Lowram: Share buffers with non-overlapping lifetimes in keygen by mkannwischer · Pull Request #1011 · pq-code-package/mldsa-native

mkannwischer · 2026-04-02T03:19:35Z

Hoisted out from PoC: Reduce large struct allocations to <= 13/17/21 KiB for ML-DSA-44/65/87 #1005

Reuse t0 as the accumulator in mld_compute_t0_t1_tr_from_sk_components, and have the caller provide s1 already in NTT form, removing two allocations (s1hat and t) from the helper.

In mld_sign_keypair_internal, share the s1 and t1 buffers via a union since s1hat is consumed before t1 is produced. Pack s1 into the secret key before the in-place NTT so the original coefficients are preserved.

Split mld_pack_sk into mld_pack_sk_s1 and mld_pack_sk_rho_key_tr_s2_t0 to support packing s1 independently before the NTT.

oqs-bot · 2026-04-02T04:27:01Z

CBMC Results (ML-DSA-44)

⚠️ Attention Required

Proof	Status	Current	Previous	Change
`sign_keypair_internal`	⚠️	22s	4s	+450%

Full Results (186 proofs)

Proof	Status	Current	Previous	Change
`TOTAL`	✅	1854s	1731s	+7.1%
`polyvecl_pointwise_acc_montgomery_c`	✅	182s	154s	+18%
`sign_verify_internal`	✅	166s	150s	+11%
`poly_pointwise_montgomery_c`	✅	163s	143s	+14%
`rej_uniform_native`	✅	145s	133s	+9%
`mld_ct_memcmp`	✅	83s	72s	+15%
`mld_invntt_layer`	✅	67s	62s	+8%
`mld_ntt_layer`	✅	56s	51s	+10%
`mld_attempt_signature_generation`	✅	55s	57s	-4%
`polymat_permute_bitrev_to_custom`	✅	29s	29s	+0%
`polyvec_matrix_expand`	✅	28s	25s	+12%
`rej_uniform`	✅	23s	23s	+0%
`sign_keypair_internal`	⚠️	22s	4s	+450%
`fqmul`	✅	20s	19s	+5%
`poly_chknorm_c`	✅	20s	19s	+5%
`poly_uniform_eta_4x`	✅	19s	19s	+0%
`sign_pk_from_sk`	✅	17s	8s	+112%
`poly_uniform_4x`	✅	16s	13s	+23%
`sign_signature_internal`	✅	16s	22s	-27%
`keccakf1600x4_permute_native`	✅	15s	16s	-6%
`polyt0_unpack`	✅	15s	14s	+7%
`rej_uniform_c`	✅	15s	16s	-6%
`polyeta_unpack`	✅	14s	14s	+0%
`polyz_unpack_c`	✅	14s	12s	+17%
`poly_add`	✅	13s	11s	+18%
`mld_ntt_butterfly_block`	✅	11s	9s	+22%
`mld_check_pct`	✅	10s	16s	-38%
`polyveck_power2round`	✅	10s	4s	+150%
`keccak_absorb_once_x4`	✅	9s	10s	-10%
`poly_decompose_c`	✅	9s	10s	-10%
`polyveck_add`	✅	9s	8s	+12%
`keccak_absorb`	✅	8s	6s	+33%
`keccak_squeezeblocks_x4`	✅	8s	6s	+33%
`keccakf1600_permute`	✅	8s	7s	+14%
`keccakf1600_permute_native`	✅	8s	8s	+0%
`poly_caddq_c`	✅	8s	7s	+14%
`polyvec_matrix_pointwise_montgomery`	✅	8s	8s	+0%
`polyveck_shiftl`	✅	8s	8s	+0%
`mld_compute_pack_z`	✅	7s	6s	+17%
`poly_invntt_tomont_c`	✅	7s	5s	+40%
`polyvec_matrix_expand_serial`	✅	7s	7s	+0%
