Tiled Thoughts: A Verbose Debug BuildCompiler passes, ML systems, and the occasional detour. Long-form notes from the path between code and silicon.https://samarthnarang.com/en-usMLIR for People Who Only Know LLVM IR: A Guided Tourhttps://samarthnarang.com/tiled-thoughts/posts/mlir-for-llvm/https://samarthnarang.com/tiled-thoughts/posts/mlir-for-llvm/A practical mental-model bridge from LLVM IR to MLIR.Sun, 22 Feb 2026 20:48:12 GMTMLIRLLVMCompilersIROptimizationCodegenCan I Beat Clang’s Auto-Vectorizer on Apple Silicon? A SAXPY Case Studyhttps://samarthnarang.com/tiled-thoughts/posts/vec-playground/https://samarthnarang.com/tiled-thoughts/posts/vec-playground/Hand-written NEON intrinsics, loop unrolling, and a spoiler you can probably guess.Sat, 22 Nov 2025 20:48:43 GMTCompilersLLVMApple-SiliconArm64VectorizationNeonQuantization in MLIR: Types, Scales, and Where to Put the qhttps://samarthnarang.com/tiled-thoughts/posts/quantization-in-mlir/https://samarthnarang.com/tiled-thoughts/posts/quantization-in-mlir/Everyone agrees it matters; the details stay fuzzy. Here are the details.Tue, 18 Nov 2025 20:48:26 GMTMLIRQuantizationCompilersMLHow GPUs Talk: A Practical Guide to Multi-GPU Training and Communicationhttps://samarthnarang.com/tiled-thoughts/posts/multi-gpu-comm/https://samarthnarang.com/tiled-thoughts/posts/multi-gpu-comm/Collectives, interconnects, and the communication patterns behind efficient distributed training.Sat, 15 Nov 2025 20:48:15 GMTGpuDistributed-TrainingSystemsData Structure and Iterator Kung Fu in LLVMhttps://samarthnarang.com/tiled-thoughts/posts/llvm-ds-iter/https://samarthnarang.com/tiled-thoughts/posts/llvm-ds-iter/Zero-copy views and safe mutation loops for faster LLVM passes.Sun, 09 Nov 2025 20:48:04 GMTLLVMData-StructuresIteratorsCompilersPerformanceC++Demystifying GPU Terminology: A Compiler Engineer’s Field Guidehttps://samarthnarang.com/tiled-thoughts/posts/demystifying-gpu-terminology/https://samarthnarang.com/tiled-thoughts/posts/demystifying-gpu-terminology/SIMD, warps, occupancy, coalescing, shared memory, spills and matrix units — mapped to real compiler decisions.Sun, 09 Nov 2025 05:00:00 GMTGpuPerformanceCompilersMLIRTritonCUDAROCmHow Rewriting works in MLIRhttps://samarthnarang.com/tiled-thoughts/posts/rewriting-mlir/https://samarthnarang.com/tiled-thoughts/posts/rewriting-mlir/How MLIR's pattern rewriting works, and why it makes optimizations composable.Sat, 08 Nov 2025 20:48:31 GMTMLIRRewritingCompilersOptimizationBooting Up: A Verbose Debug Build of Life and Compilershttps://samarthnarang.com/tiled-thoughts/posts/verbose-debug-build/https://samarthnarang.com/tiled-thoughts/posts/verbose-debug-build/Compiler passes, ML systems, and life — debug logs from the path between code and silicon.Tue, 04 Nov 2025 20:48:49 GMTReflectionCompilersMLIRLLVMTriton