BitTrace: if BITTRACE is > 0, record and print syntax bits used

* Bittrace decoding is ~3x slower.
* Binary is the same byte-wise if BITTRACE=0
* Example output:
=== Bit traces ===
global-header    :    174 bytes   	[ 0.69%] [count:    1850]
segments         :    246 bytes   	[ 0.98%] [count:    3072]
block-size       :    170 bytes   	[ 0.68%] [count:    1536]
pred-modes       :   3829 bytes   	[15.27%] [count:   51458]
pred-modes-uv    :    279 bytes   	[ 1.11%] [count:    2329]
coeffs           :  20370 bytes   	[81.27%] [count:  212914]
Total: 25065 bytes

Change-Id: Ie32569c4e54a7ec13264e68d2dae2ce45c8536cb

7 files changed, 203 insertions, 86 deletions

diff --git a/makefile.unix b/makefile.unix
index 87b6b653..f98e7662 100644
--- a/makefile.unix
+++ b/makefile.unix
@@ -118,6 +118,10 @@ GROFF = /usr/bin/groff
 COL = /usr/bin/col
 LDFLAGS = $(EXTRA_LIBS) $(EXTRA_FLAGS) -lm
 
+ifdef BITTRACE
+CFLAGS += -DBITTRACE=$(BITTRACE)
+endif
+
 ANIM_UTIL_OBJS = \
     examples/anim_util.o \
 
diff --git a/src/dec/quant_dec.c b/src/dec/quant_dec.c
index f07212ad..a0ac018b 100644
--- a/src/dec/quant_dec.c
+++ b/src/dec/quant_dec.c
@@ -61,12 +61,17 @@ static const uint16_t kAcTable[128] = {
 
 void VP8ParseQuant(VP8Decoder* const dec) {
   VP8BitReader* const br = &dec->br_;
-  const int base_q0 = VP8GetValue(br, 7);
-  const int dqy1_dc = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0;
-  const int dqy2_dc = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0;
-  const int dqy2_ac = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0;
-  const int dquv_dc = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0;
-  const int dquv_ac = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0;
+  const int base_q0 = VP8GetValue(br, 7, "global-header");
+  const int dqy1_dc = VP8Get(br, "global-header") ?
+       VP8GetSignedValue(br, 4, "global-header") : 0;
+  const int dqy2_dc = VP8Get(br, "global-header") ?
+       VP8GetSignedValue(br, 4, "global-header") : 0;
+  const int dqy2_ac = VP8Get(br, "global-header") ?
+       VP8GetSignedValue(br, 4, "global-header") : 0;
+  const int dquv_dc = VP8Get(br, "global-header") ?
+       VP8GetSignedValue(br, 4, "global-header") : 0;
+  const int dquv_ac = VP8Get(br, "global-header") ?
+       VP8GetSignedValue(br, 4, "global-header") : 0;
 
   const VP8SegmentHeader* const hdr = &dec->segment_hdr_;
   int i;
diff --git a/src/dec/tree_dec.c b/src/dec/tree_dec.c
index 3f5a957d..1c6fdea2 100644
--- a/src/dec/tree_dec.c
+++ b/src/dec/tree_dec.c
@@ -296,20 +296,21 @@ static void ParseIntraMode(VP8BitReader* const br,
   // to decode more than 1 keyframe.
   if (dec->segment_hdr_.update_map_) {
     // Hardcoded tree parsing
-    block->segment_ = !VP8GetBit(br, dec->proba_.segments_[0])
-                    ? VP8GetBit(br, dec->proba_.segments_[1])
-                    : 2 + VP8GetBit(br, dec->proba_.segments_[2]);
+    block->segment_ = !VP8GetBit(br, dec->proba_.segments_[0], "segments")
+                    ?  VP8GetBit(br, dec->proba_.segments_[1], "segments")
+                    :  VP8GetBit(br, dec->proba_.segments_[2], "segments") + 2;
   } else {
     block->segment_ = 0;  // default for intra
   }
-  if (dec->use_skip_proba_) block->skip_ = VP8GetBit(br, dec->skip_p_);
+  if (dec->use_skip_proba_) block->skip_ = VP8GetBit(br, dec->skip_p_, "skip");
 
