kernel - Module

`enum KernelArg`

Source: gpu/kernel.joule:9

`fn from_buffer<T: Copy>(buffer: &GpuBuffer<T>) -> Self`

Source: gpu/kernel.joule:31

`struct LaunchConfig`

Source: gpu/kernel.joule:38

`fn new_1d(grid_size: u32, block_size: u32) -> Self`

Create 1D launch config

Source: gpu/kernel.joule:49

`fn new_2d(grid: [u32; 2], block: [u32; 2]) -> Self`

Create 2D launch config

Source: gpu/kernel.joule:58

`fn new_3d(grid: [u32; 3], block: [u32; 3]) -> Self`

Create 3D launch config

Source: gpu/kernel.joule:67

`fn with_shared_mem(mut self, size: usize) -> Self`

Set shared memory size

Source: gpu/kernel.joule:76

`fn for_elements(total: usize, block_size: u32) -> Self`

Calculate optimal config for total work items

Source: gpu/kernel.joule:82

`fn for_image(width: u32, height: u32, block_x: u32, block_y: u32) -> Self`

Calculate optimal 2D config for image dimensions

Source: gpu/kernel.joule:88

`fn total_threads(&self) -> u64`

Total number of threads

Source: gpu/kernel.joule:95

`struct GpuKernel`

Compiled GPU kernel

Source: gpu/kernel.joule:102

`enum KernelHandle`

Internal kernel handle (backend-specific)

Source: gpu/kernel.joule:109

`fn from_source(device: &GpuDevice, source: &str, name: &str) -> Result<Self, GpuError>`

Source: gpu/kernel.joule:136

`fn from_binary(device: &GpuDevice, binary: &[u8], name: &str) -> Result<Self, GpuError>`

Source: gpu/kernel.joule:165

`fn device(&self) -> &GpuDevice`

Get the device this kernel is compiled for

Source: gpu/kernel.joule:193

`fn name(&self) -> &str`

Get kernel name

Source: gpu/kernel.joule:198

`fn launch(&self, config: &LaunchConfig, args: &[KernelArg]) -> Result<(), GpuError>`

Source: gpu/kernel.joule:204

`fn launch_async(`

Source: gpu/kernel.joule:232

`fn max_threads_per_block(&self) -> u32`

Get maximum threads per block for this kernel

Source: gpu/kernel.joule:261

`fn cuFuncGetAttribute(`

Source: gpu/kernel.joule:266

`fn shared_mem_usage(&self) -> usize`

Get shared memory usage for this kernel

Source: gpu/kernel.joule:285

`fn cuFuncGetAttribute(`

Source: gpu/kernel.joule:290

`fn register_usage(&self) -> u32`

Get register usage for this kernel

Source: gpu/kernel.joule:309

`fn cuFuncGetAttribute(`

Source: gpu/kernel.joule:314

`fn drop(&mut self)`

Source: gpu/kernel.joule:334

`fn cuModuleUnload(module: *mut std::ffi::c_void) -> i32;`

Source: gpu/kernel.joule:339

`fn hipModuleUnload(module: *mut std::ffi::c_void) -> i32;`

Source: gpu/kernel.joule:352

`fn compile_cuda_kernel(source: &str, name: &str) -> Result<KernelHandle, GpuError>`

Source: gpu/kernel.joule:364

`fn nvrtcCreateProgram(`

Source: gpu/kernel.joule:366

`fn nvrtcCompileProgram(`

Source: gpu/kernel.joule:374

`fn nvrtcGetPTXSize(prog: mut std::ffi::c_void, size: mut usize) -> i32;`

Source: gpu/kernel.joule:379

`fn nvrtcGetPTX(prog: mut std::ffi::c_void, ptx: mut i8) -> i32;`

Source: gpu/kernel.joule:380

`fn nvrtcDestroyProgram(prog: mut mut std::ffi::c_void) -> i32;`

Source: gpu/kernel.joule:381

`fn cuModuleLoadData(module: mut mut std::ffi::c_void, data: *const i8) -> i32;`

Source: gpu/kernel.joule:382

`fn cuModuleGetFunction(`

Source: gpu/kernel.joule:383

`fn load_cuda_kernel(binary: &[u8], name: &str) -> Result<KernelHandle, GpuError>`

