[libc-commits] [PATCH] D117824: [libc] Use get_round() instead of floating point tricks in generic hypot implementation.

[Tue Ly via Phabricator via libc-commits]

lntue created this revision.
lntue added a reviewer: sivachandra.
Herald added subscribers: ecnelises, tschuett, kristof.beyls, mgorny.
Herald added a project: libc-project.
lntue requested review of this revision.

The floating point tricks used to get rounding mode require -frounding-math flag, which behaves differently on aarch64.  Reverting back to use get_round instead.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D117824

Files:
  libc/src/__support/FPUtil/Hypot.h
  libc/src/math/generic/CMakeLists.txt


Index: libc/src/math/generic/CMakeLists.txt
===================================================================
--- libc/src/math/generic/CMakeLists.txt
+++ libc/src/math/generic/CMakeLists.txt
@@ -955,8 +955,6 @@
     libc.src.__support.FPUtil.fputil
   COMPILE_OPTIONS
     -O3
-    -frounding-math
-    -Wno-c++17-extensions
 )
 
 add_entrypoint_object(
@@ -1005,8 +1003,6 @@
     libc.src.__support.FPUtil.fputil
   COMPILE_OPTIONS
     -O3
-    -frounding-math
-    -Wno-c++17-extensions
 )
 
 add_entrypoint_object(
Index: libc/src/__support/FPUtil/Hypot.h
===================================================================
--- libc/src/__support/FPUtil/Hypot.h
+++ libc/src/__support/FPUtil/Hypot.h
@@ -144,9 +144,7 @@
   if ((x_bits.get_unbiased_exponent() >=
        y_bits.get_unbiased_exponent() + MantissaWidth<T>::VALUE + 2) ||
       (y == 0)) {
-    // Check if the rounding mode is FE_UPWARD, will need -frounding-math so
-    // that the compiler does not optimize it away.
-    if ((y != 0) && (0x1p0f + 0x1p-24f != 0x1p0f)) {
+    if ((y != 0) && (get_round() == FE_UPWARD)) {
       UIntType out_bits = FPBits_t(abs(x)).uintval();
       return T(FPBits_t(++out_bits));
     }
@@ -154,9 +152,7 @@
   } else if ((y_bits.get_unbiased_exponent() >=
               x_bits.get_unbiased_exponent() + MantissaWidth<T>::VALUE + 2) ||
              (x == 0)) {
-    // Check if the rounding mode is FE_UPWARD, will need -frounding-math so
-    // that the compiler does not optimize it away.
-    if ((x != 0) && (0x1p0f + 0x1p-24f != 0x1p0f)) {
+    if ((x != 0) && (get_round() == FE_UPWARD)) {
       UIntType out_bits = FPBits_t(abs(y)).uintval();
       return T(FPBits_t(++out_bits));
     }


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D117824.401726.patch
Type: text/x-patch
Size: 1710 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/libc-commits/attachments/20220120/0f19fbed/attachment-0001.bin>