b-sumner

Results 144 comments of b-sumner

j1 (both double and float) return NaN for "large" negative numbers

@VinInn, if you don't mind my asking, how did you run across this and the previous issue you raised?

j1 (both double and float) return NaN for "large" negative numbers

Thanks. We have exhaustive tests as well, or rather tests that can be made as exhaustive as desired using command line arguments, but those arguments hadn't been set up properly...

j1 (both double and float) return NaN for "large" negative numbers

@VinInn I noticed the paper mentions large errors for double precision acosh, atanh, and log1p, but for the inputs listed, I am seeing the library producing results with error below...

j1 (both double and float) return NaN for "large" negative numbers

This has been fixed.

tgammaf becomes inf for tiny values of its argument

Thanks, we'll take a look.

tgammaf becomes inf for tiny values of its argument

This has been fixed.

Use of unknown builtin

@MathiasMagnus the device libs evolve in parallel with the compiler, and the compilers used in, and provided by, ROCm 3.0 are more recent than LLVM 9.0.1. Also, the bitcode "libraries"...

Use of unknown builtin

We're working on a means to avoid the device library and compiler sources from getting out of sync and will update the build instructions. Closing this as user error due...

What constraints can be used in operand list in inline assembler for amd gcn?

The constraint for a scalar register is "s". The limitations in 6.2.1 of https://developer.amd.com/wp-content/resources/Vega_Shader_ISA_28July2017.pdf state that at most one SGPR may be read per VALU instruction. You should not have...

How to convert LDS memory address to the address can be passed into DS_READ_* and DS_WRITE_* instructions?

Accessing `__shared__` memory via inline ASM is not recommended and will probably cause more harm than good. What problem are you trying to solve? Why not simply code `temp_local...[i]`?