commit d5efc1d3ff93f16f9ee6f5c2862ffd33c21eb1f8
parent 721000be4b2df7f3d084e9dce86bb229036393dc
Author: finwo <finwo@pm.me>
Date: Sat, 10 Oct 2026 16:16:03 +0200
Add avx512 backend
Diffstat:
4 files changed, 127 insertions(+), 4 deletions(-)
diff --git a/README.md b/README.md
@@ -33,11 +33,12 @@ dep add finwo/keccak-fast
| scalar | yes | — |
| scalar+bmi | yes | — |
| AVX2 | no | x4 |
+| AVX-512 | no | x8 |
Backends register from constructors; the highest available priority wins
(scalar 0, scalar+bmi 10, avx2 20, avx512 30). `kf_backend_name()` and
-`kf_batch_name()` report the active ones. `KECCAK_FAST_NO_BMI` and
-`KECCAK_FAST_NO_AVX2` drop those backends.
+`kf_batch_name()` report the active ones. `KECCAK_FAST_NO_BMI`,
+`KECCAK_FAST_NO_AVX2` and `KECCAK_FAST_NO_AVX512` drop those backends.
The scalar backend is the donor, pinned to the x86-64 baseline
(`no-avx,no-avx2,no-avx512f,no-bmi,no-bmi2`) so a consumer's `-march=native`
diff --git a/bench/bench.c b/bench/bench.c
@@ -134,7 +134,7 @@ int main(void) {
for (size_t i = 0; i < sizeof bin; i++) bin[i] = (uint8_t)(i * 17 + 3);
printf("batch SHAKE256 32->32, per hash (%s)\n", kf_batch_name());
- case_batch(kf_batch_name(), kf_shake256_batch, 4, 32, 32, bin, bout, reps,
+ case_batch(kf_batch_name(), kf_shake256_batch, 8, 32, 32, bin, bout, reps,
target_ns);
case_batch(kf_batch_name(), kf_shake256_batch, 64, 32, 32, bin, bout, reps,
target_ns);
@@ -142,7 +142,7 @@ int main(void) {
target_ns);
printf("batch SHA3-256 32->32, per hash (%s)\n", kf_batch_name());
- case_batch(kf_batch_name(), kf_sha3_256_batch, 4, 32, 32, bin, bout, reps,
+ case_batch(kf_batch_name(), kf_sha3_256_batch, 8, 32, 32, bin, bout, reps,
target_ns);
case_batch(kf_batch_name(), kf_sha3_256_batch, 64, 32, 32, bin, bout, reps,
target_ns);
diff --git a/export.mk b/export.mk
@@ -2,3 +2,4 @@ SRC+={{module.dirname}}/src/keccak-fast.c
SRC+={{module.dirname}}/src/backend/scalar.c
SRC+={{module.dirname}}/src/backend/scalar_bmi.c
SRC+={{module.dirname}}/src/backend/avx2.c
+SRC+={{module.dirname}}/src/backend/avx512.c
diff --git a/src/backend/avx512.c b/src/backend/avx512.c
@@ -0,0 +1,121 @@
+#include <immintrin.h>
+#include <stddef.h>
+#include <stdint.h>
+
+#include "../keccak-fast-internal.h"
+
+#if (defined(__x86_64__) || defined(_M_X64) || defined(__i386__)) && \
+ !defined(KECCAK_FAST_NO_AVX512)
+
+static const uint64_t RC[24] = {
+ 1ULL, 0x8082ULL, 0x800000000000808aULL, 0x8000000080008000ULL,
+ 0x808bULL, 0x80000001ULL, 0x8000000080008081ULL, 0x8000000000008009ULL,
+ 0x8aULL, 0x88ULL, 0x80008009ULL, 0x8000000aULL,
+ 0x8000808bULL, 0x800000000000008bULL, 0x8000000000008089ULL, 0x8000000000008003ULL,
