Re: [Mesa-dev] [PATCH 08/10] radeonsi: re-enable unsafe-fp-math for LLVM 3.8

On Sun, Oct 11, 2015 at 03:29:48AM +0200, Marek Olšák wrote: > From: Marek Olšák > I don't think we should globally enable this until we are sure it does not introduce any illegal transforms. > Required for 1/sqrt ==> rsq. I think the arcp fast-math flag for instruction

Re: [Mesa-dev] [PATCH 08/10] radeonsi: re-enable unsafe-fp-math for LLVM 3.8

On Sun, Oct 11, 2015 at 4:12 AM, Connor Abbott wrote: > FWIW, this isn't quite correct with ARB_shader_precision or GL4.1 -- > it specifies that infinities should be correctly generated through > division by 0, which unsafe-fp-math doesn't guarantee. At least, > that's

Re: [Mesa-dev] [PATCH 08/10] radeonsi: re-enable unsafe-fp-math for LLVM 3.8

On Sun, Oct 11, 2015 at 7:08 AM, Marek Olšák wrote: > On Sun, Oct 11, 2015 at 4:12 AM, Connor Abbott wrote: >> FWIW, this isn't quite correct with ARB_shader_precision or GL4.1 -- >> it specifies that infinities should be correctly generated through >>

[Mesa-dev] [PATCH 08/10] radeonsi: re-enable unsafe-fp-math for LLVM 3.8

From: Marek Olšák Required for 1/sqrt ==> rsq. We should finally fix the hang instead of running away from the issue. This assumes the bug is in LLVM and we have time to fix it before the release. Include compute shaders as well, which only affects TGSI and thus OpenGL.

Re: [Mesa-dev] [PATCH 08/10] radeonsi: re-enable unsafe-fp-math for LLVM 3.8

FWIW, this isn't quite correct with ARB_shader_precision or GL4.1 -- it specifies that infinities should be correctly generated through division by 0, which unsafe-fp-math doesn't guarantee. At least, that's assuming this is similar to the "fast" per-instruction flag

Re: [Mesa-dev] [PATCH 08/10] radeonsi: re-enable unsafe-fp-math for LLVM 3.8

Hi Marek, I don't get the hang on Dota 2 Reborn with this patch and LLVM/Mesa git. Tested-by: Nick Sarnie Thanks! On Sat, Oct 10, 2015 at 10:12 PM, Connor Abbott wrote: > FWIW, this isn't quite correct with ARB_shader_precision or GL4.1 -- > it