From: Adhemerval Zanella Netto <adhemerval.zanella@linaro.org>
To: Hau Hsu <hau.hsu@sifive.com>,
libc-alpha@sourceware.org, hongrong.hsu@sifive.com,
jerry.shih@sifive.com, nick.knight@sifive.com,
kito.cheng@sifive.com
Cc: greentime.hu@sifive.com, alice.chan@sifive.com,
andrew@sifive.com, vincent.chen@sifive.com
Subject: Re: [PATCH v2 2/5] riscv: vectorized mem* functions
Date: Fri, 21 Apr 2023 09:12:03 -0300 [thread overview]
Message-ID: <61819bfe-c1cb-c898-11e6-795f1b2da0d0@linaro.org> (raw)
In-Reply-To: <20230421075405.14892-3-hau.hsu@sifive.com>
On 21/04/23 04:54, Hau Hsu via Libc-alpha wrote:
> From: Jerry Shih <jerry.shih@sifive.com>
>
> This patch proposes implementations of memchr, memcmp, memcpy, memmove,
> and memset that leverage the RISC-V V extension (RVV), version 1.0.
> These routines assumes VLEN is at least 32 bits, as is required by all
> currently defined vector extensions, and they support arbitrarily large
> VLEN. All implementations work for both RV32 and RV64 platforms, and
> make no assumptions about page size.
This is not a full review, just some remark skimming through the patch.
> ---
> sysdeps/riscv/rvv/memchr.S | 63 +++++++++++++++++++++++++++++++
> sysdeps/riscv/rvv/memcmp.S | 75 +++++++++++++++++++++++++++++++++++++
> sysdeps/riscv/rvv/memcpy.S | 51 +++++++++++++++++++++++++
> sysdeps/riscv/rvv/memmove.S | 72 +++++++++++++++++++++++++++++++++++
> sysdeps/riscv/rvv/memset.S | 51 +++++++++++++++++++++++++
> 5 files changed, 312 insertions(+)
> create mode 100644 sysdeps/riscv/rvv/memchr.S
> create mode 100644 sysdeps/riscv/rvv/memcmp.S
> create mode 100644 sysdeps/riscv/rvv/memcpy.S
> create mode 100644 sysdeps/riscv/rvv/memmove.S
> create mode 100644 sysdeps/riscv/rvv/memset.S
>
> diff --git a/sysdeps/riscv/rvv/memchr.S b/sysdeps/riscv/rvv/memchr.S
> new file mode 100644
> index 0000000000..6981a9f8b0
> --- /dev/null
> +++ b/sysdeps/riscv/rvv/memchr.S
> @@ -0,0 +1,63 @@
> +/* RVV versions memchr. RISC-V version.
> + Copyright (C) 2023 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> + Contributed by Jerry Shih <jerry.shih@sifive.com>.
We don't use 'Contributed by' anymore.
> +
> + 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
> + <http://www.gnu.org/licenses/>. */
> +
> +#include <sysdep.h>
> +#include <sys/asm.h>
> +
> +#define iResult a0
> +
> +#define pSrc a0
> +#define iValue a1
> +#define iNum a2
> +
> +#define iVL a3
> +#define iTemp a4
> +
> +#define ELEM_LMUL_SETTING m8
> +#define vData v0
> +#define vMask v8
We avoid to use camelcase, even for assembly implementations.
> +
> +ENTRY(memchr)
> +
> +L(loop):
> + vsetvli zero, iNum, e8, ELEM_LMUL_SETTING, ta, ma
> +
> + vle8ff.v vData, (pSrc)
> + /* Find the iValue inside the loaded data. */
> + vmseq.vx vMask, vData, iValue
> + vfirst.m iTemp, vMask
> +
> + /* Skip the loop if we find the matched value. */
> + bgez iTemp, L(found)
> +
> + csrr iVL, vl
> + sub iNum, iNum, iVL
> + add pSrc, pSrc, iVL
> +
> + bnez iNum, L(loop)
> +
> + li iResult, 0
> + ret
> +
> +L(found):
> + add iResult, pSrc, iTemp
> + ret
> +
> +END(memchr)
> +libc_hidden_builtin_def (memchr)
> diff --git a/sysdeps/riscv/rvv/memcmp.S b/sysdeps/riscv/rvv/memcmp.S
> new file mode 100644
> index 0000000000..b156ec524c
> --- /dev/null
> +++ b/sysdeps/riscv/rvv/memcmp.S
> @@ -0,0 +1,75 @@
> +/* RVV versions memcmp. RISC-V version.
> + Copyright (C) 2023 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> + Contributed by Jerry Shih <jerry.shih@sifive.com>.
