m4rs-mt/ILGPU

Question I experimenting with ILGPU for accelerating some computations on large point clouds. When I use Allocate3DDenseXY it throws and exception CL_INVALID_BUFFER_SIZE, which seems to be because the buffer I'm trying to allocate is too large (1024 x 1024 x 1024 floats). Is there a way that I can access CL_DEVICE_MAX_MEM_ALLOC_SIZE so that I can efficiently process the data in chunks? I don't see anything about it in the documentation. The GPU it's chosen (using ctx.GetPreferredDevice) is the integrated Intel one, which is surprising given I have a powerful NVIDIA GPU. Environment ILGPU version: 1.5.3 .NET 10 Operating system: Windows 11 Hardware (if GPU-related): Intel(R) RaptorLake-S Mobile Graphics Controller Additional context No response

Question I need to run a method created dynamically. DynamicMethod dm=new DynamicMethod("Test", typeof(void),new Type[]{typeof(Index1D),typeof(ArrayView<Single>)},typeof(Program)); var il= dm.GetILGenerator(); il.Emit(OpCodes.Ret); var del=dm.CreateDelegate<Action<Index1D,ArrayView<float>>>(); /* equals to: static void Test( Index1D index,ArrayView<float> buffer) // A view of floats to store float results from GPUMath { return; } */ using var context = Context.CreateDefault(); using var accelerator = context.GetPreferredDevice(preferCPU: true).CreateAccelerator(context); var kernel = accelerator.LoadAutoGroupedStreamKernel(del); using var buffer = accelerator.Allocate1D<float>(128); kernel((int)buffer.Length, buffer.View); However, exception System.NotSupportedException: Derived classes must provide an implementation. takes place. Is there any way to compile a dynamically created method with ILGpu? And there's no "Derived classes" I think Environment ILGPU