1 refactor prologue reserve dynamic slots which is setted by LLVM IR
2 callsiteSp is calculated by current function's callee saved registers size
3 and Add mulitple Optimized Frame:
OPTIMIZED_JS_FUNCTION_FRAME:argv[n-1],..., argv[0], argc, ...., returnaddr, Fp, type. this frame need to visit args and stackmap.
OPTIMIZED_ASM_FRAME: jscall/jsentry construct frame which is implemented by assemble, this frame is only link to previou frame, it don't need to visit args and stackmap
4 stackmap support parse const pool and llvm patch add deopt interface LLVMBuildCall3
issue: https://gitee.com/openharmony/ark_js_runtime/issues/I5AJ9U
Signed-off-by: songzhengchao <songzhengchao@huawei.com>
Change-Id: Ie6a0067169805cc6043ea94936dd578489b56c7b
description:
1 add align up for rodata section, incase movdqa crash
2 set gc leaf function attribute for no gc all
3 fix mega ic run slowpath bug for asm interpreter
4 remove trampoline unused code
issue: #I58VFA
Signed-off-by: sunzhe23 <sunzhe23@huawei.com>
1. x86 assembler replaces JSCall, JSCallArgV, JSFunctionEntry,
OptimizedCallOptimized, CallNativeTrampoline assembly code of
x86 call.S debugged and enabled to the official code.
2. Delete the relevant code in call.S.
issue:https://gitee.com/openharmony/ark_js_runtime/issues/I57UHX
Signed-off-by: wanyanglan <wanyanglan1@huawei.com>
Change-Id: I69b9aad1f52d486ba6d96248aa1a5a2d0a44f318
desc: move interpreter helper stubs from common stub to bc stub, beacuse these stubs' signatures are the same and need not construct frame pointer like bc stub.
solu:
1. delete old helper stub define in common stub, and add new helper stub define in bc stub.
2. add new targetKind in CallSignature to distinguish helper and normal stub in loading stub.
3. delete argc in stub define.
issue: https://gitee.com/openharmony/ark_js_runtime/issues/I556K1
Signed-off-by: wanghuan <wanghuan80@huawei.com>
Change-Id: I46224e4568b45794aa1bac97682afc41696d5faa
desc: add asmInter debugger to support ide debug.
solu:
1.add bc debugger stub entry;
2.check is or not debug mode in runInternal and switch release to debug or debug to release.
issue: https://gitee.com/openharmony/ark_js_runtime/issues/I53Z4V
Signed-off-by: wanghuan <wanghuan80@huawei.com>
Change-Id: I3f61bed8b4e17c43d22b5e6f103fe897cd8b41c1
Description
Cut out most of the initialization logic in Runtime::Create and disconnect the
inheritance relationship of most classes such as JSThread.
Issue:
#I53418: Runtime architecture adjustment
Signed-off-by: lifansheng <lifansheng1@huawei.com>
Change-Id: Ib9c1a40354f77d32e515e23334e7f4b4a4fa3259
when function need call another js function, it will lowering to call "JSCall" trampoline function
JSCall trampoline function will check calltarget function, then correct the arguments, finally jump
the native or optimized code address of the target function.
JSCall(glue, argc, calltarget, ....)
then proxy call(glue, argc, callTarget, argv[])
if proxy->GetHandler is function
tail call JSCall(glue, argc, callTarget, argv); // consider performance
else
//Call(trap, handler, «target, thisArgument, argArray»)
call CreateArrayFromList(glue, argc, argv[]) ---> arrHandle // maybe trigger GC, construct OptimizedWithArgvLeaveFrame, visitor argv[0]..argv[argc-1]
call JSCall(glue, argc, method, argv[])
issue:https://gitee.com/openharmony/ark_js_runtime/issues/I52T6B?from=project-issue
Signed-off-by: songzhengchao <songzhengchao@huawei.com>
Change-Id: I994aae1eb9cbf51982348ae6e203c24c326f617e
CALLARG0DYN_PREF_V8
CALLARG1DYN_PREF_V8_V8
CALLARGS2DYN_PREF_V8_V8_V8
CALLARGS3DYN_PREF_V8_V8_V8_V8
CALLITHISRANGEDYN_PREF_IMM16_V8
CALLIRANGEDYN_PREF_IMM16_V8
when function need call another js function, it will lowering to call "JSCall" trampoline function
JSCall trampoline function will check calltarget function, then correct the arguments, finally jump
the native or optimized code address of the target function
https://gitee.com/openharmony/ark_js_runtime/issues/I52KN7?from=project-issue
Signed-off-by: getingke <getingke@huawei.com>
Change-Id: I6be3f130b211ca7c51a557f9ace4bd20d46c4d04
Implement llvmIRGeneration Pass in aot compiler pipeline.
