Fast dependence analysis in a multimedia vectorizing compiler

Abstract

There are a number of data dependence tests that have been proposed in the literature. In each test there is a different trade-off between accuracy and efficiency. The most widely used approximate data dependence tests are the Banerjee inequality and the GCD test; whereas the Omega test is a well-known exact data dependence test. We consider parallelization for microprocessors with a multimedia extension (the short SIMD execution model). For the short SIMD parallelism extraction it is essential that, if dependency exists, then the dependence distance is greater than or equal to the number of data processed in the SIMD register. This implies that some loops that could not be vectorized on traditional vector processors can still be parallelized for the short SIMD execution. In all of these tests the parallelization would be prohibited when actually there is no parallelism restriction relating to the short SIMD execution model. We present a new, fast data dependence test for array references with linear subscripts, which is used in a vectorizing compiler for microprocessors with a multimedia extension. Our test is suitable for use in a dependence analyser that is organized as a series of tests, progressively increasing in accuracy, as a replacement for the GCD or Banerjee tests.