PGZ: Automatic Zero-Value Code Specialization

In prior work we proposed Zeroploit, a transform that duplicates code, specializes one path assuming certain key program operands, called versioning variables, are zero, and leaves the other path unspecialized. Dynamically, depending on the versioning variable’s value, either the specialized fast path or the default slow path will execute. We evaluated Zeroploit with hand-optimized codes in that work.

In this paper, we present PGZ, a completely automated, profile-guided compiler approach for Zeroploit. Our compiler automatically determines which versioning variables or combinations thereof are profitable, and determines the code region to duplicate and specialize. PGZ's heuristic takes operand zero value probabilities as input and it then uses classical techniques such as constant folding and dead-code elimination to estimate the potential savings of specializing a versioning variable. PGZ  transforms profitable candidates, yielding an average speedup of 21.2% for targeted shader programs, and an average frame-rate speedup of 4.4% across a collection of modern gaming applications on an NVIDIA GeForce RTX 2080 GPU.

Ram Rangan (NVIDIA)
Publication Date: 
Tuesday, March 2, 2021