[Mesa-dev] [PATCH 04/13] radeonsi: enable LLVM optimizations that assume no NaNs for non-compute shaders

[Tom Stellard]

On Wed, Jan 07, 2015 at 01:13:37AM +0100, Marek Olšák wrote:
> Neither. It's because we use DX10_CLAMP, which converts NaNs to 0.
> 

Ok, could we add a dx10_clamp bit to si_shader and make this attribute
conditional on that bit.  I'm concerned someone may remove DX10_CLAMP
and forget to also remove this attribute.

-Tom

> Marek
> 
> On Wed, Jan 7, 2015 at 12:51 AM, Tom Stellard <tom at stellard.net> wrote:
> > On Mon, Jan 05, 2015 at 12:18:43AM +0100, Marek Olšák wrote:
> >> From: Marek Olšák <marek.olsak at amd.com>
> >>
> >> ---
> >>  src/gallium/drivers/radeon/radeon_llvm_emit.c | 1 +
> >>  1 file changed, 1 insertion(+)
> >>
> >> diff --git a/src/gallium/drivers/radeon/radeon_llvm_emit.c b/src/gallium/drivers/radeon/radeon_llvm_emit.c
> >> index dc871d7..e3be72c 100644
> >> --- a/src/gallium/drivers/radeon/radeon_llvm_emit.c
> >> +++ b/src/gallium/drivers/radeon/radeon_llvm_emit.c
> >> @@ -83,6 +83,7 @@ void radeon_llvm_shader_type(LLVMValueRef F, unsigned type)
> >>
> >>       if (type != TGSI_PROCESSOR_COMPUTE) {
> >>               LLVMAddTargetDependentFunctionAttr(F, "unsafe-fp-math", "true");
> >> +             LLVMAddTargetDependentFunctionAttr(F, "enable-no-nans-fp-math", "true");
> >
> > Is this required by the OpenGL spec or is it just to fix broken/old
> > games?
> >
> > -Tom
> >
> >>       }
> >>  }
> >>
> >> --
> >> 2.1.0
> >>
> >> _______________________________________________
> >> mesa-dev mailing list
> >> mesa-dev at lists.freedesktop.org
> >> http://lists.freedesktop.org/mailman/listinfo/mesa-dev