[ARM] Change VPT state assertion

[samparker]

Summary

Just because we haven't encountered an instruction setting the VPR, it doesn't mean we can't create a VPT block - the VPR maybe a live-in.

[dmgreen]

Sounds OK I think, if the rest of this will handle live-in VPR registers correctly.

llvm/lib/Target/ARM/ARMLowOverheadLoops.cpp

class VPTState {
     SmallVector<MachineInstr *, 4> Insts;
 
     static SmallVector<VPTState, 4> Blocks;
     static SetVector<MachineInstr *> CurrentPredicates;
     static std::map<MachineInstr *,
       std::unique_ptr<PredicatedMI>> PredicatedInsts;
 
     static void CreateVPTBlock(MachineInstr *MI) {
-      assert(CurrentPredicates.size() && "Can't begin VPT without predicate");
+      assert((CurrentPredicates.size() || MI->getParent()->isLiveIn(ARM::VPR))
+             && "Can't begin VPT without predicate");
       Blocks.emplace_back(MI);
       // The execution of MI is predicated upon the current set of instructions
       // that are AND'ed together to form the VPR predicate value. In the case
       // that MI is a VPT, CurrentPredicates will also just be MI.
       PredicatedInsts.emplace(
         MI, std::make_unique<PredicatedMI>(MI, CurrentPredicates));
     }
 
     static void reset() {
       Blocks.clear();
       PredicatedInsts.clear();
       CurrentPredicates.clear();

