[PATCH] simplify-rtx.c: Change BITSIZE to UNIT_PRECISION in simplification of (extend ashiftrt (ashift ..))) Otherwise the gcc_assert can catch when dealing with partial int modes.

[PATCH] simplify-rtx.c: Change BITSIZE to UNIT_PRECISION in simplification of (extend ashiftrt (ashift ..))) Otherwise the gcc_assert can catch when dealing with partial int modes.

[John Darrington]

---
 gcc/ChangeLog      | 6 ++++++
 gcc/simplify-rtx.c | 8 ++++----
 2 files changed, 10 insertions(+), 4 deletions(-)

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index c206ab6..47035ca 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,9 @@
+2019-07-03  John Darrington <john@darrington.wattle.id.au>
+
+	simplify-rtx.c: Change BITSIZE to UNIT_PRECISION in simplification of
+	(extend ashiftrt (ashift ..)))  Otherwise the gcc_assert will catch
+	when dealing with partial int modes.
+
 2019-07-02  Eric Botcazou  <ebotcazou@adacore.com>
 
 	* cfgexpand.c (pass_expand::execute): Deal specially with instructions
diff --git a/gcc/simplify-rtx.c b/gcc/simplify-rtx.c
index 89a46a9..d74a4ba 100644
--- a/gcc/simplify-rtx.c
+++ b/gcc/simplify-rtx.c
@@ -1504,12 +1504,12 @@ simplify_unary_operation_1 (enum rtx_code code, machine_mode mode, rtx op)
 	  && CONST_INT_P (XEXP (op, 1))
 	  && XEXP (XEXP (op, 0), 1) == XEXP (op, 1)
 	  && (op_mode = as_a <scalar_int_mode> (GET_MODE (op)),
-	      GET_MODE_BITSIZE (op_mode) > INTVAL (XEXP (op, 1))))
+	      GET_MODE_UNIT_PRECISION (op_mode) > INTVAL (XEXP (op, 1))))
 	{
 	  scalar_int_mode tmode;
-	  gcc_assert (GET_MODE_BITSIZE (int_mode)
-		      > GET_MODE_BITSIZE (op_mode));
-	  if (int_mode_for_size (GET_MODE_BITSIZE (op_mode)
+	  gcc_assert (GET_MODE_UNIT_PRECISION (int_mode)
+		      > GET_MODE_UNIT_PRECISION (op_mode));
+	  if (int_mode_for_size (GET_MODE_UNIT_PRECISION (op_mode)
 				 - INTVAL (XEXP (op, 1)), 1).exists (&tmode))
 	    {
 	      rtx inner =
-- 
1.8.3.1

Re: [PATCH] simplify-rtx.c: Change BITSIZE to UNIT_PRECISION in simplification

[Richard Sandiford]

> diff --git a/gcc/simplify-rtx.c b/gcc/simplify-rtx.c
> index 89a46a9..d74a4ba 100644
> --- a/gcc/simplify-rtx.c
> +++ b/gcc/simplify-rtx.c
> @@ -1504,12 +1504,12 @@ simplify_unary_operation_1 (enum rtx_code code, machine_mode mode, rtx op)
>  	  && CONST_INT_P (XEXP (op, 1))
>  	  && XEXP (XEXP (op, 0), 1) == XEXP (op, 1)
>  	  && (op_mode = as_a <scalar_int_mode> (GET_MODE (op)),
> -	      GET_MODE_BITSIZE (op_mode) > INTVAL (XEXP (op, 1))))
> +	      GET_MODE_UNIT_PRECISION (op_mode) > INTVAL (XEXP (op, 1))))
>  	{
>  	  scalar_int_mode tmode;
> -	  gcc_assert (GET_MODE_BITSIZE (int_mode)
> -		      > GET_MODE_BITSIZE (op_mode));
> -	  if (int_mode_for_size (GET_MODE_BITSIZE (op_mode)
> +	  gcc_assert (GET_MODE_UNIT_PRECISION (int_mode)
> +		      > GET_MODE_UNIT_PRECISION (op_mode));
> +	  if (int_mode_for_size (GET_MODE_UNIT_PRECISION (op_mode)
>  				 - INTVAL (XEXP (op, 1)), 1).exists (&tmode))
>  	    {
>  	      rtx inner =

I think these should be GET_MODE_PRECISION rather than
GET_MODE_UNIT_PRECISION.  They do the same thing in this context,
but it seems odd to use GET_MODE_UNIT_... when we're specifically
dealing with scalars.

