HOL-Light: Add HOL Light poly_use_hint proofs for AArch64 by jakemas · Pull Request #1037 · pq-code-package/mldsa-native

jakemas · 2026-04-13T16:54:04Z

Resolves:

HOL-Light: Prove AArch64 poly_use_hint #486

Summary

Proof times are decent ~15min for _32 and ~22min for _88.

Port ML-DSA poly_use_hint HOL Light proofs from s2n-bignum for AArch64
- mldsa_poly_use_hint_32: GAMMA2=(Q-1)/32, for l=5,7 (s2n-bignum PR: ML-DSA Aarch64 HOL-Light proof poly_use_hint_32 awslabs/s2n-bignum#372)
- mldsa_poly_use_hint_88: GAMMA2=(Q-1)/88, for l=4 (s2n-bignum PR: ML-DSA Aarch64 HOL-Light proof poly_use_hint_88 awslabs/s2n-bignum#375)
Includes correctness proofs, subroutine forms, and constant-time/memory safety proofs for both
Adds CBMC contracts for the assembly functions

oqs-bot · 2026-04-13T18:02:53Z

CBMC Results (ML-DSA-87)

Full Results (200 proofs)

Proof	Status	Current	Previous	Change
`TOTAL`	✅	2298s	2497s	-8.0%
`polyvecl_pointwise_acc_montgomery_c`	✅	320s	411s	-22%
`sign_verify_internal`	✅	248s	272s	-9%
`polyvec_matrix_expand`	✅	163s	175s	-7%
`rej_uniform_native`	✅	115s	131s	-12%
`mld_invntt_layer`	✅	90s	95s	-5%
`poly_pointwise_montgomery_c`	✅	89s	102s	-13%
`mld_ct_memcmp`	✅	74s	81s	-9%
`mld_attempt_signature_generation`	✅	59s	60s	-2%
`polyvec_matrix_expand_serial`	✅	58s	59s	-2%
`sign_keypair_internal`	✅	50s	54s	-7%
`mld_ntt_layer`	✅	42s	45s	-7%
`sign_signature_internal`	✅	33s	34s	-3%
`polyvec_matrix_pointwise_montgomery`	✅	32s	28s	+14%
`sign_pk_from_sk`	✅	32s	33s	-3%
`polyveck_power2round`	✅	31s	36s	-14%
`fqmul`	✅	26s	31s	-16%
`keccakf1600x4_permute_native`	✅	22s	24s	-8%
`polyvec_matrix_pointwise_montgomery_yvec`	✅	18s	21s	-14%
`rej_uniform_c`	✅	18s	21s	-14%
`mld_ntt_butterfly_block`	✅	16s	17s	-6%
`poly_uniform_4x`	✅	16s	14s	+14%
`rej_uniform`	✅	16s	18s	-11%
`poly_chknorm_c`	✅	15s	16s	-6%
`poly_uniform_eta_4x`	✅	15s	11s	+36%
`mld_check_pct`	✅	13s	14s	-7%
`poly_add`	✅	13s	13s	+0%
`polyt0_unpack`	✅	13s	15s	-13%
`polyveck_add`	✅	13s	13s	+0%
`polyveck_use_hint`	✅	13s	14s	-7%
`keccak_absorb_once_x4`	✅	11s	10s	+10%
`polyveck_decompose`	✅	11s	11s	+0%
`polyveck_ntt`	✅	11s	12s	-8%
`poly_invntt_tomont_c`	✅	10s	9s	+11%
`poly_power2round`	✅	10s	8s	+25%
`polyveck_pointwise_poly_montgomery`	✅	9s	10s	-10%
`keccak_absorb`	✅	8s	6s	+33%
`keccakf1600_permute`	✅	8s	9s	-11%
`mld_sample_s1_s2`	✅	8s	7s	+14%
`pointwise_acc_native_aarch64`	✅	8s	7s	+14%
`keccakf1600_permute_native`	✅	7s	9s	-22%
`mld_compute_pack_z`	✅	7s	8s	-12%
`pointwise_acc_native_x86_64`	✅	7s	9s	-22%
`poly_decompose_c`	✅	7s	8s	-12%
`polyeta_unpack`	✅	7s	5s	+40%
`polyveck_chknorm`	✅	7s	4s	+75%
`polyveck_reduce`	✅	7s	5s	+40%
`polyveck_shiftl`	✅	7s	7s	+0%
`polyvecl_ntt`	✅	7s	6s	+17%
`sign`	✅	7s	7s	+0%
`sign_verify_pre_hash_shake256`	✅	7s	5s	+40%
`poly_shiftl`	✅	6s	3s	+100%
`polyt0_pack`	✅	6s	3s	+100%
`polyveck_sub`	✅	6s	9s	-33%
`intt_native_x86_64`	✅	5s	5s	+0%
`mld_prepare_domain_separation_prefix`	✅	5s	7s	-29%
`mld_sample_s1_s2_serial`	✅	5s	6s	-17%
`montgomery_reduce`	✅	5s	5s	+0%
`nttunpack_native_x86_64`	✅	5s	4s	+25%
`poly_chknorm_native`	✅	5s	2s	+150%
`poly_decompose`	✅	5s	3s	+67%
`poly_ntt`	✅	5s	4s	+25%
`poly_ntt_c`	✅	5s	1s	+400%
`poly_uniform_eta`	✅	5s	3s	+67%
`polyveck_caddq`	✅	5s	5s	+0%
`polyveck_invntt_tomont`	✅	5s	5s	+0%
`polyz_unpack`	✅	5s	4s	+25%
`polyz_unpack_17_native_aarch64`	✅	5s	3s	+67%
`sign_open`	✅	5s	4s	+25%
`sign_signature`	✅	5s	3s	+67%
`sign_signature_pre_hash_internal`	✅	5s	6s	-17%
`sign_signature_pre_hash_shake256`	✅	5s	4s	+25%
`sign_verify_pre_hash_internal`	✅	5s	6s	-17%
`unpack_hints`	✅	5s	4s	+25%
`decompose`	✅	4s	3s	+33%
`keccak_squeezeblocks_x4`	✅	4s	5s	-20%
`make_hint`	✅	4s	3s	+33%
`ntt_native_x86_64`	✅	4s	2s	+100%
`pack_sk_rho_key_tr_s2_t0`	✅	4s	4s	+0%
`poly_make_hint`	✅	4s	2s	+100%
`poly_pointwise_montgomery`	✅	4s	2s	+100%
`poly_sub`	✅	4s	3s	+33%
`poly_uniform_gamma1`	✅	4s	3s	+33%
`poly_use_hint_c`	✅	4s	2s	+100%
`polyvec_matrix_pointwise_montgomery_row`	✅	4s	3s	+33%
`polyveck_pack_eta`	✅	4s	1s	+300%
`polyveck_unpack_t0`	✅	4s	2s	+100%
