How verrpower/pge2m works
This repository is a C++ matrix-computation project centered on cache-tiled matrix storage and multiplication, with SIMD tile kernels and optional scheduler-backed execution. A visible client path is the top-level matrix program using the matrix API; the L3 validation program also exercises tiled matrices and vectorized random initialization. The tree includes exploratory fast-log, trigonometric, and approximation work, but no README describes product capabilities. Several implementation excerpts are partial, so the graph omits unshown internal dispatch and data-flow relationships.
Main entry points: L3test.cpp L3 matrix validation · VPRNG.cpp Vector RNG · matmul.cpp Matrix program · MLCTiledMat.cpp Tiled matrix API
Components 16
Tiled Matrices
- Tiled matrix API
PGEM2/MLCTiledMat.cpp - L3 matrix validation
validation/gemm/L3test/L3test.cpp - L2 kernel validation
validation/gemm/L2test/L2test.cpp - L1 kernel validation
validation/gemm/L1test/L1test.cpp
Compute Support
- SIMD tile kernels
PGEM2/kernel48x48.cpp - SPMC scheduler
PGEM2/SPMC.cpp - Vector RNG
PGEM2/VPRNG.cpp - Advanced math
PGEM2/adv_math.cpp
Shared Utilities
- Array deque
include/ArrayDeque.h - Iteration timer
include/AutoIterationTimer.h - Log formatting
include/log_formatter.h
Math Exploration
- Fast-log experiments
advmathlab/fastlog/fastlog_s.cpp - Remez approximation
advmathlab/remez/remez.py - Sin-cos probes
advmathlab/sincos/svml_sincos_probe.cpp
Other components
- Matrix user
- Matrix program
validation/gemm/topInt/matmul.cpp
Connections 9
- Matrix user → Matrix program: runs
- Matrix program → Tiled matrix API: uses API
validation/gemm/topInt/matmul.cpp - L3 matrix validation → Tiled matrix API: exercises
validation/gemm/L3test/L3test.cpp - L3 matrix validation → Vector RNG: generates with
validation/gemm/L3test/L3test.cpp - L1 kernel validation → SIMD tile kernels: invokes kernels
validation/gemm/L1test/L1test.cpp - L2 kernel validation → SIMD tile kernels: uses kernels
validation/gemm/L2test/L2test.cpp - SPMC scheduler → Array deque: uses queue
- Iteration timer → Log formatting: formats output
include/AutoIterationTimer.h - Vector RNG → Log formatting: formats logs
PGEM2/VPRNG.cpp