[NativeAOT] Use inline relocs for win-arm64 INLINE_GETTHREAD (#85689)

* trying to use inline reloc for win-arm64 tls

* remove arm64 INLINE_GETTHREAD_CONSTANT_POOL

* introduce INLINE_GET_TLS_VAR for win-arm64

* delete unused GetThread.asm files

diff --git a/src/coreclr/nativeaot/Runtime/amd64/GetThread.asm b/src/coreclr/nativeaot/Runtime/amd64/GetThread.asm
deleted file mode 100644 (file)
index 806f8db..0000000
--- a/src/coreclr/nativeaot/Runtime/amd64/GetThread.asm
+++ /dev/null
@@ -1,27 +0,0 @@
-;; Licensed to the .NET Foundation under one or more agreements.
-;; The .NET Foundation licenses this file to you under the MIT license.
-
-include asmmacros.inc
-
-;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
-;;
-;; RhpGetThread
-;;
-;;
-;; INPUT:
-;;
-;; OUTPUT: RAX: Thread pointer
-;;
-;; TRASHES: R10
-;;
-;; MUST PRESERVE ARGUMENT REGISTERS
-;;
-;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
-LEAF_ENTRY RhpGetThread, _TEXT
-        ;; rax = GetThread(), TRASHES r10
-        INLINE_GETTHREAD rax, r10
-        ret
-LEAF_END RhpGetThread, _TEXT
-
-
-        end

diff --git a/src/coreclr/nativeaot/Runtime/arm/GetThread.asm b/src/coreclr/nativeaot/Runtime/arm/GetThread.asm
deleted file mode 100644 (file)
index b78319f..0000000
--- a/src/coreclr/nativeaot/Runtime/arm/GetThread.asm
+++ /dev/null
@@ -1,32 +0,0 @@
-;; Licensed to the .NET Foundation under one or more agreements.
-;; The .NET Foundation licenses this file to you under the MIT license.
-
-#include "AsmMacros.h"
-
-        TEXTAREA
-
-;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
-;;
-;; RhpGetThread
-;;
-;;
-;; INPUT: none
-;;
-;; OUTPUT: r0: Thread pointer
-;;
-;; MUST PRESERVE ARGUMENT REGISTERS
-;; @todo check the actual requirements here, r0 is both return and argument register
-;;
-;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
-        LEAF_ENTRY RhpGetThread
-
-        ;; r0 = GetThread(), TRASHES r12
-        INLINE_GETTHREAD r0, r12
-        bx lr
-
-        LEAF_END
-FASTCALL_ENDFUNC
-
-        INLINE_GETTHREAD_CONSTANT_POOL
-
-        end

diff --git a/src/coreclr/nativeaot/Runtime/arm64/AllocFast.asm b/src/coreclr/nativeaot/Runtime/arm64/AllocFast.asm
index 560c6dc..e6849b8 100644 (file)
--- a/src/coreclr/nativeaot/Runtime/arm64/AllocFast.asm
+++ b/src/coreclr/nativeaot/Runtime/arm64/AllocFast.asm
@@ -48,8 +48,6 @@ RhpNewFast_RarePath
         b           RhpNewObject
     LEAF_END RhpNewFast
 
-    INLINE_GETTHREAD_CONSTANT_POOL
-
 ;; Allocate non-array object with finalizer.
 ;;  x0 == MethodTable
     LEAF_ENTRY RhpNewFinalizable
@@ -149,9 +147,6 @@ StringSizeOverflow
         b           RhExceptionHandling_FailedAllocation
     LEAF_END    RhNewString
 
-    INLINE_GETTHREAD_CONSTANT_POOL
-
-
 ;; Allocate one dimensional, zero based array (SZARRAY).
 ;;  x0 == MethodTable
 ;;  x1 == element count
@@ -213,8 +208,6 @@ ArraySizeOverflow
         b           RhExceptionHandling_FailedAllocation
     LEAF_END    RhpNewArray
 
