Revision control
Copy as Markdown
Other Tools
/* chacha20-amd64-ssse3.S - SSSE3 implementation of ChaCha20 cipher
*
* Copyright (C) 2017-2019 Jussi Kivilinna <jussi.kivilinna@iki.fi>
*
* This file is part of Libgcrypt.
*
* Libgcrypt is free software; you can redistribute it and/or modify
* it under the terms of the GNU Lesser General Public License as
* published by the Free Software Foundation; either version 2.1 of
* the License, or (at your option) any later version.
*
* Libgcrypt 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 Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
*/
/*
* Based on D. J. Bernstein reference implementation at
*
* chacha-regs.c version 20080118
* D. J. Bernstein
* Public domain.
*/
#ifdef __x86_64
#include <config.h>
#if defined(HAVE_GCC_INLINE_ASM_SSSE3) && \
(defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS) || \
defined(HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS))
.text
#include "asm-common-amd64.h"
#include "asm-poly1305-amd64.h"
/* register macros */
#define INPUT %rdi
#define DST %rsi
#define SRC %rdx
#define NBLKS %rcx
#define ROUND %eax
/* stack structure */
#define STACK_VEC_X12 (16)
#define STACK_VEC_X13 (16 + STACK_VEC_X12)
#define STACK_TMP (16 + STACK_VEC_X13)
#define STACK_TMP1 (16 + STACK_TMP)
#define STACK_TMP2 (16 + STACK_TMP1)
#define STACK_MAX (16 + STACK_TMP2)
/* vector registers */
#define X0 %xmm0
#define X1 %xmm1
#define X2 %xmm2
#define X3 %xmm3
#define X4 %xmm4
#define X5 %xmm5
#define X6 %xmm6
#define X7 %xmm7
#define X8 %xmm8
#define X9 %xmm9
#define X10 %xmm10
#define X11 %xmm11
#define X12 %xmm12
#define X13 %xmm13
#define X14 %xmm14
#define X15 %xmm15
/**********************************************************************
helper macros
**********************************************************************/
/* 4x4 32-bit integer matrix transpose */
#define transpose_4x4(x0, x1, x2, x3, t1, t2, t3) \
movdqa x0, t2; \
punpckhdq x1, t2; \
punpckldq x1, x0; \
\
movdqa x2, t1; \
punpckldq x3, t1; \
punpckhdq x3, x2; \
\
movdqa x0, x1; \
punpckhqdq t1, x1; \
punpcklqdq t1, x0; \
\
movdqa t2, x3; \
punpckhqdq x2, x3; \
punpcklqdq x2, t2; \
movdqa t2, x2;
/* fill xmm register with 32-bit value from memory */
#define pbroadcastd(mem32, xreg) \
movd mem32, xreg; \
pshufd $0, xreg, xreg;
/* xor with unaligned memory operand */
#define pxor_u(umem128, xreg, t) \
movdqu umem128, t; \
pxor t, xreg;
/* xor register with unaligned src and save to unaligned dst */
#define xor_src_dst(dst, src, offset, xreg, t) \
pxor_u(offset(src), xreg, t); \
movdqu xreg, offset(dst);
#define clear(x) pxor x,x;
/**********************************************************************
4-way chacha20
**********************************************************************/
#define ROTATE2(v1,v2,c,tmp1,tmp2) \
movdqa v1, tmp1; \
movdqa v2, tmp2; \
psrld $(32 - (c)), v1; \
pslld $(c), tmp1; \
paddb tmp1, v1; \
psrld $(32 - (c)), v2; \
pslld $(c), tmp2; \
paddb tmp2, v2;
#define ROTATE_SHUF_2(v1,v2,shuf) \
pshufb shuf, v1; \
