(simple-type-error ()
(return-from bug-407b :type-error)))))
(assert (eq :type-error (bug-407b)))
+
+;; 1.0.29.44 introduces a ton of changes for complex floats
+;; on x86-64. Huge test of doom to help catch weird corner
+;; cases.
+;; Abuse the framework to also test some float arithmetic
+;; changes wrt constant arguments in 1.0.29.54.
+(defmacro def-compute (name real-type
+ &optional (complex-type `(complex ,real-type)))
+ `(defun ,name (x y r)
+ (declare (type ,complex-type x y)
+ (type ,real-type r))
+ (flet ((reflections (x)
+ (values x
+ (conjugate x)
+ (complex (- (realpart x)) (imagpart x))
+ (- x)))
+ (compute (x y r)
+ (declare (type ,complex-type x y)
+ (type ,real-type r))
+ (list (1+ x) (* 2 x) (/ x 2) (= 1 x)
+ (+ x y) (+ r x) (+ x r)
+ (- x y) (- r x) (- x r)
+ (* x y) (* x r) (* r x)
+ (unless (zerop y)
+ (/ x y))
+ (unless (zerop r)
+ (/ x r))
+ (unless (zerop x)
+ (/ r x))
+ (conjugate x) (conjugate r)
+ (abs r) (- r) (= 1 r)
+ (- x) (1+ r) (* 2 r) (/ r 2)
+ (complex r) (complex r r) (complex 0 r)
+ (= x y) (= r x) (= y r) (= x (complex 0 r))
+ (= r (realpart x)) (= (realpart x) r)
+ (> r (realpart x)) (< r (realpart x))
+ (> (realpart x) r) (< (realpart x) r)
+ (eql x y) (eql x (complex r)) (eql y (complex r))
+ (eql x (complex r r)) (eql y (complex 0 r))
+ (eql r (realpart x)) (eql (realpart x) r))))
+ (declare (inline reflections))
+ (multiple-value-bind (x1 x2 x3 x4) (reflections x)
+ (multiple-value-bind (y1 y2 y3 y4) (reflections y)
+ #.(let ((form '(list)))
+ (dolist (x '(x1 x2 x3 x4) (reverse form))
+ (dolist (y '(y1 y2 y3 y4))
+ (push `(list ,x ,y r
+ (append (compute ,x ,y r)
+ (compute ,x ,y (- r))))
+ form)))))))))
+
+(def-compute compute-number real number)
+(def-compute compute-single single-float)
+(def-compute compute-double double-float)
+
+(labels ((equal-enough (x y)
+ (cond ((eql x y))
+ ((or (complexp x)
+ (complexp y))
+ (or (eql (coerce x '(complex double-float))
+ (coerce y '(complex double-float)))
+ (and (equal-enough (realpart x) (realpart y))
+ (equal-enough (imagpart x) (imagpart y)))))
+ ((numberp x)
+ (or (eql (coerce x 'double-float) (coerce y 'double-float))
+ (< (abs (- x y)) 1d-5))))))
+ (let* ((reals '(0 1 2))
+ (complexes '#.(let ((reals '(0 1 2))
+ (cpx '()))
+ (dolist (x reals (nreverse cpx))
+ (dolist (y reals)
+ (push (complex x y) cpx))))))
+ (declare (notinline every))
+ (dolist (r reals)
+ (dolist (x complexes)
+ (dolist (y complexes)
+ (let ((value (compute-number x y r))
+ (single (compute-single (coerce x '(complex single-float))
+ (coerce y '(complex single-float))
+ (coerce r 'single-float)))
+ (double (compute-double (coerce x '(complex double-float))
+ (coerce y '(complex double-float))
+ (coerce r 'double-float))))
+ (assert (every (lambda (pos ref single double)
+ (declare (ignorable pos))
+ (every (lambda (ref single double)
+ (or (and (equal-enough ref single)
+ (equal-enough ref double))
+ (and (not (numberp single)) ;; -ve 0s
+ (equal-enough single double))))
+ (fourth ref) (fourth single) (fourth double)))
+ '((0 0) (0 1) (0 2) (0 3)
+ (1 0) (1 1) (1 2) (1 3)
+ (2 0) (2 1) (2 2) (2 3)
+ (3 0) (3 1) (3 2) (3 3))
+ value single double))))))))
(assert (eql 0.0d0 (funcall f 123.0d0 0.0)))
(assert (eql 0.0d0 (funcall f 123.0d0 0.0d0)))
(assert (eql 0.0d0 (funcall f 123.0 0.0d0)))))
-
-;; 1.0.29.44 introduces a ton of changes for complex floats
-;; on x86-64. Huge test of doom to help catch weird corner
-;; cases.
