[llvm] [X86] Always use 128-bit V_SET0/AVX512_128_SET0 patterns, along with SUBREG_TO_REG for extension to 256/512-bit vectors (PR #212950)

[Craig Topper via llvm-commits]

topperc wrote:

@alinas try this


```
diff --git a/llvm/lib/Target/X86/X86InstrInfo.cpp b/llvm/lib/Target/X86/X86InstrInfo.cpp
index 05ee6eeabef0..c5bb86df427a 100644
--- a/llvm/lib/Target/X86/X86InstrInfo.cpp
+++ b/llvm/lib/Target/X86/X86InstrInfo.cpp
@@ -8606,7 +8606,9 @@ X86InstrInfo::foldMemoryOperandImpl(MachineFunction &MF, MachineInstr &MI,
       break;
     case X86::AVX512_512_SETALLONES:
       IsAllOnes = true;
-      [[fallthrough]];
+      Ty = FixedVectorType::get(Type::getInt32Ty(MF.getFunction().getContext()),
+                                16);
+      break;
     case X86::AVX1_SETALLONES:
     case X86::AVX2_SETALLONES:
     case X86::AVX512_256_SETALLONES:
```

https://github.com/llvm/llvm-project/pull/212950