* [PATCH] RISC-V: add multiarch RVV support for memcpy using FMV IFUNC
@ 2025-01-16 9:30 戴成荣
2025-01-16 23:46 ` Andrew Waterman
0 siblings, 1 reply; 4+ messages in thread
From: 戴成荣 @ 2025-01-16 9:30 UTC (permalink / raw)
To: Palmer Dabbelt; +Cc: libc-alpha
[-- Attachment #1: Type: text/plain, Size: 3532 bytes --]
With RISC-V GCC Function multi-versioning IFUNC support,glibc could utilize some new RISC-V CPUs with new extensions such as Vector to re-realize or speed up some functions like memcpy.
Signed-off-by: daichengrong <daichengrong@iscas.ac.cn>
---
sysdeps/riscv/multiarch/memcpy_vector.S | 36 +++++++++++++++++++
.../unix/sysv/linux/riscv/multiarch/Makefile | 2 ++
.../unix/sysv/linux/riscv/multiarch/memcpy.c | 5 +++
3 files changed, 43 insertions(+)
create mode 100644 sysdeps/riscv/multiarch/memcpy_vector.S
diff --git a/sysdeps/riscv/multiarch/memcpy_vector.S b/sysdeps/riscv/multiarch/memcpy_vector.S
new file mode 100644
index 0000000000..92e2558cd1
--- /dev/null
+++ b/sysdeps/riscv/multiarch/memcpy_vector.S
@@ -0,0 +1,36 @@
+/* memcpy for RISC-V Vector.
+ Copyright (C) 2024-2025 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <https://www.gnu.org/licenses/>. */
+
+
+#include <sysdep.h>
+#include <sys/asm.h>
+
+ENTRY (__memcpy_vector)
+ beq a2, zero, L(ret)
+ mv a6, a0
+L(loop):
+ vsetvli a3,a2,e8,m8,ta,mu
+ vle8.v v8,(a1)
+ vse8.v v8,(a6)
+ add a1,a1,a3
+ sub a2,a2,a3
+ add a6,a6,a3
+ bnez a2,L(loop)
+L(ret):
+ ret
+END (__memcpy_vector)
\ No newline at end of file
diff --git a/sysdeps/unix/sysv/linux/riscv/multiarch/Makefile b/sysdeps/unix/sysv/linux/riscv/multiarch/Makefile
index fcef5659d4..a8b6c22af1 100644
--- a/sysdeps/unix/sysv/linux/riscv/multiarch/Makefile
+++ b/sysdeps/unix/sysv/linux/riscv/multiarch/Makefile
@@ -3,7 +3,9 @@ sysdep_routines += \
memcpy \
memcpy-generic \
memcpy_noalignment \
+ memcpy_vector \
# sysdep_routines
CFLAGS-memcpy_noalignment.c += -mno-strict-align
+ASFLAGS-memcpy_vector.S += -march=rv64gcv
endif
diff --git a/sysdeps/unix/sysv/linux/riscv/multiarch/memcpy.c b/sysdeps/unix/sysv/linux/riscv/multiarch/memcpy.c
index 8544f5402a..c9879762f6 100644
--- a/sysdeps/unix/sysv/linux/riscv/multiarch/memcpy.c
+++ b/sysdeps/unix/sysv/linux/riscv/multiarch/memcpy.c
@@ -32,11 +32,16 @@ extern __typeof (__redirect_memcpy) __libc_memcpy;
extern __typeof (__redirect_memcpy) __memcpy_generic attribute_hidden;
extern __typeof (__redirect_memcpy) __memcpy_noalignment attribute_hidden;
+extern __typeof (__redirect_memcpy) __memcpy_vector attribute_hidden;
static inline __typeof (__redirect_memcpy) *
select_memcpy_ifunc (uint64_t dl_hwcap, __riscv_hwprobe_t hwprobe_func)
{
unsigned long long int v;
+ if (__riscv_hwprobe_one (hwprobe_func, RISCV_HWPROBE_KEY_IMA_EXT_0, &v) == 0
+ && (v & RISCV_HWPROBE_IMA_V) == RISCV_HWPROBE_IMA_V)
+ return __memcpy_vector;
+
if (__riscv_hwprobe_one (hwprobe_func, RISCV_HWPROBE_KEY_CPUPERF_0, &v) == 0
&& (v & RISCV_HWPROBE_MISALIGNED_MASK) == RISCV_HWPROBE_MISALIGNED_FAST)
return __memcpy_noalignment;
--
2.25.1
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH] RISC-V: add multiarch RVV support for memcpy using FMV IFUNC
2025-01-16 9:30 [PATCH] RISC-V: add multiarch RVV support for memcpy using FMV IFUNC 戴成荣
@ 2025-01-16 23:46 ` Andrew Waterman
2025-01-17 8:01 ` [PATCH v2] " daichengrong
2025-01-17 13:01 ` daichengrong
0 siblings, 2 replies; 4+ messages in thread
From: Andrew Waterman @ 2025-01-16 23:46 UTC (permalink / raw)
To: 戴成荣; +Cc: Palmer Dabbelt, libc-alpha
On Thu, Jan 16, 2025 at 1:31 AM 戴成荣 <daichengrong@iscas.ac.cn> wrote:
>
> With RISC-V GCC Function multi-versioning IFUNC support,glibc could utilize some new RISC-V CPUs with new extensions such as Vector to re-realize or speed up some functions like memcpy.
>
>
>
> Signed-off-by: daichengrong <daichengrong@iscas.ac.cn>
>
>
> ---
>
> sysdeps/riscv/multiarch/memcpy_vector.S | 36 +++++++++++++++++++
> .../unix/sysv/linux/riscv/multiarch/Makefile | 2 ++
> .../unix/sysv/linux/riscv/multiarch/memcpy.c | 5 +++
> 3 files changed, 43 insertions(+)
> create mode 100644 sysdeps/riscv/multiarch/memcpy_vector.S
>
> diff --git a/sysdeps/riscv/multiarch/memcpy_vector.S b/sysdeps/riscv/multiarch/memcpy_vector.S
> new file mode 100644
> index 0000000000..92e2558cd1
> --- /dev/null
> +++ b/sysdeps/riscv/multiarch/memcpy_vector.S
> @@ -0,0 +1,36 @@
> +/* memcpy for RISC-V Vector.
> + Copyright (C) 2024-2025 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +
> +#include <sysdep.h>
> +#include <sys/asm.h>
> +
> +ENTRY (__memcpy_vector)
> + beq a2, zero, L(ret)
This branch should be deleted; the strip-mine loop below does the
correct thing for a2=0. Including the branch is an anti-optimization:
it speeds up the uncommon case of size-0 memcpy at the expense of the
common case.
>
> + mv a6, a0
> +L(loop):
> + vsetvli a3,a2,e8,m8,ta,mu
> + vle8.v v8,(a1)
> + vse8.v v8,(a6)
> + add a1,a1,a3
> + sub a2,a2,a3
> + add a6,a6,a3
> + bnez a2,L(loop)
> +L(ret):
> + ret
> +END (__memcpy_vector)
> \ No newline at end of file
> diff --git a/sysdeps/unix/sysv/linux/riscv/multiarch/Makefile b/sysdeps/unix/sysv/linux/riscv/multiarch/Makefile
> index fcef5659d4..a8b6c22af1 100644
> --- a/sysdeps/unix/sysv/linux/riscv/multiarch/Makefile
> +++ b/sysdeps/unix/sysv/linux/riscv/multiarch/Makefile
> @@ -3,7 +3,9 @@ sysdep_routines += \
> memcpy \
> memcpy-generic \
> memcpy_noalignment \
> + memcpy_vector \
> # sysdep_routines
>
> CFLAGS-memcpy_noalignment.c += -mno-strict-align
> +ASFLAGS-memcpy_vector.S += -march=rv64gcv
> endif
> diff --git a/sysdeps/unix/sysv/linux/riscv/multiarch/memcpy.c b/sysdeps/unix/sysv/linux/riscv/multiarch/memcpy.c
> index 8544f5402a..c9879762f6 100644
> --- a/sysdeps/unix/sysv/linux/riscv/multiarch/memcpy.c
> +++ b/sysdeps/unix/sysv/linux/riscv/multiarch/memcpy.c
> @@ -32,11 +32,16 @@ extern __typeof (__redirect_memcpy) __libc_memcpy;
>
> extern __typeof (__redirect_memcpy) __memcpy_generic attribute_hidden;
> extern __typeof (__redirect_memcpy) __memcpy_noalignment attribute_hidden;
> +extern __typeof (__redirect_memcpy) __memcpy_vector attribute_hidden;
