[clang] [Clang] Enable -fextend-lifetimes at -Og (PR #118026)

Mikael Holmén via cfe-commits cfe-commits at lists.llvm.org
Fri Apr 25 00:24:37 PDT 2025


mikaelholmen wrote:

I did some downstream testing with "-Og" since this patch and noticed that e.g.
```
llc -mtriple=avr bbi-106478.ll
```
crashes with
```
SoftenFloatOperand Op #1: t2: ch = fake_use t0, ConstantFP:f64<0.000000e+00>

LLVM ERROR: Do not know how to soften this operator's operand!
```
and
```
llc -mtriple=systemz bbi-106478.ll
```
fails with
```
llc: ../lib/Target/SystemZ/SystemZLongBranch.cpp:226: unsigned int getInstSizeInBytes(const MachineInstr &, const SystemZInstrInfo *): Assertion `(Size || MI.isDebugOrPseudoInstr() || MI.isPosition() || MI.isKill() || MI.isImplicitDef() || MI.getOpcode() == TargetOpcode::MEMBARRIER || MI.getOpcode() == TargetOpcode::INIT_UNDEF || MI.isInlineAsm() || MI.getOpcode() == SystemZ::STACKMAP || MI.getOpcode() == SystemZ::PATCHPOINT || MI.getOpcode() == SystemZ::EH_SjLj_Setup) && "Missing size value for instruction."' failed.
```
for bbi-106478.ll just being
```
define double @idd() {
entry:
  notail call void (...) @llvm.fake.use(double 0.000000e+00)
  ret double 0.000000e+00
}
```
(and if removing the llvm.fake.use call they don't crash)
so I guess there is some testing and fixes around llvm.fake.use left to do.


https://github.com/llvm/llvm-project/pull/118026


More information about the cfe-commits mailing list