[AArch64] STRXui: Illegal virtual register for instruction

[sjoerdmeijer]

We are running into this verifier error with llc -verify-machineinstrs:

*** Bad machine code: Illegal virtual register for instruction ***
- function:    _Z4testbtxasbiisjaytyajxiPA23_iPA24_aPtPA24_S1_PA24_A24_sPA24_A24_bS5_PA24_A24_xPjPsS8_PA24_A24_hS8_S8_PA24_A24_tS8_SM_SE_SJ_SE_
- basic block: %bb.2 for.body41.us.preheader (0x411fd580) [272B;1328B)
- instruction: 1288B	STRXui %387:gpr64sp, %stack.12, 0 :: (store (s64) into %stack.12)
- operand 0:   %387:gpr64sp
Expected a GPR64 register, but got a GPR64sp register
LLVM ERROR: Found 1 machine code errors.

See https://godbolt.org/z/Pq9G41Eqq

I've tried reducing the input, but it's still quite big. Copying it here for completeness:

target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128-Fn32"
target triple = "aarch64-unknown-linux-gnu"

@var_32 = external global i16
@var_35 = external global i64
@var_39 = external global i64
@var_46 = external global i64
@var_50 = external global i32

define void @_Z4testbtxasbiisjaytyajxiPA23_iPA24_aPtPA24_S1_PA24_A24_sPA24_A24_bS5_PA24_A24_xPjPsS8_PA24_A24_hS8_S8_PA24_A24_tS8_SM_SE_SJ_SE_(i1 %var_0, i16 %var_1, i64 %var_2, i8 %var_3, i16 %var_4, i1 %var_5, i32 %var_6, i32 %var_7, i8 %var_10, i64 %var_11, i8 %var_14, i32 %var_15, i64 %var_16, ptr %arr_3, ptr %arr_4, ptr %arr_6, ptr %arr_7, ptr %arr_12, ptr %arr_13, ptr %arr_19, i64 %mul, i64 %conv35, i64 %idxprom138.us16, i8 %0, i8 %1, ptr %invariant.gep875.us) #0 {
entry:
  br i1 %var_5, label %for.body41.lr.ph, label %for.cond563.preheader

for.body41.lr.ph:                                 ; preds = %entry
  %arrayidx147 = getelementptr i8, ptr %arr_3, i64 120
  %tobool326.not = icmp eq i64 %var_2, 0
  %not353 = xor i64 0, -1
  %add538 = select i1 %var_0, i16 0, i16 1
  br i1 %var_0, label %for.body41.us, label %for.body41

for.body41.us:                                    ; preds = %for.cond.cleanup93.us, %for.body41.lr.ph
  %var_24.promoted9271009.us = phi i64 [ 0, %for.body41.lr.ph ], [ %6, %for.cond.cleanup93.us ]
  %var_37.promoted9301008.us = phi i64 [ 1, %for.body41.lr.ph ], [ 0, %for.cond.cleanup93.us ]
  %2 = phi i8 [ 0, %for.body41.lr.ph ], [ 1, %for.cond.cleanup93.us ]
  %add4139751001.us = phi i16 [ 0, %for.body41.lr.ph ], [ 1, %for.cond.cleanup93.us ]
  %3 = phi i8 [ 0, %for.body41.lr.ph ], [ %var_10, %for.cond.cleanup93.us ]
  store i32 %var_6, ptr %arr_3, align 4
  store i8 %var_10, ptr %arr_3, align 1
  br label %for.body67.us

for.body67.us:                                    ; preds = %for.cond.cleanup93.us, %for.body41.us
  %4 = phi i8 [ %3, %for.body41.us ], [ 0, %for.cond.cleanup93.us ]
  %add413977.us = phi i16 [ %add4139751001.us, %for.body41.us ], [ %add413.us17, %for.cond.cleanup93.us ]
  %5 = phi i8 [ %2, %for.body41.us ], [ %.sroa.speculated829.us, %for.cond.cleanup93.us ]
  %conv64922.us = phi i32 [ 1, %for.body41.us ], [ 0, %for.cond.cleanup93.us ]
  %6 = phi i64 [ %var_24.promoted9271009.us, %for.body41.us ], [ %.sroa.speculated832.us, %for.cond.cleanup93.us ]
  %mul354903918.us = phi i64 [ %var_37.promoted9301008.us, %for.body41.us ], [ 0, %for.cond.cleanup93.us ]
  %i_2.0921.us = zext i32 %var_15 to i64
  %.sroa.speculated832.us = tail call i64 @llvm.umin.i64(i64 %var_24.promoted9271009.us, i64 -30)
  %sext1023 = shl i64 %i_2.0921.us, 1
  %idxprom138.us162 = ashr i64 %sext1023, 1
  %gep889.us = getelementptr [24 x i16], ptr %arr_19, i64 %idxprom138.us16
  %arrayidx149.us = getelementptr i8, ptr %arrayidx147, i64 %idxprom138.us162
  %arrayidx319.us = getelementptr [24 x i8], ptr null, i64 %idxprom138.us162
  %7 = sext i32 %conv64922.us to i64
  %8 = getelementptr i32, ptr %arr_12, i64 %7
  %arrayidx226.us = getelementptr i8, ptr %8, i64 -12
  br label %for.cond95.preheader.us

