Add PMU cycle counting for Armv8.1-M by mkannwischer · Pull Request #1514 · pq-code-package/mlkem-native

mkannwischer · 2026-01-21T02:40:54Z

Add PMU-based cycle counting support for Armv8.1-M Cortex-M processors.
This uses the CMSIS PMU APIs for portable cycle counter access.

Resolves Armv8.1-M: Add PMU benchmarking code #1502
Hoisted out from Add native implementations for Keccak on Armv8.1-m mldsa-native#511

oqs-bot · 2026-01-21T03:14:38Z

CBMC Results (ML-KEM-512)

Full Results (139 proofs)

Proof	Status	Current	Previous	Change
`TOTAL`	✅	1236s	1109s	+11.5%
`mlk_indcpa_enc`	✅	190s	174s	+9%
`mlk_indcpa_keypair_derand`	✅	186s	172s	+8%
`mlk_keccak_squeezeblocks_x4`	✅	140s	121s	+16%
`mlk_rej_uniform_c`	✅	95s	71s	+34%
`mlk_polyvec_basemul_acc_montgomery_cached_c`	✅	54s	41s	+32%
`mlk_poly_rej_uniform`	✅	38s	33s	+15%
`poly_ntt_native`	✅	32s	24s	+33%
`mlk_ntt_layer`	✅	25s	20s	+25%
`polyvec_basemul_acc_montgomery_cached_native`	✅	24s	22s	+9%
`keccakf1600x4_permute_native_x4`	✅	21s	21s	+0%
`mlk_poly_reduce_native`	✅	16s	14s	+14%
`mlk_poly_sub`	✅	12s	8s	+50%
`mlk_polyvec_add`	✅	11s	8s	+38%
`mlk_keccak_absorb_once_x4`	✅	10s	11s	-9%
`mlk_ntt_butterfly_block`	✅	10s	8s	+25%
`mlk_indcpa_dec`	✅	9s	10s	-10%
`mlk_keccak_squeezeblocks`	✅	9s	8s	+12%
`mlk_poly_frombytes_native`	✅	9s	8s	+12%
`mlk_poly_rej_uniform_x4`	✅	8s	9s	-11%
`keccakf1600_permute_native`	✅	7s	4s	+75%
`mlk_fqmul`	✅	7s	6s	+17%
`kem_dec`	✅	6s	3s	+100%
`mlk_keccak_squeeze_once`	✅	6s	6s	+0%
`mlk_poly_frommsg`	✅	6s	8s	-25%
`mlk_keccak_absorb_once`	✅	5s	4s	+25%
`mlk_keccakf1600_extract_bytes (big endian)`	✅	5s	2s	+150%
`mlk_polymat_permute_bitrev_to_custom`	✅	5s	4s	+25%
`mlk_polyvec_tobytes`	✅	5s	4s	+25%
`mlk_scalar_compress_d1`	✅	5s	2s	+150%
`poly_getnoise_eta1122_4x_native`	✅	5s	2s	+150%
`keccak_f1600_x4_native_aarch64_v8a_v84a_scalar_hybrid`	✅	4s	2s	+100%
`kem_enc_derand`	✅	4s	4s	+0%
`mlk_gen_matrix_serial`	✅	4s	4s	+0%
`mlk_keccakf1600_permute`	✅	4s	1s	+300%
`mlk_poly_cbd_eta1`	✅	4s	2s	+100%
`mlk_poly_compress_du`	✅	4s	3s	+33%
