]> git.sesse.net Git - ffmpeg/blobdiff - libavcodec/alac.c
Fix issue 123 (Support Magic-carpet fli version,
[ffmpeg] / libavcodec / alac.c
index 3dd0ab9edb131e9b40f30c6c46c2e1ff2f68a8cc..777c720f34753be4a5026c3227c54a64807ccc44 100644 (file)
@@ -55,6 +55,7 @@
 #include "avcodec.h"
 #include "bitstream.h"
 #include "bytestream.h"
+#include "unary.h"
 
 #define ALAC_EXTRADATA_SIZE 36
 #define MAX_CHANNELS 2
@@ -139,17 +140,9 @@ static int alac_set_info(ALACContext *alac)
     return 0;
 }
 
-/* hideously inefficient. could use a bitmask search,
- * alternatively bsr on x86,
- */
-static int count_leading_zeros(int32_t input)
+static inline int count_leading_zeros(int32_t input)
 {
-    int i = 0;
-    while (!(0x80000000 & input) && i < 32) {
-        i++;
-        input = input << 1;
-    }
-    return i;
+    return 31-av_log2(input);
 }
 
 static void bastardized_rice_decompress(ALACContext *alac,
@@ -167,18 +160,15 @@ static void bastardized_rice_decompress(ALACContext *alac,
     int sign_modifier = 0;
 
     for (output_count = 0; output_count < output_size; output_count++) {
-        int32_t x = 0;
+        int32_t x;
         int32_t x_modified;
         int32_t final_val;
 
         /* read x - number of 1s before 0 represent the rice */
-        while (x <= 8 && get_bits1(&alac->gb)) {
-            x++;
-        }
-
+        x = get_unary_0_9(&alac->gb);
 
         if (x > 8) { /* RICE THRESHOLD */
-          /* use alternative encoding */
+            /* use alternative encoding */
             int32_t value;
 
             value = get_bits(&alac->gb, readsamplesize);
@@ -189,7 +179,7 @@ static void bastardized_rice_decompress(ALACContext *alac,
 
             x = value;
         } else {
-          /* standard rice encoding */
+            /* standard rice encoding */
             int extrabits;
             int k; /* size of extra bits */
 
@@ -209,10 +199,9 @@ static void bastardized_rice_decompress(ALACContext *alac,
 
                 if (extrabits > 1) {
                     x += extrabits - 1;
-                    get_bits(&alac->gb, k);
-                } else {
-                    get_bits(&alac->gb, k - 1);
-                }
+                    skip_bits(&alac->gb, k);
+                } else
+                    skip_bits(&alac->gb, k - 1);
             }
         }
 
@@ -225,8 +214,8 @@ static void bastardized_rice_decompress(ALACContext *alac,
         sign_modifier = 0;
 
         /* now update the history */
-        history += (x_modified * rice_historymult)
-                 - ((history * rice_historymult) >> 9);
+        history += x_modified * rice_historymult
+                   - ((history * rice_historymult) >> 9);
 
         if (x_modified > 0xffff)
             history = 0xffff;
@@ -237,10 +226,7 @@ static void bastardized_rice_decompress(ALACContext *alac,
 
             sign_modifier = 1;
 
-            x = 0;
-            while (x <= 8 && get_bits1(&alac->gb)) {
-                x++;
-            }
+            x = get_unary_0_9(&alac->gb);
 
             if (x > 8) {
                 block_size = get_bits(&alac->gb, 16);
@@ -259,16 +245,15 @@ static void bastardized_rice_decompress(ALACContext *alac,
                 if (extrabits < 2) {
                     x = 1 - extrabits;
                     block_size += x;
-                    get_bits(&alac->gb, k - 1);
+                    skip_bits(&alac->gb, k - 1);
                 } else {
-                    get_bits(&alac->gb, k);
+                    skip_bits(&alac->gb, k);
                 }
             }
 
             if (block_size > 0) {
                 memset(&output_buffer[output_count+1], 0, block_size * 4);
                 output_count += block_size;
-
             }
 
             if (block_size > 0xffff)
@@ -279,12 +264,15 @@ static void bastardized_rice_decompress(ALACContext *alac,
     }
 }
 
-#define SIGN_EXTENDED32(val, bits) ((val << (32 - bits)) >> (32 - bits))
+static inline int32_t extend_sign32(int32_t val, int bits)
+{
+    return (val << (32 - bits)) >> (32 - bits);
+}
 
