void GenerateVertexShader(int prim, char *buffer) {
char *p = buffer;
#if defined(USING_GLES2)
WRITE(p, "precision highp float;\n");
#elif !defined(FORCE_OPENGL_2_0)
WRITE(p, "#version 110\n");
// Remove lowp/mediump in non-mobile implementations
WRITE(p, "#define lowp\n");
WRITE(p, "#define mediump\n");
#else
// Need to remove lowp/mediump for Mac
WRITE(p, "#define lowp\n");
WRITE(p, "#define mediump\n");
#endif
const u32 vertType = gstate.vertType;
int lmode = (gstate.lmode & 1) && gstate.isLightingEnabled();
int doTexture = gstate.isTextureMapEnabled() && !gstate.isModeClear();
bool hwXForm = CanUseHardwareTransform(prim);
bool hasColor = (vertType & GE_VTYPE_COL_MASK) != 0 || !hwXForm;
bool hasNormal = (vertType & GE_VTYPE_NRM_MASK) != 0 && hwXForm;
bool enableFog = gstate.isFogEnabled() && !gstate.isModeThrough() && !gstate.isModeClear();
bool throughmode = (vertType & GE_VTYPE_THROUGH_MASK) != 0;
bool flipV = gstate_c.flipTexture;
bool doTextureProjection = gstate.getUVGenMode() == 1;
DoLightComputation doLight[4] = {LIGHT_OFF, LIGHT_OFF, LIGHT_OFF, LIGHT_OFF};
if (hwXForm) {
int shadeLight0 = gstate.getUVGenMode() == 2 ? gstate.getUVLS0() : -1;
int shadeLight1 = gstate.getUVGenMode() == 2 ? gstate.getUVLS1() : -1;
for (int i = 0; i < 4; i++) {
if (!hasNormal)
continue;
if (i == shadeLight0 || i == shadeLight1)
doLight[i] = LIGHT_SHADE;
if ((gstate.lightingEnable & 1) && (gstate.lightEnable[i] & 1))
doLight[i] = LIGHT_FULL;
}
}
if ((vertType & GE_VTYPE_WEIGHT_MASK) != GE_VTYPE_WEIGHT_NONE) {
WRITE(p, "%s", boneWeightAttrDecl[gstate.getNumBoneWeights() - 1]);
}
if (hwXForm)
WRITE(p, "attribute vec3 a_position;\n");
else
WRITE(p, "attribute vec4 a_position;\n"); // need to pass the fog coord in w
if (doTexture) {
if (!hwXForm && doTextureProjection)
WRITE(p, "attribute vec3 a_texcoord;\n");
else
WRITE(p, "attribute vec2 a_texcoord;\n");
}
if (hasColor) {
WRITE(p, "attribute lowp vec4 a_color0;\n");
if (lmode && !hwXForm) // only software transform supplies color1 as vertex data
WRITE(p, "attribute lowp vec3 a_color1;\n");
}
if (hwXForm && hasNormal)
WRITE(p, "attribute mediump vec3 a_normal;\n");
if (gstate.isModeThrough()) {
WRITE(p, "uniform mat4 u_proj_through;\n");
} else {
WRITE(p, "uniform mat4 u_proj;\n");
// Add all the uniforms we'll need to transform properly.
}
if (hwXForm || !hasColor)
WRITE(p, "uniform lowp vec4 u_matambientalpha;\n"); // matambient + matalpha
if (enableFog) {
WRITE(p, "uniform vec2 u_fogcoef;\n");
}
if (hwXForm) {
// When transforming by hardware, we need a great deal more uniforms...
