Update fast alloc pattern on arm32 to use (combined_limit - alloc_ptr) >= size, fix few typos in comments

filipnavara · filipnavara · commit 3ac08ee41d09 · 2025-06-02T13:00:14.000+02:00
diff --git a/src/coreclr/runtime/amd64/AllocFast.S b/src/coreclr/runtime/amd64/AllocFast.S
@@ -41,7 +41,7 @@ LEAF_ENTRY RhpNewFast, _TEXT
         // Calculate the new alloc pointer to account for the allocation.
         add         rdx, rsi
 
-        // Set the new objects MethodTable pointer.
+        // Set the new object's MethodTable pointer.
         mov         [rsi + OFFSETOF__Object__m_pEEType], rbx
 
         // Update the alloc pointer to the newly calculated one.
diff --git a/src/coreclr/runtime/arm/AllocFast.S b/src/coreclr/runtime/arm/AllocFast.S
@@ -10,75 +10,84 @@
 //  r0 == MethodTable
 .macro NEW_FAST Variation
         PROLOG_PUSH "{r4,lr}"
-        mov	        r4, r0 // save MethodTable
+        mov	    r4, r0 // save MethodTable
 
         // r0 = ee_alloc_context pointer; trashes volatile registers, expects saved lr
         INLINE_GET_ALLOC_CONTEXT_BASE
 
+        ldr         r2, [r4, #OFFSETOF__MethodTable__m_uBaseSize]
+
+        // Load potential new object address into r3.
+        ldr         r3, [r0, #(OFFSETOF__ee_alloc_context + OFFSETOF__ee_alloc_context__alloc_ptr)]
+
+        // Load and calculate the maximum size of object we can fit.
+        ldr         r1, [r0, #(OFFSETOF__ee_alloc_context + OFFSETOF__ee_alloc_context__combined_limit)]
+        sub         r1, r3
+
         // When doing aligned or misaligned allocation we first check
         // the alignment and skip to the regular path if it's already
         // matching the expectation.
         // Otherwise, we try to allocate size + ASM_MIN_OBJECT_SIZE and
         // then prepend a dummy free object at the beginning of the
         // allocation.
 .ifnc \Variation,
-        ldr         r3, [r0, #(OFFSETOF__ee_alloc_context + OFFSETOF__ee_alloc_context__alloc_ptr)]
         tst         r3, #0x7
 .ifc \Variation,Align8
         beq         1f // AlreadyAligned
 .else // Variation == "Misalign"
         bne         1f // AlreadyAligned
 .endif
 
-        ldr         r2, [r4, #OFFSETOF__MethodTable__m_uBaseSize]
         add         r2, ASM_MIN_OBJECT_SIZE
-        ldr         r3, [r0, #(OFFSETOF__ee_alloc_context + OFFSETOF__ee_alloc_context__alloc_ptr)]
-        add         r2, r3
-        bcs         2f // AllocFailed
-        ldr         r1, [r0, #(OFFSETOF__ee_alloc_context + OFFSETOF__ee_alloc_context__combined_limit)]
+
+        // Determine whether the end of the object is too big for the current allocation context. If so,
+        // we abandon the attempt to allocate the object directly and fall back to the slow helper.
         cmp         r2, r1
         bhi         2f // AllocFailed
 
-        // set the new alloc pointer
+        // Update the alloc pointer to account for the allocation.
+        add         r2, r3
         str         r2, [r0, #(OFFSETOF__ee_alloc_context + OFFSETOF__ee_alloc_context__alloc_ptr)]
 
-        // initialize the padding object preceeding the new object
+        // Initialize the padding object preceeding the new object.
         PREPARE_EXTERNAL_VAR_INDIRECT G_FREE_OBJECT_METHOD_TABLE, r2
         str         r2, [r3, #OFFSETOF__Object__m_pEEType]
         mov         r2, #0
         str         r2, [r3, #OFFSETOF__Array__m_Length]
 