-  block->is_i4x4_ = !VP8GetBit(br, 145);   // decide for B_PRED first
+  block->is_i4x4_ = !VP8GetBit(br, 145, "block-size");
   if (!block->is_i4x4_) {
     // Hardcoded 16x16 intra-mode decision tree.
     const int ymode =
-        VP8GetBit(br, 156) ? (VP8GetBit(br, 128) ? TM_PRED : H_PRED)
-                           : (VP8GetBit(br, 163) ? V_PRED : DC_PRED);
+        VP8GetBit(br, 156, "pred-modes") ?
+            (VP8GetBit(br, 128, "pred-modes") ? TM_PRED : H_PRED) :
+            (VP8GetBit(br, 163, "pred-modes") ? V_PRED : DC_PRED);
     block->imodes_[0] = ymode;
     memset(top, ymode, 4 * sizeof(*top));
     memset(left, ymode, 4 * sizeof(*left));
@@ -323,22 +324,25 @@ static void ParseIntraMode(VP8BitReader* const br,
         const uint8_t* const prob = kBModesProba[top[x]][ymode];
 #if (USE_GENERIC_TREE == 1)
         // Generic tree-parsing
-        int i = kYModesIntra4[VP8GetBit(br, prob[0])];
+        int i = kYModesIntra4[VP8GetBit(br, prob[0], "pred-modes")];
         while (i > 0) {
-          i = kYModesIntra4[2 * i + VP8GetBit(br, prob[i])];
+          i = kYModesIntra4[2 * i + VP8GetBit(br, prob[i], "pred-modes")];
         }
         ymode = -i;
 #else
         // Hardcoded tree parsing
-        ymode = !VP8GetBit(br, prob[0]) ? B_DC_PRED :
-                  !VP8GetBit(br, prob[1]) ? B_TM_PRED :
-                    !VP8GetBit(br, prob[2]) ? B_VE_PRED :
-                      !VP8GetBit(br, prob[3]) ?
-                        (!VP8GetBit(br, prob[4]) ? B_HE_PRED :
-                          (!VP8GetBit(br, prob[5]) ? B_RD_PRED : B_VR_PRED)) :
-                        (!VP8GetBit(br, prob[6]) ? B_LD_PRED :
-                          (!VP8GetBit(br, prob[7]) ? B_VL_PRED :
-                            (!VP8GetBit(br, prob[8]) ? B_HD_PRED : B_HU_PRED)));
+        ymode = !VP8GetBit(br, prob[0], "pred-modes") ? B_DC_PRED :
+                  !VP8GetBit(br, prob[1], "pred-modes") ? B_TM_PRED :
+                    !VP8GetBit(br, prob[2], "pred-modes") ? B_VE_PRED :
+                      !VP8GetBit(br, prob[3], "pred-modes") ?
+                        (!VP8GetBit(br, prob[4], "pred-modes") ? B_HE_PRED :
+                          (!VP8GetBit(br, prob[5], "pred-modes") ? B_RD_PRED
+                                                                 : B_VR_PRED)) :
+                        (!VP8GetBit(br, prob[6], "pred-modes") ? B_LD_PRED :
+                          (!VP8GetBit(br, prob[7], "pred-modes") ? B_VL_PRED :
+                            (!VP8GetBit(br, prob[8], "pred-modes") ? B_HD_PRED
+                                                                   : B_HU_PRED))
+                        );
 #endif  // USE_GENERIC_TREE
         top[x] = ymode;
       }
@@ -348,9 +352,9 @@ static void ParseIntraMode(VP8BitReader* const br,
     }
   }
   // Hardcoded UVMode decision tree
-  block->uvmode_ = !VP8GetBit(br, 142) ? DC_PRED
-                 : !VP8GetBit(br, 114) ? V_PRED
-                 : VP8GetBit(br, 183) ? TM_PRED : H_PRED;
+  block->uvmode_ = !VP8GetBit(br, 142, "pred-modes-uv") ? DC_PRED
+                 : !VP8GetBit(br, 114, "pred-modes-uv") ? V_PRED
+                 : VP8GetBit(br, 183, "pred-modes-uv") ? TM_PRED : H_PRED;
 }
 