`polyveck_pointwise_poly_montgomery`	✅	7s	6s	+17%
`polyveck_reduce`	✅	7s	3s	+133%
`polyveck_use_hint`	✅	7s	6s	+17%
`rej_eta_native`	✅	7s	8s	-12%
`mld_polyvecl_permute_bitrev_to_custom_native`	✅	6s	7s	-14%
`pointwise_acc_native_aarch64`	✅	6s	8s	-25%
`poly_power2round`	✅	6s	3s	+100%
`poly_uniform_eta`	✅	6s	5s	+20%
`poly_uniform_gamma1_4x`	✅	6s	2s	+200%
`poly_use_hint_c`	✅	6s	5s	+20%
`polyt1_unpack`	✅	6s	2s	+200%
`polyveck_caddq`	✅	6s	4s	+50%
`polyvecl_ntt`	✅	6s	5s	+20%
`polyz_unpack`	✅	6s	4s	+50%
`sign`	✅	6s	4s	+50%
`unpack_sk`	✅	6s	7s	-14%
`fqscale`	✅	5s	2s	+150%
`mld_h`	✅	5s	4s	+25%
`mld_prepare_domain_separation_prefix`	✅	5s	7s	-29%
`pack_sig_c`	✅	5s	4s	+25%
`pointwise_acc_native_x86_64`	✅	5s	5s	+0%
`poly_challenge`	✅	5s	3s	+67%
`poly_invntt_tomont`	✅	5s	5s	+0%
`polyveck_invntt_tomont`	✅	5s	4s	+25%
`polyveck_sub`	✅	5s	8s	-38%
`polyveck_unpack_eta`	✅	5s	4s	+25%
`polyvecl_pointwise_acc_montgomery_native`	✅	5s	4s	+25%
`sign_signature_extmu`	✅	5s	3s	+67%
`sign_verify`	✅	5s	4s	+25%
`sign_verify_extmu`	✅	5s	6s	-17%
`sign_verify_pre_hash_internal`	✅	5s	7s	-29%
`unpack_hints`	✅	5s	6s	-17%
`keccak_f1600_x4_native_aarch64_v84a`	✅	4s	3s	+33%
`keccak_finalize`	✅	4s	4s	+0%
`keccakf1600_xor_bytes`	✅	4s	2s	+100%
`keccakf1600_xor_bytes (big endian)`	✅	4s	2s	+100%
`mld_ct_cmask_nonzero_u32`	✅	4s	3s	+33%
`mld_ct_get_optblocker_u32`	✅	4s	2s	+100%
`mld_ct_sel_int32`	✅	4s	2s	+100%
`mld_sample_s1_s2`	✅	4s	4s	+0%
`mld_sample_s1_s2_serial`	✅	4s	5s	-20%
`ntt_native_aarch64`	✅	4s	4s	+0%
`pointwise_native_aarch64`	✅	4s	6s	-33%
`poly_decompose_native`	✅	4s	3s	+33%
`poly_ntt`	✅	4s	5s	-20%
`poly_ntt_native`	✅	4s	3s	+33%
`poly_reduce`	✅	4s	3s	+33%
`poly_shiftl`	✅	4s	3s	+33%
`poly_uniform`	✅	4s	3s	+33%
`poly_uniform_gamma1`	✅	4s	1s	+300%
`poly_use_hint_native`	✅	4s	2s	+100%
`polyveck_decompose`	✅	4s	12s	-67%
`polyveck_ntt`	✅	4s	5s	-20%
`polyvecl_permute_bitrev_to_custom`	✅	4s	3s	+33%
`polyvecl_uniform_gamma1`	✅	4s	6s	-33%
`rej_eta_c`	✅	4s	3s	+33%
`shake128_init`	✅	4s	2s	+100%
`shake128x4_squeezeblocks`	✅	4s	5s	-20%
`shake256_finalize`	✅	4s	4s	+0%
`sign_open`	✅	4s	4s	+0%
`sign_signature`	✅	4s	5s	-20%
`sign_signature_pre_hash_shake256`	✅	4s	4s	+0%
`sign_verify_pre_hash_shake256`	✅	4s	4s	+0%
`unpack_sig`	✅	4s	4s	+0%
`caddq`	✅	3s	2s	+50%
`decompose`	✅	3s	2s	+50%
`intt_native_x86_64`	✅	3s	4s	-25%
`keccak_f1600_x1_native_aarch64`	✅	3s	1s	+200%
`keccak_squeeze`	✅	3s	3s	+0%
`keccakf1600x4_extract_bytes`	✅	3s	1s	+200%
`keccakf1600x4_xor_bytes`	✅	3s	2s	+50%
`make_hint`	✅	3s	2s	+50%
`mld_keccakf1600_extract_bytes`	✅	3s	2s	+50%