-        // calc the new object pointer and initialize it
+        // Calculate the new object pointer and initialize it.
         add         r3, ASM_MIN_OBJECT_SIZE
         str         r4, [r3, #OFFSETOF__Object__m_pEEType]
 
+        // Return the object allocated in r0.
         mov         r0, r3
 
         EPILOG_POP  "{r4,pc}"
 .endif // Variation != ""
 
 1: // AlreadyAligned
 
-        // r4 contains MethodTable pointer
-        ldr         r2, [r4, #OFFSETOF__MethodTable__m_uBaseSize]
-
         // r0: ee_alloc_context pointer
-        // r4: MethodTable pointer
+        // r1: ee_alloc_context.combined_limit
         // r2: base size
+        // r3: ee_alloc_context.alloc_ptr
+        // r4: MethodTable pointer
 
-        ldr         r3, [r0, #(OFFSETOF__ee_alloc_context + OFFSETOF__ee_alloc_context__alloc_ptr)]
-        add         r2, r3
-        bcs         2f // AllocFailed
-        ldr         r1, [r0, #(OFFSETOF__ee_alloc_context + OFFSETOF__ee_alloc_context__combined_limit)]
+        // Determine whether the end of the object is too big for the current allocation context. If so,
+        // we abandon the attempt to allocate the object directly and fall back to the slow helper.
         cmp         r2, r1
         bhi         2f // AllocFailed
 
-        // set the new alloc pointer
-        str         r2, [r0, #(OFFSETOF__ee_alloc_context + OFFSETOF__ee_alloc_context__alloc_ptr)]
+        // Calculate the new alloc pointer to account for the allocation.
+        add         r2, r3
 
-        // Set the new object's MethodTable pointer
+        // Set the new object's MethodTable pointer.
         str         r4, [r3, #OFFSETOF__Object__m_pEEType]
 
+        // Update the alloc pointer to the newly calculated one.
+        str         r2, [r0, #(OFFSETOF__ee_alloc_context + OFFSETOF__ee_alloc_context__alloc_ptr)]
+
+        // Return the object allocated in r0.
         mov         r0, r3
 
         EPILOG_POP  "{r4,pc}"
@@ -198,27 +207,27 @@ NESTED_END RhpNewObject, _TEXT
         // r6 == string/array size
         // r0 == ee_alloc_context*
 
-        // Load potential new object address into r12.
-        ldr         r12, [r0, #(OFFSETOF__ee_alloc_context + OFFSETOF__ee_alloc_context__alloc_ptr)]
+        // Load potential new object address into r3.
+        ldr         r3, [r0, #(OFFSETOF__ee_alloc_context + OFFSETOF__ee_alloc_context__alloc_ptr)]
 
-        // Determine whether the end of the object would lie outside of the current allocation context. If so,
-        // we abandon the attempt to allocate the object directly and fall back to the slow helper.
-        adds        r6, r12
-        bcs         1f // if we get a carry here, the string/array is too large to fit below 4 GB
+        // Load and calculate the maximum size of object we can fit
+        ldr         r1, [r0, #(OFFSETOF__ee_alloc_context + OFFSETOF__ee_alloc_context__combined_limit)]
+        sub         r1, r3
 
-        ldr         r12, [r0, #(OFFSETOF__ee_alloc_context + OFFSETOF__ee_alloc_context__combined_limit)]
-        cmp         r6, r12
+        // Determine whether the end of the object is too big for the current allocation context. If so,
+        // we abandon the attempt to allocate the object directly and fall back to the slow helper.
+        cmp         r6, r1
         bhi         1f
 
-        // Reload new object address into r12.
-        ldr         r12, [r0, #(OFFSETOF__ee_alloc_context + OFFSETOF__ee_alloc_context__alloc_ptr)]
-
-        // Update the alloc pointer to account for the allocation.
-        str         r6, [r0, #(OFFSETOF__ee_alloc_context + OFFSETOF__ee_alloc_context__alloc_ptr)]
+        // Calculate the alloc pointer to account for the allocation.
+        add         r6, r3
 
