lowmem: Unpack z lazily in verification by mkannwischer · Pull Request #1025 · pq-code-package/mldsa-native

mkannwischer · 2026-04-07T06:29:38Z

Introduce mld_zvec following the lazy polyvec pattern (eager / lazy
variants with #define dispatch on MLD_CONFIG_REDUCE_RAM):

mld_zvec_init: in eager mode unpacks the full polyvecl, performs the
polyvecl-wide infinity-norm bound check, and NTTs in place. In lazy
mode it just stores a pointer to the packed signature bytes.
mld_zvec_get_poly: in eager mode copies a single polynomial from
the precomputed vector. In lazy mode unpacks one polynomial,
performs the per-poly infinity-norm bound check, and NTTs into
the caller-provided buffer.

The norm check thus moves out of mld_sign_verify_internal into the
zvec init / get_poly accessors, so the verify body no longer has to
sequence chknorm explicitly.

Add a fused matrix-vector helper
mld_polyvec_matrix_pointwise_montgomery_zvec used by verify:

The eager variant is a thin wrapper around the existing
mld_polyvec_matrix_pointwise_montgomery_eager (z is already NTT'd
by mld_zvec_init).
The lazy variant streams z via mld_zvec_get_poly_lazy and generates
the matrix on-the-fly column-by-column,
accumulating A[*,l] * z[l] into w.

In REDUCE_RAM mode this avoids holding the full unpacked polyvecl z
in memory at once, reducing verify allocation by 2-5 KiB per parameter
set.

Depends on lowram: Stream matrix A element-by-element to reduce memory #1019
Depends on Lowram: Share buffers with non-overlapping lifetimes in verify #1007
Hoisted out from PoC: Reduce large struct allocations to <= 13/17/21 KiB for ML-DSA-44/65/87 #1005

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>

Replace the row-level matrix buffer (mld_polyvecl) with a single-poly buffer in REDUCE_RAM mode. In the lazy path, matrix elements A[k][l] are sampled on demand one at a time, and the matrix-vector product accumulates element-by-element instead of row-by-row. Restructure polymat into eager/lazy variants following the same pattern as s1hat/s2hat/t0hat: - mld_polymat_eager: stores full K x L matrix - mld_polymat_lazy: stores rho + single poly_buffer + tmp - mld_polyvec_matrix_expand_eager/_lazy: separate implementations - mld_polyvec_matrix_pointwise_montgomery_eager/_lazy: separate implementations with CBMC contracts only on the eager variants Move all polymat-related code from polyvec.h/polyvec.c into polyvec_lazy.h/polyvec_lazy.c. Signed-off-by: Matthias J. Kannwischer <matthias@zerorisc.com>

Signed-off-by: Matthias J. Kannwischer <matthias@zerorisc.com>

oqs-bot · 2026-04-07T06:47:49Z

CBMC Results (ML-DSA-44)

⚠️ Attention Required

Proof	Status	Current	Previous	Change
`polyvecl_pointwise_acc_montgomery_c`	⚠️	332s	214s	+55%

Full Results (182 proofs)