pshufb shuf, v2;
#define XOR(ds,s) \
pxor s, ds;
#define PLUS(ds,s) \
paddd s, ds;
#define QUARTERROUND2(a1,b1,c1,d1,a2,b2,c2,d2,ign,tmp1,tmp2,\
interleave_op1,interleave_op2) \
movdqa .Lshuf_rol16 rRIP, tmp1; \
interleave_op1; \
PLUS(a1,b1); PLUS(a2,b2); XOR(d1,a1); XOR(d2,a2); \
ROTATE_SHUF_2(d1, d2, tmp1); \
PLUS(c1,d1); PLUS(c2,d2); XOR(b1,c1); XOR(b2,c2); \
ROTATE2(b1, b2, 12, tmp1, tmp2); \
movdqa .Lshuf_rol8 rRIP, tmp1; \
interleave_op2; \
PLUS(a1,b1); PLUS(a2,b2); XOR(d1,a1); XOR(d2,a2); \
ROTATE_SHUF_2(d1, d2, tmp1); \
PLUS(c1,d1); PLUS(c2,d2); XOR(b1,c1); XOR(b2,c2); \
ROTATE2(b1, b2, 7, tmp1, tmp2);
chacha20_data:
.align 16
.Lshuf_rol16:
.byte 2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13
.Lshuf_rol8:
.byte 3,0,1,2,7,4,5,6,11,8,9,10,15,12,13,14
.Lcounter1:
.long 1,0,0,0
.Linc_counter:
.long 0,1,2,3
.Lunsigned_cmp:
.long 0x80000000,0x80000000,0x80000000,0x80000000
.align 8
.globl _gcry_chacha20_amd64_ssse3_blocks4
ELF(.type _gcry_chacha20_amd64_ssse3_blocks4,@function;)
_gcry_chacha20_amd64_ssse3_blocks4:
/* input:
* %rdi: input
* %rsi: dst
* %rdx: src
* %rcx: nblks (multiple of 4)
*/
CFI_STARTPROC();
pushq %rbp;
CFI_PUSH(%rbp);
movq %rsp, %rbp;
CFI_DEF_CFA_REGISTER(%rbp);
subq $STACK_MAX, %rsp;
andq $~15, %rsp;
.Loop4:
mov $20, ROUND;
/* Construct counter vectors X12 and X13 */
movdqa .Linc_counter rRIP, X0;
movdqa .Lunsigned_cmp rRIP, X2;
pbroadcastd((12 * 4)(INPUT), X12);
pbroadcastd((13 * 4)(INPUT), X13);
paddd X0, X12;
movdqa X12, X1;
pxor X2, X0;
pxor X2, X1;
pcmpgtd X1, X0;
psubd X0, X13;
movdqa X12, (STACK_VEC_X12)(%rsp);
movdqa X13, (STACK_VEC_X13)(%rsp);
/* Load vectors */
pbroadcastd((0 * 4)(INPUT), X0);
pbroadcastd((1 * 4)(INPUT), X1);
pbroadcastd((2 * 4)(INPUT), X2);
pbroadcastd((3 * 4)(INPUT), X3);
pbroadcastd((4 * 4)(INPUT), X4);
pbroadcastd((5 * 4)(INPUT), X5);
pbroadcastd((6 * 4)(INPUT), X6);
pbroadcastd((7 * 4)(INPUT), X7);
pbroadcastd((8 * 4)(INPUT), X8);
pbroadcastd((9 * 4)(INPUT), X9);
pbroadcastd((10 * 4)(INPUT), X10);
pbroadcastd((11 * 4)(INPUT), X11);
pbroadcastd((14 * 4)(INPUT), X14);
pbroadcastd((15 * 4)(INPUT), X15);
movdqa X11, (STACK_TMP)(%rsp);
movdqa X15, (STACK_TMP1)(%rsp);
.Lround2_4:
QUARTERROUND2(X0, X4, X8, X12, X1, X5, X9, X13, tmp:=,X11,X15,,)
movdqa (STACK_TMP)(%rsp), X11;
movdqa (STACK_TMP1)(%rsp), X15;
movdqa X8, (STACK_TMP)(%rsp);
movdqa X9, (STACK_TMP1)(%rsp);
QUARTERROUND2(X2, X6, X10, X14, X3, X7, X11, X15, tmp:=,X8,X9,,)
QUARTERROUND2(X0, X5, X10, X15, X1, X6, X11, X12, tmp:=,X8,X9,,)
movdqa (STACK_TMP)(%rsp), X8;
movdqa (STACK_TMP1)(%rsp), X9;
movdqa X11, (STACK_TMP)(%rsp);
movdqa X15, (STACK_TMP1)(%rsp);
QUARTERROUND2(X2, X7, X8, X13, X3, X4, X9, X14, tmp:=,X11,X15,,)
sub $2, ROUND;
jnz .Lround2_4;
/* tmp := X15 */
movdqa (STACK_TMP)(%rsp), X11;
pbroadcastd((0 * 4)(INPUT), X15);
PLUS(X0, X15);
pbroadcastd((1 * 4)(INPUT), X15);
PLUS(X1, X15);
pbroadcastd((2 * 4)(INPUT), X15);
PLUS(X2, X15);
pbroadcastd((3 * 4)(INPUT), X15);
PLUS(X3, X15);
pbroadcastd((4 * 4)(INPUT), X15);
PLUS(X4, X15);
pbroadcastd((5 * 4)(INPUT), X15);
