From: Hongtao Liu <crazylht@gmail.com>
To: GCC Patches <gcc-patches@gcc.gnu.org>
Subject: Re: [PATCH] Simplify (view_convert ~a) < 0 to (view_convert a) >= 0 [PR middle-end/100738]
Date: Fri, 4 Jun 2021 16:11:21 +0800 [thread overview]
Message-ID: <CAMZc-bwKL1PMKpOgwcT+ierLv-JjATMvCmC_Ntb+J3=4t11BXA@mail.gmail.com> (raw)
In-Reply-To: <CAMZc-bwnHQhDZTZiO7D4EyL8WQ4mJ1GxaXUgy9e7M1=6LfefyA@mail.gmail.com>
[-- Attachment #1: Type: text/plain, Size: 2121 bytes --]
On Fri, Jun 4, 2021 at 1:01 PM Hongtao Liu <crazylht@gmail.com> wrote:
>
> On Tue, Jun 1, 2021 at 6:17 PM Marc Glisse <marc.glisse@inria.fr> wrote:
> >
> > On Tue, 1 Jun 2021, Hongtao Liu via Gcc-patches wrote:
> >
> > > Hi:
> > > This patch is about to simplify (view_convert:type ~a) < 0 to
> > > (view_convert:type a) >= 0 when type is signed integer. Similar for
> > > (view_convert:type ~a) >= 0.
> > > Bootstrapped and regtested on x86_64-linux-gnu{-m32,}.
> > > Ok for the trunk?
> > >
> > > gcc/ChangeLog:
> > >
> > > PR middle-end/100738
> > > * match.pd ((view_convert ~a) < 0 --> (view_convert a) >= 0,
> > > (view_convert ~a) >= 0 --> (view_convert a) < 0): New GIMPLE
> > > simplification.
> >
> > We already have
> >
> > /* Fold ~X op C as X op' ~C, where op' is the swapped comparison. */
> > (for cmp (simple_comparison)
> > scmp (swapped_simple_comparison)
> > (simplify
> > (cmp (bit_not@2 @0) CONSTANT_CLASS_P@1)
> > (if (single_use (@2)
> > && (TREE_CODE (@1) == INTEGER_CST || TREE_CODE (@1) == VECTOR_CST))
> > (scmp @0 (bit_not @1)))))
> >
> > Would it make sense to try and generalize it a bit, say with
> >
> > (cmp (nop_convert1? (bit_not @0)) CONSTANT_CLASS_P)
> >
> > (scmp (view_convert:XXX @0) (bit_not @1))
> >
> Thanks for your advice, it looks great.
> And can I use *view_convert1?* instead of *nop_convert1?* here,
> because the original case is view_convert, and nop_convert would fail
> to simplify the case.
Here is updated patch
gcc/ChangeLog:
PR middle-end/100738
* match.pd (Fold ~X op C as X op' ~C): Extend GIMPLE
simplification to handle view_convert ~X.
gcc/testsuite/ChangeLog:
PR middle-end/100738
* g++.target/i386/avx2-pr100738-1.C: New test.
* g++.target/i386/sse4_1-pr100738-1.C: New test.
> > (I still believe that it is a bad idea that SSA_NAMEs are strongly typed,
> > encoding the type in operations would be more convenient, but I think the
> > time for that choice has long gone)
> >
> > --
> > Marc Glisse
>
>
>
> --
> BR,
> Hongtao
--
BR,
Hongtao
[-- Attachment #2: 0001-Extend-gimple-simplication-of-X-op-C-X-op-C-to-hanld.patch --]
[-- Type: text/x-patch, Size: 6208 bytes --]
From 60308636a36fa7a5b96d115452a42be914ef19e7 Mon Sep 17 00:00:00 2001
From: liuhongt <hongtao.liu@intel.com>
Date: Thu, 27 May 2021 15:21:06 +0800
Subject: [PATCH] Extend gimple simplication of ((~X) op C --> (X) op' ~C) to
hanlde view_convert of ~X
gcc/ChangeLog:
PR middle-end/100738
* match.pd (Fold ~X op C as X op' ~C): Extend GIMPLE
simplification to handle view_convert ~X.
gcc/testsuite/ChangeLog:
PR middle-end/100738
* g++.target/i386/avx2-pr100738-1.C: New test.
* g++.target/i386/sse4_1-pr100738-1.C: New test.
