marcoheisig

(defun simd-sum (array &aux (n (array-total-size array)))
  "Compute the sum of the elements of the supplied simple double-float ARRAY."
  (declare (type (simple-array double-float 1) array)
           (optimize speed (safety 0)))
  (do ((index 0 (the (integer 0 #.(- array-total-size-limit 16)) (+ index 16)))
       (acc1 (make-f64.4 0 0 0 0) (f64.4+ acc1 (f64.4-row-major-aref array (+ index 0))))
       (acc2 (make-f64.4 0 0 0 0) (f64.4+ acc2 (f64.4-row-major-aref array (+ index 4))))
       (acc3 (make-f64.4 0 0 0 0) (f64.4+ acc3 (f64.4-row-major-aref array (+ index 8))))
       (acc4 (make-f64.4 0 0 0 0) (f64.4+ acc4 (f64.4-row-major-aref array (+ index 12)))))
      ((>= index (- n 16))

marcoheisig

;;; Benchmark results:

(SICL-SEQUENCE::FIND-AUX 1 SHORT-LIST NIL #'EQL NIL 0 NIL NIL) 77.93 nanoseconds
(FIND 1 SHORT-LIST) 46.56 nanoseconds
(SICL-SEQUENCE::FIND-AUX 1 LONG-LIST NIL #'EQL NIL 0 NIL NIL) 90.53 microseconds
(FIND 1 LONG-LIST) 29.86 microseconds
(SICL-SEQUENCE::FIND-AUX 1 SHORT-VECTOR NIL #'EQL NIL 0 NIL NIL) 6.55 microseconds
(FIND 1 SHORT-VECTOR) 42.00 nanoseconds
(SICL-SEQUENCE::FIND-AUX 1 LONG-VECTOR NIL #'EQL NIL 0 NIL NIL) 124.40 microseconds
(FIND 1 LONG-VECTOR) 91.50 microseconds

marcoheisig

(cl:defpackage #:avx2
  (:use :cl)
  (:export #:d4+ #:d4* #:d4ref #:d4set))

(cl:in-package #:avx2)

(sb-c:defknown (d4+ d4*) ((sb-ext:simd-pack-256 double-float)
                          (sb-ext:simd-pack-256 double-float))
    (sb-ext:simd-pack-256 double-float)
    (sb-c:movable sb-c:flushable sb-c:always-translatable)