My suggestion is to make them consistent to each other.
I prefer to use __gen_ocl_get_time_stamp().
>>> well, the naming rule for extension built-in is xxx_function(), you can
>>> check the Khronos website. I would change __gen_ocl_simd_any() to
>>> beignet_simd_any()
These two functions should be pure internal function. We should not put them in
the header file. Just put them in the ocl_misc.cl should be good enough.
>>> I agree with you.
> + Instruction READ_ARF(Type type, Register dst, uint32_t regNum, uint32_t
> subRegNum) {
> + return internal::ReadARFInstruction(type, dst, regNum,
> + subRegNum).convert(); }
This is a GEN IR level instructions, but the prototype seems too low level for
me.
It's better to define some abstract ARF types in this layer, and to
determine/choose the actual regNum/subRegNum in the instruction selection stage
according to the actual target platform.
For example, we could define some enum values as
GEN_ARF_TM0
GEN_ARF_XXX
at the ocl_misc.h file, and use this enum value in the ocl kernel.
>>> Yes, I will add some enum value in the Gen IR level, currently I don't want
>>> to add a built-in as __gen_ocl_read_arf(uint arf).
>>> because I have to define both at ocl_misc.h and Gen IR and keep them
>>> consistent then.
>>> if in the future we have that requirement, we can do it that time.
int __gen_ocl_region(short offset, int data)
>>> this seems better than current naming. I will use this one.
_______________________________________________
Beignet mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/beignet
