On Sun, May 25, 2025 at 8:12 AM H.J. Lu <hjl.to...@gmail.com> wrote:
>
> On Sun, May 25, 2025 at 7:47 AM H.J. Lu <hjl.to...@gmail.com> wrote:
> >
> > commit ef26c151c14a87177d46fd3d725e7f82e040e89f
> > Author: Roger Sayle <ro...@nextmovesoftware.com>
> > Date: Thu Dec 23 12:33:07 2021 +0000
> >
> > x86: PR target/103773: Fix wrong-code with -Oz from pop to memory.
> >
> > transformed "mov $0,mem" to the shorter and "$0,mem" for -Oz. But
> >
> > (define_insn "*mov<mode>_and"
> > [(set (match_operand:SWI248 0 "memory_operand" "=m")
> > (match_operand:SWI248 1 "const0_operand"))
> > (clobber (reg:CC FLAGS_REG))]
> > "reload_completed"
> > "and{<imodesuffix>}\t{%1, %0|%0, %1}"
> > [(set_attr "type" "alu1")
> > (set_attr "mode" "<MODE>")
> > (set_attr "length_immediate" "1")])
> >
> > isn't guarded for -Oz. As a result, "and $0,mem" is generated without
> > -Oz. Enable *mov<mode>_and only for -Oz.
> >
> > gcc/
> >
> > PR target/120427
> > * config/i386/i386.md (*mov<mode>_and): Enable only for -Oz.
> >
> > gcc/testsuite/
> >
> > PR target/120427
> > * gcc.target/i386/pr120427.c: New test.
> >
> > OK for master?
> >
>
> "mov $-1,mem" has the same issue. Here is the updated patch to also
> enable "or $-1,mem" only for -Oz.
>
> OK for master?
It doesn't work since "*mov<mode>_or" was extended from load. Here is
the v2 patch:
1. Add "*mov<mode>_or_store" for "or $-1,mem".
2. Rename "*mov<mode>_or" to "*mov<mode>_or_load", replacing
nonimmediate_operand with register_operand.
3. Enable "*mov<mode>_and" and "*mov<mode>_or_store" only for -Oz.
Tested on x86-64.
--
H.J.
From be013c2d0bde068804fda3db6b05c89d7a26d54e Mon Sep 17 00:00:00 2001
From: "H.J. Lu" <hjl.to...@gmail.com>
Date: Sun, 25 May 2025 07:40:29 +0800
Subject: [PATCH v2] x86: Enable *mov<mode>_(and|or_store) only for -Oz
commit ef26c151c14a87177d46fd3d725e7f82e040e89f
Author: Roger Sayle <ro...@nextmovesoftware.com>
Date: Thu Dec 23 12:33:07 2021 +0000
x86: PR target/103773: Fix wrong-code with -Oz from pop to memory.
added "*mov<mode>_and" and extended "*mov<mode>_or" to transform
"mov $0,mem" to the shorter "and $0,mem" and "mov $-1,mem" to the shorter
"or $-1,mem" for -Oz. But the new pattern:
(define_insn "*mov<mode>_and"
[(set (match_operand:SWI248 0 "memory_operand" "=m")
(match_operand:SWI248 1 "const0_operand"))
(clobber (reg:CC FLAGS_REG))]
"reload_completed"
"and{<imodesuffix>}\t{%1, %0|%0, %1}"
[(set_attr "type" "alu1")
(set_attr "mode" "<MODE>")
(set_attr "length_immediate" "1")])
and the extended pattern:
(define_insn "*mov<mode>_or"
[(set (match_operand:SWI248 0 "nonimmediate_operand" "=rm")
(match_operand:SWI248 1 "constm1_operand"))
(clobber (reg:CC FLAGS_REG))]
"reload_completed"
"or{<imodesuffix>}\t{%1, %0|%0, %1}"
[(set_attr "type" "alu1")
(set_attr "mode" "<MODE>")
(set_attr "length_immediate" "1")])
aren't guarded for -Oz. As a result, "and $0,mem" and "or $-1,mem" are
generated without -Oz. This patch:
1. Add "*mov<mode>_or_store" for "or $-1,mem".
2. Rename "*mov<mode>_or" to "*mov<mode>_or_load", replacing
nonimmediate_operand with register_operand.
3. Enable "*mov<mode>_and" and "*mov<mode>_or_store" only for -Oz.
gcc/
PR target/120427
* config/i386/i386.md (*mov<mode>_and): Enable only for -Oz.
(*mov<mode>_or_store): New.
(*mov<mode>_or): Renamed to ...
(*mov<mode>_or_load): This. Replace nonimmediate_operand with
register_operand.
gcc/testsuite/
PR target/120427
* gcc.target/i386/pr120427-1.c: New test.