-#define SIGN_ONLY(v) \
-                     ((v < 0) ? (-1) : \
-                                ((v > 0) ? (1) : \
-                                           (0)))
+static inline int sign_only(int v)
+{
+    return v ? FFSIGN(v) : 0;
+}
 
 static void predictor_decompress_fir_adapt(int32_t *error_buffer,
                                            int32_t *buffer_out,
@@ -300,7 +288,9 @@ static void predictor_decompress_fir_adapt(int32_t *error_buffer,
     *buffer_out = *error_buffer;
 
     if (!predictor_coef_num) {
-        if (output_size <= 1) return;
+        if (output_size <= 1)
+            return;
+
         memcpy(buffer_out+1, error_buffer+1, (output_size-1) * 4);
         return;
     }
@@ -309,31 +299,29 @@ static void predictor_decompress_fir_adapt(int32_t *error_buffer,
       /* second-best case scenario for fir decompression,
        * error describes a small difference from the previous sample only
        */
-        if (output_size <= 1) return;
+        if (output_size <= 1)
+            return;
         for (i = 0; i < output_size - 1; i++) {
             int32_t prev_value;
             int32_t error_value;
 
             prev_value = buffer_out[i];
             error_value = error_buffer[i+1];
-            buffer_out[i+1] = SIGN_EXTENDED32((prev_value + error_value), readsamplesize);
+            buffer_out[i+1] =
+                extend_sign32((prev_value + error_value), readsamplesize);
         }
         return;
     }
 
     /* read warm-up samples */
-    if (predictor_coef_num > 0) {
-        int i;
+    if (predictor_coef_num > 0)
         for (i = 0; i < predictor_coef_num; i++) {
             int32_t val;
 
             val = buffer_out[i] + error_buffer[i+1];
-
-            val = SIGN_EXTENDED32(val, readsamplesize);
-
+            val = extend_sign32(val, readsamplesize);
             buffer_out[i+1] = val;
         }
-    }
 
 #if 0
     /* 4 and 8 are very common cases (the only ones i've seen). these
@@ -350,12 +338,9 @@ static void predictor_decompress_fir_adapt(int32_t *error_buffer,
     }
 #endif
 
-
     /* general case */
     if (predictor_coef_num > 0) {
-        for (i = predictor_coef_num + 1;
-             i < output_size;
-             i++) {
+        for (i = predictor_coef_num + 1; i < output_size; i++) {
             int j;
             int sum = 0;
             int outval;
@@ -369,7 +354,7 @@ static void predictor_decompress_fir_adapt(int32_t *error_buffer,
             outval = (1 << (predictor_quantitization-1)) + sum;
             outval = outval >> predictor_quantitization;
             outval = outval + buffer_out[0] + error_val;
-            outval = SIGN_EXTENDED32(outval, readsamplesize);
+            outval = extend_sign32(outval, readsamplesize);
 
             buffer_out[predictor_coef_num+1] = outval;
 
@@ -378,7 +363,7 @@ static void predictor_decompress_fir_adapt(int32_t *error_buffer,
 
                 while (predictor_num >= 0 && error_val > 0) {
                     int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num];
-                    int sign = SIGN_ONLY(val);
+                    int sign = sign_only(val);
 
                     predictor_coef_table[predictor_num] -= sign;
 
@@ -394,7 +379,7 @@ static void predictor_decompress_fir_adapt(int32_t *error_buffer,
 
                 while (predictor_num >= 0 && error_val < 0) {
                     int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num];
-                    int sign = - SIGN_ONLY(val);
+                    int sign = - sign_only(val);
 
                     predictor_coef_table[predictor_num] -= sign;
 
@@ -412,32 +397,29 @@ static void predictor_decompress_fir_adapt(int32_t *error_buffer,
     }
 }
 
-static void deinterlace_16(int32_t *buffer_a, int32_t *buffer_b,
-                    int16_t *buffer_out,
-                    int numchannels, int numsamples,
-                    uint8_t interlacing_shift,
-                    uint8_t interlacing_leftweight)
+static void reconstruct_stereo_16(int32_t *buffer[MAX_CHANNELS],
+                                  int16_t *buffer_out,
+                                  int numchannels, int numsamples,
+                                  uint8_t interlacing_shift,
+                                  uint8_t interlacing_leftweight)
 {
     int i;
-    if (numsamples <= 0) return;
+    if (numsamples <= 0)
+        return;
 
     /* weighted interlacing */
     if (interlacing_leftweight) {
         for (i = 0; i < numsamples; i++) {
-            int32_t difference, midright;
-            int16_t left;
-            int16_t right;
+            int32_t a, b;
 
-            midright = buffer_a[i];
-            difference = buffer_b[i];
+            a = buffer[0][i];
+            b = buffer[1][i];
 
+            a -= (b * interlacing_leftweight) >> interlacing_shift;
+            b += a;
 