`mld_value_barrier_i64`	✅	3s	2s	+50%
`ntt_native_x86_64`	✅	3s	3s	+0%
`pack_pk`	✅	3s	6s	-50%
`pack_sig_h_poly`	✅	3s	4s	-25%
`poly_chknorm_native`	✅	3s	2s	+50%
`poly_ntt_c`	✅	3s	3s	+0%
`poly_use_hint`	✅	3s	1s	+200%
`polyeta_pack`	✅	3s	3s	+0%
`polyt1_pack`	✅	3s	7s	-57%
`polyveck_chknorm`	✅	3s	4s	-25%
`polyveck_pack_eta`	✅	3s	3s	+0%
`polyveck_pack_t0`	✅	3s	3s	+0%
`polyveck_pack_w1`	✅	3s	6s	-50%
`polyveck_unpack_t0`	✅	3s	2s	+50%
`polyvecl_pack_eta`	✅	3s	3s	+0%
`polyvecl_uniform_gamma1_serial`	✅	3s	2s	+50%
`polyvecl_unpack_eta`	✅	3s	5s	-40%
`polyvecl_unpack_z`	✅	3s	5s	-40%
`polyz_pack`	✅	3s	7s	-57%
`power2round`	✅	3s	2s	+50%
`shake128_absorb`	✅	3s	3s	+0%
`shake128_release`	✅	3s	1s	+200%
`shake256_init`	✅	3s	4s	-25%
`shake256x4_absorb_once`	✅	3s	2s	+50%
`shake256x4_squeezeblocks`	✅	3s	4s	-25%
`sign_keypair`	✅	3s	2s	+50%
`sign_signature_pre_hash_internal`	✅	3s	4s	-25%
`sys_check_capability`	✅	3s	6s	-50%
`unpack_pk`	✅	3s	4s	-25%
`use_hint`	✅	3s	3s	+0%
`keccak_f1600_x1_native_aarch64_v84a`	✅	2s	2s	+0%
`keccak_f1600_x4_native_aarch64_v8a_scalar_hybrid`	✅	2s	2s	+0%
`keccak_f1600_x4_native_aarch64_v8a_v84a_scalar_hybrid`	✅	2s	2s	+0%
`keccak_init`	✅	2s	2s	+0%
`keccakf1600x4_permute`	✅	2s	2s	+0%
`mld_ct_abs_i32`	✅	2s	1s	+100%
`mld_ct_cmask_neg_i32`	✅	2s	3s	-33%
`mld_ct_cmask_nonzero_u8`	✅	2s	4s	-50%
`mld_ct_get_optblocker_i64`	✅	2s	2s	+0%
`mld_value_barrier_u32`	✅	2s	2s	+0%
`mld_value_barrier_u8`	✅	2s	2s	+0%
`montgomery_reduce`	✅	2s	3s	-33%
`pack_sig_z`	✅	2s	1s	+100%
`pack_sk_rho_key_tr_s2_t0`	✅	2s	-	new
`pack_sk_s1`	✅	2s	-	new
`pointwise_native_x86_64`	✅	2s	2s	+0%
`poly_caddq`	✅	2s	4s	-50%
`poly_caddq_native`	✅	2s	2s	+0%
`poly_caddq_native_aarch64`	✅	2s	4s	-50%
`poly_chknorm`	✅	2s	4s	-50%
`poly_chknorm_native_aarch64`	✅	2s	2s	+0%
`poly_decompose`	✅	2s	3s	-33%
`poly_invntt_tomont_native`	✅	2s	5s	-60%
`poly_make_hint`	✅	2s	5s	-60%
`poly_pointwise_montgomery`	✅	2s	1s	+100%
`poly_pointwise_montgomery_native`	✅	2s	3s	-33%
`poly_sub`	✅	2s	2s	+0%
`polyt0_pack`	✅	2s	4s	-50%
`polyvecl_chknorm`	✅	2s	3s	-33%
`polyvecl_pointwise_acc_montgomery`	✅	2s	4s	-50%
`polyw1_pack`	✅	2s	3s	-33%
`reduce32`	✅	2s	1s	+100%
`shake128_squeeze`	✅	2s	2s	+0%
`shake128x4_absorb_once`	✅	2s	1s	+100%
`shake256`	✅	2s	4s	-50%
`shake256_absorb`	✅	2s	4s	-50%
`shake256_release`	✅	2s	1s	+100%
`shake256_squeeze`	✅	2s	2s	+0%
`keccakf1600_extract_bytes (big endian)`	✅	1s	2s	-50%
`mld_ct_get_optblocker_u8`	✅	1s	2s	-50%
`polyz_unpack_native`	✅	1s	3s	-67%
`rej_eta`	✅	1s	3s	-67%
`shake128_finalize`	✅	1s	1s	+0%

