
} else {
WRITE(p, " v_color0 = base.matambientalpha;\n");
if (lmode)
WRITE(p, " v_color1 = vec3(0.0);\n");
}
if (enableFog) {
WRITE(p, " v_fogdepth = position.w;\n");
}
if (isModeThrough) {
WRITE(p, " gl_Position = base.proj_through_mtx * vec4(position.xyz, 1.0);\n");
} else {
// The viewport is used in this case, so need to compensate for that.
if (gstate_c.Supports(GPU_ROUND_DEPTH_TO_16BIT)) {
WRITE(p, " gl_Position = depthRoundZVP(base.proj_mtx * vec4(position.xyz, 1.0));\n");
} else {
WRITE(p, " gl_Position = base.proj_mtx * vec4(position.xyz, 1.0);\n");
}
}
} else {
// Step 1: World Transform / Skinning
if (!enableBones) {
// No skinning, just standard T&L.
WRITE(p, " vec3 worldpos = vec4(position.xyz, 1.0) * base.world_mtx;\n");
if (hasNormal)
WRITE(p, " mediump vec3 worldnormal = normalize(vec4(%snormal, 0.0) * base.world_mtx);\n", flipNormal ? "-" : "");
else
WRITE(p, " mediump vec3 worldnormal = vec3(0.0, 0.0, 1.0);\n");
} else {
static const char *rescale[4] = { "", " * 1.9921875", " * 1.999969482421875", "" }; // 2*127.5f/128.f, 2*32767.5f/32768.f, 1.0f};
const char *factor = rescale[boneWeightScale];
static const char * const boneWeightAttr[8] = {
"w1.x", "w1.y", "w1.z", "w1.w",
"w2.x", "w2.y", "w2.z", "w2.w",
};
WRITE(p, " mat3x4 skinMatrix = w1.x * bone.m[0];\n");
if (numBoneWeights > 1) {
for (int i = 1; i < numBoneWeights; i++) {
WRITE(p, " skinMatrix += %s * bone.m[%i];\n", boneWeightAttr[i], i);
}
}
WRITE(p, ";\n");
// Trying to simplify this results in bugs in LBP...
WRITE(p, " vec3 skinnedpos = (vec4(position, 1.0) * skinMatrix) %s;\n", factor);
WRITE(p, " vec3 worldpos = vec4(skinnedpos, 1.0) * base.world_mtx;\n");
if (hasNormal) {
WRITE(p, " mediump vec3 skinnednormal = vec4(%snormal, 0.0) * skinMatrix %s;\n", flipNormal ? "-" : "", factor);
} else {
WRITE(p, " mediump vec3 skinnednormal = vec4(0.0, 0.0, %s1.0, 0.0) * skinMatrix %s;\n", flipNormal ? "-" : "", factor);
}
WRITE(p, " mediump vec3 worldnormal = normalize(vec4(skinnednormal, 0.0) * base.world_mtx);\n");
}
WRITE(p, " vec4 viewPos = vec4(vec4(worldpos, 1.0) * base.view_mtx, 1.0);\n");
// Final view and projection transforms.
if (gstate_c.Supports(GPU_ROUND_DEPTH_TO_16BIT)) {
WRITE(p, " gl_Position = depthRoundZVP(base.proj_mtx * viewPos);\n");
} else {
WRITE(p, " gl_Position = base.proj_mtx * viewPos;\n");
}
// TODO: Declare variables for dots for shade mapping if needed.
