* [PATCH] aarch64: Specify that FEAT_MOPS sequences clobber CC
@ 2022-11-30 17:40 Kyrylo Tkachov
2022-12-06 10:26 ` Kyrylo Tkachov
0 siblings, 1 reply; 2+ messages in thread
From: Kyrylo Tkachov @ 2022-11-30 17:40 UTC (permalink / raw)
To: gcc-patches
[-- Attachment #1: Type: text/plain, Size: 1369 bytes --]
Hi all,
According to the architecture pseudocode the FEAT_MOPS sequences overwrite the NZCV flags
as part of their operation, so GCC needs to model that in the relevant RTL patterns.
For the testcase:
void g();
void foo (int a, size_t N, char *__restrict__ in,
char *__restrict__ out)
{
if (a != 3)
__builtin_memcpy (out, in, N);
if (a > 3)
g ();
}
we will currently generate:
foo:
cmp w0, 3
bne .L6
.L1:
ret
.L6:
cpyfp [x3]!, [x2]!, x1!
cpyfm [x3]!, [x2]!, x1!
cpyfe [x3]!, [x2]!, x1!
ble .L1 // Flags reused after CPYF* sequence
b g
This is wrong as the result of cmp needs to be recalculated after the MOPS sequence.
With this patch we'll insert a "cmp w0, 3" before the ble, like what clang does.
Bootstrapped and tested on aarch64-none-linux-gnu.
Pushing to trunk and to the GCC 12 branch after some baking time.
Thanks,
Kyrill
gcc/ChangeLog:
* config/aarch64/aarch64.md (aarch64_cpymemdi): Specify clobber of CC reg.
(*aarch64_cpymemdi): Likewise.
(aarch64_movmemdi): Likewise.
(aarch64_setmemdi): Likewise.
(*aarch64_setmemdi): Likewise.
gcc/testsuite/ChangeLog:
* gcc.target/aarch64/mops_5.c: New test.
* gcc.target/aarch64/mops_6.c: Likewise.
* gcc.target/aarch64/mops_7.c: Likewise.
[-- Attachment #2: mops-cc.patch --]
[-- Type: application/octet-stream, Size: 3910 bytes --]
diff --git a/gcc/config/aarch64/aarch64.md b/gcc/config/aarch64/aarch64.md
index 76b6898ca048c559ff7f7fba78119161b3d382f6..8a18405b04ea0df8ad74c24719cd315b265c501c 100644
--- a/gcc/config/aarch64/aarch64.md
+++ b/gcc/config/aarch64/aarch64.md
@@ -1592,6 +1592,7 @@ (define_expand "aarch64_cpymemdi"
[(set (match_operand 2) (const_int 0))
(clobber (match_dup 3))
(clobber (match_dup 4))
+ (clobber (reg:CC CC_REGNUM))
(set (match_operand 0)
(unspec:BLK [(match_operand 1) (match_dup 2)] UNSPEC_CPYMEM))])]
"TARGET_MOPS"
@@ -1605,6 +1606,7 @@ (define_insn "*aarch64_cpymemdi"
[(set (match_operand:DI 2 "register_operand" "+&r") (const_int 0))
(clobber (match_operand:DI 0 "register_operand" "+&r"))
(clobber (match_operand:DI 1 "register_operand" "+&r"))
+ (clobber (reg:CC CC_REGNUM))
(set (mem:BLK (match_dup 0))
(unspec:BLK [(mem:BLK (match_dup 1)) (match_dup 2)] UNSPEC_CPYMEM))]
"TARGET_MOPS"
@@ -1635,6 +1637,7 @@ (define_insn "aarch64_movmemdi"
(set (match_operand:DI 2 "register_operand" "+&r") (const_int 0))
(clobber (match_operand:DI 0 "register_operand" "+&r"))
(clobber (match_operand:DI 1 "register_operand" "+&r"))
+ (clobber (reg:CC CC_REGNUM))
(set (mem:BLK (match_dup 0))
(unspec:BLK [(mem:BLK (match_dup 1)) (match_dup 2)] UNSPEC_MOVMEM))])]
"TARGET_MOPS"
@@ -1680,6 +1683,7 @@ (define_expand "aarch64_setmemdi"
[(parallel
[(set (match_operand 2) (const_int 0))
(clobber (match_dup 3))
+ (clobber (reg:CC CC_REGNUM))
(set (match_operand 0)
(unspec:BLK [(match_operand 1)
(match_dup 2)] UNSPEC_SETMEM))])]
@@ -1692,6 +1696,7 @@ (define_expand "aarch64_setmemdi"
(define_insn "*aarch64_setmemdi"
[(set (match_operand:DI 2 "register_operand" "+&r") (const_int 0))
(clobber (match_operand:DI 0 "register_operand" "+&r"))
+ (clobber (reg:CC CC_REGNUM))
(set (mem:BLK (match_dup 0))
(unspec:BLK [(match_operand:QI 1 "aarch64_reg_or_zero" "rZ")
(match_dup 2)] UNSPEC_SETMEM))]
diff --git a/gcc/testsuite/gcc.target/aarch64/mops_5.c b/gcc/testsuite/gcc.target/aarch64/mops_5.c
new file mode 100644
index 0000000000000000000000000000000000000000..8a2626786eb510ed880850fc6ab8992482a955de
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/mops_5.c
