sigrid/ffmpeg

libavcodec/arm/Makefile+2−0

@@ -21,6 +21,7 @@ OBJS-$(CONFIG_IDCTDSP) += arm/idctdsp_init_arm.o \ · 20 unmodified lines
21	21	arm/simple_idct_arm.o
22	22	OBJS-$(CONFIG_MPEGAUDIODSP) += arm/mpegaudiodsp_init_arm.o
23	23	OBJS-$(CONFIG_MPEGVIDEO) += arm/mpegvideo_arm.o
	24	+OBJS-$(CONFIG_MPEGVIDEOENC) += arm/mpegvideoencdsp_init_arm.o
24	25	OBJS-$(CONFIG_NEON_CLOBBER_TEST) += arm/neontest.o
25	26	OBJS-$(CONFIG_VIDEODSP) += arm/videodsp_init_arm.o
26	27	OBJS-$(CONFIG_VP3DSP) += arm/vp3dsp_init_arm.o
@@ -60,6 +61,7 @@ ARMV6-OBJS-$(CONFIG_IDCTDSP) += arm/idctdsp_init_armv6.o \ · 33 unmodified lines
60	61	arm/idctdsp_armv6.o \
61	62	arm/simple_idct_armv6.o
62	63	ARMV6-OBJS-$(CONFIG_MPEGAUDIODSP) += arm/mpegaudiodsp_fixed_armv6.o
	64	+ARMV6-OBJS-$(CONFIG_MPEGVIDEOENC) += arm/mpegvideoencdsp_armv6.o
63	65
64	66	ARMV6-OBJS-$(CONFIG_MLP_DECODER) += arm/mlpdsp_armv6.o
65	67	ARMV6-OBJS-$(CONFIG_VP7_DECODER) += arm/vp8_armv6.o \

libavcodec/arm/dsputil_armv6.S+0−55

@@ -297,58 +297,3 @@ function ff_sse16_armv6, export=1 · 296 unmodified lines
297	297
298	298	pop {r4-r9, pc}
299	299	endfunc
300		−
301		−function ff_pix_norm1_armv6, export=1
302		− push {r4-r6, lr}
303		− mov r12, #16
304		− mov lr, #0
305		−1:
306		− ldm r0, {r2-r5}
307		− uxtb16 r6, r2
308		− uxtb16 r2, r2, ror #8
309		− smlad lr, r6, r6, lr
310		− uxtb16 r6, r3
311		− smlad lr, r2, r2, lr
312		− uxtb16 r3, r3, ror #8
313		− smlad lr, r6, r6, lr
314		− uxtb16 r6, r4
315		− smlad lr, r3, r3, lr
316		− uxtb16 r4, r4, ror #8
317		− smlad lr, r6, r6, lr
318		− uxtb16 r6, r5
319		− smlad lr, r4, r4, lr
320		− uxtb16 r5, r5, ror #8
321		− smlad lr, r6, r6, lr
322		− subs r12, r12, #1
323		− add r0, r0, r1
324		− smlad lr, r5, r5, lr
325		− bgt 1b
326		−
327		− mov r0, lr
328		− pop {r4-r6, pc}
329		−endfunc
330		−
331		−function ff_pix_sum_armv6, export=1
332		− push {r4-r7, lr}
333		− mov r12, #16
334		− mov r2, #0
335		− mov r3, #0
336		− mov lr, #0
337		− ldr r4, [r0]
338		−1:
339		− subs r12, r12, #1
340		− ldr r5, [r0, #4]
341		− usada8 r2, r4, lr, r2
342		− ldr r6, [r0, #8]
343		− usada8 r3, r5, lr, r3
344		− ldr r7, [r0, #12]
345		− usada8 r2, r6, lr, r2
346		− beq 2f
347		− ldr_pre r4, r0, r1
348		− usada8 r3, r7, lr, r3
349		− bgt 1b
350		−2:
351		− usada8 r3, r7, lr, r3
352		− add r0, r2, r3
353		− pop {r4-r7, pc}
354		−endfunc

libavcodec/arm/dsputil_init_armv6.c+0−6

@@ -43,9 +43,6 @@ int ff_pix_abs8_armv6(MpegEncContext s, uint8_t blk1, uint8_t *blk2, · 42 unmodified lines
43	43	int ff_sse16_armv6(MpegEncContext s, uint8_t blk1, uint8_t *blk2,
44	44	int line_size, int h);
45	45
46		−int ff_pix_norm1_armv6(uint8_t *pix, int line_size);
47		−int ff_pix_sum_armv6(uint8_t *pix, int line_size);
48		−
49	46	av_cold void ff_dsputil_init_armv6(DSPContext c, AVCodecContext avctx,
50	47	unsigned high_bit_depth)
51	48	{
@@ -63,7 +60,4 @@ av_cold void ff_dsputil_init_armv6(DSPContext c, AVCodecContext avctx, · 11 unmodified lines
63	60	c->sad[1] = ff_pix_abs8_armv6;
64	61
65	62	c->sse[0] = ff_sse16_armv6;
66		−
67		− c->pix_norm1 = ff_pix_norm1_armv6;
68		− c->pix_sum = ff_pix_sum_armv6;
69	63	}