oqs-bot · 2026-04-02T04:28:24Z

CBMC Results (ML-DSA-87)

⚠️ Attention Required

Proof	Status	Current	Previous	Change
`TOTAL`	⚠️	4197s	3002s	+39.8%
`polyveck_decompose`	⚠️	22s	13s	+69%
`polyvecl_pointwise_acc_montgomery_c`	⚠️	1751s	986s	+78%
`sign_keypair_internal`	⚠️	51s	6s	+750%
`sign_pk_from_sk`	⚠️	27s	8s	+238%

Full Results (186 proofs)

Proof	Status	Current	Previous	Change
`TOTAL`	⚠️	4197s	3002s	+39.8%
`polyvecl_pointwise_acc_montgomery_c`	⚠️	1751s	986s	+78%
`sign_verify_internal`	✅	263s	235s	+12%
`poly_pointwise_montgomery_c`	✅	213s	153s	+39%
`polyvec_matrix_expand`	✅	205s	170s	+21%
`rej_uniform_native`	✅	162s	138s	+17%
`polyvec_matrix_expand_serial`	✅	141s	121s	+17%
`mld_attempt_signature_generation`	✅	99s	73s	+36%
`mld_ct_memcmp`	✅	96s	74s	+30%
`mld_invntt_layer`	✅	76s	63s	+21%
`mld_ntt_layer`	✅	60s	50s	+20%
`sign_keypair_internal`	⚠️	51s	6s	+750%
`sign_signature_internal`	✅	40s	37s	+8%
`polyveck_invntt_tomont`	✅	31s	29s	+7%
`polymat_permute_bitrev_to_custom`	✅	30s	26s	+15%
`sign_pk_from_sk`	⚠️	27s	8s	+238%
`rej_uniform`	✅	26s	21s	+24%
`fqmul`	✅	23s	19s	+21%
`poly_chknorm_c`	✅	23s	18s	+28%
`poly_uniform_eta_4x`	✅	22s	16s	+38%
`polyveck_decompose`	⚠️	22s	13s	+69%
`poly_uniform_4x`	✅	17s	13s	+31%
`polyeta_unpack`	✅	17s	14s	+21%
`keccakf1600x4_permute_native`	✅	16s	13s	+23%
`rej_uniform_c`	✅	16s	12s	+33%
`polyt0_unpack`	✅	15s	12s	+25%
`mld_ntt_butterfly_block`	✅	12s	10s	+20%
`poly_add`	✅	12s	9s	+33%
`polyveck_use_hint`	✅	12s	7s	+71%
`keccak_absorb`	✅	11s	9s	+22%
`keccak_absorb_once_x4`	✅	11s	9s	+22%
`mld_polyvecl_permute_bitrev_to_custom_native`	✅	11s	9s	+22%
`polyvec_matrix_pointwise_montgomery`	✅	11s	8s	+38%
`polyveck_add`	✅	11s	10s	+10%
`polyz_unpack_c`	✅	11s	8s	+38%
`mld_check_pct`	✅	10s	7s	+43%
`polyveck_power2round`	✅	10s	10s	+0%
`polyvecl_ntt`	✅	10s	8s	+25%
`rej_eta_native`	✅	10s	6s	+67%
`unpack_sk`	✅	10s	9s	+11%
`keccakf1600_permute`	✅	9s	11s	-18%
`keccakf1600_permute_native`	✅	9s	8s	+12%
`polyveck_caddq`	✅	9s	10s	-10%
`polyveck_chknorm`	✅	9s	6s	+50%
`polyveck_reduce`	✅	9s	6s	+50%
`keccak_squeezeblocks_x4`	✅	8s	8s	+0%
`pointwise_acc_native_aarch64`	✅	8s	8s	+0%
`poly_caddq_c`	✅	8s	8s	+0%
`poly_decompose_c`	✅	8s	5s	+60%
`poly_power2round`	✅	8s	3s	+167%
`polyveck_ntt`	✅	8s	7s	+14%
`polyvecl_chknorm`	✅	8s	7s	+14%
`sign`	✅	8s	7s	+14%
`mld_compute_pack_z`	✅	7s	8s	-12%
`mld_sample_s1_s2_serial`	✅	7s	4s	+75%
`pointwise_acc_native_x86_64`	✅	7s	8s	-12%
`polyveck_pointwise_poly_montgomery`	✅	7s	6s	+17%
`polyveck_sub`	✅	7s	5s	+40%
`polyveck_unpack_eta`	✅	7s	7s	+0%
`unpack_hints`	✅	7s	6s	+17%
`mld_h`	✅	6s	5s	+20%
`poly_chknorm_native_aarch64`	✅	6s	4s	+50%
`poly_invntt_tomont_c`	✅	6s	9s	-33%
`poly_invntt_tomont_native`	✅	6s	2s	+200%
`polyveck_shiftl`	✅	6s	6s	+0%
`polyvecl_permute_bitrev_to_custom`	✅	6s	4s	+50%
`polyvecl_pointwise_acc_montgomery`	✅	6s	3s	+100%
`sign_signature_pre_hash_internal`	✅	6s	4s	+50%
`keccakf1600_extract_bytes (big endian)`	✅	5s	3s	+67%
`mld_ct_get_optblocker_i64`	✅	5s	4s	+25%
`mld_prepare_domain_separation_prefix`	✅	5s	2s	+150%
`mld_sample_s1_s2`	✅	5s	8s	-38%
`montgomery_reduce`	✅	5s	4s	+25%
`poly_challenge`	✅	5s	3s	+67%
`poly_chknorm_native`	✅	5s	2s	+150%
`poly_uniform`	✅	5s	4s	+25%
`poly_uniform_eta`	✅	5s	3s	+67%
`poly_uniform_gamma1_4x`	✅	5s	2s	+150%
`poly_use_hint`	✅	5s	1s	+400%
`poly_use_hint_native`	✅	5s	4s	+25%
`polyveck_pack_t0`	✅	5s	2s	+150%
`polyvecl_uniform_gamma1_serial`	✅	5s	3s	+67%
`rej_eta_c`	✅	5s	4s	+25%
`shake256`	✅	5s	1s	+400%
`shake256_init`	✅	5s	2s	+150%