 int VP8ParseIntraModeRow(VP8BitReader* const br, VP8Decoder* const dec) {
@@ -514,8 +518,10 @@ void VP8ParseProba(VP8BitReader* const br, VP8Decoder* const dec) {
     for (b = 0; b < NUM_BANDS; ++b) {
       for (c = 0; c < NUM_CTX; ++c) {
         for (p = 0; p < NUM_PROBAS; ++p) {
-          const int v = VP8GetBit(br, CoeffsUpdateProba[t][b][c][p]) ?
-                        VP8GetValue(br, 8) : CoeffsProba0[t][b][c][p];
+          const int v =
+              VP8GetBit(br, CoeffsUpdateProba[t][b][c][p], "global-header") ?
+                        VP8GetValue(br, 8, "global-header") :
+                        CoeffsProba0[t][b][c][p];
           proba->bands_[t][b].probas_[c][p] = v;
         }
       }
@@ -524,9 +530,8 @@ void VP8ParseProba(VP8BitReader* const br, VP8Decoder* const dec) {
       proba->bands_ptr_[t][b] = &proba->bands_[t][kBands[b]];
     }
   }
-  dec->use_skip_proba_ = VP8Get(br);
+  dec->use_skip_proba_ = VP8Get(br, "global-header");
   if (dec->use_skip_proba_) {
-    dec->skip_p_ = VP8GetValue(br, 8);
+    dec->skip_p_ = VP8GetValue(br, 8, "global-header");
   }
 }
-
diff --git a/src/dec/vp8_dec.c b/src/dec/vp8_dec.c
index c904b529..57efb690 100644
--- a/src/dec/vp8_dec.c
+++ b/src/dec/vp8_dec.c
@@ -161,23 +161,26 @@ static int ParseSegmentHeader(VP8BitReader* br,
                               VP8SegmentHeader* hdr, VP8Proba* proba) {
   assert(br != NULL);
   assert(hdr != NULL);
-  hdr->use_segment_ = VP8Get(br);
+  hdr->use_segment_ = VP8Get(br, "global-header");
   if (hdr->use_segment_) {
-    hdr->update_map_ = VP8Get(br);
-    if (VP8Get(br)) {   // update data
+    hdr->update_map_ = VP8Get(br, "global-header");
+    if (VP8Get(br, "global-header")) {   // update data
       int s;
-      hdr->absolute_delta_ = VP8Get(br);
+      hdr->absolute_delta_ = VP8Get(br, "global-header");
       for (s = 0; s < NUM_MB_SEGMENTS; ++s) {
-        hdr->quantizer_[s] = VP8Get(br) ? VP8GetSignedValue(br, 7) : 0;
+        hdr->quantizer_[s] = VP8Get(br, "global-header") ?
+            VP8GetSignedValue(br, 7, "global-header") : 0;
       }
       for (s = 0; s < NUM_MB_SEGMENTS; ++s) {
-        hdr->filter_strength_[s] = VP8Get(br) ? VP8GetSignedValue(br, 6) : 0;
+        hdr->filter_strength_[s] = VP8Get(br, "global-header") ?
+            VP8GetSignedValue(br, 6, "global-header") : 0;
       }
     }
     if (hdr->update_map_) {
       int s;
       for (s = 0; s < MB_FEATURE_TREE_PROBS; ++s) {
-        proba->segments_[s] = VP8Get(br) ? VP8GetValue(br, 8) : 255u;
+        proba->segments_[s] = VP8Get(br, "global-header") ?
+            VP8GetValue(br, 8, "global-header") : 255u;
       }
     }
   } else {
@@ -205,7 +208,7 @@ static VP8StatusCode ParsePartitions(VP8Decoder* const dec,
   size_t last_part;
   size_t p;
 
