[GR-74060] Implement AVX2 and AVX512 paths for byte compress and shift operations

[graalvmbot]

Implements fallback paths for vectorized byte shift and compress operations when no direct AVX512 hardware instructions are available. Shifting is done by extending to a larger element size, using an appropriate shift, and narrowing back. Compress is implemented in terms of 128-bit shuffles; for larger vectors, the final result is assembled on the stack out of the results of compressing 128-bit lanes.

https://github.com/debe/libfindchars benchmark results from my old-ish i9-10920X workstation:

AVX2

                                                           ----- C2 -----  --- Graal ----
Benchmark                            (config)   Mode  Cnt   Score   Error   Score   Error  Units
SweepBenchmark.regex                 2-15-0-2  thrpt    5  15.030 ± 0.029  17.965 ± 0.077  ops/s
SweepBenchmark.regex                 4-15-0-2  thrpt    5  15.044 ± 0.021  17.812 ± 0.032  ops/s
SweepBenchmark.regex                 8-15-0-2  thrpt    5  14.830 ± 0.060  18.079 ± 0.022  ops/s
SweepBenchmark.regex                12-15-0-2  thrpt    5  14.710 ± 0.073  18.079 ± 0.064  ops/s
SweepBenchmark.regex                  8-5-0-2  thrpt    5  24.186 ± 0.066  26.095 ± 0.011  ops/s
SweepBenchmark.regex                 8-30-0-2  thrpt    5   9.997 ± 1.032  11.452 ± 0.087  ops/s
SweepBenchmark.regex                 8-50-0-2  thrpt    5   7.317 ± 0.082   8.734 ± 0.143  ops/s
SweepBenchmark.regex                 8-15-1-2  thrpt    5   9.261 ± 0.016  12.804 ± 0.026  ops/s
SweepBenchmark.regex                 8-15-0-1  thrpt    5  14.948 ± 0.008  17.901 ± 0.037  ops/s
SweepBenchmark.regex                 8-15-0-4  thrpt    5  14.864 ± 0.070  17.946 ± 0.084  ops/s
SweepBenchmark.regex                 8-15-0-8  thrpt    5  14.856 ± 0.041  18.047 ± 0.074  ops/s
SweepBenchmark.regexWithConversion   2-15-0-2  thrpt    5  13.820 ± 0.105  17.439 ± 0.192  ops/s
SweepBenchmark.regexWithConversion   4-15-0-2  thrpt    5  13.796 ± 0.084  17.439 ± 0.124  ops/s
SweepBenchmark.regexWithConversion   8-15-0-2  thrpt    5  14.004 ± 0.059  17.331 ± 0.574  ops/s
SweepBenchmark.regexWithConversion  12-15-0-2  thrpt    5  14.041 ± 0.039  17.565 ± 0.099  ops/s
SweepBenchmark.regexWithConversion    8-5-0-2  thrpt    5  22.137 ± 0.121  25.915 ± 0.090  ops/s
SweepBenchmark.regexWithConversion   8-30-0-2  thrpt    5   9.390 ± 0.233  11.483 ± 0.158  ops/s
SweepBenchmark.regexWithConversion   8-50-0-2  thrpt    5   7.151 ± 0.043   8.274 ± 1.107  ops/s
