Fix test262 expectations for fast implementation of Math.exp()

test/test262/test262.status

Index: test/test262/test262.status
diff --git a/test/test262/test262.status b/test/test262/test262.status
index 06e32ae38b403970b3e1d77ca769fdfae239bb50..8eaa3657fa3c517d8446ac25da4809651bf729d2 100644
--- a/test/test262/test262.status
+++ b/test/test262/test262.status
@@ -46,12 +46,13 @@ S15.12.2_A1: FAIL
 
 ##################### DELIBERATE INCOMPATIBILITIES #####################
 
-# This tests precision of Math.tan and Math.sin.  The implementation for those
+# This tests precision of Math functions.  The implementation for those
 # trigonometric functions are platform/compiler dependent.  Furthermore, the
 # expectation values by far deviates from the actual result given by an
 # arbitrary-precision calculator, making those tests partly bogus.
-S15.8.2.16_A7: PASS || FAIL_OK
-S15.8.2.18_A7: PASS || FAIL_OK
+S15.8.2.8_A6: PASS || FAIL_OK   # Math.exp (less precise with --fast-math)
+S15.8.2.16_A7: PASS || FAIL_OK  # Math.sin
+S15.8.2.18_A7: PASS || FAIL_OK  # Math.tan
 
 # Linux for ia32 (and therefore simulators) default to extended 80 bit floating
 # point formats, so these tests checking 64-bit FP precision fail. The other