This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] [ARM] Fix widen-sum pattern in neon.md.
- From: Xingxing Pan <xxingpan at marvell dot com>
- To: "ramrad01 at arm dot com" <ramrad01 at arm dot com>
- Cc: Julian Brown <julian at codesourcery dot com>, James Greenhalgh <james dot greenhalgh at arm dot com>, Kyrill Tkachov <kyrylo dot tkachov at arm dot com>, Ramana Radhakrishnan <ramana dot radhakrishnan at arm dot com>, Richard Earnshaw <richard dot earnshaw at arm dot com>, "nickc at redhat dot com" <nickc at redhat dot com>, Xinyu Qi <xyqi at marvell dot com>, Liping Gao <lgao1 at marvell dot com>, Joey Ye <joey dot ye at arm dot com>, "gcc-patches at gcc dot gnu dot org" <gcc-patches at gcc dot gnu dot org>
- Date: Mon, 20 Apr 2015 14:03:38 +0800
- Subject: Re: [PATCH] [ARM] Fix widen-sum pattern in neon.md.
- Authentication-results: sourceware.org; auth=none
- References: <54F85B61 dot 8090901 at marvell dot com> <CAJA7tRZBJMaHdHub+TqYPo9VzwpA9y-TjEGXSGPvbiLNK-1Xgw at mail dot gmail dot com>
On 04/15/2015 03:13 AM, Ramana Radhakrishnan wrote:
On Thu, Mar 5, 2015 at 1:34 PM, Xingxing Pan <xxingpan@marvell.com> wrote:
Hi,
The expanding of widen-sum pattern always fails. The vectorizer expects the
operands to have the same size, while the current implementation of
widen-sum pattern dose not conform to this.
This patch implements the widen-sum pattern with vpadal. Change the vaddw
pattern to anonymous. Add widen-sum test cases for neon.
Can you please respin addressing James and Kyrill's comments ?
Ramana
--
Regards,
Xingxing
Hi,
Sorry for late response.
The pattern is rewritten to utilize neon_vpadal<sup><mode>'s "0"
constraints. Have run vect.exp and neon.exp in an armv7 board.
vect.exp has two new XFAILs:
XFAIL: gcc.dg/vect/slp-reduc-3.c scan-tree-dump-times vect "vectorizing
stmts using SLP" 1
XFAIL: gcc.dg/vect/slp-reduc-3.c -flto -ffat-lto-objects
scan-tree-dump-times vect "vectorizing stmts using SLP" 1
This is because widen-sum optimization precedes SLP. The xfail predicate
vect_widen_sum_hi_to_si becomes true when widen-sum is enabled.
neon.exp has four new XFAILs:
XFAIL: gcc.target/arm/neon/vect-widen-sum-char2short-s-d.c
scan-tree-dump-times vect "pattern recognized.*w\\+" 1
XFAIL: gcc.target/arm/neon/vect-widen-sum-char2short-s-d.c
scan-rtl-dump-times expand "UNSPEC_VPADAL" 1
XFAIL: gcc.target/arm/neon/vect-widen-sum-char2short-s.c
scan-tree-dump-times vect "pattern recognized.*w\\+" 1
XFAIL: gcc.target/arm/neon/vect-widen-sum-char2short-s.c
scan-rtl-dump-times expand "UNSPEC_VPADAL" 1
If the widen-sum pattern is successfully expanded, "w+" and
"UNSPEC_VPADAL" should appear in the dump file like other
vect-widen-sum-*.c tests. But vect-widen-sum-char2short-s[-d].c is
special because at tree level the signed operations will be converted
into unsigned operations, which destroy the widen-sum pattern. That is
due to the workaround of PR tree-optimization/25125. I just add xfail
following gcc.dg/vect/vect-reduc-pattern-2c.c.
