lowram: Stream matrix A element-by-element to reduce memory by mkannwischer · Pull Request #1019 · pq-code-package/mldsa-native

mkannwischer · 2026-04-05T06:24:46Z

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.

Hoisted out from PoC: Reduce large struct allocations to <= 13/17/21 KiB for ML-DSA-44/65/87 #1005
Depends on lowram: Compute h incrementally in signing #1015

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>

oqs-bot · 2026-04-05T07:28:27Z

CBMC Results (ML-DSA-65)

⚠️ Attention Required

Proof	Status	Current	Previous	Change
`polymat_permute_bitrev_to_custom`	⚠️	337s	32s	+953%
`sign_signature_internal`	⚠️	38s	25s	+52%

Full Results (177 proofs)

Proof	Status	Current	Previous	Change
`TOTAL`	✅	2379s	2543s	-6.4%
`polymat_permute_bitrev_to_custom`	⚠️	337s	32s	+953%
`sign_verify_internal`	✅	227s	350s	-35%
`polyvec_matrix_expand_eager`	✅	148s	-	new
`poly_pointwise_montgomery_c`	✅	147s	165s	-11%
`rej_uniform_native`	✅	140s	151s	-7%
`polyvecl_pointwise_acc_montgomery_c`	✅	113s	201s	-44%
`mld_invntt_layer`	✅	94s	99s	-5%
`mld_attempt_signature_generation`	✅	82s	295s	-72%
`polyvec_matrix_expand_eager_serial`	✅	77s	-	new
`mld_ct_memcmp`	✅	66s	82s	-20%
`mld_ntt_layer`	✅	53s	58s	-9%
`sign_signature_internal`	⚠️	38s	25s	+52%
`rej_uniform`	✅	25s	23s	+9%
`mld_compute_t0_t1_tr_from_sk_components`	✅	22s	25s	-12%
`poly_chknorm_c`	✅	20s	20s	+0%
`fqmul`	✅	18s	20s	-10%
`poly_uniform_eta_4x`	✅	16s	16s	+0%
`poly_uniform_4x`	✅	15s	16s	-6%
`polyveck_decompose`	✅	14s	16s	-12%
`rej_uniform_c`	✅	14s	17s	-18%
`polyt0_unpack`	✅	13s	16s	-19%
`polyvec_matrix_pointwise_montgomery_eager`	✅	13s	-	new
`polyveck_add`	✅	13s	9s	+44%
`keccakf1600x4_permute_native`	✅	12s	14s	-14%
`mld_ntt_butterfly_block`	✅	12s	14s	-14%
`polyveck_power2round`	✅	12s	10s	+20%
`keccak_absorb_once_x4`	✅	11s	9s	+22%
`mld_polyvecl_permute_bitrev_to_custom_native`	✅	11s	9s	+22%
`polyveck_use_hint`	✅	11s	12s	-8%
`polyvecl_chknorm`	✅	11s	13s	-15%
`poly_add`	✅	10s	12s	-17%
`poly_decompose_c`	✅	10s	7s	+43%
`polyveck_ntt`	✅	10s	10s	+0%
`sign`	✅	10s	7s	+43%
`sign_pk_from_sk`	✅	10s	8s	+25%
`polyveck_pointwise_poly_montgomery`	✅	9s	8s	+12%
`keccakf1600_permute`	✅	8s	7s	+14%
`mld_check_pct`	✅	8s	9s	-11%
`sign_open`	✅	8s	5s	+60%
`keccak_absorb`	✅	7s	6s	+17%
