New Intel Tremont Optimizations Heading To The GCC Compiler
The GNU Compiler Collection has already supported Intel's Tremont cores as used by the low-power Jasper Lake platform. Now though coming to GCC are some optimizations to further enhance the performance when targeting the Tremont micro-architecture.
The new GCC Tremont patches include updating the memcpy/memset inline strategies, updated handling around vector and SSE conversions, switching to using the Haswell scheduling mountain, enabling a number of other options.
What's important is the end result and with these four patches touching just over 300 lines of code is around a 1% improvement for SPEC CPU 2017's INTrate performance but around 7.5% higher FPrate.
These Intel Tremont improvements for GCC are currently available via this patch series while should be mainlined soon for GCC 12.
The new GCC Tremont patches include updating the memcpy/memset inline strategies, updated handling around vector and SSE conversions, switching to using the Haswell scheduling mountain, enabling a number of other options.
What's important is the end result and with these four patches touching just over 300 lines of code is around a 1% improvement for SPEC CPU 2017's INTrate performance but around 7.5% higher FPrate.
These Intel Tremont improvements for GCC are currently available via this patch series while should be mainlined soon for GCC 12.
2 Comments