/* sm4-avx2-amd64.S  -  AVX2 implementation of SM4 cipher
 *
 * Copyright (C) 2020, 2022-2023 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
 * License along with this program; if not, see <http://www.gnu.org/licenses/>.
 */

/* Based on SM4 AES-NI work by Markku-Juhani O. Saarinen at:
 *  https://github.com/mjosaarinen/sm4ni
 */

#include <config.h>

#ifdef __x86_64
#if (defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS) || \
     defined(HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS)) && \
    defined(ENABLE_AESNI_SUPPORT) && defined(ENABLE_AVX2_SUPPORT)

#include "asm-common-amd64.h"

/* vector registers */
#define RX0          %ymm0
#define RX1          %ymm1
#define MASK_4BIT    %ymm2
#define RTMP0        %ymm3
#define RTMP1        %ymm4
#define RTMP2        %ymm5
#define RTMP3        %ymm6
#define RTMP4        %ymm7

#define RA0          %ymm8
#define RA1          %ymm9
#define RA2          %ymm10
#define RA3          %ymm11
#define RA0x         %xmm8
#define RA1x         %xmm9
#define RA2x         %xmm10
#define RA3x         %xmm11

#define RB0          %ymm12
#define RB1          %ymm13
#define RB2          %ymm14
#define RB3          %ymm15
#define RB0x         %xmm12
#define RB1x         %xmm13
#define RB2x         %xmm14
#define RB3x         %xmm15

#define RNOT         %ymm0
#define RBSWAP       %ymm1

#define RX0x         %xmm0
#define RX1x         %xmm1
#define MASK_4BITx   %xmm2

#define RNOTx        %xmm0
#define RBSWAPx      %xmm1

#define RTMP0x       %xmm3
#define RTMP1x       %xmm4
#define RTMP2x       %xmm5
#define RTMP3x       %xmm6
#define RTMP4x       %xmm7

/**********************************************************************
  helper macros
 **********************************************************************/

/* Transpose four 32-bit words between 128-bit vector lanes. */
#define transpose_4x4(x0, x1, x2, x3, t1, t2) \
	vpunpckhdq x1, x0, t2; \
	vpunpckldq x1, x0, x0; \
	\
	vpunpckldq x3, x2, t1; \
	vpunpckhdq x3, x2, x2; \
	\
	vpunpckhqdq t1, x0, x1; \
	vpunpcklqdq t1, x0, x0; \
	\
	vpunpckhqdq x2, t2, x3; \
	vpunpcklqdq x2, t2, x2;

/* post-SubByte transform. */
#define transform_pre(x, lo_t, hi_t, mask4bit, tmp0) \
	vpand x, mask4bit, tmp0; \
	vpandn x, mask4bit, x; \
	vpsrld $4, x, x; \
	\
	vpshufb tmp0, lo_t, tmp0; \
	vpshufb x, hi_t, x; \
	vpxor tmp0, x, x;

/* post-SubByte transform. Note: x has been XOR'ed with mask4bit by
 * 'vaeslastenc' instruction. */
#define transform_post(x, lo_t, hi_t, mask4bit, tmp0) \
	vpandn mask4bit, x, tmp0; \
	vpsrld $4, x, x; \
	vpand x, mask4bit, x; \
	\
	vpshufb tmp0, lo_t, tmp0; \
	vpshufb x, hi_t, x; \
	vpxor tmp0, x, x;

/**********************************************************************
  16-way SM4 with AES-NI and AVX
 **********************************************************************/

SECTION_RODATA
.align 16

ELF(.type _sm4_aesni_avx2_consts,@object)
_sm4_aesni_avx2_consts:

/*
 * Following four affine transform look-up tables are from work by
 * Markku-Juhani O. Saarinen, at https://github.com/mjosaarinen/sm4ni
 *
 * These allow exposing SM4 S-Box from AES SubByte.
 */

/* pre-SubByte affine transform, from SM4 field to AES field. */
.Lpre_tf_lo_s:
	.quad 0x9197E2E474720701, 0xC7C1B4B222245157
.Lpre_tf_hi_s:
	.quad 0xE240AB09EB49A200, 0xF052B91BF95BB012

/* post-SubByte affine transform, from AES field to SM4 field. */
.Lpost_tf_lo_s:
	.quad 0x5B67F2CEA19D0834, 0xEDD14478172BBE82
.Lpost_tf_hi_s:
	.quad 0xAE7201DD73AFDC00, 0x11CDBE62CC1063BF

/* For isolating SubBytes from AESENCLAST, inverse shift row */
.Linv_shift_row:
	.byte 0x00, 0x0d, 0x0a, 0x07, 0x04, 0x01, 0x0e, 0x0b
	.byte 0x08, 0x05, 0x02, 0x0f, 0x0c, 0x09, 0x06, 0x03

/* Inverse shift row + Rotate left by 8 bits on 32-bit words with vpshufb */
.Linv_shift_row_rol_8:
	.byte 0x07, 0x00, 0x0d, 0x0a, 0x0b, 0x04, 0x01, 0x0e
	.byte 0x0f, 0x08, 0x05, 0x02, 0x03, 0x0c, 0x09, 0x06

/* Inverse shift row + Rotate left by 16 bits on 32-bit words with vpshufb */
.Linv_shift_row_rol_16:
	.byte 0x0a, 0x07, 0x00, 0x0d, 0x0e, 0x0b, 0x04, 0x01
	.byte 0x02, 0x0f, 0x08, 0x05, 0x06, 0x03, 0x0c, 0x09

