Re: RFR: 8290824: Use InputStream.readAllBytes() instead of explicit loops [v2]

[Сергей Цыпанов]

On Mon, 25 Jul 2022 06:40:49 GMT, Сергей Цыпанов <d...@openjdk.org> wrote:

>> We can use `InputStream.readAllBytes()` in `ModuleHashes` and `X509CertPath`.
>
> Сергей Цыпанов has updated the pull request incrementally with one additional 
> commit since the last revision:
> 
>   8278461: Revert X509CertPath

I've measured this with benchmark

@State(Scope.Thread)
@BenchmarkMode(Mode.AverageTime)
@OutputTimeUnit(TimeUnit.MILLISECONDS)
@Fork(jvmArgsAppend = {"-Xms1g", "-Xmx1g"})
public class HashBenchmark {

  private final Path path = 
Path.of("/Users/stsypanov/Library/Java/JavaVirtualMachines/openjdk-18.0.2/Contents/Home/jmods");
  private final ModuleFinder finder = ModulePath.of(Runtime.version(), true, 
path);

  @Benchmark
  public byte[] hash() throws Exception {
    return hash(finder, "java.base");
  }

  private byte[] hash(ModuleFinder finder, String name) throws Exception {
    ModuleReference mref = finder.find(name).orElseThrow(RuntimeException::new);
    try (ModuleReader reader = mref.open()) {
      return ModuleHashes.computeHash(reader, "SHA-256");
    }
  }
}

and indeed the change makes things slower:

baseline
Benchmark                                            Mode  Cnt         Score    
     Error   Units
HashBenchmark.hash                                   avgt   50       174,632 ±  
     5,581   ms/op
HashBenchmark.hash:·gc.alloc.rate                    avgt   50       165,848 ±  
     4,574  MB/sec
HashBenchmark.hash:·gc.alloc.rate.norm               avgt   50  31773902,052 ±  
 24870,662    B/op
HashBenchmark.hash:·gc.churn.G1_Eden_Space           avgt   50       166,357 ±  
     9,399  MB/sec
HashBenchmark.hash:·gc.churn.G1_Eden_Space.norm      avgt   50  31891221,589 ± 
1725370,956    B/op
HashBenchmark.hash:·gc.churn.G1_Old_Gen              avgt   50         5,553 ±  
     0,327  MB/sec
HashBenchmark.hash:·gc.churn.G1_Old_Gen.norm         avgt   50   1064354,875 ±  
 60435,285    B/op
HashBenchmark.hash:·gc.churn.G1_Survivor_Space       avgt   50         0,034 ±  
     0,032  MB/sec
HashBenchmark.hash:·gc.churn.G1_Survivor_Space.norm  avgt   50      6754,365 ±  
  6398,008    B/op
HashBenchmark.hash:·gc.count                         avgt   50       144,000    
            counts
HashBenchmark.hash:·gc.time                          avgt   50       249,000    
                ms

patched

Benchmark                                            Mode  Cnt          Score   
      Error   Units
HashBenchmark.hash                                   avgt   50        192,068 ± 
      8,236   ms/op
HashBenchmark.hash:·gc.alloc.rate                    avgt   50        568,790 ± 
     22,912  MB/sec
HashBenchmark.hash:·gc.alloc.rate.norm               avgt   50  119441637,758 ± 
  28476,724    B/op
HashBenchmark.hash:·gc.churn.G1_Eden_Space           avgt   50        568,892 ± 
     26,936  MB/sec
HashBenchmark.hash:·gc.churn.G1_Eden_Space.norm      avgt   50  119444622,067 ± 
2746230,425    B/op
HashBenchmark.hash:·gc.churn.G1_Old_Gen              avgt   50          4,964 ± 
      0,238  MB/sec
HashBenchmark.hash:·gc.churn.G1_Old_Gen.norm         avgt   50    1042019,563 ± 
  23557,187    B/op
HashBenchmark.hash:·gc.churn.G1_Survivor_Space       avgt   50          0,043 ± 
      0,033  MB/sec
HashBenchmark.hash:·gc.churn.G1_Survivor_Space.norm  avgt   50       9130,656 ± 
   7040,858    B/op
HashBenchmark.hash:·gc.count                         avgt   50        492,000   
             counts
HashBenchmark.hash:·gc.time                          avgt   50        738,000   
                 ms

I'll close this PR then.

-------------

PR: https://git.openjdk.org/jdk/pull/9596