`keccakf1600_permute_native`	✅	7s	8s	-12%
`mld_h`	✅	7s	7s	+0%
`poly_invntt_tomont_c`	✅	7s	10s	-30%
`polyveck_caddq`	✅	7s	7s	+0%
`polyveck_invntt_tomont`	✅	7s	8s	-12%
`polyveck_reduce`	✅	7s	8s	-12%
`polyveck_shiftl`	✅	7s	7s	+0%
`polyveck_sub`	✅	7s	12s	-42%
`polyvecl_ntt`	✅	7s	9s	-22%
`unpack_sk`	✅	7s	9s	-22%
`mld_sample_s1_s2`	✅	6s	4s	+50%
`pack_sig_c`	✅	6s	-	new
`pack_sk`	✅	6s	3s	+100%
`poly_use_hint_native`	✅	6s	3s	+100%
`polyveck_pack_w1`	✅	6s	2s	+200%
`polyvecl_uniform_gamma1_serial`	✅	6s	5s	+20%
`sign_keypair_internal`	✅	6s	5s	+20%
`sign_signature_pre_hash_internal`	✅	6s	5s	+20%
`sign_signature_pre_hash_shake256`	✅	6s	5s	+20%
`unpack_hints`	✅	6s	4s	+50%
`keccak_squeezeblocks_x4`	✅	5s	7s	-29%
`keccakf1600_extract_bytes (big endian)`	✅	5s	3s	+67%
`mld_compute_pack_z`	✅	5s	7s	-29%
`mld_ct_abs_i32`	✅	5s	2s	+150%
`mld_ct_cmask_nonzero_u8`	✅	5s	1s	+400%
`ntt_native_aarch64`	✅	5s	3s	+67%
`poly_invntt_tomont_native`	✅	5s	3s	+67%
`poly_ntt_c`	✅	5s	3s	+67%
`poly_power2round`	✅	5s	5s	+0%
`poly_reduce`	✅	5s	2s	+150%
`poly_sub`	✅	5s	2s	+150%
`poly_uniform`	✅	5s	7s	-29%
`poly_uniform_gamma1`	✅	5s	4s	+25%
`poly_use_hint_c`	✅	5s	4s	+25%
`polyveck_unpack_t0`	✅	5s	4s	+25%
`polyvecl_unpack_eta`	✅	5s	4s	+25%
`sign_keypair`	✅	5s	4s	+25%
`sign_signature`	✅	5s	5s	+0%
`sign_verify`	✅	5s	5s	+0%
`sign_verify_extmu`	✅	5s	3s	+67%
`sign_verify_pre_hash_internal`	✅	5s	6s	-17%
`use_hint`	✅	5s	3s	+67%
`caddq`	✅	4s	4s	+0%
`fqscale`	✅	4s	3s	+33%
`keccakf1600_xor_bytes`	✅	4s	4s	+0%
`keccakf1600_xor_bytes (big endian)`	✅	4s	3s	+33%
`mld_sample_s1_s2_serial`	✅	4s	5s	-20%
`pack_sig_z`	✅	4s	4s	+0%
`poly_caddq_c`	✅	4s	6s	-33%
`poly_challenge`	✅	4s	5s	-20%
`poly_ntt`	✅	4s	3s	+33%
`poly_ntt_native`	✅	4s	4s	+0%
`poly_pointwise_montgomery`	✅	4s	2s	+100%
`poly_uniform_eta`	✅	4s	5s	-20%
`poly_uniform_gamma1_4x`	✅	4s	4s	+0%
`polyeta_pack`	✅	4s	3s	+33%
`polyeta_unpack`	✅	4s	4s	+0%
`polyt0_pack`	✅	4s	4s	+0%
`polyveck_chknorm`	✅	4s	4s	+0%
`polyveck_unpack_eta`	✅	4s	3s	+33%
`polyvecl_permute_bitrev_to_custom`	✅	4s	3s	+33%
`polyvecl_pointwise_acc_montgomery_native`	✅	4s	7s	-43%
`rej_eta_c`	✅	4s	3s	+33%
`rej_eta_native`	✅	4s	6s	-33%
`shake256`	✅	4s	3s	+33%
`shake256x4_squeezeblocks`	✅	4s	2s	+100%
`decompose`	✅	3s	2s	+50%
`keccak_init`	✅	3s	2s	+50%
`keccak_squeeze`	✅	3s	2s	+50%
