duke@435: /* coleenp@4037: * Copyright (c) 2003, 2012, Oracle and/or its affiliates. All rights reserved. duke@435: * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. duke@435: * duke@435: * This code is free software; you can redistribute it and/or modify it duke@435: * under the terms of the GNU General Public License version 2 only, as duke@435: * published by the Free Software Foundation. duke@435: * duke@435: * This code is distributed in the hope that it will be useful, but WITHOUT duke@435: * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or duke@435: * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License duke@435: * version 2 for more details (a copy is included in the LICENSE file that duke@435: * accompanied this code). duke@435: * duke@435: * You should have received a copy of the GNU General Public License version duke@435: * 2 along with this work; if not, write to the Free Software Foundation, duke@435: * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. duke@435: * trims@1907: * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA trims@1907: * or visit www.oracle.com if you need additional information or have any trims@1907: * questions. duke@435: * duke@435: */ duke@435: stefank@2314: #include "precompiled.hpp" twisti@4318: #include "asm/macroAssembler.hpp" twisti@4318: #include "asm/macroAssembler.inline.hpp" stefank@2314: #include "interpreter/interpreter.hpp" stefank@2314: #include "nativeInst_x86.hpp" stefank@2314: #include "oops/instanceOop.hpp" coleenp@4037: #include "oops/method.hpp" stefank@2314: #include "oops/objArrayKlass.hpp" stefank@2314: #include "oops/oop.inline.hpp" stefank@2314: #include "prims/methodHandles.hpp" stefank@2314: #include "runtime/frame.inline.hpp" stefank@2314: #include "runtime/handles.inline.hpp" stefank@2314: #include "runtime/sharedRuntime.hpp" stefank@2314: #include "runtime/stubCodeGenerator.hpp" stefank@2314: #include "runtime/stubRoutines.hpp" stefank@4299: #include "runtime/thread.inline.hpp" stefank@2314: #include "utilities/top.hpp" stefank@2314: #ifdef COMPILER2 stefank@2314: #include "opto/runtime.hpp" stefank@2314: #endif duke@435: duke@435: // Declaration and definition of StubGenerator (no .hpp file). duke@435: // For a more detailed description of the stub routine structure duke@435: // see the comment in stubRoutines.hpp duke@435: duke@435: #define __ _masm-> coleenp@548: #define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8) never@739: #define a__ ((Assembler*)_masm)-> duke@435: duke@435: #ifdef PRODUCT duke@435: #define BLOCK_COMMENT(str) /* nothing */ duke@435: #else duke@435: #define BLOCK_COMMENT(str) __ block_comment(str) duke@435: #endif duke@435: duke@435: #define BIND(label) bind(label); BLOCK_COMMENT(#label ":") duke@435: const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions duke@435: duke@435: // Stub Code definitions duke@435: duke@435: static address handle_unsafe_access() { duke@435: JavaThread* thread = JavaThread::current(); duke@435: address pc = thread->saved_exception_pc(); duke@435: // pc is the instruction which we must emulate duke@435: // doing a no-op is fine: return garbage from the load duke@435: // therefore, compute npc duke@435: address npc = Assembler::locate_next_instruction(pc); duke@435: duke@435: // request an async exception duke@435: thread->set_pending_unsafe_access_error(); duke@435: duke@435: // return address of next instruction to execute duke@435: return npc; duke@435: } duke@435: duke@435: class StubGenerator: public StubCodeGenerator { duke@435: private: duke@435: duke@435: #ifdef PRODUCT duke@435: #define inc_counter_np(counter) (0) duke@435: #else duke@435: void inc_counter_np_(int& counter) { never@3314: // This can destroy rscratch1 if counter is far from the code cache duke@435: __ incrementl(ExternalAddress((address)&counter)); duke@435: } duke@435: #define inc_counter_np(counter) \ duke@435: BLOCK_COMMENT("inc_counter " #counter); \ duke@435: inc_counter_np_(counter); duke@435: #endif duke@435: duke@435: // Call stubs are used to call Java from C duke@435: // duke@435: // Linux Arguments: duke@435: // c_rarg0: call wrapper address address duke@435: // c_rarg1: result address duke@435: // c_rarg2: result type BasicType coleenp@4037: // c_rarg3: method Method* duke@435: // c_rarg4: (interpreter) entry point address duke@435: // c_rarg5: parameters intptr_t* duke@435: // 16(rbp): parameter size (in words) int duke@435: // 24(rbp): thread Thread* duke@435: // duke@435: // [ return_from_Java ] <--- rsp duke@435: // [ argument word n ] duke@435: // ... duke@435: // -12 [ argument word 1 ] duke@435: // -11 [ saved r15 ] <--- rsp_after_call duke@435: // -10 [ saved r14 ] duke@435: // -9 [ saved r13 ] duke@435: // -8 [ saved r12 ] duke@435: // -7 [ saved rbx ] duke@435: // -6 [ call wrapper ] duke@435: // -5 [ result ] duke@435: // -4 [ result type ] duke@435: // -3 [ method ] duke@435: // -2 [ entry point ] duke@435: // -1 [ parameters ] duke@435: // 0 [ saved rbp ] <--- rbp duke@435: // 1 [ return address ] duke@435: // 2 [ parameter size ] duke@435: // 3 [ thread ] duke@435: // duke@435: // Windows Arguments: duke@435: // c_rarg0: call wrapper address address duke@435: // c_rarg1: result address duke@435: // c_rarg2: result type BasicType coleenp@4037: // c_rarg3: method Method* duke@435: // 48(rbp): (interpreter) entry point address duke@435: // 56(rbp): parameters intptr_t* duke@435: // 64(rbp): parameter size (in words) int duke@435: // 72(rbp): thread Thread* duke@435: // duke@435: // [ return_from_Java ] <--- rsp duke@435: // [ argument word n ] duke@435: // ... iveresov@2689: // -28 [ argument word 1 ] iveresov@2689: // -27 [ saved xmm15 ] <--- rsp_after_call iveresov@2689: // [ saved xmm7-xmm14 ] iveresov@2689: // -9 [ saved xmm6 ] (each xmm register takes 2 slots) iveresov@2689: // -7 [ saved r15 ] duke@435: // -6 [ saved r14 ] duke@435: // -5 [ saved r13 ] duke@435: // -4 [ saved r12 ] duke@435: // -3 [ saved rdi ] duke@435: // -2 [ saved rsi ] duke@435: // -1 [ saved rbx ] duke@435: // 0 [ saved rbp ] <--- rbp duke@435: // 1 [ return address ] duke@435: // 2 [ call wrapper ] duke@435: // 3 [ result ] duke@435: // 4 [ result type ] duke@435: // 5 [ method ] duke@435: // 6 [ entry point ] duke@435: // 7 [ parameters ] duke@435: // 8 [ parameter size ] duke@435: // 9 [ thread ] duke@435: // duke@435: // Windows reserves the callers stack space for arguments 1-4. duke@435: // We spill c_rarg0-c_rarg3 to this space. duke@435: duke@435: // Call stub stack layout word offsets from rbp duke@435: enum call_stub_layout { duke@435: #ifdef _WIN64 iveresov@2689: xmm_save_first = 6, // save from xmm6 iveresov@2689: xmm_save_last = 15, // to xmm15 iveresov@2689: xmm_save_base = -9, iveresov@2689: rsp_after_call_off = xmm_save_base - 2 * (xmm_save_last - xmm_save_first), // -27 iveresov@2689: r15_off = -7, duke@435: r14_off = -6, duke@435: r13_off = -5, duke@435: r12_off = -4, duke@435: rdi_off = -3, duke@435: rsi_off = -2, duke@435: rbx_off = -1, duke@435: rbp_off = 0, duke@435: retaddr_off = 1, duke@435: call_wrapper_off = 2, duke@435: result_off = 3, duke@435: result_type_off = 4, duke@435: method_off = 5, duke@435: entry_point_off = 6, duke@435: parameters_off = 7, duke@435: parameter_size_off = 8, duke@435: thread_off = 9 duke@435: #else duke@435: rsp_after_call_off = -12, duke@435: mxcsr_off = rsp_after_call_off, duke@435: r15_off = -11, duke@435: r14_off = -10, duke@435: r13_off = -9, duke@435: r12_off = -8, duke@435: rbx_off = -7, duke@435: call_wrapper_off = -6, duke@435: result_off = -5, duke@435: result_type_off = -4, duke@435: method_off = -3, duke@435: entry_point_off = -2, duke@435: parameters_off = -1, duke@435: rbp_off = 0, duke@435: retaddr_off = 1, duke@435: parameter_size_off = 2, duke@435: thread_off = 3 duke@435: #endif duke@435: }; duke@435: iveresov@2689: #ifdef _WIN64 iveresov@2689: Address xmm_save(int reg) { iveresov@2689: assert(reg >= xmm_save_first && reg <= xmm_save_last, "XMM register number out of range"); iveresov@2689: return Address(rbp, (xmm_save_base - (reg - xmm_save_first) * 2) * wordSize); iveresov@2689: } iveresov@2689: #endif iveresov@2689: duke@435: address generate_call_stub(address& return_address) { duke@435: assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 && duke@435: (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off, duke@435: "adjust this code"); duke@435: StubCodeMark mark(this, "StubRoutines", "call_stub"); duke@435: address start = __ pc(); duke@435: duke@435: // same as in generate_catch_exception()! duke@435: const Address rsp_after_call(rbp, rsp_after_call_off * wordSize); duke@435: duke@435: const Address call_wrapper (rbp, call_wrapper_off * wordSize); duke@435: const Address result (rbp, result_off * wordSize); duke@435: const Address result_type (rbp, result_type_off * wordSize); duke@435: const Address method (rbp, method_off * wordSize); duke@435: const Address entry_point (rbp, entry_point_off * wordSize); duke@435: const Address parameters (rbp, parameters_off * wordSize); duke@435: const Address parameter_size(rbp, parameter_size_off * wordSize); duke@435: duke@435: // same as in generate_catch_exception()! duke@435: const Address thread (rbp, thread_off * wordSize); duke@435: duke@435: const Address r15_save(rbp, r15_off * wordSize); duke@435: const Address r14_save(rbp, r14_off * wordSize); duke@435: const Address r13_save(rbp, r13_off * wordSize); duke@435: const Address r12_save(rbp, r12_off * wordSize); duke@435: const Address rbx_save(rbp, rbx_off * wordSize); duke@435: duke@435: // stub code duke@435: __ enter(); never@739: __ subptr(rsp, -rsp_after_call_off * wordSize); duke@435: duke@435: // save register parameters duke@435: #ifndef _WIN64 never@739: __ movptr(parameters, c_rarg5); // parameters never@739: __ movptr(entry_point, c_rarg4); // entry_point duke@435: #endif duke@435: never@739: __ movptr(method, c_rarg3); // method never@739: __ movl(result_type, c_rarg2); // result type never@739: __ movptr(result, c_rarg1); // result never@739: __ movptr(call_wrapper, c_rarg0); // call wrapper duke@435: duke@435: // save regs belonging to calling function never@739: __ movptr(rbx_save, rbx); never@739: __ movptr(r12_save, r12); never@739: __ movptr(r13_save, r13); never@739: __ movptr(r14_save, r14); never@739: __ movptr(r15_save, r15); duke@435: #ifdef _WIN64 iveresov@2689: for (int i = 6; i <= 15; i++) { iveresov@2689: __ movdqu(xmm_save(i), as_XMMRegister(i)); iveresov@2689: } iveresov@2689: duke@435: const Address rdi_save(rbp, rdi_off * wordSize); duke@435: const Address rsi_save(rbp, rsi_off * wordSize); duke@435: never@739: __ movptr(rsi_save, rsi); never@739: __ movptr(rdi_save, rdi); duke@435: #else duke@435: const Address mxcsr_save(rbp, mxcsr_off * wordSize); duke@435: { duke@435: Label skip_ldmx; duke@435: __ stmxcsr(mxcsr_save); duke@435: __ movl(rax, mxcsr_save); duke@435: __ andl(rax, MXCSR_MASK); // Only check control and mask bits never@739: ExternalAddress mxcsr_std(StubRoutines::x86::mxcsr_std()); duke@435: __ cmp32(rax, mxcsr_std); duke@435: __ jcc(Assembler::equal, skip_ldmx); duke@435: __ ldmxcsr(mxcsr_std); duke@435: __ bind(skip_ldmx); duke@435: } duke@435: #endif duke@435: duke@435: // Load up thread register never@739: __ movptr(r15_thread, thread); coleenp@548: __ reinit_heapbase(); duke@435: duke@435: #ifdef ASSERT duke@435: // make sure we have no pending exceptions duke@435: { duke@435: Label L; never@739: __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); duke@435: __ jcc(Assembler::equal, L); duke@435: __ stop("StubRoutines::call_stub: entered with pending exception"); duke@435: __ bind(L); duke@435: } duke@435: #endif duke@435: duke@435: // pass parameters if any duke@435: BLOCK_COMMENT("pass parameters if any"); duke@435: Label parameters_done; duke@435: __ movl(c_rarg3, parameter_size); duke@435: __ testl(c_rarg3, c_rarg3); duke@435: __ jcc(Assembler::zero, parameters_done); duke@435: duke@435: Label loop; never@739: __ movptr(c_rarg2, parameters); // parameter pointer never@739: __ movl(c_rarg1, c_rarg3); // parameter counter is in c_rarg1 duke@435: __ BIND(loop); never@739: __ movptr(rax, Address(c_rarg2, 0));// get parameter never@739: __ addptr(c_rarg2, wordSize); // advance to next parameter never@739: __ decrementl(c_rarg1); // decrement counter never@739: __ push(rax); // pass parameter duke@435: __ jcc(Assembler::notZero, loop); duke@435: duke@435: // call Java function duke@435: __ BIND(parameters_done); coleenp@4037: __ movptr(rbx, method); // get Method* never@739: __ movptr(c_rarg1, entry_point); // get entry_point never@739: __ mov(r13, rsp); // set sender sp duke@435: BLOCK_COMMENT("call Java function"); duke@435: __ call(c_rarg1); duke@435: duke@435: BLOCK_COMMENT("call_stub_return_address:"); duke@435: return_address = __ pc(); duke@435: duke@435: // store result depending on type (everything that is not duke@435: // T_OBJECT, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT) never@739: __ movptr(c_rarg0, result); duke@435: Label is_long, is_float, is_double, exit; duke@435: __ movl(c_rarg1, result_type); duke@435: __ cmpl(c_rarg1, T_OBJECT); duke@435: __ jcc(Assembler::equal, is_long); duke@435: __ cmpl(c_rarg1, T_LONG); duke@435: __ jcc(Assembler::equal, is_long); duke@435: __ cmpl(c_rarg1, T_FLOAT); duke@435: __ jcc(Assembler::equal, is_float); duke@435: __ cmpl(c_rarg1, T_DOUBLE); duke@435: __ jcc(Assembler::equal, is_double); duke@435: duke@435: // handle T_INT case duke@435: __ movl(Address(c_rarg0, 0), rax); duke@435: duke@435: __ BIND(exit); duke@435: duke@435: // pop parameters never@739: __ lea(rsp, rsp_after_call); duke@435: duke@435: #ifdef ASSERT duke@435: // verify that threads correspond duke@435: { duke@435: Label L, S; never@739: __ cmpptr(r15_thread, thread); duke@435: __ jcc(Assembler::notEqual, S); duke@435: __ get_thread(rbx); never@739: __ cmpptr(r15_thread, rbx); duke@435: __ jcc(Assembler::equal, L); duke@435: __ bind(S); duke@435: __ jcc(Assembler::equal, L); duke@435: __ stop("StubRoutines::call_stub: threads must correspond"); duke@435: __ bind(L); duke@435: } duke@435: #endif duke@435: duke@435: // restore regs belonging to calling function iveresov@2689: #ifdef _WIN64 iveresov@2689: for (int i = 15; i >= 6; i--) { iveresov@2689: __ movdqu(as_XMMRegister(i), xmm_save(i)); iveresov@2689: } iveresov@2689: #endif never@739: __ movptr(r15, r15_save); never@739: __ movptr(r14, r14_save); never@739: __ movptr(r13, r13_save); never@739: __ movptr(r12, r12_save); never@739: __ movptr(rbx, rbx_save); duke@435: duke@435: #ifdef _WIN64 never@739: __ movptr(rdi, rdi_save); never@739: __ movptr(rsi, rsi_save); duke@435: #else duke@435: __ ldmxcsr(mxcsr_save); duke@435: #endif duke@435: duke@435: // restore rsp never@739: __ addptr(rsp, -rsp_after_call_off * wordSize); duke@435: duke@435: // return never@739: __ pop(rbp); duke@435: __ ret(0); duke@435: duke@435: // handle return types different from T_INT duke@435: __ BIND(is_long); duke@435: __ movq(Address(c_rarg0, 0), rax); duke@435: __ jmp(exit); duke@435: duke@435: __ BIND(is_float); duke@435: __ movflt(Address(c_rarg0, 0), xmm0); duke@435: __ jmp(exit); duke@435: duke@435: __ BIND(is_double); duke@435: __ movdbl(Address(c_rarg0, 0), xmm0); duke@435: __ jmp(exit); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: // Return point for a Java call if there's an exception thrown in duke@435: // Java code. The exception is caught and transformed into a duke@435: // pending exception stored in JavaThread that can be tested from duke@435: // within the VM. duke@435: // duke@435: // Note: Usually the parameters are removed by the callee. In case duke@435: // of an exception crossing an activation frame boundary, that is duke@435: // not the case if the callee is compiled code => need to setup the duke@435: // rsp. duke@435: // duke@435: // rax: exception oop duke@435: duke@435: address generate_catch_exception() { duke@435: StubCodeMark mark(this, "StubRoutines", "catch_exception"); duke@435: address start = __ pc(); duke@435: duke@435: // same as in generate_call_stub(): duke@435: const Address rsp_after_call(rbp, rsp_after_call_off * wordSize); duke@435: const Address thread (rbp, thread_off * wordSize); duke@435: duke@435: #ifdef ASSERT duke@435: // verify that threads correspond duke@435: { duke@435: Label L, S; never@739: __ cmpptr(r15_thread, thread); duke@435: __ jcc(Assembler::notEqual, S); duke@435: __ get_thread(rbx); never@739: __ cmpptr(r15_thread, rbx); duke@435: __ jcc(Assembler::equal, L); duke@435: __ bind(S); duke@435: __ stop("StubRoutines::catch_exception: threads must correspond"); duke@435: __ bind(L); duke@435: } duke@435: #endif duke@435: duke@435: // set pending exception duke@435: __ verify_oop(rax); duke@435: never@739: __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax); duke@435: __ lea(rscratch1, ExternalAddress((address)__FILE__)); never@739: __ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1); duke@435: __ movl(Address(r15_thread, Thread::exception_line_offset()), (int) __LINE__); duke@435: duke@435: // complete return to VM duke@435: assert(StubRoutines::_call_stub_return_address != NULL, duke@435: "_call_stub_return_address must have been generated before"); duke@435: __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address)); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: // Continuation point for runtime calls returning with a pending duke@435: // exception. The pending exception check happened in the runtime duke@435: // or native call stub. The pending exception in Thread is duke@435: // converted into a Java-level exception. duke@435: // duke@435: // Contract with Java-level exception handlers: duke@435: // rax: exception duke@435: // rdx: throwing pc duke@435: // duke@435: // NOTE: At entry of this stub, exception-pc must be on stack !! duke@435: duke@435: address generate_forward_exception() { duke@435: StubCodeMark mark(this, "StubRoutines", "forward exception"); duke@435: address start = __ pc(); duke@435: duke@435: // Upon entry, the sp points to the return address returning into duke@435: // Java (interpreted or compiled) code; i.e., the return address duke@435: // becomes the throwing pc. duke@435: // duke@435: // Arguments pushed before the runtime call are still on the stack duke@435: // but the exception handler will reset the stack pointer -> duke@435: // ignore them. A potential result in registers can be ignored as duke@435: // well. duke@435: duke@435: #ifdef ASSERT duke@435: // make sure this code is only executed if there is a pending exception duke@435: { duke@435: Label L; never@739: __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL); duke@435: __ jcc(Assembler::notEqual, L); duke@435: __ stop("StubRoutines::forward exception: no pending exception (1)"); duke@435: __ bind(L); duke@435: } duke@435: #endif duke@435: duke@435: // compute exception handler into rbx never@739: __ movptr(c_rarg0, Address(rsp, 0)); duke@435: BLOCK_COMMENT("call exception_handler_for_return_address"); duke@435: __ call_VM_leaf(CAST_FROM_FN_PTR(address, duke@435: SharedRuntime::exception_handler_for_return_address), twisti@1730: r15_thread, c_rarg0); never@739: __ mov(rbx, rax); duke@435: duke@435: // setup rax & rdx, remove return address & clear pending exception never@739: __ pop(rdx); never@739: __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset())); xlu@947: __ movptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); duke@435: duke@435: #ifdef ASSERT duke@435: // make sure exception is set duke@435: { duke@435: Label L; never@739: __ testptr(rax, rax); duke@435: __ jcc(Assembler::notEqual, L); duke@435: __ stop("StubRoutines::forward exception: no pending exception (2)"); duke@435: __ bind(L); duke@435: } duke@435: #endif duke@435: