The vector popcount expanders use a hardcoded subreg. This might lead
to double subregs being generated which then fail to match. With this
patch simplify_gen_subreg is used instead to fold the subregs.
Boostrapped and regression-tested on IBM z14.
Committed to mainline
gcc/ChangeLog:
2020-04-20 Andreas Krebbel <kreb...@linux.ibm.com>
* config/s390/vector.md ("popcountv8hi2_vx", "popcountv4si2_vx")
("popcountv2di2_vx"): Use simplify_gen_subreg.
gcc/testsuite/ChangeLog:
2020-04-20 Andreas Krebbel <kreb...@linux.ibm.com>
* g++.dg/pr94666.C: New test.
---
gcc/ChangeLog | 5 +++++
gcc/config/s390/vector.md | 10 +++++++---
gcc/testsuite/ChangeLog | 4 ++++
gcc/testsuite/g++.dg/pr94666.C | 20 ++++++++++++++++++++
4 files changed, 36 insertions(+), 3 deletions(-)
create mode 100644 gcc/testsuite/g++.dg/pr94666.C
diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index ed3541d1264..721928d931d 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,8 @@
+2020-04-20 Andreas Krebbel <kreb...@linux.ibm.com>
+
+ * config/s390/vector.md ("popcountv8hi2_vx", "popcountv4si2_vx")
+ ("popcountv2di2_vx"): Use simplify_gen_subreg.
+
2020-04-20 Andreas Krebbel <kreb...@linux.ibm.com>
PR target/94613
diff --git a/gcc/config/s390/vector.md b/gcc/config/s390/vector.md
index 9d3df9813c9..08f2d4cbda6 100644
--- a/gcc/config/s390/vector.md
+++ b/gcc/config/s390/vector.md
@@ -871,7 +871,7 @@
(define_expand "popcountv8hi2_vx"
[(set (match_dup 2)
- (unspec:V16QI [(subreg:V16QI (match_operand:V8HI 1 "register_operand"
"v") 0)]
+ (unspec:V16QI [(match_operand:V8HI 1 "register_operand" "v")]
UNSPEC_POPCNT))
; Make a copy of the result
(set (match_dup 3) (match_dup 2))
@@ -903,6 +903,8 @@
]
"TARGET_VX && !TARGET_VXE"
{
+ operands[1] = simplify_gen_subreg (V16QImode, operands[1],
+ V8HImode, 0);
operands[2] = gen_reg_rtx (V16QImode);
operands[3] = gen_reg_rtx (V16QImode);
operands[4] = gen_reg_rtx (V16QImode);
@@ -911,20 +913,21 @@
(define_expand "popcountv4si2_vx"
[(set (match_dup 2)
- (unspec:V16QI [(subreg:V16QI (match_operand:V4SI 1 "register_operand"
"v") 0)]
+ (unspec:V16QI [(match_operand:V4SI 1 "register_operand" "v")]
UNSPEC_POPCNT))
(set (match_operand:V4SI 0 "register_operand" "=v")
(unspec:V4SI [(match_dup 2) (match_dup 3)]
UNSPEC_VEC_VSUM))]
"TARGET_VX && !TARGET_VXE"
{
+ operands[1] = simplify_gen_subreg (V16QImode, operands[1], V4SImode, 0);
operands[2] = gen_reg_rtx (V16QImode);
operands[3] = force_reg (V16QImode, CONST0_RTX (V16QImode));
})
(define_expand "popcountv2di2_vx"
[(set (match_dup 2)
- (unspec:V16QI [(subreg:V16QI (match_operand:V2DI 1 "register_operand"
"v") 0)]
+ (unspec:V16QI [(match_operand:V2DI 1 "register_operand" "v")]
UNSPEC_POPCNT))
(set (match_dup 3)
(unspec:V4SI [(match_dup 2) (match_dup 4)]
@@ -934,6 +937,7 @@
UNSPEC_VEC_VSUMG))]
"TARGET_VX && !TARGET_VXE"
{
+ operands[1] = simplify_gen_subreg (V16QImode, operands[1], V2DImode, 0);
operands[2] = gen_reg_rtx (V16QImode);
operands[3] = gen_reg_rtx (V4SImode);
operands[4] = force_reg (V16QImode, CONST0_RTX (V16QImode));
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index 117eed9e35b..ce31ed0cd44 100644
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,7 @@
+2020-04-20 Andreas Krebbel <kreb...@linux.ibm.com>
+
+ * g++.dg/pr94666.C: New test.
+
2020-04-20 Andreas Krebbel <kreb...@linux.ibm.com>
PR target/94613
diff --git a/gcc/testsuite/g++.dg/pr94666.C b/gcc/testsuite/g++.dg/pr94666.C
new file mode 100644
index 00000000000..a9bfb24f795
--- /dev/null
+++ b/gcc/testsuite/g++.dg/pr94666.C
@@ -0,0 +1,20 @@
+// { dg-do compile }
+// { dg-options "-O3" }
+// { dg-additional-options "-march=z13" { target s390*-*-* } }
+
+int a, c;
+struct A {
+ int e() {
+ int f;
+ for (int b = 0; b < 4; b++) {
+ a = __builtin_popcountl(d[b]);
+ f += a;
+ }
+ return f;
+ }
+ long d[4];
+} * g;
+void h() {
+ for (int b; b; b++)
+ c += g[b].e();
+}
--
2.17.1
