dsputil: Split audio operations off into a separate context

configure

@@ -1529,6 +1529,7 @@ CONFIG_EXTRA="
     aandcttables
     ac3dsp
     audio_frame_queue
+    audiodsp
     blockdsp
     cabac
     dsputil
@@ -1713,8 +1714,8 @@ aac_decoder_select="mdct sinewin"
 aac_encoder_select="audio_frame_queue mdct sinewin"
 aac_latm_decoder_select="aac_decoder aac_latm_parser"
 ac3_decoder_select="mdct ac3dsp ac3_parser dsputil"
-ac3_encoder_select="mdct ac3dsp dsputil"
-ac3_fixed_encoder_select="mdct ac3dsp dsputil"
+ac3_encoder_select="ac3dsp audiodsp dsputil mdct"
+ac3_fixed_encoder_select="ac3dsp audiodsp dsputil mdct"
 aic_decoder_select="dsputil golomb"
 alac_encoder_select="lpc"
 als_decoder_select="dsputil"
@@ -1735,7 +1736,7 @@ binkaudio_rdft_decoder_select="mdct rdft sinewin"
 cavs_decoder_select="blockdsp dsputil golomb h264chroma qpeldsp videodsp"
 cllc_decoder_select="dsputil"
 comfortnoise_encoder_select="lpc"
-cook_decoder_select="dsputil mdct sinewin"
+cook_decoder_select="audiodsp mdct sinewin"
 cscd_decoder_select="lzo"
 cscd_decoder_suggest="zlib"
 dca_decoder_select="mdct"
@@ -1849,7 +1850,7 @@ svq1_decoder_select="hpeldsp"
 svq1_encoder_select="aandcttables dsputil hpeldsp mpegvideoenc"
 svq3_decoder_select="h264_decoder hpeldsp tpeldsp"
 svq3_decoder_suggest="zlib"
-tak_decoder_select="dsputil"
+tak_decoder_select="audiodsp"
 theora_decoder_select="vp3_decoder"
 thp_decoder_select="mjpeg_decoder"
 tiff_decoder_suggest="zlib"

libavcodec/Makefile

@@ -28,6 +28,7 @@ OBJS = allcodecs.o                                                      \
 OBJS-$(CONFIG_AANDCTTABLES)            += aandcttab.o
 OBJS-$(CONFIG_AC3DSP)                  += ac3dsp.o
 OBJS-$(CONFIG_AUDIO_FRAME_QUEUE)       += audio_frame_queue.o
+OBJS-$(CONFIG_AUDIODSP)                += audiodsp.o
 OBJS-$(CONFIG_BLOCKDSP)                += blockdsp.o
 OBJS-$(CONFIG_CABAC)                   += cabac.o
 OBJS-$(CONFIG_DCT)                     += dct.o dct32_fixed.o dct32_float.o

libavcodec/ac3enc.c

@@ -37,6 +37,7 @@
 #include "libavutil/opt.h"
 #include "avcodec.h"
 #include "put_bits.h"
+#include "audiodsp.h"
 #include "ac3dsp.h"
 #include "ac3.h"
 #include "fft.h"
@@ -2480,6 +2481,7 @@ av_cold int ff_ac3_encode_init(AVCodecContext *avctx)
     if (ret)
         goto init_fail;
 
+    ff_audiodsp_init(&s->adsp);
     ff_dsputil_init(&s->dsp, avctx);
     ff_ac3dsp_init(&s->ac3dsp, avctx->flags & CODEC_FLAG_BITEXACT);
 

libavcodec/ac3enc.h

@@ -39,6 +39,7 @@
 #include "fft.h"
 #include "mathops.h"
 #include "put_bits.h"
+#include "audiodsp.h"
 
 #ifndef CONFIG_AC3ENC_FLOAT
 #define CONFIG_AC3ENC_FLOAT 0
@@ -162,6 +163,7 @@ typedef struct AC3EncodeContext {
     AVCodecContext *avctx;                  ///< parent AVCodecContext
     PutBitContext pb;                       ///< bitstream writer context
     DSPContext dsp;
+    AudioDSPContext adsp;
     AVFloatDSPContext fdsp;
     AC3DSPContext ac3dsp;                   ///< AC-3 optimized functions
     FFTContext mdct;                        ///< FFT context for MDCT calculation

libavcodec/ac3enc_fixed.c

@@ -29,6 +29,7 @@
 #define FFT_FLOAT 0
 #undef CONFIG_AC3ENC_FLOAT
 #include "internal.h"
+#include "audiodsp.h"
 #include "ac3enc.h"
 #include "eac3enc.h"
 
@@ -100,9 +101,10 @@ static void scale_coefficients(AC3EncodeContext *s)
 /*
  * Clip MDCT coefficients to allowable range.
  */
-static void clip_coefficients(DSPContext *dsp, int32_t *coef, unsigned int len)
+static void clip_coefficients(AudioDSPContext *adsp, int32_t *coef,
+                              unsigned int len)
 {
-    dsp->vector_clip_int32(coef, coef, COEF_MIN, COEF_MAX, len);
+    adsp->vector_clip_int32(coef, coef, COEF_MIN, COEF_MAX, len);
 }
 
 

libavcodec/ac3enc_float.c

@@ -28,6 +28,7 @@
 
 #define CONFIG_AC3ENC_FLOAT 1
 #include "internal.h"
+#include "audiodsp.h"
 #include "ac3enc.h"
 #include "eac3enc.h"
 #include "kbdwin.h"
@@ -107,9 +108,10 @@ static void scale_coefficients(AC3EncodeContext *s)
 /*
  * Clip MDCT coefficients to allowable range.
  */
-static void clip_coefficients(DSPContext *dsp, float *coef, unsigned int len)
+static void clip_coefficients(AudioDSPContext *adsp, float *coef,
+                              unsigned int len)
 {
-    dsp->vector_clipf(coef, coef, COEF_MIN, COEF_MAX, len);
+    adsp->vector_clipf(coef, coef, COEF_MIN, COEF_MAX, len);
 }
 
 

libavcodec/ac3enc_template.c

@@ -30,6 +30,8 @@
 
 #include "libavutil/attributes.h"
 #include "libavutil/internal.h"
+
+#include "audiodsp.h"
 #include "internal.h"
 #include "ac3enc.h"
 #include "eac3enc.h"
@@ -40,7 +42,8 @@ static void scale_coefficients(AC3EncodeContext *s);
 
 static int normalize_samples(AC3EncodeContext *s);
 
-static void clip_coefficients(DSPContext *dsp, CoefType *coef, unsigned int len);
+static void clip_coefficients(AudioDSPContext *adsp, CoefType *coef,
+                              unsigned int len);
 
 static CoefType calc_cpl_coord(CoefSumType energy_ch, CoefSumType energy_cpl);
 
@@ -161,7 +164,7 @@ static void apply_channel_coupling(AC3EncodeContext *s)
         }
 
         /* coefficients must be clipped in order to be encoded */
-        clip_coefficients(&s->dsp, cpl_coef, num_cpl_coefs);
+        clip_coefficients(&s->adsp, cpl_coef, num_cpl_coefs);
     }
 
     /* calculate energy in each band in coupling channel and each fbw channel */
@@ -412,7 +415,7 @@ int AC3_NAME(encode_frame)(AVCodecContext *avctx, AVPacket *avpkt,
     if (s->fixed_point)
         scale_coefficients(s);
 
