andk.asm

包含几个高效的矢量运算的数学函数

	.include "ccall.asm"

	.def _ti_andk
_ti_andk:
	pre_ccall 4,AR_X,AR_K,AR_Z,AR_N
	
	; Local variables
	; N%8 - stack[0]
	; N/8 - stack[1]
	

; Compute N%7 to make N divisible by 8 in the unrolled loop	
	sar AR_N,*			; stack[0]=N         
	lacl *+				; acc=N
	and #07h			; acc&=0111b - Unroll 8 iterations
	sacl *+				; stack[1] = N % 0111b
	sar AR_K,*			; stack[2] = k
	sbrk #2
	lacc *,13			; acch=N>>3
	sach *+				; stack[0]=N>>3
	lar AR_N,*			; AR_N = stack[1] = N%7
	mar *+,AR_N			; SP=stack[2]
	banz NEXT0,AR_X		; Skip loop if AR_N = 0
	b NEXT1

NEXT0:
	; z[AR_Z++]=x[AR_X++]*y[AR_K++]
	lacl *+,AR_STACK 		; TREG = x[AR_X++];
	and *,AR_Z 			; PREG = TREG * y[AR_K++];
	sacl *+,AR_N 		; z[AR_Z++] = PREG >> 16;
		
	; while(AR_N--)
	banz NEXT0,AR_X

NEXT1:
	mar *,AR_STACK
	sbrk #2				; SP=stack[0]
	lar AR_N,*+	 		; AR_N = stack[0]=N>>3 
	mar *+,AR_N			; SP=stack[2]
	banz NEXT2,AR_X		; Go to 8x loop if AR_X > 0
	b DONE

NEXT2:
	; z[AR_Z++]=x[AR_X++]*y[AR_K++]
	.loop 7
	lacl *+,AR_STACK			; TREG = x[AR_X++];
	and *,AR_Z 			; PREG = TREG * y[AR_K++];
	sacl *+,AR_X 		; z[AR_Z++] = PREG >> 16;
	.endloop
	lacl *+,AR_STACK 		; TREG = x[AR_X++];
	and *,AR_Z 			; PREG = TREG * y[AR_K++];
	sacl *+,AR_N 		; z[AR_Z++] = PREG >> 16;
	; while(AR_N--)
	banz NEXT2,AR_X
DONE:
	post_ccall 4,2