1.1 --- a/src/cpu/x86/vm/x86_32.ad Tue Mar 31 10:02:01 2009 -0700
1.2 +++ b/src/cpu/x86/vm/x86_32.ad Tue Mar 31 14:07:08 2009 -0700
1.3 @@ -3694,12 +3694,16 @@
1.4 }
1.5 %}
1.6
1.7 - enc_class enc_String_Compare() %{
1.8 + enc_class enc_String_Compare(eDIRegP str1, eSIRegP str2, regXD tmp1, regXD tmp2,
1.9 + eAXRegI tmp3, eBXRegI tmp4, eCXRegI result) %{
1.10 Label ECX_GOOD_LABEL, LENGTH_DIFF_LABEL,
1.11 POP_LABEL, DONE_LABEL, CONT_LABEL,
1.12 WHILE_HEAD_LABEL;
1.13 MacroAssembler masm(&cbuf);
1.14
1.15 + XMMRegister tmp1Reg = as_XMMRegister($tmp1$$reg);
1.16 + XMMRegister tmp2Reg = as_XMMRegister($tmp2$$reg);
1.17 +
1.18 // Get the first character position in both strings
1.19 // [8] char array, [12] offset, [16] count
1.20 int value_offset = java_lang_String::value_offset_in_bytes();
1.21 @@ -3717,7 +3721,6 @@
1.22 // Compute the minimum of the string lengths(rsi) and the
1.23 // difference of the string lengths (stack)
1.24
1.25 -
1.26 if (VM_Version::supports_cmov()) {
1.27 masm.movl(rdi, Address(rdi, count_offset));
1.28 masm.movl(rsi, Address(rsi, count_offset));
1.29 @@ -3731,7 +3734,7 @@
1.30 masm.movl(rsi, rdi);
1.31 masm.subl(rdi, rcx);
1.32 masm.push(rdi);
1.33 - masm.jcc(Assembler::lessEqual, ECX_GOOD_LABEL);
1.34 + masm.jccb(Assembler::lessEqual, ECX_GOOD_LABEL);
1.35 masm.movl(rsi, rcx);
1.36 // rsi holds min, rcx is unused
1.37 }
1.38 @@ -3756,7 +3759,7 @@
1.39 Label LSkip2;
1.40 // Check if the strings start at same location
1.41 masm.cmpptr(rbx,rax);
1.42 - masm.jcc(Assembler::notEqual, LSkip2);
1.43 + masm.jccb(Assembler::notEqual, LSkip2);
1.44
1.45 // Check if the length difference is zero (from stack)
1.46 masm.cmpl(Address(rsp, 0), 0x0);
1.47 @@ -3766,9 +3769,52 @@
1.48 masm.bind(LSkip2);
1.49 }
1.50
1.51 - // Shift rax, and rbx, to the end of the arrays, negate min
1.52 - masm.lea(rax, Address(rax, rsi, Address::times_2, 2));
1.53 - masm.lea(rbx, Address(rbx, rsi, Address::times_2, 2));
1.54 + // Advance to next character
1.55 + masm.addptr(rax, 2);
1.56 + masm.addptr(rbx, 2);
1.57 +
1.58 + if (UseSSE42Intrinsics) {
1.59 + // With SSE4.2, use double quad vector compare
1.60 + Label COMPARE_VECTORS, VECTOR_NOT_EQUAL, COMPARE_TAIL;
1.61 + // Setup to compare 16-byte vectors
1.62 + masm.movl(rdi, rsi);
1.63 + masm.andl(rsi, 0xfffffff8); // rsi holds the vector count
1.64 + masm.andl(rdi, 0x00000007); // rdi holds the tail count
1.65 + masm.testl(rsi, rsi);
1.66 + masm.jccb(Assembler::zero, COMPARE_TAIL);
1.67 +
1.68 + masm.lea(rax, Address(rax, rsi, Address::times_2));
1.69 + masm.lea(rbx, Address(rbx, rsi, Address::times_2));
1.70 + masm.negl(rsi);
1.71 +
1.72 + masm.bind(COMPARE_VECTORS);
1.73 + masm.movdqu(tmp1Reg, Address(rax, rsi, Address::times_2));
1.74 + masm.movdqu(tmp2Reg, Address(rbx, rsi, Address::times_2));
1.75 + masm.pxor(tmp1Reg, tmp2Reg);
1.76 + masm.ptest(tmp1Reg, tmp1Reg);
