PR #7303 added changes to improve the overhead of the global register allocator and PR #7328 fixed a bug in the newly introduced code.
Performance experiments have shown that while these changes reduce the CPU consumed by the JIT compilation threads by ~7.5%, they also cause throughput regressions on Daytrader (~1%), AcmeAIREE8 (~2%) and Ilog (~1%) benchmarks.
PR #7303 added changes to improve the overhead of the global register allocator and PR #7328 fixed a bug in the newly introduced code. Performance experiments have shown that while these changes reduce the CPU consumed by the JIT compilation threads by ~7.5%, they also cause throughput regressions on Daytrader (~1%), AcmeAIREE8 (~2%) and Ilog (~1%) benchmarks.