for.cond.cleanup93.us:                            ; preds = %for.cond.cleanup98.us
  br i1 %var_5, label %for.body67.us, label %for.body41.us

for.cond.cleanup98.us:                            ; preds = %for.cond510.preheader.us
  br i1 %var_0, label %for.cond.cleanup93.us, label %for.cond95.preheader.us

for.body99.us:                                    ; preds = %for.cond95.preheader.us, %for.cond510.preheader.us
  %mul287985.us = phi i16 [ 0, %for.cond95.preheader.us ], [ %mul287.us, %for.cond510.preheader.us ]
  %9 = phi i8 [ %29, %for.cond95.preheader.us ], [ %var_14, %for.cond510.preheader.us ]
  %add413979.us = phi i16 [ %add413978.us, %for.cond95.preheader.us ], [ %add413.us17, %for.cond510.preheader.us ]
  %10 = phi i32 [ 0, %for.cond95.preheader.us ], [ %26, %for.cond510.preheader.us ]
  %mul354905.us = phi i64 [ %mul354904.us, %for.cond95.preheader.us ], [ %mul354907.us, %for.cond510.preheader.us ]
  %sub283896.us = phi i64 [ 1, %for.cond95.preheader.us ], [ %sub283.us, %for.cond510.preheader.us ]
  %conv96880.us = phi i64 [ 1, %for.cond95.preheader.us ], [ 0, %for.cond510.preheader.us ]
  %.sroa.speculated829.us = tail call i8 @llvm.smin.i8(i8 %30, i8 0)
  br label %for.body113.us

for.body380.us:                                   ; preds = %for.cond376.preheader.us, %for.inc505.us
  %indvars.iv1018 = phi i64 [ 0, %for.cond376.preheader.us ], [ %indvars.iv.next1019, %for.inc505.us ]
  %11 = phi i8 [ 0, %for.cond376.preheader.us ], [ %13, %for.inc505.us ]
  %add413980.us = phi i16 [ %add413979.us, %for.cond376.preheader.us ], [ 0, %for.inc505.us ]
  %12 = load i16, ptr %arrayidx384.us, align 2
  store i16 %12, ptr %invariant.gep875.us, align 2
  %add413.us17 = or i16 %add413980.us, 0
  %arrayidx416.us = getelementptr i16, ptr %arr_13, i64 %indvars.iv1018
  %conv419.us = select i1 %var_0, i32 36006, i32 -7680
  store i32 %conv419.us, ptr @var_50, align 4
  %tobool435.not.us = icmp eq i64 %mul, 0
  br i1 %tobool435.not.us, label %for.inc505.us, label %if.then436.us

if.then436.us:                                    ; preds = %for.body380.us
  %.sroa.speculated817.us = tail call i8 @llvm.smax.i8(i8 %11, i8 0)
  %cond464.in.us = load i16, ptr %gep876.us, align 2
  %tobool465.not.us = icmp eq i16 %cond464.in.us, 0
  br i1 %tobool465.not.us, label %for.inc505.us, label %if.then466.us

if.then466.us:                                    ; preds = %if.then436.us
  store i64 %conv35, ptr %arr_3, align 8
  br label %for.inc505.us

for.inc505.us:                                    ; preds = %if.then466.us, %if.then436.us, %for.body380.us
  %13 = phi i8 [ %11, %for.body380.us ], [ %.sroa.speculated817.us, %if.then466.us ], [ 0, %if.then436.us ]
  %indvars.iv.next1019 = add i64 %indvars.iv1018, 1
  %cmp378.us = icmp ult i64 %indvars.iv1018, 0
  br i1 %cmp378.us, label %for.body380.us, label %for.cond510.preheader.us