--
Regards,
Xingxing
commit c44b5bd19efb029b8bbd4e3c7e2d631bdc482b7c
Author: Xingxing Pan <xxingpan@marvell.com>
Date: Sun Apr 19 15:54:43 2015 +0800
Fix widen-sum pattern in neon.md.
gcc/
2015-04-19 Xingxing Pan <xxingpan@marvell.com>
* config/arm/iterators.md (VWSD): New.
(V_widen_sum_d): New.
* config/arm/neon.md (widen_ssum<mode>3): Redefined.
(widen_usum<mode>3): Ditto.
(neon_svaddw<mode>3): New anonymous define_insn.
(neon_uvaddw<mode>3): Ditto.
gcc/testsuite/
2015-04-19 Xingxing Pan <xxingpan@marvell.com>
* gcc.target/arm/neon/vect-widen-sum-char2short-s-d.c: New.
* gcc.target/arm/neon/vect-widen-sum-char2short-s.c: New.
* gcc.target/arm/neon/vect-widen-sum-char2short-u-d.c: New.
* gcc.target/arm/neon/vect-widen-sum-char2short-u.c: New.
* gcc.target/arm/neon/vect-widen-sum-short2int-s-d.c: New.
* gcc.target/arm/neon/vect-widen-sum-short2int-s.c: New.
* gcc.target/arm/neon/vect-widen-sum-short2int-u-d.c: New.
* gcc.target/arm/neon/vect-widen-sum-short2int-u.c: New.
* lib/target-supports.exp
(check_effective_target_vect_widen_sum_hi_to_si_pattern): Return 1 for ARM NEON.
(check_effective_target_vect_widen_sum_hi_to_si): Ditto.
(check_effective_target_vect_widen_sum_qi_to_hi): Ditto.
diff --git a/gcc/config/arm/iterators.md b/gcc/config/arm/iterators.md
index f7f8ab7..f73278d 100644
--- a/gcc/config/arm/iterators.md
+++ b/gcc/config/arm/iterators.md
@@ -95,6 +95,9 @@
;; Widenable modes.
(define_mode_iterator VW [V8QI V4HI V2SI])
+;; Widenable modes. Used by widen sum.
+(define_mode_iterator VWSD [V8QI V4HI V16QI V8HI])
+
;; Narrowable modes.
(define_mode_iterator VN [V8HI V4SI V2DI])
@@ -555,9 +558,14 @@
;; Same as V_widen, but lower-case.
(define_mode_attr V_widen_l [(V8QI "v8hi") (V4HI "v4si") ( V2SI "v2di")])
-;; Widen. Result is half the number of elements, but widened to double-width.
+;; Widen. Result is half the number of elements, but widened to double-width.
(define_mode_attr V_unpack [(V16QI "V8HI") (V8HI "V4SI") (V4SI "V2DI")])
+;; Widen. Result is half the number of elements, but widened to double-width.
+;; Used by widen sum.
+(define_mode_attr V_widen_sum_d [(V8QI "V4HI") (V4HI "V2SI")
+ (V16QI "V8HI") (V8HI "V4SI")])
+
;; Conditions to be used in extend<mode>di patterns.
