4 ms·
Performance Left on the Table: An Eval of Compiler Autovectorization for RISC-V
- sposeray 4y ago[dead]
- therealcamino 4y agoI love good, careful studies of performance bottlenecks in systems with lots of moving parts (input coding style, compiler analyses, compiler optimizer and the interactions between scalar and vector sides, hardware implementation). It can be really interesting to find out what counterintuitive or unintended interactions there are. I think it's a real service to the community to publish these kinds of studies, so everyone gets the benefit of the analysis.