         // Set the new object's MethodTable pointer and element count.
-        str         r4, [r12, #OFFSETOF__Object__m_pEEType]
-        str         r5, [r12, #OFFSETOF__Array__m_Length]
+        str         r4, [r3, #OFFSETOF__Object__m_pEEType]
+        str         r5, [r3, #OFFSETOF__Array__m_Length]
+
+        // Update the alloc pointer to the newly calculated one.
+        str         r6, [r0, #(OFFSETOF__ee_alloc_context + OFFSETOF__ee_alloc_context__alloc_ptr)]
 
         // Return the object allocated in r0.
         mov         r0, r12
diff --git a/src/coreclr/runtime/arm64/AllocFast.S b/src/coreclr/runtime/arm64/AllocFast.S
@@ -37,7 +37,7 @@
         // Calculate the new alloc pointer to account for the allocation.
         add         x2, x2, x12
 
-        // Set the new objects MethodTable pointer.
+        // Set the new object's MethodTable pointer.
         str         x0, [x12, #OFFSETOF__Object__m_pEEType]
 
         // Update the alloc pointer to the newly calculated one.
@@ -79,7 +79,7 @@ LOCAL_LABEL(RhpNewFast_RarePath):
         // void* RhpGcAlloc(MethodTable *pEEType, uint32_t uFlags, uintptr_t numElements, void * pTransitionFrame)
         bl          C_FUNC(RhpGcAlloc)
 
-        // Set the new objects MethodTable pointer on success.
+        // Set the new object's MethodTable pointer on success.
         cbz         x0, LOCAL_LABEL(NewOutOfMemory)
 
         .cfi_remember_state
@@ -111,7 +111,7 @@ LOCAL_LABEL(NewOutOfMemory):
         // Load potential new object address into x12.
         ldr         x12, [x3, #(OFFSETOF__ee_alloc_context + OFFSETOF__ee_alloc_context__alloc_ptr)]
 
-        // Load and calculate the maximum size of object we can fit
+        // Load and calculate the maximum size of object we can fit.
         ldr         x13, [x3, #(OFFSETOF__ee_alloc_context + OFFSETOF__ee_alloc_context__combined_limit)]
         sub         x13, x13, x12
 
@@ -124,10 +124,10 @@ LOCAL_LABEL(NewOutOfMemory):
         bhi         C_FUNC(RhpNewVariableSizeObject)
 #endif
 
-        // Calculate the alloc pointer to account for the allocation.
+        // Calculate the new alloc pointer to account for the allocation.
         add         x2, x2, x12
 
-        // Set the new objects MethodTable pointer and element count.
+        // Set the new object's MethodTable pointer and element count.
         str         x0, [x12, #OFFSETOF__Object__m_pEEType]
         str         x1, [x12, #OFFSETOF__Array__m_Length]
 
@@ -263,7 +263,7 @@ LOCAL_LABEL(ArraySizeOverflow):
         // void* RhpGcAlloc(MethodTable *pEEType, uint32_t uFlags, uintptr_t numElements, void * pTransitionFrame)
         bl          C_FUNC(RhpGcAlloc)
 
-        // Set the new objects MethodTable pointer and length on success.
+        // Set the new object's MethodTable pointer and length on success.
         cbz         x0, LOCAL_LABEL(RhpNewVariableSizeObject_OutOfMemory)
 
         .cfi_remember_state
diff --git a/src/coreclr/runtime/arm64/AllocFast.asm b/src/coreclr/runtime/arm64/AllocFast.asm
@@ -27,7 +27,7 @@
         ;; Load potential new object address into x12.
         ldr         x12, [x3, #(OFFSETOF__ee_alloc_context + OFFSETOF__ee_alloc_context__alloc_ptr)]
 
-        ;; Load and calculate the maximum size of object we can fit
+        ;; Load and calculate the maximum size of object we can fit.
         ldr         x13, [x3, #(OFFSETOF__ee_alloc_context + OFFSETOF__ee_alloc_context__combined_limit)]
         sub         x13, x13, x12
 