`polyvecl_chknorm`	✅	4s	4s	+0%
`polyvecl_pointwise_acc_montgomery_native`	✅	4s	3s	+33%
`polyvecl_uniform_gamma1`	✅	4s	5s	-20%
`polyz_pack`	✅	4s	4s	+0%
`polyz_unpack_19_native_aarch64`	✅	4s	1s	+300%
`polyz_unpack_c`	✅	4s	5s	-20%
`rej_eta_c`	✅	4s	4s	+0%
`rej_eta_native`	✅	4s	3s	+33%
`shake256_init`	✅	4s	5s	-20%
`shake256x4_squeezeblocks`	✅	4s	1s	+300%
`sign_verify`	✅	4s	7s	-43%
`sign_verify_extmu`	✅	4s	3s	+33%
`sk_s1hat_get_poly`	✅	4s	3s	+33%
`sk_t0hat_get_poly`	✅	4s	3s	+33%
`unpack_sk_s2hat`	✅	4s	3s	+33%
`unpack_sk_t0hat`	✅	4s	3s	+33%
`caddq`	✅	3s	5s	-40%
`keccak_f1600_x4_native_aarch64_v84a`	✅	3s	2s	+50%
`keccak_f1600_x4_native_aarch64_v8a_scalar_hybrid`	✅	3s	1s	+200%
`keccak_f1600_x4_native_aarch64_v8a_v84a_scalar_hybrid`	✅	3s	2s	+50%
`keccak_init`	✅	3s	1s	+200%
`keccak_squeeze`	✅	3s	3s	+0%
`keccakf1600x4_extract_bytes`	✅	3s	1s	+200%
`keccakf1600x4_xor_bytes`	✅	3s	4s	-25%
`mld_ct_abs_i32`	✅	3s	2s	+50%
`mld_ct_cmask_nonzero_u32`	✅	3s	3s	+0%
`mld_h`	✅	3s	4s	-25%
`mld_polymat_expand_entry`	✅	3s	3s	+0%
`pack_pk`	✅	3s	3s	+0%
`pack_sk_s1`	✅	3s	2s	+50%
`pointwise_native_aarch64`	✅	3s	3s	+0%
`poly_caddq_c`	✅	3s	5s	-40%
`poly_caddq_native`	✅	3s	4s	-25%
`poly_caddq_native_aarch64`	✅	3s	3s	+0%
`poly_challenge`	✅	3s	5s	-40%
`poly_chknorm`	✅	3s	2s	+50%
`poly_chknorm_native_aarch64`	✅	3s	4s	-25%
`poly_decompose_native`	✅	3s	3s	+0%
`poly_invntt_tomont`	✅	3s	2s	+50%
`poly_invntt_tomont_native`	✅	3s	2s	+50%
`poly_ntt_native`	✅	3s	4s	-25%
`poly_permute_bitrev_to_custom_optional`	✅	3s	3s	+0%
`poly_reduce`	✅	3s	2s	+50%
`poly_uniform`	✅	3s	3s	+0%
`poly_uniform_gamma1_4x`	✅	3s	5s	-40%
`polyeta_pack`	✅	3s	2s	+50%
`polyveck_pack_t0`	✅	3s	5s	-40%
`polyveck_pack_w1`	✅	3s	2s	+50%
`polyveck_unpack_eta`	✅	3s	3s	+0%
`polyvecl_pointwise_acc_montgomery`	✅	3s	4s	-25%
`polyvecl_uniform_gamma1_serial`	✅	3s	4s	-25%
`polyvecl_unpack_eta`	✅	3s	4s	-25%
`polyz_unpack_native`	✅	3s	4s	-25%
`shake128_release`	✅	3s	3s	+0%
`shake256_absorb`	✅	3s	3s	+0%
`shake256_release`	✅	3s	4s	-25%
`shake256_squeeze`	✅	3s	2s	+50%
`sign_keypair`	✅	3s	4s	-25%
`unpack_pk`	✅	3s	4s	-25%
`unpack_sk`	✅	3s	4s	-25%
`unpack_sk_s1hat`	✅	3s	4s	-25%
`use_hint`	✅	3s	2s	+50%
`yvec_init`	✅	3s	1s	+200%
`fqscale`	✅	2s	4s	-50%
`keccak_f1600_x1_native_aarch64`	✅	2s	1s	+100%
`keccak_f1600_x1_native_aarch64_v84a`	✅	2s	2s	+0%
`keccakf1600_extract_bytes (big endian)`	✅	2s	1s	+100%
`keccakf1600_xor_bytes`	✅	2s	5s	-60%
`keccakf1600_xor_bytes (big endian)`	✅	2s	3s	-33%
`mld_ct_cmask_neg_i32`	✅	2s	3s	-33%
`mld_ct_cmask_nonzero_u8`	✅	2s	1s	+100%
`mld_ct_get_optblocker_i64`	✅	2s	3s	-33%
`mld_ct_get_optblocker_u32`	✅	2s	1s	+100%
`mld_ct_get_optblocker_u8`	✅	2s	2s	+0%
`mld_ct_sel_int32`	✅	2s	2s	+0%
`mld_keccakf1600_extract_bytes`	✅	2s	2s	+0%
`ntt_native_aarch64`	✅	2s	4s	-50%
`pack_sig_c`	✅	2s	5s	-60%
`pointwise_native_x86_64`	✅	2s	3s	-33%
`poly_permute_bitrev_to_custom_optional_native`	✅	2s	3s	-33%
`poly_pointwise_montgomery_native`	✅	2s	3s	-33%
`poly_use_hint`	✅	2s	5s	-60%
`poly_use_hint_native`	✅	2s	5s	-60%
`poly_use_hint_native_aarch64`	✅	2s	-	new
`polyt1_unpack`	✅	2s	3s	-33%
`polyvecl_pack_eta`	✅	2s	2s	+0%
`polyvecl_unpack_z`	✅	2s	3s	-33%
`polyw1_pack`	✅	2s	2s	+0%
`power2round`	✅	2s	2s	+0%
`rej_eta`	✅	2s	2s	+0%
`shake128_finalize`	✅	2s	5s	-60%
`shake128_init`	✅	2s	2s	+0%
`shake128_squeeze`	✅	2s	3s	-33%
`shake128x4_absorb_once`	✅	2s	2s	+0%
`shake128x4_squeezeblocks`	✅	2s	2s	+0%
`shake256`	✅	2s	2s	+0%
`sign_signature_extmu`	✅	2s	5s	-60%
`sk_s2hat_get_poly`	✅	2s	5s	-60%
`sys_check_capability`	✅	2s	2s	+0%
`unpack_sig`	✅	2s	5s	-60%
`yvec_get_poly`	✅	2s	3s	-33%
`keccak_finalize`	✅	1s	2s	-50%
`keccakf1600x4_permute`	✅	1s	1s	+0%
`mld_value_barrier_i64`	✅	1s	4s	-75%
`mld_value_barrier_u32`	✅	1s	2s	-50%
`mld_value_barrier_u8`	✅	1s	4s	-75%
`pack_sig_h_poly`	✅	1s	3s	-67%
`pack_sig_z`	✅	1s	2s	-50%
`poly_caddq`	✅	1s	3s	-67%
`polyt1_pack`	✅	1s	2s	-50%
`reduce32`	✅	1s	2s	-50%
`shake128_absorb`	✅	1s	4s	-75%
`shake256_finalize`	✅	1s	2s	-50%
`shake256x4_absorb_once`	✅	1s	2s	-50%

