diff options
| author | Paul Duncan <pabs@pablotron.org> | 2024-05-29 22:05:03 -0400 |
|---|---|---|
| committer | Paul Duncan <pabs@pablotron.org> | 2024-05-29 22:05:03 -0400 |
| commit | 161a4f1bedd29fa19896c44295267bbf89dd7afa (patch) | |
| tree | 48e8893e7230ec99644b21062cd1e3273caaf21c | |
| parent | f3e10c9418998efdcfda829fd43d08c875e7e528 (diff) | |
| download | sha3-161a4f1bedd29fa19896c44295267bbf89dd7afa.tar.bz2 sha3-161a4f1bedd29fa19896c44295267bbf89dd7afa.zip | |
| -rw-r--r-- | README.md | 8 |
1 files changed, 6 insertions, 2 deletions
@@ -211,6 +211,8 @@ The available backends are: - Scalar (`BACKEND=1`): Default if no faster backend is available. - [AVX-512][] (`BACKEND=2`): [AVX-512][] acceleration. Selected by default if [AVX-512][] is supported. +- [AVX2][] (`BACKEND=6`): [AVX2][] acceleration. Currently slower than + the scalar backend and not enabled by default. - [Neon][] (`BACKEND=3`): ARM [Neon][] acceleration. Currently slower than the scalar backend on ARM CPUs and not enabled by default. @@ -225,8 +227,8 @@ A minimal [libcpucycles][]-based benchmarking tool is available in byte (cpb)][cpb] for a variety of message lengths, then prints a table of results to standard output in [CSV][] format. -The results from running `bench` on a couple of my systems are available -in the tables below. +`bench` results from several of my systems are shown in the tables +below. ### Lenovo ThinkPad X1 Carbon, 6th Gen (i7-1185G7, AVX-512 Backend) @@ -352,3 +354,5 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. "Observed CPU cycles divided by the number of input bytes." [median]: https://en.wikipedia.org/wiki/Median "Median" +[avx2]: https://en.wikipedia.org/wiki/Advanced_Vector_Extensions#Advanced_Vector_Extensions_2 + "Advanced Vector Extensions 2: 256-bit SIMD vector instruction set" |