Source: gpu/kernel.joule:443

`fn cuModuleLoadData(module: mut mut std::ffi::c_void, data: *const u8) -> i32;`

Source: gpu/kernel.joule:445

`fn cuModuleGetFunction(`

Source: gpu/kernel.joule:446

`fn cuda_launch(`

Source: gpu/kernel.joule:471

`fn cuLaunchKernel(`

Source: gpu/kernel.joule:477

`fn cuda_launch_async(`

Source: gpu/kernel.joule:548

`fn cuLaunchKernel(`

Source: gpu/kernel.joule:555

`fn compile_metal_kernel(`

Source: gpu/kernel.joule:642

`fn mtl_compile_library(`

Source: gpu/kernel.joule:648

`fn mtl_library_new_function(`

Source: gpu/kernel.joule:655

`fn mtl_device_new_compute_pipeline(`

Source: gpu/kernel.joule:661

`fn load_metal_kernel(`

Source: gpu/kernel.joule:698

`fn mtl_load_library(`

Source: gpu/kernel.joule:704

`fn mtl_library_new_function(`

Source: gpu/kernel.joule:711

`fn mtl_device_new_compute_pipeline(`

Source: gpu/kernel.joule:717

`fn metal_launch(`

Source: gpu/kernel.joule:754

`fn mtl_device_new_command_queue(`

Source: gpu/kernel.joule:761

`fn mtl_command_queue_command_buffer(`

Source: gpu/kernel.joule:764

`fn mtl_command_buffer_compute_command_encoder(`

Source: gpu/kernel.joule:767

`fn mtl_compute_encoder_set_pipeline(`

Source: gpu/kernel.joule:770

`fn mtl_compute_encoder_set_buffer(`

Source: gpu/kernel.joule:774

`fn mtl_compute_encoder_set_bytes(`

Source: gpu/kernel.joule:780

`fn mtl_compute_encoder_set_threadgroup_memory_length(`

Source: gpu/kernel.joule:786

`fn mtl_compute_encoder_dispatch_threadgroups(`

Source: gpu/kernel.joule:791

`fn mtl_compute_encoder_end_encoding(encoder: *mut std::ffi::c_void);`

Source: gpu/kernel.joule:796

`fn mtl_command_buffer_commit(buffer: *mut std::ffi::c_void);`

Source: gpu/kernel.joule:797

`fn mtl_command_buffer_wait_until_completed(buffer: *mut std::ffi::c_void);`

Source: gpu/kernel.joule:798

`fn metal_launch_async(`

Source: gpu/kernel.joule:931

`fn mtl_command_queue_command_buffer(`

Source: gpu/kernel.joule:939

`fn mtl_command_buffer_compute_command_encoder(`

Source: gpu/kernel.joule:942

`fn mtl_compute_encoder_set_pipeline(`

Source: gpu/kernel.joule:945

`fn mtl_compute_encoder_set_buffer(`

Source: gpu/kernel.joule:949

`fn mtl_compute_encoder_set_bytes(`

Source: gpu/kernel.joule:955

`fn mtl_compute_encoder_set_threadgroup_memory_length(`

Source: gpu/kernel.joule:961

`fn mtl_compute_encoder_dispatch_threadgroups(`

Source: gpu/kernel.joule:966

`fn mtl_compute_encoder_end_encoding(encoder: *mut std::ffi::c_void);`

Source: gpu/kernel.joule:971

`fn mtl_command_buffer_commit(buffer: *mut std::ffi::c_void);`

Source: gpu/kernel.joule:972

`fn compile_rocm_kernel(source: &str, name: &str) -> Result<KernelHandle, GpuError>`