Rename and reconstruct LLVMStubModule and LLVMModuleAssembler class. Move aot code infos
related to describe current (code size, begin addr, entry offset of each func etc.) AOT
snippet and corresponding serialization/deserialization func to NEW class called AotCodeInfo.
Move functions related to file reading/writing to NEW class called AotFileManager.
Add new call opcode in circuit IR to differentiate call into common stubs and call into
runtime interfaces without GC.
Adapt js runtime option for above changes.
Issue: https://gitee.com/openharmony/ark_js_runtime/issues/I4ZDY5
Signed-off-by: luochuhao <luochuhao@huawei.com>
Change-Id: If0ab5927ad2551fe5567d1c7b18540fbcc38bc15
Because there are some macros that trigger compilation errors when
compiling stub-compiler and aot-compiler. The root cause is that several
classes declarations are missing from the associated macros. This PR
fixes the issues via adding the lost classes to the macros. Besides, we
also rename the FastStub to CommonStubs in order to make the stub-arch
more readable.
Issue:https://gitee.com/openharmony/ark_js_runtime/issues/I4YKBN
Signed-off-by: wuzhefeng <wuzhefeng1@huawei.com>
Change-Id: I16263fef7deeadf8a9ce4e5252a6e9898a7742bc
Refactor ID definition and callSignature initialization mechanism of runtime stub,
optimizer stub and bytecode handler stub,
reduce amount and complexity of macro used in calling those stubs.
Add stubDes struct to classify stubs both in AOT file generation phase
and AOT file loading phase.
Fix some circular dependency problems in including header files.
Fix certain inline function declaration and definition format problems, which could lead to
compiling errors.
Issue: https://gitee.com/openharmony/ark_js_runtime/issues/I4VMLD?from=project-issue
Test: stubTest(unit test), richards with asm interpreter enabled
Signed-off-by: luochuhao <luochuhao@huawei.com>
Change-Id: Ibd5fcd963347b97f8dec227f3245d2064463b0b0
Change several structures for call stub.
Related issue: #I4XK12
Signed-off-by: lichenshuai <lichenshuai@huawei.com>
Change-Id: Ie7389ff96c619451304a3f0d352b271310c4b73e
The lowering slow path requires not only the vreg information of the
instruction, but also other information such as string id, method id, etc.
For each bytecode instruction, collect the various information stored in
the instruction, and use the information as the input of the gate, and
in the lowering stage, take out the required information from the gate.
issue:https://gitee.com/openharmony/ark_js_runtime/issues/I4WQR5
Signed-off-by: wanyanglan <wanyanglan1@huawei.com>
Change-Id: I42ef36554b0b88ce3e1cd0f593e8ce9e924b83e0
1 writebarrier marking fast path
2 The design of Frame is unfriendly for low-level operations like ASM-interpreter and Compiler IR.
1 writebarrier marking flag is get by jsthread.threadStateBitField_
2 Refactor the Frame design by introducing Aligned struct
issue:https://gitee.com/openharmony/ark_js_runtime/issues/I4XJ4M?from=project-issue
Signed-off-by: songzhengchao <songzhengchao@huawei.com>
Change-Id: I8fae1af0ce3c898909f088f8963f08c2f48dba8e
This PR refactors the design of JSThread by introducing aligned struct,
which is more friendly for ASM-Interpreter and Compiler IRs. Meanwhile,
to facilitate debugging, a macro enable_bytrace is introduced to control
the lib dependencies of ark-js-runtime.
Issue: https://gitee.com/openharmony/ark_js_runtime/issues/I4X0TG
Signed-off-by: wuzhefeng <wuzhefeng1@huawei.com>
Change-Id: I56835deec5d372d7c827ded26544a87766f29002
g00416891