Optimize matrix operations.

Guile's array procedures are slow, so switch to using raw bytevectors
instead.  Additionally, converting the bytevector to a pointer is also
slow, so the pointer is now cached for redundant calls.

There's more to be done, but this was low hanging fruit.

* sly/math/transform.scm (<transform>)[ptr]: New field.
  (transform-ptr): New accessor.
  (set-transform-ptr!): New setter.
  (make-4x4-matrix): Implement using bytevectors.
  (matrix-set!, matrix-ref): New procedures.
  (make-transform, transform-vector2, transform-position): Use
  'matrix-set!' and 'matrix-ref'.
  (transpose, transform+, transform*): Use new transform constructor.
  (transform->pointer): Cache pointer.
* sly/render/context.scm (copy-transform!): Use 'bytevector-copy!'
* sly/render/shader.scm: Use 'transform->pointer' in the transform
  uniform setter.

0 files changed, 0 insertions, 0 deletions