ARM64: optimize PREPARE_EXTERNAL:VAR (dotnet#21) (dotnet#8296)

There are multiple cases were we don't need just the pointer for an external variable but the value that is stored at this location.
So far this was done with the PREPARE_EXTERNAL_VAR followed by an ldr x? [x?]. The PREPARE_EXTERNAL_VAR macro needs two instructions (adrp + add). As the ldr instruction supports an offset we can eliminate the add for this use case. The two new macros PREPARE_EXTERNAL_VAR_INDIRECT and PREPARE_EXTERNAL_VAR_INDIRECT_W make use of this.

src/coreclr/src/nativeaot/Runtime/arm64/ExceptionHandling.S

@@ -465,9 +465,8 @@ PopExInfoLoop:
 DonePopping:
         str         x3, [x1, #OFFSETOF__Thread__m_pExInfoStackHead]     // store the new head on the Thread
 
-        PREPARE_EXTERNAL_VAR RhpTrapThreads, x3 
+        PREPARE_EXTERNAL_VAR_INDIRECT_W RhpTrapThreads, 3 
 
-        ldr         w3, [x3]
         tbz         x3, #TrapThreadsFlags_AbortInProgress_Bit, NoAbort
 
         ldr         x3, [sp, #rsp_offset_is_not_handling_thread_abort]

src/coreclr/src/nativeaot/Runtime/arm64/InteropThunksHelpers.S

@@ -40,8 +40,7 @@ POINTER_SIZE = 0x08
     // IntPtr RhGetCommonStubAddress()
     //
     LEAF_ENTRY RhGetCommonStubAddress, _TEXT
-        adrp x0, RhCommonStub
-        add	x0, x0, :lo12:RhCommonStub
+        PREPARE_EXTERNAL_VAR RhCommonStub, x0
         ret
     LEAF_END RhGetCommonStubAddress, _TEXT
 

src/coreclr/src/nativeaot/Runtime/arm64/PInvoke.S

@@ -118,9 +118,8 @@ Done:
 
         PROLOG_SAVE_REG_PAIR_INDEXED    fp, lr, #-0x10
 
-        PREPARE_EXTERNAL_VAR RhpTrapThreads, x10
+        PREPARE_EXTERNAL_VAR_INDIRECT_W RhpTrapThreads, 10
 
-        ldr         w10, [x10]
         tbz         x10, #TrapThreadsFlags_TrapThreads_Bit, NoWait
         bl          RhpWaitForGCNoAbort
 NoWait:
@@ -181,9 +180,8 @@ ThreadAttached:
         str         xzr, [x10, #OFFSETOF__Thread__m_pTransitionFrame] 
         dmb         ish
 
-        PREPARE_EXTERNAL_VAR RhpTrapThreads, x11
+        PREPARE_EXTERNAL_VAR_INDIRECT_W RhpTrapThreads, 11
 
-        ldr         w11, [x11]
         tbnz        x11, #TrapThreadsFlags_TrapThreads_Bit, TrapThread
 
         ret
@@ -294,9 +292,8 @@ NESTED_ENTRY RhpPInvoke, _TEXT, NoHandler
         str     x10, [x0, #OFFSETOF__PInvokeTransitionFrame__m_pThread]
         str     x0, [x10, #OFFSETOF__Thread__m_pTransitionFrame]
 
-        PREPARE_EXTERNAL_VAR RhpTrapThreads, x9
+        PREPARE_EXTERNAL_VAR_INDIRECT_W RhpTrapThreads, 9
 
-        ldr     w9, [x9]
         cbnz    w9, InvokeRareTrapThread  // TrapThreadsFlags_None = 0
         ret
 
@@ -310,9 +307,8 @@ LEAF_ENTRY RhpPInvokeReturn, _TEXT
         mov         x10, 0
         str         x10, [x9, #OFFSETOF__Thread__m_pTransitionFrame]
 
-        PREPARE_EXTERNAL_VAR RhpTrapThreads, x9
+        PREPARE_EXTERNAL_VAR_INDIRECT_W RhpTrapThreads, 9
 
-        ldr     w9, [x9]
         cbnz    w9, 0f  // TrapThreadsFlags_None = 0
         ret
 0:

src/coreclr/src/nativeaot/Runtime/arm64/StubDispatch.S

@@ -99,8 +99,7 @@
         // Calling convention of the universal thunk is:
         //  xip0: contains target address for the thunk to call
         //  xip1: contains parameter of the thunks target
-        adrp xip0, RhpCidResolve
-        add	xip0, xip0, :lo12:RhpCidResolve
+        PREPARE_EXTERNAL_VAR RhpCidResolve, xip0
         mov xip1, x11
         b       RhpUniversalTransition_DebugStepTailCall
     LEAF_END RhpInterfaceDispatchSlow, _TEXT

src/coreclr/src/nativeaot/Runtime/arm64/WriteBarriers.S

@@ -28,26 +28,22 @@
         .macro UPDATE_GC_SHADOW destReg, refReg
 
         // If g_GCShadow is 0, don't perform the check.
-        PREPARE_EXTERNAL_VAR g_GCShadow, X9
-        ldr     x9, [x9]
+        PREPARE_EXTERNAL_VAR_INDIRECT g_GCShadow, X9
         cbz     x9, 1f
 
