--- old/src/cpu/x86/vm/stubGenerator_x86_64.cpp Fri May 31 16:03:59 2013 +++ new/src/cpu/x86/vm/stubGenerator_x86_64.cpp Fri May 31 16:03:58 2013 @@ -1245,27 +1245,28 @@ // // Input: // start - register containing starting address of destination array - // end - register containing ending address of destination array + // count - elements count // scratch - scratch register // // The input registers are overwritten. - // The ending address is inclusive. - void gen_write_ref_array_post_barrier(Register start, Register end, Register scratch) { - assert_different_registers(start, end, scratch); + // + void gen_write_ref_array_post_barrier(Register start, Register count, Register scratch) { + assert_different_registers(start, count, scratch); BarrierSet* bs = Universe::heap()->barrier_set(); switch (bs->kind()) { case BarrierSet::G1SATBCT: case BarrierSet::G1SATBCTLogging: - { - __ pusha(); // push registers (overkill) - // must compute element count unless barrier set interface is changed (other platforms supply count) - assert_different_registers(start, end, scratch); - __ lea(scratch, Address(end, BytesPerHeapOop)); - __ subptr(scratch, start); // subtract start to get #bytes - __ shrptr(scratch, LogBytesPerHeapOop); // convert to element count - __ mov(c_rarg0, start); - __ mov(c_rarg1, scratch); + __ pusha(); // push registers (overkill) + if (c_rarg0 == count) { // On win64 c_rarg0 == rcx + assert_different_registers(c_rarg1, start); + __ mov(c_rarg1, count); + __ mov(c_rarg0, start); + } else { + assert_different_registers(c_rarg0, count); + __ mov(c_rarg0, start); + __ mov(c_rarg1, count); + } __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post), 2); __ popa(); } @@ -1277,22 +1278,16 @@ assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code"); Label L_loop; + const Register end = count; - __ shrptr(start, CardTableModRefBS::card_shift); - __ addptr(end, BytesPerHeapOop); - __ shrptr(end, CardTableModRefBS::card_shift); - __ subptr(end, start); // number of bytes to copy + __ leaq(end, Address(start, count, TIMES_OOP, 0)); // end == start+count*oop_size + __ subptr(end, BytesPerHeapOop); // end - 1 to make inclusive + __ shrptr(start, CardTableModRefBS::card_shift); + __ shrptr(end, CardTableModRefBS::card_shift); + __ subptr(end, start); // end --> cards count - intptr_t disp = (intptr_t) ct->byte_map_base; - if (Assembler::is_simm32(disp)) { - Address cardtable(noreg, noreg, Address::no_scale, disp); - __ lea(scratch, cardtable); - } else { - ExternalAddress cardtable((address)disp); - __ lea(scratch, cardtable); - } - - const Register count = end; // 'end' register contains bytes count now + int64_t disp = (int64_t) ct->byte_map_base; + __ mov64(scratch, disp); __ addptr(start, scratch); __ BIND(L_loop); __ movb(Address(start, count, Address::times_1), 0); @@ -1944,8 +1939,7 @@ __ BIND(L_exit); if (is_oop) { - __ leaq(end_to, Address(saved_to, dword_count, Address::times_4, -4)); - gen_write_ref_array_post_barrier(saved_to, end_to, rax); + gen_write_ref_array_post_barrier(saved_to, dword_count, rax); } restore_arg_regs(); inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free @@ -2040,12 +2034,10 @@ // Copy in multi-bytes chunks copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes); - __ bind(L_exit); - if (is_oop) { - Register end_to = rdx; - __ leaq(end_to, Address(to, dword_count, Address::times_4, -4)); - gen_write_ref_array_post_barrier(to, end_to, rax); - } + __ BIND(L_exit); + if (is_oop) { + gen_write_ref_array_post_barrier(to, dword_count, rax); + } restore_arg_regs(); inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free __ xorptr(rax, rax); // return 0 @@ -2083,6 +2075,7 @@ const Register end_from = from; // source array end address const Register end_to = rcx; // destination array end address const Register saved_to = to; + const Register saved_count = r11; // End pointers are inclusive, and if count is not zero they point // to the last unit copied: end_to[0] := end_from[0] @@ -2100,6 +2093,8 @@ // r9 and r10 may be used to save non-volatile registers // 'from', 'to' and 'qword_count' are now valid if (is_oop) { + // Save to and count for store barrier + __ movptr(saved_count, qword_count); // no registers are destroyed by this call gen_write_ref_array_pre_barrier(to, qword_count, dest_uninitialized); } @@ -2132,7 +2127,7 @@ if (is_oop) { __ BIND(L_exit); - gen_write_ref_array_post_barrier(saved_to, end_to, rax); + gen_write_ref_array_post_barrier(saved_to, saved_count, rax); } restore_arg_regs(); if (is_oop) { @@ -2215,8 +2210,7 @@ if (is_oop) { __ BIND(L_exit); - __ lea(rcx, Address(to, saved_count, Address::times_8, -8)); - gen_write_ref_array_post_barrier(to, rcx, rax); + gen_write_ref_array_post_barrier(to, saved_count, rax); } restore_arg_regs(); if (is_oop) { @@ -2403,21 +2397,21 @@ // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops. // Emit GC store barriers for the oops we have copied (r14 + rdx), // and report their number to the caller. - assert_different_registers(rax, r14_length, count, to, end_to, rcx); - __ lea(end_to, to_element_addr); - __ addptr(end_to, -heapOopSize); // make an inclusive end pointer - gen_write_ref_array_post_barrier(to, end_to, rscratch1); - __ movptr(rax, r14_length); // original oops - __ addptr(rax, count); // K = (original - remaining) oops - __ notptr(rax); // report (-1^K) to caller - __ jmp(L_done); + assert_different_registers(rax, r14_length, count, to, end_to, rcx, rscratch1); + Label L_post_barrier; + __ addptr(r14_length, count); // K = (original - remaining) oops + __ movptr(rax, r14_length); // save the value + __ notptr(rax); // report (-1^K) to caller (does not affect flags) + __ jccb(Assembler::notZero, L_post_barrier); + __ jmp(L_done); // K == 0, nothing was copied, skip post barrier // Come here on success only. __ BIND(L_do_card_marks); - __ addptr(end_to, -heapOopSize); // make an inclusive end pointer - gen_write_ref_array_post_barrier(to, end_to, rscratch1); - __ xorptr(rax, rax); // return 0 on success + __ xorptr(rax, rax); // return 0 on success + __ BIND(L_post_barrier); + gen_write_ref_array_post_barrier(to, r14_length, rscratch1); + // Common exit point (success or failure). __ BIND(L_done); __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));