lavu/aes: add x86 AESNI optimizations

crypto_bench comparison for AES-128-ECB: lavu_aesni AES-128-ECB size: 1048576 runs: 1024 time: 0.596 +- 0.081 lavu_c AES-128-ECB size: 1048576 runs: 1024 time: 17.007 +- 2.131 crypto AES-128-ECB size: 1048576 runs: 1024 time: 0.612 +- 1.857 gcrypt AES-128-ECB size: 1048576 runs: 1024 time: 1.123 +- 0.224 tomcrypt AES-128-ECB size: 1048576 runs: 1024 time: 9.038 +- 0.790

Rodger Combs committed 11 years ago

Commit 2ea3c511 parent 2daaafa5 files · +139 −2Browse files

Unified · Split

5 files

libavutil/aes.c +2 −0
libavutil/aes_internal.h +2 −0
libavutil/x86/Makefile +4 −2
libavutil/x86/aes.asm +95 −0
libavutil/x86/aes_init.c +36 −0

libavutil/aes.c+2−0

@@ -208,6 +208,8 @@ int av_aes_init(AVAES a, const uint8_t key, int key_bits, int decrypt) · 207 unmodified lines
208	208	uint8_t alog8[512];
209	209
210	210	a->crypt = decrypt ? aes_decrypt : aes_encrypt;
	211	+ if (ARCH_X86)
	212	+ ff_init_aes_x86(a, decrypt);
211	213
212	214	if (!enc_multbl[FF_ARRAY_ELEMS(enc_multbl) - 1][FF_ARRAY_ELEMS(enc_multbl[0]) - 1]) {
213	215	j = 1;

libavutil/aes_internal.h+2−0

@@ -40,4 +40,6 @@ typedef struct AVAES { · 39 unmodified lines
40	40	void (crypt)(struct AVAES a, uint8_t dst, const uint8_t src, int count, uint8_t *iv, int rounds);
41	41	} AVAES;
42	42
	43	+void ff_init_aes_x86(AVAES *a, int decrypt);
	44	+
43	45	#endif /* AVUTIL_AES_INTERNAL_H */

libavutil/x86/Makefile+4−2

@@ -1,4 +1,5 @@
1		−OBJS += x86/cpu.o \
	1	+OBJS += x86/aes_init.o \
	2	+ x86/cpu.o \
2	3	x86/fixed_dsp_init.o \
3	4	x86/float_dsp_init.o \
4	5	x86/imgutils_init.o \
@@ -10,7 +11,8 @@ OBJS-$(CONFIG_PIXELUTILS) += x86/pixelutils_init.o \ · 5 unmodified lines
10	11
11	12	EMMS_OBJS_$(HAVE_MMX_INLINE)_$(HAVE_MMX_EXTERNAL)_$(HAVE_MM_EMPTY) = x86/emms.o
12	13
13		−X86ASM-OBJS += x86/cpuid.o \
	14	+X86ASM-OBJS += x86/aes.o \
	15	+ x86/cpuid.o \
14	16	$(EMMS_OBJS__yes_) \
15	17	x86/fixed_dsp.o \
16	18	x86/float_dsp.o \

