• file
• revisions
• annotate
• diff
• comparison
• raw

media/libyuv/source/rotate.cc@b8a032363ba2 (annotated)

media/libyuv/source/rotate.cc

Thu, 22 Jan 2015 13:21:57 +0100

author

Michael Schloh von Bennewitz <michael@schloh.com>

date

Thu, 22 Jan 2015 13:21:57 +0100

branch

TOR_BUG_9701

changeset 15

b8a032363ba2

permissions

-rw-r--r--

Incorporate requested changes from Mozilla in review:
https://bugzilla.mozilla.org/show_bug.cgi?id=1123480#c6

michael@0	1	/*
michael@0	2	* Copyright 2011 The LibYuv Project Authors. All rights reserved.
michael@0	3	*
michael@0	4	* Use of this source code is governed by a BSD-style license
michael@0	5	* that can be found in the LICENSE file in the root of the source
michael@0	6	* tree. An additional intellectual property rights grant can be found
michael@0	7	* in the file PATENTS. All contributing project authors may
michael@0	8	* be found in the AUTHORS file in the root of the source tree.
michael@0	9	*/
michael@0	10
michael@0	11	#include "libyuv/rotate.h"
michael@0	12
michael@0	13	#include "libyuv/cpu_id.h"
michael@0	14	#include "libyuv/convert.h"
michael@0	15	#include "libyuv/planar_functions.h"
michael@0	16	#include "libyuv/row.h"
michael@0	17
michael@0	18	#ifdef __cplusplus
michael@0	19	namespace libyuv {
michael@0	20	extern "C" {
michael@0	21	#endif
michael@0	22
michael@0	23	#if !defined(LIBYUV_DISABLE_X86) && \
michael@0	24	(defined(_M_IX86) || defined(__x86_64__) || defined(__i386__))
michael@0	25	#if defined(__APPLE__) && defined(__i386__)
michael@0	26	#define DECLARE_FUNCTION(name) \
michael@0	27	".text \n" \
michael@0	28	".private_extern _" #name " \n" \
michael@0	29	".align 4,0x90 \n" \
michael@0	30	"_" #name ": \n"
michael@0	31	#elif defined(__MINGW32__) || defined(__CYGWIN__) && defined(__i386__)
michael@0	32	#define DECLARE_FUNCTION(name) \
michael@0	33	".text \n" \
michael@0	34	".align 4,0x90 \n" \
michael@0	35	"_" #name ": \n"
michael@0	36	#else
michael@0	37	#define DECLARE_FUNCTION(name) \
michael@0	38	".text \n" \
michael@0	39	".align 4,0x90 \n" \
michael@0	40	#name ": \n"
michael@0	41	#endif
michael@0	42	#endif
michael@0	43
michael@0	44	#if !defined(LIBYUV_DISABLE_NEON) && !defined(__native_client__) && \
michael@0	45	(defined(__ARM_NEON__) || defined(LIBYUV_NEON))
michael@0	46	#define HAS_MIRRORROW_NEON
michael@0	47	void MirrorRow_NEON(const uint8* src, uint8* dst, int width);
michael@0	48	#define HAS_MIRRORROW_UV_NEON
michael@0	49	void MirrorUVRow_NEON(const uint8* src, uint8* dst_a, uint8* dst_b, int width);
michael@0	50	#define HAS_TRANSPOSE_WX8_NEON
michael@0	51	void TransposeWx8_NEON(const uint8* src, int src_stride,
michael@0	52	uint8* dst, int dst_stride, int width);
michael@0	53	#define HAS_TRANSPOSE_UVWX8_NEON
michael@0	54	void TransposeUVWx8_NEON(const uint8* src, int src_stride,
michael@0	55	uint8* dst_a, int dst_stride_a,
michael@0	56	uint8* dst_b, int dst_stride_b,
michael@0	57	int width);
michael@0	58	#endif // defined(__ARM_NEON__)
michael@0	59
michael@0	60	#if !defined(LIBYUV_DISABLE_MIPS) && !defined(__native_client__) && \
michael@0	61	defined(__mips__) && \
michael@0	62	defined(__mips_dsp) && (__mips_dsp_rev >= 2)
michael@0	63	#define HAS_TRANSPOSE_WX8_MIPS_DSPR2
michael@0	64	void TransposeWx8_MIPS_DSPR2(const uint8* src, int src_stride,
michael@0	65	uint8* dst, int dst_stride, int width);
michael@0	66
michael@0	67	void TransposeWx8_FAST_MIPS_DSPR2(const uint8* src, int src_stride,
michael@0	68	uint8* dst, int dst_stride, int width);
michael@0	69	#define HAS_TRANSPOSE_UVWx8_MIPS_DSPR2
michael@0	70	void TransposeUVWx8_MIPS_DSPR2(const uint8* src, int src_stride,
michael@0	71	uint8* dst_a, int dst_stride_a,
michael@0	72	uint8* dst_b, int dst_stride_b,
michael@0	73	int width);
michael@0	74	#endif // defined(__mips__)
michael@0	75
michael@0	76	#if !defined(LIBYUV_DISABLE_X86) && \
michael@0	77	defined(_M_IX86) && defined(_MSC_VER)
michael@0	78	#define HAS_TRANSPOSE_WX8_SSSE3
michael@0	79	__declspec(naked) __declspec(align(16))
michael@0	80	static void TransposeWx8_SSSE3(const uint8* src, int src_stride,
michael@0	81	uint8* dst, int dst_stride, int width) {
michael@0	82	__asm {
michael@0	83	push edi
michael@0	84	push esi
michael@0	85	push ebp
michael@0	86	mov eax, [esp + 12 + 4] // src
michael@0	87	mov edi, [esp + 12 + 8] // src_stride
michael@0	88	mov edx, [esp + 12 + 12] // dst
michael@0	89	mov esi, [esp + 12 + 16] // dst_stride
michael@0	90	mov ecx, [esp + 12 + 20] // width
michael@0	91
michael@0	92	// Read in the data from the source pointer.
michael@0	93	// First round of bit swap.
michael@0	94	align 4
michael@0	95	convertloop:
michael@0	96	movq xmm0, qword ptr [eax]
michael@0	97	lea ebp, [eax + 8]
michael@0	98	movq xmm1, qword ptr [eax + edi]
michael@0	99	lea eax, [eax + 2 * edi]
michael@0	100	punpcklbw xmm0, xmm1
michael@0	101	movq xmm2, qword ptr [eax]
michael@0	102	movdqa xmm1, xmm0
michael@0	103	palignr xmm1, xmm1, 8
michael@0	104	movq xmm3, qword ptr [eax + edi]
michael@0	105	lea eax, [eax + 2 * edi]
michael@0	106	punpcklbw xmm2, xmm3
michael@0	107	movdqa xmm3, xmm2
michael@0	108	movq xmm4, qword ptr [eax]
michael@0	109	palignr xmm3, xmm3, 8
michael@0	110	movq xmm5, qword ptr [eax + edi]
michael@0	111	punpcklbw xmm4, xmm5
michael@0	112	lea eax, [eax + 2 * edi]
michael@0	113	movdqa xmm5, xmm4
michael@0	114	movq xmm6, qword ptr [eax]
michael@0	115	palignr xmm5, xmm5, 8
michael@0	116	movq xmm7, qword ptr [eax + edi]
michael@0	117	punpcklbw xmm6, xmm7
michael@0	118	mov eax, ebp
michael@0	119	movdqa xmm7, xmm6
michael@0	120	palignr xmm7, xmm7, 8
michael@0	121	// Second round of bit swap.
michael@0	122	punpcklwd xmm0, xmm2
michael@0	123	punpcklwd xmm1, xmm3
michael@0	124	movdqa xmm2, xmm0
michael@0	125	movdqa xmm3, xmm1
michael@0	126	palignr xmm2, xmm2, 8
michael@0	127	palignr xmm3, xmm3, 8
michael@0	128	punpcklwd xmm4, xmm6
michael@0	129	punpcklwd xmm5, xmm7
michael@0	130	movdqa xmm6, xmm4
michael@0	131	movdqa xmm7, xmm5
michael@0	132	palignr xmm6, xmm6, 8
michael@0	133	palignr xmm7, xmm7, 8
michael@0	134	// Third round of bit swap.
michael@0	135	// Write to the destination pointer.
michael@0	136	punpckldq xmm0, xmm4
michael@0	137	movq qword ptr [edx], xmm0
michael@0	138	movdqa xmm4, xmm0
michael@0	139	palignr xmm4, xmm4, 8
michael@0	140	movq qword ptr [edx + esi], xmm4
michael@0	141	lea edx, [edx + 2 * esi]
michael@0	142	punpckldq xmm2, xmm6
michael@0	143	movdqa xmm6, xmm2
michael@0	144	palignr xmm6, xmm6, 8
michael@0	145	movq qword ptr [edx], xmm2
michael@0	146	punpckldq xmm1, xmm5
michael@0	147	movq qword ptr [edx + esi], xmm6