/* Inverse shift row + Rotate left by 24 bits on 32-bit words with vpshufb */
.Linv_shift_row_rol_24:
	.byte 0x0d, 0x0a, 0x07, 0x00, 0x01, 0x0e, 0x0b, 0x04
	.byte 0x05, 0x02, 0x0f, 0x08, 0x09, 0x06, 0x03, 0x0c

/* For CTR-mode IV byteswap */
.Lbswap128_mask:
	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0

/* For input word byte-swap */
.Lbswap32_mask:
	.byte 3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12

/* CTR byte addition constants */
.align 32
.Lbige_addb_0_1:
	.byte 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
	.byte 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1
.Lbige_addb_2_3:
	.byte 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2
	.byte 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 3
.Lbige_addb_4_5:
	.byte 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 4
	.byte 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 5
.Lbige_addb_6_7:
	.byte 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 6
	.byte 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 7
.Lbige_addb_8_9:
	.byte 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 8
	.byte 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 9
.Lbige_addb_10_11:
	.byte 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 10
	.byte 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 11
.Lbige_addb_12_13:
	.byte 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 12
	.byte 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 13
.Lbige_addb_14_15:
	.byte 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 14
	.byte 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 15

.align 4
/* 4-bit mask */
.L0f0f0f0f:
	.long 0x0f0f0f0f

.text

.align 16
ELF(.type   __sm4_crypt_blk16,@function;)
__sm4_crypt_blk16:
	/* input:
	 *	%rdi: ctx, CTX
	 *	RA0, RA1, RA2, RA3, RB0, RB1, RB2, RB3: sixteen parallel
	 *						plaintext blocks
	 * output:
	 *	RA0, RA1, RA2, RA3, RB0, RB1, RB2, RB3: sixteen parallel
	 * 						ciphertext blocks
	 */
	CFI_STARTPROC();

	vbroadcasti128 .Lbswap32_mask rRIP, RTMP2;
	vpshufb RTMP2, RA0, RA0;
	vpshufb RTMP2, RA1, RA1;
	vpshufb RTMP2, RA2, RA2;
	vpshufb RTMP2, RA3, RA3;
	vpshufb RTMP2, RB0, RB0;
	vpshufb RTMP2, RB1, RB1;
	vpshufb RTMP2, RB2, RB2;
	vpshufb RTMP2, RB3, RB3;

	vpbroadcastd .L0f0f0f0f rRIP, MASK_4BIT;
	transpose_4x4(RA0, RA1, RA2, RA3, RTMP0, RTMP1);
	transpose_4x4(RB0, RB1, RB2, RB3, RTMP0, RTMP1);

#define ROUND(round, s0, s1, s2, s3, r0, r1, r2, r3) \
	vpbroadcastd (4*(round))(%rdi), RX0; \
	vbroadcasti128 .Lpre_tf_lo_s rRIP, RTMP4; \
	vbroadcasti128 .Lpre_tf_hi_s rRIP, RTMP1; \
	vmovdqa RX0, RX1; \
	vpxor s1, RX0, RX0; \
	vpxor s2, RX0, RX0; \
	vpxor s3, RX0, RX0; /* s1 ^ s2 ^ s3 ^ rk */ \
	    vbroadcasti128 .Lpost_tf_lo_s rRIP, RTMP2; \
	    vbroadcasti128 .Lpost_tf_hi_s rRIP, RTMP3; \
	    vpxor r1, RX1, RX1; \
	    vpxor r2, RX1, RX1; \
	    vpxor r3, RX1, RX1; /* r1 ^ r2 ^ r3 ^ rk */ \
	\
	/* sbox, non-linear part */ \
	transform_pre(RX0, RTMP4, RTMP1, MASK_4BIT, RTMP0); \
	    transform_pre(RX1, RTMP4, RTMP1, MASK_4BIT, RTMP0); \
	vextracti128 $1, RX0, RTMP4x; \
	    vextracti128 $1, RX1, RTMP0x; \
	vaesenclast MASK_4BITx, RX0x, RX0x; \
	vaesenclast MASK_4BITx, RTMP4x, RTMP4x; \
	    vaesenclast MASK_4BITx, RX1x, RX1x; \
	    vaesenclast MASK_4BITx, RTMP0x, RTMP0x; \
	vinserti128 $1, RTMP4x, RX0, RX0; \
	vbroadcasti128 .Linv_shift_row rRIP, RTMP4; \
	    vinserti128 $1, RTMP0x, RX1, RX1; \
	transform_post(RX0, RTMP2, RTMP3, MASK_4BIT, RTMP0); \
	    transform_post(RX1, RTMP2, RTMP3, MASK_4BIT, RTMP0); \
	\
	/* linear part */ \
	vpshufb RTMP4, RX0, RTMP0; \
	vpxor RTMP0, s0, s0; /* s0 ^ x */ \
	    vpshufb RTMP4, RX1, RTMP2; \
	    vbroadcasti128 .Linv_shift_row_rol_8 rRIP, RTMP4; \
	    vpxor RTMP2, r0, r0; /* r0 ^ x */ \
	vpshufb RTMP4, RX0, RTMP1; \
	vpxor RTMP1, RTMP0, RTMP0; /* x ^ rol(x,8) */ \
	    vpshufb RTMP4, RX1, RTMP3; \
	    vbroadcasti128 .Linv_shift_row_rol_16 rRIP, RTMP4; \
	    vpxor RTMP3, RTMP2, RTMP2; /* x ^ rol(x,8) */ \
	vpshufb RTMP4, RX0, RTMP1; \
	vpxor RTMP1, RTMP0, RTMP0; /* x ^ rol(x,8) ^ rol(x,16) */ \
	    vpshufb RTMP4, RX1, RTMP3; \
	    vbroadcasti128 .Linv_shift_row_rol_24 rRIP, RTMP4; \
	    vpxor RTMP3, RTMP2, RTMP2; /* x ^ rol(x,8) ^ rol(x,16) */ \
	vpshufb RTMP4, RX0, RTMP1; \
	vpxor RTMP1, s0, s0; /* s0 ^ x ^ rol(x,24) */ \
	vpslld $2, RTMP0, RTMP1; \
	vpsrld $30, RTMP0, RTMP0; \
	vpxor RTMP0, s0, s0;  \
	vpxor RTMP1, s0, s0; /* s0 ^ x ^ rol(x,2) ^ rol(x,10) ^ rol(x,18) ^ rol(x,24) */ \
	    vpshufb RTMP4, RX1, RTMP3; \
	    vpxor RTMP3, r0, r0; /* r0 ^ x ^ rol(x,24) */ \
	    vpslld $2, RTMP2, RTMP3; \
	    vpsrld $30, RTMP2, RTMP2; \
	    vpxor RTMP2, r0, r0;  \
	    vpxor RTMP3, r0, r0; /* r0 ^ x ^ rol(x,2) ^ rol(x,10) ^ rol(x,18) ^ rol(x,24) */