oqs-bot · 2026-04-13T18:04:08Z

CBMC Results (ML-DSA-65)

Full Results (200 proofs)

Proof	Status	Current	Previous	Change
`TOTAL`	✅	2284s	2382s	-4.1%
`polyvecl_pointwise_acc_montgomery_c`	✅	509s	553s	-8%
`sign_verify_internal`	✅	178s	188s	-5%
`polyvec_matrix_expand`	✅	119s	128s	-7%
`rej_uniform_native`	✅	117s	130s	-10%
`mld_invntt_layer`	✅	93s	96s	-3%
`poly_pointwise_montgomery_c`	✅	92s	99s	-7%
`mld_ct_memcmp`	✅	78s	79s	-1%
`mld_attempt_signature_generation`	✅	48s	49s	-2%
`mld_ntt_layer`	✅	42s	47s	-11%
`sign_signature_internal`	✅	30s	30s	+0%
`sign_keypair_internal`	✅	29s	33s	-12%
`fqmul`	✅	28s	29s	-3%
`polyvec_matrix_expand_serial`	✅	26s	25s	+4%
`keccakf1600x4_permute_native`	✅	23s	23s	+0%
`polyvec_matrix_pointwise_montgomery`	✅	22s	23s	-4%
`sign_pk_from_sk`	✅	20s	20s	+0%
`polyvec_matrix_pointwise_montgomery_yvec`	✅	19s	20s	-5%
`mld_ntt_butterfly_block`	✅	18s	17s	+6%
`polyveck_decompose`	✅	17s	15s	+13%
`rej_uniform`	✅	17s	16s	+6%
`rej_uniform_c`	✅	17s	20s	-15%
`poly_chknorm_c`	✅	16s	18s	-11%
`polyt0_unpack`	✅	16s	14s	+14%
`polyveck_power2round`	✅	15s	17s	-12%
`poly_uniform_eta_4x`	✅	12s	12s	+0%
`keccak_absorb_once_x4`	✅	11s	10s	+10%
`poly_add`	✅	11s	12s	-8%
`poly_uniform_4x`	✅	11s	13s	-15%
`polyveck_add`	✅	10s	10s	+0%
`polyveck_pointwise_poly_montgomery`	✅	10s	11s	-9%
`keccakf1600_permute_native`	✅	9s	9s	+0%
`poly_invntt_tomont_c`	✅	9s	8s	+12%
`poly_power2round`	✅	9s	9s	+0%
`mld_check_pct`	✅	8s	8s	+0%
`polyveck_caddq`	✅	8s	8s	+0%
`polyveck_chknorm`	✅	8s	3s	+167%
`polyveck_reduce`	✅	8s	7s	+14%
`polyveck_shiftl`	✅	8s	9s	-11%
`polyvecl_ntt`	✅	8s	7s	+14%
`sign`	✅	8s	10s	-20%
`keccak_absorb`	✅	7s	8s	-12%
`mld_compute_pack_z`	✅	7s	6s	+17%
`mld_sample_s1_s2`	✅	7s	3s	+133%
`polyeta_pack`	✅	7s	3s	+133%
`polyveck_sub`	✅	7s	6s	+17%
`polyveck_use_hint`	✅	7s	7s	+0%
`polyvecl_chknorm`	✅	7s	5s	+40%
`polyz_unpack_c`	✅	7s	10s	-30%
`fqscale`	✅	6s	2s	+200%
`keccakf1600_permute`	✅	6s	7s	-14%
`pointwise_acc_native_x86_64`	✅	6s	7s	-14%
`polyveck_invntt_tomont`	✅	6s	5s	+20%
`polyz_unpack`	✅	6s	4s	+50%
`mld_ct_cmask_nonzero_u32`	✅	5s	3s	+67%
`mld_h`	✅	5s	3s	+67%
`mld_polymat_expand_entry`	✅	5s	4s	+25%
`mld_prepare_domain_separation_prefix`	✅	5s	3s	+67%
`pointwise_acc_native_aarch64`	✅	5s	7s	-29%
`poly_caddq`	✅	5s	3s	+67%
`poly_caddq_c`	✅	5s	6s	-17%
`poly_sub`	✅	5s	2s	+150%
`poly_uniform_gamma1`	✅	5s	5s	+0%
`polyeta_unpack`	✅	5s	2s	+150%
`polyveck_ntt`	✅	5s	7s	-29%
`polyveck_unpack_eta`	✅	5s	5s	+0%
`polyvecl_uniform_gamma1`	✅	5s	2s	+150%
`shake256_init`	✅	5s	2s	+150%
`sign_keypair`	✅	5s	4s	+25%
`sign_signature_pre_hash_shake256`	✅	5s	3s	+67%
`sign_verify_pre_hash_shake256`	✅	5s	4s	+25%
`sk_s2hat_get_poly`	✅	5s	2s	+150%
`unpack_hints`	✅	5s	5s	+0%
`yvec_get_poly`	✅	5s	2s	+150%
`keccak_squeezeblocks_x4`	✅	4s	5s	-20%
`keccakf1600x4_xor_bytes`	✅	4s	2s	+100%
`mld_ct_cmask_nonzero_u8`	✅	4s	2s	+100%
`mld_ct_sel_int32`	✅	4s	2s	+100%
`mld_sample_s1_s2_serial`	✅	4s	4s	+0%
`montgomery_reduce`	✅	4s	2s	+100%
`ntt_native_aarch64`	✅	4s	3s	+33%
`nttunpack_native_x86_64`	✅	4s	5s	-20%
`pack_sig_c`	✅	4s	5s	-20%
`pack_sk_s1`	✅	4s	3s	+33%
`pointwise_native_x86_64`	✅	4s	4s	+0%
`poly_caddq_native`	✅	4s	4s	+0%
`poly_challenge`	✅	4s	5s	-20%
`poly_decompose_native`	✅	4s	3s	+33%
`poly_invntt_tomont`	✅	4s	2s	+100%
`poly_permute_bitrev_to_custom_optional`	✅	4s	3s	+33%
`poly_pointwise_montgomery`	✅	4s	3s	+33%
`poly_pointwise_montgomery_native`	✅	4s	2s	+100%
`poly_uniform_gamma1_4x`	✅	4s	5s	-20%
`poly_use_hint_native_aarch64`	✅	4s	-	new
`polyveck_pack_eta`	✅	4s	4s	+0%
`polyveck_unpack_t0`	✅	4s	5s	-20%
`polyvecl_unpack_z`	✅	4s	4s	+0%
`rej_eta_native`	✅	4s	6s	-33%
`shake128x4_absorb_once`	✅	4s	4s	+0%
`shake256_squeeze`	✅	4s	2s	+100%
`sign_open`	✅	4s	4s	+0%
`sign_signature`	✅	4s	6s	-33%
`sign_signature_extmu`	✅	4s	6s	-33%
`sign_signature_pre_hash_internal`	✅	4s	5s	-20%
`sign_verify_pre_hash_internal`	✅	4s	4s	+0%
`sk_s1hat_get_poly`	✅	4s	4s	+0%
`unpack_sk`	✅	4s	4s	+0%
`caddq`	✅	3s	3s	+0%
`keccak_f1600_x1_native_aarch64_v84a`	✅	3s	4s	-25%
`keccak_init`	✅	3s	2s	+50%
`keccak_squeeze`	✅	3s	3s	+0%
`keccakf1600_xor_bytes`	✅	3s	3s	+0%
`keccakf1600_xor_bytes (big endian)`	✅	3s	2s	+50%
`keccakf1600x4_permute`	✅	3s	3s	+0%
`mld_ct_abs_i32`	✅	3s	3s	+0%
`mld_ct_get_optblocker_i64`	✅	3s	3s	+0%
`pack_sig_h_poly`	✅	3s	3s	+0%
`pack_sig_z`	✅	3s	4s	-25%
`pack_sk_rho_key_tr_s2_t0`	✅	3s	2s	+50%
`pointwise_native_aarch64`	✅	3s	3s	+0%
`poly_caddq_native_aarch64`	✅	3s	3s	+0%
`poly_chknorm_native_aarch64`	✅	3s	3s	+0%
`poly_decompose_c`	✅	3s	4s	-25%
`poly_ntt`	✅	3s	2s	+50%
`poly_ntt_c`	✅	3s	3s	+0%
`poly_permute_bitrev_to_custom_optional_native`	✅	3s	4s	-25%
`poly_shiftl`	✅	3s	3s	+0%
`poly_uniform`	✅	3s	7s	-57%
`poly_uniform_eta`	✅	3s	3s	+0%
`poly_use_hint`	✅	3s	2s	+50%
`polyt1_unpack`	✅	3s	4s	-25%
`polyvec_matrix_pointwise_montgomery_row`	✅	3s	2s	+50%
`polyveck_pack_t0`	✅	3s	5s	-40%
`polyvecl_pointwise_acc_montgomery`	✅	3s	3s	+0%
`polyvecl_pointwise_acc_montgomery_native`	✅	3s	5s	-40%
`polyvecl_uniform_gamma1_serial`	✅	3s	3s	+0%
`polyvecl_unpack_eta`	✅	3s	5s	-40%
`polyw1_pack`	✅	3s	1s	+200%
`polyz_pack`	✅	3s	4s	-25%
`polyz_unpack_17_native_aarch64`	✅	3s	2s	+50%
`polyz_unpack_19_native_aarch64`	✅	3s	2s	+50%
`power2round`	✅	3s	2s	+50%
`reduce32`	✅	3s	4s	-25%
`rej_eta`	✅	3s	3s	+0%
`rej_eta_c`	✅	3s	3s	+0%
`shake128_finalize`	✅	3s	3s	+0%
`shake128_release`	✅	3s	1s	+200%
`shake128x4_squeezeblocks`	✅	3s	2s	+50%
`shake256`	✅	3s	3s	+0%
`shake256_finalize`	✅	3s	3s	+0%
`shake256x4_squeezeblocks`	✅	3s	2s	+50%
`sign_verify`	✅	3s	5s	-40%
`unpack_sig`	✅	3s	2s	+50%
`unpack_sk_s2hat`	✅	3s	2s	+50%
`unpack_sk_t0hat`	✅	3s	4s	-25%
`use_hint`	✅	3s	4s	-25%
`yvec_init`	✅	3s	2s	+50%
`intt_native_x86_64`	✅	2s	3s	-33%
`keccak_f1600_x1_native_aarch64`	✅	2s	3s	-33%
`keccak_f1600_x4_native_aarch64_v84a`	✅	2s	4s	-50%
`keccak_f1600_x4_native_aarch64_v8a_scalar_hybrid`	✅	2s	3s	-33%
`keccak_f1600_x4_native_aarch64_v8a_v84a_scalar_hybrid`	✅	2s	1s	+100%
`keccak_finalize`	✅	2s	1s	+100%
`keccakf1600x4_extract_bytes`	✅	2s	2s	+0%
`make_hint`	✅	2s	3s	-33%
`mld_ct_cmask_neg_i32`	✅	2s	3s	-33%
`mld_ct_get_optblocker_u32`	✅	2s	2s	+0%
`mld_ct_get_optblocker_u8`	✅	2s	2s	+0%
`mld_keccakf1600_extract_bytes`	✅	2s	3s	-33%
`mld_value_barrier_i64`	✅	2s	2s	+0%
`pack_pk`	✅	2s	2s	+0%
`poly_chknorm`	✅	2s	4s	-50%
`poly_chknorm_native`	✅	2s	3s	-33%
`poly_invntt_tomont_native`	✅	2s	2s	+0%
`poly_make_hint`	✅	2s	4s	-50%
`poly_ntt_native`	✅	2s	3s	-33%
`poly_reduce`	✅	2s	1s	+100%
`poly_use_hint_c`	✅	2s	1s	+100%
`polyt1_pack`	✅	2s	4s	-50%
`polyvecl_pack_eta`	✅	2s	4s	-50%
`polyz_unpack_native`	✅	2s	3s	-33%
`shake128_absorb`	✅	2s	3s	-33%
`shake128_init`	✅	2s	4s	-50%
`shake128_squeeze`	✅	2s	6s	-67%
`shake256_absorb`	✅	2s	2s	+0%
`shake256_release`	✅	2s	3s	-33%
`shake256x4_absorb_once`	✅	2s	2s	+0%
`sign_verify_extmu`	✅	2s	3s	-33%
`sk_t0hat_get_poly`	✅	2s	3s	-33%
`sys_check_capability`	✅	2s	2s	+0%
`unpack_pk`	✅	2s	4s	-50%
`unpack_sk_s1hat`	✅	2s	4s	-50%
`decompose`	✅	1s	3s	-67%
`keccakf1600_extract_bytes (big endian)`	✅	1s	3s	-67%
`mld_value_barrier_u32`	✅	1s	2s	-50%
`mld_value_barrier_u8`	✅	1s	2s	-50%
`ntt_native_x86_64`	✅	1s	5s	-80%
`poly_decompose`	✅	1s	2s	-50%
`poly_use_hint_native`	✅	1s	4s	-75%
`polyt0_pack`	✅	1s	4s	-75%
`polyveck_pack_w1`	✅	1s	1s	+0%