SweepBenchmark.regexWithConversion   8-15-1-2  thrpt    5   7.795 ± 0.295  10.539 ± 0.031  ops/s 
SweepBenchmark.regexWithConversion   8-15-0-1  thrpt    5  13.578 ± 0.083  17.435 ± 0.144  ops/s
SweepBenchmark.regexWithConversion   8-15-0-4  thrpt    5  13.986 ± 0.117  17.745 ± 0.196  ops/s
SweepBenchmark.regexWithConversion   8-15-0-8  thrpt    5  14.006 ± 0.109  17.560 ± 0.156  ops/s
SweepBenchmark.simdC2Jit             2-15-0-2  thrpt    5  15.702 ± 0.184  92.406 ± 0.030  ops/s
SweepBenchmark.simdC2Jit             4-15-0-2  thrpt    5  16.182 ± 0.165  93.510 ± 0.133  ops/s
SweepBenchmark.simdC2Jit             8-15-0-2  thrpt    5  15.438 ± 0.171  80.587 ± 0.133  ops/s
SweepBenchmark.simdC2Jit            12-15-0-2  thrpt    5  15.331 ± 0.185  78.441 ± 0.167  ops/s
SweepBenchmark.simdC2Jit              8-5-0-2  thrpt    5  32.318 ± 0.325  87.664 ± 0.225  ops/s
SweepBenchmark.simdC2Jit             8-30-0-2  thrpt    5   8.511 ± 0.052  73.431 ± 0.124  ops/s
SweepBenchmark.simdC2Jit             8-50-0-2  thrpt    5   6.242 ± 0.040  65.357 ± 0.236  ops/s
SweepBenchmark.simdC2Jit             8-15-1-2  thrpt    5  11.690 ± 0.047  50.358 ± 0.656  ops/s
SweepBenchmark.simdC2Jit             8-15-0-1  thrpt    5  15.961 ± 0.134  93.151 ± 0.140  ops/s
SweepBenchmark.simdC2Jit             8-15-0-4  thrpt    5  15.987 ± 0.188  95.057 ± 0.169  ops/s
SweepBenchmark.simdC2Jit             8-15-0-8  thrpt    5  16.472 ± 0.141  93.240 ± 0.070  ops/s
SweepBenchmark.simdCompiled          2-15-0-2  thrpt    5  15.689 ± 0.190  95.202 ± 0.139  ops/s
SweepBenchmark.simdCompiled          4-15-0-2  thrpt    5  16.048 ± 0.190  93.917 ± 0.142  ops/s
SweepBenchmark.simdCompiled          8-15-0-2  thrpt    5  15.508 ± 0.142  74.007 ± 0.098  ops/s
SweepBenchmark.simdCompiled         12-15-0-2  thrpt    5  15.565 ± 0.055  72.782 ± 0.113  ops/s
SweepBenchmark.simdCompiled           8-5-0-2  thrpt    5  33.033 ± 0.366  85.474 ± 0.216  ops/s
SweepBenchmark.simdCompiled          8-30-0-2  thrpt    5   9.088 ± 0.091  70.767 ± 0.153  ops/s
SweepBenchmark.simdCompiled          8-50-0-2  thrpt    5   6.199 ± 0.051  63.415 ± 0.250  ops/s
SweepBenchmark.simdCompiled          8-15-1-2  thrpt    5  14.284 ± 0.104  48.948 ± 8.457  ops/s
SweepBenchmark.simdCompiled          8-15-0-1  thrpt    5  15.825 ± 0.181  95.638 ± 0.074  ops/s
SweepBenchmark.simdCompiled          8-15-0-4  thrpt    5  15.710 ± 0.181  94.250 ± 0.093  ops/s
SweepBenchmark.simdCompiled          8-15-0-8  thrpt    5  16.157 ± 0.217  94.970 ± 0.156  ops/s

