We present a technique that exploits close collaboration between the compiler and the speculative multithreaded hardware to explore aggressive optimizations and parallelization for scalar programs. The compiler aggressively optimizes the frequently executed code in user programs by predicting an execution path or the values of long-latency instructions. Based on the predicted hot execution path, the compiler forms regions of greatly simplified data and control flow graphs and then performs aggressive optimizations on the formed regions. Thread level speculation (TLS) helps expose program parallelism and guarantees program correctness when the prediction is incorrect. With the collaboration of compilers and speculative multithreaded support, the program performance can be significantly improved. The preliminary results with simple trace regions demonstrate that the performance gain on dynamic compiler schedule cycles can be 33% for some benchmark and about 10%, on the average, for all the eight SpecInt95 benchmarks. For SpecInt2k, the performance gain is up to 23% with the conservative execution model. With a cycle accurate simulator with the conservative execution model, the overall performance gain by considering runtime factors (e.g., cache misses and branch misprediction) for vortex and m88ksim is 12% and 14.7%, respectively. The performance gain can be higher with more sophisticated region formation and region-based optimizations.

Compiler-directed data speculation has been implemented on Itanium systems to allow for a compiler to move a load across a store even when the two operations are potentially aliased This not only breaks data dependency to reduce critical path length, but also allows a load to be scheduled far apart from its uses to hide cache miss latencies. However, the effectiveness of data speculation is affected by the sophistication of alias analysis technique as well as the aggressiveness of the instruction scheduler. In general, the more sophisticated is the alias analysis technique, the less performance gain is from data speculation, and the more aggressive is the instruction scheduler, the more opportunity is for data speculation. In this paper we evaluate in various scenarios the performance potentials of data speculation for SPEC2000C benchmarks. For each scenario, we determine the performance contributions of data speculation due to both critical path reduction and cache miss latency reduction. We also show interesting statistics about the effects of scheduling constraints, the percentage of critical dependencies, the impacts of cache miss latencies, and the distances between the load locations before and after data speculation.

Advanced microprocessors have been increasing clock rates, well beyond the Gigahertz boundary. For such high performance microprocessors, a small and fast data micro-cache (ucache) is important to overall performance, and proper management of it via load bypassing has a significant performance impact. In this paper, we propose and evaluate a hardware-software collaborative technique to manage ucache bypassing for EPIC processors. The hardware supports the ucache bypassing with a fag in the load instruction format, and the compiler employs static analysis and profiling to identify loads that should bypass the ucache. The collaborative method achieves a significant improvement in performance for the SpecInt2000 benchmarks. On average, about 40%, 30%, 24%, and 22% of load references are identified to bypass 256 B, 1 K, 4 K, and 8 K sized ucaches, respectively. This reduces the ucache miss rates by 39%, 32%, 28%, and 26%. The number of pipeline stalls from loads to their uses is reduced by 13%, 9%, 6%, and 5%. Meanwhile, the L1 and L2 cache misses remain largely unchanged. For the 256 B ucache, bypassing improves overall performance on average by 5%.

A chief source of inefficiency in program analysis using abstract interpretation comes from the fact that a large context (i.e., problem state) is propagated from node to node during the course of an analysis. This problem can be addressed and largely alleviated by a technique we call context projection, which projects an input context for a node to the portion that is actually relevant and determines whether the node should be reevaluated based on the projected context. This technique reduces the cost of an evaluation and eliminates unnecessary evaluations. Therefore, the efficiency of computing fixpoints over general lattices is greatly improved. A specific method, reachability, is presented as an example to accomplish context projection. Experimental results using reachability show very convincing speedups (more than eight for larger programs) that demonstrate the practical significance of context projection.