PLUS(X5, X15);
pbroadcastd((6 * 4)(INPUT), X15);
PLUS(X6, X15);
pbroadcastd((7 * 4)(INPUT), X15);
PLUS(X7, X15);
pbroadcastd((8 * 4)(INPUT), X15);
PLUS(X8, X15);
pbroadcastd((9 * 4)(INPUT), X15);
PLUS(X9, X15);
pbroadcastd((10 * 4)(INPUT), X15);
PLUS(X10, X15);
pbroadcastd((11 * 4)(INPUT), X15);
PLUS(X11, X15);
movdqa (STACK_VEC_X12)(%rsp), X15;
PLUS(X12, X15);
movdqa (STACK_VEC_X13)(%rsp), X15;
PLUS(X13, X15);
movdqa X13, (STACK_TMP)(%rsp);
pbroadcastd((14 * 4)(INPUT), X15);
PLUS(X14, X15);
movdqa (STACK_TMP1)(%rsp), X15;
movdqa X14, (STACK_TMP1)(%rsp);
pbroadcastd((15 * 4)(INPUT), X13);
PLUS(X15, X13);
movdqa X15, (STACK_TMP2)(%rsp);
/* Update counter */
addq $4, (12 * 4)(INPUT);
transpose_4x4(X0, X1, X2, X3, X13, X14, X15);
xor_src_dst(DST, SRC, (64 * 0 + 16 * 0), X0, X15);
xor_src_dst(DST, SRC, (64 * 1 + 16 * 0), X1, X15);
xor_src_dst(DST, SRC, (64 * 2 + 16 * 0), X2, X15);
xor_src_dst(DST, SRC, (64 * 3 + 16 * 0), X3, X15);
transpose_4x4(X4, X5, X6, X7, X0, X1, X2);
movdqa (STACK_TMP)(%rsp), X13;
movdqa (STACK_TMP1)(%rsp), X14;
movdqa (STACK_TMP2)(%rsp), X15;
xor_src_dst(DST, SRC, (64 * 0 + 16 * 1), X4, X0);
xor_src_dst(DST, SRC, (64 * 1 + 16 * 1), X5, X0);
xor_src_dst(DST, SRC, (64 * 2 + 16 * 1), X6, X0);
xor_src_dst(DST, SRC, (64 * 3 + 16 * 1), X7, X0);
transpose_4x4(X8, X9, X10, X11, X0, X1, X2);
xor_src_dst(DST, SRC, (64 * 0 + 16 * 2), X8, X0);
xor_src_dst(DST, SRC, (64 * 1 + 16 * 2), X9, X0);
xor_src_dst(DST, SRC, (64 * 2 + 16 * 2), X10, X0);
xor_src_dst(DST, SRC, (64 * 3 + 16 * 2), X11, X0);
transpose_4x4(X12, X13, X14, X15, X0, X1, X2);
xor_src_dst(DST, SRC, (64 * 0 + 16 * 3), X12, X0);
xor_src_dst(DST, SRC, (64 * 1 + 16 * 3), X13, X0);
xor_src_dst(DST, SRC, (64 * 2 + 16 * 3), X14, X0);
xor_src_dst(DST, SRC, (64 * 3 + 16 * 3), X15, X0);
sub $4, NBLKS;
lea (4 * 64)(DST), DST;
lea (4 * 64)(SRC), SRC;
jnz .Loop4;
/* clear the used vector registers and stack */
clear(X0);
movdqa X0, (STACK_VEC_X12)(%rsp);
movdqa X0, (STACK_VEC_X13)(%rsp);
movdqa X0, (STACK_TMP)(%rsp);
movdqa X0, (STACK_TMP1)(%rsp);
movdqa X0, (STACK_TMP2)(%rsp);
clear(X1);
clear(X2);
clear(X3);
clear(X4);
clear(X5);
clear(X6);
clear(X7);
clear(X8);
clear(X9);
clear(X10);
clear(X11);
clear(X12);
clear(X13);
clear(X14);
clear(X15);
/* eax zeroed by round loop. */
leave;
CFI_LEAVE();
ret;
CFI_ENDPROC();
ELF(.size _gcry_chacha20_amd64_ssse3_blocks4,
.-_gcry_chacha20_amd64_ssse3_blocks4;)
/**********************************************************************
2-way && 1-way chacha20
**********************************************************************/
#define ROTATE_SHUF(v1,shuf) \
pshufb shuf, v1;
#define ROTATE(v1,c,tmp1) \
movdqa v1, tmp1; \
psrld $(32 - (c)), v1; \
pslld $(c), tmp1; \
paddb tmp1, v1;
#define WORD_SHUF(v1,shuf) \
pshufd $shuf, v1, v1;
#define QUARTERROUND4(x0,x1,x2,x3,shuf_rol8,shuf_rol16,tmp1,shuf_x1,\
shuf_x2,shuf_x3) \
PLUS(x0, x1); XOR(x3, x0); ROTATE_SHUF(x3, shuf_rol16); \
PLUS(x2, x3); XOR(x1, x2); ROTATE(x1, 12, tmp1); \
PLUS(x0, x1); XOR(x3, x0); ROTATE_SHUF(x3, shuf_rol8); \