michael@0	148	lea edx, [edx + 2 * esi]
michael@0	149	movdqa xmm5, xmm1
michael@0	150	movq qword ptr [edx], xmm1
michael@0	151	palignr xmm5, xmm5, 8
michael@0	152	punpckldq xmm3, xmm7
michael@0	153	movq qword ptr [edx + esi], xmm5
michael@0	154	lea edx, [edx + 2 * esi]
michael@0	155	movq qword ptr [edx], xmm3
michael@0	156	movdqa xmm7, xmm3
michael@0	157	palignr xmm7, xmm7, 8
michael@0	158	sub ecx, 8
michael@0	159	movq qword ptr [edx + esi], xmm7
michael@0	160	lea edx, [edx + 2 * esi]
michael@0	161	jg convertloop
michael@0	162
michael@0	163	pop ebp
michael@0	164	pop esi
michael@0	165	pop edi
michael@0	166	ret
michael@0	167	}
michael@0	168	}
michael@0	169
michael@0	170	#define HAS_TRANSPOSE_UVWX8_SSE2
michael@0	171	__declspec(naked) __declspec(align(16))
michael@0	172	static void TransposeUVWx8_SSE2(const uint8* src, int src_stride,
michael@0	173	uint8* dst_a, int dst_stride_a,
michael@0	174	uint8* dst_b, int dst_stride_b,
michael@0	175	int w) {
michael@0	176	__asm {
michael@0	177	push ebx
michael@0	178	push esi
michael@0	179	push edi
michael@0	180	push ebp
michael@0	181	mov eax, [esp + 16 + 4] // src
michael@0	182	mov edi, [esp + 16 + 8] // src_stride
michael@0	183	mov edx, [esp + 16 + 12] // dst_a
michael@0	184	mov esi, [esp + 16 + 16] // dst_stride_a
michael@0	185	mov ebx, [esp + 16 + 20] // dst_b
michael@0	186	mov ebp, [esp + 16 + 24] // dst_stride_b
michael@0	187	mov ecx, esp
michael@0	188	sub esp, 4 + 16
michael@0	189	and esp, ~15
michael@0	190	mov [esp + 16], ecx
michael@0	191	mov ecx, [ecx + 16 + 28] // w
michael@0	192
michael@0	193	align 4
michael@0	194	convertloop:
michael@0	195	// Read in the data from the source pointer.
michael@0	196	// First round of bit swap.
michael@0	197	movdqa xmm0, [eax]
michael@0	198	movdqa xmm1, [eax + edi]
michael@0	199	lea eax, [eax + 2 * edi]
michael@0	200	movdqa xmm7, xmm0 // use xmm7 as temp register.
michael@0	201	punpcklbw xmm0, xmm1
michael@0	202	punpckhbw xmm7, xmm1
michael@0	203	movdqa xmm1, xmm7
michael@0	204	movdqa xmm2, [eax]
michael@0	205	movdqa xmm3, [eax + edi]
michael@0	206	lea eax, [eax + 2 * edi]
michael@0	207	movdqa xmm7, xmm2
michael@0	208	punpcklbw xmm2, xmm3
michael@0	209	punpckhbw xmm7, xmm3
michael@0	210	movdqa xmm3, xmm7
michael@0	211	movdqa xmm4, [eax]
michael@0	212	movdqa xmm5, [eax + edi]
michael@0	213	lea eax, [eax + 2 * edi]
michael@0	214	movdqa xmm7, xmm4
michael@0	215	punpcklbw xmm4, xmm5
michael@0	216	punpckhbw xmm7, xmm5
michael@0	217	movdqa xmm5, xmm7
michael@0	218	movdqa xmm6, [eax]
michael@0	219	movdqa xmm7, [eax + edi]
michael@0	220	lea eax, [eax + 2 * edi]
michael@0	221	movdqa [esp], xmm5 // backup xmm5
michael@0	222	neg edi
michael@0	223	movdqa xmm5, xmm6 // use xmm5 as temp register.
michael@0	224	punpcklbw xmm6, xmm7
michael@0	225	punpckhbw xmm5, xmm7
michael@0	226	movdqa xmm7, xmm5
michael@0	227	lea eax, [eax + 8 * edi + 16]
michael@0	228	neg edi
michael@0	229	// Second round of bit swap.
michael@0	230	movdqa xmm5, xmm0
michael@0	231	punpcklwd xmm0, xmm2
michael@0	232	punpckhwd xmm5, xmm2
michael@0	233	movdqa xmm2, xmm5
michael@0	234	movdqa xmm5, xmm1
michael@0	235	punpcklwd xmm1, xmm3
michael@0	236	punpckhwd xmm5, xmm3
michael@0	237	movdqa xmm3, xmm5
michael@0	238	movdqa xmm5, xmm4
michael@0	239	punpcklwd xmm4, xmm6
michael@0	240	punpckhwd xmm5, xmm6
michael@0	241	movdqa xmm6, xmm5
michael@0	242	movdqa xmm5, [esp] // restore xmm5
michael@0	243	movdqa [esp], xmm6 // backup xmm6
michael@0	244	movdqa xmm6, xmm5 // use xmm6 as temp register.
michael@0	245	punpcklwd xmm5, xmm7
michael@0	246	punpckhwd xmm6, xmm7
michael@0	247	movdqa xmm7, xmm6
michael@0	248	// Third round of bit swap.
michael@0	249	// Write to the destination pointer.
michael@0	250	movdqa xmm6, xmm0
michael@0	251	punpckldq xmm0, xmm4
michael@0	252	punpckhdq xmm6, xmm4
michael@0	253	movdqa xmm4, xmm6
michael@0	254	movdqa xmm6, [esp] // restore xmm6
michael@0	255	movlpd qword ptr [edx], xmm0
michael@0	256	movhpd qword ptr [ebx], xmm0
michael@0	257	movlpd qword ptr [edx + esi], xmm4
michael@0	258	lea edx, [edx + 2 * esi]
michael@0	259	movhpd qword ptr [ebx + ebp], xmm4
michael@0	260	lea ebx, [ebx + 2 * ebp]
michael@0	261	movdqa xmm0, xmm2 // use xmm0 as the temp register.
michael@0	262	punpckldq xmm2, xmm6
michael@0	263	movlpd qword ptr [edx], xmm2
michael@0	264	movhpd qword ptr [ebx], xmm2
michael@0	265	punpckhdq xmm0, xmm6
michael@0	266	movlpd qword ptr [edx + esi], xmm0
michael@0	267	lea edx, [edx + 2 * esi]
michael@0	268	movhpd qword ptr [ebx + ebp], xmm0
michael@0	269	lea ebx, [ebx + 2 * ebp]
michael@0	270	movdqa xmm0, xmm1 // use xmm0 as the temp register.
michael@0	271	punpckldq xmm1, xmm5
michael@0	272	movlpd qword ptr [edx], xmm1
michael@0	273	movhpd qword ptr [ebx], xmm1
michael@0	274	punpckhdq xmm0, xmm5
michael@0	275	movlpd qword ptr [edx + esi], xmm0
michael@0	276	lea edx, [edx + 2 * esi]
michael@0	277	movhpd qword ptr [ebx + ebp], xmm0
michael@0	278	lea ebx, [ebx + 2 * ebp]
michael@0	279	movdqa xmm0, xmm3 // use xmm0 as the temp register.
michael@0	280	punpckldq xmm3, xmm7
michael@0	281	movlpd qword ptr [edx], xmm3
michael@0	282	movhpd qword ptr [ebx], xmm3
michael@0	283	punpckhdq xmm0, xmm7
michael@0	284	sub ecx, 8
michael@0	285	movlpd qword ptr [edx + esi], xmm0
michael@0	286	lea edx, [edx + 2 * esi]
michael@0	287	movhpd qword ptr [ebx + ebp], xmm0
michael@0	288	lea ebx, [ebx + 2 * ebp]
michael@0	289	jg convertloop
michael@0	290
michael@0	291	mov esp, [esp + 16]
michael@0	292	pop ebp
michael@0	293	pop edi
michael@0	294	pop esi
michael@0	295	pop ebx
michael@0	296	ret
michael@0	297	}
michael@0	298	}
michael@0	299	#elif !defined(LIBYUV_DISABLE_X86) && \
michael@0	300	(defined(__i386__) || (defined(__x86_64__) && !defined(__native_client__)))
michael@0	301	#define HAS_TRANSPOSE_WX8_SSSE3
michael@0	302	static void TransposeWx8_SSSE3(const uint8* src, int src_stride,
michael@0	303	uint8* dst, int dst_stride, int width) {
michael@0	304	asm volatile (
michael@0	305	// Read in the data from the source pointer.
michael@0	306	// First round of bit swap.
michael@0	307	".p2align 2 \n"
michael@0	308	"1: \n"
michael@0	309	"movq (%0),%%xmm0 \n"
michael@0	310	"movq (%0,%3),%%xmm1 \n"
michael@0	311	"lea (%0,%3,2),%0 \n"
michael@0	312	"punpcklbw %%xmm1,%%xmm0 \n"
michael@0	313	"movq (%0),%%xmm2 \n"
michael@0	314	"movdqa %%xmm0,%%xmm1 \n"
michael@0	315	"palignr $0x8,%%xmm1,%%xmm1 \n"
michael@0	316	"movq (%0,%3),%%xmm3 \n"
michael@0	317	"lea (%0,%3,2),%0 \n"
michael@0	318	"punpcklbw %%xmm3,%%xmm2 \n"
michael@0	319	"movdqa %%xmm2,%%xmm3 \n"
michael@0	320	"movq (%0),%%xmm4 \n"