-            right = midright - ((difference * interlacing_leftweight) >> interlacing_shift);
-            left = (midright - ((difference * interlacing_leftweight) >> interlacing_shift))
-                 + difference;
-
-            buffer_out[i*numchannels] = left;
-            buffer_out[i*numchannels + 1] = right;
+            buffer_out[i*numchannels] = b;
+            buffer_out[i*numchannels + 1] = a;
         }
 
         return;
@@ -447,8 +429,8 @@ static void deinterlace_16(int32_t *buffer_a, int32_t *buffer_b,
     for (i = 0; i < numsamples; i++) {
         int16_t left, right;
 
-        left = buffer_a[i];
-        right = buffer_b[i];
+        left = buffer[0][i];
+        right = buffer[1][i];
 
         buffer_out[i*numchannels] = left;
         buffer_out[i*numchannels + 1] = right;
@@ -491,44 +473,49 @@ static int alac_decode_frame(AVCodecContext *avctx,
     init_get_bits(&alac->gb, inbuffer, input_buffer_size * 8);
 
     channels = get_bits(&alac->gb, 3) + 1;
+    if (channels > MAX_CHANNELS) {
+        av_log(avctx, AV_LOG_ERROR, "channels > %d not supported\n",
+               MAX_CHANNELS);
+        return input_buffer_size;
+    }
 
-        /* 2^result = something to do with output waiting.
-         * perhaps matters if we read > 1 frame in a pass?
-         */
-        get_bits(&alac->gb, 4);
-
-        get_bits(&alac->gb, 12); /* unknown, skip 12 bits */
+    /* 2^result = something to do with output waiting.
+     * perhaps matters if we read > 1 frame in a pass?
+     */
+    skip_bits(&alac->gb, 4);
 
-        hassize = get_bits(&alac->gb, 1); /* the output sample size is stored soon */
+    skip_bits(&alac->gb, 12); /* unknown, skip 12 bits */
 
-        wasted_bytes = get_bits(&alac->gb, 2); /* unknown ? */
+    /* the output sample size is stored soon */
+    hassize = get_bits1(&alac->gb);
 
-        isnotcompressed = get_bits(&alac->gb, 1); /* whether the frame is compressed */
+    wasted_bytes = get_bits(&alac->gb, 2); /* unknown ? */
 
-        if (hassize) {
-            /* now read the number of samples,
-             * as a 32bit integer */
-            outputsamples = get_bits(&alac->gb, 32);
-        } else
-            outputsamples = alac->setinfo_max_samples_per_frame;
+    /* whether the frame is compressed */
+    isnotcompressed = get_bits1(&alac->gb);
 
-        *outputsize = outputsamples * alac->bytespersample;
-        readsamplesize = alac->setinfo_sample_size - (wasted_bytes * 8) + channels - 1;
+    if (hassize) {
+        /* now read the number of samples as a 32bit integer */
+        outputsamples = get_bits(&alac->gb, 32);
+    } else
+        outputsamples = alac->setinfo_max_samples_per_frame;
 
-        if (!isnotcompressed) {
-         /* so it is compressed */
-            int16_t predictor_coef_table[channels][32];
-            int predictor_coef_num[channels];
-            int prediction_type[channels];
-            int prediction_quantitization[channels];
-            int ricemodifier[channels];
+    *outputsize = outputsamples * alac->bytespersample;
+    readsamplesize = alac->setinfo_sample_size - (wasted_bytes * 8) + channels - 1;
 
-            int i, chan;
+    if (!isnotcompressed) {
+        /* so it is compressed */
+        int16_t predictor_coef_table[channels][32];
+        int predictor_coef_num[channels];
+        int prediction_type[channels];
+        int prediction_quantitization[channels];
+        int ricemodifier[channels];
+        int i, chan;
 
-            interlacing_shift = get_bits(&alac->gb, 8);
-            interlacing_leftweight = get_bits(&alac->gb, 8);
+        interlacing_shift = get_bits(&alac->gb, 8);
+        interlacing_leftweight = get_bits(&alac->gb, 8);
 
-          for (chan = 0; chan < channels; chan++) {
+        for (chan = 0; chan < channels; chan++) {
             prediction_type[chan] = get_bits(&alac->gb, 4);
             prediction_quantitization[chan] = get_bits(&alac->gb, 4);
 
@@ -536,16 +523,14 @@ static int alac_decode_frame(AVCodecContext *avctx,
             predictor_coef_num[chan] = get_bits(&alac->gb, 5);
 
             /* read the predictor table */
-            for (i = 0; i < predictor_coef_num[chan]; i++) {
+            for (i = 0; i < predictor_coef_num[chan]; i++)
                 predictor_coef_table[chan][i] = (int16_t)get_bits(&alac->gb, 16);
-            }
-          }
+        }
 
-            if (wasted_bytes) {
-                av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented, unhandling of wasted_bytes\n");
-            }
+        if (wasted_bytes)
+            av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented, unhandling of wasted_bytes\n");
 
