;******************************************************
;  File: convol1.sa
;   By: S.A. Tretter
;   Date: 03/06/02
;
;   Compile using
;
;    cl6x -mv6713 -o3 convol1.sa
;
;   or by using Code Composer Studio with these options.
;  
;   This is a C callable assembly function for computing
;   one convolution iteration. The circular buffering
;   hardware of the C6000 is used. The function
;   prototype is:
;
;   extern float convolve( float *, float *, int,
;                    int, int);
;
;   The function can be called from C as
;
;       convolve(x, h, Nh, Nblock, newest)
;
;       x[]     circular input sample buffer
;       h[]     FIR filter coefficients
;       Nh      number of filter taps
;       Nblock  circular buffer size in bytes is
;               2^{Nblock+1} and in words is 2^{Nblock-1}
;       newest  index pointing to newest sample in buffer
;
;    According to the TI C Compiler conventions, the 
;    arguments on entry are found in the following
;    registers:
;
;       &x[0]   A4
;       &h[0]   B4
;       Nh      A6
;       Nblock  B6
;       newest  A8
;
;    WARNING:  The C calling function must align the
;     circular buffer, x[], on a boundary that is a
;     multiple of the buffer size in bytes, that is, a
;     multiple of BUF_LEN = 2^{Nblock+1} bytes.  This can 
;     be done by a statement in the C program of the form
;
;      #pragma DATA_ALIGN(x, BUF_LEN)
;
;     The array x[] must be a global array according to 
;     the TI C Compiler manual.
;********************************************************

          .global _convolve
_convolve .cproc  x_addr, h_addr, Nh, Nblock, newest
          .reg  sum, prod, x_value, h_value

;  Compute addressof x[newest] and put in x_addr
;  Note: The instruction ADDAW shifts the second argument,
;        newest, left 2 bits, i.e., multiplies it by 4,
;        before adding it to the first argument to get the
;        byte address of x[newest].

     ADDAW   x_addr, newest, x_addr ; &x[newest]

;-------------------------------------------------------
;  Set up circular addressing
;   Load Nblock into the BK0 field of the Address Mode
;     Register (AMR)

     SHL Nblock, 16, Nblock ; Shift Nblock to BK0 field

;   Note: The assembly optimizer will assign x_addr to
;         some register it likes.  You will have to
;         manually look at the assembled and optimized
;         code to see which register it picked and then
;         set up the circular mode using BK0 by writing
;         01 to the field for that register in AMR.
;         The assembler will give you a warning that
;         changing the AMR can give unpredicatable
;         results but you can ignore this.
;
;         Suppose B4 was chosen by the optimizer.
;
     set Nblock, 8,8, Nblock; Set mode circular, BK0, B4
;    set Nblock, 10,10, Nblock; Use this for B5.    
     MVC Nblock, AMR        ; load mode into AMR
;-------------------------------------------------------

;  Clear convolution sum registers

     ZERO  sum

;  Now compute the convolution sum.

loop:   .trip 8, 500  ; assume between 8 and 500 taps
     LDW *x_addr--, x_value ; x[newest-k] -> x_value
     LDW *h_addr++, h_value ; h[k] -> h_value
     MPYSP  x_value, h_value, prod ; h[k]*x[n-k]
     ADDSP prod, sum, sum ; sum of products

 [Nh] SUB Nh, 1, Nh  ; Decrement count by 1 tap
 [Nh] B  loop   ; Continue until all taps computed

      .return sum  ; By C convention, put sum in A4
      .endproc