Re: [Q] svref vs. aref for fixnum arrays

   From: Didier Verna <[email protected]>

   I notice an important performance difference between these two fixnum array
   implementations:
   
   1/ simple-vector / svref
   2/ simple-array / aref
   
   Note that in my test cases, the arrays don't contain anything else than
   fixnums, and the compiler is aware of that (no number comsing at all). Yet,
   the simple-vector method is faster.

Your assertion that the compiler is aware of something is presented
without any evidence.  But you seem not to understand something about
CL types:

  The type simple-vector is equivalent to (simple-array t (*))

  The type simple-array is equivalent to (simple-array * *), and
assuming that the compiler can infer the rank of the array from the
nu8mber of arguments to aref, it is equivalent to (simple-array * (*))

The first of these specifies a particular array type with a known
representational structure.  In particular, the element-type is t.
The second of these specifies all possible rank 1 arrays, including
unspecialized arrays, various specialized byte arrays, complex-number
arrays, strings, and even bit-vectors.  (Each element of a bit-vector
will necessarily be of type fixnum.)

   Is there an implementation difference between the two ? For instance, would
   fixnums be shifted in the simple-array version by any chance ?
   
his is thw wrong question,.  If you want compiled code not to have to
determine the representational type iof an array at run time, you
generally have to tell it the specific representational type at
compile time.  That declaration is implicit in the use of svref
(because svref requires that the array argument be a simple-vector,
which is a simple-array of rank 1 specialized to element-type t) but a
mere simple-array declaration doesn't say anything about the
specialization of element-type.

See http://www.franz.com/support/documentation/8.0/doc/implementation.htm#data-types-1