duke@435: // continue at exception handler (return address removed) duke@435: // rax: exception duke@435: // rbx: exception handler duke@435: // rdx: throwing pc duke@435: __ verify_oop(rax); duke@435: __ jmp(rbx); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: // Support for jint atomic::xchg(jint exchange_value, volatile jint* dest) duke@435: // duke@435: // Arguments : duke@435: // c_rarg0: exchange_value duke@435: // c_rarg0: dest duke@435: // duke@435: // Result: duke@435: // *dest <- ex, return (orig *dest) duke@435: address generate_atomic_xchg() { duke@435: StubCodeMark mark(this, "StubRoutines", "atomic_xchg"); duke@435: address start = __ pc(); duke@435: duke@435: __ movl(rax, c_rarg0); // Copy to eax we need a return value anyhow duke@435: __ xchgl(rax, Address(c_rarg1, 0)); // automatic LOCK duke@435: __ ret(0); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: // Support for intptr_t atomic::xchg_ptr(intptr_t exchange_value, volatile intptr_t* dest) duke@435: // duke@435: // Arguments : duke@435: // c_rarg0: exchange_value duke@435: // c_rarg1: dest duke@435: // duke@435: // Result: duke@435: // *dest <- ex, return (orig *dest) duke@435: address generate_atomic_xchg_ptr() { duke@435: StubCodeMark mark(this, "StubRoutines", "atomic_xchg_ptr"); duke@435: address start = __ pc(); duke@435: never@739: __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow never@739: __ xchgptr(rax, Address(c_rarg1, 0)); // automatic LOCK duke@435: __ ret(0); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: // Support for jint atomic::atomic_cmpxchg(jint exchange_value, volatile jint* dest, duke@435: // jint compare_value) duke@435: // duke@435: // Arguments : duke@435: // c_rarg0: exchange_value duke@435: // c_rarg1: dest duke@435: // c_rarg2: compare_value duke@435: // duke@435: // Result: duke@435: // if ( compare_value == *dest ) { duke@435: // *dest = exchange_value duke@435: // return compare_value; duke@435: // else duke@435: // return *dest; duke@435: address generate_atomic_cmpxchg() { duke@435: StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg"); duke@435: address start = __ pc(); duke@435: duke@435: __ movl(rax, c_rarg2); duke@435: if ( os::is_MP() ) __ lock(); duke@435: __ cmpxchgl(c_rarg0, Address(c_rarg1, 0)); duke@435: __ ret(0); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: // Support for jint atomic::atomic_cmpxchg_long(jlong exchange_value, duke@435: // volatile jlong* dest, duke@435: // jlong compare_value) duke@435: // Arguments : duke@435: // c_rarg0: exchange_value duke@435: // c_rarg1: dest duke@435: // c_rarg2: compare_value duke@435: // duke@435: // Result: duke@435: // if ( compare_value == *dest ) { duke@435: // *dest = exchange_value duke@435: // return compare_value; duke@435: // else duke@435: // return *dest; duke@435: address generate_atomic_cmpxchg_long() { duke@435: StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long"); duke@435: address start = __ pc(); duke@435: duke@435: __ movq(rax, c_rarg2); duke@435: if ( os::is_MP() ) __ lock(); duke@435: __ cmpxchgq(c_rarg0, Address(c_rarg1, 0)); duke@435: __ ret(0); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: // Support for jint atomic::add(jint add_value, volatile jint* dest) duke@435: // duke@435: // Arguments : duke@435: // c_rarg0: add_value duke@435: // c_rarg1: dest duke@435: // duke@435: // Result: duke@435: // *dest += add_value duke@435: // return *dest; duke@435: address generate_atomic_add() { duke@435: StubCodeMark mark(this, "StubRoutines", "atomic_add"); duke@435: address start = __ pc(); duke@435: duke@435: __ movl(rax, c_rarg0); duke@435: if ( os::is_MP() ) __ lock(); duke@435: __ xaddl(Address(c_rarg1, 0), c_rarg0); duke@435: __ addl(rax, c_rarg0); duke@435: __ ret(0); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: // Support for intptr_t atomic::add_ptr(intptr_t add_value, volatile intptr_t* dest) duke@435: // duke@435: // Arguments : duke@435: // c_rarg0: add_value duke@435: // c_rarg1: dest duke@435: // duke@435: // Result: duke@435: // *dest += add_value duke@435: // return *dest; duke@435: address generate_atomic_add_ptr() { duke@435: StubCodeMark mark(this, "StubRoutines", "atomic_add_ptr"); duke@435: address start = __ pc(); duke@435: never@739: __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow duke@435: if ( os::is_MP() ) __ lock(); never@739: __ xaddptr(Address(c_rarg1, 0), c_rarg0); never@739: __ addptr(rax, c_rarg0); duke@435: __ ret(0); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: // Support for intptr_t OrderAccess::fence() duke@435: // duke@435: // Arguments : duke@435: // duke@435: // Result: duke@435: address generate_orderaccess_fence() { duke@435: StubCodeMark mark(this, "StubRoutines", "orderaccess_fence"); duke@435: address start = __ pc(); never@1106: __ membar(Assembler::StoreLoad); duke@435: __ ret(0); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: // Support for intptr_t get_previous_fp() duke@435: // duke@435: // This routine is used to find the previous frame pointer for the duke@435: // caller (current_frame_guess). This is used as part of debugging duke@435: // ps() is seemingly lost trying to find frames. duke@435: // This code assumes that caller current_frame_guess) has a frame. duke@435: address generate_get_previous_fp() { duke@435: StubCodeMark mark(this, "StubRoutines", "get_previous_fp"); duke@435: const Address old_fp(rbp, 0); duke@435: const Address older_fp(rax, 0); duke@435: address start = __ pc(); duke@435: duke@435: __ enter(); never@739: __ movptr(rax, old_fp); // callers fp never@739: __ movptr(rax, older_fp); // the frame for ps() never@739: __ pop(rbp); duke@435: __ ret(0); duke@435: duke@435: return start; duke@435: } duke@435: roland@3606: // Support for intptr_t get_previous_sp() roland@3606: // roland@3606: // This routine is used to find the previous stack pointer for the roland@3606: // caller. roland@3606: address generate_get_previous_sp() { roland@3606: StubCodeMark mark(this, "StubRoutines", "get_previous_sp"); roland@3606: address start = __ pc(); roland@3606: roland@3606: __ movptr(rax, rsp); roland@3606: __ addptr(rax, 8); // return address is at the top of the stack. roland@3606: __ ret(0); roland@3606: roland@3606: return start; roland@3606: } roland@3606: duke@435: //---------------------------------------------------------------------------------------------------- duke@435: // Support for void verify_mxcsr() duke@435: // duke@435: // This routine is used with -Xcheck:jni to verify that native duke@435: // JNI code does not return to Java code without restoring the duke@435: // MXCSR register to our expected state. duke@435: duke@435: address generate_verify_mxcsr() { duke@435: StubCodeMark mark(this, "StubRoutines", "verify_mxcsr"); duke@435: address start = __ pc(); duke@435: duke@435: const Address mxcsr_save(rsp, 0); duke@435: duke@435: if (CheckJNICalls) { duke@435: Label ok_ret; never@739: __ push(rax); never@739: __ subptr(rsp, wordSize); // allocate a temp location duke@435: __ stmxcsr(mxcsr_save); duke@435: __ movl(rax, mxcsr_save); duke@435: __ andl(rax, MXCSR_MASK); // Only check control and mask bits never@739: __ cmpl(rax, *(int *)(StubRoutines::x86::mxcsr_std())); duke@435: __ jcc(Assembler::equal, ok_ret); duke@435: duke@435: __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall"); duke@435: never@739: __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std())); duke@435: duke@435: __ bind(ok_ret); never@739: __ addptr(rsp, wordSize); never@739: __ pop(rax); duke@435: } duke@435: duke@435: __ ret(0); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: address generate_f2i_fixup() { duke@435: StubCodeMark mark(this, "StubRoutines", "f2i_fixup"); duke@435: Address inout(rsp, 5 * wordSize); // return address + 4 saves duke@435: duke@435: address start = __ pc(); duke@435: duke@435: Label L; duke@435: never@739: __ push(rax); never@739: __ push(c_rarg3); never@739: __ push(c_rarg2); never@739: __ push(c_rarg1); duke@435: duke@435: __ movl(rax, 0x7f800000); duke@435: __ xorl(c_rarg3, c_rarg3); duke@435: __ movl(c_rarg2, inout); duke@435: __ movl(c_rarg1, c_rarg2); duke@435: __ andl(c_rarg1, 0x7fffffff); duke@435: __ cmpl(rax, c_rarg1); // NaN? -> 0 duke@435: __ jcc(Assembler::negative, L); duke@435: __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint duke@435: __ movl(c_rarg3, 0x80000000); duke@435: __ movl(rax, 0x7fffffff); duke@435: __ cmovl(Assembler::positive, c_rarg3, rax); duke@435: duke@435: __ bind(L); never@739: __ movptr(inout, c_rarg3); never@739: never@739: __ pop(c_rarg1); never@739: __ pop(c_rarg2); never@739: __ pop(c_rarg3); never@739: __ pop(rax); duke@435: duke@435: __ ret(0); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: address generate_f2l_fixup() { duke@435: StubCodeMark mark(this, "StubRoutines", "f2l_fixup"); duke@435: Address inout(rsp, 5 * wordSize); // return address + 4 saves duke@435: address start = __ pc(); duke@435: duke@435: Label L; duke@435: never@739: __ push(rax); never@739: __ push(c_rarg3); never@739: __ push(c_rarg2); never@739: __ push(c_rarg1); duke@435: duke@435: __ movl(rax, 0x7f800000); duke@435: __ xorl(c_rarg3, c_rarg3); duke@435: __ movl(c_rarg2, inout); duke@435: __ movl(c_rarg1, c_rarg2); duke@435: __ andl(c_rarg1, 0x7fffffff); duke@435: __ cmpl(rax, c_rarg1); // NaN? -> 0 duke@435: __ jcc(Assembler::negative, L); duke@435: __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong duke@435: __ mov64(c_rarg3, 0x8000000000000000); duke@435: __ mov64(rax, 0x7fffffffffffffff); never@739: __ cmov(Assembler::positive, c_rarg3, rax); duke@435: duke@435: __ bind(L); never@739: __ movptr(inout, c_rarg3); never@739: never@739: __ pop(c_rarg1); never@739: __ pop(c_rarg2); never@739: __ pop(c_rarg3); never@739: __ pop(rax); duke@435: duke@435: __ ret(0); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: address generate_d2i_fixup() { duke@435: StubCodeMark mark(this, "StubRoutines", "d2i_fixup"); duke@435: Address inout(rsp, 6 * wordSize); // return address + 5 saves duke@435: duke@435: address start = __ pc(); duke@435: duke@435: Label L; duke@435: never@739: __ push(rax); never@739: __ push(c_rarg3); never@739: __ push(c_rarg2); never@739: __ push(c_rarg1); never@739: __ push(c_rarg0); duke@435: duke@435: __ movl(rax, 0x7ff00000); duke@435: __ movq(c_rarg2, inout); duke@435: __ movl(c_rarg3, c_rarg2); never@739: __ mov(c_rarg1, c_rarg2); never@739: __ mov(c_rarg0, c_rarg2); duke@435: __ negl(c_rarg3); never@739: __ shrptr(c_rarg1, 0x20); duke@435: __ orl(c_rarg3, c_rarg2); duke@435: __ andl(c_rarg1, 0x7fffffff); duke@435: __ xorl(c_rarg2, c_rarg2); duke@435: __ shrl(c_rarg3, 0x1f); duke@435: __ orl(c_rarg1, c_rarg3); duke@435: __ cmpl(rax, c_rarg1); duke@435: __ jcc(Assembler::negative, L); // NaN -> 0 never@739: __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint duke@435: __ movl(c_rarg2, 0x80000000); duke@435: __ movl(rax, 0x7fffffff); never@739: __ cmov(Assembler::positive, c_rarg2, rax); duke@435: duke@435: __ bind(L); never@739: __ movptr(inout, c_rarg2); never@739: never@739: __ pop(c_rarg0); never@739: __ pop(c_rarg1); never@739: __ pop(c_rarg2); never@739: __ pop(c_rarg3); never@739: __ pop(rax); duke@435: duke@435: __ ret(0); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: address generate_d2l_fixup() { duke@435: StubCodeMark mark(this, "StubRoutines", "d2l_fixup"); duke@435: Address inout(rsp, 6 * wordSize); // return address + 5 saves duke@435: duke@435: address start = __ pc(); duke@435: duke@435: Label L; duke@435: never@739: __ push(rax); never@739: __ push(c_rarg3); never@739: __ push(c_rarg2); never@739: __ push(c_rarg1); never@739: __ push(c_rarg0); duke@435: duke@435: __ movl(rax, 0x7ff00000); duke@435: __ movq(c_rarg2, inout); duke@435: __ movl(c_rarg3, c_rarg2); never@739: __ mov(c_rarg1, c_rarg2); never@739: __ mov(c_rarg0, c_rarg2); duke@435: __ negl(c_rarg3); never@739: __ shrptr(c_rarg1, 0x20); duke@435: __ orl(c_rarg3, c_rarg2); duke@435: __ andl(c_rarg1, 0x7fffffff); duke@435: __ xorl(c_rarg2, c_rarg2); duke@435: __ shrl(c_rarg3, 0x1f); duke@435: __ orl(c_rarg1, c_rarg3); duke@435: __ cmpl(rax, c_rarg1); duke@435: __ jcc(Assembler::negative, L); // NaN -> 0 duke@435: __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong duke@435: __ mov64(c_rarg2, 0x8000000000000000); duke@435: __ mov64(rax, 0x7fffffffffffffff); duke@435: __ cmovq(Assembler::positive, c_rarg2, rax); duke@435: duke@435: __ bind(L); duke@435: __ movq(inout, c_rarg2); duke@435: never@739: __ pop(c_rarg0); never@739: __ pop(c_rarg1); never@739: __ pop(c_rarg2); never@739: __ pop(c_rarg3); never@739: __ pop(rax); duke@435: duke@435: __ ret(0); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: address generate_fp_mask(const char *stub_name, int64_t mask) { kvn@1800: __ align(CodeEntryAlignment); duke@435: StubCodeMark mark(this, "StubRoutines", stub_name); duke@435: address start = __ pc(); duke@435: duke@435: __ emit_data64( mask, relocInfo::none ); duke@435: __ emit_data64( mask, relocInfo::none ); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: // The following routine generates a subroutine to throw an duke@435: // asynchronous UnknownError when an unsafe access gets a fault that duke@435: // could not be reasonably prevented by the programmer. (Example: duke@435: // SIGBUS/OBJERR.) duke@435: address generate_handler_for_unsafe_access() { duke@435: StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access"); duke@435: address start = __ pc(); duke@435: never@739: __ push(0); // hole for return address-to-be never@739: __ pusha(); // push registers duke@435: Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord); duke@435: never@3136: // FIXME: this probably needs alignment logic never@3136: never@739: __ subptr(rsp, frame::arg_reg_save_area_bytes); duke@435: BLOCK_COMMENT("call handle_unsafe_access"); duke@435: __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access))); never@739: __ addptr(rsp, frame::arg_reg_save_area_bytes); never@739: never@739: __ movptr(next_pc, rax); // stuff next address never@739: __ popa(); duke@435: __ ret(0); // jump to next address duke@435: duke@435: return start; duke@435: } duke@435: duke@435: // Non-destructive plausibility checks for oops duke@435: // duke@435: // Arguments: duke@435: // all args on stack! duke@435: // duke@435: // Stack after saving c_rarg3: duke@435: // [tos + 0]: saved c_rarg3 duke@435: // [tos + 1]: saved c_rarg2 kvn@559: // [tos + 2]: saved r12 (several TemplateTable methods use it) kvn@559: // [tos + 3]: saved flags kvn@559: // [tos + 4]: return address kvn@559: // * [tos + 5]: error message (char*) kvn@559: // * [tos + 6]: object to verify (oop) kvn@559: // * [tos + 7]: saved rax - saved by caller and bashed kvn@1938: // * [tos + 8]: saved r10 (rscratch1) - saved by caller duke@435: // * = popped on exit duke@435: address generate_verify_oop() { duke@435: StubCodeMark mark(this, "StubRoutines", "verify_oop"); duke@435: address start = __ pc(); duke@435: duke@435: Label exit, error; duke@435: never@739: __ pushf(); duke@435: __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr())); duke@435: never@739: __ push(r12); kvn@559: duke@435: // save c_rarg2 and c_rarg3 never@739: __ push(c_rarg2); never@739: __ push(c_rarg3); duke@435: kvn@559: enum { kvn@559: // After previous pushes. kvn@559: oop_to_verify = 6 * wordSize, kvn@559: saved_rax = 7 * wordSize, kvn@1938: saved_r10 = 8 * wordSize, kvn@559: kvn@559: // Before the call to MacroAssembler::debug(), see below. kvn@559: return_addr = 16 * wordSize, kvn@559: error_msg = 17 * wordSize kvn@559: }; kvn@559: duke@435: // get object never@739: __ movptr(rax, Address(rsp, oop_to_verify)); duke@435: duke@435: // make sure object is 'reasonable' never@739: __ testptr(rax, rax); duke@435: __ jcc(Assembler::zero, exit); // if obj is NULL it is OK duke@435: // Check if the oop is in the right area of memory never@739: __ movptr(c_rarg2, rax); xlu@947: __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_mask()); never@739: __ andptr(c_rarg2, c_rarg3); xlu@947: __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_bits()); never@739: __ cmpptr(c_rarg2, c_rarg3); duke@435: __ jcc(Assembler::notZero, error); duke@435: kvn@559: // set r12 to heapbase for load_klass() kvn@559: __ reinit_heapbase(); kvn@559: coleenp@4037: // make sure klass is 'reasonable', which is not zero. coleenp@548: __ load_klass(rax, rax); // get klass never@739: __ testptr(rax, rax); duke@435: __ jcc(Assembler::zero, error); // if klass is NULL it is broken coleenp@4037: // TODO: Future assert that klass is lower 4g memory for UseCompressedKlassPointers duke@435: duke@435: // return if everything seems ok duke@435: __ bind(exit); never@739: __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back kvn@1938: __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back never@739: __ pop(c_rarg3); // restore c_rarg3 never@739: __ pop(c_rarg2); // restore c_rarg2 never@739: __ pop(r12); // restore r12 never@739: __ popf(); // restore flags kvn@1938: __ ret(4 * wordSize); // pop caller saved stuff duke@435: duke@435: // handle errors duke@435: __ bind(error); never@739: __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back kvn@1938: __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back never@739: __ pop(c_rarg3); // get saved c_rarg3 back never@739: __ pop(c_rarg2); // get saved c_rarg2 back never@739: __ pop(r12); // get saved r12 back never@739: __ popf(); // get saved flags off stack -- duke@435: // will be ignored duke@435: never@739: __ pusha(); // push registers duke@435: // (rip is already duke@435: // already pushed) kvn@559: // debug(char* msg, int64_t pc, int64_t regs[]) duke@435: // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and duke@435: // pushed all the registers, so now the stack looks like: duke@435: // [tos + 0] 16 saved registers duke@435: // [tos + 16] return address kvn@559: // * [tos + 17] error message (char*) kvn@559: // * [tos + 18] object to verify (oop) kvn@559: // * [tos + 19] saved rax - saved by caller and bashed kvn@1938: // * [tos + 20] saved r10 (rscratch1) - saved by caller kvn@559: // * = popped on exit kvn@559: never@739: __ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message never@739: __ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address never@739: __ movq(c_rarg2, rsp); // pass address of regs on stack never@739: __ mov(r12, rsp); // remember rsp never@739: __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows never@739: __ andptr(rsp, -16); // align stack as required by ABI duke@435: BLOCK_COMMENT("call MacroAssembler::debug"); never@739: __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64))); never@739: __ mov(rsp, r12); // restore rsp never@739: __ popa(); // pop registers (includes r12) kvn@1938: __ ret(4 * wordSize); // pop caller saved stuff duke@435: duke@435: return start; duke@435: } duke@435: duke@435: // duke@435: // Verify that a register contains clean 32-bits positive value duke@435: // (high 32-bits are 0) so it could be used in 64-bits shifts. duke@435: // duke@435: // Input: duke@435: // Rint - 32-bits value duke@435: // Rtmp - scratch duke@435: // duke@435: void assert_clean_int(Register Rint, Register Rtmp) { duke@435: #ifdef ASSERT duke@435: Label L; duke@435: assert_different_registers(Rtmp, Rint); duke@435: __ movslq(Rtmp, Rint); duke@435: __ cmpq(Rtmp, Rint); kvn@559: __ jcc(Assembler::equal, L); duke@435: __ stop("high 32-bits of int value are not 0"); duke@435: __ bind(L); duke@435: #endif duke@435: } duke@435: duke@435: // Generate overlap test for array copy stubs duke@435: // duke@435: // Input: duke@435: // c_rarg0 - from duke@435: // c_rarg1 - to duke@435: // c_rarg2 - element count duke@435: // duke@435: // Output: duke@435: // rax - &from[element count - 