const char *ambientStr = ((matUpdate & 1) && hasColor) ? "color0" : "base.matambientalpha";
const char *diffuseStr = ((matUpdate & 2) && hasColor) ? "color0.rgb" : "light.matdiffuse";
const char *specularStr = ((matUpdate & 4) && hasColor) ? "color0.rgb" : "light.matspecular.rgb";
bool diffuseIsZero = true;
bool specularIsZero = true;
bool distanceNeeded = false;
if (enableLighting) {
WRITE(p, " vec4 lightSum0 = light.u_ambient * %s + vec4(light.matemissive, 0.0);\n", ambientStr);
for (int i = 0; i < 4; i++) {
GELightType type = static_cast<GELightType>(id.Bits(VS_BIT_LIGHT0_TYPE + 4 * i, 2));
GELightComputation comp = static_cast<GELightComputation>(id.Bits(VS_BIT_LIGHT0_COMP + 4 * i, 2));
if (doLight[i] != LIGHT_FULL)
continue;
diffuseIsZero = false;
if (comp != GE_LIGHTCOMP_ONLYDIFFUSE)
specularIsZero = false;
if (type != GE_LIGHTTYPE_DIRECTIONAL)
distanceNeeded = true;
}
if (!specularIsZero) {
WRITE(p, " vec3 lightSum1 = vec3(0.0);\n");
}
if (!diffuseIsZero) {
WRITE(p, " vec3 toLight;\n");
WRITE(p, " vec3 diffuse;\n");
}
if (distanceNeeded) {
WRITE(p, " float distance;\n");
WRITE(p, " float lightScale;\n");
}
}
// Calculate lights if needed. If shade mapping is enabled, lights may need to be
// at least partially calculated.
for (int i = 0; i < 4; i++) {
if (doLight[i] != LIGHT_FULL)
continue;
GELightType type = static_cast<GELightType>(id.Bits(VS_BIT_LIGHT0_TYPE + 4 * i, 2));
GELightComputation comp = static_cast<GELightComputation>(id.Bits(VS_BIT_LIGHT0_COMP + 4 * i, 2));
if (type == GE_LIGHTTYPE_DIRECTIONAL) {
// We prenormalize light positions for directional lights.
WRITE(p, " toLight = light.pos[%i];\n", i);
} else {
WRITE(p, " toLight = light.pos[%i] - worldpos;\n", i);
WRITE(p, " distance = length(toLight);\n");
WRITE(p, " toLight /= distance;\n");
}
bool doSpecular = comp != GE_LIGHTCOMP_ONLYDIFFUSE;
bool poweredDiffuse = comp == GE_LIGHTCOMP_BOTHWITHPOWDIFFUSE;
WRITE(p, " mediump float dot%i = max(dot(toLight, worldnormal), 0.0);\n", i);
if (poweredDiffuse) {
// pow(0.0, 0.0) may be undefined, but the PSP seems to treat it as 1.0.
// Seen in Tales of the World: Radiant Mythology (#2424.)
WRITE(p, " if (dot%i == 0.0 && light.matspecular.a == 0.0) {\n", i);
WRITE(p, " dot%i = 1.0;\n", i);
WRITE(p, " } else {\n");
WRITE(p, " dot%i = pow(dot%i, light.matspecular.a);\n", i, i);
WRITE(p, " }\n");
}
const char *timesLightScale = " * lightScale";
// Attenuation
switch (type) {
case GE_LIGHTTYPE_DIRECTIONAL:
timesLightScale = "";
break;
case GE_LIGHTTYPE_POINT:
WRITE(p, " lightScale = clamp(1.0 / dot(light.att[%i], vec3(1.0, distance, distance*distance)), 0.0, 1.0);\n", i);
break;
case GE_LIGHTTYPE_SPOT:
case GE_LIGHTTYPE_UNKNOWN:
WRITE(p, " float angle%i = dot(normalize(light.dir[%i]), toLight);\n", i, i);
WRITE(p, " if (angle%i >= light.angle[%i]) {\n", i, i);
WRITE(p, " lightScale = clamp(1.0 / dot(light.att[%i], vec3(1.0, distance, distance*distance)), 0.0, 1.0) * pow(angle%i, light.spotCoef[%i]);\n", i, i, i);
WRITE(p, " } else {\n");
WRITE(p, " lightScale = 0.0;\n");
WRITE(p, " }\n");
break;
default:
// ILLEGAL
break;
}
WRITE(p, " diffuse = (light.diffuse[%i] * %s) * dot%i;\n", i, diffuseStr, i);
if (doSpecular) {
WRITE(p, " dot%i = dot(normalize(toLight + vec3(0.0, 0.0, 1.0)), worldnormal);\n", i);
WRITE(p, " if (dot%i > 0.0)\n", i);
WRITE(p, " lightSum1 += light.specular[%i] * %s * (pow(dot%i, light.matspecular.a) %s);\n", i, specularStr, i, timesLightScale);
}
WRITE(p, " lightSum0.rgb += (light.ambient[%i] * %s.rgb + diffuse)%s;\n", i, ambientStr, timesLightScale);
}
if (enableLighting) {
// Sum up ambient, emissive here.
if (lmode) {
WRITE(p, " v_color0 = clamp(lightSum0, 0.0, 1.0);\n");
// v_color1 only exists when lmode = 1.
if (specularIsZero) {
WRITE(p, " v_color1 = vec3(0.0);\n");
} else {
WRITE(p, " v_color1 = clamp(lightSum1, 0.0, 1.0);\n");
}
} else {
if (specularIsZero) {
WRITE(p, " v_color0 = clamp(lightSum0, 0.0, 1.0);\n");
} else {
WRITE(p, " v_color0 = clamp(clamp(lightSum0, 0.0, 1.0) + vec4(lightSum1, 0.0), 0.0, 1.0);\n");
}
}
} else {
// Lighting doesn't affect color.
if (hasColor) {
WRITE(p, " v_color0 = color0;\n");
} else {
WRITE(p, " v_color0 = base.matambientalpha;\n");
}
if (lmode) {
WRITE(p, " v_color1 = vec3(0.0);\n");
}
}
bool scaleUV = !throughmode && (uvGenMode == GE_TEXMAP_TEXTURE_COORDS || uvGenMode == GE_TEXMAP_UNKNOWN);
// Step 3: UV generation
if (doTexture) {
switch (uvGenMode) {
case GE_TEXMAP_TEXTURE_COORDS: // Scale-offset. Easy.
case GE_TEXMAP_UNKNOWN: // Not sure what this is, but Riviera uses it. Treating as coords works.
if (scaleUV) {
if (hasTexcoord) {
WRITE(p, " v_texcoord = vec3(texcoord.xy, 0.0);\n");
} else {
WRITE(p, " v_texcoord = vec3(0.0);\n");
}
} else {
if (hasTexcoord) {
WRITE(p, " v_texcoord = vec3(texcoord.xy * base.uvscaleoffset.xy + base.uvscaleoffset.zw, 0.0);\n");
} else {
WRITE(p, " v_texcoord = vec3(base.uvscaleoffset.zw, 0.0);\n");
}
}
break;
case GE_TEXMAP_TEXTURE_MATRIX: // Projection mapping.
{
std::string temp_tc;
switch (uvProjMode) {
case GE_PROJMAP_POSITION: // Use model space XYZ as source
temp_tc = "vec4(position.xyz, 1.0)";
break;
case GE_PROJMAP_UV: // Use unscaled UV as source
{
// scaleUV is false here.
if (hasTexcoord) {
temp_tc = "vec4(texcoord.xy, 0.0, 1.0)";
} else {
temp_tc = "vec4(0.0, 0.0, 0.0, 1.0)";
}
}
break;
case GE_PROJMAP_NORMALIZED_NORMAL: // Use normalized transformed normal as source
if (hasNormal)
temp_tc = flipNormal ? "vec4(normalize(-normal), 1.0)" : "vec4(normalize(normal), 1.0)";
else
temp_tc = "vec4(0.0, 0.0, 1.0, 1.0)";
break;
case GE_PROJMAP_NORMAL: // Use non-normalized transformed normal as source
if (hasNormal)
temp_tc = flipNormal ? "vec4(-normal, 1.0)" : "vec4(normal, 1.0)";
else
temp_tc = "vec4(0.0, 0.0, 1.0, 1.0)";
break;
}
// Transform by texture matrix. XYZ as we are doing projection mapping.
WRITE(p, " v_texcoord = (%s * base.tex_mtx).xyz * vec3(base.uvscaleoffset.xy, 1.0);\n", temp_tc.c_str());
}
break;
case GE_TEXMAP_ENVIRONMENT_MAP: // Shade mapping - use dots from light sources.
