Fix clang build.dev/fix-clang

Clang warns (and errors due to -Werror) about these comparisons
losing precision as a result of the constants kXMaxInt and kXMaxUInt
being unrepresentable as a double. Although the warning could be
silenced by explicitly casting the constants to doubles, this relies
on implementation-defined rounding behavior so just write out the
floating point literals explicitly.

1 files changed, 6 insertions, 2 deletions

diff --git a/src/aarch64/logic-aarch64.cc b/src/aarch64/logic-aarch64.cc
index cb82f715..de77e58a 100644
--- a/src/aarch64/logic-aarch64.cc
+++ b/src/aarch64/logic-aarch64.cc
@@ -4755,7 +4755,9 @@ int32_t Simulator::FPToInt32(double value, FPRounding rmode) {
 
 int64_t Simulator::FPToInt64(double value, FPRounding rmode) {
   value = FPRoundInt(value, rmode);
-  if (value >= kXMaxInt) {
+  // This is equivalent to "if (value >= kXMaxInt)" but avoids rounding issues
+  // as a result of kMaxInt not being representable as a double.
+  if (value >= 9223372036854775808.) {
     return kXMaxInt;
   } else if (value < kXMinInt) {
     return kXMinInt;
@@ -4788,7 +4790,9 @@ uint32_t Simulator::FPToUInt32(double value, FPRounding rmode) {
 
 uint64_t Simulator::FPToUInt64(double value, FPRounding rmode) {
   value = FPRoundInt(value, rmode);
-  if (value >= kXMaxUInt) {
+  // This is equivalent to "if (value >= kXMaxUInt)" but avoids rounding issues
+  // as a result of kMaxUInt not being representable as a double.
+  if (value >= 18446744073709551616.) {
     return kXMaxUInt;
   } else if (value < 0.0) {
     return 0;