libavcodec/arm/mpegvideoencdsp_armv6.S+76−0

@@ -0,0 +1,76 @@
	1	+/*
	2	+ * Copyright (c) 2009 Mans Rullgard <mans@mansr.com>
	3	+ *
	4	+ * This file is part of Libav.
	5	+ *
	6	+ * Libav 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	+ * Libav 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 Libav; if not, write to the Free Software
	18	+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
	19	+ */
	20	+
	21	+#include "libavutil/arm/asm.S"
	22	+
	23	+function ff_pix_norm1_armv6, export=1
	24	+ push {r4-r6, lr}
	25	+ mov r12, #16
	26	+ mov lr, #0
	27	+1:
	28	+ ldm r0, {r2-r5}
	29	+ uxtb16 r6, r2
	30	+ uxtb16 r2, r2, ror #8
	31	+ smlad lr, r6, r6, lr
	32	+ uxtb16 r6, r3
	33	+ smlad lr, r2, r2, lr
	34	+ uxtb16 r3, r3, ror #8
	35	+ smlad lr, r6, r6, lr
	36	+ uxtb16 r6, r4
	37	+ smlad lr, r3, r3, lr
	38	+ uxtb16 r4, r4, ror #8
	39	+ smlad lr, r6, r6, lr
	40	+ uxtb16 r6, r5
	41	+ smlad lr, r4, r4, lr
	42	+ uxtb16 r5, r5, ror #8
	43	+ smlad lr, r6, r6, lr
	44	+ subs r12, r12, #1
	45	+ add r0, r0, r1
	46	+ smlad lr, r5, r5, lr
	47	+ bgt 1b
	48	+
	49	+ mov r0, lr
	50	+ pop {r4-r6, pc}
	51	+endfunc
	52	+
	53	+function ff_pix_sum_armv6, export=1
	54	+ push {r4-r7, lr}
	55	+ mov r12, #16
	56	+ mov r2, #0
	57	+ mov r3, #0
	58	+ mov lr, #0
	59	+ ldr r4, [r0]
	60	+1:
	61	+ subs r12, r12, #1
	62	+ ldr r5, [r0, #4]
	63	+ usada8 r2, r4, lr, r2
	64	+ ldr r6, [r0, #8]
	65	+ usada8 r3, r5, lr, r3
	66	+ ldr r7, [r0, #12]
	67	+ usada8 r2, r6, lr, r2
	68	+ beq 2f
	69	+ ldr_pre r4, r0, r1
	70	+ usada8 r3, r7, lr, r3
	71	+ bgt 1b
	72	+2:
	73	+ usada8 r3, r7, lr, r3
	74	+ add r0, r2, r3
	75	+ pop {r4-r7, pc}
	76	+endfunc

libavcodec/arm/mpegvideoencdsp_init_arm.c+38−0

@@ -0,0 +1,38 @@
	1	+/*
	2	+ * This file is part of Libav.
	3	+ *
	4	+ * Libav is free software; you can redistribute it and/or
	5	+ * modify it under the terms of the GNU Lesser General Public
	6	+ * License as published by the Free Software Foundation; either
	7	+ * version 2.1 of the License, or (at your option) any later version.
	8	+ *
	9	+ * Libav is distributed in the hope that it will be useful,
	10	+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
	11	+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	12	+ * Lesser General Public License for more details.
	13	+ *
	14	+ * You should have received a copy of the GNU Lesser General Public
	15	+ * License along with Libav; if not, write to the Free Software
	16	+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
	17	+ */
	18	+
	19	+#include <stdint.h>
	20	+
	21	+#include "libavutil/cpu.h"
	22	+#include "libavutil/arm/cpu.h"
	23	+#include "libavcodec/avcodec.h"
	24	+#include "libavcodec/mpegvideoencdsp.h"
	25	+
	26	+int ff_pix_norm1_armv6(uint8_t *pix, int line_size);
	27	+int ff_pix_sum_armv6(uint8_t *pix, int line_size);
	28	+
	29	+av_cold void ff_mpegvideoencdsp_init_arm(MpegvideoEncDSPContext *c,
	30	+ AVCodecContext *avctx)
	31	+{
	32	+ int cpu_flags = av_get_cpu_flags();
	33	+
	34	+ if (have_armv6(cpu_flags)) {
	35	+ c->pix_norm1 = ff_pix_norm1_armv6;
	36	+ c->pix_sum = ff_pix_sum_armv6;
	37	+ }
	38	+}

libavcodec/dnxhdenc.c+3−2

@@ -309,6 +309,7 @@ static av_cold int dnxhd_encode_init(AVCodecContext *avctx) · 308 unmodified lines
309	309	ff_blockdsp_init(&ctx->bdsp, avctx);
310	310	ff_dsputil_init(&ctx->m.dsp, avctx);
311	311	ff_idctdsp_init(&ctx->m.idsp, avctx);
	312	+ ff_mpegvideoencdsp_init(&ctx->m.mpvencdsp, avctx);
312	313	ff_dct_common_init(&ctx->m);
313	314	if (!ctx->m.dct_quantize)
314	315	ctx->m.dct_quantize = ff_dct_quantize_c;
@@ -719,8 +720,8 @@ static int dnxhd_mb_var_thread(AVCodecContext avctx, void arg, · 404 unmodified lines
719	720	int varc;
720	721
721	722	if (!partial_last_row && mb_x * 16 <= avctx->width - 16) {
722		− sum = ctx->m.dsp.pix_sum(pix, ctx->m.linesize);
723		− varc = ctx->m.dsp.pix_norm1(pix, ctx->m.linesize);
	723	+ sum = ctx->m.mpvencdsp.pix_sum(pix, ctx->m.linesize);
	724	+ varc = ctx->m.mpvencdsp.pix_norm1(pix, ctx->m.linesize);
724	725	} else {
725	726	int bw = FFMIN(avctx->width - 16 * mb_x, 16);
726	727	int bh = FFMIN((avctx->height >> ctx->interlaced) - 16 * mb_y, 16);

