K&R C compilation failure

[llvmbot]

Bugzilla Link	1083
Resolution	FIXED
Resolved on	Nov 07, 2018 00:17
Version	1.9
OS	Linux
Reporter	LLVM Bugzilla Contributor
CC	@asl

Extended Description

$:~/tmp$ llvm-gcc -emit-llvm -c svc.c -o svc.o
cc1: ../../dst-directory/gcc/llvm-convert.cpp:216:
void::FunctionPrologArgumentConversion::HandleScalarArgument(const
llvm::Type*): Assertion `ArgVal->getType()->isIntegral() && LLVMTy->isIntegral()
&& "Lowerings don't match?"' failed.
svc.c: In function ‘svc_register’:
svc.c:11: internal compiler error: Aborted
Please submit a full bug report,

Source: [svc.c from dietlib v0.30]

#include <rpc/rpc.h>

static SVCXPRT **xports;

bool_t svc_register(xprt, prog, vers, dispatch, protocol)
SVCXPRT *xprt;
unsigned long prog;
unsigned long vers;
void (*dispatch) ();
rpcprot_t protocol;
{
return (TRUE);
}

Can be reproduced with:
llvm-gcc [-emit-llvm] -c svc.c -o svc.o

Domagoj

[llvmbot]

assigned to @lattner

[llvmbot]

aKor has just pointed me to
http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20070104/5c2b9cf1/attachment.txt
suggesting that that patch is the most likely source of the problem.
However, I don't have that patch applied. So, with, or without the patch,
the code seems to be buggy.

To clear any possible confusion, here's my current code:

void HandleScalarArgument(const llvm::Type *LLVMTy) {
  Value *ArgVal = AI;
  if (ArgVal->getType() != LLVMTy) {
    // If this is just a mismatch between integer types, this could be due
    // to K&R prototypes, where the forward proto defines the arg as int and
    // the actual impls is a short or char.
    assert(ArgVal->getType()->isIntegral() && LLVMTy->isIntegral() &&
           "Lowerings don't match?");
    ArgVal =
      CastInst::createInferredCast(ArgVal, LLVMTy, NameStack.back(), CurBB);
  }
  assert(!LocStack.empty());
  Value *Loc = LocStack.back();
  if (cast<PointerType>(Loc->getType())->getElementType() != LLVMTy)
    // This cast only involves pointers, therefore BitCast
    Loc = CastInst::create(Instruction::BitCast, Loc,
                           PointerType::get(LLVMTy), "tmp", CurBB);

  new StoreInst(ArgVal, Loc, CurBB);
  AI->setName(NameStack.back());
  ++AI;
}

---

$:~/llvm/dst-directory/gcc$ svn info llvm-convert.cpp
Path: llvm-convert.cpp
Name: llvm-convert.cpp
URL: svn://anonsvn.opensource.apple.com/svn/llvm/trunk/gcc/llvm-convert.cpp
Repository Root: svn://anonsvn.opensource.apple.com/svn/llvm
Repository UUID: 51af6f4a-8115-0410-a7cd-c0d7a57f2c62
Revision: 224
Node Kind: file
Schedule: normal
Last Changed Author: jlaskey
Last Changed Rev: 219
Last Changed Date: 2006-12-01 01:54:57 -0800 (Fri, 01 Dec 2006)
Text Last Updated: 2006-12-09 13:16:41 -0800 (Sat, 09 Dec 2006)

      Domagoj

[asl]

Supplied source code includes some platform-dependent code (because it isn't
preprocessed).

However, it fails here too (on Linux/X86) but with another assertion:

cc1: /home/asl/proj/llvm/src/lib/VMCore/Instructions.cpp:1568: static
llvm::CastInst* llvm::CastInst::createIntegerCast(llvm::Value*, const
llvm::Type*, bool, const std::string&, llvm::BasicBlock*): Assertion
`C->getType()->isIntegral() && Ty->isIntegral() && "Invalid cast"' failed.

Will supply preprocessed & reduced source soon

[asl]

delta-reduced testcase
This file fails for me with TOT with assertion message printed before.

[llvmbot]

just to confirm that Anton's reduced testcase produces the same msg.

Domagoj

[lattner]

Verified, the reduced testcase crashes with TOT, investigating.

[lattner]

Further reduced testcase:

int svc_register (void (*dispatch) (int));

int svc_register (dispatch)
void (*dispatch) ();
{}

[lattner]

Fixed. Testcase here: test/Regression/CFrontend/2007-01-06-KNR-Proto.c

Patch sent to llvm-commits, but the mailing lists are apparently blocked, so no URL provided.

-Chris

[lattner]

Patch here:
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20070101/042147.html