Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss
MIPS/MFLOPS and the Transputer
70 views
Skip to first unread message
Mike B.
Sep 27, 2012, 9:42:13 AM9/27/12
to
During a transputer performance measurement meditation session, I
found some oblique points which I want to share with you.
MIPS = Million Instructions Per Second.
On the transputer, not every instruction is really an fully-fledged
instruction. It would be pointless to count the pfix/nfix instruction.
Every “operate” instruction itself is worthy of being counted except
one – the “fpuentry” instruction (and therefor all fpu* mnemonics).
MFLOPS = Million FLoating-point Operations Per Second
For me - beyond all question are the following instructions:
fpadd, fpsub, fpmul, fpdiv
fpusqrtfirst, fpremfirst
fpumulby2, fpudivby2
fpuexpinc32,fpuexpdec32
fpldnladdsn, fpldnladddb
fpldnlmulsn, fpldnlmuldb
I’m unsure about the comparison. Is this really an operation?
fpuabs, fpgt, fpeq, fpordered
Outside are all load/store and conversion functions, even if they are
far from easy and fast.
Comments are welcome!
-Mike
found some oblique points which I want to share with you.
MIPS = Million Instructions Per Second.
On the transputer, not every instruction is really an fully-fledged
instruction. It would be pointless to count the pfix/nfix instruction.
Every “operate” instruction itself is worthy of being counted except
one – the “fpuentry” instruction (and therefor all fpu* mnemonics).
MFLOPS = Million FLoating-point Operations Per Second
For me - beyond all question are the following instructions:
fpadd, fpsub, fpmul, fpdiv
fpusqrtfirst, fpremfirst
fpumulby2, fpudivby2
fpuexpinc32,fpuexpdec32
fpldnladdsn, fpldnladddb
fpldnlmulsn, fpldnlmuldb
I’m unsure about the comparison. Is this really an operation?
fpuabs, fpgt, fpeq, fpordered
Outside are all load/store and conversion functions, even if they are
far from easy and fast.
Comments are welcome!
-Mike
0 new messages