PLUS(x2, x3); \
WORD_SHUF(x3, shuf_x3); \
XOR(x1, x2); \
WORD_SHUF(x2, shuf_x2); \
ROTATE(x1, 7, tmp1); \
WORD_SHUF(x1, shuf_x1);
.align 8
.globl _gcry_chacha20_amd64_ssse3_blocks1
ELF(.type _gcry_chacha20_amd64_ssse3_blocks1,@function;)
_gcry_chacha20_amd64_ssse3_blocks1:
/* input:
* %rdi: input
* %rsi: dst
* %rdx: src
* %rcx: nblks
*/
CFI_STARTPROC();
/* Load constants */
movdqa .Lcounter1 rRIP, X4;
movdqa .Lshuf_rol8 rRIP, X5;
movdqa .Lshuf_rol16 rRIP, X6;
/* Load state */
movdqu (0 * 4)(INPUT), X10;
movdqu (4 * 4)(INPUT), X11;
movdqu (8 * 4)(INPUT), X12;
movdqu (12 * 4)(INPUT), X13;
cmp $2, NBLKS;
jb .Loop1;
mov $20, ROUND;
movdqa X10, X0;
movdqa X11, X1;
movdqa X12, X2;
movdqa X13, X3;
movdqa X10, X8;
movdqa X11, X9;
movdqa X12, X14;
movdqa X13, X15;
paddq X4, X15;
.Lround2_2:
QUARTERROUND4(X0, X1, X2, X3, X5, X6, X7, 0x39, 0x4e, 0x93);
QUARTERROUND4(X8, X9, X14, X15, X5, X6, X7, 0x39, 0x4e, 0x93);
QUARTERROUND4(X0, X1, X2, X3, X5, X6, X7, 0x93, 0x4e, 0x39);
QUARTERROUND4(X8, X9, X14, X15, X5, X6, X7, 0x93, 0x4e, 0x39);
sub $2, ROUND;
jnz .Lround2_2;
PLUS(X0, X10);
PLUS(X1, X11);
PLUS(X2, X12);
PLUS(X3, X13);
/* Update counter */
paddq X4, X13;
PLUS(X8, X10);
PLUS(X9, X11);
PLUS(X14, X12);
PLUS(X15, X13);
/* Update counter */
paddq X4, X13;
xor_src_dst(DST, SRC, 0 * 4, X0, X7);
xor_src_dst(DST, SRC, 4 * 4, X1, X7);
xor_src_dst(DST, SRC, 8 * 4, X2, X7);
xor_src_dst(DST, SRC, 12 * 4, X3, X7);
xor_src_dst(DST, SRC, 16 * 4, X8, X7);
xor_src_dst(DST, SRC, 20 * 4, X9, X7);
xor_src_dst(DST, SRC, 24 * 4, X14, X7);
xor_src_dst(DST, SRC, 28 * 4, X15, X7);
lea (2 * 64)(DST), DST;
lea (2 * 64)(SRC), SRC;
clear(X8);
clear(X9);
clear(X14);
clear(X15);
sub $2, NBLKS;
jz .Ldone1;
.Loop1:
mov $20, ROUND;
movdqa X10, X0;
movdqa X11, X1;
movdqa X12, X2;
movdqa X13, X3;
.Lround2_1:
QUARTERROUND4(X0, X1, X2, X3, X5, X6, X7, 0x39, 0x4e, 0x93);
QUARTERROUND4(X0, X1, X2, X3, X5, X6, X7, 0x93, 0x4e, 0x39);
sub $2, ROUND;
jnz .Lround2_1;
PLUS(X0, X10);
PLUS(X1, X11);
PLUS(X2, X12);
PLUS(X3, X13);
/* Update counter */
paddq X4, X13;
xor_src_dst(DST, SRC, 0 * 4, X0, X7);
xor_src_dst(DST, SRC, 4 * 4, X1, X7);
xor_src_dst(DST, SRC, 8 * 4, X2, X7);
xor_src_dst(DST, SRC, 12 * 4, X3, X7);
lea (64)(DST), DST;
lea (64)(SRC), SRC;
sub $1, NBLKS;
jnz .Loop1;
.Ldone1:
/* Store counter */
movdqu X13, (12 * 4)(INPUT);
/* clear the used vector registers */
clear(X0);
clear(X1);
clear(X2);
clear(X3);
clear(X4);
clear(X5);
clear(X6);
clear(X7);
clear(X10);
clear(X11);
clear(X12);
clear(X13);
/* eax zeroed by round loop. */
ret;
CFI_ENDPROC();
ELF(.size _gcry_chacha20_amd64_ssse3_blocks1,
.-_gcry_chacha20_amd64_ssse3_blocks1;)
/**********************************************************************
4-way stitched chacha20-poly1305
**********************************************************************/
#define _ /*_*/
.align 8
.globl _gcry_chacha20_poly1305_amd64_ssse3_blocks4
ELF(.type _gcry_chacha20_poly1305_amd64_ssse3_blocks4,@function;)
_gcry_chacha20_poly1305_amd64_ssse3_blocks4:
/* input:
* %rdi: input
* %rsi: dst
