src/hotspot/cpu/ppc/stubGenerator_ppc.cpp

rev 53130 : 8216060: [PPC64] Vector CRC implementation should be used by interpreter and be faster for short arrays
Reviewed-by: gromero

*** 3186,3214 ****
      return start;
    }
  
  
    // Compute CRC32/CRC32C function.
!   void generate_CRC_updateBytes(const char* name, Register table, bool invertCRC) {
  
        // arguments to kernel_crc32:
        const Register crc     = R3_ARG1;  // Current checksum, preset by caller or result from previous call.
        const Register data    = R4_ARG2;  // source byte array
        const Register dataLen = R5_ARG3;  // #bytes to process
  
-       const Register t0      = R2;
-       const Register t1      = R7;
-       const Register t2      = R8;
-       const Register t3      = R9;
-       const Register tc0     = R10;
-       const Register tc1     = R11;
-       const Register tc2     = R12;
- 
        BLOCK_COMMENT("Stub body {");
        assert_different_registers(crc, data, dataLen, table);
  
!       __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3, tc0, tc1, tc2, table, invertCRC);
  
        BLOCK_COMMENT("return");
        __ mr_if_needed(R3_RET, crc);      // Updated crc is function result. No copying required (R3_ARG1 == R3_RET).
        __ blr();
  
--- 3186,3210 ----
      return start;
    }
  
  
    // Compute CRC32/CRC32C function.
!   void generate_CRC_updateBytes(const char* name, Register table, Register vector_constants, bool invertCRC) {
  
        // arguments to kernel_crc32:
        const Register crc     = R3_ARG1;  // Current checksum, preset by caller or result from previous call.
        const Register data    = R4_ARG2;  // source byte array
        const Register dataLen = R5_ARG3;  // #bytes to process
  
        BLOCK_COMMENT("Stub body {");
        assert_different_registers(crc, data, dataLen, table);
  
!       if (vector_constants == noreg) {
!         __ kernel_crc32_1word(crc, data, dataLen, table, R2, R7, R8, R9, R10, R11, R12, table, invertCRC);
!       } else {
!         __ kernel_crc32_vpmsum(crc, data, dataLen, table, vector_constants, R7, R8, R9, R10, R11, R12, invertCRC);
!       }
  
        BLOCK_COMMENT("return");
        __ mr_if_needed(R3_RET, crc);      // Updated crc is function result. No copying required (R3_ARG1 == R3_RET).
        __ blr();
  
*** 3495,3540 ****
    address generate_CRC32_updateBytes(const char* name) {
      __ align(CodeEntryAlignment);
      StubCodeMark mark(this, "StubRoutines", name);
      address start = __ function_entry();  // Remember stub start address (is rtn value).
  
-     const Register table   = R6;       // crc table address
- 
      // arguments to kernel_crc32:
      const Register crc     = R3_ARG1;  // Current checksum, preset by caller or result from previous call.
      const Register data    = R4_ARG2;  // source byte array
      const Register dataLen = R5_ARG3;  // #bytes to process
  
!     if (VM_Version::has_vpmsumb()) {
!       const Register constants    = R2;  // constants address
!       const Register bconstants   = R8;  // barret table address
! 
!       const Register t0      = R9;
!       const Register t1      = R10;
!       const Register t2      = R11;
!       const Register t3      = R12;
!       const Register t4      = R7;
! 
!       BLOCK_COMMENT("Stub body {");
!       assert_different_registers(crc, data, dataLen, table);
! 
!       StubRoutines::ppc64::generate_load_crc_table_addr(_masm, table);
!       StubRoutines::ppc64::generate_load_crc_constants_addr(_masm, constants);
!       StubRoutines::ppc64::generate_load_crc_barret_constants_addr(_masm, bconstants);
! 
!       __ kernel_crc32_1word_vpmsum(crc, data, dataLen, table, constants, bconstants, t0, t1, t2, t3, t4, true);
! 
!       BLOCK_COMMENT("return");
!       __ mr_if_needed(R3_RET, crc);      // Updated crc is function result. No copying required (R3_ARG1 == R3_RET).
!       __ blr();
  
-       BLOCK_COMMENT("} Stub body");
-     } else {
        StubRoutines::ppc64::generate_load_crc_table_addr(_masm, table);
!       generate_CRC_updateBytes(name, table, true);
      }
! 
      return start;
    }
  
  
    /**
--- 3491,3514 ----
    address generate_CRC32_updateBytes(const char* name) {
      __ align(CodeEntryAlignment);
      StubCodeMark mark(this, "StubRoutines", name);
      address start = __ function_entry();  // Remember stub start address (is rtn value).
  
