Memento mori: dynamic allocation-site-based optimizations

Abstract

Languages that lack static typing are ubiquitous in the world of mobile and web applications. The rapid rise of larger applications like interactive web GUIs, games, and cryptography presents a new range of implementation challenges for modern virtual machines to close the performance gap between typed and untyped languages. While all languages can benefit from efficient automatic memory management, languages like JavaScript present extra thrill with innocentlooking but difficult features like dynamically-sized arrays, deletable properties, and prototypes. Optimizing such languages requires complex dynamic techniques with more radical object layout strategies such as dynamically evolving representations for arrays. This paper presents a general approach for gathering temporal allocation site feedback that tackles both the general problem of object lifetime estimation and improves optimization of these problematic language features. We introduce a new implementation technique where allocation mementos processed by the garbage collector and runtime system efficiently tie objects back to allocation sites in the program and dynamically estimate object lifetime, representation, and size to inform three optimizations: pretenuring, pretransitioning, and presizing. Unlike previous work on pretenuring, our system utilizes allocation mementos to achieve fully dynamic allocation-sitebased pretenuring in a production system. We implement all of our techniques in V8, a high performance virtual machine for JavaScript, and demonstrate solid performance improvements across a range of benchmarks.

Efficient and thread-safe objects for dynamically-typed languages

“…Objects added to the graph will be pre-shared as well if most of the objects allocated at the same allocation sites are later added to a shared object graph. This technique is similar to what is done by Domani et al [9] and Clifford et al [6]. Domani et al use such a technique to allocate objects directly in the global heap as described in further detail in Section 8.1.…”

Phase-Aware Web Browser Power Management on HMP Platforms

“…A number of recent publications have targeted the performance of browsers [2,7,10,22,26] with the aim of improving the user experience. However, very few studies have addressed the issue of web browser power consumption, although battery lifetime is an important metric when measuring the usability of mobile devices.…”

Applying Optimizations for Dynamically-typed Languages to Java

“…With respect to the example in Listing 3, the bytecode dispatch node transfers execution between the seven bytecode nodes with indices [1,2,3,4,5,6,7]. Figure 2 shows the bytecode dispatch node for the bytecode in Listing 3 and illustrates the control low with red arrows.…”

“…Next, the compiler peels the execution of iinc, iload_0, and sipush out of the loop. At bytecode if_icmplt the compiler sees that this bytecode has two possible successors, namely bci = 3 (i.e., the loop body) // iinc 0, 1 5 bcNodes [4].execute(frame); // iload_0 6 bcNodes [5].execute(frame); // sipush 1000 7 int next = bcNodes [6].execute(frame);// if_icmplt 2 8 int bci = bcNodes [6]. Step 2: Unrolled loop of the basic block dispatch node.…”

“…1 bcNodes [2].execute(frame); // iconst_0 2 bcNodes [3].execute(frame); // istore_0 3 merge_bci_3: 4 bcNodes [3].execute(frame); // iinc 0, 1 5 bcNodes [4].execute(frame); // iload_0 6 bcNodes [5].execute(frame); // sipush 1000 7 int next = bcNodes [6].execute(frame);// if_icmplt 2 8 int bci = bcNodes [6].successors[next]; 9 if (bci == 3) { 10 goto merge_bci_3; 11 } else if (bci == 7) { 12 bcNodes [7].execute(frame); // return 13 return; 14 } Listing 7: Final stage: Unrolled loop of the basic block dispatch node. and bci = 7 (i.e., the loop exit).…”