[MSP430] Fix gcc.dg/pr85180.c and gcc.dg/pr87985.c timeouts for msp430-elf -mlarge

Compilation of gcc.dg/pr85180.c and gcc.dg/pr87985.c times out after 5 minutes
for msp430 with -mlarge.

nonzero_bits1 (from rtlanal.c), recurses many times for each reg
because reg_nonzero_bits_for_combine (combine.c) never considers using 
last_set_nonzero_bits for the given reg when the reg is PSImode (i.e. Pmode for
msp430-elf -mlarge).

nonzero bits for a mode of class MODE_PARTIAL_INT are valid for a mode of class
MODE_INT, and vice-versa. The existing comment in reg_nonzero_bits_for_combine
explaining why last_set_nonzero_bits is valid even when the precision of the
last set mode is less than the current mode, also explains why
MODE_PARTIAL_INT and MODE_INT can be used interchangeably here:

> record_value_for_reg invoked nonzero_bits on the register
> with nonzero_bits_mode (because last_set_mode is necessarily integral
> and HWI_COMPUTABLE_MODE_P in this case) so bits in nonzero_bits_mode
> are all valid, hence in mode too since nonzero_bits_mode is defined
> to the largest HWI_COMPUTABLE_MODE_P mode.

The attached patch fixes the timeout with mlarge (compilation takes only a
couple of seconds) by allowing the last set nonzero bits for a 
reg to be used if either the current mode or last mode is
MODE_PARTIAL_INT or MODE_INT. Currently only MODE_INT is considered.

Successfully bootstrapped and regtested x86_64-pc-linux-gnu and msp430-elf
-msmall and -mlarge.

Ok for trunk?

On Wed, Dec 12, 2018 at 12:09:19PM +0000, Jozef Lawrynowicz wrote:
> Compilation of gcc.dg/pr85180.c and gcc.dg/pr87985.c times out after 5 minutes
> for msp430 with -mlarge.
> 
> nonzero_bits1 (from rtlanal.c), recurses many times for each reg
> because reg_nonzero_bits_for_combine (combine.c) never considers using 
> last_set_nonzero_bits for the given reg when the reg is PSImode (i.e. Pmode for
> msp430-elf -mlarge).
> 
> nonzero bits for a mode of class MODE_PARTIAL_INT are valid for a mode of class
> MODE_INT, and vice-versa.

The unused bits in a MODE_PARTIAL_INT value are undefined, so nonzero_bits
isn't valid for conversion in either direction.

And *which* bits are undefined isn't defined anywhere either, so we cannot
convert to/from smaller MODE_INT modes, either.

> The existing comment in reg_nonzero_bits_for_combine
> explaining why last_set_nonzero_bits is valid even when the precision of the
> last set mode is less than the current mode, also explains why
> MODE_PARTIAL_INT and MODE_INT can be used interchangeably here:
> 
> > record_value_for_reg invoked nonzero_bits on the register
> > with nonzero_bits_mode (because last_set_mode is necessarily integral
> > and HWI_COMPUTABLE_MODE_P in this case) so bits in nonzero_bits_mode
> > are all valid, hence in mode too since nonzero_bits_mode is defined
> > to the largest HWI_COMPUTABLE_MODE_P mode.

I don't see how that follows; not all bits in MODE_PARTIAL_INT modes
are necessarily valid.

Perhaps it is true that you can return whatever you want here, for the
undefined bits in a partial int var; but you'll need to justify that
then, it isn't obvious to me at least.

> +	      && (GET_MODE_CLASS (mode) == MODE_INT
> +		  || GET_MODE_CLASS (mode) == MODE_PARTIAL_INT)))

That's SCALAR_INT_MODE_P fwiw.

Thanks,


Segher

Hi Segher,

Thanks for the review.

On Wed, 12 Dec 2018 19:47:53 -0600
Segher Boessenkool <segher@kernel.crashing.org> wrote:

> The unused bits in a MODE_PARTIAL_INT value are undefined, so nonzero_bits
> isn't valid for conversion in either direction.
>
> And *which* bits are undefined isn't defined anywhere either, so we cannot
> convert to/from smaller MODE_INT modes, either.

Can't we use the last_set_nonzero_bits if last_set_mode was MODE_INT and the
current mode is MODE_PARTIAL_INT? As long as the current mode bitsize is less
than the bitsize of nonzero_bits_mode, which it will be if we've gotten to this
point.

If the current mode is MODE_PARTIAL_INT, then on entry to
reg_nonzero_bits_for_combine, the current nonzero_bits has already been
initialized to GET_MODE_MASK(mode), so we are not at risk of disturbing the
undefined bits, as we are only ever doing &= with GET_MODE_MASK(mode).

