Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

(0.20.0) AArch64: Implement Nestmate interface call #8854

Merged
merged 1 commit into from
Mar 18, 2020
Merged

(0.20.0) AArch64: Implement Nestmate interface call #8854

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
76 changes: 59 additions & 17 deletions runtime/compiler/aarch64/codegen/CallSnippet.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -508,44 +508,74 @@ uint32_t TR::ARM64VirtualUnresolvedSnippet::getLength(int32_t estimatedSnippetSt

uint8_t *TR::ARM64InterfaceCallSnippet::emitSnippetBody()
{
TR::Compilation *comp = cg()->comp();
uint8_t *cursor = cg()->getBinaryBufferCursor();
TR::Node *callNode = getNode();
TR::SymbolReference *methodSymRef = getNode()->getSymbolReference();
TR::SymbolReference *glueRef = cg()->symRefTab()->findOrCreateRuntimeHelper(TR_ARM64interfaceCallHelper, false, false, false);
TR_J9VMBase *fej9 = (TR_J9VMBase *)(comp->fe());
void* thunk = fej9->getJ2IThunk(callNode->getSymbolReference()->getSymbol()->castToMethodSymbol()->getMethod(), comp);

getSnippetLabel()->setCodeLocation(cursor);

// bl glueRef
*(int32_t *)cursor = cg()->encodeHelperBranchAndLink(glueRef, cursor, getNode());
cursor += 4;
*(int32_t *)cursor = cg()->encodeHelperBranchAndLink(glueRef, cursor, callNode);
cursor += ARM64_INSTRUCTION_LENGTH;

// Store the code cache RA
*(intptrj_t *)cursor = (intptrj_t)getReturnLabel()->getCodeLocation();
cg()->addExternalRelocation(new (cg()->trHeapMemory()) TR::ExternalRelocation(
cursor,
NULL,
TR_AbsoluteMethodAddress, cg()),
__FILE__, __LINE__, getNode());
cursor += 8;
__FILE__, __LINE__, callNode);
cursor += sizeof(intptrj_t);

// CP
*(intptrj_t *)cursor = (intptrj_t)methodSymRef->getOwningMethod(cg()->comp())->constantPool();
cg()->addExternalRelocation(new (cg()->trHeapMemory()) TR::ExternalRelocation(
cursor,
*(uint8_t **)cursor,
getNode() ? (uint8_t *)getNode()->getInlinedSiteIndex() : (uint8_t *)-1,
TR_Thunks, cg()),
__FILE__, __LINE__, getNode());
cursor += 8;
intptrj_t cpAddr = (intptrj_t)methodSymRef->getOwningMethod(comp)->constantPool();
*(intptrj_t *)cursor = cpAddr;
uint8_t *j2iThunkRelocationPoint = cursor;
cursor += sizeof(intptrj_t);

// CP index
*(intptrj_t *)cursor = methodSymRef->getCPIndexForVM();
cursor += 8;
cursor += sizeof(intptrj_t);

// Add 2 more slots for resolved values (interface class and iTable offset)
// 2 slots for resolved values (interface class and iTable index)
*(intptrj_t *)cursor = 0;
cursor += 8;
cursor += sizeof(intptrj_t);
*(intptrj_t *)cursor = 0;
cursor += 8;
cursor += sizeof(intptrj_t);

/*
* J2I thunk address.
* This is used for private nestmate calls.
*/
*(intptrj_t*)cursor = (intptrj_t)thunk;
if (comp->compileRelocatableCode())
{
auto info =
(TR_RelocationRecordInformation *)comp->trMemory()->allocateMemory(
sizeof(TR_RelocationRecordInformation),
heapAlloc);

// data1 = constantPool
info->data1 = cpAddr;

// data2 = inlined site index
info->data2 = callNode ? callNode->getInlinedSiteIndex() : (uintptr_t)-1;

// data3 = distance in bytes from Constant Pool Pointer to J2I Thunk
info->data3 = (intptrj_t)cursor - (intptrj_t)j2iThunkRelocationPoint;

cg()->addExternalRelocation(new (cg()->trHeapMemory()) TR::ExternalRelocation(
j2iThunkRelocationPoint,
(uint8_t *)info,
NULL,
TR_J2IVirtualThunkPointer, cg()),
__FILE__, __LINE__, callNode);
}
cursor += sizeof(intptrj_t);

return cursor;
}
Expand Down Expand Up @@ -582,11 +612,23 @@ TR_Debug::print(TR::FILE *pOutFile, TR::ARM64InterfaceCallSnippet * snippet)

printPrefix(pOutFile, NULL, cursor, sizeof(intptr_t));
trfprintf(pOutFile, ".dword \t0x%08x\t\t; itable index", *(intptr_t *)cursor);
cursor += sizeof(intptr_t);

