Changelog

All notable changes to tropical-gemm.

[Unreleased]

Added

• CUDA: on-disk CUBIN cache (issue #41). Kernels are compiled straight to a CUBIN (native SASS) for the device’s architecture and cached on disk ($XDG_CACHE_HOME / ~/.cache/tropical-gemm/), keyed on a stable hash of source + compile flags + GPU arch + NVRTC version. A warm CudaContext::new() drops from ~10.7 s to ~0.13 s (skips both the NVRTC compile and the driver’s PTX→SASS JIT). The cache is load-validated and self-healing.

Changed

• CUDA: interior/boundary-split tile loads (issue #40). The global→shared tile copies now give fully-interior blocks a bounds-check-free fast path; only edge blocks (last block-row/col, final K-tile) run the per-element row < M && col < K predicate. The old code ran that predicate on every element of every block — the entire throughput gap vs the C reference. On A40/CUDA 12.8 the kernels go from ~3-6% slower to ~3% faster than TropicalGemm_Cuda, with identical results (full CUDA suite passes). The shared LOAD_A_TILE / LOAD_B_TILE / STORE_C*_TILE macros apply it across all f32/f64/i32/i64, argmax, and batched kernels.
• CUDA: migrated to cudarc 0.19.7. CUDARC_CUDA_VERSION selects the CUDA toolkit at build time, so cargo check works with no toolkit installed and no nvcc is needed at build time.
• CUDA: device_name() now reports the real GPU model (via cuDeviceGetName).

Notes

• Arch flags do not affect CUDA throughput (issue #40). Compiling to an arch-targeted CUBIN (-arch=sm_XX) runs identically to driver-JIT’d PTX; the CUBIN build is purely a startup-latency win (issue #41). The throughput gap vs the C reference was the per-element bounds check above, not the toolchain or the CUDA version.

[0.2.0]

Added

• 2D output functions: New *_matmul_2d variants that return properly shaped 2D arrays instead of flattened 1D output. Available for all semirings (maxplus, minplus, maxmul) and data types (f32, f64, i32, i64):
  • maxplus_matmul_2d, minplus_matmul_2d, maxmul_matmul_2d (f32)
  • maxplus_matmul_2d_f64, minplus_matmul_2d_f64, maxmul_matmul_2d_f64
  • maxplus_matmul_2d_i32, minplus_matmul_2d_i32, maxmul_matmul_2d_i32
  • maxplus_matmul_2d_i64, minplus_matmul_2d_i64, maxmul_matmul_2d_i64
• mdBook documentation
• Comprehensive architecture documentation
• Performance tuning guide
• Troubleshooting guide

Changed

• GIL release during compute: All CPU functions now release Python’s GIL during heavy computation, allowing other Python threads to run concurrently. This improves performance in multi-threaded Python applications.

Fixed

• Batched CPU path copies: Fixed unnecessary memory copies in batched PyTorch operations by using np.asarray() instead of np.array() for zero-copy array creation when possible.

[0.1.0] - Initial Release

Features

• High-performance tropical matrix multiplication
• Support for three semirings: MaxPlus, MinPlus, MaxMul
• SIMD acceleration (AVX-512, AVX2, SSE4.1, NEON)
• CUDA GPU acceleration
• Argmax tracking for backpropagation
• Python bindings with NumPy support
• PyTorch autograd integration

Crates

• tropical-gemm: Core CPU implementation
• tropical-gemm-cuda: CUDA GPU backend
• tropical-gemm-python: Python bindings

Performance

• BLIS-style 5-loop cache blocking
• Runtime SIMD dispatch
• GPU speedup up to 800x for large matrices

Version History

Migration Guides

From NumPy Implementation

If migrating from a pure NumPy tropical matrix multiplication:

# Before (NumPy)
def maxplus_matmul_numpy(a, b):
    m, k = a.shape
    n = b.shape[1]
    c = np.full((m, n), -np.inf)
    for i in range(m):
        for j in range(n):
            for kk in range(k):
                c[i, j] = max(c[i, j], a[i, kk] + b[kk, j])
    return c

# After (tropical-gemm)
import tropical_gemm
c = tropical_gemm.maxplus_matmul(a, b)

API Changes

No breaking changes yet (this is the first release).