-    clip_coefficients(&s->dsp, s->blocks[0].mdct_coef[1],
+    clip_coefficients(&s->adsp, s->blocks[0].mdct_coef[1],
                       AC3_MAX_COEFS * s->num_blocks * s->channels);
 
     s->cpl_on = s->cpl_enabled;

libavcodec/acelp_pitch_delay.c

@@ -26,6 +26,7 @@
 #include "avcodec.h"
 #include "acelp_pitch_delay.h"
 #include "celp_math.h"
+#include "audiodsp.h"
 
 int ff_acelp_decode_8bit_to_1st_delay3(int ac_index)
 {
@@ -90,7 +91,7 @@ void ff_acelp_update_past_gain(
 }
 
 int16_t ff_acelp_decode_gain_code(
-    DSPContext *dsp,
+    AudioDSPContext *adsp,
     int gain_corr_factor,
     const int16_t* fc_v,
     int mr_energy,
@@ -107,7 +108,7 @@ int16_t ff_acelp_decode_gain_code(
         mr_energy += quant_energy[i] * ma_prediction_coeff[i];
 
     mr_energy = gain_corr_factor * exp(M_LN10 / (20 << 23) * mr_energy) /
-                sqrt(dsp->scalarproduct_int16(fc_v, fc_v, subframe_size));
+                sqrt(adsp->scalarproduct_int16(fc_v, fc_v, subframe_size));
     return mr_energy >> 12;
 }
 

libavcodec/acelp_pitch_delay.h

@@ -24,7 +24,8 @@
 #define AVCODEC_ACELP_PITCH_DELAY_H
 
 #include <stdint.h>
-#include "dsputil.h"
+
+#include "audiodsp.h"
 
 #define PITCH_DELAY_MIN             20
 #define PITCH_DELAY_MAX             143
@@ -139,7 +140,7 @@ void ff_acelp_update_past_gain(
 /**
  * @brief Decode the adaptive codebook gain and add
  *        correction (4.1.5 and 3.9.1 of G.729).
- * @param dsp initialized dsputil context
+ * @param adsp initialized audio DSP context
  * @param gain_corr_factor gain correction factor (2.13)
  * @param fc_v fixed-codebook vector (2.13)
  * @param mr_energy mean innovation energy and fixed-point correction (7.13)
@@ -208,7 +209,7 @@ void ff_acelp_update_past_gain(
  * @remark The routine is used in G.729 and AMR (all modes).
  */
 int16_t ff_acelp_decode_gain_code(
-    DSPContext *dsp,
+    AudioDSPContext *adsp,
     int gain_corr_factor,
     const int16_t* fc_v,
     int mr_energy,

libavcodec/arm/Makefile

@@ -4,6 +4,7 @@ OBJS                                   += arm/fmtconvert_init_arm.o
 
 OBJS-$(CONFIG_AC3DSP)                  += arm/ac3dsp_init_arm.o         \
                                           arm/ac3dsp_arm.o
+OBJS-$(CONFIG_AUDIODSP)                += arm/audiodsp_init_arm.o
 OBJS-$(CONFIG_BLOCKDSP)                += arm/blockdsp_init_arm.o
 OBJS-$(CONFIG_DSPUTIL)                 += arm/dsputil_init_arm.o        \
                                           arm/dsputil_arm.o             \
@@ -77,11 +78,13 @@ VFP-OBJS-$(CONFIG_DCA_DECODER)         += arm/dcadsp_vfp.o              \
 NEON-OBJS                              += arm/fmtconvert_neon.o
 
 NEON-OBJS-$(CONFIG_AC3DSP)             += arm/ac3dsp_neon.o
+NEON-OBJS-$(CONFIG_AUDIODSP)           += arm/audiodsp_init_neon.o      \
+                                          arm/audiodsp_neon.o           \
+                                          arm/int_neon.o
 NEON-OBJS-$(CONFIG_BLOCKDSP)           += arm/blockdsp_init_neon.o      \
                                           arm/blockdsp_neon.o
 NEON-OBJS-$(CONFIG_DSPUTIL)            += arm/dsputil_init_neon.o       \
                                           arm/dsputil_neon.o            \
-                                          arm/int_neon.o                \
                                           arm/simple_idct_neon.o
 NEON-OBJS-$(CONFIG_FFT)                += arm/fft_neon.o                \
                                           arm/fft_fixed_neon.o

libavcodec/arm/audiodsp_arm.h

@@ -0,0 +1,26 @@
+/*
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVCODEC_ARM_AUDIODSP_ARM_H
+#define AVCODEC_ARM_AUDIODSP_ARM_H
+
+#include "libavcodec/audiodsp.h"
+
+void ff_audiodsp_init_neon(AudioDSPContext *c);
+
+#endif /* AVCODEC_ARM_AUDIODSP_ARM_H */

libavcodec/arm/audiodsp_init_arm.c

@@ -0,0 +1,33 @@
+/*
+ * ARM optimized audio functions
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/attributes.h"
+#include "libavutil/cpu.h"
+#include "libavutil/arm/cpu.h"
+#include "libavcodec/audiodsp.h"
+#include "audiodsp_arm.h"
+
+av_cold void ff_audiodsp_init_arm(AudioDSPContext *c)
+{
+    int cpu_flags = av_get_cpu_flags();
+
+    if (have_neon(cpu_flags))
+        ff_audiodsp_init_neon(c);
+}

libavcodec/arm/audiodsp_init_neon.c

@@ -0,0 +1,41 @@
+/*
+ * ARM NEON optimised audio functions
+ * Copyright (c) 2008 Mans Rullgard <mans@mansr.com>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include <stdint.h>
+
+#include "libavutil/attributes.h"
+#include "libavcodec/audiodsp.h"
+#include "audiodsp_arm.h"
+
+void ff_vector_clipf_neon(float *dst, const float *src, float min, float max,
+                          int len);
+void ff_vector_clip_int32_neon(int32_t *dst, const int32_t *src, int32_t min,
+                               int32_t max, unsigned int len);
+
+int32_t ff_scalarproduct_int16_neon(const int16_t *v1, const int16_t *v2, int len);
+
+av_cold void ff_audiodsp_init_neon(AudioDSPContext *c)
+{
+    c->vector_clip_int32 = ff_vector_clip_int32_neon;
+    c->vector_clipf      = ff_vector_clipf_neon;
+
+    c->scalarproduct_int16 = ff_scalarproduct_int16_neon;
+}

libavcodec/arm/audiodsp_neon.S

@@ -0,0 +1,64 @@
+/*
+ * ARM NEON optimised audio functions
+ * Copyright (c) 2008 Mans Rullgard <mans@mansr.com>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/arm/asm.S"
+
+function ff_vector_clipf_neon, export=1
+VFP     vdup.32         q1,  d0[1]
+VFP     vdup.32         q0,  d0[0]
+NOVFP   vdup.32         q0,  r2
+NOVFP   vdup.32         q1,  r3
+NOVFP   ldr             r2,  [sp]
+        vld1.f32        {q2},[r1,:128]!
+        vmin.f32        q10, q2,  q1
+        vld1.f32        {q3},[r1,:128]!
+        vmin.f32        q11, q3,  q1
+1:      vmax.f32        q8,  q10, q0
+        vmax.f32        q9,  q11, q0
+        subs            r2,  r2,  #8
+        beq             2f
+        vld1.f32        {q2},[r1,:128]!
+        vmin.f32        q10, q2,  q1
+        vld1.f32        {q3},[r1,:128]!
+        vmin.f32        q11, q3,  q1
+        vst1.f32        {q8},[r0,:128]!
+        vst1.f32        {q9},[r0,:128]!
+        b               1b
+2:      vst1.f32        {q8},[r0,:128]!
+        vst1.f32        {q9},[r0,:128]!
+        bx              lr
+endfunc
+
+function ff_vector_clip_int32_neon, export=1
+        vdup.32         q0,  r2
+        vdup.32         q1,  r3
+        ldr             r2,  [sp]
+1:
+        vld1.32         {q2-q3},  [r1,:128]!
+        vmin.s32        q2,  q2,  q1
+        vmin.s32        q3,  q3,  q1
+        vmax.s32        q2,  q2,  q0
+        vmax.s32        q3,  q3,  q0
+        vst1.32         {q2-q3},  [r0,:128]!
+        subs            r2,  r2,  #8
+        bgt             1b
+        bx              lr
+endfunc

libavcodec/arm/dsputil_init_neon.c

@@ -34,13 +34,6 @@ void ff_add_pixels_clamped_neon(const int16_t *, uint8_t *, int);
 void ff_put_pixels_clamped_neon(const int16_t *, uint8_t *, int);
 void ff_put_signed_pixels_clamped_neon(const int16_t *, uint8_t *, int);
 
