libretro/ppsspp
1
0
Fork 0
mirror of https://github.com/libretro/ppsspp.git synced 2024-11-24 08:39:51 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Use arrays and for loops in GLSL shaders for bone processing, to shrink the shaders which are becoming too big for some mobile GPUs. Might fix #1816.

Browse Source
This commit is contained in:
Henrik Rydgard 2013-05-18 20:44:25 +02:00
parent dca2406374
commit 1b210fab14
3 changed files with 70 additions and 37 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
GPU/GLES/ShaderManager.cpp
View File

@ -54,6 +54,9 @@ Shader::Shader(const char *code, uint32_t shaderType) {
ERROR_LOG(G3D, "Info log: %s\n", infoLog);
ERROR_LOG(G3D, "Shader source:\n%s\n", (const char *)code);
Reporting::ReportMessage("Error in shader compilation: info: %s / code: %s", infoLog, (const char *)code);
#ifdef SHADERLOG
OutputDebugString(infoLog);
#endif
} else {
DEBUG_LOG(G3D, "Compiled shader:\n%s\n", (const char *)code);
}
@ -102,11 +105,8 @@ LinkedShader::LinkedShader(Shader *vs, Shader *fs)
u_view = glGetUniformLocation(program, "u_view");
u_world = glGetUniformLocation(program, "u_world");
u_texmtx = glGetUniformLocation(program, "u_texmtx");
for (int i = 0; i < 8; i++) {
char name[64];
sprintf(name, "u_bone%i", i);
u_bone[i] = glGetUniformLocation(program, name);
}
u_bone = glGetUniformLocation(program, "u_bone");
numBones = gstate.getNumBoneWeights();
// Lighting, texturing
u_ambient = glGetUniformLocation(program, "u_ambient");
@ -141,8 +141,8 @@ LinkedShader::LinkedShader(Shader *vs, Shader *fs)
a_color1 = glGetAttribLocation(program, "a_color1");
a_texcoord = glGetAttribLocation(program, "a_texcoord");
a_normal = glGetAttribLocation(program, "a_normal");
a_weight0123 = glGetAttribLocation(program, "a_weight0123");
a_weight4567 = glGetAttribLocation(program, "a_weight4567");
a_weight0123 = glGetAttribLocation(program, "a_w1");
a_weight4567 = glGetAttribLocation(program, "a_w2");
glUseProgram(program);
@ -317,9 +317,13 @@ void LinkedShader::updateUniforms() {
if (u_texmtx != -1 && (dirtyUniforms & DIRTY_TEXMATRIX)) {
SetMatrix4x3(u_texmtx, gstate.tgenMatrix);
}
for (int i = 0; i < 8; i++) {
if (u_bone[i] != -1 && (dirtyUniforms & (DIRTY_BONEMATRIX0 << i))) {
SetMatrix4x3(u_bone[i], gstate.boneMatrix + 12 * i);
// TODO: Could even set all bones in one go if they're all dirty.
if (u_bone != -1) {
for (int i = 0; i < numBones; i++) {
if (dirtyUniforms & (DIRTY_BONEMATRIX0 << i)) {
SetMatrix4x3(u_bone + i, gstate.boneMatrix + 12 * i);
}
}
}

3
GPU/GLES/ShaderManager.h
View File

@ -54,7 +54,8 @@ public:
int u_view;
int u_texmtx;
int u_world;
int u_bone[8];
int u_bone; // array, size is numBones
int numBones;
// Fragment processing inputs
int u_alphacolorref;

