[LLVMdev] Possible bug in getCallPreservedMask for CallingConv::Intel_OCL_BI

[Philip Reames]

Not sure who owns this bit of code, so sending this to the general list.

It looks like there may be an unintentional fall through happening in 
the X86RegisterInfo::getCallPreservedMask function.

http://llvm.org/docs/doxygen/html/X86RegisterInfo_8cpp_source.html

case  CallingConv::Intel_OCL_BI  <http://llvm.org/docs/doxygen/html/namespacellvm_1_1CallingConv.html#a4f861731fc6dbfdccc05af5968d98974ad47327c131a0990283111588b89587cb>: {
      if  (IsWin64 && HasAVX512)
        return  CSR_Win64_Intel_OCL_BI_AVX512_RegMask;
      if  (Is64Bit && HasAVX512)
        return  CSR_64_Intel_OCL_BI_AVX512_RegMask;
      if  (IsWin64 && HasAVX)
        return  CSR_Win64_Intel_OCL_BI_AVX_RegMask;
      if  (Is64Bit && HasAVX)
        return  CSR_64_Intel_OCL_BI_AVX_RegMask;
      if  (!HasAVX && !IsWin64 && Is64Bit)
        return  CSR_64_Intel_OCL_BI_RegMask;
      // Missing break/return?
  }

It _may_ be intentional - it falls through to the Cold calling 
convention and then general C convention, but it really looks suspicious.

If it is intentional, please add an explicit note about the desired fall 
through behaviour.  I found this by inserting a new case in the switch 
and having tests fail.  The two tests that seem to hit the fallthrough 
behavor are: CodeGen/X86/avx-intel-ocl.ll and CodeGen/X86/sse-intel-ocl.ll

Philip
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20140313/4dffdfa7/attachment.html>