Mitosis: A Speculative Multithreaded Processor Based on Precomputation Slices

Abstract

Abstract-This paper presents the Mitosis framework, which is a combined hardware-software approach to speculative multithreading, even in the presence of frequent dependences among threads. Speculative multithreading increases single-threaded application performance by exploiting thread-level parallelism speculatively, that is, executing code in parallel, even when the compiler or runtime system cannot guarantee that the parallelism exists. The proposed approach is based on predicting/computing thread input values via software through a piece of code that is added at the beginning of each thread (the precomputation slice). A precomputation slice is expected to compute the correct thread input values most of the time but not necessarily always. This allows aggressive optimization techniques to be applied to the slice to make it very short. This paper focuses on the microarchitecture that supports this execution model. The primary novelty of the microarchitecture is the hardware support for the execution and validation of precomputation slices. Additionally, this paper presents new architectures for the register file and the cache memory in order to support multiple versions of each variable and allow for efficient rollback in case of misspeculation. We show that the proposed microarchitecture, together with the compiler support, achieves an average speedup of 2.2 for applications that conventional nonspeculative approaches are not able to parallelize at all.

P-slice based efficient speculative multithreading

“…Previously proposed techniques to improve the memory locality in SpMT architectures can be classified into two categories: the ones that propose additional state bits in the cache lines to identify data that is retained across thread executions [11]; and those which attempt to snoop the bus for data used by other threads and replicate them in special local buffers [13].…”

“…It belongs to the latter category as mentioned above. Unlike [13], instead of replicating every data available on the bus, the compiler is responsible for selectively identifying the memory operations for which replication is done. Using profile information, the compiler inserts a hint in all those static memory instructions whose data, if replicated is likely to benefit the memory locality across the threads.…”

“…Code regions that can be speculatively parallelized are loop iterations, loop continuations, subroutines, modules or more complex schemes based on profiling. To handle inter-thread dependences there are several proposals like assuming no dependences, use of hardware/software value prediction [13] [15], synchronization [26] or speculative loop fission [27].…”

Prophet: A Speculative Multi-threading Execution Model with Architectural Support Based on CMP

“…This method introduces the false RAW violation while decreasing register transfer overhead and makes the hardware more complex. Using pre-computation technology proposed in paper [5,9], we design the Prophet model with architecture support based on CMP, where the program's correctness is guaranteed by the thread validation and commition mechanism. The thread commits its data to memory only when speculation successes and this is supported by hardware.…”

“…Most SpMT system relies on compiler to generate parallel threads [4,5,10,11]. We develop Prophet compiler to divided the sequential program into parallel threads despite uncertainty whether these thread are actually independent or not.…”