OpenGL FS/VS Shader Program Compiling Flow
由于GLSL的跨平台性，所以OpenGL没有编译器，由显卡驱动来完成着色器的编译工作
Related API
GLuint fs = glCreateShader(GL_FRAGMENT_SHADER);
glShaderSource(fs, 1, fs_source, NULL);
glCompileShader(fs);
| OpenGL API | gl3w | |
|---|---|---|
| glCreateShader | gl3wCreateShader | _mesa_CreateShader |
| glShaderSource | gl3wShaderSource | _mesa_ShaderSource |
| glCompileShader | gl3wCompileShader | _mesa_CompileShader |
Compiling flow
LLVM IR -> MVP Assemble
Source code
/home/icube/mesa3d/src/gallium/drivers/mvp/mvp_shader_lib
Understanding OpenGL Shader Compiling Flow
GLSL330内置类型分布
Vertex Shader
|sys in(in stack)|attr in(13 slot)|attr out(18 slot)|
|:-- |:-- |:-- |
|int gl_VertexID <br>int gl_InstanceID | vec4 gl_Color <br>vec4 gl_SecondaryColor <br>vec4 gl_Normal <br>vec4 gl_Vertex <br>vec4 gl_MultiTexCoord0 <br>vec4 gl_MultiTexCoord1 <br>vec4 gl_MultiTexCoord2 <br>vec4 gl_MultiTexCoord3 <br> vec4 gl_MultiTexCoord4 <br>vec4 gl_MultiTexCoord5 <br>vec4 gl_MultiTexCoord6 <br> vec4 gl_MultiTexCoord7 <br>float gl_FogCoord |vec4 gl_Position <br>vec4 gl_FrontColor <br>vec4 gl_BackColor <br>float gl_FogFragCoord <br>float gl_PointSize <br>int gl_Layer <br> ... |
对于VS attr in，插槽不做固定分配，按照定义顺序逐一分配，直到插槽上限
Fragment Shader
|sys in(1 slot)|attr in(15 slot)|attr out(10 slot)|
|:-- |:-- |:-- |
| int gl_SampleID <br> vec2 gl_SamplePosition <br> int gl_SampleMaskIn | vec4 gl_FragCoord <br> vec4 gl_Color<br>int gl_PrimitiveID <br>bool gl_FrontFacing <br> vec2 gl_PointCoord <br> float gl_FogFragCoord <br>int gl_Layer | float gl_FragDepth <br> vec4 gl_FragColor <br> vec4 gl_FragData[gl_MaxDrawBuffers] ... |
下面是shader编译过程中用到的一些自定义库函数，这些函数在shader执行之前帮助完成了相关的数据准备工作。
VS
shader.vs
#version 330 core
layout (location = 0) in vec3 aPos;
layout (location = 1) in vec3 aColor;
out vec3 ourColor;
void main()
{
gl_Position = vec4(aPos, 1.0);
ourColor = aColor;
}
#define CORE_REG_TID 0xc2600c84
#define CORE_REG_CMD 0xc2600c88
#define CORE_REG_VTX_DONE 0xc2601000
#define IC1000_IO_BASE 0xC2600000
// attribute count + draw_start index + 8 * (attribute stride + addr)
#define ATTRIBUTE_SIZE ((1 + 1 + 8 * 2) * 4)
// save misc info address /* attributes addr + vs_constant_buffer addr + ps_constant_buffer addr */
#define IC1000_CORE_REGS_GP (1024*3 + 0x30 + IC1000_IO_BASE)
#define MVP_LM_BASE 0xC2000000
#define ATTR_FORMAT_VEC1 0
#define ATTR_FORMAT_VEC2 1
#define ATTR_FORMAT_VEC3 2
#define ATTR_FORMAT_VEC4 3
extern void vs_main(float* gl_in, float* gl_out, float* gl_const, unsigned int index, unsigned int instance_id);
void calc_in_out(float** gl_in, float** gl_out, unsigned int* vtxIndex, unsigned int *instanceId)
{
unsigned int lm_base = MVP_LM_BASE;
unsigned int lm_size = 1024;
unsigned int tid = *( unsigned int *)(CORE_REG_TID);
unsigned int cmd = *( unsigned int *)(CORE_REG_CMD);
unsigned int bid = (cmd >> 28) & 1;
unsigned int varying_num = 8;
*gl_in = *gl_out = (float *)(lm_base + (tid * lm_size) + bid * varying_num * 16);
*vtxIndex = *(unsigned int *)(*gl_in);
*instanceId = *((unsigned int *)(*gl_in) + 4);
}
void load_attributes(float* gl_in, unsigned int index)
{
unsigned int* misc_info = (unsigned int *)(*( unsigned int*)(IC1000_CORE_REGS_GP));
unsigned int* attr_header = (unsigned int *)(*misc_info);
unsigned int attr_count = attr_header[0];