However, the above suggestion doesn't actually provide any visible benefit to
testresults, so it doesn't need to be included.

> I don't see how that follows; not all bits in MODE_PARTIAL_INT modes
> are necessarily valid.

Yes, this was an oversight on my part. 
nonzero_bits_mode is only used to calculate last_set_nonzero_bits if
last_set_mode is in the MODE_INT class.
If last_set_mode was MODE_PARTIAL_INT class, then last_set_mode was just that
partial int mode; it wasn't calculated in the wider nonzero_bits_mode.

After some further investigation, it seems we can attribute the recursion to
an inconsistency with how nonzero_bits() is invoked.
The mode passed to nonzero_bits(rtx, mode) is normally the mode of rtx
itself. But there are two cases where nonzero_bits_mode is used instead, even
if that is wider than the mode of the rtx.

In record_value_for_reg:
>   rsp->last_set_mode = mode;
>   if (GET_MODE_CLASS (mode) == MODE_INT
>       && HWI_COMPUTABLE_MODE_P (mode))
>     mode = nonzero_bits_mode;
>   rsp->last_set_nonzero_bits = nonzero_bits (value, mode);

In update_rsp_from_reg_equal: 
>  if (rsp->nonzero_bits != HOST_WIDE_INT_M1U)
>    {
>      bits = nonzero_bits (src, nonzero_bits_mode);

Note that the the mode of src in update_rsp_from_reg_equal is not
checked to see if it is a MODE_INT class and HWI_COMPUTABLE, nonzero_bits_mode
is always used.

This mode passed to nonzero_bits() eventually makes its way to
reg_nonzero_bits_for_combine. rsp->last_set_mode is always the true mode of the
reg (i.e. not nonzero_bits_mode) from when it is set in record_value_for_reg.
So the recursion happens because update_rsp_from_reg_equal has asked for the
nonzero_bits in nonzero_bits_mode, but the last_set_mode was PSImode.
nonzero_bits_mode is not equal to PSImode, nor is it in the same class, so the
nonzero bits will never be reused.

So my revised patch (attached) instead modifies update_rsp_from_reg_equal to
only request nonzero_bits in nonzero_bits_mode if the mode class is MODE_INT
and HWI_COMPUTABLE. This gives it parity with how last_set_nonzero_bits are set
in record_value_for_reg.

I've regtested the attached patch for msp430-elf, currently bootstrapping and
testing on x86_64-pc-linux-gnu.
Is this ok for trunk if bootstrap and regtest for x86_64-pc-linux-gnu is
successful?

Jozef
2018-12-14  Jozef Lawrynowicz  <jozef.l@mittosystems.com>

	gcc/ChangeLog:
	* combine.c (update_rsp_from_reg_equal): Only look for the nonzero bits
	of src in nonzero_bits_mode if the mode of src is MODE_INT and
	HWI_COMPUTABLE.
	(reg_nonzero_bits_for_combine): Add clarification to comment.

diff --git a/gcc/combine.c b/gcc/combine.c
index 7e61139..c93aaed 100644
--- a/gcc/combine.c
+++ b/gcc/combine.c
@@ -1698,9 +1698,13 @@ update_rsp_from_reg_equal (reg_stat_type *rsp, rtx_insn *insn, const_rtx set,
   /* Don't call nonzero_bits if it cannot change anything.  */
   if (rsp->nonzero_bits != HOST_WIDE_INT_M1U)
     {
-      bits = nonzero_bits (src, nonzero_bits_mode);
+      machine_mode mode = GET_MODE (x);
+      if (GET_MODE_CLASS (mode) == MODE_INT
+	  && HWI_COMPUTABLE_MODE_P (mode))
+	mode = nonzero_bits_mode;
+      bits = nonzero_bits (src, mode);
       if (reg_equal && bits)
-	bits &= nonzero_bits (reg_equal, nonzero_bits_mode);
+	bits &= nonzero_bits (reg_equal, mode);
       rsp->nonzero_bits |= bits;
     }
 
@@ -10224,6 +10228,7 @@ simplify_and_const_int (rtx x, scalar_int_mode mode, rtx varop,
 
 /* Given a REG X of mode XMODE, compute which bits in X can be nonzero.
    We don't care about bits outside of those defined in MODE.
+   We DO care about all the bits in MODE, even if XMODE is smaller than MODE.
 
    For most X this is simply GET_MODE_MASK (GET_MODE (MODE)), but if X is
    a shift, AND, or zero_extract, we can do better.  */

Hi!

