[2/2] libstdc++: Fix endianness issue with IBM long double [PR98384]

Message ID 20210222215559.1834935-2-ppalka@redhat.com
State New
Headers show
Series
  • [1/2] libstdc++: Robustify long double std::to_chars testcase [PR98384]
Related show

Commit Message

Marek Polacek via Gcc-patches Feb. 22, 2021, 9:55 p.m.
The logic in std::to_chars for extracting the high- and low-order parts
of a IBM long double value does the right thing on powerpc64le, but not
on powerpc64be.  This patch makes the extraction endian-agnostic, which
fixes the execution FAIL of to_chars/long_double.cc on powerpc64be.

Tested on powerpc64le and powerpc64be.  Does this look OK for trunk?

libstdc++-v3/ChangeLog:

	PR libstdc++/98384
	* src/c++17/floating_to_chars.cc (get_ieee_repr): Extract
	the high- and low-order parts from an IBM long double value
	in an endian-agnostic way.
---
 libstdc++-v3/src/c++17/floating_to_chars.cc | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

-- 
2.30.1.559.g2283e0e9af

Comments

Marek Polacek via Gcc-patches Feb. 22, 2021, 10:22 p.m. | #1
On Mon, 22 Feb 2021, 22:13 Patrick Palka via Libstdc++, <
libstdc++@gcc.gnu.org> wrote:

> The logic in std::to_chars for extracting the high- and low-order parts

> of a IBM long double value does the right thing on powerpc64le, but not

> on powerpc64be.  This patch makes the extraction endian-agnostic, which

> fixes the execution FAIL of to_chars/long_double.cc on powerpc64be.

>

> Tested on powerpc64le and powerpc64be.  Does this look OK for trunk?

>



I was going to suggest __builtin_unpack_longdouble but actually since you
want the bits this is better. You'd need another step to get the bits from
each double.

OK for trunk.

Patch

diff --git a/libstdc++-v3/src/c++17/floating_to_chars.cc b/libstdc++-v3/src/c++17/floating_to_chars.cc
index a50548acae7..4b2f85c1c1a 100644
--- a/libstdc++-v3/src/c++17/floating_to_chars.cc
+++ b/libstdc++-v3/src/c++17/floating_to_chars.cc
@@ -395,11 +395,11 @@  namespace
       // of the high part, and we merge the mantissa of the high part with the
       // mantissa (and the implicit leading bit) of the low part.
       using uint_t = unsigned __int128;
-      uint_t value_bits = 0;
-      memcpy(&value_bits, &value, sizeof(value_bits));
+      uint64_t value_bits[2] = {};
+      memcpy(value_bits, &value, sizeof(value_bits));
 
-      const uint64_t value_hi = value_bits;
-      const uint64_t value_lo = value_bits >> 64;
+      const uint64_t value_hi = value_bits[0];
+      const uint64_t value_lo = value_bits[1];
 
       uint64_t mantissa_hi = value_hi & ((1ull << 52) - 1);
       unsigned exponent_hi = (value_hi >> 52) & ((1ull << 11) - 1);