libavcodec/dsputil.c+0−79

@@ -26,15 +26,12 @@ · 25 unmodified lines
26	26	*/
27	27
28	28	#include "libavutil/attributes.h"
29		−#include "libavutil/imgutils.h"
30	29	#include "avcodec.h"
31	30	#include "copy_block.h"
32	31	#include "dct.h"
33	32	#include "dsputil.h"
34	33	#include "simple_idct.h"
35	34	#include "faandct.h"
36		−#include "imgconvert.h"
37		−#include "mathops.h"
38	35	#include "mpegvideo.h"
39	36	#include "config.h"
40	37
@@ -47,74 +44,6 @@ uint32_t ff_square_tab[512] = { 0, }; · 6 unmodified lines
47	44	#define BIT_DEPTH 8
48	45	#include "dsputilenc_template.c"
49	46
50		−static int pix_sum_c(uint8_t *pix, int line_size)
51		−{
52		− int s = 0, i, j;
53		−
54		− for (i = 0; i < 16; i++) {
55		− for (j = 0; j < 16; j += 8) {
56		− s += pix[0];
57		− s += pix[1];
58		− s += pix[2];
59		− s += pix[3];
60		− s += pix[4];
61		− s += pix[5];
62		− s += pix[6];
63		− s += pix[7];
64		− pix += 8;
65		− }
66		− pix += line_size - 16;
67		− }
68		− return s;
69		−}
70		−
71		−static int pix_norm1_c(uint8_t *pix, int line_size)
72		−{
73		− int s = 0, i, j;
74		− uint32_t *sq = ff_square_tab + 256;
75		−
76		− for (i = 0; i < 16; i++) {
77		− for (j = 0; j < 16; j += 8) {
78		−#if 0
79		− s += sq[pix[0]];
80		− s += sq[pix[1]];
81		− s += sq[pix[2]];
82		− s += sq[pix[3]];
83		− s += sq[pix[4]];
84		− s += sq[pix[5]];
85		− s += sq[pix[6]];
86		− s += sq[pix[7]];
87		−#else
88		−#if HAVE_FAST_64BIT
89		− register uint64_t x = (uint64_t ) pix;
90		− s += sq[x & 0xff];
91		− s += sq[(x >> 8) & 0xff];
92		− s += sq[(x >> 16) & 0xff];
93		− s += sq[(x >> 24) & 0xff];
94		− s += sq[(x >> 32) & 0xff];
95		− s += sq[(x >> 40) & 0xff];
96		− s += sq[(x >> 48) & 0xff];
97		− s += sq[(x >> 56) & 0xff];
98		−#else
99		− register uint32_t x = (uint32_t ) pix;
100		− s += sq[x & 0xff];
101		− s += sq[(x >> 8) & 0xff];
102		− s += sq[(x >> 16) & 0xff];
103		− s += sq[(x >> 24) & 0xff];
104		− x = (uint32_t ) (pix + 4);
105		− s += sq[x & 0xff];
106		− s += sq[(x >> 8) & 0xff];
107		− s += sq[(x >> 16) & 0xff];
108		− s += sq[(x >> 24) & 0xff];
109		−#endif
110		−#endif
111		− pix += 8;
112		− }
113		− pix += line_size - 16;
114		− }
115		− return s;
116		−}
117		−
118	47	static int sse4_c(MpegEncContext v, uint8_t pix1, uint8_t *pix2,
119	48	int line_size, int h)
120	49	{
@@ -1055,9 +984,6 @@ av_cold void ff_dsputil_init(DSPContext c, AVCodecContext avctx) · 934 unmodified lines
1055	984
1056	985	c->sum_abs_dctelem = sum_abs_dctelem_c;
1057	986
1058		− c->pix_sum = pix_sum_c;
1059		− c->pix_norm1 = pix_norm1_c;
1060		−
1061	987	/* TODO [0] 16 [1] 8 */
1062	988	c->pix_abs[0][0] = pix_abs16_c;
1063	989	c->pix_abs[0][1] = pix_abs16_x2_c;
@@ -1097,11 +1023,6 @@ av_cold void ff_dsputil_init(DSPContext c, AVCodecContext avctx) · 33 unmodified lines
1097	1023	c->nsse[0] = nsse16_c;
1098	1024	c->nsse[1] = nsse8_c;
1099	1025
1100		− c->shrink[0] = av_image_copy_plane;
1101		− c->shrink[1] = ff_shrink22;
1102		− c->shrink[2] = ff_shrink44;
1103		− c->shrink[3] = ff_shrink88;
1104		−
1105	1026	c->draw_edges = draw_edges_8_c;
1106	1027
1107	1028	switch (avctx->bits_per_raw_sample) {

libavcodec/dsputil.h+0−6

@@ -58,9 +58,6 @@ typedef struct DSPContext { · 57 unmodified lines
58	58	int stride);
59	59	int (sum_abs_dctelem)(int16_t block /* align 16 */);
60	60
61		− int (pix_sum)(uint8_t pix, int line_size);
62		− int (pix_norm1)(uint8_t pix, int line_size);
63		−
64	61	me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
65	62	me_cmp_func sse[6];
66	63	me_cmp_func hadamard8_diff[6];
@@ -92,9 +89,6 @@ typedef struct DSPContext { · 25 unmodified lines
92	89	#define EDGE_WIDTH 16
93	90	#define EDGE_TOP 1
94	91	#define EDGE_BOTTOM 2
95		−
96		− void (shrink[4])(uint8_t dst, int dst_wrap, const uint8_t *src,
97		− int src_wrap, int width, int height);
98	92	} DSPContext;
99	93
100	94	void ff_dsputil_static_init(void);

libavcodec/motion_est.c+3−2

@@ -881,8 +881,9 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, · 880 unmodified lines
881	881
882	882	/* intra / predictive decision */
883	883	pix = c->src[0][0];
884		− sum = s->dsp.pix_sum(pix, s->linesize);
885		− varc = s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)sum*sum)>>8) + 500;
	884	+ sum = s->mpvencdsp.pix_sum(pix, s->linesize);
	885	+ varc = s->mpvencdsp.pix_norm1(pix, s->linesize) -
	886	+ (((unsigned) sum * sum) >> 8) + 500;
886	887
887	888	pic->mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
888	889	pic->mb_var [s->mb_stride * mb_y + mb_x] = (varc+128)>>8;

