Go to the documentation of this file.
28 #ifndef AVCODEC_SIMPLE_IDCT_H
29 #define AVCODEC_SIMPLE_IDCT_H
void ff_simple_idct_put_int16_10bit(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
void ff_prores_idct_10(int16_t *block, const int16_t *qmat)
Special version of ff_simple_idct_int16_10bit() which does dequantization and scales by a factor of 2...
void ff_simple_idct_add_int16_12bit(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
void ff_simple_idct_int16_10bit(int16_t *block)
void ff_simple_idct_add_int32_10bit(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
void ff_simple_idct84_add(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
void ff_simple_idct_put_int16_8bit(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
void ff_simple_idct_put_int16_12bit(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
void ff_prores_idct_12(int16_t *block, const int16_t *qmat)
void ff_simple_idct_add_int16_8bit(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
void ff_simple_idct_put_int32_10bit(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
void ff_simple_idct44_add(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
void ff_simple_idct_int16_8bit(int16_t *block)
void ff_simple_idct48_add(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
void ff_simple_idct_int16_12bit(int16_t *block)
void ff_simple_idct_add_int16_10bit(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
void ff_simple_idct248_put(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
void ff_simple_idct_int32_10bit(int16_t *block)
The exact code depends on how similar the blocks are and how related they are to the block