1] duke@435: // duke@435: void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) { duke@435: assert(no_overlap_target != NULL, "must be generated"); duke@435: array_overlap_test(no_overlap_target, NULL, sf); duke@435: } duke@435: void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) { duke@435: array_overlap_test(NULL, &L_no_overlap, sf); duke@435: } duke@435: void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) { duke@435: const Register from = c_rarg0; duke@435: const Register to = c_rarg1; duke@435: const Register count = c_rarg2; duke@435: const Register end_from = rax; duke@435: never@739: __ cmpptr(to, from); never@739: __ lea(end_from, Address(from, count, sf, 0)); duke@435: if (NOLp == NULL) { duke@435: ExternalAddress no_overlap(no_overlap_target); duke@435: __ jump_cc(Assembler::belowEqual, no_overlap); never@739: __ cmpptr(to, end_from); duke@435: __ jump_cc(Assembler::aboveEqual, no_overlap); duke@435: } else { duke@435: __ jcc(Assembler::belowEqual, (*NOLp)); never@739: __ cmpptr(to, end_from); duke@435: __ jcc(Assembler::aboveEqual, (*NOLp)); duke@435: } duke@435: } duke@435: duke@435: // Shuffle first three arg regs on Windows into Linux/Solaris locations. duke@435: // duke@435: // Outputs: duke@435: // rdi - rcx duke@435: // rsi - rdx duke@435: // rdx - r8 duke@435: // rcx - r9 duke@435: // duke@435: // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter duke@435: // are non-volatile. r9 and r10 should not be used by the caller. duke@435: // duke@435: void setup_arg_regs(int nargs = 3) { duke@435: const Register saved_rdi = r9; duke@435: const Register saved_rsi = r10; duke@435: assert(nargs == 3 || nargs == 4, "else fix"); duke@435: #ifdef _WIN64 duke@435: assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9, duke@435: "unexpected argument registers"); duke@435: if (nargs >= 4) never@739: __ mov(rax, r9); // r9 is also saved_rdi never@739: __ movptr(saved_rdi, rdi); never@739: __ movptr(saved_rsi, rsi); never@739: __ mov(rdi, rcx); // c_rarg0 never@739: __ mov(rsi, rdx); // c_rarg1 never@739: __ mov(rdx, r8); // c_rarg2 duke@435: if (nargs >= 4) never@739: __ mov(rcx, rax); // c_rarg3 (via rax) duke@435: #else duke@435: assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx, duke@435: "unexpected argument registers"); duke@435: #endif duke@435: } duke@435: duke@435: void restore_arg_regs() { duke@435: const Register saved_rdi = r9; duke@435: const Register saved_rsi = r10; duke@435: #ifdef _WIN64 never@739: __ movptr(rdi, saved_rdi); never@739: __ movptr(rsi, saved_rsi); duke@435: #endif duke@435: } duke@435: duke@435: // Generate code for an array write pre barrier duke@435: // duke@435: // addr - starting address iveresov@2606: // count - element count iveresov@2606: // tmp - scratch register duke@435: // duke@435: // Destroy no registers! duke@435: // iveresov@2606: void gen_write_ref_array_pre_barrier(Register addr, Register count, bool dest_uninitialized) { duke@435: BarrierSet* bs = Universe::heap()->barrier_set(); duke@435: switch (bs->kind()) { duke@435: case BarrierSet::G1SATBCT: duke@435: case BarrierSet::G1SATBCTLogging: iveresov@2606: // With G1, don't generate the call if we statically know that the target in uninitialized iveresov@2606: if (!dest_uninitialized) { iveresov@2606: __ pusha(); // push registers iveresov@2606: if (count == c_rarg0) { iveresov@2606: if (addr == c_rarg1) { iveresov@2606: // exactly backwards!! iveresov@2606: __ xchgptr(c_rarg1, c_rarg0); iveresov@2606: } else { iveresov@2606: __ movptr(c_rarg1, count); iveresov@2606: __ movptr(c_rarg0, addr); iveresov@2606: } iveresov@2606: } else { iveresov@2606: __ movptr(c_rarg0, addr); iveresov@2606: __ movptr(c_rarg1, count); iveresov@2606: } iveresov@2606: __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre), 2); iveresov@2606: __ popa(); duke@435: } iveresov@2606: break; duke@435: case BarrierSet::CardTableModRef: duke@435: case BarrierSet::CardTableExtension: duke@435: case BarrierSet::ModRef: duke@435: break; ysr@777: default: duke@435: ShouldNotReachHere(); duke@435: duke@435: } duke@435: } duke@435: duke@435: // duke@435: // Generate code for an array write post barrier duke@435: // duke@435: // Input: duke@435: // start - register containing starting address of destination array duke@435: // end - register containing ending address of destination array duke@435: // scratch - scratch register duke@435: // duke@435: // The input registers are overwritten. duke@435: // The ending address is inclusive. duke@435: void gen_write_ref_array_post_barrier(Register start, Register end, Register scratch) { duke@435: assert_different_registers(start, end, scratch); duke@435: BarrierSet* bs = Universe::heap()->barrier_set(); duke@435: switch (bs->kind()) { duke@435: case BarrierSet::G1SATBCT: duke@435: case BarrierSet::G1SATBCTLogging: duke@435: duke@435: { never@739: __ pusha(); // push registers (overkill) duke@435: // must compute element count unless barrier set interface is changed (other platforms supply count) duke@435: assert_different_registers(start, end, scratch); ysr@1280: __ lea(scratch, Address(end, BytesPerHeapOop)); ysr@1280: __ subptr(scratch, start); // subtract start to get #bytes ysr@1280: __ shrptr(scratch, LogBytesPerHeapOop); // convert to element count never@739: __ mov(c_rarg0, start); never@739: __ mov(c_rarg1, scratch); apetrusenko@1627: __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post), 2); never@739: __ popa(); duke@435: } duke@435: break; duke@435: case BarrierSet::CardTableModRef: duke@435: case BarrierSet::CardTableExtension: duke@435: { duke@435: CardTableModRefBS* ct = (CardTableModRefBS*)bs; duke@435: assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code"); duke@435: duke@435: Label L_loop; duke@435: never@739: __ shrptr(start, CardTableModRefBS::card_shift); ysr@1280: __ addptr(end, BytesPerHeapOop); never@739: __ shrptr(end, CardTableModRefBS::card_shift); never@739: __ subptr(end, start); // number of bytes to copy duke@435: never@684: intptr_t disp = (intptr_t) ct->byte_map_base; twisti@3310: if (Assembler::is_simm32(disp)) { never@684: Address cardtable(noreg, noreg, Address::no_scale, disp); never@684: __ lea(scratch, cardtable); never@684: } else { never@684: ExternalAddress cardtable((address)disp); never@684: __ lea(scratch, cardtable); never@684: } never@684: duke@435: const Register count = end; // 'end' register contains bytes count now never@739: __ addptr(start, scratch); duke@435: __ BIND(L_loop); duke@435: __ movb(Address(start, count, Address::times_1), 0); never@739: __ decrement(count); duke@435: __ jcc(Assembler::greaterEqual, L_loop); duke@435: } ysr@777: break; ysr@777: default: ysr@777: ShouldNotReachHere(); ysr@777: ysr@777: } ysr@777: } duke@435: kvn@840: duke@435: // Copy big chunks forward duke@435: // duke@435: // Inputs: duke@435: // end_from - source arrays end address duke@435: // end_to - destination array end address duke@435: // qword_count - 64-bits element count, negative duke@435: // to - scratch kvn@4411: // L_copy_bytes - entry label duke@435: // L_copy_8_bytes - exit label duke@435: // kvn@4411: void copy_bytes_forward(Register end_from, Register end_to, duke@435: Register qword_count, Register to, kvn@4411: Label& L_copy_bytes, Label& L_copy_8_bytes) { duke@435: DEBUG_ONLY(__ stop("enter at entry label, not here")); duke@435: Label L_loop; kvn@1800: __ align(OptoLoopAlignment); kvn@4411: if (UseUnalignedLoadStores) { kvn@4411: Label L_end; kvn@4411: // Copy 64-bytes per iteration kvn@4411: __ BIND(L_loop); kvn@4411: if (UseAVX >= 2) { kvn@4411: __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56)); kvn@4411: __ vmovdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0); kvn@4411: __ vmovdqu(xmm1, Address(end_from, qword_count, Address::times_8, -24)); kvn@4411: __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm1); kvn@4411: } else { kvn@4411: __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56)); kvn@4411: __ movdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0); kvn@4411: __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, -40)); kvn@4411: __ movdqu(Address(end_to, qword_count, Address::times_8, -40), xmm1); kvn@4411: __ movdqu(xmm2, Address(end_from, qword_count, Address::times_8, -24)); kvn@4411: __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm2); kvn@4411: __ movdqu(xmm3, Address(end_from, qword_count, Address::times_8, - 8)); kvn@4411: __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm3); kvn@4411: } kvn@4411: __ BIND(L_copy_bytes); kvn@4411: __ addptr(qword_count, 8); kvn@4411: __ jcc(Assembler::lessEqual, L_loop); kvn@4411: __ subptr(qword_count, 4); // sub(8) and add(4) kvn@4411: __ jccb(Assembler::greater, L_end); kvn@4411: // Copy trailing 32 bytes kvn@4411: if (UseAVX >= 2) { kvn@4411: __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24)); kvn@4411: __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0); kvn@4411: } else { kvn@4411: __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24)); kvn@4411: __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0); kvn@4411: __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, - 8)); kvn@4411: __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm1); kvn@4411: } kvn@4411: __ addptr(qword_count, 4); kvn@4411: __ BIND(L_end); kvn@4873: if (UseAVX >= 2) { kvn@4873: // clean upper bits of YMM registers kvn@4873: __ vzeroupper(); kvn@4873: } kvn@840: } else { kvn@4411: // Copy 32-bytes per iteration kvn@4411: __ BIND(L_loop); kvn@840: __ movq(to, Address(end_from, qword_count, Address::times_8, -24)); kvn@840: __ movq(Address(end_to, qword_count, Address::times_8, -24), to); kvn@840: __ movq(to, Address(end_from, qword_count, Address::times_8, -16)); kvn@840: __ movq(Address(end_to, qword_count, Address::times_8, -16), to); kvn@840: __ movq(to, Address(end_from, qword_count, Address::times_8, - 8)); kvn@840: __ movq(Address(end_to, qword_count, Address::times_8, - 8), to); kvn@840: __ movq(to, Address(end_from, qword_count, Address::times_8, - 0)); kvn@840: __ movq(Address(end_to, qword_count, Address::times_8, - 0), to); kvn@4411: kvn@4411: __ BIND(L_copy_bytes); kvn@4411: __ addptr(qword_count, 4); kvn@4411: __ jcc(Assembler::lessEqual, L_loop); kvn@840: } never@739: __ subptr(qword_count, 4); duke@435: __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords duke@435: } duke@435: duke@435: // Copy big chunks backward duke@435: // duke@435: // Inputs: duke@435: // from - source arrays address duke@435: // dest - destination array address duke@435: // qword_count - 64-bits element count duke@435: // to - scratch kvn@4411: // L_copy_bytes - entry label duke@435: // L_copy_8_bytes - exit label duke@435: // kvn@4411: void copy_bytes_backward(Register from, Register dest, duke@435: Register qword_count, Register to, kvn@4411: Label& L_copy_bytes, Label& L_copy_8_bytes) { duke@435: DEBUG_ONLY(__ stop("enter at entry label, not here")); duke@435: Label L_loop; kvn@1800: __ align(OptoLoopAlignment); kvn@4411: if (UseUnalignedLoadStores) { kvn@4411: Label L_end; kvn@4411: // Copy 64-bytes per iteration kvn@4411: __ BIND(L_loop); kvn@4411: if (UseAVX >= 2) { kvn@4411: __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 32)); kvn@4411: __ vmovdqu(Address(dest, qword_count, Address::times_8, 32), xmm0); kvn@4411: __ vmovdqu(xmm1, Address(from, qword_count, Address::times_8, 0)); kvn@4411: __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm1); kvn@4411: } else { kvn@4411: __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 48)); kvn@4411: __ movdqu(Address(dest, qword_count, Address::times_8, 48), xmm0); kvn@4411: __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 32)); kvn@4411: __ movdqu(Address(dest, qword_count, Address::times_8, 32), xmm1); kvn@4411: __ movdqu(xmm2, Address(from, qword_count, Address::times_8, 16)); kvn@4411: __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm2); kvn@4411: __ movdqu(xmm3, Address(from, qword_count, Address::times_8, 0)); kvn@4411: __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm3); kvn@4411: } kvn@4411: __ BIND(L_copy_bytes); kvn@4411: __ subptr(qword_count, 8); kvn@4411: __ jcc(Assembler::greaterEqual, L_loop); kvn@4411: kvn@4411: __ addptr(qword_count, 4); // add(8) and sub(4) kvn@4411: __ jccb(Assembler::less, L_end); kvn@4411: // Copy trailing 32 bytes kvn@4411: if (UseAVX >= 2) { kvn@4411: __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 0)); kvn@4411: __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm0); kvn@4411: } else { kvn@4411: __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 16)); kvn@4411: __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm0); kvn@4411: __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 0)); kvn@4411: __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm1); kvn@4411: } kvn@4411: __ subptr(qword_count, 4); kvn@4411: __ BIND(L_end); kvn@4873: if (UseAVX >= 2) { kvn@4873: // clean upper bits of YMM registers kvn@4873: __ vzeroupper(); kvn@4873: } kvn@840: } else { kvn@4411: // Copy 32-bytes per iteration kvn@4411: __ BIND(L_loop); kvn@840: __ movq(to, Address(from, qword_count, Address::times_8, 24)); kvn@840: __ movq(Address(dest, qword_count, Address::times_8, 24), to); kvn@840: __ movq(to, Address(from, qword_count, Address::times_8, 16)); kvn@840: __ movq(Address(dest, qword_count, Address::times_8, 16), to); kvn@840: __ movq(to, Address(from, qword_count, Address::times_8, 8)); kvn@840: __ movq(Address(dest, qword_count, Address::times_8, 8), to); kvn@840: __ movq(to, Address(from, qword_count, Address::times_8, 0)); kvn@840: __ movq(Address(dest, qword_count, Address::times_8, 0), to); kvn@4411: kvn@4411: __ BIND(L_copy_bytes); kvn@4411: __ subptr(qword_count, 4); kvn@4411: __ jcc(Assembler::greaterEqual, L_loop); kvn@840: } never@739: __ addptr(qword_count, 4); duke@435: __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords duke@435: } duke@435: duke@435: duke@435: // Arguments: duke@435: // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary duke@435: // ignored duke@435: // name - stub name string duke@435: // duke@435: // Inputs: duke@435: // c_rarg0 - source array address duke@435: // c_rarg1 - destination array address duke@435: // c_rarg2 - element count, treated as ssize_t, can be zero duke@435: // duke@435: // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries, duke@435: // we let the hardware handle it. The one to eight bytes within words, duke@435: // dwords or qwords that span cache line boundaries will still be loaded duke@435: // and stored atomically. duke@435: // duke@435: // Side Effects: duke@435: // disjoint_byte_copy_entry is set to the no-overlap entry point duke@435: // used by generate_conjoint_byte_copy(). duke@435: // iveresov@2595: address generate_disjoint_byte_copy(bool aligned, address* entry, const char *name) { duke@435: __ align(CodeEntryAlignment); duke@435: StubCodeMark mark(this, "StubRoutines", name); duke@435: address start = __ pc(); duke@435: kvn@4411: Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes; duke@435: Label L_copy_byte, L_exit; duke@435: const Register from = rdi; // source array address duke@435: const Register to = rsi; // destination array address duke@435: const Register count = rdx; // elements count duke@435: const Register byte_count = rcx; duke@435: const Register qword_count = count; duke@435: const Register end_from = from; // source array end address duke@435: const Register end_to = to; // destination array end address duke@435: // End pointers are inclusive, and if count is not zero they point duke@435: // to the last unit copied: end_to[0] := end_from[0] duke@435: duke@435: __ enter(); // required for proper stackwalking of RuntimeStub frame duke@435: assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. duke@435: iveresov@2595: if (entry != NULL) { iveresov@2595: *entry = __ pc(); iveresov@2595: // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) iveresov@2595: BLOCK_COMMENT("Entry:"); iveresov@2595: } duke@435: duke@435: setup_arg_regs(); // from => rdi, to => rsi, count => rdx duke@435: // r9 and r10 may be used to save non-volatile registers duke@435: duke@435: // 'from', 'to' and 'count' are now valid never@739: __ movptr(byte_count, count); never@739: __ shrptr(count, 3); // count => qword_count duke@435: duke@435: // Copy from low to high addresses. Use 'to' as scratch. never@739: __ lea(end_from, Address(from, qword_count, Address::times_8, -8)); never@739: __ lea(end_to, Address(to, qword_count, Address::times_8, -8)); never@739: __ negptr(qword_count); // make the count negative kvn@4411: __ jmp(L_copy_bytes); duke@435: duke@435: // Copy trailing qwords duke@435: __ BIND(L_copy_8_bytes); duke@435: __ movq(rax, Address(end_from, qword_count, Address::times_8, 8)); duke@435: __ movq(Address(end_to, qword_count, Address::times_8, 8), rax); never@739: __ increment(qword_count); duke@435: __ jcc(Assembler::notZero, L_copy_8_bytes); duke@435: duke@435: // Check for and copy trailing dword duke@435: __ BIND(L_copy_4_bytes); never@739: __ testl(byte_count, 4); duke@435: __ jccb(Assembler::zero, L_copy_2_bytes); duke@435: __ movl(rax, Address(end_from, 8)); duke@435: __ movl(Address(end_to, 8), rax); duke@435: never@739: __ addptr(end_from, 4); never@739: __ addptr(end_to, 4); duke@435: duke@435: // Check for and copy trailing word duke@435: __ BIND(L_copy_2_bytes); never@739: __ testl(byte_count, 2); duke@435: __ jccb(Assembler::zero, L_copy_byte); duke@435: __ movw(rax, Address(end_from, 8)); duke@435: __ movw(Address(end_to, 8), rax); duke@435: never@739: __ addptr(end_from, 2); never@739: __ addptr(end_to, 2); duke@435: duke@435: // Check for and copy trailing byte duke@435: __ BIND(L_copy_byte); never@739: __ testl(byte_count, 1); duke@435: __ jccb(Assembler::zero, L_exit); duke@435: __ movb(rax, Address(end_from, 8)); duke@435: __ movb(Address(end_to, 8), rax); duke@435: duke@435: __ BIND(L_exit); duke@435: restore_arg_regs(); never@3314: inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free never@739: __ xorptr(rax, rax); // return 0 duke@435: __ leave(); // required for proper stackwalking of RuntimeStub frame duke@435: __ ret(0); duke@435: kvn@4411: // Copy in multi-bytes chunks kvn@4411: copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes); duke@435: __ jmp(L_copy_4_bytes); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: // Arguments: duke@435: // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary duke@435: // ignored duke@435: // name - stub name string duke@435: // duke@435: // Inputs: duke@435: // c_rarg0 - source array address duke@435: // c_rarg1 - destination array address duke@435: // c_rarg2 - element count, treated as ssize_t, can be zero duke@435: // duke@435: // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries, duke@435: // we let the hardware handle it. The one to eight bytes within words, duke@435: // dwords or qwords that span cache line boundaries will still be loaded duke@435: // and stored atomically. duke@435: // iveresov@2595: address generate_conjoint_byte_copy(bool aligned, address nooverlap_target, iveresov@2595: address* entry, const char *name) { duke@435: __ align(CodeEntryAlignment); duke@435: StubCodeMark mark(this, "StubRoutines", name); duke@435: address start = __ pc(); duke@435: kvn@4411: Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes; duke@435: const Register from = rdi; // source array address duke@435: const Register to = rsi; // destination array address duke@435: const Register count = rdx; // elements count duke@435: const Register byte_count = rcx; duke@435: const Register qword_count = count; duke@435: duke@435: __ enter(); // required for proper stackwalking of RuntimeStub frame duke@435: assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. duke@435: iveresov@2595: if (entry != NULL) { iveresov@2595: *entry = __ pc(); iveresov@2595: // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) iveresov@2595: BLOCK_COMMENT("Entry:"); iveresov@2595: } iveresov@2595: iveresov@2595: array_overlap_test(nooverlap_target, Address::times_1); duke@435: setup_arg_regs(); // from => rdi, to => rsi, count => rdx duke@435: // r9 and r10 may be used to save non-volatile registers duke@435: duke@435: // 'from', 'to' and 'count' are now valid never@739: __ movptr(byte_count, count); never@739: __ shrptr(count, 3); // count => qword_count duke@435: duke@435: // Copy from high to low addresses. duke@435: duke@435: // Check for and copy trailing byte never@739: __ testl(byte_count, 1); duke@435: __ jcc(Assembler::zero, L_copy_2_bytes); duke@435: __ movb(rax, Address(from, byte_count, Address::times_1, -1)); duke@435: __ movb(Address(to, byte_count, Address::times_1, -1), rax); never@739: __ decrement(byte_count); // Adjust for possible trailing word duke@435: duke@435: // Check for and copy trailing word duke@435: __ BIND(L_copy_2_bytes); never@739: __ testl(byte_count, 2); duke@435: __ jcc(Assembler::zero, L_copy_4_bytes); duke@435: __ movw(rax, Address(from, byte_count, Address::times_1, -2)); duke@435: __ movw(Address(to, byte_count, Address::times_1, -2), rax); duke@435: duke@435: // Check for and copy trailing dword duke@435: __ BIND(L_copy_4_bytes); never@739: __ testl(byte_count, 4); kvn@4411: __ jcc(Assembler::zero, L_copy_bytes); duke@435: __ movl(rax, Address(from, qword_count, Address::times_8)); duke@435: __ movl(Address(to, qword_count, Address::times_8), rax); kvn@4411: __ jmp(L_copy_bytes); duke@435: duke@435: // Copy trailing qwords duke@435: __ BIND(L_copy_8_bytes); duke@435: __ movq(rax, Address(from, qword_count, Address::times_8, -8)); duke@435: __ movq(Address(to, qword_count, Address::times_8, -8), rax); never@739: __ decrement(qword_count); duke@435: __ jcc(Assembler::notZero, L_copy_8_bytes); duke@435: duke@435: restore_arg_regs(); never@3314: inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free never@739: __ xorptr(rax, rax); // return 0 duke@435: __ leave(); // required for proper stackwalking of RuntimeStub frame duke@435: __ ret(0); duke@435: kvn@4411: // Copy in multi-bytes chunks kvn@4411: copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes); duke@435: duke@435: restore_arg_regs(); never@3314: inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free never@739: __ xorptr(rax, rax); // return 0 duke@435: __ leave(); // required for proper stackwalking of RuntimeStub frame duke@435: __ ret(0); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: // Arguments: duke@435: // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary duke@435: // ignored duke@435: // name - stub name string duke@435: // duke@435: // Inputs: duke@435: // c_rarg0 - source array address duke@435: // c_rarg1 - destination array address duke@435: // c_rarg2 - element count, treated as ssize_t, can be zero duke@435: // duke@435: // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we duke@435: // let the hardware handle it. The two or four words within dwords duke@435: // or qwords that span cache line boundaries will still be loaded duke@435: // and stored atomically. duke@435: // duke@435: // Side Effects: duke@435: // disjoint_short_copy_entry is set to the no-overlap entry point duke@435: // used by generate_conjoint_short_copy(). duke@435: // iveresov@2595: address generate_disjoint_short_copy(bool aligned, address *entry, const char *name) { duke@435: __ align(CodeEntryAlignment); duke@435: StubCodeMark mark(this, "StubRoutines", name); duke@435: address start = __ pc(); duke@435: kvn@4411: Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit; duke@435: const Register from = rdi; // source array address duke@435: const Register to = rsi; // destination array address duke@435: const Register count = rdx; // elements count duke@435: const Register word_count = rcx; duke@435: const Register qword_count = count; duke@435: const Register end_from = from; // source array end address duke@435: const Register end_to = to; // destination array end address duke@435: // End pointers are inclusive, and if count is not zero they point duke@435: // to the last unit copied: end_to[0] := end_from[0] duke@435: duke@435: __ enter(); // required for proper stackwalking of RuntimeStub frame duke@435: assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. duke@435: iveresov@2595: if (entry != NULL) { iveresov@2595: *entry = __ pc(); iveresov@2595: // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) iveresov@2595: BLOCK_COMMENT("Entry:"); iveresov@2595: } duke@435: duke@435: setup_arg_regs(); // from => rdi, to => rsi, count => rdx duke@435: // r9 and r10 may be used to save non-volatile registers duke@435: duke@435: // 'from', 'to' and 'count' are now valid never@739: __ movptr(word_count, count); never@739: __ shrptr(count, 2); // count => qword_count duke@435: duke@435: // Copy from low to high addresses. Use 'to' as scratch. never@739: __ lea(end_from, Address(from, qword_count, Address::times_8, -8)); never@739: __ lea(end_to, Address(to, qword_count, Address::times_8, -8)); never@739: __ negptr(qword_count); kvn@4411: __ jmp(L_copy_bytes); duke@435: duke@435: // Copy trailing qwords duke@435: __ BIND(L_copy_8_bytes); duke@435: __ movq(rax, Address(end_from, qword_count, Address::times_8, 8)); duke@435: __ movq(Address(end_to, qword_count, Address::times_8, 8), rax); never@739: __ increment(qword_count); duke@435: __ jcc(Assembler::notZero, L_copy_8_bytes); duke@435: duke@435: // Original 'dest' is trashed, so we can't use it as a duke@435: // base register for a possible trailing word copy duke@435: duke@435: // Check for and copy trailing dword duke@435: __ BIND(L_copy_4_bytes); never@739: __ testl(word_count, 2); duke@435: __ jccb(Assembler::zero, L_copy_2_bytes); duke@435: __ movl(rax, Address(end_from, 8)); duke@435: __ movl(Address(end_to, 8), rax); duke@435: never@739: __ addptr(end_from, 4); never@739: __ addptr(end_to, 4); duke@435: duke@435: // Check for and copy trailing word duke@435: __ BIND(L_copy_2_bytes); never@739: __ testl(word_count, 1); duke@435: __ jccb(Assembler::zero, L_exit); duke@435: __ movw(rax, Address(end_from, 8)); duke@435: __ movw(Address(end_to, 8), rax); duke@435: duke@435: __ BIND(L_exit); duke@435: restore_arg_regs(); never@3314: inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free never@739: __ xorptr(rax, rax); // return 0 duke@435: __ leave(); // required for proper stackwalking of RuntimeStub frame duke@435: __ ret(0); duke@435: kvn@4411: // Copy in multi-bytes chunks kvn@4411: copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes); duke@435: __ jmp(L_copy_4_bytes); duke@435: duke@435: return start; duke@435: } duke@435: never@2118: address generate_fill(BasicType t, bool aligned, const char *name) { never@2118: __ align(CodeEntryAlignment); never@2118: StubCodeMark mark(this, "StubRoutines", name); never@2118: address start = __ pc(); never@2118: never@2118: BLOCK_COMMENT("Entry:"); never@2118: never@2118: const Register to = c_rarg0; // source array address never@2118: const Register value = c_rarg1; // value never@2118: const Register count = c_rarg2; // elements count never@2118: never@2118: __ enter(); // required for proper stackwalking of RuntimeStub frame never@2118: never@2118: __ generate_fill(t, aligned, to, value, count, rax, xmm0); never@2118: never@2118: __ leave(); // required for proper stackwalking of RuntimeStub frame never@2118: __ ret(0); never@2118: return start; never@2118: } never@2118: duke@435: // Arguments: duke@435: // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary duke@435: // ignored duke@435: // name - stub name string duke@435: // duke@435: // Inputs: duke@435: // c_rarg0 - source array address duke@435: // c_rarg1 - destination array address duke@435: // c_rarg2 - element count, treated as ssize_t, can be zero duke@435: // duke@435: // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we duke@435: // let the hardware handle it. The two or four words within dwords duke@435: // or qwords that span cache line boundaries will still be loaded duke@435: // and stored atomically. duke@435: // iveresov@2595: address generate_conjoint_short_copy(bool aligned, address nooverlap_target, iveresov@2595: address *entry, const char *name) { duke@435: __ align(CodeEntryAlignment); duke@435: StubCodeMark mark(this, "StubRoutines", name); duke@435: address start = __ pc(); duke@435: kvn@4411: Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes; duke@435: const Register from = rdi; // source array address duke@435: const Register to = rsi; // destination array address duke@435: const Register count = rdx; // elements count duke@435: const Register word_count = rcx; duke@435: const Register qword_count = count; duke@435: duke@435: __ enter(); // required for proper stackwalking of RuntimeStub frame duke@435: assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. duke@435: iveresov@2595: if (entry != NULL) { iveresov@2595: *entry = __ pc(); iveresov@2595: // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) iveresov@2595: BLOCK_COMMENT("Entry:"); iveresov@2595: } iveresov@2595: iveresov@2595: array_overlap_test(nooverlap_target, Address::times_2); duke@435: setup_arg_regs(); // from => rdi, to => rsi, count => rdx duke@435: // r9 and r10 may be used to save non-volatile registers duke@435: duke@435: // 'from', 'to' and 'count' are now valid never@739: __ movptr(word_count, count); never@739: __ shrptr(count, 2); // count => qword_count duke@435: duke@435: // Copy from high to low addresses. Use 'to' as scratch. duke@435: duke@435: // Check for and copy trailing word never@739: __ testl(word_count, 1); duke@435: __ jccb(Assembler::zero, L_copy_4_bytes); duke@435: __ movw(rax, Address(from, word_count, Address::times_2, -2)); duke@435: __ movw(Address(to, word_count, Address::times_2, -2), rax); duke@435: duke@435: // Check for and copy trailing dword duke@435: __ BIND(L_copy_4_bytes); never@739: __ testl(word_count, 2); kvn@4411: __ jcc(Assembler::zero, L_copy_bytes); duke@435: __ movl(rax, Address(from, qword_count, Address::times_8)); duke@435: __ movl(Address(to, qword_count, Address::times_8), rax); kvn@4411: __ jmp(L_copy_bytes); duke@435: duke@435: // Copy trailing qwords duke@435: __ BIND(L_copy_8_bytes); duke@435: __ movq(rax, Address(from, qword_count, Address::times_8, -8)); duke@435: __ movq(Address(to, qword_count, Address::times_8, -8), rax); never@739: __ decrement(qword_count); duke@435: __ jcc(Assembler::notZero, L_copy_8_bytes); duke@435: duke@435: restore_arg_regs(); never@3314: inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free never@739: __ xorptr(rax, rax); // return 0 duke@435: __ leave(); // required for proper stackwalking of RuntimeStub frame duke@435: __ ret(0); duke@435: kvn@4411: // Copy in multi-bytes chunks kvn@4411: copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes); duke@435: duke@435: restore_arg_regs(); never@3314: inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free never@739: __ xorptr(rax, rax); // return 0 duke@435: __ leave(); // required for proper stackwalking of RuntimeStub frame duke@435: __ ret(0); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: // Arguments: duke@435: // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary duke@435: // ignored coleenp@548: // is_oop - true => oop array, so generate store check code duke@435: // name - stub name string duke@435: // duke@435: // Inputs: duke@435: // c_rarg0 - source array address duke@435: // c_rarg1 - destination array address duke@435: // c_rarg2 - element count, treated as ssize_t, can be zero duke@435: // duke@435: // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let duke@435: // the hardware handle it. The two dwords within qwords that span duke@435: // cache line boundaries will still be loaded and stored atomicly. duke@435: // duke@435: // Side Effects: duke@435: // disjoint_int_copy_entry is set to the no-overlap entry point coleenp@548: // used by generate_conjoint_int_oop_copy(). duke@435: // iveresov@2606: address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, address* entry, iveresov@2606: const char *name, bool dest_uninitialized = false) { duke@435: __ align(CodeEntryAlignment); duke@435: StubCodeMark mark(this, "StubRoutines", name); duke@435: address start = __ pc(); duke@435: kvn@4411: Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit; duke@435: const Register from = rdi; // source array address duke@435: const Register to = rsi; // destination array address duke@435: const Register count = rdx; // elements count duke@435: const Register dword_count = rcx; duke@435: const Register qword_count = count; duke@435: const Register end_from = from; // source array end address duke@435: const Register end_to = to; // destination array end address coleenp@548: const Register saved_to = r11; // saved destination array address duke@435: // End pointers are inclusive, and if count is not zero they point duke@435: // to the last unit copied: end_to[0] := end_from[0] duke@435: duke@435: __ enter(); // required for proper stackwalking of RuntimeStub frame duke@435: assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. duke@435: iveresov@2595: if (entry != NULL) { iveresov@2595: *entry = __ pc(); iveresov@2595: // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) iveresov@2595: BLOCK_COMMENT("Entry:"); coleenp@548: } coleenp@548: duke@435: setup_arg_regs(); // from => rdi, to => rsi, count => rdx duke@435: // r9 and r10 may be used to save non-volatile registers coleenp@548: if (is_oop) { coleenp@548: __ movq(saved_to, to); iveresov@2606: gen_write_ref_array_pre_barrier(to, count, dest_uninitialized); coleenp@548: } coleenp@548: duke@435: // 'from', 'to' and 'count' are now valid never@739: __ movptr(dword_count, count); never@739: __ shrptr(count, 1); // count => qword_count duke@435: duke@435: // Copy from low to high addresses. Use 'to' as scratch. never@739: __ lea(end_from, Address(from, qword_count, Address::times_8, -8)); never@739: __ lea(end_to, Address(to, qword_count, Address::times_8, -8)); never@739: __ negptr(qword_count); kvn@4411: __ jmp(L_copy_bytes); duke@435: duke@435: // Copy trailing qwords duke@435: __ BIND(L_copy_8_bytes); duke@435: __ movq(rax, Address(end_from, qword_count, Address::times_8, 8)); duke@435: __ movq(Address(end_to, qword_count, Address::times_8, 8), rax); never@739: __ increment(qword_count); duke@435: __ jcc(Assembler::notZero, L_copy_8_bytes); duke@435: duke@435: // Check for and copy trailing dword duke@435: __ BIND(L_copy_4_bytes); never@739: __ testl(dword_count, 1); // Only byte test since the value is 0 or 1 duke@435: __ jccb(Assembler::zero, L_exit); duke@435: __ movl(rax, Address(end_from, 8)); duke@435: __ movl(Address(end_to, 8), rax); duke@435: duke@435: __ BIND(L_exit); coleenp@548: if (is_oop) { coleenp@548: __ leaq(end_to, Address(saved_to, dword_count, Address::times_4, -4)); coleenp@548: gen_write_ref_array_post_barrier(saved_to, end_to, rax); coleenp@548: } duke@435: restore_arg_regs(); never@3314: inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free never@739: __ xorptr(rax, rax); // return 0 duke@435: __ leave(); // required for proper stackwalking of RuntimeStub frame duke@435: __ ret(0); duke@435: kvn@4411: // Copy in multi-bytes chunks kvn@4411: copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes); duke@435: __ jmp(L_copy_4_bytes); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: // Arguments: duke@435: // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary duke@435: // ignored coleenp@548: // is_oop - true => oop array, so generate store check code duke@435: // name - stub name string duke@435: // duke@435: // Inputs: duke@435: // c_rarg0 - source array address duke@435: // c_rarg1 - destination array address duke@435: // c_rarg2 - element count, treated as ssize_t, can be zero duke@435: // duke@435: // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let duke@435: // the hardware handle it. The two dwords within qwords that span duke@435: // cache line boundaries will still be loaded and stored atomicly. duke@435: // iveresov@2595: address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, address nooverlap_target, iveresov@2606: address *entry, const char *name, iveresov@2606: bool dest_uninitialized = false) { duke@435: __ align(CodeEntryAlignment); duke@435: StubCodeMark mark(this, "StubRoutines", name); duke@435: address start = __ pc(); duke@435: kvn@4411: Label L_copy_bytes, L_copy_8_bytes, L_copy_2_bytes, L_exit; duke@435: const Register from = rdi; // source array address duke@435: const Register to = rsi; // destination array address duke@435: const Register count = rdx; // elements count duke@435: const Register dword_count = rcx; duke@435: const Register qword_count = count; duke@435: duke@435: __ enter(); // required for proper stackwalking of RuntimeStub frame duke@435: assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. duke@435: iveresov@2595: if (entry != NULL) { iveresov@2595: *entry = __ pc(); iveresov@2595: // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) iveresov@2595: BLOCK_COMMENT("Entry:"); iveresov@2595: } iveresov@2595: iveresov@2595: array_overlap_test(nooverlap_target, Address::times_4); iveresov@2595: setup_arg_regs(); // from => rdi, to => rsi, count => rdx iveresov@2595: // r9 and r10 may be used to save non-volatile registers iveresov@2595: coleenp@548: if (is_oop) { coleenp@548: // no registers are destroyed by this call iveresov@2606: gen_write_ref_array_pre_barrier(to, count, dest_uninitialized); coleenp@548: } coleenp@548: coleenp@548: assert_clean_int(count, rax); // Make sure 'count' is clean int. duke@435: // 'from', 'to' and 'count' are now valid never@739: __ movptr(dword_count, count); never@739: __ shrptr(count, 1); // count => qword_count duke@435: duke@435: // Copy from high to low addresses. Use 'to' as scratch. duke@435: duke@435: // Check for and copy trailing dword never@739: __ testl(dword_count, 1); kvn@4411: __ jcc(Assembler::zero, L_copy_bytes); duke@435: __ movl(rax, Address(from, dword_count, Address::times_4, -4)); duke@435: __ movl(Address(to, dword_count, Address::times_4, -4), rax); kvn@4411: __ jmp(L_copy_bytes); duke@435: duke@435: // Copy trailing qwords duke@435: __ BIND(L_copy_8_bytes); duke@435: __ movq(rax, Address(from, qword_count, Address::times_8, -8)); duke@435: __ movq(Address(to, qword_count, Address::times_8, -8), rax); never@739: __ decrement(qword_count); duke@435: __ jcc(Assembler::notZero, L_copy_8_bytes); duke@435: coleenp@548: if (is_oop) { coleenp@548: __ jmp(L_exit); coleenp@548: } duke@435: restore_arg_regs(); never@3314: inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free never@739: __ xorptr(rax, rax); // return 0 duke@435: __ leave(); // required for proper stackwalking of RuntimeStub frame duke@435: __ ret(0); duke@435: kvn@4411: // Copy in multi-bytes chunks kvn@4411: copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes); duke@435: coleenp@548: __ bind(L_exit); coleenp@548: if (is_oop) { coleenp@548: Register end_to = rdx; coleenp@548: __ leaq(end_to, Address(to, dword_count, Address::times_4, -4)); coleenp@548: gen_write_ref_array_post_barrier(to, end_to, rax); coleenp@548: } duke@435: restore_arg_regs(); never@3314: inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free never@739: __ xorptr(rax, rax); // return 0 duke@435: __ leave(); // required for proper stackwalking of RuntimeStub frame duke@435: __ ret(0); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: // Arguments: duke@435: // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes duke@435: // ignored duke@435: // is_oop - true => oop array, so generate store check code duke@435: // name - stub name string duke@435: // duke@435: // Inputs: duke@435: // c_rarg0 - source array address duke@435: // c_rarg1 - destination array address duke@435: // c_rarg2 - element count, treated as ssize_t, can be zero duke@435: // coleenp@548: // Side Effects: duke@435: // disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the duke@435: // no-overlap entry point used by generate_conjoint_long_oop_copy(). duke@435: // iveresov@2606: address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, address *entry, iveresov@2606: const char *name, bool dest_uninitialized = false) { duke@435: __ align(CodeEntryAlignment); duke@435: StubCodeMark mark(this, "StubRoutines", name); duke@435: address start = __ pc(); duke@435: kvn@4411: Label L_copy_bytes, L_copy_8_bytes, L_exit; duke@435: const Register from = rdi; // source array address duke@435: const Register to = rsi; // destination array address duke@435: const Register qword_count = rdx; // elements count duke@435: const Register end_from = from; // source array end address duke@435: const Register end_to = rcx; // destination array end address duke@435: const Register saved_to = to; duke@435: // End pointers are inclusive, and if count is not zero they point duke@435: // to the last unit copied: end_to[0] := end_from[0] duke@435: duke@435: __ enter(); // required for proper stackwalking of RuntimeStub frame duke@435: // Save no-overlap entry point for generate_conjoint_long_oop_copy() duke@435: assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. duke@435: iveresov@2595: if (entry != NULL) { iveresov@2595: *entry = __ pc(); iveresov@2595: // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) iveresov@2595: BLOCK_COMMENT("Entry:"); duke@435: } duke@435: duke@435: setup_arg_regs(); // from => rdi, to => rsi, count => rdx duke@435: // r9 and r10 may be used to save non-volatile registers duke@435: // 'from', 'to' and 'qword_count' are now valid iveresov@2595: if (is_oop) { iveresov@2595: // no registers are destroyed by this call iveresov@2606: gen_write_ref_array_pre_barrier(to, qword_count, dest_uninitialized); iveresov@2595: } duke@435: duke@435: // Copy from low to high addresses. Use 'to' as scratch. never@739: __ lea(end_from, Address(from, qword_count, Address::times_8, -8)); never@739: __ lea(end_to, Address(to, qword_count, Address::times_8, -8)); never@739: __ negptr(qword_count); kvn@4411: __ jmp(L_copy_bytes); duke@435: duke@435: // Copy trailing qwords duke@435: __ BIND(L_copy_8_bytes); duke@435: __ movq(rax, Address(end_from, qword_count, Address::times_8, 8)); duke@435: __ movq(Address(end_to, qword_count, Address::times_8, 8), rax); never@739: __ increment(qword_count); duke@435: __ jcc(Assembler::notZero, L_copy_8_bytes); duke@435: duke@435: if (is_oop) { duke@435: __ jmp(L_exit); duke@435: } else { duke@435: restore_arg_regs(); never@3314: inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free never@739: __ xorptr(rax, rax); // return 0 duke@435: __ leave(); // required for proper stackwalking of RuntimeStub frame duke@435: __ ret(0); duke@435: } duke@435: kvn@4411: // Copy in multi-bytes chunks kvn@4411: copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes); duke@435: duke@435: if (is_oop) { duke@435: __ BIND(L_exit); duke@435: gen_write_ref_array_post_barrier(saved_to, end_to, rax); duke@435: } duke@435: restore_arg_regs(); never@3314: if (is_oop) { never@3314: inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free never@3314: } else { never@3314: inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free never@3314: } never@739: __ xorptr(rax, rax); // return 0 duke@435: __ leave(); // required for proper stackwalking of RuntimeStub frame duke@435: __ ret(0); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: // Arguments: duke@435: // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes duke@435: // ignored duke@435: // is_oop - true => oop array, so generate store check code duke@435: // name - stub name string duke@435: // duke@435: // Inputs: duke@435: // c_rarg0 - source array address duke@435: // c_rarg1 - destination array address duke@435: // c_rarg2 - element count, treated as ssize_t, can be zero duke@435: // iveresov@2606: address generate_conjoint_long_oop_copy(bool aligned, bool is_oop, iveresov@2606: address nooverlap_target, address *entry, iveresov@2606: const char *name, bool dest_uninitialized = false) { duke@435: __ align(CodeEntryAlignment); duke@435: StubCodeMark mark(this, "StubRoutines", name); duke@435: address start = __ pc(); duke@435: kvn@4411: Label L_copy_bytes, L_copy_8_bytes, L_exit; duke@435: const Register from = rdi; // source array address duke@435: const Register to = rsi; // destination array address duke@435: const Register qword_count = rdx; // elements count duke@435: const Register saved_count = rcx; duke@435: duke@435: __ enter(); // required for proper stackwalking of RuntimeStub frame duke@435: assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. duke@435: iveresov@2595: if (entry != NULL) { iveresov@2595: *entry = __ pc(); iveresov@2595: // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) iveresov@2595: BLOCK_COMMENT("Entry:"); duke@435: } iveresov@2595: iveresov@2595: array_overlap_test(nooverlap_target, Address::times_8); duke@435: setup_arg_regs(); // from => rdi, to => rsi, count => rdx duke@435: // r9 and r10 may be used to save non-volatile registers duke@435: // 'from', 'to' and 'qword_count' are now valid duke@435: if (is_oop) { duke@435: // Save to and count for store barrier never@739: __ movptr(saved_count, qword_count); duke@435: // No registers are destroyed by this call iveresov@2606: gen_write_ref_array_pre_barrier(to, saved_count, dest_uninitialized); duke@435: } duke@435: kvn@4411: __ jmp(L_copy_bytes); duke@435: duke@435: // Copy trailing qwords duke@435: __ BIND(L_copy_8_bytes); duke@435: __ movq(rax, Address(from, qword_count, Address::times_8, -8)); duke@435: __ movq(Address(to, qword_count, Address::times_8, -8), rax); never@739: __ decrement(qword_count); duke@435: __ jcc(Assembler::notZero, L_copy_8_bytes); duke@435: duke@435: if (is_oop) { duke@435: __ jmp(L_exit); duke@435: } else { duke@435: restore_arg_regs(); never@3314: inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free never@739: __ xorptr(rax, rax); // return 0 duke@435: __ leave(); // required for proper stackwalking of RuntimeStub frame duke@435: __ ret(0); duke@435: } duke@435: kvn@4411: // Copy in multi-bytes chunks kvn@4411: copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes); duke@435: duke@435: if (is_oop) { duke@435: __ BIND(L_exit); never@739: __ lea(rcx, Address(to, saved_count, Address::times_8, -8)); duke@435: gen_write_ref_array_post_barrier(to, rcx, rax); duke@435: } duke@435: restore_arg_regs(); never@3314: if (is_oop) { never@3314: inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free never@3314: } else { never@3314: inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free never@3314: } never@739: __ xorptr(rax, rax); // return 0 duke@435: __ leave(); // required for proper stackwalking of RuntimeStub frame duke@435: __ ret(0); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: duke@435: // Helper for generating a dynamic type check. duke@435: // Smashes no registers. duke@435: void generate_type_check(Register sub_klass, duke@435: Register super_check_offset, duke@435: Register super_klass, duke@435: Label& L_success) { duke@435: assert_different_registers(sub_klass, super_check_offset, super_klass); duke@435: duke@435: BLOCK_COMMENT("type_check:"); duke@435: duke@435: Label L_miss; duke@435: jrose@1079: __ check_klass_subtype_fast_path(sub_klass, super_klass, noreg, &L_success, &L_miss, NULL, jrose@1079: super_check_offset); jrose@1079: __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, &L_success, NULL); duke@435: duke@435: // Fall through on failure! duke@435: __ BIND(L_miss); duke@435: } duke@435: duke@435: // duke@435: // Generate checkcasting array copy stub duke@435: // duke@435: // Input: duke@435: // c_rarg0 - source array address duke@435: // c_rarg1 - destination array address duke@435: // c_rarg2 - element count, treated as ssize_t, can be zero duke@435: // c_rarg3 - size_t ckoff (super_check_offset) duke@435: // not Win64 duke@435: // c_rarg4 - oop ckval (super_klass) duke@435: // Win64 duke@435: // rsp+40 - oop ckval (super_klass) duke@435: // duke@435: // Output: duke@435: // rax == 0 - success duke@435: // rax == -1^K - failure, where K is partial transfer count duke@435: // iveresov@2606: address generate_checkcast_copy(const char *name, address *entry, iveresov@2606: bool dest_uninitialized = false) { duke@435: duke@435: Label L_load_element, L_store_element, L_do_card_marks, L_done; duke@435: duke@435: // Input registers (after setup_arg_regs) duke@435: const Register from = rdi; // source array address duke@435: const Register to = rsi; // destination array address duke@435: const Register length = rdx; // elements count duke@435: const Register ckoff = rcx; // super_check_offset duke@435: const Register ckval = r8; // super_klass duke@435: duke@435: // Registers used as temps (r13, r14 are save-on-entry) duke@435: const Register end_from = from; // source array end address duke@435: const Register end_to = r13; // destination array end address duke@435: const Register count = rdx; // -(count_remaining) duke@435: const Register r14_length = r14; // saved copy of length duke@435: // End pointers are inclusive, and if length is not zero they point duke@435: // to the last unit copied: end_to[0] := end_from[0] duke@435: duke@435: const Register rax_oop = rax; // actual oop copied duke@435: const Register r11_klass = r11; // oop._klass duke@435: duke@435: //--------------------------------------------------------------- duke@435: // Assembler stub will be used for this call to arraycopy duke@435: // if the two arrays are subtypes of Object[] but the duke@435: // destination array type is not equal to or a supertype duke@435: // of the source type. Each element must be separately duke@435: // checked. duke@435: duke@435: __ align(CodeEntryAlignment); duke@435: StubCodeMark mark(this, "StubRoutines", name); duke@435: address start = __ pc(); duke@435: duke@435: __ enter(); // required for proper stackwalking of RuntimeStub frame duke@435: duke@435: #ifdef ASSERT duke@435: // caller guarantees that the arrays really are different duke@435: // otherwise, we would have to make conjoint checks duke@435: { Label L; coleenp@548: array_overlap_test(L, TIMES_OOP); duke@435: __ stop("checkcast_copy within a single array"); duke@435: __ bind(L); duke@435: } duke@435: #endif //ASSERT duke@435: duke@435: setup_arg_regs(4); // from => rdi, to => rsi, length => rdx duke@435: // ckoff => rcx, ckval => r8 duke@435: // r9 and r10 may be used to save non-volatile registers duke@435: #ifdef _WIN64 duke@435: // last argument (#4) is on stack on Win64 twisti@2348: __ movptr(ckval, Address(rsp, 6 * wordSize)); duke@435: #endif duke@435: twisti@2348: // Caller of this entry point must set up the argument registers. iveresov@2595: if (entry != NULL) { iveresov@2595: *entry = __ pc(); iveresov@2595: BLOCK_COMMENT("Entry:"); iveresov@2595: } twisti@2348: twisti@2348: // allocate spill slots for r13, r14 twisti@2348: enum { twisti@2348: saved_r13_offset, twisti@2348: saved_r14_offset, twisti@2348: saved_rbp_offset twisti@2348: }; twisti@2348: __ subptr(rsp, saved_rbp_offset * wordSize); twisti@2348: __ movptr(Address(rsp, saved_r13_offset * wordSize), r13); twisti@2348: __ movptr(Address(rsp, saved_r14_offset * wordSize), r14); twisti@2348: duke@435: // check that int operands are properly extended to size_t duke@435: assert_clean_int(length, rax); duke@435: assert_clean_int(ckoff, rax); duke@435: duke@435: #ifdef ASSERT duke@435: BLOCK_COMMENT("assert consistent ckoff/ckval"); duke@435: // The ckoff and ckval must be mutually consistent, duke@435: // even though caller generates both. duke@435: { Label L; stefank@3391: int sco_offset = in_bytes(Klass::super_check_offset_offset()); duke@435: __ cmpl(ckoff, Address(ckval, sco_offset)); duke@435: __ jcc(Assembler::equal, L); duke@435: __ stop("super_check_offset inconsistent"); duke@435: __ bind(L); duke@435: } duke@435: #endif //ASSERT duke@435: duke@435: // Loop-invariant addresses. They are exclusive end pointers. coleenp@548: Address end_from_addr(from, length, TIMES_OOP, 0); coleenp@548: Address end_to_addr(to, length, TIMES_OOP, 0); duke@435: // Loop-variant addresses. They assume post-incremented count < 0. coleenp@548: Address from_element_addr(end_from, count, TIMES_OOP, 0); coleenp@548: Address to_element_addr(end_to, count, TIMES_OOP, 0); duke@435: iveresov@2606: gen_write_ref_array_pre_barrier(to, count, dest_uninitialized); duke@435: duke@435: // Copy from low to high addresses, indexed from the end of each array. never@739: __ lea(end_from, end_from_addr); never@739: __ lea(end_to, end_to_addr); never@739: __ movptr(r14_length, length); // save a copy of the length never@739: assert(length == count, ""); // else fix next line: never@739: __ negptr(count); // negate and test the length duke@435: __ jcc(Assembler::notZero, L_load_element); duke@435: duke@435: // Empty array: Nothing to do. never@739: __ xorptr(rax, rax); // return 0 on (trivial) success duke@435: __ jmp(L_done); duke@435: duke@435: // ======== begin loop ======== duke@435: // (Loop is rotated; its entry is L_load_element.) duke@435: // Loop control: duke@435: // for (count = -count; count != 0; count++) duke@435: // Base pointers src, dst are biased by 8*(count-1),to last element. kvn@1800: __ align(OptoLoopAlignment); duke@435: duke@435: __ BIND(L_store_element); coleenp@548: __ store_heap_oop(to_element_addr, rax_oop); // store the oop never@739: __ increment(count); // increment the count toward zero duke@435: __ jcc(Assembler::zero, L_do_card_marks); duke@435: duke@435: // ======== loop entry is here ======== duke@435: __ BIND(L_load_element); coleenp@548: __ load_heap_oop(rax_oop, from_element_addr); // load the oop never@739: __ testptr(rax_oop, rax_oop); duke@435: __ jcc(Assembler::zero, L_store_element); duke@435: coleenp@548: __ load_klass(r11_klass, rax_oop);// query the object klass duke@435: generate_type_check(r11_klass, ckoff, ckval, L_store_element); duke@435: // ======== end loop ======== duke@435: duke@435: // It was a real error; we must depend on the caller to finish the job. duke@435: // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops. duke@435: // Emit GC store barriers for the oops we have copied (r14 + rdx), duke@435: // and report their number to the caller. duke@435: assert_different_registers(rax, r14_length, count, to, end_to, rcx); never@739: __ lea(end_to, to_element_addr); ysr@1280: __ addptr(end_to, -heapOopSize); // make an inclusive end pointer apetrusenko@797: gen_write_ref_array_post_barrier(to, end_to, rscratch1); never@739: __ movptr(rax, r14_length); // original oops never@739: __ addptr(rax, count); // K = (original - remaining) oops never@739: __ notptr(rax); // report (-1^K) to caller duke@435: __ jmp(L_done); duke@435: duke@435: // Come here on success only. duke@435: __ BIND(L_do_card_marks); ysr@1280: __ addptr(end_to, -heapOopSize); // make an inclusive end pointer apetrusenko@797: gen_write_ref_array_post_barrier(to, end_to, rscratch1); never@739: __ xorptr(rax, rax); // return 0 on success duke@435: duke@435: // Common exit point (success or failure). duke@435: __ BIND(L_done); never@739: __ movptr(r13, Address(rsp, saved_r13_offset * wordSize)); never@739: __ movptr(r14, Address(rsp, saved_r14_offset * wordSize)); duke@435: restore_arg_regs(); never@3314: inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); // Update counter after rscratch1 is free duke@435: __ leave(); // required for proper stackwalking of RuntimeStub frame duke@435: __ ret(0); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: // duke@435: // Generate 'unsafe' array copy stub duke@435: // Though just as safe as the other stubs, it takes an unscaled duke@435: // size_t argument instead of an element count. duke@435: // duke@435: // Input: duke@435: // c_rarg0 - source array address duke@435: // c_rarg1 - destination array address duke@435: // c_rarg2 - byte count, treated as ssize_t, can be zero duke@435: // duke@435: // Examines the alignment of the operands and dispatches duke@435: // to a long, int, short, or byte copy loop. duke@435: // iveresov@2595: address generate_unsafe_copy(const char *name, iveresov@2595: address byte_copy_entry, address short_copy_entry, iveresov@2595: address int_copy_entry, address long_copy_entry) { duke@435: duke@435: Label L_long_aligned, L_int_aligned, L_short_aligned; duke@435: duke@435: // Input registers (before setup_arg_regs) duke@435: const Register from = c_rarg0; // source array address duke@435: const Register to = c_rarg1; // destination array address duke@435: const Register size = c_rarg2; // byte count (size_t) duke@435: duke@435: // Register used as a temp duke@435: const Register bits = rax; // test copy of low bits duke@435: duke@435: __ align(CodeEntryAlignment); duke@435: StubCodeMark mark(this, "StubRoutines", name); duke@435: address start = __ pc(); duke@435: duke@435: __ enter(); // required for proper stackwalking of RuntimeStub frame duke@435: duke@435: // bump this on entry, not on exit: duke@435: inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr); duke@435: never@739: __ mov(bits, from); never@739: __ orptr(bits, to); never@739: __ orptr(bits, size); duke@435: duke@435: __ testb(bits, BytesPerLong-1); duke@435: __ jccb(Assembler::zero, L_long_aligned); duke@435: duke@435: __ testb(bits, BytesPerInt-1); duke@435: __ jccb(Assembler::zero, L_int_aligned); duke@435: duke@435: __ testb(bits, BytesPerShort-1); duke@435: __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry)); duke@435: duke@435: __ BIND(L_short_aligned); never@739: __ shrptr(size, LogBytesPerShort); // size => short_count duke@435: __ jump(RuntimeAddress(short_copy_entry)); duke@435: duke@435: __ BIND(L_int_aligned); never@739: __ shrptr(size, LogBytesPerInt); // size => int_count duke@435: __ jump(RuntimeAddress(int_copy_entry)); duke@435: duke@435: __ BIND(L_long_aligned); never@739: __ shrptr(size, LogBytesPerLong); // size => qword_count duke@435: __ jump(RuntimeAddress(long_copy_entry)); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: // Perform range checks on the proposed arraycopy. duke@435: // Kills temp, but nothing else. duke@435: // Also, clean the sign bits of src_pos and dst_pos. duke@435: void arraycopy_range_checks(Register src, // source array oop (c_rarg0) duke@435: Register src_pos, // source position (c_rarg1) duke@435: Register dst, // destination array oo (c_rarg2) duke@435: Register dst_pos, // destination position (c_rarg3) duke@435: Register length, duke@435: Register temp, duke@435: Label& L_failed) { duke@435: BLOCK_COMMENT("arraycopy_range_checks:"); duke@435: duke@435: // if (src_pos + length > arrayOop(src)->length()) FAIL; duke@435: __ movl(temp, length); duke@435: __ addl(temp, src_pos); // src_pos + length duke@435: __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes())); duke@435: __ jcc(Assembler::above, L_failed); duke@435: duke@435: // if (dst_pos + length > arrayOop(dst)->length()) FAIL; duke@435: __ movl(temp, length); duke@435: __ addl(temp, dst_pos); // dst_pos + length duke@435: __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes())); duke@435: __ jcc(Assembler::above, L_failed); duke@435: duke@435: // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'. duke@435: // Move with sign extension can be used since they are positive. duke@435: __ movslq(src_pos, src_pos); duke@435: __ movslq(dst_pos, dst_pos); duke@435: duke@435: BLOCK_COMMENT("arraycopy_range_checks done"); duke@435: } duke@435: duke@435: // duke@435: // Generate generic array copy stubs duke@435: // duke@435: // Input: duke@435: // c_rarg0 - src oop duke@435: // c_rarg1 - src_pos (32-bits) duke@435: // c_rarg2 - dst oop duke@435: // c_rarg3 - dst_pos (32-bits) duke@435: // not Win64 duke@435: // c_rarg4 - element count (32-bits) duke@435: // Win64 duke@435: // rsp+40 - element count (32-bits) duke@435: // duke@435: // Output: duke@435: // rax == 0 - success duke@435: // rax == -1^K - failure, where K is partial transfer count duke@435: // iveresov@2595: address generate_generic_copy(const char *name, iveresov@2595: address byte_copy_entry, address short_copy_entry, iveresov@2691: address int_copy_entry, address oop_copy_entry, iveresov@2691: address long_copy_entry, address checkcast_copy_entry) { duke@435: duke@435: Label L_failed, L_failed_0, L_objArray; duke@435: Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs; duke@435: duke@435: // Input registers duke@435: const Register src = c_rarg0; // source array oop duke@435: const Register src_pos = c_rarg1; // source position duke@435: const Register dst = c_rarg2; // destination array oop duke@435: const Register dst_pos = c_rarg3; // destination position twisti@2348: #ifndef _WIN64 twisti@2348: const Register length = c_rarg4; duke@435: #else twisti@2348: const Address length(rsp, 6 * wordSize); // elements count is on stack on Win64 duke@435: #endif duke@435: duke@435: { int modulus = CodeEntryAlignment; duke@435: int target = modulus - 5; // 5 = sizeof jmp(L_failed) duke@435: int advance = target - (__ offset() % modulus); duke@435: if (advance < 0) advance += modulus; duke@435: if (advance > 0) __ nop(advance); duke@435: } duke@435: StubCodeMark mark(this, "StubRoutines", name); duke@435: duke@435: // Short-hop target to L_failed. Makes for denser prologue code. duke@435: __ BIND(L_failed_0); duke@435: __ jmp(L_failed); duke@435: assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed"); duke@435: duke@435: __ align(CodeEntryAlignment); duke@435: address start = __ pc(); duke@435: duke@435: __ enter(); // required for proper stackwalking of RuntimeStub frame duke@435: duke@435: // bump this on entry, not on exit: duke@435: inc_counter_np(SharedRuntime::_generic_array_copy_ctr); duke@435: duke@435: //----------------------------------------------------------------------- duke@435: // Assembler stub will be used for this call to arraycopy duke@435: // if the following conditions are met: duke@435: // duke@435: // (1) src and dst must not be null. duke@435: // (2) src_pos must not be negative. duke@435: // (3) dst_pos must not be negative. duke@435: // (4) length must not be negative. duke@435: // (5) src klass and dst klass should be the same and not NULL. duke@435: // (6) src and dst should be arrays. duke@435: // (7) src_pos + length must not exceed length of src. duke@435: // (8) dst_pos + length must not exceed length of dst. duke@435: // duke@435: duke@435: // if (src == NULL) return -1; never@739: __ testptr(src, src); // src oop duke@435: size_t j1off = __ offset(); duke@435: __ jccb(Assembler::zero, L_failed_0); duke@435: duke@435: // if (src_pos < 0) return -1; duke@435: __ testl(src_pos, src_pos); // src_pos (32-bits) duke@435: __ jccb(Assembler::negative, L_failed_0); duke@435: duke@435: // if (dst == NULL) return -1; never@739: __ testptr(dst, dst); // dst oop duke@435: __ jccb(Assembler::zero, L_failed_0); duke@435: duke@435: // if (dst_pos < 0) return -1; duke@435: __ testl(dst_pos, dst_pos); // dst_pos (32-bits) duke@435: size_t j4off = __ offset(); duke@435: __ jccb(Assembler::negative, L_failed_0); duke@435: duke@435: // The first four tests are very dense code, duke@435: // but not quite dense enough to put four duke@435: // jumps in a 16-byte instruction fetch buffer. duke@435: // That's good, because some branch predicters duke@435: // do not like jumps so close together. duke@435: // Make sure of this. duke@435: guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps"); duke@435: duke@435: // registers used as temp duke@435: const Register r11_length = r11; // elements count to copy duke@435: const Register r10_src_klass = r10; // array klass duke@435: duke@435: // if (length < 0) return -1; twisti@2348: __ movl(r11_length, length); // length (elements count, 32-bits value) duke@435: __ testl(r11_length, r11_length); duke@435: __ jccb(Assembler::negative, L_failed_0); duke@435: coleenp@548: __ load_klass(r10_src_klass, src); duke@435: #ifdef ASSERT duke@435: // assert(src->klass() != NULL); twisti@2348: { twisti@2348: BLOCK_COMMENT("assert klasses not null {"); twisti@2348: Label L1, L2; never@739: __ testptr(r10_src_klass, r10_src_klass); duke@435: __ jcc(Assembler::notZero, L2); // it is broken if klass is NULL duke@435: __ bind(L1); duke@435: __ stop("broken null klass"); duke@435: __ bind(L2); twisti@2348: __ load_klass(rax, dst); twisti@2348: __ cmpq(rax, 0); duke@435: __ jcc(Assembler::equal, L1); // this would be broken also twisti@2348: BLOCK_COMMENT("} assert klasses not null done"); duke@435: } duke@435: #endif duke@435: duke@435: // Load layout helper (32-bits) duke@435: // duke@435: // |array_tag| | header_size | element_type | |log2_element_size| duke@435: // 32 30 24 16 8 2 0 duke@435: // duke@435: // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0 duke@435: // duke@435: stefank@3391: const int lh_offset = in_bytes(Klass::layout_helper_offset()); twisti@2348: twisti@2348: // Handle objArrays completely differently... twisti@2348: const jint objArray_lh = Klass::array_layout_helper(T_OBJECT); twisti@2348: __ cmpl(Address(r10_src_klass, lh_offset), objArray_lh); twisti@2348: __ jcc(Assembler::equal, L_objArray); twisti@2348: twisti@2348: // if (src->klass() != dst->klass()) return -1; twisti@2348: __ load_klass(rax, dst); twisti@2348: __ cmpq(r10_src_klass, rax); twisti@2348: __ jcc(Assembler::notEqual, L_failed); duke@435: duke@435: const Register rax_lh = rax; // layout helper duke@435: __ movl(rax_lh, Address(r10_src_klass, lh_offset)); duke@435: duke@435: // if (!src->is_Array()) return -1; duke@435: __ cmpl(rax_lh, Klass::_lh_neutral_value); duke@435: __ jcc(Assembler::greaterEqual, L_failed); duke@435: duke@435: // At this point, it is known to be a typeArray (array_tag 0x3). duke@435: #ifdef ASSERT twisti@2348: { twisti@2348: BLOCK_COMMENT("assert primitive array {"); twisti@2348: Label L; duke@435: __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift)); duke@435: __ jcc(Assembler::greaterEqual, L); duke@435: __ stop("must be a primitive array"); duke@435: __ bind(L); twisti@2348: BLOCK_COMMENT("} assert primitive array done"); duke@435: } duke@435: #endif duke@435: duke@435: arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length, duke@435: r10, L_failed); duke@435: coleenp@4142: // TypeArrayKlass duke@435: // duke@435: // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize); duke@435: // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize); duke@435: // duke@435: duke@435: const Register r10_offset = r10; // array offset duke@435: const Register rax_elsize = rax_lh; // element size duke@435: duke@435: __ movl(r10_offset, rax_lh); duke@435: __ shrl(r10_offset, Klass::_lh_header_size_shift); never@739: __ andptr(r10_offset, Klass::_lh_header_size_mask); // array_offset never@739: __ addptr(src, r10_offset); // src array offset never@739: __ addptr(dst, r10_offset); // dst array offset duke@435: BLOCK_COMMENT("choose copy loop based on element size"); duke@435: __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize duke@435: duke@435: // next registers should be set before the jump to corresponding stub duke@435: const Register from = c_rarg0; // source array address duke@435: const Register to = c_rarg1; // destination array address duke@435: const Register count = c_rarg2; // elements count duke@435: duke@435: // 'from', 'to', 'count' registers should be set in such order duke@435: // since they are the same as 'src', 'src_pos', 'dst'. duke@435: duke@435: __ BIND(L_copy_bytes); duke@435: __ cmpl(rax_elsize, 0); duke@435: __ jccb(Assembler::notEqual, L_copy_shorts); never@739: __ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr never@739: __ lea(to, Address(dst, dst_pos, Address::times_1, 0));// dst_addr never@739: __ movl2ptr(count, r11_length); // length duke@435: __ jump(RuntimeAddress(byte_copy_entry)); duke@435: duke@435: __ BIND(L_copy_shorts); duke@435: __ cmpl(rax_elsize, LogBytesPerShort); duke@435: __ jccb(Assembler::notEqual, L_copy_ints); never@739: __ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr never@739: __ lea(to, Address(dst, dst_pos, Address::times_2, 0));// dst_addr never@739: __ movl2ptr(count, r11_length); // length duke@435: __ jump(RuntimeAddress(short_copy_entry)); duke@435: duke@435: __ BIND(L_copy_ints); duke@435: __ cmpl(rax_elsize, LogBytesPerInt); duke@435: __ jccb(Assembler::notEqual, L_copy_longs); never@739: __ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr never@739: __ lea(to, Address(dst, dst_pos, Address::times_4, 0));// dst_addr never@739: __ movl2ptr(count, r11_length); // length duke@435: __ jump(RuntimeAddress(int_copy_entry)); duke@435: duke@435: __ BIND(L_copy_longs); duke@435: #ifdef ASSERT twisti@2348: { twisti@2348: BLOCK_COMMENT("assert long copy {"); twisti@2348: Label L; duke@435: __ cmpl(rax_elsize, LogBytesPerLong); duke@435: __ jcc(Assembler::equal, L); duke@435: __ stop("must be long copy, but elsize is wrong"); duke@435: __ bind(L); twisti@2348: BLOCK_COMMENT("} assert long copy done"); duke@435: } duke@435: #endif never@739: __ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr never@739: __ lea(to, Address(dst, dst_pos, Address::times_8, 0));// dst_addr never@739: __ movl2ptr(count, r11_length); // length duke@435: __ jump(RuntimeAddress(long_copy_entry)); duke@435: coleenp@4142: // ObjArrayKlass duke@435: __ BIND(L_objArray); twisti@2348: // live at this point: r10_src_klass, r11_length, src[_pos], dst[_pos] duke@435: duke@435: Label L_plain_copy, L_checkcast_copy; duke@435: // test array classes for subtyping twisti@2348: __ load_klass(rax, dst); twisti@2348: __ cmpq(r10_src_klass, rax); // usual case is exact equality duke@435: __ jcc(Assembler::notEqual, L_checkcast_copy); duke@435: duke@435: // Identically typed arrays can be copied without element-wise checks. duke@435: arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length, duke@435: r10, L_failed); duke@435: never@739: __ lea(from, Address(src, src_pos, TIMES_OOP, duke@435: arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr never@739: __ lea(to, Address(dst, dst_pos, TIMES_OOP, never@739: arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr never@739: __ movl2ptr(count, r11_length); // length duke@435: __ BIND(L_plain_copy); duke@435: __ jump(RuntimeAddress(oop_copy_entry)); duke@435: duke@435: __ BIND(L_checkcast_copy); twisti@2348: // live at this point: r10_src_klass, r11_length, rax (dst_klass) duke@435: { duke@435: // Before looking at dst.length, make sure dst is also an objArray. twisti@2348: __ cmpl(Address(rax, lh_offset), objArray_lh); duke@435: __ jcc(Assembler::notEqual, L_failed); duke@435: duke@435: // It is safe to examine both src.length and dst.length. duke@435: arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length, duke@435: rax, L_failed); twisti@2348: twisti@2348: const Register r11_dst_klass = r11; coleenp@548: __ load_klass(r11_dst_klass, dst); // reload duke@435: duke@435: // Marshal the base address arguments now, freeing registers. never@739: __ lea(from, Address(src, src_pos, TIMES_OOP, duke@435: arrayOopDesc::base_offset_in_bytes(T_OBJECT))); never@739: __ lea(to, Address(dst, dst_pos, TIMES_OOP, duke@435: arrayOopDesc::base_offset_in_bytes(T_OBJECT))); twisti@2348: __ movl(count, length); // length (reloaded) duke@435: Register sco_temp = c_rarg3; // this register is free now duke@435: assert_different_registers(from, to, count, sco_temp, duke@435: r11_dst_klass, r10_src_klass); duke@435: assert_clean_int(count, sco_temp); duke@435: duke@435: // Generate the type check. stefank@3391: const int sco_offset = in_bytes(Klass::super_check_offset_offset()); duke@435: __ movl(sco_temp, Address(r11_dst_klass, sco_offset)); duke@435: assert_clean_int(sco_temp, rax); duke@435: generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy); duke@435: coleenp@4142: // Fetch destination element klass from the ObjArrayKlass header. coleenp@4142: int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset()); never@739: __ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset)); twisti@2348: __ movl( sco_temp, Address(r11_dst_klass, sco_offset)); duke@435: assert_clean_int(sco_temp, rax); duke@435: duke@435: // the checkcast_copy loop needs two extra arguments: duke@435: assert(c_rarg3 == sco_temp, "#3 already in place"); twisti@2348: // Set up arguments for checkcast_copy_entry. twisti@2348: setup_arg_regs(4); twisti@2348: __ movptr(r8, r11_dst_klass); // dst.klass.element_klass, r8 is c_rarg4 on Linux/Solaris duke@435: __ jump(RuntimeAddress(checkcast_copy_entry)); duke@435: } duke@435: duke@435: __ BIND(L_failed); never@739: __ xorptr(rax, rax); never@739: __ notptr(rax); // return -1 duke@435: __ leave(); // required for proper stackwalking of RuntimeStub frame duke@435: __ ret(0); duke@435: duke@435: return start; duke@435: } duke@435: duke@435: void generate_arraycopy_stubs() { iveresov@2595: address entry; iveresov@2595: address entry_jbyte_arraycopy; iveresov@2595: address entry_jshort_arraycopy; iveresov@2595: address entry_jint_arraycopy; iveresov@2595: address entry_oop_arraycopy; iveresov@2595: address entry_jlong_arraycopy; iveresov@2595: address entry_checkcast_arraycopy; iveresov@2595: iveresov@2595: StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, &entry, iveresov@2595: "jbyte_disjoint_arraycopy"); iveresov@2595: StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, entry, &entry_jbyte_arraycopy, iveresov@2595: "jbyte_arraycopy"); iveresov@2595: iveresov@2595: StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, &entry, iveresov@2595: "jshort_disjoint_arraycopy"); iveresov@2595: StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, entry, &entry_jshort_arraycopy, iveresov@2595: "jshort_arraycopy"); iveresov@2595: iveresov@2595: StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, false, &entry, iveresov@2595: "jint_disjoint_arraycopy"); iveresov@2595: StubRoutines::_jint_arraycopy = generate_conjoint_int_oop_copy(false, false, entry, iveresov@2595: &entry_jint_arraycopy, "jint_arraycopy"); iveresov@2595: iveresov@2595: StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, false, &entry, iveresov@2595: "jlong_disjoint_arraycopy"); iveresov@2595: StubRoutines::_jlong_arraycopy = generate_conjoint_long_oop_copy(false, false, entry, iveresov@2595: &entry_jlong_arraycopy, "jlong_arraycopy"); duke@435: coleenp@548: coleenp@548: if (UseCompressedOops) { iveresov@2595: StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, true, &entry, iveresov@2595: "oop_disjoint_arraycopy"); iveresov@2595: StubRoutines::_oop_arraycopy = generate_conjoint_int_oop_copy(false, true, entry, iveresov@2595: &entry_oop_arraycopy, "oop_arraycopy"); iveresov@2606: StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_int_oop_copy(false, true, &entry, iveresov@2606: "oop_disjoint_arraycopy_uninit", iveresov@2606: /*dest_uninitialized*/true); iveresov@2606: StubRoutines::_oop_arraycopy_uninit = generate_conjoint_int_oop_copy(false, true, entry, iveresov@2606: NULL, "oop_arraycopy_uninit", iveresov@2606: /*dest_uninitialized*/true); coleenp@548: } else { iveresov@2595: StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, true, &entry, iveresov@2595: "oop_disjoint_arraycopy"); iveresov@2595: StubRoutines::_oop_arraycopy = generate_conjoint_long_oop_copy(false, true, entry, iveresov@2595: &entry_oop_arraycopy, "oop_arraycopy"); iveresov@2606: StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_long_oop_copy(false, true, &entry, iveresov@2606: "oop_disjoint_arraycopy_uninit", iveresov@2606: /*dest_uninitialized*/true); iveresov@2606: StubRoutines::_oop_arraycopy_uninit = generate_conjoint_long_oop_copy(false, true, entry, iveresov@2606: NULL, "oop_arraycopy_uninit", iveresov@2606: /*dest_uninitialized*/true); coleenp@548: } duke@435: iveresov@2606: StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy); iveresov@2606: StubRoutines::_checkcast_arraycopy_uninit = generate_checkcast_copy("checkcast_arraycopy_uninit", NULL, iveresov@2606: /*dest_uninitialized*/true); iveresov@2606: iveresov@2595: StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy", iveresov@2595: entry_jbyte_arraycopy, iveresov@2595: entry_jshort_arraycopy, iveresov@2595: entry_jint_arraycopy, iveresov@2595: entry_jlong_arraycopy); iveresov@2595: StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy", iveresov@2595: entry_jbyte_arraycopy, iveresov@2595: entry_jshort_arraycopy, iveresov@2595: entry_jint_arraycopy, iveresov@2595: entry_oop_arraycopy, iveresov@2595: entry_jlong_arraycopy, iveresov@2595: entry_checkcast_arraycopy); duke@435: never@2118: StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill"); never@2118: StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill"); never@2118: StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill"); never@2118: StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill"); never@2118: StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill"); never@2118: StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill"); never@2118: duke@435: // We don't generate specialized code for HeapWord-aligned source duke@435: // arrays, so just use the code we've already generated duke@435: StubRoutines::_arrayof_jbyte_disjoint_arraycopy = StubRoutines::_jbyte_disjoint_arraycopy; duke@435: StubRoutines::_arrayof_jbyte_arraycopy = StubRoutines::_jbyte_arraycopy; duke@435: duke@435: StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy; duke@435: StubRoutines::_arrayof_jshort_arraycopy = StubRoutines::_jshort_arraycopy; duke@435: duke@435: StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy; duke@435: StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy; duke@435: duke@435: StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy; duke@435: StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy; duke@435: duke@435: StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy; duke@435: StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy; iveresov@2606: iveresov@2606: StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit; iveresov@2606: StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit; duke@435: } duke@435: never@1609: void generate_math_stubs() { never@1609: { never@1609: StubCodeMark mark(this, "StubRoutines", "log"); never@1609: StubRoutines::_intrinsic_log = (double (*)(double)) __ pc(); never@1609: never@1609: __ subq(rsp, 8); never@1609: __ movdbl(Address(rsp, 0), xmm0); never@1609: __ fld_d(Address(rsp, 0)); never@1609: __ flog(); never@1609: __ fstp_d(Address(rsp, 0)); never@1609: __ movdbl(xmm0, Address(rsp, 0)); never@1609: __ addq(rsp, 8); never@1609: __ ret(0); never@1609: } never@1609: { never@1609: StubCodeMark mark(this, "StubRoutines", "log10"); never@1609: StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc(); never@1609: never@1609: __ subq(rsp, 8); never@1609: __ movdbl(Address(rsp, 0), xmm0); never@1609: __ fld_d(Address(rsp, 0)); never@1609: __ flog10(); never@1609: __ fstp_d(Address(rsp, 0)); never@1609: __ movdbl(xmm0, Address(rsp, 0)); never@1609: __ addq(rsp, 8); never@1609: __ ret(0); never@1609: } never@1609: { never@1609: StubCodeMark mark(this, "StubRoutines", "sin"); never@1609: StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc(); never@1609: never@1609: __ subq(rsp, 8); never@1609: __ movdbl(Address(rsp, 0), xmm0); never@1609: __ fld_d(Address(rsp, 0)); never@1609: __ trigfunc('s'); never@1609: __ fstp_d(Address(rsp, 0)); never@1609: __ movdbl(xmm0, Address(rsp, 0)); never@1609: __ addq(rsp, 8); never@1609: __ ret(0); never@1609: } never@1609: { never@1609: StubCodeMark mark(this, "StubRoutines", "cos"); never@1609: StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc(); never@1609: never@1609: __ subq(rsp, 8); never@1609: __ movdbl(Address(rsp, 0), xmm0); never@1609: __ fld_d(Address(rsp, 0)); never@1609: __ trigfunc('c'); never@1609: __ fstp_d(Address(rsp, 0)); never@1609: __ movdbl(xmm0, Address(rsp, 0)); never@1609: __ addq(rsp, 8); never@1609: __ ret(0); never@1609: } never@1609: { never@1609: StubCodeMark mark(this, "StubRoutines", "tan"); never@1609: StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc(); never@1609: never@1609: __ subq(rsp, 8); never@1609: __ movdbl(Address(rsp, 0), xmm0); never@1609: __ fld_d(Address(rsp, 0)); never@1609: __ trigfunc('t'); never@1609: __ fstp_d(Address(rsp, 0)); never@1609: __ movdbl(xmm0, Address(rsp, 0)); never@1609: __ addq(rsp, 8); never@1609: __ ret(0); never@1609: } roland@3787: { roland@3787: StubCodeMark mark(this, "StubRoutines", "exp"); roland@3787: StubRoutines::_intrinsic_exp = (double (*)(double)) __ pc(); roland@3787: roland@3787: __ subq(rsp, 8); roland@3787: __ movdbl(Address(rsp, 0), xmm0); roland@3787: __ fld_d(Address(rsp, 0)); roland@3787: __ exp_with_fallback(0); roland@3787: __ fstp_d(Address(rsp, 0)); roland@3787: __ movdbl(xmm0, Address(rsp, 0)); roland@3787: __ addq(rsp, 8); roland@3787: __ ret(0); roland@3787: } roland@3787: { roland@3787: StubCodeMark mark(this, "StubRoutines", "pow"); roland@3787: StubRoutines::_intrinsic_pow = (double (*)(double,double)) __ pc(); roland@3787: roland@3787: __ subq(rsp, 8); roland@3787: __ movdbl(Address(rsp, 0), xmm1); roland@3787: __ fld_d(Address(rsp, 0)); roland@3787: __ movdbl(Address(rsp, 0), xmm0); roland@3787: __ fld_d(Address(rsp, 0)); roland@3787: __ pow_with_fallback(0); roland@3787: __ fstp_d(Address(rsp, 0)); roland@3787: __ movdbl(xmm0, Address(rsp, 0)); roland@3787: __ addq(rsp, 8); roland@3787: __ ret(0); roland@3787: } never@1609: } never@1609: kvn@4205: // AES intrinsic stubs kvn@4205: enum {AESBlockSize = 16}; kvn@4205: kvn@4205: address generate_key_shuffle_mask() { kvn@4205: __ align(16); kvn@4205: StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask"); kvn@4205: address start = __ pc(); kvn@4205: __ emit_data64( 0x0405060700010203, relocInfo::none ); kvn@4205: __ emit_data64( 0x0c0d0e0f08090a0b, relocInfo::none ); kvn@4205: return start; kvn@4205: } kvn@4205: kvn@4205: // Utility routine for loading a 128-bit key word in little endian format kvn@4205: // can optionally specify that the shuffle mask is already in an xmmregister kvn@4205: void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) { kvn@4205: __ movdqu(xmmdst, Address(key, offset)); kvn@4205: if (xmm_shuf_mask != NULL) { kvn@4205: __ pshufb(xmmdst, xmm_shuf_mask); kvn@4205: } else { kvn@4205: __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); kvn@4205: } kvn@4205: } kvn@4205: kvn@4205: // Arguments: kvn@4205: // kvn@4205: // Inputs: kvn@4205: // c_rarg0 - source byte array address kvn@4205: // c_rarg1 - destination byte array address kvn@4205: // c_rarg2 - K (key) in little endian int array kvn@4205: // kvn@4205: address generate_aescrypt_encryptBlock() { kvn@4363: assert(UseAES, "need AES instructions and misaligned SSE support"); kvn@4205: __ align(CodeEntryAlignment); kvn@4205: StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock"); kvn@4205: Label L_doLast; kvn@4205: address start = __ pc(); kvn@4205: kvn@4205: const Register from = c_rarg0; // source array address kvn@4205: const Register to = c_rarg1; // destination array address kvn@4205: const Register key = c_rarg2; // key array address kvn@4205: const Register keylen = rax; kvn@4205: kvn@4205: const XMMRegister xmm_result = xmm0; kvn@4363: const XMMRegister xmm_key_shuf_mask = xmm1; kvn@4363: // On win64 xmm6-xmm15 must be preserved so don't use them. kvn@4363: const XMMRegister xmm_temp1 = xmm2; kvn@4363: const XMMRegister xmm_temp2 = xmm3; kvn@4363: const XMMRegister xmm_temp3 = xmm4; kvn@4363: const XMMRegister xmm_temp4 = xmm5; kvn@4205: kvn@4205: __ enter(); // required for proper stackwalking of RuntimeStub frame kvn@4205: kvn@4363: // keylen could be only {11, 13, 15} * 4 = {44, 52, 60} kvn@4205: __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT))); kvn@4205: kvn@4205: __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); kvn@4205: __ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input kvn@4205: kvn@4205: // For encryption, the java expanded key ordering is just what we need kvn@4205: // we don't know if the key is aligned, hence not using load-execute form kvn@4205: kvn@4363: load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask); kvn@4363: __ pxor(xmm_result, xmm_temp1); kvn@4363: kvn@4363: load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask); kvn@4363: load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask); kvn@4363: load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask); kvn@4363: load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask); kvn@4363: kvn@4363: __ aesenc(xmm_result, xmm_temp1); kvn@4363: __ aesenc(xmm_result, xmm_temp2); kvn@4363: __ aesenc(xmm_result, xmm_temp3); kvn@4363: __ aesenc(xmm_result, xmm_temp4); kvn@4363: kvn@4363: load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask); kvn@4363: load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask); kvn@4363: load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask); kvn@4363: load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask); kvn@4363: kvn@4363: __ aesenc(xmm_result, xmm_temp1); kvn@4363: __ aesenc(xmm_result, xmm_temp2); kvn@4363: __ aesenc(xmm_result, xmm_temp3); kvn@4363: __ aesenc(xmm_result, xmm_temp4); kvn@4363: kvn@4363: load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask); kvn@4363: load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask); kvn@4363: kvn@4363: __ cmpl(keylen, 44); kvn@4363: __ jccb(Assembler::equal, L_doLast); kvn@4363: kvn@4363: __ aesenc(xmm_result, xmm_temp1); kvn@4363: __ aesenc(xmm_result, xmm_temp2); kvn@4363: kvn@4363: load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask); kvn@4363: load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask); kvn@4363: kvn@4363: __ cmpl(keylen, 52); kvn@4363: __ jccb(Assembler::equal, L_doLast); kvn@4363: kvn@4363: __ aesenc(xmm_result, xmm_temp1); kvn@4363: __ aesenc(xmm_result, xmm_temp2); kvn@4363: kvn@4363: load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask); kvn@4363: load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask); kvn@4205: kvn@4205: __ BIND(L_doLast); kvn@4363: __ aesenc(xmm_result, xmm_temp1); kvn@4363: __ aesenclast(xmm_result, xmm_temp2); kvn@4205: __ movdqu(Address(to, 0), xmm_result); // store the result kvn@4205: __ xorptr(rax, rax); // return 0 kvn@4205: __ leave(); // required for proper stackwalking of RuntimeStub frame kvn@4205: __ ret(0); kvn@4205: kvn@4205: return start; kvn@4205: } kvn@4205: kvn@4205: kvn@4205: // Arguments: kvn@4205: // kvn@4205: // Inputs: kvn@4205: // c_rarg0 - source byte array address kvn@4205: // c_rarg1 - destination byte array address kvn@4205: // c_rarg2 - K (key) in little endian int array kvn@4205: // kvn@4205: address generate_aescrypt_decryptBlock() { kvn@4363: assert(UseAES, "need AES instructions and misaligned SSE support"); kvn@4205: __ align(CodeEntryAlignment); kvn@4205: StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock"); kvn@4205: Label L_doLast; kvn@4205: address start = __ pc(); kvn@4205: kvn@4205: const Register from = c_rarg0; // source array address kvn@4205: const Register to = c_rarg1; // destination array address kvn@4205: const Register key = c_rarg2; // key array address kvn@4205: const Register keylen = rax; kvn@4205: kvn@4205: const XMMRegister xmm_result = xmm0; kvn@4363: const XMMRegister xmm_key_shuf_mask = xmm1; kvn@4363: // On win64 xmm6-xmm15 must be preserved so don't use them. kvn@4363: const XMMRegister xmm_temp1 = xmm2; kvn@4363: const XMMRegister xmm_temp2 = xmm3; kvn@4363: const XMMRegister xmm_temp3 = xmm4; kvn@4363: const XMMRegister xmm_temp4 = xmm5; kvn@4205: kvn@4205: __ enter(); // required for proper stackwalking of RuntimeStub frame kvn@4205: kvn@4363: // keylen could be only {11, 13, 15} * 4 = {44, 52, 60} kvn@4205: __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT))); kvn@4205: kvn@4205: __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); kvn@4205: __ movdqu(xmm_result, Address(from, 0)); kvn@4205: kvn@4205: // for decryption java expanded key ordering is rotated one position from what we want kvn@4205: // so we start from 0x10 here and hit 0x00 last kvn@4205: // we don't know if the key is aligned, hence not using load-execute form kvn@4363: load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask); kvn@4363: load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask); kvn@4363: load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask); kvn@4363: load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask); kvn@4363: kvn@4363: __ pxor (xmm_result, xmm_temp1); kvn@4363: __ aesdec(xmm_result, xmm_temp2); kvn@4363: __ aesdec(xmm_result, xmm_temp3); kvn@4363: __ aesdec(xmm_result, xmm_temp4); kvn@4363: kvn@4363: load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask); kvn@4363: load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask); kvn@4363: load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask); kvn@4363: load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask); kvn@4363: kvn@4363: __ aesdec(xmm_result, xmm_temp1); kvn@4363: __ aesdec(xmm_result, xmm_temp2); kvn@4363: __ aesdec(xmm_result, xmm_temp3); kvn@4363: __ aesdec(xmm_result, xmm_temp4); kvn@4363: kvn@4363: load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask); kvn@4363: load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask); kvn@4363: load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask); kvn@4363: kvn@4363: __ cmpl(keylen, 44); kvn@4363: __ jccb(Assembler::equal, L_doLast); kvn@4363: kvn@4363: __ aesdec(xmm_result, xmm_temp1); kvn@4363: __ aesdec(xmm_result, xmm_temp2); kvn@4363: kvn@4363: load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask); kvn@4363: load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask); kvn@4363: kvn@4363: __ cmpl(keylen, 52); kvn@4363: __ jccb(Assembler::equal, L_doLast); kvn@4363: kvn@4363: __ aesdec(xmm_result, xmm_temp1); kvn@4363: __ aesdec(xmm_result, xmm_temp2); kvn@4363: kvn@4363: load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask); kvn@4363: load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask); kvn@4205: kvn@4205: __ BIND(L_doLast); kvn@4363: __ aesdec(xmm_result, xmm_temp1); kvn@4363: __ aesdec(xmm_result, xmm_temp2); kvn@4363: kvn@4205: // for decryption the aesdeclast operation is always on key+0x00 kvn@4363: __ aesdeclast(xmm_result, xmm_temp3); kvn@4205: __ movdqu(Address(to, 0), xmm_result); // store the result kvn@4205: __ xorptr(rax, rax); // return 0 kvn@4205: __ leave(); // required for proper stackwalking of RuntimeStub frame kvn@4205: __ ret(0); kvn@4205: kvn@4205: return start; kvn@4205: } kvn@4205: kvn@4205: kvn@4205: // Arguments: kvn@4205: // kvn@4205: // Inputs: kvn@4205: // c_rarg0 - source byte array address kvn@4205: // c_rarg1 - destination byte array address kvn@4205: // c_rarg2 - K (key) in little endian int array kvn@4205: // c_rarg3 - r vector byte array address kvn@4205: // c_rarg4 - input length kvn@4205: // kvn@4205: address generate_cipherBlockChaining_encryptAESCrypt() { kvn@4363: assert(UseAES, "need AES instructions and misaligned SSE support"); kvn@4205: __ align(CodeEntryAlignment); kvn@4205: StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt"); kvn@4205: address start = __ pc(); kvn@4205: kvn@4205: Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256; kvn@4205: const Register from = c_rarg0; // source array address kvn@4205: const Register to = c_rarg1; // destination array address kvn@4205: const Register key = c_rarg2; // key array address kvn@4205: const Register rvec = c_rarg3; // r byte array initialized from initvector array address kvn@4205: // and left with the results of the last encryption block kvn@4205: #ifndef _WIN64 kvn@4205: const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16) kvn@4205: #else kvn@4205: const Address len_mem(rsp, 6 * wordSize); // length is on stack on Win64 kvn@4205: const Register len_reg = r10; // pick the first volatile windows register kvn@4205: #endif kvn@4205: const Register pos = rax; kvn@4205: kvn@4205: // xmm register assignments for the loops below kvn@4205: const XMMRegister xmm_result = xmm0; kvn@4205: const XMMRegister xmm_temp = xmm1; kvn@4205: // keys 0-10 preloaded into xmm2-xmm12 kvn@4205: const int XMM_REG_NUM_KEY_FIRST = 2; kvn@4363: const int XMM_REG_NUM_KEY_LAST = 15; kvn@4205: const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST); kvn@4363: const XMMRegister xmm_key10 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+10); kvn@4363: const XMMRegister xmm_key11 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+11); kvn@4363: const XMMRegister xmm_key12 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+12); kvn@4363: const XMMRegister xmm_key13 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+13); kvn@4205: kvn@4205: __ enter(); // required for proper stackwalking of RuntimeStub frame kvn@4205: kvn@4205: #ifdef _WIN64 kvn@4205: // on win64, fill len_reg from stack position kvn@4205: __ movl(len_reg, len_mem); kvn@4363: // save the xmm registers which must be preserved 6-15 kvn@4205: __ subptr(rsp, -rsp_after_call_off * wordSize); kvn@4205: for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) { kvn@4205: __ movdqu(xmm_save(i), as_XMMRegister(i)); kvn@4205: } kvn@4205: #endif kvn@4205: kvn@4205: const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front kvn@4205: __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); kvn@4363: // load up xmm regs xmm2 thru xmm12 with key 0x00 - 0xa0 kvn@4363: for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_FIRST+10; rnum++) { kvn@4205: load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask); kvn@4205: offset += 0x10; kvn@4205: } kvn@4205: __ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec kvn@4205: kvn@4205: // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256)) kvn@4205: __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT))); kvn@4205: __ cmpl(rax, 44); kvn@4205: __ jcc(Assembler::notEqual, L_key_192_256); kvn@4205: kvn@4205: // 128 bit code follows here kvn@4205: __ movptr(pos, 0); kvn@4205: __ align(OptoLoopAlignment); kvn@4363: kvn@4205: __ BIND(L_loopTop_128); kvn@4205: __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input kvn@4205: __ pxor (xmm_result, xmm_temp); // xor with the current r vector kvn@4205: __ pxor (xmm_result, xmm_key0); // do the aes rounds kvn@4363: for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 9; rnum++) { kvn@4205: __ aesenc(xmm_result, as_XMMRegister(rnum)); kvn@4205: } kvn@4205: __ aesenclast(xmm_result, xmm_key10); kvn@4205: __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output kvn@4205: // no need to store r to memory until we exit kvn@4205: __ addptr(pos, AESBlockSize); kvn@4205: __ subptr(len_reg, AESBlockSize); kvn@4205: __ jcc(Assembler::notEqual, L_loopTop_128); kvn@4205: kvn@4205: __ BIND(L_exit); kvn@4205: __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object kvn@4205: kvn@4205: #ifdef _WIN64 kvn@4205: // restore xmm regs belonging to calling function kvn@4205: for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) { kvn@4205: __ movdqu(as_XMMRegister(i), xmm_save(i)); kvn@4205: } kvn@4205: #endif kvn@4205: __ movl(rax, 0); // return 0 (why?) kvn@4205: __ leave(); // required for proper stackwalking of RuntimeStub frame kvn@4205: __ ret(0); kvn@4205: kvn@4205: __ BIND(L_key_192_256); kvn@4205: // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256) kvn@4363: load_key(xmm_key11, key, 0xb0, xmm_key_shuf_mask); kvn@4363: load_key(xmm_key12, key, 0xc0, xmm_key_shuf_mask); kvn@4205: __ cmpl(rax, 52); kvn@4205: __ jcc(Assembler::notEqual, L_key_256); kvn@4205: kvn@4205: // 192-bit code follows here (could be changed to use more xmm registers) kvn@4205: __ movptr(pos, 0); kvn@4205: __ align(OptoLoopAlignment); kvn@4363: kvn@4205: __ BIND(L_loopTop_192); kvn@4205: __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input kvn@4205: __ pxor (xmm_result, xmm_temp); // xor with the current r vector kvn@4205: __ pxor (xmm_result, xmm_key0); // do the aes rounds kvn@4363: for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 11; rnum++) { kvn@4205: __ aesenc(xmm_result, as_XMMRegister(rnum)); kvn@4205: } kvn@4363: __ aesenclast(xmm_result, xmm_key12); kvn@4205: __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output kvn@4205: // no need to store r to memory until we exit kvn@4205: __ addptr(pos, AESBlockSize); kvn@4205: __ subptr(len_reg, AESBlockSize); kvn@4205: __ jcc(Assembler::notEqual, L_loopTop_192); kvn@4205: __ jmp(L_exit); kvn@4205: kvn@4205: __ BIND(L_key_256); kvn@4205: // 256-bit code follows here (could be changed to use more xmm registers) kvn@4363: load_key(xmm_key13, key, 0xd0, xmm_key_shuf_mask); kvn@4205: __ movptr(pos, 0); kvn@4205: __ align(OptoLoopAlignment); kvn@4363: kvn@4205: __ BIND(L_loopTop_256); kvn@4205: __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input kvn@4205: __ pxor (xmm_result, xmm_temp); // xor with the current r vector kvn@4205: __ pxor (xmm_result, xmm_key0); // do the aes rounds kvn@4363: for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 13; rnum++) { kvn@4205: __ aesenc(xmm_result, as_XMMRegister(rnum)); kvn@4205: } kvn@4205: load_key(xmm_temp, key, 0xe0); kvn@4205: __ aesenclast(xmm_result, xmm_temp); kvn@4205: __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output kvn@4205: // no need to store r to memory until we exit kvn@4205: __ addptr(pos, AESBlockSize); kvn@4205: __ subptr(len_reg, AESBlockSize); kvn@4205: __ jcc(Assembler::notEqual, L_loopTop_256); kvn@4205: __ jmp(L_exit); kvn@4205: kvn@4205: return start; kvn@4205: } kvn@4205: kvn@4205: kvn@4205: kvn@4205: // This is a version of CBC/AES Decrypt which does 4 blocks in a loop at a time kvn@4205: // to hide instruction latency kvn@4205: // kvn@4205: // Arguments: kvn@4205: // kvn@4205: // Inputs: kvn@4205: // c_rarg0 - source byte array address kvn@4205: // c_rarg1 - destination byte array address kvn@4205: // c_rarg2 - K (key) in little endian int array kvn@4205: // c_rarg3 - r vector byte array address kvn@4205: // c_rarg4 - input length kvn@4205: // kvn@4205: kvn@4205: address generate_cipherBlockChaining_decryptAESCrypt_Parallel() { kvn@4363: assert(UseAES, "need AES instructions and misaligned SSE support"); kvn@4205: __ align(CodeEntryAlignment); kvn@4205: StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt"); kvn@4205: address start = __ pc(); kvn@4205: kvn@4205: Label L_exit, L_key_192_256, L_key_256; kvn@4205: Label L_singleBlock_loopTop_128, L_multiBlock_loopTop_128; kvn@4205: Label L_singleBlock_loopTop_192, L_singleBlock_loopTop_256; kvn@4205: const Register from = c_rarg0; // source array address kvn@4205: const Register to = c_rarg1; // destination array address kvn@4205: const Register key = c_rarg2; // key array address kvn@4205: const Register rvec = c_rarg3; // r byte array initialized from initvector array address kvn@4205: // and left with the results of the last encryption block kvn@4205: #ifndef _WIN64 kvn@4205: const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16) kvn@4205: #else kvn@4205: const Address len_mem(rsp, 6 * wordSize); // length is on stack on Win64 kvn@4205: const Register len_reg = r10; // pick the first volatile windows register kvn@4205: #endif kvn@4205: const Register pos = rax; kvn@4205: kvn@4205: // keys 0-10 preloaded into xmm2-xmm12 kvn@4205: const int XMM_REG_NUM_KEY_FIRST = 5; kvn@4205: const int XMM_REG_NUM_KEY_LAST = 15; kvn@4363: const XMMRegister xmm_key_first = as_XMMRegister(XMM_REG_NUM_KEY_FIRST); kvn@4205: const XMMRegister xmm_key_last = as_XMMRegister(XMM_REG_NUM_KEY_LAST); kvn@4205: kvn@4205: __ enter(); // required for proper stackwalking of RuntimeStub frame kvn@4205: kvn@4205: #ifdef _WIN64 kvn@4205: // on win64, fill len_reg from stack position kvn@4205: __ movl(len_reg, len_mem); kvn@4205: // save the xmm registers which must be preserved 6-15 kvn@4205: __ subptr(rsp, -rsp_after_call_off * wordSize); kvn@4205: for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) { kvn@4205: __ movdqu(xmm_save(i), as_XMMRegister(i)); kvn@4205: } kvn@4205: #endif kvn@4205: // the java expanded key ordering is rotated one position from what we want kvn@4205: // so we start from 0x10 here and hit 0x00 last kvn@4205: const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front kvn@4205: __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); kvn@4205: // load up xmm regs 5 thru 15 with key 0x10 - 0xa0 - 0x00 kvn@4363: for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x10; rnum < XMM_REG_NUM_KEY_LAST; rnum++) { kvn@4205: load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask); kvn@4205: offset += 0x10; kvn@4205: } kvn@4363: load_key(xmm_key_last, key, 0x00, xmm_key_shuf_mask); kvn@4205: kvn@4205: const XMMRegister xmm_prev_block_cipher = xmm1; // holds cipher of previous block kvn@4363: kvn@4205: // registers holding the four results in the parallelized loop kvn@4205: const XMMRegister xmm_result0 = xmm0; kvn@4205: const XMMRegister xmm_result1 = xmm2; kvn@4205: const XMMRegister xmm_result2 = xmm3; kvn@4205: const XMMRegister xmm_result3 = xmm4; kvn@4205: kvn@4205: __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // initialize with initial rvec kvn@4205: kvn@4205: // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256)) kvn@4205: __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT))); kvn@4205: __ cmpl(rax, 44); kvn@4205: __ jcc(Assembler::notEqual, L_key_192_256); kvn@4205: kvn@4205: kvn@4205: // 128-bit code follows here, parallelized kvn@4205: __ movptr(pos, 0); kvn@4205: __ align(OptoLoopAlignment); kvn@4205: __ BIND(L_multiBlock_loopTop_128); kvn@4205: __ cmpptr(len_reg, 4*AESBlockSize); // see if at least 4 blocks left kvn@4205: __ jcc(Assembler::less, L_singleBlock_loopTop_128); kvn@4205: kvn@4205: __ movdqu(xmm_result0, Address(from, pos, Address::times_1, 0*AESBlockSize)); // get next 4 blocks into xmmresult registers kvn@4205: __ movdqu(xmm_result1, Address(from, pos, Address::times_1, 1*AESBlockSize)); kvn@4205: __ movdqu(xmm_result2, Address(from, pos, Address::times_1, 2*AESBlockSize)); kvn@4205: __ movdqu(xmm_result3, Address(from, pos, Address::times_1, 3*AESBlockSize)); kvn@4205: kvn@4205: #define DoFour(opc, src_reg) \ kvn@4205: __ opc(xmm_result0, src_reg); \ kvn@4205: __ opc(xmm_result1, src_reg); \ kvn@4205: __ opc(xmm_result2, src_reg); \ kvn@4205: __ opc(xmm_result3, src_reg); kvn@4205: kvn@4205: DoFour(pxor, xmm_key_first); kvn@4205: for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST - 1; rnum++) { kvn@4205: DoFour(aesdec, as_XMMRegister(rnum)); kvn@4205: } kvn@4205: DoFour(aesdeclast, xmm_key_last); kvn@4205: // for each result, xor with the r vector of previous cipher block kvn@4205: __ pxor(xmm_result0, xmm_prev_block_cipher); kvn@4205: __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 0*AESBlockSize)); kvn@4205: __ pxor(xmm_result1, xmm_prev_block_cipher); kvn@4205: __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 1*AESBlockSize)); kvn@4205: __ pxor(xmm_result2, xmm_prev_block_cipher); kvn@4205: __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 2*AESBlockSize)); kvn@4205: __ pxor(xmm_result3, xmm_prev_block_cipher); kvn@4205: __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 3*AESBlockSize)); // this will carry over to next set of blocks kvn@4205: kvn@4205: __ movdqu(Address(to, pos, Address::times_1, 0*AESBlockSize), xmm_result0); // store 4 results into the next 64 bytes of output kvn@4205: __ movdqu(Address(to, pos, Address::times_1, 1*AESBlockSize), xmm_result1); kvn@4205: __ movdqu(Address(to, pos, Address::times_1, 2*AESBlockSize), xmm_result2); kvn@4205: __ movdqu(Address(to, pos, Address::times_1, 3*AESBlockSize), xmm_result3); kvn@4205: kvn@4205: __ addptr(pos, 4*AESBlockSize); kvn@4205: __ subptr(len_reg, 4*AESBlockSize); kvn@4205: __ jmp(L_multiBlock_loopTop_128); kvn@4205: kvn@4205: // registers used in the non-parallelized loops kvn@4363: // xmm register assignments for the loops below kvn@4363: const XMMRegister xmm_result = xmm0; kvn@4205: const XMMRegister xmm_prev_block_cipher_save = xmm2; kvn@4363: const XMMRegister xmm_key11 = xmm3; kvn@4363: const XMMRegister xmm_key12 = xmm4; kvn@4363: const XMMRegister xmm_temp = xmm4; kvn@4205: kvn@4205: __ align(OptoLoopAlignment); kvn@4205: __ BIND(L_singleBlock_loopTop_128); kvn@4205: __ cmpptr(len_reg, 0); // any blocks left?? kvn@4205: __ jcc(Assembler::equal, L_exit); kvn@4205: __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input kvn@4205: __ movdqa(xmm_prev_block_cipher_save, xmm_result); // save for next r vector kvn@4205: __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds kvn@4205: for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST - 1; rnum++) { kvn@4205: __ aesdec(xmm_result, as_XMMRegister(rnum)); kvn@4205: } kvn@4205: __ aesdeclast(xmm_result, xmm_key_last); kvn@4205: __ pxor (xmm_result, xmm_prev_block_cipher); // xor with the current r vector kvn@4205: __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output kvn@4205: // no need to store r to memory until we exit kvn@4205: __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save); // set up next r vector with cipher input from this block kvn@4205: kvn@4205: __ addptr(pos, AESBlockSize); kvn@4205: __ subptr(len_reg, AESBlockSize); kvn@4205: __ jmp(L_singleBlock_loopTop_128); kvn@4205: kvn@4205: kvn@4205: __ BIND(L_exit); kvn@4205: __ movdqu(Address(rvec, 0), xmm_prev_block_cipher); // final value of r stored in rvec of CipherBlockChaining object kvn@4205: #ifdef _WIN64 kvn@4205: // restore regs belonging to calling function kvn@4205: for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) { kvn@4205: __ movdqu(as_XMMRegister(i), xmm_save(i)); kvn@4205: } kvn@4205: #endif kvn@4205: __ movl(rax, 0); // return 0 (why?) kvn@4205: __ leave(); // required for proper stackwalking of RuntimeStub frame kvn@4205: __ ret(0); kvn@4205: kvn@4205: kvn@4205: __ BIND(L_key_192_256); kvn@4205: // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256) kvn@4363: load_key(xmm_key11, key, 0xb0); kvn@4205: __ cmpl(rax, 52); kvn@4205: __ jcc(Assembler::notEqual, L_key_256); kvn@4205: kvn@4205: // 192-bit code follows here (could be optimized to use parallelism) kvn@4363: load_key(xmm_key12, key, 0xc0); // 192-bit key goes up to c0 kvn@4205: __ movptr(pos, 0); kvn@4205: __ align(OptoLoopAlignment); kvn@4363: kvn@4205: __ BIND(L_singleBlock_loopTop_192); kvn@4205: __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input kvn@4205: __ movdqa(xmm_prev_block_cipher_save, xmm_result); // save for next r vector kvn@4205: __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds kvn@4205: for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST - 1; rnum++) { kvn@4205: __ aesdec(xmm_result, as_XMMRegister(rnum)); kvn@4205: } kvn@4363: __ aesdec(xmm_result, xmm_key11); kvn@4363: __ aesdec(xmm_result, xmm_key12); kvn@4205: __ aesdeclast(xmm_result, xmm_key_last); // xmm15 always came from key+0 kvn@4205: __ pxor (xmm_result, xmm_prev_block_cipher); // xor with the current r vector kvn@4363: __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output kvn@4205: // no need to store r to memory until we exit kvn@4363: __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save); // set up next r vector with cipher input from this block kvn@4205: __ addptr(pos, AESBlockSize); kvn@4205: __ subptr(len_reg, AESBlockSize); kvn@4205: __ jcc(Assembler::notEqual,L_singleBlock_loopTop_192); kvn@4205: __ jmp(L_exit); kvn@4205: kvn@4205: __ BIND(L_key_256); kvn@4205: // 256-bit code follows here (could be optimized to use parallelism) kvn@4205: __ movptr(pos, 0); kvn@4205: __ align(OptoLoopAlignment); kvn@4363: kvn@4205: __ BIND(L_singleBlock_loopTop_256); kvn@4363: __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input kvn@4205: __ movdqa(xmm_prev_block_cipher_save, xmm_result); // save for next r vector kvn@4205: __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds kvn@4205: for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST - 1; rnum++) { kvn@4205: __ aesdec(xmm_result, as_XMMRegister(rnum)); kvn@4205: } kvn@4363: __ aesdec(xmm_result, xmm_key11); kvn@4363: load_key(xmm_temp, key, 0xc0); kvn@4363: __ aesdec(xmm_result, xmm_temp); kvn@4363: load_key(xmm_temp, key, 0xd0); kvn@4363: __ aesdec(xmm_result, xmm_temp); kvn@4363: load_key(xmm_temp, key, 0xe0); // 256-bit key goes up to e0 kvn@4363: __ aesdec(xmm_result, xmm_temp); kvn@4363: __ aesdeclast(xmm_result, xmm_key_last); // xmm15 came from key+0 kvn@4205: __ pxor (xmm_result, xmm_prev_block_cipher); // xor with the current r vector kvn@4363: __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output kvn@4205: // no need to store r to memory until we exit kvn@4363: __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save); // set up next r vector with cipher input from this block kvn@4205: __ addptr(pos, AESBlockSize); kvn@4205: __ subptr(len_reg, AESBlockSize); kvn@4205: __ jcc(Assembler::notEqual,L_singleBlock_loopTop_256); kvn@4205: __ jmp(L_exit); kvn@4205: kvn@4205: return start; kvn@4205: } kvn@4205: kvn@4205: kvn@4205: duke@435: #undef __ duke@435: #define __ masm-> duke@435: duke@435: // Continuation point for throwing of implicit exceptions that are duke@435: // not handled in the current activation. Fabricates an exception duke@435: // oop and initiates normal exception dispatching in this duke@435: // frame. Since we need to preserve callee-saved values (currently duke@435: // only for C2, but done for C1 as well) we need a callee-saved oop duke@435: // map and therefore have to make these stubs into RuntimeStubs duke@435: // rather than BufferBlobs. If the compiler needs all registers to duke@435: // be preserved between the fault point and the exception handler duke@435: // then it must assume responsibility for that in duke@435: // AbstractCompiler::continuation_for_implicit_null_exception or duke@435: // continuation_for_implicit_division_by_zero_exception. All other duke@435: // implicit exceptions (e.g., NullPointerException or duke@435: // AbstractMethodError on entry) are either at call sites or duke@435: // otherwise assume that stack unwinding will be initiated, so duke@435: // caller saved registers were assumed volatile in the compiler. duke@435: address generate_throw_exception(const char* name, duke@435: address runtime_entry, never@2978: Register arg1 = noreg, never@2978: Register arg2 = noreg) { duke@435: // Information about frame layout at time of blocking runtime call. duke@435: // Note that we only have to preserve callee-saved registers since duke@435: // the compilers are responsible for supplying a continuation point duke@435: // if they expect all registers to be preserved. duke@435: enum layout { duke@435: rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt, duke@435: rbp_off2, duke@435: return_off, duke@435: return_off2, duke@435: framesize // inclusive of return address duke@435: }; duke@435: duke@435: int insts_size = 512; duke@435: int locs_size = 64; duke@435: duke@435: CodeBuffer code(name, insts_size, locs_size); duke@435: OopMapSet* oop_maps = new OopMapSet(); duke@435: MacroAssembler* masm = new MacroAssembler(&code); duke@435: duke@435: address start = __ pc(); duke@435: duke@435: // This is an inlined and slightly modified version of call_VM duke@435: // which has the ability to fetch the return PC out of duke@435: // thread-local storage and also sets up last_Java_sp slightly duke@435: // differently than the real call_VM duke@435: duke@435: __ enter(); // required for proper stackwalking of RuntimeStub frame duke@435: duke@435: assert(is_even(framesize/2), "sp not 16-byte aligned"); duke@435: duke@435: // return address and rbp are already in place never@739: __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog duke@435: duke@435: int frame_complete = __ pc() - start; duke@435: duke@435: // Set up last_Java_sp and last_Java_fp roland@3522: address the_pc = __ pc(); roland@3522: __ set_last_Java_frame(rsp, rbp, the_pc); roland@3522: __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack duke@435: duke@435: // Call runtime never@2978: if (arg1 != noreg) { never@2978: assert(arg2 != c_rarg1, "clobbered"); never@2978: __ movptr(c_rarg1, arg1); never@2978: } never@2978: if (arg2 != noreg) { never@2978: __ movptr(c_rarg2, arg2); never@2978: } never@739: __ movptr(c_rarg0, r15_thread); duke@435: BLOCK_COMMENT("call runtime_entry"); duke@435: __ call(RuntimeAddress(runtime_entry)); duke@435: duke@435: // Generate oop map duke@435: OopMap* map = new OopMap(framesize, 0); duke@435: roland@3568: oop_maps->add_gc_map(the_pc - start, map); duke@435: roland@3522: __ reset_last_Java_frame(true, true); duke@435: duke@435: __ leave(); // required for proper stackwalking of RuntimeStub frame duke@435: duke@435: // check for pending exceptions duke@435: #ifdef ASSERT duke@435: Label L; never@739: __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), never@739: (int32_t) NULL_WORD); duke@435: __ jcc(Assembler::notEqual, L); duke@435: __ should_not_reach_here(); duke@435: __ bind(L); duke@435: #endif // ASSERT duke@435: __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); duke@435: duke@435: duke@435: // codeBlob framesize is in words (not VMRegImpl::slot_size) duke@435: RuntimeStub* stub = duke@435: RuntimeStub::new_runtime_stub(name, duke@435: &code, duke@435: frame_complete, duke@435: (framesize >> (LogBytesPerWord - LogBytesPerInt)), duke@435: oop_maps, false); duke@435: return stub->entry_point(); duke@435: } duke@435: duke@435: // Initialization duke@435: void generate_initial() { duke@435: // Generates all stubs and initializes the entry points duke@435: duke@435: // This platform-specific stub is needed by generate_call_stub() never@739: StubRoutines::x86::_mxcsr_std = generate_fp_mask("mxcsr_std", 0x0000000000001F80); duke@435: duke@435: // entry points that exist in all platforms Note: This is code duke@435: // that could be shared among different platforms - however the duke@435: // benefit seems to be smaller than the disadvantage of having a duke@435: // much more complicated generator structure. See also comment in duke@435: // stubRoutines.hpp. duke@435: duke@435: StubRoutines::_forward_exception_entry = generate_forward_exception(); duke@435: duke@435: StubRoutines::_call_stub_entry = duke@435: generate_call_stub(StubRoutines::_call_stub_return_address); duke@435: duke@435: // is referenced by megamorphic call duke@435: StubRoutines::_catch_exception_entry = generate_catch_exception(); duke@435: duke@435: // atomic calls duke@435: StubRoutines::_atomic_xchg_entry = generate_atomic_xchg(); duke@435: StubRoutines::_atomic_xchg_ptr_entry = generate_atomic_xchg_ptr(); duke@435: StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg(); duke@435: StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long(); duke@435: StubRoutines::_atomic_add_entry = generate_atomic_add(); duke@435: StubRoutines::_atomic_add_ptr_entry = generate_atomic_add_ptr(); duke@435: StubRoutines::_fence_entry = generate_orderaccess_fence(); duke@435: duke@435: StubRoutines::_handler_for_unsafe_access_entry = duke@435: generate_handler_for_unsafe_access(); duke@435: duke@435: // platform dependent never@739: StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp(); roland@3606: StubRoutines::x86::_get_previous_sp_entry = generate_get_previous_sp(); never@739: never@739: StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr(); never@2978: bdelsart@3372: // Build this early so it's available for the interpreter. bdelsart@3372: StubRoutines::_throw_StackOverflowError_entry = bdelsart@3372: generate_throw_exception("StackOverflowError throw_exception", bdelsart@3372: CAST_FROM_FN_PTR(address, bdelsart@3372: SharedRuntime:: bdelsart@3372: throw_StackOverflowError)); duke@435: } duke@435: duke@435: void generate_all() { duke@435: // Generates all stubs and initializes the entry points duke@435: duke@435: // These entry points require SharedInfo::stack0 to be set up in duke@435: // non-core builds and need to be relocatable, so they each duke@435: // fabricate a RuntimeStub internally. duke@435: StubRoutines::_throw_AbstractMethodError_entry = duke@435: generate_throw_exception("AbstractMethodError throw_exception", duke@435: CAST_FROM_FN_PTR(address, duke@435: SharedRuntime:: never@3136: throw_AbstractMethodError)); duke@435: dcubed@451: StubRoutines::_throw_IncompatibleClassChangeError_entry = dcubed@451: generate_throw_exception("IncompatibleClassChangeError throw_exception", dcubed@451: CAST_FROM_FN_PTR(address, dcubed@451: SharedRuntime:: never@3136: throw_IncompatibleClassChangeError)); duke@435: duke@435: StubRoutines::_throw_NullPointerException_at_call_entry = duke@435: generate_throw_exception("NullPointerException at call throw_exception", duke@435: CAST_FROM_FN_PTR(address, duke@435: SharedRuntime:: never@3136: throw_NullPointerException_at_call)); duke@435: duke@435: // entry points that are platform specific never@739: StubRoutines::x86::_f2i_fixup = generate_f2i_fixup(); never@739: StubRoutines::x86::_f2l_fixup = generate_f2l_fixup(); never@739: StubRoutines::x86::_d2i_fixup = generate_d2i_fixup(); never@739: StubRoutines::x86::_d2l_fixup = generate_d2l_fixup(); never@739: never@739: StubRoutines::x86::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF); never@739: StubRoutines::x86::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000); never@739: StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF); never@739: StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000); duke@435: duke@435: // support for verify_oop (must happen after universe_init) duke@435: StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop(); duke@435: duke@435: // arraycopy stubs used by compilers duke@435: generate_arraycopy_stubs(); twisti@1543: never@1609: generate_math_stubs(); kvn@4205: kvn@4205: // don't bother generating these AES intrinsic stubs unless global flag is set kvn@4205: if (UseAESIntrinsics) { kvn@4205: StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // needed by the others kvn@4205: kvn@4205: StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock(); kvn@4205: StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock(); kvn@4205: StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt(); kvn@4205: StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt_Parallel(); kvn@4205: } duke@435: } duke@435: duke@435: public: duke@435: StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) { duke@435: if (all) { duke@435: generate_all(); duke@435: } else { duke@435: generate_initial(); duke@435: } duke@435: } duke@435: }; // end class declaration duke@435: duke@435: void StubGenerator_generate(CodeBuffer* code, bool all) { duke@435: StubGenerator g(code, all); duke@435: }