Move lpc_compute_autocorr() from DSPContext to a new struct LPCContext.

Signed-off-by: Mans Rullgard <mans@mansr.com>

12 files changed, 63 insertions, 43 deletions

diff --git a/libavcodec/alacenc.c b/libavcodec/alacenc.c
index d1369c4859..5706ae9f4d 100644
--- a/libavcodec/alacenc.c
+++ b/libavcodec/alacenc.c
@@ -51,11 +51,11 @@ typedef struct RiceContext {
     int rice_modifier;
 } RiceContext;
 
-typedef struct LPCContext {
+typedef struct AlacLPCContext {
     int lpc_order;
     int lpc_coeff[ALAC_MAX_LPC_ORDER+1];
     int lpc_quant;
-} LPCContext;
+} AlacLPCContext;
 
 typedef struct AlacEncodeContext {
     int compression_level;
@@ -69,8 +69,8 @@ typedef struct AlacEncodeContext {
     int interlacing_leftweight;
     PutBitContext pbctx;
     RiceContext rc;
-    LPCContext lpc[MAX_CHANNELS];
-    DSPContext dspctx;
+    AlacLPCContext lpc[MAX_CHANNELS];
+    LPCContext lpc_ctx;
     AVCodecContext *avctx;
 } AlacEncodeContext;
 
@@ -141,7 +141,7 @@ static void calc_predictor_params(AlacEncodeContext *s, int ch)
         s->lpc[ch].lpc_coeff[4] =   80;
         s->lpc[ch].lpc_coeff[5] =  -25;
     } else {
-        opt_order = ff_lpc_calc_coefs(&s->dspctx, s->sample_buf[ch],
+        opt_order = ff_lpc_calc_coefs(&s->lpc_ctx, s->sample_buf[ch],
                                       s->avctx->frame_size,
                                       s->min_prediction_order,
                                       s->max_prediction_order,
@@ -237,7 +237,7 @@ static void alac_stereo_decorrelation(AlacEncodeContext *s)
 static void alac_linear_predictor(AlacEncodeContext *s, int ch)
 {
     int i;
-    LPCContext lpc = s->lpc[ch];
+    AlacLPCContext lpc = s->lpc[ch];
 
     if(lpc.lpc_order == 31) {
         s->predictor_buf[0] = s->sample_buf[ch][0];
@@ -455,7 +455,7 @@ static av_cold int alac_encode_init(AVCodecContext *avctx)
     avctx->coded_frame->key_frame = 1;
 
     s->avctx = avctx;
-    dsputil_init(&s->dspctx, avctx);
+    ff_lpc_init(&s->lpc_ctx);
 
     return 0;
 }
diff --git a/libavcodec/arm/asm-offsets.h b/libavcodec/arm/asm-offsets.h
index 174b5aadcb..f616f808f1 100644
--- a/libavcodec/arm/asm-offsets.h
+++ b/libavcodec/arm/asm-offsets.h
@@ -33,16 +33,16 @@
 #define Y_DC_SCALE               0xab4
 #define C_DC_SCALE               0xab8
 #define AC_PRED                  0xae0
-#define BLOCK_LAST_INDEX         0x21c0
-#define INTER_SCANTAB_RASTER_END 0x23c0
-#define H263_AIC                 0x2670
+#define BLOCK_LAST_INDEX         0x21bc
+#define INTER_SCANTAB_RASTER_END 0x23bc
+#define H263_AIC                 0x2668
 #elif defined(__APPLE__)
 #define Y_DC_SCALE               0xa70
 #define C_DC_SCALE               0xa74
 #define AC_PRED                  0xa9c
-#define BLOCK_LAST_INDEX         0x217c
-#define INTER_SCANTAB_RASTER_END 0x237c
-#define H263_AIC                 0x2620
+#define BLOCK_LAST_INDEX         0x2178
+#define INTER_SCANTAB_RASTER_END 0x2378
+#define H263_AIC                 0x261c
 #endif
 