1.77 + masm.jccb(Assembler::notZero, VECTOR_NOT_EQUAL);
1.78 + masm.addl(rsi, 8);
1.79 + masm.jcc(Assembler::notZero, COMPARE_VECTORS);
1.80 + masm.jmpb(COMPARE_TAIL);
1.81 +
1.82 + // Mismatched characters in the vectors
1.83 + masm.bind(VECTOR_NOT_EQUAL);
1.84 + masm.lea(rax, Address(rax, rsi, Address::times_2));
1.85 + masm.lea(rbx, Address(rbx, rsi, Address::times_2));
1.86 + masm.movl(rdi, 8);
1.87 +
1.88 + // Compare tail (< 8 chars), or rescan last vectors to
1.89 + // find 1st mismatched characters
1.90 + masm.bind(COMPARE_TAIL);
1.91 + masm.testl(rdi, rdi);
1.92 + masm.jccb(Assembler::zero, LENGTH_DIFF_LABEL);
1.93 + masm.movl(rsi, rdi);
1.94 + // Fallthru to tail compare
1.95 + }
1.96 +
1.97 + //Shift rax, and rbx, to the end of the arrays, negate min
1.98 + masm.lea(rax, Address(rax, rsi, Address::times_2, 0));
1.99 + masm.lea(rbx, Address(rbx, rsi, Address::times_2, 0));
1.100 masm.negl(rsi);
1.101
1.102 // Compare the rest of the characters
1.103 @@ -3776,93 +3822,329 @@
1.104 masm.load_unsigned_short(rcx, Address(rbx, rsi, Address::times_2, 0));
1.105 masm.load_unsigned_short(rdi, Address(rax, rsi, Address::times_2, 0));
1.106 masm.subl(rcx, rdi);
1.107 - masm.jcc(Assembler::notZero, POP_LABEL);
1.108 + masm.jccb(Assembler::notZero, POP_LABEL);
1.109 masm.incrementl(rsi);
1.110 masm.jcc(Assembler::notZero, WHILE_HEAD_LABEL);
1.111
1.112 // Strings are equal up to min length. Return the length difference.
1.113 masm.bind(LENGTH_DIFF_LABEL);
1.114 masm.pop(rcx);
1.115 - masm.jmp(DONE_LABEL);
1.116 + masm.jmpb(DONE_LABEL);
1.117
1.118 // Discard the stored length difference
1.119 masm.bind(POP_LABEL);
1.120 masm.addptr(rsp, 4);
1.121 -
1.122 +
1.123 // That's it
1.124 masm.bind(DONE_LABEL);
1.125 %}
1.126
1.127 - enc_class enc_Array_Equals(eDIRegP ary1, eSIRegP ary2, eAXRegI tmp1, eBXRegI tmp2, eCXRegI result) %{
1.128 - Label TRUE_LABEL, FALSE_LABEL, DONE_LABEL, COMPARE_LOOP_HDR, COMPARE_LOOP;
1.129 + enc_class enc_String_Equals(eDIRegP str1, eSIRegP str2, regXD tmp1, regXD tmp2,
1.130 + eBXRegI tmp3, eCXRegI tmp4, eAXRegI result) %{
1.131 + Label RET_TRUE, RET_FALSE, DONE, COMPARE_VECTORS, COMPARE_CHAR;
1.132 MacroAssembler masm(&cbuf);
1.133
1.134 - Register ary1Reg = as_Register($ary1$$reg);
1.135 - Register ary2Reg = as_Register($ary2$$reg);
1.136 - Register tmp1Reg = as_Register($tmp1$$reg);
1.137 - Register tmp2Reg = as_Register($tmp2$$reg);
1.138 - Register resultReg = as_Register($result$$reg);
1.139 + XMMRegister tmp1Reg = as_XMMRegister($tmp1$$reg);
1.140 + XMMRegister tmp2Reg = as_XMMRegister($tmp2$$reg);
1.141 +
1.142 + int value_offset = java_lang_String::value_offset_in_bytes();
1.143 + int offset_offset = java_lang_String::offset_offset_in_bytes();
1.144 + int count_offset = java_lang_String::count_offset_in_bytes();
1.145 + int base_offset = arrayOopDesc::base_offset_in_bytes(T_CHAR);
1.146 +
1.147 + // does source == target string?
