Re: [PATCH] Fix altivec easy_vector_constant_add_self splitter for V4SFmode (PR target/37168)

[Jakub Jelinek <jakub at redhat dot com>]

On Sat, Aug 30, 2008 at 05:04:39PM -0400, David Edelsohn wrote:
> > +  enum machine_mode op_mode = <MODE>mode;
> 
> The line above sets op_mode to <MODE>mode.
> 
> > +  if (op_mode == <MODE>mode)
> > +    {
> > +      op_mode = V4SImode;
> > +      operands[0] = gen_lowpart (op_mode, operands[0]);
> > +    }
> 
> Then the block of code above is conditional on
> op_mode == <MODE>mode , which always
> will be true.  Maybe you intended
> 
> if (op_mode == V4SFmode)

Yeah, of course, that's what I meant to write, but apparently screwed up. 
The testcase was cured even with the errorneous patch, but
it would break V8HImode and V16QImode handling.
I'll bootstrap/regtest the fixed patch tomorrow.

	Jakub