`keccakf1600x4_permute`	✅	3s	2s	+50%
`mld_ct_get_optblocker_u32`	✅	3s	2s	+50%
`mld_ct_sel_int32`	✅	3s	2s	+50%
`mld_keccakf1600_extract_bytes`	✅	3s	3s	+0%
`pack_pk`	✅	3s	4s	-25%
`pack_sig_h_poly`	✅	3s	-	new
`poly_caddq_native_aarch64`	✅	3s	4s	-25%
`poly_chknorm`	✅	3s	1s	+200%
`poly_invntt_tomont`	✅	3s	5s	-40%
`poly_make_hint`	✅	3s	3s	+0%
`poly_pointwise_montgomery_native`	✅	3s	3s	+0%
`poly_shiftl`	✅	3s	4s	-25%
`poly_use_hint`	✅	3s	3s	+0%
`polyt1_unpack`	✅	3s	5s	-40%
`polyveck_pack_eta`	✅	3s	2s	+50%
`polyvecl_pack_eta`	✅	3s	4s	-25%
`polyvecl_uniform_gamma1`	✅	3s	2s	+50%
`polyvecl_unpack_z`	✅	3s	3s	+0%
`polyw1_pack`	✅	3s	3s	+0%
`polyz_unpack_native`	✅	3s	3s	+0%
`power2round`	✅	3s	1s	+200%
`reduce32`	✅	3s	1s	+200%
`shake128_squeeze`	✅	3s	3s	+0%
`shake256_absorb`	✅	3s	3s	+0%
`shake256_release`	✅	3s	3s	+0%
`sign_signature_extmu`	✅	3s	5s	-40%
`sign_verify_pre_hash_shake256`	✅	3s	3s	+0%
`sys_check_capability`	✅	3s	4s	-25%
`unpack_sig`	✅	3s	4s	-25%
`intt_native_x86_64`	✅	2s	6s	-67%
`keccakf1600x4_extract_bytes`	✅	2s	2s	+0%
`keccakf1600x4_xor_bytes`	✅	2s	2s	+0%
`make_hint`	✅	2s	2s	+0%
`mld_ct_cmask_neg_i32`	✅	2s	1s	+100%
`mld_ct_cmask_nonzero_u32`	✅	2s	3s	-33%
`mld_ct_get_optblocker_i64`	✅	2s	1s	+100%
`mld_ct_get_optblocker_u8`	✅	2s	2s	+0%
`mld_prepare_domain_separation_prefix`	✅	2s	4s	-50%
`mld_value_barrier_u8`	✅	2s	2s	+0%
`montgomery_reduce`	✅	2s	4s	-50%
`ntt_native_x86_64`	✅	2s	5s	-60%
`poly_caddq`	✅	2s	5s	-60%
`poly_caddq_native`	✅	2s	2s	+0%
`poly_chknorm_native`	✅	2s	5s	-60%
`poly_chknorm_native_aarch64`	✅	2s	3s	-33%
`poly_decompose_native`	✅	2s	4s	-50%
`polyt1_pack`	✅	2s	4s	-50%
`polyveck_pack_t0`	✅	2s	3s	-33%
`polyvecl_pointwise_acc_montgomery`	✅	2s	4s	-50%
`polyz_pack`	✅	2s	3s	-33%
`polyz_unpack`	✅	2s	4s	-50%
`polyz_unpack_c`	✅	2s	3s	-33%
`rej_eta`	✅	2s	2s	+0%
`shake128_finalize`	✅	2s	2s	+0%
`shake128_init`	✅	2s	3s	-33%
`shake128_release`	✅	2s	2s	+0%
`shake128x4_absorb_once`	✅	2s	2s	+0%
`shake128x4_squeezeblocks`	✅	2s	3s	-33%
`shake256_finalize`	✅	2s	2s	+0%
`shake256_init`	✅	2s	4s	-50%
`shake256x4_absorb_once`	✅	2s	5s	-60%
`unpack_pk`	✅	2s	5s	-60%
`keccak_finalize`	✅	1s	3s	-67%
`mld_value_barrier_i64`	✅	1s	1s	+0%
`mld_value_barrier_u32`	✅	1s	4s	-75%
`poly_decompose`	✅	1s	4s	-75%
`shake128_absorb`	✅	1s	2s	-50%
`shake256_squeeze`	✅	1s	3s	-67%