1.148 + masm.cmpptr(rdi, rsi);
1.149 + masm.jcc(Assembler::equal, RET_TRUE);
1.150 +
1.151 + // get and compare counts
1.152 + masm.movl(rcx, Address(rdi, count_offset));
1.153 + masm.movl(rax, Address(rsi, count_offset));
1.154 + masm.cmpl(rcx, rax);
1.155 + masm.jcc(Assembler::notEqual, RET_FALSE);
1.156 + masm.testl(rax, rax);
1.157 + masm.jcc(Assembler::zero, RET_TRUE);
1.158 +
1.159 + // get source string offset and value
1.160 + masm.movptr(rbx, Address(rsi, value_offset));
1.161 + masm.movl(rax, Address(rsi, offset_offset));
1.162 + masm.leal(rsi, Address(rbx, rax, Address::times_2, base_offset));
1.163 +
1.164 + // get compare string offset and value
1.165 + masm.movptr(rbx, Address(rdi, value_offset));
1.166 + masm.movl(rax, Address(rdi, offset_offset));
1.167 + masm.leal(rdi, Address(rbx, rax, Address::times_2, base_offset));
1.168 +
1.169 + // Set byte count
1.170 + masm.shll(rcx, 1);
1.171 + masm.movl(rax, rcx);
1.172 +
1.173 + if (UseSSE42Intrinsics) {
1.174 + // With SSE4.2, use double quad vector compare
1.175 + Label COMPARE_WIDE_VECTORS, COMPARE_TAIL;
1.176 + // Compare 16-byte vectors
1.177 + masm.andl(rcx, 0xfffffff0); // vector count (in bytes)
1.178 + masm.andl(rax, 0x0000000e); // tail count (in bytes)
1.179 + masm.testl(rcx, rcx);
1.180 + masm.jccb(Assembler::zero, COMPARE_TAIL);
1.181 + masm.lea(rdi, Address(rdi, rcx, Address::times_1));
1.182 + masm.lea(rsi, Address(rsi, rcx, Address::times_1));
1.183 + masm.negl(rcx);
1.184 +
1.185 + masm.bind(COMPARE_WIDE_VECTORS);
1.186 + masm.movdqu(tmp1Reg, Address(rdi, rcx, Address::times_1));
1.187 + masm.movdqu(tmp2Reg, Address(rsi, rcx, Address::times_1));
1.188 + masm.pxor(tmp1Reg, tmp2Reg);
1.189 + masm.ptest(tmp1Reg, tmp1Reg);
1.190 + masm.jccb(Assembler::notZero, RET_FALSE);
1.191 + masm.addl(rcx, 16);
1.192 + masm.jcc(Assembler::notZero, COMPARE_WIDE_VECTORS);
1.193 + masm.bind(COMPARE_TAIL);
1.194 + masm.movl(rcx, rax);
1.195 + // Fallthru to tail compare
1.196 + }
1.197 +
1.198 + // Compare 4-byte vectors
1.199 + masm.andl(rcx, 0xfffffffc); // vector count (in bytes)
1.200 + masm.andl(rax, 0x00000002); // tail char (in bytes)
1.201 + masm.testl(rcx, rcx);
1.202 + masm.jccb(Assembler::zero, COMPARE_CHAR);
1.203 + masm.lea(rdi, Address(rdi, rcx, Address::times_1));
1.204 + masm.lea(rsi, Address(rsi, rcx, Address::times_1));
1.205 + masm.negl(rcx);
1.206 +
1.207 + masm.bind(COMPARE_VECTORS);
1.208 + masm.movl(rbx, Address(rdi, rcx, Address::times_1));
1.209 + masm.cmpl(rbx, Address(rsi, rcx, Address::times_1));
1.210 + masm.jccb(Assembler::notEqual, RET_FALSE);
1.211 + masm.addl(rcx, 4);
1.212 + masm.jcc(Assembler::notZero, COMPARE_VECTORS);
1.213 +
1.214 + // Compare trailing char (final 2 bytes), if any
