A method for decompilation of AMD GCN kernels to OpenCL

Introduction: Decompilers are useful tools for software analysis and support in the absence of source code. They available many hardware architectures programming languages. However, none existing decompilers modern AMD GPU such as GCN RDNA. Purpose: We aim at developing first assembly decompiler tool a architecture that generates code OpenCL language, which is widely used GPGPUs. Results: developed algorithms following operations: preprocessing code, searching data accesses, extracting systemvalues, decompiling arithmetic operations recovering types. also templates decompilation branching operations. Practical relevance: implemented presented Python called OpenCLDecompiler, supports large subset instructions. This automatically converts disassembled GPGPU into equivalent reduces effort required to analyze