[mlir][GPU] Implement ValueBoundsOpInterface for GPU ID operations #122190
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,19 @@ | ||
| //===- ValueBoundsOpInterfaceImpl.h - Impl. of ValueBoundsOpInterface -----===// | ||
| // | ||
| // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. | ||
| // See https://llvm.org/LICENSE.txt for license information. | ||
| // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception | ||
| // | ||
| //===----------------------------------------------------------------------===// | ||
|
|
||
| #ifndef MLIR_DIALECT_GPU_IR_VALUEBOUNDSOPINTERFACEIMPL_H | ||
| #define MLIR_DIALECT_GPU_IR_VALUEBOUNDSOPINTERFACEIMPL_H | ||
|
|
||
| namespace mlir { | ||
| class DialectRegistry; | ||
|
|
||
| namespace gpu { | ||
| void registerValueBoundsOpInterfaceExternalModels(DialectRegistry ®istry); | ||
| } // namespace gpu | ||
| } // namespace mlir | ||
| #endif // MLIR_DIALECT_GPU_IR_VALUEBOUNDSOPINTERFACEIMPL_H |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,114 @@ | ||
| //===- ValueBoundsOpInterfaceImpl.cpp - Impl. of ValueBoundsOpInterface ---===// | ||
| // | ||
| // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. | ||
| // See https://llvm.org/LICENSE.txt for license information. | ||
| // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception | ||
| // | ||
| //===----------------------------------------------------------------------===// | ||
|
|
||
| #include "mlir/Dialect/GPU/IR/ValueBoundsOpInterfaceImpl.h" | ||
|
|
||
| #include "mlir/Dialect/GPU/IR/GPUDialect.h" | ||
| #include "mlir/Interfaces/InferIntRangeInterface.h" | ||
| #include "mlir/Interfaces/ValueBoundsOpInterface.h" | ||
|
|
||
| using namespace mlir; | ||
| using namespace mlir::gpu; | ||
|
|
||
| namespace { | ||
| /// Implement ValueBoundsOpInterface (which only works on index-typed values, | ||
| /// gathers a set of constraint expressions, and is used for affine analyses) | ||
| /// in terms of InferIntRangeInterface (which works | ||
| /// on arbitrary integer types, creates [min, max] ranges, and is used in for | ||
| /// arithmetic simplification). | ||
| template <typename Op> | ||
| struct GpuIdOpInterface | ||
| : public ValueBoundsOpInterface::ExternalModel<GpuIdOpInterface<Op>, Op> { | ||
| void populateBoundsForIndexValue(Operation *op, Value value, | ||
| ValueBoundsConstraintSet &cstr) const { | ||
| auto inferrable = cast<InferIntRangeInterface>(op); | ||
| assert(value == op->getResult(0) && | ||
| "inferring for value that isn't the GPU op's result"); | ||
| auto translateConstraint = [&](Value v, const ConstantIntRanges &range) { | ||
| assert(v == value && | ||
| "GPU ID op inferring values for something that's not its result"); | ||
| cstr.bound(v) >= range.smin().getSExtValue(); | ||
| cstr.bound(v) <= range.smax().getSExtValue(); | ||
| }; | ||
| // No arguments, so we don't need to pass in their ranges. | ||
| inferrable.inferResultRanges({}, translateConstraint); | ||
| } | ||
| }; | ||
|
|
||
| struct GpuLaunchOpInterface | ||