oqs-bot · 2026-04-05T07:28:33Z

CBMC Results (ML-DSA-44)

Full Results (177 proofs)

Proof	Status	Current	Previous	Change
`TOTAL`	✅	1953s	2020s	-3.3%
`sign_verify_internal`	✅	209s	179s	+17%
`polyvecl_pointwise_acc_montgomery_c`	✅	178s	205s	-13%
`poly_pointwise_montgomery_c`	✅	163s	150s	+9%
`rej_uniform_native`	✅	146s	144s	+1%
`mld_invntt_layer`	✅	92s	87s	+6%
`mld_attempt_signature_generation`	✅	89s	251s	-65%
`mld_ct_memcmp`	✅	82s	77s	+6%
`mld_ntt_layer`	✅	59s	54s	+9%
`rej_uniform`	✅	24s	21s	+14%
`sign_signature_internal`	✅	23s	19s	+21%
`poly_chknorm_c`	✅	22s	21s	+5%
`polyvec_matrix_expand_eager`	✅	21s	-	new
`fqmul`	✅	20s	18s	+11%
`rej_uniform_c`	✅	20s	14s	+43%
`mld_compute_t0_t1_tr_from_sk_components`	✅	18s	12s	+50%
`polymat_permute_bitrev_to_custom`	✅	18s	14s	+29%
`poly_uniform_eta_4x`	✅	17s	20s	-15%
`polyeta_unpack`	✅	17s	16s	+6%
`poly_uniform_4x`	✅	15s	16s	-6%
`polyt0_unpack`	✅	14s	14s	+0%
`keccakf1600x4_permute_native`	✅	13s	14s	-7%
`mld_ntt_butterfly_block`	✅	13s	12s	+8%
`poly_add`	✅	13s	14s	-7%
`polyvec_matrix_expand_eager_serial`	✅	13s	-	new
`polyz_unpack_c`	✅	13s	11s	+18%
`mld_check_pct`	✅	12s	8s	+50%
`polyveck_power2round`	✅	11s	10s	+10%
`keccak_absorb_once_x4`	✅	10s	11s	-9%
`polyveck_chknorm`	✅	10s	6s	+67%
`polyveck_decompose`	✅	10s	6s	+67%
`keccakf1600_permute`	✅	9s	8s	+12%
`unpack_hints`	✅	9s	5s	+80%
`keccak_absorb`	✅	8s	7s	+14%
`keccakf1600_permute_native`	✅	8s	9s	-11%
`mld_polyvecl_permute_bitrev_to_custom_native`	✅	8s	8s	+0%
`mld_sample_s1_s2_serial`	✅	8s	4s	+100%
`poly_uniform_gamma1_4x`	✅	8s	6s	+33%
`poly_use_hint_c`	✅	8s	6s	+33%
`polyvec_matrix_pointwise_montgomery_eager`	✅	8s	-	new
`polyveck_shiftl`	✅	8s	7s	+14%
`sign`	✅	8s	5s	+60%
`sign_signature_pre_hash_internal`	✅	8s	3s	+167%
`poly_use_hint_native`	✅	7s	3s	+133%
`polyveck_reduce`	✅	7s	6s	+17%
`polyvecl_chknorm`	✅	7s	5s	+40%
`sign_pk_from_sk`	✅	7s	7s	+0%
`sign_signature_pre_hash_shake256`	✅	7s	4s	+75%
`unpack_sk`	✅	7s	6s	+17%
`keccak_squeezeblocks_x4`	✅	6s	6s	+0%
`mld_compute_pack_z`	✅	6s	4s	+50%
`mld_sample_s1_s2`	✅	6s	4s	+50%
`poly_invntt_tomont_c`	✅	6s	7s	-14%
`poly_sub`	✅	6s	4s	+50%
`polyveck_add`	✅	6s	8s	-25%
`polyveck_caddq`	✅	6s	4s	+50%
`polyvecl_ntt`	✅	6s	4s	+50%
`polyvecl_pack_eta`	✅	6s	6s	+0%
`rej_eta_native`	✅	6s	6s	+0%
`shake256_finalize`	✅	6s	2s	+200%
`sign_verify_extmu`	✅	6s	4s	+50%
`sign_verify_pre_hash_internal`	✅	6s	6s	+0%
`caddq`	✅	5s	4s	+25%
`mld_ct_cmask_nonzero_u32`	✅	5s	2s	+150%
`mld_h`	✅	5s	3s	+67%
`ntt_native_aarch64`	✅	5s	2s	+150%
`poly_caddq_c`	✅	5s	5s	+0%
`poly_pointwise_montgomery_native`	✅	5s	4s	+25%
`poly_uniform`	✅	5s	4s	+25%
`poly_uniform_eta`	✅	5s	6s	-17%
`polyveck_ntt`	✅	5s	5s	+0%
`polyveck_pack_eta`	✅	5s	3s	+67%
`polyveck_sub`	✅	5s	5s	+0%
`polyveck_use_hint`	✅	5s	6s	-17%
`sign_keypair_internal`	✅	5s	5s	+0%
`sign_verify_pre_hash_shake256`	✅	5s	4s	+25%
`unpack_sig`	✅	5s	4s	+25%
`fqscale`	✅	4s	4s	+0%
`keccak_squeeze`	✅	4s	1s	+300%
`keccakf1600x4_extract_bytes`	✅	4s	3s	+33%
`make_hint`	✅	4s	4s	+0%
`mld_ct_abs_i32`	✅	4s	3s	+33%
`mld_ct_sel_int32`	✅	4s	3s	+33%
`mld_prepare_domain_separation_prefix`	✅	4s	7s	-43%
`ntt_native_x86_64`	✅	4s	2s	+100%
