Changes in v3:
1)Change title for patch2.
2)Apply correct patch version for patch2.
Thanks to Sunyuechi for the reminder.
Changes in v2:
1)Modify some codes and descriptions according to Sunyuechi's
suggestions.
2)Removed benchmark_report.txt to avoid warnings.
[PATCH v1]
I've implemented optimizations to rte_memcpy targeting RISC-V
architectures, achieving an average 10%~15% reduction in execution time
for data sizes between 129 to 1024 bytes( 1025~1600 gains little).
These enhancements draw inspiration from x86 implementations,
specifically focusing on:
1)Alignment Handling for Unaligned Scenarios
2)Vector Configuration Tuning
3)Strategic Prefetching with zicbop
- Patch 1: Cover letter
- Patch 2: Base implementation
- Patch 3: Benchmark report
Tested on Tested on SG2044 (VLEN=128)
Qiguo Chen (1):
lib/eal/riscv: optimize rte_memcpy with RISCV vector and zicbop
extensions
.mailmap | 1 +
config/riscv/meson.build | 14 ++
lib/eal/riscv/include/rte_memcpy.h | 303 ++++++++++++++++++++++++++++-
3 files changed, 316 insertions(+), 2 deletions(-)
--
2.21.0.windows.1