GLSL Trace

Features

shader debugging
shader profiling (using extensions EXT_shader_realtime_clock and ARB_shader_clock)
supports mesh and ray tracing shaders (Vulkan only)

Dependencies

glslang - required
SPIRV-Cross - required for OpenGL
GLFW - for tests
GLEW - for tests

How to use

Setup:

(OpenGL) Create shader and shader program for reqular rendering.
Use glslang to parse GLSL or HLSL source code and build AST
(Vulkan) Convert glslang AST to SPIRV and create pipeline for reqular rendering.
Create ShaderTrace object to store debug information
Get glslang AST TProgram::getIntermediate(EShLanguage stage)
Insert trace recording ShaderTrace::InsertTraceRecording(TIntermediate &intermediate, uint32_t setIndex) or ShaderTrace::InsertFunctionProfiler(TIntermediate &intermediate, uint32_t setIndex, bool shaderSubgroupClock, bool shaderDeviceClock), where:
shaderSubgroupClock - requires OpenGL extension GL_ARB_shader_clock, requires Vulkan extension VK_KHR_shader_clock and VkPhysicalDeviceShaderClockFeaturesKHR::shaderSubgroupClock must be true.
shaderDeviceClock - requires OpenGL extension GL_EXT_shader_realtime_clock, requires Vulkan extension VK_KHR_shader_clock and VkPhysicalDeviceShaderClockFeaturesKHR::shaderDeviceClock must be true.
descSetIndex - descriptor set index that will be reserved for storage buffer (Vulkan only)
(Vulkan) Convert AST with trace recording to SPIRV and create pipeline for debugging.
(OpenGL) Convert AST to SPIRV, use SPIRV-Cross to get GLSL code and create shader program for debugging.

To debug draw or compute or ray tracing invocation:

(Vulkan) bind pipeline that contains shader with inserted trace recording
(Vulkan) bind descriptor set with storage buffer to index descSetIndex
(OpenGL) bind shader program that contains shader with inserted trace recording
(OpenGL) get shader storage block index of dbg_ShaderTraceStorageand bind storage buffer to that index.
after invocation map storage buffer and pass pointer to ShaderTrace::ParseShaderTrace (const void *ptr, uint64_t maxSize, vector<string> &result)

How its works

Debugging:

AST processing: log all function results, log all function calls, log some operator results.
Result parsing: write modified values, write unmodified values that are used, write line number and line from source code.

Pixel/invocation profiling:

AST processing: insert time measurement into user-defined functions.
Result parsing: calculate average time and a fraction of the total shader execution time.

Frame profiling:

AST processing: insert time measurement into entry function.
Result: sum of shader invocation time per pixel or tile.

All:

Shader trace recorded to the storage buffer as forward list using atomic operations.
Each ShaderTrace object has unique number to determine different shaders when parsing results.
Allowed debugging on different shader stages in single draw/dispatch call.
Many invocations of same or different shaders can be recoreded into one storage buffer.

Debugging

Enable pixel/invocation debugging from code (OpenGL)

// use glslang to compile shader from source
// for full source code see 'CreateShader()' in 'tests/OpenGL/ShaderCompiler.cpp'
glslang::TProgram  program;
...

// after program compilation get AST
TIntermediate*  intermediate = program.getIntermediate( ... );

// insert trace recording into glslang AST
ShaderTrace  shaderTrace;
shaderTrace.InsertTraceRecording( *intermediate, /*unused*/0 );

// Parts of shader source code will be inserted into shader trace results
shaderTrace.SetSource( ... );

// add included source files if used '#include' directive
shaderTrace.IncludeSource( ... );

// convert AST to SPIRV binary
glslang::GlslangToSpv( *intermediate, spirvData, ... );

// use SPIRV-Cross to convert SPIRV binary into GLSL code
spirv_cross::CompilerGLSL  compiler {spirvData.data(), spirvData.size()};
string  shaderSrc = compiler.compile();

// create shaders and program
...

