Array bounds check elimination for the Java HotSpot#8482; client compiler

Array bounds check elimination in the context of deoptimization

As a type-safe program language, Java requires bounds checks of array accesses. Whenever an array element is accessed, a cmp (compare) instruction is executed to check whether the index value is within the valid bounds. Array bounds checks may prevent many useful optimizations because of precise exception. We present a new ABCE (Array Bounds Check Elimination) algorithm to eliminate redundant checks based on sparse representation for a Java static compiler. In contrast to other approaches performing in JVMs, we adhere to the design principle of the static compiler to optimize scientific Java applications. The algorithm is a light-weight algorithm working on an intermediate representation in Static Single Assignment form. It fully removes bounds checks if it can be proven that they never fail. Whenever possible, it moves bounds checks out of loops to reduce the total number of executed checks. If such a check fails, the executing program branches into the unmodified loop to preserve the exception semantics of Java. For the scientific SciMark 2.0 benchmark suite, this algorithm removes on average 76% of bounds check instructions. The evaluation shows a speedup near to the theoretical maximum for LU test case.

Optimizing compilers, including those in virtual machines, commonly utilize Static Single Assignment Form as their intermediate representation, but interpreters typically implement stack-oriented virtual machines. This paper introduces an easily interpreted variant of Static Single Assignment Form. Each instruction of this Interpretable Static Single Assignment Form, including the Phi Instruction, has self-contained operational semantics facilitating efficient interpretation. Even the array manipulation instructions possess directly-executable single-assignment semantics. In addition, this paper describes the construction of a prototype virtual machine realizing Interpretable Static Single Assignment Form and reports on its performance.

Static single assignment (SSA) form for scalars has been a significant advance. It has simplified the way we think about scalar variables. It has simplified the design of some optimizations and has made other optimizations more effective. Unfortunately none of this can be be said for SSA form for arrays. The current SSA processing of arrays views an array as a single object. But the kinds of analyses that sophisticated compilers need to perform on arrays, for example those that drive loop parallelization, are at the element level. Current SSA form for arrays is incapable of providing the element-level data flow information required for such analyses.In this paper, we introduce an Array SSA form that captures precise element-level data flow information for array variables in all cases. It is general and simple, and coincides with standard SSA form when applied to scalar variables. It can also be used for structures and other variable types that can be modeled as arrays. An important application of Array SSA form is in automatic parallelization. We show how Array SSA form can enable parallelization of any loop that is free of loop-carried true data dependences. This includes loops with loop-carried anti and output dependences, unanalyzable subscript expressions, and arbitrary control flow within an iteration. Array SSA form achieves this level of generality by making manifest its - functions as runtime computations in cases that are not amenable to compile-time analysis.

More precise construction of static single assignment programs using reaching definitions

This paper deals with the binary analysis of executable programs, with the goal of understanding how they access memory. It explains how to statically build a formal model of all memory accesses. Starting with a control-flow graph of each procedure, well-known techniques are used to structure this graph into a hierarchy of loops in all cases. The paper shows that much more information can be extracted by performing a complete data-flow analysis over machine registers after the program has been put in static single assignment (SSA) form. By using the SSA form, registers used in addressing memory can be symbolically expressed in terms of other, previously set registers. By including the loop structures in the analysis, loop indices and trip counts can also often be expressed symbolically. The whole process produces a formal model made of loops where memory accesses are linear expressions of loop counters and registers. The paper provides a quantitative evaluation of the results when applied to several dozens of SPEC benchmark programs. Because static analysis is often incomplete, the paper ends by describing a lightweight instrumentation strategy that collects at run time enough information to complete the program's symbolic description.

Static Single Assignment form is an intermediate representation that uses /spl Phi/ instructions to merge values at each confluent point of the control flow graph. /spl Phi/ instructions are not machine instructions and must be renamed back to move instructions when translating out of SSA form. Without a coalescing algorithm, the out of SSA translation generates many move instructions. Leung and George [A. L. Leung et al., (1999)] use a SSA form for programs represented as native machine instructions, including the use of machine dedicated registers. For this purpose, they handle renaming constraints thanks to a pinning mechanism. Pinning /spl Phi/ arguments and their corresponding definition to a common resource is also a very attractive technique for coalescing variables. Extending this idea, we propose a method to reduce the /spl Phi/-related copies during the out of SSA translation, thanks to a pinning-based coalescing algorithm that is aware of renaming constraints. We implemented our algorithm in the STMicroelectronics Linear Assembly Optimizer [B. Dupont de Dinechin et al., (2000)]. Our experiments show interesting results when comparing to the existing approaches of Leung and George [A. L. Leung et al., (1999)], Sreedhar et al. [V. Sreedhar et al., (1999)], and Appel and George for register coalescing [L. George et al., (1996)].

Register sharing for high-level synthesis of programs represented in static single assignment (SSA) form is proven to have a polynomial-time solution. Register sharing is modeled as a graph-coloring problem. Although graph coloring is NP-Complete in the general case, an interference graph constructed for a program in SSA form probably belongs to the class of chordal graphs that have an optimal O(|V|+|E|) time algorithm. Chordal graph coloring reduces the number of registers allocated to the program by as much as 86% and 64.93% on average compared to linear scan register allocation.

