Crash from llvm-types

[llvmbot]

Bugzilla Link	2987
Resolution	FIXED
Resolved on	Nov 07, 2018 00:22
Version	trunk
OS	MacOS X
Attachments	.i file
Reporter	LLVM Bugzilla Contributor

Extended Description

This happens on llvm-gcc on x86 darwin (or ppc32). Was introduced by 58387.

now what? $ld/gcc/llvm-gcc-4.2.SL.obj/gcc/xgcc -B $ld/gcc/llvm-gcc-4.2.SL.obj/gcc/ t024_x.i -w
Assertion failed: ((isBitfield(Field) || FieldNo == ~0U || FieldOffsetInBits == 8*Info->ElementOffsetInBytes[FieldNo]) && "Wrong LLVM field offset!"), function ConvertRECORD, file ../../llvm-gcc-4.2/gcc/llvm-types.cpp, line 2150.
In file included from /Volumes/MacOS9/gcc/llvm-gcc-4.2.SL.obj/gcc/testsuite/gcc/gcc.dg-struct-layout-1/t024_x.c:6:
/Volumes/MacOS9/gcc/llvm-gcc-4.2.SL.obj/gcc/testsuite/gcc/gcc.dg-struct-layout-1/t024_test.h: In function ‘test2045’:
/Volumes/MacOS9/gcc/llvm-gcc-4.2.SL.obj/gcc/testsuite/gcc/gcc.dg-struct-layout-1/t024_test.h:46: internal compiler error: Abort trap

Test from gcc testsuite.

[llvmbot]

The shift operands have different types (i32 and i8):

0x19e25a8: i32 = zero_extend 0x19fcfa8
0x19e25a8:
0x19e26a8: i8 = Constant <8>
0x19e27a8: i32 = shl 0x19e25a8, 0x19e26a8

According to LangRef this is not allowed at the level
of bitcode, so it might be worth looking into who is
creating these nodes. That said, I'm going to add
support for this to LegalizeTypes.

[llvmbot]

Now I come to think of it, isn't operand 1 always supposed
to have type TLI.getShiftAmountTy()?

[llvmbot]

Sorry, these comments for #3358 :)

[llvmbot]

This was an old bug caught by recently added sanity checking.
Fixed here:
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20081027/069152.html

[llvmbot]

Verified.