WRITE(p, " v_texcoord = vec3(base.uvscaleoffset.xy * vec2(1.0 + dot(normalize(light.pos[%i]), worldnormal), 1.0 + dot(normalize(light.pos[%i]), worldnormal)) * 0.5, 1.0);\n", ls0, ls1);
break;
default:
// ILLEGAL
break;
}
}
// Compute fogdepth
if (enableFog)
WRITE(p, " v_fogdepth = (viewPos.z + base.fogcoef_stencilreplace.x) * base.fogcoef_stencilreplace.y;\n");
}
WRITE(p, "}\n");
return true;
}
// Missing: Z depth range
bool GenerateFragmentShader(const ShaderID &id, 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 *fragColor1 = "fragColor1";
const char *texture = "texture2D";
const char *texelFetch = NULL;
bool highpFog = false;
bool highpTexcoord = false;
bool bitwiseOps = false;
const char *lastFragData = nullptr;
if (gl_extensions.IsGLES) {
// ES doesn't support dual source alpha :(
if (gstate_c.featureFlags & GPU_SUPPORTS_GLSL_ES_300) {
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";
}
if (gl_extensions.EXT_blend_func_extended) {
// Oldy moldy GLES, so use the fixed output name.
fragColor1 = "gl_SecondaryFragColorEXT";
}
}
// 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;
if (gstate_c.featureFlags & GPU_SUPPORTS_ANY_FRAMEBUFFER_FETCH) {
if (gl_extensions.GLES3 && gl_extensions.EXT_shader_framebuffer_fetch) {
WRITE(p, "#extension GL_EXT_shader_framebuffer_fetch : require\n");
lastFragData = "fragColor0";
} else if (gl_extensions.EXT_shader_framebuffer_fetch) {
WRITE(p, "#extension GL_EXT_shader_framebuffer_fetch : require\n");
lastFragData = "gl_LastFragData[0]";
} else if (gl_extensions.NV_shader_framebuffer_fetch) {
// GL_NV_shader_framebuffer_fetch is available on mobile platform and ES 2.0 only but not on desktop.
WRITE(p, "#extension GL_NV_shader_framebuffer_fetch : require\n");
lastFragData = "gl_LastFragData[0]";
} else if (gl_extensions.ARM_shader_framebuffer_fetch) {
WRITE(p, "#extension GL_ARM_shader_framebuffer_fetch : require\n");
lastFragData = "gl_LastFragColorARM";
}
}
WRITE(p, "precision lowp float;\n");
} else {
// TODO: Handle this in VersionGEThan?
#if !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");
} 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");
}
} 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";
}
}
#endif
// We remove these everywhere - GL4, GL3, Mac-forced-GL2, etc.
WRITE(p, "#define lowp\n");
WRITE(p, "#define mediump\n");
WRITE(p, "#define highp\n");
}
if (glslES30) {
varying = "in";
}
bool lmode = id.Bit(FS_BIT_LMODE);
bool doTexture = id.Bit(FS_BIT_DO_TEXTURE);
bool enableFog = id.Bit(FS_BIT_ENABLE_FOG);
bool enableAlphaTest = id.Bit(FS_BIT_ALPHA_TEST);
bool alphaTestAgainstZero = id.Bit(FS_BIT_ALPHA_AGAINST_ZERO);
bool enableColorTest = id.Bit(FS_BIT_COLOR_TEST);
bool colorTestAgainstZero = id.Bit(FS_BIT_COLOR_AGAINST_ZERO);
bool enableColorDoubling = id.Bit(FS_BIT_COLOR_DOUBLE);
bool doTextureProjection = id.Bit(FS_BIT_DO_TEXTURE_PROJ);
bool doTextureAlpha = id.Bit(FS_BIT_TEXALPHA);
bool doFlatShading = id.Bit(FS_BIT_FLATSHADE);
GEComparison alphaTestFunc = (GEComparison)id.Bits(FS_BIT_ALPHA_TEST_FUNC, 3);
GEComparison colorTestFunc = (GEComparison)id.Bits(FS_BIT_COLOR_TEST_FUNC, 2);
bool needShaderTexClamp = id.Bit(FS_BIT_SHADER_TEX_CLAMP);
GETexFunc texFunc = (GETexFunc)id.Bits(FS_BIT_TEXFUNC, 3);
bool textureAtOffset = id.Bit(FS_BIT_TEXTURE_AT_OFFSET);
ReplaceBlendType replaceBlend = static_cast<ReplaceBlendType>(id.Bits(FS_BIT_REPLACE_BLEND, 3));
