lowram: Compute h incrementally in signing by mkannwischer · Pull Request #1015 · pq-code-package/mldsa-native

mkannwischer · 2026-04-03T11:17:39Z

Instead of allocating a full polyveck for h in attempt_signature_generation,
compute cs2, ct0, and hints one polynomial at a time using scratch polys.

This eliminates the polyveck h from the yh_u union, replacing
mld_pack_sig_c_h with incremental packing via mld_pack_sig_c and
mld_pack_sig_h_poly.
This is a prerequisite for also eliminating y in a follow up.

Depends on Sign: Unpack s1, s2, and t0 on the fly in REDUCE_RAM mode #1002
Hoisted out from PoC: Reduce large struct allocations to <= 13/17/21 KiB for ML-DSA-44/65/87 #1005

oqs-bot · 2026-04-03T11:34:24Z

CBMC Results (ML-DSA-44)

⚠️ Attention Required

Proof	Status	Current	Previous	Change
`polyvec_matrix_expand_serial`	⚠️	24s	16s	+50%

Full Results (177 proofs)

Proof	Status	Current	Previous	Change
`TOTAL`	✅	1916s	2072s	-7.5%
`sign_verify_internal`	✅	219s	184s	+19%
`polyvecl_pointwise_acc_montgomery_c`	✅	171s	216s	-21%
`poly_pointwise_montgomery_c`	✅	160s	155s	+3%
`rej_uniform_native`	✅	148s	145s	+2%
`mld_attempt_signature_generation`	✅	109s	253s	-57%
`mld_invntt_layer`	✅	89s	88s	+1%
`mld_ct_memcmp`	✅	78s	78s	+0%
`mld_ntt_layer`	✅	55s	56s	-2%
`polyvec_matrix_expand_serial`	⚠️	24s	16s	+50%
`poly_chknorm_c`	✅	22s	20s	+10%
`polyvec_matrix_expand`	✅	22s	19s	+16%
`rej_uniform`	✅	22s	21s	+5%
`fqmul`	✅	21s	21s	+0%
`sign_signature_internal`	✅	21s	18s	+17%
`poly_uniform_eta_4x`	✅	20s	18s	+11%
`polyeta_unpack`	✅	18s	16s	+12%
`rej_uniform_c`	✅	17s	16s	+6%
`poly_uniform_4x`	✅	16s	17s	-6%
`polyt0_unpack`	✅	14s	14s	+0%
`keccakf1600x4_permute_native`	✅	13s	14s	-7%
`mld_ntt_butterfly_block`	✅	13s	12s	+8%
`mld_compute_t0_t1_tr_from_sk_components`	✅	12s	15s	-20%
`polyz_unpack_c`	✅	12s	12s	+0%
`keccak_absorb_once_x4`	✅	11s	12s	-8%
`poly_add`	✅	11s	12s	-8%
`mld_check_pct`	✅	10s	6s	+67%
`polymat_permute_bitrev_to_custom`	✅	9s	17s	-47%
`keccakf1600_permute`	✅	8s	8s	+0%
`polyveck_caddq`	✅	8s	6s	+33%
`sign`	✅	8s	11s	-27%
`sign_keypair_internal`	✅	8s	6s	+33%
`unpack_sk`	✅	8s	8s	+0%
`keccak_absorb`	✅	7s	7s	+0%
`keccak_squeezeblocks_x4`	✅	7s	7s	+0%
`keccakf1600_permute_native`	✅	7s	8s	-12%
`mld_compute_pack_z`	✅	7s	4s	+75%
`mld_polyvecl_permute_bitrev_to_custom_native`	✅	7s	10s	-30%