llvm/test/CodeGen/Thumb2/LowOverheadLoops/begin-vpt-without-inst.mir

+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=thumbv8.1m.main -mattr=+mve -run-pass=arm-low-overhead-loops %s -o - | FileCheck %s
+
+--- |
+  @arr = external dso_local local_unnamed_addr global [0 x i32], align 4
+
+  define dso_local arm_aapcs_vfpcc void @foo(i32 %i) {
+  entry:
+    %tobool.not11 = icmp eq i32 %i, 0
+    br i1 %tobool.not11, label %for.end5, label %vector.ph.preheader
+
+  vector.ph.preheader:                              ; preds = %entry
+    %active.lane.mask = call <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32 0, i32 3)
+    br label %vector.ph
+
+  vector.ph:                                        ; preds = %vector.ph.preheader, %vector.ph
+    %i.addr.012 = phi i32 [ %math, %vector.ph ], [ %i, %vector.ph.preheader ]
+    call void @llvm.masked.store.v4i32.p0v4i32(<4 x i32> <i32 2, i32 2, i32 2, i32 2>, <4 x i32>* bitcast ([0 x i32]* @arr to <4 x i32>*), i32 4, <4 x i1> %active.lane.mask)
+    %0 = call { i32, i1 } @llvm.uadd.with.overflow.i32(i32 %i.addr.012, i32 1)
+    %math = extractvalue { i32, i1 } %0, 0
+    %ov = extractvalue { i32, i1 } %0, 1
+    br i1 %ov, label %for.end5, label %vector.ph
+
+  for.end5:                                         ; preds = %vector.ph, %entry
+    ret void
+  }
+
+  declare <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32, i32)
+  declare void @llvm.masked.store.v4i32.p0v4i32(<4 x i32>, <4 x i32>*, i32 immarg, <4 x i1>)
+  declare { i32, i1 } @llvm.uadd.with.overflow.i32(i32, i32)
+
+...
+---
+name:            foo
+alignment:       8
+tracksRegLiveness: true
+registers:       []
+liveins:
+  - { reg: '$r0', virtual-reg: '' }
+frameInfo:
+  maxAlignment:    1
+  stackProtector:  ''
+  maxCallFrameSize: 0
+  cvBytesOfCalleeSavedRegisters: 0
+  localFrameSize:  0
+  savePoint:       ''
+  restorePoint:    ''
+fixedStack:      []
+stack:           []
+callSites:       []
+constants:
+  - id:              0
+    value:           '<4 x i32> <i32 0, i32 1, i32 2, i32 3>'
+    alignment:       8
+    isTargetSpecific: false
+machineFunctionInfo: {}
+body:             |
+  ; CHECK-LABEL: name: foo
+  ; CHECK: bb.0.entry:
+  ; CHECK:   successors: %bb.3(0x30000000), %bb.1(0x50000000)
+  ; CHECK:   liveins: $r0
+  ; CHECK:   tCBZ $r0, %bb.3
+  ; CHECK: bb.1.vector.ph.preheader:
+  ; CHECK:   successors: %bb.2(0x80000000)
+  ; CHECK:   liveins: $r0
+  ; CHECK:   renamable $r1 = tLEApcrel %const.0, 14 /* CC::al */, $noreg
+  ; CHECK:   renamable $q0 = MVE_VMOVimmi32 3, 0, $noreg, undef renamable $q0
+  ; CHECK:   renamable $q1 = MVE_VLDRWU32 killed renamable $r1, 0, 0, $noreg :: (load 16 from constant-pool, align 8)
+  ; CHECK:   $r1 = t2MOVi16 target-flags(arm-lo16) @arr, 14 /* CC::al */, $noreg
+  ; CHECK:   $r1 = t2MOVTi16 killed $r1, target-flags(arm-hi16) @arr, 14 /* CC::al */, $noreg
+  ; CHECK:   renamable $vpr = MVE_VCMPu32 killed renamable $q0, killed renamable $q1, 8, 0, $noreg
+  ; CHECK:   renamable $q0 = MVE_VMOVimmi32 2, 0, $noreg, undef renamable $q0
+  ; CHECK: bb.2.vector.ph:
+  ; CHECK:   successors: %bb.3(0x04000000), %bb.2(0x7c000000)
+  ; CHECK:   liveins: $vpr, $q0, $r0, $r1
+  ; CHECK:   renamable $r0, $cpsr = tADDi8 killed renamable $r0, 1, 14 /* CC::al */, $noreg
+  ; CHECK:   MVE_VPST 8, implicit $vpr
+  ; CHECK:   MVE_VSTRWU32 renamable $q0, renamable $r1, 0, 1, renamable $vpr :: (store 16 into `<4 x i32>* bitcast ([0 x i32]* @arr to <4 x i32>*)`, align 4)
+  ; CHECK:   tBcc %bb.2, 3 /* CC::lo */, killed $cpsr
+  ; CHECK: bb.3.for.end5:
+  ; CHECK:   tBX_RET 14 /* CC::al */, $noreg
+  ; CHECK: bb.4 (align 8):
+  ; CHECK:   CONSTPOOL_ENTRY 0, %const.0, 16
+  bb.0.entry:
+    successors: %bb.3(0x30000000), %bb.1(0x50000000)
+    liveins: $r0
+
+    tCBZ $r0, %bb.3
+
+  bb.1.vector.ph.preheader:
+    successors: %bb.2(0x80000000)
+    liveins: $r0
+
+    renamable $r1 = tLEApcrel %const.0, 14 /* CC::al */, $noreg
+    renamable $q0 = MVE_VMOVimmi32 3, 0, $noreg, undef renamable $q0
+    renamable $q1 = MVE_VLDRWU32 killed renamable $r1, 0, 0, $noreg :: (load 16 from constant-pool, align 8)
+    $r1 = t2MOVi16 target-flags(arm-lo16) @arr, 14 /* CC::al */, $noreg
+    $r1 = t2MOVTi16 killed $r1, target-flags(arm-hi16) @arr, 14 /* CC::al */, $noreg
+    renamable $vpr = MVE_VCMPu32 killed renamable $q0, killed renamable $q1, 8, 0, $noreg
+    renamable $q0 = MVE_VMOVimmi32 2, 0, $noreg, undef renamable $q0
+
+  bb.2.vector.ph:
+    successors: %bb.3(0x04000000), %bb.2(0x7c000000)
+    liveins: $vpr, $q0, $r0, $r1
+
+    renamable $r0, $cpsr = tADDi8 killed renamable $r0, 1, 14 /* CC::al */, $noreg
+    MVE_VPST 8, implicit $vpr
+    MVE_VSTRWU32 renamable $q0, renamable $r1, 0, 1, renamable $vpr :: (store 16 into `<4 x i32>* bitcast ([0 x i32]* @arr to <4 x i32>*)`, align 4)
+    tBcc %bb.2, 3 /* CC::lo */, killed $cpsr
+
+  bb.3.for.end5:
+    tBX_RET 14 /* CC::al */, $noreg
+
+  bb.4 (align 8):
+    CONSTPOOL_ENTRY 0, %const.0, 16
+
+...