-  dec->num_parts_minus_one_ = (1 << VP8GetValue(br, 2)) - 1;
+  dec->num_parts_minus_one_ = (1 << VP8GetValue(br, 2, "global-header")) - 1;
   last_part = dec->num_parts_minus_one_;
   if (size < 3 * last_part) {
     // we can't even read the sizes with sz[]! That's a failure.
@@ -229,21 +232,21 @@ static VP8StatusCode ParsePartitions(VP8Decoder* const dec,
 // Paragraph 9.4
 static int ParseFilterHeader(VP8BitReader* br, VP8Decoder* const dec) {
   VP8FilterHeader* const hdr = &dec->filter_hdr_;
-  hdr->simple_    = VP8Get(br);
-  hdr->level_     = VP8GetValue(br, 6);
-  hdr->sharpness_ = VP8GetValue(br, 3);
-  hdr->use_lf_delta_ = VP8Get(br);
+  hdr->simple_    = VP8Get(br, "global-header");
+  hdr->level_     = VP8GetValue(br, 6, "global-header");
+  hdr->sharpness_ = VP8GetValue(br, 3, "global-header");
+  hdr->use_lf_delta_ = VP8Get(br, "global-header");
   if (hdr->use_lf_delta_) {
-    if (VP8Get(br)) {   // update lf-delta?
+    if (VP8Get(br, "global-header")) {   // update lf-delta?
       int i;
       for (i = 0; i < NUM_REF_LF_DELTAS; ++i) {
-        if (VP8Get(br)) {
-          hdr->ref_lf_delta_[i] = VP8GetSignedValue(br, 6);
+        if (VP8Get(br, "global-header")) {
+          hdr->ref_lf_delta_[i] = VP8GetSignedValue(br, 6, "global-header");
         }
       }
       for (i = 0; i < NUM_MODE_LF_DELTAS; ++i) {
-        if (VP8Get(br)) {
-          hdr->mode_lf_delta_[i] = VP8GetSignedValue(br, 6);
+        if (VP8Get(br, "global-header")) {
+          hdr->mode_lf_delta_[i] = VP8GetSignedValue(br, 6, "global-header");
         }
       }
     }
@@ -352,8 +355,8 @@ int VP8GetHeaders(VP8Decoder* const dec, VP8Io* const io) {
   buf_size -= frm_hdr->partition_length_;
 
   if (frm_hdr->key_frame_) {
-    pic_hdr->colorspace_ = VP8Get(br);
-    pic_hdr->clamp_type_ = VP8Get(br);
+    pic_hdr->colorspace_ = VP8Get(br, "global-header");
+    pic_hdr->clamp_type_ = VP8Get(br, "global-header");
   }
   if (!ParseSegmentHeader(br, &dec->segment_hdr_, &dec->proba_)) {
     return VP8SetError(dec, VP8_STATUS_BITSTREAM_ERROR,
@@ -378,7 +381,7 @@ int VP8GetHeaders(VP8Decoder* const dec, VP8Io* const io) {
                        "Not a key frame.");
   }
 
-  VP8Get(br);   // ignore the value of update_proba_
+  VP8Get(br, "global-header");   // ignore the value of update_proba_
 
   VP8ParseProba(br, dec);
 
