Machine Learning Compilers

Project Overview

• Introduction
  • What are Machine Learning Compilers?
  • Project Background
  • Project Structure
  • Where to Go Next?
• Project Report
• User Guide
  • Installing Dependencies
  • Building from Source
  • Using our Tensor Compiler
• Documentation Setup

Submissions

• 1. Assembly
  • 1.1 Hello Assembly
  • 1.2 Assembly Function
• 2. Base
  • 2.1 Copying Data
  • 2.2 Instruction Throughput and Latency
• 3. Neon
  • 3.1 Execution Throughput and Latency
  • 3.2 Microkernel
  • 3.3 Loops
  • 3.4 SIMD Lanes
  • 3.5 Accumulator Block Shapes
  • 3.6 Batch-Reduce GEMM
  • 3.7 Transposition
• 4. Code Generation
  • 4.1 BRGEMM Primitive
  • 4.2 Unary Primitives
• 5. Tensor Operation Backend
  • 5.1 User Interface
  • 5.2 Recursive Loops over Primitives
  • 5.3 Performance Benchmarking
  • 5.4 Shared Memory Parallelization
  • 5.5 Optimization Passes
  • 5.6 Unary Operations
• 6. Einsum Trees
  • 6.1 Lowering
  • 6.2 Einsum Tree Optimization
• 7. Individual Phase
  • 7.1 Our Pitch
  • 7.2 Sketch
  • 7.3 Progress of Week 1
  • 7.4 Progress of Week 2
  • 7.5 Wrap-Up

API

• mini_jit
  • exec_t
  • ptype_t
  • dim_t
  • dtype_t
  • error_t
  • to_string()
  • to_string()
  • to_string()
  • to_string()
  • to_string()
  • mini_jit::Benchmark
  • mini_jit::Binary
  • mini_jit::Brgemm
  • mini_jit::Kernel
  • mini_jit::TensorOperation
  • mini_jit::Unary
• mini_jit::benchmarks
  • mini_jit::benchmarks::EinsumTreeBench
  • mini_jit::benchmarks::FastSigmoidPrimitiveBench
  • mini_jit::benchmarks::IdentityPrimitiveBench
  • mini_jit::benchmarks::IdentityTransPrimitiveBench
  • mini_jit::benchmarks::MatmulBrMNKBench
  • mini_jit::benchmarks::MatmulMNKBench
  • mini_jit::benchmarks::NaiveMatmulBrMNKBench
  • mini_jit::benchmarks::NaiveMatmulMNKBench
  • mini_jit::benchmarks::NaiveSigmoidPrimitiveBench
  • mini_jit::benchmarks::ReciprocalPrimitiveBench
  • mini_jit::benchmarks::ReLUPrimitiveBench
  • mini_jit::benchmarks::ReLUTransPrimitiveBench
  • mini_jit::benchmarks::SigmoidInterpolationPrimitiveBench
  • mini_jit::benchmarks::SigmoidTaylorPrimitiveBench
  • mini_jit::benchmarks::SquarePrimitiveBench
  • mini_jit::benchmarks::SquareTransPrimitiveBench
  • mini_jit::benchmarks::TensorOperationBench
  • mini_jit::benchmarks::ZeroEorPrimitiveBench
  • mini_jit::benchmarks::ZeroXZRPrimitiveBench
• mini_jit::converters
  • to_string_bin()
  • to_string_hex()
• mini_jit::einsum
  • mini_jit::einsum::EinsumNode
  • mini_jit::einsum::EinsumTree
• mini_jit::instructions
  • mini_jit::instructions::base
  • mini_jit::instructions::simd_fp
• mini_jit::ir
  • mini_jit::ir::Dimension
  • mini_jit::ir::IRConverter
  • mini_jit::ir::Optimizer
• mini_jit::kernels
  • mini_jit::kernels::matmul
  • mini_jit::kernels::unary
  • mini_jit::kernels::binary
• mini_jit::registers
  • gpr_t
  • simd_fp_t
  • arr_spec_t
  • neon_size_spec_t