[SVE] Don't use LocalStackAllocation for SVE objects

I have introduced a new TargetFrameLowering query function:

  isStackIdSafeForLocalArea

that queries whether or not it is safe for objects of a given stack
id to be bundled into the local area. The default behaviour is to
always bundle regardless of the stack id, however for AArch64 this is
overriden so that it's only safe for fixed-size stack objects.
There is future work here to extend this algorithm for multiple local
areas so that SVE stack objects can be bundled together and accessed
from their own virtual base-pointer.

Differential Revision: https://reviews.llvm.org/D83859

llvm/include/llvm/CodeGen/TargetFrameLowering.h

@@ -134,6 +134,12 @@ class TargetFrameLowering {
   /// was called).
   virtual unsigned getStackAlignmentSkew(const MachineFunction &MF) const;
 
+  /// This method returns whether or not it is safe for an object with the
+  /// given stack id to be bundled into the local area.
+  virtual bool isStackIdSafeForLocalArea(unsigned StackId) const {
+    return true;
+  }
+
   /// getOffsetOfLocalArea - This method returns the offset of the local area
   /// from the stack pointer on entrance to a function.
   ///

llvm/lib/CodeGen/LocalStackSlotAllocation.cpp

@@ -220,6 +220,8 @@ void LocalStackSlotPass::calculateFrameObjectOffsets(MachineFunction &Fn) {
         continue;
       if (StackProtectorFI == (int)i)
         continue;
+      if (!TFI.isStackIdSafeForLocalArea(MFI.getStackID(i)))
+        continue;
 
       switch (MFI.getObjectSSPLayout(i)) {
       case MachineFrameInfo::SSPLK_None:
@@ -254,6 +256,8 @@ void LocalStackSlotPass::calculateFrameObjectOffsets(MachineFunction &Fn) {
       continue;
     if (ProtectedObjs.count(i))
       continue;
+    if (!TFI.isStackIdSafeForLocalArea(MFI.getStackID(i)))
+      continue;
 
     AdjustStackOffset(MFI, i, Offset, StackGrowsDown, MaxAlign);
   }

llvm/lib/Target/AArch64/AArch64FrameLowering.h

@@ -105,6 +105,12 @@ class AArch64FrameLowering : public TargetFrameLowering {
     }
   }
 
+  bool isStackIdSafeForLocalArea(unsigned StackId) const override {
+    // We don't support putting SVE objects into the pre-allocated local
+    // frame block at the moment.
+    return StackId != TargetStackID::SVEVector;
+  }
+
 private:
   bool shouldCombineCSRLocalStackBump(MachineFunction &MF,
                                       uint64_t StackBumpBytes) const;

llvm/test/CodeGen/AArch64/sve-localstackalloc.mir

@@ -0,0 +1,61 @@
+# RUN: llc -mtriple=aarch64--linux-gnu -mattr=+sve -run-pass=localstackalloc -o - %s | FileCheck %s
+
+--- |
+  ; ModuleID = '<stdin>'
+  source_filename = "<stdin>"
+  target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
+  target triple = "aarch64-unknown-linux-gnu"
+
+  define <vscale x 32 x i8> @insert_32i8_idx(<vscale x 32 x i8> %a, i8 %elt, i64 %idx) #0 {
+    %ins = insertelement <vscale x 32 x i8> %a, i8 %elt, i64 %idx
+    ret <vscale x 32 x i8> %ins
+  }
+
+  attributes #0 = { "target-features"="+sve" }
+
+...
+---
+name:            insert_32i8_idx
+alignment:       4
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: zpr, preferred-register: '' }
+  - { id: 1, class: zpr, preferred-register: '' }
+  - { id: 2, class: gpr32, preferred-register: '' }
+  - { id: 3, class: gpr64, preferred-register: '' }
+  - { id: 5, class: ppr_3b, preferred-register: '' }
+  - { id: 6, class: gpr64sp, preferred-register: '' }
+  - { id: 7, class: zpr, preferred-register: '' }
+  - { id: 8, class: zpr, preferred-register: '' }
+liveins:
+  - { reg: '$z0', virtual-reg: '%0' }
+  - { reg: '$z1', virtual-reg: '%1' }
+  - { reg: '$w0', virtual-reg: '%2' }
+frameInfo:
+  maxAlignment:    1
+  maxCallFrameSize: 0
+# CHECK-LABEL: name: insert_32i8_idx
+# CHECK: localFrameSize:  0
+stack:
+  - { id: 0, name: '', type: default, offset: 0, size: 32, alignment: 16,
+      stack-id: sve-vec, callee-saved-register: '', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+machineFunctionInfo: {}
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: $z0, $z1, $w0
+
+    %2:gpr32 = COPY $w0
+    %1:zpr = COPY $z1
+    %0:zpr = COPY $z0
+    %5:ppr_3b = PTRUE_B 31
+    %6:gpr64sp = ADDXri %stack.0, 0, 0
+    ST1B_IMM %1, %5, %6, 1 :: (store unknown-size, align 16)
+    ST1B_IMM %0, %5, %stack.0, 0 :: (store unknown-size into %stack.0, align 16)
+    %7:zpr = LD1B_IMM %5, %6, 1 :: (load unknown-size from %stack.0 + 16, align 16)
+    %8:zpr = LD1B_IMM %5, %stack.0, 0 :: (load unknown-size from %stack.0, align 16)
+    $z0 = COPY %8
+    $z1 = COPY %7
+    RET_ReallyLR implicit $z0, implicit $z1
+
+...