New file poly1305-blocks.asm

diff --git a/configure.ac b/configure.ac
index 040b772f5467270c874f8e55bf796da1712846de..5ee7239eaaf1188748216053cc6f97b169d87e54 100644 (file)
--- a/configure.ac
+++ b/configure.ac
@@ -612,7 +612,7 @@ asm_nettle_optional_list="cpuid.asm cpu-facility.asm \
   aes256-encrypt-2.asm aes256-decrypt-2.asm \
   cbc-aes128-encrypt-2.asm cbc-aes192-encrypt-2.asm cbc-aes256-encrypt-2.asm \
   chacha-2core.asm chacha-3core.asm chacha-4core.asm chacha-core-internal-2.asm \
-  poly1305-internal-2.asm \
+  poly1305-blocks.asm poly1305-internal-2.asm \
   ghash-set-key-2.asm ghash-update-2.asm \
   salsa20-2core.asm salsa20-core-internal-2.asm \
   sha1-compress-2.asm sha256-compress-n-2.asm \
@@ -759,10 +759,7 @@ AH_VERBATIM([HAVE_NATIVE],
 #undef HAVE_NATIVE_ecc_secp384r1_redc
 #undef HAVE_NATIVE_ecc_secp521r1_modp
 #undef HAVE_NATIVE_ecc_secp521r1_redc
-#undef HAVE_NATIVE_poly1305_set_key
-#undef HAVE_NATIVE_poly1305_block
 #undef HAVE_NATIVE_poly1305_blocks
-#undef HAVE_NATIVE_poly1305_digest
 #undef HAVE_NATIVE_ghash_set_key
 #undef HAVE_NATIVE_ghash_update
 #undef HAVE_NATIVE_salsa20_core

diff --git a/poly1305-internal.c b/poly1305-internal.c
index cd9583f57e995a091d43faf38cd3149d5a8e93fb..380b934eed72d512283dd7394df8fa62dca1f5cf 100644 (file)
--- a/poly1305-internal.c
+++ b/poly1305-internal.c
@@ -169,15 +169,6 @@ _nettle_poly1305_block (struct poly1305_ctx *ctx, const uint8_t *m, unsigned t4)
   ctx->h0 += b * 5;
 }
 
-const uint8_t *
-_nettle_poly1305_blocks (struct poly1305_ctx *ctx, size_t blocks, const uint8_t *m)
-{
-  for (; blocks > 0; blocks--, m += POLY1305_BLOCK_SIZE)
-    _nettle_poly1305_block (ctx, m, 1);
-
-  return m;
-}
-
 /* Adds digest to the nonce */
 void
 _nettle_poly1305_digest (struct poly1305_ctx *ctx, union nettle_block16 *s)

diff --git a/poly1305-update.c b/poly1305-update.c
index aa39185996dba5af2d8d0946b45b62cdcb9a711f..fdc725581a3a47818bbac609b031f9e41ad96cad 100644 (file)
--- a/poly1305-update.c
+++ b/poly1305-update.c
@@ -49,8 +49,14 @@ _nettle_poly1305_update (struct poly1305_ctx *ctx,
                               length, m);
       _nettle_poly1305_block(ctx, block, 1);
     }
+#if HAVE_NATIVE_poly1305_blocks
   m = _nettle_poly1305_blocks (ctx, length >> 4, m);
   length &= 15;
+#else
+  for (; length >= POLY1305_BLOCK_SIZE;
+       length -= POLY1305_BLOCK_SIZE, m += POLY1305_BLOCK_SIZE)
+    _nettle_poly1305_block (ctx, m, 1);
+#endif
 
   memcpy (block, m, length);
   return length;

