[PATCH] D89360: Treat constant contexts as being in the default rounding mode.

This revision was landed with ongoing or failed builds. This revision was automatically updated to reflect the committed changes. Closed by commit rG7e801ca0efa9: Treat constant contexts as being in the default rounding mode. (authored by rsmith). Repository: rG LLVM Github Monorepo CHANGES SI

[PATCH] D89360: Treat constant contexts as being in the default rounding mode.

rjmccall accepted this revision. rjmccall added a comment. This revision is now accepted and ready to land. Okay, thanks. This LGTM. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D89360/new/ https://reviews.llvm.org/D89360

[PATCH] D89360: Treat constant contexts as being in the default rounding mode.

rsmith added a comment. In D89360#2329167 , @rjmccall wrote: > Okay. I can accept the fairly unfortunate thing with `const int` > initializers. When is `IsConstantContext` set in C, just static initializers? This affects static initializers, enumerator

[PATCH] D89360: Treat constant contexts as being in the default rounding mode.

rsmith updated this revision to Diff 298724. rsmith added a comment. - Add more testcases to demonstrate behavior in C. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D89360/new/ https://reviews.llvm.org/D89360 Files: clang/lib/AST/ExprConstant.cp

[PATCH] D89360: Treat constant contexts as being in the default rounding mode.

sepavloff added a comment. Probably there is an issue with code generation. The source: constexpr float func_01(float x, float y) { return x + y; } float V1 = func_01(1.0F, 0x0.01p0F); compiled with '-frounding-math' must produce dynamic initializer. It however is evaluated at

[PATCH] D89360: Treat constant contexts as being in the default rounding mode.

sepavloff added a comment. As C++ case is much more complicated and we need a ruleset based on essential properties rather than implementation viewpoint, I am convinced that the solution based on manifestly constant-evaluated property is the best. Thank you for the discussion! Repository: r

[PATCH] D89360: Treat constant contexts as being in the default rounding mode.

sepavloff added a comment. In D89360#2330522 , @rsmith wrote: > In D89360#2329856 , @sepavloff wrote: > >> I would propose to consider solution in D88498 >> . It tries to fix the re

[PATCH] D89360: Treat constant contexts as being in the default rounding mode.

rsmith added inline comments. Comment at: clang/test/SemaCXX/rounding-math.cpp:9 + +constexpr int f(int n) { return int(n * (1.0 / 3.0)); } + rsmith wrote: > sepavloff wrote: > > This code requires additional solution. The function body is built using > > dynami

[PATCH] D89360: Treat constant contexts as being in the default rounding mode.

rsmith added a comment. In D89360#2329856 , @sepavloff wrote: > I would propose to consider solution in D88498 > . It tries to fix the real reason of the > malfunction - using dynamic rounding mode for evaluation of globa

[PATCH] D89360: Treat constant contexts as being in the default rounding mode.

sepavloff added a comment. I would propose to consider solution in D88498 . It tries to fix the real reason of the malfunction - using dynamic rounding mode for evaluation of global variable initializers. With the workaround for constexpr functions it allows to

[PATCH] D89360: Treat constant contexts as being in the default rounding mode.

rjmccall added a comment. Okay. I can accept the fairly unfortunate thing with `const int` initializers. When is `IsConstantContext` set in C, just static initializers? Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D89360/new/ https://reviews.llv

[PATCH] D89360: Treat constant contexts as being in the default rounding mode.

rsmith created this revision. rsmith added reviewers: rjmccall, mibintc, sepavloff. Herald added a project: clang. rsmith requested review of this revision. This addresses a regression where pretty much all C++ compilations using -frounding-math now fail, due to rounding being performed in constex