`mlk_poly_frombytes`	✅	4s	2s	+100%
`mlk_poly_invntt_tomont_c`	✅	4s	4s	+0%
`mlk_poly_ntt_c`	✅	4s	1s	+300%
`mlk_polyvec_basemul_acc_montgomery_cached`	✅	4s	3s	+33%
`mlk_polyvec_invntt_tomont`	✅	4s	4s	+0%
`mlk_polyvec_permute_bitrev_to_custom_native`	✅	4s	1s	+300%
`mlk_scalar_decompress_d11`	✅	4s	2s	+100%
`mlk_scalar_signed_to_unsigned_q`	✅	4s	2s	+100%
`mlk_shake256x4`	✅	4s	3s	+33%
`polyvec_basemul_acc_montgomery_cached_k4_native_aarch64`	✅	4s	1s	+300%
`sys_check_capability`	✅	4s	4s	+0%
`keccak_f1600_x4_native_aarch64_v8a_scalar_hybrid`	✅	3s	2s	+50%
`kem_check_sk`	✅	3s	3s	+0%
`kem_enc`	✅	3s	1s	+200%
`kem_keypair_derand`	✅	3s	4s	-25%
`mlk_ct_cmask_nonzero_u16`	✅	3s	3s	+0%
`mlk_ct_cmov_zero`	✅	3s	2s	+50%
`mlk_ct_get_optblocker_i32`	✅	3s	3s	+0%
`mlk_ct_get_optblocker_u32`	✅	3s	1s	+200%
`mlk_ct_get_optblocker_u8`	✅	3s	2s	+50%
`mlk_invntt_layer`	✅	3s	6s	-50%
`mlk_keccakf1600_extract_bytes`	✅	3s	3s	+0%
`mlk_keccakf1600x4_xor_bytes`	✅	3s	2s	+50%
`mlk_matvec_mul`	✅	3s	2s	+50%
`mlk_poly_add`	✅	3s	1s	+200%
`mlk_poly_compress_dv`	✅	3s	3s	+0%
`mlk_poly_getnoise_eta1122_4x`	✅	3s	2s	+50%
`mlk_poly_getnoise_eta2`	✅	3s	4s	-25%
`mlk_poly_invntt_tomont`	✅	3s	2s	+50%
`mlk_poly_mulcache_compute_c`	✅	3s	3s	+0%
`mlk_poly_mulcache_compute_native`	✅	3s	6s	-50%
`mlk_poly_reduce`	✅	3s	2s	+50%
`mlk_poly_tobytes`	✅	3s	3s	+0%
`mlk_poly_tobytes_c`	✅	3s	3s	+0%
`mlk_poly_tobytes_native`	✅	3s	4s	-25%
`mlk_poly_tomsg`	✅	3s	3s	+0%
`mlk_polyvec_compress_du`	✅	3s	4s	-25%
`mlk_polyvec_frombytes`	✅	3s	2s	+50%
`mlk_polyvec_mulcache_compute`	✅	3s	2s	+50%
`mlk_polyvec_ntt`	✅	3s	4s	-25%
`mlk_scalar_compress_d10`	✅	3s	4s	-25%
`mlk_scalar_compress_d4`	✅	3s	2s	+50%
`mlk_scalar_compress_d5`	✅	3s	4s	-25%
`mlk_scalar_decompress_d4`	✅	3s	3s	+0%
`mlk_scalar_decompress_d5`	✅	3s	4s	-25%
`mlk_sha3_256`	✅	3s	2s	+50%
`poly_reduce_native_aarch64`	✅	3s	1s	+200%
`poly_tobytes_native_aarch64`	✅	3s	2s	+50%
`keccak_f1600_x1_native_aarch64`	✅	2s	1s	+100%
`keccak_f1600_x4_native_aarch64_v84a`	✅	2s	1s	+100%
`kem_check_pk`	✅	2s	4s	-50%
