[Mono-bugs] [Bug 54467][Wis] Changed - Incorrect values when subtracting doubles
Wed, 19 May 2004 09:08:51 -0400 (EDT)
Please do not reply to this email- if you want to comment on the bug, go to the
URL shown below and enter your comments there.
Changed by email@example.com.
--- shadow/54467 2004-05-19 08:51:20.000000000 -0400
+++ shadow/54467.tmp.6000 2004-05-19 09:08:51.000000000 -0400
@@ -87,6 +87,29 @@
affects with NaNs etc.
------- Additional Comments From firstname.lastname@example.org 2004-05-19 08:51 -------
Created an attachment (id=7761)
More test cases that fail (in the form of regression tests)
+------- Additional Comments From email@example.com 2004-05-19 09:08 -------
+After the comment from Paolo, I investigated a bit, and created more
+tests that fail (see previous attachment, at the bottom).
+I have verified that adding 0.0 works, and the fastes way to do it
+seems to be with a fldz and a faddp in sequence.
+They take four bytes in code, but do not touch the memory otherwise,
+and with that correction all the "easy" tests pass.
+On the other hand, they could cause some more spilling (they use a fp
+register). An alternative would be to push a fp 0.0 on the stack (I
+mean the regular stack in memory), and have the faddp use it instead
+of using a fp register.
+This is probably slower, but would avoid any added spill, and would
+therefore be easier to "inline" in the relevant places in mini-x86.c.
+However, I am puzzled by the last test that fails (it seems that fsqrt
+does honor the precision bit, but things get strange with that complex
+expression at the end of the file).
+Moreover, I am going to investigate what will happen with NaNs (I'll
+create more test cases).