`mld_sample_s1_s2_serial`	✅	7s	5s	+40%
`poly_caddq_c`	✅	7s	4s	+75%
`poly_use_hint_native`	✅	7s	2s	+250%
`polyveck_use_hint`	✅	7s	7s	+0%
`sign_open`	✅	7s	4s	+75%
`poly_challenge`	✅	6s	5s	+20%
`poly_invntt_tomont_c`	✅	6s	6s	+0%
`poly_power2round`	✅	6s	4s	+50%
`poly_shiftl`	✅	6s	3s	+100%
`polyvec_matrix_pointwise_montgomery`	✅	6s	12s	-50%
`polyveck_power2round`	✅	6s	13s	-54%
`polyvecl_chknorm`	✅	6s	3s	+100%
`polyz_pack`	✅	6s	4s	+50%
`sign_verify`	✅	6s	6s	+0%
`sign_verify_extmu`	✅	6s	6s	+0%
`decompose`	✅	5s	2s	+150%
`mld_h`	✅	5s	2s	+150%
`pack_sk`	✅	5s	4s	+25%
`poly_uniform`	✅	5s	4s	+25%
`poly_use_hint`	✅	5s	2s	+150%
`poly_use_hint_c`	✅	5s	3s	+67%
`polyt0_pack`	✅	5s	5s	+0%
`polyveck_add`	✅	5s	9s	-44%
`polyveck_decompose`	✅	5s	4s	+25%
`polyveck_invntt_tomont`	✅	5s	4s	+25%
`polyveck_pointwise_poly_montgomery`	✅	5s	5s	+0%
`polyveck_shiftl`	✅	5s	6s	-17%
`polyveck_unpack_t0`	✅	5s	4s	+25%
`polyvecl_pointwise_acc_montgomery_native`	✅	5s	4s	+25%
`polyz_unpack_native`	✅	5s	3s	+67%
`rej_eta_c`	✅	5s	3s	+67%
`shake128_absorb`	✅	5s	1s	+400%
`sign_pk_from_sk`	✅	5s	5s	+0%
`sign_signature_pre_hash_shake256`	✅	5s	7s	-29%
`sys_check_capability`	✅	5s	3s	+67%
`unpack_sig`	✅	5s	2s	+150%
`keccak_squeeze`	✅	4s	3s	+33%
`mld_prepare_domain_separation_prefix`	✅	4s	2s	+100%
`mld_sample_s1_s2`	✅	4s	7s	-43%
`ntt_native_x86_64`	✅	4s	3s	+33%
`poly_caddq_native`	✅	4s	6s	-33%
`poly_chknorm_native`	✅	4s	5s	-20%
`poly_decompose`	✅	4s	5s	-20%
`poly_make_hint`	✅	4s	2s	+100%
`poly_ntt`	✅	4s	5s	-20%
`poly_ntt_c`	✅	4s	4s	+0%
`poly_uniform_eta`	✅	4s	4s	+0%
`poly_uniform_gamma1_4x`	✅	4s	5s	-20%
`polyveck_chknorm`	✅	4s	4s	+0%
`polyveck_pack_eta`	✅	4s	5s	-20%
`polyveck_reduce`	✅	4s	5s	-20%
`polyveck_sub`	✅	4s	6s	-33%
`polyvecl_ntt`	✅	4s	5s	-20%
`polyvecl_permute_bitrev_to_custom`	✅	4s	4s	+0%
`polyvecl_uniform_gamma1`	✅	4s	2s	+100%
`polyvecl_unpack_eta`	✅	4s	2s	+100%
`polyvecl_unpack_z`	✅	4s	3s	+33%
`polyw1_pack`	✅	4s	4s	+0%
`reduce32`	✅	4s	3s	+33%
`rej_eta`	✅	4s	3s	+33%
`rej_eta_native`	✅	4s	5s	-20%
`shake128_release`	✅	4s	4s	+0%
`shake128x4_squeezeblocks`	✅	4s	5s	-20%
`sign_keypair`	✅	4s	4s	+0%
`sign_signature_extmu`	✅	4s	4s	+0%
`sign_verify_pre_hash_internal`	✅	4s	5s	-20%
`sign_verify_pre_hash_shake256`	✅	4s	5s	-20%
`unpack_hints`	✅	4s	6s	-33%
`use_hint`	✅	4s	4s	+0%
`fqscale`	✅	3s	2s	+50%
`intt_native_x86_64`	✅	3s	3s	+0%
`keccak_init`	✅	3s	1s	+200%