	leaq (32*4)(%rdi), %rax;
.align 16
.Lroundloop_blk16:
	ROUND(0, RA0, RA1, RA2, RA3, RB0, RB1, RB2, RB3);
	ROUND(1, RA1, RA2, RA3, RA0, RB1, RB2, RB3, RB0);
	ROUND(2, RA2, RA3, RA0, RA1, RB2, RB3, RB0, RB1);
	ROUND(3, RA3, RA0, RA1, RA2, RB3, RB0, RB1, RB2);
	leaq (4*4)(%rdi), %rdi;
	cmpq %rax, %rdi;
	jne .Lroundloop_blk16;

#undef ROUND

	vbroadcasti128 .Lbswap128_mask rRIP, RTMP2;

	transpose_4x4(RA0, RA1, RA2, RA3, RTMP0, RTMP1);
	transpose_4x4(RB0, RB1, RB2, RB3, RTMP0, RTMP1);
	vpshufb RTMP2, RA0, RA0;
	vpshufb RTMP2, RA1, RA1;
	vpshufb RTMP2, RA2, RA2;
	vpshufb RTMP2, RA3, RA3;
	vpshufb RTMP2, RB0, RB0;
	vpshufb RTMP2, RB1, RB1;
	vpshufb RTMP2, RB2, RB2;
	vpshufb RTMP2, RB3, RB3;

	ret_spec_stop;
	CFI_ENDPROC();
ELF(.size __sm4_crypt_blk16,.-__sm4_crypt_blk16;)

.align 16
.globl _gcry_sm4_aesni_avx2_crypt_blk1_16
ELF(.type   _gcry_sm4_aesni_avx2_crypt_blk1_16,@function;)
_gcry_sm4_aesni_avx2_crypt_blk1_16:
	/* input:
	 *	%rdi: round key array, CTX
	 *	%rsi: dst (1..16 blocks)
	 *	%rdx: src (1..16 blocks)
	 *	%rcx: num blocks (1..16)
	 */
	CFI_STARTPROC();

#define LOAD_INPUT(offset, yreg) \
	cmpq $(1 + 2 * (offset)), %rcx; \
	jb .Lblk16_load_input_done; \
	ja 1f; \
	  vmovdqu (offset) * 32(%rdx), yreg##x; \
	  jmp .Lblk16_load_input_done; \
	1: \
	  vmovdqu (offset) * 32(%rdx), yreg;

	LOAD_INPUT(0, RA0);
	LOAD_INPUT(1, RA1);
	LOAD_INPUT(2, RA2);
	LOAD_INPUT(3, RA3);
	LOAD_INPUT(4, RB0);
	LOAD_INPUT(5, RB1);
	LOAD_INPUT(6, RB2);
	LOAD_INPUT(7, RB3);
#undef LOAD_INPUT

.Lblk16_load_input_done:
	call __sm4_crypt_blk16;

#define STORE_OUTPUT(yreg, offset) \
	cmpq $(1 + 2 * (offset)), %rcx; \
	jb .Lblk16_store_output_done; \
	ja 1f; \
	  vmovdqu yreg##x, (offset) * 32(%rsi); \
	  jmp .Lblk16_store_output_done; \
	1: \
	  vmovdqu yreg, (offset) * 32(%rsi);

	STORE_OUTPUT(RA0, 0);
	STORE_OUTPUT(RA1, 1);
	STORE_OUTPUT(RA2, 2);
	STORE_OUTPUT(RA3, 3);
	STORE_OUTPUT(RB0, 4);
	STORE_OUTPUT(RB1, 5);
	STORE_OUTPUT(RB2, 6);
	STORE_OUTPUT(RB3, 7);
#undef STORE_OUTPUT

.Lblk16_store_output_done:
	vzeroall;
	xorl %eax, %eax;
	ret_spec_stop;
	CFI_ENDPROC();
ELF(.size _gcry_sm4_aesni_avx2_crypt_blk1_16,.-_gcry_sm4_aesni_avx2_crypt_blk1_16;)

#define inc_le128(x, minus_one, tmp) \
	vpcmpeqq minus_one, x, tmp; \
	vpsubq minus_one, x, x; \
	vpslldq $8, tmp, tmp; \
	vpsubq tmp, x, x;