-void ff_vector_clipf_neon(float *dst, const float *src, float min, float max,
-                          int len);
-void ff_vector_clip_int32_neon(int32_t *dst, const int32_t *src, int32_t min,
-                               int32_t max, unsigned int len);
-
-int32_t ff_scalarproduct_int16_neon(const int16_t *v1, const int16_t *v2, int len);
-
 av_cold void ff_dsputil_init_neon(DSPContext *c, AVCodecContext *avctx,
                                   unsigned high_bit_depth)
 {
@@ -57,9 +50,4 @@ av_cold void ff_dsputil_init_neon(DSPContext *c, AVCodecContext *avctx,
     c->add_pixels_clamped        = ff_add_pixels_clamped_neon;
     c->put_pixels_clamped        = ff_put_pixels_clamped_neon;
     c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_neon;
-
-    c->vector_clipf      = ff_vector_clipf_neon;
-    c->vector_clip_int32 = ff_vector_clip_int32_neon;
-
-    c->scalarproduct_int16 = ff_scalarproduct_int16_neon;
 }

libavcodec/arm/dsputil_neon.S

@@ -126,45 +126,3 @@ function ff_add_pixels_clamped_neon, export=1
         vst1.8          {d6},    [r3,:64], r2
         bx              lr
 endfunc
-
-function ff_vector_clipf_neon, export=1
-VFP     vdup.32         q1,  d0[1]
-VFP     vdup.32         q0,  d0[0]
-NOVFP   vdup.32         q0,  r2
-NOVFP   vdup.32         q1,  r3
-NOVFP   ldr             r2,  [sp]
-        vld1.f32        {q2},[r1,:128]!
-        vmin.f32        q10, q2,  q1
-        vld1.f32        {q3},[r1,:128]!
-        vmin.f32        q11, q3,  q1
-1:      vmax.f32        q8,  q10, q0
-        vmax.f32        q9,  q11, q0
-        subs            r2,  r2,  #8
-        beq             2f
-        vld1.f32        {q2},[r1,:128]!
-        vmin.f32        q10, q2,  q1
-        vld1.f32        {q3},[r1,:128]!
-        vmin.f32        q11, q3,  q1
-        vst1.f32        {q8},[r0,:128]!
-        vst1.f32        {q9},[r0,:128]!
-        b               1b
-2:      vst1.f32        {q8},[r0,:128]!
-        vst1.f32        {q9},[r0,:128]!
-        bx              lr
-endfunc
-
-function ff_vector_clip_int32_neon, export=1
-        vdup.32         q0,  r2
-        vdup.32         q1,  r3
-        ldr             r2,  [sp]
-1:
-        vld1.32         {q2-q3},  [r1,:128]!
-        vmin.s32        q2,  q2,  q1
-        vmin.s32        q3,  q3,  q1
-        vmax.s32        q2,  q2,  q0
-        vmax.s32        q3,  q3,  q0
-        vst1.32         {q2-q3},  [r0,:128]!
-        subs            r2,  r2,  #8
-        bgt             1b
-        bx              lr
-endfunc

libavcodec/audiodsp.c

@@ -0,0 +1,118 @@
+/*
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include <stdint.h>
+
+#include "libavutil/attributes.h"
+#include "libavutil/common.h"
+#include "audiodsp.h"
+
+static inline uint32_t clipf_c_one(uint32_t a, uint32_t mini,
+                                   uint32_t maxi, uint32_t maxisign)
+{
+    if (a > mini)
+        return mini;
+    else if ((a ^ (1U << 31)) > maxisign)
+        return maxi;
+    else
+        return a;
+}
+
+static void vector_clipf_c_opposite_sign(float *dst, const float *src,
+                                         float *min, float *max, int len)
+{
+    int i;
+    uint32_t mini        = *(uint32_t *) min;
+    uint32_t maxi        = *(uint32_t *) max;
+    uint32_t maxisign    = maxi ^ (1U << 31);
+    uint32_t *dsti       = (uint32_t *) dst;
+    const uint32_t *srci = (const uint32_t *) src;
+
+    for (i = 0; i < len; i += 8) {
+        dsti[i + 0] = clipf_c_one(srci[i + 0], mini, maxi, maxisign);
+        dsti[i + 1] = clipf_c_one(srci[i + 1], mini, maxi, maxisign);
+        dsti[i + 2] = clipf_c_one(srci[i + 2], mini, maxi, maxisign);
+        dsti[i + 3] = clipf_c_one(srci[i + 3], mini, maxi, maxisign);
+        dsti[i + 4] = clipf_c_one(srci[i + 4], mini, maxi, maxisign);
+        dsti[i + 5] = clipf_c_one(srci[i + 5], mini, maxi, maxisign);
+        dsti[i + 6] = clipf_c_one(srci[i + 6], mini, maxi, maxisign);
+        dsti[i + 7] = clipf_c_one(srci[i + 7], mini, maxi, maxisign);
+    }
+}
+
+static void vector_clipf_c(float *dst, const float *src,
+                           float min, float max, int len)
+{
+    int i;
+
+    if (min < 0 && max > 0) {
+        vector_clipf_c_opposite_sign(dst, src, &min, &max, len);
+    } else {
+        for (i = 0; i < len; i += 8) {
+            dst[i]     = av_clipf(src[i], min, max);
+            dst[i + 1] = av_clipf(src[i + 1], min, max);
+            dst[i + 2] = av_clipf(src[i + 2], min, max);
+            dst[i + 3] = av_clipf(src[i + 3], min, max);
+            dst[i + 4] = av_clipf(src[i + 4], min, max);
+            dst[i + 5] = av_clipf(src[i + 5], min, max);
+            dst[i + 6] = av_clipf(src[i + 6], min, max);
+            dst[i + 7] = av_clipf(src[i + 7], min, max);
+        }
+    }
+}
+
+static int32_t scalarproduct_int16_c(const int16_t *v1, const int16_t *v2,
+                                     int order)
+{
+    int res = 0;
+
+    while (order--)
+        res += *v1++ **v2++;
+
+    return res;
+}
+
+static void vector_clip_int32_c(int32_t *dst, const int32_t *src, int32_t min,
+                                int32_t max, unsigned int len)
+{
+    do {
+        *dst++ = av_clip(*src++, min, max);
+        *dst++ = av_clip(*src++, min, max);
+        *dst++ = av_clip(*src++, min, max);
+        *dst++ = av_clip(*src++, min, max);
+        *dst++ = av_clip(*src++, min, max);
+        *dst++ = av_clip(*src++, min, max);
+        *dst++ = av_clip(*src++, min, max);
+        *dst++ = av_clip(*src++, min, max);
+        len   -= 8;
+    } while (len > 0);
+}
+
+av_cold void ff_audiodsp_init(AudioDSPContext *c)
+{
+    c->scalarproduct_int16 = scalarproduct_int16_c;
+    c->vector_clip_int32   = vector_clip_int32_c;
+    c->vector_clipf        = vector_clipf_c;
+
+    if (ARCH_ARM)
+        ff_audiodsp_init_arm(c);
+    if (ARCH_PPC)
+        ff_audiodsp_init_ppc(c);
+    if (ARCH_X86)
+        ff_audiodsp_init_x86(c);
+}