* %rdx: src
* %rcx: nblks (multiple of 4)
* %r9: poly1305-state
* %r8: poly1305-src
*/
CFI_STARTPROC();
pushq %rbp;
CFI_PUSH(%rbp);
movq %rsp, %rbp;
CFI_DEF_CFA_REGISTER(%rbp);
subq $(9 * 8) + STACK_MAX + 16, %rsp;
andq $~15, %rsp;
movq %rbx, (STACK_MAX + 0 * 8)(%rsp);
movq %r12, (STACK_MAX + 1 * 8)(%rsp);
movq %r13, (STACK_MAX + 2 * 8)(%rsp);
movq %r14, (STACK_MAX + 3 * 8)(%rsp);
movq %r15, (STACK_MAX + 4 * 8)(%rsp);
CFI_REG_ON_STACK(rbx, STACK_MAX + 0 * 8);
CFI_REG_ON_STACK(r12, STACK_MAX + 1 * 8);
CFI_REG_ON_STACK(r13, STACK_MAX + 2 * 8);
CFI_REG_ON_STACK(r14, STACK_MAX + 3 * 8);
CFI_REG_ON_STACK(r15, STACK_MAX + 4 * 8);
movq %rdx, (STACK_MAX + 5 * 8)(%rsp); # SRC
movq %rsi, (STACK_MAX + 6 * 8)(%rsp); # DST
movq %rcx, (STACK_MAX + 7 * 8)(%rsp); # NBLKS
/* Load state */
POLY1305_LOAD_STATE();
.Loop_poly4:
/* Construct counter vectors X12 and X13 */
movdqa .Linc_counter rRIP, X0;
movdqa .Lunsigned_cmp rRIP, X2;
pbroadcastd((12 * 4)(INPUT), X12);
pbroadcastd((13 * 4)(INPUT), X13);
paddd X0, X12;
movdqa X12, X1;
pxor X2, X0;
pxor X2, X1;
pcmpgtd X1, X0;
psubd X0, X13;
movdqa X12, (STACK_VEC_X12)(%rsp);
movdqa X13, (STACK_VEC_X13)(%rsp);
/* Load vectors */
pbroadcastd((0 * 4)(INPUT), X0);
pbroadcastd((1 * 4)(INPUT), X1);
pbroadcastd((2 * 4)(INPUT), X2);
pbroadcastd((3 * 4)(INPUT), X3);
pbroadcastd((4 * 4)(INPUT), X4);
pbroadcastd((5 * 4)(INPUT), X5);
pbroadcastd((6 * 4)(INPUT), X6);
pbroadcastd((7 * 4)(INPUT), X7);
pbroadcastd((8 * 4)(INPUT), X8);
pbroadcastd((9 * 4)(INPUT), X9);
pbroadcastd((10 * 4)(INPUT), X10);
pbroadcastd((11 * 4)(INPUT), X11);
pbroadcastd((14 * 4)(INPUT), X14);
pbroadcastd((15 * 4)(INPUT), X15);
movdqa X11, (STACK_TMP)(%rsp);
movdqa X15, (STACK_TMP1)(%rsp);
/* Process four ChaCha20 blocks and sixteen Poly1305 blocks. */
movl $20, (STACK_MAX + 8 * 8 + 4)(%rsp);
.Lround4_with_poly1305_outer:
movl $6, (STACK_MAX + 8 * 8)(%rsp);
.Lround4_with_poly1305_inner1:
/* rounds 0-5 & 10-15 */
POLY1305_BLOCK_PART1(0 * 16)
QUARTERROUND2(X0, X4, X8, X12, X1, X5, X9, X13, tmp:=,X11,X15,
POLY1305_BLOCK_PART2(),
POLY1305_BLOCK_PART3())
movdqa (STACK_TMP)(%rsp), X11;
movdqa (STACK_TMP1)(%rsp), X15;
movdqa X8, (STACK_TMP)(%rsp);
movdqa X9, (STACK_TMP1)(%rsp);
QUARTERROUND2(X2, X6, X10, X14, X3, X7, X11, X15, tmp:=,X8,X9,
POLY1305_BLOCK_PART4(),
POLY1305_BLOCK_PART5())
POLY1305_BLOCK_PART1(1 * 16)
lea (2 * 16)(POLY_RSRC), POLY_RSRC;
QUARTERROUND2(X0, X5, X10, X15, X1, X6, X11, X12, tmp:=,X8,X9,
POLY1305_BLOCK_PART2(),
POLY1305_BLOCK_PART3())
movdqa (STACK_TMP)(%rsp), X8;
movdqa (STACK_TMP1)(%rsp), X9;
movdqa X11, (STACK_TMP)(%rsp);
movdqa X15, (STACK_TMP1)(%rsp);
QUARTERROUND2(X2, X7, X8, X13, X3, X4, X9, X14, tmp:=,X11,X15,
POLY1305_BLOCK_PART4(),
POLY1305_BLOCK_PART5())
subl $2, (STACK_MAX + 8 * 8)(%rsp);
jnz .Lround4_with_poly1305_inner1;
movl $4, (STACK_MAX + 8 * 8)(%rsp);
.Lround4_with_poly1305_inner2:
/* rounds 6-9 & 16-19 */
POLY1305_BLOCK_PART1(0 * 16)
lea (1 * 16)(POLY_RSRC), POLY_RSRC;
QUARTERROUND2(X0, X4, X8, X12, X1, X5, X9, X13, tmp:=,X11,X15,
POLY1305_BLOCK_PART2(),