`kem_keypair`	✅	2s	3s	-33%
`mlk_barrett_reduce`	✅	2s	1s	+100%
`mlk_ct_cmask_neg_i16`	✅	2s	1s	+100%
`mlk_ct_memcmp`	✅	2s	3s	-33%
`mlk_ct_sel_int16`	✅	2s	3s	-33%
`mlk_gen_matrix`	✅	2s	5s	-60%
`mlk_keccakf1600_xor_bytes`	✅	2s	1s	+100%
`mlk_keccakf1600x4_extract_bytes`	✅	2s	2s	+0%
`mlk_keccakf1600x4_permute`	✅	2s	2s	+0%
`mlk_montgomery_reduce`	✅	2s	2s	+0%
`mlk_poly_cbd_eta2`	✅	2s	2s	+0%
`mlk_poly_decompress_du`	✅	2s	2s	+0%
`mlk_poly_decompress_dv`	✅	2s	2s	+0%
`mlk_poly_getnoise_eta1_4x`	✅	2s	4s	-50%
`mlk_poly_getnoise_eta1_4x_native`	✅	2s	3s	-33%
`mlk_poly_mulcache_compute`	✅	2s	2s	+0%
`mlk_poly_ntt`	✅	2s	1s	+100%
`mlk_poly_reduce_c`	✅	2s	5s	-60%
`mlk_poly_tomont`	✅	2s	2s	+0%
`mlk_poly_tomont_c`	✅	2s	3s	-33%
`mlk_polyvec_decompress_du`	✅	2s	2s	+0%
`mlk_polyvec_permute_bitrev_to_custom`	✅	2s	1s	+100%
`mlk_polyvec_reduce`	✅	2s	4s	-50%
`mlk_polyvec_tomont`	✅	2s	2s	+0%
`mlk_rej_uniform`	✅	2s	3s	-33%
`mlk_scalar_compress_d11`	✅	2s	1s	+100%
`mlk_sha3_512`	✅	2s	1s	+100%
`mlk_shake128_absorb_once`	✅	2s	2s	+0%
`mlk_shake128_squeezeblocks`	✅	2s	2s	+0%
`mlk_shake128x4_squeezeblocks`	✅	2s	4s	-50%
`mlk_shake256`	✅	2s	1s	+100%
`mlk_value_barrier_u32`	✅	2s	3s	-33%
`mlk_value_barrier_u8`	✅	2s	3s	-33%
`ntt_native_aarch64`	✅	2s	6s	-67%
`poly_invntt_tomont_native`	✅	2s	1s	+100%
`poly_mulcache_compute_native_aarch64`	✅	2s	1s	+100%
`polyvec_basemul_acc_montgomery_cached_k3_native_aarch64`	✅	2s	3s	-33%
`rej_uniform_native_aarch64`	✅	2s	3s	-33%
`intt_native_aarch64`	✅	1s	2s	-50%
`keccak_f1600_x1_native_aarch64_v84a`	✅	1s	2s	-50%
`mlk_check_pct`	✅	1s	4s	-75%
`mlk_ct_cmask_nonzero_u8`	✅	1s	2s	-50%
`mlk_ct_sel_uint8`	✅	1s	3s	-67%
`mlk_keccakf1600_xor_bytes (big endian)`	✅	1s	2s	-50%
`mlk_poly_frombytes_c`	✅	1s	2s	-50%
`mlk_poly_tomont_native`	✅	1s	3s	-67%
`mlk_scalar_decompress_d10`	✅	1s	2s	-50%
`mlk_shake128x4_absorb_once`	✅	1s	2s	-50%
`mlk_value_barrier_i32`	✅	1s	6s	-83%
`poly_tomont_native_aarch64`	✅	1s	1s	+0%
`polyvec_basemul_acc_montgomery_cached_k2_native_aarch64`	✅	1s	3s	-67%
`rej_uniform_native`	✅	1s	2s	-50%