AVX512, no VBMI2

                                                           ----- C2 -----  ---- Graal ----
Benchmark                            (config)   Mode  Cnt   Score   Error    Score   Error  Units
SweepBenchmark.regex                 2-15-0-2  thrpt    5  13.867 ± 0.072   17.972 ± 0.019  ops/s
SweepBenchmark.regex                 4-15-0-2  thrpt    5  13.910 ± 0.073   18.985 ± 0.081  ops/s
SweepBenchmark.regex                 8-15-0-2  thrpt    5  13.466 ± 0.110   19.600 ± 0.100  ops/s
SweepBenchmark.regex                12-15-0-2  thrpt    5  14.032 ± 0.057   17.703 ± 0.085  ops/s
SweepBenchmark.regex                20-15-0-2  thrpt    5  14.432 ± 0.153   17.970 ± 0.091  ops/s
SweepBenchmark.regex                  8-5-0-2  thrpt    5  20.954 ± 0.048   27.721 ± 0.031  ops/s
SweepBenchmark.regex                 8-30-0-2  thrpt    5  10.757 ± 0.031   12.027 ± 0.174  ops/s
SweepBenchmark.regex                 8-50-0-2  thrpt    5   7.457 ± 0.108    9.093 ± 0.054  ops/s
SweepBenchmark.regex                 8-15-1-2  thrpt    5   9.365 ± 0.025   12.176 ± 0.019  ops/s
SweepBenchmark.regex                 8-15-2-2  thrpt    5   6.195 ± 0.020    8.592 ± 0.018  ops/s
SweepBenchmark.regex                 8-15-3-2  thrpt    5   5.641 ± 0.010    8.134 ± 0.016  ops/s
SweepBenchmark.regex                 8-15-0-1  thrpt    5  13.649 ± 0.105   19.248 ± 0.083  ops/s
SweepBenchmark.regex                 8-15-0-4  thrpt    5  13.462 ± 0.029   17.964 ± 0.037  ops/s
SweepBenchmark.regex                 8-15-0-8  thrpt    5  13.150 ± 0.096   18.890 ± 0.080  ops/s
SweepBenchmark.regexWithConversion   2-15-0-2  thrpt    5  13.744 ± 0.168   17.801 ± 0.078  ops/s
SweepBenchmark.regexWithConversion   4-15-0-2  thrpt    5  13.723 ± 0.232   17.546 ± 0.054  ops/s
SweepBenchmark.regexWithConversion   8-15-0-2  thrpt    5  13.748 ± 0.169   17.334 ± 0.268  ops/s
SweepBenchmark.regexWithConversion  12-15-0-2  thrpt    5  13.379 ± 0.097   17.484 ± 0.040  ops/s
SweepBenchmark.regexWithConversion  20-15-0-2  thrpt    5  13.746 ± 0.166   17.539 ± 0.057  ops/s
SweepBenchmark.regexWithConversion    8-5-0-2  thrpt    5  21.639 ± 0.147   25.946 ± 0.040  ops/s
SweepBenchmark.regexWithConversion   8-30-0-2  thrpt    5  10.268 ± 0.283   11.591 ± 0.043  ops/s
SweepBenchmark.regexWithConversion   8-50-0-2  thrpt    5   7.309 ± 0.231    8.498 ± 0.148  ops/s
SweepBenchmark.regexWithConversion   8-15-1-2  thrpt    5   7.923 ± 0.229   10.909 ± 0.054  ops/s
SweepBenchmark.regexWithConversion   8-15-2-2  thrpt    5   5.568 ± 0.074    7.703 ± 0.016  ops/s
SweepBenchmark.regexWithConversion   8-15-3-2  thrpt    5   5.091 ± 0.253    6.704 ± 0.015  ops/s
SweepBenchmark.regexWithConversion   8-15-0-1  thrpt    5  13.401 ± 0.078   17.532 ± 0.063  ops/s
SweepBenchmark.regexWithConversion   8-15-0-4  thrpt    5  13.439 ± 0.087   17.458 ± 0.192  ops/s
SweepBenchmark.regexWithConversion   8-15-0-8  thrpt    5  13.364 ± 0.055   17.533 ± 0.075  ops/s