@@ -0,0 +1,17 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -march=armv8.6-a+mops" } */
+
+#include <stddef.h>
+
+void g();
+void foo (int a, size_t N, char *__restrict__ in,
+ char *__restrict__ out)
+{
+ if (a != 3)
+ __builtin_memcpy (out, in, N);
+ if (a > 3)
+ g ();
+}
+
+/* { dg-final { scan-assembler-times {cmp\tw0, *} 2 } } */
+
diff --git a/gcc/testsuite/gcc.target/aarch64/mops_6.c b/gcc/testsuite/gcc.target/aarch64/mops_6.c
new file mode 100644
index 0000000000000000000000000000000000000000..c6c9a548351e4e8e12960138c55630b125994ed7
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/mops_6.c
@@ -0,0 +1,17 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -march=armv8.6-a+mops" } */
+
+#include <stddef.h>
+
+void g();
+void foo (int a, size_t N, char *__restrict__ in,
+ char *__restrict__ out)
+{
+ if (a != 3)
+ __builtin_memmove (out, in, N);
+ if (a > 3)
+ g ();
+}
+
+/* { dg-final { scan-assembler-times {cmp\tw0, *} 2 } } */
+
diff --git a/gcc/testsuite/gcc.target/aarch64/mops_7.c b/gcc/testsuite/gcc.target/aarch64/mops_7.c
new file mode 100644
index 0000000000000000000000000000000000000000..79720ca164c400061da1fc2d0ef7a969b603190c
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/mops_7.c
@@ -0,0 +1,16 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -march=armv8.6-a+mops" } */
+
+#include <stddef.h>
+
+void g();
+void foo (int a, size_t N, char *__restrict__ out)
+{
+ if (a != 3)
+ __builtin_memset (out, 0, N);
+ if (a > 3)
+ g ();
+}
+
+/* { dg-final { scan-assembler-times {cmp\tw0, *} 2 } } */
+
^ permalink raw reply [flat|nested] 2+ messages in thread
* RE: [PATCH] aarch64: Specify that FEAT_MOPS sequences clobber CC
2022-11-30 17:40 [PATCH] aarch64: Specify that FEAT_MOPS sequences clobber CC Kyrylo Tkachov
@ 2022-12-06 10:26 ` Kyrylo Tkachov
0 siblings, 0 replies; 2+ messages in thread
From: Kyrylo Tkachov @ 2022-12-06 10:26 UTC (permalink / raw)
To: gcc-patches
> -----Original Message-----
> From: Kyrylo Tkachov
> Sent: Wednesday, November 30, 2022 5:40 PM
> To: gcc-patches@gcc.gnu.org
> Subject: [PATCH] aarch64: Specify that FEAT_MOPS sequences clobber CC
>
> Hi all,
>
> According to the architecture pseudocode the FEAT_MOPS sequences
> overwrite the NZCV flags
> as part of their operation, so GCC needs to model that in the relevant RTL
> patterns.
> For the testcase:
> void g();
> void foo (int a, size_t N, char *__restrict__ in,
> char *__restrict__ out)
> {
> if (a != 3)
> __builtin_memcpy (out, in, N);
> if (a > 3)
> g ();
> }
>
> we will currently generate:
> foo:
> cmp w0, 3
> bne .L6
> .L1:
> ret
> .L6:
> cpyfp [x3]!, [x2]!, x1!
> cpyfm [x3]!, [x2]!, x1!
> cpyfe [x3]!, [x2]!, x1!
> ble .L1 // Flags reused after CPYF* sequence
> b g
>
> This is wrong as the result of cmp needs to be recalculated after the MOPS
> sequence.
> With this patch we'll insert a "cmp w0, 3" before the ble, like what clang
> does.
>
> Bootstrapped and tested on aarch64-none-linux-gnu.
> Pushing to trunk and to the GCC 12 branch after some baking time.
Pushed to releases/gcc-12 now.
Thanks,
Kyrill
>
> Thanks,
> Kyrill
>
> gcc/ChangeLog:
>
> * config/aarch64/aarch64.md (aarch64_cpymemdi): Specify clobber
> of CC reg.
> (*aarch64_cpymemdi): Likewise.
> (aarch64_movmemdi): Likewise.
> (aarch64_setmemdi): Likewise.
> (*aarch64_setmemdi): Likewise.
>
> gcc/testsuite/ChangeLog:
>
> * gcc.target/aarch64/mops_5.c: New test.
> * gcc.target/aarch64/mops_6.c: Likewise.
> * gcc.target/aarch64/mops_7.c: Likewise.
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2022-12-06 10:26 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-11-30 17:40 [PATCH] aarch64: Specify that FEAT_MOPS sequences clobber CC Kyrylo Tkachov
2022-12-06 10:26 ` Kyrylo Tkachov
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).