Update Runtime to support Apple Silicon

- Update entry point call for Apple arm64 function calling convention
- New Assembler routines for function calls

lib/Runtime/Debug/TTEventLog.cpp

@@ -1,5 +1,6 @@
 //-------------------------------------------------------------------------------------------------------
 // Copyright (C) Microsoft. All rights reserved.
+// Copyright (c) ChakraCore Project Contributors. All rights reserved.
 // Licensed under the MIT license. See LICENSE.txt file in the project root for full license information.
 //-------------------------------------------------------------------------------------------------------
 #include "RuntimeDebugPch.h"
@@ -2816,7 +2817,9 @@ namespace TTD
         TTDAssert(wcscmp(_u("x86"), archString.Contents) == 0, "Mismatch in arch between record and replay!!!");
 #elif defined(_M_X64)
         TTDAssert(wcscmp(_u("x64"), archString.Contents) == 0, "Mismatch in arch between record and replay!!!");
-#elif defined(_M_ARM)
+#elif defined(_M_ARM) // #TODO investigate why this is checking for "arm64" instead of "arm"
+        TTDAssert(wcscmp(_u("arm64"), archString.Contents) == 0, "Mismatch in arch between record and replay!!!");
+#elif defined(_M_ARM64)
         TTDAssert(wcscmp(_u("arm64"), archString.Contents) == 0, "Mismatch in arch between record and replay!!!");
 #else
         TTDAssert(false, "Unknown arch!!!");

lib/Runtime/Language/Arguments.h

@@ -1,5 +1,6 @@
 //-------------------------------------------------------------------------------------------------------
 // Copyright (C) Microsoft. All rights reserved.
+// Copyright (c) ChakraCore Project Contributors. All rights reserved.
 // Licensed under the MIT license. See LICENSE.txt file in the project root for full license information.
 //-------------------------------------------------------------------------------------------------------
 #pragma once
@@ -82,6 +83,9 @@ inline int _count_args(const T1&, const T2&, const T3&, const T4&, const T5&, Js
 // xplat-todo: fix me ARM
 #define CALL_ENTRYPOINT_NOASSERT(entryPoint, function, callInfo, ...) \
     entryPoint(function, callInfo, ##__VA_ARGS__)
+#elif defined (_ARM64_)
+#define CALL_ENTRYPOINT_NOASSERT(entryPoint, function, callInfo, ...) \
+    entryPoint(function, callInfo, function, callInfo, ##__VA_ARGS__)
 #else
 #error CALL_ENTRYPOINT_NOASSERT not yet implemented
 #endif

lib/Runtime/Language/CMakeLists.txt

@@ -78,6 +78,12 @@ elseif(CC_TARGETS_X86)
         i386/AsmJsJitTemplate.cpp
         i386/StackFrame.cpp
     )
+elseif(CC_TARGETS_ARM64)
+    set (CRL_SOURCE_FILES ${CRL_SOURCE_FILES}
+        arm64/StackFrame.cpp
+        arm64/arm64_Thunks.S
+        arm64/arm64_CallEhFrame.S
+    )
 elseif(CC_TARGETS_ARM)
     set (CRL_SOURCE_FILES ${CRL_SOURCE_FILES}
         arm/StackFrame.cpp

lib/Runtime/Language/InterpreterStackFrame.cpp

@@ -1,6 +1,6 @@
 //-------------------------------------------------------------------------------------------------------
 // Copyright (C) Microsoft Corporation and contributors. All rights reserved.
-// Copyright (c) 2021 ChakraCore Project Contributors. All rights reserved.
+// Copyright (c) ChakraCore Project Contributors. All rights reserved.
 // Licensed under the MIT license. See LICENSE.txt file in the project root for full license information.
 //-------------------------------------------------------------------------------------------------------
 