| : public ValueBoundsOpInterface::ExternalModel<GpuLaunchOpInterface, | ||
| LaunchOp> { | ||
| void populateBoundsForIndexValue(Operation *op, Value value, | ||
| ValueBoundsConstraintSet &cstr) const { | ||
| auto launchOp = cast<LaunchOp>(op); | ||
|
|
||
| Value sizeArg = nullptr; | ||
| bool isSize = false; | ||
| KernelDim3 gridSizeArgs = launchOp.getGridSizeOperandValues(); | ||
| KernelDim3 blockSizeArgs = launchOp.getBlockSizeOperandValues(); | ||
|
|
||
| auto match = [&](KernelDim3 bodyArgs, KernelDim3 externalArgs, | ||
| bool areSizeArgs) { | ||
| if (value == bodyArgs.x) { | ||
| sizeArg = externalArgs.x; | ||
| isSize = areSizeArgs; | ||
| } | ||
| if (value == bodyArgs.y) { | ||
| sizeArg = externalArgs.y; | ||
| isSize = areSizeArgs; | ||
| } | ||
| if (value == bodyArgs.z) { | ||
| sizeArg = externalArgs.z; | ||
| isSize = areSizeArgs; | ||
| } | ||
| }; | ||
| match(launchOp.getThreadIds(), blockSizeArgs, false); | ||
| match(launchOp.getBlockSize(), blockSizeArgs, true); | ||
| match(launchOp.getBlockIds(), gridSizeArgs, false); | ||
| match(launchOp.getGridSize(), gridSizeArgs, true); | ||
| if (launchOp.hasClusterSize()) { | ||
| KernelDim3 clusterSizeArgs = *launchOp.getClusterSizeOperandValues(); | ||
| match(*launchOp.getClusterIds(), clusterSizeArgs, false); | ||
| match(*launchOp.getClusterSize(), clusterSizeArgs, true); | ||
| } | ||
|
|
||
| if (!sizeArg) | ||
| return; | ||
| if (isSize) { | ||
| cstr.bound(value) == cstr.getExpr(sizeArg); | ||
| cstr.bound(value) >= 1; | ||
| } else { | ||
| cstr.bound(value) < cstr.getExpr(sizeArg); | ||
| cstr.bound(value) >= 0; | ||
| } | ||
| } | ||
| }; | ||
| } // namespace | ||
|
|
||
| void mlir::gpu::registerValueBoundsOpInterfaceExternalModels( | ||
| DialectRegistry ®istry) { | ||
| registry.addExtension(+[](MLIRContext *ctx, GPUDialect *dialect) { | ||
| #define REGISTER(X) X::attachInterface<GpuIdOpInterface<X>>(*ctx); | ||
| REGISTER(ClusterDimOp) | ||
| REGISTER(ClusterDimBlocksOp) | ||
| REGISTER(ClusterIdOp) | ||
| REGISTER(ClusterBlockIdOp) | ||
| REGISTER(BlockDimOp) | ||
| REGISTER(BlockIdOp) | ||
| REGISTER(GridDimOp) | ||
| REGISTER(ThreadIdOp) | ||
| REGISTER(LaneIdOp) | ||
| REGISTER(SubgroupIdOp) | ||
| REGISTER(GlobalIdOp) | ||
| REGISTER(NumSubgroupsOp) | ||
| REGISTER(SubgroupSizeOp) | ||
| #undef REGISTER | ||
|
|
||
| LaunchOp::attachInterface<GpuLaunchOpInterface>(*ctx); | ||
| }); | ||
| } | ||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,150 @@ | ||
| // RUN: mlir-opt %s -pass-pipeline='builtin.module( \ | ||
| // RUN: func.func(test-affine-reify-value-bounds), \ | ||
| // RUN: gpu.module(llvm.func(test-affine-reify-value-bounds)), \ | ||
| // RUN: gpu.module(gpu.func(test-affine-reify-value-bounds)))' \ | ||
| // RUN: -verify-diagnostics \ | ||
| // RUN: -split-input-file | FileCheck %s | ||
|
|
||
| // CHECK-LABEL: func @launch_func | ||
| func.func @launch_func(%arg0 : index) { | ||
| %c0 = arith.constant 0 : index | ||
| %c1 = arith.constant 1 : index | ||