oqs-bot · 2026-01-21T03:17:51Z

CBMC Results (ML-KEM-768)

Full Results (139 proofs)

Proof	Status	Current	Previous	Change
`TOTAL`	✅	1448s	1701s	-14.9%
`mlk_indcpa_keypair_derand`	✅	357s	420s	-15%
`mlk_indcpa_enc`	✅	218s	263s	-17%
`mlk_keccak_squeezeblocks_x4`	✅	126s	142s	-11%
`mlk_rej_uniform_c`	✅	71s	100s	-29%
`mlk_polyvec_basemul_acc_montgomery_cached_c`	✅	68s	104s	-35%
`polyvec_basemul_acc_montgomery_cached_native`	✅	57s	61s	-7%
`poly_ntt_native`	✅	46s	59s	-22%
`mlk_poly_rej_uniform`	✅	40s	45s	-11%
`mlk_ntt_layer`	✅	24s	32s	-25%
`keccakf1600x4_permute_native_x4`	✅	17s	22s	-23%
`mlk_indcpa_dec`	✅	17s	21s	-19%
`mlk_poly_reduce_native`	✅	12s	19s	-37%
`mlk_polyvec_add`	✅	12s	13s	-8%
`mlk_keccak_absorb_once_x4`	✅	11s	10s	+10%
`mlk_ntt_butterfly_block`	✅	9s	11s	-18%
`mlk_poly_frombytes_native`	✅	9s	9s	+0%
`mlk_poly_sub`	✅	9s	14s	-36%
`mlk_poly_frommsg`	✅	8s	7s	+14%
`mlk_invntt_layer`	✅	7s	6s	+17%
`mlk_keccak_squeezeblocks`	✅	7s	7s	+0%
`mlk_poly_rej_uniform_x4`	✅	7s	10s	-30%
`keccakf1600_permute_native`	✅	6s	5s	+20%
`kem_dec`	✅	6s	5s	+20%
`mlk_fqmul`	✅	6s	6s	+0%
`mlk_keccak_squeeze_once`	✅	6s	6s	+0%
`mlk_poly_getnoise_eta1_4x_native`	✅	6s	3s	+100%
`mlk_poly_mulcache_compute`	✅	5s	2s	+150%
`mlk_polymat_permute_bitrev_to_custom`	✅	5s	6s	-17%
`mlk_polyvec_mulcache_compute`	✅	5s	2s	+150%
`mlk_scalar_compress_d5`	✅	5s	1s	+400%
`poly_tobytes_native_aarch64`	✅	5s	2s	+150%
`kem_check_pk`	✅	4s	4s	+0%
`kem_check_sk`	✅	4s	3s	+33%
`kem_enc_derand`	✅	4s	4s	+0%
`mlk_ct_get_optblocker_u8`	✅	4s	1s	+300%
`mlk_gen_matrix_serial`	✅	4s	4s	+0%
`mlk_keccak_absorb_once`	✅	4s	4s	+0%
`mlk_keccakf1600_permute`	✅	4s	4s	+0%
`mlk_matvec_mul`	✅	4s	3s	+33%
`mlk_poly_getnoise_eta1_4x`	✅	4s	6s	-33%
`mlk_poly_reduce_c`	✅	4s	4s	+0%
`mlk_poly_tobytes_native`	✅	4s	5s	-20%
`mlk_value_barrier_i32`	✅	4s	3s	+33%
`poly_getnoise_eta1122_4x_native`	✅	4s	3s	+33%
`poly_mulcache_compute_native_aarch64`	✅	4s	4s	+0%
`intt_native_aarch64`	✅	3s	3s	+0%
`kem_enc`	✅	3s	3s	+0%
`kem_keypair`	✅	3s	2s	+50%
`kem_keypair_derand`	✅	3s	4s	-25%
`mlk_check_pct`	✅	3s	5s	-40%
`mlk_ct_sel_uint8`	✅	3s	2s	+50%
`mlk_gen_matrix`	✅	3s	4s	-25%
`mlk_keccakf1600x4_xor_bytes`	✅	3s	1s	+200%
`mlk_poly_cbd_eta2`	✅	3s	2s	+50%
`mlk_poly_compress_du`	✅	3s	4s	-25%
`mlk_poly_compress_dv`	✅	3s	3s	+0%
`mlk_poly_frombytes`	✅	3s	1s	+200%
`mlk_poly_frombytes_c`	✅	3s	3s	+0%
`mlk_poly_invntt_tomont_c`	✅	3s	1s	+200%
`mlk_poly_ntt`	✅	3s	3s	+0%
`mlk_poly_tobytes`	✅	3s	1s	+200%
`mlk_poly_tobytes_c`	✅	3s	2s	+50%
`mlk_poly_tomont_native`	✅	3s	2s	+50%
`mlk_poly_tomsg`	✅	3s	3s	+0%
`mlk_polyvec_decompress_du`	✅	3s	2s	+50%
`mlk_polyvec_invntt_tomont`	✅	3s	2s	+50%
`mlk_polyvec_ntt`	✅	3s	4s	-25%
`mlk_polyvec_permute_bitrev_to_custom`	✅	3s	1s	+200%
`mlk_polyvec_tomont`	✅	3s	2s	+50%
`mlk_scalar_compress_d1`	✅	3s	1s	+200%
`mlk_scalar_compress_d10`	✅	3s	2s	+50%