Partial redundancy elimination (PRE) is an interesting compiler optimization because of its effectiveness and generality. Among many PRE algorithms, the one in static single assignment form (SSAPRE) has benefits over other bit-vector-based PRE algorithms. It preserves the properties of the SSA form after PRE and exploits the sparsity of the SSA form, resulting in reduced analysis and optimization time. This paper presents a practical improvement of the SSAPRE algorithm that further reduces the analysis and optimization time. The underlying idea is removing unnecessary Φ’s during the Φ-Insertion phase that is the first step of SSAPRE. We classify the expressions into three categories: confined expressions, local expressions, and the others. We show that unnecessary Φ’s for confined and local expressions can be easily detected and removed. We implement our locality-based SSAPRE algorithm in a C compiler and evaluate its effectiveness with 20 applications from SPEC benchmark suites. In our measurements, on average 91% of Φ’s identified by the original demand-driven SSAPRE algorithm are unnecessary for PRE. Pruning these unnecessary Φ’s in the Φ-Insertion phase makes our locality-based SSAPRE algorithm 1.8 times faster, on average, than the original SSAPRE algorithm.

We present an algorithm for incrementally including may-alias information into Static Single Assignment form by computing a sequence of increasingly precise (and correspondingly larger) partial SSA forms. Our experiments show significant speedup of our method over exhaustive use of may-alias information, as optimization problems converge well before most may-aliases are needed.

The SSA form (Static Single Assignment form) is used in compilers as an intermediate language as an alternative to traditional three-address code because code in SSA form is easier to analyse and optimize using data-flow analysis such as common-subexpression elimination, value numbering, register allocation and so on.

Loop splitting is a key compiler optimization technique for reducing control overhead and improving pipeline efficiency. To address the limited applicability of the existing loop splitting strategy in GCC compiler, an improved algorithm is proposed based on Static Single Assignment form. It categorizes condition variables by their positions in PHI nodes into loop-header PHI node condition variables and merge PHI node condition variables. For these two types of condition variables, the algorithm performs semi-invariance analysis separately and selects partitioning points accordingly, thereby achieving more general loop splitting optimization. The algorithm is implemented in the Sunway GCC compiler. Experimental results show that, compared to the original loop splitting algorithm, the algorithm improves the performance of the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$470. \text{lbm}$</tex> benchmark in SPEC CPU 2006 test suite by 8.8%, and the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$620.0 m n e t p p_{s}$</tex> benchmark in SPEC CPU 2017 test suite by 4.3%. This method expands the scope of optimizable loop structures, improves the optimization efficiency of the Sunway GCC compiler, and can provide assistance for the basic software ecosystem construction of the domestic Sunway platform.

Scalar Compiler Optimizations on the Static Single Assignment Form and the Flow Graph Y.N. Srikant

This paper presents a new algorithm for performing global value numbering on a routine in static single assignment form. Our algorithm has all the strengths of the most powerful existing practical methods of global value numbering; it unifies optimistic value numbering with constant folding, algebraic simplification and unreachable code elimination. It goes beyond existing methods by unifying optimistic value numbering with further analyses: it canonicalizes the structure of expressions in order to expose more congruences by performing global reassociation, it exploits the congruences induced by the predicates of conditional jumps (predicate inference and value inference), and it associates the arguments of acyclic ø functions with the predicates controlling their arrival (ø predication), thus enabling congruence finding on conditional control structures. Finally, it implements an efficient sparse formulation and offers a range of tradeoffs between compilation time and optimization strength. We describe an implementation of the algorithm and present measurements of its strength and efficiency collected when optimizing the SPEC CINT2000 C benchmarks.

Recent developments in register allocation, mostly linked to static single assignment (SSA) form, have shown the benefits of decoupling the problem in two phases: a first spilling phase places load and store instructions so that the register pressure at all program points is small enough, and a second assignment and coalescing phase maps the variables to physical registers and reduces the number of move instructions among registers. This article focuses on the first phase, for which many open questions remain: in particular, we study the notion of optimal spilling (what can be expressed?) and the impact of SSA form (does it help?). To identify the important features for optimal spilling on load-store architectures, we develop a new integer linear programming formulation, more accurate and expressive than previous approaches. Among other features, we can express SSA ϕ-functions, memory-to-memory copies, and the fact that a value can be stored simultaneously in a register and in memory. Based on this formulation, we present a thorough analysis of the results obtained for the SPECINT 2000 and EEMBC 1.1 benchmarks, from which we draw, among others, the following conclusions: (1) rematerialization is extremely important; (2) SSA complicates the formulation of optimal spilling, especially because of memory coalescing when the code is not in conventional SSA (CSSA); (3) microarchitectural features are significant and thus have to be accounted for; and (4) significant savings can be obtained in terms of static spill costs, cache miss rates, and dynamic instruction counts.