`poly_caddq`	✅	4s	4s	+0%
`poly_caddq_native`	✅	4s	5s	-20%
`poly_challenge`	✅	4s	4s	+0%
`poly_decompose`	✅	4s	2s	+100%
`poly_ntt_c`	✅	4s	2s	+100%
`poly_pointwise_montgomery`	✅	4s	1s	+300%
`poly_reduce`	✅	4s	2s	+100%
`poly_shiftl`	✅	4s	5s	-20%
`poly_uniform_gamma1`	✅	4s	3s	+33%
`polyeta_pack`	✅	4s	3s	+33%
`polyveck_invntt_tomont`	✅	4s	3s	+33%
`polyveck_pack_t0`	✅	4s	4s	+0%
`polyveck_pack_w1`	✅	4s	2s	+100%
`polyvecl_uniform_gamma1`	✅	4s	4s	+0%
`polyvecl_uniform_gamma1_serial`	✅	4s	4s	+0%
`polyz_pack`	✅	4s	4s	+0%
`polyz_unpack`	✅	4s	2s	+100%
`polyz_unpack_native`	✅	4s	3s	+33%
`shake256x4_absorb_once`	✅	4s	2s	+100%
`sign_keypair`	✅	4s	3s	+33%
`sign_signature`	✅	4s	4s	+0%
`decompose`	✅	3s	4s	-25%
`intt_native_x86_64`	✅	3s	2s	+50%
`keccak_init`	✅	3s	2s	+50%
`keccakf1600_extract_bytes (big endian)`	✅	3s	1s	+200%
`keccakf1600_xor_bytes`	✅	3s	4s	-25%
`mld_ct_cmask_neg_i32`	✅	3s	2s	+50%
`mld_ct_get_optblocker_i64`	✅	3s	3s	+0%
`mld_keccakf1600_extract_bytes`	✅	3s	2s	+50%
`mld_value_barrier_u8`	✅	3s	3s	+0%
`montgomery_reduce`	✅	3s	2s	+50%
`pack_sig_c`	✅	3s	-	new
`pack_sig_z`	✅	3s	4s	-25%
`pack_sk`	✅	3s	2s	+50%
`poly_chknorm_native`	✅	3s	4s	-25%
`poly_chknorm_native_aarch64`	✅	3s	3s	+0%
`poly_decompose_native`	✅	3s	5s	-40%
`poly_invntt_tomont`	✅	3s	4s	-25%
`poly_make_hint`	✅	3s	3s	+0%
`poly_ntt_native`	✅	3s	4s	-25%
`poly_power2round`	✅	3s	5s	-40%
`poly_use_hint`	✅	3s	3s	+0%
`polyt0_pack`	✅	3s	3s	+0%
`polyt1_pack`	✅	3s	1s	+200%
`polyt1_unpack`	✅	3s	1s	+200%
`polyveck_pointwise_poly_montgomery`	✅	3s	4s	-25%
`polyvecl_permute_bitrev_to_custom`	✅	3s	3s	+0%
`polyvecl_pointwise_acc_montgomery`	✅	3s	4s	-25%
`polyvecl_pointwise_acc_montgomery_native`	✅	3s	3s	+0%
`polyw1_pack`	✅	3s	3s	+0%
`rej_eta_c`	✅	3s	6s	-50%
`shake128_absorb`	✅	3s	6s	-50%
`shake128x4_squeezeblocks`	✅	3s	1s	+200%
`shake256`	✅	3s	2s	+50%
`shake256_release`	✅	3s	2s	+50%
`shake256x4_squeezeblocks`	✅	3s	4s	-25%
`sign_open`	✅	3s	4s	-25%
`sign_signature_extmu`	✅	3s	2s	+50%
`sign_verify`	✅	3s	7s	-57%
`sys_check_capability`	✅	3s	4s	-25%
`keccak_finalize`	✅	2s	2s	+0%
`keccakf1600x4_permute`	✅	2s	2s	+0%
`keccakf1600x4_xor_bytes`	✅	2s	1s	+100%
`mld_ct_cmask_nonzero_u8`	✅	2s	5s	-60%
`mld_ct_get_optblocker_u32`	✅	2s	2s	+0%
`mld_value_barrier_u32`	✅	2s	5s	-60%
`pack_pk`	✅	2s	6s	-67%
`pack_sig_h_poly`	✅	2s	-	new
`poly_caddq_native_aarch64`	✅	2s	4s	-50%
`poly_chknorm`	✅	2s	4s	-50%
`poly_decompose_c`	✅	2s	4s	-50%
`poly_invntt_tomont_native`	✅	2s	2s	+0%
`poly_ntt`	✅	2s	4s	-50%
`polyveck_unpack_eta`	✅	2s	2s	+0%
`polyveck_unpack_t0`	✅	2s	4s	-50%
`polyvecl_unpack_eta`	✅	2s	4s	-50%
`polyvecl_unpack_z`	✅	2s	4s	-50%
`power2round`	✅	2s	3s	-33%
`reduce32`	✅	2s	2s	+0%
`rej_eta`	✅	2s	3s	-33%
`shake128_finalize`	✅	2s	2s	+0%
`shake128_init`	✅	2s	2s	+0%
`shake128_squeeze`	✅	2s	1s	+100%
`shake128x4_absorb_once`	✅	2s	2s	+0%
`shake256_absorb`	✅	2s	2s	+0%
`shake256_init`	✅	2s	2s	+0%
`shake256_squeeze`	✅	2s	1s	+100%
`unpack_pk`	✅	2s	5s	-60%
`use_hint`	✅	2s	1s	+100%
`keccakf1600_xor_bytes (big endian)`	✅	1s	2s	-50%
`mld_ct_get_optblocker_u8`	✅	1s	3s	-67%
`mld_value_barrier_i64`	✅	1s	1s	+0%
`shake128_release`	✅	1s	2s	-50%