>
> static inline __typeof (__redirect_memcpy) *
> select_memcpy_ifunc (uint64_t dl_hwcap, __riscv_hwprobe_t hwprobe_func)
> {
> unsigned long long int v;
> + if (__riscv_hwprobe_one (hwprobe_func, RISCV_HWPROBE_KEY_IMA_EXT_0, &v) == 0
> + && (v & RISCV_HWPROBE_IMA_V) == RISCV_HWPROBE_IMA_V)
> + return __memcpy_vector;
> +
> if (__riscv_hwprobe_one (hwprobe_func, RISCV_HWPROBE_KEY_CPUPERF_0, &v) == 0
> && (v & RISCV_HWPROBE_MISALIGNED_MASK) == RISCV_HWPROBE_MISALIGNED_FAST)
> return __memcpy_noalignment;
> --
> 2.25.1
^ permalink raw reply [flat|nested] 4+ messages in thread
* [PATCH v2] RISC-V: add multiarch RVV support for memcpy using FMV IFUNC
2025-01-16 23:46 ` Andrew Waterman
@ 2025-01-17 8:01 ` daichengrong
2025-01-17 13:01 ` daichengrong
1 sibling, 0 replies; 4+ messages in thread
From: daichengrong @ 2025-01-17 8:01 UTC (permalink / raw)
To: Andrew Waterman; +Cc: Palmer Dabbelt, libc-alpha
> On Thu, Jan 16, 2025 at 1:31 AM 戴成荣 <daichengrong@iscas.ac.cn> wrote:
> >
> > With RISC-V GCC Function multi-versioning IFUNC support,glibc could utilize some new RISC-V CPUs with new extensions such as Vector to re-realize or speed up some functions like memcpy.
> > +
> > +ENTRY (__memcpy_vector)
> > + beq a2, zero, L(ret)
>
> This branch should be deleted; the strip-mine loop below does the
> correct thing for a2=0. Including the branch is an anti-optimization:
> it speeds up the uncommon case of size-0 memcpy at the expense of the
> common case.
>
thanks! This branch has been deleted.
> >
> > + mv a6, a0
> > +L(loop):
> > + vsetvli a3,a2,e8,m8,ta,mu
> > + vle8.v v8,(a1)
> > + vse8.v v8,(a6)
> > + add a1,a1,a3
> > + sub a2,a2,a3
> > + add a6,a6,a3
> > + bnez a2,L(loop)
> > +L(ret):
> > + ret
> > +END (__memcpy_vector)
Signed-off-by: daichengrong <daichengrong@iscas.ac.cn>
---
sysdeps/riscv/multiarch/memcpy_vector.S | 35 +++++++++++++++++++
.../unix/sysv/linux/riscv/multiarch/Makefile | 2 ++
.../unix/sysv/linux/riscv/multiarch/memcpy.c | 5 +++
3 files changed, 42 insertions(+)
create mode 100644 sysdeps/riscv/multiarch/memcpy_vector.S
diff --git a/sysdeps/riscv/multiarch/memcpy_vector.S b/sysdeps/riscv/multiarch/memcpy_vector.S
new file mode 100644
index 0000000000..8fddab8432
--- /dev/null
+++ b/sysdeps/riscv/multiarch/memcpy_vector.S
@@ -0,0 +1,35 @@
+/* memcpy for RISC-V Vector.
