| 1 | //===- AMDGPUUnifyDivergentExitNodes.h ------------------------------------===// |
| 2 | // |
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | // |
| 9 | // This is a variant of the UnifyFunctionExitNodes pass. Rather than ensuring |
| 10 | // there is at most one ret and one unreachable instruction, it ensures there is |
| 11 | // at most one divergent exiting block. |
| 12 | // |
| 13 | // StructurizeCFG can't deal with multi-exit regions formed by branches to |
| 14 | // multiple return nodes. It is not desirable to structurize regions with |
| 15 | // uniform branches, so unifying those to the same return block as divergent |
| 16 | // branches inhibits use of scalar branching. It still can't deal with the case |
| 17 | // where one branch goes to return, and one unreachable. Replace unreachable in |
| 18 | // this case with a return. |
| 19 | // |
| 20 | //===----------------------------------------------------------------------===// |
| 21 | |
| 22 | #ifndef LLVM_LIB_TARGET_AMDGPU_AMDGPUUNIFYDIVERGENTEXITNODES_H |
| 23 | #define LLVM_LIB_TARGET_AMDGPU_AMDGPUUNIFYDIVERGENTEXITNODES_H |
| 24 | |
| 25 | #include "AMDGPU.h" |
| 26 | |
| 27 | namespace llvm { |
| 28 | class AMDGPUUnifyDivergentExitNodesPass |
| 29 | : public PassInfoMixin<AMDGPUUnifyDivergentExitNodesPass> { |
| 30 | public: |
| 31 | PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM); |
| 32 | }; |
| 33 | |
| 34 | } // end namespace llvm |
| 35 | |
| 36 | #endif // LLVM_LIB_TARGET_AMDGPU_AMDGPUUNIFYDIVERGENTEXITNODES_H |
| 37 | |