8065618: C2 RA incorrectly removes kill projections
Summary: Don't remove KILL projections if their "defining" nodes have SCMemProj projection (memory side effects).
Reviewed-by: iveresov
/*
* Copyright (c) 1999, 2013, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License version 2 only, as
* published by the Free Software Foundation.
*
* This code is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
* version 2 for more details (a copy is included in the LICENSE file that
* accompanied this code).
*
* You should have received a copy of the GNU General Public License version
* 2 along with this work; if not, write to the Free Software Foundation,
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
*
* Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
* or visit www.oracle.com if you need additional information or have any
* questions.
*
*/
#include "precompiled.hpp"
#include "opto/addnode.hpp"
#include "opto/callnode.hpp"
#include "opto/cfgnode.hpp"
#include "opto/compile.hpp"
#include "opto/convertnode.hpp"
#include "opto/locknode.hpp"
#include "opto/memnode.hpp"
#include "opto/mulnode.hpp"
#include "opto/node.hpp"
#include "opto/parse.hpp"
#include "opto/phaseX.hpp"
#include "opto/rootnode.hpp"
#include "opto/runtime.hpp"
#include "opto/type.hpp"
//--------------------gen_stub-------------------------------
void GraphKit::gen_stub(address C_function,
const char *name,
int is_fancy_jump,
bool pass_tls,
bool return_pc) {
ResourceMark rm;
const TypeTuple *jdomain = C->tf()->domain();
const TypeTuple *jrange = C->tf()->range();
// The procedure start
StartNode* start = new StartNode(root(), jdomain);
_gvn.set_type_bottom(start);
// Make a map, with JVM state
uint parm_cnt = jdomain->cnt();
uint max_map = MAX2(2*parm_cnt+1, jrange->cnt());
// %%% SynchronizationEntryBCI is redundant; use InvocationEntryBci in interfaces
assert(SynchronizationEntryBCI == InvocationEntryBci, "");
JVMState* jvms = new (C) JVMState(0);
jvms->set_bci(InvocationEntryBci);
jvms->set_monoff(max_map);
jvms->set_scloff(max_map);
jvms->set_endoff(max_map);
{
SafePointNode *map = new SafePointNode( max_map, jvms );
jvms->set_map(map);
set_jvms(jvms);
assert(map == this->map(), "kit.map is set");
}
// Make up the parameters
uint i;
for( i = 0; i < parm_cnt; i++ )
map()->init_req(i, _gvn.transform(new ParmNode(start, i)));
for( ; i<map()->req(); i++ )
map()->init_req(i, top()); // For nicer debugging
// GraphKit requires memory to be a MergeMemNode:
set_all_memory(map()->memory());
// Get base of thread-local storage area
Node* thread = _gvn.transform( new ThreadLocalNode() );
const int NoAlias = Compile::AliasIdxBot;
Node* adr_last_Java_pc = basic_plus_adr(top(),
thread,
in_bytes(JavaThread::frame_anchor_offset()) +
in_bytes(JavaFrameAnchor::last_Java_pc_offset()));
#if defined(SPARC)
Node* adr_flags = basic_plus_adr(top(),
thread,
in_bytes(JavaThread::frame_anchor_offset()) +
in_bytes(JavaFrameAnchor::flags_offset()));
#endif /* defined(SPARC) */
// Drop in the last_Java_sp. last_Java_fp is not touched.
// Always do this after the other "last_Java_frame" fields are set since
// as soon as last_Java_sp != NULL the has_last_Java_frame is true and
// users will look at the other fields.
//
Node *adr_sp = basic_plus_adr(top(), thread, in_bytes(JavaThread::last_Java_sp_offset()));
Node *last_sp = basic_plus_adr(top(), frameptr(), (intptr_t) STACK_BIAS);
store_to_memory(NULL, adr_sp, last_sp, T_ADDRESS, NoAlias, MemNode::unordered);
// Set _thread_in_native
// The order of stores into TLS is critical! Setting _thread_in_native MUST
// be last, because a GC is allowed at any time after setting it and the GC
// will require last_Java_pc and last_Java_sp.