@@ -1814,13 +1814,13 @@ namespace Js
     }
 #endif
 
-#if !defined(_M_ARM64)
+#if defined(_M_ARM64) && defined(_WIN32)
+    // Language\arm64\arm64_Thunks.asm
+#else
     Var InterpreterStackFrame::StaticInterpreterThunk(RecyclableObject* function, CallInfo callInfo, ...)
     {
         return InterpreterThunk((JavascriptCallStackLayout*)&function);
     }
-#else
-    // Language\arm64\arm64_Thunks.asm
 #endif
 #pragma optimize("", on)
 
@@ -6352,7 +6352,11 @@ namespace Js
         // For ARM we need to make sure that pipeline is synchronized with memory/cache for newly jitted code.
         // Note: this does not seem to affect perf, but if it was, we could add a boolean isCalled to EntryPointInfo
         //       and do ISB only for 1st time this entry point is called (potential working set regression though).
+        #if defined(_InstructionSynchronizationBarrier)
         _InstructionSynchronizationBarrier();
+        #else
+        asm("isb");
+        #endif
 #endif
         uint newOffset = ::Math::PointerCastToIntegral<uint>(
             CALL_ENTRYPOINT_NOASSERT(address, function, CallInfo(CallFlags_InternalFrame, 1), this));
@@ -6386,7 +6390,11 @@ namespace Js
         // For ARM we need to make sure that pipeline is synchronized with memory/cache for newly jitted code.
         // Note: this does not seem to affect perf, but if it was, we could add a boolean isCalled to EntryPointInfo
         //       and do ISB only for 1st time this entry point is called (potential working set regression though).
+        #if defined(_InstructionSynchronizationBarrier)
         _InstructionSynchronizationBarrier();
+        #else
+        asm("isb");
+        #endif
 #endif
         uint newOffset = ::Math::PointerCastToIntegral<uint>(
             CALL_ENTRYPOINT_NOASSERT(address, function, CallInfo(CallFlags_InternalFrame, 1), this));

lib/Runtime/Language/JavascriptStackWalker.cpp

@@ -1,5 +1,6 @@
 //-------------------------------------------------------------------------------------------------------
 // Copyright (C) Microsoft. All rights reserved.
+// Copyright (c) ChakraCore Project Contributors. All rights reserved.
 // Licensed under the MIT license. See LICENSE.txt file in the project root for full license information.
 //-------------------------------------------------------------------------------------------------------
 #include "RuntimeLanguagePch.h"
@@ -278,7 +279,7 @@ namespace Js
         }
         else
 #endif
-            if (this->GetCurrentFunction()->GetFunctionInfo()->IsCoroutine())
+        if (this->GetCurrentFunction()->GetFunctionInfo()->IsCoroutine())
         {
             JavascriptGenerator* gen = VarTo<JavascriptGenerator>(this->GetCurrentArgv()[JavascriptFunctionArgIndex_This]);
             return gen->GetArguments().Values;