+ Copyright (C) 2024-2025 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <https: www.gnu.org="" licenses=""></https:>. */
+
+
+#include <sysdep.h>
+#include <sys asm.h="">
+
+ENTRY (__memcpy_vector)
+ mv a6, a0
+L(loop):
+ vsetvli a3,a2,e8,m8,ta,mu
+ vle8.v v8,(a1)
+ vse8.v v8,(a6)
+ add a1,a1,a3
+ sub a2,a2,a3
+ add a6,a6,a3
+ bnez a2,L(loop)
+L(ret):
+ ret
+END (__memcpy_vector)
diff --git a/sysdeps/unix/sysv/linux/riscv/multiarch/Makefile b/sysdeps/unix/sysv/linux/riscv/multiarch/Makefile
index fcef5659d4..a8b6c22af1 100644
--- a/sysdeps/unix/sysv/linux/riscv/multiarch/Makefile
+++ b/sysdeps/unix/sysv/linux/riscv/multiarch/Makefile
@@ -3,7 +3,9 @@ sysdep_routines += \
memcpy \
memcpy-generic \
memcpy_noalignment \
+ memcpy_vector \
# sysdep_routines
CFLAGS-memcpy_noalignment.c += -mno-strict-align
+ASFLAGS-memcpy_vector.S += -march=rv64gcv
endif
diff --git a/sysdeps/unix/sysv/linux/riscv/multiarch/memcpy.c b/sysdeps/unix/sysv/linux/riscv/multiarch/memcpy.c
index 8544f5402a..c9879762f6 100644
--- a/sysdeps/unix/sysv/linux/riscv/multiarch/memcpy.c
+++ b/sysdeps/unix/sysv/linux/riscv/multiarch/memcpy.c
@@ -32,11 +32,16 @@ extern __typeof (__redirect_memcpy) __libc_memcpy;
extern __typeof (__redirect_memcpy) __memcpy_generic attribute_hidden;
extern __typeof (__redirect_memcpy) __memcpy_noalignment attribute_hidden;
+extern __typeof (__redirect_memcpy) __memcpy_vector attribute_hidden;
static inline __typeof (__redirect_memcpy) *
select_memcpy_ifunc (uint64_t dl_hwcap, __riscv_hwprobe_t hwprobe_func)
{
unsigned long long int v;
+ if (__riscv_hwprobe_one (hwprobe_func, RISCV_HWPROBE_KEY_IMA_EXT_0, &v) == 0
+ && (v & RISCV_HWPROBE_IMA_V) == RISCV_HWPROBE_IMA_V)
+ return __memcpy_vector;
+
if (__riscv_hwprobe_one (hwprobe_func, RISCV_HWPROBE_KEY_CPUPERF_0, &v) == 0
&& (v & RISCV_HWPROBE_MISALIGNED_MASK) == RISCV_HWPROBE_MISALIGNED_FAST)
return __memcpy_noalignment;
--
2.25.1</sys></sysdep.h></daichengrong@iscas.ac.cn></daichengrong@iscas.ac.cn>
^ permalink raw reply [flat|nested] 4+ messages in thread
* [PATCH v2] RISC-V: add multiarch RVV support for memcpy using FMV IFUNC
2025-01-16 23:46 ` Andrew Waterman
2025-01-17 8:01 ` [PATCH v2] " daichengrong
@ 2025-01-17 13:01 ` daichengrong
1 sibling, 0 replies; 4+ messages in thread
From: daichengrong @ 2025-01-17 13:01 UTC (permalink / raw)
To: Andrew Waterman; +Cc: Palmer Dabbelt, libc-alpha
[-- Attachment #1: Type: text/plain, Size: 4077 bytes --]
On 2025/1/17 07:46, Andrew Waterman wrote:
> On Thu, Jan 16, 2025 at 1:31 AM 戴成荣<daichengrong@iscas.ac.cn> wrote:
>> With RISC-V GCC Function multi-versioning IFUNC support,glibc could utilize some new RISC-V CPUs with new extensions such as Vector to re-realize or speed up some functions like memcpy.
>>
>> +ENTRY (__memcpy_vector)
>> + beq a2, zero, L(ret)
> This branch should be deleted; the strip-mine loop below does the
> correct thing for a2=0. Including the branch is an anti-optimization:
> it speeds up the uncommon case of size-0 memcpy at the expense of the
> common case.
Thanks! This branch has been deleted.
>> + mv a6, a0
>> +L(loop):
>> + vsetvli a3,a2,e8,m8,ta,mu
>> + vle8.v v8,(a1)
>> + vse8.v v8,(a6)
>> + add a1,a1,a3
>> + sub a2,a2,a3
>> + add a6,a6,a3
>> + bnez a2,L(loop)
>> +L(ret):
>> + ret
>> +END (__memcpy_vector)
Add riscv vector support for memcpy.
