diff mbox

PR 61084: SPARC fallout from wide-int merge

Message ID 87fvkm549i.fsf@talisman.default
State New
Headers show

Commit Message

Richard Sandiford May 7, 2014, 9:26 a.m. UTC 
The DImode constant spliiter assigned the result of trunc_int_for_mode
to an unsigned int rather than a HOST_WIDE_INT.  This then produced const_ints
that were zero-extended rather than sign-extended and tripped the assert:

	gcc_checking_assert (INTVAL (x.first)
			     == sext_hwi (INTVAL (x.first), precision)
			     || (x.second == BImode && INTVAL (x.first) == 1));

The other hunks are just by inspection, but I think gen_int_mode is
preferred over GEN_INT when the mode is obvious.

Tested by Rainer, who says that the boostrap now completes.
OK to install?

Thanks,
Richard


gcc/
	PR target/61084
	* config/sparc/sparc.md: Fix types of low and high in DI constant
	splitter.  Use gen_int_mode in some other splitters.

Comments

Mike Stump May 7, 2014, 3:43 p.m. UTC | #1 
On May 7, 2014, at 2:26 AM, Richard Sandiford <rdsandiford@googlemail.com> wrote:
> The DImode constant spliiter assigned the result of trunc_int_for_mode
> to an unsigned int rather than a HOST_WIDE_INT.  This then produced const_ints
> that were zero-extended rather than sign-extended and tripped the assert:
> 
> 	gcc_checking_assert (INTVAL (x.first)
> 			     == sext_hwi (INTVAL (x.first), precision)
> 			     || (x.second == BImode && INTVAL (x.first) == 1));
> 
> The other hunks are just by inspection, but I think gen_int_mode is
> preferred over GEN_INT when the mode is obvious.
> 
> Tested by Rainer, who says that the boostrap now completes.
> OK to install?

Ok.
Richard Sandiford May 7, 2014, 3:56 p.m. UTC | #2 
Mike Stump <mikestump@comcast.net> writes:
> On May 7, 2014, at 2:26 AM, Richard Sandiford
> <rdsandiford@googlemail.com> wrote:
>> The DImode constant spliiter assigned the result of trunc_int_for_mode
>> to an unsigned int rather than a HOST_WIDE_INT.  This then produced const_ints
>> that were zero-extended rather than sign-extended and tripped the assert:
>> 
>> 	gcc_checking_assert (INTVAL (x.first)
>> 			     == sext_hwi (INTVAL (x.first), precision)
>> 			     || (x.second == BImode && INTVAL (x.first) == 1));
>> 
>> The other hunks are just by inspection, but I think gen_int_mode is
>> preferred over GEN_INT when the mode is obvious.
>> 
>> Tested by Rainer, who says that the boostrap now completes.
>> OK to install?
>
> Ok.

I think this needs a backend maintainer.  Although it was exposed by
the wide-int assert, it isn't really wide-int-related as such.

Thanks,
Richard
Eric Botcazou May 14, 2014, 8:38 a.m. UTC | #3 
> 	PR target/61084
> 	* config/sparc/sparc.md: Fix types of low and high in DI constant
> 	splitter.  Use gen_int_mode in some other splitters.

OK, thanks.
diff mbox

Patch

Index: gcc/config/sparc/sparc.md
===================================================================
--- gcc/config/sparc/sparc.md	2014-05-07 10:15:23.051156294 +0100
+++ gcc/config/sparc/sparc.md	2014-05-07 10:15:27.922201361 +0100
@@ -1886,7 +1886,7 @@  (define_split
   emit_insn (gen_movsi (gen_lowpart (SImode, operands[0]),
 			operands[1]));
 #else
-  unsigned int low, high;
+  HOST_WIDE_INT low, high;
 
   low = trunc_int_for_mode (INTVAL (operands[1]), SImode);
   high = trunc_int_for_mode (INTVAL (operands[1]) >> 32, SImode);
@@ -4822,7 +4822,7 @@  (define_split
   [(set (match_dup 3) (match_dup 4))
    (set (match_dup 0) (ior:SI (not:SI (match_dup 3)) (match_dup 1)))]
 {
-  operands[4] = GEN_INT (~INTVAL (operands[2]));
+  operands[4] = gen_int_mode (~INTVAL (operands[2]), SImode);
 })
 
 (define_insn_and_split "*or_not_di_sp32"
@@ -4899,7 +4899,7 @@  (define_split
   [(set (match_dup 3) (match_dup 4))
    (set (match_dup 0) (not:SI (xor:SI (match_dup 3) (match_dup 1))))]
 {
-  operands[4] = GEN_INT (~INTVAL (operands[2]));
+  operands[4] = gen_int_mode (~INTVAL (operands[2]), SImode);
 })
 
 (define_split
@@ -4911,7 +4911,7 @@  (define_split
   [(set (match_dup 3) (match_dup 4))
    (set (match_dup 0) (xor:SI (match_dup 3) (match_dup 1)))]
 {
-  operands[4] = GEN_INT (~INTVAL (operands[2]));
+  operands[4] = gen_int_mode (~INTVAL (operands[2]), SImode);
 })
 
 ;; Split DImode logical operations requiring two instructions.