[Bug rtl-optimization/60763] [4.9 Regression] ICE in extract_insn starting with rev 208984

[meissner at linux dot vnet.ibm.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60763

--- Comment #14 from Michael Meissner <meissner at linux dot vnet.ibm.com> ---
On Tue, Apr 08, 2014 at 03:21:13PM +0000, dje at gcc dot gnu.org wrote:
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60763
> 
> --- Comment #13 from David Edelsohn <dje at gcc dot gnu.org> ---
> Gentlemen, I really do not understand this discussion.
> 
> I used the term "crude" and receive a response that it is not crude, but it is
> dangerous. WTF? Why is anyone trying to "sell" the patch when I repeatedly have
> said that I do not disagree with it in principle? I only am trying to ensure
> that this is the right use of the GCC API. i386 uses this idiom a lot; rs6000
> does not.
> 
> I see multiple responses trying to convince me of a patch with which I already
> agree and no one writing a one line comment.

When you are doing a subreg type operation, before reload, you must NOT use the
register number directly.  This is the dangerous part that was mentioned.
Hence using gen_highpart, gen_lowpart, or simplify_gen_subreg is the
appropriate solution.

If you are doing splits after reload, and are dealing with whole registers, the
preferred solution is to create a new REG with the appropriate register number.
In particular, gen_highpart and gen_lowpart must not be used after reload.
However, up until the patch to add more checking, simplify_gen_subreg could be
used after reload.

The alpha, arm, i386, ia64, iq2000, m32k, m68k, mips, msp430, pa, pdp11,
picochip, rl78, rs6000, s390, sh, sparc, spu, and xtensa all generate a call to
gen_rtx_REG using REGNO of an operand to change types.

Out of force of habit, I've tended to use simplify_gen_subreg when doing
splitting, no matter whether it is before or after reload.  I thought with
'simplify' in the name, it would do the right thing after reload, but evidently
it does not.