//-----------------------------
// Compute signature for C call. Varies from the Java signature!
const Type **fields = TypeTuple::fields(2*parm_cnt+2);
uint cnt = TypeFunc::Parms;
// The C routines gets the base of thread-local storage passed in as an
// extra argument. Not all calls need it, but its cheap to add here.
for (uint pcnt = cnt; pcnt < parm_cnt; pcnt++, cnt++) {
// Convert ints to longs if required.
if (CCallingConventionRequiresIntsAsLongs && jdomain->field_at(pcnt)->isa_int()) {
fields[cnt++] = TypeLong::LONG;
fields[cnt] = Type::HALF; // must add an additional half for a long
} else {
fields[cnt] = jdomain->field_at(pcnt);
}
}
fields[cnt++] = TypeRawPtr::BOTTOM; // Thread-local storage
// Also pass in the caller's PC, if asked for.
if( return_pc )
fields[cnt++] = TypeRawPtr::BOTTOM; // Return PC
const TypeTuple* domain = TypeTuple::make(cnt,fields);
// The C routine we are about to call cannot return an oop; it can block on
// exit and a GC will trash the oop while it sits in C-land. Instead, we
// return the oop through TLS for runtime calls.
// Also, C routines returning integer subword values leave the high
// order bits dirty; these must be cleaned up by explicit sign extension.
const Type* retval = (jrange->cnt() == TypeFunc::Parms) ? Type::TOP : jrange->field_at(TypeFunc::Parms);
// Make a private copy of jrange->fields();
const Type **rfields = TypeTuple::fields(jrange->cnt() - TypeFunc::Parms);
// Fixup oop returns
int retval_ptr = retval->isa_oop_ptr();
if( retval_ptr ) {
assert( pass_tls, "Oop must be returned thru TLS" );
// Fancy-jumps return address; others return void
rfields[TypeFunc::Parms] = is_fancy_jump ? TypeRawPtr::BOTTOM : Type::TOP;
} else if( retval->isa_int() ) { // Returning any integer subtype?
// "Fatten" byte, char & short return types to 'int' to show that
// the native C code can return values with junk high order bits.
// We'll sign-extend it below later.
rfields[TypeFunc::Parms] = TypeInt::INT; // It's "dirty" and needs sign-ext
} else if( jrange->cnt() >= TypeFunc::Parms+1 ) { // Else copy other types
rfields[TypeFunc::Parms] = jrange->field_at(TypeFunc::Parms);
if( jrange->cnt() == TypeFunc::Parms+2 )
rfields[TypeFunc::Parms+1] = jrange->field_at(TypeFunc::Parms+1);
}
const TypeTuple* range = TypeTuple::make(jrange->cnt(),rfields);
// Final C signature
const TypeFunc *c_sig = TypeFunc::make(domain,range);
//-----------------------------
// Make the call node
CallRuntimeNode *call = new CallRuntimeNode(c_sig, C_function, name, TypePtr::BOTTOM);
//-----------------------------
// Fix-up the debug info for the call
call->set_jvms( new (C) JVMState(0) );
call->jvms()->set_bci(0);
call->jvms()->set_offsets(cnt);
// Set fixed predefined input arguments
cnt = 0;
for (i = 0; i < TypeFunc::Parms; i++)
call->init_req(cnt++, map()->in(i));
// A little too aggressive on the parm copy; return address is not an input
call->set_req(TypeFunc::ReturnAdr, top());
for (; i < parm_cnt; i++) { // Regular input arguments
// Convert ints to longs if required.
if (CCallingConventionRequiresIntsAsLongs && jdomain->field_at(i)->isa_int()) {
Node* int_as_long = _gvn.transform(new ConvI2LNode(map()->in(i)));
call->init_req(cnt++, int_as_long); // long
call->init_req(cnt++, top()); // half
} else {
call->init_req(cnt++, map()->in(i));
}
}
call->init_req( cnt++, thread );
if( return_pc ) // Return PC, if asked for
call->init_req( cnt++, returnadr() );
_gvn.transform_no_reclaim(call);
//-----------------------------
// Now set up the return results
set_control( _gvn.transform( new ProjNode(call,TypeFunc::Control)) );
set_i_o( _gvn.transform( new ProjNode(call,TypeFunc::I_O )) );
set_all_memory_call(call);
if (range->cnt() > TypeFunc::Parms) {
Node* retnode = _gvn.transform( new ProjNode(call,TypeFunc::Parms) );
// C-land is allowed to return sub-word values. Convert to integer type.