| %c2 = arith.constant 2 : index | ||
| %c4 = arith.constant 4 : index | ||
| %c64 = arith.constant 64 : index | ||
| gpu.launch blocks(%block_id_x, %block_id_y, %block_id_z) in (%grid_dim_x = %arg0, %grid_dim_y = %c4, %grid_dim_z = %c2) | ||
| threads(%thread_id_x, %thread_id_y, %thread_id_z) in (%block_dim_x = %c64, %block_dim_y = %c4, %block_dim_z = %c2) { | ||
|
|
||
| // Sanity checks: | ||
| // expected-error @below{{unknown}} | ||
| "test.compare" (%thread_id_x, %c1) {cmp = "EQ"} : (index, index) -> () | ||
| // expected-remark @below{{false}} | ||
| "test.compare" (%thread_id_x, %c64) {cmp = "GE"} : (index, index) -> () | ||
|
|
||
| // expected-remark @below{{true}} | ||
| "test.compare" (%grid_dim_x, %c1) {cmp = "GE"} : (index, index) -> () | ||
| // expected-remark @below{{true}} | ||
| "test.compare" (%grid_dim_x, %arg0) {cmp = "EQ"} : (index, index) -> () | ||
| // expected-remark @below{{true}} | ||
| "test.compare" (%grid_dim_y, %c4) {cmp = "EQ"} : (index, index) -> () | ||
| // expected-remark @below{{true}} | ||
| "test.compare" (%grid_dim_z, %c2) {cmp = "EQ"} : (index, index) -> () | ||
|
|
||
| // expected-remark @below{{true}} | ||
| "test.compare"(%block_id_x, %c0) {cmp = "GE"} : (index, index) -> () | ||
| // expected-remark @below{{true}} | ||
| "test.compare"(%block_id_x, %arg0) {cmp = "LT"} : (index, index) -> () | ||
| // expected-remark @below{{true}} | ||
| "test.compare"(%block_id_y, %c0) {cmp = "GE"} : (index, index) -> () | ||
| // expected-remark @below{{true}} | ||
| "test.compare"(%block_id_y, %c4) {cmp = "LT"} : (index, index) -> () | ||
| // expected-remark @below{{true}} | ||
| "test.compare"(%block_id_z, %c0) {cmp = "GE"} : (index, index) -> () | ||
| // expected-remark @below{{true}} | ||
| "test.compare"(%block_id_z, %c2) {cmp = "LT"} : (index, index) -> () | ||
|
|
||
| // expected-remark @below{{true}} | ||
| "test.compare" (%block_dim_x, %c64) {cmp = "EQ"} : (index, index) -> () | ||
| // expected-remark @below{{true}} | ||
| "test.compare" (%block_dim_y, %c4) {cmp = "EQ"} : (index, index) -> () | ||
| // expected-remark @below{{true}} | ||
| "test.compare" (%block_dim_z, %c2) {cmp = "EQ"} : (index, index) -> () | ||
|
|
||
| // expected-remark @below{{true}} | ||
| "test.compare"(%thread_id_x, %c0) {cmp = "GE"} : (index, index) -> () | ||
| // expected-remark @below{{true}} | ||
| "test.compare"(%thread_id_x, %c64) {cmp = "LT"} : (index, index) -> () | ||
| // expected-remark @below{{true}} | ||
| "test.compare"(%thread_id_y, %c0) {cmp = "GE"} : (index, index) -> () | ||
| // expected-remark @below{{true}} | ||
| "test.compare"(%thread_id_y, %c4) {cmp = "LT"} : (index, index) -> () | ||
| // expected-remark @below{{true}} | ||
| "test.compare"(%thread_id_z, %c0) {cmp = "GE"} : (index, index) -> () | ||
| // expected-remark @below{{true}} | ||
| "test.compare"(%thread_id_z, %c2) {cmp = "LT"} : (index, index) -> () | ||
| gpu.terminator | ||
|
There was a problem hiding this comment. Can you add another check for |
