Examples
The example projects that come with ParetoStack 0.3.13. paretostack examples lists them; paretostack examples run NAME --yes copies one to ~/ParetoStack/sample/NAME and optimizes it.
| Example | Output | Needs | Strategy |
|---|---|---|---|
word_filter |
bit-exact | Python only | L2 set membership |
text_stats |
bit-exact | Python only | L1 bind (little to gain) |
quickstart |
bit-exact | Python only | L1 inline / bind |
vector_loops |
bit-exact | numpy | L1 vectorize |
parallel_items |
bit-exact | numpy | L1 parallelize |
assert_heavy |
bit-exact | Python only | L0 python -O (search) |
matmul_lib |
contract | numpy | LIB library swap |
image_fixed_point |
contract | numpy | L3 fixed point (accuracy trade) |
fft_convolution |
contract | numpy | L3 FFT convolution |
box_sums |
contract | numpy | L3 summed-area table (NumPy) |
opencv_filters |
contract | OpenCV 5 | L3 summed-area table (OpenCV) |
opencv_pixels |
bit-exact | OpenCV | OpenCV: pixel loops -> cv2 functions |
opencv_cse |
bit-exact | OpenCV | OpenCV: repeated work done once |
opencv_metrics |
bit-exact | OpenCV | OpenCV: SAD / SSD / histogram loops |
opencv_color |
contract | OpenCV | OpenCV: grey / blend (rounding differs) |
nbody_sim |
contract | a C compiler | L4 C port |
s3_ocr_rescan |
bit-exact | Python only | Cloud-style job: fake S3 bucket + OCR (L1 / PAR) |
cpp_matmul |
bit-exact | a C++ compiler | C++ L2 tiled matrix multiply |
cpp_loops |
bit-exact | a C++ compiler | C++ L2 interchange + OpenMP |
java_box_sums |
contract | a JDK | Java L3 summed-area table |
log_report |
bit-exact | Python only | L1 text rules: regex compiled once + a string grown on an attribute joined once (str_build, re_hoist) |
word_filter
Word filter demo (list membership in a loop)
text_stats
Text statistics demo (pure Python string processing)
quickstart
Quick start (a small project that passes steps 1 to 5)
vector_loops
vector_loops (strategy sample)
parallel_items
parallel_items (strategy sample)
assert_heavy
assert_heavy (strategy sample)
matmul_lib
Matrix-multiply demo (library swap to NumPy / BLAS)
image_fixed_point
Fixed-point image maths (NOT bit-exact): grey, blend and contrast on 1080p frames
fft_convolution
FIR filter by direct convolution (NOT bit-exact): the FFT version rounds differently
box_sums
Window sums by slicing (NOT bit-exact): a summed-area table does it in four additions
opencv_filters
OpenCV filters demo (separable filter2D + local window means)
opencv_pixels
OpenCV demo: hand-written pixel loops (threshold, difference, blur, Sobel, erode, dilate, flip, histogram, sums)
opencv_cse
OpenCV demo: the same work repeated on the same image (grey, bilateral, median, Canny, template match)
opencv_metrics
OpenCV demo: SAD / SSD / energy / histogram / extremes written as loops
opencv_color
OpenCV demo (NOT bit-exact): grey conversion and blending loops, judged by PSNR / SSIM
nbody_sim
N-body gravity simulation demo (physics, NumPy arrays with scalar loops)
s3_ocr_rescan
Re-scan and OCR documents kept in S3 (a made-up project: fake bucket, no network)
cpp_matmul
C++ matrix multiply: loop order and tiling (bit-exact)
cpp_loops
C++ loops: interchange and OpenMP (bit-exact)
java_box_sums
Java window sums: summed-area table (NOT bit-exact)
log_report
A log report: lines parsed with re.match in a loop, the report text grown on an attribute (str_build + re_hoist, 0.3.12)