-;; Abuse the framework to also test some float arithmetic
-;; changes wrt constant arguments in 1.0.29.54.
-(with-test (:name :float-arithmetic)
- (labels ((equal-enough (x y)
- (cond ((eql x y))
- ((or (complexp x)
- (complexp y))
- (or (eql (coerce x '(complex double-float))
- (coerce y '(complex double-float)))
- (and (equal-enough (realpart x) (realpart y))
- (equal-enough (imagpart x) (imagpart y)))))
- ((numberp x)
- (or (eql (coerce x 'double-float) (coerce y 'double-float))
- (< (abs (- x y)) 1d-5)))))
- (reflections (x)
- (values x
- (conjugate x)
- (complex (- (realpart x)) (imagpart x))
- (- x)))
- (compute (x y r)
- (list (1+ x) (* 2 x) (/ x 2) (= 1 x)
- (+ x y) (+ r x) (+ x r)
- (- x y) (- r x) (- x r)
- (* x y) (* x r) (* r x)
- (unless (zerop y)
- (/ x y))
- (unless (zerop r)
- (/ x r))
- (unless (zerop x)
- (/ r x))
- (conjugate x) (conjugate r)
- (abs r) (- r) (= 1 r)
- (- x) (1+ r) (* 2 r) (/ r 2)
- (complex r) (complex r r) (complex 0 r)
- (= x y) (= r x) (= y r) (= x (complex 0 r))
- (= r (realpart x)) (= (realpart x) r)
- (> r (realpart x)) (< r (realpart x))
- (> (realpart x) r) (< (realpart x) r)
- (eql x y) (eql x (complex r)) (eql y (complex r))
- (eql x (complex r r)) (eql y (complex 0 r))
- (eql r (realpart x)) (eql (realpart x) r)))
- (compute-all (x y r)
- (multiple-value-bind (x1 x2 x3 x4) (reflections x)
- (multiple-value-bind (y1 y2 y3 y4) (reflections y)
- #.(let ((form '(list)))
- (dolist (x '(x1 x2 x3 x4) (reverse form))
- (dolist (y '(y1 y2 y3 y4))
- (push `(list ,x ,y r
- (append (compute ,x ,y r)
- (compute ,x ,y (- r))))
- form))))))))
- (declare (inline reflections compute compute-all))
- (let* ((reals '(0 1 2))
- (complexes '#.(let ((reals '(0 1 2))
- (cpx '()))
- (dolist (x reals (nreverse cpx))
- (dolist (y reals)
- (push (complex x y) cpx)))))
- (val ()))
- (declare (notinline every))
- (dolist (r reals (nreverse val))
- (dolist (x complexes)
- (dolist (y complexes)
- (let ((value (compute-all x y r))
- (single (compute-all (coerce x '(complex single-float))
- (coerce y '(complex single-float))
- (coerce r 'single-float)))
- (double (compute-all (coerce x '(complex double-float))
- (coerce y '(complex double-float))
- (coerce r 'double-float))))
- (assert (every (lambda (pos ref single double)
- (declare (ignorable pos))
- (every (lambda (ref single double)
- (or (and (equal-enough ref single)
- (equal-enough ref double))
- (and (not (numberp single)) ;; -ve 0s
- (equal-enough single double))))
- (fourth ref) (fourth single) (fourth double)))
- '((0 0) (0 1) (0 2) (0 3)
- (1 0) (1 1) (1 2) (1 3)
- (2 0) (2 1) (2 2) (2 3)
- (3 0) (3 1) (3 2) (3 3))
- value single double)))))))))