oqs-bot · 2026-04-05T07:33:03Z

CBMC Results (ML-DSA-87)

Full Results (177 proofs)

Proof	Status	Current	Previous	Change
`TOTAL`	✅	2412s	2688s	-10.3%
`polyvecl_pointwise_acc_montgomery_c`	✅	419s	310s	+35%
`polyvec_matrix_expand_eager`	✅	160s	-	new
`poly_pointwise_montgomery_c`	✅	155s	174s	-11%
`rej_uniform_native`	✅	141s	153s	-8%
`mld_attempt_signature_generation`	✅	125s	239s	-48%
`polyvec_matrix_expand_eager_serial`	✅	115s	-	new
`sign_verify_internal`	✅	110s	215s	-49%
`mld_invntt_layer`	✅	92s	101s	-9%
`mld_ct_memcmp`	✅	72s	84s	-14%
`mld_ntt_layer`	✅	54s	61s	-11%
`sign_signature_internal`	✅	47s	39s	+21%
`mld_compute_t0_t1_tr_from_sk_components`	✅	29s	26s	+12%
`polymat_permute_bitrev_to_custom`	✅	25s	47s	-47%
`poly_chknorm_c`	✅	20s	22s	-9%
`rej_uniform`	✅	20s	24s	-17%
`polyeta_unpack`	✅	19s	18s	+6%
`fqmul`	✅	18s	22s	-18%
`mld_check_pct`	✅	17s	7s	+143%
`polyveck_decompose`	✅	16s	62s	-74%
`poly_uniform_4x`	✅	15s	17s	-12%
`poly_uniform_eta_4x`	✅	15s	17s	-12%
`rej_uniform_c`	✅	15s	15s	+0%
`keccakf1600x4_permute_native`	✅	13s	14s	-7%
`polyvec_matrix_pointwise_montgomery_eager`	✅	13s	-	new
`polyt0_unpack`	✅	12s	15s	-20%
`polyveck_add`	✅	12s	12s	+0%
`mld_ntt_butterfly_block`	✅	11s	14s	-21%
`mld_polyvecl_permute_bitrev_to_custom_native`	✅	11s	12s	-8%
`polyveck_power2round`	✅	11s	10s	+10%
`polyvecl_ntt`	✅	11s	12s	-8%
`poly_add`	✅	10s	14s	-29%
`polyz_unpack_c`	✅	10s	9s	+11%
`sign_pk_from_sk`	✅	10s	9s	+11%
`keccak_absorb_once_x4`	✅	9s	11s	-18%
`keccakf1600_permute_native`	✅	9s	9s	+0%
`polyveck_caddq`	✅	9s	10s	-10%
`polyveck_ntt`	✅	9s	9s	+0%
`polyveck_use_hint`	✅	9s	10s	-10%
`keccak_absorb`	✅	8s	9s	-11%
`keccakf1600_permute`	✅	8s	7s	+14%
`polyveck_shiftl`	✅	8s	6s	+33%
`sign_open`	✅	8s	6s	+33%
`keccak_squeezeblocks_x4`	✅	7s	5s	+40%
`poly_caddq_c`	✅	7s	8s	-12%
`poly_challenge`	✅	7s	4s	+75%
`poly_decompose_c`	✅	7s	8s	-12%
`poly_uniform_eta`	✅	7s	4s	+75%
`polyt0_pack`	✅	7s	2s	+250%
`polyveck_sub`	✅	7s	7s	+0%
`sign`	✅	7s	7s	+0%
`sign_verify_extmu`	✅	7s	3s	+133%
`unpack_sk`	✅	7s	8s	-12%
`mld_h`	✅	6s	6s	+0%
`mld_sample_s1_s2`	✅	6s	6s	+0%
`mld_sample_s1_s2_serial`	✅	6s	7s	-14%
`poly_decompose_native`	✅	6s	4s	+50%
`poly_invntt_tomont_c`	✅	6s	8s	-25%
`poly_use_hint_native`	✅	6s	3s	+100%
`polyveck_invntt_tomont`	✅	6s	6s	+0%
`polyveck_pointwise_poly_montgomery`	✅	6s	7s	-14%
`sign_keypair_internal`	✅	6s	6s	+0%
`sign_verify`	✅	6s	5s	+20%
`sign_verify_pre_hash_internal`	✅	6s	4s	+50%
`make_hint`	✅	5s	3s	+67%
`mld_compute_pack_z`	✅	5s	5s	+0%
`montgomery_reduce`	✅	5s	4s	+25%
`poly_chknorm`	✅	5s	4s	+25%
`poly_chknorm_native_aarch64`	✅	5s	3s	+67%
`poly_decompose`	✅	5s	2s	+150%
`poly_power2round`	✅	5s	5s	+0%
`polyveck_chknorm`	✅	5s	3s	+67%
`polyveck_reduce`	✅	5s	10s	-50%
`polyveck_unpack_eta`	✅	5s	5s	+0%
`polyvecl_pack_eta`	✅	5s	2s	+150%
`polyvecl_uniform_gamma1_serial`	✅	5s	4s	+25%
`polyvecl_unpack_eta`	✅	5s	7s	-29%
`shake128_finalize`	✅	5s	3s	+67%
`shake128x4_squeezeblocks`	✅	5s	1s	+400%
`sign_keypair`	✅	5s	5s	+0%
`sign_signature_pre_hash_internal`	✅	5s	4s	+25%
`sign_signature_pre_hash_shake256`	✅	5s	7s	-29%
`unpack_hints`	✅	5s	6s	-17%
`decompose`	✅	4s	3s	+33%
`keccak_squeeze`	✅	4s	4s	+0%
