[PATCH] D138208: [AArch64][SME] Always allocate a lazy-save buffer if a function has ZA state.

[Sander de Smalen via Phabricator via llvm-commits]

sdesmalen created this revision.
sdesmalen added reviewers: paulwalker-arm, kmclaughlin, aemerson.
Herald added subscribers: hiraditya, kristof.beyls.
Herald added a project: All.
sdesmalen requested review of this revision.
Herald added a project: LLVM.
Herald added a subscriber: llvm-commits.

We already do this for most cases, with the exception of instructions that
get expanded to function calls (e.g. for lowering operations on fp128
values), in which case we temporarily allocate a lazy-save buffer.

The code that is generated in this case, is however incorrect, as it seems
to pass an incorrect address for the TPIDR2 object to the ZA restore
function. By always allocating the lazy-save buffer once, we avoid this
issue entirely.

The cost is that we also allocate such a buffer when it is not
needed. We could fix that in a follow-up patch, where we remove the
lazy-save buffer when it isn't used.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D138208

Files:
  llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
  llvm/test/CodeGen/AArch64/sme-disable-gisel-fisel.ll
  llvm/test/CodeGen/AArch64/sme-shared-za-interface.ll


Index: llvm/test/CodeGen/AArch64/sme-shared-za-interface.ll
===================================================================
--- llvm/test/CodeGen/AArch64/sme-shared-za-interface.ll
+++ llvm/test/CodeGen/AArch64/sme-shared-za-interface.ll
@@ -48,12 +48,12 @@
 ; CHECK-NEXT:    sub x9, x9, x8
 ; CHECK-NEXT:    mov sp, x9
 ; CHECK-NEXT:    sub x10, x29, #16
-; CHECK-NEXT:    sturh w8, [x29, #-8]
 ; CHECK-NEXT:    stur x9, [x29, #-16]
+; CHECK-NEXT:    sturh w8, [x29, #-8]
 ; CHECK-NEXT:    msr TPIDR2_EL0, x10
 ; CHECK-NEXT:    bl __addtf3
 ; CHECK-NEXT:    smstart za
-; CHECK-NEXT:    add x0, x29, #0
+; CHECK-NEXT:    sub x0, x29, #16
 ; CHECK-NEXT:    mrs x8, TPIDR2_EL0
 ; CHECK-NEXT:    cbnz x8, .LBB1_2
 ; CHECK-NEXT:  // %bb.1:
Index: llvm/test/CodeGen/AArch64/sme-disable-gisel-fisel.ll
===================================================================
--- llvm/test/CodeGen/AArch64/sme-disable-gisel-fisel.ll
+++ llvm/test/CodeGen/AArch64/sme-disable-gisel-fisel.ll
@@ -310,12 +310,12 @@
 ; CHECK-COMMON-NEXT:    sub x9, x9, x8
 ; CHECK-COMMON-NEXT:    mov sp, x9
 ; CHECK-COMMON-NEXT:    sub x10, x29, #16
-; CHECK-COMMON-NEXT:    sturh w8, [x29, #-8]
 ; CHECK-COMMON-NEXT:    stur x9, [x29, #-16]
+; CHECK-COMMON-NEXT:    sturh w8, [x29, #-8]
 ; CHECK-COMMON-NEXT:    msr TPIDR2_EL0, x10
 ; CHECK-COMMON-NEXT:    bl __addtf3
 ; CHECK-COMMON-NEXT:    smstart za
-; CHECK-COMMON-NEXT:    add x0, x29, #0
+; CHECK-COMMON-NEXT:    sub x0, x29, #16
 ; CHECK-COMMON-NEXT:    mrs x8, TPIDR2_EL0
 ; CHECK-COMMON-NEXT:    cbnz x8, .LBB8_2
 ; CHECK-COMMON-NEXT:  // %bb.1:
Index: llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
===================================================================
--- llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -5997,21 +5997,6 @@
 }
 
 
-/// Returns true if the Function has ZA state and contains at least one call to
-/// a function that requires setting up a lazy-save buffer.
-static bool requiresBufferForLazySave(const Function &F) {
-  SMEAttrs CallerAttrs(F);
-  if (!CallerAttrs.hasZAState())
-    return false;
-
-  for (const BasicBlock &BB : F)
-    for (const Instruction &I : BB)
-      if (const CallInst *Call = dyn_cast<CallInst>(&I))
-        if (CallerAttrs.requiresLazySave(SMEAttrs(*Call)))
-          return true;
-  return false;
-}
-
 int
 AArch64TargetLowering::allocateLazySaveBuffer(SDValue &Chain, const SDLoc &DL,
                                               SelectionDAG &DAG) const {
@@ -6427,8 +6412,8 @@
   if (Subtarget->hasCustomCallingConv())
     Subtarget->getRegisterInfo()->UpdateCustomCalleeSavedRegs(MF);
 
-  if (requiresBufferForLazySave(MF.getFunction())) {
-    // Set up a buffer once and store the buffer in the MachineFunctionInfo.
+  // Conservatively assume that the function may require the lazy-save mechanism.
+  if (SMEAttrs(MF.getFunction()).hasZAState()) {
     int TPIDR2Obj = allocateLazySaveBuffer(Chain, DL, DAG);
     FuncInfo->setLazySaveTPIDR2Obj(TPIDR2Obj);
   }
@@ -7025,9 +7010,6 @@
     SDValue NN = DAG.getNode(ISD::MUL, DL, MVT::i64, N, N);
     int TPIDR2Obj = *FuncInfo->getLazySaveTPIDR2Obj();
 
-    if (!TPIDR2Obj)
-      TPIDR2Obj = allocateLazySaveBuffer(Chain, DL, DAG);
-
     MachinePointerInfo MPI = MachinePointerInfo::getStack(MF, TPIDR2Obj);
     SDValue TPIDR2ObjAddr = DAG.getFrameIndex(TPIDR2Obj,
         DAG.getTargetLoweringInfo().getFrameIndexTy(DAG.getDataLayout()));


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D138208.476116.patch
Type: text/x-patch
Size: 3463 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20221117/13528b96/attachment.bin>