ReplaceAlphaType stencilToAlpha = static_cast<ReplaceAlphaType>(id.Bits(FS_BIT_STENCIL_TO_ALPHA, 2));
GEBlendSrcFactor replaceBlendFuncA = (GEBlendSrcFactor)id.Bits(FS_BIT_BLENDFUNC_A, 4);
GEBlendDstFactor replaceBlendFuncB = (GEBlendDstFactor)id.Bits(FS_BIT_BLENDFUNC_B, 4);
GEBlendMode replaceBlendEq = (GEBlendMode)id.Bits(FS_BIT_BLENDEQ, 3);
bool isModeClear = id.Bit(FS_BIT_CLEARMODE);
const char *shading = "";
if (glslES30)
shading = doFlatShading ? "flat" : "";
if (doTexture)
WRITE(p, "uniform sampler2D tex;\n");
if (!isModeClear && replaceBlend > REPLACE_BLEND_STANDARD) {
if (!gstate_c.Supports(GPU_SUPPORTS_ANY_FRAMEBUFFER_FETCH) && replaceBlend == REPLACE_BLEND_COPY_FBO) {
if (!texelFetch) {
WRITE(p, "uniform vec2 u_fbotexSize;\n");
}
WRITE(p, "uniform sampler2D fbotex;\n");
}
if (replaceBlendFuncA >= GE_SRCBLEND_FIXA) {
WRITE(p, "uniform vec3 u_blendFixA;\n");
}
if (replaceBlendFuncB >= GE_DSTBLEND_FIXB) {
WRITE(p, "uniform vec3 u_blendFixB;\n");
}
}
if (needShaderTexClamp && doTexture) {
WRITE(p, "uniform vec4 u_texclamp;\n");
if (id.Bit(FS_BIT_TEXTURE_AT_OFFSET)) {
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 && !colorTestAgainstZero) || (enableAlphaTest && !alphaTestAgainstZero))) {
WRITE(p, "uniform ivec4 u_alphacolormask;\n");
}
}
}
StencilValueType replaceAlphaWithStencilType = (StencilValueType)id.Bits(FS_BIT_REPLACE_ALPHA_WITH_STENCIL_TYPE, 4);
if (stencilToAlpha && replaceAlphaWithStencilType == STENCIL_VALUE_UNIFORM) {
WRITE(p, "uniform float u_stencilReplaceValue;\n");
}
if (doTexture && texFunc == 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 && !colorTestAgainstZero) {
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 (!strcmp(fragColor0, "fragColor0")) {
const char *qualifierColor0 = "out";
if (lastFragData && !strcmp(lastFragData, fragColor0)) {
qualifierColor0 = "inout";
}
// Output the output color definitions.
if (stencilToAlpha == REPLACE_ALPHA_DUALSOURCE) {
WRITE(p, "%s vec4 fragColor0;\n", qualifierColor0);
WRITE(p, "out vec4 fragColor1;\n");
} else {
WRITE(p, "%s vec4 fragColor0;\n", qualifierColor0);
}
}
// 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) && needShaderTexClamp) {
WRITE(p, "float mymod(float a, float b) { return a - b * floor(a / b); }\n");
}
WRITE(p, "void main() {\n");
if (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 (doTexture) {
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 (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.x / v_texcoord.z)";
vcoord = "(v_texcoord.y / v_texcoord.z)";
}
std::string modulo = (gl_extensions.bugs & BUG_PVR_SHADER_PRECISION_BAD) ? "mymod" : "mod";
if (id.Bit(FS_BIT_CLAMP_S)) {
ucoord = "clamp(" + ucoord + ", u_texclamp.z, u_texclamp.x - u_texclamp.z)";
} else {
ucoord = modulo + "(" + ucoord + ", u_texclamp.x)";
}
if (id.Bit(FS_BIT_CLAMP_T)) {
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)";
}
WRITE(p, " vec2 fixedcoord = vec2(%s, %s);\n", ucoord.c_str(), vcoord.c_str());
texcoord = "fixedcoord";
// We already projected it.
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 (texFunc) {
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 (texFunc) {
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 && !colorTestAgainstZero) {
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) {
// 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 {
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 (colorTestAgainstZero) {
// When testing against 0 (common), we can avoid some math.