On Fri, Dec 14, 2018 at 03:22:13PM +0000, Jozef Lawrynowicz wrote:
> 2018-12-14  Jozef Lawrynowicz  <jozef.l@mittosystems.com>
> 
> 	gcc/ChangeLog:
> 	* combine.c (update_rsp_from_reg_equal): Only look for the nonzero bits
> 	of src in nonzero_bits_mode if the mode of src is MODE_INT and
> 	HWI_COMPUTABLE.
> 	(reg_nonzero_bits_for_combine): Add clarification to comment.

Is there some PR this fixes?

> 
> diff --git a/gcc/combine.c b/gcc/combine.c
> index 7e61139..c93aaed 100644
> --- a/gcc/combine.c
> +++ b/gcc/combine.c
> @@ -1698,9 +1698,13 @@ update_rsp_from_reg_equal (reg_stat_type *rsp, rtx_insn *insn, const_rtx set,
>    /* Don't call nonzero_bits if it cannot change anything.  */
>    if (rsp->nonzero_bits != HOST_WIDE_INT_M1U)
>      {
> -      bits = nonzero_bits (src, nonzero_bits_mode);
> +      machine_mode mode = GET_MODE (x);
> +      if (GET_MODE_CLASS (mode) == MODE_INT
> +	  && HWI_COMPUTABLE_MODE_P (mode))
> +	mode = nonzero_bits_mode;
> +      bits = nonzero_bits (src, mode);
>        if (reg_equal && bits)
> -	bits &= nonzero_bits (reg_equal, nonzero_bits_mode);
> +	bits &= nonzero_bits (reg_equal, mode);
>        rsp->nonzero_bits |= bits;
>      }
>  
> @@ -10224,6 +10228,7 @@ simplify_and_const_int (rtx x, scalar_int_mode mode, rtx varop,
>  
>  /* Given a REG X of mode XMODE, compute which bits in X can be nonzero.
>     We don't care about bits outside of those defined in MODE.
> +   We DO care about all the bits in MODE, even if XMODE is smaller than MODE.
>  
>     For most X this is simply GET_MODE_MASK (GET_MODE (MODE)), but if X is
>     a shift, AND, or zero_extract, we can do better.  */

I think this is okay for trunk, and for backports after waiting a week
or so for fallout.  Thanks!


Segher

On Tue, 18 Dec 2018 03:08:51 -0600
Segher Boessenkool <segher@kernel.crashing.org> wrote:

> Hi!
> 
> On Fri, Dec 14, 2018 at 03:22:13PM +0000, Jozef Lawrynowicz wrote:
> > 2018-12-14  Jozef Lawrynowicz  <jozef.l@mittosystems.com>
> > 
> > 	gcc/ChangeLog:
> > 	* combine.c (update_rsp_from_reg_equal): Only look for the nonzero bits
> > 	of src in nonzero_bits_mode if the mode of src is MODE_INT and
> > 	HWI_COMPUTABLE.
> > 	(reg_nonzero_bits_for_combine): Add clarification to comment.  
> 
> Is there some PR this fixes?

No not for this one, I just spotted the timeouts in the GCC testsuite.

> > 
> > diff --git a/gcc/combine.c b/gcc/combine.c
> > index 7e61139..c93aaed 100644
> > --- a/gcc/combine.c
> > +++ b/gcc/combine.c
> > @@ -1698,9 +1698,13 @@ update_rsp_from_reg_equal (reg_stat_type *rsp, rtx_insn *insn, const_rtx set,
> >    /* Don't call nonzero_bits if it cannot change anything.  */
> >    if (rsp->nonzero_bits != HOST_WIDE_INT_M1U)
> >      {
> > -      bits = nonzero_bits (src, nonzero_bits_mode);
> > +      machine_mode mode = GET_MODE (x);
> > +      if (GET_MODE_CLASS (mode) == MODE_INT
> > +	  && HWI_COMPUTABLE_MODE_P (mode))
> > +	mode = nonzero_bits_mode;
> > +      bits = nonzero_bits (src, mode);
> >        if (reg_equal && bits)
> > -	bits &= nonzero_bits (reg_equal, nonzero_bits_mode);
> > +	bits &= nonzero_bits (reg_equal, mode);
> >        rsp->nonzero_bits |= bits;
> >      }
> >  
> > @@ -10224,6 +10228,7 @@ simplify_and_const_int (rtx x, scalar_int_mode mode, rtx varop,
> >  
> >  /* Given a REG X of mode XMODE, compute which bits in X can be nonzero.
> >     We don't care about bits outside of those defined in MODE.
> > +   We DO care about all the bits in MODE, even if XMODE is smaller than MODE.
> >  
> >     For most X this is simply GET_MODE_MASK (GET_MODE (MODE)), but if X is
> >     a shift, AND, or zero_extract, we can do better.  */  
> 
> I think this is okay for trunk, and for backports after waiting a week
> or so for fallout.  Thanks!

Thanks, applied to trunk.

Jozef