// create buffer for shader output
uint32_t bufferSize = 4u << 20;
glGenBuffers( 1, &dbgBuffer );
glBindBuffer( GL_SHADER_STORAGE_BUFFER, dbgBuffer );
glBufferStorage( GL_SHADER_STORAGE_BUFFER, bufferSize, nullptr, GL_MAP_READ_BIT | GL_DYNAMIC_STORAGE_BIT );

// bind buffer to shader
GLuint	sb_index = glGetProgramResourceIndex( program, GL_SHADER_STORAGE_BLOCK, "dbg_ShaderTraceStorage" );
glShaderStorageBlockBinding( program, sb_index, /*binding*/0 );
glBindBufferBase( GL_SHADER_STORAGE_BUFFER, /*binding*/0, dbgBuffer );
		
// clear buffer
uint32_t  zero = 0;
glClearBufferData( GL_SHADER_STORAGE_BUFFER, GL_R32UI, GL_RED_INTEGER, GL_UNSIGNED_INT, &zero );

// set pixel which you need to debug (2 components)
// record if {pixel_x, pixel_y} == floor(gl_FragCoord.xy)
uint32_t  data[] = { pixel_x, pixel_y };
glBufferSubData( GL_SHADER_STORAGE_BUFFER, 0, sizeof(data), data );

// draw
...

// ... or which compute invocation or ray tracing launch (3 components)
// record if {thread_x, thread_y, thread_z} == gl_GlobalInvocationID
// record if {thread_x, thread_y, thread_z} == gl_LaunchID
uint32_t  data[] = { thread_x, thread_y, thread_z };
glBufferSubData( GL_SHADER_STORAGE_BUFFER, 0, sizeof(data), data );

// dispatch or trace
...

// read buffer data
void* ptr = glMapBuffer( GL_SHADER_STORAGE_BUFFER, GL_READ_ONLY );

// get shader trace as a string
vector<string>  result;
shaderTrace.ParseShaderTrace( ptr, bufferSize, result );

Enable pixel/invocation debugging from code (Vulkan)

// use glslang to compile shader from source
// for full source code see 'Device::_Compile()' in 'tests/Vulkan/Device.cpp' 
glslang::TProgram  program;
...

// after program compilation get AST
TIntermediate*  intermediate = program.getIntermediate( ... );

// insert trace recording into glslang AST
// descSetIndex - any free descriptor set index
ShaderTrace  shaderTrace;
shaderTrace.InsertTraceRecording( *intermediate, descSetIndex );

// Parts of shader source code will be inserted into shader trace results
shaderTrace.SetSource( ... );

// add included source files if used '#include' directive
shaderTrace.IncludeSource( ... );

// convert AST to SPIRV binary
glslang::GlslangToSpv( *intermediate, spirvData, ... );
	
// create pipeline
...

// create buffer for shader output
VkBufferCreateInfo	info = {};
info.sType  = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
info.size   = 4u << 20;
info.usage  = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT;
vkCreateBuffer( device, &info, nullptr, &debugOutputBuffer );

info.usage  = VK_BUFFER_USAGE_TRANSFER_DST_BIT;
vkCreateBuffer( device, &info, nullptr, &stagingBuffer );

// bind descriptor set with 'debugOutputBuffer'
vkCmdBindDescriptorSets( cmdBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, ppln_layout, descSetIndex, 1, &dbg_desc_set, 0, nullptr );
	
// set pixel which you need to debug (2 components)
// record if {pixel_x, pixel_y} == floor(gl_FragCoord.xy)
uint32_t  data[] = { pixel_x, pixel_y };
vkCmdUpdateBuffer( cmdBuffer, debugOutputBuffer, 0, sizeof(data), data );
vkCmdFillBuffer( cmdBuffer, debugOutputBuffer, sizeof(data), VK_WHOLE_SIZE, 0 );

// draw
...

// ... or which compute invocation or ray tracing launch (3 components)
// record if {thread_x, thread_y, thread_z} == gl_GlobalInvocationID
// record if {thread_x, thread_y, thread_z} == gl_LaunchID
uint32_t  data[] = { thread_x, thread_y, thread_z };
vkCmdUpdateBuffer( cmdBuffer, debugOutputBuffer, 0, sizeof(data), data );
vkCmdFillBuffer( cmdBuffer, debugOutputBuffer, sizeof(data), VK_WHOLE_SIZE, 0 );

// dispatch or trace
...

// copy buffer data to staging buffer and map
...
void* ptr;
vkMapMemory( device, stagingBuffer, 0, info.size, 0, &ptr );

// get shader trace as a string
vector<string>  result;
shaderTrace.ParseShaderTrace( ptr, info.size, result );

Enable debugging from shader source

// empty function will be replaced during shader compilation
void dbg_EnableTraceRecording (bool b) {}

void main ()
{
    bool condition = ...
        
    // if 'condition' is true then trace recording will start here
    dbg_EnableTraceRecording( condition );
    ...
}

Pause trace recording

// empty functions will be replaced during shader compilation
void dbg_EnableTraceRecording (bool b) {}
void dbg_PauseTraceRecording (bool b) {}

void main ()
{
    bool condition = ...
    
