[Mlir-commits] [mlir] 3556114 - [mlir][gpu] Allow gpu.launch_func to be async.
Christian Sigg
llvmlistbot at llvm.org
Thu Oct 29 13:48:46 PDT 2020
Author: Christian Sigg
Date: 2020-10-29T21:48:38+01:00
New Revision: 355611408384b87faaf3e5b54c2cf0938baa15be
URL: https://github.com/llvm/llvm-project/commit/355611408384b87faaf3e5b54c2cf0938baa15be
DIFF: https://github.com/llvm/llvm-project/commit/355611408384b87faaf3e5b54c2cf0938baa15be.diff
LOG: [mlir][gpu] Allow gpu.launch_func to be async.
This is a roll-forward of rGec7780ebdab4, now that the remaining
gpu.launch_func have been converted to custom form in rGb22f111023ba.
Reviewed By: antiagainst
Differential Revision: https://reviews.llvm.org/D90420
Added:
Modified:
mlir/include/mlir/Dialect/GPU/GPUOps.td
mlir/lib/Dialect/GPU/IR/GPUDialect.cpp
mlir/test/Dialect/GPU/invalid.mlir
mlir/test/Dialect/GPU/ops.mlir
Removed:
################################################################################
diff --git a/mlir/include/mlir/Dialect/GPU/GPUOps.td b/mlir/include/mlir/Dialect/GPU/GPUOps.td
index 8ad3c9abe6cc..593b735f01c9 100644
--- a/mlir/include/mlir/Dialect/GPU/GPUOps.td
+++ b/mlir/include/mlir/Dialect/GPU/GPUOps.td
@@ -291,12 +291,14 @@ def GPU_GPUFuncOp : GPU_Op<"func", [HasParent<"GPUModuleOp">,
let parser = [{ return parseGPUFuncOp(parser, result); }];
}
-def GPU_LaunchFuncOp : GPU_Op<"launch_func">,
- Arguments<(ins SymbolRefAttr:$kernel,
+def GPU_LaunchFuncOp : GPU_Op<"launch_func",
+ [GPU_AsyncOpInterface, AttrSizedOperandSegments]>,
+ Arguments<(ins Variadic<GPU_AsyncToken>:$asyncDependencies,
+ SymbolRefAttr:$kernel,
Index:$gridSizeX, Index:$gridSizeY, Index:$gridSizeZ,
Index:$blockSizeX, Index:$blockSizeY, Index:$blockSizeZ,
Variadic<AnyType>:$operands)>,
- Results<(outs)> {
+ Results<(outs Optional<GPU_AsyncToken>:$asyncToken)> {
let summary = "Launches a function as a GPU kernel";
let description = [{
@@ -308,14 +310,22 @@ def GPU_LaunchFuncOp : GPU_Op<"launch_func">,
function is required to be a gpu.module. And finally, the module containing
the kernel module (which thus cannot be the top-level module) is required
to have the `gpu.container_module` attribute. The `gpu.launch_func`
- operation has a symbol attribute named `kernel` to identify the fully
+ operation has a symbol attribute named `kernel` to identify the fully
specified kernel function to launch (both the gpu.module and func).
- 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
- 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
- passed as arguments to the kernel function.
+ The `gpu.launch_func` supports async dependencies: the kernel does not start
+ executing until the ops producing those async dependencies have completed.
+
+ By the default, the host implicitly blocks until kernel execution has
+ completed. If the `async` keyword is present, the host does not block but
+ instead a `!gpu.async.token` is returned. Other async GPU ops can take this
+ token as dependency.
+
+ The operation requires at least the grid and block sizes along the x,y,z
+ dimensions as arguments. When a lower-dimensional kernel is required,
+ unused sizes must be explicitly set to `1`.
+
+ The remaining operands are passed as arguments to the kernel function.
Example:
@@ -351,11 +361,15 @@ def GPU_LaunchFuncOp : GPU_Op<"launch_func">,
}
}
+ %t0 = gpu.wait async
gpu.launch_func
- @kernels::@kernel_1 // Kernel function.
- blocks in (%cst, %cst, %cst) // Grid size.
- threads in (%cst, %cst, %cst) // Block size.
- args(%arg0 : f32, %arg1 : memref<?xf32, 1>) // Kernel arguments.
+ async // (Optional) Don't block host, return token.
+ [%t0] // (Optional) Execute only after %t0 has completed.
+ @kernels::@kernel_1 // Kernel function.
+ blocks in (%cst, %cst, %cst) // Grid size.
+ threads in (%cst, %cst, %cst) // Block size.
+ args(%arg0 : f32, // (Optional) Kernel arguments.
+ %arg1 : memref<?xf32, 1>)
}
```
}];
@@ -401,6 +415,7 @@ def GPU_LaunchFuncOp : GPU_Op<"launch_func">,
let verifier = [{ return ::verify(*this); }];
let assemblyFormat = [{
+ custom<AsyncDependencies>(type($asyncToken), $asyncDependencies)
$kernel
`blocks` `in` ` ` `(`$gridSizeX`,` $gridSizeY`,` $gridSizeZ`)`
`threads` `in` ` ` `(`$blockSizeX`,` $blockSizeY`,` $blockSizeZ`)`
diff --git a/mlir/lib/Dialect/GPU/IR/GPUDialect.cpp b/mlir/lib/Dialect/GPU/IR/GPUDialect.cpp
index 9514d3e0a04b..8b0f7a3829d2 100644
--- a/mlir/lib/Dialect/GPU/IR/GPUDialect.cpp
+++ b/mlir/lib/Dialect/GPU/IR/GPUDialect.cpp
@@ -438,10 +438,15 @@ void LaunchFuncOp::build(OpBuilder &builder, OperationState &result,
auto kernelSymbol = builder.getSymbolRefAttr(
kernelModule.getName(), {builder.getSymbolRefAttr(kernelFunc.getName())});
result.addAttribute(getKernelAttrName(), kernelSymbol);
+ SmallVector<int32_t, 8> segmentSizes(8, 1);
+ segmentSizes.front() = 0; // Initially no async dependencies.