`shake256x4_squeezeblocks`	✅	5s	2s	+150%
`sign_keypair`	✅	5s	5s	+0%
`sign_signature`	✅	5s	6s	-17%
`sign_signature_extmu`	✅	5s	5s	+0%
`sign_signature_pre_hash_shake256`	✅	5s	6s	-17%
`sign_verify_extmu`	✅	5s	4s	+25%
`sign_verify_pre_hash_shake256`	✅	5s	6s	-17%
`decompose`	✅	4s	4s	+0%
`intt_native_x86_64`	✅	4s	2s	+100%
`mld_ct_abs_i32`	✅	4s	3s	+33%
`pack_pk`	✅	4s	2s	+100%
`pack_sig_h_poly`	✅	4s	2s	+100%
`poly_chknorm`	✅	4s	4s	+0%
`poly_ntt`	✅	4s	4s	+0%
`poly_ntt_native`	✅	4s	4s	+0%
`poly_pointwise_montgomery_native`	✅	4s	4s	+0%
`poly_shiftl`	✅	4s	3s	+33%
`poly_uniform_gamma1`	✅	4s	5s	-20%
`poly_use_hint_c`	✅	4s	2s	+100%
`polyt1_unpack`	✅	4s	4s	+0%
`polyveck_pack_eta`	✅	4s	3s	+33%
`polyvecl_unpack_eta`	✅	4s	5s	-20%
`polyvecl_unpack_z`	✅	4s	4s	+0%
`polyz_pack`	✅	4s	3s	+33%
`polyz_unpack`	✅	4s	2s	+100%
`polyz_unpack_native`	✅	4s	2s	+100%
`shake256_finalize`	✅	4s	3s	+33%
`sign_open`	✅	4s	6s	-33%
`unpack_pk`	✅	4s	3s	+33%
`unpack_sig`	✅	4s	5s	-20%
`fqscale`	✅	3s	3s	+0%
`keccak_f1600_x4_native_aarch64_v84a`	✅	3s	3s	+0%
`keccak_f1600_x4_native_aarch64_v8a_v84a_scalar_hybrid`	✅	3s	2s	+50%
`keccak_finalize`	✅	3s	3s	+0%
`keccak_squeeze`	✅	3s	3s	+0%
`keccakf1600_xor_bytes`	✅	3s	2s	+50%
`keccakf1600x4_xor_bytes`	✅	3s	2s	+50%
`make_hint`	✅	3s	2s	+50%
`mld_ct_cmask_neg_i32`	✅	3s	2s	+50%
`mld_ct_cmask_nonzero_u8`	✅	3s	3s	+0%
`mld_ct_get_optblocker_u32`	✅	3s	2s	+50%
`mld_keccakf1600_extract_bytes`	✅	3s	3s	+0%
`mld_value_barrier_i64`	✅	3s	3s	+0%
`mld_value_barrier_u32`	✅	3s	5s	-40%
`mld_value_barrier_u8`	✅	3s	1s	+200%
`ntt_native_x86_64`	✅	3s	2s	+50%
`pack_sig_c`	✅	3s	5s	-40%
`pack_sig_z`	✅	3s	3s	+0%
`pack_sk_rho_key_tr_s2_t0`	✅	3s	-	new
`pack_sk_s1`	✅	3s	-	new
`pointwise_native_aarch64`	✅	3s	5s	-40%
`pointwise_native_x86_64`	✅	3s	3s	+0%
`poly_caddq_native`	✅	3s	3s	+0%
`poly_decompose`	✅	3s	3s	+0%
`poly_decompose_native`	✅	3s	5s	-40%
`poly_invntt_tomont`	✅	3s	3s	+0%
`poly_ntt_c`	✅	3s	3s	+0%
`poly_pointwise_montgomery`	✅	3s	5s	-40%
`poly_reduce`	✅	3s	2s	+50%
`polyeta_pack`	✅	3s	3s	+0%
`polyt0_pack`	✅	3s	2s	+50%
`polyt1_pack`	✅	3s	2s	+50%
`polyveck_pack_w1`	✅	3s	2s	+50%
`polyveck_unpack_t0`	✅	3s	3s	+0%
`polyvecl_pack_eta`	✅	3s	4s	-25%
`polyvecl_pointwise_acc_montgomery_native`	✅	3s	3s	+0%
`polyvecl_uniform_gamma1`	✅	3s	3s	+0%
`polyw1_pack`	✅	3s	3s	+0%
`power2round`	✅	3s	3s	+0%
`rej_eta`	✅	3s	3s	+0%
`shake128_finalize`	✅	3s	5s	-40%
`shake128_init`	✅	3s	3s	+0%
`shake128_squeeze`	✅	3s	3s	+0%
`shake256_release`	✅	3s	2s	+50%
`shake256x4_absorb_once`	✅	3s	3s	+0%
`sign_verify_pre_hash_internal`	✅	3s	2s	+50%
`caddq`	✅	2s	3s	-33%
`keccak_f1600_x1_native_aarch64`	✅	2s	2s	+0%
`keccak_f1600_x1_native_aarch64_v84a`	✅	2s	2s	+0%
`keccak_f1600_x4_native_aarch64_v8a_scalar_hybrid`	✅	2s	1s	+100%
`keccakf1600_xor_bytes (big endian)`	✅	2s	2s	+0%
`keccakf1600x4_extract_bytes`	✅	2s	3s	-33%
`mld_ct_cmask_nonzero_u32`	✅	2s	3s	-33%
`mld_ct_get_optblocker_u8`	✅	2s	2s	+0%
`ntt_native_aarch64`	✅	2s	2s	+0%
`poly_caddq_native_aarch64`	✅	2s	3s	-33%
`poly_make_hint`	✅	2s	4s	-50%
`reduce32`	✅	2s	1s	+100%
`shake128_absorb`	✅	2s	2s	+0%
`shake128_release`	✅	2s	2s	+0%
`shake128x4_absorb_once`	✅	2s	3s	-33%
`shake128x4_squeezeblocks`	✅	2s	2s	+0%
`shake256_absorb`	✅	2s	2s	+0%
`shake256_squeeze`	✅	2s	3s	-33%
`sign_verify`	✅	2s	4s	-50%
`sys_check_capability`	✅	2s	2s	+0%
`use_hint`	✅	2s	2s	+0%
`keccak_init`	✅	1s	3s	-67%
`keccakf1600x4_permute`	✅	1s	1s	+0%
`mld_ct_sel_int32`	✅	1s	2s	-50%
`poly_caddq`	✅	1s	5s	-80%
`poly_sub`	✅	1s	3s	-67%

