[SYSTEMDS-3874] Java17 Vectorized LibMM

[Baunsgaard]

This PR contains code for vectorized MM It further include some vectorized instructions for CLA.

[Baunsgaard]

The results show that the vector api improve performance, for single threaded execution our AMD box improve by ~80% and Intel ~60% for dense mm. These improvements, are with allocation overhead of the output, and in ideal cases where the input are cached properly, and the JIT compilation is done.

--- The failing tests are related to some sparse kernels with wrong results, will be fixed shortly.

Script to evaluate mm performance:

./src/test/scripts/performance/matrixMultiplication.sh

After:

SU1
MM Perf : rep 5000 -- [1009, 100, 100, 1000, 1, 1]
                    mm SingleThread,    2.170+-  0.129 ms,           
                mm MultiThread: 128,    1.675+-  0.198 ms,           
MM Perf : rep 100 -- [1009, 1000, 1000, 1000, 1, 1]
                    mm SingleThread,  215.661+- 44.191 ms,           
                mm MultiThread: 128,   23.836+-  1.566 ms,           
MM Perf : rep 100 -- [1009, 10000, 1000, 1000, 1, 1]
                    mm SingleThread, 2189.817+-407.982 ms,           
                mm MultiThread: 128,   62.026+-  2.399 ms,           
MM Perf : rep 100 -- [1009, 1000, 10000, 1000, 1, 1]
                    mm SingleThread, 2084.534+- 42.736 ms,           
                mm MultiThread: 128,   57.026+-  2.680 ms,           
MM Perf : rep 100 -- [1009, 1000, 1000, 10000, 1, 1]
                    mm SingleThread, 2128.684+- 40.128 ms,           
                mm MultiThread: 128,   72.801+-  6.310 ms,           
MM Perf : rep 100 -- [1009, 10000, 10000, 10000, 1, 1]
                    mm SingleThread, 212909.303+-225.702 ms,           
                mm MultiThread: 128, 4725.499+- 51.779 ms, 
            

SO009
MM Perf : rep 5000 -- [1009, 100, 100, 1000, 1, 1]
                    mm SingleThread,    2.239+-  0.289 ms,           
                 mm MultiThread: 48,    0.360+-  0.052 ms,           
MM Perf : rep 100 -- [1009, 1000, 1000, 1000, 1, 1]
                    mm SingleThread,  189.776+- 74.425 ms,           
                 mm MultiThread: 48,   12.043+-  0.635 ms,           
MM Perf : rep 100 -- [1009, 10000, 1000, 1000, 1, 1]
                    mm SingleThread, 1124.289+-273.464 ms,           
                 mm MultiThread: 48,   77.630+-  8.706 ms,           
MM Perf : rep 100 -- [1009, 1000, 10000, 1000, 1, 1]
                    mm SingleThread, 1152.994+-181.044 ms,           
                 mm MultiThread: 48,   74.188+- 15.157 ms,           
MM Perf : rep 100 -- [1009, 1000, 1000, 10000, 1, 1]
                    mm SingleThread, 1171.379+- 48.588 ms,           
                 mm MultiThread: 48,   87.808+- 15.199 ms,           
MM Perf : rep 100 -- [1009, 10000, 10000, 10000, 1, 1]
                    mm SingleThread, 112851.281+-1033.173 ms,           
                 mm MultiThread: 48, 6611.407+- 77.775 ms, 

Before:

SU1
MM Perf : rep 5000 -- [1009, 100, 100, 1000, 1, 1]
                    mm SingleThread,    3.991+-  0.843 ms,           
                mm MultiThread: 128,    1.214+-  0.368 ms,           
MM Perf : rep 100 -- [1009, 1000, 1000, 1000, 1, 1]
                    mm SingleThread,  331.139+- 13.338 ms,           
                mm MultiThread: 128,   32.462+-  3.078 ms,           
MM Perf : rep 100 -- [1009, 10000, 1000, 1000, 1, 1]
                    mm SingleThread, 3489.530+-549.035 ms,           
                mm MultiThread: 128,  115.628+-  3.087 ms,           
MM Perf : rep 100 -- [1009, 1000, 10000, 1000, 1, 1]
                    mm SingleThread, 3396.771+- 18.842 ms,           
                mm MultiThread: 128,  107.051+-  3.761 ms,           
MM Perf : rep 100 -- [1009, 1000, 1000, 10000, 1, 1]
                    mm SingleThread, 3513.700+-370.610 ms,           
                mm MultiThread: 128,  117.059+-  3.606 ms,           
MM Perf : rep 100 -- [1009, 10000, 10000, 10000, 1, 1] 
                    mm MultiThread: 128, 8428.286+- 71.497 ms,     

SO009
MM Perf : rep 5000 -- [1009, 100, 100, 1000, 1, 1]
                    mm SingleThread,    3.409+-  0.111 ms,           
                 mm MultiThread: 48,    0.548+-  0.043 ms,           
MM Perf : rep 100 -- [1009, 1000, 1000, 1000, 1, 1]
                    mm SingleThread,  283.795+-108.979 ms,           
                 mm MultiThread: 48,   21.052+-  0.719 ms,           
MM Perf : rep 100 -- [1009, 10000, 1000, 1000, 1, 1]
                    mm SingleThread, 2016.267+-199.360 ms,           
                 mm MultiThread: 48,  125.341+- 15.542 ms,           
MM Perf : rep 100 -- [1009, 1000, 10000, 1000, 1, 1]
                    mm SingleThread, 2075.305+-190.909 ms,           
                 mm MultiThread: 48,  123.638+- 24.349 ms,           
MM Perf : rep 100 -- [1009, 1000, 1000, 10000, 1, 1]
                    mm SingleThread, 2100.731+-189.725 ms,           
                 mm MultiThread: 48,  132.040+- 24.115 ms,           
MM Perf : rep 100 -- [1009, 10000, 10000, 10000, 1, 1]
                 mm MultiThread: 48, 11441.370+-103.849 ms,  

[codecov]

Codecov Report

Attention: Patch coverage is 90.51095% with 13 lines in your changes missing coverage. Please review.

Project coverage is 72.95%. Comparing base (499b6e3) to head (1a819a3).
Report is 4 commits behind head on main.

Files with missing lines	Patch %	Lines
...e/sysds/runtime/compress/colgroup/ColGroupDDC.java	41.66%	7 Missing ⚠️
...pache/sysds/runtime/matrix/data/LibMatrixMult.java	92.50%	6 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff            @@
##               main    #2216   +/-   ##
=========================================
  Coverage     72.94%   72.95%           
- Complexity    46061    46076   +15     
=========================================
  Files          1479     1479           
  Lines        172576   172613   +37     
  Branches      33776    33783    +7     
=========================================
+ Hits         125893   125928   +35     
- Misses        37186    37194    +8     
+ Partials       9497     9491    -6     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.