diff --git a/x86_64/poly1305-blocks.asm b/x86_64/poly1305-blocks.asm
new file mode 100644 (file)
index 0000000..63bfed3
--- /dev/null
+++ b/x86_64/poly1305-blocks.asm
@@ -0,0 +1,128 @@
+C x86_64/poly1305-blocks.asm
+
+ifelse(`
+   Copyright (C) 2022 Niels Möller
+
+   This file is part of GNU Nettle.
+
+   GNU Nettle is free software: you can redistribute it and/or
+   modify it under the terms of either:
+
+     * the GNU Lesser General Public License as published by the Free
+       Software Foundation; either version 3 of the License, or (at your
+       option) any later version.
+
+   or
+
+     * the GNU General Public License as published by the Free
+       Software Foundation; either version 2 of the License, or (at your
+       option) any later version.
+
+   or both in parallel, as here.
+
+   GNU Nettle is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   General Public License for more details.
+
+   You should have received copies of the GNU General Public License and
+   the GNU Lesser General Public License along with this program.  If
+   not, see http://www.gnu.org/licenses/.
+')
+
+       .file "poly1305-blocks.asm"
+
+define(`CTX', `%rdi') C First argument to all functions
+define(`BLOCKS', `%rsi')
+define(`MP_PARAM', `%rdx')     C Moved to MP, to not collide with mul instruction.
+
+define(`MP', `%r8')            C May clobber, both with unix and windows conventions.
+define(`T0', `%rbx')
+define(`T1', `%rcx')
+define(`H0', `%rbp')
+define(`H1', `%r9')
+define(`H2', `%r10')
+define(`F0', `%r11')
+define(`F1', `%r12')
+
+C const uint8_t *
+C _nettle_poly1305_blocks (struct poly1305_ctx *ctx, size_t blocks, const uint8_t *m)
+
+PROLOGUE(_nettle_poly1305_blocks)
+       W64_ENTRY(3, 0)
+       mov     MP_PARAM, MP
+       test    BLOCKS, BLOCKS
+       jz      .Lend
+
+       push    %rbx
+       push    %rbp
+       push    %r12
+       mov     P1305_H0 (CTX), H0
+       mov     P1305_H1 (CTX), H1
+       mov     P1305_H2 (CTX), H2
+       ALIGN(16)
+.Loop:
+       mov     (MP), T0
+       mov     8(MP), T1
+       add     $16, MP
+
+       add     H0, T0
+       adc     H1, T1
+       adc     $1, H2
+
+       mov     P1305_R1 (CTX), %rax
+       mul     T0                      C R1*T0
+       mov     %rax, F0
+       mov     %rdx, F1
+
+       mov     T0, %rax                C Last use of T0 input
+       mov     P1305_R0 (CTX), T0
+       mul     T0                      C R0*T0
+       mov     %rax, H0
+       mov     %rdx, H1
+
+       mov     T1, %rax
+       mul     T0                      C R0*T1
+       add     %rax, F0
+       adc     %rdx, F1
+
+       mov     P1305_S1 (CTX), T0
+       mov     T1, %rax                C Last use of T1 input
+       mul     T0                      C S1*T1
+       add     %rax, H0
+       adc     %rdx, H1
+
+       mov     H2, %rax
+       mul     T0                      C S1*H2
+       add     %rax, F0
+       adc     %rdx, F1
+
+       mov     H2, T0
+       and     $3, H2
+
+       shr     $2, T0
+       mov     P1305_S0 (CTX), %rax
+       mul     T0                      C S0*(H2 >> 2)
+       add     %rax, H0
+       adc     %rdx, H1
+
+       imul    P1305_R0 (CTX), H2      C R0*(H2 & 3)
+       add     F0, H1
+       adc     F1, H2
+
+       dec     BLOCKS
+       jnz     .Loop
+
+       mov     H0, P1305_H0 (CTX)
+       mov     H1, P1305_H1 (CTX)
+       mov     H2, P1305_H2 (CTX)
+
+       pop     %r12
+       pop     %rbp
+       pop     %rbx
+
+.Lend:
+       mov     MP, %rax
+       W64_EXIT(3, 0)
+       ret
+EPILOGUE(_nettle_poly1305_blocks)