         // Save destReg since we're about to modify it (and we need the original value both within the macro and
         // once we exit the macro).
         mov     x10, \destReg
 
         // Transform destReg into the equivalent address in the shadow heap.
-        PREPARE_EXTERNAL_VAR g_lowest_address, X9
-        ldr     x9, [x9]
+        PREPARE_EXTERNAL_VAR_INDIRECT g_lowest_address, X9
         subs    \destReg, \destReg, x9
         blt     0f
 
-        PREPARE_EXTERNAL_VAR g_GCShadow, X9
-        ldr     x9, [x9]
+        PREPARE_EXTERNAL_VAR_INDIRECT g_GCShadow, X9
         add     \destReg, \destReg, x9
 
-        PREPARE_EXTERNAL_VAR g_GCShadowEnd, X9
-        ldr     x9, [x9]
+        PREPARE_EXTERNAL_VAR_INDIRECT g_GCShadowEnd, X9
         cmp     \destReg, x9
         bgt     0f
 
@@ -109,8 +105,8 @@
 
 #ifdef FEATURE_USE_SOFTWARE_WRITE_WATCH_FOR_GC_HEAP
         // Update the write watch table if necessary
-        PREPARE_EXTERNAL_VAR g_write_watch_table, x\trash
-        ldr  x\trash, [x\trash]
+        PREPARE_EXTERNAL_VAR_INDIRECT g_write_watch_table, x\trash
+
         cbz  x\trash, 2f
         add  x\trash, x\trash, \destReg, lsr #0xc  // SoftwareWriteWatch::AddressToTableByteIndexShift
         ldrb w17, [x\trash]
@@ -122,20 +118,17 @@
 2:
         // We can skip the card table write if the reference is to
         // an object not on the epehemeral segment.
-        PREPARE_EXTERNAL_VAR g_ephemeral_low, x\trash
-        ldr     x\trash, [x\trash]
+        PREPARE_EXTERNAL_VAR_INDIRECT g_ephemeral_low, x\trash
         cmp     \refReg, x\trash
         blt     0f
 
-        PREPARE_EXTERNAL_VAR g_ephemeral_high, x\trash
-        ldr     x\trash, [x\trash]
+        PREPARE_EXTERNAL_VAR_INDIRECT g_ephemeral_high, x\trash
         cmp     \refReg, x\trash
         bge     0f
 
         // Set this objects card, if it has not already been set.
 
-        PREPARE_EXTERNAL_VAR g_card_table, x\trash
-        ldr     x\trash, [x\trash]
+        PREPARE_EXTERNAL_VAR_INDIRECT g_card_table, x\trash
         add     \trash2, x\trash, \destReg, lsr #11
 
         // Check that this card has not already been written. Avoiding useless writes is a big win on
@@ -149,8 +142,7 @@
 
 #ifdef FEATURE_MANUALLY_MANAGED_CARD_BUNDLES
         // Check if we need to update the card bundle table
-        PREPARE_EXTERNAL_VAR g_card_bundle_table, x\trash
-        ldr     x\trash, [x\trash]
+        PREPARE_EXTERNAL_VAR_INDIRECT g_card_bundle_table, x\trash
         add  \trash2, x\trash, \destReg, lsr #21
         ldrb w\trash, [\trash2]
         cmp  x\trash, 0xFF
@@ -178,13 +170,11 @@
         // The "check" of this checked write barrier - is destReg
         // within the heap? if no, early out.
 
-        PREPARE_EXTERNAL_VAR g_lowest_address, x\trash
-        ldr     x\trash, [x\trash]
+        PREPARE_EXTERNAL_VAR_INDIRECT g_lowest_address, x\trash
         cmp     \destReg, x\trash
         blt     0f
 
-        PREPARE_EXTERNAL_VAR g_highest_address, x\trash
-        ldr     x\trash, [x\trash]
+        PREPARE_EXTERNAL_VAR_INDIRECT g_highest_address, x\trash
         cmp     \destReg, x\trash
         bgt     0f
 

src/coreclr/src/nativeaot/Runtime/unix/unixasmmacrosarm64.inc

@@ -46,6 +46,17 @@ C_FUNC(\Name):
         add	\HelperReg, \HelperReg, :lo12:C_FUNC(\Name)
 .endm 
 
+.macro PREPARE_EXTERNAL_VAR_INDIRECT Name, HelperReg
+        adrp \HelperReg, C_FUNC(\Name)
+        ldr	\HelperReg, [\HelperReg, :lo12:C_FUNC(\Name)]
+.endm 
+
+.macro PREPARE_EXTERNAL_VAR_INDIRECT_W Name, HelperReg
+        adrp x\HelperReg, C_FUNC(\Name)
+        ldr	w\HelperReg, [x\HelperReg, :lo12:C_FUNC(\Name)]
+.endm 
+
+
 .macro PROLOG_STACK_ALLOC Size
         sub sp, sp, \Size
         .cfi_adjust_cfa_offset \Size