(define_mode_attr qhs_zextenddi_cond [(SI "") (HI "&& arm_arch6") (QI "")])
(define_mode_attr qhs_sextenddi_cond [(SI "") (HI "&& arm_arch6")
diff --git a/gcc/config/arm/neon.md b/gcc/config/arm/neon.md
index 63c327e..839883f 100644
--- a/gcc/config/arm/neon.md
+++ b/gcc/config/arm/neon.md
@@ -1174,7 +1174,29 @@
;; Widening operations
-(define_insn "widen_ssum<mode>3"
+(define_expand "widen_usum<mode>3"
+ [(match_operand:<V_widen_sum_d> 0 "s_register_operand" "")
+ (match_operand:VWSD 1 "s_register_operand" "")
+ (match_operand:<V_widen_sum_d> 2 "s_register_operand" "")]
+ "TARGET_NEON"
+ {
+ emit_insn (gen_neon_vpadalu<mode> (operands[0], operands[2], operands[1]));
+ DONE;
+ }
+)
+
+(define_expand "widen_ssum<mode>3"
+ [(match_operand:<V_widen_sum_d> 0 "s_register_operand" "")
+ (match_operand:VWSD 1 "s_register_operand" "")
+ (match_operand:<V_widen_sum_d> 2 "s_register_operand" "")]
+ "TARGET_NEON"
+ {
+ emit_insn (gen_neon_vpadals<mode> (operands[0], operands[2], operands[1]));
+ DONE;
+ }
+)
+
+(define_insn "*neon_svaddw<mode>3"
[(set (match_operand:<V_widen> 0 "s_register_operand" "=w")
(plus:<V_widen> (sign_extend:<V_widen>
(match_operand:VW 1 "s_register_operand" "%w"))
@@ -1184,7 +1206,7 @@
[(set_attr "type" "neon_add_widen")]
)
-(define_insn "widen_usum<mode>3"
+(define_insn "*neon_uvaddw<mode>3"
[(set (match_operand:<V_widen> 0 "s_register_operand" "=w")
(plus:<V_widen> (zero_extend:<V_widen>
(match_operand:VW 1 "s_register_operand" "%w"))
diff --git a/gcc/testsuite/gcc.target/arm/neon/vect-widen-sum-char2short-s-d.c b/gcc/testsuite/gcc.target/arm/neon/vect-widen-sum-char2short-s-d.c
new file mode 100644
index 0000000..8d0278c
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/neon/vect-widen-sum-char2short-s-d.c
@@ -0,0 +1,63 @@
+/* { dg-do run } */
+/* { dg-require-effective-target arm_neon_hw } */
+/* { dg-options "-O2 -ffast-math -ftree-vectorize -mvectorize-with-neon-double -fdump-tree-vect-details -fdump-rtl-expand" } */
+/* { dg-add-options arm_neon } */
+
+/* { dg-final { scan-tree-dump-times "pattern recognized.*w\\\+" 1 "vect" { xfail *-*-* } } } */
+/* { dg-final { cleanup-tree-dump "vect" } } */
+/* { dg-final { scan-rtl-dump-times "UNSPEC_VPADAL" 1 "expand" { xfail *-*-* } } } */
+/* { dg-final { cleanup-rtl-dump "expand" } } */
+
+typedef signed char STYPE1;
+typedef signed short STYPE2;
+extern void abort (void);
+
+#define N 128
+STYPE1 sdata[N];
+
+volatile int y = 0;
+
+__attribute__ ((noinline)) int
+ssum ()
+{
+ int i;
+ STYPE2 sum = 0;
+ STYPE2 check_sum = 0;
+
+ /* widenning sum: sum chars into short.