-    INLINE_GETTHREAD_CONSTANT_POOL
-
 ;; Allocate one dimensional, zero based array (SZARRAY) using the slow path that calls a runtime helper.
 ;;  x0 == MethodTable
 ;;  x1 == element count

diff --git a/src/coreclr/nativeaot/Runtime/arm64/AsmMacros.h b/src/coreclr/nativeaot/Runtime/arm64/AsmMacros.h
index 3df6973..6fc40e8 100644 (file)
--- a/src/coreclr/nativeaot/Runtime/arm64/AsmMacros.h
+++ b/src/coreclr/nativeaot/Runtime/arm64/AsmMacros.h
@@ -211,18 +211,16 @@ $ReturnAddressName
 
 ;; -----------------------------------------------------------------------------
 ;;
-;; Macro to get a pointer to the Thread* object for the currently executing thread
+;; Macro to get a pointer to a threadlocal symbol for the currently executing thread
 ;;
 
 __tls_array     equ 0x58    ;; offsetof(TEB, ThreadLocalStoragePointer)
 
-    EXTERN _tls_index
-
-    GBLS __SECTIONREL_tls_CurrentThread
-__SECTIONREL_tls_CurrentThread SETS "SECTIONREL_tls_CurrentThread"
-
     MACRO
-        INLINE_GETTHREAD $destReg, $trashReg
+        INLINE_GET_TLS_VAR $destReg, $trashReg, $variable
+
+        EXTERN _tls_index
+        EXTERN $variable
 
         ;; The following macro variables are just some assembler magic to get the name of the 32-bit version
         ;; of $trashReg. It does it by string manipulation. Replaces something like x3 with w3.
@@ -230,32 +228,27 @@ __SECTIONREL_tls_CurrentThread SETS "SECTIONREL_tls_CurrentThread"
 TrashRegister32Bit SETS "$trashReg"
 TrashRegister32Bit SETS "w":CC:("$TrashRegister32Bit":RIGHT:((:LEN:TrashRegister32Bit) - 1))
 
-        ldr         $trashReg, =_tls_index
-        ldr         $TrashRegister32Bit, [$trashReg]
+        adrp        $destReg, _tls_index
+        ldr         $TrashRegister32Bit, [$destReg, _tls_index]
         ldr         $destReg, [xpr, #__tls_array]
-        ldr         $destReg, [$destReg, $trashReg lsl #3]
-        ldr         $trashReg, =$__SECTIONREL_tls_CurrentThread
-        ldr         $trashReg, [$trashReg]
-        add         $destReg, $destReg, $trashReg
+        ldr         $destReg, [$destReg, $TrashRegister32Bit uxtw #3]
+        add         $destReg, $destReg, #0, lsl #0xC
+        RELOC       0xA, $variable                          ;; IMAGE_REL_ARM64_SECREL_HIGH12A
+        add         $destReg, $destReg, #0, lsl #0
+        RELOC       0x9, $variable                          ;; IMAGE_REL_ARM64_SECREL_LOW12A
     MEND
 
-    ;; INLINE_GETTHREAD_CONSTANT_POOL macro has to be used after the last function in the .asm file that used
-    ;; INLINE_GETTHREAD. Optionally, it can be also used after any function that used INLINE_GETTHREAD
-    ;; to improve density, or to reduce distance between the constant pool and its use.
+;; -----------------------------------------------------------------------------
+;;
+;; Macro to get a pointer to the Thread* object for the currently executing thread
+;;
     MACRO
-        INLINE_GETTHREAD_CONSTANT_POOL
-        EXTERN tls_CurrentThread
-
-    ;; Section relocs are 32 bits. Using an extra DCD initialized to zero for 8-byte alignment.
-$__SECTIONREL_tls_CurrentThread
-        DCD tls_CurrentThread
-        RELOC 8, tls_CurrentThread      ;; SECREL
-        DCD 0
-
-__SECTIONREL_tls_CurrentThread SETS "$__SECTIONREL_tls_CurrentThread":CC:"_"
+        INLINE_GETTHREAD $destReg, $trashReg
 