Proof	Status	Current	Previous	Change
`TOTAL`	✅	2216s	2059s	+7.6%
`polyvecl_pointwise_acc_montgomery_c`	⚠️	332s	214s	+55%
`sign_verify_internal`	✅	268s	184s	+46%
`mld_attempt_signature_generation`	✅	216s	255s	-15%
`poly_pointwise_montgomery_c`	✅	164s	163s	+1%
`rej_uniform_native`	✅	147s	148s	-1%
`mld_invntt_layer`	✅	84s	88s	-5%
`mld_ct_memcmp`	✅	73s	72s	+1%
`mld_ntt_layer`	✅	56s	56s	+0%
`fqmul`	✅	22s	19s	+16%
`polymat_permute_bitrev_to_custom`	✅	22s	17s	+29%
`rej_uniform`	✅	22s	23s	-4%
`poly_chknorm_c`	✅	21s	21s	+0%
`sign_signature_internal`	✅	19s	19s	+0%
`polyeta_unpack`	✅	17s	16s	+6%
`polyvec_matrix_expand_eager`	✅	17s	-	new
`poly_uniform_4x`	✅	15s	17s	-12%
`poly_uniform_eta_4x`	✅	15s	18s	-17%
`mld_compute_t0_t1_tr_from_sk_components`	✅	14s	13s	+8%
`poly_add`	✅	14s	11s	+27%
`rej_uniform_c`	✅	14s	15s	-7%
`keccakf1600x4_permute_native`	✅	13s	12s	+8%
`polyt0_unpack`	✅	13s	14s	-7%
`polyvec_matrix_expand_eager_serial`	✅	13s	-	new
`mld_ntt_butterfly_block`	✅	12s	16s	-25%
`polyz_unpack_c`	✅	11s	12s	-8%
`sign_pk_from_sk`	✅	11s	7s	+57%
`mld_check_pct`	✅	10s	7s	+43%
`mld_polyvecl_permute_bitrev_to_custom_native`	✅	10s	8s	+25%
`keccak_absorb_once_x4`	✅	9s	9s	+0%
`keccakf1600_permute`	✅	9s	9s	+0%
`keccakf1600_permute_native`	✅	9s	10s	-10%
`polyvec_matrix_pointwise_montgomery_eager`	✅	9s	-	new
`polyveck_use_hint`	✅	9s	8s	+12%
`polyvecl_chknorm`	✅	9s	6s	+50%
`poly_caddq_c`	✅	8s	4s	+100%
`polyveck_decompose`	✅	8s	5s	+60%
`unpack_sk`	✅	8s	7s	+14%
`keccak_absorb`	✅	7s	6s	+17%
`keccak_squeezeblocks_x4`	✅	7s	7s	+0%
`polyveck_sub`	✅	7s	4s	+75%
`sign`	✅	7s	6s	+17%
`sign_verify`	✅	7s	4s	+75%
`unpack_hints`	✅	7s	5s	+40%
`mld_compute_pack_z`	✅	6s	4s	+50%
`mld_h`	✅	6s	4s	+50%
`mld_sample_s1_s2`	✅	6s	4s	+50%
`pack_sig_h_poly`	✅	6s	-	new
`poly_invntt_tomont_c`	✅	6s	7s	-14%
`poly_ntt`	✅	6s	2s	+200%
`polyveck_chknorm`	✅	6s	5s	+20%
`polyveck_power2round`	✅	6s	12s	-50%
`decompose`	✅	5s	4s	+25%
`mld_prepare_domain_separation_prefix`	✅	5s	3s	+67%
`poly_challenge`	✅	5s	6s	-17%
`poly_chknorm`	✅	5s	2s	+150%
`poly_chknorm_native`	✅	5s	5s	+0%
`poly_decompose`	✅	5s	3s	+67%
`poly_ntt_native`	✅	5s	3s	+67%
`poly_power2round`	✅	5s	5s	+0%
`poly_uniform_eta`	✅	5s	7s	-29%
`poly_use_hint_c`	✅	5s	6s	-17%
`polyt0_pack`	✅	5s	5s	+0%
`polyveck_pointwise_poly_montgomery`	✅	5s	4s	+25%
`polyw1_pack`	✅	5s	4s	+25%
`sign_open`	✅	5s	4s	+25%
`sign_verify_extmu`	✅	5s	3s	+67%
`caddq`	✅	4s	2s	+100%
`keccak_f1600_x4_native_aarch64_v8a_v84a_scalar_hybrid`	✅	4s	1s	+300%
`keccak_squeeze`	✅	4s	3s	+33%
`mld_ct_get_optblocker_u8`	✅	4s	1s	+300%
`mld_sample_s1_s2_serial`	✅	4s	6s	-33%
`montgomery_reduce`	✅	4s	3s	+33%
`ntt_native_aarch64`	✅	4s	5s	-20%
`poly_invntt_tomont`	✅	4s	4s	+0%
`poly_make_hint`	✅	4s	3s	+33%
`poly_ntt_c`	✅	4s	4s	+0%
`poly_pointwise_montgomery_native`	✅	4s	3s	+33%
`poly_reduce`	✅	4s	3s	+33%
`poly_sub`	✅	4s	4s	+0%
`poly_uniform`	✅	4s	4s	+0%
`poly_uniform_gamma1`	✅	4s	3s	+33%
`polyveck_add`	✅	4s	7s	-43%
`polyveck_caddq`	✅	4s	5s	-20%
`polyveck_invntt_tomont`	✅	4s	6s	-33%
`polyveck_ntt`	✅	4s	4s	+0%
`polyveck_shiftl`	✅	4s	7s	-43%
`polyvecl_ntt`	✅	4s	4s	+0%
`polyvecl_uniform_gamma1`	✅	4s	3s	+33%
`polyvecl_uniform_gamma1_serial`	✅	4s	4s	+0%
`polyz_unpack`	✅	4s	2s	+100%
`power2round`	✅	4s	4s	+0%
