[LLVM][IR] Add support for vector ConstantInt/FP to ConstandFolding:FoldBitCast.

[paulwalker-arm]

No description provided.

[llvmbot]

@llvm/pr-subscribers-llvm-analysis

@llvm/pr-subscribers-llvm-transforms

Author: Paul Walker (paulwalker-arm)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/117163.diff

2 Files Affected:

• (modified) llvm/lib/Analysis/ConstantFolding.cpp (+7-1)
• (modified) llvm/test/Transforms/InstCombine/cast.ll (+2)

diff --git a/llvm/lib/Analysis/ConstantFolding.cpp b/llvm/lib/Analysis/ConstantFolding.cpp
index 1971c28fc4c4de..f66078e0aa545d 100644
--- a/llvm/lib/Analysis/ConstantFolding.cpp
+++ b/llvm/lib/Analysis/ConstantFolding.cpp
@@ -151,8 +151,14 @@ Constant *FoldBitCast(Constant *C, Type *DestTy, const DataLayout &DL) {
     return FoldBitCast(ConstantVector::get(Ops), DestTy, DL);
   }
 
+  // Some of what follows may extend to cover scalable vectors but the current
+  // implementation is fixed length specific.
+  if (!isa<FixedVectorType>(C->getType()))
+    return ConstantExpr::getBitCast(C, DestTy);
+
   // If this is a bitcast from constant vector -> vector, fold it.
-  if (!isa<ConstantDataVector>(C) && !isa<ConstantVector>(C))
+  if (!isa<ConstantDataVector>(C) && !isa<ConstantVector>(C) &&
+      !isa<ConstantInt>(C) && !isa<ConstantFP>(C))
     return ConstantExpr::getBitCast(C, DestTy);
 
   // If the element types match, IR can fold it.
diff --git a/llvm/test/Transforms/InstCombine/cast.ll b/llvm/test/Transforms/InstCombine/cast.ll
index ca748a9483e9b2..0f957e22ad17bc 100644
--- a/llvm/test/Transforms/InstCombine/cast.ll
+++ b/llvm/test/Transforms/InstCombine/cast.ll
@@ -2,6 +2,8 @@
 ; Tests to make sure elimination of casts is working correctly
 ; RUN: opt < %s -passes=instcombine -S -data-layout="E-p:64:64:64-p1:32:32:32-p2:64:64:64-p3:64:64:64-a0:0:8-f32:32:32-f64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-v64:64:64-v128:128:128-n8:16:32:64" | FileCheck %s --check-prefixes=ALL,BE
 ; RUN: opt < %s -passes=instcombine -S -data-layout="e-p:64:64:64-p1:32:32:32-p2:64:64:64-p3:64:64:64-a0:0:8-f32:32:32-f64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-v64:64:64-v128:128:128-n8:16:32:64" | FileCheck %s --check-prefixes=ALL,LE
+; RUN: opt < %s -passes=instcombine -S -data-layout="E-p:64:64:64-p1:32:32:32-p2:64:64:64-p3:64:64:64-a0:0:8-f32:32:32-f64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-v64:64:64-v128:128:128-n8:16:32:64" -use-constant-fp-for-fixed-length-splat -use-constant-int-for-fixed-length-splat | FileCheck %s --check-prefixes=ALL,BE
+; RUN: opt < %s -passes=instcombine -S -data-layout="e-p:64:64:64-p1:32:32:32-p2:64:64:64-p3:64:64:64-a0:0:8-f32:32:32-f64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-v64:64:64-v128:128:128-n8:16:32:64" -use-constant-fp-for-fixed-length-splat -use-constant-int-for-fixed-length-splat | FileCheck %s --check-prefixes=ALL,LE
 
 declare void @use_i32(i32)
 declare void @use_v2i32(<2 x i32>)

[david-arm]

Refers to line 206: Do we need to add ConstantInt and ConstantFP checks here too?

[paulwalker-arm]

I was going to say no because it should be safe to not perform anymore transformations. However, if I'm reading it correctly I think that block of code is just broken upstream?

For the case where constant folding does not happen it is effectively returning ConstantExpr::getBitCast(C, SrcAsIntTy) and by this point we already know SrcAsIntTy != DestVTy?

[david-arm]

LGTM!