[PATCH] Simplifying RTL shifts to narrower modes

[Roger Sayle]

On Thu, 6 Jan 2005, Richard Henderson wrote:
> Indeed it should.
> > 	* simplify-rtx.c (simplify_subreg): Simplify truncations of shifts
> > 	of sign or zero extended values.
>
> Ok, but

Thanks.


> > +   /* Simplify (subreg:QI (lshiftrt:SI (sign_extend:SI (x:QI)) C), 0) into
> > +      to (ashiftrt:QI (x:QI) C), where C is a suitable small constant and
> > +      the outer subreg is effectively a truncation to the original mode.  */
> > +   if ((GET_CODE (op) == LSHIFTRT
> > +        || GET_CODE (op) == ASHIFTRT)
> > +       && SCALAR_INT_MODE_P (outermode)
> > +       && (2 * GET_MODE_BITSIZE (outermode)) <= GET_MODE_BITSIZE (innermode)
>
> Could do with a word or two about this last line.  Took me a moment
> or five to figure it out.

Hehe :).

I've now committed the patch with the following additional comment
that should make it much clearer why in this one case we test for
"2 * BITSIZE (outer) <= BITSIZE (inner)" instead of the more obvious
"BITSIZE (outer) < BITSIZE (inner)" used in the other two clauses.


+   /* Simplify (subreg:QI (lshiftrt:SI (sign_extend:SI (x:QI)) C), 0) into
+      to (ashiftrt:QI (x:QI) C), where C is a suitable small constant and
+      the outer subreg is effectively a truncation to the original mode.  */
+   if ((GET_CODE (op) == LSHIFTRT
+        || GET_CODE (op) == ASHIFTRT)
+       && SCALAR_INT_MODE_P (outermode)
+       /* Ensure that INNERMODE is at least twice as wide as the OUTERMODE
+ 	 to avoid the possibility that an outer LSHIFTRT shifts by more
+ 	 than the sign extension's sign_bit_copies and introduces zeros
+ 	 into the high bits of the result.  */
+       && (2 * GET_MODE_BITSIZE (outermode)) <= GET_MODE_BITSIZE (innermode)
...


I thought it better to use a single "at least twice as wide" test, which
shouldn't normally make a difference for typical integer machine modes,
than duplicate or specialize this test for LSHIFTRT and ASHIFTRT.

Thanks again,

Roger
--