* [PATCH, rs6000] Fix PR78604
@ 2017-02-08 18:45 Pat Haugen
2017-02-08 19:53 ` Segher Boessenkool
0 siblings, 1 reply; 2+ messages in thread
From: Pat Haugen @ 2017-02-08 18:45 UTC (permalink / raw)
To: GCC Patches; +Cc: Segher Boessenkool, David Edelsohn, Bill Schmidt
The following patch fixes the pr by adding code for integer comparisons to invert the condition and swap operands to avoid generation of a compound test. Bootstrap/regtest on powerpc64-unknown-linux-gnu with no new regressions. Ok for trunk?
-Pat
2017-02-08 Pat Haugen <pthaugen@us.ibm.com>
PR target/78604
* config/rs6000/rs6000.c (rs6000_emit_vector_cond_expr): Invert
condition/operands for integer GE/LE/GEU/LEU operations.
testsuite/ChangeLog:
2017-02-08 Pat Haugen <pthaugen@us.ibm.com>
PR target/78604
* gcc.target/powerpc/pr78604.c: New.
Index: config/rs6000/rs6000.c
===================================================================
--- config/rs6000/rs6000.c (revision 245252)
+++ config/rs6000/rs6000.c (working copy)
@@ -25150,12 +25150,29 @@ rs6000_emit_vector_cond_expr (rtx dest,
return 0;
break;
- /* Mark unsigned tests with CCUNSmode. */
+ case GE:
+ case LE:
+ if (GET_MODE_CLASS (mask_mode) == MODE_VECTOR_INT)
+ {
+ /* Invert condition to avoid compound test. */
+ invert_move = true;
+ rcode = reverse_condition (rcode);
+ }
+ break;
+
case GTU:
case GEU:
case LTU:
case LEU:
+ /* Mark unsigned tests with CCUNSmode. */
cc_mode = CCUNSmode;
+
+ /* Invert condition to avoid compound test if necessary. */
+ if (rcode == GEU || rcode == LEU)
+ {
+ invert_move = true;
+ rcode = reverse_condition (rcode);
+ }
break;
default:
Index: testsuite/gcc.target/powerpc/pr78604.c
===================================================================
--- testsuite/gcc.target/powerpc/pr78604.c (revision 0)
+++ testsuite/gcc.target/powerpc/pr78604.c (working copy)
@@ -0,0 +1,112 @@
+/* { dg-do compile { target { powerpc*-*-* } } } */
+/* { dg-skip-if "" { powerpc*-*-darwin* } { "*" } { "" } } */
+/* { dg-require-effective-target powerpc_p8vector_ok } */
+/* { dg-skip-if "do not override -mcpu" { powerpc*-*-* } { "-mcpu=*" } { "-mcpu=power8" } } */
+/* { dg-options "-mcpu=power8 -O2 -ftree-vectorize" } */
+
+#ifndef SIZE
+#define SIZE 1024
+#endif
+
+#ifndef ALIGN
+#define ALIGN 32
+#endif
+
+#ifndef TYPE
+#define TYPE long long
+#endif
+
+#ifndef SIGN_TYPE
+#define SIGN_TYPE signed TYPE
+#endif
+
+#ifndef UNS_TYPE
+#define UNS_TYPE unsigned TYPE
+#endif
+
+#define ALIGN_ATTR __attribute__((__aligned__(ALIGN)))
+
+SIGN_TYPE sa[SIZE] ALIGN_ATTR;
+SIGN_TYPE sb[SIZE] ALIGN_ATTR;
+SIGN_TYPE sc[SIZE] ALIGN_ATTR;
+
+UNS_TYPE ua[SIZE] ALIGN_ATTR;
+UNS_TYPE ub[SIZE] ALIGN_ATTR;
+UNS_TYPE uc[SIZE] ALIGN_ATTR;
+
+void
+sign_lt (SIGN_TYPE val1, SIGN_TYPE val2)