libavcodec/audiodsp.h

@@ -0,0 +1,59 @@
+/*
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVCODEC_AUDIODSP_H
+#define AVCODEC_AUDIODSP_H
+
+#include <stdint.h>
+
+typedef struct AudioDSPContext {
+    /**
+     * Calculate scalar product of two vectors.
+     * @param len length of vectors, should be multiple of 16
+     */
+    int32_t (*scalarproduct_int16)(const int16_t *v1,
+                                   const int16_t *v2 /* align 16 */, int len);
+
+    /**
+     * Clip each element in an array of int32_t to a given minimum and
+     * maximum value.
+     * @param dst  destination array
+     *             constraints: 16-byte aligned
+     * @param src  source array
+     *             constraints: 16-byte aligned
+     * @param min  minimum value
+     *             constraints: must be in the range [-(1 << 24), 1 << 24]
+     * @param max  maximum value
+     *             constraints: must be in the range [-(1 << 24), 1 << 24]
+     * @param len  number of elements in the array
+     *             constraints: multiple of 32 greater than zero
+     */
+    void (*vector_clip_int32)(int32_t *dst, const int32_t *src, int32_t min,
+                              int32_t max, unsigned int len);
+    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
+    void (*vector_clipf)(float *dst /* align 16 */,
+                         const float *src /* align 16 */,
+                         float min, float max, int len /* align 16 */);
+} AudioDSPContext;
+
+void ff_audiodsp_init(AudioDSPContext *c);
+void ff_audiodsp_init_arm(AudioDSPContext *c);
+void ff_audiodsp_init_ppc(AudioDSPContext *c);
+void ff_audiodsp_init_x86(AudioDSPContext *c);
+
+#endif /* AVCODEC_AUDIODSP_H */

libavcodec/cook.c

@@ -44,9 +44,10 @@
 
 #include "libavutil/channel_layout.h"
 #include "libavutil/lfg.h"
+
+#include "audiodsp.h"
 #include "avcodec.h"
 #include "get_bits.h"
-#include "dsputil.h"
 #include "bytestream.h"
 #include "fft.h"
 #include "internal.h"
@@ -122,7 +123,7 @@ typedef struct cook {
     void (*saturate_output)(struct cook *q, float *out);
 
     AVCodecContext*     avctx;
-    DSPContext          dsp;
+    AudioDSPContext     adsp;
     GetBitContext       gb;
     /* stream data */
     int                 num_vectors;
@@ -865,8 +866,8 @@ static inline void decode_bytes_and_gain(COOKContext *q, COOKSubpacket *p,
  */
 static void saturate_output_float(COOKContext *q, float *out)
 {
-    q->dsp.vector_clipf(out, q->mono_mdct_output + q->samples_per_channel,
-                        -1.0f, 1.0f, FFALIGN(q->samples_per_channel, 8));
+    q->adsp.vector_clipf(out, q->mono_mdct_output + q->samples_per_channel,
+                         -1.0f, 1.0f, FFALIGN(q->samples_per_channel, 8));
 }
 
 
@@ -1065,7 +1066,7 @@ static av_cold int cook_decode_init(AVCodecContext *avctx)
     /* Initialize RNG. */
     av_lfg_init(&q->random_state, 0);
 
-    ff_dsputil_init(&q->dsp, avctx);
+    ff_audiodsp_init(&q->adsp);
 
     while (edata_ptr < edata_ptr_end) {
         /* 8 for mono, 16 for stereo, ? for multichannel

libavcodec/dsputil.c

@@ -1267,87 +1267,6 @@ WRAPPER8_16_SQ(quant_psnr8x8_c, quant_psnr16_c)
 WRAPPER8_16_SQ(rd8x8_c, rd16_c)
 WRAPPER8_16_SQ(bit8x8_c, bit16_c)
 
-static inline uint32_t clipf_c_one(uint32_t a, uint32_t mini,
-                                   uint32_t maxi, uint32_t maxisign)
-{
-    if (a > mini)
-        return mini;
-    else if ((a ^ (1U << 31)) > maxisign)
-        return maxi;
-    else
-        return a;
-}
-
-static void vector_clipf_c_opposite_sign(float *dst, const float *src,
-                                         float *min, float *max, int len)
-{
-    int i;
-    uint32_t mini        = *(uint32_t *) min;
-    uint32_t maxi        = *(uint32_t *) max;
-    uint32_t maxisign    = maxi ^ (1U << 31);
-    uint32_t *dsti       = (uint32_t *) dst;
-    const uint32_t *srci = (const uint32_t *) src;
-
-    for (i = 0; i < len; i += 8) {
-        dsti[i + 0] = clipf_c_one(srci[i + 0], mini, maxi, maxisign);
-        dsti[i + 1] = clipf_c_one(srci[i + 1], mini, maxi, maxisign);
-        dsti[i + 2] = clipf_c_one(srci[i + 2], mini, maxi, maxisign);
-        dsti[i + 3] = clipf_c_one(srci[i + 3], mini, maxi, maxisign);
-        dsti[i + 4] = clipf_c_one(srci[i + 4], mini, maxi, maxisign);
-        dsti[i + 5] = clipf_c_one(srci[i + 5], mini, maxi, maxisign);
-        dsti[i + 6] = clipf_c_one(srci[i + 6], mini, maxi, maxisign);
-        dsti[i + 7] = clipf_c_one(srci[i + 7], mini, maxi, maxisign);
-    }
-}
-
-static void vector_clipf_c(float *dst, const float *src,
-                           float min, float max, int len)
-{
-    int i;
-
-    if (min < 0 && max > 0) {
-        vector_clipf_c_opposite_sign(dst, src, &min, &max, len);
-    } else {
-        for (i = 0; i < len; i += 8) {
-            dst[i]     = av_clipf(src[i], min, max);
-            dst[i + 1] = av_clipf(src[i + 1], min, max);
-            dst[i + 2] = av_clipf(src[i + 2], min, max);
-            dst[i + 3] = av_clipf(src[i + 3], min, max);
-            dst[i + 4] = av_clipf(src[i + 4], min, max);
-            dst[i + 5] = av_clipf(src[i + 5], min, max);
-            dst[i + 6] = av_clipf(src[i + 6], min, max);
-            dst[i + 7] = av_clipf(src[i + 7], min, max);
-        }
-    }
-}
-
-static int32_t scalarproduct_int16_c(const int16_t *v1, const int16_t *v2,
-                                     int order)
-{
-    int res = 0;
-
-    while (order--)
-        res += *v1++ **v2++;
-
-    return res;
-}
-
-static void vector_clip_int32_c(int32_t *dst, const int32_t *src, int32_t min,
-                                int32_t max, unsigned int len)
-{
-    do {
-        *dst++ = av_clip(*src++, min, max);
-        *dst++ = av_clip(*src++, min, max);
-        *dst++ = av_clip(*src++, min, max);
-        *dst++ = av_clip(*src++, min, max);
-        *dst++ = av_clip(*src++, min, max);
-        *dst++ = av_clip(*src++, min, max);
-        *dst++ = av_clip(*src++, min, max);
-        *dst++ = av_clip(*src++, min, max);
-        len   -= 8;
-    } while (len > 0);
-}
-
 static void jref_idct_put(uint8_t *dest, int line_size, int16_t *block)
 {
     ff_j_rev_dct(block);
@@ -1502,10 +1421,6 @@ av_cold void ff_dsputil_init(DSPContext *c, AVCodecContext *avctx)
     c->try_8x8basis = try_8x8basis_c;
     c->add_8x8basis = add_8x8basis_c;
 
-    c->scalarproduct_int16 = scalarproduct_int16_c;
-    c->vector_clip_int32   = vector_clip_int32_c;
-    c->vector_clipf        = vector_clipf_c;
-
     c->shrink[0] = av_image_copy_plane;
     c->shrink[1] = ff_shrink22;
     c->shrink[2] = ff_shrink44;

libavcodec/dsputil.h

@@ -125,11 +125,6 @@ typedef struct DSPContext {
     void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
     void (*bswap16_buf)(uint16_t *dst, const uint16_t *src, int len);
 