michael@0	321	"palignr $0x8,%%xmm3,%%xmm3 \n"
michael@0	322	"movq (%0,%3),%%xmm5 \n"
michael@0	323	"lea (%0,%3,2),%0 \n"
michael@0	324	"punpcklbw %%xmm5,%%xmm4 \n"
michael@0	325	"movdqa %%xmm4,%%xmm5 \n"
michael@0	326	"movq (%0),%%xmm6 \n"
michael@0	327	"palignr $0x8,%%xmm5,%%xmm5 \n"
michael@0	328	"movq (%0,%3),%%xmm7 \n"
michael@0	329	"lea (%0,%3,2),%0 \n"
michael@0	330	"punpcklbw %%xmm7,%%xmm6 \n"
michael@0	331	"neg %3 \n"
michael@0	332	"movdqa %%xmm6,%%xmm7 \n"
michael@0	333	"lea 0x8(%0,%3,8),%0 \n"
michael@0	334	"palignr $0x8,%%xmm7,%%xmm7 \n"
michael@0	335	"neg %3 \n"
michael@0	336	// Second round of bit swap.
michael@0	337	"punpcklwd %%xmm2,%%xmm0 \n"
michael@0	338	"punpcklwd %%xmm3,%%xmm1 \n"
michael@0	339	"movdqa %%xmm0,%%xmm2 \n"
michael@0	340	"movdqa %%xmm1,%%xmm3 \n"
michael@0	341	"palignr $0x8,%%xmm2,%%xmm2 \n"
michael@0	342	"palignr $0x8,%%xmm3,%%xmm3 \n"
michael@0	343	"punpcklwd %%xmm6,%%xmm4 \n"
michael@0	344	"punpcklwd %%xmm7,%%xmm5 \n"
michael@0	345	"movdqa %%xmm4,%%xmm6 \n"
michael@0	346	"movdqa %%xmm5,%%xmm7 \n"
michael@0	347	"palignr $0x8,%%xmm6,%%xmm6 \n"
michael@0	348	"palignr $0x8,%%xmm7,%%xmm7 \n"
michael@0	349	// Third round of bit swap.
michael@0	350	// Write to the destination pointer.
michael@0	351	"punpckldq %%xmm4,%%xmm0 \n"
michael@0	352	"movq %%xmm0,(%1) \n"
michael@0	353	"movdqa %%xmm0,%%xmm4 \n"
michael@0	354	"palignr $0x8,%%xmm4,%%xmm4 \n"
michael@0	355	"movq %%xmm4,(%1,%4) \n"
michael@0	356	"lea (%1,%4,2),%1 \n"
michael@0	357	"punpckldq %%xmm6,%%xmm2 \n"
michael@0	358	"movdqa %%xmm2,%%xmm6 \n"
michael@0	359	"movq %%xmm2,(%1) \n"
michael@0	360	"palignr $0x8,%%xmm6,%%xmm6 \n"
michael@0	361	"punpckldq %%xmm5,%%xmm1 \n"
michael@0	362	"movq %%xmm6,(%1,%4) \n"
michael@0	363	"lea (%1,%4,2),%1 \n"
michael@0	364	"movdqa %%xmm1,%%xmm5 \n"
michael@0	365	"movq %%xmm1,(%1) \n"
michael@0	366	"palignr $0x8,%%xmm5,%%xmm5 \n"
michael@0	367	"movq %%xmm5,(%1,%4) \n"
michael@0	368	"lea (%1,%4,2),%1 \n"
michael@0	369	"punpckldq %%xmm7,%%xmm3 \n"
michael@0	370	"movq %%xmm3,(%1) \n"
michael@0	371	"movdqa %%xmm3,%%xmm7 \n"
michael@0	372	"palignr $0x8,%%xmm7,%%xmm7 \n"
michael@0	373	"sub $0x8,%2 \n"
michael@0	374	"movq %%xmm7,(%1,%4) \n"
michael@0	375	"lea (%1,%4,2),%1 \n"
michael@0	376	"jg 1b \n"
michael@0	377	: "+r"(src), // %0
michael@0	378	"+r"(dst), // %1
michael@0	379	"+r"(width) // %2
michael@0	380	: "r"((intptr_t)(src_stride)), // %3
michael@0	381	"r"((intptr_t)(dst_stride)) // %4
michael@0	382	: "memory", "cc"
michael@0	383	#if defined(__SSE2__)
michael@0	384	, "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
michael@0	385	#endif
michael@0	386	);
michael@0	387	}
michael@0	388
michael@0	389	#if !defined(LIBYUV_DISABLE_X86) && defined(__i386__)
michael@0	390	#define HAS_TRANSPOSE_UVWX8_SSE2
michael@0	391	extern "C" void TransposeUVWx8_SSE2(const uint8* src, int src_stride,
michael@0	392	uint8* dst_a, int dst_stride_a,
michael@0	393	uint8* dst_b, int dst_stride_b,
michael@0	394	int w);
michael@0	395	asm (
michael@0	396	DECLARE_FUNCTION(TransposeUVWx8_SSE2)
michael@0	397	"push %ebx \n"
michael@0	398	"push %esi \n"
michael@0	399	"push %edi \n"
michael@0	400	"push %ebp \n"
michael@0	401	"mov 0x14(%esp),%eax \n"
michael@0	402	"mov 0x18(%esp),%edi \n"
michael@0	403	"mov 0x1c(%esp),%edx \n"
michael@0	404	"mov 0x20(%esp),%esi \n"
michael@0	405	"mov 0x24(%esp),%ebx \n"
michael@0	406	"mov 0x28(%esp),%ebp \n"
michael@0	407	"mov %esp,%ecx \n"
michael@0	408	"sub $0x14,%esp \n"
michael@0	409	"and $0xfffffff0,%esp \n"
michael@0	410	"mov %ecx,0x10(%esp) \n"
michael@0	411	"mov 0x2c(%ecx),%ecx \n"
michael@0	412
michael@0	413	"1: \n"
michael@0	414	"movdqa (%eax),%xmm0 \n"
michael@0	415	"movdqa (%eax,%edi,1),%xmm1 \n"
michael@0	416	"lea (%eax,%edi,2),%eax \n"
michael@0	417	"movdqa %xmm0,%xmm7 \n"
michael@0	418	"punpcklbw %xmm1,%xmm0 \n"
michael@0	419	"punpckhbw %xmm1,%xmm7 \n"
michael@0	420	"movdqa %xmm7,%xmm1 \n"
michael@0	421	"movdqa (%eax),%xmm2 \n"
michael@0	422	"movdqa (%eax,%edi,1),%xmm3 \n"
michael@0	423	"lea (%eax,%edi,2),%eax \n"
michael@0	424	"movdqa %xmm2,%xmm7 \n"
michael@0	425	"punpcklbw %xmm3,%xmm2 \n"
michael@0	426	"punpckhbw %xmm3,%xmm7 \n"
michael@0	427	"movdqa %xmm7,%xmm3 \n"
michael@0	428	"movdqa (%eax),%xmm4 \n"
michael@0	429	"movdqa (%eax,%edi,1),%xmm5 \n"
michael@0	430	"lea (%eax,%edi,2),%eax \n"
michael@0	431	"movdqa %xmm4,%xmm7 \n"
michael@0	432	"punpcklbw %xmm5,%xmm4 \n"
michael@0	433	"punpckhbw %xmm5,%xmm7 \n"
michael@0	434	"movdqa %xmm7,%xmm5 \n"
michael@0	435	"movdqa (%eax),%xmm6 \n"
michael@0	436	"movdqa (%eax,%edi,1),%xmm7 \n"
michael@0	437	"lea (%eax,%edi,2),%eax \n"
michael@0	438	"movdqa %xmm5,(%esp) \n"
michael@0	439	"neg %edi \n"
michael@0	440	"movdqa %xmm6,%xmm5 \n"
michael@0	441	"punpcklbw %xmm7,%xmm6 \n"
michael@0	442	"punpckhbw %xmm7,%xmm5 \n"
michael@0	443	"movdqa %xmm5,%xmm7 \n"
michael@0	444	"lea 0x10(%eax,%edi,8),%eax \n"
michael@0	445	"neg %edi \n"
michael@0	446	"movdqa %xmm0,%xmm5 \n"
michael@0	447	"punpcklwd %xmm2,%xmm0 \n"
michael@0	448	"punpckhwd %xmm2,%xmm5 \n"
michael@0	449	"movdqa %xmm5,%xmm2 \n"
michael@0	450	"movdqa %xmm1,%xmm5 \n"
michael@0	451	"punpcklwd %xmm3,%xmm1 \n"
michael@0	452	"punpckhwd %xmm3,%xmm5 \n"
michael@0	453	"movdqa %xmm5,%xmm3 \n"
michael@0	454	"movdqa %xmm4,%xmm5 \n"
michael@0	455	"punpcklwd %xmm6,%xmm4 \n"
michael@0	456	"punpckhwd %xmm6,%xmm5 \n"
michael@0	457	"movdqa %xmm5,%xmm6 \n"
michael@0	458	"movdqa (%esp),%xmm5 \n"
michael@0	459	"movdqa %xmm6,(%esp) \n"
michael@0	460	"movdqa %xmm5,%xmm6 \n"
michael@0	461	"punpcklwd %xmm7,%xmm5 \n"
michael@0	462	"punpckhwd %xmm7,%xmm6 \n"
michael@0	463	"movdqa %xmm6,%xmm7 \n"
michael@0	464	"movdqa %xmm0,%xmm6 \n"
michael@0	465	"punpckldq %xmm4,%xmm0 \n"
michael@0	466	"punpckhdq %xmm4,%xmm6 \n"
michael@0	467	"movdqa %xmm6,%xmm4 \n"
michael@0	468	"movdqa (%esp),%xmm6 \n"
michael@0	469	"movlpd %xmm0,(%edx) \n"
michael@0	470	"movhpd %xmm0,(%ebx) \n"
michael@0	471	"movlpd %xmm4,(%edx,%esi,1) \n"
michael@0	472	"lea (%edx,%esi,2),%edx \n"
michael@0	473	"movhpd %xmm4,(%ebx,%ebp,1) \n"
michael@0	474	"lea (%ebx,%ebp,2),%ebx \n"
michael@0	475	"movdqa %xmm2,%xmm0 \n"
michael@0	476	"punpckldq %xmm6,%xmm2 \n"
michael@0	477	"movlpd %xmm2,(%edx) \n"
michael@0	478	"movhpd %xmm2,(%ebx) \n"
michael@0	479	"punpckhdq %xmm6,%xmm0 \n"
michael@0	480	"movlpd %xmm0,(%edx,%esi,1) \n"
michael@0	481	"lea (%edx,%esi,2),%edx \n"
michael@0	482	"movhpd %xmm0,(%ebx,%ebp,1) \n"
michael@0	483	"lea (%ebx,%ebp,2),%ebx \n"
michael@0	484	"movdqa %xmm1,%xmm0 \n"
michael@0	485	"punpckldq %xmm5,%xmm1 \n"
michael@0	486	"movlpd %xmm1,(%edx) \n"