+{
+ unsigned long i;
+
+ for (i = 0; i < SIZE; i++)
+ sa[i] = (sb[i] < sc[i]) ? val1 : val2;
+}
+
+void
+sign_lte (SIGN_TYPE val1, SIGN_TYPE val2)
+{
+ unsigned long i;
+
+ for (i = 0; i < SIZE; i++)
+ sa[i] = (sb[i] <= sc[i]) ? val1 : val2;
+}
+
+void
+sign_gt (SIGN_TYPE val1, SIGN_TYPE val2)
+{
+ unsigned long i;
+
+ for (i = 0; i < SIZE; i++)
+ sa[i] = (sb[i] > sc[i]) ? val1 : val2;
+}
+
+void
+sign_gte (SIGN_TYPE val1, SIGN_TYPE val2)
+{
+ unsigned long i;
+
+ for (i = 0; i < SIZE; i++)
+ sa[i] = (sb[i] >= sc[i]) ? val1 : val2;
+}
+
+
+void
+uns_lt (UNS_TYPE val1, UNS_TYPE val2)
+{
+ unsigned long i;
+
+ for (i = 0; i < SIZE; i++)
+ ua[i] = (ub[i] < uc[i]) ? val1 : val2;
+}
+
+void
+uns_lte (UNS_TYPE val1, UNS_TYPE val2)
+{
+ unsigned long i;
+
+ for (i = 0; i < SIZE; i++)
+ ua[i] = (ub[i] <= uc[i]) ? val1 : val2;
+}
+
+void
+uns_gt (UNS_TYPE val1, UNS_TYPE val2)
+{
+ unsigned long i;
+
+ for (i = 0; i < SIZE; i++)
+ ua[i] = (ub[i] > uc[i]) ? val1 : val2;
+}
+
+void
+uns_gte (UNS_TYPE val1, UNS_TYPE val2)
+{
+ unsigned long i;
+
+ for (i = 0; i < SIZE; i++)
+ ua[i] = (ub[i] >= uc[i]) ? val1 : val2;
+}
+
+/* { dg-final { scan-assembler-times {\mvcmpgtsd\M} 4 } } */
+/* { dg-final { scan-assembler-times {\mvcmpgtud\M} 4 } } */
+/* { dg-final { scan-assembler-not {\mvcmpequd\M} } } */
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH, rs6000] Fix PR78604
2017-02-08 18:45 [PATCH, rs6000] Fix PR78604 Pat Haugen
@ 2017-02-08 19:53 ` Segher Boessenkool
0 siblings, 0 replies; 2+ messages in thread
From: Segher Boessenkool @ 2017-02-08 19:53 UTC (permalink / raw)
To: Pat Haugen; +Cc: GCC Patches, David Edelsohn, Bill Schmidt
On Wed, Feb 08, 2017 at 12:45:22PM -0600, Pat Haugen wrote:
> The following patch fixes the pr by adding code for integer comparisons to invert the condition and swap operands to avoid generation of a compound test. Bootstrap/regtest on powerpc64-unknown-linux-gnu with no new regressions. Ok for trunk?
This had me confused for a bit, but you're not swapping the operands of
the condition, you're swapping the true/false operands of the cond
expression :-)
Okay for trunk. Thanks,
Segher
> 2017-02-08 Pat Haugen <pthaugen@us.ibm.com>
>
> PR target/78604
> * config/rs6000/rs6000.c (rs6000_emit_vector_cond_expr): Invert
> condition/operands for integer GE/LE/GEU/LEU operations.
>
>
> testsuite/ChangeLog:
> 2017-02-08 Pat Haugen <pthaugen@us.ibm.com>
>
> PR target/78604
> * gcc.target/powerpc/pr78604.c: New.
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2017-02-08 19:53 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-02-08 18:45 [PATCH, rs6000] Fix PR78604 Pat Haugen
2017-02-08 19:53 ` Segher Boessenkool
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).