    // if 'condition' is true then trace recording will start here
    dbg_EnableTraceRecording( condition );
    ...
	
	// pause
	dbg_PauseTraceRecording( true );
	
	// trace will not be recorded
	...
	
	// resume
	dbg_PauseTraceRecording( false );
	...
}

Example of shader trace

//> gl_GlobalInvocationID: uint3 {8, 8, 0}
//> gl_LocalInvocationID: uint3 {0, 0, 0}
//> gl_WorkGroupID: uint3 {1, 1, 0}
no source

//> index: uint {136}
//  gl_GlobalInvocationID: uint3 {8, 8, 0}
11. index = gl_GlobalInvocationID.x + gl_GlobalInvocationID.y * gl_NumWorkGroups.x * gl_WorkGroupSize.x;

//> size: uint {256}
12. size = gl_NumWorkGroups.x * gl_NumWorkGroups.y * gl_WorkGroupSize.x * gl_WorkGroupSize.y;

//> value: float {0.506611}
//  index: uint {136}
//  size: uint {256}
13. value = sin( float(index) / size );

//> imageStore(): void
//  gl_GlobalInvocationID: uint3 {8, 8, 0}
//  value: float {0.506611}
14.     imageStore( un_OutImage, ivec2(gl_GlobalInvocationID.xy), vec4(value) );

The //> symbol marks the modified variable or function result.

Pixel profiling

Enable pixel/invocation profiling from code (OpenGL)

// use glslang to compile shader from source
// for full source code see 'CreateShader()' in 'tests/OpenGL/ShaderCompiler.cpp'
glslang::TProgram  program;
...

// after program compilation get AST
TIntermediate*  intermediate = program.getIntermediate( ... );

// insert profiling code into glslang AST
ShaderTrace  shaderTrace;
shaderTrace.InsertFunctionProfiler( *intermediate, /*unused*/0, true, true ); // TODO: check extensions

// convert AST to SPIRV binary
glslang::GlslangToSpv( *intermediate, spirvData, ... );

// use SPIRV-Cross to convert SPIRV binary into GLSL code
spirv_cross::CompilerGLSL  compiler {spirvData.data(), spirvData.size()};
string  shaderSrc = compiler.compile();

// create shaders and program
...

// create buffer for shader output
uint32_t bufferSize = 4u << 20;
glGenBuffers( 1, &dbgBuffer );
glBindBuffer( GL_SHADER_STORAGE_BUFFER, dbgBuffer );
glBufferStorage( GL_SHADER_STORAGE_BUFFER, bufferSize, nullptr, GL_MAP_READ_BIT | GL_DYNAMIC_STORAGE_BIT );

// bind buffer to shader
GLuint	sb_index = glGetProgramResourceIndex( program, GL_SHADER_STORAGE_BLOCK, "dbg_ShaderTraceStorage" );
glShaderStorageBlockBinding( program, sb_index, /*binding*/0 );
glBindBufferBase( GL_SHADER_STORAGE_BUFFER, /*binding*/0, dbgBuffer );

// clear buffer
uint32_t  zero = 0;
glClearBufferData( GL_SHADER_STORAGE_BUFFER, GL_R32UI, GL_RED_INTEGER, GL_UNSIGNED_INT, &zero );

// set pixel which you need to profile (2 components)
// record if {pixel_x, pixel_y} == floor(gl_FragCoord.xy)
uint32_t  data[] = { pixel_x, pixel_y };
glBufferSubData( GL_SHADER_STORAGE_BUFFER, 0, sizeof(data), data );

// draw
...

// ... or which compute invocation or ray tracing launch (3 components)
// record if {thread_x, thread_y, thread_z} == gl_GlobalInvocationID
// record if {thread_x, thread_y, thread_z} == gl_LaunchID
uint32_t  data[] = { thread_x, thread_y, thread_z };
glBufferSubData( GL_SHADER_STORAGE_BUFFER, 0, sizeof(data), data );

// dispatch or trace
...

// read buffer data
void* ptr = glMapBuffer( GL_SHADER_STORAGE_BUFFER, GL_READ_ONLY );

// get profiling info as a string
vector<string>  result;
shaderTrace.ParseShaderTrace( ptr, bufferSize, result );

Enable pixel/invocation profiling from code (Vulkan)

// get shader clock extension features
VkPhysicalDeviceShaderClockFeaturesKHR	shaderClockFeat;
...

// use glslang to compile shader from source
// for full source code see 'Device::_Compile()' in 'tests/Vulkan/Device.cpp'  
glslang::TProgram  program;
...

