> Numerically, op< is less expensive than op<=.  Assuming that a fixnum
> (lambda (x y) (< x y)) compiles to a signed-comparison sequence like
> (x86)
>
> MOV ax, [esp-4]    ; y
> SUB ax, [esp-8]    ; (< x y)
> JL g01
> MOV ax, BOOLEAN_T  ; #t
> JMP g02
> g01:
> MOV ax, BOOLEAN_F  ; #f
> g02:
>
> On the x86 and ARM, JL (BLT) tests two flag bits, whereas JLE (BLE)
> tests three.  (For unsigned comparisons, JB (BCC) tests one flag bit,
> whereas JBE (BLS) tests two.)

Ahem...
Do you really think this makes a measurable difference in execution speed?

cheers,
felix