public inbox for gcc-cvs@sourceware.org
help / color / mirror / Atom feed
* [gcc(refs/users/marxin/heads/PR93274-make-resolver-static-v2)] i386: Fix up *avx_vperm_broadcast_v4df [PR93430]
@ 2020-01-27 10:16 Martin Liska
0 siblings, 0 replies; only message in thread
From: Martin Liska @ 2020-01-27 10:16 UTC (permalink / raw)
To: gcc-cvs
https://gcc.gnu.org/g:322db86f4b4df1261308e8a02e69018d9cea98e9
commit 322db86f4b4df1261308e8a02e69018d9cea98e9
Author: Jakub Jelinek <jakub@redhat.com>
Date: Sun Jan 26 12:10:48 2020 +0100
i386: Fix up *avx_vperm_broadcast_v4df [PR93430]
Apparently my recent patch which moved the *avx_vperm_broadcast* and
*vpermil* patterns before vpermpd broke the following testcase, the
define_insn_and_split matched always but the splitter condition only split
it if not -mavx2 for V4DFmode, basically relying on the vpermpd pattern to
come first.
The following patch fixes it by moving that part of SPLIT-CONDITION into
CONDITION, so that when it is not met, we just don't match the pattern
and thus match the later vpermpd pattern in that case.
Except, for { 0, 0, 0, 0 } permutation, there is actually no reason to do
that, vbroadcastsd from memory seems to be slightly cheaper than vpermpd $0.
2020-01-26 Jakub Jelinek <jakub@redhat.com>
PR target/93430
* config/i386/sse.md (*avx_vperm_broadcast_<mode>): Disallow for
TARGET_AVX2 and V4DFmode not in the split condition, but in the
pattern condition, though allow { 0, 0, 0, 0 } broadcast always.
* gcc.dg/pr93430.c: New test.
* gcc.target/i386/avx2-pr93430.c: New test.
Diff:
---
gcc/ChangeLog | 7 ++++++
gcc/config/i386/sse.md | 5 +++--
gcc/testsuite/ChangeLog | 4 ++++
gcc/testsuite/gcc.dg/pr93430.c | 33 ++++++++++++++++++++++++++++
gcc/testsuite/gcc.target/i386/avx2-pr93430.c | 5 +++++
5 files changed, 52 insertions(+), 2 deletions(-)
diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index e35efd6..db62a7a 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,10 @@
+2020-01-26 Jakub Jelinek <jakub@redhat.com>
+
+ PR target/93430
+ * config/i386/sse.md (*avx_vperm_broadcast_<mode>): Disallow for
+ TARGET_AVX2 and V4DFmode not in the split condition, but in the
+ pattern condition, though allow { 0, 0, 0, 0 } broadcast always.
+
2020-01-25 Feng Xue <fxue@os.amperecomputing.com>
PR ipa/93166
diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
index f2f4a4e..04a8c5e 100644
--- a/gcc/config/i386/sse.md
+++ b/gcc/config/i386/sse.md
@@ -19912,9 +19912,10 @@
(match_operand:VF_256 1 "nonimmediate_operand" "m,o,?v")
(match_parallel 2 "avx_vbroadcast_operand"
[(match_operand 3 "const_int_operand" "C,n,n")])))]
- "TARGET_AVX"
+ "TARGET_AVX
+ && (<MODE>mode != V4DFmode || !TARGET_AVX2 || operands[3] == const0_rtx)"
"#"
- "&& reload_completed && (<MODE>mode != V4DFmode || !TARGET_AVX2)"
+ "&& reload_completed"
[(set (match_dup 0) (vec_duplicate:VF_256 (match_dup 1)))]
{
rtx op0 = operands[0], op1 = operands[1];
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index 99da3cc..860d56d 100644
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,5 +1,9 @@
2020-01-26 Jakub Jelinek <jakub@redhat.com>
+ PR target/93430
+ * gcc.dg/pr93430.c: New test.
+ * gcc.target/i386/avx2-pr93430.c: New test.
+
PR ipa/93166
* g++.dg/pr93166.C: Move to ...
* g++.dg/pr93166_0.C: ... here. Turn it into a proper lto test.
diff --git a/gcc/testsuite/gcc.dg/pr93430.c b/gcc/testsuite/gcc.dg/pr93430.c
new file mode 100644
index 0000000..92c0b9f
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/pr93430.c
@@ -0,0 +1,33 @@
+/* PR target/93430 */
+/* { dg-do compile } */
+/* { dg-options "-O2" } */
+/* { dg-additional-options "-mavx -mno-avx2" { target avx } } */
+
+typedef double V __attribute__((vector_size (4 * sizeof (double))));
+typedef long long VI __attribute__((vector_size (4 * sizeof (long long))));
+
+#if __SIZEOF_DOUBLE__ == __SIZEOF_LONG_LONG__
+void
+foo (V *x, V *y)
+{
+ y[0] = __builtin_shuffle (x[0], x[0], (VI) { 0, 0, 0, 0 });
+}
+
+void
+bar (V *x, V *y)
+{
+ y[0] = __builtin_shuffle (x[0], x[0], (VI) { 1, 1, 1, 1 });
+}
+
+void
+baz (V *x, V *y)
+{
+ y[0] = __builtin_shuffle (x[0], x[0], (VI) { 2, 2, 2, 2 });
+}
+
+void
+qux (V *x, V *y)
+{
+ y[0] = __builtin_shuffle (x[0], x[0], (VI) { 3, 3, 3, 3 });
+}
+#endif
diff --git a/gcc/testsuite/gcc.target/i386/avx2-pr93430.c b/gcc/testsuite/gcc.target/i386/avx2-pr93430.c
new file mode 100644
index 0000000..d7c1a51
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/avx2-pr93430.c
@@ -0,0 +1,5 @@
+/* PR target/93430 */
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx2" } */
+
+#include "../../gcc.dg/pr93430.c"
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2020-01-27 10:16 UTC | newest]
Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-01-27 10:16 [gcc(refs/users/marxin/heads/PR93274-make-resolver-static-v2)] i386: Fix up *avx_vperm_broadcast_v4df [PR93430] Martin Liska
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).