libavcodec/mpegvideo_enc.c+19−12

@@ -895,7 +895,7 @@ static int get_intra_count(MpegEncContext s, uint8_t src, · 894 unmodified lines
895	895	int offset = x + y * stride;
896	896	int sad = s->dsp.sad[0](NULL, src + offset, ref + offset, stride,
897	897	16);
898		− int mean = (s->dsp.pix_sum(src + offset, stride) + 128) >> 8;
	898	+ int mean = (s->mpvencdsp.pix_sum(src + offset, stride) + 128) >> 8;
899	899	int sae = get_sae(src + offset, mean, stride);
900	900
901	901	acc += sae + 500 < sad;
@@ -1138,15 +1138,21 @@ static int estimate_best_b_count(MpegEncContext *s) · 236 unmodified lines
1138	1138	pre_input.f->data[2] += INPLACE_OFFSET;
1139	1139	}
1140	1140
1141		− s->dsp.shrink[scale](s->tmp_frames[i]->data[0], s->tmp_frames[i]->linesize[0],
1142		− pre_input.f->data[0], pre_input.f->linesize[0],
1143		− c->width, c->height);
1144		− s->dsp.shrink[scale](s->tmp_frames[i]->data[1], s->tmp_frames[i]->linesize[1],
1145		− pre_input.f->data[1], pre_input.f->linesize[1],
1146		− c->width >> 1, c->height >> 1);
1147		− s->dsp.shrink[scale](s->tmp_frames[i]->data[2], s->tmp_frames[i]->linesize[2],
1148		− pre_input.f->data[2], pre_input.f->linesize[2],
1149		− c->width >> 1, c->height >> 1);
	1141	+ s->mpvencdsp.shrink[scale](s->tmp_frames[i]->data[0],
	1142	+ s->tmp_frames[i]->linesize[0],
	1143	+ pre_input.f->data[0],
	1144	+ pre_input.f->linesize[0],
	1145	+ c->width, c->height);
	1146	+ s->mpvencdsp.shrink[scale](s->tmp_frames[i]->data[1],
	1147	+ s->tmp_frames[i]->linesize[1],
	1148	+ pre_input.f->data[1],
	1149	+ pre_input.f->linesize[1],
	1150	+ c->width >> 1, c->height >> 1);
	1151	+ s->mpvencdsp.shrink[scale](s->tmp_frames[i]->data[2],
	1152	+ s->tmp_frames[i]->linesize[2],
	1153	+ pre_input.f->data[2],
	1154	+ pre_input.f->linesize[2],
	1155	+ c->width >> 1, c->height >> 1);
1150	1156	}
1151	1157	}
1152	1158
@@ -2420,9 +2426,10 @@ static int mb_var_thread(AVCodecContext c, void arg){ · 1267 unmodified lines
2420	2426	int yy = mb_y * 16;
2421	2427	uint8_t pix = s->new_picture.f->data[0] + (yy s->linesize) + xx;
2422	2428	int varc;
2423		− int sum = s->dsp.pix_sum(pix, s->linesize);
	2429	+ int sum = s->mpvencdsp.pix_sum(pix, s->linesize);
2424	2430
2425		− varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)sum*sum)>>8) + 500 + 128)>>8;
	2431	+ varc = (s->mpvencdsp.pix_norm1(pix, s->linesize) -
	2432	+ (((unsigned) sum * sum) >> 8) + 500 + 128) >> 8;
2426	2433
2427	2434	s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
2428	2435	s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;