lib/Runtime/Language/arm64/arm64_CallEhFrame.S

@@ -0,0 +1,136 @@
+;-------------------------------------------------------------------------------------------------------
+; Copyright (C) Microsoft. All rights reserved.
+; Copyright (c) ChakraCore Project Contributors. All rights reserved.
+; Licensed under the MIT license. See LICENSE.txt file in the project root for full license information.
+;-------------------------------------------------------------------------------------------------------
+
+;
+;   arm64_CallEhFrame() and arm64_CallCatch() both thunk into jitted code at the
+;   start of an EH region. The purpose is to restore the frame pointer (fp)
+;   and locals pointer (x28) to the appropriate values for executing the parent
+;   function and to create a local frame that can be unwound using the parent
+;   function's pdata. The parent's frame looks like this:
+;
+;-------------------
+;   {x0-x7}     -- homed parameters
+;   lr          -- address from which parent was called
+;   fp          -- saved frame pointer, pointed to by current fp
+;   arg obj
+;   {x19-x28}   -- non-volatile registers: all of them are saved
+;   {q8-q15}    -- non-volatile double registers: all of them are saved
+;   locals area -- pointed to by x28
+;   pointer to non-volatile register area above
+;   stack args
+;-------------------
+;
+;   The reason for the "pointer to non-volatile register area" is to allow the
+;   unwinder to deallocate the locals area regardless of its size. So this thunk can skip
+;   the allocation of the locals area altogether, and unwinding still works.
+;   The unwind pseudo-codes for the above prolog look like:
+;
+;   1. Deallocate stack args (sp now points to "pointer to non-volatile register area")
+;   2. Restore rN (rN now points to first saved register)
+;   3. Copy rN to sp (sp now points to first saved register)
+;   4. Restore {q8-q15} (non-volatile double registers restored)
+;   5. Restore {x19-x28} (non-volatile registers restored, sp points to saved r11)
+;   6. Restore fp
+;   7. Load lr into pc and deallocate remaining stack.
+;
+;   The prologs for the assembly thunks allocate a frame that can be unwound by executing
+;   the above steps, although we don't allocate a locals area and don't know the size of the
+;   stack args. The caller doesn't return to this thunk; it executes its own epilog and
+;   returns to the caller of the thunk (one of the runtime try helpers).
+
+
+#include "unixasmmacros.inc"
+
+.global C_FUNC(arm64_CallEhFrame)
+.global C_FUNC(arm64_CallCatch)
+
+.macro STANDARD_PROLOG
+
+    ;
+    ; Generate a prolog that will match the original function's, with all
+    ; parameters homed and all non-volatile registers saved:
+    ;
+    ;    Size  Offset
+    ;    ----  ------ 
+    ;     64    176    Homed parameters
+    ;     16    160    Saved FP/LR
+    ;     16    144    ArgOut / stack function list
+    ;     80     64    Saved x19-x28
+    ;     64      0    Saved d8-d15
+    ;  = 240 total
+    ;
+    ; The try/catch/finally blocks will jump to the epilog code skipping
+    ; the instruction that deallocates the locals, in order to allow these
+    ; thunks to skip re-allocating locals space.
+    ;
+
+    ; Params:
+    ; x0 -- thunk target
+    ; x1 -- frame pointer
+    ; x2 -- locals pointer
+    ; x3 -- size of stack args area
+    ; x4 -- exception object (for arm64_CallCatch only)
+
+    PROLOG_SAVE_REG_PAIR d8, d9, -240
+    PROLOG_SAVE_REG_PAIR d10, d11, 16
+    PROLOG_SAVE_REG_PAIR d12, d13, 32
+    PROLOG_SAVE_REG_PAIR d14, d15, 48
+    PROLOG_SAVE_REG_PAIR x19, x20, 64
+    PROLOG_SAVE_REG_PAIR x21, x22, 80
+    PROLOG_SAVE_REG_PAIR x23, x24, 96
+    PROLOG_SAVE_REG_PAIR x25, x26, 112
+    PROLOG_SAVE_REG_PAIR x27, x28, 128
+    PROLOG_SAVE_REG fp, 160 ; TODO: verify that this works the same as PROLOG_SAVE_REG_PAIR_NO_FP
+    PROLOG_SAVE_REG lr, 168
+
+    sub     x15, x1, x2         ; x15 = frame pointer minus locals pointer
+    sub     x15, x15, #160      ; x15 -= space we already allocated
+    add     x15, x15, x3        ; x15 += argout area = same stack allocation as original function
+    lsr     x15, x15, #4        ; x15 /= 16
+    sub     sp, sp, x15, lsl #4 ; allocate the stack
+
+.endm
+
+
+
+NESTED_ENTRY arm64_CallEhFrame, _TEXT, NoHandler
+
+    STANDARD_PROLOG
+
+    ; Set up the locals pointer and frame pointer
+    mov     x28, x2
+    mov     fp, x1
+
+    ; Thunk to the jitted code (and don't return)
+    br      x0
+
+NESTED_END arm64_CallEhFrame
+
+
+
+    ; arm64_CallCatch() is similar to arm64_CallEhFrame() except that we also pass the catch object to the jitted code
+
+NESTED_ENTRY arm64_CallCatch, _TEXT, NoHandler
+
+    ; Params:
+    ; x0 -- thunk target
+    ; x1 -- frame pointer
+    ; x2 -- locals pointer
+    ; x3 -- size of stack args area
+    ; x4 -- exception object
+
+    STANDARD_PROLOG
+
+    ; Set up the locals pointer and frame pointer and catch object handler
+    mov     x28, x2
+    mov     fp, x1
+    mov     x1, x4
+
+    ; Thunk to the jitted code (and don't return)
+    br      x0
+
+NESTED_END arm64_CallCatch
+