SweepBenchmark.simdC2Jit             2-15-0-2  thrpt    5  14.730 ± 0.179  105.139 ± 1.267  ops/s
SweepBenchmark.simdC2Jit             4-15-0-2  thrpt    5  14.871 ± 0.153  105.135 ± 0.088  ops/s
SweepBenchmark.simdC2Jit             8-15-0-2  thrpt    5  14.684 ± 0.133  105.114 ± 0.257  ops/s
SweepBenchmark.simdC2Jit            12-15-0-2  thrpt    5  14.239 ± 0.127   84.657 ± 0.089  ops/s
SweepBenchmark.simdC2Jit            20-15-0-2  thrpt    5  14.238 ± 0.125   84.225 ± 0.079  ops/s
SweepBenchmark.simdC2Jit              8-5-0-2  thrpt    5  32.286 ± 0.332  118.308 ± 0.197  ops/s
SweepBenchmark.simdC2Jit             8-30-0-2  thrpt    5   8.011 ± 0.064   92.980 ± 0.178  ops/s
SweepBenchmark.simdC2Jit             8-50-0-2  thrpt    5   5.489 ± 0.009   80.916 ± 0.819  ops/s
SweepBenchmark.simdC2Jit             8-15-1-2  thrpt    5  13.505 ± 0.139   64.742 ± 1.431  ops/s
SweepBenchmark.simdC2Jit             8-15-2-2  thrpt    5  13.564 ± 0.136   65.330 ± 0.638  ops/s
SweepBenchmark.simdC2Jit             8-15-3-2  thrpt    5  11.590 ± 0.052   60.246 ± 0.588  ops/s
SweepBenchmark.simdC2Jit             8-15-0-1  thrpt    5  14.887 ± 0.143  105.403 ± 0.254  ops/s
SweepBenchmark.simdC2Jit             8-15-0-4  thrpt    5  14.805 ± 0.151  105.299 ± 0.129  ops/s
SweepBenchmark.simdC2Jit             8-15-0-8  thrpt    5  14.694 ± 0.133  105.443 ± 0.035  ops/s
SweepBenchmark.simdCompiled          2-15-0-2  thrpt    5  14.680 ± 0.123  104.267 ± 0.052  ops/s
SweepBenchmark.simdCompiled          4-15-0-2  thrpt    5  14.740 ± 0.128  104.226 ± 0.049  ops/s
SweepBenchmark.simdCompiled          8-15-0-2  thrpt    5  14.806 ± 0.105  104.264 ± 0.238  ops/s
SweepBenchmark.simdCompiled         12-15-0-2  thrpt    5  14.080 ± 0.102   84.752 ± 0.145  ops/s
SweepBenchmark.simdCompiled         20-15-0-2  thrpt    5  14.146 ± 0.133   84.702 ± 0.036  ops/s
SweepBenchmark.simdCompiled           8-5-0-2  thrpt    5  32.280 ± 0.268  118.720 ± 0.753  ops/s
SweepBenchmark.simdCompiled          8-30-0-2  thrpt    5   7.986 ± 0.053   92.323 ± 0.075  ops/s
SweepBenchmark.simdCompiled          8-50-0-2  thrpt    5   5.435 ± 0.008   80.346 ± 0.212  ops/s
SweepBenchmark.simdCompiled          8-15-1-2  thrpt    5  13.528 ± 0.238   65.521 ± 1.039  ops/s
SweepBenchmark.simdCompiled          8-15-2-2  thrpt    5  13.436 ± 0.118   59.264 ± 2.396  ops/s
SweepBenchmark.simdCompiled          8-15-3-2  thrpt    5  12.888 ± 0.124   58.988 ± 0.403  ops/s
SweepBenchmark.simdCompiled          8-15-0-1  thrpt    5  14.762 ± 0.096  104.201 ± 0.026  ops/s
SweepBenchmark.simdCompiled          8-15-0-4  thrpt    5  14.696 ± 0.141  104.151 ± 0.159  ops/s
SweepBenchmark.simdCompiled          8-15-0-8  thrpt    5  14.765 ± 0.126  104.177 ± 1.330  ops/s

The cases where we beat C2 by 5-10x seem to be cases where C2 fails to inline VectorSupport::compressExpandOp into ByteVector::compressTemplate (callee is too large), at least according to the agent that I asked to look at its logs. Without that, we're broadly on par.