+
+ Like gcc.dg/vect/vect-reduc-pattern-2c.c, the widening-summation pattern
+ is currently not detected because of this patch:
+
+ 2005-12-26 Kazu Hirata <kazu@codesourcery.com>
+ PR tree-optimization/25125
+ */
+
+ for (i = 0; i < N; i++)
+ {
+ sdata[i] = i*2;
+ check_sum += sdata[i];
+ /* Avoid vectorization. */
+ if (y)
+ abort ();
+ }
+
+ /* widenning sum: sum chars into int. */
+ for (i = 0; i < N; i++)
+ {
+ sum += sdata[i];
+ }
+
+ /* check results: */
+ if (sum != check_sum)
+ abort ();
+
+ return 0;
+}
+
+int
+main (void)
+{
+ ssum ();
+ return 0;
+}
diff --git a/gcc/testsuite/gcc.target/arm/neon/vect-widen-sum-char2short-s.c b/gcc/testsuite/gcc.target/arm/neon/vect-widen-sum-char2short-s.c
new file mode 100644
index 0000000..f7384c3
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/neon/vect-widen-sum-char2short-s.c
@@ -0,0 +1,63 @@
+/* { dg-do run } */
+/* { dg-require-effective-target arm_neon_hw } */
+/* { dg-options "-O2 -ffast-math -ftree-vectorize -fdump-tree-vect-details -fdump-rtl-expand" } */
+/* { dg-add-options arm_neon } */
+
+/* { dg-final { scan-tree-dump-times "pattern recognized.*w\\\+" 1 "vect" { xfail *-*-* } } } */
+/* { dg-final { cleanup-tree-dump "vect" } } */
+/* { dg-final { scan-rtl-dump-times "UNSPEC_VPADAL" 1 "expand" { xfail *-*-* } } } */
+/* { dg-final { cleanup-rtl-dump "expand" } } */
+
+typedef signed char STYPE1;
+typedef signed short STYPE2;
+extern void abort (void);
+
+#define N 128
+STYPE1 sdata[N];
+
+volatile int y = 0;
+
+__attribute__ ((noinline)) int
+ssum ()
+{
+ int i;
+ STYPE2 sum = 0;
+ STYPE2 check_sum = 0;
+
+ /* widenning sum: sum chars into short.
+
+ Like gcc.dg/vect/vect-reduc-pattern-2c.c, the widening-summation pattern
+ is currently not detected because of this patch:
+
+ 2005-12-26 Kazu Hirata <kazu@codesourcery.com>
+ PR tree-optimization/25125
+ */
+
+ for (i = 0; i < N; i++)
+ {
+ sdata[i] = i*2;
+ check_sum += sdata[i];
+ /* Avoid vectorization. */
+ if (y)
+ abort ();
+ }
+
+ /* widenning sum: sum chars into int. */
+ for (i = 0; i < N; i++)
+ {
+ sum += sdata[i];
+ }
+
+ /* check results: */
+ if (sum != check_sum)
+ abort ();
+
+ return 0;
+}
+
+int
+main (void)
+{
+ ssum ();
+ return 0;
+}
diff --git a/gcc/testsuite/gcc.target/arm/neon/vect-widen-sum-char2short-u-d.c b/gcc/testsuite/gcc.target/arm/neon/vect-widen-sum-char2short-u-d.c
new file mode 100644
index 0000000..35f8fa7
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/neon/vect-widen-sum-char2short-u-d.c
@@ -0,0 +1,54 @@
+/* { dg-do run } */
+/* { dg-require-effective-target arm_neon_hw } */
+/* { dg-options "-O2 -ffast-math -ftree-vectorize -mvectorize-with-neon-double -fdump-tree-vect-details -fdump-rtl-expand" } */
+/* { dg-add-options arm_neon } */
+
+/* { dg-final { scan-tree-dump-times "pattern recognized.*w\\\+" 1 "vect" { target { arm_neon } } } } */
+/* { dg-final { cleanup-tree-dump "vect" } } */
+/* { dg-final { scan-rtl-dump-times "UNSPEC_VPADAL" 1 "expand" { target { arm_neon } } } } */
