Anton Smirnov

Results 213 comments of Anton Smirnov

We now use much newer versions so this should be fine.

We now have better support for `gfx103x` other than `gfx1030`. Although now we mostly require system-wide installation of ROCm for full support.

Will close. Feel free to try with latest AMDGPU.jl.

I think we can close it now as the issue has been resolved: - we now use correct ROCm & LLVM versions - alloca issue was also fixed

We now have stream-ordered allocations and on top use the refcounting mechanism from GPUArrays. As well as `unsafe_free!`.

Scalar indexing no longer happens.

> which allocated new memory instead of wrapping over the passed pointer Can you show an example of when it allocated? Both `HostBuffer` and `HIPBuffer` when passed a pointer and...

I don't have any objections regarding these changes (as long as they work). `lock` keyword is more of a remnant from the HSA era. FYI, current implementation also does not...

Hm... actually yes, just tried it on 1.10-beta2.