example 5-8.asm

《基于TI DSP的通用算法实现》程序代码

; Example 5 - 8. FIR Filter Implementation ASM Listing for the TMS320C62x DSP


* ========================================================================= *
*                                                                           *
*     TEXAS INSTRUMENTS, INC.                                               *
*                                                                           *
*     NAME                                                                  *
*           DSP_fir_gen                                                     *
*                                                                           *
*     REVISION DATE                                                         *
*         15-Feb-2002                                                       *
*                                                                           *
*   USAGE                                                                   *
*       This routine is C-callable and can be called as:                    *
*                                                                           *
*       void DSP_fir_gen                                                    *
*       (                                                                   *
*           const short *restrict x,                                        *
*           const short *restrict h,                                        *
*           short       *restrict r,                                        *
*           int nh,                                                         *
*           int nr                                                          *
*       )                                                                   *
*                                                                           *
*       x[nr+nh-1] : Input array                                            *
*       h[nh]      : Coefficient array. Must be in reverse order.           *
*       r[nr]      : Output array                                           *
*       nh         : Number of coefficients. Must be >= 5.                  *
*       nr         : Number of output samples                               *
*                                                                           *
*   DESCRIPTION                                                             *
*       Computes a real FIR filter (direct-form) using coefficients stored  *
*       in vector h. The coefficients have to be arranged in reverse        *
*       order. The real data input is stored in vector x. The filter        *
*       output result is stored in vector r. It operates on 16-bit data     *
*       with a 32-bit accumulate. The filter is nr output samples and nh    *
*       coefficients.                                                       *
*                                                                           *
*   C CODE                                                                  *
*       void DSP_fir_gen                                                    *
*       (                                                                   *
*           const short *restrict x,                                        *
*           const short *restrict h,                                        *
*           short       *restrict r,                                        *
*           int nh,                                                         *
*           int nr                                                          *
*       )                                                                   *
*       {                                                                   *
*           int i, j, sum;                                                  *
*                                                                           *
*           for (j = 0; j < nr; j++)                                        *
*           {                                                               *
*               sum = 0;                                                    *
*               for (i = 0; i < nh; i++)                                    *
*                   sum += x[i + j] * h[i];                                 *
*               r[j] = sum >> 15;                                           *
*           }                                                               *
*       }                                                                   *
*                                                                           *
*   TECHNIQUES                                                              *
*       The inner loop is unrolled four times, but the last three           *
*       accumulates are executed conditionally to allow for a number of     *
*       coefficients that is not a multiple of four. The outer loop is      *
*       unrolled twice, but the last store is executed conditionally to     *
*       allow for a number of output samples that is not a multiple of      *
*       two. Both the inner and outer loops are software pipelined.         *
*                                                                           *
*   ASSUMPTIONS                                                             *
*       nh must be >= 5.                                                    *
*                                                                           *
*   MEMORY NOTE                                                             *
*       No memory bank hits under any conditions.                           *
*                                                                           *
*   NOTES                                                                   *
*      This function is interrupt-tolerant but not interruptible.           *
*      This function is endian neutral.                                     *
*                                                                           *
*   CYCLES                                                                  *
*       (4 * ceil(nh/4) + 9) *  ceil(nr/2) + 18                             *
*                                                                           *
*       For nh = 13, nr = 19: 268 cycles                                    *
*                                                                           *
*   CODESIZE                                                                *
*       640 bytes                                                           *
*                                                                           *
* ------------------------------------------------------------------------- *
*             Copyright (c) 2002 Texas Instruments, Incorporated.           *
*                            All Rights Reserved.                           *
* ========================================================================= *


        .text
        .global _DSP_fir_gen
_DSP_fir_gen:

        STW     .D2T1   A14,*B15--[8]    ; push register (for c-callable func)
||      MV      .L1X    B15,A14

        STW     .D2T1   A13,*+B15[6]    ; push register (for c-callable func)
||      STW     .D1T2   B13,*-A14[4]    ; push register (for c-callable func)
||      SUB     .L2     B6,1,B6         ; used to do last h mults if necessary

        STW     .D1T1   A12,*-A14[3]    ; push register (for c-callable func)
||      STW     .D2T2   B11,*+B15[2]    ; push register (for c-callable func)
||      ADD     .S2     B6,4,B6         ; set up inner loop counter
||      AND     .L2     B6,3,B13        ; used to do last h mults if necessary
||      MVK     .S1     7,A13           ; used to do last h mults if necessary

        STW     .D2T2   B12,*+B15[3]    ; push register (for c-callable func)
||      SHR     .S2     B6,2,B12        ; set up inner loop counter
||      AND     .S1     A8,1,A12        ; used to do last r store if necessary
||      ADD     .L1     A8,1,A8         ; set up outer loop counter
||      XOR     .L2     3,B13,B13       ; used to do last h mults if necessary

        STW     .D1T2   B10,*-A14[7]    ; push register (for c-callable func)
||      ADD     .L2X    A6,2,B11        ; set up pointer to r[1]
||      SHR     .S1     A8,1,A2         ; set up outer loop counter

        LDH     .D1     *A4++,B8        ; x0 = x[j]
||      ADD     .L2X    A4,4,B1         ; set up pointer to x[j+2]
||      ADD     .L1X    B4,2,A8         ; set up pointer to h[1]
||[A2]  SUB     .S1     A2,1,A2         ; decrement outer loop counter

        LDH     .D2     *B1++[2],B0     ; x2 = x[j+i+2]
||      LDH     .D1     *A4++[2],A0     ; x1 = x[j+i+1]
||      SHR     .S2X    A13,B13,B13     ; used to do last h mults if necessary
||      XOR     .S1     A12,1,A12       ; used to do last r store if necessary

        LDH     .D1     *A8++[2],B6     ; h1 = h[i+1]
||      LDH     .D2     *B4++[2],A14    ; h0 = h[i]
||      SHL     .S1     B12,3,A3        ; used to rst x pointer each outer loop
||      SHL     .S2     B12,3,B10       ; used to rst h pointer each outer loop