lib/Runtime/Language/arm64/arm64_Thunks.S

@@ -0,0 +1,96 @@
+;-------------------------------------------------------------------------------------------------------
+; Copyright (C) Microsoft. All rights reserved.
+; Copyright (c) ChakraCore Project Contributors. All rights reserved.
+; Licensed under the MIT license. See LICENSE.txt file in the project root for full license information.
+;-------------------------------------------------------------------------------------------------------
+
+#include "unixasmmacros.inc"
+
+.global C_FUNC(_ZN2Js13ScriptContext31ProfileModeDeferredParsingThunkEPNS_16RecyclableObjectENS_8CallInfoEz)
+.global C_FUNC(_ZN2Js13ScriptContext35ProfileModeDeferredDeserializeThunkEPNS_16RecyclableObjectENS_8CallInfoEz)
+
+#ifdef _ENABLE_DYNAMIC_THUNKS
+
+.global C_FUNC(_ZN2Js21InterpreterStackFrame28DelayDynamicInterpreterThunkEPNS_16RecyclableObjectENS_8CallInfoEz)
+.global C_FUNC(_ZN2Js18DynamicProfileInfo29EnsureDynamicProfileInfoThunkEPNS_16RecyclableObjectENS_8CallInfoEz)
+
+;;============================================================================================================
+;; InterpreterStackFrame::DelayDynamicInterpreterThunk
+;;============================================================================================================
+    ;Var InterpreterStackFrame::DelayDynamicInterpreterThunk(RecyclableObject* function, CallInfo callInfo, ...)
+
+NESTED_ENTRY _ZN2Js21InterpreterStackFrame28DelayDynamicInterpreterThunkEPNS_16RecyclableObjectENS_8CallInfoEz, _TEXT, NoHandler
+
+    PROLOG_SAVE_REG_PAIR_INDEXED fp, lr, -32   ; save parameters and volatile registers
+    stp     x0, x1, [sp, #16]
+
+    bl      C_FUNC(_ZN2Js21InterpreterStackFrame29EnsureDynamicInterpreterThunkEPNS_14ScriptFunctionE)  ; call InterpreterStackFrame::EnsureDynamicInterpreterThunk
+    mov     x16, x0                  ; back up entryPoint in x16
+
+    ldp     x0, x1, [sp, #16]        ; restore parameters and volatile registers
+
+    EPILOG_RESTORE_REG_PAIR_INDEXED fp, lr, 32
+    br      x16                      ; jump (tail call) to new entryPoint
+
+NESTED_END _ZN2Js21InterpreterStackFrame28DelayDynamicInterpreterThunkEPNS_16RecyclableObjectENS_8CallInfoEz, _TEXT
+
+;;============================================================================================================
+;; DynamicProfileInfo::EnsureDynamicProfileInfoThunk
+;;============================================================================================================
+    ;Var DynamicProfileInfo::EnsureDynamicProfileInfoThunk(RecyclableObject* function, CallInfo callInfo, ...)
+NESTED_ENTRY _ZN2Js18DynamicProfileInfo29EnsureDynamicProfileInfoThunkEPNS_16RecyclableObjectENS_8CallInfoEz, _TEXT, NoHandler
+
+    PROLOG_SAVE_REG_PAIR_INDEXED fp, lr, -32   ; save parameters and volatile registers
+    stp     x0, x1, [sp, #16]
+
+    bl      C_FUNC(_ZN2Js18DynamicProfileInfo24EnsureDynamicProfileInfoEPNS_14ScriptFunctionE) ; call DynamicProfileInfo::EnsureDynamicProfileInfo
+    mov     x16, x0                  ; back up entryPoint in x16
+
+    ldp     x0, x1, [sp, #16]        ; restore parameters and volatile registers
+
+    EPILOG_RESTORE_REG_PAIR_INDEXED fp, lr, 32