`keccakf1600x4_permute`	✅	4s	3s	+33%
`mld_ct_get_optblocker_u32`	✅	4s	3s	+33%
`mld_prepare_domain_separation_prefix`	✅	4s	5s	-20%
`mld_value_barrier_i64`	✅	4s	5s	-20%
`ntt_native_aarch64`	✅	4s	4s	+0%
`pack_sig_z`	✅	4s	5s	-20%
`poly_caddq_native`	✅	4s	6s	-33%
`poly_caddq_native_aarch64`	✅	4s	3s	+33%
`poly_invntt_tomont_native`	✅	4s	1s	+300%
`poly_ntt_c`	✅	4s	4s	+0%
`poly_pointwise_montgomery_native`	✅	4s	5s	-20%
`poly_sub`	✅	4s	5s	-20%
`poly_uniform`	✅	4s	4s	+0%
`poly_uniform_gamma1`	✅	4s	2s	+100%
`polyeta_pack`	✅	4s	3s	+33%
`polyveck_pack_eta`	✅	4s	2s	+100%
`polyvecl_chknorm`	✅	4s	4s	+0%
`polyvecl_uniform_gamma1`	✅	4s	4s	+0%
`polyw1_pack`	✅	4s	3s	+33%
`shake256_absorb`	✅	4s	4s	+0%
`shake256_init`	✅	4s	3s	+33%
`shake256x4_absorb_once`	✅	4s	2s	+100%
`sys_check_capability`	✅	4s	3s	+33%
`use_hint`	✅	4s	2s	+100%
`caddq`	✅	3s	4s	-25%
`fqscale`	✅	3s	4s	-25%
`keccak_init`	✅	3s	2s	+50%
`keccakf1600_extract_bytes (big endian)`	✅	3s	3s	+0%
`keccakf1600_xor_bytes`	✅	3s	1s	+200%
`keccakf1600_xor_bytes (big endian)`	✅	3s	2s	+50%
`keccakf1600x4_xor_bytes`	✅	3s	5s	-40%
`mld_ct_cmask_nonzero_u32`	✅	3s	6s	-50%
`mld_ct_get_optblocker_i64`	✅	3s	3s	+0%
`mld_ct_get_optblocker_u8`	✅	3s	3s	+0%
`mld_ct_sel_int32`	✅	3s	1s	+200%
`ntt_native_x86_64`	✅	3s	4s	-25%
`pack_sig_c`	✅	3s	-	new
`pack_sig_h_poly`	✅	3s	-	new
`pack_sk`	✅	3s	4s	-25%
`poly_chknorm_native`	✅	3s	4s	-25%
`poly_make_hint`	✅	3s	5s	-40%
`poly_ntt_native`	✅	3s	2s	+50%
`poly_pointwise_montgomery`	✅	3s	2s	+50%
`poly_reduce`	✅	3s	4s	-25%
`poly_use_hint_c`	✅	3s	3s	+0%
`polyt1_unpack`	✅	3s	3s	+0%
`polyveck_pack_w1`	✅	3s	5s	-40%
`polyveck_unpack_t0`	✅	3s	5s	-40%
`polyvecl_permute_bitrev_to_custom`	✅	3s	3s	+0%
`polyvecl_pointwise_acc_montgomery`	✅	3s	4s	-25%
`polyvecl_unpack_z`	✅	3s	3s	+0%
`polyz_unpack_native`	✅	3s	3s	+0%
`power2round`	✅	3s	3s	+0%
`reduce32`	✅	3s	2s	+50%
`rej_eta_native`	✅	3s	8s	-62%
`shake128_init`	✅	3s	2s	+50%
`shake128_release`	✅	3s	3s	+0%
`shake256_finalize`	✅	3s	3s	+0%
`shake256_release`	✅	3s	2s	+50%
`shake256_squeeze`	✅	3s	4s	-25%
`shake256x4_squeezeblocks`	✅	3s	4s	-25%
`sign_signature`	✅	3s	7s	-57%
`sign_signature_extmu`	✅	3s	4s	-25%
`unpack_pk`	✅	3s	3s	+0%
`unpack_sig`	✅	3s	4s	-25%
`intt_native_x86_64`	✅	2s	3s	-33%
`keccak_finalize`	✅	2s	4s	-50%
`keccakf1600x4_extract_bytes`	✅	2s	1s	+100%
`mld_ct_abs_i32`	✅	2s	3s	-33%
`mld_ct_cmask_neg_i32`	✅	2s	2s	+0%
`mld_ct_cmask_nonzero_u8`	✅	2s	4s	-50%
`mld_keccakf1600_extract_bytes`	✅	2s	2s	+0%
`mld_value_barrier_u8`	✅	2s	2s	+0%
`pack_pk`	✅	2s	4s	-50%
`poly_caddq`	✅	2s	5s	-60%
`poly_invntt_tomont`	✅	2s	5s	-60%
`poly_ntt`	✅	2s	5s	-60%
`poly_shiftl`	✅	2s	5s	-60%
`poly_uniform_gamma1_4x`	✅	2s	5s	-60%
`poly_use_hint`	✅	2s	2s	+0%
`polyt1_pack`	✅	2s	3s	-33%
`polyveck_pack_t0`	✅	2s	2s	+0%
`polyvecl_pointwise_acc_montgomery_native`	✅	2s	3s	-33%
`polyz_pack`	✅	2s	4s	-50%
`rej_eta`	✅	2s	4s	-50%
`rej_eta_c`	✅	2s	3s	-33%
`shake128_absorb`	✅	2s	4s	-50%
`shake128_squeeze`	✅	2s	3s	-33%
`shake128x4_absorb_once`	✅	2s	2s	+0%
`shake256`	✅	2s	4s	-50%
`sign_verify_pre_hash_shake256`	✅	2s	4s	-50%
`mld_value_barrier_u32`	✅	1s	5s	-80%
`polyz_unpack`	✅	1s	5s	-80%

