h263_vld_h.asm

h263,jpeg,mpeg2编解码核心程序(TI DSP C64xx)

*       The code word used in this context refers to the run, level         *
*       associated with each Huffman code word that is decoded and not      *
*       to the Huffman code word itself.                                    *
*                                                                           *
*          Last(1/0)  RUNVALUE   LEVEL                                      *
*          14         13....8    7....0                                     *
*                                                                           *
*       Each code word thus has a pad of 2 bits and is stored as a half     *
*       word                                                                *
*                                                                           *
*       Example:                                                            *
*                                                                           *
*          1) Code: 0000 0000 110s Run: 0   Level: 11                       *
*                                                                           *
*       Clearly for this code f5 the first 5 bits from the left are zero.   *
*       first 12 bits from the left are extracted in idx1. If the upper     *
*       5 bits are zero then idx1 can have a maximum value of upto 127.     *
*       In this case this value will be either 13 or 14 depending on the    *
*       sign bit.                                                           *
*                                                                           *
*          0         000000 001011  ------------->   0x0b                   *
*                                                                           *
*       The length table is a set of values constructed as follows. The     *
*       code words are taken and divided into two categories for this       *
*       purpose:                                                            *
*                   Class 1:  Codes 00000xxxxx...00011xxxxx                 *
*                   Class 2:  Codes 001xxxxx...1xxxxx                       *
*       Therfore it can be seen that under class 2 there can be a set of    *
*       32 codes and under class 1 there are 60 enties that the given       *
*       code table maps into. Therfore there are a total of 32 entries.     *
*       This can be seen by way of example:                                 *
*                                                                           *
*       Code at Index: 50 of the Table 16 in H.263 spec                     *
*                                                                           *
*       Run: 19 Level: 1 Code: 0000 1110 1s and falls under Class 1         *
*       When the 9 left most bits are examined this gives us 29.  The       *
*       29th entry in the length table gives us the length of the code      *
*       word as 10.                                                         *
*                                                                           *
*       Similarly a code at Index 40 of the Table 16 in H.263 spec          *
*                                                                           *
*       Run: 10 Level: 1 Code: 0010110s. This comes under Class 2 and       *
*       hence the 5 left most bits are extracted to give an index of 5      *
*       The code lengths for these class of codes is stored with an         *
*       implicit offset of 60 and hence this code's length is actually      *
*       stored at location 65. The 65 th entry from the length table is     *
*       8 and corresponds to the length of the code.                        *
*                                                                           *
*       The entries in the Zig-Zag table are expected to be in the range    *
*       0 .. 63, with no repeated entries.  The VLD starts reading the      *
*       Zig-Zag table at zz[1], and under normal circumstances will not     *
*       read past element zz[63].  However, when an invalid bitstream is    *
*       encountered, elements as high as zz[126] may be accessed.  See      *
*       the MEMORY NOTE below for more information.                         *
*                                                                           *
*   ASSUMPTIONS                                                             *
*       Speculative reads are OK.  See MEMORY NOTE below.                   *
*                                                                           *
*       Short (as in, invalid due to truncation) bitstreams are followed    *
*       by sentinel values to halt VLD with an error condition, to prevent  *
*       underflow.                                                          *
*                                                                           *
*   MEMORY NOTE                                                             *
*       The VLD routine will read up to two full words beyond the final     *
*       position of the bitstream word pointer, as part of its lookahead.   *
*       The user must ensure that two words of valid memory exist there.    *
*                                                                           *
*       Speculative reads are performed to each of the following arrays:    *
*       (Ranges are in bytes, and are inclusive.)                           *
*                                                                           *
*          Table/Array            Valid Range   Accessed Range              *
*                                                                           *
*          Zig-Zag Table            0..63           0..127                  *
*          Length Table             0..91           0..511                  *
*          Coef/Run Table           0..607          0..607                  *
*                                                                           *
*       The user should ensure that valid (eg. CPU readable) memory         *
*       exists in the required ranges past the end of each table.           *
*                                                                           *
*       Alternately, to ensure that these reads do not reduce cache         *
*       effectiveness and do not access invalid memory, the following       *
*       table layout is recommended, as it causes all speculative reads     *
*       to land inside other valid table entries:                           *
*                                                                           *
*          Bytes   0 .. 63    'zz'    Zig-Zag table                         *
*          Bytes  64 .. 159   'len'   Huffman-code Length table             *
*          Bytes 160 .. 767   'tab'   Huffman-code Coefficient/Run Table    *
*                                                                           *
*       No bank conflicts occur in this code at any time.                   *
*                                                                           *
*   TECHNIQUES                                                              *
*       The branch back to the main code and the small set of               *
*       operations therin are scheduled right within the kernel to          *
*       minimize code size and increase execution speed.                    *
*                                                                           *
*   NOTES                                                                   *
*       The input bitstream must be "word order", meaning that it is        *
*       accessed with word-wide memory accesses.                            *
*                                                                           *
*       This code is fully interruptible and does not mask interrupts.      *
*                                                                           *
*   CYCLES                                                                  *
*       cycles = 12 * N + 30, including 6 cycles of call overhead.          *
*       (N is the number of symbols in the block, including EOB.)           *
*                                                                           *
*       For N = 15, cycles = 210.                                           *
*       For N = 25, cycles = 330.                                           *
*                                                                           *
*   CODESIZE                                                                *
*       312 bytes.                                                          *
*                                                                           *
* ------------------------------------------------------------------------- *
*             Copyright (c) 2000 Texas Instruments, Incorporated.           *
*                            All Rights Reserved.                           *
* ========================================================================= *

                .sect ".text:hand"
                .global  _h263_vld_asm
_h263_vld_asm:
* =============== SYMBOLIC REGISTER ASSIGNMENTS:  ARGUMENTS =============== *
            .asg            A4,         A_bufPtr    ; Ptr to ptr to buffer
            .asg            B4,         B_bitPtr    ; Ptr to bit-pointer
            .asg            A6,         A_tab       ; VLD code table ptr
            .asg            B6,         B_len       ; VLD length table ptr
            .asg            A8,         A_Q         ; Q value for IQuant
            .asg            B8,         B_dq        ; Delta-Q value for IQ
            .asg            A10,        A_zz_in     ; Zig-zag table ptr
            .asg            B10,        B_idct      ; IDCT block pointer
            .asg            A12,        A_inter     ; INTER vs INTRA flag
            .asg            B3,         B_ret       ; Return address
            .asg            A4,         A_err_ret   ; Error code return val
* ===================== SYMBOLIC REGISTER ASSIGNMENTS ===================== *
            .asg            A0,         A_EOB       ; End of Block flag
            .asg            A1,         A_s_tmp     ; Sign bit temp-val
            .asg            A1,         A_tmp       ; Shift amt for sign-bit
            .asg            A1,         A_err       ; Error code
            .asg            A2,         A_err_      ; Error code
            .asg            A2,         A_s         ; Sign bit
            .asg            A2,         A_nbit      ; Number of bits (copy)
            .asg            A3,         A_zz_end    ; End of zig-zag tbl.
            .asg            A5,         A_bit25     ; Constant: 1 << 25
            .asg            A7,         A_levelQ    ; Inv. Quant'd level
            .asg            A9,         A_idx_      ; Table index
            .asg            A9,         A_level     ; Decoded level
            .asg            A16,        A_idx       ; Table index
            .asg            A16,        A_lrl       ; Last/Run/Lvl from tbl
            .asg            A17,        A_vld_buf   ; Top 32 bits of stream