`shake128_absorb`	✅	4s	2s	+100%
`shake128x4_squeezeblocks`	✅	4s	1s	+300%
`sign_keypair_internal`	✅	4s	6s	-33%
`sign_signature`	✅	4s	6s	-33%
`sign_verify_pre_hash_internal`	✅	4s	5s	-20%
`intt_native_x86_64`	✅	3s	1s	+200%
`keccak_f1600_x1_native_aarch64_v84a`	✅	3s	3s	+0%
`keccak_f1600_x4_native_aarch64_v84a`	✅	3s	2s	+50%
`keccak_f1600_x4_native_aarch64_v8a_scalar_hybrid`	✅	3s	2s	+50%
`keccak_init`	✅	3s	2s	+50%
`keccakf1600_xor_bytes (big endian)`	✅	3s	2s	+50%
`make_hint`	✅	3s	2s	+50%
`mld_ct_cmask_neg_i32`	✅	3s	3s	+0%
`mld_ct_cmask_nonzero_u32`	✅	3s	2s	+50%
`mld_ct_cmask_nonzero_u8`	✅	3s	3s	+0%
`ntt_native_x86_64`	✅	3s	2s	+50%
`pack_pk`	✅	3s	2s	+50%
`pack_sig_c`	✅	3s	-	new
`pack_sig_z`	✅	3s	3s	+0%
`pack_sk`	✅	3s	2s	+50%
`poly_caddq_native`	✅	3s	2s	+50%
`poly_chknorm_native_aarch64`	✅	3s	3s	+0%
`poly_invntt_tomont_native`	✅	3s	3s	+0%
`poly_uniform_gamma1_4x`	✅	3s	4s	-25%
`poly_use_hint`	✅	3s	3s	+0%
`polyveck_reduce`	✅	3s	4s	-25%
`polyveck_unpack_eta`	✅	3s	4s	-25%
`polyvecl_permute_bitrev_to_custom`	✅	3s	1s	+200%
`polyvecl_pointwise_acc_montgomery_native`	✅	3s	3s	+0%
`polyvecl_unpack_z`	✅	3s	2s	+50%
`polyz_pack`	✅	3s	3s	+0%
`reduce32`	✅	3s	2s	+50%
`rej_eta`	✅	3s	2s	+50%
`rej_eta_c`	✅	3s	4s	-25%
`rej_eta_native`	✅	3s	5s	-40%
`shake128_finalize`	✅	3s	3s	+0%
`shake128_init`	✅	3s	1s	+200%
`shake128_squeeze`	✅	3s	3s	+0%
`shake128x4_absorb_once`	✅	3s	1s	+200%
`shake256_finalize`	✅	3s	2s	+50%
`shake256_squeeze`	✅	3s	2s	+50%
`shake256x4_squeezeblocks`	✅	3s	3s	+0%
`sign_keypair`	✅	3s	4s	-25%
`sign_signature_extmu`	✅	3s	3s	+0%
`sign_signature_pre_hash_internal`	✅	3s	3s	+0%
`sign_signature_pre_hash_shake256`	✅	3s	6s	-50%
`sign_verify_pre_hash_shake256`	✅	3s	4s	-25%
`sys_check_capability`	✅	3s	3s	+0%
`fqscale`	✅	2s	3s	-33%
`keccak_finalize`	✅	2s	2s	+0%
`keccakf1600_extract_bytes (big endian)`	✅	2s	2s	+0%
`keccakf1600_xor_bytes`	✅	2s	1s	+100%
`keccakf1600x4_extract_bytes`	✅	2s	2s	+0%
`keccakf1600x4_permute`	✅	2s	4s	-50%
`keccakf1600x4_xor_bytes`	✅	2s	3s	-33%
`mld_ct_abs_i32`	✅	2s	3s	-33%
`mld_ct_get_optblocker_i64`	✅	2s	4s	-50%
`mld_ct_get_optblocker_u32`	✅	2s	3s	-33%
`mld_keccakf1600_extract_bytes`	✅	2s	3s	-33%
`mld_value_barrier_i64`	✅	2s	1s	+100%
`mld_value_barrier_u8`	✅	2s	4s	-50%
`poly_caddq`	✅	2s	3s	-33%
`poly_decompose_c`	✅	2s	5s	-60%
`poly_decompose_native`	✅	2s	2s	+0%
`poly_pointwise_montgomery`	✅	2s	2s	+0%
`poly_shiftl`	✅	2s	2s	+0%
`poly_use_hint_native`	✅	2s	5s	-60%
`polyeta_pack`	✅	2s	1s	+100%
`polyt1_pack`	✅	2s	3s	-33%
`polyt1_unpack`	✅	2s	4s	-50%
`polyvec_matrix_pointwise_montgomery_zvec_eager`	✅	2s	-	new
`polyveck_pack_eta`	✅	2s	3s	-33%
`polyveck_unpack_t0`	✅	2s	5s	-60%
`polyvecl_pack_eta`	✅	2s	4s	-50%
`polyvecl_pointwise_acc_montgomery`	✅	2s	2s	+0%
`polyvecl_unpack_eta`	✅	2s	3s	-33%
`polyz_unpack_native`	✅	2s	3s	-33%
`shake128_release`	✅	2s	3s	-33%
`shake256`	✅	2s	2s	+0%
`shake256_absorb`	✅	2s	3s	-33%
`shake256_init`	✅	2s	2s	+0%
`shake256_release`	✅	2s	2s	+0%
`shake256x4_absorb_once`	✅	2s	3s	-33%
`unpack_pk`	✅	2s	4s	-50%
`keccak_f1600_x1_native_aarch64`	✅	1s	7s	-86%
`mld_ct_sel_int32`	✅	1s	3s	-67%
`mld_value_barrier_u32`	✅	1s	3s	-67%
`poly_caddq_native_aarch64`	✅	1s	3s	-67%
`polyveck_pack_t0`	✅	1s	3s	-67%
`polyveck_pack_w1`	✅	1s	3s	-67%
`use_hint`	✅	1s	4s	-75%