      // arguments to kernel_crc32:
      const Register crc     = R3_ARG1;  // Current checksum, preset by caller or result from previous call.
      const Register data    = R4_ARG2;  // source byte array
      const Register dataLen = R5_ARG3;  // #bytes to process
  
!     const Register table   = R6;       // crc table address
!     Register vector_constants = noreg;
  
      StubRoutines::ppc64::generate_load_crc_table_addr(_masm, table);
!     if (VM_Version::has_vpmsumb()) {
!       vector_constants = R2;
!       StubRoutines::ppc64::generate_load_crc_constants_addr(_masm, vector_constants);
      }
!     generate_CRC_updateBytes(name, table, vector_constants, true);
      return start;
    }
  
  
    /**
*** 3555,3600 ****
    address generate_CRC32C_updateBytes(const char* name) {
      __ align(CodeEntryAlignment);
      StubCodeMark mark(this, "StubRoutines", name);
      address start = __ function_entry();  // Remember stub start address (is rtn value).
  
-     const Register table   = R6;       // crc table address
- 
      // arguments to kernel_crc32:
      const Register crc     = R3_ARG1;  // Current checksum, preset by caller or result from previous call.
      const Register data    = R4_ARG2;  // source byte array
      const Register dataLen = R5_ARG3;  // #bytes to process
  
!     if (VM_Version::has_vpmsumb()) {
!       const Register constants    = R2;  // constants address
!       const Register bconstants   = R8;  // barret table address
! 
!       const Register t0      = R9;
!       const Register t1      = R10;
!       const Register t2      = R11;
!       const Register t3      = R12;
!       const Register t4      = R7;
! 
!       BLOCK_COMMENT("Stub body {");
!       assert_different_registers(crc, data, dataLen, table);
! 
!       StubRoutines::ppc64::generate_load_crc32c_table_addr(_masm, table);
!       StubRoutines::ppc64::generate_load_crc32c_constants_addr(_masm, constants);
!       StubRoutines::ppc64::generate_load_crc32c_barret_constants_addr(_masm, bconstants);
! 
!       __ kernel_crc32_1word_vpmsum(crc, data, dataLen, table, constants, bconstants, t0, t1, t2, t3, t4, false);
! 
!       BLOCK_COMMENT("return");
!       __ mr_if_needed(R3_RET, crc);      // Updated crc is function result. No copying required (R3_ARG1 == R3_RET).
!       __ blr();
  
-       BLOCK_COMMENT("} Stub body");
-     } else {
        StubRoutines::ppc64::generate_load_crc32c_table_addr(_masm, table);
!       generate_CRC_updateBytes(name, table, false);
      }
! 
      return start;
    }
  
  
    // Initialization
--- 3529,3552 ----
    address generate_CRC32C_updateBytes(const char* name) {
      __ align(CodeEntryAlignment);
      StubCodeMark mark(this, "StubRoutines", name);
      address start = __ function_entry();  // Remember stub start address (is rtn value).
  
      // arguments to kernel_crc32:
      const Register crc     = R3_ARG1;  // Current checksum, preset by caller or result from previous call.
      const Register data    = R4_ARG2;  // source byte array
      const Register dataLen = R5_ARG3;  // #bytes to process
  
!     const Register table   = R6;       // crc table address
!     Register vector_constants = noreg;
  
      StubRoutines::ppc64::generate_load_crc32c_table_addr(_masm, table);
!     if (VM_Version::has_vpmsumb()) {
!       vector_constants = R2;
!       StubRoutines::ppc64::generate_load_crc32c_constants_addr(_masm, vector_constants);
      }
!     generate_CRC_updateBytes(name, table, vector_constants, false);
      return start;
    }
  
  
    // Initialization