oqs-bot · 2026-04-02T04:30:11Z

CBMC Results (ML-DSA-65)

⚠️ Attention Required

Proof	Status	Current	Previous	Change
`sign_keypair_internal`	⚠️	29s	6s	+383%

Full Results (186 proofs)

Proof	Status	Current	Previous	Change
`TOTAL`	✅	2940s	2556s	+15.0%
`polyvecl_pointwise_acc_montgomery_c`	✅	860s	655s	+31%
`sign_verify_internal`	✅	244s	222s	+10%
`poly_pointwise_montgomery_c`	✅	201s	161s	+25%
`rej_uniform_native`	✅	162s	146s	+11%
`mld_ct_memcmp`	✅	101s	80s	+26%
`polyvec_matrix_expand`	✅	95s	87s	+9%
`mld_invntt_layer`	✅	75s	67s	+12%
`mld_attempt_signature_generation`	✅	69s	67s	+3%
`mld_ntt_layer`	✅	59s	56s	+5%
`polyvec_matrix_expand_serial`	✅	55s	51s	+8%
`polymat_permute_bitrev_to_custom`	✅	40s	37s	+8%
`sign_keypair_internal`	⚠️	29s	6s	+383%
`sign_signature_internal`	✅	27s	27s	+0%
`fqmul`	✅	24s	22s	+9%
`poly_chknorm_c`	✅	24s	24s	+0%
`rej_uniform`	✅	24s	23s	+4%
`polyveck_power2round`	✅	19s	14s	+36%
`sign_pk_from_sk`	✅	19s	8s	+138%
`poly_uniform_4x`	✅	18s	17s	+6%
`poly_uniform_eta_4x`	✅	16s	16s	+0%
`rej_uniform_c`	✅	16s	14s	+14%
`keccakf1600x4_permute_native`	✅	15s	15s	+0%
`poly_add`	✅	15s	11s	+36%
`polyt0_unpack`	✅	15s	14s	+7%
`polyvec_matrix_pointwise_montgomery`	✅	14s	13s	+8%
`polyveck_add`	✅	14s	10s	+40%
`mld_ntt_butterfly_block`	✅	13s	13s	+0%
`polyveck_decompose`	✅	13s	18s	-28%
`keccak_absorb_once_x4`	✅	11s	11s	+0%
`mld_sample_s1_s2`	✅	11s	5s	+120%
`mld_check_pct`	✅	10s	15s	-33%
`polyveck_caddq`	✅	10s	10s	+0%
`polyveck_pointwise_poly_montgomery`	✅	10s	5s	+100%
`sign`	✅	10s	9s	+11%
`unpack_sk`	✅	10s	8s	+25%
`keccakf1600_permute_native`	✅	9s	8s	+12%
`mld_compute_pack_z`	✅	9s	9s	+0%
`polyveck_sub`	✅	9s	9s	+0%
`keccak_absorb`	✅	8s	6s	+33%
`keccakf1600_permute`	✅	8s	8s	+0%
`pointwise_acc_native_x86_64`	✅	8s	7s	+14%
`poly_invntt_tomont_c`	✅	8s	5s	+60%
`polyveck_ntt`	✅	8s	7s	+14%
`polyveck_reduce`	✅	8s	6s	+33%
`polyveck_shiftl`	✅	8s	6s	+33%
`polyvecl_ntt`	✅	8s	10s	-20%
`mld_polyvecl_permute_bitrev_to_custom_native`	✅	7s	7s	+0%
`poly_decompose_c`	✅	7s	6s	+17%
`poly_power2round`	✅	7s	6s	+17%
`sign_signature_pre_hash_shake256`	✅	7s	4s	+75%
`keccak_f1600_x1_native_aarch64`	✅	6s	4s	+50%
`keccak_squeezeblocks_x4`	✅	6s	6s	+0%
`mld_h`	✅	6s	5s	+20%
`mld_prepare_domain_separation_prefix`	✅	6s	5s	+20%
`pointwise_acc_native_aarch64`	✅	6s	10s	-40%
`poly_uniform_eta`	✅	6s	5s	+20%
`poly_use_hint_c`	✅	6s	6s	+0%
`polyveck_invntt_tomont`	✅	6s	5s	+20%
`polyveck_pack_t0`	✅	6s	2s	+200%
`polyz_unpack_c`	✅	6s	2s	+200%
`rej_eta_native`	✅	6s	3s	+100%
`sign_keypair`	✅	6s	7s	-14%
`sign_open`	✅	6s	7s	-14%
`sign_verify`	✅	6s	3s	+100%
`unpack_hints`	✅	6s	9s	-33%
`keccak_f1600_x4_native_aarch64_v84a`	✅	5s	3s	+67%
`montgomery_reduce`	✅	5s	3s	+67%
`poly_caddq_c`	✅	5s	4s	+25%
`poly_invntt_tomont_native`	✅	5s	3s	+67%
`poly_reduce`	✅	5s	3s	+67%
`poly_shiftl`	✅	5s	5s	+0%
`poly_uniform`	✅	5s	7s	-29%
`polyeta_unpack`	✅	5s	5s	+0%
`polyveck_use_hint`	✅	5s	10s	-50%
`polyvecl_pointwise_acc_montgomery_native`	✅	5s	6s	-17%
`rej_eta`	✅	5s	2s	+150%
`shake256_finalize`	✅	5s	2s	+150%
`shake256_init`	✅	5s	3s	+67%
`sign_signature_pre_hash_internal`	✅	5s	4s	+25%
`fqscale`	✅	4s	1s	+300%
`intt_native_x86_64`	✅	4s	3s	+33%
`keccak_finalize`	✅	4s	4s	+0%
`make_hint`	✅	4s	3s	+33%
`mld_ct_cmask_nonzero_u8`	✅	4s	3s	+33%
`mld_keccakf1600_extract_bytes`	✅	4s	4s	+0%
`mld_sample_s1_s2_serial`	✅	4s	7s	-43%
`pack_sk_s1`	✅	4s	-	new
`pointwise_native_x86_64`	✅	4s	4s	+0%