.align 16
.globl _gcry_sm4_aesni_avx2_ctr_enc
ELF(.type   _gcry_sm4_aesni_avx2_ctr_enc,@function;)
_gcry_sm4_aesni_avx2_ctr_enc:
	/* input:
	 *	%rdi: ctx, CTX
	 *	%rsi: dst (16 blocks)
	 *	%rdx: src (16 blocks)
	 *	%rcx: iv (big endian, 128bit)
	 */
	CFI_STARTPROC();

	cmpb $(0x100 - 16), 15(%rcx);
	jbe .Lctr_byteadd;

	movq 8(%rcx), %rax;
	bswapq %rax;

	vbroadcasti128 .Lbswap128_mask rRIP, RTMP3;
	vpcmpeqd RNOT, RNOT, RNOT;
	vpsrldq $8, RNOT, RNOT;   /* ab: -1:0 ; cd: -1:0 */
	vpaddq RNOT, RNOT, RTMP2; /* ab: -2:0 ; cd: -2:0 */

	/* load IV and byteswap */
	vmovdqu (%rcx), RTMP4x;
	vpshufb RTMP3x, RTMP4x, RTMP4x;
	vmovdqa RTMP4x, RTMP0x;
	inc_le128(RTMP4x, RNOTx, RTMP1x);
	vinserti128 $1, RTMP4x, RTMP0, RTMP0;
	vpshufb RTMP3, RTMP0, RA0; /* +1 ; +0 */

	/* check need for handling 64-bit overflow and carry */
	cmpq $(0xffffffffffffffff - 16), %rax;
	ja .Lhandle_ctr_carry;

	/* construct IVs */
	vpsubq RTMP2, RTMP0, RTMP0; /* +3 ; +2 */
	vpshufb RTMP3, RTMP0, RA1;
	vpsubq RTMP2, RTMP0, RTMP0; /* +5 ; +4 */
	vpshufb RTMP3, RTMP0, RA2;
	vpsubq RTMP2, RTMP0, RTMP0; /* +7 ; +6 */
	vpshufb RTMP3, RTMP0, RA3;
	vpsubq RTMP2, RTMP0, RTMP0; /* +9 ; +8 */
	vpshufb RTMP3, RTMP0, RB0;
	vpsubq RTMP2, RTMP0, RTMP0; /* +11 ; +10 */
	vpshufb RTMP3, RTMP0, RB1;
	vpsubq RTMP2, RTMP0, RTMP0; /* +13 ; +12 */
	vpshufb RTMP3, RTMP0, RB2;
	vpsubq RTMP2, RTMP0, RTMP0; /* +15 ; +14 */
	vpshufb RTMP3, RTMP0, RB3;
	vpsubq RTMP2, RTMP0, RTMP0; /* +16 */
	vpshufb RTMP3x, RTMP0x, RTMP0x;

	jmp .Lctr_carry_done;

.Lhandle_ctr_carry:
	/* construct IVs */
	inc_le128(RTMP0, RNOT, RTMP1);
	inc_le128(RTMP0, RNOT, RTMP1);
	vpshufb RTMP3, RTMP0, RA1; /* +3 ; +2 */
	inc_le128(RTMP0, RNOT, RTMP1);
	inc_le128(RTMP0, RNOT, RTMP1);
	vpshufb RTMP3, RTMP0, RA2; /* +5 ; +4 */
	inc_le128(RTMP0, RNOT, RTMP1);
	inc_le128(RTMP0, RNOT, RTMP1);
	vpshufb RTMP3, RTMP0, RA3; /* +7 ; +6 */
	inc_le128(RTMP0, RNOT, RTMP1);
	inc_le128(RTMP0, RNOT, RTMP1);
	vpshufb RTMP3, RTMP0, RB0; /* +9 ; +8 */
	inc_le128(RTMP0, RNOT, RTMP1);
	inc_le128(RTMP0, RNOT, RTMP1);
	vpshufb RTMP3, RTMP0, RB1; /* +11 ; +10 */
	inc_le128(RTMP0, RNOT, RTMP1);
	inc_le128(RTMP0, RNOT, RTMP1);
	vpshufb RTMP3, RTMP0, RB2; /* +13 ; +12 */
	inc_le128(RTMP0, RNOT, RTMP1);
	inc_le128(RTMP0, RNOT, RTMP1);
	vpshufb RTMP3, RTMP0, RB3; /* +15 ; +14 */
	inc_le128(RTMP0, RNOT, RTMP1);
	vextracti128 $1, RTMP0, RTMP0x;
	vpshufb RTMP3x, RTMP0x, RTMP0x; /* +16 */

.Lctr_carry_done:
	/* store new IV */
	vmovdqu RTMP0x, (%rcx);

.align 8
.Lload_ctr_done:
	call __sm4_crypt_blk16;

	vpxor (0 * 32)(%rdx), RA0, RA0;
	vpxor (1 * 32)(%rdx), RA1, RA1;
	vpxor (2 * 32)(%rdx), RA2, RA2;
	vpxor (3 * 32)(%rdx), RA3, RA3;
	vpxor (4 * 32)(%rdx), RB0, RB0;
	vpxor (5 * 32)(%rdx), RB1, RB1;
	vpxor (6 * 32)(%rdx), RB2, RB2;
	vpxor (7 * 32)(%rdx), RB3, RB3;

	vmovdqu RA0, (0 * 32)(%rsi);
	vmovdqu RA1, (1 * 32)(%rsi);
	vmovdqu RA2, (2 * 32)(%rsi);
	vmovdqu RA3, (3 * 32)(%rsi);
	vmovdqu RB0, (4 * 32)(%rsi);
	vmovdqu RB1, (5 * 32)(%rsi);
	vmovdqu RB2, (6 * 32)(%rsi);
	vmovdqu RB3, (7 * 32)(%rsi);

	vzeroall;

	ret_spec_stop;