Source: gpu/kernel.joule:1099

`fn hiprtcCreateProgram(`

Source: gpu/kernel.joule:1101

`fn hiprtcCompileProgram(`

Source: gpu/kernel.joule:1109

`fn hiprtcGetCodeSize(prog: mut std::ffi::c_void, size: mut usize) -> i32;`

Source: gpu/kernel.joule:1114

`fn hiprtcGetCode(prog: mut std::ffi::c_void, code: mut i8) -> i32;`

Source: gpu/kernel.joule:1115

`fn hiprtcDestroyProgram(prog: mut mut std::ffi::c_void) -> i32;`

Source: gpu/kernel.joule:1116

`fn hipModuleLoadData(module: mut mut std::ffi::c_void, data: *const i8) -> i32;`

Source: gpu/kernel.joule:1117

`fn hipModuleGetFunction(`

Source: gpu/kernel.joule:1118

`fn load_rocm_kernel(binary: &[u8], name: &str) -> Result<KernelHandle, GpuError>`

Source: gpu/kernel.joule:1174

`fn hipModuleLoadData(module: mut mut std::ffi::c_void, data: *const u8) -> i32;`

Source: gpu/kernel.joule:1176

`fn hipModuleGetFunction(`

Source: gpu/kernel.joule:1177

`fn rocm_launch(`

Source: gpu/kernel.joule:1202

`fn rocm_launch_async(`

Source: gpu/kernel.joule:1211

`fn rocm_launch_on_stream(`

Source: gpu/kernel.joule:1222

`fn hipModuleLaunchKernel(`

Source: gpu/kernel.joule:1229

`fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result`

Source: gpu/kernel.joule:1315

enum KernelArg

fn from_buffer<T: Copy>(buffer: &GpuBuffer<T>) -> Self

struct LaunchConfig

fn new_1d(grid_size: u32, block_size: u32) -> Self

fn new_2d(grid: [u32; 2], block: [u32; 2]) -> Self

fn new_3d(grid: [u32; 3], block: [u32; 3]) -> Self

fn with_shared_mem(mut self, size: usize) -> Self

fn for_elements(total: usize, block_size: u32) -> Self

fn for_image(width: u32, height: u32, block_x: u32, block_y: u32) -> Self

fn total_threads(&self) -> u64

struct GpuKernel

enum KernelHandle

fn from_source(device: &GpuDevice, source: &str, name: &str) -> Result<Self, GpuError>

fn from_binary(device: &GpuDevice, binary: &[u8], name: &str) -> Result<Self, GpuError>

fn device(&self) -> &GpuDevice

fn name(&self) -> &str

fn launch(&self, config: &LaunchConfig, args: &[KernelArg]) -> Result<(), GpuError>

fn launch_async(

fn max_threads_per_block(&self) -> u32

fn cuFuncGetAttribute(

fn shared_mem_usage(&self) -> usize

fn cuFuncGetAttribute(

fn register_usage(&self) -> u32

fn cuFuncGetAttribute(

fn drop(&mut self)

fn cuModuleUnload(module: *mut std::ffi::c_void) -> i32;

fn hipModuleUnload(module: *mut std::ffi::c_void) -> i32;

fn compile_cuda_kernel(source: &str, name: &str) -> Result<KernelHandle, GpuError>

fn nvrtcCreateProgram(

fn nvrtcCompileProgram(

fn nvrtcGetPTXSize(prog: *mut std::ffi::c_void, size: *mut usize) -> i32;

fn nvrtcGetPTX(prog: *mut std::ffi::c_void, ptx: *mut i8) -> i32;

fn nvrtcDestroyProgram(prog: *mut *mut std::ffi::c_void) -> i32;

fn cuModuleLoadData(module: *mut *mut std::ffi::c_void, data: *const i8) -> i32;