`poly_chknorm_native_aarch64`	✅	4s	2s	+100%
`poly_decompose`	✅	4s	3s	+33%
`poly_invntt_tomont`	✅	4s	3s	+33%
`poly_pointwise_montgomery`	✅	4s	3s	+33%
`poly_uniform_gamma1`	✅	4s	3s	+33%
`poly_use_hint`	✅	4s	3s	+33%
`polyt0_pack`	✅	4s	4s	+0%
`polyt1_pack`	✅	4s	3s	+33%
`polyveck_pack_w1`	✅	4s	3s	+33%
`polyveck_unpack_t0`	✅	4s	4s	+0%
`polyvecl_uniform_gamma1`	✅	4s	4s	+0%
`polyvecl_unpack_z`	✅	4s	3s	+33%
`polyz_pack`	✅	4s	4s	+0%
`rej_eta_c`	✅	4s	2s	+100%
`shake256_absorb`	✅	4s	2s	+100%
`sign_signature_extmu`	✅	4s	5s	-20%
`sign_verify_pre_hash_shake256`	✅	4s	6s	-33%
`sys_check_capability`	✅	4s	2s	+100%
`caddq`	✅	3s	4s	-25%
`keccak_f1600_x1_native_aarch64_v84a`	✅	3s	2s	+50%
`keccak_f1600_x4_native_aarch64_v8a_scalar_hybrid`	✅	3s	1s	+200%
`keccak_squeeze`	✅	3s	3s	+0%
`keccakf1600_xor_bytes`	✅	3s	2s	+50%
`keccakf1600x4_extract_bytes`	✅	3s	2s	+50%
`mld_ct_cmask_neg_i32`	✅	3s	2s	+50%
`mld_ct_get_optblocker_i64`	✅	3s	1s	+200%
`mld_ct_get_optblocker_u32`	✅	3s	3s	+0%
`mld_ct_sel_int32`	✅	3s	1s	+200%
`ntt_native_aarch64`	✅	3s	2s	+50%
`ntt_native_x86_64`	✅	3s	4s	-25%
`pack_pk`	✅	3s	4s	-25%
`pack_sig_z`	✅	3s	1s	+200%
`pack_sk_rho_key_tr_s2_t0`	✅	3s	-	new
`poly_caddq`	✅	3s	2s	+50%
`poly_caddq_native_aarch64`	✅	3s	4s	-25%
`poly_chknorm`	✅	3s	4s	-25%
`poly_chknorm_native`	✅	3s	2s	+50%
`poly_ntt`	✅	3s	2s	+50%
`poly_ntt_c`	✅	3s	3s	+0%
`poly_pointwise_montgomery_native`	✅	3s	5s	-40%
`poly_sub`	✅	3s	4s	-25%
`poly_use_hint_native`	✅	3s	2s	+50%
`polyeta_pack`	✅	3s	3s	+0%
`polyt1_unpack`	✅	3s	4s	-25%
`polyveck_chknorm`	✅	3s	8s	-62%
`polyveck_pack_eta`	✅	3s	3s	+0%
`polyveck_unpack_eta`	✅	3s	3s	+0%
`polyvecl_chknorm`	✅	3s	7s	-57%
`polyvecl_pack_eta`	✅	3s	2s	+50%
`polyvecl_permute_bitrev_to_custom`	✅	3s	3s	+0%
`polyvecl_uniform_gamma1_serial`	✅	3s	5s	-40%
`polyvecl_unpack_eta`	✅	3s	2s	+50%
`polyw1_pack`	✅	3s	3s	+0%
`polyz_unpack`	✅	3s	2s	+50%
`power2round`	✅	3s	2s	+50%
`reduce32`	✅	3s	1s	+200%
`shake128_absorb`	✅	3s	2s	+50%
`shake128_squeeze`	✅	3s	2s	+50%
`shake128x4_squeezeblocks`	✅	3s	2s	+50%
`shake256`	✅	3s	3s	+0%
`shake256_squeeze`	✅	3s	2s	+50%
`sign_signature`	✅	3s	5s	-40%
`sign_verify_pre_hash_internal`	✅	3s	5s	-40%
`unpack_pk`	✅	3s	2s	+50%
`unpack_sig`	✅	3s	2s	+50%
`decompose`	✅	2s	3s	-33%
`keccak_f1600_x4_native_aarch64_v8a_v84a_scalar_hybrid`	✅	2s	3s	-33%
`keccak_init`	✅	2s	2s	+0%
`keccakf1600_extract_bytes (big endian)`	✅	2s	3s	-33%
`keccakf1600_xor_bytes (big endian)`	✅	2s	3s	-33%
`keccakf1600x4_permute`	✅	2s	3s	-33%
`keccakf1600x4_xor_bytes`	✅	2s	3s	-33%
`mld_ct_abs_i32`	✅	2s	2s	+0%
`mld_ct_cmask_nonzero_u32`	✅	2s	3s	-33%
`mld_ct_get_optblocker_u8`	✅	2s	1s	+100%
`mld_value_barrier_u32`	✅	2s	3s	-33%
`mld_value_barrier_u8`	✅	2s	5s	-60%
`pack_sig_c`	✅	2s	3s	-33%
`pack_sig_h_poly`	✅	2s	4s	-50%
`pointwise_native_aarch64`	✅	2s	2s	+0%
`poly_caddq_native`	✅	2s	4s	-50%
`poly_challenge`	✅	2s	3s	-33%
`poly_decompose_native`	✅	2s	4s	-50%
`poly_make_hint`	✅	2s	5s	-60%
`poly_ntt_native`	✅	2s	4s	-50%
`poly_uniform_gamma1_4x`	✅	2s	4s	-50%
`polyvecl_pointwise_acc_montgomery`	✅	2s	4s	-50%
`polyz_unpack_native`	✅	2s	6s	-67%
`shake128_finalize`	✅	2s	3s	-33%
`shake128_init`	✅	2s	1s	+100%
`shake128_release`	✅	2s	3s	-33%
`shake128x4_absorb_once`	✅	2s	4s	-50%
`shake256_release`	✅	2s	1s	+100%
`shake256x4_absorb_once`	✅	2s	3s	-33%
`sign_verify_extmu`	✅	2s	3s	-33%
`use_hint`	✅	2s	2s	+0%
`mld_value_barrier_i64`	✅	1s	2s	-50%
`shake256x4_squeezeblocks`	✅	1s	3s	-67%