_)
movdqa (STACK_TMP)(%rsp), X11;
movdqa (STACK_TMP1)(%rsp), X15;
movdqa X8, (STACK_TMP)(%rsp);
movdqa X9, (STACK_TMP1)(%rsp);
QUARTERROUND2(X2, X6, X10, X14, X3, X7, X11, X15, tmp:=,X8,X9,
POLY1305_BLOCK_PART3(),
_)
QUARTERROUND2(X0, X5, X10, X15, X1, X6, X11, X12, tmp:=,X8,X9,
POLY1305_BLOCK_PART4(),
_)
movdqa (STACK_TMP)(%rsp), X8;
movdqa (STACK_TMP1)(%rsp), X9;
movdqa X11, (STACK_TMP)(%rsp);
movdqa X15, (STACK_TMP1)(%rsp);
QUARTERROUND2(X2, X7, X8, X13, X3, X4, X9, X14, tmp:=,X11,X15,
POLY1305_BLOCK_PART5(),
_)
subl $2, (STACK_MAX + 8 * 8)(%rsp);
jnz .Lround4_with_poly1305_inner2;
subl $10, (STACK_MAX + 8 * 8 + 4)(%rsp);
jnz .Lround4_with_poly1305_outer;
/* tmp := X15 */
movdqa (STACK_TMP)(%rsp), X11;
pbroadcastd((0 * 4)(INPUT), X15);
PLUS(X0, X15);
pbroadcastd((1 * 4)(INPUT), X15);
PLUS(X1, X15);
pbroadcastd((2 * 4)(INPUT), X15);
PLUS(X2, X15);
pbroadcastd((3 * 4)(INPUT), X15);
PLUS(X3, X15);
pbroadcastd((4 * 4)(INPUT), X15);
PLUS(X4, X15);
pbroadcastd((5 * 4)(INPUT), X15);
PLUS(X5, X15);
pbroadcastd((6 * 4)(INPUT), X15);
PLUS(X6, X15);
pbroadcastd((7 * 4)(INPUT), X15);
PLUS(X7, X15);
pbroadcastd((8 * 4)(INPUT), X15);
PLUS(X8, X15);
pbroadcastd((9 * 4)(INPUT), X15);
PLUS(X9, X15);
pbroadcastd((10 * 4)(INPUT), X15);
PLUS(X10, X15);
pbroadcastd((11 * 4)(INPUT), X15);
PLUS(X11, X15);
movdqa (STACK_VEC_X12)(%rsp), X15;
PLUS(X12, X15);
movdqa (STACK_VEC_X13)(%rsp), X15;
PLUS(X13, X15);
movdqa X13, (STACK_TMP)(%rsp);
pbroadcastd((14 * 4)(INPUT), X15);
PLUS(X14, X15);
movdqa (STACK_TMP1)(%rsp), X15;
movdqa X14, (STACK_TMP1)(%rsp);
pbroadcastd((15 * 4)(INPUT), X13);
PLUS(X15, X13);
movdqa X15, (STACK_TMP2)(%rsp);
/* Update counter */
addq $4, (12 * 4)(INPUT);
movq (STACK_MAX + 5 * 8)(%rsp), SRC;
movq (STACK_MAX + 6 * 8)(%rsp), DST;
transpose_4x4(X0, X1, X2, X3, X13, X14, X15);
xor_src_dst(DST, SRC, (64 * 0 + 16 * 0), X0, X15);
xor_src_dst(DST, SRC, (64 * 1 + 16 * 0), X1, X15);
xor_src_dst(DST, SRC, (64 * 2 + 16 * 0), X2, X15);
xor_src_dst(DST, SRC, (64 * 3 + 16 * 0), X3, X15);
transpose_4x4(X4, X5, X6, X7, X0, X1, X2);
movdqa (STACK_TMP)(%rsp), X13;
movdqa (STACK_TMP1)(%rsp), X14;
movdqa (STACK_TMP2)(%rsp), X15;
xor_src_dst(DST, SRC, (64 * 0 + 16 * 1), X4, X0);
xor_src_dst(DST, SRC, (64 * 1 + 16 * 1), X5, X0);
xor_src_dst(DST, SRC, (64 * 2 + 16 * 1), X6, X0);
xor_src_dst(DST, SRC, (64 * 3 + 16 * 1), X7, X0);
transpose_4x4(X8, X9, X10, X11, X0, X1, X2);
xor_src_dst(DST, SRC, (64 * 0 + 16 * 2), X8, X0);
xor_src_dst(DST, SRC, (64 * 1 + 16 * 2), X9, X0);
xor_src_dst(DST, SRC, (64 * 2 + 16 * 2), X10, X0);
xor_src_dst(DST, SRC, (64 * 3 + 16 * 2), X11, X0);
transpose_4x4(X12, X13, X14, X15, X0, X1, X2);
xor_src_dst(DST, SRC, (64 * 0 + 16 * 3), X12, X0);
xor_src_dst(DST, SRC, (64 * 1 + 16 * 3), X13, X0);
xor_src_dst(DST, SRC, (64 * 2 + 16 * 3), X14, X0);
xor_src_dst(DST, SRC, (64 * 3 + 16 * 3), X15, X0);
subq $4, (STACK_MAX + 7 * 8)(%rsp); # NBLKS
lea (4 * 64)(DST), DST;
lea (4 * 64)(SRC), SRC;
movq SRC, (STACK_MAX + 5 * 8)(%rsp);
movq DST, (STACK_MAX + 6 * 8)(%rsp);
jnz .Loop_poly4;