Looks good otherwise (and small enough not to need a copyright assignment,
in case you haven't signed one).

Thanks,
Richard

Re: [PATCH] simplify-rtx.c: Change BITSIZE to UNIT_PRECISION in simplification

[Jeff Law]

On 7/3/19 2:17 PM, Richard Sandiford wrote:
>> diff --git a/gcc/simplify-rtx.c b/gcc/simplify-rtx.c
>> index 89a46a9..d74a4ba 100644
>> --- a/gcc/simplify-rtx.c
>> +++ b/gcc/simplify-rtx.c
>> @@ -1504,12 +1504,12 @@ simplify_unary_operation_1 (enum rtx_code code, machine_mode mode, rtx op)
>>  	  && CONST_INT_P (XEXP (op, 1))
>>  	  && XEXP (XEXP (op, 0), 1) == XEXP (op, 1)
>>  	  && (op_mode = as_a <scalar_int_mode> (GET_MODE (op)),
>> -	      GET_MODE_BITSIZE (op_mode) > INTVAL (XEXP (op, 1))))
>> +	      GET_MODE_UNIT_PRECISION (op_mode) > INTVAL (XEXP (op, 1))))
>>  	{
>>  	  scalar_int_mode tmode;
>> -	  gcc_assert (GET_MODE_BITSIZE (int_mode)
>> -		      > GET_MODE_BITSIZE (op_mode));
>> -	  if (int_mode_for_size (GET_MODE_BITSIZE (op_mode)
>> +	  gcc_assert (GET_MODE_UNIT_PRECISION (int_mode)
>> +		      > GET_MODE_UNIT_PRECISION (op_mode));
>> +	  if (int_mode_for_size (GET_MODE_UNIT_PRECISION (op_mode)
>>  				 - INTVAL (XEXP (op, 1)), 1).exists (&tmode))
>>  	    {
>>  	      rtx inner =
> 
> I think these should be GET_MODE_PRECISION rather than
> GET_MODE_UNIT_PRECISION.  They do the same thing in this context,
> but it seems odd to use GET_MODE_UNIT_... when we're specifically
> dealing with scalars.
> 
> Looks good otherwise (and small enough not to need a copyright assignment,
> in case you haven't signed one).
John, I assume you're doing this for an out of tree port (s12z?)?
Otherwise it'd also be useful if you could include a test which triggers
the assert.

If you could confirm that Richard's suggestion of using
GET_MODE_PRECISION rather than GET_MODE_UNIT_PRECISION works it'd be
appreciated.

Thanks,
jeff

Re: [PATCH] simplify-rtx.c: Change BITSIZE to UNIT_PRECISION in simplification

[John Darrington]

On Wed, Jul 03, 2019 at 04:32:54PM -0600, Jeff Law wrote:

     John, I assume you're doing this for an out of tree port (s12z?)?

That is correct.

     Otherwise it'd also be useful if you could include a test which triggers
     the assert.

Once I have the other aspects of the port sorted out I can create a test
case (although it wouldn't be necessary because it happens building
libgcc).
     
     If you could confirm that Richard's suggestion of using
     GET_MODE_PRECISION rather than GET_MODE_UNIT_PRECISION works it'd be
     appreciated.

I had wondered about the difference between the two.   Unfortunately, 
neither are mentioned in gccint.texinfo   - I ended up deciding upon
GET_MOE_UNIT_PRECISION because there is a similar situation a few lines
above in simplify-rtx.c which uses it.   

If GET_MODE_UNIT_PRECISION is inappropriate, then please change it.


Regards,

John

-- 
Avoid eavesdropping.  Send strong encrypted email.
PGP Public key ID: 1024D/2DE827B3 
fingerprint = 8797 A26D 0854 2EAB 0285  A290 8A67 719C 2DE8 27B3
See http://sks-keyservers.net or any PGP keyserver for public key.