More Like this

1. Decreasing the Miss Rate and Eliminating the Performance Penalty of a Data Filter Cache

   https://doi.org/10.1145/3449043  

   Stokes, Michael; Whalley, David; Onder, Soner (June 2021, ACM Transactions on Architecture and Code Optimization)

   null (Ed.)

   While data filter caches (DFCs) have been shown to be effective at reducing data access energy, they have not been adopted in processors due to the associated performance penalty caused by high DFC miss rates. In this article, we present a design that both decreases the DFC miss rate and completely eliminates the DFC performance penalty even for a level-one data cache (L1 DC) with a single cycle access time. First, we show that a DFC that lazily fills each word in a DFC line from an L1 DC only when the word is referenced is more energy-efficient than eagerly filling the entire DFC line. For a 512B DFC, we are able to eliminate loads of words into the DFC that are never referenced before being evicted, which occurred for about 75% of the words in 32B lines. Second, we demonstrate that a lazily word filled DFC line can effectively share and pack data words from multiple L1 DC lines to lower the DFC miss rate. For a 512B DFC, we completely avoid accessing the L1 DC for loads about 23% of the time and avoid a fully associative L1 DC access for loads 50% of the time, where the DFC only requires about 2.5% of the size of the L1 DC. Finally, we present a method that completely eliminates the DFC performance penalty by speculatively performing DFC tag checks early and only accessing DFC data when a hit is guaranteed. For a 512B DFC, we improve data access energy usage for the DTLB and L1 DC by 33% with no performance degradation. 

   more » « less
2. Decreasing the Miss Rate and Eliminating the Performance Penalty of a Data Filter Cache

   https://doi.org/3449043  

   Stokes, Michael; Whalley, David; Onder, Soner (June 2021, ACM transactions on architecture and code optimization)

   Kaeli, David (Ed.)

   While data filter caches (DFCs) have been shown to be effective at reducing data access energy, they have not been adopted in processors due to the associated performance penalty caused by high DFC miss rates. In this article, we present a design that both decreases the DFC miss rate and completely eliminates the DFC performance penalty even for a level-one data cache (L1 DC) with a single cycle access time. First, we show that a DFC that lazily fills each word in a DFC line from an L1 DC only when the word is referenced is more energy-efficient than eagerly filling the entire DFC line. For a 512B DFC, we are able to eliminate loads of words into the DFC that are never referenced before being evicted, which occurred for about 75% of the words in 32B lines. Second, we demonstrate that a lazily word filled DFC line can effectively share and pack data words from multiple L1 DC lines to lower the DFC miss rate. For a 512B DFC, we completely avoid accessing the L1 DC for loads about 23% of the time and avoid a fully associative L1 DC access for loads 50% of the time, where the DFC only requires about 2.5% of the size of the L1 DC. Finally, we present a method that completely eliminates the DFC performance penalty by speculatively performing DFC tag checks early and only accessing DFC data when a hit is guaranteed. For a 512B DFC, we improve data access energy usage for the DTLB and L1 DC by 33% with no performance degradation. 

   more » « less
3. Improving First Level Cache Efficiency for GPUs Using Dynamic Line Protection

   https://doi.org/10.1145/3225058.3225104  

   Zhu, Xian; Wernsman, Robert; Zambreno, Joseph (August 2018, Proceedings of the International Conference on Parallel Processing (ICPP))

   A modern Graphics Processing Unit (GPU) utilizes L1 Data (L1D) caches to reduce memory bandwidth requirements and latencies. However, the L1D cache can easily be overwhelmed by many memory requests from GPU function units, which can bottleneck GPU performance. It has been shown that the performance of L1D caches is greatly reduced for many GPU applications as a large amount of L1D cache lines are replaced before they are re-referenced. By examining the cache access patterns of these applications, we observe L1D caches with low associativity have difficulty capturing data locality for GPU applications with diverse reuse patterns. These patterns result in frequent line replacements and low data re-usage. To improve the efficiency of L1D caches, we design a Dynamic Line Protection scheme (DLP) that can both preserve valuable cache lines and increase cache line utilization. DLP collects data reuse information from the L1D cache. This information is used to predict protection distances for each memory instruction at runtime, which helps maintain a balance between exploitation of data locality and over-protection of cache lines with long reuse distances. When all cache lines are protected in a set, redundant cache misses are bypassed to reduce the contention for the set. The evaluation result shows that our proposed solution improves cache hits while reducing cache traffic for cache-insufficient applications, achieving up to 137% and an average of 43% IPC improvement over the baseline. 

   more » « less
4. Reducing Load Latency with Cache Level Prediction

   https://doi.org/10.1109/HPCA53966.2022.00054  

   Jalili, Majid; Erez, Mattan (April 2022, Proceedings of the 2022 IEEE International Symposium on High Performance Computer Architecture (HPCA))

   High load latency that results from deep cache hierarchies and relatively slow main memory is an important limiter of single-thread performance. Data prefetch helps reduce this latency by fetching data up the hierarchy before it is requested by load instructions. However, data prefetching has shown to be imperfect in many situations. We propose cache-level prediction to complement prefetchers. Our method predicts which memory hierarchy level a load will access allowing the memory loads to start earlier, and thereby saves many cycles. The predictor provides high prediction accuracy at the cost of just one cycle added latency to L1 misses. Level prediction reduces the memory access latency by 20% on average, and provides speedup of 10.3% over a conventional baseline, and 6.1% over a boosted baseline on generic, graph, and HPC applications. 

   more » « less
5. Stream Floating: Enabling Proactive and Decentralized Cache Optimizations

   https://doi.org/10.1109/HPCA51647.2021.00060  

   Wang, Zhengrong; Weng, Jian; Lowe-Power, Jason; Gaur, Jayesh; Nowatzki, Tony (February 2021, 2021 IEEE International Symposium on High-Performance Computer Architecture (HPCA))

   null (Ed.)

   As multicore systems continue to grow in scale and on-chip memory capacity, the on-chip network bandwidth and latency become problematic bottlenecks. Because of this, overheads in data transfer, the coherence protocol and replacement policies become increasingly important. Unfortunately, even in well-structured programs, many natural optimizations are difficult to implement because of the reactive and centralized nature of traditional cache hierarchies, where all requests are initiated by the core for short, cache line granularity accesses. For example, long-lasting access patterns could be streamed from shared caches without requests from the core. Indirect memory access can be performed by chaining requests made from within the cache, rather than constantly returning to the core. Our primary insight is that if programs can embed information about long-term memory stream behavior in their ISAs, then these streams can be floated to the appropriate level of the memory hierarchy. This decentralized approach to address generation and cache requests can lead to better cache policies and lower request and data traffic by proactively sending data before the cores even request it. To evaluate the opportunities of stream floating, we enhance a tiled multicore cache hierarchy with stream engines to process stream requests in last-level cache banks. We develop several novel optimizations that are facilitated by stream exposure in the ISA, and subsequent exposure to caches. We evaluate using a cycle-level execution-driven gem5-based simulator, using 10 data-processing workloads from Rodinia and 2 streaming kernels written in OpenMP. We find that stream floating enables 52% and 39% speedup over an inorder and OOO core with state of art prefetcher design respectively, with 64% and 49% energy efficiency advantage. 

   more » « less