Skip to content

SLOTHY: Superoptimize AArch64 INTT #748

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
mkannwischer merged 1 commit into from
Dec 1, 2025
Merged

SLOTHY: Superoptimize AArch64 INTT #748

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 11 additions & 11 deletions dev/aarch64_clean/src/intt.S
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -230,17 +230,6 @@ MLD_ASM_FN_SYMBOL(intt_asm)
movk wtmp, #0x7f, lsl #16
dup consts.4s, wtmp

/* check-magic: 16382 == pow(2, 32-8, MLDSA_Q) */
// Load scaling factor 1/256*R
mov wtmp, #16382 // 2^(32 - 8) mod Q
dup ninv.4s, wtmp

/* check-magic: 4197891 == round(16382 * 2**31 / MLDSA_Q) */
// Load Barrett constant for 1/256*R
movz wtmp, #0x0e03
movk wtmp, #0x40, lsl #16
dup ninv_tw.4s, wtmp

mov inp, in
mov count, #16

Expand Down Expand Up @@ -308,6 +297,17 @@ intt_layer5678_start:

// ---------------------------------------------------------------------

/* check-magic: 16382 == pow(2, 32-8, MLDSA_Q) */
// Load scaling factor 1/256*R
mov wtmp, #16382 // 2^(32 - 8) mod Q
dup ninv.4s, wtmp

/* check-magic: 4197891 == round(16382 * 2**31 / MLDSA_Q) */
// Load Barrett constant for 1/256*R
movz wtmp, #0x0e03
movk wtmp, #0x40, lsl #16
dup ninv_tw.4s, wtmp

mov count, #4
load_roots_1234

Expand Down
9 changes: 6 additions & 3 deletions dev/aarch64_opt/src/Makefile
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,7 +28,6 @@ SLOTHY_FLAGS_SPLIT= -c inputs_are_outputs \
-c variable_size \
-c constraints.stalls_first_attempt=64 \
-c split_heuristic=true \
-c split_heuristic_factor=1.5 \
-c split_heuristic_repeat=2 \
-c sw_pipelining.enabled=true \
-c sw_pipelining.halving_heuristic=True \
Expand Down Expand Up @@ -67,11 +66,15 @@ ntt.S: ../../aarch64_clean/src/ntt.S
$(eval TMPFILE := $(shell mktemp))
slothy-cli $(TARGET_ISA) $(TARGET_MICROARCH) $< -o $(TMPFILE) -l ntt_layer123_start $(SLOTHY_FLAGS) $(RESERVE_X_ONLY_FLAG)
# optimize second loop using split heuristic
slothy-cli $(TARGET_ISA) $(TARGET_MICROARCH) $(TMPFILE) -o $@ -l ntt_layer45678_start $(SLOTHY_FLAGS_SPLIT) $(RESERVE_X_ONLY_FLAG)
slothy-cli $(TARGET_ISA) $(TARGET_MICROARCH) $(TMPFILE) -o $@ -l ntt_layer45678_start $(SLOTHY_FLAGS_SPLIT) -c split_heuristic_factor=1.5 $(RESERVE_X_ONLY_FLAG)

# Copy remaining files without optimization for now
intt.S: ../../aarch64_clean/src/intt.S
cp $< $@
# optimize first loop in one go and write to temp file
$(eval TMPFILE := $(shell mktemp))
slothy-cli $(TARGET_ISA) $(TARGET_MICROARCH) $< -o $(TMPFILE) -l intt_layer5678_start $(SLOTHY_FLAGS) -c reserved_regs="[x0,x18--x30,sp]"
# optimize second loop using split heuristic
slothy-cli $(TARGET_ISA) $(TARGET_MICROARCH) $(TMPFILE) -o $@ -l intt_layer1234_start $(SLOTHY_FLAGS_SPLIT) -c split_heuristic_factor=2.5 $(RESERVE_X_ONLY_FLAG)

mld_polyvecl_pointwise_acc_montgomery_l4.S: ../../aarch64_clean/src/mld_polyvecl_pointwise_acc_montgomery_l4.S
cp $< $@
Expand Down
Loading