@@ -403,28 +406,28 @@ static const uint8_t kZigzag[16] = {
 // See section 13-2: http://tools.ietf.org/html/rfc6386#section-13.2
 static int GetLargeValue(VP8BitReader* const br, const uint8_t* const p) {
   int v;
-  if (!VP8GetBit(br, p[3])) {
-    if (!VP8GetBit(br, p[4])) {
+  if (!VP8GetBit(br, p[3], "coeffs")) {
+    if (!VP8GetBit(br, p[4], "coeffs")) {
       v = 2;
     } else {
-      v = 3 + VP8GetBit(br, p[5]);
+      v = 3 + VP8GetBit(br, p[5], "coeffs");
     }
   } else {
-    if (!VP8GetBit(br, p[6])) {
-      if (!VP8GetBit(br, p[7])) {
-        v = 5 + VP8GetBit(br, 159);
+    if (!VP8GetBit(br, p[6], "coeffs")) {
+      if (!VP8GetBit(br, p[7], "coeffs")) {
+        v = 5 + VP8GetBit(br, 159, "coeffs");
       } else {
-        v = 7 + 2 * VP8GetBit(br, 165);
-        v += VP8GetBit(br, 145);
+        v = 7 + 2 * VP8GetBit(br, 165, "coeffs");
+        v += VP8GetBit(br, 145, "coeffs");
       }
     } else {
       const uint8_t* tab;
-      const int bit1 = VP8GetBit(br, p[8]);
-      const int bit0 = VP8GetBit(br, p[9 + bit1]);
+      const int bit1 = VP8GetBit(br, p[8], "coeffs");
+      const int bit0 = VP8GetBit(br, p[9 + bit1], "coeffs");
       const int cat = 2 * bit1 + bit0;
       v = 0;
       for (tab = kCat3456[cat]; *tab; ++tab) {
-        v += v + VP8GetBit(br, *tab);
+        v += v + VP8GetBit(br, *tab, "coeffs");
       }
       v += 3 + (8 << cat);
     }
@@ -438,24 +441,24 @@ static int GetCoeffsFast(VP8BitReader* const br,
                          int ctx, const quant_t dq, int n, int16_t* out) {
   const uint8_t* p = prob[n]->probas_[ctx];
   for (; n < 16; ++n) {
-    if (!VP8GetBit(br, p[0])) {
+    if (!VP8GetBit(br, p[0], "coeffs")) {
       return n;  // previous coeff was last non-zero coeff
     }
-    while (!VP8GetBit(br, p[1])) {       // sequence of zero coeffs
+    while (!VP8GetBit(br, p[1], "coeffs")) {       // sequence of zero coeffs
       p = prob[++n]->probas_[0];
       if (n == 16) return 16;
     }
     {        // non zero coeff
       const VP8ProbaArray* const p_ctx = &prob[n + 1]->probas_[0];
       int v;
-      if (!VP8GetBit(br, p[2])) {
+      if (!VP8GetBit(br, p[2], "coeffs")) {
         v = 1;
         p = p_ctx[1];
       } else {
         v = GetLargeValue(br, p);
         p = p_ctx[2];
       }
-      out[kZigzag[n]] = VP8GetSigned(br, v) * dq[n > 0];
+      out[kZigzag[n]] = VP8GetSigned(br, v, "coeffs") * dq[n > 0];
     }
   }
   return 16;
@@ -468,24 +471,24 @@ static int GetCoeffsAlt(VP8BitReader* const br,
                         int ctx, const quant_t dq, int n, int16_t* out) {
   const uint8_t* p = prob[n]->probas_[ctx];
   for (; n < 16; ++n) {
-    if (!VP8GetBitAlt(br, p[0])) {
+    if (!VP8GetBitAlt(br, p[0], "coeffs")) {
       return n;  // previous coeff was last non-zero coeff
     }
-    while (!VP8GetBitAlt(br, p[1])) {       // sequence of zero coeffs
+    while (!VP8GetBitAlt(br, p[1], "coeffs")) {       // sequence of zero coeffs
       p = prob[++n]->probas_[0];
       if (n == 16) return 16;
     }
     {        // non zero coeff
       const VP8ProbaArray* const p_ctx = &prob[n + 1]->probas_[0];
       int v;
-      if (!VP8GetBitAlt(br, p[2])) {
+      if (!VP8GetBitAlt(br, p[2], "coeffs")) {
         v = 1;
         p = p_ctx[1];
       } else {
         v = GetLargeValue(br, p);
         p = p_ctx[2];
       }
-      out[kZigzag[n]] = VP8GetSigned(br, v) * dq[n > 0];
+      out[kZigzag[n]] = VP8GetSigned(br, v, "coeffs") * dq[n > 0];
     }
   }
   return 16;
diff --git a/src/utils/bit_reader_inl_utils.h b/src/utils/bit_reader_inl_utils.h
index 7e607f37..46b38807 100644
--- a/src/utils/bit_reader_inl_utils.h
+++ b/src/utils/bit_reader_inl_utils.h
@@ -104,7 +104,8 @@ void VP8LoadNewBytes(VP8BitReader* const br) {
 }
 
 // Read a bit with proba 'prob'. Speed-critical function!
-static WEBP_INLINE int VP8GetBit(VP8BitReader* const br, int prob) {
+static WEBP_INLINE int VP8GetBit(VP8BitReader* const br,
+                                 int prob, const char label[]) {
   // Don't move this declaration! It makes a big speed difference to store
   // 'range' *before* calling VP8LoadNewBytes(), even if this function doesn't
   // alter br->range_ value.
@@ -129,13 +130,14 @@ static WEBP_INLINE int VP8GetBit(VP8BitReader* const br, int prob) {
       br->bits_ -= shift;
     }
     br->range_ = range - 1;
+    BT_TRACK(br);
     return bit;
   }
 }
 