// 0.002 is approximately half of 1.0 / 255.0.
if (colorTestFunc == GE_COMP_NOTEQUAL) {
WRITE(p, " if (v.r < 0.002 && v.g < 0.002 && v.b < 0.002) discard;\n");
} else if (colorTestFunc != GE_COMP_NEVER) {
// Anything else is a test for == 0.
WRITE(p, " if (v.r > 0.002 || v.g > 0.002 || v.b > 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 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);
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 {
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) {
const char *srcFactor = "ERROR";
switch (replaceBlendFuncA) {
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;
case GE_SRCBLEND_DOUBLEINVSRCALPHA:
srcFactor = "vec3(1.0 - v.a * 2.0)";
break;
// PRE_SRC for REPLACE_BLEND_PRE_SRC_2X_ALPHA means "double the src."
// It's close to the same, but clamping can still be an issue.
case GE_SRCBLEND_DOUBLEDSTALPHA:
srcFactor = "vec3(2.0)";
break;
case GE_SRCBLEND_DOUBLEINVDSTALPHA:
srcFactor = "ERROR";
break;
case GE_SRCBLEND_FIXA:
srcFactor = "u_blendFixA";
break;
default:
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.
if (gstate_c.featureFlags & GPU_SUPPORTS_ANY_FRAMEBUFFER_FETCH) {
WRITE(p, " lowp vec4 destColor = %s;\n", lastFragData);
} 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);
}
const char *srcFactor = "vec3(1.0)";
const char *dstFactor = "vec3(0.0)";
switch (replaceBlendFuncA) {
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;
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;
default:
srcFactor = "u_blendFixA";
break;
}
switch (replaceBlendFuncB) {
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;
default:
srcFactor = "u_blendFixB";
break;
}
switch (replaceBlendEq) {
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, " %s = vec4(v.rgb, %s);\n", fragColor0, replacedAlpha.c_str());
WRITE(p, " %s = vec4(0.0, 0.0, 0.0, v.a);\n", fragColor1);
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;
default:
ERROR_LOG(G3D, "Bad stencil-to-alpha type, corrupt ID?");
return false;
}
LogicOpReplaceType replaceLogicOpType = (LogicOpReplaceType)id.Bits(FS_BIT_REPLACE_LOGIC_OP_TYPE, 2);
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;
default:
ERROR_LOG(G3D, "Bad logic op type, corrupt ID?");
return false;
}
#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
if (gstate_c.Supports(GPU_ROUND_FRAGMENT_DEPTH_TO_16BIT)) {
WRITE(p, " highp float z = gl_FragCoord.z;\n");
WRITE(p, " z = (1.0/65535.0) * floor(z * 65535.0);\n");
WRITE(p, " gl_FragDepth = z;\n");
}
WRITE(p, "}\n");
return true;
}
// Missing: Z depth range
bool GenerateVulkanGLSLFragmentShader(const ShaderID &id, char *buffer) {
char *p = buffer;
const char *lastFragData = nullptr;
WRITE(p, "%s", vulkan_glsl_preamble);
bool lmode = id.Bit(FS_BIT_LMODE);
bool doTexture = id.Bit(FS_BIT_DO_TEXTURE);
bool enableFog = id.Bit(FS_BIT_ENABLE_FOG);
bool enableAlphaTest = id.Bit(FS_BIT_ALPHA_TEST);
bool alphaTestAgainstZero = id.Bit(FS_BIT_ALPHA_AGAINST_ZERO);
bool enableColorTest = id.Bit(FS_BIT_COLOR_TEST);
bool colorTestAgainstZero = id.Bit(FS_BIT_COLOR_AGAINST_ZERO);
bool enableColorDoubling = id.Bit(FS_BIT_COLOR_DOUBLE);
bool doTextureProjection = id.Bit(FS_BIT_DO_TEXTURE_PROJ);
bool doTextureAlpha = id.Bit(FS_BIT_TEXALPHA);
bool doFlatShading = id.Bit(FS_BIT_FLATSHADE);