+/* { dg-final { cleanup-rtl-dump "expand" } } */
+
+typedef unsigned char UTYPE1;
+typedef unsigned short UTYPE2;
+extern void abort (void);
+
+#define N 128
+UTYPE1 udata[N];
+
+volatile int y = 0;
+
+__attribute__ ((noinline)) int
+usum ()
+{
+ int i;
+ UTYPE2 sum = 0;
+ UTYPE2 check_sum = 0;
+
+ for (i = 0; i < N; i++)
+ {
+ udata[i] = i*2;
+ check_sum += udata[i];
+ /* Avoid vectorization. */
+ if (y)
+ abort ();
+ }
+
+ /* widenning sum: sum chars into int. */
+ for (i = 0; i < N; i++)
+ {
+ sum += udata[i];
+ }
+
+ /* check results: */
+ if (sum != check_sum)
+ abort ();
+
+ return 0;
+}
+
+int
+main (void)
+{
+ usum ();
+ return 0;
+}
diff --git a/gcc/testsuite/gcc.target/arm/neon/vect-widen-sum-char2short-u.c b/gcc/testsuite/gcc.target/arm/neon/vect-widen-sum-char2short-u.c
new file mode 100644
index 0000000..38af5f0
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/neon/vect-widen-sum-char2short-u.c
@@ -0,0 +1,54 @@
+/* { dg-do run } */
+/* { dg-require-effective-target arm_neon_hw } */
+/* { dg-options "-O2 -ffast-math -ftree-vectorize -fdump-tree-vect-details -fdump-rtl-expand" } */
+/* { dg-add-options arm_neon } */
+
+/* { dg-final { scan-tree-dump-times "pattern recognized.*w\\\+" 1 "vect" { target { arm_neon } } } } */
+/* { dg-final { cleanup-tree-dump "vect" } } */
+/* { dg-final { scan-rtl-dump-times "UNSPEC_VPADAL" 1 "expand" { target { arm_neon } } } } */
+/* { dg-final { cleanup-rtl-dump "expand" } } */
+
+typedef unsigned char UTYPE1;
+typedef unsigned short UTYPE2;
+extern void abort (void);
+
+#define N 128
+UTYPE1 udata[N];
+
+volatile int y = 0;
+
+__attribute__ ((noinline)) int
+usum ()
+{
+ int i;
+ UTYPE2 sum = 0;
+ UTYPE2 check_sum = 0;
+
+ for (i = 0; i < N; i++)
+ {
+ udata[i] = i*2;
+ check_sum += udata[i];
+ /* Avoid vectorization. */
+ if (y)
+ abort ();
+ }
+
+ /* widenning sum: sum chars into int. */
+ for (i = 0; i < N; i++)
+ {
+ sum += udata[i];
+ }
+
+ /* check results: */
+ if (sum != check_sum)
+ abort ();
+
+ return 0;
+}
+
+int
+main (void)
+{
+ usum ();
+ return 0;
+}
diff --git a/gcc/testsuite/gcc.target/arm/neon/vect-widen-sum-short2int-s-d.c b/gcc/testsuite/gcc.target/arm/neon/vect-widen-sum-short2int-s-d.c
new file mode 100644
index 0000000..ef765de
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/neon/vect-widen-sum-short2int-s-d.c
@@ -0,0 +1,54 @@
+/* { dg-do run } */
+/* { dg-require-effective-target arm_neon_hw } */
+/* { dg-options "-O2 -ffast-math -ftree-vectorize -mvectorize-with-neon-double -fdump-tree-vect-details -fdump-rtl-expand" } */
+/* { dg-add-options arm_neon } */
+
+/* { dg-final { scan-tree-dump-times "pattern recognized.*w\\\+" 1 "vect" { target { arm_neon } } } } */
+/* { dg-final { cleanup-tree-dump "vect" } } */
+/* { dg-final { scan-rtl-dump-times "UNSPEC_VPADAL" 1 "expand" { target arm_neon } } } */
+/* { dg-final { cleanup-rtl-dump "expand" } } */
+
+typedef signed short STYPE1;
+typedef signed int STYPE2;
+extern void abort (void);
+