 // simplified version of VP8GetBit() for prob=0x80 (note shift is always 1 here)
 static WEBP_UBSAN_IGNORE_UNSIGNED_OVERFLOW WEBP_INLINE
-int VP8GetSigned(VP8BitReader* const br, int v) {
+int VP8GetSigned(VP8BitReader* const br, int v, const char label[]) {
   if (br->bits_ < 0) {
     VP8LoadNewBytes(br);
   }
@@ -148,11 +150,13 @@ int VP8GetSigned(VP8BitReader* const br, int v) {
     br->range_ += mask;
     br->range_ |= 1;
     br->value_ -= (bit_t)((split + 1) & mask) << pos;
+    BT_TRACK(br);
     return (v ^ mask) - mask;
   }
 }
 
-static WEBP_INLINE int VP8GetBitAlt(VP8BitReader* const br, int prob) {
+static WEBP_INLINE int VP8GetBitAlt(VP8BitReader* const br,
+                                    int prob, const char label[]) {
   // Don't move this declaration! It makes a big speed difference to store
   // 'range' *before* calling VP8LoadNewBytes(), even if this function doesn't
   // alter br->range_ value.
@@ -179,6 +183,7 @@ static WEBP_INLINE int VP8GetBitAlt(VP8BitReader* const br, int prob) {
       br->bits_ -= shift;
     }
     br->range_ = range;
+    BT_TRACK(br);
     return bit;
   }
 }
diff --git a/src/utils/bit_reader_utils.c b/src/utils/bit_reader_utils.c
index 1500354d..857cd609 100644
--- a/src/utils/bit_reader_utils.c
+++ b/src/utils/bit_reader_utils.c
@@ -102,17 +102,18 @@ void VP8LoadFinalBytes(VP8BitReader* const br) {
 //------------------------------------------------------------------------------
 // Higher-level calls
 
-uint32_t VP8GetValue(VP8BitReader* const br, int bits) {
+uint32_t VP8GetValue(VP8BitReader* const br, int bits, const char label[]) {
   uint32_t v = 0;
   while (bits-- > 0) {
-    v |= VP8GetBit(br, 0x80) << bits;
+    v |= VP8GetBit(br, 0x80, label) << bits;
   }
   return v;
 }
 
-int32_t VP8GetSignedValue(VP8BitReader* const br, int bits) {
-  const int value = VP8GetValue(br, bits);
-  return VP8Get(br) ? -value : value;
+int32_t VP8GetSignedValue(VP8BitReader* const br, int bits,
+                          const char label[]) {
+  const int value = VP8GetValue(br, bits, label);
+  return VP8Get(br, label) ? -value : value;
 }
 
 //------------------------------------------------------------------------------
@@ -220,3 +221,78 @@ uint32_t VP8LReadBits(VP8LBitReader* const br, int n_bits) {
 }
 