GEComparison alphaTestFunc = (GEComparison)id.Bits(FS_BIT_ALPHA_TEST_FUNC, 3);
GEComparison colorTestFunc = (GEComparison)id.Bits(FS_BIT_COLOR_TEST_FUNC, 2);
bool needShaderTexClamp = id.Bit(FS_BIT_SHADER_TEX_CLAMP);
GETexFunc texFunc = (GETexFunc)id.Bits(FS_BIT_TEXFUNC, 3);
bool textureAtOffset = id.Bit(FS_BIT_TEXTURE_AT_OFFSET);
ReplaceBlendType replaceBlend = static_cast<ReplaceBlendType>(id.Bits(FS_BIT_REPLACE_BLEND, 3));
ReplaceAlphaType stencilToAlpha = static_cast<ReplaceAlphaType>(id.Bits(FS_BIT_STENCIL_TO_ALPHA, 2));
GEBlendSrcFactor replaceBlendFuncA = (GEBlendSrcFactor)id.Bits(FS_BIT_BLENDFUNC_A, 4);
GEBlendDstFactor replaceBlendFuncB = (GEBlendDstFactor)id.Bits(FS_BIT_BLENDFUNC_B, 4);
GEBlendMode replaceBlendEq = (GEBlendMode)id.Bits(FS_BIT_BLENDEQ, 3);
StencilValueType replaceAlphaWithStencilType = (StencilValueType)id.Bits(FS_BIT_REPLACE_ALPHA_WITH_STENCIL_TYPE, 4);
bool isModeClear = id.Bit(FS_BIT_CLEARMODE);
const char *shading = doFlatShading ? "flat" : "";
WRITE(p, "layout (std140, set = 0, binding = 2) uniform baseUBO {\n%s} base;\n", ub_baseStr);
if (doTexture) {
WRITE(p, "layout (binding = 0) uniform sampler2D tex;\n");
}
if (!isModeClear && replaceBlend > REPLACE_BLEND_STANDARD) {
if (replaceBlend == REPLACE_BLEND_COPY_FBO) {
WRITE(p, "layout (binding = 1) uniform sampler2D fbotex;\n");
}
}
WRITE(p, "layout (location = 1) %s in vec4 v_color0;\n", shading);
if (lmode)
WRITE(p, "layout (location = 2) %s in vec3 v_color1;\n", shading);
if (enableFog) {
WRITE(p, "layout (location = 3) in float v_fogdepth;\n");
}
if (doTexture) {
WRITE(p, "layout (location = 0) in vec3 v_texcoord;\n");
}
if (enableAlphaTest && !alphaTestAgainstZero) {
WRITE(p, "int roundAndScaleTo255i(in float x) { return int(floor(x * 255.0 + 0.5)); }\n");
}
if (enableColorTest && !colorTestAgainstZero) {
WRITE(p, "ivec3 roundAndScaleTo255iv(in vec3 x) { return ivec3(floor(x * 255.0 + 0.5)); }\n");
}
WRITE(p, "layout (location = 0, index = 0) out vec4 fragColor0;\n");
if (stencilToAlpha == REPLACE_ALPHA_DUALSOURCE) {
WRITE(p, "layout (location = 0, index = 1) out vec4 fragColor1;\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) && needShaderTexClamp) {
WRITE(p, "float mymod(float a, float b) { return a - b * floor(a / b); }\n");
}
WRITE(p, "void main() {\n");
if (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 (doTexture) {
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 (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.x / v_texcoord.z)";
vcoord = "(v_texcoord.y / v_texcoord.z)";
}
std::string modulo = (gl_extensions.bugs & BUG_PVR_SHADER_PRECISION_BAD) ? "mymod" : "mod";
if (id.Bit(FS_BIT_CLAMP_S)) {
ucoord = "clamp(" + ucoord + ", base.texclamp.z, base.texclamp.x - base.texclamp.z)";
} else {
ucoord = modulo + "(" + ucoord + ", base.texclamp.x)";
}
if (id.Bit(FS_BIT_CLAMP_T)) {
vcoord = "clamp(" + vcoord + ", base.texclamp.w, base.texclamp.y - base.texclamp.w)";
} else {
vcoord = modulo + "(" + vcoord + ", base.texclamp.y)";
}
if (textureAtOffset) {
ucoord = "(" + ucoord + " + base.texclampoff.x)";
vcoord = "(" + vcoord + " + base.texclampoff.y)";
}
WRITE(p, " vec2 fixedcoord = vec2(%s, %s);\n", ucoord.c_str(), vcoord.c_str());
texcoord = "fixedcoord";
// We already projected it.
