[clang] Align -ffp-model=fast denormal handling with -ffast-math (PR #89477)

[Matt Arsenault via cfe-commits]

arsenm wrote:

> Some of the GPU targets, IIRC, want daz/ftz by default. Not all targets have DAZ/FTZ bits that can be set; I think RISC-V is in this category, although to be honest, trying to track down all the ISA extensions to make sure is a bit beyond my ken.
> 

OpenCL allows you to flush float denorms by default and has an explicit flag for it, which is allowed to only change the f32 mode. For AMDGPU we default to no denormals on old targets without fast support. This is also somewhat forced because unfortunately OpenCL only provided a flag to enable flushing and not go in the other direction.

> It's a real mess since some targets flush to positive zero and some targets flush to signed zero and some manuals I couldn't make heads or tails out of what they actually did.

I believe ARM has a control bit to set if you want the flush to positive zero or preserve sign mode. I don't see why anyone would ever want the positive zero mode. I'm not sure there are any other users 

https://github.com/llvm/llvm-project/pull/89477