oqs-bot · 2026-04-07T06:48:40Z

CBMC Results (ML-DSA-65)

⚠️ Attention Required

Proof	Status	Current	Previous	Change
`polyvecl_pointwise_acc_montgomery_c`	⚠️	300s	167s	+80%

Full Results (182 proofs)

Proof	Status	Current	Previous	Change
`TOTAL`	✅	2345s	2314s	+1.3%
`polyvecl_pointwise_acc_montgomery_c`	⚠️	300s	167s	+80%
`sign_verify_internal`	✅	236s	319s	-26%
`mld_attempt_signature_generation`	✅	223s	269s	-17%
`poly_pointwise_montgomery_c`	✅	160s	142s	+13%
`rej_uniform_native`	✅	143s	145s	-1%
`mld_invntt_layer`	✅	99s	92s	+8%
`polyvec_matrix_expand_eager`	✅	99s	-	new
`mld_ct_memcmp`	✅	78s	71s	+10%
`mld_ntt_layer`	✅	55s	52s	+6%
`polyvec_matrix_expand_eager_serial`	✅	54s	-	new
`sign_signature_internal`	✅	33s	25s	+32%
`mld_compute_t0_t1_tr_from_sk_components`	✅	22s	24s	-8%
`rej_uniform`	✅	22s	20s	+10%
`poly_chknorm_c`	✅	20s	19s	+5%
`fqmul`	✅	18s	18s	+0%
`poly_uniform_eta_4x`	✅	17s	16s	+6%
`polyveck_decompose`	✅	17s	15s	+13%
`poly_uniform_4x`	✅	16s	17s	-6%
`polymat_permute_bitrev_to_custom`	✅	15s	27s	-44%
`polyt0_unpack`	✅	15s	13s	+15%
`polyveck_add`	✅	14s	9s	+56%
`rej_uniform_c`	✅	14s	14s	+0%
`keccakf1600x4_permute_native`	✅	12s	14s	-14%
`polyvec_matrix_pointwise_montgomery_eager`	✅	12s	-	new
`polyveck_power2round`	✅	12s	12s	+0%
`polyvecl_chknorm`	✅	12s	12s	+0%
`mld_ntt_butterfly_block`	✅	11s	12s	-8%
`poly_add`	✅	11s	10s	+10%
`keccak_absorb_once_x4`	✅	10s	10s	+0%
`keccakf1600_permute`	✅	10s	8s	+25%
`keccakf1600_permute_native`	✅	9s	8s	+12%
`mld_check_pct`	✅	9s	10s	-10%
`mld_polyvecl_permute_bitrev_to_custom_native`	✅	9s	8s	+12%
`mld_compute_pack_z`	✅	8s	8s	+0%
`poly_decompose_c`	✅	8s	8s	+0%
`polyveck_caddq`	✅	8s	6s	+33%
`polyveck_ntt`	✅	8s	6s	+33%
`polyveck_pointwise_poly_montgomery`	✅	8s	6s	+33%
`polyveck_use_hint`	✅	8s	7s	+14%
`sign_open`	✅	8s	4s	+100%
`sign_pk_from_sk`	✅	8s	7s	+14%
`unpack_sk`	✅	8s	7s	+14%
`keccak_absorb`	✅	7s	9s	-22%
`keccak_squeezeblocks_x4`	✅	7s	6s	+17%
`mld_h`	✅	7s	5s	+40%
`mld_prepare_domain_separation_prefix`	✅	7s	6s	+17%
`poly_caddq_c`	✅	7s	7s	+0%
`polyveck_chknorm`	✅	7s	6s	+17%
`polyveck_invntt_tomont`	✅	7s	8s	-12%
`sign`	✅	7s	7s	+0%
`keccak_squeeze`	✅	6s	4s	+50%
`poly_invntt_tomont_c`	✅	6s	4s	+50%
`poly_power2round`	✅	6s	3s	+100%
`poly_reduce`	✅	6s	1s	+500%
`poly_uniform_eta`	✅	6s	3s	+100%
`poly_use_hint_c`	✅	6s	4s	+50%
`polyeta_unpack`	✅	6s	4s	+50%
`polyveck_shiftl`	✅	6s	9s	-33%
`polyveck_sub`	✅	6s	12s	-50%
`keccak_f1600_x4_native_aarch64_v84a`	✅	5s	1s	+400%
`mld_sample_s1_s2`	✅	5s	7s	-29%
`mld_sample_s1_s2_serial`	✅	5s	6s	-17%
`poly_invntt_tomont_native`	✅	5s	3s	+67%
`poly_sub`	✅	5s	4s	+25%
`polyt0_pack`	✅	5s	3s	+67%
`polyveck_reduce`	✅	5s	7s	-29%
`polyvecl_ntt`	✅	5s	8s	-38%
`rej_eta_native`	✅	5s	3s	+67%
`shake128_release`	✅	5s	2s	+150%
`shake256_squeeze`	✅	5s	3s	+67%
`sign_keypair_internal`	✅	5s	5s	+0%
`sign_verify`	✅	5s	6s	-17%
`sign_verify_pre_hash_internal`	✅	5s	4s	+25%
`sign_verify_pre_hash_shake256`	✅	5s	6s	-17%
`fqscale`	✅	4s	4s	+0%
`intt_native_x86_64`	✅	4s	3s	+33%
`keccakf1600_extract_bytes (big endian)`	✅	4s	1s	+300%
`mld_ct_sel_int32`	✅	4s	1s	+300%
`pack_pk`	✅	4s	3s	+33%
`pack_sig_h_poly`	✅	4s	-	new
`pack_sk`	✅	4s	3s	+33%
`poly_caddq_native`	✅	4s	5s	-20%
`poly_challenge`	✅	4s	2s	+100%
`poly_pointwise_montgomery`	✅	4s	2s	+100%
`poly_pointwise_montgomery_native`	✅	4s	3s	+33%
`poly_uniform`	✅	4s	4s	+0%
`polyt1_unpack`	✅	4s	2s	+100%
