zlib/intel/adler32vec.s

16 	adler and sum2 (in 32-bit representation), will never overflow for the next NMAX=5552 bytes. This simplifies the
19 		for (i=0;i<N;i+=NMAX) {
20 			for (k=0;k<NMAX;k++) {
30 			for (k=0;k<NMAX;k++) {
56 	For i386, NMAX/16 = 347, whenever possible (NMAX-bytes block), it calls 173 times of macro code DO32 (K=32),
59 	For x86_64 (where more xmm registers are available), NMAX/64 = 86, whenever possible (NMAX-bytes block),
75 #define NMAX 5552 	/* NMAX is the largest n such that 255n(n+1)/2 + (n+1)(BASE-1) <= 2^32-1 */
79 //    while (len >= NMAX) {
80 //        len -= NMAX;
81 //        n = NMAX / 16;          /* NMAX is divisible by 16 */
267 	cmpl	$NMAX, len				// len vs NMAX
268 	jl		len_lessthan_NMAX		// if (len < NMAX), skip the following NMAX batches processing
270 len_ge_NMAX_loop:					// while (len>=NMAX) {
272 	subl	$NMAX, len				// 		len -= NMAX
273 	movl	$(NMAX/32), %eax		// 		n = NMAX/32
281 	cmpl	$NMAX, len				//
282 	jge		len_ge_NMAX_loop		// }	/* len>=NMAX */
361 	cmpl	$NMAX, len				// len vs NMAX
362 	jl		3f						// if (len < NMAX), skip the following NMAX batches processing
364 0:									// while (len>=NMAX) {
366 	subl	$NMAX, len				// 		len -= NMAX
367 	movl	$(NMAX/16), %eax		// 		n = NMAX/16
376 	cmpl	$NMAX, len				//
377 	jge		0b						// }	/* len>=NMAX */
564 	cmp		$NMAX, len				// len vs NMAX
565 	jl		3f						// if (len < NMAX), skip the following NMAX batches processing
567 0:									// while (len>=NMAX) {
569 	sub		$NMAX, len				// 		len -= NMAX
570 	mov		$(NMAX/16), %eax		// 		n = NMAX/16
579 	cmp		$NMAX, len				//
580 	jge		0b						// }	/* len>=NMAX */
881 	cmpq	$NMAX, len				// len vs NMAX
882 	jl		len_lessthan_NMAX		// if (len < NMAX), skip the following NMAX batches processing
884 len_ge_NMAX_loop:					// while (len>=NMAX) {
886 	subq	$NMAX, len				// 		len -= NMAX
887 	movq	$(NMAX/64), %rax		// 		n = NMAX/64
898 	cmpq	$NMAX, len				//
899 	jge		len_ge_NMAX_loop		// }	/* len>=NMAX */