[AArch64][SME] Don't mark 'smstart za' as using/defining VG.

[sdesmalen-arm]

VG is only used/defined when changing the streaming mode, using 'smstart sm' or plainly 'smstart' (same for smstop).

[llvmbot]

@llvm/pr-subscribers-backend-aarch64

Author: Sander de Smalen (sdesmalen-arm)

Changes

VG is only used/defined when changing the streaming mode, using 'smstart sm' or plainly 'smstart' (same for smstop).

---

Full diff: https://github.com/llvm/llvm-project/pull/84775.diff

3 Files Affected:

• (modified) llvm/lib/Target/AArch64/AArch64ISelLowering.cpp (+11-1)
• (modified) llvm/lib/Target/AArch64/SMEInstrFormats.td (-2)
• (added) llvm/test/CodeGen/AArch64/sme-write-vg.ll (+24)

diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index 054311d39e7b83..ddbda4d515fcba 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -7751,13 +7751,23 @@ void AArch64TargetLowering::AdjustInstrPostInstrSelection(MachineInstr &MI,
   // register allocator to pass call args in callee saved regs, without extra
   // copies to avoid these fake clobbers of actually-preserved GPRs.
   if (MI.getOpcode() == AArch64::MSRpstatesvcrImm1 ||
-      MI.getOpcode() == AArch64::MSRpstatePseudo)
+      MI.getOpcode() == AArch64::MSRpstatePseudo) {
     for (unsigned I = MI.getNumOperands() - 1; I > 0; --I)
       if (MachineOperand &MO = MI.getOperand(I);
           MO.isReg() && MO.isImplicit() && MO.isDef() &&
           (AArch64::GPR32RegClass.contains(MO.getReg()) ||
            AArch64::GPR64RegClass.contains(MO.getReg())))
         MI.removeOperand(I);
+    // If the instruction or pseudo changes PSTATE.SM, then add VG as a Use and
+    // Def.
+    if (MI.getOperand(0).getImm() == AArch64SVCR::SVCRSM ||
+        MI.getOperand(0).getImm() == AArch64SVCR::SVCRSMZA) {
+      MI.addOperand(MachineOperand::CreateReg(AArch64::VG, /*IsDef=*/false,
+                                              /*IsImplicit=*/true));
+      MI.addOperand(MachineOperand::CreateReg(AArch64::VG, /*IsDef=*/true,
+                                              /*IsImplicit=*/true));
+    }
+  }
 
   // Add an implicit use of 'VG' for ADDXri/SUBXri, which are instructions that
   // have nothing to do with VG, were it not that they are used to materialise a
diff --git a/llvm/lib/Target/AArch64/SMEInstrFormats.td b/llvm/lib/Target/AArch64/SMEInstrFormats.td
index 33cb5f9734b819..44d9a8ac7cb677 100644
--- a/llvm/lib/Target/AArch64/SMEInstrFormats.td
+++ b/llvm/lib/Target/AArch64/SMEInstrFormats.td
@@ -223,8 +223,6 @@ def MSRpstatesvcrImm1
   let Inst{8} = imm;
   let Inst{7-5} = 0b011; // op2
   let hasPostISelHook = 1;
-  let Uses = [VG];
-  let Defs = [VG];
 }
 
 def : InstAlias<"smstart",    (MSRpstatesvcrImm1 0b011, 0b1)>;
diff --git a/llvm/test/CodeGen/AArch64/sme-write-vg.ll b/llvm/test/CodeGen/AArch64/sme-write-vg.ll
new file mode 100644
index 00000000000000..577606d45484f1
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/sme-write-vg.ll
@@ -0,0 +1,24 @@
+; RUN: llc -mattr=+sme -stop-after=finalize-isel < %s | FileCheck %s
+
+target triple = "aarch64"
+
+; Check that we don't define VG for 'smstart za' and 'smstop za'
+define void @smstart_za() "aarch64_new_za" nounwind {
+  ; CHECK-LABEL:    name: smstart_za
+  ; CHECK-NOT:        implicit-def {{[^,]*}}$vg
+  ret void
+}
+
+; Check that we do define VG for 'smstart sm' and 'smstop sm'
+define void @smstart_sm() nounwind {
+  ; CHECK-LABEL: name: smstart_sm
+  ; CHECK:          MSRpstatesvcrImm1 1, 1,
+  ; CHECK-SAME:       implicit-def {{[^,]*}}$vg
+  ; CHECK:          MSRpstatesvcrImm1 1, 0,
+  ; CHECK-SAME:       implicit-def {{[^,]*}}$vg
+  call void @require_sm()
+  ret void
+}
+
+declare void @require_sm() "aarch64_pstate_sm_enabled"
+declare void @require_za() "aarch64_inout_za"

[paulwalker-arm]

Whilst true, it's effectively what the code says. Perhaps "The SVE vector length can change when entering/leaving streaming mode."?