oqs-bot · 2026-04-13T18:05:44Z

CBMC Results (ML-DSA-44)

Full Results (200 proofs)

Proof	Status	Current	Previous	Change
`TOTAL`	✅	1663s	1771s	-6.1%
`polyvecl_pointwise_acc_montgomery_c`	✅	125s	134s	-7%
`rej_uniform_native`	✅	125s	134s	-7%
`sign_verify_internal`	✅	110s	117s	-6%
`poly_pointwise_montgomery_c`	✅	98s	113s	-13%
`mld_invntt_layer`	✅	90s	100s	-10%
`mld_ct_memcmp`	✅	79s	87s	-9%
`mld_attempt_signature_generation`	✅	51s	49s	+4%
`mld_ntt_layer`	✅	46s	47s	-2%
`fqmul`	✅	29s	31s	-6%
`polyvec_matrix_expand`	✅	26s	25s	+4%
`polyvec_matrix_pointwise_montgomery`	✅	25s	25s	+0%
`sign_keypair_internal`	✅	24s	25s	-4%
`keccakf1600x4_permute_native`	✅	20s	22s	-9%
`rej_uniform_c`	✅	19s	19s	+0%
`poly_chknorm_c`	✅	18s	17s	+6%
`sign_pk_from_sk`	✅	18s	20s	-10%
`mld_ntt_butterfly_block`	✅	17s	16s	+6%
`polyt0_unpack`	✅	16s	16s	+0%
`rej_uniform`	✅	16s	17s	-6%
`polyveck_chknorm`	✅	15s	17s	-12%
`sign_signature_internal`	✅	15s	18s	-17%
`poly_uniform_4x`	✅	12s	12s	+0%
`poly_uniform_eta_4x`	✅	12s	13s	-8%
`polyvec_matrix_pointwise_montgomery_yvec`	✅	12s	13s	-8%
`poly_add`	✅	11s	11s	+0%
`mld_compute_pack_z`	✅	10s	8s	+25%
`poly_invntt_tomont_c`	✅	10s	10s	+0%
`polyz_unpack_c`	✅	10s	13s	-23%
`keccak_absorb_once_x4`	✅	9s	8s	+12%
`poly_power2round`	✅	9s	13s	-31%
`polyveck_decompose`	✅	8s	9s	-11%
`keccakf1600_permute_native`	✅	7s	7s	+0%
`mld_check_pct`	✅	7s	10s	-30%
`mld_h`	✅	7s	2s	+250%
`pointwise_acc_native_aarch64`	✅	7s	6s	+17%
`polyvec_matrix_expand_serial`	✅	7s	7s	+0%
`polyveck_pointwise_poly_montgomery`	✅	7s	5s	+40%
`sign_open`	✅	7s	4s	+75%
`keccakf1600_permute`	✅	6s	9s	-33%
`ntt_native_x86_64`	✅	6s	2s	+200%
`poly_caddq_c`	✅	6s	4s	+50%
`poly_uniform`	✅	6s	7s	-14%
`poly_uniform_gamma1_4x`	✅	6s	3s	+100%
`polyeta_unpack`	✅	6s	7s	-14%
`polyveck_invntt_tomont`	✅	6s	4s	+50%
`polyveck_pack_eta`	✅	6s	6s	+0%
`polyvecl_ntt`	✅	6s	5s	+20%
`rej_eta_c`	✅	6s	5s	+20%
`sign`	✅	6s	6s	+0%
`keccak_absorb`	✅	5s	6s	-17%
`keccak_squeezeblocks_x4`	✅	5s	5s	+0%
`nttunpack_native_x86_64`	✅	5s	3s	+67%
`pack_pk`	✅	5s	3s	+67%
`pointwise_acc_native_x86_64`	✅	5s	5s	+0%
`poly_decompose`	✅	5s	3s	+67%
`poly_use_hint_c`	✅	5s	4s	+25%
`poly_use_hint_native_aarch64`	✅	5s	-	new
`polyveck_add`	✅	5s	6s	-17%
`polyveck_pack_w1`	✅	5s	4s	+25%
`polyveck_power2round`	✅	5s	7s	-29%
`polyvecl_chknorm`	✅	5s	5s	+0%
`polyvecl_uniform_gamma1_serial`	✅	5s	2s	+150%
`polyvecl_unpack_z`	✅	5s	2s	+150%
`polyw1_pack`	✅	5s	2s	+150%
`shake128_finalize`	✅	5s	2s	+150%
`sign_signature_extmu`	✅	5s	3s	+67%
`sign_signature_pre_hash_internal`	✅	5s	4s	+25%
`sys_check_capability`	✅	5s	3s	+67%
`yvec_init`	✅	5s	5s	+0%
`caddq`	✅	4s	3s	+33%
`keccak_f1600_x4_native_aarch64_v8a_scalar_hybrid`	✅	4s	2s	+100%
`keccak_squeeze`	✅	4s	3s	+33%
`keccakf1600x4_xor_bytes`	✅	4s	3s	+33%
`mld_sample_s1_s2`	✅	4s	5s	-20%
`mld_sample_s1_s2_serial`	✅	4s	3s	+33%
`mld_value_barrier_u32`	✅	4s	3s	+33%
`pack_sig_c`	✅	4s	2s	+100%
`pack_sig_z`	✅	4s	3s	+33%
`pack_sk_s1`	✅	4s	3s	+33%
`pointwise_native_x86_64`	✅	4s	3s	+33%
`poly_challenge`	✅	4s	4s	+0%
`poly_chknorm`	✅	4s	2s	+100%
`poly_ntt_native`	✅	4s	4s	+0%
`polyveck_shiftl`	✅	4s	2s	+100%
`polyveck_use_hint`	✅	4s	6s	-33%
`polyvecl_pointwise_acc_montgomery`	✅	4s	5s	-20%
`polyz_unpack`	✅	4s	3s	+33%
`polyz_unpack_17_native_aarch64`	✅	4s	4s	+0%
`polyz_unpack_19_native_aarch64`	✅	4s	3s	+33%
`shake256_squeeze`	✅	4s	2s	+100%
`sign_keypair`	✅	4s	7s	-43%
`sign_signature`	✅	4s	5s	-20%
`sign_signature_pre_hash_shake256`	✅	4s	7s	-43%
`sign_verify_pre_hash_internal`	✅	4s	3s	+33%
`unpack_hints`	✅	4s	8s	-50%
`unpack_sk`	✅	4s	2s	+100%
`unpack_sk_t0hat`	✅	4s	2s	+100%
`yvec_get_poly`	✅	4s	1s	+300%
`decompose`	✅	3s	4s	-25%
`keccak_f1600_x4_native_aarch64_v8a_v84a_scalar_hybrid`	✅	3s	2s	+50%
`keccakf1600_extract_bytes (big endian)`	✅	3s	4s	-25%
`keccakf1600_xor_bytes (big endian)`	✅	3s	1s	+200%
`keccakf1600x4_extract_bytes`	✅	3s	3s	+0%
`keccakf1600x4_permute`	✅	3s	2s	+50%
`make_hint`	✅	3s	3s	+0%
`mld_ct_get_optblocker_u8`	✅	3s	2s	+50%
`mld_keccakf1600_extract_bytes`	✅	3s	2s	+50%
`mld_prepare_domain_separation_prefix`	✅	3s	5s	-40%
`montgomery_reduce`	✅	3s	3s	+0%
`ntt_native_aarch64`	✅	3s	4s	-25%
`pack_sig_h_poly`	✅	3s	4s	-25%
`pointwise_native_aarch64`	✅	3s	3s	+0%
`poly_caddq`	✅	3s	5s	-40%
`poly_caddq_native`	✅	3s	4s	-25%
`poly_decompose_c`	✅	3s	3s	+0%
`poly_decompose_native`	✅	3s	6s	-50%
`poly_make_hint`	✅	3s	4s	-25%
`poly_ntt`	✅	3s	2s	+50%
`poly_permute_bitrev_to_custom_optional`	✅	3s	4s	-25%
