Benchmarks and optimization

[beepy0]

The goals are as follows:

• [x] #17 #18 Benchmark both algorithms’ baseline speed
• [x] #19 #20 Analyze algorithms for hotspots
• [x] #21 #22 Determine existence of memory bounds
• [x] #23 #24 Determine existence of compute bounds
• [ ] #6 Adapt to use SIMD instructions where suitable
• [ ] #25 #26 Consider memory-tuning if feasable
• [ ] #27 #28 Benchmark improved algorithms

[beepy0]

• throughput as num inputs/sec or even mb/sec
• AGMS/Fast-AGMS: vector size vs. actual throughput graph. Check what happens with SIMD after that
• H3 : could vectorize computation for a single value instead of multiple inputs.
• plot for relative acc/throughput for Fast-AGMS based on rows/buckets size ratio
• Chart experiments and make diagrams in advance so you understand the test before you get the actual values.
• throughput chart for both algorithms based on rows/buckets size, display throughput change after update vector/matrix no longer fits into L1/L2 cache.