Signed-off-by: daichengrong <daichengrong@iscas.ac.cn>
---
sysdeps/riscv/multiarch/memcpy_vector.S | 35 +++++++++++++++++++
.../unix/sysv/linux/riscv/multiarch/Makefile | 2 ++
.../unix/sysv/linux/riscv/multiarch/memcpy.c | 5 +++
3 files changed, 42 insertions(+)
create mode 100644 sysdeps/riscv/multiarch/memcpy_vector.S
diff --git a/sysdeps/riscv/multiarch/memcpy_vector.S
b/sysdeps/riscv/multiarch/memcpy_vector.S
new file mode 100644
index 0000000000..8fddab8432
--- /dev/null
+++ b/sysdeps/riscv/multiarch/memcpy_vector.S
@@ -0,0 +1,35 @@
+/* memcpy for RISC-V Vector.
+ Copyright (C) 2024-2025 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <https://www.gnu.org/licenses/>. */
+
+
+#include <sysdep.h>
+#include <sys/asm.h>
+
+ENTRY (__memcpy_vector)
+ mv a6, a0
+L(loop):
+ vsetvli a3,a2,e8,m8,ta,mu
+ vle8.v v8,(a1)
+ vse8.v v8,(a6)
+ add a1,a1,a3
+ sub a2,a2,a3
+ add a6,a6,a3
+ bnez a2,L(loop)
+L(ret):
+ ret
+END (__memcpy_vector)
diff --git a/sysdeps/unix/sysv/linux/riscv/multiarch/Makefile
b/sysdeps/unix/sysv/linux/riscv/multiarch/Makefile
index fcef5659d4..a8b6c22af1 100644
--- a/sysdeps/unix/sysv/linux/riscv/multiarch/Makefile
+++ b/sysdeps/unix/sysv/linux/riscv/multiarch/Makefile
@@ -3,7 +3,9 @@ sysdep_routines += \
memcpy \
memcpy-generic \
memcpy_noalignment \
+ memcpy_vector \
# sysdep_routines
CFLAGS-memcpy_noalignment.c += -mno-strict-align
+ASFLAGS-memcpy_vector.S += -march=rv64gcv
endif
diff --git a/sysdeps/unix/sysv/linux/riscv/multiarch/memcpy.c
b/sysdeps/unix/sysv/linux/riscv/multiarch/memcpy.c
index 8544f5402a..c9879762f6 100644
--- a/sysdeps/unix/sysv/linux/riscv/multiarch/memcpy.c
+++ b/sysdeps/unix/sysv/linux/riscv/multiarch/memcpy.c
@@ -32,11 +32,16 @@ extern __typeof (__redirect_memcpy) __libc_memcpy;
extern __typeof (__redirect_memcpy) __memcpy_generic attribute_hidden;
extern __typeof (__redirect_memcpy) __memcpy_noalignment attribute_hidden;
+extern __typeof (__redirect_memcpy) __memcpy_vector attribute_hidden;
static inline __typeof (__redirect_memcpy) *
select_memcpy_ifunc (uint64_t dl_hwcap, __riscv_hwprobe_t hwprobe_func)
{
unsigned long long int v;
+ if (__riscv_hwprobe_one (hwprobe_func, RISCV_HWPROBE_KEY_IMA_EXT_0,
&v) == 0
+ && (v & RISCV_HWPROBE_IMA_V) == RISCV_HWPROBE_IMA_V)
+ return __memcpy_vector;
+
if (__riscv_hwprobe_one (hwprobe_func, RISCV_HWPROBE_KEY_CPUPERF_0, &v) == 0
&& (v & RISCV_HWPROBE_MISALIGNED_MASK) == RISCV_HWPROBE_MISALIGNED_FAST)
return __memcpy_noalignment;
--
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2025-01-17 13:01 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2025-01-16 9:30 [PATCH] RISC-V: add multiarch RVV support for memcpy using FMV IFUNC 戴成荣
2025-01-16 23:46 ` Andrew Waterman
2025-01-17 8:01 ` [PATCH v2] " daichengrong
2025-01-17 13:01 ` daichengrong
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).