doTextureProjection = false;
}
if (doTextureProjection) {
WRITE(p, " vec4 t = textureProj(tex, %s);\n", texcoord);
} else {
WRITE(p, " vec4 t = texture(tex, %s.xy);\n", texcoord);
}
WRITE(p, " vec4 p = v_color0;\n");
if (doTextureAlpha) { // texfmt == RGBA
switch (texFunc) {
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, base.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 (texFunc) {
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, base.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 (enableAlphaTest) {
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 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 {
const char *alphaTestFuncs[] = { "#", "#", " != ", " == ", " >= ", " > ", " <= ", " < " };
if (alphaTestFuncs[alphaTestFunc][0] != '#') {
WRITE(p, " if ((roundAndScaleTo255i(v.a) & base.alphacolormask.a) %s base.alphacolorref.a) discard;\n", alphaTestFuncs[alphaTestFunc]);
} else {
// This means NEVER. See above.
WRITE(p, " discard;\n");
}
}
}
if (enableColorTest) {
if (colorTestAgainstZero) {
// When testing against 0 (common), we can avoid some math.
// 0.002 is approximately half of 1.0 / 255.0.
if (colorTestFunc == GE_COMP_NOTEQUAL) {
WRITE(p, " if (v.r < 0.002 && v.g < 0.002 && v.b < 0.002) discard;\n");
} else if (colorTestFunc != GE_COMP_NEVER) {
// Anything else is a test for == 0.
WRITE(p, " if (v.r > 0.002 || v.g > 0.002 || v.b > 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 {
const char *colorTestFuncs[] = { "#", "#", " != ", " == " };
if (colorTestFuncs[colorTestFunc][0] != '#') {
WRITE(p, " ivec3 v_scaled = roundAndScaleTo255iv(v.rgb);\n");
WRITE(p, " if ((v_scaled & base.alphacolormask.rgb) %s (base.alphacolorref.rgb & base.alphacolormask.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(base.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) {
const char *srcFactor = "ERROR";
switch (replaceBlendFuncA) {
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;
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 = "base.blendFixA"; break;
}
WRITE(p, " v.rgb = v.rgb * %s;\n", srcFactor);
}
if (replaceBlend == REPLACE_BLEND_COPY_FBO) {
WRITE(p, " lowp vec4 destColor = texelFetch(fbotex, ivec2(gl_FragCoord.x, gl_FragCoord.y), 0);\n");
const char *srcFactor = "vec3(1.0)";
const char *dstFactor = "vec3(0.0)";
switch (replaceBlendFuncA) {
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;
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 = "base.blendFixA"; break;
}
switch (replaceBlendFuncB) {
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 = "base.blendFixB"; break;
}
switch (replaceBlendEq) {
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 = "base.fogcoef_stencilreplace.z";
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, " fragColor0 = vec4(v.rgb, %s);\n", replacedAlpha.c_str());
break;
case REPLACE_ALPHA_NO:
WRITE(p, " fragColor0 = v;\n");
break;
default:
ERROR_LOG(G3D, "Bad stencil-to-alpha type, corrupt ID?");
return false;
}
LogicOpReplaceType replaceLogicOpType = (LogicOpReplaceType)id.Bits(FS_BIT_REPLACE_LOGIC_OP_TYPE, 2);
switch (replaceLogicOpType) {
case LOGICOPTYPE_ONE:
WRITE(p, " fragColor0.rgb = vec3(1.0, 1.0, 1.0);\n");
break;
case LOGICOPTYPE_INVERT:
WRITE(p, " fragColor0.rgb = vec3(1.0, 1.0, 1.0) - fragColor0.rgb;\n");
break;
case LOGICOPTYPE_NORMAL:
break;
default:
ERROR_LOG(G3D, "Bad logic op type, corrupt ID?");
return false;
}
if (gstate_c.Supports(GPU_ROUND_FRAGMENT_DEPTH_TO_16BIT)) {
WRITE(p, " highp float z = gl_FragCoord.z;\n");
WRITE(p, " z = (1.0/65535.0) * floor(z * 65535.0);\n");
WRITE(p, " gl_FragDepth = z;\n");
}
WRITE(p, "}\n");
return true;
}