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

libavcodec/arm/dsputil_init_neon.c+0−2

 void ff_avg_h264_chroma_mc4_neon(uint8_t *, uint8_t *, int, int, int, int);
 void ff_avg_h264_chroma_mc2_neon(uint8_t *, uint8_t *, int, int, int, int);
-void ff_butterflies_float_neon(float *v1, float *v2, int len);
 float ff_scalarproduct_float_neon(const float *v1, const float *v2, int len);
 void ff_vector_clipf_neon(float *dst, const float *src, float min, float max,
         c->avg_h264_qpel_pixels_tab[1][15] = ff_avg_h264_qpel8_mc33_neon;
     }
-    c->butterflies_float          = ff_butterflies_float_neon;
     c->scalarproduct_float        = ff_scalarproduct_float_neon;
     c->vector_clipf               = ff_vector_clipf_neon;
     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	bx lr	531	bx lr
532	endfunc	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	function ff_scalarproduct_float_neon, export=1	534	function ff_scalarproduct_float_neon, export=1
547	vmov.f32 q2, #0.0	535	vmov.f32 q2, #0.0
548	1: vld1.32 {q0},[r0,:128]!	536	1: vld1.32 {q0},[r0,:128]!

libavcodec/dsputil.c+0−12

 WRAPPER8_16_SQ(rd8x8_c, rd16_c)
 WRAPPER8_16_SQ(bit8x8_c, bit16_c)
-static void butterflies_float_c(float *restrict v1, float *restrict v2,
-                                int len)
-{
-    int i;
-    for (i = 0; i < len; i++) {
-        float t = v1[i] - v2[i];
-        v1[i] += v2[i];
-        v2[i] = t;
-    }
-}
 float ff_scalarproduct_float_c(const float *v1, const float *v2, int len)
 {
     float p = 0.0;
     c->apply_window_int16 = apply_window_int16_c;
     c->vector_clip_int32 = vector_clip_int32_c;
     c->scalarproduct_float = ff_scalarproduct_float_c;
-    c->butterflies_float = butterflies_float_c;
     c->shrink[0]= av_image_copy_plane;
     c->shrink[1]= ff_shrink22;

libavcodec/dsputil.h+0−7

@@ -349,13 +349,6 @@ typedef struct DSPContext { · 348 unmodified lines
349	* @param len length of vectors, multiple of 4	349	* @param len length of vectors, multiple of 4
350	*/	350	*/
351	float (scalarproduct_float)(const float v1, const float *v2, int len);	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	/* (I)DCT */	353	/* (I)DCT */
361	void (fdct)(DCTELEM block/* align 16*/);	354	void (fdct)(DCTELEM block/* align 16*/);

libavcodec/imc.c+5−2

 #include <stdio.h>
 #include "libavutil/channel_layout.h"
+#include "libavutil/float_dsp.h"
 #include "avcodec.h"
 #include "get_bits.h"
 #include "dsputil.h"
     GetBitContext gb;
     DSPContext dsp;
+    AVFloatDSPContext fdsp;
     FFTContext fft;
     DECLARE_ALIGNED(32, FFTComplex, samples)[COEFFS / 2];
     float *out_samples;
         return ret;
     }
     ff_dsputil_init(&q->dsp, avctx);
+    avpriv_float_dsp_init(&q->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
     avctx->sample_fmt     = AV_SAMPLE_FMT_FLTP;
     avctx->channel_layout = avctx->channels == 1 ? AV_CH_LAYOUT_MONO
                                                  : AV_CH_LAYOUT_STEREO;
     }
     if (avctx->channels == 2) {
-        q->dsp.butterflies_float((float *)q->frame.extended_data[0],
+        q->fdsp.butterflies_float((float *)q->frame.extended_data[0],
-                                 (float *)q->frame.extended_data[1], COEFFS);
+                                  (float *)q->frame.extended_data[1], COEFFS);
     }
     *got_frame_ptr   = 1;

libavcodec/mpegaudiodec.c+4−3

  */
 #include "libavutil/channel_layout.h"
+#include "libavutil/float_dsp.h"
 #include "avcodec.h"
 #include "get_bits.h"
 #include "internal.h"
     int err_recognition;
     AVCodecContext* avctx;
     MPADSPContext mpadsp;
-    DSPContext dsp;
+    AVFloatDSPContext fdsp;
     AVFrame frame;
 } MPADecodeContext;
     s->avctx = avctx;
+    avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
     ff_mpadsp_init(&s->mpadsp);
-    ff_dsputil_init(&s->dsp, avctx);
     if (avctx->request_sample_fmt == OUT_FMT &&
         avctx->codec_id != AV_CODEC_ID_MP3ON4)
         /* NOTE: the 1/sqrt(2) normalization factor is included in the
            global gain */
 #if CONFIG_FLOAT
-       s-> dsp.butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576);
+       s->fdsp.butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576);
 #else
         tab0 = g0->sb_hybrid;
         tab1 = g1->sb_hybrid;