assert( retval != Type::TOP, "" );
if (retval == TypeInt::BOOL) {
retnode = _gvn.transform( new AndINode(retnode, intcon(0xFF)) );
} else if (retval == TypeInt::CHAR) {
retnode = _gvn.transform( new AndINode(retnode, intcon(0xFFFF)) );
} else if (retval == TypeInt::BYTE) {
retnode = _gvn.transform( new LShiftINode(retnode, intcon(24)) );
retnode = _gvn.transform( new RShiftINode(retnode, intcon(24)) );
} else if (retval == TypeInt::SHORT) {
retnode = _gvn.transform( new LShiftINode(retnode, intcon(16)) );
retnode = _gvn.transform( new RShiftINode(retnode, intcon(16)) );
}
map()->set_req( TypeFunc::Parms, retnode );
}
//-----------------------------
// Clear last_Java_sp
store_to_memory(NULL, adr_sp, null(), T_ADDRESS, NoAlias, MemNode::unordered);
// Clear last_Java_pc and (optionally)_flags
store_to_memory(NULL, adr_last_Java_pc, null(), T_ADDRESS, NoAlias, MemNode::unordered);
#if defined(SPARC)
store_to_memory(NULL, adr_flags, intcon(0), T_INT, NoAlias, MemNode::unordered);
#endif /* defined(SPARC) */
#if (defined(IA64) && !defined(AIX))
Node* adr_last_Java_fp = basic_plus_adr(top(), thread, in_bytes(JavaThread::last_Java_fp_offset()));
store_to_memory(NULL, adr_last_Java_fp, null(), T_ADDRESS, NoAlias, MemNode::unordered);
#endif
// For is-fancy-jump, the C-return value is also the branch target
Node* target = map()->in(TypeFunc::Parms);
// Runtime call returning oop in TLS? Fetch it out
if( pass_tls ) {
Node* adr = basic_plus_adr(top(), thread, in_bytes(JavaThread::vm_result_offset()));
Node* vm_result = make_load(NULL, adr, TypeOopPtr::BOTTOM, T_OBJECT, NoAlias, MemNode::unordered);
map()->set_req(TypeFunc::Parms, vm_result); // vm_result passed as result
// clear thread-local-storage(tls)
store_to_memory(NULL, adr, null(), T_ADDRESS, NoAlias, MemNode::unordered);
}
//-----------------------------
// check exception
Node* adr = basic_plus_adr(top(), thread, in_bytes(Thread::pending_exception_offset()));
Node* pending = make_load(NULL, adr, TypeOopPtr::BOTTOM, T_OBJECT, NoAlias, MemNode::unordered);
Node* exit_memory = reset_memory();
Node* cmp = _gvn.transform( new CmpPNode(pending, null()) );
Node* bo = _gvn.transform( new BoolNode(cmp, BoolTest::ne) );
IfNode *iff = create_and_map_if(control(), bo, PROB_MIN, COUNT_UNKNOWN);
Node* if_null = _gvn.transform( new IfFalseNode(iff) );
Node* if_not_null = _gvn.transform( new IfTrueNode(iff) );
assert (StubRoutines::forward_exception_entry() != NULL, "must be generated before");
Node *exc_target = makecon(TypeRawPtr::make( StubRoutines::forward_exception_entry() ));
Node *to_exc = new TailCallNode(if_not_null,
i_o(),
exit_memory,
frameptr(),
returnadr(),
exc_target, null());
root()->add_req(_gvn.transform(to_exc)); // bind to root to keep live
C->init_start(start);
//-----------------------------
// If this is a normal subroutine return, issue the return and be done.
Node *ret;
switch( is_fancy_jump ) {
case 0: // Make a return instruction
// Return to caller, free any space for return address
ret = new ReturnNode(TypeFunc::Parms, if_null,
i_o(),
exit_memory,
frameptr(),
returnadr());
if (C->tf()->range()->cnt() > TypeFunc::Parms)
ret->add_req( map()->in(TypeFunc::Parms) );
break;
case 1: // This is a fancy tail-call jump. Jump to computed address.
// Jump to new callee; leave old return address alone.
ret = new TailCallNode(if_null,
i_o(),
exit_memory,
frameptr(),
returnadr(),
target, map()->in(TypeFunc::Parms));
break;
case 2: // Pop return address & jump
// Throw away old return address; jump to new computed address
//assert(C_function == CAST_FROM_FN_PTR(address, OptoRuntime::rethrow_C), "fancy_jump==2 only for rethrow");
ret = new TailJumpNode(if_null,
i_o(),
exit_memory,
frameptr(),
target, map()->in(TypeFunc::Parms));
break;
default:
ShouldNotReachHere();
}
root()->add_req(_gvn.transform(ret));
}