WRITE(p, "uniform mat4 u_world;\n");
WRITE(p, "uniform mat4 u_view;\n");
if (gstate.getUVGenMode() == 0)
WRITE(p, "uniform vec4 u_uvscaleoffset;\n");
else if (gstate.getUVGenMode() == 1)
WRITE(p, "uniform mat4 u_texmtx;\n");
if ((vertType & GE_VTYPE_WEIGHT_MASK) != GE_VTYPE_WEIGHT_NONE) {
int numBones = 1 + ((vertType & GE_VTYPE_WEIGHTCOUNT_MASK) >> GE_VTYPE_WEIGHTCOUNT_SHIFT);
for (int i = 0; i < numBones; i++) {
WRITE(p, "uniform mat4 u_bone%i;\n", i);
}
}
if (gstate.isLightingEnabled()) {
WRITE(p, "uniform lowp vec4 u_ambient;\n");
if ((gstate.materialupdate & 2) == 0)
WRITE(p, "uniform lowp vec3 u_matdiffuse;\n");
// if ((gstate.materialupdate & 4) == 0)
WRITE(p, "uniform lowp vec4 u_matspecular;\n"); // Specular coef is contained in alpha
WRITE(p, "uniform lowp vec3 u_matemissive;\n");
}
for (int i = 0; i < 4; i++) {
if (doLight[i] != LIGHT_OFF) {
// This is needed for shade mapping
WRITE(p, "uniform vec3 u_lightpos%i;\n", i);
}
if (doLight[i] == LIGHT_FULL) {
// These are needed for the full thing
WRITE(p, "uniform vec3 u_lightdir%i;\n", i);
WRITE(p, "uniform vec3 u_lightatt%i;\n", i);
WRITE(p, "uniform float u_lightangle%i;\n", i);
WRITE(p, "uniform float u_lightspotCoef%i;\n", i);
WRITE(p, "uniform lowp vec3 u_lightambient%i;\n", i);
WRITE(p, "uniform lowp vec3 u_lightdiffuse%i;\n", i);
WRITE(p, "uniform lowp vec3 u_lightspecular%i;\n", i);
}
}
}
VSShader *ShaderManagerDX9::ApplyShader(int prim, u32 vertType) {
bool useHWTransform = CanUseHardwareTransform(prim);
ShaderID VSID;
ComputeVertexShaderID(&VSID, vertType, useHWTransform);
ShaderID FSID;
ComputeFragmentShaderID(&FSID);
// Just update uniforms if this is the same shader as last time.
if (lastVShader_ != nullptr && lastPShader_ != nullptr && VSID == lastVSID_ && FSID == lastFSID_) {
if (globalDirty_) {
PSUpdateUniforms(globalDirty_);
VSUpdateUniforms(globalDirty_);
globalDirty_ = 0;
}
return lastVShader_; // Already all set.
}
VSCache::iterator vsIter = vsCache_.find(VSID);
VSShader *vs;
if (vsIter == vsCache_.end()) {
// Vertex shader not in cache. Let's compile it.
GenerateVertexShaderDX9(VSID, codeBuffer_);
vs = new VSShader(VSID, codeBuffer_, useHWTransform);
if (vs->Failed()) {
ERROR_LOG(HLE, "Shader compilation failed, falling back to software transform");
osm.Show("hardware transform error - falling back to software", 2.5f, 0xFF3030FF, -1, true);
delete vs;
ComputeVertexShaderID(&VSID, vertType, false);
// TODO: Look for existing shader with the appropriate ID, use that instead of generating a new one - however, need to make sure
// that that shader ID is not used when computing the linked shader ID below, because then IDs won't match
// next time and we'll do this over and over...
// Can still work with software transform.
GenerateVertexShaderDX9(VSID, codeBuffer_);
vs = new VSShader(VSID, codeBuffer_, false);
}
vsCache_[VSID] = vs;
} else {
vs = vsIter->second;
}
lastVSID_ = VSID;
FSCache::iterator fsIter = fsCache_.find(FSID);
PSShader *fs;
if (fsIter == fsCache_.end()) {
// Fragment shader not in cache. Let's compile it.
GenerateFragmentShaderDX9(FSID, codeBuffer_);
fs = new PSShader(FSID, codeBuffer_);
fsCache_[FSID] = fs;
} else {
fs = fsIter->second;
}
lastFSID_ = FSID;
if (globalDirty_) {
PSUpdateUniforms(globalDirty_);
VSUpdateUniforms(globalDirty_);
globalDirty_ = 0;
}
pD3Ddevice->SetPixelShader(fs->shader);
pD3Ddevice->SetVertexShader(vs->shader);
lastPShader_ = fs;
lastVShader_ = vs;
return vs;
}