1.215 + masm.bind(COMPARE_CHAR);
1.216 + masm.testl(rax, rax);
1.217 + masm.jccb(Assembler::zero, RET_TRUE);
1.218 + masm.load_unsigned_short(rbx, Address(rdi, 0));
1.219 + masm.load_unsigned_short(rcx, Address(rsi, 0));
1.220 + masm.cmpl(rbx, rcx);
1.221 + masm.jccb(Assembler::notEqual, RET_FALSE);
1.222 +
1.223 + masm.bind(RET_TRUE);
1.224 + masm.movl(rax, 1); // return true
1.225 + masm.jmpb(DONE);
1.226 +
1.227 + masm.bind(RET_FALSE);
1.228 + masm.xorl(rax, rax); // return false
1.229 +
1.230 + masm.bind(DONE);
1.231 + %}
1.232 +
1.233 + enc_class enc_String_IndexOf(eSIRegP str1, eDIRegP str2, regXD tmp1, eAXRegI tmp2,
1.234 + eCXRegI tmp3, eDXRegI tmp4, eBXRegI result) %{
1.235 + // SSE4.2 version
1.236 + Label LOAD_SUBSTR, PREP_FOR_SCAN, SCAN_TO_SUBSTR,
1.237 + SCAN_SUBSTR, RET_NEG_ONE, RET_NOT_FOUND, CLEANUP, DONE;
1.238 + MacroAssembler masm(&cbuf);
1.239 +
1.240 + XMMRegister tmp1Reg = as_XMMRegister($tmp1$$reg);
1.241 +
1.242 + // Get the first character position in both strings
1.243 + // [8] char array, [12] offset, [16] count
1.244 + int value_offset = java_lang_String::value_offset_in_bytes();
1.245 + int offset_offset = java_lang_String::offset_offset_in_bytes();
1.246 + int count_offset = java_lang_String::count_offset_in_bytes();
1.247 + int base_offset = arrayOopDesc::base_offset_in_bytes(T_CHAR);
1.248 +
1.249 + // Get counts for string and substr
1.250 + masm.movl(rdx, Address(rsi, count_offset));
1.251 + masm.movl(rax, Address(rdi, count_offset));
1.252 + // Check for substr count > string count
1.253 + masm.cmpl(rax, rdx);
1.254 + masm.jcc(Assembler::greater, RET_NEG_ONE);
1.255 +
1.256 + // Start the indexOf operation
1.257 + // Get start addr of string
1.258 + masm.movptr(rbx, Address(rsi, value_offset));
1.259 + masm.movl(rcx, Address(rsi, offset_offset));
1.260 + masm.lea(rsi, Address(rbx, rcx, Address::times_2, base_offset));
1.261 + masm.push(rsi);
1.262 +
1.263 + // Get start addr of substr
1.264 + masm.movptr(rbx, Address(rdi, value_offset));
1.265 + masm.movl(rcx, Address(rdi, offset_offset));
1.266 + masm.lea(rdi, Address(rbx, rcx, Address::times_2, base_offset));
1.267 + masm.push(rdi);
1.268 + masm.push(rax);
1.269 + masm.jmpb(PREP_FOR_SCAN);
1.270 +
1.271 + // Substr count saved at sp
1.272 + // Substr saved at sp+4
1.273 + // String saved at sp+8
1.274 +
1.275 + // Prep to load substr for scan
1.276 + masm.bind(LOAD_SUBSTR);
1.277 + masm.movptr(rdi, Address(rsp, 4));
1.278 + masm.movl(rax, Address(rsp, 0));
1.279 +
1.280 + // Load substr
1.281 + masm.bind(PREP_FOR_SCAN);
1.282 + masm.movdqu(tmp1Reg, Address(rdi, 0));
1.283 + masm.addl(rdx, 8); // prime the loop
1.284 + masm.subptr(rsi, 16);
1.285 +
1.286 + // Scan string for substr in 16-byte vectors
