Browse categories
Explore
Fiverr Pro
English
$
USD
Looks Like This Service Is On Hold
I will optimize c and cpp code for performance and low latency
Pakistan
I speak Urdu, English
Low Level Systems and Performance Engineer
I build and optimize performance-critical software in C, C++ and Rust, with a focus on low latency, memory efficiency and measurable results.
My work covers profiling and hot-path optimization, cache...
About this Gig
Is your C/C++ code correct, but missing its latency, throughput, or runtime target?
I profile first, isolate where the CPU actually spends time, then optimize the measured bottleneck and verify the result with reproducible before/after benchmarks.
I work on:
- CPU cache & memory layout: SoA vs AoS, alignment, arena allocation, heap-churn reduction
- SIMD/vectorization: AVX2, AVX-512, ARM NEON
- Multithreading, atomics, lock contention & false sharing
- Disassembly inspection, compiler vectorization and targeted inline asm when justified
Tools: perf, FlameGraph, Callgrind, Google Benchmark, Compiler Explorer
Deliverables include a clean Git patch/commit, benchmark evidence, and technical notes. Modified code is checked with ASan/UBSan where supported and kept warning-clean under strict compiler flags.Your code stays private and is never pasted into public tools. NDA welcome.
Before ordering, feel free to send your benchmark, perf output, or relevant code section. Ill review the context and confirm whether the workload is a good fit and which scope makes sense.
Development technology:
C/C++
My Portfolio
FAQ
How do you know the speedup is real and not benchmark noise?
I use repeated measurements under controlled conditions, with CPU affinity/isolation where the platform supports it. I report reproducible before/after results rather than a single best run, and performance claims are tied to the agreed workload and benchmark command.
What if my code cannot be vectorized with AVX2, AVX-512, or NEON?
Then I will not force SIMD into it. If branching, dependencies, alignment, data layout, or workload size make vectorization ineffective, I document that and target higher-value areas such as cache locality, allocation overhead, branching, synchronization, or algorithmic changes.
How do you verify that optimization does not break correctness?
Changes are validated against your existing tests, reference outputs, deterministic results, or an agreed correctness check. Where supported, modified code is also checked with ASan and UBSan, and touched code is kept warning-clean under strict compiler flags.
How much of my code does the LOC limit cover?
The LOC limit applies to the performance-critical region under investigation, not your entire repository. Large codebases can still fit a package if profiling isolates the bottleneck to a smaller subsystem or kernel. Send the workload before ordering and I’ll confirm the appropriate scope.
. What if the bottleneck is I/O, waiting, or the algorithm rather than CPU execution?
I will identify that in the audit rather than force a CPU-level optimization that will not move your metric. You will receive the evidence, the actual bottleneck, and the highest-value next step.
Can you work with proprietary code or under NDA?
Yes. I can work with proprietary code and reasonable NDA/security requirements. Source is handled in an isolated local development environment and is not pasted into public tools. Please remove credentials, API keys, certificates, customer data, and unrelated secrets before sharing.