 //------------------------------------------------------------------------------
+// Bit-tracing tool
+
+#if (BITTRACE > 0)
+
+#include <stdlib.h>   // for atexit()
+#include <stdio.h>
+#include <string.h>
+
+#define MAX_NUM_LABELS 32
+static struct {
+  const char* label;
+  int size;
+  int count;
+} kLabels[MAX_NUM_LABELS];
+
+static int last_label = 0;
+static int last_pos = 0;
+static const uint8_t* buf_start = NULL;
+static int init_done = 0;
+
+static void PrintBitTraces(void) {
+  int i;
+  int scale = 1;
+  int total = 0;
+  const char* units = "bits";
+#if (BITTRACE == 2)
+  scale = 8;
+  units = "bytes";
+#endif
+  for (i = 0; i < last_label; ++i) total += kLabels[i].size;
+  if (total < 1) total = 1;   // avoid rounding errors
+  printf("=== Bit traces ===\n");
+  for (i = 0; i < last_label; ++i) {
+    const int skip = 16 - (int)strlen(kLabels[i].label);
+    const int value = (kLabels[i].size + scale - 1) / scale;
+    assert(skip > 0);
+    printf("%s \%*s: %6d %s   \t[%5.2f%%] [count: %7d]\n",
+           kLabels[i].label, skip, "", value, units,
+           100.f * kLabels[i].size / total,
+           kLabels[i].count);
+  }
+  total = (total + scale - 1) / scale;
+  printf("Total: %d %s\n", total, units);
+}
+
+void BitTrace(const struct VP8BitReader* const br, const char label[]) {
+  int i, pos;
+  if (!init_done) {
+    memset(kLabels, 0, sizeof(kLabels));
+    atexit(PrintBitTraces);
+    buf_start = br->buf_;
+    init_done = 1;
+  }
+  pos = (int)(br->buf_ - buf_start) * 8 - br->bits_;
+  // if there's a too large jump, we've changed partition -> reset counter
+  if (abs(pos - last_pos) > 32) {
+    buf_start = br->buf_;
+    pos = 0;
+    last_pos = 0;
+  }
+  if (br->range_ >= 0x7f) pos += kVP8Log2Range[br->range_ - 0x7f];
+  for (i = 0; i < last_label; ++i) {
+    if (!strcmp(label, kLabels[i].label)) break;
+  }
+  if (i == MAX_NUM_LABELS) abort();   // overflow!
+  kLabels[i].label = label;
+  kLabels[i].size += pos - last_pos;
+  kLabels[i].count += 1;
+  if (i == last_label) ++last_label;
+  last_pos = pos;
+}
+
+#endif  // BITTRACE > 0
+
+//------------------------------------------------------------------------------
diff --git a/src/utils/bit_reader_utils.h b/src/utils/bit_reader_utils.h
index 0f8db2ca..e64156e3 100644
--- a/src/utils/bit_reader_utils.h
+++ b/src/utils/bit_reader_utils.h
@@ -21,6 +21,27 @@
 #endif
 #include "src/webp/types.h"
 
+// Warning! This macro triggers quite some MACRO wizardry around func signature!
+#if !defined(BITTRACE)
+#define BITTRACE 0    // 0 = off, 1 = print bits, 2 = print bytes
+#endif
+
+#if (BITTRACE > 0)
+struct VP8BitReader;
+extern void BitTrace(const struct VP8BitReader* const br, const char label[]);
+#define BT_TRACK(br) BitTrace(br, label)
+#define VP8Get(BR, L) VP8GetValue(BR, 1, L)
+#else
+#define BT_TRACK(br)
+// We'll REMOVE the 'const char label[]' from all signatures and calls (!!):
+#define VP8GetValue(BR, N, L) VP8GetValue(BR, N)
+#define VP8Get(BR, L) VP8GetValue(BR, 1, L)
+#define VP8GetSignedValue(BR, N, L) VP8GetSignedValue(BR, N)
+#define VP8GetBit(BR, P, L) VP8GetBit(BR, P)
+#define VP8GetBitAlt(BR, P, L) VP8GetBitAlt(BR, P)
+#define VP8GetSigned(BR, V, L) VP8GetSigned(BR, V)
+#endif
+
 #ifdef __cplusplus
 extern "C" {
 #endif
@@ -92,17 +113,15 @@ void VP8BitReaderSetBuffer(VP8BitReader* const br,
 void VP8RemapBitReader(VP8BitReader* const br, ptrdiff_t offset);
 
 // return the next value made of 'num_bits' bits
-uint32_t VP8GetValue(VP8BitReader* const br, int num_bits);
-static WEBP_INLINE uint32_t VP8Get(VP8BitReader* const br) {
-  return VP8GetValue(br, 1);
-}
+uint32_t VP8GetValue(VP8BitReader* const br, int num_bits, const char label[]);
 
 // return the next value with sign-extension.
-int32_t VP8GetSignedValue(VP8BitReader* const br, int num_bits);
+int32_t VP8GetSignedValue(VP8BitReader* const br, int num_bits,
+                          const char label[]);
 
 // bit_reader_inl.h will implement the following methods:
-//   static WEBP_INLINE int VP8GetBit(VP8BitReader* const br, int prob)
-//   static WEBP_INLINE int VP8GetSigned(VP8BitReader* const br, int v)
+//   static WEBP_INLINE int VP8GetBit(VP8BitReader* const br, int prob, ...)
+//   static WEBP_INLINE int VP8GetSigned(VP8BitReader* const br, int v, ...)
 // and should be included by the .c files that actually need them.
 // This is to avoid recompiling the whole library whenever this file is touched,
 // and also allowing platform-specific ad-hoc hacks.