llvm / llvm-project

The LLVM Project is a collection of modular and reusable compiler and toolchain technologies.
http://llvm.org
Other
27.64k stars 11.37k forks source link

[SelectionDAG] LegalizeTypes hits asserts in PerformExpensiveChecks when compiled with XDEBUG #27857

Open llvmbot opened 8 years ago

llvmbot commented 8 years ago
Bugzilla Link 27483
Version trunk
OS Linux
Blocks llvm/llvm-project#31494
Reporter LLVM Bugzilla Contributor
CC @fhahn,@RKSimon,@sanjoy

Extended Description

The following lit tests hit asserts when run with toolchain compiled with XDEBUG enabled. Judging from the test names this looks to be a problem with fp16/fp128

CodeGen/AMDGPU/fp16_to_fp.ll CodeGen/AMDGPU/fp32_to_fp16.ll CodeGen/AMDGPU/half.ll CodeGen/ARM/fp16-v3.ll CodeGen/ARM/fp16-args.ll CodeGen/ARM/fp16-promote.ll CodeGen/ARM/fp16.ll CodeGen/ARM/half.ll CodeGen/Mips/fp16-promote.ll CodeGen/NVPTX/fp16.ll CodeGen/NVPTX/half.ll CodeGen/Thumb2/float-intrinsics-double.ll CodeGen/Thumb2/float-intrinsics-float.ll CodeGen/X86/atomic-non-integer.ll CodeGen/X86/cvt16.ll CodeGen/X86/fastmath-float-half-conversion.ll CodeGen/X86/fp128-i128.ll CodeGen/X86/fp128-load.ll CodeGen/X86/fp128-compare.ll CodeGen/X86/fp128-libcalls.ll CodeGen/X86/fp128-cast.ll CodeGen/X86/half.ll CodeGen/X86/soft-fp.ll CodeGen/X86/vector-half-conversions.ll

fhahn commented 2 years ago

mentioned in issue llvm/llvm-project#31494

fhahn commented 7 years ago

I've committed a patch recently that fixed some fp16 failures with XDEBUG and there's another one up for review https://reviews.llvm.org/D28195

RKSimon commented 8 years ago

As of rL268050, XDEBUG functionality is now enabled by the LLVM_ENABLE_EXPENSIVE_CHECKS cmake option