---
gcc/match.pd | 5 +-
.../g++.target/i386/avx2-pr100738-1.C | 120 ++++++++++++++++++
.../g++.target/i386/sse4_1-pr100738-1.C | 120 ++++++++++++++++++
3 files changed, 243 insertions(+), 2 deletions(-)
create mode 100644 gcc/testsuite/g++.target/i386/avx2-pr100738-1.C
create mode 100644 gcc/testsuite/g++.target/i386/sse4_1-pr100738-1.C
diff --git a/gcc/match.pd b/gcc/match.pd
index cdb87636951..cbb76d67dc5 100644
--- a/gcc/match.pd
+++ b/gcc/match.pd
@@ -4144,10 +4144,11 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
(for cmp (simple_comparison)
scmp (swapped_simple_comparison)
(simplify
- (cmp (bit_not@2 @0) CONSTANT_CLASS_P@1)
+ (cmp (view_convert1? (bit_not@2 @0)) CONSTANT_CLASS_P@1)
+ (with {tree ttype = TREE_TYPE (@1);}
(if (single_use (@2)
&& (TREE_CODE (@1) == INTEGER_CST || TREE_CODE (@1) == VECTOR_CST))
- (scmp @0 (bit_not @1)))))
+ (scmp (view_convert:ttype @0) (bit_not @1))))))
(for cmp (simple_comparison)
/* Fold (double)float1 CMP (double)float2 into float1 CMP float2. */
diff --git a/gcc/testsuite/g++.target/i386/avx2-pr100738-1.C b/gcc/testsuite/g++.target/i386/avx2-pr100738-1.C
new file mode 100644
index 00000000000..80fdad3e5f0
--- /dev/null
+++ b/gcc/testsuite/g++.target/i386/avx2-pr100738-1.C
@@ -0,0 +1,120 @@
+/* { dg-do compile } */
+/* { dg-options "-mavx2 -std=c++14 -O2 -mno-avx512f -mno-xop" } */
+/* { dg-final { scan-assembler-not "pxor" } } */
+/* { dg-final { scan-assembler-not "pcmpgt\[bdq]" } } */
+/* { dg-final { scan-assembler-times "pblendvb" 6 } } */
+/* { dg-final { scan-assembler-times "blendvps" 6 } } */
+/* { dg-final { scan-assembler-times "blendvpd" 6 } } */
+
+typedef char v32qi __attribute__ ((vector_size (32)));
+typedef short v16hi __attribute__ ((vector_size (32)));
+typedef int v8si __attribute__ ((vector_size (32)));
+typedef long long v4di __attribute__ ((vector_size (32)));
+
+v8si
+f1 (v32qi a, v8si b, v8si c)
+{
+ return ((v8si)~a) < 0 ? b : c;
+}
+
+v4di
+f2 (v32qi a, v4di b, v4di c)
+{
+ return ((v4di)~a) < 0 ? b : c;
+}
+
+v32qi
+f3 (v16hi a, v32qi b, v32qi c)
+{
+ return ((v32qi)~a) < 0 ? b : c;
+}
+
+v8si
+f4 (v16hi a, v8si b, v8si c)
+{
+ return ((v8si)~a) < 0 ? b : c;
+}
+
+v4di
+f5 (v16hi a, v4di b, v4di c)
+{
+ return ((v4di)~a) < 0 ? b : c;
+}
+
+v32qi
+f6 (v8si a, v32qi b, v32qi c)
+{
+ return ((v32qi)~a) < 0 ? b : c;
+}
+
+v4di
+f7 (v8si a, v4di b, v4di c)
+{
+ return ((v4di)~a) < 0 ? b : c;
+}
+
+v32qi
+f8 (v4di a, v32qi b, v32qi c)
+{
+ return ((v32qi)~a) < 0 ? b : c;
+}
+
+v8si
+f9 (v4di a, v8si b, v8si c)
+{
+ return ((v8si)~a) < 0 ? b : c;
+}
+
+v8si
+f10 (v32qi a, v8si b, v8si c)
+{
+ return ((v8si)~a) >= 0 ? b : c;
+}
+
+v4di
+f11 (v32qi a, v4di b, v4di c)
+{
+ return ((v4di)~a) >= 0 ? b : c;
+}
+
+v32qi
+f12 (v16hi a, v32qi b, v32qi c)
+{
+ return ((v32qi)~a) >= 0 ? b : c;
+}
+
+v8si
+f13 (v16hi a, v8si b, v8si c)
+{
+ return ((v8si)~a) >= 0 ? b : c;
+}
+
+v4di
+f14 (v16hi a, v4di b, v4di c)
+{
+ return ((v4di)~a) >= 0 ? b : c;
+}
+
+v32qi
+f15 (v8si a, v32qi b, v32qi c)
+{
+ return ((v32qi)~a) >= 0 ? b : c;
+}
+
+v4di
+f16 (v8si a, v4di b, v4di c)
+{
+ return ((v4di)~a) >= 0 ? b : c;
+}
+
+v32qi
+f17 (v4di a, v32qi b, v32qi c)