fn cuModuleGetFunction(

fn load_cuda_kernel(binary: &[u8], name: &str) -> Result<KernelHandle, GpuError>

fn cuModuleLoadData(module: *mut *mut std::ffi::c_void, data: *const u8) -> i32;

fn cuModuleGetFunction(

fn cuda_launch(

fn cuLaunchKernel(

fn cuda_launch_async(

fn cuLaunchKernel(

fn compile_metal_kernel(

fn mtl_compile_library(

fn mtl_library_new_function(

fn mtl_device_new_compute_pipeline(

fn load_metal_kernel(

fn mtl_load_library(

fn mtl_library_new_function(

fn mtl_device_new_compute_pipeline(

fn metal_launch(

fn mtl_device_new_command_queue(

fn mtl_command_queue_command_buffer(

fn mtl_command_buffer_compute_command_encoder(

fn mtl_compute_encoder_set_pipeline(

fn mtl_compute_encoder_set_buffer(

fn mtl_compute_encoder_set_bytes(

fn mtl_compute_encoder_set_threadgroup_memory_length(

fn mtl_compute_encoder_dispatch_threadgroups(

fn mtl_compute_encoder_end_encoding(encoder: *mut std::ffi::c_void);

fn mtl_command_buffer_commit(buffer: *mut std::ffi::c_void);

fn mtl_command_buffer_wait_until_completed(buffer: *mut std::ffi::c_void);

fn metal_launch_async(

fn mtl_command_queue_command_buffer(

fn mtl_command_buffer_compute_command_encoder(

fn mtl_compute_encoder_set_pipeline(

fn mtl_compute_encoder_set_buffer(

fn mtl_compute_encoder_set_bytes(

fn mtl_compute_encoder_set_threadgroup_memory_length(

fn mtl_compute_encoder_dispatch_threadgroups(

fn mtl_compute_encoder_end_encoding(encoder: *mut std::ffi::c_void);

fn mtl_command_buffer_commit(buffer: *mut std::ffi::c_void);

fn compile_rocm_kernel(source: &str, name: &str) -> Result<KernelHandle, GpuError>

fn hiprtcCreateProgram(

fn hiprtcCompileProgram(

fn hiprtcGetCodeSize(prog: *mut std::ffi::c_void, size: *mut usize) -> i32;

fn hiprtcGetCode(prog: *mut std::ffi::c_void, code: *mut i8) -> i32;

fn hiprtcDestroyProgram(prog: *mut *mut std::ffi::c_void) -> i32;

fn hipModuleLoadData(module: *mut *mut std::ffi::c_void, data: *const i8) -> i32;

fn hipModuleGetFunction(

`enum KernelArg`

`fn from_buffer<T: Copy>(buffer: &GpuBuffer<T>) -> Self`

`struct LaunchConfig`

`fn new_1d(grid_size: u32, block_size: u32) -> Self`

`fn new_2d(grid: [u32; 2], block: [u32; 2]) -> Self`

`fn new_3d(grid: [u32; 3], block: [u32; 3]) -> Self`

`fn with_shared_mem(mut self, size: usize) -> Self`

`fn for_elements(total: usize, block_size: u32) -> Self`

`fn for_image(width: u32, height: u32, block_x: u32, block_y: u32) -> Self`

`fn total_threads(&self) -> u64`

`struct GpuKernel`

`enum KernelHandle`

`fn from_source(device: &GpuDevice, source: &str, name: &str) -> Result<Self, GpuError>`

`fn from_binary(device: &GpuDevice, binary: &[u8], name: &str) -> Result<Self, GpuError>`

`fn device(&self) -> &GpuDevice`

`fn name(&self) -> &str`

`fn launch(&self, config: &LaunchConfig, args: &[KernelArg]) -> Result<(), GpuError>`

