Draft: Refactored IPWRTBaseSerialSumFac with cached quadmetrics

Issue/feature addressed

The current implementation of IPWRTBaseSerialSumFac does not perform quadrature metric (weights + jacobian) caching leading to 30-50% slower runtime performance compared to legacy's implementation with caching.

Proposed solution

Allocate a pointer which caches quadrature metric (weight+jacobian) in an initial compute (i.e first time-step) and which is used to perform a single Vmul with and an input vector later.

Implementation

1. Performs kernel functions MultiplyByQuadMetricSeg/Tri/Quad/Tet/Pyr/Prism/HexKernel (in IProductWRTBaseSerialSumFac.hpp:118,163,207,261,315,370,423,) that caches the corresponding quadrature metric into an array QuadMetric (IProductWRTBaseSerialSumFac.hpp:474) which is accessed by the pointer quadmetricptr (IProductWRTBaseSerialSumFac.hpp:476)
2. When the quadrature metric is cached, MetricCount = True. Next, we perform a vector multiplication (vmul) between the input vector and the cached metric and stores the results into wsp (IProductWRTBaseSerialSumFac.hpp:123,169,212,268,321,376,429), which will be used to perform the inner product with bases.

Tests

Passed all tests with test_ipwrtbase_serial_sumfac.cpp

Suggested reviewers

Jacques

Checklist

• Suitable tests added for new functionality.
• Contributed code is correctly formatted. (See the contributing guidelines).
• License added to any new files.