+ segmentSizes.back() = static_cast<int32_t>(kernelOperands.size());
+ result.addAttribute(getOperandSegmentSizeAttr(),
+ builder.getI32VectorAttr(segmentSizes));
}
unsigned LaunchFuncOp::getNumKernelOperands() {
- return getNumOperands() - kNumConfigOperands;
+ return getNumOperands() - asyncDependencies().size() - kNumConfigOperands;
}
StringRef LaunchFuncOp::getKernelModuleName() {
@@ -451,15 +456,17 @@ StringRef LaunchFuncOp::getKernelModuleName() {
StringRef LaunchFuncOp::getKernelName() { return kernel().getLeafReference(); }
Value LaunchFuncOp::getKernelOperand(unsigned i) {
- return getOperation()->getOperand(i + kNumConfigOperands);
+ return getOperand(asyncDependencies().size() + kNumConfigOperands + i);
}
KernelDim3 LaunchFuncOp::getGridSizeOperandValues() {
- return KernelDim3{getOperand(0), getOperand(1), getOperand(2)};
+ auto operands = getOperands().drop_front(asyncDependencies().size());
+ return KernelDim3{operands[0], operands[1], operands[2]};
}
KernelDim3 LaunchFuncOp::getBlockSizeOperandValues() {
- return KernelDim3{getOperand(3), getOperand(4), getOperand(5)};
+ auto operands = getOperands().drop_front(asyncDependencies().size());
+ return KernelDim3{operands[3], operands[4], operands[5]};
}
static LogicalResult verify(LaunchFuncOp op) {
diff --git a/mlir/test/Dialect/GPU/invalid.mlir b/mlir/test/Dialect/GPU/invalid.mlir
index 3612b8e0dcc1..3dc5be405aac 100644
--- a/mlir/test/Dialect/GPU/invalid.mlir
+++ b/mlir/test/Dialect/GPU/invalid.mlir
@@ -37,6 +37,7 @@ func @launch_requires_gpu_return(%sz : index) {
func @launch_func_too_few_operands(%sz : index) {
// expected-error at +1 {{expected 6 or more operands}}
"gpu.launch_func"(%sz, %sz, %sz, %sz, %sz)
+ {operand_segment_sizes = dense<[0, 1, 1, 1, 1, 1, 0, 0]> : vector<8xi32>}
: (index, index, index, index, index) -> ()
return
}
@@ -55,6 +56,7 @@ module attributes {gpu.container_module} {
func @launch_func_missing_callee_attribute(%sz : index) {
// expected-error at +1 {{'gpu.launch_func' op requires attribute 'kernel'}}
"gpu.launch_func"(%sz, %sz, %sz, %sz, %sz, %sz)
+ {operand_segment_sizes = dense<[0, 1, 1, 1, 1, 1, 1, 0]> : vector<8xi32>}
: (index, index, index, index, index, index) -> ()
return
}
diff --git a/mlir/test/Dialect/GPU/ops.mlir b/mlir/test/Dialect/GPU/ops.mlir
index e81b233abfbc..a3b781afdfbc 100644
--- a/mlir/test/Dialect/GPU/ops.mlir
+++ b/mlir/test/Dialect/GPU/ops.mlir
@@ -73,6 +73,7 @@ module attributes {gpu.container_module} {
%1 = "op"() : () -> (memref<?xf32, 1>)
// CHECK: %{{.*}} = constant 8
%cst = constant 8 : index
+ %t0 = gpu.wait async
// CHECK: gpu.launch_func @kernels::@kernel_1 blocks in (%{{.*}}, %{{.*}}, %{{.*}}) threads in (%{{.*}}, %{{.*}}, %{{.*}}) args(%{{.*}} : f32, %{{.*}} : memref<?xf32, 1>)
gpu.launch_func @kernels::@kernel_1 blocks in (%cst, %cst, %cst) threads in (%cst, %cst, %cst) args(%0 : f32, %1 : memref<?xf32, 1>)
@@ -80,6 +81,9 @@ module attributes {gpu.container_module} {
// CHECK: gpu.launch_func @kernels::@kernel_2 blocks in (%{{.*}}, %{{.*}}, %{{.*}}) threads in (%{{.*}}, %{{.*}}, %{{.*}})
gpu.launch_func @kernels::@kernel_2 blocks in (%cst, %cst, %cst) threads in (%cst, %cst, %cst)
+ // CHECK: %{{.*}} = gpu.launch_func async [%{{.*}}] @kernels::@kernel_2 blocks in (%{{.*}}, %{{.*}}, %{{.*}}) threads in (%{{.*}}, %{{.*}}, %{{.*}})
+ %t1 = gpu.launch_func async [%t0] @kernels::@kernel_2 blocks in (%cst, %cst, %cst) threads in (%cst, %cst, %cst)
+
return
}
More information about the Mlir-commits
mailing list