-    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
-    void (*vector_clipf)(float *dst /* align 16 */,
-                         const float *src /* align 16 */,
-                         float min, float max, int len /* align 16 */);
-
     /* (I)DCT */
     void (*fdct)(int16_t *block /* align 16 */);
     void (*fdct248)(int16_t *block /* align 16 */);
@@ -189,30 +184,6 @@ typedef struct DSPContext {
 
     void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src,
                       int src_wrap, int width, int height);
-
-    /**
-     * Calculate scalar product of two vectors.
-     * @param len length of vectors, should be multiple of 16
-     */
-    int32_t (*scalarproduct_int16)(const int16_t *v1,
-                                   const int16_t *v2 /* align 16 */, int len);
-
-    /**
-     * Clip each element in an array of int32_t to a given minimum and
-     * maximum value.
-     * @param dst  destination array
-     *             constraints: 16-byte aligned
-     * @param src  source array
-     *             constraints: 16-byte aligned
-     * @param min  minimum value
-     *             constraints: must be in the range [-(1 << 24), 1 << 24]
-     * @param max  maximum value
-     *             constraints: must be in the range [-(1 << 24), 1 << 24]
-     * @param len  number of elements in the array
-     *             constraints: multiple of 32 greater than zero
-     */
-    void (*vector_clip_int32)(int32_t *dst, const int32_t *src, int32_t min,
-                              int32_t max, unsigned int len);
 } DSPContext;
 
 void ff_dsputil_static_init(void);

libavcodec/ppc/Makefile

@@ -1,5 +1,6 @@
 OBJS                                   += ppc/fmtconvert_altivec.o      \
 
+OBJS-$(CONFIG_AUDIODSP)                += ppc/audiodsp.o
 OBJS-$(CONFIG_BLOCKDSP)                += ppc/blockdsp.o
 OBJS-$(CONFIG_DSPUTIL)                 += ppc/dsputil_ppc.o
 OBJS-$(CONFIG_FFT)                     += ppc/fft_altivec.o
@@ -24,7 +25,6 @@ ALTIVEC-OBJS-$(CONFIG_DSPUTIL)         += ppc/dsputil_altivec.o         \
                                           ppc/fdct_altivec.o            \
                                           ppc/gmc_altivec.o             \
                                           ppc/idct_altivec.o            \
-                                          ppc/int_altivec.o             \
 
 FFT-OBJS-$(HAVE_GNU_AS)                += ppc/fft_altivec_s.o
 ALTIVEC-OBJS-$(CONFIG_FFT)             += $(FFT-OBJS-yes)

libavcodec/ppc/int_altivec.c → libavcodec/ppc/audiodsp.c

@@ -20,7 +20,7 @@
 
 /**
  * @file
- * miscellaneous integer operations
+ * miscellaneous audio operations
  */
 
 #include "config.h"
@@ -29,10 +29,13 @@
 #endif
 
 #include "libavutil/attributes.h"
+#include "libavutil/cpu.h"
+#include "libavutil/ppc/cpu.h"
 #include "libavutil/ppc/types_altivec.h"
 #include "libavutil/ppc/util_altivec.h"
-#include "libavcodec/dsputil.h"
-#include "dsputil_altivec.h"
+#include "libavcodec/audiodsp.h"
+
+#if HAVE_ALTIVEC
 
 static int32_t scalarproduct_int16_altivec(const int16_t *v1, const int16_t *v2,
                                            int order)
@@ -56,7 +59,14 @@ static int32_t scalarproduct_int16_altivec(const int16_t *v1, const int16_t *v2,
     return ires;
 }
 
-av_cold void ff_int_init_altivec(DSPContext *c, AVCodecContext *avctx)
+#endif /* HAVE_ALTIVEC */
+
+av_cold void ff_audiodsp_init_ppc(AudioDSPContext *c)
 {
+#if HAVE_ALTIVEC
+    if (!PPC_ALTIVEC(av_get_cpu_flags()))
+        return;
+
     c->scalarproduct_int16 = scalarproduct_int16_altivec;
+#endif /* HAVE_ALTIVEC */
 }

libavcodec/ppc/dsputil_altivec.h

@@ -35,6 +35,5 @@ void ff_idct_add_altivec(uint8_t *dest, int line_size, int16_t *block);
 
 void ff_dsputil_init_altivec(DSPContext *c, AVCodecContext *avctx,
                              unsigned high_bit_depth);
-void ff_int_init_altivec(DSPContext *c, AVCodecContext *avctx);
 
 #endif /* AVCODEC_PPC_DSPUTIL_ALTIVEC_H */

libavcodec/ppc/dsputil_ppc.c

@@ -34,7 +34,7 @@ av_cold void ff_dsputil_init_ppc(DSPContext *c, AVCodecContext *avctx,
 {
     if (PPC_ALTIVEC(av_get_cpu_flags())) {
         ff_dsputil_init_altivec(c, avctx, high_bit_depth);
-        ff_int_init_altivec(c, avctx);
+
         c->gmc1 = ff_gmc1_altivec;
 
         if (!high_bit_depth) {

libavcodec/takdec.c

@@ -28,8 +28,8 @@
 #include "libavutil/internal.h"
 #include "libavutil/samplefmt.h"
 #include "tak.h"
+#include "audiodsp.h"
 #include "avcodec.h"
-#include "dsputil.h"
 #include "internal.h"
 #include "unary.h"
 
@@ -45,7 +45,7 @@ typedef struct MCDParam {
 
 typedef struct TAKDecContext {
     AVCodecContext *avctx;                          // parent AVCodecContext
-    DSPContext      dsp;
+    AudioDSPContext adsp;
     TAKStreamInfo   ti;
     GetBitContext   gb;                             // bitstream reader initialized to start at the current frame
 
@@ -172,7 +172,7 @@ static av_cold int tak_decode_init(AVCodecContext *avctx)
 {
     TAKDecContext *s = avctx->priv_data;
 
-    ff_dsputil_init(&s->dsp, avctx);
+    ff_audiodsp_init(&s->adsp);
 
     s->avctx = avctx;
 
@@ -484,8 +484,8 @@ static int decode_subframe(TAKDecContext *s, int32_t *decoded,
     for (i = 0; i < subframe_size - filter_order; i++) {
         int v = 1 << (filter_quant - 1);
 
-        v += s->dsp.scalarproduct_int16(&s->residues[i], filter,
-                                        FFALIGN(filter_order, 16));
+        v += s->adsp.scalarproduct_int16(&s->residues[i], filter,
+                                         FFALIGN(filter_order, 16));
 
         v = (av_clip(v >> filter_quant, -8192, 8191) << dshift) - *decoded;
         *decoded++ = v;
@@ -654,8 +654,8 @@ static int decorrelate(TAKDecContext *s, int c1, int c2, int length)
         for (i = 0; i < length2; i++) {
             int v = 1 << 9;
 
-            v += s->dsp.scalarproduct_int16(&s->residues[i], filter,
-                                            FFALIGN(filter_order, 16));
+            v += s->adsp.scalarproduct_int16(&s->residues[i], filter,
+                                             FFALIGN(filter_order, 16));
 
             p1[i] = (av_clip(v >> 10, -8192, 8191) << dshift) - p1[i];
         }

libavcodec/x86/Makefile

@@ -2,6 +2,7 @@ OBJS                                   += x86/constants.o               \
                                           x86/fmtconvert_init.o         \
 
 OBJS-$(CONFIG_AC3DSP)                  += x86/ac3dsp_init.o
+OBJS-$(CONFIG_AUDIODSP)                += x86/audiodsp_init.o
 OBJS-$(CONFIG_DCT)                     += x86/dct_init.o
 OBJS-$(CONFIG_DSPUTIL)                 += x86/dsputil_init.o
 OBJS-$(CONFIG_ENCODERS)                += x86/dsputilenc_mmx.o          \
@@ -44,6 +45,7 @@ OBJS-$(CONFIG_VP7_DECODER)             += x86/vp8dsp_init.o
 OBJS-$(CONFIG_VP8_DECODER)             += x86/vp8dsp_init.o
 OBJS-$(CONFIG_VP9_DECODER)             += x86/vp9dsp_init.o
 
