AMDGPU.jl
AMDGPU.jl copied to clipboard
JuliaGPU
AMD GPU (ROCm) programming in Julia
Update doc section referring to queue handling to reflect latest syntax.
Some libraries, like rocSPARSE, call HIP functions which expect to be passed allocations generated from `hipMalloc` and friends. Because `hipMalloc` just ends up calling HSA allocation functions, we should be...
Previously, a call to `killqueue()` would not clean up the global `QUEUES` list. On my device, a new call to `Queue()` would reuse the same queue pointer, and this would...
I've noticed a number of competing techniques used to manage data races throughout the codebase. There is the RT_LOCK (I assume RT = runtime) used to manage global state access...
This test currently fails (Julia 1.8, MI250x) https://github.com/JuliaGPU/AMDGPU.jl/blob/e0a48dd9aadc0329e176a983ff0d7ee0e824b252/test/hsa/memory.jl#L184 with following error ``` Region API Queries: Test Failed at /pfs/lustrep4/scratch/project_465000139/lurass/AMDGPU.jl/test/hsa/memory.jl:184 Expression: all(Runtime.region_host_accessible, regions_finegrained) Stacktrace: [1] macro expansion @ /pfs/lustrep4/scratch/project_465000139/lurass/julia_local/julia-1.8.0/share/julia/stdlib/v1.8/Test/src/Test.jl:464 [inlined] [2]...
Hello, I just got a new AMD RX 6700 XT and have been testing out some AMDGPU features with it. when using it for the first time, I got the...
CUDA has a great feature for sizing threads and blocks, namely launch_configuration(). I rarely manually size my kernel, instead something like: ``` kernel = @cuda launch=false myfunc(args...) config = launch_configuration(kernel.fun)...