 #endif
diff --git a/libavcodec/dsputil.c b/libavcodec/dsputil.c
index cc60524107..57b2640686 100644
--- a/libavcodec/dsputil.c
+++ b/libavcodec/dsputil.c
@@ -36,7 +36,6 @@
 #include "mathops.h"
 #include "mpegvideo.h"
 #include "config.h"
-#include "lpc.h"
 #include "ac3dec.h"
 #include "vorbis.h"
 #include "png.h"
@@ -4431,9 +4430,6 @@ av_cold void dsputil_init(DSPContext* c, AVCodecContext *avctx)
 #if CONFIG_AC3_DECODER
     c->ac3_downmix = ff_ac3_downmix_c;
 #endif
-#if CONFIG_LPC
-    c->lpc_compute_autocorr = ff_lpc_compute_autocorr;
-#endif
     c->vector_fmul = vector_fmul_c;
     c->vector_fmul_reverse = vector_fmul_reverse_c;
     c->vector_fmul_add = vector_fmul_add_c;
diff --git a/libavcodec/dsputil.h b/libavcodec/dsputil.h
index 0efbad918a..c619b78529 100644
--- a/libavcodec/dsputil.h
+++ b/libavcodec/dsputil.h
@@ -374,8 +374,6 @@ typedef struct DSPContext {
     /* assume len is a multiple of 4, and arrays are 16-byte aligned */
     void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
     void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
-    /* no alignment needed */
-    void (*lpc_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);
     /* assume len is a multiple of 8, and arrays are 16-byte aligned */
     void (*vector_fmul)(float *dst, const float *src, int len);
     void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
diff --git a/libavcodec/flacenc.c b/libavcodec/flacenc.c
index 272d446b29..d6f0f87376 100644
--- a/libavcodec/flacenc.c
+++ b/libavcodec/flacenc.c
@@ -23,7 +23,6 @@
 #include "libavutil/md5.h"
 #include "avcodec.h"
 #include "get_bits.h"
-#include "dsputil.h"
 #include "golomb.h"
 #include "lpc.h"
 #include "flac.h"
@@ -95,7 +94,7 @@ typedef struct FlacEncodeContext {
     FlacFrame frame;
     CompressionOptions options;
     AVCodecContext *avctx;
-    DSPContext dsp;
+    LPCContext lpc_ctx;
     struct AVMD5 *md5ctx;
 } FlacEncodeContext;
 
@@ -217,7 +216,7 @@ static av_cold int flac_encode_init(AVCodecContext *avctx)
 
     s->avctx = avctx;
 
-    dsputil_init(&s->dsp, avctx);
+    ff_lpc_init(&s->lpc_ctx);
 
     if (avctx->sample_fmt != AV_SAMPLE_FMT_S16)
         return -1;
@@ -902,7 +901,7 @@ static int encode_residual_ch(FlacEncodeContext *s, int ch)
 
     /* LPC */
     sub->type = FLAC_SUBFRAME_LPC;
-    opt_order = ff_lpc_calc_coefs(&s->dsp, smp, n, min_order, max_order,
+    opt_order = ff_lpc_calc_coefs(&s->lpc_ctx, smp, n, min_order, max_order,
                                   s->options.lpc_coeff_precision, coefs, shift, s->options.lpc_type,
                                   s->options.lpc_passes, omethod,
                                   MAX_LPC_SHIFT, 0);
diff --git a/libavcodec/lpc.c b/libavcodec/lpc.c
index 112a78d4b9..3a93c9f673 100644
--- a/libavcodec/lpc.c
+++ b/libavcodec/lpc.c
@@ -20,7 +20,6 @@
  */
 
 #include "libavutil/lls.h"
-#include "dsputil.h"
 
