floatdsp: move butterflies_float from dsputil to avfloatdsp.

This makes wmadec/enc, twinvq and mpegaudiodec (i.e. mp2/mp3) independent of dsputil.

Ronald S. Bultje committed 13 years ago

Commit 5959bfa1 parent 42d324615 files · +51 −47Browse files

Unified · Split

15 files

libavcodec/aacdec.c +3 −3
libavcodec/arm/dsputil_init_neon.c +0 −2
libavcodec/arm/dsputil_neon.S +0 −12
libavcodec/dsputil.c +0 −12
libavcodec/dsputil.h +0 −7
libavcodec/imc.c +5 −2
libavcodec/mpegaudiodec.c +4 −3
libavcodec/twinvq.c +1 −3
libavcodec/wma.c +0 −1
libavcodec/wma.h +0 −1
libavcodec/wmadec.c +1 −1
libavutil/arm/float_dsp_init_neon.c +3 −0
libavutil/arm/float_dsp_neon.S +12 −0
libavutil/float_dsp.c +13 −0
libavutil/float_dsp.h +9 −0

@@ -1693,9 +1693,9 @@ static void apply_mid_side_stereo(AACContext ac, ChannelElement cpe) · 1692 unmodified lines
1693	1693	if (cpe->ms_mask[idx] &&
1694	1694	cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1695	1695	for (group = 0; group < ics->group_len[g]; group++) {
1696		− ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1697		− ch1 + group * 128 + offsets[i],
1698		− offsets[i+1] - offsets[i]);
	1696	+ ac->fdsp.butterflies_float(ch0 + group * 128 + offsets[i],
	1697	+ ch1 + group * 128 + offsets[i],
	1698	+ offsets[i+1] - offsets[i]);
1699	1699	}
1700	1700	}
1701	1701	}

libavcodec/arm/dsputil_init_neon.c+0−2

@@ -142,7 +142,6 @@ void ff_avg_h264_chroma_mc8_neon(uint8_t , uint8_t , int, int, int, int); · 141 unmodified lines
142	142	void ff_avg_h264_chroma_mc4_neon(uint8_t , uint8_t , int, int, int, int);
143	143	void ff_avg_h264_chroma_mc2_neon(uint8_t , uint8_t , int, int, int, int);
144	144
145		−void ff_butterflies_float_neon(float v1, float v2, int len);
146	145	float ff_scalarproduct_float_neon(const float v1, const float v2, int len);
147	146
148	147	void ff_vector_clipf_neon(float dst, const float src, float min, float max,
@@ -294,7 +293,6 @@ void ff_dsputil_init_neon(DSPContext c, AVCodecContext avctx) · 145 unmodified lines
294	293	c->avg_h264_qpel_pixels_tab[1][15] = ff_avg_h264_qpel8_mc33_neon;
295	294	}
296	295
297		− c->butterflies_float = ff_butterflies_float_neon;
298	296	c->scalarproduct_float = ff_scalarproduct_float_neon;
299	297	c->vector_clipf = ff_vector_clipf_neon;
300	298	c->vector_clip_int32 = ff_vector_clip_int32_neon;

libavcodec/arm/dsputil_neon.S+0−12

@@ -531,18 +531,6 @@ function ff_add_pixels_clamped_neon, export=1 · 530 unmodified lines
531	531	bx lr
532	532	endfunc
533	533
534		−function ff_butterflies_float_neon, export=1
535		−1: vld1.32 {q0},[r0,:128]
536		− vld1.32 {q1},[r1,:128]
537		− vsub.f32 q2, q0, q1
538		− vadd.f32 q1, q0, q1
539		− vst1.32 {q2},[r1,:128]!
540		− vst1.32 {q1},[r0,:128]!
541		− subs r2, r2, #4
542		− bgt 1b
543		− bx lr
544		−endfunc
545		−
546	534	function ff_scalarproduct_float_neon, export=1
547	535	vmov.f32 q2, #0.0
548	536	1: vld1.32 {q0},[r0,:128]!