+MMX-OBJS-$(CONFIG_AUDIODSP)            += x86/audiodsp_mmx.o
 MMX-OBJS-$(CONFIG_BLOCKDSP)            += x86/blockdsp_mmx.o
 MMX-OBJS-$(CONFIG_DSPUTIL)             += x86/dsputil_mmx.o             \
                                           x86/idct_mmx_xvid.o           \
@@ -61,6 +63,7 @@ YASM-OBJS                              += x86/deinterlace.o             \
                                           x86/fmtconvert.o              \
 
 YASM-OBJS-$(CONFIG_AC3DSP)             += x86/ac3dsp.o
+YASM-OBJS-$(CONFIG_AUDIODSP)           += x86/audiodsp.o
 YASM-OBJS-$(CONFIG_DCT)                += x86/dct32.o
 YASM-OBJS-$(CONFIG_DNXHD_ENCODER)      += x86/dnxhdenc.o
 YASM-OBJS-$(CONFIG_DSPUTIL)            += x86/dsputil.o

libavcodec/x86/audiodsp.asm

@@ -0,0 +1,137 @@
+;******************************************************************************
+;* optimized audio functions
+;* Copyright (c) 2008 Loren Merritt
+;*
+;* This file is part of Libav.
+;*
+;* Libav is free software; you can redistribute it and/or
+;* modify it under the terms of the GNU Lesser General Public
+;* License as published by the Free Software Foundation; either
+;* version 2.1 of the License, or (at your option) any later version.
+;*
+;* Libav is distributed in the hope that it will be useful,
+;* but WITHOUT ANY WARRANTY; without even the implied warranty of
+;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+;* Lesser General Public License for more details.
+;*
+;* You should have received a copy of the GNU Lesser General Public
+;* License along with Libav; if not, write to the Free Software
+;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+;******************************************************************************
+
+%include "libavutil/x86/x86util.asm"
+
+SECTION_TEXT
+
+%macro SCALARPRODUCT 0
+; int ff_scalarproduct_int16(int16_t *v1, int16_t *v2, int order)
+cglobal scalarproduct_int16, 3,3,3, v1, v2, order
+    shl orderq, 1
+    add v1q, orderq
+    add v2q, orderq
+    neg orderq
+    pxor    m2, m2
+.loop:
+    movu    m0, [v1q + orderq]
+    movu    m1, [v1q + orderq + mmsize]
+    pmaddwd m0, [v2q + orderq]
+    pmaddwd m1, [v2q + orderq + mmsize]
+    paddd   m2, m0
+    paddd   m2, m1
+    add     orderq, mmsize*2
+    jl .loop
+%if mmsize == 16
+    movhlps m0, m2
+    paddd   m2, m0
+    pshuflw m0, m2, 0x4e
+%else
+    pshufw  m0, m2, 0x4e
+%endif
+    paddd   m2, m0
+    movd   eax, m2
+    RET
+%endmacro
+
+INIT_MMX mmxext
+SCALARPRODUCT
+INIT_XMM sse2
+SCALARPRODUCT
+
+
+;-----------------------------------------------------------------------------
+; void ff_vector_clip_int32(int32_t *dst, const int32_t *src, int32_t min,
+;                           int32_t max, unsigned int len)
+;-----------------------------------------------------------------------------
+
+; %1 = number of xmm registers used
+; %2 = number of inline load/process/store loops per asm loop
+; %3 = process 4*mmsize (%3=0) or 8*mmsize (%3=1) bytes per loop
+; %4 = CLIPD function takes min/max as float instead of int (CLIPD_SSE2)
+; %5 = suffix
+%macro VECTOR_CLIP_INT32 4-5
+cglobal vector_clip_int32%5, 5,5,%1, dst, src, min, max, len
+%if %4
+    cvtsi2ss  m4, minm
+    cvtsi2ss  m5, maxm
+%else
+    movd      m4, minm
+    movd      m5, maxm
+%endif
+    SPLATD    m4
+    SPLATD    m5
+.loop:
+%assign %%i 1
+%rep %2
+    mova      m0,  [srcq+mmsize*0*%%i]
+    mova      m1,  [srcq+mmsize*1*%%i]
+    mova      m2,  [srcq+mmsize*2*%%i]
+    mova      m3,  [srcq+mmsize*3*%%i]
+%if %3
+    mova      m7,  [srcq+mmsize*4*%%i]
+    mova      m8,  [srcq+mmsize*5*%%i]
+    mova      m9,  [srcq+mmsize*6*%%i]
+    mova      m10, [srcq+mmsize*7*%%i]
+%endif
+    CLIPD  m0,  m4, m5, m6
+    CLIPD  m1,  m4, m5, m6
+    CLIPD  m2,  m4, m5, m6
+    CLIPD  m3,  m4, m5, m6
+%if %3
+    CLIPD  m7,  m4, m5, m6
+    CLIPD  m8,  m4, m5, m6
+    CLIPD  m9,  m4, m5, m6
+    CLIPD  m10, m4, m5, m6
+%endif
+    mova  [dstq+mmsize*0*%%i], m0
+    mova  [dstq+mmsize*1*%%i], m1
+    mova  [dstq+mmsize*2*%%i], m2
+    mova  [dstq+mmsize*3*%%i], m3
+%if %3
+    mova  [dstq+mmsize*4*%%i], m7
+    mova  [dstq+mmsize*5*%%i], m8
+    mova  [dstq+mmsize*6*%%i], m9
+    mova  [dstq+mmsize*7*%%i], m10
+%endif
+%assign %%i %%i+1
+%endrep
+    add     srcq, mmsize*4*(%2+%3)
+    add     dstq, mmsize*4*(%2+%3)
+    sub     lend, mmsize*(%2+%3)
+    jg .loop
+    REP_RET
+%endmacro
+
+INIT_MMX mmx
+%define CLIPD CLIPD_MMX
+VECTOR_CLIP_INT32 0, 1, 0, 0
+INIT_XMM sse2
+VECTOR_CLIP_INT32 6, 1, 0, 0, _int
+%define CLIPD CLIPD_SSE2
+VECTOR_CLIP_INT32 6, 2, 0, 1
+INIT_XMM sse4
+%define CLIPD CLIPD_SSE41
+%ifdef m8
+VECTOR_CLIP_INT32 11, 1, 1, 0
+%else
+VECTOR_CLIP_INT32 6, 1, 0, 0
+%endif

libavcodec/x86/audiodsp.h

@@ -0,0 +1,25 @@
+/*
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVCODEC_X86_AUDIODSP_H
+#define AVCODEC_X86_AUDIODSP_H
+
+void ff_vector_clipf_sse(float *dst, const float *src,
+                         float min, float max, int len);
+
+#endif /* AVCODEC_X86_AUDIODSP_H */

libavcodec/x86/audiodsp_init.c

@@ -0,0 +1,66 @@
+/*
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include <stdint.h>
+
+#include "config.h"
+#include "libavutil/attributes.h"
+#include "libavutil/cpu.h"
+#include "libavutil/x86/asm.h"
+#include "libavutil/x86/cpu.h"
+#include "libavcodec/audiodsp.h"
+#include "audiodsp.h"
+
+int32_t ff_scalarproduct_int16_mmxext(const int16_t *v1, const int16_t *v2,
+                                      int order);
+int32_t ff_scalarproduct_int16_sse2(const int16_t *v1, const int16_t *v2,
+                                    int order);
+
+void ff_vector_clip_int32_mmx(int32_t *dst, const int32_t *src,
+                              int32_t min, int32_t max, unsigned int len);
+void ff_vector_clip_int32_sse2(int32_t *dst, const int32_t *src,
+                               int32_t min, int32_t max, unsigned int len);
+void ff_vector_clip_int32_int_sse2(int32_t *dst, const int32_t *src,
+                                   int32_t min, int32_t max, unsigned int len);
+void ff_vector_clip_int32_sse4(int32_t *dst, const int32_t *src,
+                               int32_t min, int32_t max, unsigned int len);
+
+av_cold void ff_audiodsp_init_x86(AudioDSPContext *c)
+{
+    int cpu_flags = av_get_cpu_flags();
+
+    if (EXTERNAL_MMX(cpu_flags))
+        c->vector_clip_int32 = ff_vector_clip_int32_mmx;
+
+    if (EXTERNAL_MMXEXT(cpu_flags))
+        c->scalarproduct_int16 = ff_scalarproduct_int16_mmxext;
+
+    if (INLINE_SSE(cpu_flags))
+        c->vector_clipf = ff_vector_clipf_sse;
+
+    if (EXTERNAL_SSE2(cpu_flags)) {
+        c->scalarproduct_int16 = ff_scalarproduct_int16_sse2;
+        if (cpu_flags & AV_CPU_FLAG_ATOM)
+            c->vector_clip_int32 = ff_vector_clip_int32_int_sse2;
+        else
+            c->vector_clip_int32 = ff_vector_clip_int32_sse2;
+    }
+
+    if (EXTERNAL_SSE4(cpu_flags))
+        c->vector_clip_int32 = ff_vector_clip_int32_sse4;
+}