`polyveck_pack_eta`	✅	4s	3s	+33%
`polyveck_pack_t0`	✅	4s	3s	+33%
`polyveck_unpack_t0`	✅	4s	4s	+0%
`polyvecl_permute_bitrev_to_custom`	✅	4s	5s	-20%
`polyvecl_pointwise_acc_montgomery_native`	✅	4s	5s	-20%
`polyz_unpack_c`	✅	4s	5s	-20%
`power2round`	✅	4s	1s	+300%
`shake128x4_absorb_once`	✅	4s	2s	+100%
`shake256_release`	✅	4s	3s	+33%
`sign_signature_extmu`	✅	4s	5s	-20%
`sign_signature_pre_hash_internal`	✅	4s	4s	+0%
`sign_signature_pre_hash_shake256`	✅	4s	2s	+100%
`unpack_hints`	✅	4s	5s	-20%
`decompose`	✅	3s	3s	+0%
`keccak_finalize`	✅	3s	3s	+0%
`keccakf1600x4_xor_bytes`	✅	3s	1s	+200%
`mld_ct_cmask_neg_i32`	✅	3s	1s	+200%
`mld_ct_cmask_nonzero_u32`	✅	3s	3s	+0%
`mld_ct_get_optblocker_i64`	✅	3s	2s	+50%
`poly_caddq`	✅	3s	1s	+200%
`poly_caddq_native_aarch64`	✅	3s	3s	+0%
`poly_chknorm`	✅	3s	4s	-25%
`poly_chknorm_native`	✅	3s	2s	+50%
`poly_chknorm_native_aarch64`	✅	3s	2s	+50%
`poly_decompose`	✅	3s	5s	-40%
`poly_invntt_tomont`	✅	3s	4s	-25%
`poly_ntt`	✅	3s	4s	-25%
`poly_ntt_c`	✅	3s	3s	+0%
`poly_ntt_native`	✅	3s	2s	+50%
`poly_shiftl`	✅	3s	3s	+0%
`polyeta_pack`	✅	3s	3s	+0%
`polyt1_pack`	✅	3s	3s	+0%
`polyvec_matrix_pointwise_montgomery_zvec_eager`	✅	3s	-	new
`polyveck_pack_w1`	✅	3s	2s	+50%
`polyveck_unpack_eta`	✅	3s	2s	+50%
`polyvecl_pack_eta`	✅	3s	3s	+0%
`polyvecl_uniform_gamma1_serial`	✅	3s	3s	+0%
`polyz_pack`	✅	3s	2s	+50%
`polyz_unpack`	✅	3s	4s	-25%
`reduce32`	✅	3s	3s	+0%
`shake128_absorb`	✅	3s	3s	+0%
`shake128_init`	✅	3s	2s	+50%
`shake128_squeeze`	✅	3s	3s	+0%
`shake256_absorb`	✅	3s	2s	+50%
`shake256_finalize`	✅	3s	1s	+200%
`shake256x4_absorb_once`	✅	3s	2s	+50%
`shake256x4_squeezeblocks`	✅	3s	5s	-40%
`sign_keypair`	✅	3s	6s	-50%
`sign_signature`	✅	3s	2s	+50%
`sign_verify_extmu`	✅	3s	5s	-40%
`sys_check_capability`	✅	3s	1s	+200%
`unpack_pk`	✅	3s	2s	+50%
`caddq`	✅	2s	3s	-33%
`keccak_f1600_x1_native_aarch64`	✅	2s	2s	+0%
`keccak_f1600_x1_native_aarch64_v84a`	✅	2s	1s	+100%
`keccak_f1600_x4_native_aarch64_v8a_scalar_hybrid`	✅	2s	5s	-60%
`keccak_init`	✅	2s	3s	-33%
`keccakf1600_xor_bytes (big endian)`	✅	2s	3s	-33%
`keccakf1600x4_extract_bytes`	✅	2s	4s	-50%
`keccakf1600x4_permute`	✅	2s	1s	+100%
`make_hint`	✅	2s	5s	-60%
`mld_ct_abs_i32`	✅	2s	4s	-50%
`mld_ct_cmask_nonzero_u8`	✅	2s	1s	+100%
`mld_ct_get_optblocker_u32`	✅	2s	2s	+0%
`mld_ct_get_optblocker_u8`	✅	2s	2s	+0%
`mld_value_barrier_i64`	✅	2s	2s	+0%
`mld_value_barrier_u8`	✅	2s	2s	+0%
`montgomery_reduce`	✅	2s	2s	+0%
`ntt_native_aarch64`	✅	2s	6s	-67%
`ntt_native_x86_64`	✅	2s	4s	-50%
`pack_sig_c`	✅	2s	-	new
`pack_sig_z`	✅	2s	1s	+100%
`poly_decompose_native`	✅	2s	4s	-50%
`poly_make_hint`	✅	2s	2s	+0%
`poly_uniform_gamma1`	✅	2s	4s	-50%
`poly_use_hint`	✅	2s	1s	+100%
`poly_use_hint_native`	✅	2s	2s	+0%
`polyvecl_pointwise_acc_montgomery`	✅	2s	3s	-33%
`polyvecl_uniform_gamma1`	✅	2s	2s	+0%
`polyvecl_unpack_eta`	✅	2s	4s	-50%
`polyw1_pack`	✅	2s	3s	-33%
`polyz_unpack_native`	✅	2s	2s	+0%
`rej_eta`	✅	2s	3s	-33%
`rej_eta_c`	✅	2s	4s	-50%
`shake128_finalize`	✅	2s	1s	+100%
`shake128x4_squeezeblocks`	✅	2s	2s	+0%
`shake256`	✅	2s	1s	+100%
`shake256_init`	✅	2s	4s	-50%
`use_hint`	✅	2s	2s	+0%
`keccak_f1600_x4_native_aarch64_v8a_v84a_scalar_hybrid`	✅	1s	3s	-67%
`keccakf1600_xor_bytes`	✅	1s	5s	-80%
`mld_keccakf1600_extract_bytes`	✅	1s	1s	+0%
`mld_value_barrier_u32`	✅	1s	1s	+0%
`poly_uniform_gamma1_4x`	✅	1s	3s	-67%
`polyvecl_unpack_z`	✅	1s	2s	-50%

