Borsalino v0.2.1

Borsalino v0.2.1

Hotfix release — removes accidentally published early-research content.
No other changes from v0.2.0.

Borsalino v0.2.0

New Features

• Async dispatch — dispatch_async() returns Pulse handle for non-blocking GPU execution. VkFence (Vulkan), MTLCommandBuffer (Metal). Drop performs implicit join.
• Persistent buffers — create_device_buffer() keeps data on GPU across dispatches. VRAM on discrete GPUs, zero-copy on unified memory.
• GPU timestamps — gpu.timestamp() for profiling. Vulkan: vkCmdWriteTimestamp query pool.
• 2D/3D tiled dispatch — WGSL shared memory + barriers for tiled matmul.
• Candle integration — custom element-wise GPU kernel pattern for complementing ML frameworks.

Benchmarks

Platform	Tiled Matmul 8192	Batched SAXPY 1M	Dispatch
GB10 (RTX Spark)	1,403 GFLOPS	372 GFLOPS	0.4 µs
RTX 5080	523 GFLOPS	477 GFLOPS	0.5 µs
M3 Pro	186 GFLOPS	42 GFLOPS	142 µs

Breaking Changes

None. All additions are backward-compatible trait methods with default implementations.

Full details: CHANGELOG.md

Borsalino v0.1.0

Borsalino v0.1.0

Thin GPU compute abstraction for the Industrial Algebra ecosystem.
One trait, two backends (Metal + Vulkan), zero ceremony.

Highlights

• Vulkan backend — full GpuBackend trait via ash 0.38 FFI. WGSL->SPIR-V via naga
• WGSL shader language — write once, run on Metal + Vulkan
• Device-local memory — auto-detects VRAM vs unified, RTX 5080: 15x improvement
• Batched dispatch — dispatch_many() amortises overhead, 0.5 us/dispatch, 577 GFLOPS
• Metal Apple Silicon M3 — seven root causes resolved
• Dual license — AGPL-3.0 + commercial

Benchmarks

Platform	SAXPY 16M	Batched SAXPY 1M	Per-dispatch
RTX 5080	92 GFLOPS	577 GFLOPS	0.5 us
GB10	49 GFLOPS	208 GFLOPS	1.0 us
M3 Pro	30 GFLOPS	—	136 us
AMD iGPU	17 GFLOPS	—	45 us

Full details: CHANGELOG.md