libavcodec/x86/audiodsp_mmx.c

@@ -0,0 +1,58 @@
+/*
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "config.h"
+#include "libavutil/x86/asm.h"
+#include "audiodsp.h"
+
+#if HAVE_INLINE_ASM
+
+void ff_vector_clipf_sse(float *dst, const float *src,
+                         float min, float max, int len)
+{
+    x86_reg i = (len - 16) * 4;
+    __asm__ volatile (
+        "movss          %3, %%xmm4      \n\t"
+        "movss          %4, %%xmm5      \n\t"
+        "shufps $0, %%xmm4, %%xmm4      \n\t"
+        "shufps $0, %%xmm5, %%xmm5      \n\t"
+        "1:                             \n\t"
+        "movaps   (%2, %0), %%xmm0      \n\t" // 3/1 on intel
+        "movaps 16(%2, %0), %%xmm1      \n\t"
+        "movaps 32(%2, %0), %%xmm2      \n\t"
+        "movaps 48(%2, %0), %%xmm3      \n\t"
+        "maxps      %%xmm4, %%xmm0      \n\t"
+        "maxps      %%xmm4, %%xmm1      \n\t"
+        "maxps      %%xmm4, %%xmm2      \n\t"
+        "maxps      %%xmm4, %%xmm3      \n\t"
+        "minps      %%xmm5, %%xmm0      \n\t"
+        "minps      %%xmm5, %%xmm1      \n\t"
+        "minps      %%xmm5, %%xmm2      \n\t"
+        "minps      %%xmm5, %%xmm3      \n\t"
+        "movaps     %%xmm0,   (%1, %0)  \n\t"
+        "movaps     %%xmm1, 16(%1, %0)  \n\t"
+        "movaps     %%xmm2, 32(%1, %0)  \n\t"
+        "movaps     %%xmm3, 48(%1, %0)  \n\t"
+        "sub           $64, %0          \n\t"
+        "jge            1b              \n\t"
+        : "+&r" (i)
+        : "r" (dst), "r" (src), "m" (min), "m" (max)
+        : "memory");
+}
+
+#endif /* HAVE_INLINE_ASM */

libavcodec/x86/dsputil.asm

@@ -26,119 +26,6 @@ pb_bswap32: db 3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12
 
 SECTION_TEXT
 
-%macro SCALARPRODUCT 0
-; int ff_scalarproduct_int16(int16_t *v1, int16_t *v2, int order)
-cglobal scalarproduct_int16, 3,3,3, v1, v2, order
-    shl orderq, 1
-    add v1q, orderq
-    add v2q, orderq
-    neg orderq
-    pxor    m2, m2
-.loop:
-    movu    m0, [v1q + orderq]
-    movu    m1, [v1q + orderq + mmsize]
-    pmaddwd m0, [v2q + orderq]
-    pmaddwd m1, [v2q + orderq + mmsize]
-    paddd   m2, m0
-    paddd   m2, m1
-    add     orderq, mmsize*2
-    jl .loop
-%if mmsize == 16
-    movhlps m0, m2
-    paddd   m2, m0
-    pshuflw m0, m2, 0x4e
-%else
-    pshufw  m0, m2, 0x4e
-%endif
-    paddd   m2, m0
-    movd   eax, m2
-    RET
-%endmacro
-
-INIT_MMX mmxext
-SCALARPRODUCT
-INIT_XMM sse2
-SCALARPRODUCT
-
-
-;-----------------------------------------------------------------------------
-; void ff_vector_clip_int32(int32_t *dst, const int32_t *src, int32_t min,
-;                           int32_t max, unsigned int len)
-;-----------------------------------------------------------------------------
-
-; %1 = number of xmm registers used
-; %2 = number of inline load/process/store loops per asm loop
-; %3 = process 4*mmsize (%3=0) or 8*mmsize (%3=1) bytes per loop
-; %4 = CLIPD function takes min/max as float instead of int (CLIPD_SSE2)
-; %5 = suffix
-%macro VECTOR_CLIP_INT32 4-5
-cglobal vector_clip_int32%5, 5,5,%1, dst, src, min, max, len
-%if %4
-    cvtsi2ss  m4, minm
-    cvtsi2ss  m5, maxm
-%else
-    movd      m4, minm
-    movd      m5, maxm
-%endif
-    SPLATD    m4
-    SPLATD    m5
-.loop:
-%assign %%i 1
-%rep %2
-    mova      m0,  [srcq+mmsize*0*%%i]
-    mova      m1,  [srcq+mmsize*1*%%i]
-    mova      m2,  [srcq+mmsize*2*%%i]
-    mova      m3,  [srcq+mmsize*3*%%i]
-%if %3
-    mova      m7,  [srcq+mmsize*4*%%i]
-    mova      m8,  [srcq+mmsize*5*%%i]
-    mova      m9,  [srcq+mmsize*6*%%i]
-    mova      m10, [srcq+mmsize*7*%%i]
-%endif
-    CLIPD  m0,  m4, m5, m6
-    CLIPD  m1,  m4, m5, m6
-    CLIPD  m2,  m4, m5, m6
-    CLIPD  m3,  m4, m5, m6
-%if %3
-    CLIPD  m7,  m4, m5, m6
-    CLIPD  m8,  m4, m5, m6
-    CLIPD  m9,  m4, m5, m6
-    CLIPD  m10, m4, m5, m6
-%endif
-    mova  [dstq+mmsize*0*%%i], m0
-    mova  [dstq+mmsize*1*%%i], m1
-    mova  [dstq+mmsize*2*%%i], m2
-    mova  [dstq+mmsize*3*%%i], m3
-%if %3
-    mova  [dstq+mmsize*4*%%i], m7
-    mova  [dstq+mmsize*5*%%i], m8
-    mova  [dstq+mmsize*6*%%i], m9
-    mova  [dstq+mmsize*7*%%i], m10
-%endif
-%assign %%i %%i+1
-%endrep
-    add     srcq, mmsize*4*(%2+%3)
-    add     dstq, mmsize*4*(%2+%3)
-    sub     lend, mmsize*(%2+%3)
-    jg .loop
-    REP_RET
-%endmacro
-
-INIT_MMX mmx
-%define CLIPD CLIPD_MMX
-VECTOR_CLIP_INT32 0, 1, 0, 0
-INIT_XMM sse2
-VECTOR_CLIP_INT32 6, 1, 0, 0, _int
-%define CLIPD CLIPD_SSE2
-VECTOR_CLIP_INT32 6, 2, 0, 1
-INIT_XMM sse4
-%define CLIPD CLIPD_SSE41
-%ifdef m8
-VECTOR_CLIP_INT32 11, 1, 1, 0
-%else
-VECTOR_CLIP_INT32 6, 1, 0, 0
-%endif
-
 ; %1 = aligned/unaligned
 %macro BSWAP_LOOPS  1
     mov      r3, r2

libavcodec/x86/dsputil_init.c

@@ -26,23 +26,9 @@
 #include "dsputil_x86.h"
 #include "idct_xvid.h"
 