@@ -36,10 +36,10 @@
         cmp         x2, x13
         bhi         RhpNewFast_RarePath
 
-        ;; Calculate the alloc pointer to account for the allocation.
+        ;; Calculate the new alloc pointer to account for the allocation.
         add         x2, x2, x12
 
-        ;; Set the new objects MethodTable pointer.
+        ;; Set the new object's MethodTable pointer.
         str         x0, [x12, #OFFSETOF__Object__m_pEEType]
 
         ;; Update the alloc pointer to the newly calculated one.
@@ -119,7 +119,7 @@ NewOutOfMemory
         ;; Calculate the new alloc pointer to account for the allocation.
         add         x2, x2, x12
 
-        ;; Set the new objects MethodTable pointer and element count.
+        ;; Set the new object's MethodTable pointer and element count.
         str         x0, [x12, #OFFSETOF__Object__m_pEEType]
         str         x1, [x12, #OFFSETOF__Array__m_Length]
 
diff --git a/src/coreclr/runtime/loongarch64/AllocFast.S b/src/coreclr/runtime/loongarch64/AllocFast.S
@@ -36,7 +36,7 @@
         // Calculate the new alloc pointer to account for the allocation.
         add.d  $a2, $a2, $t3
 
-        // Set the new objects MethodTable pointer.
+        // Set the new object's MethodTable pointer.
         st.d  $a0, $t3, OFFSETOF__Object__m_pEEType
 
         // Update the alloc pointer to the newly calculated one.
@@ -75,7 +75,7 @@ LOCAL_LABEL(RhpNewFast_RarePath):
         // void* RhpGcAlloc(MethodTable *pEEType, uint32_t uFlags, uintptr_t numElements, void * pTransitionFrame)
         bl  C_FUNC(RhpGcAlloc)
 
-        // Set the new objects MethodTable pointer on success.
+        // Set the new object's MethodTable pointer on success.
         beqz  $a0, LOCAL_LABEL(NewOutOfMemory)
 
         .cfi_remember_state
@@ -117,7 +117,7 @@ LOCAL_LABEL(NewOutOfMemory):
         // Calculate the new alloc pointer to account for the allocation.
         add.d  $a2, $a2, $t3
 
-        // Set the new objects MethodTable pointer and element count.
+        // Set the new object's MethodTable pointer and element count.
         st.d  $a0, $t3, OFFSETOF__Object__m_pEEType
         st.d  $a1, $t3, OFFSETOF__Array__m_Length
 
@@ -233,7 +233,7 @@ LOCAL_LABEL(ArraySizeOverflow):
         // void* RhpGcAlloc(MethodTable *pEEType, uint32_t uFlags, uintptr_t numElements, void * pTransitionFrame)
         bl  C_FUNC(RhpGcAlloc)
 
-        // Set the new objects MethodTable pointer and length on success.
+        // Set the new object's MethodTable pointer and length on success.
         beqz  $a0, LOCAL_LABEL(RhpNewVariableSizeObject_OutOfMemory)
 
         .cfi_remember_state
diff --git a/src/coreclr/runtime/riscv64/AllocFast.S b/src/coreclr/runtime/riscv64/AllocFast.S
@@ -41,7 +41,7 @@
         // Calculate the new alloc pointer to account for the allocation.
         add   t0, t0, t1
 
-        // Set the new objects MethodTable pointer.
+        // Set the new object's MethodTable pointer.
         sd    s1, OFFSETOF__Object__m_pEEType(t1)
 
         // Update the alloc pointer to the newly calculated one.
@@ -144,7 +144,7 @@ LOCAL_LABEL(NewOutOfMemory):
         // Calculate the new alloc pointer to account for the allocation.
         add   t0, t0, t1
 
-        // Set the new objects MethodTable pointer.
+        // Set the new object's MethodTable pointer.
         sd    s1, OFFSETOF__Object__m_pEEType(t1)
         sd    s2, OFFSETOF__Array__m_Length(t1)
 