+    br      x16                      ; jump (tail call) to new entryPoint
+
+NESTED_END _ZN2Js18DynamicProfileInfo29EnsureDynamicProfileInfoThunkEPNS_16RecyclableObjectENS_8CallInfoEz, _TEXT
+
+#endif
+
+;;============================================================================================================
+;; ScriptContext::ProfileModeDeferredParsingThunk
+;;============================================================================================================
+    ;; Var ScriptContext::ProfileModeDeferredParsingThunk(RecyclableObject* function, CallInfo callInfo, ...)
+NESTED_ENTRY _ZN2Js13ScriptContext31ProfileModeDeferredParsingThunkEPNS_16RecyclableObjectENS_8CallInfoEz, _TEXT, NoHandler
+
+    PROLOG_SAVE_REG_PAIR_INDEXED fp, lr, -32   ; save parameters and volatile registers
+    stp     x0, x1, [sp, #16]
+
+    mov     x0, sp                   ; Pass the address of the function at the saved x0 in case it need to be boxed
+    add     x0, x0, #16              ; 16 is subtracted from the stack pointer when the a function is called, add it back here.
+    bl      C_FUNC(_ZN2Js13ScriptContext24ProfileModeDeferredParseEPPNS_14ScriptFunctionE) ; call ScriptContext::ProfileModeDeferredParse
+    mov     x16, x0                  ; back up entryPoint in x16
+
+    ldp     x0, x1, [sp, #16]        ; restore parameters and volatile registers
+
+    EPILOG_RESTORE_REG_PAIR_INDEXED fp, lr, 32
+    br      x16                      ; jump (tail call) to new entryPoint
+
+NESTED_END _ZN2Js13ScriptContext31ProfileModeDeferredParsingThunkEPNS_16RecyclableObjectENS_8CallInfoEz, _TEXT
+
+;;============================================================================================================
+;; ScriptContext::ProfileModeDeferredDeserializeThunk
+;;============================================================================================================
+    ;; Var ScriptContext::ProfileModeDeferredDeserializeThunk(RecyclableObject* function, CallInfo callInfo, ...)
+NESTED_ENTRY _ZN2Js13ScriptContext35ProfileModeDeferredDeserializeThunkEPNS_16RecyclableObjectENS_8CallInfoEz, _TEXT, NoHandler
+
+    PROLOG_SAVE_REG_PAIR_INDEXED fp, lr, -32   ; save parameters and volatile registers
+    stp     x0, x1, [sp, #16]
+
+    bl      C_FUNC(_ZN2Js13ScriptContext30ProfileModeDeferredDeserializeEPNS_14ScriptFunctionE) ; call ScriptContext::ProfileModeDeferredDeserialize
+    mov     x16, x0                     ; back up entryPoint in x16
+
+    ldp     x0, x1, [sp, #16]        ; restore parameters and volatile registers
+
+    EPILOG_RESTORE_REG_PAIR_INDEXED fp, lr, 32
+    br      x16                      ; jump (tail call) to new entryPoint
+
+NESTED_END _ZN2Js13ScriptContext35ProfileModeDeferredDeserializeThunkEPNS_16RecyclableObjectENS_8CallInfoEz, _TEXT

lib/Runtime/Library/CMakeLists.txt

@@ -133,6 +133,12 @@ elseif(CC_TARGETS_ARM)
     set (CRLIB_SOURCE_CODES ${CRLIB_SOURCE_CODES}
         arm/arm_JavascriptFunctionA.S
     )
+elseif(CC_TARGETS_ARM64)
+    set (CRLIB_SOURCE_CODES ${CRLIB_SOURCE_CODES}
+        arm64/arm64_CallFunction.S
+        arm64/arm64_DeferredDeserializeThunk.S
+        arm64/arm64_DeferredParsingThunk.S
+    )
 endif()
 
 add_library (Chakra.Runtime.Library OBJECT ${CRLIB_SOURCE_CODES})