michael@0	487	"movhpd %xmm1,(%ebx) \n"
michael@0	488	"punpckhdq %xmm5,%xmm0 \n"
michael@0	489	"movlpd %xmm0,(%edx,%esi,1) \n"
michael@0	490	"lea (%edx,%esi,2),%edx \n"
michael@0	491	"movhpd %xmm0,(%ebx,%ebp,1) \n"
michael@0	492	"lea (%ebx,%ebp,2),%ebx \n"
michael@0	493	"movdqa %xmm3,%xmm0 \n"
michael@0	494	"punpckldq %xmm7,%xmm3 \n"
michael@0	495	"movlpd %xmm3,(%edx) \n"
michael@0	496	"movhpd %xmm3,(%ebx) \n"
michael@0	497	"punpckhdq %xmm7,%xmm0 \n"
michael@0	498	"sub $0x8,%ecx \n"
michael@0	499	"movlpd %xmm0,(%edx,%esi,1) \n"
michael@0	500	"lea (%edx,%esi,2),%edx \n"
michael@0	501	"movhpd %xmm0,(%ebx,%ebp,1) \n"
michael@0	502	"lea (%ebx,%ebp,2),%ebx \n"
michael@0	503	"jg 1b \n"
michael@0	504	"mov 0x10(%esp),%esp \n"
michael@0	505	"pop %ebp \n"
michael@0	506	"pop %edi \n"
michael@0	507	"pop %esi \n"
michael@0	508	"pop %ebx \n"
michael@0	509	#if defined(__native_client__)
michael@0	510	"pop %ecx \n"
michael@0	511	"and $0xffffffe0,%ecx \n"
michael@0	512	"jmp *%ecx \n"
michael@0	513	#else
michael@0	514	"ret \n"
michael@0	515	#endif
michael@0	516	);
michael@0	517	#elif !defined(LIBYUV_DISABLE_X86) && !defined(__native_client__) && \
michael@0	518	defined(__x86_64__)
michael@0	519	// 64 bit version has enough registers to do 16x8 to 8x16 at a time.
michael@0	520	#define HAS_TRANSPOSE_WX8_FAST_SSSE3
michael@0	521	static void TransposeWx8_FAST_SSSE3(const uint8* src, int src_stride,
michael@0	522	uint8* dst, int dst_stride, int width) {
michael@0	523	asm volatile (
michael@0	524	// Read in the data from the source pointer.
michael@0	525	// First round of bit swap.
michael@0	526	".p2align 2 \n"
michael@0	527	"1: \n"
michael@0	528	"movdqa (%0),%%xmm0 \n"
michael@0	529	"movdqa (%0,%3),%%xmm1 \n"
michael@0	530	"lea (%0,%3,2),%0 \n"
michael@0	531	"movdqa %%xmm0,%%xmm8 \n"
michael@0	532	"punpcklbw %%xmm1,%%xmm0 \n"
michael@0	533	"punpckhbw %%xmm1,%%xmm8 \n"
michael@0	534	"movdqa (%0),%%xmm2 \n"
michael@0	535	"movdqa %%xmm0,%%xmm1 \n"
michael@0	536	"movdqa %%xmm8,%%xmm9 \n"
michael@0	537	"palignr $0x8,%%xmm1,%%xmm1 \n"
michael@0	538	"palignr $0x8,%%xmm9,%%xmm9 \n"
michael@0	539	"movdqa (%0,%3),%%xmm3 \n"
michael@0	540	"lea (%0,%3,2),%0 \n"
michael@0	541	"movdqa %%xmm2,%%xmm10 \n"
michael@0	542	"punpcklbw %%xmm3,%%xmm2 \n"
michael@0	543	"punpckhbw %%xmm3,%%xmm10 \n"
michael@0	544	"movdqa %%xmm2,%%xmm3 \n"
michael@0	545	"movdqa %%xmm10,%%xmm11 \n"
michael@0	546	"movdqa (%0),%%xmm4 \n"
michael@0	547	"palignr $0x8,%%xmm3,%%xmm3 \n"
michael@0	548	"palignr $0x8,%%xmm11,%%xmm11 \n"
michael@0	549	"movdqa (%0,%3),%%xmm5 \n"
michael@0	550	"lea (%0,%3,2),%0 \n"
michael@0	551	"movdqa %%xmm4,%%xmm12 \n"
michael@0	552	"punpcklbw %%xmm5,%%xmm4 \n"
michael@0	553	"punpckhbw %%xmm5,%%xmm12 \n"
michael@0	554	"movdqa %%xmm4,%%xmm5 \n"
michael@0	555	"movdqa %%xmm12,%%xmm13 \n"
michael@0	556	"movdqa (%0),%%xmm6 \n"
michael@0	557	"palignr $0x8,%%xmm5,%%xmm5 \n"
michael@0	558	"palignr $0x8,%%xmm13,%%xmm13 \n"
michael@0	559	"movdqa (%0,%3),%%xmm7 \n"
michael@0	560	"lea (%0,%3,2),%0 \n"
michael@0	561	"movdqa %%xmm6,%%xmm14 \n"
michael@0	562	"punpcklbw %%xmm7,%%xmm6 \n"
michael@0	563	"punpckhbw %%xmm7,%%xmm14 \n"
michael@0	564	"neg %3 \n"
michael@0	565	"movdqa %%xmm6,%%xmm7 \n"
michael@0	566	"movdqa %%xmm14,%%xmm15 \n"
michael@0	567	"lea 0x10(%0,%3,8),%0 \n"
michael@0	568	"palignr $0x8,%%xmm7,%%xmm7 \n"
michael@0	569	"palignr $0x8,%%xmm15,%%xmm15 \n"
michael@0	570	"neg %3 \n"
michael@0	571	// Second round of bit swap.
michael@0	572	"punpcklwd %%xmm2,%%xmm0 \n"
michael@0	573	"punpcklwd %%xmm3,%%xmm1 \n"
michael@0	574	"movdqa %%xmm0,%%xmm2 \n"
michael@0	575	"movdqa %%xmm1,%%xmm3 \n"
michael@0	576	"palignr $0x8,%%xmm2,%%xmm2 \n"
michael@0	577	"palignr $0x8,%%xmm3,%%xmm3 \n"
michael@0	578	"punpcklwd %%xmm6,%%xmm4 \n"
michael@0	579	"punpcklwd %%xmm7,%%xmm5 \n"
michael@0	580	"movdqa %%xmm4,%%xmm6 \n"
michael@0	581	"movdqa %%xmm5,%%xmm7 \n"
michael@0	582	"palignr $0x8,%%xmm6,%%xmm6 \n"
michael@0	583	"palignr $0x8,%%xmm7,%%xmm7 \n"
michael@0	584	"punpcklwd %%xmm10,%%xmm8 \n"
michael@0	585	"punpcklwd %%xmm11,%%xmm9 \n"
michael@0	586	"movdqa %%xmm8,%%xmm10 \n"
michael@0	587	"movdqa %%xmm9,%%xmm11 \n"
michael@0	588	"palignr $0x8,%%xmm10,%%xmm10 \n"
michael@0	589	"palignr $0x8,%%xmm11,%%xmm11 \n"
michael@0	590	"punpcklwd %%xmm14,%%xmm12 \n"
michael@0	591	"punpcklwd %%xmm15,%%xmm13 \n"
michael@0	592	"movdqa %%xmm12,%%xmm14 \n"
michael@0	593	"movdqa %%xmm13,%%xmm15 \n"
michael@0	594	"palignr $0x8,%%xmm14,%%xmm14 \n"
michael@0	595	"palignr $0x8,%%xmm15,%%xmm15 \n"
michael@0	596	// Third round of bit swap.
michael@0	597	// Write to the destination pointer.
michael@0	598	"punpckldq %%xmm4,%%xmm0 \n"
michael@0	599	"movq %%xmm0,(%1) \n"
michael@0	600	"movdqa %%xmm0,%%xmm4 \n"
michael@0	601	"palignr $0x8,%%xmm4,%%xmm4 \n"
michael@0	602	"movq %%xmm4,(%1,%4) \n"
michael@0	603	"lea (%1,%4,2),%1 \n"
michael@0	604	"punpckldq %%xmm6,%%xmm2 \n"
michael@0	605	"movdqa %%xmm2,%%xmm6 \n"
michael@0	606	"movq %%xmm2,(%1) \n"
michael@0	607	"palignr $0x8,%%xmm6,%%xmm6 \n"
michael@0	608	"punpckldq %%xmm5,%%xmm1 \n"
michael@0	609	"movq %%xmm6,(%1,%4) \n"
michael@0	610	"lea (%1,%4,2),%1 \n"
michael@0	611	"movdqa %%xmm1,%%xmm5 \n"
michael@0	612	"movq %%xmm1,(%1) \n"
michael@0	613	"palignr $0x8,%%xmm5,%%xmm5 \n"
michael@0	614	"movq %%xmm5,(%1,%4) \n"
michael@0	615	"lea (%1,%4,2),%1 \n"
michael@0	616	"punpckldq %%xmm7,%%xmm3 \n"
michael@0	617	"movq %%xmm3,(%1) \n"
michael@0	618	"movdqa %%xmm3,%%xmm7 \n"
michael@0	619	"palignr $0x8,%%xmm7,%%xmm7 \n"
michael@0	620	"movq %%xmm7,(%1,%4) \n"
michael@0	621	"lea (%1,%4,2),%1 \n"
michael@0	622	"punpckldq %%xmm12,%%xmm8 \n"
michael@0	623	"movq %%xmm8,(%1) \n"
michael@0	624	"movdqa %%xmm8,%%xmm12 \n"
michael@0	625	"palignr $0x8,%%xmm12,%%xmm12 \n"
michael@0	626	"movq %%xmm12,(%1,%4) \n"
michael@0	627	"lea (%1,%4,2),%1 \n"
michael@0	628	"punpckldq %%xmm14,%%xmm10 \n"
michael@0	629	"movdqa %%xmm10,%%xmm14 \n"
michael@0	630	"movq %%xmm10,(%1) \n"
michael@0	631	"palignr $0x8,%%xmm14,%%xmm14 \n"
michael@0	632	"punpckldq %%xmm13,%%xmm9 \n"
michael@0	633	"movq %%xmm14,(%1,%4) \n"
michael@0	634	"lea (%1,%4,2),%1 \n"
michael@0	635	"movdqa %%xmm9,%%xmm13 \n"
michael@0	636	"movq %%xmm9,(%1) \n"
michael@0	637	"palignr $0x8,%%xmm13,%%xmm13 \n"
michael@0	638	"movq %%xmm13,(%1,%4) \n"
michael@0	639	"lea (%1,%4,2),%1 \n"
michael@0	640	"punpckldq %%xmm15,%%xmm11 \n"
michael@0	641	"movq %%xmm11,(%1) \n"
michael@0	642	"movdqa %%xmm11,%%xmm15 \n"
michael@0	643	"palignr $0x8,%%xmm15,%%xmm15 \n"
michael@0	644	"sub $0x10,%2 \n"
michael@0	645	"movq %%xmm15,(%1,%4) \n"