> +
> + 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
> + <http://www.gnu.org/licenses/>. */
> +
> +#include <sysdep.h>
> +#include <sys/asm.h>
> +
> +#define iResult a0
> +
> +#define pSrc1 a0
> +#define pSrc2 a1
> +#define iNum a2
> +
> +#define iVL a3
> +#define iTemp a4
> +#define iTemp1 a5
> +#define iTemp2 a6
> +
> +#define ELEM_LMUL_SETTING m8
> +#define vData1 v0
> +#define vData2 v8
> +#define vMask v16
> +
> +ENTRY(memcmp)
> +
> +L(loop):
> + vsetvli iVL, iNum, e8, ELEM_LMUL_SETTING, ta, ma
> +
> + vle8.v vData1, (pSrc1)
> + vle8.v vData2, (pSrc2)
> +
> + vmsne.vv vMask, vData1, vData2
> + sub iNum, iNum, iVL
> + vfirst.m iTemp, vMask
> +
> + /* Skip the loop if we find the different value between pSrc1 and pSrc2. */
> + bgez iTemp, L(found)
> +
> + add pSrc1, pSrc1, iVL
> + add pSrc2, pSrc2, iVL
> +
> + bnez iNum, L(loop)
> +
> + li iResult, 0
> + ret
> +
> +L(found):
> + add pSrc1, pSrc1, iTemp
> + add pSrc2, pSrc2, iTemp
> + lbu iTemp1, 0(pSrc1)
> + lbu iTemp2, 0(pSrc2)
> + sub iResult, iTemp1, iTemp2
> + ret
> +
> +END(memcmp)
> +libc_hidden_builtin_def (memcmp)
> +weak_alias (memcmp,bcmp)
> +strong_alias (memcmp, __memcmpeq)
> +libc_hidden_def (__memcmpeq)
> +
> diff --git a/sysdeps/riscv/rvv/memcpy.S b/sysdeps/riscv/rvv/memcpy.S
> new file mode 100644
> index 0000000000..de790fbe51
> --- /dev/null
> +++ b/sysdeps/riscv/rvv/memcpy.S
> @@ -0,0 +1,51 @@
> +/* RVV versions memcpy. RISC-V version.
> + Copyright (C) 2023 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> + Contributed by Jerry Shih <jerry.shih@sifive.com>.
> +
> + 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
> + <http://www.gnu.org/licenses/>. */
> +
> +#include <sysdep.h>
> +#include <sys/asm.h>
> +
> +#define pDst a0
> +#define pSrc a1
> +#define iNum a2
> +
> +#define iVL a3
> +#define pDstPtr a4
> +
> +#define ELEM_LMUL_SETTING m8
> +#define vData v0
> +
> +ENTRY(memcpy)
> +
> + mv pDstPtr, pDst
> +
> +L(loop):
> + vsetvli iVL, iNum, e8, ELEM_LMUL_SETTING, ta, ma
> +
> + vle8.v vData, (pSrc)
> + sub iNum, iNum, iVL
> + add pSrc, pSrc, iVL
> + vse8.v vData, (pDstPtr)
> + add pDstPtr, pDstPtr, iVL
> +
> + bnez iNum, L(loop)
> +
> + ret
> +
> +END(memcpy)
> +libc_hidden_builtin_def (memcpy)
> diff --git a/sysdeps/riscv/rvv/memmove.S b/sysdeps/riscv/rvv/memmove.S
> new file mode 100644
> index 0000000000..ed12744064
> --- /dev/null
> +++ b/sysdeps/riscv/rvv/memmove.S
> @@ -0,0 +1,72 @@
> +/* RVV versions memmove. RISC-V version.
> + Copyright (C) 2023 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> + Contributed by Jerry Shih <jerry.shih@sifive.com>.
