Add experimental SLOTHY-optimized NTTs

So far, clean ASM is only optimized according to Cortex-A72 model, and we do not yet explore algorithmic variations from Fast and Clean: Auditable high-performance assembly via constraint solving https://eprint.iacr.org/2022/1303 This will come at a later point. Both the clean and the optimized code are added to the repository, as well as the SLOTHY script. Signed-off-by: Hanno Becker <[email protected]>
pq-code-package · Sep 19, 2024 · 6d9391a · 6d9391a
1 parent 2ede9d4
commit 6d9391a
Show file tree

Hide file tree

Showing 12 changed files with 2,378 additions and 17 deletions.
diff --git a/...m/aarch64/intt_kyber_123_45_67_twiddles.S → mlkem/asm/aarch64/intt_123_45_67_twiddles.S b/...m/aarch64/intt_kyber_123_45_67_twiddles.S → mlkem/asm/aarch64/intt_123_45_67_twiddles.S
diff --git a/mlkem/asm/aarch64/intt_123_4567.S → mlkem/asm/aarch64/intt_clean.S b/mlkem/asm/aarch64/intt_123_4567.S → mlkem/asm/aarch64/intt_clean.S
@@ -159,11 +159,11 @@
 .data
 .p2align 4
 roots:
-#include "intt_kyber_123_45_67_twiddles.S"
+#include "intt_123_45_67_twiddles.S"
 .text
 
-        .global intt_kyber_123_4567
-        .global _intt_kyber_123_4567
+        .global intt_asm_clean
+        .global _intt_asm_clean
 
 .p2align 4
 const_addr:       .short 3329
@@ -191,8 +191,8 @@ ninv_tw_addr:     .short 5040
                   .short 5040
                   .short 5040
 
-intt_kyber_123_4567:
-_intt_kyber_123_4567:
+intt_asm_clean:
+_intt_asm_clean:
         push_stack
 
         in      .req x0

diff --git a/mlkem/asm/aarch64/intt_opt.S b/mlkem/asm/aarch64/intt_opt.S
diff --git a/...sm/aarch64/ntt_kyber_123_45_67_twiddles.S → mlkem/asm/aarch64/ntt_123_45_67_twiddles.S b/...sm/aarch64/ntt_kyber_123_45_67_twiddles.S → mlkem/asm/aarch64/ntt_123_45_67_twiddles.S
diff --git a/mlkem/asm/aarch64/ntt_123_4567.S → mlkem/asm/aarch64/ntt_clean.S b/mlkem/asm/aarch64/ntt_123_4567.S → mlkem/asm/aarch64/ntt_clean.S
@@ -157,7 +157,7 @@
 .data
 .p2align 4
 roots:
-        #include "ntt_kyber_123_45_67_twiddles.S"
+        #include "ntt_123_45_67_twiddles.S"
 
         in      .req x0
         inp     .req x1
@@ -208,8 +208,8 @@ roots:
         t3  .req v28
 
         .text
-        .global ntt_kyber_123_4567
-        .global _ntt_kyber_123_4567
+        .global ntt_asm_clean
+        .global _ntt_asm_clean
 
 .p2align 4
 const_addr:
@@ -222,8 +222,8 @@ const_addr:
         .short 0
         .short 0
 
-ntt_kyber_123_4567:
-_ntt_kyber_123_4567:
+ntt_asm_clean:
+_ntt_asm_clean:
         push_stack
 
         ASM_LOAD(r_ptr0, roots)