`keccakf1600_extract_bytes (big endian)`	✅	3s	3s	+0%
`keccakf1600x4_permute`	✅	3s	3s	+0%
`mld_ct_cmask_nonzero_u32`	✅	3s	3s	+0%
`mld_ct_cmask_nonzero_u8`	✅	3s	3s	+0%
`mld_ct_get_optblocker_i64`	✅	3s	5s	-40%
`mld_ct_get_optblocker_u32`	✅	3s	2s	+50%
`mld_value_barrier_i64`	✅	3s	2s	+50%
`ntt_native_aarch64`	✅	3s	4s	-25%
`pack_sig_h_poly`	✅	3s	-	new
`pack_sig_z`	✅	3s	2s	+50%
`poly_caddq`	✅	3s	5s	-40%
`poly_caddq_native_aarch64`	✅	3s	3s	+0%
`poly_chknorm_native_aarch64`	✅	3s	4s	-25%
`poly_invntt_tomont`	✅	3s	2s	+50%
`poly_pointwise_montgomery`	✅	3s	5s	-40%
`poly_sub`	✅	3s	3s	+0%
`poly_uniform_gamma1`	✅	3s	2s	+50%
`polyt1_pack`	✅	3s	3s	+0%
`polyt1_unpack`	✅	3s	3s	+0%
`polyveck_ntt`	✅	3s	5s	-40%
`polyveck_pack_t0`	✅	3s	4s	-25%
`polyvecl_pack_eta`	✅	3s	4s	-25%
`polyvecl_pointwise_acc_montgomery`	✅	3s	2s	+50%
`power2round`	✅	3s	3s	+0%
`shake128_init`	✅	3s	2s	+50%
`shake128_squeeze`	✅	3s	2s	+50%
`shake128x4_absorb_once`	✅	3s	2s	+50%
`shake256`	✅	3s	2s	+50%
`shake256_absorb`	✅	3s	2s	+50%
`shake256_init`	✅	3s	1s	+200%
`shake256_release`	✅	3s	2s	+50%
`sign_signature`	✅	3s	5s	-40%
`sign_signature_pre_hash_internal`	✅	3s	7s	-57%
`unpack_pk`	✅	3s	3s	+0%
`caddq`	✅	2s	5s	-60%
`keccak_finalize`	✅	2s	2s	+0%
`keccakf1600_xor_bytes (big endian)`	✅	2s	3s	-33%
`keccakf1600x4_extract_bytes`	✅	2s	2s	+0%
`keccakf1600x4_xor_bytes`	✅	2s	2s	+0%
`make_hint`	✅	2s	2s	+0%
`mld_ct_abs_i32`	✅	2s	2s	+0%
`mld_ct_cmask_neg_i32`	✅	2s	2s	+0%
`mld_ct_sel_int32`	✅	2s	4s	-50%
`mld_keccakf1600_extract_bytes`	✅	2s	2s	+0%
`mld_value_barrier_u32`	✅	2s	3s	-33%
`mld_value_barrier_u8`	✅	2s	3s	-33%
`montgomery_reduce`	✅	2s	3s	-33%
`pack_pk`	✅	2s	3s	-33%
`poly_chknorm`	✅	2s	3s	-33%
`poly_decompose_c`	✅	2s	2s	+0%
`poly_invntt_tomont_native`	✅	2s	4s	-50%
`poly_ntt_native`	✅	2s	4s	-50%
`poly_pointwise_montgomery_native`	✅	2s	2s	+0%
`poly_reduce`	✅	2s	3s	-33%
`polyeta_pack`	✅	2s	3s	-33%
`polyveck_unpack_eta`	✅	2s	6s	-67%
`polyz_unpack`	✅	2s	3s	-33%
`shake128_finalize`	✅	2s	3s	-33%
`shake256_squeeze`	✅	2s	3s	-33%
`shake256x4_absorb_once`	✅	2s	2s	+0%
`shake256x4_squeezeblocks`	✅	2s	4s	-50%
`keccakf1600_xor_bytes`	✅	1s	1s	+0%
`mld_ct_get_optblocker_u8`	✅	1s	2s	-50%
`pack_sig_c`	✅	1s	-	new
`poly_decompose_native`	✅	1s	4s	-75%
`polyveck_pack_w1`	✅	1s	3s	-67%
`polyvecl_uniform_gamma1_serial`	✅	1s	3s	-67%
`shake256_finalize`	✅	1s	3s	-67%

oqs-bot · 2026-04-03T11:40:20Z

CBMC Results (ML-DSA-65)

⚠️ Attention Required

Proof	Status	Current	Previous	Change
`polymat_permute_bitrev_to_custom`	⚠️	364s	31s	+1074%
`polyveck_use_hint`	⚠️	26s	11s	+136%
`polyvecl_pointwise_acc_montgomery_c`	⚠️	591s	191s	+209%

Full Results (177 proofs)

Proof	Status	Current	Previous	Change
`TOTAL`	✅	2967s	2509s	+18.3%
`polyvecl_pointwise_acc_montgomery_c`	⚠️	591s	191s	+209%
`polymat_permute_bitrev_to_custom`	⚠️	364s	31s	+1074%
`sign_verify_internal`	✅	287s	346s	-17%
`poly_pointwise_montgomery_c`	✅	160s	166s	-4%
`rej_uniform_native`	✅	147s	149s	-1%
`mld_attempt_signature_generation`	✅	109s	285s	-62%
`polyvec_matrix_expand`	✅	102s	126s	-19%
`mld_invntt_layer`	✅	93s	98s	-5%
`polyvec_matrix_expand_serial`	✅	83s	72s	+15%
`mld_ct_memcmp`	✅	74s	80s	-7%
`mld_ntt_layer`	✅	56s	57s	-2%
`mld_compute_t0_t1_tr_from_sk_components`	✅	28s	27s	+4%
`sign_signature_internal`	✅	28s	27s	+4%
`polyveck_use_hint`	⚠️	26s	11s	+136%
`fqmul`	✅	21s	23s	-9%
`poly_chknorm_c`	✅	21s	24s	-12%
`rej_uniform`	✅	19s	20s	-5%
`polyveck_decompose`	✅	17s	13s	+31%
`poly_uniform_eta_4x`	✅	16s	17s	-6%
`rej_uniform_c`	✅	16s	17s	-6%
`keccakf1600x4_permute_native`	✅	15s	13s	+15%
`mld_check_pct`	✅	14s	8s	+75%
`poly_uniform_4x`	✅	14s	16s	-12%
`polyt0_unpack`	✅	14s	13s	+8%
`mld_ntt_butterfly_block`	✅	13s	12s	+8%
`poly_add`	✅	12s	11s	+9%
`polyvec_matrix_pointwise_montgomery`	✅	12s	10s	+20%
`polyveck_power2round`	✅	11s	12s	-8%
`keccak_absorb_once_x4`	✅	9s	11s	-18%
`mld_polyvecl_permute_bitrev_to_custom_native`	✅	9s	8s	+12%
`polyveck_caddq`	✅	9s	8s	+12%
`unpack_sk`	✅	9s	8s	+12%
`mld_sample_s1_s2`	✅	8s	5s	+60%
`polyveck_add`	✅	8s	9s	-11%
`polyveck_reduce`	✅	8s	11s	-27%
`polyvecl_chknorm`	✅	8s	12s	-33%
`keccak_absorb`	✅	7s	11s	-36%
`keccak_squeezeblocks_x4`	✅	7s	7s	+0%
`keccakf1600_permute`	✅	7s	9s	-22%
`keccakf1600_permute_native`	✅	7s	8s	-12%
`poly_use_hint_c`	✅	7s	5s	+40%
`polyeta_unpack`	✅	7s	5s	+40%
`polyveck_ntt`	✅	7s	10s	-30%
`polyveck_pointwise_poly_montgomery`	✅	7s	8s	-12%
`polyveck_shiftl`	✅	7s	6s	+17%
`polyvecl_ntt`	✅	7s	8s	-12%
`sign`	✅	7s	6s	+17%
`sign_keypair_internal`	✅	7s	7s	+0%
`sign_pk_from_sk`	✅	7s	8s	-12%
`sign_verify_pre_hash_internal`	✅	7s	4s	+75%
`poly_caddq_native`	✅	6s	4s	+50%
`poly_decompose_c`	✅	6s	8s	-25%
`poly_invntt_tomont_c`	✅	6s	7s	-14%
`poly_pointwise_montgomery`	✅	6s	4s	+50%
`poly_power2round`	✅	6s	6s	+0%
`poly_use_hint`	✅	6s	2s	+200%
`polyveck_invntt_tomont`	✅	6s	8s	-25%
`polyveck_sub`	✅	6s	13s	-54%
`polyvecl_unpack_z`	✅	6s	5s	+20%
`polyz_unpack_c`	✅	6s	3s	+100%
`shake128_squeeze`	✅	6s	1s	+500%
`sign_open`	✅	6s	2s	+200%
`sign_signature_extmu`	✅	6s	3s	+100%
`keccak_squeeze`	✅	5s	5s	+0%
`make_hint`	✅	5s	4s	+25%
`mld_ct_get_optblocker_u32`	✅	5s	3s	+67%
`mld_sample_s1_s2_serial`	✅	5s	7s	-29%
`poly_invntt_tomont_native`	✅	5s	3s	+67%
`poly_uniform_gamma1`	✅	5s	3s	+67%
`poly_uniform_gamma1_4x`	✅	5s	5s	+0%
`polyveck_chknorm`	✅	5s	3s	+67%
`polyveck_unpack_t0`	✅	5s	3s	+67%
`polyz_unpack`	✅	5s	2s	+150%
`rej_eta_c`	✅	5s	4s	+25%
`shake256x4_absorb_once`	✅	5s	2s	+150%
`sign_signature`	✅	5s	7s	-29%
`sign_signature_pre_hash_internal`	✅	5s	3s	+67%
`sign_verify_extmu`	✅	5s	3s	+67%
`intt_native_x86_64`	✅	4s	4s	+0%
`keccak_init`	✅	4s	2s	+100%
