Add hybrid Neon/Neon Keccak-f1600, don't use scalar ASM on Cortex-A72 (…

…#139) * FIPS202: Only pick lazy-rotation assembly for v8.2-A+ Signed-off-by: Hanno Becker <[email protected]> * Add hybrid Neon/Neon Keccak-f1600 ASM Signed-off-by: Hanno Becker <[email protected]> * Add SHA3 extension flag to Graviton3 build Signed-off-by: Hanno Becker <[email protected]> * Don't try to detect slow barrel shifter The `major*100 + minor` semantics of `__ARM_ARCH` is not well-supported it seems. Signed-off-by: Hanno Becker <[email protected]> --------- Signed-off-by: Hanno Becker <[email protected]>
pq-code-package · Sep 19, 2024 · fc539c3 · fc539c3
1 parent efadcf9
commit fc539c3
Show file tree

Hide file tree

Showing 4 changed files with 467 additions and 28 deletions.
diff --git a/.github/workflows/bench.yml b/.github/workflows/bench.yml
@@ -17,7 +17,7 @@ jobs:
         - system: rpi4
           name: Arm Cortex-A72 (Raspberry Pi 4) benchmarks
           bench_pmu: PMU
-          archflags: -mcpu=cortex-a72
+          archflags: -mcpu=cortex-a72 -DSYS_AARCH64_SLOW_BARREL_SHIFTER
           cflags: -DFORCE_AARCH64
           bench_extra_args: ""
         - system: rpi5

diff --git a/.github/workflows/bench_ec2_all.yml b/.github/workflows/bench_ec2_all.yml
@@ -31,7 +31,7 @@ jobs:
           - name: Graviton3
             ec2_instance_type: c7g.medium
             ec2_ami: ubuntu-latest (aarch64)
-            archflags: -mcpu=neoverse-v1 -march=armv8.4-a
+            archflags: -march=armv8.4-a+sha3
             cflags: -DFORCE_AARCH64
     uses: ./.github/workflows/bench_ec2_reusable.yml
     if: github.repository_owner == 'pq-code-package' && (github.event.label.name == 'benchmark' || github.ref == 'refs/heads/main')