80
GPU/GLES/VertexShaderGenerator.cpp
View File

@ -106,25 +106,25 @@ void ComputeVertexShaderID(VertexShaderID *id, int prim) {
}
static const char * const boneWeightAttrDecl[8] = {
"attribute mediump float a_weight0123;\n",
"attribute mediump vec2 a_weight0123;\n",
"attribute mediump vec3 a_weight0123;\n",
"attribute mediump vec4 a_weight0123;\n",
"attribute mediump vec4 a_weight0123;\nattribute mediump float a_weight4567;\n",
"attribute mediump vec4 a_weight0123;\nattribute mediump vec2 a_weight4567;\n",
"attribute mediump vec4 a_weight0123;\nattribute mediump vec3 a_weight4567;\n",
"attribute mediump vec4 a_weight0123;\nattribute mediump vec4 a_weight4567;\n",
"attribute mediump float a_w1;\n",
"attribute mediump vec2 a_w1;\n",
"attribute mediump vec3 a_w1;\n",
"attribute mediump vec4 a_w1;\n",
"attribute mediump vec4 a_w1;\nattribute mediump float a_w2;\n",
"attribute mediump vec4 a_w1;\nattribute mediump vec2 a_w2;\n",
"attribute mediump vec4 a_w1;\nattribute mediump vec3 a_w2;\n",
"attribute mediump vec4 a_w1;\nattribute mediump vec4 a_w2;\n",
};
static const char * const boneWeightAttr[8] = {
"a_weight0123.x",
"a_weight0123.y",
"a_weight0123.z",
"a_weight0123.w",
"a_weight4567.x",
"a_weight4567.y",
"a_weight4567.z",
"a_weight4567.w",
"a_w1.x",
"a_w1.y",
"a_w1.z",
"a_w1.w",
"a_w2.x",
"a_w2.y",
"a_w2.z",
"a_w2.w",
};
enum DoLightComputation {
@ -135,8 +135,13 @@ enum DoLightComputation {
void GenerateVertexShader(int prim, char *buffer) {
char *p = buffer;
WRITE(p, "precision highp float;\n");
#define USE_FOR_LOOP
#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
@ -220,9 +225,7 @@ void GenerateVertexShader(int prim, char *buffer) {
WRITE(p, "uniform mediump 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 mediump mat4 u_bone%i;\n", i);
}
WRITE(p, "uniform mediump mat4 u_bone[%i];\n", numBones);
}
if (gstate.isLightingEnabled()) {
WRITE(p, "uniform lowp vec4 u_ambient;\n");
@ -301,19 +304,44 @@ void GenerateVertexShader(int prim, char *buffer) {
WRITE(p, " vec3 worldnormal = vec3(0.0, 0.0, 1.0);\n");
int numWeights = 1 + ((vertType & GE_VTYPE_WEIGHTCOUNT_MASK) >> GE_VTYPE_WEIGHTCOUNT_SHIFT);
static const float rescale[4] = {0, 2*127.5f/128.f, 2*32767.5f/32768.f, 2.0f};
float factor = rescale[(vertType & GE_VTYPE_WEIGHT_MASK) >> GE_VTYPE_WEIGHT_SHIFT];
#ifdef USE_FOR_LOOP
// To loop through the weights, we unfortunately need to put them in a float array.
// GLSL ES sucks - no way to directly initialize an array!
switch (numWeights) {
case 1: WRITE(p, " float w[1]; w[0] = a_w1;\n"); break;
case 2: WRITE(p, " float w[2]; w[0] = a_w1.x; w[1] = a_w1.y;\n"); break;
case 3: WRITE(p, " float w[3]; w[0] = a_w1.x; w[1] = a_w1.y; w[2] = a_w1.z;\n"); break;
case 4: WRITE(p, " float w[4]; w[0] = a_w1.x; w[1] = a_w1.y; w[2] = a_w1.z; w[3] = a_w1.w;\n"); break;
case 5: WRITE(p, " float w[5]; w[0] = a_w1.x; w[1] = a_w1.y; w[2] = a_w1.z; w[3] = a_w1.w; w[4] = a_w2;\n"); break;
case 6: WRITE(p, " float w[6]; w[0] = a_w1.x; w[1] = a_w1.y; w[2] = a_w1.z; w[3] = a_w1.w; w[4] = a_w2.x; w[5] = a_w2.y;\n"); break;
case 7: WRITE(p, " float w[7]; w[0] = a_w1.x; w[1] = a_w1.y; w[2] = a_w1.z; w[3] = a_w1.w; w[4] = a_w2.x; w[5] = a_w2.y; w[6] = a_w2.z;\n"); break;
case 8: WRITE(p, " float w[8]; w[0] = a_w1.x; w[1] = a_w1.y; w[2] = a_w1.z; w[3] = a_w1.w; w[4] = a_w2.x; w[5] = a_w2.y; w[6] = a_w2.z; w[7] = a_w2.w;\n"); break;
}
WRITE(p, " for (int i = 0; i < %i; i++) {\n", numWeights);
WRITE(p, " worldpos += w[i] * (u_bone[i] * vec4(a_position.xyz, 1.0)).xyz;\n");
if (hasNormal)
WRITE(p, " worldnormal += w[i] * (u_bone[i] * vec4(a_normal, 0.0)).xyz;\n");
WRITE(p, " }\n");
#else
for (int i = 0; i < numWeights; i++) {
const char *weightAttr = boneWeightAttr[i];
// workaround for "cant do .x of scalar" issue
if (numWeights == 1 && i == 0) weightAttr = "a_weight0123";
if (numWeights == 5 && i == 4) weightAttr = "a_weight4567";
WRITE(p, " worldpos += %s * (u_bone%i * vec4(a_position.xyz, 1.0)).xyz;\n", weightAttr, i);
if (numWeights == 1 && i == 0) weightAttr = "a_w1";
if (numWeights == 5 && i == 4) weightAttr = "a_w2";
WRITE(p, " worldpos += %s * (u_bone[%i] * vec4(a_position.xyz, 1.0)).xyz;\n", weightAttr, i);
if (hasNormal)
WRITE(p, " worldnormal += %s * (u_bone%i * vec4(a_normal, 0.0)).xyz;\n", weightAttr, i);
WRITE(p, " worldnormal += %s * (u_bone[%i] * vec4(a_normal, 0.0)).xyz;\n", weightAttr, i);
}
#endif
// Finally, multiply by world matrix (yes, we have to).
static const float rescale[4] = {0, 2*127.5f/128.f, 2*32767.5f/32768.f, 2.0f};
float factor = rescale[(vertType & GE_VTYPE_WEIGHT_MASK) >> GE_VTYPE_WEIGHT_SHIFT];
// Finally, multiply by world matrix and compensate for quantization at the same time.
WRITE(p, " worldpos = (u_world * vec4(worldpos * %f, 1.0)).xyz;\n", factor);
if (hasNormal)
WRITE(p, " worldnormal = (u_world * vec4(worldnormal, 0.0)).xyz;\n");