The patterns for the svcmp_wide intrinsics used a VNx16BI
input predicate for all modes, instead of the usual <VPRED>.
That unnecessarily made some input bits significant, but more
importantly, it triggered an ICE in aarch64_sve_same_pred_for_ptest_p
when testing whether a comparison pattern could be fused with a PTEST.
A later patch will add tests for other comparisons.
gcc/
* config/aarch64/aarch64-sve.md (@aarch64_pred_cmp<cmp_op><mode>_wide)
(*aarch64_pred_cmp<cmp_op><mode>_wide_cc): Use <VPRED> instead of
VNx16BI for the governing predicate.
(*aarch64_pred_cmp<cmp_op><mode>_wide_ptest): Likewise.
gcc/testsuite/
* gcc.target/aarch64/sve/acle/general/cmpeq_1.c: Add more tests.
---
gcc/config/aarch64/aarch64-sve.md | 6 +--
.../aarch64/sve/acle/general/cmpeq_1.c | 53 ++++++++++++++++++-
2 files changed, 55 insertions(+), 4 deletions(-)
diff --git a/gcc/config/aarch64/aarch64-sve.md
b/gcc/config/aarch64/aarch64-sve.md
index 5486bafb4a6..997c340a725 100644
--- a/gcc/config/aarch64/aarch64-sve.md
+++ b/gcc/config/aarch64/aarch64-sve.md
@@ -8609,7 +8609,7 @@ (define_insn_and_split "*cmp<cmp_op><mode>_acle_and"
(define_insn "@aarch64_pred_cmp<cmp_op><mode>_wide"
[(set (match_operand:<VPRED> 0 "register_operand")
(unspec:<VPRED>
- [(match_operand:VNx16BI 1 "register_operand")
+ [(match_operand:<VPRED> 1 "register_operand")
(match_operand:SI 2 "aarch64_sve_ptrue_flag")
(unspec:<VPRED>
[(match_operand:SVE_FULL_BHSI 3 "register_operand")
@@ -8634,7 +8634,7 @@ (define_insn "*aarch64_pred_cmp<cmp_op><mode>_wide_cc"
(match_operand 4)
(match_operand:SI 5 "aarch64_sve_ptrue_flag")
(unspec:<VPRED>
- [(match_operand:VNx16BI 6 "register_operand")
+ [(match_operand:<VPRED> 6 "register_operand")
(match_operand:SI 7 "aarch64_sve_ptrue_flag")
(unspec:<VPRED>
[(match_operand:SVE_FULL_BHSI 2 "register_operand")
@@ -8669,7 +8669,7 @@ (define_insn "*aarch64_pred_cmp<cmp_op><mode>_wide_ptest"
(match_operand 4)
(match_operand:SI 5 "aarch64_sve_ptrue_flag")
(unspec:<VPRED>
- [(match_operand:VNx16BI 6 "register_operand")
+ [(match_operand:<VPRED> 6 "register_operand")
(match_operand:SI 7 "aarch64_sve_ptrue_flag")
(unspec:<VPRED>
[(match_operand:SVE_FULL_BHSI 2 "register_operand")
diff --git a/gcc/testsuite/gcc.target/aarch64/sve/acle/general/cmpeq_1.c
b/gcc/testsuite/gcc.target/aarch64/sve/acle/general/cmpeq_1.c
index dd8f6c49424..c73d10959fa 100644
--- a/gcc/testsuite/gcc.target/aarch64/sve/acle/general/cmpeq_1.c
+++ b/gcc/testsuite/gcc.target/aarch64/sve/acle/general/cmpeq_1.c
@@ -18,5 +18,56 @@ test2 (svbool_t pg, svint8_t x, svint64_t y, int *any)
return svptest_any (pg, res);
}
-/* { dg-final { scan-assembler-times {\tcmpeq\t} 2 } } */
+void
+test3 (svint8_t x, svint64_t y, int *any, svbool_t *ptr)
+{
+ svbool_t pg = svptrue_b8 ();
+ svbool_t res = svcmpeq_wide (pg, x, y);
+ *any = svptest_any (pg, res);
+ *ptr = res;
+}
+
+int
+test4 (svint8_t x, svint64_t y, int *any)
+{
+ svbool_t pg = svptrue_b8 ();
+ svbool_t res = svcmpeq_wide (pg, x, y);
+ return svptest_any (pg, res);
+}
+
+void
+test5 (svint16_t x, svint64_t y, int *any, svbool_t *ptr)
+{
+ svbool_t pg = svptrue_b16 ();
+ svbool_t res = svcmpeq_wide (pg, x, y);
+ *any = svptest_any (pg, res);
+ *ptr = res;
+}
+
+int
+test6 (svint16_t x, svint64_t y, int *any)
+{
+ svbool_t pg = svptrue_b16 ();
+ svbool_t res = svcmpeq_wide (pg, x, y);
+ return svptest_any (pg, res);
+}
+
+void
+test7 (svint32_t x, svint64_t y, int *any, svbool_t *ptr)
+{
+ svbool_t pg = svptrue_b32 ();
+ svbool_t res = svcmpeq_wide (pg, x, y);
+ *any = svptest_any (pg, res);
+ *ptr = res;
+}
+
+int
+test8 (svint32_t x, svint64_t y, int *any)
+{
+ svbool_t pg = svptrue_b32 ();
+ svbool_t res = svcmpeq_wide (pg, x, y);
+ return svptest_any (pg, res);
+}
+
+/* { dg-final { scan-assembler-times {\tcmpeq\t} 8 } } */
/* { dg-final { scan-assembler-not {\tptest\t} } } */
--
2.43.0
