RFR (L): 7088419 : Use x86 Hardware CRC32 Instruction with java.util.zip.CRC32 and java.util.zip.Adler32

[Vladimir Kozlov]

David,


I think you forgot to revert changes in jvm.cpp. Changes are good in 
general.

About test:

'@summary' should bug synopsis: "Use x86 Hardware CRC32 Instruction with 
java.util.zip.CRC32 and java.util.zip.Adler32".

What test does is a separate comment.
And you forgot '@run' command:

* @run main CRCTest

Thanks,
Vladimir

On 6/26/13 10:52 AM, David Chase wrote:
> Bug: There is a lovely instruction available on late-model Intel chips
> that can be used to accelerate some CRC calculations.
>
> Fix: Enhance the compiler to have intrinsics for the CRC methods
> that use this instruction, where appropriate.
>
> Testing:
> Hand testing on Sparc (to ensure no harm was done)
> and x86 (to ensure no harm was done, and that it was faster)
> plus with JPRT across a range of x86 (32/64, Windows/MacOS/Linux/Solaris)
> and other targets to be sure all was well.
>
> This is also all based on C code that had already been well-tested.
>
> Webrev: http://cr.openjdk.java.net/~drchase/7088419/webrev.05/
> No changes to JDK: this is all in the compiler.
> It does not interfere with the possibility of parallelizing CRC32 and Adler32,
> but that seemed like a tall order for a compiler intrinsic.
>
> Guide to the webrev changes:
>
>      - Add new instructions to assembler
>          - src/cpu/x86/vm/assembler_x86.hpp (declare entrypoints:
>            movdqa, pinsrd, pinsrq, pextrd, pextrq, vpclmulqdq)
>          - src/cpu/x86/vm/assembler_x86.cpp (emit the bits)
>      - Add new operations to macro-assembler
>          - src/cpu/x86/vm/macroAssembler_x86.cpp (this includes the
>            truly "macro" instructions update_byte_crc32,
>            fold_128bit_crc32, fold_8bit_crc32, kernel_crc32)
>          - src/cpu/x86/vm/macroAssembler_x86.hpp
>      - Add flag and feature checks
>          - src/cpu/x86/vm/globals_x86.hpp (arch-specific UseCLMUL)
>          - src/share/vm/runtime/globals.hpp (UseCRC32Intrinsics  flag
>            and default value)
>          - src/cpu/x86/vm/vm_version_x86.hpp (pull feature from cpu info)
>          - src/cpu/x86/vm/vm_version_x86.cpp (process flag, print
>            feature)
>          - src/share/vm/prims/jvm.cpp (set property for jdk lib -- is
>            this needed for crc?)
>      - Register intrinsic
>          - src/share/vm/classfile/vmSymbols.hpp (declare it with
>            signature etc)
>      - Write stub data (CRC constants and tables)
>          - src/share/vm/runtime/stubRoutines.{c,h}pp (declaration of
>            statics and accessors for crc32 generated method and
>            constants).
>          - src/cpu/x86/vm/stubRoutines_x86.{c,h}pp (what are
>            _verify_mxcsr_entry and _key_shuffle_mask_addr ? -- answer,
>            these are refactored from 32/64-specific files, see below.)
>          - src/cpu/x86/vm/stubRoutines_x86_{32,64}.{c,h}pp (code
>            refactored into common stubRoutines.{c,h}pp)
>          - src/cpu/x86/vm/stubGenerator_x86_{32,64}.cpp (write out
>            generated stub procedure, initialize pointer to method and
>            data)
>      - Add stub call substitution to interpreter
>          - src/share/vm/interpreter/templateInterpreter.cpp (declare
>            method entrypoints)
>          - src/cpu/x86/vm/interpreterGenerator_x86.hpp,
>            src/cpu/x86/vm/templateInterpreter_x86_{32,64}.cpp
>            (declaration and definition of 32/64-specific glue to call
>            the intrinsic stub routine)
>          - src/share/vm/interpreter/interpreter.cpp (enum lookup and
>            debugging output)
>          - src/share/vm/interpreter/abstractInterpreter.hpp (extend
>            enumeration of method kinds)
>      - Add to C1
>          - src/share/vm/c1/c1_LIR.{c,h}pp  (declare new nodes
>            LIR_OpUpdateCRC32)
>          - src/cpu/sparc/vm/c1_LIRGenerator_sparc.cpp (unimplemented
>            stubs for unsupported arch)
>          - src/cpu/sparc/vm/c1_LIRAssembler_sparc.cpp (unimplemented
>            stubs for unsupported arch)
>          - src/share/vm/c1/c1_GraphBuilder.cpp (try inline intrinsic)
>          - src/share/vm/c1/c1_Runtime1.cpp (stub routine looker-upper)
>          - src/share/vm/c1/c1_LIRGenerator.{c,h}pp (Add case to
>            do_Intrinsic for one-byte update)
>          - src/cpu/x86/vm/c1_LIRGenerator_x86.cpp (Add cases for x86 CRC
>            updates -- byte, array, buffer)
>          - src/cpu/x86/vm/c1_LIRAssembler_x86.cpp (Add emit_updatecrc32
>            for the the one-byte case from new node).
>          - src/share/vm/c1/c1_LIRAssembler.hpp (Add decl for
>            emit_updatecrc32)
>      - Add to C2
>          - src/share/vm/opto/runtime.{c,h}pp (new intrinsic function has
>            a type; add calls to generate that type).
>          - src/share/vm/opto/escape.cpp (add special case for escape
>            analysis assertion)
>          - src/share/vm/opto/library_call.cpp (substitution of inline
>            glue code to intrinsic stub call).
>