// after program compilation get AST
TIntermediate*  intermediate = program.getIntermediate( ... );
	
// insert profiling code into glslang AST
// descSetIndex - any free descriptor set index
ShaderTrace  shaderTrace;
shaderTrace.InsertFunctionProfiler( *intermediate, descSetIndex, shaderClockFeat.shaderSubgroupClock, shaderClockFeat.shaderDeviceClock );

// convert AST to SPIRV binary
glslang::GlslangToSpv( *intermediate, spirvData, ... );
	
// create pipeline
...

// create buffer for shader output
VkBufferCreateInfo	info = {};
info.sType  = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
info.size   = 4u << 20;
info.usage  = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT;
vkCreateBuffer( device, &info, nullptr, &debugOutputBuffer );

info.usage  = VK_BUFFER_USAGE_TRANSFER_DST_BIT;
vkCreateBuffer( device, &info, nullptr, &stagingBuffer );

// bind descriptor set with 'debugOutputBuffer'
vkCmdBindDescriptorSets( cmdBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, ppln_layout, descSetIndex, 1, &dbg_desc_set, 0, nullptr );

// set pixel which you need to debug (2 components)
// record if {pixel_x, pixel_y} == floor(gl_FragCoord.xy)
uint32_t  data[] = { pixel_x, pixel_y };
vkCmdUpdateBuffer( cmdBuffer, debugOutputBuffer, 0, sizeof(data), data );
vkCmdFillBuffer( cmdBuffer, debugOutputBuffer, sizeof(data), VK_WHOLE_SIZE, 0 );

// draw
...

// ... or which compute invocation or ray tracing launch (3 components)
// record if {thread_x, thread_y, thread_z} == gl_GlobalInvocationID
// record if {thread_x, thread_y, thread_z} == gl_LaunchID
uint32_t  data[] = { thread_x, thread_y, thread_z };
vkCmdUpdateBuffer( cmdBuffer, debugOutputBuffer, 0, sizeof(data), data );
vkCmdFillBuffer( cmdBuffer, debugOutputBuffer, sizeof(data), VK_WHOLE_SIZE, 0 );

// dispatch or trace
...

// copy buffer data to staging buffer and map
...
void* ptr;
vkMapMemory( device, stagingBuffer, 0, info.size, 0, &ptr );

// get profiling info as a string
vector<string>  result;
shaderTrace.ParseShaderTrace( ptr, info.size, result );

Enable profiling from shader source

// empty function will be replaced during shader compilation
void dbg_EnableProfiling (bool b) {}
    
void main ()
{
    bool condition = ...
        
    // if condition is true then profiling will start here
    dbg_EnableProfiling( condition );
    ...
}

Example of shader profiling output

//> gl_GlobalInvocationID: uint3 {512, 512, 0}
//> gl_LocalInvocationID: uint3 {0, 0, 0}
//> gl_WorkGroupID: uint3 {64, 64, 0}
no source

// subgroup total: 100.00%,  avr: 100.00%,  (95108.00)
// device   total: 100.00%,  avr: 100.00%,  (2452.00)
// invocations:    1
106. void main ()

// subgroup total: 89.57%,  avr: 89.57%,  (85192.00)
// device   total: 89.56%,  avr: 89.56%,  (2196.00)
// invocations:    1
29. float FBM (in float3 coord)

// subgroup total: 84.67%,  avr: 12.10%,  (11504.57)
// device   total: 84.18%,  avr: 12.03%,  (294.86)
// invocations:    7
56. float GradientNoise (const float3 pos)

// subgroup total: 45.15%,  avr: 0.81%,  (766.86)
// device   total: 44.54%,  avr: 0.80%,  (19.50)
// invocations:    56
72. float3 DHash33 (const float3 p)

Render pass profiling

Enable render pass profiling (OpenGL)

// use glslang to compile shader from source
// for full source code see 'CreateShader()' in 'tests/OpenGL/ShaderCompiler.cpp'
glslang::TProgram  program;
...

// after program compilation get AST
TIntermediate*  intermediate = program.getIntermediate( ... );

// insert profiling code into glslang AST
ShaderTrace  shaderTrace;
shaderTrace.InsertFunctionProfiler( *intermediate, /*unused*/0, true, true ); // TODO: check extensions

// convert AST to SPIRV binary
glslang::GlslangToSpv( *intermediate, spirvData, ... );

// use SPIRV-Cross to convert SPIRV binary into GLSL code
spirv_cross::CompilerGLSL  compiler {spirvData.data(), spirvData.size()};
string  shaderSrc = compiler.compile();

// create shaders and program
...