/* Store state */
POLY1305_STORE_STATE();
/* clear the used vector registers and stack */
clear(X0);
movdqa X0, (STACK_VEC_X12)(%rsp);
movdqa X0, (STACK_VEC_X13)(%rsp);
movdqa X0, (STACK_TMP)(%rsp);
movdqa X0, (STACK_TMP1)(%rsp);
movdqa X0, (STACK_TMP2)(%rsp);
clear(X1);
clear(X2);
clear(X3);
clear(X4);
clear(X5);
clear(X6);
clear(X7);
clear(X8);
clear(X9);
clear(X10);
clear(X11);
clear(X12);
clear(X13);
clear(X14);
clear(X15);
movq (STACK_MAX + 0 * 8)(%rsp), %rbx;
movq (STACK_MAX + 1 * 8)(%rsp), %r12;
movq (STACK_MAX + 2 * 8)(%rsp), %r13;
movq (STACK_MAX + 3 * 8)(%rsp), %r14;
movq (STACK_MAX + 4 * 8)(%rsp), %r15;
CFI_RESTORE(%rbx);
CFI_RESTORE(%r12);
CFI_RESTORE(%r13);
CFI_RESTORE(%r14);
CFI_RESTORE(%r15);
xorl %eax, %eax;
leave;
CFI_LEAVE();
ret;
CFI_ENDPROC();
ELF(.size _gcry_chacha20_poly1305_amd64_ssse3_blocks4,
.-_gcry_chacha20_poly1305_amd64_ssse3_blocks4;)
/**********************************************************************
2-way && 1-way stitched chacha20-poly1305
**********************************************************************/
.align 8
.globl _gcry_chacha20_poly1305_amd64_ssse3_blocks1
ELF(.type _gcry_chacha20_poly1305_amd64_ssse3_blocks1,@function;)
_gcry_chacha20_poly1305_amd64_ssse3_blocks1:
/* input:
* %rdi: chacha20-state
* %rsi: dst
* %rdx: src
* %rcx: nblks
* %r9: poly1305-state
* %r8: poly1305-src
*/
CFI_STARTPROC();
pushq %rbp;
CFI_PUSH(%rbp);
movq %rsp, %rbp;
CFI_DEF_CFA_REGISTER(%rbp);
subq $(9 * 8), %rsp;
movq %rbx, (0 * 8)(%rsp);
movq %r12, (1 * 8)(%rsp);
movq %r13, (2 * 8)(%rsp);
movq %r14, (3 * 8)(%rsp);
movq %r15, (4 * 8)(%rsp);
CFI_REG_ON_STACK(rbx, 0 * 8);
CFI_REG_ON_STACK(r12, 1 * 8);
CFI_REG_ON_STACK(r13, 2 * 8);
CFI_REG_ON_STACK(r14, 3 * 8);
CFI_REG_ON_STACK(r15, 4 * 8);
movq %rdx, (5 * 8)(%rsp); # SRC
movq %rsi, (6 * 8)(%rsp); # DST
movq %rcx, (7 * 8)(%rsp); # NBLKS
/* Load constants */
movdqa .Lcounter1 rRIP, X4;
movdqa .Lshuf_rol8 rRIP, X5;
movdqa .Lshuf_rol16 rRIP, X6;
/* Load state */
movdqu (0 * 4)(INPUT), X10;
movdqu (4 * 4)(INPUT), X11;
movdqu (8 * 4)(INPUT), X12;
movdqu (12 * 4)(INPUT), X13;
POLY1305_LOAD_STATE();
cmpq $2, (7 * 8)(%rsp); #NBLKS
jb .Loop_poly1;
movdqa X10, X0;
movdqa X11, X1;
movdqa X12, X2;
movdqa X13, X3;
movdqa X10, X8;
movdqa X11, X9;
movdqa X12, X14;
movdqa X13, X15;
paddq X4, X15;
/* Process two ChaCha20 blocks and eight Poly1305 blocks. */
movl $20, (8 * 8 + 4)(%rsp);
.Lround2_with_poly1305_outer:
movl $8, (8 * 8)(%rsp);
.Lround2_with_poly1305_inner:
POLY1305_BLOCK_PART1(0 * 16);
QUARTERROUND4(X0, X1, X2, X3, X5, X6, X7, 0x39, 0x4e, 0x93);
lea (1 * 16)(POLY_RSRC), POLY_RSRC;
POLY1305_BLOCK_PART2();
QUARTERROUND4(X8, X9, X14, X15, X5, X6, X7, 0x39, 0x4e, 0x93);
POLY1305_BLOCK_PART3();
QUARTERROUND4(X0, X1, X2, X3, X5, X6, X7, 0x93, 0x4e, 0x39);
POLY1305_BLOCK_PART4();
QUARTERROUND4(X8, X9, X14, X15, X5, X6, X7, 0x93, 0x4e, 0x39);
POLY1305_BLOCK_PART5();
subl $2, (8 * 8)(%rsp);
jnz .Lround2_with_poly1305_inner;
QUARTERROUND4(X0, X1, X2, X3, X5, X6, X7, 0x39, 0x4e, 0x93);
QUARTERROUND4(X8, X9, X14, X15, X5, X6, X7, 0x39, 0x4e, 0x93);
