[Ada] Fix off-by-one bug in underflow handling of Scaling

Message ID 20210506075812.GA125647@adacore.com
State New
Headers show
Series
  • [Ada] Fix off-by-one bug in underflow handling of Scaling
Related show

Commit Message

Pierre-Marie de Rodat May 6, 2021, 7:58 a.m.
Gradual underflow needs to be used for one more exponent.

Tested on x86_64-pc-linux-gnu, committed on trunk

gcc/ada/

	* libgnat/s-fatgen.adb (Scaling): Fix off-by-one bug for underflow.

Patch

diff --git a/gcc/ada/libgnat/s-fatgen.adb b/gcc/ada/libgnat/s-fatgen.adb
--- a/gcc/ada/libgnat/s-fatgen.adb
+++ b/gcc/ada/libgnat/s-fatgen.adb
@@ -784,7 +784,7 @@  package body System.Fat_Gen is
             --  Check for gradual underflow
 
             if T'Denorm
-              and then Adjustment >= IEEE_Emin - (Mantissa - 1) - Exp
+              and then Adjustment >= IEEE_Emin - Mantissa - Exp
             then
                Expf := IEEE_Emin;
                Expi := Exp + Adjustment - Expf;
@@ -807,6 +807,13 @@  package body System.Fat_Gen is
                         Float_Word (IEEE_Ebias + Expf) * Exp_Factor;
 
          if Expi < 0 then
+            --  Given that Expi >= -Mantissa, only -64 is problematic
+
+            if Expi = -64 then
+               XX := XX / 2.0;
+               Expi := -63;
+            end if;
+
             XX := XX / T (UST.Long_Long_Unsigned (2) ** (-Expi));
          end if;