printPrefix(pOutFile, NULL, cursor, sizeof(intptrj_t));
trfprintf(pOutFile, ".dword \t" POINTER_PRINTF_FORMAT "\t\t; J2I thunk address for private", *(intptrj_t *)cursor);
}

uint32_t TR::ARM64InterfaceCallSnippet::getLength(int32_t estimatedSnippetStart)
{
return 44;
/* 6 address fields:
* - Code cache RA
* - CP Pointer
* - CP Index
* - Interface Class Pointer
* - ITable Index (may also contain a tagged J9Method* when handling nestmates)
* - J2I thunk address
*/
return ARM64_INSTRUCTION_LENGTH + sizeof(intptrj_t)*6;
}

uint8_t *TR::ARM64CallSnippet::generateVIThunk(TR::Node *callNode, int32_t argSize, TR::CodeGenerator *cg)
Expand Down
68 changes: 59 additions & 9 deletions runtime/compiler/aarch64/runtime/PicBuilder.spp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -72,6 +72,7 @@
.extern jitResolveInterfaceMethod
.extern jitLookupInterfaceMethod
.extern jitCallCFunction
.extern jitInstanceOf
.extern jitThrowException
.extern mcc_reservationAdjustment_unwrapper
.extern mcc_callPointPatching_unwrapper
Expand Down Expand Up @@ -108,7 +109,8 @@
.set J9TR_UICSnippet_CP, 8
.set J9TR_UICSnippet_CPIndex, 16
.set J9TR_ICSnippet_InterfaceClass, 24
.set J9TR_ICSnippet_MethodIndex, 32
.set J9TR_ICSnippet_ITableIndex, 32
.set J9TR_ICSnippet_J2IThunk, 40

// Unresolved virtual call snippet

Expand Down Expand Up @@ -332,7 +334,7 @@ L_mergedDataResolve:
ldp x14, x15, [J9SP, #112]
ldp x16, x17, [J9SP, #128]
ldp x18, x19, [J9SP, #144]
ldr x21, [J9SP, #168] // not restoring J9SP
ldr x21, [J9SP, #168] // not restoring J9SP
ldp x22, x23, [J9SP, #176]
ldp x24, x25, [J9SP, #192]
ldp x26, x27, [J9SP, #208]
Expand Down Expand Up @@ -519,7 +521,7 @@ _interpreterUnresolvedStaticDataGlue:
stp x24, x25, [J9SP, #192]
stp x26, x27, [J9SP, #208]
str x28, [J9SP, #224]
ldr x3, const_jitResolveStaticField // load resolve helper address
ldr x3, const_jitResolveStaticField // load resolve helper address
b L_mergedDataResolve

_interpreterUnresolvedStaticDataStoreGlue:
Expand Down Expand Up @@ -576,7 +578,7 @@ _interpreterUnresolvedInstanceDataStoreGlue:
stp x24, x25, [J9SP, #192]
stp x26, x27, [J9SP, #208]
str x28, [J9SP, #224]
ldr x3, const_jitResolveFieldSetter // load resolve helper address
ldr x3, const_jitResolveFieldSetter // load resolve helper address
b L_mergedDataResolve