`poly_permute_bitrev_to_custom_optional_native`	✅	3s	2s	+50%
`poly_pointwise_montgomery`	✅	3s	6s	-50%
`poly_pointwise_montgomery_native`	✅	3s	4s	-25%
`poly_reduce`	✅	3s	3s	+0%
`poly_shiftl`	✅	3s	3s	+0%
`poly_uniform_eta`	✅	3s	4s	-25%
`poly_uniform_gamma1`	✅	3s	3s	+0%
`polyeta_pack`	✅	3s	4s	-25%
`polyt0_pack`	✅	3s	5s	-40%
`polyt1_pack`	✅	3s	2s	+50%
`polyvec_matrix_pointwise_montgomery_row`	✅	3s	4s	-25%
`polyveck_caddq`	✅	3s	7s	-57%
`polyveck_reduce`	✅	3s	3s	+0%
`polyveck_sub`	✅	3s	6s	-50%
`polyveck_unpack_t0`	✅	3s	4s	-25%
`polyvecl_unpack_eta`	✅	3s	2s	+50%
`polyz_pack`	✅	3s	3s	+0%
`power2round`	✅	3s	3s	+0%
`rej_eta`	✅	3s	4s	-25%
`rej_eta_native`	✅	3s	3s	+0%
`shake128_absorb`	✅	3s	4s	-25%
`shake128x4_absorb_once`	✅	3s	4s	-25%
`shake256`	✅	3s	5s	-40%
`shake256_init`	✅	3s	3s	+0%
`sign_verify_pre_hash_shake256`	✅	3s	5s	-40%
`sk_s2hat_get_poly`	✅	3s	2s	+50%
`unpack_pk`	✅	3s	3s	+0%
`unpack_sig`	✅	3s	5s	-40%
`unpack_sk_s1hat`	✅	3s	3s	+0%
`use_hint`	✅	3s	3s	+0%
`intt_native_x86_64`	✅	2s	3s	-33%
`keccak_f1600_x1_native_aarch64`	✅	2s	3s	-33%
`keccak_f1600_x1_native_aarch64_v84a`	✅	2s	1s	+100%
`keccak_finalize`	✅	2s	3s	-33%
`keccak_init`	✅	2s	2s	+0%
`mld_ct_abs_i32`	✅	2s	4s	-50%
`mld_ct_cmask_neg_i32`	✅	2s	2s	+0%
`mld_ct_cmask_nonzero_u32`	✅	2s	3s	-33%
`mld_ct_cmask_nonzero_u8`	✅	2s	2s	+0%
`mld_ct_get_optblocker_u32`	✅	2s	4s	-50%
`mld_ct_sel_int32`	✅	2s	1s	+100%
`mld_polymat_expand_entry`	✅	2s	5s	-60%
`mld_value_barrier_i64`	✅	2s	2s	+0%
`pack_sk_rho_key_tr_s2_t0`	✅	2s	4s	-50%
`poly_caddq_native_aarch64`	✅	2s	4s	-50%
`poly_chknorm_native`	✅	2s	3s	-33%
`poly_invntt_tomont`	✅	2s	2s	+0%
`poly_invntt_tomont_native`	✅	2s	2s	+0%
`poly_ntt_c`	✅	2s	4s	-50%
`poly_sub`	✅	2s	4s	-50%
`poly_use_hint_native`	✅	2s	5s	-60%
`polyt1_unpack`	✅	2s	3s	-33%
`polyveck_ntt`	✅	2s	3s	-33%
`polyveck_pack_t0`	✅	2s	4s	-50%
`polyveck_unpack_eta`	✅	2s	4s	-50%
`polyvecl_pack_eta`	✅	2s	4s	-50%
`polyvecl_pointwise_acc_montgomery_native`	✅	2s	2s	+0%
`polyvecl_uniform_gamma1`	✅	2s	3s	-33%
`polyz_unpack_native`	✅	2s	4s	-50%
`reduce32`	✅	2s	2s	+0%
`shake128_init`	✅	2s	3s	-33%
`shake128_release`	✅	2s	3s	-33%
`shake128_squeeze`	✅	2s	3s	-33%
`shake128x4_squeezeblocks`	✅	2s	3s	-33%
`shake256_absorb`	✅	2s	2s	+0%
`shake256_finalize`	✅	2s	2s	+0%
`shake256x4_absorb_once`	✅	2s	2s	+0%
`shake256x4_squeezeblocks`	✅	2s	2s	+0%
`sign_verify`	✅	2s	6s	-67%
`sign_verify_extmu`	✅	2s	4s	-50%
`sk_s1hat_get_poly`	✅	2s	4s	-50%
`sk_t0hat_get_poly`	✅	2s	4s	-50%
`unpack_sk_s2hat`	✅	2s	3s	-33%
`fqscale`	✅	1s	3s	-67%
`keccak_f1600_x4_native_aarch64_v84a`	✅	1s	3s	-67%
`keccakf1600_xor_bytes`	✅	1s	3s	-67%
`mld_ct_get_optblocker_i64`	✅	1s	3s	-67%
`mld_value_barrier_u8`	✅	1s	2s	-50%
`poly_chknorm_native_aarch64`	✅	1s	3s	-67%
`poly_use_hint`	✅	1s	2s	-50%
`shake256_release`	✅	1s	2s	-50%

mkannwischer

Thanks @jakemas. Please add the CBMC proofs.

jakemas · 2026-04-14T07:46:04Z

Thanks @jakemas. Please add the CBMC proofs.

Thank you, added the harness as a single file with split functionality on GAMMA2

hanno-becker

Thank you @jakemas for working on this!

The specification seems far too weak to claim functional correctness? We seem to only say something about the bounds of the output.

jakemas · 2026-04-30T04:48:35Z

Update: FIPS 204-aligned specs for _32 (commit `2c8a1e6`)

Addressed the review comments for the _32 variant:

FIPS 204 alignment (Hanno + Matthias)

Added mldsa_cmod, mldsa_decompose_32, mldsa_use_hint_32 definitions in mldsa_specs.ml, closely following FIPS 204 Algorithms 36 and 40
Proved MLDSA_USE_HINT_32_EQUIV: mldsa_use_hint_32 h r = mldsa_use_hint_32_spec r h — no CHEAT_TAC
SUBROUTINE_CORRECT postcondition now uses mldsa_use_hint_32 (FIPS 204-aligned)
Key sub-lemmas: DECOMPOSE_32_R1_EQUIV (17-interval Barrett analysis), DECOMPOSE_32_R0_SIGN (r0 sign equivalence)

Other fixes

Removed unused MLDSA_USE_HINT_32_SPEC_BOUND (Matthias)
Safety proof uses SUBROUTINE_CORRECT_INTERNAL (code-aligned spec with val < 16 bound)
All specs in common/mldsa_specs.ml for x86 sharing

Pending

_88 variant: same treatment in progress

jakemas · 2026-05-01T01:16:46Z

woo got it! testing _88 in CI now then will merge shared code ( DIV_SANDWICH, ENSURES_STRENGTHEN_POST, and INT_MOD_RESIDUE) maybe go eat a div sandwich myself 🥪

mkannwischer

Thank you so much @jakemas for getting this to work so quickly! The use_hint and decompose specs now look great to me!