libavcodec/dsputil.c+0−12

@@ -2353,17 +2353,6 @@ WRAPPER8_16_SQ(quant_psnr8x8_c, quant_psnr16_c) · 2352 unmodified lines
2353	2353	WRAPPER8_16_SQ(rd8x8_c, rd16_c)
2354	2354	WRAPPER8_16_SQ(bit8x8_c, bit16_c)
2355	2355
2356		−static void butterflies_float_c(float restrict v1, float restrict v2,
2357		− int len)
2358		−{
2359		− int i;
2360		− for (i = 0; i < len; i++) {
2361		− float t = v1[i] - v2[i];
2362		− v1[i] += v2[i];
2363		− v2[i] = t;
2364		− }
2365		−}
2366		−
2367	2356	float ff_scalarproduct_float_c(const float v1, const float v2, int len)
2368	2357	{
2369	2358	float p = 0.0;
@@ -2706,7 +2695,6 @@ av_cold void ff_dsputil_init(DSPContext* c, AVCodecContext *avctx) · 336 unmodified lines
2706	2695	c->apply_window_int16 = apply_window_int16_c;
2707	2696	c->vector_clip_int32 = vector_clip_int32_c;
2708	2697	c->scalarproduct_float = ff_scalarproduct_float_c;
2709		− c->butterflies_float = butterflies_float_c;
2710	2698
2711	2699	c->shrink[0]= av_image_copy_plane;
2712	2700	c->shrink[1]= ff_shrink22;

libavcodec/dsputil.h+0−7

@@ -349,13 +349,6 @@ typedef struct DSPContext { · 348 unmodified lines
349	349	* @param len length of vectors, multiple of 4
350	350	*/
351	351	float (scalarproduct_float)(const float v1, const float *v2, int len);
352		− /**
353		− * Calculate the sum and difference of two vectors of floats.
354		− * @param v1 first input vector, sum output, 16-byte aligned
355		− * @param v2 second input vector, difference output, 16-byte aligned
356		− * @param len length of vectors, multiple of 4
357		− */
358		− void (butterflies_float)(float restrict v1, float *restrict v2, int len);
359	352
360	353	/* (I)DCT */
361	354	void (fdct)(DCTELEM block/* align 16*/);

libavcodec/imc.c+5−2

@@ -36,6 +36,7 @@ · 35 unmodified lines
36	36	#include <stdio.h>
37	37
38	38	#include "libavutil/channel_layout.h"
	39	+#include "libavutil/float_dsp.h"
39	40	#include "avcodec.h"
40	41	#include "get_bits.h"
41	42	#include "dsputil.h"
@@ -95,6 +96,7 @@ typedef struct { · 53 unmodified lines
95	96	GetBitContext gb;
96	97
97	98	DSPContext dsp;
	99	+ AVFloatDSPContext fdsp;
98	100	FFTContext fft;
99	101	DECLARE_ALIGNED(32, FFTComplex, samples)[COEFFS / 2];
100	102	float *out_samples;
@@ -244,6 +246,7 @@ static av_cold int imc_decode_init(AVCodecContext *avctx) · 143 unmodified lines
244	246	return ret;
245	247	}
246	248	ff_dsputil_init(&q->dsp, avctx);
	249	+ avpriv_float_dsp_init(&q->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
247	250	avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
248	251	avctx->channel_layout = avctx->channels == 1 ? AV_CH_LAYOUT_MONO
249	252	: AV_CH_LAYOUT_STEREO;
@@ -959,8 +962,8 @@ static int imc_decode_frame(AVCodecContext avctx, void data, · 709 unmodified lines
959	962	}
960	963
961	964	if (avctx->channels == 2) {
962		− q->dsp.butterflies_float((float *)q->frame.extended_data[0],
963		− (float *)q->frame.extended_data[1], COEFFS);
	965	+ q->fdsp.butterflies_float((float *)q->frame.extended_data[0],
	966	+ (float *)q->frame.extended_data[1], COEFFS);
964	967	}
965	968
966	969	*got_frame_ptr = 1;