`keccakf1600_extract_bytes (big endian)`	✅	4s	3s	+33%
`keccakf1600_xor_bytes`	✅	4s	3s	+33%
`keccakf1600x4_extract_bytes`	✅	4s	4s	+0%
`mld_compute_pack_z`	✅	4s	8s	-50%
`mld_h`	✅	4s	5s	-20%
`mld_prepare_domain_separation_prefix`	✅	4s	3s	+33%
`montgomery_reduce`	✅	4s	2s	+100%
`ntt_native_aarch64`	✅	4s	2s	+100%
`pack_pk`	✅	4s	2s	+100%
`pack_sig_c`	✅	4s	-	new
`pack_sk`	✅	4s	4s	+0%
`poly_caddq_c`	✅	4s	4s	+0%
`poly_challenge`	✅	4s	5s	-20%
`poly_ntt_c`	✅	4s	6s	-33%
`poly_ntt_native`	✅	4s	3s	+33%
`poly_reduce`	✅	4s	3s	+33%
`poly_uniform_eta`	✅	4s	4s	+0%
`polyt0_pack`	✅	4s	5s	-20%
`polyveck_unpack_eta`	✅	4s	5s	-20%
`polyvecl_pointwise_acc_montgomery_native`	✅	4s	6s	-33%
`polyvecl_uniform_gamma1_serial`	✅	4s	2s	+100%
`polyvecl_unpack_eta`	✅	4s	3s	+33%
`polyz_pack`	✅	4s	3s	+33%
`rej_eta_native`	✅	4s	5s	-20%
`shake128_finalize`	✅	4s	2s	+100%
`shake128_release`	✅	4s	2s	+100%
`sign_signature_pre_hash_shake256`	✅	4s	3s	+33%
`sign_verify`	✅	4s	4s	+0%
`sign_verify_pre_hash_shake256`	✅	4s	5s	-20%
`unpack_hints`	✅	4s	4s	+0%
`unpack_pk`	✅	4s	5s	-20%
`fqscale`	✅	3s	2s	+50%
`keccakf1600x4_permute`	✅	3s	1s	+200%
`mld_ct_cmask_nonzero_u8`	✅	3s	2s	+50%
`ntt_native_x86_64`	✅	3s	5s	-40%
`poly_caddq`	✅	3s	5s	-40%
`poly_chknorm`	✅	3s	3s	+0%
`poly_chknorm_native_aarch64`	✅	3s	4s	-25%
`poly_invntt_tomont`	✅	3s	2s	+50%
`poly_make_hint`	✅	3s	1s	+200%
`poly_pointwise_montgomery_native`	✅	3s	3s	+0%
`poly_shiftl`	✅	3s	2s	+50%
`poly_sub`	✅	3s	2s	+50%
`poly_uniform`	✅	3s	3s	+0%
`poly_use_hint_native`	✅	3s	5s	-40%
`polyt1_pack`	✅	3s	3s	+0%
`polyt1_unpack`	✅	3s	3s	+0%
`polyveck_pack_eta`	✅	3s	2s	+50%
`polyveck_pack_w1`	✅	3s	5s	-40%
`polyvecl_permute_bitrev_to_custom`	✅	3s	3s	+0%
`polyvecl_pointwise_acc_montgomery`	✅	3s	3s	+0%
`polyz_unpack_native`	✅	3s	2s	+50%
`shake256`	✅	3s	3s	+0%
`shake256_squeeze`	✅	3s	3s	+0%
`use_hint`	✅	3s	4s	-25%
`caddq`	✅	2s	5s	-60%
`decompose`	✅	2s	2s	+0%
`keccak_finalize`	✅	2s	2s	+0%
`keccakf1600_xor_bytes (big endian)`	✅	2s	2s	+0%
`keccakf1600x4_xor_bytes`	✅	2s	3s	-33%
`mld_ct_abs_i32`	✅	2s	2s	+0%
`mld_ct_cmask_nonzero_u32`	✅	2s	4s	-50%
`mld_ct_get_optblocker_i64`	✅	2s	4s	-50%
`mld_ct_get_optblocker_u8`	✅	2s	2s	+0%
`mld_ct_sel_int32`	✅	2s	4s	-50%
`mld_keccakf1600_extract_bytes`	✅	2s	5s	-60%
`mld_value_barrier_i64`	✅	2s	1s	+100%
`mld_value_barrier_u32`	✅	2s	2s	+0%
`mld_value_barrier_u8`	✅	2s	1s	+100%
`pack_sig_h_poly`	✅	2s	-	new
`pack_sig_z`	✅	2s	2s	+0%
`poly_caddq_native_aarch64`	✅	2s	4s	-50%
`poly_chknorm_native`	✅	2s	3s	-33%
`poly_decompose`	✅	2s	3s	-33%
`poly_decompose_native`	✅	2s	5s	-60%
`poly_ntt`	✅	2s	2s	+0%
`polyeta_pack`	✅	2s	2s	+0%
`polyveck_pack_t0`	✅	2s	4s	-50%
`polyvecl_pack_eta`	✅	2s	2s	+0%
`polyvecl_uniform_gamma1`	✅	2s	3s	-33%
`polyw1_pack`	✅	2s	3s	-33%
`power2round`	✅	2s	3s	-33%
`reduce32`	✅	2s	3s	-33%
`rej_eta`	✅	2s	6s	-67%
`shake128_absorb`	✅	2s	3s	-33%
`shake128_init`	✅	2s	2s	+0%
`shake128x4_absorb_once`	✅	2s	2s	+0%
`shake128x4_squeezeblocks`	✅	2s	2s	+0%
`shake256_absorb`	✅	2s	3s	-33%
`shake256_finalize`	✅	2s	1s	+100%
`shake256_release`	✅	2s	1s	+100%
`shake256x4_squeezeblocks`	✅	2s	4s	-50%
`sign_keypair`	✅	2s	7s	-71%
`sys_check_capability`	✅	2s	3s	-33%
`mld_ct_cmask_neg_i32`	✅	1s	2s	-50%
`shake256_init`	✅	1s	4s	-75%
`unpack_sig`	✅	1s	4s	-75%