* gcc.target/i386/pr120427-2.c: Likewise.
Signed-off-by: H.J. Lu <hjl.to...@gmail.com>
---
gcc/config/i386/i386.md | 18 +++++++++++---
gcc/testsuite/gcc.target/i386/pr120427-1.c | 28 ++++++++++++++++++++++
gcc/testsuite/gcc.target/i386/pr120427-2.c | 28 ++++++++++++++++++++++
3 files changed, 71 insertions(+), 3 deletions(-)
create mode 100644 gcc/testsuite/gcc.target/i386/pr120427-1.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr120427-2.c
diff --git a/gcc/config/i386/i386.md b/gcc/config/i386/i386.md
index b7a18d583da..e55dd27cfcf 100644
--- a/gcc/config/i386/i386.md
+++ b/gcc/config/i386/i386.md
@@ -2442,14 +2442,26 @@ (define_insn "*mov<mode>_and"
[(set (match_operand:SWI248 0 "memory_operand" "=m")
(match_operand:SWI248 1 "const0_operand"))
(clobber (reg:CC FLAGS_REG))]
- "reload_completed"
+ "reload_completed
+ && optimize_insn_for_size_p () && optimize_size > 1"
"and{<imodesuffix>}\t{%1, %0|%0, %1}"
[(set_attr "type" "alu1")
(set_attr "mode" "<MODE>")
(set_attr "length_immediate" "1")])
-(define_insn "*mov<mode>_or"
- [(set (match_operand:SWI248 0 "nonimmediate_operand" "=rm")
+(define_insn "*mov<mode>_or_store"
+ [(set (match_operand:SWI248 0 "memory_operand" "=m")
+ (match_operand:SWI248 1 "constm1_operand"))
+ (clobber (reg:CC FLAGS_REG))]
+ "reload_completed
+ && optimize_insn_for_size_p () && optimize_size > 1"
+ "or{<imodesuffix>}\t{%1, %0|%0, %1}"
+ [(set_attr "type" "alu1")
+ (set_attr "mode" "<MODE>")
+ (set_attr "length_immediate" "1")])
+
+(define_insn "*mov<mode>_or_load"
+ [(set (match_operand:SWI248 0 "register_operand" "=r")
(match_operand:SWI248 1 "constm1_operand"))
(clobber (reg:CC FLAGS_REG))]
"reload_completed"
diff --git a/gcc/testsuite/gcc.target/i386/pr120427-1.c b/gcc/testsuite/gcc.target/i386/pr120427-1.c
new file mode 100644
index 00000000000..7f1690e49b4
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr120427-1.c
@@ -0,0 +1,28 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -mtune=sapphirerapids" } */
+/* { dg-final { scan-assembler-not "and\[lq\]?\[\\t \]+\\\$0, \[0-9\]*\\(" } } */
+
+struct __pthread_mutex_s
+{
+ int __lock;
+ unsigned int __count;
+ int __owner;
+ unsigned int __nusers;
+ int __kind;
+ short __spins;
+ short __elision;
+ void *p[2];
+};
+typedef union
+{
+ struct __pthread_mutex_s __data;
+ char __size[40];
+ long int __align;
+} pthread_mutex_t;
+typedef struct { pthread_mutex_t mutex; } __rtld_lock_recursive_t;
+void
+foo (__rtld_lock_recursive_t *lock, int i)
+{
+ lock[i] = (__rtld_lock_recursive_t) {{ { 0, 0, 0, 0, 1,
+ 0, 0, { ((void *)0) , ((void *)0) } } }};
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr120427-2.c b/gcc/testsuite/gcc.target/i386/pr120427-2.c
new file mode 100644
index 00000000000..a380c128ccb
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr120427-2.c
@@ -0,0 +1,28 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -mtune=sapphirerapids" } */
+/* { dg-final { scan-assembler-not "or\[lq\]?\[\\t \]+\\\$-1, \[0-9\]*\\(" } } */
+
+struct __pthread_mutex_s
+{
+ int __lock;
+ unsigned int __count;
+ int __owner;
+ unsigned int __nusers;
+ int __kind;
+ short __spins;
+ short __elision;
+ void *p[2];
+};
+typedef union
+{
+ struct __pthread_mutex_s __data;
+ char __size[40];
+ long int __align;
+} pthread_mutex_t;
+typedef struct { pthread_mutex_t mutex; } __rtld_lock_recursive_t;
+void
+foo (__rtld_lock_recursive_t *lock, int i)
+{
+ lock[i] = (__rtld_lock_recursive_t) {{ { -1, -1, -1, -1, 1,
+ -1, -1, { ((void *)-1) , ((void *)-1) } } }};
+}
--
2.49.0