libavcodec/mpegvideoencdsp.c+83−0

@@ -21,7 +21,10 @@ · 20 unmodified lines
21	21
22	22	#include "config.h"
23	23	#include "libavutil/attributes.h"
	24	+#include "libavutil/imgutils.h"
24	25	#include "avcodec.h"
	26	+#include "dsputil.h"
	27	+#include "imgconvert.h"
25	28	#include "mpegvideoencdsp.h"
26	29
27	30	static int try_8x8basis_c(int16_t rem[64], int16_t weight[64],
@@ -53,12 +56,92 @@ static void add_8x8basis_c(int16_t rem[64], int16_t basis[64], int scale) · 25 unmodified lines
53	56	(BASIS_SHIFT - RECON_SHIFT);
54	57	}
55	58
	59	+static int pix_sum_c(uint8_t *pix, int line_size)
	60	+{
	61	+ int s = 0, i, j;
	62	+
	63	+ for (i = 0; i < 16; i++) {
	64	+ for (j = 0; j < 16; j += 8) {
	65	+ s += pix[0];
	66	+ s += pix[1];
	67	+ s += pix[2];
	68	+ s += pix[3];
	69	+ s += pix[4];
	70	+ s += pix[5];
	71	+ s += pix[6];
	72	+ s += pix[7];
	73	+ pix += 8;
	74	+ }
	75	+ pix += line_size - 16;
	76	+ }
	77	+ return s;
	78	+}
	79	+
	80	+static int pix_norm1_c(uint8_t *pix, int line_size)
	81	+{
	82	+ int s = 0, i, j;
	83	+ uint32_t *sq = ff_square_tab + 256;
	84	+
	85	+ for (i = 0; i < 16; i++) {
	86	+ for (j = 0; j < 16; j += 8) {
	87	+#if 0
	88	+ s += sq[pix[0]];
	89	+ s += sq[pix[1]];
	90	+ s += sq[pix[2]];
	91	+ s += sq[pix[3]];
	92	+ s += sq[pix[4]];
	93	+ s += sq[pix[5]];
	94	+ s += sq[pix[6]];
	95	+ s += sq[pix[7]];
	96	+#else
	97	+#if HAVE_FAST_64BIT
	98	+ register uint64_t x = (uint64_t ) pix;
	99	+ s += sq[x & 0xff];
	100	+ s += sq[(x >> 8) & 0xff];
	101	+ s += sq[(x >> 16) & 0xff];
	102	+ s += sq[(x >> 24) & 0xff];
	103	+ s += sq[(x >> 32) & 0xff];
	104	+ s += sq[(x >> 40) & 0xff];
	105	+ s += sq[(x >> 48) & 0xff];
	106	+ s += sq[(x >> 56) & 0xff];
	107	+#else
	108	+ register uint32_t x = (uint32_t ) pix;
	109	+ s += sq[x & 0xff];
	110	+ s += sq[(x >> 8) & 0xff];
	111	+ s += sq[(x >> 16) & 0xff];
	112	+ s += sq[(x >> 24) & 0xff];
	113	+ x = (uint32_t ) (pix + 4);
	114	+ s += sq[x & 0xff];
	115	+ s += sq[(x >> 8) & 0xff];
	116	+ s += sq[(x >> 16) & 0xff];
	117	+ s += sq[(x >> 24) & 0xff];
	118	+#endif
	119	+#endif
	120	+ pix += 8;
	121	+ }
	122	+ pix += line_size - 16;
	123	+ }
	124	+ return s;
	125	+}
	126	+
56	127	av_cold void ff_mpegvideoencdsp_init(MpegvideoEncDSPContext *c,
57	128	AVCodecContext *avctx)
58	129	{
59	130	c->try_8x8basis = try_8x8basis_c;
60	131	c->add_8x8basis = add_8x8basis_c;
61	132
	133	+ c->shrink[0] = av_image_copy_plane;
	134	+ c->shrink[1] = ff_shrink22;
	135	+ c->shrink[2] = ff_shrink44;
	136	+ c->shrink[3] = ff_shrink88;
	137	+
	138	+ c->pix_sum = pix_sum_c;
	139	+ c->pix_norm1 = pix_norm1_c;
	140	+
	141	+ if (ARCH_ARM)
	142	+ ff_mpegvideoencdsp_init_arm(c, avctx);
	143	+ if (ARCH_PPC)
	144	+ ff_mpegvideoencdsp_init_ppc(c, avctx);
62	145	if (ARCH_X86)
63	146	ff_mpegvideoencdsp_init_x86(c, avctx);
64	147	}

libavcodec/mpegvideoencdsp.h+9−0

@@ -31,10 +31,19 @@ typedef struct MpegvideoEncDSPContext { · 30 unmodified lines
31	31	int16_t basis[64], int scale);
32	32	void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
33	33
	34	+ int (pix_sum)(uint8_t pix, int line_size);
	35	+ int (pix_norm1)(uint8_t pix, int line_size);
	36	+
	37	+ void (shrink[4])(uint8_t dst, int dst_wrap, const uint8_t *src,
	38	+ int src_wrap, int width, int height);
34	39	} MpegvideoEncDSPContext;
35	40
36	41	void ff_mpegvideoencdsp_init(MpegvideoEncDSPContext *c,
37	42	AVCodecContext *avctx);
	43	+void ff_mpegvideoencdsp_init_arm(MpegvideoEncDSPContext *c,
	44	+ AVCodecContext *avctx);
	45	+void ff_mpegvideoencdsp_init_ppc(MpegvideoEncDSPContext *c,
	46	+ AVCodecContext *avctx);
38	47	void ff_mpegvideoencdsp_init_x86(MpegvideoEncDSPContext *c,
39	48	AVCodecContext *avctx);
40	49

libavcodec/ppc/Makefile+1−0

@@ -13,6 +13,7 @@ OBJS-$(CONFIG_IDCTDSP) += ppc/idctdsp.o · 12 unmodified lines
13	13	OBJS-$(CONFIG_MPEGAUDIODSP) += ppc/mpegaudiodsp_altivec.o
14	14	OBJS-$(CONFIG_MPEGVIDEO) += ppc/mpegvideo_altivec.o \
15	15	ppc/mpegvideodsp.o
	16	+OBJS-$(CONFIG_MPEGVIDEOENC) += ppc/mpegvideoencdsp.o
16	17	OBJS-$(CONFIG_VIDEODSP) += ppc/videodsp_ppc.o
17	18	OBJS-$(CONFIG_VP3DSP) += ppc/vp3dsp_altivec.o
18	19

