[llvm] [VectorCombine][AMDGPU] Narrow Phi of Shuffles. (PR #140188)

[Simon Pilgrim via llvm-commits]

================
@@ -0,0 +1,6459 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -passes=vector-combine -S %s | FileCheck %s --check-prefixes=CHECK
+; RUN: opt -mtriple=aarch64 -passes=vector-combine -S %s | FileCheck %s --check-prefixes=CHECK-AARCH64
+; RUN: opt -mtriple=amdgcn-amd-amdhsa -passes=vector-combine -S %s | FileCheck %s --check-prefixes=CHECK-AMDGPU
+; RUN: opt -mtriple=arm -passes=vector-combine -S %s | FileCheck %s --check-prefixes=CHECK-ARM
+; RUN: opt -mtriple=hexagon-- -passes=vector-combine -S %s | FileCheck %s --check-prefixes=CHECK-HEXAGON
+; RUN: opt -mtriple=riscv64 -passes=vector-combine -S %s | FileCheck %s --check-prefixes=CHECK-RISCV
+; RUN: opt -mtriple=x86_64-- -mcpu=x86-64 -passes=vector-combine -S %s | FileCheck %s --check-prefixes=CHECK-X86-V1
+; RUN: opt -mtriple=x86_64-- -mcpu=x86-64-v2 -passes=vector-combine -S %s | FileCheck %s --check-prefixes=CHECK-X86-V2
+; RUN: opt -mtriple=x86_64-- -mcpu=x86-64-v3 -passes=vector-combine -S %s | FileCheck %s --check-prefixes=CHECK-X86-V3
+; RUN: opt -mtriple=x86_64-- -mcpu=x86-64-v4 -passes=vector-combine -S %s | FileCheck %s --check-prefixes=CHECK-X86-V4
----------------
RKSimon wrote:

You can't add target specific tests in the VectorCombine test root folder as they will may fail on builds that don't have those targets - that's why most tests are in the target subfolders as VectorCombine is mainly cost driven

https://github.com/llvm/llvm-project/pull/140188