-          for (chan = 0; chan < channels; chan++) {
+        for (chan = 0; chan < channels; chan++) {
             bastardized_rice_decompress(alac,
                                         alac->predicterror_buffer[chan],
                                         outputsamples,
@@ -556,7 +541,7 @@ static int alac_decode_frame(AVCodecContext *avctx,
                                         (1 << alac->setinfo_rice_kmodifier) - 1);
 
             if (prediction_type[chan] == 0) {
-              /* adaptive fir */
+                /* adaptive fir */
                 predictor_decompress_fir_adapt(alac->predicterror_buffer[chan],
                                                alac->outputsamples_buffer[chan],
                                                outputsamples,
@@ -566,31 +551,30 @@ static int alac_decode_frame(AVCodecContext *avctx,
                                                prediction_quantitization[chan]);
             } else {
                 av_log(avctx, AV_LOG_ERROR, "FIXME: unhandled prediction type: %i\n", prediction_type[chan]);
-                /* i think the only other prediction type (or perhaps this is just a
-                 * boolean?) runs adaptive fir twice.. like:
+                /* I think the only other prediction type (or perhaps this is
+                 * just a boolean?) runs adaptive fir twice.. like:
                  * predictor_decompress_fir_adapt(predictor_error, tempout, ...)
                  * predictor_decompress_fir_adapt(predictor_error, outputsamples ...)
                  * little strange..
                  */
             }
-          }
-        } else {
-         /* not compressed, easy case */
-            if (alac->setinfo_sample_size <= 16) {
-                int i, chan;
-              for (chan = 0; chan < channels; chan++) {
+        }
+    } else {
+        /* not compressed, easy case */
+        if (alac->setinfo_sample_size <= 16) {
+            int i, chan;
+            for (chan = 0; chan < channels; chan++)
                 for (i = 0; i < outputsamples; i++) {
                     int32_t audiobits;
 
                     audiobits = get_bits(&alac->gb, alac->setinfo_sample_size);
-                    audiobits = SIGN_EXTENDED32(audiobits, readsamplesize);
+                    audiobits = extend_sign32(audiobits, readsamplesize);
 
                     alac->outputsamples_buffer[chan][i] = audiobits;
                 }
-              }
-            } else {
-                int i, chan;
-              for (chan = 0; chan < channels; chan++) {
+        } else {
+            int i, chan;
+            for (chan = 0; chan < channels; chan++)
                 for (i = 0; i < outputsamples; i++) {
                     int32_t audiobits;
 
@@ -603,41 +587,39 @@ static int alac_decode_frame(AVCodecContext *avctx,
 
                     alac->outputsamples_buffer[chan][i] = audiobits;
                 }
-              }
-            }
-            /* wasted_bytes = 0; */
-            interlacing_shift = 0;
-            interlacing_leftweight = 0;
         }
+        /* wasted_bytes = 0; */
+        interlacing_shift = 0;
+        interlacing_leftweight = 0;
+    }
 
-        switch(alac->setinfo_sample_size) {
-        case 16: {
-          if (channels == 2) {
-            deinterlace_16(alac->outputsamples_buffer[0],
-                           alac->outputsamples_buffer[1],
-                           (int16_t*)outbuffer,
-                           alac->numchannels,
-                           outputsamples,
-                           interlacing_shift,
-                           interlacing_leftweight);
-          } else {
-              int i;
-              for (i = 0; i < outputsamples; i++) {
-                  int16_t sample = alac->outputsamples_buffer[0][i];
-                  ((int16_t*)outbuffer)[i * alac->numchannels] = sample;
-              }
-          }
-            break;
-        }
-        case 20:
-        case 24:
-        case 32:
-            av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented sample size %i\n", alac->setinfo_sample_size);
-            break;
-        default:
-            break;
+    switch(alac->setinfo_sample_size) {
+    case 16:
+        if (channels == 2) {
+            reconstruct_stereo_16(alac->outputsamples_buffer,
+                                  (int16_t*)outbuffer,
+                                  alac->numchannels,
+                                  outputsamples,
+                                  interlacing_shift,
+                                  interlacing_leftweight);
+        } else {
+            int i;
+            for (i = 0; i < outputsamples; i++) {
+                int16_t sample = alac->outputsamples_buffer[0][i];
+                ((int16_t*)outbuffer)[i * alac->numchannels] = sample;
+            }
         }
-
+        break;
+    case 20:
+    case 24:
+        // It is not clear if there exist any encoder that creates 24 bit ALAC
+        // files. iTunes convert 24 bit raw files to 16 bit before encoding.
+    case 32:
+        av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented sample size %i\n", alac->setinfo_sample_size);
+        break;
+    default:
+        break;
+    }
 
     return input_buffer_size;
 }