Citation Context

...ax 42f543: lea 0xffffffffffffffff(%rdi),%rdx 42f547: and %rdx,%rdi 42f54a: jne 42f540 42f54c: pop %rcx # unnecessary 42f54d: retq Figure 8. PopCnt() code generated by compiler A. lenge using SimPoint =-=[28]-=-—a methodology that uses phase analysis for finding representative regions for simulation. For the second challenge, we use Pin to collect SimPoint profiles (which we call PinPoints) and instruction t...

Citation Context

...runs. 1 Researchers predominantly skip the initial 250 million to two billion instructions and then measure a single section of 100 million to one billion instructions. Unfortunately, several studies =-=[4,10,12,17]-=- have concluded that results based only on a single abbreviated execution stream are inaccurate or misleading because they fail to capture global variations in program behavior and performance. Anothe...

Citation Context

...end of each interval), and thus is not nearly as performance critical. We note that the hashing function we use is fundamentally the same as the random projection method we used to generate phases in =-=[21]-=-. In this prior work, we make use of random projections of the data to reduce the dimensionality of the samples being taken. A random projection takes trace data in the form of amatrixofsizeL×B, where...

Citation Context

...chnique, we demonstrate here how we can use component-based power breakdowns to identify power phases of programs. Several prior papers have proposed methods for detecting or exploiting program phases=-=[1, 5, 9, 22, 23, 29]-=-. Our example here is distinct because we focus on power phases rather than performance phases. A more detailed description of our power phase research can be found in [14]. We use the similarity matr...

Citation Context

...The variations in application behavior are often repetitive because of loop-oriented execution semantics. The different regions of execution are commonly referred to as phases of application behavior =-=[39]-=-. Adaptive responses to such phases for power-efficient computing in a uniprocessor setting have been well studied in the past (e.g. [1]). In a multi-core setting, the global power manager must observ...

Citation Context

... used the SimpleScalar simulator version 3.0b [4] to model AM architectures for a four-issue out-of-order superscalar processor. The Simplescalar simulation parameters are given in Table 4. SimPoints =-=[9]-=- were used to fast-forward benchmark initialization phase and choose a representative 100 Million-Cycle section of each program. To further reduce simulation time, a subset of the SPEC2000 benchmarks ...

Citation Context

...thods that define and use metrics to dynamically identify phases for adaptive optimization [5, 6, 8, 13, 26], and into techniques that identify appropriate simulation points for the desired workloads =-=[10, 16, 17, 24, 25]-=-. Balasubramonian et al. [5, 6] use interval-based exploration mechanisms in order to dynamically reconfigure the underlying micro-architecture to meet the need of an application. They identify a chan...

Citation Context

...n information for recurring phases, thereby improving performance [11][12][19]. There have been several proposals to implicitly [13] or explicitly detect program phase changes [2][10][11][12] [19][20]=-=[21]-=-. Recently, several researchers have proposed hardware techniques aimed specifically at detecting phase changes, identifying phases and predicting phases [2][10] [11][12][19]. In this work, we focus o...