michael@0	646	"lea (%1,%4,2),%1 \n"
michael@0	647	"jg 1b \n"
michael@0	648	: "+r"(src), // %0
michael@0	649	"+r"(dst), // %1
michael@0	650	"+r"(width) // %2
michael@0	651	: "r"((intptr_t)(src_stride)), // %3
michael@0	652	"r"((intptr_t)(dst_stride)) // %4
michael@0	653	: "memory", "cc",
michael@0	654	"xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7",
michael@0	655	"xmm8", "xmm9", "xmm10", "xmm11", "xmm12", "xmm13", "xmm14", "xmm15"
michael@0	656	);
michael@0	657	}
michael@0	658
michael@0	659	#define HAS_TRANSPOSE_UVWX8_SSE2
michael@0	660	static void TransposeUVWx8_SSE2(const uint8* src, int src_stride,
michael@0	661	uint8* dst_a, int dst_stride_a,
michael@0	662	uint8* dst_b, int dst_stride_b,
michael@0	663	int w) {
michael@0	664	asm volatile (
michael@0	665	// Read in the data from the source pointer.
michael@0	666	// First round of bit swap.
michael@0	667	".p2align 2 \n"
michael@0	668	"1: \n"
michael@0	669	"movdqa (%0),%%xmm0 \n"
michael@0	670	"movdqa (%0,%4),%%xmm1 \n"
michael@0	671	"lea (%0,%4,2),%0 \n"
michael@0	672	"movdqa %%xmm0,%%xmm8 \n"
michael@0	673	"punpcklbw %%xmm1,%%xmm0 \n"
michael@0	674	"punpckhbw %%xmm1,%%xmm8 \n"
michael@0	675	"movdqa %%xmm8,%%xmm1 \n"
michael@0	676	"movdqa (%0),%%xmm2 \n"
michael@0	677	"movdqa (%0,%4),%%xmm3 \n"
michael@0	678	"lea (%0,%4,2),%0 \n"
michael@0	679	"movdqa %%xmm2,%%xmm8 \n"
michael@0	680	"punpcklbw %%xmm3,%%xmm2 \n"
michael@0	681	"punpckhbw %%xmm3,%%xmm8 \n"
michael@0	682	"movdqa %%xmm8,%%xmm3 \n"
michael@0	683	"movdqa (%0),%%xmm4 \n"
michael@0	684	"movdqa (%0,%4),%%xmm5 \n"
michael@0	685	"lea (%0,%4,2),%0 \n"
michael@0	686	"movdqa %%xmm4,%%xmm8 \n"
michael@0	687	"punpcklbw %%xmm5,%%xmm4 \n"
michael@0	688	"punpckhbw %%xmm5,%%xmm8 \n"
michael@0	689	"movdqa %%xmm8,%%xmm5 \n"
michael@0	690	"movdqa (%0),%%xmm6 \n"
michael@0	691	"movdqa (%0,%4),%%xmm7 \n"
michael@0	692	"lea (%0,%4,2),%0 \n"
michael@0	693	"movdqa %%xmm6,%%xmm8 \n"
michael@0	694	"punpcklbw %%xmm7,%%xmm6 \n"
michael@0	695	"neg %4 \n"
michael@0	696	"lea 0x10(%0,%4,8),%0 \n"
michael@0	697	"punpckhbw %%xmm7,%%xmm8 \n"
michael@0	698	"movdqa %%xmm8,%%xmm7 \n"
michael@0	699	"neg %4 \n"
michael@0	700	// Second round of bit swap.
michael@0	701	"movdqa %%xmm0,%%xmm8 \n"
michael@0	702	"movdqa %%xmm1,%%xmm9 \n"
michael@0	703	"punpckhwd %%xmm2,%%xmm8 \n"
michael@0	704	"punpckhwd %%xmm3,%%xmm9 \n"
michael@0	705	"punpcklwd %%xmm2,%%xmm0 \n"
michael@0	706	"punpcklwd %%xmm3,%%xmm1 \n"
michael@0	707	"movdqa %%xmm8,%%xmm2 \n"
michael@0	708	"movdqa %%xmm9,%%xmm3 \n"
michael@0	709	"movdqa %%xmm4,%%xmm8 \n"
michael@0	710	"movdqa %%xmm5,%%xmm9 \n"
michael@0	711	"punpckhwd %%xmm6,%%xmm8 \n"
michael@0	712	"punpckhwd %%xmm7,%%xmm9 \n"
michael@0	713	"punpcklwd %%xmm6,%%xmm4 \n"
michael@0	714	"punpcklwd %%xmm7,%%xmm5 \n"
michael@0	715	"movdqa %%xmm8,%%xmm6 \n"
michael@0	716	"movdqa %%xmm9,%%xmm7 \n"
michael@0	717	// Third round of bit swap.
michael@0	718	// Write to the destination pointer.
michael@0	719	"movdqa %%xmm0,%%xmm8 \n"
michael@0	720	"punpckldq %%xmm4,%%xmm0 \n"
michael@0	721	"movlpd %%xmm0,(%1) \n" // Write back U channel
michael@0	722	"movhpd %%xmm0,(%2) \n" // Write back V channel
michael@0	723	"punpckhdq %%xmm4,%%xmm8 \n"
michael@0	724	"movlpd %%xmm8,(%1,%5) \n"
michael@0	725	"lea (%1,%5,2),%1 \n"
michael@0	726	"movhpd %%xmm8,(%2,%6) \n"
michael@0	727	"lea (%2,%6,2),%2 \n"
michael@0	728	"movdqa %%xmm2,%%xmm8 \n"
michael@0	729	"punpckldq %%xmm6,%%xmm2 \n"
michael@0	730	"movlpd %%xmm2,(%1) \n"
michael@0	731	"movhpd %%xmm2,(%2) \n"
michael@0	732	"punpckhdq %%xmm6,%%xmm8 \n"
michael@0	733	"movlpd %%xmm8,(%1,%5) \n"
michael@0	734	"lea (%1,%5,2),%1 \n"
michael@0	735	"movhpd %%xmm8,(%2,%6) \n"
michael@0	736	"lea (%2,%6,2),%2 \n"
michael@0	737	"movdqa %%xmm1,%%xmm8 \n"
michael@0	738	"punpckldq %%xmm5,%%xmm1 \n"
michael@0	739	"movlpd %%xmm1,(%1) \n"
michael@0	740	"movhpd %%xmm1,(%2) \n"
michael@0	741	"punpckhdq %%xmm5,%%xmm8 \n"
michael@0	742	"movlpd %%xmm8,(%1,%5) \n"
michael@0	743	"lea (%1,%5,2),%1 \n"
michael@0	744	"movhpd %%xmm8,(%2,%6) \n"
michael@0	745	"lea (%2,%6,2),%2 \n"
michael@0	746	"movdqa %%xmm3,%%xmm8 \n"
michael@0	747	"punpckldq %%xmm7,%%xmm3 \n"
michael@0	748	"movlpd %%xmm3,(%1) \n"
michael@0	749	"movhpd %%xmm3,(%2) \n"
michael@0	750	"punpckhdq %%xmm7,%%xmm8 \n"
michael@0	751	"sub $0x8,%3 \n"
michael@0	752	"movlpd %%xmm8,(%1,%5) \n"
michael@0	753	"lea (%1,%5,2),%1 \n"
michael@0	754	"movhpd %%xmm8,(%2,%6) \n"
michael@0	755	"lea (%2,%6,2),%2 \n"
michael@0	756	"jg 1b \n"
michael@0	757	: "+r"(src), // %0
michael@0	758	"+r"(dst_a), // %1
michael@0	759	"+r"(dst_b), // %2
michael@0	760	"+r"(w) // %3
michael@0	761	: "r"((intptr_t)(src_stride)), // %4
michael@0	762	"r"((intptr_t)(dst_stride_a)), // %5
michael@0	763	"r"((intptr_t)(dst_stride_b)) // %6
michael@0	764	: "memory", "cc",
michael@0	765	"xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7",
michael@0	766	"xmm8", "xmm9"
michael@0	767	);
michael@0	768	}
michael@0	769	#endif
michael@0	770	#endif
michael@0	771
michael@0	772	static void TransposeWx8_C(const uint8* src, int src_stride,
michael@0	773	uint8* dst, int dst_stride,
michael@0	774	int width) {
michael@0	775	int i;
michael@0	776	for (i = 0; i < width; ++i) {
michael@0	777	dst[0] = src[0 * src_stride];
michael@0	778	dst[1] = src[1 * src_stride];
michael@0	779	dst[2] = src[2 * src_stride];
michael@0	780	dst[3] = src[3 * src_stride];
michael@0	781	dst[4] = src[4 * src_stride];
michael@0	782	dst[5] = src[5 * src_stride];
michael@0	783	dst[6] = src[6 * src_stride];
michael@0	784	dst[7] = src[7 * src_stride];
michael@0	785	++src;
michael@0	786	dst += dst_stride;
michael@0	787	}
michael@0	788	}
michael@0	789
michael@0	790	static void TransposeWxH_C(const uint8* src, int src_stride,
michael@0	791	uint8* dst, int dst_stride,
michael@0	792	int width, int height) {
michael@0	793	int i;
michael@0	794	for (i = 0; i < width; ++i) {
michael@0	795	int j;
michael@0	796	for (j = 0; j < height; ++j) {
michael@0	797	dst[i * dst_stride + j] = src[j * src_stride + i];
michael@0	798	}
michael@0	799	}
michael@0	800	}
michael@0	801
michael@0	802	LIBYUV_API
michael@0	803	void TransposePlane(const uint8* src, int src_stride,
michael@0	804	uint8* dst, int dst_stride,
michael@0	805	int width, int height) {
michael@0	806	int i = height;
michael@0	807	void (*TransposeWx8)(const uint8* src, int src_stride,
michael@0	808	uint8* dst, int dst_stride,
michael@0	809	int width) = TransposeWx8_C;
michael@0	810	#if defined(HAS_TRANSPOSE_WX8_NEON)
michael@0	811	if (TestCpuFlag(kCpuHasNEON)) {
michael@0	812	TransposeWx8 = TransposeWx8_NEON;
michael@0	813	}
