stan/jdk-24
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

8296170: Refactor stack-locking path in C2_MacroAssembler::fast_unlock()

Browse Source 
Reviewed-by: thartmann, phh
This commit is contained in:
Roman Kennke 2022-11-17 08:23:57 +00:00
parent 502fa3eeea
commit e81359f148
1 changed files with 5 additions and 29 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

34
src/hotspot/cpu/x86/c2_MacroAssembler_x86.cpp
View File

@ -704,11 +704,6 @@ void C2_MacroAssembler::fast_lock(Register objReg, Register boxReg, Register tmp
#if INCLUDE_RTM_OPT
} // use_rtm()
#endif
// DONE_LABEL is a hot target - we'd really like to place it at the
// start of cache line by padding with NOPs.
// See the AMD and Intel software optimization manuals for the
// most efficient "long" NOP encodings.
// Unfortunately none of our alignment mechanisms suffice.
bind(DONE_LABEL);
// ZFlag == 1 count in fast path
@ -770,7 +765,7 @@ void C2_MacroAssembler::fast_unlock(Register objReg, Register boxReg, Register t
assert(boxReg == rax, "");
assert_different_registers(objReg, boxReg, tmpReg);
Label DONE_LABEL, Stacked, CheckSucc, COUNT, NO_COUNT;
Label DONE_LABEL, Stacked, COUNT, NO_COUNT;
#if INCLUDE_RTM_OPT
if (UseRTMForStackLocks && use_rtm) {
@ -829,8 +824,6 @@ void C2_MacroAssembler::fast_unlock(Register objReg, Register boxReg, Register t
// each other and there's no need for an explicit barrier (fence).
// See also http://gee.cs.oswego.edu/dl/jmm/cookbook.html.
#ifndef _LP64
get_thread (boxReg);
// Note that we could employ various encoding schemes to reduce
// the number of loads below (currently 4) to just 2 or 3.
// Refer to the comments in synchronizer.cpp.
@ -840,30 +833,12 @@ void C2_MacroAssembler::fast_unlock(Register objReg, Register boxReg, Register t
jccb (Assembler::notZero, DONE_LABEL);
movptr(boxReg, Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(EntryList)));
orptr(boxReg, Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(cxq)));
jccb (Assembler::notZero, CheckSucc);
jccb (Assembler::notZero, DONE_LABEL);
movptr(Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(owner)), NULL_WORD);
jmpb (DONE_LABEL);
bind (Stacked);
// It's not inflated and it's not recursively stack-locked.
// It must be stack-locked.
// Try to reset the header to displaced header.
// The "box" value on the stack is stable, so we can reload
// and be assured we observe the same value as above.
movptr(tmpReg, Address(boxReg, 0));
lock();
cmpxchgptr(tmpReg, Address(objReg, oopDesc::mark_offset_in_bytes())); // Uses RAX which is box
// Intention fall-thru into DONE_LABEL
// DONE_LABEL is a hot target - we'd really like to place it at the
// start of cache line by padding with NOPs.
// See the AMD and Intel software optimization manuals for the
// most efficient "long" NOP encodings.
// Unfortunately none of our alignment mechanisms suffice.
bind (CheckSucc);
#else // _LP64
// It's inflated
Label LNotRecursive, LSuccess, LGoSlowPath;
Label CheckSucc, LNotRecursive, LSuccess, LGoSlowPath;
cmpptr(Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(recursions)), 0);
jccb(Assembler::equal, LNotRecursive);
@ -936,13 +911,14 @@ void C2_MacroAssembler::fast_unlock(Register objReg, Register boxReg, Register t
testl (boxReg, 0); // set ICC.ZF=1 to indicate success
jmpb (DONE_LABEL);
#endif
if (!UseHeavyMonitors) {
bind (Stacked);
movptr(tmpReg, Address (boxReg, 0)); // re-fetch
lock();
cmpxchgptr(tmpReg, Address(objReg, oopDesc::mark_offset_in_bytes())); // Uses RAX which is box
// Intentional fall-thru into DONE_LABEL
}
#endif
bind(DONE_LABEL);
// ZFlag == 1 count in fast path