oqs-bot · 2026-04-03T11:49:37Z

CBMC Results (ML-DSA-87)

⚠️ Attention Required

Proof	Status	Current	Previous	Change
`TOTAL`	⚠️	3703s	2668s	+38.8%
`polyvecl_pointwise_acc_montgomery_c`	⚠️	1497s	311s	+381%

Full Results (177 proofs)

Proof	Status	Current	Previous	Change
`TOTAL`	⚠️	3703s	2668s	+38.8%
`polyvecl_pointwise_acc_montgomery_c`	⚠️	1497s	311s	+381%
`polyvec_matrix_expand`	✅	254s	195s	+30%
`poly_pointwise_montgomery_c`	✅	191s	176s	+9%
`rej_uniform_native`	✅	156s	158s	-1%
`sign_verify_internal`	✅	149s	216s	-31%
`mld_invntt_layer`	✅	105s	104s	+1%
`mld_attempt_signature_generation`	✅	100s	239s	-58%
`mld_ct_memcmp`	✅	89s	83s	+7%
`polyvec_matrix_expand_serial`	✅	89s	85s	+5%
`mld_ntt_layer`	✅	59s	56s	+5%
`sign_signature_internal`	✅	42s	42s	+0%
`polymat_permute_bitrev_to_custom`	✅	26s	47s	-45%
`mld_compute_t0_t1_tr_from_sk_components`	✅	23s	25s	-8%
`fqmul`	✅	22s	21s	+5%
`poly_chknorm_c`	✅	22s	24s	-8%
`rej_uniform`	✅	21s	24s	-12%
`polyeta_unpack`	✅	19s	20s	-5%
`poly_uniform_4x`	✅	18s	16s	+12%
`polyt0_unpack`	✅	18s	15s	+20%
`mld_check_pct`	✅	17s	9s	+89%
`polyveck_power2round`	✅	17s	8s	+112%
`poly_uniform_eta_4x`	✅	16s	17s	-6%
`rej_uniform_c`	✅	16s	16s	+0%
`keccakf1600x4_permute_native`	✅	15s	15s	+0%
`mld_ntt_butterfly_block`	✅	15s	13s	+15%
`mld_polyvecl_permute_bitrev_to_custom_native`	✅	14s	13s	+8%
`keccak_absorb_once_x4`	✅	12s	9s	+33%
`mld_sample_s1_s2_serial`	✅	12s	6s	+100%
`poly_add`	✅	11s	12s	-8%
`polyveck_add`	✅	11s	9s	+22%
`polyveck_decompose`	✅	11s	60s	-82%
`polyveck_use_hint`	✅	11s	9s	+22%
`keccakf1600_permute`	✅	10s	7s	+43%
`polyveck_shiftl`	✅	10s	7s	+43%
`polyvecl_ntt`	✅	10s	8s	+25%
`polyz_unpack_c`	✅	10s	8s	+25%
`keccakf1600_permute_native`	✅	9s	7s	+29%
`polyvec_matrix_pointwise_montgomery`	✅	9s	11s	-18%
`polyveck_chknorm`	✅	9s	4s	+125%
`polyveck_invntt_tomont`	✅	9s	8s	+12%
`unpack_sk`	✅	9s	6s	+50%
`mld_compute_pack_z`	✅	8s	6s	+33%
`mld_sample_s1_s2`	✅	8s	5s	+60%
`poly_challenge`	✅	8s	5s	+60%
`poly_decompose_c`	✅	8s	8s	+0%
`poly_invntt_tomont_c`	✅	8s	9s	-11%
`polyveck_ntt`	✅	8s	11s	-27%
`polyveck_reduce`	✅	8s	10s	-20%
`poly_caddq_c`	✅	7s	6s	+17%
`polyveck_pack_w1`	✅	7s	4s	+75%
`polyveck_sub`	✅	7s	6s	+17%
`polyvecl_chknorm`	✅	7s	4s	+75%
`sign`	✅	7s	7s	+0%
`sign_pk_from_sk`	✅	7s	9s	-22%
`keccak_absorb`	✅	6s	7s	-14%
`keccak_squeezeblocks_x4`	✅	6s	7s	-14%
`poly_pointwise_montgomery_native`	✅	6s	4s	+50%
`poly_uniform_eta`	✅	6s	4s	+50%
`polyveck_pointwise_poly_montgomery`	✅	6s	5s	+20%
`polyz_pack`	✅	6s	3s	+100%
`shake128_init`	✅	6s	1s	+500%
`sign_keypair_internal`	✅	6s	5s	+20%
`sign_signature_extmu`	✅	6s	6s	+0%
`sign_signature_pre_hash_shake256`	✅	6s	4s	+50%
`sign_verify`	✅	6s	5s	+20%
`sign_verify_pre_hash_internal`	✅	6s	2s	+200%
`unpack_hints`	✅	6s	6s	+0%
`fqscale`	✅	5s	7s	-29%
`keccak_squeeze`	✅	5s	2s	+150%
`keccakf1600_extract_bytes (big endian)`	✅	5s	2s	+150%
`keccakf1600x4_extract_bytes`	✅	5s	5s	+0%
`keccakf1600x4_permute`	✅	5s	4s	+25%
`ntt_native_aarch64`	✅	5s	4s	+25%
`pack_pk`	✅	5s	3s	+67%
`pack_sk`	✅	5s	2s	+150%
`poly_decompose`	✅	5s	2s	+150%
`poly_invntt_tomont_native`	✅	5s	2s	+150%
`poly_power2round`	✅	5s	8s	-38%
`poly_uniform`	✅	5s	4s	+25%
`poly_use_hint`	✅	5s	5s	+0%
`polyveck_caddq`	✅	5s	11s	-55%