.align 8
.Lctr_byteadd_full_ctr_carry:
	movq 8(%rcx), %r11;
	movq (%rcx), %r10;
	bswapq %r11;
	bswapq %r10;
	addq $16, %r11;
	adcq $0, %r10;
	bswapq %r11;
	bswapq %r10;
	movq %r11, 8(%rcx);
	movq %r10, (%rcx);
	jmp .Lctr_byteadd_ymm;
.align 8
.Lctr_byteadd:
	vbroadcasti128 (%rcx), RB3;
	je .Lctr_byteadd_full_ctr_carry;
	addb $16, 15(%rcx);
.Lctr_byteadd_ymm:
	vpaddb .Lbige_addb_0_1 rRIP, RB3, RA0;
	vpaddb .Lbige_addb_2_3 rRIP, RB3, RA1;
	vpaddb .Lbige_addb_4_5 rRIP, RB3, RA2;
	vpaddb .Lbige_addb_6_7 rRIP, RB3, RA3;
	vpaddb .Lbige_addb_8_9 rRIP, RB3, RB0;
	vpaddb .Lbige_addb_10_11 rRIP, RB3, RB1;
	vpaddb .Lbige_addb_12_13 rRIP, RB3, RB2;
	vpaddb .Lbige_addb_14_15 rRIP, RB3, RB3;

	jmp .Lload_ctr_done;
	CFI_ENDPROC();
ELF(.size _gcry_sm4_aesni_avx2_ctr_enc,.-_gcry_sm4_aesni_avx2_ctr_enc;)

.align 16
.globl _gcry_sm4_aesni_avx2_cbc_dec
ELF(.type   _gcry_sm4_aesni_avx2_cbc_dec,@function;)
_gcry_sm4_aesni_avx2_cbc_dec:
	/* input:
	 *	%rdi: ctx, CTX
	 *	%rsi: dst (16 blocks)
	 *	%rdx: src (16 blocks)
	 *	%rcx: iv
	 */
	CFI_STARTPROC();

	vmovdqu (0 * 32)(%rdx), RA0;
	vmovdqu (1 * 32)(%rdx), RA1;
	vmovdqu (2 * 32)(%rdx), RA2;
	vmovdqu (3 * 32)(%rdx), RA3;
	vmovdqu (4 * 32)(%rdx), RB0;
	vmovdqu (5 * 32)(%rdx), RB1;
	vmovdqu (6 * 32)(%rdx), RB2;
	vmovdqu (7 * 32)(%rdx), RB3;

	call __sm4_crypt_blk16;

	vmovdqu (%rcx), RNOTx;
	vinserti128 $1, (%rdx), RNOT, RNOT;
	vpxor RNOT, RA0, RA0;
	vpxor (0 * 32 + 16)(%rdx), RA1, RA1;
	vpxor (1 * 32 + 16)(%rdx), RA2, RA2;
	vpxor (2 * 32 + 16)(%rdx), RA3, RA3;
	vpxor (3 * 32 + 16)(%rdx), RB0, RB0;
	vpxor (4 * 32 + 16)(%rdx), RB1, RB1;
	vpxor (5 * 32 + 16)(%rdx), RB2, RB2;
	vpxor (6 * 32 + 16)(%rdx), RB3, RB3;
	vmovdqu (7 * 32 + 16)(%rdx), RNOTx;
	vmovdqu RNOTx, (%rcx); /* store new IV */

	vmovdqu RA0, (0 * 32)(%rsi);
	vmovdqu RA1, (1 * 32)(%rsi);
	vmovdqu RA2, (2 * 32)(%rsi);
	vmovdqu RA3, (3 * 32)(%rsi);
	vmovdqu RB0, (4 * 32)(%rsi);
	vmovdqu RB1, (5 * 32)(%rsi);
	vmovdqu RB2, (6 * 32)(%rsi);
	vmovdqu RB3, (7 * 32)(%rsi);

	vzeroall;

	ret_spec_stop;
	CFI_ENDPROC();
ELF(.size _gcry_sm4_aesni_avx2_cbc_dec,.-_gcry_sm4_aesni_avx2_cbc_dec;)

.align 16
.globl _gcry_sm4_aesni_avx2_cfb_dec
ELF(.type   _gcry_sm4_aesni_avx2_cfb_dec,@function;)
_gcry_sm4_aesni_avx2_cfb_dec:
	/* input:
	 *	%rdi: ctx, CTX
	 *	%rsi: dst (16 blocks)
	 *	%rdx: src (16 blocks)
	 *	%rcx: iv
	 */
	CFI_STARTPROC();

	/* Load input */
	vmovdqu (%rcx), RNOTx;
	vinserti128 $1, (%rdx), RNOT, RA0;
	vmovdqu (0 * 32 + 16)(%rdx), RA1;
	vmovdqu (1 * 32 + 16)(%rdx), RA2;
	vmovdqu (2 * 32 + 16)(%rdx), RA3;
	vmovdqu (3 * 32 + 16)(%rdx), RB0;
	vmovdqu (4 * 32 + 16)(%rdx), RB1;
	vmovdqu (5 * 32 + 16)(%rdx), RB2;
	vmovdqu (6 * 32 + 16)(%rdx), RB3;