||
| } | ||
|
|
||
| func.return | ||
| } | ||
|
|
||
| // ----- | ||
|
|
||
| // The tests for what the ranges are are located in int-range-interface.mlir, | ||
| // so here we just make sure that the results of that interface propagate into | ||
| // constraints. | ||
|
|
||
| // CHECK-LABEL: func @kernel | ||
| module attributes {gpu.container_module} { | ||
| gpu.module @gpu_module { | ||
| llvm.func @kernel() attributes {gpu.kernel} { | ||
|
|
||
| %c0 = arith.constant 0 : index | ||
| %ctid_max = arith.constant 4294967295 : index | ||
| %thread_id_x = gpu.thread_id x | ||
|
|
||
| // expected-remark @below{{true}} | ||
| "test.compare" (%thread_id_x, %c0) {cmp = "GE"} : (index, index) -> () | ||
| // expected-remark @below{{true}} | ||
| "test.compare" (%thread_id_x, %ctid_max) {cmp = "LT"} : (index, index) -> () | ||
| llvm.return | ||
| } | ||
| } | ||
| } | ||
|
|
||
| // ----- | ||
|
|
||
| // CHECK-LABEL: func @annotated_kernel | ||
| module attributes {gpu.container_module} { | ||
| gpu.module @gpu_module { | ||
| gpu.func @annotated_kernel() kernel | ||
| attributes {known_block_size = array<i32: 8, 12, 16>, | ||
| known_grid_size = array<i32: 20, 24, 28>} { | ||
|
|
||
| %c0 = arith.constant 0 : index | ||
| %c8 = arith.constant 8 : index | ||
| %thread_id_x = gpu.thread_id x | ||
|
|
||
| // expected-remark @below{{true}} | ||
| "test.compare"(%thread_id_x, %c0) {cmp = "GE"} : (index, index) -> () | ||
| // expected-remark @below{{true}} | ||
| "test.compare"(%thread_id_x, %c8) {cmp = "LT"} : (index, index) -> () | ||
|
|
||
| %block_dim_x = gpu.block_dim x | ||
| // expected-remark @below{{true}} | ||
| "test.compare"(%block_dim_x, %c8) {cmp = "EQ"} : (index, index) -> () | ||
|
|
||
| gpu.return | ||
| } | ||
| } | ||
| } | ||
|
|
||
| // ----- | ||
|
|
||
| // CHECK-LABEL: func @local_bounds_kernel | ||
| module attributes {gpu.container_module} { | ||
| gpu.module @gpu_module { | ||
| gpu.func @local_bounds_kernel() kernel { | ||
|
|
||
| %c0 = arith.constant 0 : index | ||
| %c1 = arith.constant 1 : index | ||
| %c8 = arith.constant 8 : index | ||
|
|
||
| %block_dim_x = gpu.block_dim x upper_bound 8 | ||
| // expected-remark @below{{true}} | ||
| "test.compare"(%block_dim_x, %c1) {cmp = "GE"} : (index, index) -> () | ||
| // expected-remark @below{{true}} | ||
| "test.compare"(%block_dim_x, %c8) {cmp = "LE"} : (index, index) -> () | ||
| // expected-error @below{{unknown}} | ||
| "test.compare"(%block_dim_x, %c8) {cmp = "EQ"} : (index, index) -> () | ||
|
|
||
| %thread_id_x = gpu.thread_id x upper_bound 8 | ||
| // expected-remark @below{{true}} | ||
| "test.compare"(%thread_id_x, %c0) {cmp = "GE"} : (index, index) -> () | ||
| // expected-remark @below{{true}} | ||
| "test.compare"(%thread_id_x, %c8) {cmp = "LT"} : (index, index) -> () | ||
|
|
||
| gpu.return | ||
| } | ||
| } | ||
| } | ||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Would it make sense to add an
assert(op->getNumOperands() == 0 && "expected op with no operands");?