+{
+ return ((v32qi)~a) >= 0 ? b : c;
+}
+
+v8si
+f18 (v4di a, v8si b, v8si c)
+{
+ return ((v8si)~a) >= 0 ? b : c;
+}
diff --git a/gcc/testsuite/g++.target/i386/sse4_1-pr100738-1.C b/gcc/testsuite/g++.target/i386/sse4_1-pr100738-1.C
new file mode 100644
index 00000000000..d3454c264cd
--- /dev/null
+++ b/gcc/testsuite/g++.target/i386/sse4_1-pr100738-1.C
@@ -0,0 +1,120 @@
+/* { dg-do compile } */
+/* { dg-options "-msse4 -std=c++14 -mno-avx2 -O2 -mno-xop" } */
+/* { dg-final { scan-assembler-not "pxor" } } */
+/* { dg-final { scan-assembler-not "pcmpgt\[bdq]" } } */
+/* { dg-final { scan-assembler-times "pblendvb" 6 } } */
+/* { dg-final { scan-assembler-times "blendvps" 6 } } */
+/* { dg-final { scan-assembler-times "blendvpd" 6 } } */
+
+typedef char v16qi __attribute__ ((vector_size (16)));
+typedef short v8hi __attribute__ ((vector_size (16)));
+typedef int v4si __attribute__ ((vector_size (16)));
+typedef long long v2di __attribute__ ((vector_size (16)));
+
+v4si
+f1 (v16qi a, v4si b, v4si c)
+{
+ return ((v4si)~a) < 0 ? b : c;
+}
+
+v2di
+f2 (v16qi a, v2di b, v2di c)
+{
+ return ((v2di)~a) < 0 ? b : c;
+}
+
+v16qi
+f3 (v8hi a, v16qi b, v16qi c)
+{
+ return ((v16qi)~a) < 0 ? b : c;
+}
+
+v4si
+f4 (v8hi a, v4si b, v4si c)
+{
+ return ((v4si)~a) < 0 ? b : c;
+}
+
+v2di
+f5 (v8hi a, v2di b, v2di c)
+{
+ return ((v2di)~a) < 0 ? b : c;
+}
+
+v16qi
+f6 (v4si a, v16qi b, v16qi c)
+{
+ return ((v16qi)~a) < 0 ? b : c;
+}
+
+v2di
+f7 (v4si a, v2di b, v2di c)
+{
+ return ((v2di)~a) < 0 ? b : c;
+}
+
+v16qi
+f8 (v2di a, v16qi b, v16qi c)
+{
+ return ((v16qi)~a) < 0 ? b : c;
+}
+
+v4si
+f9 (v2di a, v4si b, v4si c)
+{
+ return ((v4si)~a) < 0 ? b : c;
+}
+
+v4si
+f10 (v16qi a, v4si b, v4si c)
+{
+ return ((v4si)~a) >= 0 ? b : c;
+}
+
+v2di
+f11 (v16qi a, v2di b, v2di c)
+{
+ return ((v2di)~a) >= 0 ? b : c;
+}
+
+v16qi
+f12 (v8hi a, v16qi b, v16qi c)
+{
+ return ((v16qi)~a) >= 0 ? b : c;
+}
+
+v4si
+f13 (v8hi a, v4si b, v4si c)
+{
+ return ((v4si)~a) >= 0 ? b : c;
+}
+
+v2di
+f14 (v8hi a, v2di b, v2di c)
+{
+ return ((v2di)~a) >= 0 ? b : c;
+}
+
+v16qi
+f15 (v4si a, v16qi b, v16qi c)
+{
+ return ((v16qi)~a) >= 0 ? b : c;
+}
+
+v2di
+f16 (v4si a, v2di b, v2di c)
+{
+ return ((v2di)~a) >= 0 ? b : c;
+}
+
+v16qi
+f17 (v2di a, v16qi b, v16qi c)
+{
+ return ((v16qi)~a) >= 0 ? b : c;
+}
+
+v4si
+f18 (v2di a, v4si b, v4si c)
+{
+ return ((v4si)~a) >= 0 ? b : c;
+}
--
2.18.1
next prev parent reply other threads:[~2021-06-04 8:07 UTC|newest]
Thread overview: 10+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-06-01 5:22 Hongtao Liu
2021-06-01 5:29 ` Andrew Pinski
2021-06-01 11:52 ` Hongtao Liu
2021-06-01 10:17 ` Marc Glisse
2021-06-04 5:01 ` Hongtao Liu
2021-06-04 8:11 ` Hongtao Liu [this message]
2021-06-04 8:18 ` Marc Glisse
2021-06-07 6:22 ` Hongtao Liu
2021-06-07 7:06 ` Hongtao Liu
2021-06-07 13:19 ` Richard Biener
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to='CAMZc-bwKL1PMKpOgwcT+ierLv-JjATMvCmC_Ntb+J3=4t11BXA@mail.gmail.com' \
--to=crazylht@gmail.com \
--cc=gcc-patches@gcc.gnu.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
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).