About the service
I'm a low-level developer who works close to the hardware. For over ten years I've written and hand-optimized assembler for x86 and ARM, squeezing performance where every clock cycle and every cache line matters. When a hot loop is slowing down the whole product, I break it down at the instruction level, understand exactly how the processor executes it, and rewrite it so the pipeline doesn't stall, branches predict well, and memory access is cache-friendly. This is not typical application development: I read disassembled code as freely as others read sources, and reason about registers, SIMD lanes, alignment and microarchitecture, not frameworks.
My everyday work is hand-optimizing performance-critical sections, careful profiling and honest measurements before and after, so the gain is visible in numbers, not just words. I debug at the instruction level when a crash or subtle bug only manifests in an optimized build, and conduct legitimate reverse engineering: restore lost documentation, understand undocumented behavior of your own component, or maintain software whose binary diverged from sources. I write clean, commented assembler that an ordinary engineer can read too, and always keep a portable reference version nearby so the code stays maintainable.
Show me where it hurts: codec, math kernel, parser, inner physics loop, firmware procedure. I'll measure, find the real bottleneck instead of guesses, and deliver an optimized, tested version with numbers to prove it.
— Assembler x86 and ARM, hand-written and optimized
— SIMD vectorization: SSE, AVX and NEON
— Profiling, benchmarks and hot spot analysis
— Debugging optimized builds at the instruction level
— Reverse engineering and disassembly for maintenance
— Cache, alignment and branch prediction tuning
My everyday work is hand-optimizing performance-critical sections, careful profiling and honest measurements before and after, so the gain is visible in numbers, not just words. I debug at the instruction level when a crash or subtle bug only manifests in an optimized build, and conduct legitimate reverse engineering: restore lost documentation, understand undocumented behavior of your own component, or maintain software whose binary diverged from sources. I write clean, commented assembler that an ordinary engineer can read too, and always keep a portable reference version nearby so the code stays maintainable.
Show me where it hurts: codec, math kernel, parser, inner physics loop, firmware procedure. I'll measure, find the real bottleneck instead of guesses, and deliver an optimized, tested version with numbers to prove it.
— Assembler x86 and ARM, hand-written and optimized
— SIMD vectorization: SSE, AVX and NEON
— Profiling, benchmarks and hot spot analysis
— Debugging optimized builds at the instruction level
— Reverse engineering and disassembly for maintenance
— Cache, alignment and branch prediction tuning