1.287 + masm.bind(SCAN_TO_SUBSTR);
1.288 + masm.subl(rdx, 8);
1.289 + masm.addptr(rsi, 16);
1.290 + masm.pcmpestri(tmp1Reg, Address(rsi, 0), 0x0d);
1.291 + masm.jcc(Assembler::above, SCAN_TO_SUBSTR); // CF == 0 && ZF == 0
1.292 + masm.jccb(Assembler::aboveEqual, RET_NOT_FOUND); // CF == 0
1.293 +
1.294 + // Fallthru: found a potential substr
1.295 +
1.296 + // Make sure string is still long enough
1.297 + masm.subl(rdx, rcx);
1.298 + masm.cmpl(rdx, rax);
1.299 + masm.jccb(Assembler::negative, RET_NOT_FOUND);
1.300 + // Compute start addr of substr
1.301 + masm.lea(rsi, Address(rsi, rcx, Address::times_2));
1.302 + masm.movptr(rbx, rsi);
1.303 +
1.304 + // Compare potential substr
1.305 + masm.addl(rdx, 8); // prime the loop
1.306 + masm.addl(rax, 8);
1.307 + masm.subptr(rsi, 16);
1.308 + masm.subptr(rdi, 16);
1.309 +
1.310 + // Scan 16-byte vectors of string and substr
1.311 + masm.bind(SCAN_SUBSTR);
1.312 + masm.subl(rax, 8);
1.313 + masm.subl(rdx, 8);
1.314 + masm.addptr(rsi, 16);
1.315 + masm.addptr(rdi, 16);
1.316 + masm.movdqu(tmp1Reg, Address(rdi, 0));
1.317 + masm.pcmpestri(tmp1Reg, Address(rsi, 0), 0x0d);
1.318 + masm.jcc(Assembler::noOverflow, LOAD_SUBSTR); // OF == 0
1.319 + masm.jcc(Assembler::positive, SCAN_SUBSTR); // SF == 0
1.320 +
1.321 + // Compute substr offset
1.322 + masm.movptr(rsi, Address(rsp, 8));
1.323 + masm.subptr(rbx, rsi);
1.324 + masm.shrl(rbx, 1);
1.325 + masm.jmpb(CLEANUP);
1.326 +
1.327 + masm.bind(RET_NEG_ONE);
1.328 + masm.movl(rbx, -1);
1.329 + masm.jmpb(DONE);
1.330 +
1.331 + masm.bind(RET_NOT_FOUND);
1.332 + masm.movl(rbx, -1);
1.333 +
1.334 + masm.bind(CLEANUP);
1.335 + masm.addptr(rsp, 12);
1.336 +
1.337 + masm.bind(DONE);
1.338 + %}
1.339 +
1.340 + enc_class enc_Array_Equals(eDIRegP ary1, eSIRegP ary2, regXD tmp1, regXD tmp2,
1.341 + eBXRegI tmp3, eDXRegI tmp4, eAXRegI result) %{
1.342 + Label TRUE_LABEL, FALSE_LABEL, DONE, COMPARE_VECTORS, COMPARE_CHAR;
1.343 + MacroAssembler masm(&cbuf);
1.344 +
1.345 + XMMRegister tmp1Reg = as_XMMRegister($tmp1$$reg);
1.346 + XMMRegister tmp2Reg = as_XMMRegister($tmp2$$reg);
1.347 + Register ary1Reg = as_Register($ary1$$reg);
1.348 + Register ary2Reg = as_Register($ary2$$reg);
1.349 + Register tmp3Reg = as_Register($tmp3$$reg);
1.350 + Register tmp4Reg = as_Register($tmp4$$reg);
1.351 + Register resultReg = as_Register($result$$reg);
1.352
1.353 int length_offset = arrayOopDesc::length_offset_in_bytes();
1.354 int base_offset = arrayOopDesc::base_offset_in_bytes(T_CHAR);
1.355
1.356 // Check the input args
1.357 - masm.cmpl(ary1Reg, ary2Reg);
1.358 + masm.cmpptr(ary1Reg, ary2Reg);
1.359 masm.jcc(Assembler::equal, TRUE_LABEL);
1.360 - masm.testl(ary1Reg, ary1Reg);
1.361 + masm.testptr(ary1Reg, ary1Reg);
