SEQUENTIAL PROGRAM SOFTWARE
SEQUENTIAL PROGRAM CODE

The construction of compilers to convert sequential programs for parallel execution on multicore and multiprocessor systems raises some questions. IEEE Transactions on Parallel and Distributed Systems, pages 1084-1096, Jan 2006. An adaptive algorithm selection framework for reduction parallelization. Dynamic optimization option search in GCC. High-level adaptive program optimization with ADAPT. In ISCA Workshop on Parallel Execution of Sequential Programs on Multi-core Architectures (PESPMA'08), pages 1-9, 2008.

SEQUENTIAL PROGRAM SOFTWARE

PLR: A software approach to transient fault tolerance for multicore architectures. Orchestra: intrusion detection using parallel execution and monitoring of program variants in user-space.

SEQUENTIAL PROGRAM CODE

SPIRAL: Code generation for DSP transforms. Fast and effective orchestration of compiler optimizations for automatic performance tuning. Online performance auditing: using hot optimizations without getting burned. Synthesis Lectures on Computer Architecture.

Multilisp: a language for concurrent symbolic computation. A practical method for quickly evaluating program optimizations. MiDataSets: Creating the conditions for a more realistic evaluation of iterative optimization. Dynamic feedback: an effective technique for adaptive computing. Self-adapting linear algebra algorithms and software. MapReduce: Simplified data processing on large clusters. Communications of the ACM, 5(7), Jul 1962. Competitive execution: a method to exploit idle workstations. Rapidly selecting good compiler optimizations using performance counters. Black box approach for selecting optimization options using budget-limited genetic algorithms. PetaBricks: a language and compiler for algorithmic choice. Computation-driven CPE provides super-linear speedups for 5 out of 31 data sets (with a maximum speedup of 7.4) and at most a slow-down of 1% for two data sets. Using a state-of-the-art SAT solver as an illustrative example, we report for compiler-based CPE speedups of 4-6% for many data sets, with a maximum slowdown of 2%. The paper discusses two strategies for the generation of variants and investigates the applicability of CPE for two usage scenarios: i) computation-driven CPE: a simple and straightforward parallelization of heuristic algorithms, and ii) compiler-driven CPE: generation of CPE-enabled programs as part of the compilation process using different optimization strategies. We present and evaluate a run-time system that is implemented as a user-space library and that closely interacts with the operating system. The run-time system ensures that the behavior and outcome of a CPE-enabled program is not distinguishable from the one of its original sequential counterpart.

Variants compete at run-time under the control of a CPE-aware run-time system. The performance of different variants depends on runtime conditions, such as program input or the execution platform, and the execution time of a CPE-enabled program is the sum of the shortest variants. A sequential program is transformed into a CPE-enabled program by introducing multiple variants for parts of the program. Competitive parallel execution ( CPE) is a simple yet attractive technique to improve the performance of sequential programs on multi-core and multi-processor systems.

Sequential program