oqs-bot · 2026-04-07T06:48:52Z

CBMC Results (ML-DSA-87)

Full Results (182 proofs)

Proof	Status	Current	Previous	Change
`TOTAL`	✅	2210s	2628s	-15.9%
`sign_verify_internal`	✅	243s	221s	+10%
`mld_attempt_signature_generation`	✅	186s	238s	-22%
`polyvecl_pointwise_acc_montgomery_c`	✅	182s	303s	-40%
`polyvec_matrix_expand_eager`	✅	162s	-	new
`rej_uniform_native`	✅	142s	154s	-8%
`poly_pointwise_montgomery_c`	✅	139s	175s	-21%
`mld_invntt_layer`	✅	92s	98s	-6%
`mld_ct_memcmp`	✅	65s	80s	-19%
`polyvec_matrix_expand_eager_serial`	✅	61s	-	new
`mld_ntt_layer`	✅	52s	54s	-4%
`sign_signature_internal`	✅	35s	39s	-10%
`mld_compute_t0_t1_tr_from_sk_components`	✅	26s	26s	+0%
`polymat_permute_bitrev_to_custom`	✅	23s	47s	-51%
`fqmul`	✅	19s	23s	-17%
`rej_uniform`	✅	19s	21s	-10%
`poly_chknorm_c`	✅	18s	22s	-18%
`poly_uniform_eta_4x`	✅	18s	18s	+0%
`polyeta_unpack`	✅	16s	16s	+0%
`poly_uniform_4x`	✅	15s	15s	+0%
`keccakf1600x4_permute_native`	✅	14s	13s	+8%
`rej_uniform_c`	✅	14s	15s	-7%
`polyveck_power2round`	✅	13s	12s	+8%
`mld_ntt_butterfly_block`	✅	12s	14s	-14%
`polyt0_unpack`	✅	12s	15s	-20%
`poly_add`	✅	11s	13s	-15%
`polyveck_add`	✅	11s	10s	+10%
`polyveck_ntt`	✅	11s	7s	+57%
`keccak_absorb_once_x4`	✅	10s	10s	+0%
`keccakf1600_permute_native`	✅	10s	8s	+25%
`mld_polyvecl_permute_bitrev_to_custom_native`	✅	10s	13s	-23%
`mld_check_pct`	✅	9s	9s	+0%
`poly_invntt_tomont_c`	✅	9s	6s	+50%
`polyveck_decompose`	✅	9s	59s	-85%
`polyveck_shiftl`	✅	9s	5s	+80%
`polyvecl_ntt`	✅	9s	11s	-18%
`sign_open`	✅	9s	3s	+200%
`sign_pk_from_sk`	✅	9s	7s	+29%
`poly_decompose_c`	✅	8s	7s	+14%
`polyveck_caddq`	✅	8s	7s	+14%
`polyz_unpack_c`	✅	8s	11s	-27%
`sign_keypair_internal`	✅	8s	7s	+14%
`unpack_sk`	✅	8s	7s	+14%
`keccak_absorb`	✅	7s	6s	+17%
`keccakf1600_permute`	✅	7s	8s	-12%
`mld_sample_s1_s2`	✅	7s	8s	-12%
`poly_power2round`	✅	7s	6s	+17%
`sign`	✅	7s	7s	+0%
`keccak_squeezeblocks_x4`	✅	6s	5s	+20%
`mld_sample_s1_s2_serial`	✅	6s	7s	-14%
`poly_challenge`	✅	6s	5s	+20%
`poly_sub`	✅	6s	3s	+100%
`polyvec_matrix_pointwise_montgomery_eager`	✅	6s	-	new
`polyveck_invntt_tomont`	✅	6s	4s	+50%
`polyveck_use_hint`	✅	6s	8s	-25%
`polyvecl_unpack_z`	✅	6s	3s	+100%
`shake256_squeeze`	✅	6s	2s	+200%
`sign_signature`	✅	6s	5s	+20%
`sign_verify_pre_hash_internal`	✅	6s	4s	+50%
`unpack_hints`	✅	6s	5s	+20%
`mld_compute_pack_z`	✅	5s	6s	-17%
`polyt0_pack`	✅	5s	4s	+25%
`polyveck_pointwise_poly_montgomery`	✅	5s	8s	-38%
`polyveck_sub`	✅	5s	8s	-38%
`polyveck_unpack_eta`	✅	5s	4s	+25%
`sign_signature_extmu`	✅	5s	4s	+25%
`sign_verify_extmu`	✅	5s	7s	-29%
`sign_verify_pre_hash_shake256`	✅	5s	3s	+67%
`intt_native_x86_64`	✅	4s	2s	+100%
`keccakf1600_extract_bytes (big endian)`	✅	4s	2s	+100%
`keccakf1600_xor_bytes (big endian)`	✅	4s	4s	+0%
`keccakf1600x4_permute`	✅	4s	1s	+300%
`make_hint`	✅	4s	2s	+100%
`mld_ct_cmask_neg_i32`	✅	4s	2s	+100%
`mld_ct_get_optblocker_u32`	✅	4s	3s	+33%
`mld_h`	✅	4s	5s	-20%
`mld_prepare_domain_separation_prefix`	✅	4s	3s	+33%
`pack_pk`	✅	4s	2s	+100%
`pack_sk`	✅	4s	3s	+33%
`poly_caddq_c`	✅	4s	6s	-33%
`poly_caddq_native`	✅	4s	2s	+100%
`poly_caddq_native_aarch64`	✅	4s	4s	+0%
`poly_chknorm`	✅	4s	3s	+33%
`poly_ntt_c`	✅	4s	5s	-20%
`poly_ntt_native`	✅	4s	2s	+100%
`poly_pointwise_montgomery`	✅	4s	5s	-20%
`poly_uniform_eta`	✅	4s	5s	-20%
`polyveck_reduce`	✅	4s	8s	-50%
`polyvecl_pack_eta`	✅	4s	4s	+0%