	/* Update IV */
	vmovdqu (7 * 32 + 16)(%rdx), RNOTx;
	vmovdqu RNOTx, (%rcx);

	call __sm4_crypt_blk16;

	vpxor (0 * 32)(%rdx), RA0, RA0;
	vpxor (1 * 32)(%rdx), RA1, RA1;
	vpxor (2 * 32)(%rdx), RA2, RA2;
	vpxor (3 * 32)(%rdx), RA3, RA3;
	vpxor (4 * 32)(%rdx), RB0, RB0;
	vpxor (5 * 32)(%rdx), RB1, RB1;
	vpxor (6 * 32)(%rdx), RB2, RB2;
	vpxor (7 * 32)(%rdx), RB3, RB3;

	vmovdqu RA0, (0 * 32)(%rsi);
	vmovdqu RA1, (1 * 32)(%rsi);
	vmovdqu RA2, (2 * 32)(%rsi);
	vmovdqu RA3, (3 * 32)(%rsi);
	vmovdqu RB0, (4 * 32)(%rsi);
	vmovdqu RB1, (5 * 32)(%rsi);
	vmovdqu RB2, (6 * 32)(%rsi);
	vmovdqu RB3, (7 * 32)(%rsi);

	vzeroall;

	ret_spec_stop;
	CFI_ENDPROC();
ELF(.size _gcry_sm4_aesni_avx2_cfb_dec,.-_gcry_sm4_aesni_avx2_cfb_dec;)

.align 16
.globl _gcry_sm4_aesni_avx2_ocb_enc
ELF(.type _gcry_sm4_aesni_avx2_ocb_enc,@function;)

_gcry_sm4_aesni_avx2_ocb_enc:
	/* input:
	 *	%rdi: ctx, CTX
	 *	%rsi: dst (16 blocks)
	 *	%rdx: src (16 blocks)
	 *	%rcx: offset
	 *	%r8 : checksum
	 *	%r9 : L pointers (void *L[16])
	 */
	CFI_STARTPROC();

	subq $(4 * 8), %rsp;
	CFI_ADJUST_CFA_OFFSET(4 * 8);

	movq %r10, (0 * 8)(%rsp);
	movq %r11, (1 * 8)(%rsp);
	movq %r12, (2 * 8)(%rsp);
	movq %r13, (3 * 8)(%rsp);
	CFI_REL_OFFSET(%r10, 0 * 8);
	CFI_REL_OFFSET(%r11, 1 * 8);
	CFI_REL_OFFSET(%r12, 2 * 8);
	CFI_REL_OFFSET(%r13, 3 * 8);

	vmovdqu (%rcx), RTMP0x;
	vmovdqu (%r8), RTMP1x;

	/* Offset_i = Offset_{i-1} xor L_{ntz(i)} */
	/* Checksum_i = Checksum_{i-1} xor P_i  */
	/* C_i = Offset_i xor ENCIPHER(K, P_i xor Offset_i)  */

#define OCB_INPUT(n, l0reg, l1reg, yreg) \
	  vmovdqu (n * 32)(%rdx), yreg; \
	  vpxor (l0reg), RTMP0x, RNOTx; \
	  vpxor (l1reg), RNOTx, RTMP0x; \
	  vinserti128 $1, RTMP0x, RNOT, RNOT; \
	  vpxor yreg, RTMP1, RTMP1; \
	  vpxor yreg, RNOT, yreg; \
	  vmovdqu RNOT, (n * 32)(%rsi);

	movq (0 * 8)(%r9), %r10;
	movq (1 * 8)(%r9), %r11;
	movq (2 * 8)(%r9), %r12;
	movq (3 * 8)(%r9), %r13;
	OCB_INPUT(0, %r10, %r11, RA0);
	OCB_INPUT(1, %r12, %r13, RA1);
	movq (4 * 8)(%r9), %r10;
	movq (5 * 8)(%r9), %r11;
	movq (6 * 8)(%r9), %r12;
	movq (7 * 8)(%r9), %r13;
	OCB_INPUT(2, %r10, %r11, RA2);
	OCB_INPUT(3, %r12, %r13, RA3);
	movq (8 * 8)(%r9), %r10;
	movq (9 * 8)(%r9), %r11;
	movq (10 * 8)(%r9), %r12;
	movq (11 * 8)(%r9), %r13;
	OCB_INPUT(4, %r10, %r11, RB0);
	OCB_INPUT(5, %r12, %r13, RB1);
	movq (12 * 8)(%r9), %r10;
	movq (13 * 8)(%r9), %r11;
	movq (14 * 8)(%r9), %r12;
	movq (15 * 8)(%r9), %r13;
	OCB_INPUT(6, %r10, %r11, RB2);
	OCB_INPUT(7, %r12, %r13, RB3);
#undef OCB_INPUT

	vextracti128 $1, RTMP1, RNOTx;
	vmovdqu RTMP0x, (%rcx);
	vpxor RNOTx, RTMP1x, RTMP1x;
	vmovdqu RTMP1x, (%r8);

	movq (0 * 8)(%rsp), %r10;
	movq (1 * 8)(%rsp), %r11;
	movq (2 * 8)(%rsp), %r12;
	movq (3 * 8)(%rsp), %r13;
	CFI_RESTORE(%r10);
	CFI_RESTORE(%r11);
	CFI_RESTORE(%r12);
	CFI_RESTORE(%r13);

	call __sm4_crypt_blk16;

	addq $(4 * 8), %rsp;
	CFI_ADJUST_CFA_OFFSET(-4 * 8);

	vpxor (0 * 32)(%rsi), RA0, RA0;
	vpxor (1 * 32)(%rsi), RA1, RA1;
	vpxor (2 * 32)(%rsi), RA2, RA2;
	vpxor (3 * 32)(%rsi), RA3, RA3;
	vpxor (4 * 32)(%rsi), RB0, RB0;
	vpxor (5 * 32)(%rsi), RB1, RB1;
	vpxor (6 * 32)(%rsi), RB2, RB2;
	vpxor (7 * 32)(%rsi), RB3, RB3;