+#define N 128
+STYPE1 sdata[N];
+
+volatile int y = 0;
+
+__attribute__ ((noinline)) int
+ssum ()
+{
+ int i;
+ STYPE2 sum = 0;
+ STYPE2 check_sum = 0;
+
+ for (i = 0; i < N; i++)
+ {
+ sdata[i] = i*2;
+ check_sum += sdata[i];
+ /* Avoid vectorization. */
+ if (y)
+ abort ();
+ }
+
+ /* widenning sum: sum chars into int. */
+ for (i = 0; i < N; i++)
+ {
+ sum += sdata[i];
+ }
+
+ /* check results: */
+ if (sum != check_sum)
+ abort ();
+
+ return 0;
+}
+
+int
+main (void)
+{
+ ssum ();
+ return 0;
+}
diff --git a/gcc/testsuite/gcc.target/arm/neon/vect-widen-sum-short2int-s.c b/gcc/testsuite/gcc.target/arm/neon/vect-widen-sum-short2int-s.c
new file mode 100644
index 0000000..fb38d56
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/neon/vect-widen-sum-short2int-s.c
@@ -0,0 +1,54 @@
+/* { dg-do run } */
+/* { dg-require-effective-target arm_neon_hw } */
+/* { dg-options "-O2 -ffast-math -ftree-vectorize -fdump-tree-vect-details -fdump-rtl-expand" } */
+/* { dg-add-options arm_neon } */
+
+/* { dg-final { scan-tree-dump-times "pattern recognized.*w\\\+" 1 "vect" { target { arm_neon } } } } */
+/* { dg-final { cleanup-tree-dump "vect" } } */
+/* { dg-final { scan-rtl-dump-times "UNSPEC_VPADAL" 1 "expand" { target { arm_neon } } } } */
+/* { dg-final { cleanup-rtl-dump "expand" } } */
+
+typedef signed short STYPE1;
+typedef signed int STYPE2;
+extern void abort (void);
+
+#define N 128
+STYPE1 sdata[N];
+
+volatile int y = 0;
+
+__attribute__ ((noinline)) int
+ssum ()
+{
+ int i;
+ STYPE2 sum = 0;
+ STYPE2 check_sum = 0;
+
+ for (i = 0; i < N; i++)
+ {
+ sdata[i] = i*2;
+ check_sum += sdata[i];
+ /* Avoid vectorization. */
+ if (y)
+ abort ();
+ }
+
+ /* widenning sum: sum chars into int. */
+ for (i = 0; i < N; i++)
+ {
+ sum += sdata[i];
+ }
+
+ /* check results: */
+ if (sum != check_sum)
+ abort ();
+
+ return 0;
+}
+
+int
+main (void)
+{
+ ssum ();
+ return 0;
+}
diff --git a/gcc/testsuite/gcc.target/arm/neon/vect-widen-sum-short2int-u-d.c b/gcc/testsuite/gcc.target/arm/neon/vect-widen-sum-short2int-u-d.c
new file mode 100644
index 0000000..5a3dfd6
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/neon/vect-widen-sum-short2int-u-d.c
@@ -0,0 +1,54 @@
+/* { dg-do run } */
+/* { dg-require-effective-target arm_neon_hw } */
+/* { dg-options "-O2 -ffast-math -ftree-vectorize -mvectorize-with-neon-double -fdump-tree-vect-details -fdump-rtl-expand" } */
+/* { dg-add-options arm_neon } */
+
+/* { dg-final { scan-tree-dump-times "pattern recognized.*w\\\+" 1 "vect" { target { arm_neon } } } } */
+/* { dg-final { cleanup-tree-dump "vect" } } */
+/* { dg-final { scan-rtl-dump-times "UNSPEC_VPADAL" 1 "expand" { target { arm_neon } } } } */
+/* { dg-final { cleanup-rtl-dump "expand" } } */
+
+typedef unsigned short UTYPE1;
+typedef unsigned int UTYPE2;
+extern void abort (void);
+
+#define N 128
+UTYPE1 udata[N];
+
+volatile int y = 0;
+
+__attribute__ ((noinline)) int
+usum ()
+{
+ int i;
+ UTYPE2 sum = 0;
+ UTYPE2 check_sum = 0;