libavcodec/twinvq.c+1−3

 typedef struct TwinContext {
     AVCodecContext *avctx;
     AVFrame frame;
-    DSPContext      dsp;
     AVFloatDSPContext fdsp;
     FFTContext mdct_ctx[3];
     if (tctx->avctx->channels == 2) {
         memcpy(&out[1][0],     &prev_buf[2*mtab->size],         size1 * sizeof(out[1][0]));
         memcpy(&out[1][size1], &tctx->curr_frame[2*mtab->size], size2 * sizeof(out[1][0]));
-        tctx->dsp.butterflies_float(out[0], out[1], mtab->size);
+        tctx->fdsp.butterflies_float(out[0], out[1], mtab->size);
     }
 }
         return -1;
     }
-    ff_dsputil_init(&tctx->dsp, avctx);
     avpriv_float_dsp_init(&tctx->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
     if ((ret = init_mdct_win(tctx))) {
         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	\|\| avctx->bit_rate <= 0)	82	\|\| avctx->bit_rate <= 0)
83	return -1;	83	return -1;
84		84
85	ff_dsputil_init(&s->dsp, avctx);
86	ff_fmt_convert_init(&s->fmt_conv, avctx);	85	ff_fmt_convert_init(&s->fmt_conv, avctx);
87	avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);	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	float lsp_pow_e_table[256];	132	float lsp_pow_e_table[256];
133	float lsp_pow_m_table1[(1 << LSP_POW_BITS)];	133	float lsp_pow_m_table1[(1 << LSP_POW_BITS)];
134	float lsp_pow_m_table2[(1 << LSP_POW_BITS)];	134	float lsp_pow_m_table2[(1 << LSP_POW_BITS)];
135	DSPContext dsp;
136	FmtConvertContext fmt_conv;	135	FmtConvertContext fmt_conv;
137	AVFloatDSPContext fdsp;	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	s->channel_coded[0] = 1;	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	next:	725	next:

libavutil/arm/float_dsp_init_neon.c+3−0

 void ff_vector_fmul_reverse_neon(float *dst, const float *src0,
                                  const float *src1, int len);
+void ff_butterflies_float_neon(float *v1, float *v2, int len);
 void ff_float_dsp_init_neon(AVFloatDSPContext *fdsp)
 {
     fdsp->vector_fmul = ff_vector_fmul_neon;
     fdsp->vector_fmul_window = ff_vector_fmul_window_neon;
     fdsp->vector_fmul_add    = ff_vector_fmul_add_neon;
     fdsp->vector_fmul_reverse = ff_vector_fmul_reverse_neon;
+    fdsp->butterflies_float = ff_butterflies_float_neon;
 }

libavutil/arm/float_dsp_neon.S+12−0

@@ -244,3 +244,15 @@ function ff_vector_fmul_reverse_neon, export=1 · 243 unmodified lines
244	2: vst1.32 {q8-q9}, [r0,:128]!	244	2: vst1.32 {q8-q9}, [r0,:128]!
245	bx lr	245	bx lr
246	endfunc	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	dst[i] = src0[i] * src1[-i];	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	void avpriv_float_dsp_init(AVFloatDSPContext *fdsp, int bit_exact)	104	void avpriv_float_dsp_init(AVFloatDSPContext *fdsp, int bit_exact)
93	{	105	{
94	fdsp->vector_fmul = vector_fmul_c;	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	fdsp->vector_fmul_window = vector_fmul_window_c;	110	fdsp->vector_fmul_window = vector_fmul_window_c;
99	fdsp->vector_fmul_add = vector_fmul_add_c;	111	fdsp->vector_fmul_add = vector_fmul_add_c;
100	fdsp->vector_fmul_reverse = vector_fmul_reverse_c;	112	fdsp->vector_fmul_reverse = vector_fmul_reverse_c;
		113	fdsp->butterflies_float = butterflies_float_c;
101		114
102	#if ARCH_ARM	115	#if ARCH_ARM
103	ff_float_dsp_init_arm(fdsp);	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	void (vector_fmul_reverse)(float dst, const float *src0,	138	void (vector_fmul_reverse)(float dst, const float *src0,
139	const float *src1, int len);	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	} AVFloatDSPContext;	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