On 12/16/24 7:15 AM, Mariam Arutunian wrote:
On Mon, Dec 16, 2024 at 5:20 PM Xi Ruoyao <xry...@xry111.site> wrote:
A generic CRC optimization pass has been implemented in r15-5850. But
without target-specific code, it'll only optimize the CRC loop to a
table lookup. With LoongArch-specific code we can do it better: for
64-bit LoongArch and the IEEE 802.3 polynomial or the Castagnoli
polynomial, we can directly invoke the hardware instruction; even if a
hardware instruction isn't available, we can still use LoongArch bit
reversion instructions to optimize the bit reverse before looking up the
table.
Xi Ruoyao (5):
LoongArch: Remove QHSD and use QHWD instead
LoongArch: Add bit reverse operations
LoongArch: Add CRC expander to generate faster CRC
LoongArch: Combine xor and crc instructions
LoongArch: Add crc tests
gcc/config/loongarch/loongarch.md | 138 +++++++++++++++++-
gcc/testsuite/g++.target/loongarch/crc-scan.C | 13 ++
gcc/testsuite/g++.target/loongarch/crc.C | 113 ++++++++++++++
3 files changed, 261 insertions(+), 3 deletions(-)
create mode 100644 gcc/testsuite/g++.target/loongarch/crc-scan.C
create mode 100644 gcc/testsuite/g++.target/loongarch/crc.C
--
2.47.1
Thanks for your work. It's nice to see added support for more architectures!
I’ve looked over the changes, and everything looks good to me.
If any updates are needed on my side, please let me know.
Definitely good news. I didn't know loongarch had suitable instructions
for this. If I did I would have reached out to Xi earlier :-)
Jeff