`mlk_scalar_compress_d11`	✅	3s	3s	+0%
`mlk_scalar_decompress_d10`	✅	3s	6s	-50%
`mlk_scalar_decompress_d5`	✅	3s	3s	+0%
`mlk_shake256`	✅	3s	2s	+50%
`mlk_shake256x4`	✅	3s	4s	-25%
`polyvec_basemul_acc_montgomery_cached_k2_native_aarch64`	✅	3s	2s	+50%
`rej_uniform_native`	✅	3s	5s	-40%
`keccak_f1600_x1_native_aarch64`	✅	2s	2s	+0%
`keccak_f1600_x4_native_aarch64_v84a`	✅	2s	3s	-33%
`keccak_f1600_x4_native_aarch64_v8a_scalar_hybrid`	✅	2s	2s	+0%
`keccak_f1600_x4_native_aarch64_v8a_v84a_scalar_hybrid`	✅	2s	2s	+0%
`mlk_barrett_reduce`	✅	2s	2s	+0%
`mlk_ct_cmask_neg_i16`	✅	2s	4s	-50%
`mlk_ct_cmask_nonzero_u16`	✅	2s	2s	+0%
`mlk_ct_cmask_nonzero_u8`	✅	2s	1s	+100%
`mlk_ct_cmov_zero`	✅	2s	2s	+0%
`mlk_ct_get_optblocker_u32`	✅	2s	2s	+0%
`mlk_ct_sel_int16`	✅	2s	3s	-33%
`mlk_keccakf1600_extract_bytes`	✅	2s	2s	+0%
`mlk_keccakf1600_extract_bytes (big endian)`	✅	2s	1s	+100%
`mlk_keccakf1600_xor_bytes`	✅	2s	2s	+0%
`mlk_keccakf1600_xor_bytes (big endian)`	✅	2s	2s	+0%
`mlk_keccakf1600x4_extract_bytes`	✅	2s	1s	+100%
`mlk_keccakf1600x4_permute`	✅	2s	2s	+0%
`mlk_montgomery_reduce`	✅	2s	2s	+0%
`mlk_poly_add`	✅	2s	3s	-33%
`mlk_poly_cbd_eta1`	✅	2s	4s	-50%
`mlk_poly_decompress_du`	✅	2s	3s	-33%
`mlk_poly_decompress_dv`	✅	2s	2s	+0%
`mlk_poly_getnoise_eta1122_4x`	✅	2s	4s	-50%
`mlk_poly_getnoise_eta2`	✅	2s	2s	+0%
`mlk_poly_invntt_tomont`	✅	2s	2s	+0%
`mlk_poly_mulcache_compute_c`	✅	2s	1s	+100%
`mlk_poly_mulcache_compute_native`	✅	2s	3s	-33%
`mlk_poly_ntt_c`	✅	2s	3s	-33%
`mlk_poly_reduce`	✅	2s	5s	-60%
`mlk_polyvec_basemul_acc_montgomery_cached`	✅	2s	2s	+0%
`mlk_polyvec_compress_du`	✅	2s	3s	-33%
`mlk_polyvec_frombytes`	✅	2s	3s	-33%
`mlk_polyvec_permute_bitrev_to_custom_native`	✅	2s	4s	-50%
`mlk_rej_uniform`	✅	2s	1s	+100%
`mlk_scalar_compress_d4`	✅	2s	1s	+100%
`mlk_scalar_signed_to_unsigned_q`	✅	2s	3s	-33%
`mlk_shake128x4_absorb_once`	✅	2s	4s	-50%
`mlk_shake128x4_squeezeblocks`	✅	2s	3s	-33%
`mlk_value_barrier_u32`	✅	2s	4s	-50%
`poly_invntt_tomont_native`	✅	2s	2s	+0%
`poly_reduce_native_aarch64`	✅	2s	2s	+0%
`poly_tomont_native_aarch64`	✅	2s	2s	+0%
`polyvec_basemul_acc_montgomery_cached_k3_native_aarch64`	✅	2s	3s	-33%
`rej_uniform_native_aarch64`	✅	2s	3s	-33%
`sys_check_capability`	✅	2s	3s	-33%
`keccak_f1600_x1_native_aarch64_v84a`	✅	1s	2s	-50%
`mlk_ct_get_optblocker_i32`	✅	1s	2s	-50%
`mlk_ct_memcmp`	✅	1s	2s	-50%
`mlk_poly_tomont`	✅	1s	4s	-75%
`mlk_poly_tomont_c`	✅	1s	2s	-50%
`mlk_polyvec_reduce`	✅	1s	2s	-50%
`mlk_polyvec_tobytes`	✅	1s	2s	-50%
`mlk_scalar_decompress_d11`	✅	1s	1s	+0%
`mlk_scalar_decompress_d4`	✅	1s	5s	-80%
`mlk_sha3_256`	✅	1s	2s	-50%
`mlk_sha3_512`	✅	1s	3s	-67%
`mlk_shake128_absorb_once`	✅	1s	2s	-50%
`mlk_shake128_squeezeblocks`	✅	1s	1s	+0%
`mlk_value_barrier_u8`	✅	1s	2s	-50%
`ntt_native_aarch64`	✅	1s	2s	-50%
`polyvec_basemul_acc_montgomery_cached_k4_native_aarch64`	✅	1s	2s	-50%

