vp10: code sign bit before absolute value in non-arithcoded header.

For reading, this makes the operation branchless, although it still
requires two shifts. For writing, this makes the operation as fast
as writing an unsigned value, branchlessly. This is also how other
codecs typically code signed, non-arithcoded bitstream elements.

See issue 1039.

Change-Id: I6a8182cc88a16842fb431688c38f6b52d7f24ead

6 files changed, 34 insertions, 9 deletions

diff --git a/vp10/decoder/decodeframe.c b/vp10/decoder/decodeframe.c
index d6a4e6f6a..387204bfa 100644
--- a/vp10/decoder/decodeframe.c
+++ b/vp10/decoder/decodeframe.c
@@ -1095,17 +1095,17 @@ static void setup_loopfilter(struct loopfilter *lf,
 
       for (i = 0; i < MAX_REF_FRAMES; i++)
         if (vpx_rb_read_bit(rb))
-          lf->ref_deltas[i] = vpx_rb_read_signed_literal(rb, 6);
+          lf->ref_deltas[i] = vpx_rb_read_inv_signed_literal(rb, 6);
 
       for (i = 0; i < MAX_MODE_LF_DELTAS; i++)
         if (vpx_rb_read_bit(rb))
-          lf->mode_deltas[i] = vpx_rb_read_signed_literal(rb, 6);
+          lf->mode_deltas[i] = vpx_rb_read_inv_signed_literal(rb, 6);
     }
   }
 }
 
 static INLINE int read_delta_q(struct vpx_read_bit_buffer *rb) {
-  return vpx_rb_read_bit(rb) ? vpx_rb_read_signed_literal(rb, 4) : 0;
+  return vpx_rb_read_bit(rb) ? vpx_rb_read_inv_signed_literal(rb, 4) : 0;
 }
 
 static void setup_quantization(VP10_COMMON *const cm, MACROBLOCKD *const xd,
diff --git a/vp10/encoder/bitstream.c b/vp10/encoder/bitstream.c
index 8d84a8590..c0a7d597c 100644
--- a/vp10/encoder/bitstream.c
+++ b/vp10/encoder/bitstream.c
@@ -714,8 +714,7 @@ static void encode_loopfilter(struct loopfilter *lf,
         vpx_wb_write_bit(wb, changed);
         if (changed) {
           lf->last_ref_deltas[i] = delta;
-          vpx_wb_write_literal(wb, abs(delta) & 0x3F, 6);
-          vpx_wb_write_bit(wb, delta < 0);
+          vpx_wb_write_inv_signed_literal(wb, delta, 6);
         }
       }
 
@@ -725,8 +724,7 @@ static void encode_loopfilter(struct loopfilter *lf,
         vpx_wb_write_bit(wb, changed);
         if (changed) {
           lf->last_mode_deltas[i] = delta;
-          vpx_wb_write_literal(wb, abs(delta) & 0x3F, 6);
-          vpx_wb_write_bit(wb, delta < 0);
+          vpx_wb_write_inv_signed_literal(wb, delta, 6);
         }
       }
     }
@@ -736,8 +734,7 @@ static void encode_loopfilter(struct loopfilter *lf,
 static void write_delta_q(struct vpx_write_bit_buffer *wb, int delta_q) {
   if (delta_q != 0) {
     vpx_wb_write_bit(wb, 1);
-    vpx_wb_write_literal(wb, abs(delta_q), 4);
-    vpx_wb_write_bit(wb, delta_q < 0);
+    vpx_wb_write_inv_signed_literal(wb, delta_q, 4);
   } else {
     vpx_wb_write_bit(wb, 0);
   }
diff --git a/vpx_dsp/bitreader_buffer.c b/vpx_dsp/bitreader_buffer.c
index fb04ee634..bb917263e 100644
--- a/vpx_dsp/bitreader_buffer.c
+++ b/vpx_dsp/bitreader_buffer.c
@@ -7,6 +7,7 @@
  *  in the file PATENTS.  All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
+#include "./vpx_config.h"
 #include "./bitreader_buffer.h"
 
 size_t vpx_rb_bytes_read(struct vpx_read_bit_buffer *rb) {
@@ -39,3 +40,14 @@ int vpx_rb_read_signed_literal(struct vpx_read_bit_buffer *rb,
   const int value = vpx_rb_read_literal(rb, bits);
   return vpx_rb_read_bit(rb) ? -value : value;
 }
+
+int vpx_rb_read_inv_signed_literal(struct vpx_read_bit_buffer *rb,
+                                   int bits) {
+#if CONFIG_MISC_FIXES
+  const int nbits = sizeof(unsigned) * 8 - bits - 1;
+  const unsigned value = vpx_rb_read_literal(rb, bits + 1) << nbits;
+  return ((int) value) >> nbits;
+#else
+  return vpx_rb_read_signed_literal(rb, bits);
+#endif
+}
diff --git a/vpx_dsp/bitreader_buffer.h b/vpx_dsp/bitreader_buffer.h
index 03b156ba2..8a48a95ed 100644
--- a/vpx_dsp/bitreader_buffer.h
+++ b/vpx_dsp/bitreader_buffer.h
@@ -38,6 +38,8 @@ int vpx_rb_read_literal(struct vpx_read_bit_buffer *rb, int bits);
 
 int vpx_rb_read_signed_literal(struct vpx_read_bit_buffer *rb, int bits);
 
+int vpx_rb_read_inv_signed_literal(struct vpx_read_bit_buffer *rb, int bits);
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/vpx_dsp/bitwriter_buffer.c b/vpx_dsp/bitwriter_buffer.c
index 0dfb859db..6182a7222 100644
--- a/vpx_dsp/bitwriter_buffer.c
+++ b/vpx_dsp/bitwriter_buffer.c
@@ -9,7 +9,9 @@
  */
 
 #include <limits.h>
+#include <stdlib.h>
 
+#include "./vpx_config.h"
 #include "./bitwriter_buffer.h"
 
 size_t vpx_wb_bytes_written(const struct vpx_write_bit_buffer *wb) {
@@ -34,3 +36,13 @@ void vpx_wb_write_literal(struct vpx_write_bit_buffer *wb, int data, int bits) {
   for (bit = bits - 1; bit >= 0; bit--)
     vpx_wb_write_bit(wb, (data >> bit) & 1);
 }
+
+void vpx_wb_write_inv_signed_literal(struct vpx_write_bit_buffer *wb,
+                                     int data, int bits) {
+#if CONFIG_MISC_FIXES
+  vpx_wb_write_literal(wb, data, bits + 1);
+#else
+  vpx_wb_write_literal(wb, abs(data), bits);
+  vpx_wb_write_bit(wb, data < 0);
+#endif
+}
diff --git a/vpx_dsp/bitwriter_buffer.h b/vpx_dsp/bitwriter_buffer.h
index 9397668ee..a123a2fe8 100644
--- a/vpx_dsp/bitwriter_buffer.h
+++ b/vpx_dsp/bitwriter_buffer.h
@@ -28,6 +28,8 @@ void vpx_wb_write_bit(struct vpx_write_bit_buffer *wb, int bit);
 
 void vpx_wb_write_literal(struct vpx_write_bit_buffer *wb, int data, int bits);
 
+void vpx_wb_write_inv_signed_literal(struct vpx_write_bit_buffer *wb, int data,
+                                     int bits);
 
 #ifdef __cplusplus
 }  // extern "C"