vpx_dsp/inv_txfm.h

   1 /*
   2  *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
   3  *
   4  *  Use of this source code is governed by a BSD-style license
   5  *  that can be found in the LICENSE file in the root of the source
   6  *  tree. An additional intellectual property rights grant can be found
   7  *  in the file PATENTS.  All contributing project authors may
   8  *  be found in the AUTHORS file in the root of the source tree.
   9  */
  10
  11 #ifndef VPX_DSP_INV_TXFM_H_
  12 #define VPX_DSP_INV_TXFM_H_
  13
  14 #include <assert.h>
  15
  16 #include "./vpx_config.h"
  17 #include "vpx_dsp/txfm_common.h"
  18 #include "vpx_ports/mem.h"
  19
  20 #ifdef __cplusplus
  21 extern "C" {
  22 #endif
  23
  24 static INLINE tran_low_t check_range(tran_high_t input) {
  25 #if CONFIG_COEFFICIENT_RANGE_CHECKING
  26   // For valid VP9 input streams, intermediate stage coefficients should always
  27   // stay within the range of a signed 16 bit integer. Coefficients can go out
  28   // of this range for invalid/corrupt VP9 streams. However, strictly checking
  29   // this range for every intermediate coefficient can burdensome for a decoder,
  30   // therefore the following assertion is only enabled when configured with
  31   // --enable-coefficient-range-checking.
  32   assert(INT16_MIN <= input);
  33   assert(input <= INT16_MAX);
  34 #endif  // CONFIG_COEFFICIENT_RANGE_CHECKING
  35   return (tran_low_t)input;
  36 }
  37
  38 static INLINE tran_low_t dct_const_round_shift(tran_high_t input) {
  39   tran_high_t rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
  40   return check_range(rv);
  41 }
  42
  43 #if CONFIG_VP9_HIGHBITDEPTH
  44 static INLINE tran_low_t highbd_check_range(tran_high_t input,
  45                                             int bd) {
  46 #if CONFIG_COEFFICIENT_RANGE_CHECKING
  47   // For valid highbitdepth VP9 streams, intermediate stage coefficients will
  48   // stay within the ranges:
  49   // - 8 bit: signed 16 bit integer
  50   // - 10 bit: signed 18 bit integer
  51   // - 12 bit: signed 20 bit integer
  52   const int32_t int_max = (1 << (7 + bd)) - 1;
  53   const int32_t int_min = -int_max - 1;
  54   assert(int_min <= input);
  55   assert(input <= int_max);
  56   (void) int_min;
  57 #endif  // CONFIG_COEFFICIENT_RANGE_CHECKING
  58   (void) bd;
  59   return (tran_low_t)input;
  60 }
  61
  62 static INLINE tran_low_t highbd_dct_const_round_shift(tran_high_t input,
  63                                                       int bd) {
  64   tran_high_t rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
  65   return highbd_check_range(rv, bd);
  66 }
  67 #endif  // CONFIG_VP9_HIGHBITDEPTH
  68
  69 #if CONFIG_EMULATE_HARDWARE
  70 // When CONFIG_EMULATE_HARDWARE is 1 the transform performs a
  71 // non-normative method to handle overflows. A stream that causes
  72 // overflows  in the inverse transform is considered invalid in VP9,
  73 // and a hardware implementer is free to choose any reasonable
  74 // method to handle overflows. However to aid in hardware
  75 // verification they can use a specific implementation of the
  76 // WRAPLOW() macro below that is identical to their intended
  77 // hardware implementation (and also use configure options to trigger
  78 // the C-implementation of the transform).
  79 //
  80 // The particular WRAPLOW implementation below performs strict
  81 // overflow wrapping to match common hardware implementations.
  82 // bd of 8 uses trans_low with 16bits, need to remove 16bits
  83 // bd of 10 uses trans_low with 18bits, need to remove 14bits
  84 // bd of 12 uses trans_low with 20bits, need to remove 12bits
  85 // bd of x uses trans_low with 8+x bits, need to remove 24-x bits
  86 #define WRAPLOW(x, bd) ((((int32_t)(x)) << (24 - bd)) >> (24 - bd))
  87 #else
  88 #define WRAPLOW(x, bd) ((int32_t)(x))
  89 #endif  // CONFIG_EMULATE_HARDWARE
  90
  91 void idct4_c(const tran_low_t *input, tran_low_t *output);
  92 void idct8_c(const tran_low_t *input, tran_low_t *output);
  93 void idct16_c(const tran_low_t *input, tran_low_t *output);
  94 void idct32_c(const tran_low_t *input, tran_low_t *output);
  95 void iadst4_c(const tran_low_t *input, tran_low_t *output);
  96 void iadst8_c(const tran_low_t *input, tran_low_t *output);
  97 void iadst16_c(const tran_low_t *input, tran_low_t *output);
  98
  99 #if CONFIG_VP9_HIGHBITDEPTH
 100 void vpx_highbd_idct4_c(const tran_low_t *input, tran_low_t *output, int bd);
 101 void vpx_highbd_idct8_c(const tran_low_t *input, tran_low_t *output, int bd);
 102 void vpx_highbd_idct16_c(const tran_low_t *input, tran_low_t *output, int bd);
 103
 104 void vpx_highbd_iadst4_c(const tran_low_t *input, tran_low_t *output, int bd);
 105 void vpx_highbd_iadst8_c(const tran_low_t *input, tran_low_t *output, int bd);
 106 void vpx_highbd_iadst16_c(const tran_low_t *input, tran_low_t *output, int bd);
 107
 108 static INLINE uint16_t highbd_clip_pixel_add(uint16_t dest, tran_high_t trans,
 109                                              int bd) {
 110   trans = WRAPLOW(trans, bd);
 111   return clip_pixel_highbd(WRAPLOW(dest + trans, bd), bd);
 112 }
 113 #endif
 114
 115 static INLINE uint8_t clip_pixel_add(uint8_t dest, tran_high_t trans) {
 116   trans = WRAPLOW(trans, 8);
 117   return clip_pixel(WRAPLOW(dest + trans, 8));
 118 }
 119 #ifdef __cplusplus
 120 }  // extern "C"
 121 #endif
 122
 123 #endif  // VPX_DSP_INV_TXFM_H_