1.1 --- a/src/cpu/sparc/vm/stubGenerator_sparc.cpp Tue Feb 04 17:38:01 2020 +0800
1.2 +++ b/src/cpu/sparc/vm/stubGenerator_sparc.cpp Tue Feb 04 18:13:14 2020 +0800
1.3 @@ -4788,6 +4788,130 @@
1.4 return start;
1.5 }
1.6
1.7 + /* Single and multi-block ghash operations */
1.8 + address generate_ghash_processBlocks() {
1.9 + __ align(CodeEntryAlignment);
1.10 + Label L_ghash_loop, L_aligned, L_main;
1.11 + StubCodeMark mark(this, "StubRoutines", "ghash_processBlocks");
1.12 + address start = __ pc();
1.13 +
1.14 + Register state = I0;
1.15 + Register subkeyH = I1;
1.16 + Register data = I2;
1.17 + Register len = I3;
1.18 +
1.19 + __ save_frame(0);
1.20 +
1.21 + __ ldx(state, 0, O0);
1.22 + __ ldx(state, 8, O1);
1.23 +
1.24 + // Loop label for multiblock operations
1.25 + __ BIND(L_ghash_loop);
1.26 +
1.27 + // Check if 'data' is unaligned
1.28 + __ andcc(data, 7, G1);
1.29 + __ br(Assembler::zero, false, Assembler::pt, L_aligned);
1.30 + __ delayed()->nop();
1.31 +
1.32 + Register left_shift = L1;
1.33 + Register right_shift = L2;
1.34 + Register data_ptr = L3;
1.35 +
1.36 + // Get left and right shift values in bits
1.37 + __ sll(G1, LogBitsPerByte, left_shift);
1.38 + __ mov(64, right_shift);
1.39 + __ sub(right_shift, left_shift, right_shift);
1.40 +
1.41 + // Align to read 'data'
1.42 + __ sub(data, G1, data_ptr);
1.43 +
1.44 + // Load first 8 bytes of 'data'
1.45 + __ ldx(data_ptr, 0, O4);
1.46 + __ sllx(O4, left_shift, O4);
1.47 + __ ldx(data_ptr, 8, O5);
1.48 + __ srlx(O5, right_shift, G4);
1.49 + __ bset(G4, O4);
1.50 +
1.51 + // Load second 8 bytes of 'data'
1.52 + __ sllx(O5, left_shift, O5);
1.53 + __ ldx(data_ptr, 16, G4);
1.54 + __ srlx(G4, right_shift, G4);
1.55 + __ ba(L_main);
1.56 + __ delayed()->bset(G4, O5);
1.57 +
1.58 + // If 'data' is aligned, load normally
1.59 + __ BIND(L_aligned);
1.60 + __ ldx(data, 0, O4);
1.61 + __ ldx(data, 8, O5);
1.62 +
1.63 + __ BIND(L_main);
1.64 + __ ldx(subkeyH, 0, O2);
1.65 + __ ldx(subkeyH, 8, O3);
1.66 +
1.67 + __ xor3(O0, O4, O0);
1.68 + __ xor3(O1, O5, O1);
1.69 +
1.70 + __ xmulxhi(O0, O3, G3);
1.71 + __ xmulx(O0, O2, O5);
1.72 + __ xmulxhi(O1, O2, G4);
1.73 + __ xmulxhi(O1, O3, G5);
1.74 + __ xmulx(O0, O3, G1);
1.75 + __ xmulx(O1, O3, G2);
1.76 + __ xmulx(O1, O2, O3);
1.77 + __ xmulxhi(O0, O2, O4);
1.78 +
1.79 + __ mov(0xE1, O0);
1.80 + __ sllx(O0, 56, O0);
1.81 +
1.82 + __ xor3(O5, G3, O5);
1.83 + __ xor3(O5, G4, O5);
1.84 + __ xor3(G5, G1, G1);
1.85 + __ xor3(G1, O3, G1);
1.86 + __ srlx(G2, 63, O1);
1.87 + __ srlx(G1, 63, G3);
1.88 + __ sllx(G2, 63, O3);
1.89 + __ sllx(G2, 58, O2);
1.90 + __ xor3(O3, O2, O2);
1.91 +
1.92 + __ sllx(G1, 1, G1);
1.93 + __ or3(G1, O1, G1);
1.94 +
1.95 + __ xor3(G1, O2, G1);
1.96 +
1.97 + __ sllx(G2, 1, G2);
1.98 +
1.99 + __ xmulxhi(G1, O0, O1);
1.100 + __ xmulx(G1, O0, O2);
1.101 + __ xmulxhi(G2, O0, O3);
1.102 + __ xmulx(G2, O0, G1);
1.103 +
1.104 + __ xor3(O4, O1, O4);
1.105 + __ xor3(O5, O2, O5);
1.106 + __ xor3(O5, O3, O5);
1.107 +
1.108 + __ sllx(O4, 1, O2);
1.109 + __ srlx(O5, 63, O3);
1.110 +
1.111 + __ or3(O2, O3, O0);
1.112 +
1.113 + __ sllx(O5, 1, O1);
1.114 + __ srlx(G1, 63, O2);
1.115 + __ or3(O1, O2, O1);
1.116 + __ xor3(O1, G3, O1);
1.117 +
1.118 + __ deccc(len);
1.119 + __ br(Assembler::notZero, true, Assembler::pt, L_ghash_loop);
1.120 + __ delayed()->add(data, 16, data);
1.121 +
1.122 + __ stx(O0, I0, 0);
1.123 + __ stx(O1, I0, 8);
1.124 +
1.125 + __ ret();
1.126 + __ delayed()->restore();
1.127 +
1.128 + return start;
1.129 + }
1.130 +
1.131 void generate_initial() {
1.132 // Generates all stubs and initializes the entry points
1.133
1.134 @@ -4860,6 +4984,10 @@
1.135 StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
1.136 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt_Parallel();
1.137 }
1.138 + // generate GHASH intrinsics code
1.139 + if (UseGHASHIntrinsics) {
1.140 + StubRoutines::_ghash_processBlocks = generate_ghash_processBlocks();
1.141 + }
1.142
1.143 // generate SHA1/SHA256/SHA512 intrinsics code
1.144 if (UseSHA1Intrinsics) {