oqs-bot · 2026-01-21T03:19:42Z

CBMC Results (ML-KEM-1024)

Full Results (139 proofs)

Proof	Status	Current	Previous	Change
`TOTAL`	✅	1826s	1903s	-4.0%
`mlk_indcpa_enc`	✅	379s	407s	-7%
`mlk_indcpa_keypair_derand`	✅	280s	294s	-5%
`mlk_keccak_squeezeblocks_x4`	✅	141s	145s	-3%
`mlk_polyvec_add`	✅	132s	142s	-7%
`polyvec_basemul_acc_montgomery_cached_native`	✅	124s	130s	-5%
`mlk_rej_uniform_c`	✅	92s	94s	-2%
`mlk_polyvec_basemul_acc_montgomery_cached_c`	✅	57s	59s	-3%
`poly_ntt_native`	✅	45s	51s	-12%
`mlk_poly_rej_uniform`	✅	40s	45s	-11%
`mlk_poly_decompress_dv`	✅	37s	36s	+3%
`mlk_ntt_layer`	✅	33s	32s	+3%
`keccakf1600x4_permute_native_x4`	✅	20s	20s	+0%
`mlk_indcpa_dec`	✅	20s	21s	-5%
`mlk_poly_reduce_native`	✅	17s	15s	+13%
`mlk_keccak_absorb_once_x4`	✅	12s	12s	+0%
`mlk_ntt_butterfly_block`	✅	12s	11s	+9%
`mlk_poly_frombytes_native`	✅	11s	9s	+22%
`mlk_poly_sub`	✅	11s	10s	+10%
`mlk_keccak_squeezeblocks`	✅	9s	7s	+29%
`keccakf1600_permute_native`	✅	8s	7s	+14%
`kem_dec`	✅	8s	7s	+14%
`mlk_gen_matrix`	✅	8s	8s	+0%
`mlk_poly_frommsg`	✅	8s	10s	-20%
`mlk_poly_rej_uniform_x4`	✅	8s	8s	+0%
`mlk_keccak_squeeze_once`	✅	7s	6s	+17%
`mlk_shake256x4`	✅	7s	7s	+0%
`mlk_fqmul`	✅	6s	6s	+0%
`mlk_gen_matrix_serial`	✅	6s	6s	+0%
`mlk_invntt_layer`	✅	6s	6s	+0%
`mlk_poly_compress_du`	✅	6s	6s	+0%
`mlk_ct_cmask_nonzero_u8`	✅	5s	2s	+150%
`mlk_keccak_absorb_once`	✅	5s	5s	+0%
`mlk_montgomery_reduce`	✅	5s	1s	+400%
`mlk_polymat_permute_bitrev_to_custom`	✅	5s	5s	+0%
`kem_keypair`	✅	4s	3s	+33%
`mlk_keccakf1600_permute`	✅	4s	4s	+0%
`mlk_poly_getnoise_eta1_4x_native`	✅	4s	4s	+0%
`mlk_poly_getnoise_eta2`	✅	4s	2s	+100%
`mlk_poly_reduce_c`	✅	4s	3s	+33%
`mlk_poly_tobytes`	✅	4s	2s	+100%
`mlk_poly_tobytes_c`	✅	4s	4s	+0%
`mlk_polyvec_permute_bitrev_to_custom_native`	✅	4s	4s	+0%
`mlk_sha3_256`	✅	4s	2s	+100%
`poly_tobytes_native_aarch64`	✅	4s	3s	+33%
`rej_uniform_native`	✅	4s	3s	+33%
`intt_native_aarch64`	✅	3s	3s	+0%
`keccak_f1600_x4_native_aarch64_v84a`	✅	3s	1s	+200%
`kem_check_pk`	✅	3s	6s	-50%
`kem_enc_derand`	✅	3s	5s	-40%
`kem_keypair_derand`	✅	3s	3s	+0%
`mlk_ct_get_optblocker_i32`	✅	3s	1s	+200%
`mlk_keccakf1600_extract_bytes`	✅	3s	3s	+0%
`mlk_keccakf1600_extract_bytes (big endian)`	✅	3s	2s	+50%
`mlk_keccakf1600_xor_bytes (big endian)`	✅	3s	1s	+200%
`mlk_keccakf1600x4_xor_bytes`	✅	3s	2s	+50%
`mlk_matvec_mul`	✅	3s	1s	+200%
`mlk_poly_add`	✅	3s	2s	+50%
`mlk_poly_cbd_eta1`	✅	3s	2s	+50%
`mlk_poly_cbd_eta2`	✅	3s	2s	+50%
`mlk_poly_decompress_du`	✅	3s	3s	+0%
`mlk_poly_frombytes`	✅	3s	1s	+200%
`mlk_poly_frombytes_c`	✅	3s	2s	+50%
`mlk_poly_getnoise_eta1_4x`	✅	3s	4s	-25%
`mlk_poly_invntt_tomont`	✅	3s	5s	-40%
`mlk_poly_mulcache_compute_c`	✅	3s	3s	+0%
`mlk_poly_ntt`	✅	3s	4s	-25%
`mlk_poly_ntt_c`	✅	3s	2s	+50%
`mlk_poly_tomont`	✅	3s	2s	+50%
`mlk_poly_tomsg`	✅	3s	4s	-25%
`mlk_polyvec_invntt_tomont`	✅	3s	5s	-40%
