Differential D78803 Diff 260233 mlir/include/mlir/Dialect/GPU/GPUOps.td

Changeset View

Standalone View

mlir/include/mlir/Dialect/GPU/GPUOps.td

Show First 20 Lines • Show All 254 Lines • ▼ Show 20 Lines	def GPU_LaunchFuncOp : GPU_Op<"launch_func">,
let summary = "Launches a function as a GPU kerneel";		let summary = "Launches a function as a GPU kerneel";

let description = [{		let description = [{
Launch a kernel function on the specified grid of thread blocks.		Launch a kernel function on the specified grid of thread blocks.
`gpu.launch` operations are lowered to `gpu.launch_func` operations by		`gpu.launch` operations are lowered to `gpu.launch_func` operations by
outlining the kernel body into a function in a dedicated module, which		outlining the kernel body into a function in a dedicated module, which
reflects the separate compilation process. The kernel function is required		reflects the separate compilation process. The kernel function is required
to have the `gpu.kernel` attribute. The module containing the kernel		to have the `gpu.kernel` attribute. The module containing the kernel
function is required to have the `gpu.kernel_module` attribute and must be		function is required to be a gpu.module. And finally, the module containing
named. And finally, the module containing the kernel module (which thus		the kernel module (which thus cannot be the top-level module) is required
cannot be the top-level module) is required to have the		to have the `gpu.container_module` attribute. The `gpu.launch_func`
`gpu.container_module` attribute. The `gpu.launch_func` operation has a		operation has a symbol attribute named `kernel` to identify the fully specified
		frgossenUnsubmitted Done Reply Inline Actions The kernel is now identified with a single nested symbol attribute `kernel` (https://github.com/llvm/llvm-project/commit/0372db05bb1552c2b39fc735f949977e0a863a25). I should have updated this, sorry. frgossen: The kernel is now identified with a single nested symbol attribute `kernel` (https://github.
string attribute named `kernel` to specify the name of the kernel function		kernel function to launch (both the gpu.module and func).
to launch and an attribute named `kernel_module` to specify the name of the
module containing that kernel function.

The operation takes at least six operands, with the first three operands		The operation takes at least six operands, with the first three operands
being grid sizes along x,y,z dimensions and the following three being block		being grid sizes along x,y,z dimensions and the following three being block
sizes along x,y,z dimensions. When a lower-dimensional kernel is required,		sizes along x,y,z dimensions. When a lower-dimensional kernel is required,
unused sizes must be explicitly set to `1`. The remaining operands are		unused sizes must be explicitly set to `1`. The remaining operands are
passed as arguments to the kernel function.		passed as arguments to the kernel function.

A custom syntax for this operation is currently not available.		A custom syntax for this operation is currently not available.

Example:		Example:

```mlir		```mlir
module attributes {gpu.container_module} {		module attributes {gpu.container_module} {

// This module creates a separate compilation unit for the GPU compiler.		// This module creates a separate compilation unit for the GPU compiler.
module @kernels attributes {gpu.kernel_module} {		gpu.module @kernels {
func @kernel_1(%arg0 : f32, %arg1 : !llvm<"float*">)		func @kernel_1(%arg0 : f32, %arg1 : !llvm<"float*">)
attributes { nvvm.kernel = true } {		attributes { nvvm.kernel = true } {

// Operations that produce block/thread IDs and dimensions are		// Operations that produce block/thread IDs and dimensions are
// injected when outlining the `gpu.launch` body to a function called		// injected when outlining the `gpu.launch` body to a function called
// by `gpu.launch_func`.		// by `gpu.launch_func`.
%tIdX = "gpu.thread_id"() {dimension = "x"} : () -> (index)		%tIdX = "gpu.thread_id"() {dimension = "x"} : () -> (index)
%tIdY = "gpu.thread_id"() {dimension = "y"} : () -> (index)		%tIdY = "gpu.thread_id"() {dimension = "y"} : () -> (index)
▲ Show 20 Lines • Show All 402 Lines • Show Last 20 Lines