Deconstructing iterative optimization

Abstract

Iterative optimization is a popular compiler optimization approach that has been studied extensively over the past decade. In this article, we deconstruct iterative optimization by evaluating whether it works across datasets and by analyzing why it works.Up to now, most iterative optimization studies are based on a premise which was never truly evaluated: that it is possible to learn the best compiler optimizations across datasets. In this article, we evaluate this question for the first time with a very large number of datasets. We therefore compose KDataSets, a dataset suite with 1000 datasets for 32 programs, which we release to the public. We characterize the diversity of KDataSets, and subsequently use it to evaluate iterative optimization. For all 32 programs, we find that there exists at least one combination of compiler optimizations that achieves at least 83% or more of the best possible speedup across all datasets on two widely used compilers (Intel's ICC and GNU's GCC). This optimal combination is program-specific and yields speedups up to 3.75× (averaged across datasets of a program) over the highest optimization level of the compilers (-O3 for GCC and -fast for ICC). This finding suggests that optimizing programs across datasets might be much easier than previously anticipated.In addition, we evaluate the idea of introducing compiler choice as part of iterative optimization. We find that it can further improve the performance of iterative optimization because different programs favor different compilers. We also investigate why iterative optimization works by analyzing the optimal combinations. We find that only a handful optimizations yield most of the speedup. Finally, we show that optimizations interact in a complex and sometimes counterintuitive way through two case studies, which confirms that iterative optimization is an irreplaceable and important compiler strategy.

A Language and Preprocessor for User-Controlled Generation of Synthetic Programs

“…end (9) end (10) (11) feature read (y_offset,x_offset,weight,array) (12) temp += ${weight} * ${array} (13) genif ${dims} == 2 (14) [y+${y_offset}] (15) end (15) [x+${x_offset}]; (17) end processed and a copy of the template program is created. The code in each copy is scanned for any feature references as regular expressions.…”

“…//For brevity, the code that loads from global memory into local memory is omitted (10) : (11) (12) for (int x_block = x_start; x_block < x_base + 16; x_block+=4) { (13) int x = x_block; (14) double temp = 0; user attempts to sample from an empty varlist. When this happens, the program is not generated and that program instance number is skipped.…”

“…This case study is inspired by the work on cTuning with its MilepostGCC compiler [10], an autotuning compiler that extracts characteristics of a program [11,12], and uses them with a machine learning model to tune programs for performance. Many of these characteristics come from low-level properties of a program's intermediate representation such as the number of basic blocks (BBs), the number of instructions per BB, the number of back edges, and the number of BBs with two successors.…”

“…varlist can_be_defined from temp (12) end (13) (14) feature suminsn (15) variable src1, src2 from temp (16) add src1,src2 to can_be_defined (17) variable dest from can_be_defined (18) remove dest from can_be_defined (19) ${dest} = ${src1} + ${src2}; (20) end (21) (22) feature cpinsn (23) variable src from temp (24) add src to can_be_defined (25) variable dest from can_be_defined (26) remove dest from can_be_defined (27) ${dest} = ${src}; (28) end ( backgroundDist. Depending on the sampled value, the filename from which the background is loaded varies.…”

Cobayn

“…So far, researchers have proposed two main approaches for tackling the problem of identifying the best compiler optimizations: (i) iterative compilation [Chen et al 2012] and (ii) machine-learning (ML) predictive modeling [Agakov et al 2006]. The former approach relies on several recompilation phases, then selects the best set of optimizations.…”