Re: PR middle-end/52141: ICE due to asm statement

[Aldy Hernandez <aldyh at redhat dot com>]

On 02/15/12 12:07, Jakub Jelinek wrote:
> On Wed, Feb 15, 2012 at 11:59:15AM -0600, Aldy Hernandez wrote:
> > Hmmm, isn't %K for trees?  We're talking gimple, and FUNCTION_DECLs,
> > which don't have a TREE_BLOCK, here.
>
> Gimple stmts have gimple_block, I guess you'd need to create some tree
> and set its TREE_BLOCK and EXPR_LOCATION from gimple_block/gimple_location.
> Or add something similar to %K that would take the same info from gimple
> stmt and pass a stmt instead of a tree.
>
> 	Jakub

This may not actually work.  I am getting a different backtrace 
depending on whether I use -O0 or -O[123].  With the attached patch I 
get a proper backtrace with -O0, but none whatsoever with -O1:

#GOOD
houston:/build/t2/gcc$ ./cc1 a.c -fgnu-tm -O0 -quiet -w
In function 'asmfunc',
    inlined from 'f' at a.c:13:10:
a.c:7:3: error: asm not allowed in 'transaction_safe' function

#BAD
houston:/build/t2/gcc$ ./cc1 a.c -fgnu-tm -O1 -quiet -w
a.c: In function 'f':
a.c:7:3: error: asm not allowed in 'transaction_safe' function
houston:/build/t2/gcc$

With -O1, the TREE_BLOCK looks like this:

(gdb) ptg block
BLOCK #3
  SUPERCONTEXT: BLOCK #0
  ABSTRACT_ORIGIN: BLOCK #0

and the BLOCK_SUPERCONTEXT looks like this:

(gdb) ptg block
BLOCK #0
  SUPERCONTEXT: f
  SUBBLOCKS: BLOCK #3

Notice there is no BLOCK_ABSTRACT_ORIGIN in this last one, so we don't 
print any context.

I would rather have a consistent error message (though possibly 
confusing), than have an error message that sometimes makes sense.

To make matters worse, I have just noticed that at -O2 and -O3 we 
inlined all the functions into main, so the GIMPLE_ASM ends up in the 
transaction itself (see test), not in f(), so the compiler is able to 
determine we have a GIMPLE_ASM inside a transaction and issues a 
different error:

houston:/build/t2/gcc$ ./cc1 a.c -fgnu-tm -O2 -quiet -w
a.c:7:3: error: asm not allowed in atomic transaction

Here also we could've used the %K format modifier, if indeed we had full 
backtrace information.  But, like -O1, we don't.

I think I'm back to learning towards rth's suggestion of disabling early 
inlining.

Thoughts?

Attachment: curr
Description: Text document