// create buffer for shader output
// image_width, image_height - size of render target
uint32_t bufferSize = 16 + (image_width * image_height * 8);
glGenBuffers( 1, &dbgBuffer );
glBindBuffer( GL_SHADER_STORAGE_BUFFER, dbgBuffer );
glBufferStorage( GL_SHADER_STORAGE_BUFFER, bufferSize, nullptr, GL_MAP_READ_BIT | GL_DYNAMIC_STORAGE_BIT );

// bind buffer to shader
GLuint	sb_index = glGetProgramResourceIndex( program, GL_SHADER_STORAGE_BLOCK, "dbg_ShaderTraceStorage" );
glShaderStorageBlockBinding( program, sb_index, /*binding*/0 );
glBindBufferBase( GL_SHADER_STORAGE_BUFFER, /*binding*/0, dbgBuffer );

// clear buffer
uint32_t  zero = 0;
glClearBufferData( GL_SHADER_STORAGE_BUFFER, GL_R32UI, GL_RED_INTEGER, GL_UNSIGNED_INT, &zero );

// set 'scale' and 'dimension'
uint32_t  data[] = { std::bit_cast<uint32_t>(1.0f), std::bit_cast<uint32_t>(1.0f), image_width, image_height };
glBufferSubData( GL_SHADER_STORAGE_BUFFER, 0, sizeof(data), data );

// set of draw or dispatch or trace
...

// read buffer data
uint64_t* bufferData = static_cast<uint64_t*>( glMapBuffer( GL_SHADER_STORAGE_BUFFER, GL_READ_ONLY )) + 2;

for (uint32_t y = 0; y < image_height; ++y)
for (uint32_t x = 0; x < image_width; ++x)
{
	uint64_t  pixel_time = bufferData[x + y * image_width];
	
	// process results
	...
}

Enable render pass profiling (Vulkan)

// check shader clock extension
VkPhysicalDeviceShaderClockFeaturesKHR	shaderClockFeat;
...
if ( !shaderClockFeat.shaderDeviceClock )
	return; // not supported

// use glslang to compile shader from source
// for full source code see 'Device::_Compile()' in 'tests/Vulkan/Device.cpp'  
glslang::TProgram  program;
...

// after program compilation get AST
TIntermediate*  intermediate = program.getIntermediate( ... );
	
// insert profiling code into glslang AST
// descSetIndex - any free descriptor set index
ShaderTrace  shaderTrace;
shaderTrace.InsertShaderClockMap( *intermediate, descSetIndex );

// convert AST to SPIRV binary
glslang::GlslangToSpv( *intermediate, spirvData, ... );
	
// create pipeline
...

// create buffer for shader output
// image_width, image_height - size of render target
VkBufferCreateInfo	info = {};
info.sType  = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
info.size   = 16 + (image_width * image_height * 8);
info.usage  = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT;
vkCreateBuffer( device, &info, nullptr, &debugOutputBuffer );

info.usage  = VK_BUFFER_USAGE_TRANSFER_DST_BIT;
vkCreateBuffer( device, &info, nullptr, &stagingBuffer );

// bind descriptor set with 'debugOutputBuffer'
vkCmdBindDescriptorSets( cmdBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, ppln_layout, descSetIndex, 1, &dbg_desc_set, 0, nullptr );

// set 'scale' and 'dimension'
uint32_t  data[] = { std::bit_cast<uint32_t>(1.0f), std::bit_cast<uint32_t>(1.0f), image_width, image_height };
vkCmdUpdateBuffer( cmdBuffer, debugOutputBuffer, 0, sizeof(data), data );
vkCmdFillBuffer( cmdBuffer, debugOutputBuffer, sizeof(data), VK_WHOLE_SIZE, 0 );

// set of draw or dispatch or trace
...

// copy buffer data to staging buffer and map
...
void* ptr;
vkMapMemory( device, stagingBuffer, 0, info.size, 0, &ptr );
uint64_t* bufferData = static_cast<uint64_t*>( ptr ) + 2;

for (uint32_t y = 0; y < image_height; ++y)
for (uint32_t x = 0; x < image_width; ++x)
{
	uint64_t  pixel_time = bufferData[x + y * image_width];
	
	// process results
	...
}

Example of shader profiling output

Cheap and reliable Node.js hosting starts at $3/month, and $1/month static HTML hosting

azhirnov / glsl_trace

Programming Languages

Labels

Projects that are alternatives of or similar to glsl trace