+ 0x8000000000008002ULL, 0x8000000000000080ULL, 0x800aULL, 0x800000008000000aULL,
+ 0x8000000080008081ULL, 0x8000000000008080ULL, 0x80000001ULL, 0x8000000080008008ULL};
+
+__attribute__((target("avx512f")))
+static inline void kf_avx512_round_x8(__m512i v[25], uint64_t rc) {
+ __m512i c[5];
+ __m512i d[5];
+ __m512i t, u;
+
+ for (int x = 0; x < 5; x++) {
+ c[x] = _mm512_ternarylogic_epi64(v[x], v[x + 5], v[x + 10], 0x96);
+ c[x] = _mm512_ternarylogic_epi64(c[x], v[x + 15], v[x + 20], 0x96);
+ }
+ for (int x = 0; x < 5; x++) {
+ d[x] = _mm512_xor_si512(c[(x + 4) % 5], _mm512_rol_epi64(c[(x + 1) % 5], 1));
+ }
+ for (int y = 0; y < 5; y++) {
+ for (int x = 0; x < 5; x++) {
+ v[x + 5 * y] = _mm512_xor_si512(v[x + 5 * y], d[x]);
+ }
+ }
+
+ t = v[1];
+#define KF_RHOPI8(dest, rot) \
+ do { \
+ u = v[dest]; \
+ v[dest] = _mm512_rol_epi64(t, rot); \
+ t = u; \
+ } while (0)
+ KF_RHOPI8(10, 1);
+ KF_RHOPI8(7, 3);
+ KF_RHOPI8(11, 6);
+ KF_RHOPI8(17, 10);
+ KF_RHOPI8(18, 15);
+ KF_RHOPI8(3, 21);
+ KF_RHOPI8(5, 28);
+ KF_RHOPI8(16, 36);
+ KF_RHOPI8(8, 45);
+ KF_RHOPI8(21, 55);
+ KF_RHOPI8(24, 2);
+ KF_RHOPI8(4, 14);
+ KF_RHOPI8(15, 27);
+ KF_RHOPI8(23, 41);
+ KF_RHOPI8(19, 56);
+ KF_RHOPI8(13, 8);
+ KF_RHOPI8(12, 25);
+ KF_RHOPI8(2, 43);
+ KF_RHOPI8(20, 62);
+ KF_RHOPI8(14, 18);
+ KF_RHOPI8(22, 39);
+ KF_RHOPI8(9, 61);
+ KF_RHOPI8(6, 20);
+ KF_RHOPI8(1, 44);
+#undef KF_RHOPI8
+
+ for (int y = 0; y < 5; y++) {
+ __m512i row[5];
+ for (int x = 0; x < 5; x++) row[x] = v[x + 5 * y];
+ for (int x = 0; x < 5; x++) {
+ v[x + 5 * y] =
+ _mm512_ternarylogic_epi64(row[x], row[(x + 1) % 5], row[(x + 2) % 5], 0xD2);
+ }
+ }
+
+ v[0] = _mm512_xor_si512(v[0], _mm512_set1_epi64((long long)rc));
+}
+
+__attribute__((target("avx512f")))
+static void kf_permute_x8(uint64_t s[8][25]) {
+ __m512i v[25];
+ uint64_t tmp[8];
+
+ for (int j = 0; j < 25; j++) {
+ for (int i = 0; i < 8; i++) tmp[i] = s[i][j];
+ v[j] = _mm512_loadu_si512((const __m512i *)tmp);
+ }
+ for (int i = 0; i < 24; i++) kf_avx512_round_x8(v, RC[i]);
+ for (int j = 0; j < 25; j++) {
+ _mm512_storeu_si512((__m512i *)tmp, v[j]);
+ for (int i = 0; i < 8; i++) s[i][j] = tmp[i];
+ }
+}
+
+__attribute__((target("avx512f")))
+static void kf_avx512_batch(uint64_t states[][25], size_t count) {
+ size_t i = 0;
+ for (; i + 8 <= count; i += 8) {
+ kf_permute_x8(&states[i]);
+ }
+ for (; i < count; i++) {
+ kf_scalar_permute(states[i]);
+ }
+}
+
+__attribute__((constructor)) static void kf_avx512_register(void) {
+ if (!__builtin_cpu_supports("avx512f")) {
+ return;
+ }
+ if (kf_batch_priority <= KF_PRIO_AVX512) {
+ kf_batch_perm = kf_avx512_batch;
+ kf_batch_priority = KF_PRIO_AVX512;
+ kf_batch_lanes = 8;
+ kf_batch_label = "avx512";
+ }
+}
+
+#endif