libavcodec/ppc/dsputil_altivec.c+0−60

@@ -308,34 +308,6 @@ static int sad8_altivec(MpegEncContext v, uint8_t pix1, uint8_t *pix2, · 307 unmodified lines
308	308	return s;
309	309	}
310	310
311		−static int pix_norm1_altivec(uint8_t *pix, int line_size)
312		−{
313		− int i, s = 0;
314		− const vector unsigned int zero =
315		− (const vector unsigned int) vec_splat_u32(0);
316		− vector unsigned char perm = vec_lvsl(0, pix);
317		− vector unsigned int sv = (vector unsigned int) vec_splat_u32(0);
318		− vector signed int sum;
319		−
320		− for (i = 0; i < 16; i++) {
321		− /* Read the potentially unaligned pixels. */
322		− vector unsigned char pixl = vec_ld(0, pix);
323		− vector unsigned char pixr = vec_ld(15, pix);
324		− vector unsigned char pixv = vec_perm(pixl, pixr, perm);
325		−
326		− /* Square the values, and add them to our sum. */
327		− sv = vec_msum(pixv, pixv, sv);
328		−
329		− pix += line_size;
330		− }
331		− /* Sum up the four partial sums, and put the result into s. */
332		− sum = vec_sums((vector signed int) sv, (vector signed int) zero);
333		− sum = vec_splat(sum, 3);
334		− vec_ste(sum, 0, &s);
335		−
336		− return s;
337		−}
338		−
339	311	/* Sum of Squared Errors for an 8x8 block, AltiVec-enhanced.
340	312	* It's the sad8_altivec code above w/ squaring added. */
341	313	static int sse8_altivec(MpegEncContext v, uint8_t pix1, uint8_t *pix2,
@@ -430,35 +402,6 @@ static int sse16_altivec(MpegEncContext v, uint8_t pix1, uint8_t *pix2, · 88 unmodified lines
430	402	return s;
431	403	}
432	404
433		−static int pix_sum_altivec(uint8_t *pix, int line_size)
434		−{
435		− int i, s;
436		− const vector unsigned int zero =
437		− (const vector unsigned int) vec_splat_u32(0);
438		− vector unsigned char perm = vec_lvsl(0, pix);
439		− vector unsigned int sad = (vector unsigned int) vec_splat_u32(0);
440		− vector signed int sumdiffs;
441		−
442		− for (i = 0; i < 16; i++) {
443		− /* Read the potentially unaligned 16 pixels into t1. */
444		− vector unsigned char pixl = vec_ld(0, pix);
445		− vector unsigned char pixr = vec_ld(15, pix);
446		− vector unsigned char t1 = vec_perm(pixl, pixr, perm);
447		−
448		− /* Add each 4 pixel group together and put 4 results into sad. */
449		− sad = vec_sum4s(t1, sad);
450		−
451		− pix += line_size;
452		− }
453		−
454		− /* Sum up the four partial sums, and put the result into s. */
455		− sumdiffs = vec_sums((vector signed int) sad, (vector signed int) zero);
456		− sumdiffs = vec_splat(sumdiffs, 3);
457		− vec_ste(sumdiffs, 0, &s);
458		−
459		− return s;
460		−}
461		−
462	405	static void get_pixels_altivec(int16_t restrict block, const uint8_t pixels,
463	406	int line_size)
464	407	{
@@ -911,9 +854,6 @@ av_cold void ff_dsputil_init_altivec(DSPContext c, AVCodecContext avctx, · 446 unmodified lines
911	854	c->sse[0] = sse16_altivec;
912	855	c->sse[1] = sse8_altivec;
913	856
914		− c->pix_norm1 = pix_norm1_altivec;
915		− c->pix_sum = pix_sum_altivec;
916		−
917	857	c->diff_pixels = diff_pixels_altivec;
918	858
919	859	if (!high_bit_depth) {

libavcodec/ppc/mpegvideoencdsp.c+103−0

@@ -0,0 +1,103 @@
	1	+/*
	2	+ * This file is part of Libav.
	3	+ *
	4	+ * Libav is free software; you can redistribute it and/or
	5	+ * modify it under the terms of the GNU Lesser General Public
	6	+ * License as published by the Free Software Foundation; either
	7	+ * version 2.1 of the License, or (at your option) any later version.
	8	+ *
	9	+ * Libav is distributed in the hope that it will be useful,
	10	+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
	11	+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	12	+ * Lesser General Public License for more details.
	13	+ *
	14	+ * You should have received a copy of the GNU Lesser General Public
	15	+ * License along with Libav; if not, write to the Free Software
	16	+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
	17	+ */
	18	+
	19	+#include "config.h"
	20	+#include <stdint.h>
	21	+#if HAVE_ALTIVEC_H
	22	+#include <altivec.h>
	23	+#endif
	24	+
	25	+#include "libavutil/attributes.h"
	26	+#include "libavutil/cpu.h"
	27	+#include "libavutil/ppc/cpu.h"
	28	+#include "libavutil/ppc/types_altivec.h"
	29	+#include "libavutil/ppc/util_altivec.h"
	30	+#include "libavcodec/mpegvideoencdsp.h"
	31	+
	32	+#if HAVE_ALTIVEC
	33	+
	34	+static int pix_norm1_altivec(uint8_t *pix, int line_size)
	35	+{
	36	+ int i, s = 0;
	37	+ const vector unsigned int zero =
	38	+ (const vector unsigned int) vec_splat_u32(0);
	39	+ vector unsigned char perm = vec_lvsl(0, pix);
	40	+ vector unsigned int sv = (vector unsigned int) vec_splat_u32(0);
	41	+ vector signed int sum;
	42	+
	43	+ for (i = 0; i < 16; i++) {
	44	+ /* Read the potentially unaligned pixels. */
	45	+ vector unsigned char pixl = vec_ld(0, pix);
	46	+ vector unsigned char pixr = vec_ld(15, pix);
	47	+ vector unsigned char pixv = vec_perm(pixl, pixr, perm);
	48	+
	49	+ /* Square the values, and add them to our sum. */
	50	+ sv = vec_msum(pixv, pixv, sv);
	51	+
	52	+ pix += line_size;
	53	+ }
	54	+ /* Sum up the four partial sums, and put the result into s. */
	55	+ sum = vec_sums((vector signed int) sv, (vector signed int) zero);
	56	+ sum = vec_splat(sum, 3);
	57	+ vec_ste(sum, 0, &s);
	58	+
	59	+ return s;
	60	+}
	61	+
	62	+static int pix_sum_altivec(uint8_t *pix, int line_size)
	63	+{
	64	+ int i, s;
	65	+ const vector unsigned int zero =
	66	+ (const vector unsigned int) vec_splat_u32(0);
	67	+ vector unsigned char perm = vec_lvsl(0, pix);
	68	+ vector unsigned int sad = (vector unsigned int) vec_splat_u32(0);
	69	+ vector signed int sumdiffs;
	70	+
	71	+ for (i = 0; i < 16; i++) {
	72	+ /* Read the potentially unaligned 16 pixels into t1. */
	73	+ vector unsigned char pixl = vec_ld(0, pix);
	74	+ vector unsigned char pixr = vec_ld(15, pix);
	75	+ vector unsigned char t1 = vec_perm(pixl, pixr, perm);
	76	+
	77	+ /* Add each 4 pixel group together and put 4 results into sad. */
	78	+ sad = vec_sum4s(t1, sad);
	79	+
	80	+ pix += line_size;
	81	+ }
	82	+
	83	+ /* Sum up the four partial sums, and put the result into s. */
	84	+ sumdiffs = vec_sums((vector signed int) sad, (vector signed int) zero);
	85	+ sumdiffs = vec_splat(sumdiffs, 3);
	86	+ vec_ste(sumdiffs, 0, &s);
	87	+
	88	+ return s;
	89	+}
	90	+
	91	+#endif /* HAVE_ALTIVEC */
	92	+
	93	+av_cold void ff_mpegvideoencdsp_init_ppc(MpegvideoEncDSPContext *c,
	94	+ AVCodecContext *avctx)
	95	+{
	96	+#if HAVE_ALTIVEC
	97	+ if (!PPC_ALTIVEC(av_get_cpu_flags()))
	98	+ return;
	99	+
	100	+ c->pix_norm1 = pix_norm1_altivec;
	101	+ c->pix_sum = pix_sum_altivec;
	102	+#endif /* HAVE_ALTIVEC */
	103	+}

