About This Website

Hello, thanks for your interest in my website. 😊 I try to log down my learnings and thoughts about compilers (SPIR-V, MLIR, etc.), GPU (Vulkan, Metal, etc.), toolchain (ML frameworks, productivity tools, etc.), and others sporadically.

About Me

I’m Lei Zhang. I work on machine learning compilers and runtime within Google.

Currently my main focus is exploring a standards- and compilation-based approach towards ML inference on various GPUs, leveraging GPU APIs like Vulkan and IRs like SPIR-V and the MLIR infrastructure. It’s joint work with many Google colleagues and open source contributors, involving developing a compiler stack and runtime toolbox. I maintain the SPIR-V dialect and conversions in MLIR, and contribute to IREE, a MLIR-based end-to-end compiler and runtime solution. I also created µVkCompute and hardware-perfcounter along the way to make cross-vendor benchmarking and profiling easier.

Previously I contributed to various general Vulkan and SPIR-V toolchain projects; for example, tech lead and initial enabler of HLSL to SPIR-V compilation in DirectXShaderCompiler, many fundamental features in SPIRV-Tools, and also SPIRV-Headers, Shaderc, Glslang, VK-GL-CTS.

I also created a few SPIR-V projects in the Rust programming language: rspirv—Rust implementation of SPIR-V module processing functionalities, and shaderc-rs—Rust bindings for the shaderc library.