bmp-conv

High-performance BMP image convolution tool with multiple parallelization strategies.

bmp-conv explores different computation models for image convolution:

• on-CPU (single and multi-threaded) (static & queue-based)
• MPI-based distributed processing
• on-GPU WIP

Features

• CPU, (GPU — WIP)
• Multiple convolution filters (blur, sharpen, gaussian, median, etc.)
• Several workload distribution strategies
• Queue-based pipeline for batch processing
• MPI-based on-CPU pipeline
• Built-in benchmarking & logging

Quick Start

Build

cmake -S . -B build
cmake --build build

Minimal usage

./bmp-conv image.bmp --filter=sh --mode=by_column --block=30

Execution Modes

Platforms

Mode	Description
-cpu	Default CPU execution (including queue mode and MPI)
-gpu	GPU mode (WIP)

Distribution strategies

• by_row
• by_column
• by_pixel
• by_grid (block-based)

Examples

CPU Multi-threaded grid convolution

./bmp-conv -cpu image.bmp --filter=gg --mode=by_grid --threadnum=4 --block=16

CPU Queue-based batch processing

./bmp-conv -queue-mode img1.bmp img2.bmp img3.bmp \
  --filter=bb --mode=by_row --block=5 --rww=1,2,1

CPU MPI execution (4 processes)

mpirun -np 4 ./bmp-conv -cpu -mpi-mode image.bmp \
  --filter=em --mode=by_column --block=5

GPU execution (auto-defined work-group size)

./bmp-conv -gpu image.bmp --filter=gg

Benchmarking & Analysis

The project includes scripts for:

• single vs multi-thread comparison
• queue-based pipeline analysis
• MPI scalability

See:

• docs/ST-MT-mode-analysis.md
• docs/QMT-mode-analysis.md
• docs/MPI-mode-analysis.md
• docs/Benchmark-setup.md

Run benchmarks:

./tests/st-mt-benchmark.sh
./tests/qmt-benchmark.sh
./tests/mpi-benchmark.sh

Documentation

• Command-line reference
• Filters overview
• Parallel models explained

License

GPL-3.0 — see LICENSE