libavcodec/svq1enc.c+1−0

@@ -511,6 +511,7 @@ static av_cold int svq1_encode_init(AVCodecContext *avctx) · 510 unmodified lines
511	511
512	512	ff_dsputil_init(&s->dsp, avctx);
513	513	ff_hpeldsp_init(&s->hdsp, avctx->flags);
	514	+ ff_mpegvideoencdsp_init(&s->m.mpvencdsp, avctx);
514	515
515	516	avctx->coded_frame = av_frame_alloc();
516	517	s->current_picture = av_frame_alloc();

libavcodec/x86/Makefile+1−0

@@ -92,6 +92,7 @@ YASM-OBJS-$(CONFIG_HPELDSP) += x86/fpel.o \ · 91 unmodified lines
92	92	x86/hpeldsp.o
93	93	YASM-OBJS-$(CONFIG_HUFFYUVDSP) += x86/huffyuvdsp.o
94	94	YASM-OBJS-$(CONFIG_MPEGAUDIODSP) += x86/imdct36.o
	95	+YASM-OBJS-$(CONFIG_MPEGVIDEOENC) += x86/mpegvideoencdsp.o
95	96	YASM-OBJS-$(CONFIG_QPELDSP) += x86/qpeldsp.o \
96	97	x86/fpel.o \
97	98	x86/qpel.o

libavcodec/x86/dsputilenc.asm+0−69

@@ -418,72 +418,3 @@ cglobal diff_pixels, 4,5 · 417 unmodified lines
418	418	add r4, 16
419	419	jne .loop
420	420	REP_RET
421		−
422		−INIT_MMX mmx
423		−; int ff_pix_sum16_mmx(uint8_t *pix, int line_size)
424		−cglobal pix_sum16, 2, 3
425		− movsxdifnidn r1, r1d
426		− mov r2, r1
427		− neg r2
428		− shl r2, 4
429		− sub r0, r2
430		− pxor m7, m7
431		− pxor m6, m6
432		−.loop:
433		− mova m0, [r0+r2+0]
434		− mova m1, [r0+r2+0]
435		− mova m2, [r0+r2+8]
436		− mova m3, [r0+r2+8]
437		− punpcklbw m0, m7
438		− punpckhbw m1, m7
439		− punpcklbw m2, m7
440		− punpckhbw m3, m7
441		− paddw m1, m0
442		− paddw m3, m2
443		− paddw m3, m1
444		− paddw m6, m3
445		− add r2, r1
446		− js .loop
447		− mova m5, m6
448		− psrlq m6, 32
449		− paddw m6, m5
450		− mova m5, m6
451		− psrlq m6, 16
452		− paddw m6, m5
453		− movd eax, m6
454		− and eax, 0xffff
455		− RET
456		−
457		−INIT_MMX mmx
458		−; int ff_pix_norm1_mmx(uint8_t *pix, int line_size)
459		−cglobal pix_norm1, 2, 4
460		− movsxdifnidn r1, r1d
461		− mov r2, 16
462		− pxor m0, m0
463		− pxor m7, m7
464		−.loop:
465		− mova m2, [r0+0]
466		− mova m3, [r0+8]
467		− mova m1, m2
468		− punpckhbw m1, m0
469		− punpcklbw m2, m0
470		− mova m4, m3
471		− punpckhbw m3, m0
472		− punpcklbw m4, m0
473		− pmaddwd m1, m1
474		− pmaddwd m2, m2
475		− pmaddwd m3, m3
476		− pmaddwd m4, m4
477		− paddd m2, m1
478		− paddd m4, m3
479		− paddd m7, m2
480		− add r0, r1
481		− paddd m7, m4
482		− dec r2
483		− jne .loop
484		− mova m1, m7
485		− psrlq m7, 32
486		− paddd m1, m7
487		− movd eax, m1
488		− RET
489		−

libavcodec/x86/dsputilenc_mmx.c+0−4

@@ -35,8 +35,6 @@ void ff_get_pixels_mmx(int16_t block, const uint8_t pixels, int line_size); · 34 unmodified lines
35	35	void ff_get_pixels_sse2(int16_t block, const uint8_t pixels, int line_size);
36	36	void ff_diff_pixels_mmx(int16_t block, const uint8_t s1, const uint8_t *s2,
37	37	int stride);
38		−int ff_pix_sum16_mmx(uint8_t *pix, int line_size);
39		−int ff_pix_norm1_mmx(uint8_t *pix, int line_size);
40	38
41	39	#if HAVE_INLINE_ASM
42	40
@@ -831,8 +829,6 @@ av_cold void ff_dsputilenc_init_mmx(DSPContext c, AVCodecContext avctx, · 788 unmodified lines
831	829	if (!high_bit_depth)
832	830	c->get_pixels = ff_get_pixels_mmx;
833	831	c->diff_pixels = ff_diff_pixels_mmx;
834		− c->pix_sum = ff_pix_sum16_mmx;
835		− c->pix_norm1 = ff_pix_norm1_mmx;
836	832	}
837	833
838	834	if (EXTERNAL_SSE2(cpu_flags))