`mlk_polyvec_reduce`	✅	3s	3s	+0%
`mlk_scalar_compress_d10`	✅	3s	4s	-25%
`mlk_scalar_decompress_d10`	✅	3s	3s	+0%
`mlk_scalar_decompress_d11`	✅	3s	4s	-25%
`mlk_scalar_decompress_d4`	✅	3s	2s	+50%
`mlk_scalar_decompress_d5`	✅	3s	2s	+50%
`mlk_scalar_signed_to_unsigned_q`	✅	3s	3s	+0%
`mlk_sha3_512`	✅	3s	3s	+0%
`mlk_shake128_absorb_once`	✅	3s	2s	+50%
`mlk_shake128x4_squeezeblocks`	✅	3s	2s	+50%
`mlk_value_barrier_i32`	✅	3s	2s	+50%
`mlk_value_barrier_u32`	✅	3s	3s	+0%
`poly_getnoise_eta1122_4x_native`	✅	3s	1s	+200%
`poly_reduce_native_aarch64`	✅	3s	3s	+0%
`polyvec_basemul_acc_montgomery_cached_k3_native_aarch64`	✅	3s	3s	+0%
`keccak_f1600_x1_native_aarch64_v84a`	✅	2s	3s	-33%
`keccak_f1600_x4_native_aarch64_v8a_v84a_scalar_hybrid`	✅	2s	2s	+0%
`kem_check_sk`	✅	2s	2s	+0%
`mlk_ct_cmov_zero`	✅	2s	3s	-33%
`mlk_ct_get_optblocker_u32`	✅	2s	1s	+100%
`mlk_ct_get_optblocker_u8`	✅	2s	2s	+0%
`mlk_keccakf1600_xor_bytes`	✅	2s	2s	+0%
`mlk_keccakf1600x4_permute`	✅	2s	2s	+0%
`mlk_poly_compress_dv`	✅	2s	2s	+0%
`mlk_poly_getnoise_eta1122_4x`	✅	2s	3s	-33%
`mlk_poly_invntt_tomont_c`	✅	2s	3s	-33%
`mlk_poly_mulcache_compute`	✅	2s	2s	+0%
`mlk_poly_mulcache_compute_native`	✅	2s	3s	-33%
`mlk_poly_reduce`	✅	2s	1s	+100%
`mlk_poly_tobytes_native`	✅	2s	1s	+100%
`mlk_poly_tomont_c`	✅	2s	3s	-33%
`mlk_poly_tomont_native`	✅	2s	2s	+0%
`mlk_polyvec_basemul_acc_montgomery_cached`	✅	2s	1s	+100%
`mlk_polyvec_compress_du`	✅	2s	1s	+100%
`mlk_polyvec_decompress_du`	✅	2s	3s	-33%
`mlk_polyvec_frombytes`	✅	2s	2s	+0%
`mlk_polyvec_mulcache_compute`	✅	2s	2s	+0%
`mlk_polyvec_ntt`	✅	2s	5s	-60%
`mlk_polyvec_permute_bitrev_to_custom`	✅	2s	2s	+0%
`mlk_polyvec_tobytes`	✅	2s	2s	+0%
`mlk_rej_uniform`	✅	2s	2s	+0%
`mlk_scalar_compress_d1`	✅	2s	3s	-33%
`mlk_scalar_compress_d11`	✅	2s	3s	-33%
`mlk_scalar_compress_d4`	✅	2s	2s	+0%
`mlk_shake128_squeezeblocks`	✅	2s	2s	+0%
`mlk_shake256`	✅	2s	2s	+0%
`mlk_value_barrier_u8`	✅	2s	3s	-33%
`ntt_native_aarch64`	✅	2s	3s	-33%
`poly_invntt_tomont_native`	✅	2s	3s	-33%
`poly_mulcache_compute_native_aarch64`	✅	2s	3s	-33%
`poly_tomont_native_aarch64`	✅	2s	1s	+100%
`polyvec_basemul_acc_montgomery_cached_k2_native_aarch64`	✅	2s	2s	+0%
`sys_check_capability`	✅	2s	2s	+0%
`keccak_f1600_x1_native_aarch64`	✅	1s	3s	-67%
`keccak_f1600_x4_native_aarch64_v8a_scalar_hybrid`	✅	1s	2s	-50%
`kem_enc`	✅	1s	4s	-75%
`mlk_barrett_reduce`	✅	1s	2s	-50%
`mlk_check_pct`	✅	1s	3s	-67%
`mlk_ct_cmask_neg_i16`	✅	1s	3s	-67%
`mlk_ct_cmask_nonzero_u16`	✅	1s	1s	+0%
`mlk_ct_memcmp`	✅	1s	4s	-75%
`mlk_ct_sel_int16`	✅	1s	4s	-75%
`mlk_ct_sel_uint8`	✅	1s	3s	-67%
`mlk_keccakf1600x4_extract_bytes`	✅	1s	2s	-50%
`mlk_polyvec_tomont`	✅	1s	2s	-50%
`mlk_scalar_compress_d5`	✅	1s	2s	-50%
`mlk_shake128x4_absorb_once`	✅	1s	3s	-67%
`polyvec_basemul_acc_montgomery_cached_k4_native_aarch64`	✅	1s	2s	-50%
`rej_uniform_native_aarch64`	✅	1s	4s	-75%

