bob@reed.UUCP (Bob Ankeney) (08/23/89)
A quick question: Which (if either) is faster in execution:
movw foo(r1),bar(r1)
or
movw foo[r1:b],bar[r1:b]
I don't have a manual with an instruction timing guide handy (and they're
such a pain to use anyway!) Thanx for the help!
Bob Ankeney
---------------------------------------------------------------------------
Bob Ankeney | "Wise men talk because they have |
...!tektronix!reed!bob | something to say; fools, because |
...!tektronix!bob@reed.BITNET | they have to say something." - Plato |
...!percival!bob@agravain.UUCP | |
| "He who talks by the yard and acts by |
| the inch should be kicked by the foot." |
---------------------------------------------------------------------------gideony@microsoft.UUCP (Gideon Yuvall) (08/26/89)
In article <13223@reed.UUCP> bob@reed.UUCP (Bob Ankeney) writes: > > A quick question: Which (if either) is faster in execution: ... A quick answer: compile the equivalent "C",under "-O -S -KC<processor>", using NSC's compilers; you'll get NSC's best guess as to what is fastest. That guess is probably MORE accurate than the manuals, and is certainly no less accurate. -- Gideon Yuval, gideony@microsof.UUCP, 206-882-8080 (fax:206-883-8101;TWX:160520)
kls@ditka.UUCP (Karl Swartz) (08/31/89)
In article <13223@reed.UUCP> bob@reed.UUCP (Bob Ankeney) writes: > > A quick question: Which (if either) is faster in execution: According to my (ancient) manuals, for a 32032 ... movw foo(r1),bar(r1) ; 19 cycles movw foo[r1:b],bar[r1:b] ; 27 cycles This assumes the source and destination words are aligned and there are MMU delays (subtract 2 cycles from each case for no MMU). Plus there's the time to fetch an extra two bytes of instruction for the the indexed case. In other words, index mode is a loser where a simpler addressing mode can be used, certainly in this case, and probably in all cases. -- Karl Swartz |UUCP uunet!lll-winken!ames!hc!rt1!ditka!kls 1-505/667-7777 (work) |Internet kls@rt1.lanl.gov 1-505/672-3113 (home) |BIX kswartz "I never let my schooling get in the way of my education." (Twain)
george@wombat.UUCP (George Scolaro) (09/03/89)
In article <3878@ditka.UUCP> kls@ditka.UUCP (Karl Swartz) writes: >In article <13223@reed.UUCP> bob@reed.UUCP (Bob Ankeney) writes: >> >> A quick question: Which (if either) is faster in execution: > >According to my (ancient) manuals, for a 32032 ... > > movw foo(r1),bar(r1) ; 19 cycles > movw foo[r1:b],bar[r1:b] ; 27 cycles On a 32532 (or 32gx32) reading the 32gx32 data 'sheet' the timing is: movw foo(r1),bar(r1) ; 4 cycles (5 x 32032) movw foo[r1:b],bar[r1:b] ; 8 cycles (3.5 x 32032) The 32532 has hardware for the addressing modes but the following still add to the overall instruction time: Memory relative 3 clocks (bit cisc) External (yuk) 8 clocks (very cisc) Scaled Indexing 2 clocks (little cisc) Note: the above times are added for source and or destination. -- George Scolaro george@wombat (try {pyramid|sun|vsi1|killer} !daver!wombat!george)