Landing: 3c5f42d64a73

← All landings

Project / Subsystem

gcc / gcc/testsuite

Date

2026-09-18

Author

Jakub Jelinek

Commit

3c5f42d64a7318745a712ef1cc53e8853e32961b

Source

github

Perf win

Breaking

View on GitHub →

All attributes

project: gcc
subsystem: gcc/testsuite
patch_id: —
commit_hash: 3c5f42d64a7318745a712ef1cc53e8853e32961b
source_type: github
headline: Defer floating-point constant evaluation for IBM double-double
tldr: Testcase avoids evaluating problematic floating-point operations on powerpc64.
author: Jakub Jelinek
outcome: committed
performance_win: false
breaking_change: false
series_id: —
series_parts: []
tags: • testsuite
• floating-point
• powerpc
• long double
discussion_id_link: —
bugzilla_pr: —
url: https://github.com/gcc-mirror/gcc/commit/3c5f42d64a7318745a712ef1cc53e8853e32961b
date: 2026-09-18T00:00:00.000Z

A testcase for C++23 floating-point semantics (g++.dg/cpp29/std-floating1.C) has been modified to handle issues on powerpc64 targets where long double uses the IBM double-double format. Due to limitations in GCC’s real.cc emulation of this type, certain floating-point operations (like min/max with or without multiplication) can be imprecise. The test now xfails these specific cases, indicating they are known issues rather than outright failures.