libavcodec/mpegaudiodec.c+4−3

@@ -25,6 +25,7 @@ · 24 unmodified lines
25	25	*/
26	26
27	27	#include "libavutil/channel_layout.h"
	28	+#include "libavutil/float_dsp.h"
28	29	#include "avcodec.h"
29	30	#include "get_bits.h"
30	31	#include "internal.h"
@@ -82,7 +83,7 @@ typedef struct MPADecodeContext { · 51 unmodified lines
82	83	int err_recognition;
83	84	AVCodecContext* avctx;
84	85	MPADSPContext mpadsp;
85		− DSPContext dsp;
	86	+ AVFloatDSPContext fdsp;
86	87	AVFrame frame;
87	88	} MPADecodeContext;
88	89
@@ -434,8 +435,8 @@ static av_cold int decode_init(AVCodecContext * avctx) · 345 unmodified lines
434	435
435	436	s->avctx = avctx;
436	437
	438	+ avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
437	439	ff_mpadsp_init(&s->mpadsp);
438		− ff_dsputil_init(&s->dsp, avctx);
439	440
440	441	if (avctx->request_sample_fmt == OUT_FMT &&
441	442	avctx->codec_id != AV_CODEC_ID_MP3ON4)
@@ -1157,7 +1158,7 @@ found2: · 715 unmodified lines
1157	1158	/* NOTE: the 1/sqrt(2) normalization factor is included in the
1158	1159	global gain */
1159	1160	#if CONFIG_FLOAT
1160		− s-> dsp.butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576);
	1161	+ s->fdsp.butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576);
1161	1162	#else
1162	1163	tab0 = g0->sb_hybrid;
1163	1164	tab1 = g1->sb_hybrid;

libavcodec/twinvq.c+1−3

@@ -178,7 +178,6 @@ static const ModeTab mode_44_48 = { · 177 unmodified lines
178	178	typedef struct TwinContext {
179	179	AVCodecContext *avctx;
180	180	AVFrame frame;
181		− DSPContext dsp;
182	181	AVFloatDSPContext fdsp;
183	182	FFTContext mdct_ctx[3];
184	183
@@ -693,7 +692,7 @@ static void imdct_output(TwinContext *tctx, enum FrameType ftype, int wtype, · 508 unmodified lines
693	692	if (tctx->avctx->channels == 2) {
694	693	memcpy(&out[1][0], &prev_buf[2mtab->size], size1 sizeof(out[1][0]));
695	694	memcpy(&out[1][size1], &tctx->curr_frame[2mtab->size], size2 sizeof(out[1][0]));
696		− tctx->dsp.butterflies_float(out[0], out[1], mtab->size);
	695	+ tctx->fdsp.butterflies_float(out[0], out[1], mtab->size);
697	696	}
698	697	}
699	698
@@ -1157,7 +1156,6 @@ static av_cold int twin_decode_init(AVCodecContext *avctx) · 457 unmodified lines
1157	1156	return -1;
1158	1157	}
1159	1158
1160		− ff_dsputil_init(&tctx->dsp, avctx);
1161	1159	avpriv_float_dsp_init(&tctx->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
1162	1160	if ((ret = init_mdct_win(tctx))) {
1163	1161	av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");

libavcodec/wma.c+0−1

@@ -82,7 +82,6 @@ int ff_wma_init(AVCodecContext *avctx, int flags2) · 81 unmodified lines
82	82	\|\| avctx->bit_rate <= 0)
83	83	return -1;
84	84
85		− ff_dsputil_init(&s->dsp, avctx);
86	85	ff_fmt_convert_init(&s->fmt_conv, avctx);
87	86	avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
88	87

libavcodec/wma.h+0−1

@@ -132,7 +132,6 @@ typedef struct WMACodecContext { · 131 unmodified lines
132	132	float lsp_pow_e_table[256];
133	133	float lsp_pow_m_table1[(1 << LSP_POW_BITS)];
134	134	float lsp_pow_m_table2[(1 << LSP_POW_BITS)];
135		− DSPContext dsp;
136	135	FmtConvertContext fmt_conv;
137	136	AVFloatDSPContext fdsp;
138	137

libavcodec/wmadec.c+1−1

@@ -719,7 +719,7 @@ static int wma_decode_block(WMACodecContext *s) · 718 unmodified lines
719	719	s->channel_coded[0] = 1;
720	720	}
721	721
722		− s->dsp.butterflies_float(s->coefs[0], s->coefs[1], s->block_len);
	722	+ s->fdsp.butterflies_float(s->coefs[0], s->coefs[1], s->block_len);
723	723	}
724	724
725	725	next:

libavutil/arm/float_dsp_init_neon.c+3−0

@@ -41,6 +41,8 @@ void ff_vector_fmul_add_neon(float dst, const float src0, const float *src1, · 40 unmodified lines
41	41	void ff_vector_fmul_reverse_neon(float dst, const float src0,
42	42	const float *src1, int len);
43	43
	44	+void ff_butterflies_float_neon(float v1, float v2, int len);
	45	+
44	46	void ff_float_dsp_init_neon(AVFloatDSPContext *fdsp)
45	47	{
46	48	fdsp->vector_fmul = ff_vector_fmul_neon;
@@ -49,4 +51,5 @@ void ff_float_dsp_init_neon(AVFloatDSPContext *fdsp) · 2 unmodified lines
49	51	fdsp->vector_fmul_window = ff_vector_fmul_window_neon;
50	52	fdsp->vector_fmul_add = ff_vector_fmul_add_neon;
51	53	fdsp->vector_fmul_reverse = ff_vector_fmul_reverse_neon;
	54	+ fdsp->butterflies_float = ff_butterflies_float_neon;
52	55	}

libavutil/arm/float_dsp_neon.S+12−0

@@ -244,3 +244,15 @@ function ff_vector_fmul_reverse_neon, export=1 · 243 unmodified lines
244	244	2: vst1.32 {q8-q9}, [r0,:128]!
245	245	bx lr
246	246	endfunc
	247	+
	248	+function ff_butterflies_float_neon, export=1
	249	+1: vld1.32 {q0},[r0,:128]
	250	+ vld1.32 {q1},[r1,:128]
	251	+ vsub.f32 q2, q0, q1
	252	+ vadd.f32 q1, q0, q1
	253	+ vst1.32 {q2},[r1,:128]!
	254	+ vst1.32 {q1},[r0,:128]!
	255	+ subs r2, r2, #4
	256	+ bgt 1b
	257	+ bx lr
	258	+endfunc

libavutil/float_dsp.c+13−0

@@ -89,6 +89,18 @@ static void vector_fmul_reverse_c(float dst, const float src0, · 88 unmodified lines
89	89	dst[i] = src0[i] * src1[-i];
90	90	}
91	91
	92	+static void butterflies_float_c(float restrict v1, float restrict v2,
	93	+ int len)
	94	+{
	95	+ int i;
	96	+
	97	+ for (i = 0; i < len; i++) {
	98	+ float t = v1[i] - v2[i];
	99	+ v1[i] += v2[i];
	100	+ v2[i] = t;
	101	+ }
	102	+}
	103	+
92	104	void avpriv_float_dsp_init(AVFloatDSPContext *fdsp, int bit_exact)
93	105	{
94	106	fdsp->vector_fmul = vector_fmul_c;
@@ -98,6 +110,7 @@ void avpriv_float_dsp_init(AVFloatDSPContext *fdsp, int bit_exact) · 3 unmodified lines
98	110	fdsp->vector_fmul_window = vector_fmul_window_c;
99	111	fdsp->vector_fmul_add = vector_fmul_add_c;
100	112	fdsp->vector_fmul_reverse = vector_fmul_reverse_c;
	113	+ fdsp->butterflies_float = butterflies_float_c;
101	114
102	115	#if ARCH_ARM
103	116	ff_float_dsp_init_arm(fdsp);

libavutil/float_dsp.h+9−0

@@ -137,6 +137,15 @@ typedef struct AVFloatDSPContext { · 136 unmodified lines
137	137	*/
138	138	void (vector_fmul_reverse)(float dst, const float *src0,
139	139	const float *src1, int len);
	140	+
	141	+ /**
	142	+ * Calculate the sum and difference of two vectors of floats.
	143	+ *
	144	+ * @param v1 first input vector, sum output, 16-byte aligned
	145	+ * @param v2 second input vector, difference output, 16-byte aligned
	146	+ * @param len length of vectors, multiple of 4
	147	+ */
	148	+ void (butterflies_float)(float restrict v1, float *restrict v2, int len);
140	149	} AVFloatDSPContext;
141	150
142	151	/**

floatdsp: move butterflies_float from dsputil to avfloatdsp.

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code