libavutil/x86/aes.asm+95−0

@@ -0,0 +1,95 @@
	1	+;*****************************************************************************
	2	+;* Copyright (c) 2015 Rodger Combs <rodger.combs@gmail.com>
	3	+;*
	4	+;* This file is part of FFmpeg.
	5	+;*
	6	+;* FFmpeg is free software; you can redistribute it and/or
	7	+;* modify it under the terms of the GNU Lesser General Public
	8	+;* License as published by the Free Software Foundation; either
	9	+;* version 2.1 of the License, or (at your option) any later version.
	10	+;*
	11	+;* FFmpeg is distributed in the hope that it will be useful,
	12	+;* but WITHOUT ANY WARRANTY; without even the implied warranty of
	13	+;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	14	+;* Lesser General Public License for more details.
	15	+;*
	16	+;* You should have received a copy of the GNU Lesser General Public
	17	+;* License along with FFmpeg; if not, write to the Free Software
	18	+;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
	19	+;******************************************************************************
	20	+
	21	+%include "x86util.asm"
	22	+
	23	+SECTION .text
	24	+
	25	+;-----------------------------------------------------------------------------
	26	+; void ff_aes_decrypt(AVAES a, uint8_t dst, const uint8_t *src,
	27	+; int count, uint8_t *iv, int rounds)
	28	+;-----------------------------------------------------------------------------
	29	+%macro AES_CRYPT 1
	30	+cglobal aes_%1rypt, 6,6,2
	31	+ test r3, r3
	32	+ je .ret
	33	+ shl r3d, 4
	34	+ add r5d, r5d
	35	+ add r0, 0x60
	36	+ add r2, r3
	37	+ add r1, r3
	38	+ neg r3
	39	+ pxor m1, m1
	40	+ test r4, r4
	41	+ je .block
	42	+ movu m1, [r4] ; iv
	43	+.block:
	44	+ movu m0, [r2+r3] ; state
	45	+%ifidn %1, enc
	46	+ pxor m0, m1
	47	+%endif
	48	+ pxor m0, [r0+8*r5-0x60]
	49	+ cmp r5d, 24
	50	+ je .rounds12
	51	+ jl .rounds10
	52	+ aes%1 m0, [r0+0x70]
	53	+ aes%1 m0, [r0+0x60]
	54	+.rounds12:
	55	+ aes%1 m0, [r0+0x50]
	56	+ aes%1 m0, [r0+0x40]
	57	+.rounds10:
	58	+ aes%1 m0, [r0+0x30]
	59	+ aes%1 m0, [r0+0x20]
	60	+ aes%1 m0, [r0+0x10]
	61	+ aes%1 m0, [r0+0x00]
	62	+ aes%1 m0, [r0-0x10]
	63	+ aes%1 m0, [r0-0x20]
	64	+ aes%1 m0, [r0-0x30]
	65	+ aes%1 m0, [r0-0x40]
	66	+ aes%1 m0, [r0-0x50]
	67	+ aes%1last m0, [r0-0x60]
	68	+ test r4, r4
	69	+ je .noiv
	70	+%ifidn %1, enc
	71	+ mova m1, m0
	72	+%else
	73	+ pxor m0, m1
	74	+ movu m1, [r2+r3]
	75	+%endif
	76	+.noiv:
	77	+ movu [r1+r3], m0
	78	+ add r3, 16
	79	+ jl .block
	80	+ test r4, r4
	81	+ je .ret
	82	+%ifidn %1, dec
	83	+ movu [r4], m1
	84	+%else
	85	+ movu [r4], m0
	86	+%endif
	87	+.ret:
	88	+ REP_RET
	89	+%endmacro
	90	+
	91	+%if HAVE_AESNI_EXTERNAL
	92	+INIT_XMM aesni
	93	+AES_CRYPT enc
	94	+AES_CRYPT dec
	95	+%endif

libavutil/x86/aes_init.c+36−0

@@ -0,0 +1,36 @@
	1	+/*
	2	+ * Copyright (c) 2015 Rodger Combs <rodger.combs@gmail.com>
	3	+ *
	4	+ * This file is part of FFmpeg.
	5	+ *
	6	+ * FFmpeg is free software; you can redistribute it and/or
	7	+ * modify it under the terms of the GNU Lesser General Public
	8	+ * License as published by the Free Software Foundation; either
	9	+ * version 2.1 of the License, or (at your option) any later version.
	10	+ *
	11	+ * FFmpeg is distributed in the hope that it will be useful,
	12	+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
	13	+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	14	+ * Lesser General Public License for more details.
	15	+ *
	16	+ * You should have received a copy of the GNU Lesser General Public
	17	+ * License along with FFmpeg; if not, write to the Free Software
	18	+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
	19	+ */
	20	+
	21	+#include <stddef.h>
	22	+#include "libavutil/aes_internal.h"
	23	+#include "libavutil/x86/cpu.h"
	24	+
	25	+void ff_aes_decrypt_aesni(AVAES a, uint8_t dst, const uint8_t *src,
	26	+ int count, uint8_t *iv, int rounds);
	27	+void ff_aes_encrypt_aesni(AVAES a, uint8_t dst, const uint8_t *src,
	28	+ int count, uint8_t *iv, int rounds);
	29	+
	30	+void ff_init_aes_x86(AVAES *a, int decrypt)
	31	+{
	32	+ int cpu_flags = av_get_cpu_flags();
	33	+
	34	+ if (EXTERNAL_AESNI(cpu_flags))
	35	+ a->crypt = decrypt ? ff_aes_decrypt_aesni : ff_aes_encrypt_aesni;
	36	+}

lavu/aes: add x86 AESNI optimizations

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code