[Clang] Fix a non-effective assertion

[shiltian]

PTy here is literally FTy->getParamType(i), which makes this assertion not
work as expected.

[llvmbot]

@llvm/pr-subscribers-clang

Author: Shilei Tian (shiltian)

Changes

PTy here is literally FTy->getParamType(i), which makes this assertion not
work as expected.

---

Full diff: https://github.com/llvm/llvm-project/pull/81083.diff

1 Files Affected:

• (modified) clang/lib/CodeGen/CGBuiltin.cpp (+1-1)

diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index e051cbc6486353..a7a410dab1a018 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -5908,7 +5908,7 @@ RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID,
           }
         }
 
-        assert(PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) &&
+        assert(ArgValue->getType()->canLosslesslyBitCastTo(PTy) &&
                "Must be able to losslessly bit cast to param");
         // Cast vector type (e.g., v256i32) to x86_amx, this only happen
         // in amx intrinsics.

[llvmbot]

@llvm/pr-subscribers-clang-codegen

Author: Shilei Tian (shiltian)

Changes

PTy here is literally FTy->getParamType(i), which makes this assertion not
work as expected.

---

Full diff: https://github.com/llvm/llvm-project/pull/81083.diff

1 Files Affected:

• (modified) clang/lib/CodeGen/CGBuiltin.cpp (+1-1)

diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index e051cbc6486353..a7a410dab1a018 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -5908,7 +5908,7 @@ RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID,
           }
         }
 
-        assert(PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) &&
+        assert(ArgValue->getType()->canLosslesslyBitCastTo(PTy) &&
                "Must be able to losslessly bit cast to param");
         // Cast vector type (e.g., v256i32) to x86_amx, this only happen
         // in amx intrinsics.

[arsenm]

Is there much point to this assertion, given that the BitCast constructor will also assert castIsValid? canLosslesslyBitCastTo looks arbitrarily different from castIsValid

[shiltian]

canLosslesslyBitCastTo is supposed to be stricter than castIsValid but it looks too conservative. For example, it doesn't allow bitcast between float and int.