QUARTERROUND4(X0, X1, X2, X3, X5, X6, X7, 0x93, 0x4e, 0x39);
QUARTERROUND4(X8, X9, X14, X15, X5, X6, X7, 0x93, 0x4e, 0x39);
subl $10, (8 * 8 + 4)(%rsp);
jnz .Lround2_with_poly1305_outer;
movq (5 * 8)(%rsp), SRC;
movq (6 * 8)(%rsp), DST;
PLUS(X0, X10);
PLUS(X1, X11);
PLUS(X2, X12);
PLUS(X3, X13);
/* Update counter */
paddq X4, X13;
PLUS(X8, X10);
PLUS(X9, X11);
PLUS(X14, X12);
PLUS(X15, X13);
/* Update counter */
paddq X4, X13;
xor_src_dst(DST, SRC, 0 * 4, X0, X7);
xor_src_dst(DST, SRC, 4 * 4, X1, X7);
xor_src_dst(DST, SRC, 8 * 4, X2, X7);
xor_src_dst(DST, SRC, 12 * 4, X3, X7);
xor_src_dst(DST, SRC, 16 * 4, X8, X7);
xor_src_dst(DST, SRC, 20 * 4, X9, X7);
xor_src_dst(DST, SRC, 24 * 4, X14, X7);
xor_src_dst(DST, SRC, 28 * 4, X15, X7);
clear(X8);
clear(X9);
clear(X14);
clear(X15);
subq $2, (7 * 8)(%rsp); # NBLKS
lea (2 * 64)(SRC), SRC;
lea (2 * 64)(DST), DST;
movq SRC, (5 * 8)(%rsp);
movq DST, (6 * 8)(%rsp);
jz .Ldone_poly1;
.Loop_poly1:
movdqa X10, X0;
movdqa X11, X1;
movdqa X12, X2;
movdqa X13, X3;
/* Process one ChaCha20 block and four Poly1305 blocks. */
movl $20, (8 * 8 + 4)(%rsp);
.Lround1_with_poly1305_outer:
movl $8, (8 * 8)(%rsp);
.Lround1_with_poly1305_inner:
POLY1305_BLOCK_PART1(0 * 16);
QUARTERROUND4(X0, X1, X2, X3, X5, X6, X7, 0x39, 0x4e, 0x93);
POLY1305_BLOCK_PART2();
QUARTERROUND4(X0, X1, X2, X3, X5, X6, X7, 0x93, 0x4e, 0x39);
lea (1 * 16)(POLY_RSRC), POLY_RSRC;
POLY1305_BLOCK_PART3();
QUARTERROUND4(X0, X1, X2, X3, X5, X6, X7, 0x39, 0x4e, 0x93);
POLY1305_BLOCK_PART4();
QUARTERROUND4(X0, X1, X2, X3, X5, X6, X7, 0x93, 0x4e, 0x39);
POLY1305_BLOCK_PART5();
subl $4, (8 * 8)(%rsp);
jnz .Lround1_with_poly1305_inner;
QUARTERROUND4(X0, X1, X2, X3, X5, X6, X7, 0x39, 0x4e, 0x93);
QUARTERROUND4(X0, X1, X2, X3, X5, X6, X7, 0x93, 0x4e, 0x39);
subl $10, (8 * 8 + 4)(%rsp);
jnz .Lround1_with_poly1305_outer;
movq (5 * 8)(%rsp), SRC;
movq (6 * 8)(%rsp), DST;
PLUS(X0, X10);
PLUS(X1, X11);
PLUS(X2, X12);
PLUS(X3, X13);
/* Update counter */
paddq X4, X13;
xor_src_dst(DST, SRC, 0 * 4, X0, X7);
xor_src_dst(DST, SRC, 4 * 4, X1, X7);
xor_src_dst(DST, SRC, 8 * 4, X2, X7);
xor_src_dst(DST, SRC, 12 * 4, X3, X7);
subq $1, (7 * 8)(%rsp); # NBLKS
lea (64)(SRC), SRC;
lea (64)(DST), DST;
movq SRC, (5 * 8)(%rsp);
movq DST, (6 * 8)(%rsp);
jnz .Loop_poly1;
.Ldone_poly1:
/* Store state */
POLY1305_STORE_STATE();
movdqu X13, (12 * 4)(INPUT);
/* clear the used vector registers */
clear(X0);
clear(X1);
clear(X2);
clear(X3);
clear(X4);
clear(X5);
clear(X6);
clear(X7);
clear(X10);
clear(X11);
clear(X12);
clear(X13);
movq (0 * 8)(%rsp), %rbx;
movq (1 * 8)(%rsp), %r12;
movq (2 * 8)(%rsp), %r13;
movq (3 * 8)(%rsp), %r14;
movq (4 * 8)(%rsp), %r15;
CFI_RESTORE(%rbx);
CFI_RESTORE(%r12);
CFI_RESTORE(%r13);
CFI_RESTORE(%r14);
CFI_RESTORE(%r15);
xorl %eax, %eax;
leave;
CFI_LEAVE();
ret;
CFI_ENDPROC();
ELF(.size _gcry_chacha20_poly1305_amd64_ssse3_blocks1,
.-_gcry_chacha20_poly1305_amd64_ssse3_blocks1;)
#endif /*defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS)*/
#endif /*__x86_64*/