第三章 DSP的指令系统.ppt

LMS Xmem, Ymem 550000+8000+A=0077CD0888h stores result in A 55hxAAh+b send to B 最小均方运算LMS 基于LMS的自适应滤波程序 数据归一化相关指令EXP、NORM 归一化一个数是先求其指数，然后把它调整到最大精度格式。实现方法如下： EXP src LEADING BIT IS 33, 33-8=25 This instruction computes the exponent value, which is a signed 2s-complement value in the –8 to 31 range, and stores the result in T. The exponent is computed by calculating the number of leading bits in src and subtracting 8 from this value. The number of leading bits is equivalent to the number of left shifts needed to eliminate the significant bits from the 40-bit src with the exception of the sign bit. The src is not modified after this instruction. The result of subtracting 8 from the number of leading bits produces a negative exponent for accumulator values that have significant bits in the guard bits (the eight MSBs of the accumulator used in error detection and correction). See the normalization instruction NORM src [, dst] The signed number contained in src is normalized and the value is stored in dst or src, if dst is not specified. Normalizing a fixed-point number separates the number into a mantissa and an exponent by finding the magnitude of the sign-extended number. 1: ST src, Ymem || LD Xmem, dst 2: ST src, Ymem || LD Xmem, T STRCD Xmem, cond If the condition is true, this instruction stores the content of T into the datamemory location Xmem. If the condition is false, the instruction reads Xmem and writes the value in Xmem back to the same address; thus, Xmem remains the same. Regardless of the condition, Xmem is always read and updated. If A=T, T save to (ar5) BACK 并行装入和乘法指令 dst_: If dst = A, then dst_ = B; if dst = B, then dst_ = A LD Xmem, dst || MAC[R] Ymem [ , dst_ ] This instruction loads the high part of dst (bits 31–16) with a 16-bit dual datamemory operand Xmem shifted left 16-bits. In parallel, this instruction multiplies a dual data-memory operand Ymem by the content of T, adds the result of the multiplication to dst_, and stores th