libavcodec/x86/mpegvideoencdsp.asm+95−0

@@ -0,0 +1,95 @@
	1	+;*****************************************************************************
	2	+;* SIMD-optimized MPEG encoding functions
	3	+;*****************************************************************************
	4	+;* Copyright (c) 2000, 2001 Fabrice Bellard
	5	+;* Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
	6	+;*
	7	+;* This file is part of Libav.
	8	+;*
	9	+;* Libav is free software; you can redistribute it and/or
	10	+;* modify it under the terms of the GNU Lesser General Public
	11	+;* License as published by the Free Software Foundation; either
	12	+;* version 2.1 of the License, or (at your option) any later version.
	13	+;*
	14	+;* Libav is distributed in the hope that it will be useful,
	15	+;* but WITHOUT ANY WARRANTY; without even the implied warranty of
	16	+;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	17	+;* Lesser General Public License for more details.
	18	+;*
	19	+;* You should have received a copy of the GNU Lesser General Public
	20	+;* License along with Libav; if not, write to the Free Software
	21	+;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
	22	+;*****************************************************************************
	23	+
	24	+%include "libavutil/x86/x86util.asm"
	25	+
	26	+SECTION .text
	27	+
	28	+INIT_MMX mmx
	29	+; int ff_pix_sum16_mmx(uint8_t *pix, int line_size)
	30	+cglobal pix_sum16, 2, 3
	31	+ movsxdifnidn r1, r1d
	32	+ mov r2, r1
	33	+ neg r2
	34	+ shl r2, 4
	35	+ sub r0, r2
	36	+ pxor m7, m7
	37	+ pxor m6, m6
	38	+.loop:
	39	+ mova m0, [r0+r2+0]
	40	+ mova m1, [r0+r2+0]
	41	+ mova m2, [r0+r2+8]
	42	+ mova m3, [r0+r2+8]
	43	+ punpcklbw m0, m7
	44	+ punpckhbw m1, m7
	45	+ punpcklbw m2, m7
	46	+ punpckhbw m3, m7
	47	+ paddw m1, m0
	48	+ paddw m3, m2
	49	+ paddw m3, m1
	50	+ paddw m6, m3
	51	+ add r2, r1
	52	+ js .loop
	53	+ mova m5, m6
	54	+ psrlq m6, 32
	55	+ paddw m6, m5
	56	+ mova m5, m6
	57	+ psrlq m6, 16
	58	+ paddw m6, m5
	59	+ movd eax, m6
	60	+ and eax, 0xffff
	61	+ RET
	62	+
	63	+INIT_MMX mmx
	64	+; int ff_pix_norm1_mmx(uint8_t *pix, int line_size)
	65	+cglobal pix_norm1, 2, 4
	66	+ movsxdifnidn r1, r1d
	67	+ mov r2, 16
	68	+ pxor m0, m0
	69	+ pxor m7, m7
	70	+.loop:
	71	+ mova m2, [r0+0]
	72	+ mova m3, [r0+8]
	73	+ mova m1, m2
	74	+ punpckhbw m1, m0
	75	+ punpcklbw m2, m0
	76	+ mova m4, m3
	77	+ punpckhbw m3, m0
	78	+ punpcklbw m4, m0
	79	+ pmaddwd m1, m1
	80	+ pmaddwd m2, m2
	81	+ pmaddwd m3, m3
	82	+ pmaddwd m4, m4
	83	+ paddd m2, m1
	84	+ paddd m4, m3
	85	+ paddd m7, m2
	86	+ add r0, r1
	87	+ paddd m7, m4
	88	+ dec r2
	89	+ jne .loop
	90	+ mova m1, m7
	91	+ psrlq m7, 32
	92	+ paddd m1, m7
	93	+ movd eax, m1
	94	+ RET
	95	+

libavcodec/x86/mpegvideoencdsp_init.c+10−1

@@ -22,6 +22,9 @@ · 21 unmodified lines
22	22	#include "libavcodec/avcodec.h"
23	23	#include "libavcodec/mpegvideoencdsp.h"
24	24
	25	+int ff_pix_sum16_mmx(uint8_t *pix, int line_size);
	26	+int ff_pix_norm1_mmx(uint8_t *pix, int line_size);
	27	+
25	28	#if HAVE_INLINE_ASM
26	29
27	30	#define PHADDD(a, t) \
@@ -95,9 +98,15 @@ · 67 unmodified lines
95	98	av_cold void ff_mpegvideoencdsp_init_x86(MpegvideoEncDSPContext *c,
96	99	AVCodecContext *avctx)
97	100	{
98		−#if HAVE_INLINE_ASM
99	101	int cpu_flags = av_get_cpu_flags();
100	102
	103	+ if (EXTERNAL_MMX(cpu_flags)) {
	104	+ c->pix_sum = ff_pix_sum16_mmx;
	105	+ c->pix_norm1 = ff_pix_norm1_mmx;
	106	+ }
	107	+
	108	+#if HAVE_INLINE_ASM
	109	+
101	110	if (INLINE_MMX(cpu_flags)) {
102	111	if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {
103	112	c->try_8x8basis = try_8x8basis_mmx;

dsputil: Move pix_sum, pix_norm1, shrink function pointers to mpegvideoenc

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code