michael@0	814	#endif
michael@0	815	#if defined(HAS_TRANSPOSE_WX8_SSSE3)
michael@0	816	if (TestCpuFlag(kCpuHasSSSE3) && IS_ALIGNED(width, 8)) {
michael@0	817	TransposeWx8 = TransposeWx8_SSSE3;
michael@0	818	}
michael@0	819	#endif
michael@0	820	#if defined(HAS_TRANSPOSE_WX8_FAST_SSSE3)
michael@0	821	if (TestCpuFlag(kCpuHasSSSE3) &&
michael@0	822	IS_ALIGNED(width, 16) &&
michael@0	823	IS_ALIGNED(src, 16) && IS_ALIGNED(src_stride, 16)) {
michael@0	824	TransposeWx8 = TransposeWx8_FAST_SSSE3;
michael@0	825	}
michael@0	826	#endif
michael@0	827	#if defined(HAS_TRANSPOSE_WX8_MIPS_DSPR2)
michael@0	828	if (TestCpuFlag(kCpuHasMIPS_DSPR2)) {
michael@0	829	if (IS_ALIGNED(width, 4) &&
michael@0	830	IS_ALIGNED(src, 4) && IS_ALIGNED(src_stride, 4)) {
michael@0	831	TransposeWx8 = TransposeWx8_FAST_MIPS_DSPR2;
michael@0	832	} else {
michael@0	833	TransposeWx8 = TransposeWx8_MIPS_DSPR2;
michael@0	834	}
michael@0	835	}
michael@0	836	#endif
michael@0	837
michael@0	838	// Work across the source in 8x8 tiles
michael@0	839	while (i >= 8) {
michael@0	840	TransposeWx8(src, src_stride, dst, dst_stride, width);
michael@0	841	src += 8 * src_stride; // Go down 8 rows.
michael@0	842	dst += 8; // Move over 8 columns.
michael@0	843	i -= 8;
michael@0	844	}
michael@0	845
michael@0	846	TransposeWxH_C(src, src_stride, dst, dst_stride, width, i);
michael@0	847	}
michael@0	848
michael@0	849	LIBYUV_API
michael@0	850	void RotatePlane90(const uint8* src, int src_stride,
michael@0	851	uint8* dst, int dst_stride,
michael@0	852	int width, int height) {
michael@0	853	// Rotate by 90 is a transpose with the source read
michael@0	854	// from bottom to top. So set the source pointer to the end
michael@0	855	// of the buffer and flip the sign of the source stride.
michael@0	856	src += src_stride * (height - 1);
michael@0	857	src_stride = -src_stride;
michael@0	858	TransposePlane(src, src_stride, dst, dst_stride, width, height);
michael@0	859	}
michael@0	860
michael@0	861	LIBYUV_API
michael@0	862	void RotatePlane270(const uint8* src, int src_stride,
michael@0	863	uint8* dst, int dst_stride,
michael@0	864	int width, int height) {
michael@0	865	// Rotate by 270 is a transpose with the destination written
michael@0	866	// from bottom to top. So set the destination pointer to the end
michael@0	867	// of the buffer and flip the sign of the destination stride.
michael@0	868	dst += dst_stride * (width - 1);
michael@0	869	dst_stride = -dst_stride;
michael@0	870	TransposePlane(src, src_stride, dst, dst_stride, width, height);
michael@0	871	}
michael@0	872
michael@0	873	LIBYUV_API
michael@0	874	void RotatePlane180(const uint8* src, int src_stride,
michael@0	875	uint8* dst, int dst_stride,
michael@0	876	int width, int height) {
michael@0	877	// Swap first and last row and mirror the content. Uses a temporary row.
michael@0	878	align_buffer_64(row, width);
michael@0	879	const uint8* src_bot = src + src_stride * (height - 1);
michael@0	880	uint8* dst_bot = dst + dst_stride * (height - 1);
michael@0	881	int half_height = (height + 1) >> 1;
michael@0	882	int y;
michael@0	883	void (*MirrorRow)(const uint8* src, uint8* dst, int width) = MirrorRow_C;
michael@0	884	void (*CopyRow)(const uint8* src, uint8* dst, int width) = CopyRow_C;
michael@0	885	#if defined(HAS_MIRRORROW_NEON)
michael@0	886	if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 16)) {
michael@0	887	MirrorRow = MirrorRow_NEON;
michael@0	888	}
michael@0	889	#endif
michael@0	890	#if defined(HAS_MIRRORROW_SSE2)
michael@0	891	if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(width, 16) &&
michael@0	892	IS_ALIGNED(src, 16) && IS_ALIGNED(src_stride, 16) &&
michael@0	893	IS_ALIGNED(dst, 16) && IS_ALIGNED(dst_stride, 16)) {
michael@0	894	MirrorRow = MirrorRow_SSE2;
michael@0	895	}
michael@0	896	#endif
michael@0	897	#if defined(HAS_MIRRORROW_SSSE3)
michael@0	898	if (TestCpuFlag(kCpuHasSSSE3) && IS_ALIGNED(width, 16) &&
michael@0	899	IS_ALIGNED(src, 16) && IS_ALIGNED(src_stride, 16) &&
michael@0	900	IS_ALIGNED(dst, 16) && IS_ALIGNED(dst_stride, 16)) {
michael@0	901	MirrorRow = MirrorRow_SSSE3;
michael@0	902	}
michael@0	903	#endif
michael@0	904	#if defined(HAS_MIRRORROW_AVX2)
michael@0	905	if (TestCpuFlag(kCpuHasAVX2) && IS_ALIGNED(width, 32)) {
michael@0	906	MirrorRow = MirrorRow_AVX2;
michael@0	907	}
michael@0	908	#endif
michael@0	909	#if defined(HAS_MIRRORROW_MIPS_DSPR2)
michael@0	910	if (TestCpuFlag(kCpuHasMIPS_DSPR2) &&
michael@0	911	IS_ALIGNED(src, 4) && IS_ALIGNED(src_stride, 4) &&
michael@0	912	IS_ALIGNED(dst, 4) && IS_ALIGNED(dst_stride, 4)) {
michael@0	913	MirrorRow = MirrorRow_MIPS_DSPR2;
michael@0	914	}
michael@0	915	#endif
michael@0	916	#if defined(HAS_COPYROW_NEON)
michael@0	917	if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 32)) {
michael@0	918	CopyRow = CopyRow_NEON;
michael@0	919	}
michael@0	920	#endif
michael@0	921	#if defined(HAS_COPYROW_X86)
michael@0	922	if (TestCpuFlag(kCpuHasX86) && IS_ALIGNED(width, 4)) {
michael@0	923	CopyRow = CopyRow_X86;
michael@0	924	}
michael@0	925	#endif
michael@0	926	#if defined(HAS_COPYROW_SSE2)
michael@0	927	if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(width, 32) &&
michael@0	928	IS_ALIGNED(src, 16) && IS_ALIGNED(src_stride, 16) &&
michael@0	929	IS_ALIGNED(dst, 16) && IS_ALIGNED(dst_stride, 16)) {
michael@0	930	CopyRow = CopyRow_SSE2;
michael@0	931	}
michael@0	932	#endif
michael@0	933	#if defined(HAS_COPYROW_ERMS)
michael@0	934	if (TestCpuFlag(kCpuHasERMS)) {
michael@0	935	CopyRow = CopyRow_ERMS;
michael@0	936	}
michael@0	937	#endif
michael@0	938	#if defined(HAS_COPYROW_MIPS)
michael@0	939	if (TestCpuFlag(kCpuHasMIPS)) {
michael@0	940	CopyRow = CopyRow_MIPS;
michael@0	941	}
michael@0	942	#endif
michael@0	943
michael@0	944	// Odd height will harmlessly mirror the middle row twice.
michael@0	945	for (y = 0; y < half_height; ++y) {
michael@0	946	MirrorRow(src, row, width); // Mirror first row into a buffer
michael@0	947	src += src_stride;
michael@0	948	MirrorRow(src_bot, dst, width); // Mirror last row into first row
michael@0	949	dst += dst_stride;
michael@0	950	CopyRow(row, dst_bot, width); // Copy first mirrored row into last
michael@0	951	src_bot -= src_stride;
michael@0	952	dst_bot -= dst_stride;
michael@0	953	}
michael@0	954	free_aligned_buffer_64(row);
michael@0	955	}
michael@0	956
michael@0	957	static void TransposeUVWx8_C(const uint8* src, int src_stride,
michael@0	958	uint8* dst_a, int dst_stride_a,
michael@0	959	uint8* dst_b, int dst_stride_b,
michael@0	960	int width) {
michael@0	961	int i;
michael@0	962	for (i = 0; i < width; ++i) {
michael@0	963	dst_a[0] = src[0 * src_stride + 0];
michael@0	964	dst_b[0] = src[0 * src_stride + 1];
michael@0	965	dst_a[1] = src[1 * src_stride + 0];
michael@0	966	dst_b[1] = src[1 * src_stride + 1];
michael@0	967	dst_a[2] = src[2 * src_stride + 0];
michael@0	968	dst_b[2] = src[2 * src_stride + 1];
michael@0	969	dst_a[3] = src[3 * src_stride + 0];
michael@0	970	dst_b[3] = src[3 * src_stride + 1];
michael@0	971	dst_a[4] = src[4 * src_stride + 0];
michael@0	972	dst_b[4] = src[4 * src_stride + 1];
michael@0	973	dst_a[5] = src[5 * src_stride + 0];
michael@0	974	dst_b[5] = src[5 * src_stride + 1];
michael@0	975	dst_a[6] = src[6 * src_stride + 0];
michael@0	976	dst_b[6] = src[6 * src_stride + 1];
michael@0	977	dst_a[7] = src[7 * src_stride + 0];
michael@0	978	dst_b[7] = src[7 * src_stride + 1];
michael@0	979	src += 2;
michael@0	980	dst_a += dst_stride_a;
michael@0	981	dst_b += dst_stride_b;
michael@0	982	}
michael@0	983	}
michael@0	984
michael@0	985	static void TransposeUVWxH_C(const uint8* src, int src_stride,
michael@0	986	uint8* dst_a, int dst_stride_a,
michael@0	987	uint8* dst_b, int dst_stride_b,
michael@0	988	int width, int height) {
michael@0	989	int i;
michael@0	990	for (i = 0; i < width * 2; i += 2) {
michael@0	991	int j;
michael@0	992	for (j = 0; j < height; ++j) {
michael@0	993	dst_a[j + ((i >> 1) * dst_stride_a)] = src[i + (j * src_stride)];
michael@0	994	dst_b[j + ((i >> 1) * dst_stride_b)] = src[i + (j * src_stride) + 1];
michael@0	995	}
michael@0	996	}
michael@0	997	}
michael@0	998
michael@0	999	LIBYUV_API
michael@0	1000	void TransposeUV(const uint8* src, int src_stride,
michael@0	1001	uint8* dst_a, int dst_stride_a,
michael@0	1002	uint8* dst_b, int dst_stride_b,
michael@0	1003	int width, int height) {
michael@0	1004	int i = height;
michael@0	1005	void (*TransposeUVWx8)(const uint8* src, int src_stride,
michael@0	1006	uint8* dst_a, int dst_stride_a,
michael@0	1007	uint8* dst_b, int dst_stride_b,
michael@0	1008	int width) = TransposeUVWx8_C;
michael@0	1009	#if defined(HAS_TRANSPOSE_UVWX8_NEON)
michael@0	1010	if (TestCpuFlag(kCpuHasNEON)) {
michael@0	1011	TransposeUVWx8 = TransposeUVWx8_NEON;
michael@0	1012	}
michael@0	1013	#elif defined(HAS_TRANSPOSE_UVWX8_SSE2)
michael@0	1014	if (TestCpuFlag(kCpuHasSSE2) &&
michael@0	1015	IS_ALIGNED(width, 8) &&
michael@0	1016	IS_ALIGNED(src, 16) && IS_ALIGNED(src_stride, 16)) {
michael@0	1017	TransposeUVWx8 = TransposeUVWx8_SSE2;
michael@0	1018	}
michael@0	1019	#elif defined(HAS_TRANSPOSE_UVWx8_MIPS_DSPR2)
michael@0	1020	if (TestCpuFlag(kCpuHasMIPS_DSPR2) && IS_ALIGNED(width, 2) &&
michael@0	1021	IS_ALIGNED(src, 4) && IS_ALIGNED(src_stride, 4)) {
michael@0	1022	TransposeUVWx8 = TransposeUVWx8_MIPS_DSPR2;
michael@0	1023	}
michael@0	1024	#endif
michael@0	1025
michael@0	1026	// Work through the source in 8x8 tiles.
michael@0	1027	while (i >= 8) {
michael@0	1028	TransposeUVWx8(src, src_stride,
michael@0	1029	dst_a, dst_stride_a,
michael@0	1030	dst_b, dst_stride_b,
michael@0	1031	width);
michael@0	1032	src += 8 * src_stride; // Go down 8 rows.
michael@0	1033	dst_a += 8; // Move over 8 columns.
michael@0	1034	dst_b += 8; // Move over 8 columns.
michael@0	1035	i -= 8;
michael@0	1036	}
michael@0	1037
michael@0	1038	TransposeUVWxH_C(src, src_stride,
michael@0	1039	dst_a, dst_stride_a,
michael@0	1040	dst_b, dst_stride_b,
michael@0	1041	width, i);
michael@0	1042	}
michael@0	1043
michael@0	1044	LIBYUV_API
michael@0	1045	void RotateUV90(const uint8* src, int src_stride,
michael@0	1046	uint8* dst_a, int dst_stride_a,
michael@0	1047	uint8* dst_b, int dst_stride_b,
michael@0	1048	int width, int height) {
michael@0	1049	src += src_stride * (height - 1);
michael@0	1050	src_stride = -src_stride;
michael@0	1051
michael@0	1052	TransposeUV(src, src_stride,
michael@0	1053	dst_a, dst_stride_a,
michael@0	1054	dst_b, dst_stride_b,
michael@0	1055	width, height);
michael@0	1056	}
michael@0	1057
michael@0	1058	LIBYUV_API
michael@0	1059	void RotateUV270(const uint8* src, int src_stride,
michael@0	1060	uint8* dst_a, int dst_stride_a,
michael@0	1061	uint8* dst_b, int dst_stride_b,
michael@0	1062	int width, int height) {
michael@0	1063	dst_a += dst_stride_a * (width - 1);
michael@0	1064	dst_b += dst_stride_b * (width - 1);
michael@0	1065	dst_stride_a = -dst_stride_a;
michael@0	1066	dst_stride_b = -dst_stride_b;
michael@0	1067
michael@0	1068	TransposeUV(src, src_stride,
michael@0	1069	dst_a, dst_stride_a,
michael@0	1070	dst_b, dst_stride_b,
michael@0	1071	width, height);
michael@0	1072	}
michael@0	1073
michael@0	1074	// Rotate 180 is a horizontal and vertical flip.
michael@0	1075	LIBYUV_API
michael@0	1076	void RotateUV180(const uint8* src, int src_stride,
michael@0	1077	uint8* dst_a, int dst_stride_a,
michael@0	1078	uint8* dst_b, int dst_stride_b,
michael@0	1079	int width, int height) {
michael@0	1080	int i;
michael@0	1081	void (*MirrorRowUV)(const uint8* src, uint8* dst_u, uint8* dst_v, int width) =
michael@0	1082	MirrorUVRow_C;
michael@0	1083	#if defined(HAS_MIRRORUVROW_NEON)
michael@0	1084	if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 8)) {
michael@0	1085	MirrorRowUV = MirrorUVRow_NEON;
michael@0	1086	}
michael@0	1087	#elif defined(HAS_MIRRORROW_UV_SSSE3)
michael@0	1088	if (TestCpuFlag(kCpuHasSSSE3) && IS_ALIGNED(width, 16) &&
michael@0	1089	IS_ALIGNED(src, 16) && IS_ALIGNED(src_stride, 16)) {
michael@0	1090	MirrorRowUV = MirrorUVRow_SSSE3;
michael@0	1091	}
michael@0	1092	#elif defined(HAS_MIRRORUVROW_MIPS_DSPR2)
michael@0	1093	if (TestCpuFlag(kCpuHasMIPS_DSPR2) &&
michael@0	1094	IS_ALIGNED(src, 4) && IS_ALIGNED(src_stride, 4)) {
michael@0	1095	MirrorRowUV = MirrorUVRow_MIPS_DSPR2;
michael@0	1096	}
michael@0	1097	#endif
michael@0	1098
michael@0	1099	dst_a += dst_stride_a * (height - 1);
michael@0	1100	dst_b += dst_stride_b * (height - 1);
michael@0	1101
michael@0	1102	for (i = 0; i < height; ++i) {
michael@0	1103	MirrorRowUV(src, dst_a, dst_b, width);
michael@0	1104	src += src_stride;
michael@0	1105	dst_a -= dst_stride_a;
michael@0	1106	dst_b -= dst_stride_b;
michael@0	1107	}
michael@0	1108	}
michael@0	1109
michael@0	1110	LIBYUV_API
michael@0	1111	int RotatePlane(const uint8* src, int src_stride,
michael@0	1112	uint8* dst, int dst_stride,
michael@0	1113	int width, int height,
michael@0	1114	enum RotationMode mode) {
michael@0	1115	if (!src || width <= 0 || height == 0 || !dst) {
michael@0	1116	return -1;
michael@0	1117	}
michael@0	1118
michael@0	1119	// Negative height means invert the image.
michael@0	1120	if (height < 0) {
michael@0	1121	height = -height;
michael@0	1122	src = src + (height - 1) * src_stride;
