Changeset 5032

Timestamp:

Aug 26, 2006, 2:08:55 AM (18 years ago)

Author:

Gary Byers

Message:

Check fpu overflow bit (not underflow) when deciding whether or not
to signal overflow.

File:

• trunk/ccl/lib/numbers.lisp (modified) (3 diffs)

trunk/ccl/lib/numbers.lisp

@@ -121 +121 @@
          (expt (setq expt (%i+ expt (* esign eexp))))
          (t (return-from parse-float nil)))))
-    ; if ppc read s as double vs error
     (fide sign integer expt (subtypep type 'short-float))))
 
@@ -243 +242 @@
                (error 'floating-point-underflow
                       :operation 'scale
-                      :operand (list sign integer power-of-10)))
-             (if (get-fpu-mode :underflow)
+                      :operands (list sign integer power-of-10)))
+             (if (get-fpu-mode :overflow)
                (error 'floating-point-overflow
                       :operation 'scale
-                      :operand (list sign integer power-of-10))))
+                      :operands (list sign integer power-of-10))))
            (return-from try-harder
              (if under
@@ -265 +264 @@
             (when (< poo -335) (ovf t))
             ;; this case occurs if 600+ digits 
-            (when (> poo 335)(ovf))))
+            (when (> poo 335) (ovf))))
         (let* ((divisor (5-to-e (- power-of-10)))
                ;; make sure we will have enough bits in the quotient