bremoran

LGTM

hanno-becker

I cannot test that the PMU code is correct on HW, but the CI and a local check confirm that it at least builds and runs in QEMU, which is good enough for now.

Make the benchmark parameters (NWARMUP, NITERATIONS, NTESTS) configurable via CFLAGS by wrapping them in #ifndef guards and renaming to MLK_BENCHMARK_NWARMUP, MLK_BENCHMARK_NITERATIONS, and MLK_BENCHMARK_NTESTS. Signed-off-by: Matthias J. Kannwischer <matthias@kannwischer.eu>

Add PMU-based cycle counting support for Armv8.1-M Cortex-M processors. This uses the CMSIS PMU APIs for portable cycle counter access. - Resolves #1502 Co-Authored-By: Brendan Moran <brendan.moran@arm.com> Signed-off-by: Matthias J. Kannwischer <matthias@kannwischer.eu>

The cycle counts will be zero, but it still tests the PMU code builds. Signed-off-by: Matthias J. Kannwischer <matthias@kannwischer.eu>

mkannwischer force-pushed the armv8-bench branch 2 times, most recently from e46f5da to 8ab59cf Compare January 21, 2026 02:54

mkannwischer marked this pull request as ready for review January 21, 2026 03:12

mkannwischer requested a review from a team as a code owner January 21, 2026 03:12

mkannwischer requested a review from bremoran January 21, 2026 03:12

mkannwischer added the needs-mldsa-native-port label Jan 21, 2026

mkannwischer mentioned this pull request Jan 21, 2026

Armv8.1-M: Add MVE Keccak-f1600 x4 implementation #1518

Merged

bremoran approved these changes Jan 21, 2026

View reviewed changes

hanno-becker reviewed Jan 21, 2026

View reviewed changes

Comment thread .github/workflows/baremetal.yml

hanno-becker approved these changes Jan 21, 2026

View reviewed changes

mkannwischer and others added 3 commits January 21, 2026 15:07

CI: Test benchmarking binary for M55-AN547 using qemu

690c190

The cycle counts will be zero, but it still tests the PMU code builds. Signed-off-by: Matthias J. Kannwischer <matthias@kannwischer.eu>

hanno-becker force-pushed the armv8-bench branch from 8ab59cf to 690c190 Compare January 21, 2026 15:07

hanno-becker merged commit 51d942a into main Jan 21, 2026
780 of 781 checks passed

hanno-becker deleted the armv8-bench branch January 21, 2026 19:27

mkannwischer mentioned this pull request Jan 21, 2026

Port: Add PMU cycle counting for Armv8.1-M pq-code-package/mldsa-native#903

Closed

willieyz mentioned this pull request Jan 22, 2026

Port: Add PMU cycle counting for Armv8.1-M pq-code-package/mldsa-native#909

Merged

mkannwischer removed the needs-mldsa-native-port label Feb 10, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add PMU cycle counting for Armv8.1-M #1514

Add PMU cycle counting for Armv8.1-M #1514
hanno-becker merged 3 commits intomainfrom
armv8-bench

mkannwischer commented Jan 21, 2026

Uh oh!

oqs-bot commented Jan 21, 2026 •

edited

Loading

Uh oh!

oqs-bot commented Jan 21, 2026 •

edited

Loading

Uh oh!

oqs-bot commented Jan 21, 2026 •

edited

Loading

Uh oh!

bremoran left a comment

Uh oh!

Uh oh!

hanno-becker left a comment •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

mkannwischer commented Jan 21, 2026

Uh oh!

oqs-bot commented Jan 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

CBMC Results (ML-KEM-512)

Uh oh!

oqs-bot commented Jan 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

CBMC Results (ML-KEM-768)

Uh oh!

oqs-bot commented Jan 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

CBMC Results (ML-KEM-1024)

Uh oh!

bremoran left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

hanno-becker left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

oqs-bot commented Jan 21, 2026 •

edited

Loading

oqs-bot commented Jan 21, 2026 •

edited

Loading

oqs-bot commented Jan 21, 2026 •

edited

Loading

hanno-becker left a comment •

edited

Loading