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>

mkannwischer added the low-ram label Apr 5, 2026

mkannwischer force-pushed the lowram-stream-a branch 2 times, most recently from 7013003 to 5ccd3c2 Compare April 5, 2026 07:10

mkannwischer force-pushed the lowram-stream-a branch 3 times, most recently from 9fc4e08 to 3acbf60 Compare April 5, 2026 10:59

mkannwischer force-pushed the lowram-stream-a branch from 3acbf60 to 9ba57ff Compare April 5, 2026 12:59

mkannwischer marked this pull request as ready for review April 6, 2026 00:48

mkannwischer requested a review from a team as a code owner April 6, 2026 00:48

mkannwischer assigned hanno-becker Apr 6, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

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

lowram: Stream matrix A element-by-element to reduce memory#1019
mkannwischer wants to merge 2 commits intomainfrom
lowram-stream-a

mkannwischer commented Apr 5, 2026

Uh oh!

oqs-bot commented Apr 5, 2026 •

edited

Loading

Uh oh!

oqs-bot commented Apr 5, 2026 •

edited

Loading

Uh oh!

oqs-bot commented Apr 5, 2026 •

edited

Loading

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

mkannwischer commented Apr 5, 2026

Uh oh!

oqs-bot commented Apr 5, 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 5, 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 5, 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 5, 2026 •

edited

Loading

oqs-bot commented Apr 5, 2026 •

edited

Loading

oqs-bot commented Apr 5, 2026 •

edited

Loading