_interpreterUnresolvedConstantDynamicGlue:
Expand Down Expand Up @@ -681,7 +683,7 @@ L_callVirtual:
mov x2, x0
sub x0, x10, #20 // get the address of the movz instruction
ldr w1, [x0] // fetch the movz instruction
ubfx x3, x2, #0, #16 // lower 16 bits of the index
ubfx x3, x2, #0, #16 // lower 16 bits of the index
orr w1, w1, w3, LSL #5 // encode the index in the movz instruction
str w1, [x0] // store the movz instruction
ldr w1, [x0, #4] // fetch the movk instruction
Expand Down Expand Up @@ -723,11 +725,59 @@ _interfaceCallHelper:
stp x3, x2, [J9SP, #32]
stp x1, x0, [J9SP, #48]
mov x7, x30 // preserve LR
ldr x10, [x30, #J9TR_ICSnippet_codeCacheReturnAddress] // protect code cache RA in x10 (in L_commonLookupException, it is expected)
ldr x0, [x30, #J9TR_ICSnippet_ITableIndex] // Load ITable Index
tst x0, #J9TR_J9_ITABLE_OFFSET_DIRECT // Check if J9TR_J9_ITABLE_OFFSET_DIRECT flag is set
beq L_callResolve // If not set, need to call jitResolveInterfaceMethod
ldr x0, [x30, #J9TR_ICSnippet_InterfaceClass] // Load Interface Class Pointer
cbnz x0, L_typeCheckAndDirectDispatch // If it is not null, this is a known private interface call
L_callResolve:
add x0, x30, #J9TR_UICSnippet_CP // get CP/index pair pointer
ldr x1, [x30, #J9TR_ICSnippet_codeCacheReturnAddress] // get code cache RA
mov x10, x1 // protect RA in x10 (in L_commonLookupException, it is expected)
mov x1, x10 // get code cache RA
bl jitResolveInterfaceMethod // call the helper
cbz x0, L_commonLookupException // if resolve failed, throw the exception
ldr x0, [x7, #J9TR_ICSnippet_ITableIndex] // Load ITable Index
tst x0, #J9TR_J9_ITABLE_OFFSET_DIRECT // Check if J9TR_J9_ITABLE_OFFSET_DIRECT flag is set
beq L_callInterface // If not set, this does not need a direct dispatch
L_typeCheckAndDirectDispatch:
ldr x0, [x7, #J9TR_ICSnippet_InterfaceClass] // Load Interface Class Pointer
ldr x1, [J9SP, #56] // Load 'this' pointer
bl jitInstanceOf
cbnz x0, L_directDispatchInterface // If jitInstanceOf did not return null, continue on to direct dispatch
ldr x0, [J9SP, #56] // Load 'this' pointer
#ifdef OMR_GC_COMPRESSED_POINTERS
ldr w0, [x0, #J9TR_J9Object_class] // Load the class offset
#else
ldr x0, [x0, #J9TR_J9Object_class] // Load the class
#endif
and x0, x0, #~(J9TR_RequiredClassAlignment-1) // mask VFT bits
add x1, x7, J9TR_ICSnippet_InterfaceClass // Address of InterfaceClass/ITableIndex pair
mov x2, x7 // Load original RA for use inside jitLookupInterfaceMethod
bl jitLookupInterfaceMethod // Branch to jitLookupInterfaceMethod to trigger exception
// The code will not return here after the branch.
L_directDispatchInterface:
ldr x0, [x7, #J9TR_ICSnippet_ITableIndex] // Load ITable Index. This is actually a J9Method.
eor x0, x0, #J9TR_J9_ITABLE_OFFSET_DIRECT // Clear J9TR_J9_ITABLE_OFFSET_DIRECT flag.
ldr x1, [x0, #J9TR_MethodPCStartOffset] // load startPC/extra field
tst x1, J9TR_MethodNotCompiledBit // Check to see if the method has already been compiled
bne L_interpretedDispatch // If not compiled, handle interpreted case
ldr w11, [x1, -4] // Load offset of JIT-to-JIT
lsr w11, w11, 16 // shift right to get the bits we want
add x11, x11, x1 // Addr of JIT-to-JIT in x11
b L_interfaceCallout
L_interpretedDispatch:
orr x9, x0, #J9TR_J9_VTABLE_INDEX_DIRECT_METHOD_FLAG // put tagged J9Method in x9
ldr x11, [x7, #J9TR_ICSnippet_J2IThunk] // put thunk addr in x11
L_interfaceCallout:
ldr x0, [J9SP, #56] // Restore 'this' pointer
mov x30, x10 // set LR = code cache RA
ldp x7, x6, [J9SP, #0] // restore other parameter regs
ldp x5, x4, [J9SP, #16]
ldp x3, x2, [J9SP, #32]
ldr x1, [J9SP, #48]
add J9SP, J9SP, #64
br x11 // Call: does not return here
L_callInterface:
add x0, x7, #J9TR_Snippet_CallInstruction // get address of BL instruction in snippet
ldr x1, const_interfaceDispatch // get new snippet branch target
mov x2, #TR_ARM64interfaceDispatch
Expand All @@ -747,14 +797,14 @@ L_continueInterfaceSend:
ldr x0, [x0, #J9TR_ObjectHeader_class] // load class of receiver
#endif
and x0, x0, #~(J9TR_RequiredClassAlignment-1) // mask VFT bits
add x1, x30, #J9TR_ICSnippet_InterfaceClass // get InterfaceClass/MethodIndex pair pointer
add x1, x30, #J9TR_ICSnippet_InterfaceClass // get InterfaceClass/ITableIndex pair pointer
ldr x2, [x30, #J9TR_ICSnippet_codeCacheReturnAddress] // get code cache RA
mov x10, x2 // protect LR in x10 (in L_commonLookupException, it is expected)
bl jitLookupInterfaceMethod // call the helper
cbz x0, L_commonLookupException // if resolve failed, throw the exception
mov x9, #J9TR_InterpVTableOffset
sub x9, x9, x0 // convert interp vTableIndex to jit index (must be in x9 for patch virtual)
mov x30, x10 // set LR = code cache RA
mov x30, x10 // set LR = code cache RA
ldr x0, [J9SP, #56] // refetch 'this'
#ifdef OMR_GC_COMPRESSED_POINTERS
ldr w11, [x0, #J9TR_ObjectHeader_class] // load class offset of receiver
Expand Down