-int32_t ff_scalarproduct_int16_mmxext(const int16_t *v1, const int16_t *v2,
-                                      int order);
-int32_t ff_scalarproduct_int16_sse2(const int16_t *v1, const int16_t *v2,
-                                    int order);
-
 void ff_bswap32_buf_ssse3(uint32_t *dst, const uint32_t *src, int w);
 void ff_bswap32_buf_sse2(uint32_t *dst, const uint32_t *src, int w);
 
-void ff_vector_clip_int32_mmx(int32_t *dst, const int32_t *src,
-                              int32_t min, int32_t max, unsigned int len);
-void ff_vector_clip_int32_sse2(int32_t *dst, const int32_t *src,
-                               int32_t min, int32_t max, unsigned int len);
-void ff_vector_clip_int32_int_sse2(int32_t *dst, const int32_t *src,
-                                   int32_t min, int32_t max, unsigned int len);
-void ff_vector_clip_int32_sse4(int32_t *dst, const int32_t *src,
-                               int32_t min, int32_t max, unsigned int len);
-
 static av_cold void dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx,
                                      int cpu_flags, unsigned high_bit_depth)
 {
@@ -72,10 +58,6 @@ static av_cold void dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx,
 
     c->gmc = ff_gmc_mmx;
 #endif /* HAVE_MMX_INLINE */
-
-#if HAVE_MMX_EXTERNAL
-    c->vector_clip_int32 = ff_vector_clip_int32_mmx;
-#endif /* HAVE_MMX_EXTERNAL */
 }
 
 static av_cold void dsputil_init_mmxext(DSPContext *c, AVCodecContext *avctx,
@@ -88,18 +70,6 @@ static av_cold void dsputil_init_mmxext(DSPContext *c, AVCodecContext *avctx,
         c->idct     = ff_idct_xvid_mmxext;
     }
 #endif /* HAVE_MMXEXT_INLINE */
-
-#if HAVE_MMXEXT_EXTERNAL
-    c->scalarproduct_int16          = ff_scalarproduct_int16_mmxext;
-#endif /* HAVE_MMXEXT_EXTERNAL */
-}
-
-static av_cold void dsputil_init_sse(DSPContext *c, AVCodecContext *avctx,
-                                     int cpu_flags, unsigned high_bit_depth)
-{
-#if HAVE_SSE_INLINE
-    c->vector_clipf = ff_vector_clipf_sse;
-#endif /* HAVE_SSE_INLINE */
 }
 
 static av_cold void dsputil_init_sse2(DSPContext *c, AVCodecContext *avctx,
@@ -115,12 +85,6 @@ static av_cold void dsputil_init_sse2(DSPContext *c, AVCodecContext *avctx,
 #endif /* HAVE_SSE2_INLINE */
 
 #if HAVE_SSE2_EXTERNAL
-    c->scalarproduct_int16          = ff_scalarproduct_int16_sse2;
-    if (cpu_flags & AV_CPU_FLAG_ATOM) {
-        c->vector_clip_int32 = ff_vector_clip_int32_int_sse2;
-    } else {
-        c->vector_clip_int32 = ff_vector_clip_int32_sse2;
-    }
     c->bswap_buf = ff_bswap32_buf_sse2;
 #endif /* HAVE_SSE2_EXTERNAL */
 }
@@ -133,14 +97,6 @@ static av_cold void dsputil_init_ssse3(DSPContext *c, AVCodecContext *avctx,
 #endif /* HAVE_SSSE3_EXTERNAL */
 }
 
-static av_cold void dsputil_init_sse4(DSPContext *c, AVCodecContext *avctx,
-                                      int cpu_flags, unsigned high_bit_depth)
-{
-#if HAVE_SSE4_EXTERNAL
-    c->vector_clip_int32 = ff_vector_clip_int32_sse4;
-#endif /* HAVE_SSE4_EXTERNAL */
-}
-
 av_cold void ff_dsputil_init_x86(DSPContext *c, AVCodecContext *avctx,
                                  unsigned high_bit_depth)
 {
@@ -152,18 +108,12 @@ av_cold void ff_dsputil_init_x86(DSPContext *c, AVCodecContext *avctx,
     if (X86_MMXEXT(cpu_flags))
         dsputil_init_mmxext(c, avctx, cpu_flags, high_bit_depth);
 
-    if (X86_SSE(cpu_flags))
-        dsputil_init_sse(c, avctx, cpu_flags, high_bit_depth);
-
     if (X86_SSE2(cpu_flags))
         dsputil_init_sse2(c, avctx, cpu_flags, high_bit_depth);
 
     if (EXTERNAL_SSSE3(cpu_flags))
         dsputil_init_ssse3(c, avctx, cpu_flags, high_bit_depth);
 
-    if (EXTERNAL_SSE4(cpu_flags))
-        dsputil_init_sse4(c, avctx, cpu_flags, high_bit_depth);
-
     if (CONFIG_ENCODERS)
         ff_dsputilenc_init_mmx(c, avctx, high_bit_depth);
 }

libavcodec/x86/dsputil_mmx.c

@@ -25,7 +25,6 @@
 #include "config.h"
 #include "libavutil/cpu.h"
 #include "libavutil/x86/asm.h"
-#include "constants.h"
 #include "dsputil_x86.h"
 #include "inline_asm.h"
 
@@ -375,37 +374,4 @@ void ff_gmc_mmx(uint8_t *dst, uint8_t *src,
     }
 }
 
-void ff_vector_clipf_sse(float *dst, const float *src,
-                         float min, float max, int len)
-{
-    x86_reg i = (len - 16) * 4;
-    __asm__ volatile (
-        "movss          %3, %%xmm4      \n\t"
-        "movss          %4, %%xmm5      \n\t"
-        "shufps $0, %%xmm4, %%xmm4      \n\t"
-        "shufps $0, %%xmm5, %%xmm5      \n\t"
-        "1:                             \n\t"
-        "movaps   (%2, %0), %%xmm0      \n\t" // 3/1 on intel
-        "movaps 16(%2, %0), %%xmm1      \n\t"
-        "movaps 32(%2, %0), %%xmm2      \n\t"
-        "movaps 48(%2, %0), %%xmm3      \n\t"
-        "maxps      %%xmm4, %%xmm0      \n\t"
-        "maxps      %%xmm4, %%xmm1      \n\t"
-        "maxps      %%xmm4, %%xmm2      \n\t"
-        "maxps      %%xmm4, %%xmm3      \n\t"
-        "minps      %%xmm5, %%xmm0      \n\t"
-        "minps      %%xmm5, %%xmm1      \n\t"
-        "minps      %%xmm5, %%xmm2      \n\t"
-        "minps      %%xmm5, %%xmm3      \n\t"
-        "movaps     %%xmm0,   (%1, %0)  \n\t"
-        "movaps     %%xmm1, 16(%1, %0)  \n\t"
-        "movaps     %%xmm2, 32(%1, %0)  \n\t"
-        "movaps     %%xmm3, 48(%1, %0)  \n\t"
-        "sub           $64, %0          \n\t"
-        "jge            1b              \n\t"
-        : "+&r" (i)
-        : "r" (dst), "r" (src), "m" (min), "m" (max)
-        : "memory");
-}
-
 #endif /* HAVE_INLINE_ASM */

libavcodec/x86/dsputil_x86.h

@@ -46,7 +46,4 @@ void ff_gmc_mmx(uint8_t *dst, uint8_t *src,
                 int dxx, int dxy, int dyx, int dyy,
                 int shift, int r, int width, int height);
 
-void ff_vector_clipf_sse(float *dst, const float *src,
-                         float min, float max, int len);
-
 #endif /* AVCODEC_X86_DSPUTIL_X86_H */