📄 mpeg2_vld_inter_n.sa
字号:
* ========================================================================= ** TEXAS INSTRUMENTS, INC. ** ** NAME ** mpeg2_vld_inter ** ** PLATFORM ** C6400 ** ** DESCRIPTION * * This routine takes a bitstream of an MPEG-2 non-intra coded * * macroblock and returns the decoded IDCT coefficients. The routine * * is implemented as specified in the MPEG-2 standard text (ISO/IEC * * 13818-2). The routine checks the coded block pattern (cbp), * * performs coefficient decoding inlcuding, variable length decode, * * run-length expansion, inverse zigzag, dequantization, saturation * * and mismatch control. * * * * USAGE * * This routine is C callable, and has the following C prototype: * * * * void mpeg2_vld_inter * * ( * * const short *restrict Wptr, * * short *restrict outi, * * IMG_mpeg2_vld *restrict Mpeg2v, * * int mode_12Q4, * * int num_blocks, * * int bsbuf_words * * ); * * * * Wptr: Pointer to array that contains quantization matrix. The * * elements of the quantization matrix in *Wptr must be * * ordered according to the scan pattern used (zigzag or * * alternate scan). Video format 4:2:0 requires one * * quantization matrix (64 array elements). For formats * * 4:2:2 and 4:4:4 two quantization matrices (one for luma * * and one for chroma) must specified in the array (128 * * array elements). * * * * outi: Pointer to the IDCT coefficients output array * * (6*64 elements), elements must be set to zero prior to * * function call. * * * * Mpeg2v: Pointer to the context object containing the coding * * parameters of the MB to be decoded and the current state * * of the bitstream buffer. The structure is described * * below. * * * * mode_12Q4: 0: Coefficients are returned in normal 16-bit integer * * format. * * Otherwise: Coefficients are returned in 12Q4 format * * (normal 16-bit integer format left shifted by 4). This * * mode is useful for directly passing the coefficients * * into the IMG_idct_8x8 routine. * * * * num_blocks: Number of blocks that the MB contains. Valid values are * * 6 for 4:2:0, 8 for 4:2:2 and 12 for 4:4:4 format. * * * * bsbuf_words: Size of bitstream buffer in words. Must be a power of 2. * * Bitstream buffer must be aligned at an address boundary * * equal to its size in bytes (bitstream buffer is * * addressed circularly by this routine.) * * * * The structure Mpeg2v is defined as follows: * * * *C #ifndef IMG_MPEG2_VLD_STRUCT_ * *C #define IMG_MPEG2_VLD_STRUCT_ 1 * *C * *C typedef struct { * *C unsigned int *bsbuf; // pointer to bitstream buffer * *C unsigned int next_wptr; // next word to read from buffer * *C unsigned int bptr; // bit position within word * *C unsigned int word1; // word aligned buffer * *C unsigned int word2; // word aligned buffer * *C unsigned int top0; // top 32 bits of bitstream * *C unsigned int top1; // next 32 bits of bitstream * *C const unsigned char *scan; // inverse zigzag scan matrix * *C unsigned int intravlc; // intra_vlc_format * *C unsigned int quant_scale; // quant_scale * *C unsigned int dc_prec; // intra_dc_precision * *C unsigned int cbp; // coded_block_pattern * *C unsigned int fault; // fault condition (returned) * *C unsigned int reserved; // reserved * *C } IMG_mpeg2_vld; * *C * *C #endif * * * * The Mpeg2v variables should have a fixed layout since they are * * accessed by this routine. If the layout is changed, the * * corresponding changes have to be made in the assembly code too. * * * * The routine sets the fault flag Mpeg2v.fault to 1 if an invalid * * VLC code was encountered or the total run went beyond 63. In * * theses cases the decoder has to resynchronize. * *S * *S The following files contain the required VLD look-up tables: * *S b14s_tbl.c : run-level VLD table (1152 bytes, Table B-14) * *S b14_len_tbl.c : code word length table (640 bytes, Table B-14) * * * * Before calling the routine the bitstream varaibles in Mpeg2v * * have to be initialized. If bsbuf is a circular buffer and bsptr * * contains the number of bits in the buffer that already have * * been consumed, then next_wptr, bptr, word1, word2, top0 and * * top1 are initialized as follows: * * * * 1. nextwptr: bsptr may not be a multiple of 32, therefore obtain * * the next lower multiple of 32. * * * * next_wptr = (bsptr >> 5); * * * * 2. bptr: bptr is the bit pointer which points to the current * * bit WITHIN the word pointed to by next_wptr. * * * * bptr = bsptr & 31; * * bptr_cmpl = 32 - bptr; * * * * 3. word1 and word2: read next 3 words from the bitstream buffer * * (word0 is a temporary variable). bsbuf_words is the size of the * * bitstream buffer in words. * * * * word0 = bsbuf[next_wptr]; * * next_wptr = (next_wptr + 1) & (bsbuf_words-1); * * * * word1 = bsbuf[next_wptr]; * * next_wptr = (next_wptr + 1) & (bsbuf_words-1); * * * * word2 = bsbuf[next_wptr]; * * next_wptr = (next_wptr + 1) & (bsbuf_words-1); * * * * 4. top0 and top1: Shift words word0, word1, word2 by bptr to the * * left so that the current bit becomes the MSB in word0. word0 can * * simply be shifted by bptr; the then empty LSBs of word0 have to be * * filled with the MSBs of word1. To do that the required MSBs are * * brought into the position of empty LSBs of word0 by shifting word1 * * to the right by (32-bptr). The result is then copied into word0 by * * an addition. Rather than overwriting word0, top0 is used to hold * * the new bit aligned word. The same procedure is used to obtain * * top1. top0 and top1 contain the next 64 bits of the bitstream. * * * * s1 = word0 << bptr; * * s2 = word1 >> bptr_cmpl; // unsigned right-shift // * * top0 = s1 + s2; * * * * s3 = word1 << bptr; * * s4 = word2 >> bptr_cmpl; // unsigned right-shift // * * top1 = s3 + s4; * * * * Note that the routine returns the updated state of the bitstream * * buffer variables, top0, top1, word1, word2, bptr and next_wptr. If * * all other functions which access the bitstream in a decoder system * * maintain the buffer variables in the same way, then the above * * initialization procedure has to be performed only once at the * * beginning. * * * * * * TECHNIQUES * * The instruction NORM is used to detect the number of leading zeros * * or ones in a code word. This value together with additional bits * * extracted from the codeword is then used as an index into look-up * * tables to determine the length, run, level and sign. Escape code * * sequences are directly extracted from the code word. * * * * ASSUMPTIONS * * The bitstream must be stored in memory in 32-bit words which are * * in little endian byte order. * * * * Wptr is allowed to overrun once (to detect total run overrun), so * * maximum overrun that can occur is 66 (Error mark). Therefore, * * in memory 66+1 halfwords behind the weighting matrix should be * * valid (e.g. peripherals). No memory is overwritten, * * only loads occurr. * * * * Note that the AMR register is set to zero on exit. * * * * MEMORY REQUIREMENTS * * 1792 bytes for the lookup tables * * (can be shared with mpeg2_vld_intra) * * * * * * ------------------------------------------------------------------------- ** Copyright (c) 2002 Texas Instruments, Incorporated. ** All Rights Reserved. ** ========================================================================= * .sect ".text:nsa" .global _mpeg2_vld_inter_nsa .global _IMG_len_tbl0 .global _IMG_rld_table0_mpeg2_vld_inter_nsa .cproc Wptr, outi, Mpeg2v, m12Q4, num_blocks, bsbuf_size .no_mdep; A4, B4, A6, B6, A8, B8* ========================================================================= ** Registers* ========================================================================= * .reg block ; block number (0-3: lum, 4, 5: chrom) .reg cnt, cnt_sav .reg flag .reg const1, const24, const31, const32, const63, const128, const36 .reg const65 .reg amr_config, csr_bk, no_gie* ------------------------------------------------------------------------- *
⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -