This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[Committed] S/390: Invalid slgfi instruction emitted with -march=z9-109
- From: "Andreas Krebbel" <Andreas dot Krebbel at de dot ibm dot com>
- To: gcc-patches at gcc dot gnu dot org
- Cc: Ulrich dot Weigand at de dot ibm dot com
- Date: Wed, 12 Dec 2007 10:21:39 +0100
- Subject: [Committed] S/390: Invalid slgfi instruction emitted with -march=z9-109
Hello,
the attached patch fixes a problem with the 'On' constraint in the
s390 back end.
I've fixed this problem already last year:
http://gcc.gnu.org/ml/gcc-patches/2006-05/msg00499.html
But accidently the patch has been reverted by:
http://gcc.gnu.org/ml/gcc-patches/2006-10/msg00073.html
So I better try it with a testcase this time ;)
Boostrapped on s390 and s390x with --with-arch=z9-ec.
No testsuite regressions.
Bye,
-Andreas-
2007-12-12 Andreas Krebbel <krebbel1@de.ibm.com>
* config/s390/s390.c (s390_O_constraint_str): Limit the range
for 'On' to -4G+1..-1.
* config/s390/constraints.md: Adjust comment accordingly.
2007-12-12 Andreas Krebbel <krebbel1@de.ibm.com>
* gcc.target/s390/20071212-1.c: New testcase.
Index: gcc/config/s390/s390.c
===================================================================
*** gcc/config/s390/s390.c.orig 2007-12-12 09:40:51.000000000 +0100
--- gcc/config/s390/s390.c 2007-12-12 09:45:40.000000000 +0100
*************** s390_O_constraint_str (const char c, HOS
*** 2056,2063 ****
|| s390_single_part (GEN_INT (value), DImode, SImode, 0) == 1;
case 'n':
! return value == -1
! || s390_single_part (GEN_INT (value), DImode, SImode, -1) == 1;
default:
gcc_unreachable ();
--- 2056,2062 ----
|| s390_single_part (GEN_INT (value), DImode, SImode, 0) == 1;
case 'n':
! return s390_single_part (GEN_INT (value - 1), DImode, SImode, -1) == 1;
default:
gcc_unreachable ();
Index: gcc/config/s390/constraints.md
===================================================================
*** gcc/config/s390/constraints.md.orig 2007-12-12 09:40:51.000000000 +0100
--- gcc/config/s390/constraints.md 2007-12-12 09:45:40.000000000 +0100
***************
*** 49,55 ****
;; O -- Multiple letter constraint followed by 1 parameter.
;; s: Signed extended immediate value (-2G .. 2G-1).
;; p: Positive extended immediate value (0 .. 4G-1).
! ;; n: Negative extended immediate value (-4G .. -1).
;; These constraints do not accept any operand if the machine does
;; not provide the extended-immediate facility.
;; P -- Any integer constant that can be loaded without literal pool.
--- 49,55 ----
;; O -- Multiple letter constraint followed by 1 parameter.
;; s: Signed extended immediate value (-2G .. 2G-1).
;; p: Positive extended immediate value (0 .. 4G-1).
! ;; n: Negative extended immediate value (-4G+1 .. -1).
;; These constraints do not accept any operand if the machine does
;; not provide the extended-immediate facility.
;; P -- Any integer constant that can be loaded without literal pool.
***************
*** 337,343 ****
(define_constraint "On"
"@internal
! Negative extended immediate value (-4G .. -1).
This constraint will only match if the machine provides
the extended-immediate facility."
(and (match_code "const_int")
--- 337,343 ----
(define_constraint "On"
"@internal
! Negative extended immediate value (-4G+1 .. -1).
This constraint will only match if the machine provides
the extended-immediate facility."
(and (match_code "const_int")
Index: gcc/testsuite/gcc.target/s390/20071212-1.c
===================================================================
*** /dev/null 1970-01-01 00:00:00.000000000 +0000
--- gcc/testsuite/gcc.target/s390/20071212-1.c 2007-12-12 09:56:28.000000000 +0100
***************
*** 0 ****
--- 1,11 ----
+ /* This used to fail due to bug in the On constraint causing a slgfi
+ to be emitted with an immediate not fitting into 32bit. */
+
+ /* { dg-do compile } */
+ /* { dg-options "-O3 -march=z9-109" } */
+
+ long
+ foo (long a)
+ {
+ return a - (1ULL << 32);
+ }