1.362 masm.jcc(Assembler::zero, FALSE_LABEL);
1.363 - masm.testl(ary2Reg, ary2Reg);
1.364 + masm.testptr(ary2Reg, ary2Reg);
1.365 masm.jcc(Assembler::zero, FALSE_LABEL);
1.366
1.367 // Check the lengths
1.368 - masm.movl(tmp2Reg, Address(ary1Reg, length_offset));
1.369 + masm.movl(tmp4Reg, Address(ary1Reg, length_offset));
1.370 masm.movl(resultReg, Address(ary2Reg, length_offset));
1.371 - masm.cmpl(tmp2Reg, resultReg);
1.372 + masm.cmpl(tmp4Reg, resultReg);
1.373 masm.jcc(Assembler::notEqual, FALSE_LABEL);
1.374 masm.testl(resultReg, resultReg);
1.375 masm.jcc(Assembler::zero, TRUE_LABEL);
1.376
1.377 - // Get the number of 4 byte vectors to compare
1.378 - masm.shrl(resultReg, 1);
1.379 -
1.380 - // Check for odd-length arrays
1.381 - masm.andl(tmp2Reg, 1);
1.382 - masm.testl(tmp2Reg, tmp2Reg);
1.383 - masm.jcc(Assembler::zero, COMPARE_LOOP_HDR);
1.384 -
1.385 - // Compare 2-byte "tail" at end of arrays
1.386 - masm.load_unsigned_short(tmp1Reg, Address(ary1Reg, resultReg, Address::times_4, base_offset));
1.387 - masm.load_unsigned_short(tmp2Reg, Address(ary2Reg, resultReg, Address::times_4, base_offset));
1.388 - masm.cmpl(tmp1Reg, tmp2Reg);
1.389 - masm.jcc(Assembler::notEqual, FALSE_LABEL);
1.390 + // Load array addrs
1.391 + masm.lea(ary1Reg, Address(ary1Reg, base_offset));
1.392 + masm.lea(ary2Reg, Address(ary2Reg, base_offset));
1.393 +
1.394 + // Set byte count
1.395 + masm.shll(tmp4Reg, 1);
1.396 + masm.movl(resultReg, tmp4Reg);
1.397 +
1.398 + if (UseSSE42Intrinsics) {
1.399 + // With SSE4.2, use double quad vector compare
1.400 + Label COMPARE_WIDE_VECTORS, COMPARE_TAIL;
1.401 + // Compare 16-byte vectors
1.402 + masm.andl(tmp4Reg, 0xfffffff0); // vector count (in bytes)
1.403 + masm.andl(resultReg, 0x0000000e); // tail count (in bytes)
1.404 + masm.testl(tmp4Reg, tmp4Reg);
1.405 + masm.jccb(Assembler::zero, COMPARE_TAIL);
1.406 + masm.lea(ary1Reg, Address(ary1Reg, tmp4Reg, Address::times_1));
1.407 + masm.lea(ary2Reg, Address(ary2Reg, tmp4Reg, Address::times_1));
1.408 + masm.negl(tmp4Reg);
1.409 +
1.410 + masm.bind(COMPARE_WIDE_VECTORS);
1.411 + masm.movdqu(tmp1Reg, Address(ary1Reg, tmp4Reg, Address::times_1));
1.412 + masm.movdqu(tmp2Reg, Address(ary2Reg, tmp4Reg, Address::times_1));
1.413 + masm.pxor(tmp1Reg, tmp2Reg);
1.414 + masm.ptest(tmp1Reg, tmp1Reg);
1.415 +
1.416 + masm.jccb(Assembler::notZero, FALSE_LABEL);
1.417 + masm.addl(tmp4Reg, 16);
1.418 + masm.jcc(Assembler::notZero, COMPARE_WIDE_VECTORS);
1.419 + masm.bind(COMPARE_TAIL);
1.420 + masm.movl(tmp4Reg, resultReg);
1.421 + // Fallthru to tail compare
1.422 + }
1.423 +
1.424 + // Compare 4-byte vectors
1.425 + masm.andl(tmp4Reg, 0xfffffffc); // vector count (in bytes)
1.426 + masm.andl(resultReg, 0x00000002); // tail char (in bytes)