The only blocker left for me is that the SUBROUTINE_CORRECT_FIPS204 specs are missing the output bounds.
Also, can we rename SUBROUTINE_CORRECT_FIPS204 to SUBROUTINE_CORRECT for consistency with the other proofs and remove the old POLY_USE_HINT_32_AARCH64_ASM_SUBROUTINE_CORRECT? That is otherwise quite confusing to read.

mkannwischer

Thanks @jakemas for the updates.
The final specs now look great to me!

It would be nice if we could eliminate the CORRECT_CODE spec as there is no other proof in mlkem-native/mldsa-native that has multiple specs. I'd prefer it if we can keep this uniform. This can be done in a follow-up.

Two nits that should still be fixed regarding the CBMC contracts, but I'll take the liberty to fix those myself shortly.

Add formal verification proofs for poly_use_hint_32 and poly_use_hint_88 AArch64 NEON implementations using HOL Light. Each proof file contains: - Assembly-level functional correctness - Subroutine correctness with code-aligned spec - Constant-time and memory safety - FIPS 204 equivalence proof via Barrett interval cascade - FIPS 204-aligned subroutine correctness FIPS 204 definitions (mldsa_cmod, mldsa_decompose, mldsa_use_hint) and shared helpers (DIV_SANDWICH, INT_MOD_RESIDUE) are in mldsa_specs.ml. ENSURES_STRENGTHEN_POST lives in aarch64_utils.ml. The public SUBROUTINE_CORRECT theorem in each proof uses the FIPS 204 spec mldsa_use_hint_{32,88} in its postcondition and carries the output bound (< 16 or < 44) as a corollary. Code-aligned intermediates use *_CODE suffixes to mark them internal. Signed-off-by: Jake Massimo <jakemas@amazon.com>

hanno-becker · 2026-05-02T05:31:55Z

@jakemas This is the pattern I had in mind: You write the spec as you want it, but start the proof by massaging it into a shape that's better aligned to the code, including a) adding the automatic bounds, b) moving from FIPS204-aligned to code-aligned.

Could you internalize this pattern, and open a follow-up PR removing POLY_USE_HINT_88_AARCH64_ASM_CORRECT_CODE and POLY_USE_HINT_88_AARCH64_ASM_CORRECT_BOUND_CODE?

let MLDSA_USE_HINT_88_EQUIV = prove(
  `!r h. r < 8380417 /\ h <= 1
         ==> mldsa_use_hint_88 h r = mldsa_use_hint_88_code r h`, 
   (* FROM YOUR CODE *) );;

let POLY_USE_HINT_88_AARCH64_ASM_CORRECT = prove
 (`!b a h x y pc.
    nonoverlapping (word pc, LENGTH poly_use_hint_88_aarch64_asm_mc) (b, 1024) /\
    nonoverlapping (b, 1024) (a, 1024) /\
    nonoverlapping (b, 1024) (h, 1024)
    ==> ensures arm
          (\s. aligned_bytes_loaded s (word pc) poly_use_hint_88_aarch64_asm_mc /\
               read PC s = word pc /\
               C_ARGUMENTS [b; a; h] s /\
               (!i. i < 256 ==> val(x i) < 8380417) /\
               (!i. i < 256 ==> val(y i) <= 1) /\
               (!i. i < 256 ==>
                 read(memory :> bytes32(word_add a (word(4 * i)))) s = x i) /\
               (!i. i < 256 ==>
                 read(memory :> bytes32(word_add h (word(4 * i)))) s = y i))
          (\s. read PC s = word(pc + LENGTH poly_use_hint_88_aarch64_asm_mc - 4) /\
               (!i. i < 256 ==>
                 read(memory :> bytes32(word_add b (word(4 * i)))) s =
                   word(mldsa_use_hint_88 (val(y i)) (val(x i)))) /\
               (!i. i < 256 ==>
                 val(read(memory :> bytes32(word_add b (word(4 * i)))) s) < 44))
          (MAYCHANGE_REGS_AND_FLAGS_PERMITTED_BY_ABI ,,
           MAYCHANGE [memory :> bytes(b, 1024)])`,

  (* First, globalize the variables and assumptions. In particular, this will
    allow us to refer to the pure preconditions when rewriting the post condition. *)
  MAP_EVERY X_GEN_TAC
    [`b:int64`; `a:int64`; `h:int64`;
     `x:num->int32`; `y:num->int32`; `pc:num`] THEN
  REWRITE_TAC[MAYCHANGE_REGS_AND_FLAGS_PERMITTED_BY_ABI; C_ARGUMENTS;
              NONOVERLAPPING_CLAUSES; ALL;
              fst POLY_USE_HINT_88_AARCH64_ASM_EXEC] THEN
  DISCH_THEN(REPEAT_TCL CONJUNCTS_THEN ASSUME_TAC) THEN
  GLOBALIZE_PRECONDITION_TAC THEN
  
  (* First, show that the bounds in the post-condition are automatic once the
     functional description is established. *)
   MATCH_MP_TAC ENSURES_STRENGTHEN_POST THEN
   EXISTS_TAC
    `\s. read PC s = word(pc + LENGTH poly_use_hint_88_aarch64_asm_mc - 4) /\
         (!i. i < 256 ==>
           read(memory :> bytes32(word_add b (word(4 * i)))) s =
             word(mldsa_use_hint_88_code (val(x i:int32)) (val(y i:int32))))` THEN
   CONJ_TAC THENL 
  [ (* Defer main proof *)
    ALL_TAC;
    (* Rewrite to code-aligned functional description *)
    REPEAT (IMP_REWRITE_TAC[MLDSA_USE_HINT_88_EQUIV]) THEN
    (* Show that bounds are automatic *)
   REWRITE_TAC[fst POLY_USE_HINT_88_AARCH64_ASM_EXEC] THEN REPEAT STRIP_TAC THEN ASM_REWRITE_TAC[] THEN
   FIRST_X_ASSUM(MP_TAC o SPEC `i:num`) THEN ASM_REWRITE_TAC[] THEN
   DISCH_THEN SUBST1_TAC THEN REWRITE_TAC[VAL_WORD; DIMINDEX_32] THEN
   CONV_TAC NUM_REDUCE_CONV THEN
   MATCH_MP_TAC(ARITH_RULE `x < 44 ==> x MOD 4294967296 < 44`) THEN
   REWRITE_TAC[mldsa_use_hint_88_code] THEN
   CONV_TAC(TOP_DEPTH_CONV let_CONV) THEN
   REPEAT(COND_CASES_TAC THEN ASM_REWRITE_TAC[]) THEN ASM_ARITH_TAC] THEN

   (* Now do the actual proof -- literally as before *)
  CONV_TAC(RATOR_CONV(LAND_CONV(ONCE_DEPTH_CONV EXPAND_CASES_CONV))) THEN
  CONV_TAC NUM_REDUCE_CONV THEN
  REPEAT STRIP_TAC THEN
  REWRITE_TAC[SOME_FLAGS; MODIFIABLE_SIMD_REGS] THEN

  ENSURES_INIT_TAC "s0" THEN
  MEMORY_128_FROM_32_TAC "a" 0 64 THEN
  ASM_REWRITE_TAC[WORD_ADD_0] THEN CONV_TAC WORD_REDUCE_CONV THEN
  STRIP_TAC THEN
  MEMORY_128_FROM_32_TAC "h" 0 64 THEN
  ASM_REWRITE_TAC[WORD_ADD_0] THEN CONV_TAC WORD_REDUCE_CONV THEN
  STRIP_TAC THEN
  DISCARD_MATCHING_ASSUMPTIONS [`read (memory :> bytes32 a) s = x`] THEN

