Branch trace compression for snapshot-based simulation

Abstract

We present a scheme to compress branch trace information for use in snapshot-based microarchitecture simulation. The compressed trace can be used to warm any arbitrary branch predictor's state before detailed microarchitecture simulation of the snapshot. We show that compressed branch traces require less space than snapshots of concrete predictor state. Our branch-predictor based compression (BPC) technique uses a software branch predictor to provide an accurate model of the input branch trace, requiring only mispredictions to be stored in the compressed trace file. The decompressor constructs a matching software branch predictor to help reconstruct the original branch trace from the record of mispredictions. Evaluations using traces from the Journal of ILP branch predictor competition show we achieve compression rates of 0.013-0.72 bits/branch (depending on workload), which is up to 210× better than gzip; up to 52× better than the best general-purpose compression techniques; and up to 4.4× better than recently published, more general trace compression techniques. Moreover, BPC-compressed traces can be decompressed in less time than corresponding traces compressed with other methods.

Online compression of cache-filtered address traces

“…It was shown that the compressors generated by TCgen have a very high compression ratio [5,16]. Another predictor-based compressor was proposed by Barr and Asanović but specialized for control-flow traces [2]. The compression method proposed by Marathe et al targets trace records consisting of one instruction address and one data address [15].…”

“…We have chosen a VPC-like compressor/decompressor that matches approximately the amount of memory used by the big bytesort. We used the following TCgen specification : 0-Bit Header; 64-Bit Field 1 = L1 = 1, L2 = 1048576: DFCM3 [2], FCM3 [3], FCM2 [3], FCM1 [3]; ID = Field 1; Compressor = 'bzip2 -c -z -9'; Decompressor = 'bzip2 -c -d'; . This compressor/decompressor uses 232 Mbytes of memory.…”

“…We describe in Section 5.1 a method that is geared to our goals. The question of choosing a "good" value for L is actually more problematic, 2 Disk accesses contribute little to the total decompression time.…”

“…At the end of the first interval, we compute the histograms for the interval and we store them in the histogram table. Then we compute the histograms for the second interval and we compute the distance between intervals 1 and 2 using formula (2). If the distance is greater than the threshold, we create a chunk for the second interval and we store the histograms of the second interval in the histogram table.…”

Hardware-based data value and address trace filtering techniques

“…A number of trace-specific compression techniques have been proposed [8][9][10]. However, they are mainly focused on software-based compression, combining trace-specific compression with a general-purpose compression algorithm, such as gzip or bzip2.…”