1.427 + masm.testl(tmp4Reg, tmp4Reg);
1.428 + masm.jccb(Assembler::zero, COMPARE_CHAR);
1.429 + masm.lea(ary1Reg, Address(ary1Reg, tmp4Reg, Address::times_1));
1.430 + masm.lea(ary2Reg, Address(ary2Reg, tmp4Reg, Address::times_1));
1.431 + masm.negl(tmp4Reg);
1.432 +
1.433 + masm.bind(COMPARE_VECTORS);
1.434 + masm.movl(tmp3Reg, Address(ary1Reg, tmp4Reg, Address::times_1));
1.435 + masm.cmpl(tmp3Reg, Address(ary2Reg, tmp4Reg, Address::times_1));
1.436 + masm.jccb(Assembler::notEqual, FALSE_LABEL);
1.437 + masm.addl(tmp4Reg, 4);
1.438 + masm.jcc(Assembler::notZero, COMPARE_VECTORS);
1.439 +
1.440 + // Compare trailing char (final 2 bytes), if any
1.441 + masm.bind(COMPARE_CHAR);
1.442 masm.testl(resultReg, resultReg);
1.443 - masm.jcc(Assembler::zero, TRUE_LABEL);
1.444 -
1.445 - // Setup compare loop
1.446 - masm.bind(COMPARE_LOOP_HDR);
1.447 - // Shift tmp1Reg and tmp2Reg to the last 4-byte boundary of the arrays
1.448 - masm.leal(tmp1Reg, Address(ary1Reg, resultReg, Address::times_4, base_offset));
1.449 - masm.leal(tmp2Reg, Address(ary2Reg, resultReg, Address::times_4, base_offset));
1.450 - masm.negl(resultReg);
1.451 -
1.452 - // 4-byte-wide compare loop
1.453 - masm.bind(COMPARE_LOOP);
1.454 - masm.movl(ary1Reg, Address(tmp1Reg, resultReg, Address::times_4, 0));
1.455 - masm.movl(ary2Reg, Address(tmp2Reg, resultReg, Address::times_4, 0));
1.456 - masm.cmpl(ary1Reg, ary2Reg);
1.457 - masm.jcc(Assembler::notEqual, FALSE_LABEL);
1.458 - masm.increment(resultReg);
1.459 - masm.jcc(Assembler::notZero, COMPARE_LOOP);
1.460 + masm.jccb(Assembler::zero, TRUE_LABEL);
1.461 + masm.load_unsigned_short(tmp3Reg, Address(ary1Reg, 0));
1.462 + masm.load_unsigned_short(tmp4Reg, Address(ary2Reg, 0));
1.463 + masm.cmpl(tmp3Reg, tmp4Reg);
1.464 + masm.jccb(Assembler::notEqual, FALSE_LABEL);
1.465
1.466 masm.bind(TRUE_LABEL);
1.467 masm.movl(resultReg, 1); // return true
1.468 - masm.jmp(DONE_LABEL);
1.469 + masm.jmpb(DONE);
1.470
1.471 masm.bind(FALSE_LABEL);
1.472 masm.xorl(resultReg, resultReg); // return false
1.473
1.474 // That's it
1.475 - masm.bind(DONE_LABEL);
1.476 + masm.bind(DONE);
1.477 %}
1.478
1.479 enc_class enc_pop_rdx() %{
1.480 @@ -12074,11 +12356,8 @@
1.481 ins_pipe( fpu_reg_reg );
1.482 %}
1.483
1.484 -
1.485 -
1.486 // =======================================================================
1.487 // fast clearing of an array
1.488 -
1.489 instruct rep_stos(eCXRegI cnt, eDIRegP base, eAXRegI zero, Universe dummy, eFlagsReg cr) %{
1.490 match(Set dummy (ClearArray cnt base));
1.491 effect(USE_KILL cnt, USE_KILL base, KILL zero, KILL cr);
1.492 @@ -12092,24 +12371,48 @@
1.493 ins_pipe( pipe_slow );
1.494 %}
1.495