`fn launch_async(`

`fn max_threads_per_block(&self) -> u32`

`fn cuFuncGetAttribute(`

`fn shared_mem_usage(&self) -> usize`

`fn cuFuncGetAttribute(`

`fn register_usage(&self) -> u32`

`fn cuFuncGetAttribute(`

`fn drop(&mut self)`

`fn cuModuleUnload(module: *mut std::ffi::c_void) -> i32;`

`fn hipModuleUnload(module: *mut std::ffi::c_void) -> i32;`

`fn compile_cuda_kernel(source: &str, name: &str) -> Result<KernelHandle, GpuError>`

`fn nvrtcCreateProgram(`

`fn nvrtcCompileProgram(`

`fn nvrtcGetPTXSize(prog: mut std::ffi::c_void, size: mut usize) -> i32;`

`fn nvrtcGetPTX(prog: mut std::ffi::c_void, ptx: mut i8) -> i32;`

`fn nvrtcDestroyProgram(prog: mut mut std::ffi::c_void) -> i32;`

`fn cuModuleLoadData(module: mut mut std::ffi::c_void, data: *const i8) -> i32;`

`fn cuModuleGetFunction(`

`fn load_cuda_kernel(binary: &[u8], name: &str) -> Result<KernelHandle, GpuError>`

`fn cuModuleLoadData(module: mut mut std::ffi::c_void, data: *const u8) -> i32;`

`fn cuModuleGetFunction(`

`fn cuda_launch(`

`fn cuLaunchKernel(`

`fn cuda_launch_async(`

`fn cuLaunchKernel(`

`fn compile_metal_kernel(`

`fn mtl_compile_library(`

`fn mtl_library_new_function(`

`fn mtl_device_new_compute_pipeline(`

`fn load_metal_kernel(`

`fn mtl_load_library(`

`fn mtl_library_new_function(`

`fn mtl_device_new_compute_pipeline(`

`fn metal_launch(`

`fn mtl_device_new_command_queue(`

`fn mtl_command_queue_command_buffer(`

`fn mtl_command_buffer_compute_command_encoder(`

`fn mtl_compute_encoder_set_pipeline(`

`fn mtl_compute_encoder_set_buffer(`

`fn mtl_compute_encoder_set_bytes(`

`fn mtl_compute_encoder_set_threadgroup_memory_length(`

`fn mtl_compute_encoder_dispatch_threadgroups(`

`fn mtl_compute_encoder_end_encoding(encoder: *mut std::ffi::c_void);`

`fn mtl_command_buffer_commit(buffer: *mut std::ffi::c_void);`

`fn mtl_command_buffer_wait_until_completed(buffer: *mut std::ffi::c_void);`

`fn metal_launch_async(`

`fn mtl_command_queue_command_buffer(`

`fn mtl_command_buffer_compute_command_encoder(`

`fn mtl_compute_encoder_set_pipeline(`

`fn mtl_compute_encoder_set_buffer(`

`fn mtl_compute_encoder_set_bytes(`

`fn mtl_compute_encoder_set_threadgroup_memory_length(`

`fn mtl_compute_encoder_dispatch_threadgroups(`

`fn mtl_compute_encoder_end_encoding(encoder: *mut std::ffi::c_void);`

`fn mtl_command_buffer_commit(buffer: *mut std::ffi::c_void);`

`fn compile_rocm_kernel(source: &str, name: &str) -> Result<KernelHandle, GpuError>`

`fn hiprtcCreateProgram(`

`fn hiprtcCompileProgram(`

`fn hiprtcGetCodeSize(prog: mut std::ffi::c_void, size: mut usize) -> i32;`

`fn hiprtcGetCode(prog: mut std::ffi::c_void, code: mut i8) -> i32;`

`fn hiprtcDestroyProgram(prog: mut mut std::ffi::c_void) -> i32;`

`fn hipModuleLoadData(module: mut mut std::ffi::c_void, data: *const i8) -> i32;`

`fn hipModuleGetFunction(`