+
+ for (i = 0; i < N; i++)
+ {
+ udata[i] = i*2;
+ check_sum += udata[i];
+ /* Avoid vectorization. */
+ if (y)
+ abort ();
+ }
+
+ /* widenning sum: sum chars into int. */
+ for (i = 0; i < N; i++)
+ {
+ sum += udata[i];
+ }
+
+ /* check results: */
+ if (sum != check_sum)
+ abort ();
+
+ return 0;
+}
+
+int
+main (void)
+{
+ usum ();
+ return 0;
+}
diff --git a/gcc/testsuite/gcc.target/arm/neon/vect-widen-sum-short2int-u.c b/gcc/testsuite/gcc.target/arm/neon/vect-widen-sum-short2int-u.c
new file mode 100644
index 0000000..770b08d
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/neon/vect-widen-sum-short2int-u.c
@@ -0,0 +1,54 @@
+/* { dg-do run } */
+/* { dg-require-effective-target arm_neon_hw } */
+/* { dg-options "-O2 -ffast-math -ftree-vectorize -fdump-tree-vect-details -fdump-rtl-expand" } */
+/* { dg-add-options arm_neon } */
+
+/* { dg-final { scan-tree-dump-times "pattern recognized.*w\\\+" 1 "vect" { target { arm_neon } } } } */
+/* { dg-final { cleanup-tree-dump "vect" } } */
+/* { dg-final { scan-rtl-dump-times "UNSPEC_VPADAL" 1 "expand" { target { arm_neon } } } } */
+/* { dg-final { cleanup-rtl-dump "expand" } } */
+
+typedef unsigned short UTYPE1;
+typedef unsigned int UTYPE2;
+extern void abort (void);
+
+#define N 128
+UTYPE1 udata[N];
+
+volatile int y = 0;
+
+__attribute__ ((noinline)) int
+usum ()
+{
+ int i;
+ UTYPE2 sum = 0;
+ UTYPE2 check_sum = 0;
+
+ for (i = 0; i < N; i++)
+ {
+ udata[i] = i*2;
+ check_sum += udata[i];
+ /* Avoid vectorization. */
+ if (y)
+ abort ();
+ }
+
+ /* widenning sum: sum chars into int. */
+ for (i = 0; i < N; i++)
+ {
+ sum += udata[i];
+ }
+
+ /* check results: */
+ if (sum != check_sum)
+ abort ();
+
+ return 0;
+}
+
+int
+main (void)
+{
+ usum ();
+ return 0;
+}
diff --git a/gcc/testsuite/lib/target-supports.exp b/gcc/testsuite/lib/target-supports.exp
index f632d00..477ab53 100644
--- a/gcc/testsuite/lib/target-supports.exp
+++ b/gcc/testsuite/lib/target-supports.exp
@@ -3795,6 +3795,7 @@ proc check_effective_target_vect_widen_sum_hi_to_si_pattern { } {
} else {
set et_vect_widen_sum_hi_to_si_pattern_saved 0
if { [istarget powerpc*-*-*]
+ || ([istarget arm*-*-*] && [check_effective_target_arm_neon_ok])
|| [istarget ia64-*-*] } {
set et_vect_widen_sum_hi_to_si_pattern_saved 1
}
@@ -3818,7 +3819,8 @@ proc check_effective_target_vect_widen_sum_hi_to_si { } {
} else {
set et_vect_widen_sum_hi_to_si_saved [check_effective_target_vect_unpack]
if { [istarget powerpc*-*-*]
- || [istarget ia64-*-*] } {
+ || ([istarget arm*-*-*] && [check_effective_target_arm_neon_ok])
+ || [istarget ia64-*-*] } {
set et_vect_widen_sum_hi_to_si_saved 1
}
}
@@ -3841,7 +3843,7 @@ proc check_effective_target_vect_widen_sum_qi_to_hi { } {
} else {
set et_vect_widen_sum_qi_to_hi_saved 0
if { [check_effective_target_vect_unpack]
- || [check_effective_target_arm_neon_ok]
+ || ([istarget arm*-*-*] && [check_effective_target_arm_neon_ok])
|| [istarget ia64-*-*] } {
set et_vect_widen_sum_qi_to_hi_saved 1
}