 #define LPC_USE_DOUBLE
 #include "lpc.h"
@@ -55,7 +54,7 @@ static void apply_welch_window(const int32_t *data, int len, double *w_data)
  * Calculate autocorrelation data from audio samples
  * A Welch window function is applied before calculation.
  */
-void ff_lpc_compute_autocorr(const int32_t *data, int len, int lag,
+static void lpc_compute_autocorr_c(const int32_t *data, int len, int lag,
                              double *autoc)
 {
     int i, j;
@@ -162,7 +161,7 @@ static int estimate_best_order(double *ref, int min_order, int max_order)
  * 1  = LPC with coeffs determined by Levinson-Durbin recursion
  * 2+ = LPC with coeffs determined by Cholesky factorization using (use_lpc-1) passes.
  */
-int ff_lpc_calc_coefs(DSPContext *s,
+int ff_lpc_calc_coefs(LPCContext *s,
                       const int32_t *samples, int blocksize, int min_order,
                       int max_order, int precision,
                       int32_t coefs[][MAX_LPC_ORDER], int *shift,
@@ -236,3 +235,11 @@ int ff_lpc_calc_coefs(DSPContext *s,
 
     return opt_order;
 }
+
+av_cold void ff_lpc_init(LPCContext *s)
+{
+    s->lpc_compute_autocorr = lpc_compute_autocorr_c;
+
+    if (HAVE_MMX)
+        ff_lpc_init_x86(s);
+}
diff --git a/libavcodec/lpc.h b/libavcodec/lpc.h
index 1c595f6b9a..a307793374 100644
--- a/libavcodec/lpc.h
+++ b/libavcodec/lpc.h
@@ -36,18 +36,36 @@
 #define MAX_LPC_ORDER       32
 
 
+typedef struct LPCContext {
+    /**
+     * Perform autocorrelation on input samples with delay of 0 to lag.
+     * @param data  input samples.
+     *              no alignment needed.
+     * @param len   number of input samples to process
+     * @param lag   maximum delay to calculate
+     * @param autoc output autocorrelation coefficients.
+     *              constraints: array size must be at least lag+1.
+     */
+    void (*lpc_compute_autocorr)(const int32_t *data, int len, int lag,
+                                 double *autoc);
+} LPCContext;
+
+
 /**
  * Calculate LPC coefficients for multiple orders
  */
-int ff_lpc_calc_coefs(DSPContext *s,
+int ff_lpc_calc_coefs(LPCContext *s,
                       const int32_t *samples, int blocksize, int min_order,
                       int max_order, int precision,
                       int32_t coefs[][MAX_LPC_ORDER], int *shift,
                       enum AVLPCType lpc_type, int lpc_passes,
                       int omethod, int max_shift, int zero_shift);
 
-void ff_lpc_compute_autocorr(const int32_t *data, int len, int lag,
-                             double *autoc);
+/**
+ * Initialize LPCContext.
+ */
+void ff_lpc_init(LPCContext *s);
+void ff_lpc_init_x86(LPCContext *s);
 
 #ifdef LPC_USE_DOUBLE
 #define LPC_TYPE double
diff --git a/libavcodec/ra144.h b/libavcodec/ra144.h
index 536b5bbe72..11c66a46e0 100644
--- a/libavcodec/ra144.h
+++ b/libavcodec/ra144.h
@@ -23,7 +23,7 @@
 #define AVCODEC_RA144_H
 
 #include <stdint.h>
-#include "dsputil.h"
+#include "lpc.h"
 
 #define NBLOCKS         4       ///< number of subblocks within a block
 #define BLOCKSIZE       40      ///< subblock size in 16-bit words
@@ -34,7 +34,7 @@
 
 typedef struct {
     AVCodecContext *avctx;
-    DSPContext dsp;
+    LPCContext lpc_ctx;
 
     unsigned int     old_energy;        ///< previous frame energy
 
diff --git a/libavcodec/ra144enc.c b/libavcodec/ra144enc.c
index 9865dc9c04..3f8694eb8f 100644
--- a/libavcodec/ra144enc.c
+++ b/libavcodec/ra144enc.c
@@ -29,7 +29,6 @@
 
 #include "avcodec.h"
 #include "put_bits.h"
-#include "lpc.h"
 #include "celp_filters.h"
 #include "ra144.h"
 
@@ -53,7 +52,7 @@ static av_cold int ra144_encode_init(AVCodecContext * avctx)
     ractx->lpc_coef[0] = ractx->lpc_tables[0];
     ractx->lpc_coef[1] = ractx->lpc_tables[1];
     ractx->avctx = avctx;
-    dsputil_init(&ractx->dsp, avctx);
+    ff_lpc_init(&ractx->lpc_ctx);
     return 0;
 }
 