	vmovdqu RA0, (0 * 32)(%rsi);
	vmovdqu RA1, (1 * 32)(%rsi);
	vmovdqu RA2, (2 * 32)(%rsi);
	vmovdqu RA3, (3 * 32)(%rsi);
	vmovdqu RB0, (4 * 32)(%rsi);
	vmovdqu RB1, (5 * 32)(%rsi);
	vmovdqu RB2, (6 * 32)(%rsi);
	vmovdqu RB3, (7 * 32)(%rsi);

	vzeroall;

	ret_spec_stop;
	CFI_ENDPROC();
ELF(.size _gcry_sm4_aesni_avx2_ocb_enc,.-_gcry_sm4_aesni_avx2_ocb_enc;)

.align 16
.globl _gcry_sm4_aesni_avx2_ocb_dec
ELF(.type _gcry_sm4_aesni_avx2_ocb_dec,@function;)

_gcry_sm4_aesni_avx2_ocb_dec:
	/* input:
	 *	%rdi: ctx, CTX
	 *	%rsi: dst (16 blocks)
	 *	%rdx: src (16 blocks)
	 *	%rcx: offset
	 *	%r8 : checksum
	 *	%r9 : L pointers (void *L[16])
	 */
	CFI_STARTPROC();

	subq $(4 * 8), %rsp;
	CFI_ADJUST_CFA_OFFSET(4 * 8);

	movq %r10, (0 * 8)(%rsp);
	movq %r11, (1 * 8)(%rsp);
	movq %r12, (2 * 8)(%rsp);
	movq %r13, (3 * 8)(%rsp);
	CFI_REL_OFFSET(%r10, 0 * 8);
	CFI_REL_OFFSET(%r11, 1 * 8);
	CFI_REL_OFFSET(%r12, 2 * 8);
	CFI_REL_OFFSET(%r13, 3 * 8);

	vmovdqu (%rcx), RTMP0x;

	/* Offset_i = Offset_{i-1} xor L_{ntz(i)} */
	/* C_i = Offset_i xor ENCIPHER(K, P_i xor Offset_i)  */

#define OCB_INPUT(n, l0reg, l1reg, yreg) \
	  vmovdqu (n * 32)(%rdx), yreg; \
	  vpxor (l0reg), RTMP0x, RNOTx; \
	  vpxor (l1reg), RNOTx, RTMP0x; \
	  vinserti128 $1, RTMP0x, RNOT, RNOT; \
	  vpxor yreg, RNOT, yreg; \
	  vmovdqu RNOT, (n * 32)(%rsi);

	movq (0 * 8)(%r9), %r10;
	movq (1 * 8)(%r9), %r11;
	movq (2 * 8)(%r9), %r12;
	movq (3 * 8)(%r9), %r13;
	OCB_INPUT(0, %r10, %r11, RA0);
	OCB_INPUT(1, %r12, %r13, RA1);
	movq (4 * 8)(%r9), %r10;
	movq (5 * 8)(%r9), %r11;
	movq (6 * 8)(%r9), %r12;
	movq (7 * 8)(%r9), %r13;
	OCB_INPUT(2, %r10, %r11, RA2);
	OCB_INPUT(3, %r12, %r13, RA3);
	movq (8 * 8)(%r9), %r10;
	movq (9 * 8)(%r9), %r11;
	movq (10 * 8)(%r9), %r12;
	movq (11 * 8)(%r9), %r13;
	OCB_INPUT(4, %r10, %r11, RB0);
	OCB_INPUT(5, %r12, %r13, RB1);
	movq (12 * 8)(%r9), %r10;
	movq (13 * 8)(%r9), %r11;
	movq (14 * 8)(%r9), %r12;
	movq (15 * 8)(%r9), %r13;
	OCB_INPUT(6, %r10, %r11, RB2);
	OCB_INPUT(7, %r12, %r13, RB3);
#undef OCB_INPUT

	vmovdqu RTMP0x, (%rcx);

	movq (0 * 8)(%rsp), %r10;
	movq (1 * 8)(%rsp), %r11;
	movq (2 * 8)(%rsp), %r12;
	movq (3 * 8)(%rsp), %r13;
	CFI_RESTORE(%r10);
	CFI_RESTORE(%r11);
	CFI_RESTORE(%r12);
	CFI_RESTORE(%r13);

	call __sm4_crypt_blk16;

	addq $(4 * 8), %rsp;
	CFI_ADJUST_CFA_OFFSET(-4 * 8);

	vmovdqu (%r8), RTMP1x;

	vpxor (0 * 32)(%rsi), RA0, RA0;
	vpxor (1 * 32)(%rsi), RA1, RA1;
	vpxor (2 * 32)(%rsi), RA2, RA2;
	vpxor (3 * 32)(%rsi), RA3, RA3;
	vpxor (4 * 32)(%rsi), RB0, RB0;
	vpxor (5 * 32)(%rsi), RB1, RB1;
	vpxor (6 * 32)(%rsi), RB2, RB2;
	vpxor (7 * 32)(%rsi), RB3, RB3;

