Skip to content
Jin Wang edited this page Mar 27, 2015 · 1 revision

Milestones

  1. AMD IL Intermediate representation in Ocelot
  2. PTX Re-structuring pass to express PTX programs only in terms of Hammock Graphs
  3. PTX to CAL translator
  4. AMD GPU Device using the new Device interface

Blocking Issues

  1. Complete the branch of ocelot with the new device interface
  2. Buying some AMD cards to test locally
  3. Assigning tasks to project members

What am I working on?

Week Apr 11 - Apr 17

  1. Implementing ATIGPUDevice::allocate and understanding how the CudaRuntime uses the MemoryAllocation abstract object.
  2. Implementing ATIGPUDevice::launch and creating ATIExecutableKernel (I ended up not creating an ATIExecutableKernel object yet. Instead, I am launching the kernel directly from ATIGPUDevice).
  3. Looking at Ocelot's test infrastructure. I want to write the first test case for the AMD Backend (vectorScale.cu).

Week Apr 18 - Apr 24

  1. Starting to implement the PTXtoILTranslator.
  2. Looking at PTXtoLLVMTranslator and the way it works.
  3. UAV0 Memory manager. Handle the different semantics between cudaMalloc (returns a pointer) and calResAlloc (returns a handle)

Week Apr 25 - May 1

  1. Use CB0 as the param memory
  2. Continue with PTXToILTranslator