Sameeranjoshi

STEPS

1. Goto .dacecache/folder/
2. g++ sample/poplar_matmul_main.cpp src/cpu/poplar_matmul.cpp include/poplar_matmul.h include/hash.h -I../../../../../dace/runtime/include/ -lpoplar -lpopops -lpoplin -lpoputil

Sameeranjoshi

Project idea updates:

1. Found interest in following idea:

A. Deducing the consistency model of systems using litmus testing Idea : This can be extended to test the Accelerators. * Talked with Prof. Ganesh and he mentioned GPUHarbor is the recent paper in this GG mentioned.** * The GPU Concurrency paper seems more of what we discussed in the office hours after first class. This looks a really useful tool and the idea was quite understandable from the paper, although I didn't get into the weeds of the paper yet. TODO:

1. Understand what is needed from the accelerators space to adapt this.

Sameeranjoshi

035a5-5dc86f1d2c-29bb333675-7a1814c8ee-f372935fdb-35abaf40dc-3e20920ca2-096cd5ae0a-b5ee3fedee-573545fedd-058945afec-4e96ce18d9-6369e8b1c6-9281309306-b4306186a0-dc95751e2a-3939e

Sameeranjoshi

//make variable

!A_t = tensor<510x510xf32>
!B_t = tensor<510x510xf32>
!C_t = tensor<510x510xf32>

func.func @linalg_matmul(
    %A : !A_t, %B : !B_t, %C : !C_t) -> !C_t {
  %0 = linalg.matmul ins(%A, %B : !A_t, !B_t)
                     outs(%C : !C_t) -> !C_t

Sameeranjoshi

#!/bin/bash
module load ccache ninja
module load cmake/3.26.0

ccache --max-size=20G
cmake \
    -G Ninja  \
    -B ../build/ \
    -S . \
    -DCMAKE_INSTALL_PREFIX:PATH=<PATH TO>/iree-install \

Sameeranjoshi

*PASS IN MLIR*

* How to about PASS - https://github.com/joker-eph/llvm-project-with-mlir/blob/master/mlir/g3doc/WritingAPass.md
* List of Passes - https://mlir.llvm.org/docs/Passes/


# TableGen how to 

/uufs/chpc.utah.edu/common/home/u1418973/other/llvm_stuff/IREE/iree-build/third_party/llvm-project/llvm/bin/mlir-tblgen --gen-pass-doc -I/uufs/chpc.utah.edu/common/home/u1418973/other/llvm_stuff/IREE/iree/third_party/llvm-project/mlir/include       /uufs/chpc.utah.edu/common/home/u1418973/other/llvm_stuff/IREE/iree/compiler/src/iree/compiler/Codegen/Passes.td

Sameeranjoshi

Links I referred while learning CUDA.

• Linearized index mapping for permutations of grid X threads.
  https://cs.calvin.edu/courses/cs/374/CUDA/CUDA-Thread-Indexing-Cheatsheet.pdf
  https://anuradha-15.medium.com/cuda-thread-indexing-fb9910cba084

• CUDA ORNL training
  https://www.olcf.ornl.gov/cuda-training-series/

• Github Codes https://github.com/olcf/cuda-training-series