Load-Store Conflicts
a year ago
- #compression
- #microarchitecture
- #performance
- meshoptimizer implements geometry compression algorithms for efficient mesh data decompression.
- Performance variance in index decoder across compilers linked to microarchitectural details.
- Edge FIFO structure used for redundancy in encoding/decoding triangle indices.
- Store-to-load forwarding critical for performance; issues arise with mismatched load/store sizes.
- GCC-14 outperforms Clang-20 by using vector operations for FIFO updates.
- GCC-15 introduces significant performance regression due to store-load conflicts.
- Apple M4 shows impressive performance with Clang-17, leveraging efficient load/store pairs.
- Store-load forwarding issues can cause unexpected performance cliffs in high-performance code.