news@daver.bungi.com (09/13/90)
Using the optimized strcpy.s and strcmp.s that Dave just posted I now get 9250 dhrystones / second. This is up from 8771 and 7692 previously reported using the assembly string instructions and C routines respectively. After converting to compatible assembler format, Dave's routines seem to work fine with one expection. The .align 4 directives must be commented out any place the code falls through to the next section. Instructions beginning with 0x00 do not seem to work very well :-). Best regards, johnc --
dlr@daver.bungi.com (Dave Rand) (09/14/90)
[In the message entitled "Dhrystone 2.1 with improved strcpy / strcmp" on Sep 13, 11:32, John Connin writes:] > After converting to compatible assembler format, Dave's routines > seem to work fine with one expection. The .align 4 directives > must be commented out any place the code falls through to > the next section. Instructions beginning with 0x00 do not seem > to work very well :-). Yeah - the nice part about the NSC assembler is that it "optimizes" the .align's by using a 1, 2 or 3 byte "nop" instruction. I think it uses: 1 nop 2 movb r7,r7 3 orb $0,r7 This is really good on the 532, since all of these instructions execute in 2 clocks - eating up that space quickly. You will notice a win if you at least align the top of each loop... -- Dave Rand {pyramid|mips|bct|vsi1}!daver!dlr Internet: dlr@daver.bungi.com