Smaller adjustments to buy us capacity

[rschwietzke]

These changes won't change the runtime but the resource need, aka freeing things up. Less branches, less instructions, less cycles, less cache misses.

Old Performance Data

        660,414.10 msec task-clock                       #    7.386 CPUs utilized             
           261,219      context-switches                 #  395.538 /sec                      
            27,304      cpu-migrations                   #   41.344 /sec                      
         1,559,640      page-faults                      #    2.362 K/sec                     
 2,155,072,417,633      cycles                           #    3.263 GHz                       
   394,882,775,879      stalled-cycles-frontend          #   18.32% frontend cycles idle      
 4,062,894,368,682      instructions                     #    1.89  insn per cycle            
                                                  #    0.10  stalled cycles per insn   
   751,461,828,221      branches                         #    1.138 G/sec                     
    14,439,955,921      branch-misses                    #    1.92% of all branches           

      89.417514945 seconds time elapsed

     593.675541000 seconds user
      62.852506000 seconds sys

New Performance Data

        645,923.27 msec task-clock                       #    7.331 CPUs utilized             
           274,570      context-switches                 #  425.081 /sec                      
            29,801      cpu-migrations                   #   46.137 /sec                      
         1,582,803      page-faults                      #    2.450 K/sec                     
 2,096,071,549,319      cycles                           #    3.245 GHz                       
   372,558,046,965      stalled-cycles-frontend          #   17.77% frontend cycles idle      
 3,883,824,436,885      instructions                     #    1.85  insn per cycle            
                                                  #    0.10  stalled cycles per insn   
   707,559,172,285      branches                         #    1.095 G/sec                     
    14,447,594,308      branch-misses                    #    2.04% of all branches           

      88.107659677 seconds time elapsed

     576.587159000 seconds user
      64.967407000 seconds sys

[Copilot]

Pull Request Overview

This PR implements several performance optimizations aimed at reducing CPU cycles, instructions, branches, and cache misses without changing the runtime behavior. The changes focus on optimizing hot code paths in data processing and hash computation.

Key changes:

• Replace reflection-based object creation with direct instantiation using switch expressions for common data types
• Optimize hash computation algorithms to reduce operations and improve cache efficiency
• Remove unnecessary hash computation calls during object initialization

Reviewed Changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

File	Description
DataRecordFactory.java	Replaces reflection with switch-based direct instantiation for common data record types
XltCharBuffer.java	Refactors hash computation to reduce operations and improve performance
AbstractData.java	Removes premature hash computation during object initialization
StringHasher.java	Adds optimized hash method for XltCharBuffer using native String operations

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[Copilot]

The hash computation formula (src[i] - h) + (h << 5) differs significantly from the standard multiplication-based approach used in the main loop. This inconsistency makes the algorithm harder to understand and may produce different hash distribution characteristics. Consider using a consistent approach like h = 31 * h + src[i] or document why this specific formula was chosen.

Suggested change

	h = (src[i] - h) + (h << 5);
	h = 31 * h + src[i];

[Copilot]

The condition pos > 0 will skip hashing when the limiting character is at position 0, returning the full string's hash instead of an empty string's hash. This should be pos >= 0 to handle the case where the limiting character is the first character correctly.