@@ -451,7 +450,7 @@ static int ra144_encode_frame(AVCodecContext *avctx, uint8_t *frame,
     energy = ff_energy_tab[quantize(ff_t_sqrt(energy >> 5) >> 10, ff_energy_tab,
                                     32)];
 
-    ff_lpc_calc_coefs(&ractx->dsp, lpc_data, NBLOCKS * BLOCKSIZE, LPC_ORDER,
+    ff_lpc_calc_coefs(&ractx->lpc_ctx, lpc_data, NBLOCKS * BLOCKSIZE, LPC_ORDER,
                       LPC_ORDER, 16, lpc_coefs, shift, AV_LPC_TYPE_LEVINSON,
                       0, ORDER_METHOD_EST, 12, 0);
     for (i = 0; i < LPC_ORDER; i++)
diff --git a/libavcodec/x86/dsputil_mmx.h b/libavcodec/x86/dsputil_mmx.h
index c57bbbf3a7..a095e1ef3d 100644
--- a/libavcodec/x86/dsputil_mmx.h
+++ b/libavcodec/x86/dsputil_mmx.h
@@ -200,9 +200,6 @@ void ff_vc1dsp_init_mmx(DSPContext* dsp, AVCodecContext *avctx);
 void ff_put_vc1_mspel_mc00_mmx(uint8_t *dst, const uint8_t *src, int stride, int rnd);
 void ff_avg_vc1_mspel_mc00_mmx2(uint8_t *dst, const uint8_t *src, int stride, int rnd);
 
-void ff_lpc_compute_autocorr_sse2(const int32_t *data, int len, int lag,
-                                   double *autoc);
-
 void ff_mmx_idct(DCTELEM *block);
 void ff_mmxext_idct(DCTELEM *block);
 
diff --git a/libavcodec/x86/dsputilenc_mmx.c b/libavcodec/x86/dsputilenc_mmx.c
index cb8080aa0a..bd31205a6b 100644
--- a/libavcodec/x86/dsputilenc_mmx.c
+++ b/libavcodec/x86/dsputilenc_mmx.c
@@ -1166,10 +1166,6 @@ void dsputilenc_init_mmx(DSPContext* c, AVCodecContext *avctx)
 #endif
         }
 
-        if (CONFIG_LPC && mm_flags & (AV_CPU_FLAG_SSE2|AV_CPU_FLAG_SSE2SLOW)) {
-            c->lpc_compute_autocorr = ff_lpc_compute_autocorr_sse2;
-        }
-
 #if HAVE_SSSE3
         if(mm_flags & AV_CPU_FLAG_SSSE3){
             if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
diff --git a/libavcodec/x86/lpc_mmx.c b/libavcodec/x86/lpc_mmx.c
index 49eb569eea..19aad9860f 100644
--- a/libavcodec/x86/lpc_mmx.c
+++ b/libavcodec/x86/lpc_mmx.c
@@ -20,7 +20,8 @@
  */
 
 #include "libavutil/x86_cpu.h"
-#include "dsputil_mmx.h"
+#include "libavutil/cpu.h"
+#include "libavcodec/lpc.h"
 
 static void apply_welch_window_sse2(const int32_t *data, int len, double *w_data)
 {
@@ -68,7 +69,7 @@ static void apply_welch_window_sse2(const int32_t *data, int len, double *w_data
 #undef WELCH
 }
 
-void ff_lpc_compute_autocorr_sse2(const int32_t *data, int len, int lag,
+static void lpc_compute_autocorr_sse2(const int32_t *data, int len, int lag,
                                    double *autoc)
 {
     double tmp[len + lag + 2];
@@ -141,3 +142,12 @@ void ff_lpc_compute_autocorr_sse2(const int32_t *data, int len, int lag,
         }
     }
 }
+
+av_cold void ff_lpc_init_x86(LPCContext *c)
+{
+    int mm_flags = av_get_cpu_flags();
+
+    if (mm_flags & (AV_CPU_FLAG_SSE2|AV_CPU_FLAG_SSE2SLOW)) {
+        c->lpc_compute_autocorr = lpc_compute_autocorr_sse2;
+    }
+}