`polyveck_unpack_t0`	✅	5s	7s	-29%
`polyvecl_pointwise_acc_montgomery_native`	✅	5s	5s	+0%
`sign_signature`	✅	5s	2s	+150%
`sign_signature_pre_hash_internal`	✅	5s	4s	+25%
`intt_native_x86_64`	✅	4s	3s	+33%
`keccak_finalize`	✅	4s	3s	+33%
`mld_h`	✅	4s	5s	-20%
`mld_keccakf1600_extract_bytes`	✅	4s	2s	+100%
`mld_prepare_domain_separation_prefix`	✅	4s	4s	+0%
`pack_sig_c`	✅	4s	-	new
`pack_sig_z`	✅	4s	4s	+0%
`poly_chknorm_native_aarch64`	✅	4s	6s	-33%
`poly_decompose_native`	✅	4s	5s	-20%
`poly_ntt`	✅	4s	2s	+100%
`poly_uniform_gamma1_4x`	✅	4s	4s	+0%
`poly_use_hint_native`	✅	4s	4s	+0%
`polyeta_pack`	✅	4s	4s	+0%
`polyvecl_uniform_gamma1`	✅	4s	4s	+0%
`polyz_unpack_native`	✅	4s	5s	-20%
`power2round`	✅	4s	4s	+0%
`rej_eta`	✅	4s	4s	+0%
`rej_eta_c`	✅	4s	4s	+0%
`rej_eta_native`	✅	4s	4s	+0%
`shake128_absorb`	✅	4s	3s	+33%
`shake256_release`	✅	4s	4s	+0%
`shake256x4_absorb_once`	✅	4s	2s	+100%
`sign_keypair`	✅	4s	7s	-43%
`sign_verify_pre_hash_shake256`	✅	4s	6s	-33%
`use_hint`	✅	4s	2s	+100%
`caddq`	✅	3s	4s	-25%
`decompose`	✅	3s	3s	+0%
`keccakf1600_xor_bytes`	✅	3s	4s	-25%
`keccakf1600_xor_bytes (big endian)`	✅	3s	3s	+0%
`keccakf1600x4_xor_bytes`	✅	3s	3s	+0%
`make_hint`	✅	3s	4s	-25%
`mld_ct_cmask_neg_i32`	✅	3s	1s	+200%
`mld_ct_cmask_nonzero_u32`	✅	3s	3s	+0%
`mld_ct_cmask_nonzero_u8`	✅	3s	2s	+50%
`mld_ct_get_optblocker_u32`	✅	3s	2s	+50%
`mld_value_barrier_i64`	✅	3s	2s	+50%
`mld_value_barrier_u32`	✅	3s	2s	+50%
`ntt_native_x86_64`	✅	3s	3s	+0%
`pack_sig_h_poly`	✅	3s	-	new
`poly_caddq`	✅	3s	4s	-25%
`poly_caddq_native`	✅	3s	3s	+0%
`poly_caddq_native_aarch64`	✅	3s	2s	+50%
`poly_chknorm_native`	✅	3s	3s	+0%
`poly_invntt_tomont`	✅	3s	4s	-25%
`poly_make_hint`	✅	3s	3s	+0%
`poly_ntt_c`	✅	3s	4s	-25%
`poly_ntt_native`	✅	3s	4s	-25%
`poly_reduce`	✅	3s	3s	+0%
`poly_shiftl`	✅	3s	5s	-40%
`poly_sub`	✅	3s	3s	+0%
`poly_use_hint_c`	✅	3s	2s	+50%
`polyt0_pack`	✅	3s	3s	+0%
`polyveck_pack_t0`	✅	3s	3s	+0%
`polyveck_unpack_eta`	✅	3s	4s	-25%
`polyvecl_pack_eta`	✅	3s	4s	-25%
`polyvecl_permute_bitrev_to_custom`	✅	3s	3s	+0%
`polyvecl_pointwise_acc_montgomery`	✅	3s	4s	-25%
`polyvecl_uniform_gamma1_serial`	✅	3s	4s	-25%
`polyw1_pack`	✅	3s	2s	+50%
`reduce32`	✅	3s	4s	-25%
`shake128_finalize`	✅	3s	2s	+50%
`shake128_release`	✅	3s	3s	+0%
`shake128_squeeze`	✅	3s	4s	-25%
`shake128x4_absorb_once`	✅	3s	3s	+0%
`shake256`	✅	3s	2s	+50%
`shake256_finalize`	✅	3s	2s	+50%
`sys_check_capability`	✅	3s	3s	+0%
`unpack_pk`	✅	3s	4s	-25%
`mld_ct_abs_i32`	✅	2s	2s	+0%
`mld_ct_get_optblocker_i64`	✅	2s	4s	-50%
`mld_ct_get_optblocker_u8`	✅	2s	2s	+0%
`mld_ct_sel_int32`	✅	2s	4s	-50%
`mld_value_barrier_u8`	✅	2s	2s	+0%
`montgomery_reduce`	✅	2s	2s	+0%
`poly_chknorm`	✅	2s	2s	+0%
`poly_pointwise_montgomery`	✅	2s	6s	-67%
`poly_uniform_gamma1`	✅	2s	5s	-60%
`polyt1_pack`	✅	2s	4s	-50%
`polyt1_unpack`	✅	2s	3s	-33%
`polyveck_pack_eta`	✅	2s	4s	-50%
`polyvecl_unpack_eta`	✅	2s	6s	-67%
`polyvecl_unpack_z`	✅	2s	3s	-33%
`polyz_unpack`	✅	2s	2s	+0%
`shake128x4_squeezeblocks`	✅	2s	2s	+0%
`shake256_absorb`	✅	2s	3s	-33%
`shake256_init`	✅	2s	1s	+100%
`sign_open`	✅	2s	2s	+0%
`sign_verify_extmu`	✅	2s	2s	+0%
`unpack_sig`	✅	2s	4s	-50%
`keccak_init`	✅	1s	2s	-50%
`shake256_squeeze`	✅	1s	2s	-50%
`shake256x4_squeezeblocks`	✅	1s	4s	-75%