hanno-becker

Similar to what we did with poly_add, I think we should make the now-destructive nature of power2round explicit in the signature and documentation.

We'd also need to adjust the CBMC spec for compute_t0_t1_tr_from_sk_components to account for s1hat and t1 no longer being disjoint in memory. Can we do this already now (even if the call-site still uses a struct)?

Reuse t0 as the accumulator in mld_compute_t0_t1_tr_from_sk_components, and have the caller provide s1 already in NTT form, removing two allocations (s1hat and t) from the helper. In mld_sign_keypair_internal, share the s1 and t1 buffers via a union since s1hat is consumed before t1 is produced. Pack s1 into the secret key before the in-place NTT so the original coefficients are preserved. Split mld_pack_sk into mld_pack_sk_s1 and mld_pack_sk_rho_key_tr_s2_t0 to support packing s1 independently before the NTT. Aliasing s1 and t1 makes mld_compute_t0_t1_tr_from_sk_components's memory_no_alias contract on both arguments incompatible with the keygen call site. As a workaround, drop the contract and inline it into the proofs of both call sites. This will go away in a follow-up PR that eliminates mld_compute_t0_t1_tr_from_sk_components altogether (#1030). Signed-off-by: Matthias J. Kannwischer <matthias@zerorisc.com>

mkannwischer force-pushed the keygen-buffer-sharing branch from 0b96d02 to 0db7b80 Compare April 2, 2026 04:00

mkannwischer marked this pull request as ready for review April 2, 2026 04:31

mkannwischer requested a review from a team as a code owner April 2, 2026 04:31

mkannwischer assigned hanno-becker Apr 2, 2026

hanno-becker added the low-ram label Apr 3, 2026

mkannwischer force-pushed the keygen-buffer-sharing branch 3 times, most recently from 93856fc to 05fc00a Compare April 8, 2026 07:08

mkannwischer mentioned this pull request Apr 8, 2026

lowram: Per-row t0/t1 computation in keygen #1030

Open

mkannwischer force-pushed the keygen-buffer-sharing branch from 05fc00a to 035e9aa Compare April 8, 2026 09:17

hanno-becker reviewed Apr 14, 2026

View reviewed changes

Comment thread mldsa/src/poly.h

hanno-becker reviewed Apr 14, 2026

View reviewed changes

Comment thread mldsa/src/polyvec.h

hanno-becker reviewed Apr 14, 2026

View reviewed changes

Comment thread mldsa/src/sign.c Outdated

hanno-becker requested changes Apr 14, 2026

View reviewed changes

hanno-becker mentioned this pull request Apr 14, 2026

CBMC: Restore polyvecl_pointwise_acc_montgomery_c proof flags #1039

Merged

mkannwischer force-pushed the keygen-buffer-sharing branch 5 times, most recently from 4f8c24c to b839b38 Compare April 17, 2026 10:45

hanno-becker reviewed Apr 17, 2026

View reviewed changes

Comment thread mldsa/src/sign.c Outdated

mkannwischer force-pushed the keygen-buffer-sharing branch from b839b38 to ea474d8 Compare April 17, 2026 10:59

hanno-becker reviewed Apr 17, 2026

View reviewed changes

Comment thread mldsa/src/sign.c Outdated

mkannwischer force-pushed the keygen-buffer-sharing branch from ea474d8 to b0a4b1a Compare April 17, 2026 13:41

mkannwischer force-pushed the keygen-buffer-sharing branch from b0a4b1a to c9443ad Compare April 17, 2026 13:44

hanno-becker approved these changes Apr 17, 2026

View reviewed changes

mkannwischer merged commit b21396d into main Apr 17, 2026
403 checks passed

mkannwischer deleted the keygen-buffer-sharing branch April 17, 2026 14:28

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Lowram: Share buffers with non-overlapping lifetimes in keygen#1011

Lowram: Share buffers with non-overlapping lifetimes in keygen#1011
mkannwischer merged 1 commit intomainfrom
keygen-buffer-sharing

mkannwischer commented Apr 2, 2026

Uh oh!

oqs-bot commented Apr 2, 2026 •

edited

Loading

Uh oh!

oqs-bot commented Apr 2, 2026 •

edited

Loading

Uh oh!

oqs-bot commented Apr 2, 2026 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

Uh oh!

hanno-becker left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

mkannwischer commented Apr 2, 2026

Uh oh!

oqs-bot commented Apr 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

CBMC Results (ML-DSA-44)

Uh oh!

oqs-bot commented Apr 2, 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 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

CBMC Results (ML-DSA-65)

Uh oh!

Uh oh!

Uh oh!

Uh oh!

hanno-becker left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

oqs-bot commented Apr 2, 2026 •

edited

Loading

oqs-bot commented Apr 2, 2026 •

edited

Loading

oqs-bot commented Apr 2, 2026 •

edited

Loading