index = attr_header[1] + index;
unsigned int attr_stride = 0;
unsigned int attr_format = ATTR_FORMAT_VEC4;
unsigned int attr_vbo_base = 0;
float *attr_data_base;
unsigned int i = 0;
for(i = 0; i < attr_count; i++){
attr_stride = attr_header[ 2 + i*2] & 0x0000ffff;
attr_format = (attr_header[ 2 + i*2] >> 16) & 0x0000ffff;
attr_vbo_base = attr_header[ 2 + i*2 + 1];
attr_data_base = (float *)(attr_vbo_base + index * attr_stride);
//here just load (4 * float) bytes for one attribute, compiler/mesa will handle the case that attibute is not fvec4 type
if (attr_format == ATTR_FORMAT_VEC1){
*(gl_in + i * 4 + 0) = *( attr_data_base + 0);
*(gl_in + i * 4 + 1) = 0;
*(gl_in + i * 4 + 2) = 0;
*(gl_in + i * 4 + 3) = 1;
}else if(attr_format == ATTR_FORMAT_VEC2){
*(gl_in + i * 4 + 0) = *( attr_data_base + 0);
*(gl_in + i * 4 + 1) = *( attr_data_base + 1);
*(gl_in + i * 4 + 2) = 0;
*(gl_in + i * 4 + 3) = 1;
}else if(attr_format == ATTR_FORMAT_VEC3){
*(gl_in + i * 4 + 0) = *( attr_data_base + 0);
*(gl_in + i * 4 + 1) = *( attr_data_base + 1);
*(gl_in + i * 4 + 2) = *( attr_data_base + 2);
*(gl_in + i * 4 + 3) = 1;
}else { // ATTR_FORMAT_VEC4
*(gl_in + i * 4 + 0) = *( attr_data_base + 0);
*(gl_in + i * 4 + 1) = *( attr_data_base + 1);
*(gl_in + i * 4 + 2) = *( attr_data_base + 2);
*(gl_in + i * 4 + 3) = *( attr_data_base + 3);
}
}
}
void wrapper_vs(void)
{
float *gl_in, *gl_out;
unsigned int index = 0;
unsigned int instance_id = 0;
unsigned int *misc_info = (unsigned int *)(*( unsigned int *)(IC1000_CORE_REGS_GP));
unsigned int gl_const = *(misc_info + 1);
calc_in_out(&gl_in, &gl_out, &index, &instance_id);
load_attributes(gl_in, index);
vs_main(gl_in, gl_out, (float *)gl_const, index, instance_id);
*(unsigned int *)(CORE_REG_VTX_DONE) = index;
}
FS
shader.fs
#version 330 core
out vec4 FragColor;
in vec3 ourColor;
void main()
{
FragColor = vec4(ourColor, 1.0f);
}
#define CORE_REG_TID 0xc2600c84
#define CORE_REG_CMD 0xc2600c88
#define CORE_REG_VTX_DONE 0xc2601000
#define MVP_LM_BASE 0xC2000000
#define IC1000_IO_BASE 0xC2600000
// attribute count + draw_start index + 8 * (attribute stride + addr)
#define ATTRIBUTE_SIZE ((1 + 1 + 8 * 2) * 4)
// save misc info address /* attributes + vs_constant_buffer addr + ps_constant_buffer addr */
#define IC1000_CORE_REGS_GP (1024*3 + 0x30 + IC1000_IO_BASE)
extern void ps_main(float* gl_in, float* gl_out, float* gl_const);
void calc_in_out(float** gl_in, float** gl_out, unsigned int* vtxIndex)
{
unsigned int lm_base = MVP_LM_BASE;
unsigned int lm_size = 1024;
unsigned int tid = *(volatile unsigned int*)(CORE_REG_TID);
unsigned int cmd = *(volatile unsigned int*)(CORE_REG_CMD);
unsigned int bid = (cmd >> 28) & 1;
unsigned int varying_num = 8;
*gl_in = *gl_out = (float*)(lm_base + (tid*lm_size) + bid*varying_num*16);
}
void wrapper_ps(void)
{
float *gl_in, *gl_out;
unsigned int* misc_info = (unsigned int *)(*( unsigned int*)(IC1000_CORE_REGS_GP));
unsigned int gl_const = *(misc_info + 2);
calc_in_out(&gl_in, &gl_out, (void*)0);
ps_main(gl_in, gl_out, (float*)gl_const);
}
mvp_llvm_create_func() mvp_create_function() create_function() mvp_compile_tgsi_main() mvp_compile_tgsi_shader() mvp_shader_create() mvp_build_shader_variant() mvp_build_shader_variant_low_priority()