Instead of allocating a full polyveck for h in attempt_signature_generation, compute cs2, ct0, and hints one polynomial at a time using scratch polys. This eliminates the polyveck h from the yh_u union, replacing mld_pack_sig_c_h with incremental packing via mld_pack_sig_c and mld_pack_sig_h_poly. This is a prerequisite for also eliminating y in a follow up. Signed-off-by: Matthias J. Kannwischer <matthias@zerorisc.com>

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

mkannwischer force-pushed the sign-incremental-h branch from 6bc07cf to 1def327 Compare April 3, 2026 13:02

mkannwischer force-pushed the sign-incremental-h branch from 1def327 to d0bb6c5 Compare April 3, 2026 13:50

mkannwischer marked this pull request as ready for review April 3, 2026 14:17

mkannwischer requested a review from a team as a code owner April 3, 2026 14:17

mkannwischer mentioned this pull request Apr 5, 2026

lowram: Stream matrix A element-by-element to reduce memory #1019

Open

mkannwischer assigned hanno-becker Apr 5, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

lowram: Compute h incrementally in signing#1015

lowram: Compute h incrementally in signing#1015
mkannwischer wants to merge 1 commit intomainfrom
sign-incremental-h

mkannwischer commented Apr 3, 2026

Uh oh!

oqs-bot commented Apr 3, 2026 •

edited

Loading

Uh oh!

oqs-bot commented Apr 3, 2026 •

edited

Loading

Uh oh!

oqs-bot commented Apr 3, 2026 •

edited

Loading

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

mkannwischer commented Apr 3, 2026

Uh oh!

oqs-bot commented Apr 3, 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 3, 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 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

CBMC Results (ML-DSA-87)

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

oqs-bot commented Apr 3, 2026 •

edited

Loading

oqs-bot commented Apr 3, 2026 •

edited

Loading

oqs-bot commented Apr 3, 2026 •

edited

Loading