`polyvecl_uniform_gamma1`	✅	4s	4s	+0%
`polyvecl_unpack_eta`	✅	4s	4s	+0%
`polyw1_pack`	✅	4s	2s	+100%
`rej_eta_c`	✅	4s	4s	+0%
`shake256x4_squeezeblocks`	✅	4s	2s	+100%
`sign_signature_pre_hash_internal`	✅	4s	4s	+0%
`sign_signature_pre_hash_shake256`	✅	4s	6s	-33%
`sign_verify`	✅	4s	2s	+100%
`sys_check_capability`	✅	4s	4s	+0%
`caddq`	✅	3s	4s	-25%
`fqscale`	✅	3s	3s	+0%
`keccak_f1600_x1_native_aarch64_v84a`	✅	3s	2s	+50%
`keccak_f1600_x4_native_aarch64_v8a_scalar_hybrid`	✅	3s	3s	+0%
`keccak_f1600_x4_native_aarch64_v8a_v84a_scalar_hybrid`	✅	3s	3s	+0%
`keccak_init`	✅	3s	1s	+200%
`mld_ct_cmask_nonzero_u32`	✅	3s	3s	+0%
`mld_ct_cmask_nonzero_u8`	✅	3s	4s	-25%
`mld_ct_get_optblocker_u8`	✅	3s	4s	-25%
`pack_sig_c`	✅	3s	-	new
`pack_sig_z`	✅	3s	3s	+0%
`poly_decompose_native`	✅	3s	3s	+0%
`poly_make_hint`	✅	3s	2s	+50%
`poly_ntt`	✅	3s	3s	+0%
`poly_uniform`	✅	3s	1s	+200%
`poly_uniform_gamma1`	✅	3s	4s	-25%
`poly_uniform_gamma1_4x`	✅	3s	5s	-40%
`poly_use_hint`	✅	3s	3s	+0%
`poly_use_hint_native`	✅	3s	4s	-25%
`polyt1_pack`	✅	3s	3s	+0%
`polyveck_chknorm`	✅	3s	5s	-40%
`polyveck_pack_eta`	✅	3s	5s	-40%
`polyveck_pack_t0`	✅	3s	5s	-40%
`polyveck_pack_w1`	✅	3s	4s	-25%
`polyveck_unpack_t0`	✅	3s	5s	-40%
`polyvecl_chknorm`	✅	3s	3s	+0%
`polyvecl_pointwise_acc_montgomery`	✅	3s	2s	+50%
`polyvecl_pointwise_acc_montgomery_native`	✅	3s	3s	+0%
`polyvecl_uniform_gamma1_serial`	✅	3s	3s	+0%
`polyz_unpack_native`	✅	3s	4s	-25%
`power2round`	✅	3s	2s	+50%
`shake128_absorb`	✅	3s	3s	+0%
`shake128_finalize`	✅	3s	2s	+50%
`shake128_init`	✅	3s	2s	+50%
`shake128x4_absorb_once`	✅	3s	3s	+0%
`shake256`	✅	3s	2s	+50%
`shake256_finalize`	✅	3s	1s	+200%
`sign_keypair`	✅	3s	3s	+0%
`decompose`	✅	2s	5s	-60%
`keccak_f1600_x1_native_aarch64`	✅	2s	3s	-33%
`keccak_finalize`	✅	2s	3s	-33%
`keccak_squeeze`	✅	2s	4s	-50%
`keccakf1600_xor_bytes`	✅	2s	3s	-33%
`keccakf1600x4_extract_bytes`	✅	2s	5s	-60%
`keccakf1600x4_xor_bytes`	✅	2s	4s	-50%
`mld_ct_sel_int32`	✅	2s	3s	-33%
`mld_keccakf1600_extract_bytes`	✅	2s	3s	-33%
`mld_value_barrier_i64`	✅	2s	2s	+0%
`ntt_native_aarch64`	✅	2s	3s	-33%
`ntt_native_x86_64`	✅	2s	3s	-33%
`pack_sig_h_poly`	✅	2s	-	new
`poly_caddq`	✅	2s	3s	-33%
`poly_chknorm_native`	✅	2s	2s	+0%
`poly_chknorm_native_aarch64`	✅	2s	4s	-50%
`poly_decompose`	✅	2s	3s	-33%
`poly_invntt_tomont`	✅	2s	5s	-60%
`poly_invntt_tomont_native`	✅	2s	4s	-50%
`poly_pointwise_montgomery_native`	✅	2s	1s	+100%
`poly_reduce`	✅	2s	2s	+0%
`poly_shiftl`	✅	2s	2s	+0%
`poly_use_hint_c`	✅	2s	6s	-67%
`polyt1_unpack`	✅	2s	4s	-50%
`polyvec_matrix_pointwise_montgomery_zvec_eager`	✅	2s	-	new
`polyvecl_permute_bitrev_to_custom`	✅	2s	4s	-50%
`polyz_pack`	✅	2s	4s	-50%
`polyz_unpack`	✅	2s	2s	+0%
`reduce32`	✅	2s	3s	-33%
`rej_eta`	✅	2s	2s	+0%
`shake128_release`	✅	2s	2s	+0%
`shake128_squeeze`	✅	2s	3s	-33%
`shake128x4_squeezeblocks`	✅	2s	3s	-33%
`shake256_absorb`	✅	2s	3s	-33%
`shake256_init`	✅	2s	3s	-33%
`shake256x4_absorb_once`	✅	2s	3s	-33%
`unpack_pk`	✅	2s	4s	-50%
`use_hint`	✅	2s	2s	+0%
`keccak_f1600_x4_native_aarch64_v84a`	✅	1s	3s	-67%
`mld_ct_abs_i32`	✅	1s	3s	-67%
`mld_ct_get_optblocker_i64`	✅	1s	2s	-50%
`mld_value_barrier_u32`	✅	1s	3s	-67%
`mld_value_barrier_u8`	✅	1s	2s	-50%
`montgomery_reduce`	✅	1s	1s	+0%
`polyeta_pack`	✅	1s	2s	-50%
`rej_eta_native`	✅	1s	4s	-75%
`shake256_release`	✅	1s	4s	-75%