1.496 -instruct string_compare(eDIRegP str1, eSIRegP str2, eAXRegI tmp1, eBXRegI tmp2, eCXRegI result, eFlagsReg cr) %{
1.497 +instruct string_compare(eDIRegP str1, eSIRegP str2, regXD tmp1, regXD tmp2,
1.498 + eAXRegI tmp3, eBXRegI tmp4, eCXRegI result, eFlagsReg cr) %{
1.499 match(Set result (StrComp str1 str2));
1.500 - effect(USE_KILL str1, USE_KILL str2, KILL tmp1, KILL tmp2, KILL cr);
1.501 + effect(TEMP tmp1, TEMP tmp2, USE_KILL str1, USE_KILL str2, KILL tmp3, KILL tmp4, KILL cr);
1.502 //ins_cost(300);
1.503
1.504 format %{ "String Compare $str1,$str2 -> $result // KILL EAX, EBX" %}
1.505 - ins_encode( enc_String_Compare() );
1.506 + ins_encode( enc_String_Compare(str1, str2, tmp1, tmp2, tmp3, tmp4, result) );
1.507 + ins_pipe( pipe_slow );
1.508 +%}
1.509 +
1.510 +// fast string equals
1.511 +instruct string_equals(eDIRegP str1, eSIRegP str2, regXD tmp1, regXD tmp2,
1.512 + eBXRegI tmp3, eCXRegI tmp4, eAXRegI result, eFlagsReg cr) %{
1.513 + match(Set result (StrEquals str1 str2));
1.514 + effect(TEMP tmp1, TEMP tmp2, USE_KILL str1, USE_KILL str2, KILL tmp3, KILL tmp4, KILL cr);
1.515 +
1.516 + format %{ "String Equals $str1,$str2 -> $result // KILL EBX, ECX" %}
1.517 + ins_encode( enc_String_Equals(tmp1, tmp2, str1, str2, tmp3, tmp4, result) );
1.518 + ins_pipe( pipe_slow );
1.519 +%}
1.520 +
1.521 +instruct string_indexof(eSIRegP str1, eDIRegP str2, regXD tmp1, eAXRegI tmp2,
1.522 + eCXRegI tmp3, eDXRegI tmp4, eBXRegI result, eFlagsReg cr) %{
1.523 + predicate(UseSSE42Intrinsics);
1.524 + match(Set result (StrIndexOf str1 str2));
1.525 + effect(TEMP tmp1, USE_KILL str1, USE_KILL str2, KILL tmp2, KILL tmp3, KILL tmp4, KILL cr);
1.526 +
1.527 + format %{ "String IndexOf $str1,$str2 -> $result // KILL EAX, ECX, EDX" %}
1.528 + ins_encode( enc_String_IndexOf(str1, str2, tmp1, tmp2, tmp3, tmp4, result) );
1.529 ins_pipe( pipe_slow );
1.530 %}
1.531
1.532 // fast array equals
1.533 -instruct array_equals(eDIRegP ary1, eSIRegP ary2, eAXRegI tmp1, eBXRegI tmp2, eCXRegI result, eFlagsReg cr) %{
1.534 +instruct array_equals(eDIRegP ary1, eSIRegP ary2, regXD tmp1, regXD tmp2, eBXRegI tmp3,
1.535 + eDXRegI tmp4, eAXRegI result, eFlagsReg cr) %{
1.536 match(Set result (AryEq ary1 ary2));
1.537 - effect(USE_KILL ary1, USE_KILL ary2, KILL tmp1, KILL tmp2, KILL cr);
1.538 + effect(TEMP tmp1, TEMP tmp2, USE_KILL ary1, USE_KILL ary2, KILL tmp3, KILL tmp4, KILL cr);
1.539 //ins_cost(300);
1.540
1.541 - format %{ "Array Equals $ary1,$ary2 -> $result // KILL EAX, EBX" %}
1.542 - ins_encode( enc_Array_Equals(ary1, ary2, tmp1, tmp2, result) );
1.543 + format %{ "Array Equals $ary1,$ary2 -> $result // KILL EBX, EDX" %}
1.544 + ins_encode( enc_Array_Equals(ary1, ary2, tmp1, tmp2, tmp3, tmp4, result) );
1.545 ins_pipe( pipe_slow );
1.546 %}
1.547
