(2) Modified both MatrixMultiplication test cases to have empty return structs
    types
(3) visc-rt changed to reflect the same

(2) Changes to visc-rt to now have just one function to set a kernel parameter,
    no matter input or output.
(3) Matrix Multiplication example working.
(4) Changes to DFG2LLVM_NVPTX pass to generate calls to runtime api to read
    pointer parameters with out attribute.

(2) Made all the local functions static so that they are not visible outside
    this file
(3) Runs correctly. Now have to finish support for attributes

Point to note: Writing a module messes it up in memory, hence it should always
be done when it is not to be used anymore

name, local and global sizes.

    binary
(2) Several hacks in visc-rt to estimate the work required to get
    MatrixMultiplication to work. 3 issues need to be resolved
    i)    The metadata for kernel followed by number of kernels has to be one
          single metadata
    ii)   Get Kernel name and work group sizes in NVPTX pass
    iii)  Major: Cannot use device memory pointer as cl_mem. Fundamentally
          this is a design compatibility issue with VISC and OpenCL 1.1.
          OpenCL 2.0 might resolve it

is incompatible with visc-rt
(2) Set GenFunc pointer to NULL. Completely removed FMap from both backend
passes. Now generated func info is contained in DFNode itself.
(3) Added Makefile to visc-rt for easy compilation to generate visc-rt.ll

Removed unneeded(for now) metadata.

the default.

They are not entirely similar to those we have observed.
They contain the same information, but in a different setting.
We still need to see if the nvptx backend will accept this format.

Global and Generic memory in NVPTX backend. Still commiting the code with
functions to change address space of arguments, but not calling them. Might need
in future if things break.

M    llvm/projects/visc-rt/visc-rt.cpp

pointer, size ... format for PTX nodes
(2) refactoring DFG2LLVM passes. Moving all the ptx code to ptx pass.
M    llvm/projects/visc-rt/visc-rt.cpp
M    llvm/lib/Transforms/DFG2LLVM_X86/DFG2LLVM_X86.cpp
M    llvm/lib/Transforms/DFG2LLVM_NVPTX/DFG2LLVM_NVPTX.cpp

the graphID as reference. It is returned by the launch functions. Makes it
easier to write code
(2) BuildDFG support for data transfer to PTX nodes
(3) ClearDFG bugs fixed
(4) visc-rt modified
(5) Test case 3level.ll modified to match the new launch and wait intrinsic
format

M    llvm/test/VISC/unitTests/3level.ll
M    llvm/include/llvm/IR/DFGraph.h
M    llvm/include/llvm/IR/IntrinsicsVISC.td
M    llvm/projects/visc-rt/visc-rt.cpp
M    llvm/lib/Transforms/BuildDFG/BuildDFG.cpp
M    llvm/lib/Transforms/DFG2LLVM_X86/DFG2LLVM_X86.cpp
M    llvm/lib/Transforms/ClearDFG/ClearDFG.cpp

(2) Modified BuildDFG and DFG2LLVM_X86 to generate the correct calls to visc
runtime
(3) Modified the test cases to the new launch wait semantics of visc
(4) Added Wait intrinsic to VISC and modified the argument list of launch and
wait

changed to load a kernel binary (ptx) instead of kernel source

Now we add an argument that is a pointer to the function's (struct) return type.