+        INLINE_GET_TLS_VAR $destReg, $trashReg, tls_CurrentThread
     MEND
 
+
     MACRO
         INLINE_THREAD_UNHIJACK $threadReg, $trashReg1, $trashReg2
         ;;

diff --git a/src/coreclr/nativeaot/Runtime/arm64/ExceptionHandling.asm b/src/coreclr/nativeaot/Runtime/arm64/ExceptionHandling.asm
index 3db1d64..1af4fd1 100644 (file)
--- a/src/coreclr/nativeaot/Runtime/arm64/ExceptionHandling.asm
+++ b/src/coreclr/nativeaot/Runtime/arm64/ExceptionHandling.asm
@@ -624,6 +624,4 @@ SetSuccess
 
     NESTED_END RhpCallFilterFunclet
 
-    INLINE_GETTHREAD_CONSTANT_POOL
-
     end

diff --git a/src/coreclr/nativeaot/Runtime/arm64/GcProbe.asm b/src/coreclr/nativeaot/Runtime/arm64/GcProbe.asm
index ef01e43..d95ab7a 100644 (file)
--- a/src/coreclr/nativeaot/Runtime/arm64/GcProbe.asm
+++ b/src/coreclr/nativeaot/Runtime/arm64/GcProbe.asm
@@ -366,7 +366,5 @@ Success
     LEAF_END RhpSuppressGcStress
 #endif ;; FEATURE_GC_STRESS
 
-    INLINE_GETTHREAD_CONSTANT_POOL
-
     end
 

diff --git a/src/coreclr/nativeaot/Runtime/arm64/GetThread.asm b/src/coreclr/nativeaot/Runtime/arm64/GetThread.asm
deleted file mode 100644 (file)
index 7c01e66..0000000
--- a/src/coreclr/nativeaot/Runtime/arm64/GetThread.asm
+++ /dev/null
@@ -1,29 +0,0 @@
-;; Licensed to the .NET Foundation under one or more agreements.
-;; The .NET Foundation licenses this file to you under the MIT license.
-
-#include "AsmMacros.h"
-
-        TEXTAREA
-
-;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
-;;
-;; RhpGetThread
-;;
-;;
-;; INPUT: none
-;;
-;; OUTPUT: x9: Thread pointer
-;;
-;; MUST PRESERVE ARGUMENT REGISTERS
-;;
-;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
-    LEAF_ENTRY RhpGetThread
-        ;; x9 = GetThread(), TRASHES xip0 (which can be used as an intra-procedure-call scratch register)
-        INLINE_GETTHREAD x9, xip0
-        ret
-    LEAF_END
-FASTCALL_ENDFUNC
-
-    INLINE_GETTHREAD_CONSTANT_POOL
-
-    end

diff --git a/src/coreclr/nativeaot/Runtime/arm64/PInvoke.asm b/src/coreclr/nativeaot/Runtime/arm64/PInvoke.asm
index 475737f..52d55ad 100644 (file)
--- a/src/coreclr/nativeaot/Runtime/arm64/PInvoke.asm
+++ b/src/coreclr/nativeaot/Runtime/arm64/PInvoke.asm
@@ -34,8 +34,6 @@
         ret
     NESTED_END RhpPInvoke
 
-    INLINE_GETTHREAD_CONSTANT_POOL
-
 
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ;;

diff --git a/src/coreclr/nativeaot/Runtime/i386/GetThread.asm b/src/coreclr/nativeaot/Runtime/i386/GetThread.asm
deleted file mode 100644 (file)
index b330406..0000000
--- a/src/coreclr/nativeaot/Runtime/i386/GetThread.asm
+++ /dev/null
@@ -1,31 +0,0 @@
-;; Licensed to the .NET Foundation under one or more agreements.
-;; The .NET Foundation licenses this file to you under the MIT license.
-
-        .586
-        .model  flat
-        option  casemap:none
-        .code
-
-
-include AsmMacros.inc
-
-;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
-;;
-;; RhpGetThread
-;;
-;;
-;; INPUT: none
-;;
-;; OUTPUT: EAX: Thread pointer
-;;
-;; MUST PRESERVE ARGUMENT REGISTERS
-;;
-;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
-FASTCALL_FUNC RhpGetThread, 0
-        push    ecx
-        INLINE_GETTHREAD eax, ecx ; eax dest, ecx trash
-        pop     ecx
-        ret
-FASTCALL_ENDFUNC
-
-        end