Advanced SIMD Algorithms in Pictures
summary
Slides and demos show concrete SIMD wins: an EVE library implementation uses actual SIMD/register intrinsics on x86 and M1 with 32‑byte registers, yielding concrete numbers — strcmp/mismatch variants hit about 3.5× speedup over scalar in some cases and microbenchmarks report up to ~20× for set_intersect in sparse workloads. Code discusses BMI2/bit‑mask tricks, compress/copy_if strategies, register sorting networks and where implementations fall back or don’t fit (tails, alignment, power/frequency effects).
AI-written summary · about
chapters (7)
- 0:00SIMD and performance trade-offs
- 6:21SIMD registers and operations
- 15:53Mismatch algorithm with SIMD
- 24:50Copy-if algorithm and SIMD
- 33:50Sorting algorithms with SIMD
- 39:55Set intersection with SIMD
- 43:37Q&A on SIMD hardware
related talks
01h 06mSIMD in C++20: eve of a new era
01h 08mAn implementer's view of Boost.SIMD
42mFaster programs with your compilers autovectorization feature
01h 06mUtilize your CPU power
13mBaby Demuxed's First Assembly Language Function
01h 07mWriting Data Parallel Algorithms on GPUs
01h 05mThe Most Important Optimizations to Apply in Your C++ Programs
04mInterview