首页 > 代码库 > sharc dsp 学习记录1---2014-07-30

sharc dsp 学习记录1---2014-07-30

从今天开始记录学习sharc dsp过程中的点点滴滴吧。

 

image

DPI:Digital Peripheral Interface

DAI:Digital Audio Interface

 

SHARC SIMD Core

SIMD 单指令多数据

ADSP-2148x包含两个用作单指令、多数据(SIMD)引擎的计算处理器元件,分别称为PEX和PEY,各元件均由ALU、乘法器、移位器和寄存器文件组成。

PEX始终有效,PEY可通过将MODE1寄存器的PEYEN模式位设为1来使能。SIMD模式允许处理器在两个处理元件中执行同一指令,但各处理元件处理不

同的数据。这种架构对于执行计算密集型DSP算法非常有效

 

SIMD Enabled:

When any computation or data access is executed, it will be performed automatically in both processing elements

F0 = F1 + F2; explicitly defined in source code Will execute in PEx

S0 = S1 + S2; implicit operation not defined in source code Will be automatically executed in PEy in the same instruction cycle

 

SIMD Disabled (SISD Mode):

Only the explicit instruction will be executed in PEx

image

 

PM地址总线的宽度 = 24bits 。可最多访问16Mbit的程序/数据。

PM数据总线的宽度 = 48bits。 用以存取48 位字长的指令,当用来存放数据时,32 位单精度浮点数或32位定点数将位于48 位的高32 位中

 

48位指令字支持各种并行操作,可实现简练编程。例如,处理器可以有条件地在两个处理元件中执行乘法、加法和减法,同时进行分支并从存储

器获取最多4个32位数据值,所有这些只需一个指令。

 

 

image

 

 

汇编程序分析(FIR)


#include    "def21364.h"                    /* Symbol Definition File */

#define     TAPS    64                        /* length of filter */
#define     N       128                        /* number of samples */

.extern    ss_fir;

/* DM data */
.section/dm seg_dmda;                      /* Segments are declared in the .ldf file*/
.ALIGN 2;                                         /* Set alignment to long-word boundary for next variable */
.var    dline[TAPS+1];                        /* delay line compensate for circ buffer, see comments in SS_FIR.asm */
.ALIGN 2;
.var    input[N] = "input.dat";            /* array of samples */


/* PM data */
.section/dm seg_pmda;
.ALIGN 2;
.var    coeffs[TAPS] = "ssfcoeffs.dat";        /* Filter coefficients */
.ALIGN 2;
.var    output[N];                                    /* Output array */


/* PM interrupt vector code */
.section/pm   seg_rth;
Reserved_1:     rti; nop; nop; nop;
Chip_Reset:     idle; jump start; nop; nop;


/* program memory code */
.segment/pm seg_pmco;

start:
/*--------------------- Setup modify registers for arrays --------------------*/
    m1 = 1;
    m2 = -1;
    m3 = 2;

      m9 = 2;
    m10= 1;
/*----------------------------------------------------------------------------*/

 

/*--------------------- Initialization delay line ----------------------------*/
    b0 = dline;
    l0 = @dline-1;

    f8=0.0;
    lcntr = TAPS, do clear_fir until lce;
clear_fir:  dm(i0,m1) = f8;

    i0 = dline;
/*----------------------------------------------------------------------------*/

 

/*-------------- Setup DAGs for input/output/coeffs and call ss_fir ----------*/
    r3 = TAPS;
    r3 = lshift r3 by -1;                     /* r3 = taps/2 due to SIMD mode */
    r0 = 3;                                      /* 3 macs outside of fir mac loop */
    r3 = r3 - r0;                              /* r3 = taps/2 - 3 for fir mac loop counter */
   
    b1 = input;
    l1 = 0;

    b9 = output;
    l9 = 0;

    b8 = coeffs;
    l8 = @coeffs;
   
lcntr = N, do fir_loop until lce;
            call ss_fir (db);                /* Call fir */
            f0 = dm(i1,m1);              /* Read one sample */
            nop;                              /* Call can‘t be in last three locations of a loop */
fir_loop:    pm(i9,m10)=f8;          /* Write result to output */
/*----------------------------------------------------------------------------*/
 
     

/* Terminate and wait */
wait1:    idle;
        jump wait1;

 

 

#include    "def21364.h"                /* Symbol Definition File */

.global ss_fir;

/* program memory code */
.section/pm seg_pmco;

ss_fir:
    bit set MODE1 CBUFEN;               /* Circular Buffer Enable, one cycle effect latency */
    nop;                                          /* Circular Buffering not in effect until next cycle */
   
    s0 = dm(i0, m1);                        /* move pointer to delay[1] */
   
    bit set MODE1 PEYEN;                 /* SIMD Mode Enable, one cycle effect latency */
    s0 = dm(i0, m2);                        /* load s0 with the value of delay[1] for SIMD store, move pointer to delay[0] */
   
    dm(i0,m3)=f0, f4 = pm(i8,m9);    /* transfer sample to delayline, done in SIMD to load end of buffer + 1 */
                                                     /* to compensate for circular buffer issue described above, read 2 coeffs */                                       

        f8=f0*f4, f0=dm(i0,m3), f4=pm(i8,m9);                          /* samples * coeffs, read 2 samples, read 2 coeffs */
        f12=f0*f4, f0=dm(i0,m3), f4=pm(i8,m9);                        /* samples * coeffs, read 2 samples, read 2 coeffs */
        lcntr=r3, do macs until lce;                                             /* FIR loop */
macs:f12=f0*f4, f8=f8+f12, f0=dm(i0,m3), f4=pm(i8,m9);       /* samples * coeffs, accum, read 2 samples, read 2 coeffs */
        f12=f0*f4, f8=f8+f12, s0=dm(i0,m2);                            /* samples * coeffs, accum, dummy read to move pointer to oldest sample */
        f8=f8+f12;                                                                   /* final SIMD accum */
        r12=s8;                                                                       /* move PEy total into PEx register file */
       
    rts (db);
    bit clr MODE1 CBUFEN | PEYEN;                                         /* Circular Buffer Disable, SIMD Mode Disable */
    f8=f8+f12;                                                                      /* last accum */

 

dline的读写

image

红色表示进入macs 循环前 i0指针指向的位置。

 

该程序使用了循环寻址的方式

Bx、Lx、Ix 三个寄存器的序号x 必须一致,而Mx 寄存器可以在同一个DAG 组中任意选取

R3 = lshift R3 by -1;  //右移1bit