  MAP_EVERY (fun n -> ARM_STEPS_TAC POLY_USE_HINT_88_AARCH64_ASM_EXEC [n] THEN
                      SIMD_SIMPLIFY_TAC[])
        (1--1006) THEN
  ENSURES_FINAL_STATE_TAC THEN ASM_REWRITE_TAC[] THEN

  REPEAT(FIRST_X_ASSUM(STRIP_ASSUME_TAC o
    CONV_RULE (SIMD_SIMPLIFY_CONV []) o
    CONV_RULE(READ_MEMORY_SPLIT_CONV 2) o
    check (can (term_match [] `read qqq s:int128 = xxx`) o concl))) THEN

  CONV_TAC(TOP_DEPTH_CONV EXPAND_CASES_CONV) THEN
  CONV_TAC(DEPTH_CONV NUM_MULT_CONV THENC DEPTH_CONV NUM_ADD_CONV) THEN
  REWRITE_TAC[WORD_ADD_0] THEN
  ASM_REWRITE_TAC[WORD_ADD_0] THEN ASM_REWRITE_TAC[] THEN

  (* Push word_subword through SIMD ops to per-element form *)
  REWRITE_TAC[WORD_SUBWORD_AND; WORD_SUBWORD_OR] THEN
  let WSN_TAC = REWRITE_TAC(map (fun n -> prove(
    subst [mk_small_numeral n, `n:num`]
      `!x:int128. word_subword(word_not x) (n,32):int32 = word_not(word_subword x (n,32))`,
    GEN_TAC THEN MATCH_MP_TAC WORD_SUBWORD_NOT THEN
    REWRITE_TAC[DIMINDEX_32; DIMINDEX_128] THEN ARITH_TAC)) [0;32;64;96]) in
  WSN_TAC THEN
  CONV_TAC(DEPTH_CONV WORD_SIMPLE_SUBWORD_CONV) THEN
  CONV_TAC(DEPTH_CONV WORD_NUM_RED_CONV) THEN
  let EC_DEEP = CONV_RULE(DEPTH_CONV WORD_NUM_RED_CONV)
    (CONV_RULE(DEPTH_CONV(INT_RED_CONV ORELSEC NUM_RED_CONV))
      (CONV_RULE(TOP_DEPTH_CONV let_CONV)
        (REWRITE_RULE[mldsa_use_hint_88_asm; word_2smulh; word_ishr_round;
                       DIMINDEX_32] ELEMENT_CORRECT_WORD_88))) in
  let EC_OR = ONCE_REWRITE_RULE[WORD_OR_SYM] EC_DEEP in
  REPEAT CONJ_TAC THEN
  (MATCH_MP_TAC EC_OR ORELSE MATCH_MP_TAC EC_DEEP) THEN
  CONJ_TAC THEN FIRST_X_ASSUM MATCH_MP_TAC THEN ARITH_TAC);;

jakemas · 2026-05-02T05:40:15Z

@jakemas This is the pattern I had in mind: You write the spec as you want it, but start the proof by massaging it into a shape that's better aligned to the code, including a) adding the automatic bounds, b) moving from FIPS204-aligned to code-aligned.

Could you internalize this pattern, and open a follow-up PR removing POLY_USE_HINT_88_AARCH64_ASM_CORRECT_CODE and POLY_USE_HINT_88_AARCH64_ASM_CORRECT_BOUND_CODE?

let MLDSA_USE_HINT_88_EQUIV = prove(
  `!r h. r < 8380417 /\ h <= 1
         ==> mldsa_use_hint_88 h r = mldsa_use_hint_88_code r h`, 
   (* FROM YOUR CODE *) );;

let POLY_USE_HINT_88_AARCH64_ASM_CORRECT = prove
 (`!b a h x y pc.
    nonoverlapping (word pc, LENGTH poly_use_hint_88_aarch64_asm_mc) (b, 1024) /\
    nonoverlapping (b, 1024) (a, 1024) /\
    nonoverlapping (b, 1024) (h, 1024)
    ==> ensures arm
          (\s. aligned_bytes_loaded s (word pc) poly_use_hint_88_aarch64_asm_mc /\
               read PC s = word pc /\
               C_ARGUMENTS [b; a; h] s /\
               (!i. i < 256 ==> val(x i) < 8380417) /\
               (!i. i < 256 ==> val(y i) <= 1) /\
               (!i. i < 256 ==>
                 read(memory :> bytes32(word_add a (word(4 * i)))) s = x i) /\
               (!i. i < 256 ==>
                 read(memory :> bytes32(word_add h (word(4 * i)))) s = y i))
          (\s. read PC s = word(pc + LENGTH poly_use_hint_88_aarch64_asm_mc - 4) /\
               (!i. i < 256 ==>
                 read(memory :> bytes32(word_add b (word(4 * i)))) s =
                   word(mldsa_use_hint_88 (val(y i)) (val(x i)))) /\
               (!i. i < 256 ==>
                 val(read(memory :> bytes32(word_add b (word(4 * i)))) s) < 44))
          (MAYCHANGE_REGS_AND_FLAGS_PERMITTED_BY_ABI ,,
           MAYCHANGE [memory :> bytes(b, 1024)])`,

  (* First, globalize the variables and assumptions. In particular, this will
    allow us to refer to the pure preconditions when rewriting the post condition. *)
  MAP_EVERY X_GEN_TAC
    [`b:int64`; `a:int64`; `h:int64`;
     `x:num->int32`; `y:num->int32`; `pc:num`] THEN
  REWRITE_TAC[MAYCHANGE_REGS_AND_FLAGS_PERMITTED_BY_ABI; C_ARGUMENTS;
              NONOVERLAPPING_CLAUSES; ALL;
              fst POLY_USE_HINT_88_AARCH64_ASM_EXEC] THEN
  DISCH_THEN(REPEAT_TCL CONJUNCTS_THEN ASSUME_TAC) THEN
  GLOBALIZE_PRECONDITION_TAC THEN
  
  (* First, show that the bounds in the post-condition are automatic once the
     functional description is established. *)
   MATCH_MP_TAC ENSURES_STRENGTHEN_POST THEN
   EXISTS_TAC
    `\s. read PC s = word(pc + LENGTH poly_use_hint_88_aarch64_asm_mc - 4) /\
         (!i. i < 256 ==>
           read(memory :> bytes32(word_add b (word(4 * i)))) s =
             word(mldsa_use_hint_88_code (val(x i:int32)) (val(y i:int32))))` THEN
   CONJ_TAC THENL 
  [ (* Defer main proof *)
    ALL_TAC;
    (* Rewrite to code-aligned functional description *)
    REPEAT (IMP_REWRITE_TAC[MLDSA_USE_HINT_88_EQUIV]) THEN
    (* Show that bounds are automatic *)
   REWRITE_TAC[fst POLY_USE_HINT_88_AARCH64_ASM_EXEC] THEN REPEAT STRIP_TAC THEN ASM_REWRITE_TAC[] THEN
   FIRST_X_ASSUM(MP_TAC o SPEC `i:num`) THEN ASM_REWRITE_TAC[] THEN
   DISCH_THEN SUBST1_TAC THEN REWRITE_TAC[VAL_WORD; DIMINDEX_32] THEN
   CONV_TAC NUM_REDUCE_CONV THEN
   MATCH_MP_TAC(ARITH_RULE `x < 44 ==> x MOD 4294967296 < 44`) THEN
   REWRITE_TAC[mldsa_use_hint_88_code] THEN
   CONV_TAC(TOP_DEPTH_CONV let_CONV) THEN
   REPEAT(COND_CASES_TAC THEN ASM_REWRITE_TAC[]) THEN ASM_ARITH_TAC] THEN

   (* Now do the actual proof -- literally as before *)
  CONV_TAC(RATOR_CONV(LAND_CONV(ONCE_DEPTH_CONV EXPAND_CASES_CONV))) THEN
  CONV_TAC NUM_REDUCE_CONV THEN
  REPEAT STRIP_TAC THEN
  REWRITE_TAC[SOME_FLAGS; MODIFIABLE_SIMD_REGS] THEN

  ENSURES_INIT_TAC "s0" THEN
  MEMORY_128_FROM_32_TAC "a" 0 64 THEN
  ASM_REWRITE_TAC[WORD_ADD_0] THEN CONV_TAC WORD_REDUCE_CONV THEN
  STRIP_TAC THEN
  MEMORY_128_FROM_32_TAC "h" 0 64 THEN
  ASM_REWRITE_TAC[WORD_ADD_0] THEN CONV_TAC WORD_REDUCE_CONV THEN
  STRIP_TAC THEN
  DISCARD_MATCHING_ASSUMPTIONS [`read (memory :> bytes32 a) s = x`] THEN

  MAP_EVERY (fun n -> ARM_STEPS_TAC POLY_USE_HINT_88_AARCH64_ASM_EXEC [n] THEN
                      SIMD_SIMPLIFY_TAC[])
        (1--1006) THEN
  ENSURES_FINAL_STATE_TAC THEN ASM_REWRITE_TAC[] THEN

  REPEAT(FIRST_X_ASSUM(STRIP_ASSUME_TAC o
    CONV_RULE (SIMD_SIMPLIFY_CONV []) o
    CONV_RULE(READ_MEMORY_SPLIT_CONV 2) o
    check (can (term_match [] `read qqq s:int128 = xxx`) o concl))) THEN

  CONV_TAC(TOP_DEPTH_CONV EXPAND_CASES_CONV) THEN
  CONV_TAC(DEPTH_CONV NUM_MULT_CONV THENC DEPTH_CONV NUM_ADD_CONV) THEN
  REWRITE_TAC[WORD_ADD_0] THEN
  ASM_REWRITE_TAC[WORD_ADD_0] THEN ASM_REWRITE_TAC[] THEN

  (* Push word_subword through SIMD ops to per-element form *)
  REWRITE_TAC[WORD_SUBWORD_AND; WORD_SUBWORD_OR] THEN
  let WSN_TAC = REWRITE_TAC(map (fun n -> prove(
    subst [mk_small_numeral n, `n:num`]
      `!x:int128. word_subword(word_not x) (n,32):int32 = word_not(word_subword x (n,32))`,
    GEN_TAC THEN MATCH_MP_TAC WORD_SUBWORD_NOT THEN
    REWRITE_TAC[DIMINDEX_32; DIMINDEX_128] THEN ARITH_TAC)) [0;32;64;96]) in
  WSN_TAC THEN
  CONV_TAC(DEPTH_CONV WORD_SIMPLE_SUBWORD_CONV) THEN
  CONV_TAC(DEPTH_CONV WORD_NUM_RED_CONV) THEN
  let EC_DEEP = CONV_RULE(DEPTH_CONV WORD_NUM_RED_CONV)
    (CONV_RULE(DEPTH_CONV(INT_RED_CONV ORELSEC NUM_RED_CONV))
      (CONV_RULE(TOP_DEPTH_CONV let_CONV)
        (REWRITE_RULE[mldsa_use_hint_88_asm; word_2smulh; word_ishr_round;
                       DIMINDEX_32] ELEMENT_CORRECT_WORD_88))) in
  let EC_OR = ONCE_REWRITE_RULE[WORD_OR_SYM] EC_DEEP in
  REPEAT CONJ_TAC THEN
  (MATCH_MP_TAC EC_OR ORELSE MATCH_MP_TAC EC_DEEP) THEN
  CONJ_TAC THEN FIRST_X_ASSUM MATCH_MP_TAC THEN ARITH_TAC);;

ok, will do!

hanno-becker

Many thanks @jakemas !

mkannwischer · 2026-05-02T05:46:33Z

I opened #1085 for @hanno-becker's suggestion so we don't forget.

I will merge this PR shortly.

jakemas requested a review from a team as a code owner April 13, 2026 16:54

jakemas force-pushed the hol-light-aarch64-use-hint branch from 2053888 to 0b42385 Compare April 13, 2026 17:02

jakemas marked this pull request as draft April 13, 2026 17:12

jakemas force-pushed the hol-light-aarch64-use-hint branch from 0b42385 to c5c76af Compare April 13, 2026 17:21

jakemas marked this pull request as ready for review April 13, 2026 17:52

mkannwischer requested changes Apr 13, 2026

View reviewed changes

jakemas force-pushed the hol-light-aarch64-use-hint branch 5 times, most recently from e7ca229 to e7a3a45 Compare April 14, 2026 07:44

jakemas added the hol-light label Apr 14, 2026

jakemas force-pushed the hol-light-aarch64-use-hint branch 4 times, most recently from 1ea7797 to 817aa3d Compare April 17, 2026 10:22

hanno-becker reviewed Apr 17, 2026

View reviewed changes

Comment thread proofs/hol_light/aarch64/proofs/mldsa_poly_use_hint_32.ml Outdated

hanno-becker reviewed Apr 17, 2026

View reviewed changes

Comment thread proofs/hol_light/aarch64/proofs/mldsa_poly_use_hint_88.ml Outdated

hanno-becker requested changes Apr 17, 2026

View reviewed changes

jakemas force-pushed the hol-light-aarch64-use-hint branch 7 times, most recently from 40ec688 to de080de Compare April 17, 2026 15:39

mkannwischer requested changes Apr 29, 2026

View reviewed changes

Comment thread proofs/hol_light/common/mldsa_specs.ml Outdated

mkannwischer force-pushed the hol-light-aarch64-use-hint branch from c97d731 to e96df0d Compare April 29, 2026 13:10

mkannwischer reviewed Apr 29, 2026

View reviewed changes

Comment thread proofs/hol_light/aarch64/proofs/poly_use_hint_32_aarch64_asm.ml Outdated

jakemas force-pushed the hol-light-aarch64-use-hint branch from f2f03df to 2c8a1e6 Compare April 30, 2026 04:42

jakemas force-pushed the hol-light-aarch64-use-hint branch 7 times, most recently from 67b89be to a879424 Compare May 1, 2026 00:43

jakemas force-pushed the hol-light-aarch64-use-hint branch 5 times, most recently from 3c1d5a1 to 999243b Compare May 1, 2026 08:01

mkannwischer requested changes May 1, 2026

View reviewed changes

jakemas force-pushed the hol-light-aarch64-use-hint branch from 999243b to 1e91386 Compare May 1, 2026 08:26

mkannwischer mentioned this pull request May 2, 2026

CBMC: Consider combining polyz_unpack_{17,19}_native_aarch64 into one proof #1084

Open

mkannwischer approved these changes May 2, 2026

View reviewed changes

Comment thread dev/aarch64_opt/src/arith_native_aarch64.h

Comment thread dev/aarch64_opt/src/arith_native_aarch64.h Outdated

hanno-becker approved these changes May 2, 2026

View reviewed changes

mkannwischer mentioned this pull request May 2, 2026

HOL-Light: Consolidate AArch64 poly_use_hint proofs #1085

Open

mkannwischer mentioned this pull request May 6, 2026

HOL-Light: Prove AArch64 poly_use_hint #486

Closed

jakemas mentioned this pull request May 29, 2026

HOL-Light: Consolidate AArch64 poly_use_hint_32 _CORRECT theorems #1136

Open

2 tasks

Conversation

jakemas commented Apr 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Uh oh!

oqs-bot commented Apr 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

CBMC Results (ML-DSA-87)

Uh oh!

oqs-bot commented Apr 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

CBMC Results (ML-DSA-65)

Uh oh!

oqs-bot commented Apr 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

CBMC Results (ML-DSA-44)

Uh oh!

mkannwischer left a comment

Choose a reason for hiding this comment

Uh oh!

jakemas commented Apr 14, 2026

Uh oh!

Uh oh!

Uh oh!

hanno-becker left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

jakemas commented Apr 30, 2026

Update: FIPS 204-aligned specs for _32 (commit 2c8a1e6)

FIPS 204 alignment (Hanno + Matthias)

Other fixes

Pending

Uh oh!

jakemas commented May 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

mkannwischer left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

mkannwischer left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

hanno-becker commented May 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jakemas commented May 2, 2026

Uh oh!

hanno-becker left a comment

Choose a reason for hiding this comment

Uh oh!

mkannwischer commented May 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

jakemas commented Apr 13, 2026 •

edited

Loading

oqs-bot commented Apr 13, 2026 •

edited

Loading

oqs-bot commented Apr 13, 2026 •

edited

Loading

oqs-bot commented Apr 13, 2026 •

edited

Loading

Update: FIPS 204-aligned specs for _32 (commit `2c8a1e6`)

jakemas commented May 1, 2026 •

edited

Loading

hanno-becker commented May 2, 2026 •

edited

Loading

mkannwischer commented May 2, 2026 •

edited

Loading