Add SST integration into gpgpusim

[William-An]

Add SST integration into gpgpusim

Allow SST (Structural Simulation Toolkit) to utilize GPGPUSim during simulation.

Usage

To enable SST mode, make sure you have -SST_mode 1 in the gpgpusim.config file. You should also use SST specific CUDA functions.

Changes

1. Incorporate the previous sst-gpgpusim-external repo to current mainstream.
2. Changes related to SST are encapsulated with #ifdefine __SST__ directives with flag passed from makefile
3. SST specific cluster, mem interface, gpgpu_sim are created to handle SST requests.
4. Major changes were made to the following folders/files:
   1. libcuda: mostly with cuda_runtime_api.cc to add SST specific APIs as with SST we won't have access to /proc/self/exe to extract PTX file info.
   2. src/gpgpu-sim/gpu-sim.cc: for creating simt cluster and SST specific cycle function
   3. src/gpgpu-sim/shader.cc: for handling memory transactions with SST memory system instead of GPGPUSim's

[William-An]

The Jenkins run might fail a few times as we are still waiting for devel branch in sst-elements to be merged into default branch. Plus there is a bug in sst-core which prevent its build under balar configuration.

[William-An]

I guess there should also be a CI test for testing GPGPU-Sim with SST unit test?

[William-An]

We properly would want to have this merged some time so I don't have to point people to my personal repo in SST docs haha.

[JRPan]

Cool. I have some questions, but not neccesary needs changes. Take a look

• Resolve the conflicts, I created PR several days ago to address gcc-10+ problem. You fixed that as well. \
• Would it make sense to move the changes in shader.cc and shader.h and create a file sst-shader.cc? Just saying. If it's too much trouble then I'm fine withcurrect state.

[William-An]

I fixed most of the places except for moving the changes out of shader.cc/.h. Some of the code are closely mingled together, and there isn't a clean way to separate the changes unless we have a completely different chain from gpu->core ctx->cluster->shader core.