Introduce mld_zvec following the lazy polyvec pattern (eager / lazy variants with #define dispatch on MLD_CONFIG_REDUCE_RAM): - mld_zvec_init: in eager mode unpacks the full polyvecl, performs the polyvecl-wide infinity-norm bound check, and NTTs in place. In lazy mode it just stores a pointer to the packed signature bytes. - mld_zvec_get_poly: in eager mode copies a single polynomial from the precomputed vector. In lazy mode unpacks one polynomial, performs the per-poly infinity-norm bound check, and NTTs into the caller-provided buffer. The norm check thus moves out of mld_sign_verify_internal into the zvec init / get_poly accessors, so the verify body no longer has to sequence chknorm explicitly. Add a fused matrix-vector helper mld_polyvec_matrix_pointwise_montgomery_zvec used by verify: - The eager variant is a thin wrapper around the existing mld_polyvec_matrix_pointwise_montgomery_eager (z is already NTT'd by mld_zvec_init). - The lazy variant streams z via mld_zvec_get_poly_lazy and generates the matrix on-the-fly column-by-column, accumulating A[*,l] * z[l] into w. In REDUCE_RAM mode this avoids holding the full unpacked polyvecl z in memory at once, reducing verify allocation by 2-5 KiB per parameter set. Signed-off-by: Matthias J. Kannwischer <matthias@zerorisc.com>

mkannwischer added 3 commits April 7, 2026 14:16

Lowram: Share buffers with non-overlapping lifetimes in verify

9fb9059

Signed-off-by: Matthias J. Kannwischer <matthias@zerorisc.com>

mkannwischer added the low-ram label Apr 7, 2026

mkannwischer force-pushed the verify-stream-z branch from d7de92e to 0f60231 Compare April 7, 2026 06:37

mkannwischer force-pushed the verify-stream-z branch from 0f60231 to 947662d Compare April 7, 2026 07:21

mkannwischer marked this pull request as ready for review April 7, 2026 08:01

mkannwischer requested a review from a team as a code owner April 7, 2026 08:01

mkannwischer assigned hanno-becker Apr 7, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

lowmem: Unpack z lazily in verification#1025

lowmem: Unpack z lazily in verification#1025
mkannwischer wants to merge 4 commits intomainfrom
verify-stream-z

mkannwischer commented Apr 7, 2026

Uh oh!

oqs-bot commented Apr 7, 2026 •

edited

Loading

Uh oh!

oqs-bot commented Apr 7, 2026 •

edited

Loading

Uh oh!

oqs-bot commented Apr 7, 2026 •

edited

Loading

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

mkannwischer commented Apr 7, 2026

Uh oh!

oqs-bot commented Apr 7, 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 7, 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 7, 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 7, 2026 •

edited

Loading

oqs-bot commented Apr 7, 2026 •

edited

Loading

oqs-bot commented Apr 7, 2026 •

edited

Loading