> +
> + 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
> + <http://www.gnu.org/licenses/>. */
> +
> +#include <sysdep.h>
> +#include <sys/asm.h>
> +
> +#define pDst a0
> +#define pSrc a1
> +#define iNum a2
> +
> +#define iVL a3
> +#define pDstPtr a4
> +#define pSrcBackwardPtr a5
> +#define pDstBackwardPtr a6
> +
> +#define ELEM_LMUL_SETTING m8
> +#define vData v0
> +
> +ENTRY(memmove)
> +
> + mv pDstPtr, pDst
> +
> + /* If pSrc is equal or after pDst, all data in pSrc will be loaded before
> + overwrited for the overlapping case. We could use faster `forward-copy`. */
> + bgeu pSrc, pDst, L(forward_copy_loop)
> + add pSrcBackwardPtr, pSrc, iNum
> + add pDstBackwardPtr, pDst, iNum
> + /* If pDst inside source data range, we need to use `backward_copy_loop` to
> + handle the overlapping issue. */
> + bltu pDst, pSrcBackwardPtr, L(backward_copy_loop)
> +
> +L(forward_copy_loop):
> + vsetvli iVL, iNum, e8, ELEM_LMUL_SETTING, ta, ma
> +
> + vle8.v vData, (pSrc)
> + sub iNum, iNum, iVL
> + add pSrc, pSrc, iVL
> + vse8.v vData, (pDstPtr)
> + add pDstPtr, pDstPtr, iVL
> +
> + bnez iNum, L(forward_copy_loop)
> + ret
> +
> +L(backward_copy_loop):
> + vsetvli iVL, iNum, e8, ELEM_LMUL_SETTING, ta, ma
> +
> + sub pSrcBackwardPtr, pSrcBackwardPtr, iVL
> + vle8.v vData, (pSrcBackwardPtr)
> + sub iNum, iNum, iVL
> + sub pDstBackwardPtr, pDstBackwardPtr, iVL
> + vse8.v vData, (pDstBackwardPtr)
> + bnez iNum, L(backward_copy_loop)
> + ret
> +
> +END(memmove)
> +libc_hidden_builtin_def (memmove)
> diff --git a/sysdeps/riscv/rvv/memset.S b/sysdeps/riscv/rvv/memset.S
> new file mode 100644
> index 0000000000..3a6c3d0afd
> --- /dev/null
> +++ b/sysdeps/riscv/rvv/memset.S
> @@ -0,0 +1,51 @@
> +/* RVV versions memset. RISC-V version.
> + Copyright (C) 2023 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> + Contributed by Jerry Shih <jerry.shih@sifive.com>.
> +
> + 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
> + <http://www.gnu.org/licenses/>. */
> +
> +#include <sysdep.h>
> +#include <sys/asm.h>
> +
> +#define pDst a0
> +#define iValue a1
> +#define iNum a2
> +
> +#define iVL a3
> +#define iTemp a4
> +#define pDstPtr a5
> +
> +#define ELEM_LMUL_SETTING m8
> +#define vData v0
> +
> +ENTRY(memset)
> +
> + mv pDstPtr, pDst
> +
> + vsetvli iVL, iNum, e8, ELEM_LMUL_SETTING, ta, ma
> + vmv.v.x vData, iValue
> +
> +L(loop):
> + vse8.v vData, (pDstPtr)
> + sub iNum, iNum, iVL
> + add pDstPtr, pDstPtr, iVL
> + vsetvli iVL, iNum, e8, ELEM_LMUL_SETTING, ta, ma
> + bnez iNum, L(loop)
> +
> + ret
> +
> +END(memset)
> +libc_hidden_builtin_def (memset)
next prev parent reply other threads:[~2023-04-21 12:12 UTC|newest]
Thread overview: 11+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-04-21 7:54 [PATCH v2 0/5] riscv: Vectorized mem*/str* function Hau Hsu
2023-04-21 7:54 ` [PATCH v2 1/5] riscv: Enabling vectorized mem*/str* functions in build time Hau Hsu
2023-04-21 7:54 ` [PATCH v2 2/5] riscv: vectorized mem* functions Hau Hsu
2023-04-21 12:12 ` Adhemerval Zanella Netto [this message]
2023-04-21 7:54 ` [PATCH v2 3/5] riscv: vectorized str* functions Hau Hsu
2023-04-21 12:14 ` Adhemerval Zanella Netto
2023-04-21 7:54 ` [PATCH v2 4/5] riscv: vectorized strchr and strnlen functions Hau Hsu
2023-04-21 7:54 ` [PATCH v2 5/5] riscv: add vectorized __memcmpeq Hau Hsu
2023-04-21 12:09 ` [PATCH v2 0/5] riscv: Vectorized mem*/str* function Adhemerval Zanella Netto
2023-04-26 3:11 ` Hau Hsu
-- strict thread matches above, loose matches on Subject: below --
2023-04-21 7:29 [PATCH v2 2/5] riscv: vectorized mem* functions Hau Hsu
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=61819bfe-c1cb-c898-11e6-795f1b2da0d0@linaro.org \
--to=adhemerval.zanella@linaro.org \
--cc=alice.chan@sifive.com \
--cc=andrew@sifive.com \
--cc=greentime.hu@sifive.com \
--cc=hau.hsu@sifive.com \
--cc=hongrong.hsu@sifive.com \
--cc=jerry.shih@sifive.com \
--cc=kito.cheng@sifive.com \
--cc=libc-alpha@sourceware.org \
--cc=nick.knight@sifive.com \
--cc=vincent.chen@sifive.com \
/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).