musl - [PATCH v3 0/1] riscv64: Add RVV optimized memset implementation

Follow @Openwall on Twitter for new release announcements and other news

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-Id: <20251030120908.47909-1-pincheng.plct@isrc.iscas.ac.cn>
Date: Thu, 30 Oct 2025 20:09:07 +0800
From: Pincheng Wang <pincheng.plct@...c.iscas.ac.cn>
To: musl@...ts.openwall.com
Cc: pincheng.plct@...c.iscas.ac.cn
Subject: [PATCH v3 0/1] riscv64: Add RVV optimized memset implementation

Hi all,

This is v3 of the RISC-V Vector (RVV) optimized memset patch.

Changes from v2:
- Drop global vector disable in arch.mak; explicitly enable 'V'
  extension via `.option arch` in memset_vector.S.
- Use __getauxval instead of getauxval and rename memset_{scalar,vect}
  to __memset_{scalar,vect} to avoid symbol conflicts.
- Remove the small-size fast path; now use vector code uniformly for all
  sizes to avoid branch mispredictions and simplify control flow.

Changes from v1:
- Replaced compile-time detection (__riscv_vetor macro) with runtime
  detection using AT_HWCAP, addressing the main concern from v1
  feedback.
- Introduced a dispatch mechanism (memset_dispatch.c) that selects
  appropriate implementation at process startup.

Implementation details:
- memset.c is renamed internally to __memset_scalar via macro,
  preserving the generic C implementation scalar fallback.
- memset_vector.S provides the optimized __memset_vector symbol.
- memset_dispatch.c exports the public memset() symbol, which dispatches
  via function pointer.
- __init_riscv_string_optimizations is called in __libc_start_main to
  initialize the function pointer based on AT_HWCAP.
- The vector implementation uses vsetvli for bulk fills.

Performance:
- On Spacemit X60: up to ~3.36x faster (256B), with consistent gains
  across medium and large sizes.

For very small sizes (<8 bytes), there can be minor regressions compared
to the generic C version. This is trade off for the significant gains on
larger sizes.

Testing:
- QEMU with QEMU_CPU="rv64,v=true" and "rv64,v=false".
- Spacemit X60 with V extension support.
- CFLAGS="-march=rv64gc" and "-march=rv64gcv".
Functional behavior matches generic memset.

Thanks,
Pincheng Wang

Pincheng Wang (1):
  riscv64: add runtime-detected vector optimized memset

 src/env/__libc_start_main.c          |  3 +++
 src/internal/libc.h                  |  3 +++
 src/string/riscv64/memset.c          |  4 ++++
 src/string/riscv64/memset_dispatch.c | 36 ++++++++++++++++++++++++++++
 src/string/riscv64/memset_vector.S   | 28 ++++++++++++++++++++++
 5 files changed, 74 insertions(+)
 create mode 100644 src/string/riscv64/memset.c
 create mode 100644 src/string/riscv64/memset_dispatch.c
 create mode 100644 src/string/riscv64/memset_vector.S

-- 
2.39.5

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.