	/* Checksum_i = Checksum_{i-1} xor P_i  */

	vmovdqu RA0, (0 * 32)(%rsi);
	vpxor RA0, RTMP1, RTMP1;
	vmovdqu RA1, (1 * 32)(%rsi);
	vpxor RA1, RTMP1, RTMP1;
	vmovdqu RA2, (2 * 32)(%rsi);
	vpxor RA2, RTMP1, RTMP1;
	vmovdqu RA3, (3 * 32)(%rsi);
	vpxor RA3, RTMP1, RTMP1;
	vmovdqu RB0, (4 * 32)(%rsi);
	vpxor RB0, RTMP1, RTMP1;
	vmovdqu RB1, (5 * 32)(%rsi);
	vpxor RB1, RTMP1, RTMP1;
	vmovdqu RB2, (6 * 32)(%rsi);
	vpxor RB2, RTMP1, RTMP1;
	vmovdqu RB3, (7 * 32)(%rsi);
	vpxor RB3, RTMP1, RTMP1;

	vextracti128 $1, RTMP1, RNOTx;
	vpxor RNOTx, RTMP1x, RTMP1x;
	vmovdqu RTMP1x, (%r8);

	vzeroall;

	ret_spec_stop;
	CFI_ENDPROC();
ELF(.size _gcry_sm4_aesni_avx2_ocb_dec,.-_gcry_sm4_aesni_avx2_ocb_dec;)

.align 16
.globl _gcry_sm4_aesni_avx2_ocb_auth
ELF(.type _gcry_sm4_aesni_avx2_ocb_auth,@function;)

_gcry_sm4_aesni_avx2_ocb_auth:
	/* input:
	 *	%rdi: ctx, CTX
	 *	%rsi: abuf (16 blocks)
	 *	%rdx: offset
	 *	%rcx: checksum
	 *	%r8 : L pointers (void *L[16])
	 */
	CFI_STARTPROC();

	subq $(4 * 8), %rsp;
	CFI_ADJUST_CFA_OFFSET(4 * 8);

	movq %r10, (0 * 8)(%rsp);
	movq %r11, (1 * 8)(%rsp);
	movq %r12, (2 * 8)(%rsp);
	movq %r13, (3 * 8)(%rsp);
	CFI_REL_OFFSET(%r10, 0 * 8);
	CFI_REL_OFFSET(%r11, 1 * 8);
	CFI_REL_OFFSET(%r12, 2 * 8);
	CFI_REL_OFFSET(%r13, 3 * 8);

	vmovdqu (%rdx), RTMP0x;

	/* Offset_i = Offset_{i-1} xor L_{ntz(i)} */
	/* Sum_i = Sum_{i-1} xor ENCIPHER(K, A_i xor Offset_i)  */

#define OCB_INPUT(n, l0reg, l1reg, yreg) \
	  vmovdqu (n * 32)(%rsi), yreg; \
	  vpxor (l0reg), RTMP0x, RNOTx; \
	  vpxor (l1reg), RNOTx, RTMP0x; \
	  vinserti128 $1, RTMP0x, RNOT, RNOT; \
	  vpxor yreg, RNOT, yreg;

	movq (0 * 8)(%r8), %r10;
	movq (1 * 8)(%r8), %r11;
	movq (2 * 8)(%r8), %r12;
	movq (3 * 8)(%r8), %r13;
	OCB_INPUT(0, %r10, %r11, RA0);
	OCB_INPUT(1, %r12, %r13, RA1);
	movq (4 * 8)(%r8), %r10;
	movq (5 * 8)(%r8), %r11;
	movq (6 * 8)(%r8), %r12;
	movq (7 * 8)(%r8), %r13;
	OCB_INPUT(2, %r10, %r11, RA2);
	OCB_INPUT(3, %r12, %r13, RA3);
	movq (8 * 8)(%r8), %r10;
	movq (9 * 8)(%r8), %r11;
	movq (10 * 8)(%r8), %r12;
	movq (11 * 8)(%r8), %r13;
	OCB_INPUT(4, %r10, %r11, RB0);
	OCB_INPUT(5, %r12, %r13, RB1);
	movq (12 * 8)(%r8), %r10;
	movq (13 * 8)(%r8), %r11;
	movq (14 * 8)(%r8), %r12;
	movq (15 * 8)(%r8), %r13;
	OCB_INPUT(6, %r10, %r11, RB2);
	OCB_INPUT(7, %r12, %r13, RB3);
#undef OCB_INPUT

	vmovdqu RTMP0x, (%rdx);

	movq (0 * 8)(%rsp), %r10;
	movq (1 * 8)(%rsp), %r11;
	movq (2 * 8)(%rsp), %r12;
	movq (3 * 8)(%rsp), %r13;
	CFI_RESTORE(%r10);
	CFI_RESTORE(%r11);
	CFI_RESTORE(%r12);
	CFI_RESTORE(%r13);

	call __sm4_crypt_blk16;

	addq $(4 * 8), %rsp;
	CFI_ADJUST_CFA_OFFSET(-4 * 8);

	vpxor RA0, RB0, RA0;
	vpxor RA1, RB1, RA1;
	vpxor RA2, RB2, RA2;
	vpxor RA3, RB3, RA3;

	vpxor RA1, RA0, RA0;
	vpxor RA3, RA2, RA2;

	vpxor RA2, RA0, RTMP1;

	vextracti128 $1, RTMP1, RNOTx;
	vpxor (%rcx), RTMP1x, RTMP1x;
	vpxor RNOTx, RTMP1x, RTMP1x;
	vmovdqu RTMP1x, (%rcx);

	vzeroall;

	ret_spec_stop;
	CFI_ENDPROC();
ELF(.size _gcry_sm4_aesni_avx2_ocb_auth,.-_gcry_sm4_aesni_avx2_ocb_auth;)

#endif /*defined(ENABLE_AESNI_SUPPORT) && defined(ENABLE_AVX_SUPPORT)*/
#endif /*__x86_64*/