michael@0	1123	src_stride = -src_stride;
michael@0	1124	}
michael@0	1125
michael@0	1126	switch (mode) {
michael@0	1127	case kRotate0:
michael@0	1128	// copy frame
michael@0	1129	CopyPlane(src, src_stride,
michael@0	1130	dst, dst_stride,
michael@0	1131	width, height);
michael@0	1132	return 0;
michael@0	1133	case kRotate90:
michael@0	1134	RotatePlane90(src, src_stride,
michael@0	1135	dst, dst_stride,
michael@0	1136	width, height);
michael@0	1137	return 0;
michael@0	1138	case kRotate270:
michael@0	1139	RotatePlane270(src, src_stride,
michael@0	1140	dst, dst_stride,
michael@0	1141	width, height);
michael@0	1142	return 0;
michael@0	1143	case kRotate180:
michael@0	1144	RotatePlane180(src, src_stride,
michael@0	1145	dst, dst_stride,
michael@0	1146	width, height);
michael@0	1147	return 0;
michael@0	1148	default:
michael@0	1149	break;
michael@0	1150	}
michael@0	1151	return -1;
michael@0	1152	}
michael@0	1153
michael@0	1154	LIBYUV_API
michael@0	1155	int I420Rotate(const uint8* src_y, int src_stride_y,
michael@0	1156	const uint8* src_u, int src_stride_u,
michael@0	1157	const uint8* src_v, int src_stride_v,
michael@0	1158	uint8* dst_y, int dst_stride_y,
michael@0	1159	uint8* dst_u, int dst_stride_u,
michael@0	1160	uint8* dst_v, int dst_stride_v,
michael@0	1161	int width, int height,
michael@0	1162	enum RotationMode mode) {
michael@0	1163	int halfwidth = (width + 1) >> 1;
michael@0	1164	int halfheight = (height + 1) >> 1;
michael@0	1165	if (!src_y || !src_u || !src_v || width <= 0 || height == 0 ||
michael@0	1166	!dst_y || !dst_u || !dst_v) {
michael@0	1167	return -1;
michael@0	1168	}
michael@0	1169
michael@0	1170	// Negative height means invert the image.
michael@0	1171	if (height < 0) {
michael@0	1172	height = -height;
michael@0	1173	halfheight = (height + 1) >> 1;
michael@0	1174	src_y = src_y + (height - 1) * src_stride_y;
michael@0	1175	src_u = src_u + (halfheight - 1) * src_stride_u;
michael@0	1176	src_v = src_v + (halfheight - 1) * src_stride_v;
michael@0	1177	src_stride_y = -src_stride_y;
michael@0	1178	src_stride_u = -src_stride_u;
michael@0	1179	src_stride_v = -src_stride_v;
michael@0	1180	}
michael@0	1181
michael@0	1182	switch (mode) {
michael@0	1183	case kRotate0:
michael@0	1184	// copy frame
michael@0	1185	return I420Copy(src_y, src_stride_y,
michael@0	1186	src_u, src_stride_u,
michael@0	1187	src_v, src_stride_v,
michael@0	1188	dst_y, dst_stride_y,
michael@0	1189	dst_u, dst_stride_u,
michael@0	1190	dst_v, dst_stride_v,
michael@0	1191	width, height);
michael@0	1192	case kRotate90:
michael@0	1193	RotatePlane90(src_y, src_stride_y,
michael@0	1194	dst_y, dst_stride_y,
michael@0	1195	width, height);
michael@0	1196	RotatePlane90(src_u, src_stride_u,
michael@0	1197	dst_u, dst_stride_u,
michael@0	1198	halfwidth, halfheight);
michael@0	1199	RotatePlane90(src_v, src_stride_v,
michael@0	1200	dst_v, dst_stride_v,
michael@0	1201	halfwidth, halfheight);
michael@0	1202	return 0;
michael@0	1203	case kRotate270:
michael@0	1204	RotatePlane270(src_y, src_stride_y,
michael@0	1205	dst_y, dst_stride_y,
michael@0	1206	width, height);
michael@0	1207	RotatePlane270(src_u, src_stride_u,
michael@0	1208	dst_u, dst_stride_u,
michael@0	1209	halfwidth, halfheight);
michael@0	1210	RotatePlane270(src_v, src_stride_v,
michael@0	1211	dst_v, dst_stride_v,
michael@0	1212	halfwidth, halfheight);
michael@0	1213	return 0;
michael@0	1214	case kRotate180:
michael@0	1215	RotatePlane180(src_y, src_stride_y,
michael@0	1216	dst_y, dst_stride_y,
michael@0	1217	width, height);
michael@0	1218	RotatePlane180(src_u, src_stride_u,
michael@0	1219	dst_u, dst_stride_u,
michael@0	1220	halfwidth, halfheight);
michael@0	1221	RotatePlane180(src_v, src_stride_v,
michael@0	1222	dst_v, dst_stride_v,
michael@0	1223	halfwidth, halfheight);
michael@0	1224	return 0;
michael@0	1225	default:
michael@0	1226	break;
michael@0	1227	}
michael@0	1228	return -1;
michael@0	1229	}
michael@0	1230
michael@0	1231	LIBYUV_API
michael@0	1232	int NV12ToI420Rotate(const uint8* src_y, int src_stride_y,
michael@0	1233	const uint8* src_uv, int src_stride_uv,
michael@0	1234	uint8* dst_y, int dst_stride_y,
michael@0	1235	uint8* dst_u, int dst_stride_u,
michael@0	1236	uint8* dst_v, int dst_stride_v,
michael@0	1237	int width, int height,
michael@0	1238	enum RotationMode mode) {
michael@0	1239	int halfwidth = (width + 1) >> 1;
michael@0	1240	int halfheight = (height + 1) >> 1;
michael@0	1241	if (!src_y || !src_uv || width <= 0 || height == 0 ||
michael@0	1242	!dst_y || !dst_u || !dst_v) {
michael@0	1243	return -1;
michael@0	1244	}
michael@0	1245
michael@0	1246	// Negative height means invert the image.
michael@0	1247	if (height < 0) {
michael@0	1248	height = -height;
michael@0	1249	halfheight = (height + 1) >> 1;
michael@0	1250	src_y = src_y + (height - 1) * src_stride_y;
michael@0	1251	src_uv = src_uv + (halfheight - 1) * src_stride_uv;
michael@0	1252	src_stride_y = -src_stride_y;
michael@0	1253	src_stride_uv = -src_stride_uv;
michael@0	1254	}
michael@0	1255
michael@0	1256	switch (mode) {
michael@0	1257	case kRotate0:
michael@0	1258	// copy frame
michael@0	1259	return NV12ToI420(src_y, src_stride_y,
michael@0	1260	src_uv, src_stride_uv,
michael@0	1261	dst_y, dst_stride_y,
michael@0	1262	dst_u, dst_stride_u,
michael@0	1263	dst_v, dst_stride_v,
michael@0	1264	width, height);
michael@0	1265	case kRotate90:
michael@0	1266	RotatePlane90(src_y, src_stride_y,
michael@0	1267	dst_y, dst_stride_y,
michael@0	1268	width, height);
michael@0	1269	RotateUV90(src_uv, src_stride_uv,
michael@0	1270	dst_u, dst_stride_u,
michael@0	1271	dst_v, dst_stride_v,
michael@0	1272	halfwidth, halfheight);
michael@0	1273	return 0;
michael@0	1274	case kRotate270:
michael@0	1275	RotatePlane270(src_y, src_stride_y,
michael@0	1276	dst_y, dst_stride_y,
michael@0	1277	width, height);
michael@0	1278	RotateUV270(src_uv, src_stride_uv,
michael@0	1279	dst_u, dst_stride_u,
michael@0	1280	dst_v, dst_stride_v,
michael@0	1281	halfwidth, halfheight);
michael@0	1282	return 0;
michael@0	1283	case kRotate180:
michael@0	1284	RotatePlane180(src_y, src_stride_y,
michael@0	1285	dst_y, dst_stride_y,
michael@0	1286	width, height);
michael@0	1287	RotateUV180(src_uv, src_stride_uv,
michael@0	1288	dst_u, dst_stride_u,
michael@0	1289	dst_v, dst_stride_v,
michael@0	1290	halfwidth, halfheight);
michael@0	1291	return 0;
michael@0	1292	default:
michael@0	1293	break;
michael@0	1294	}
michael@0	1295	return -1;
michael@0	1296	}
michael@0	1297
michael@0	1298	#ifdef __cplusplus
michael@0	1299	} // extern "C"
michael@0	1300	} // namespace libyuv
michael@0	1301	#endif