for.body194.us:                                   ; preds = %if.end.us.7, %for.inc371.us
  %indvars.iv = phi i64 [ 0, %if.end.us.7 ], [ %indvars.iv.next, %for.inc371.us ]
  %mul287986.us = phi i16 [ %mul287985.us, %if.end.us.7 ], [ %mul287.us, %for.inc371.us ]
  %14 = phi i8 [ %9, %if.end.us.7 ], [ %16, %for.inc371.us ]
  %mul354906.us = phi i64 [ %mul354905.us, %if.end.us.7 ], [ %var_11, %for.inc371.us ]
  %sub283897.us = phi i64 [ %sub283896.us, %if.end.us.7 ], [ 0, %for.inc371.us ]
  %tobool221.not.us = icmp eq i32 1, 0
  br i1 %tobool221.not.us, label %if.end239.us, label %if.then222.us

if.then222.us:                                    ; preds = %for.body194.us
  %15 = load i32, ptr %arrayidx226.us, align 4
  %conv227.us = trunc i32 %15 to i16
  store i16 %conv227.us, ptr @var_32, align 2
  %.sroa.speculated820.us = tail call i8 @llvm.smax.i8(i8 %14, i8 0)
  br label %if.end239.us

if.end239.us:                                     ; preds = %if.then222.us, %for.body194.us
  %16 = phi i8 [ %.sroa.speculated820.us, %if.then222.us ], [ 0, %for.body194.us ]
  store i8 -107, ptr %arr_7, align 1
  %tobool253.not.us = icmp eq i8 %0, 0
  br i1 %tobool253.not.us, label %if.end282.us, label %if.then254.us

if.then254.us:                                    ; preds = %if.end239.us
  %17 = load i16, ptr %arrayidx259.us, align 2
  %tobool261.not.us = icmp eq i16 %17, 0
  %conv268.us = select i1 %tobool261.not.us, i64 0, i64 %var_16
  store i64 %conv268.us, ptr @var_35, align 8
  %gep867.us = getelementptr [24 x [24 x i64]], ptr null, i64 %indvars.iv
  store i16 %var_1, ptr %arr_6, align 2
  br label %if.end282.us

if.end282.us:                                     ; preds = %if.then254.us, %if.end239.us
  %sub283.us = or i64 %sub283897.us, -434259939
  store i64 %sub283.us, ptr %arr_4, align 8
  %mul287.us = mul i16 %mul287986.us, -18978
  store i64 0, ptr @var_39, align 8
  %18 = load i8, ptr %arrayidx321.us, align 1
  %conv322.us = zext i8 %18 to i64
  store i64 %conv322.us, ptr %arr_4, align 8
  br i1 %tobool326.not, label %if.then327.us, label %for.inc371.us

if.then327.us:                                    ; preds = %if.end282.us
  %tobool330.not.us = icmp eq i32 0, 0
  br i1 %tobool330.not.us, label %cond.end345.us, label %cond.true331.us

cond.true331.us:                                  ; preds = %if.then327.us
  %19 = load i8, ptr null, align 1
  %20 = sext i8 %19 to i16
  br label %cond.end345.us

cond.end345.us:                                   ; preds = %cond.true331.us, %if.then327.us
  %cond346.us = phi i16 [ %20, %cond.true331.us ], [ 0, %if.then327.us ]
  store i16 %cond346.us, ptr %arr_4, align 2
  %mul354.us = mul i64 %mul354906.us, %not353
  store i64 %mul354.us, ptr @var_46, align 8
  br label %for.inc371.us

for.inc371.us:                                    ; preds = %cond.end345.us, %if.end282.us
  %mul354907.us = phi i64 [ 1, %if.end282.us ], [ 0, %cond.end345.us ]
  %indvars.iv.next = or i64 %indvars.iv, 1
  br i1 %var_0, label %for.body194.us, label %for.cond376.preheader.us

cond.true146.us:                                  ; preds = %for.cond131.preheader.us
  %21 = load i8, ptr %arrayidx149.us, align 1
  %conv150.us = sext i8 %21 to i32
  br label %cond.end154.us

cond.end154.us:                                   ; preds = %for.cond131.preheader.us, %cond.true146.us
  %cond155.us = phi i32 [ %conv150.us, %cond.true146.us ], [ 0, %for.cond131.preheader.us ]
  %shl.us = shl i32 %div.us, %cond155.us
  %tobool157.not.us = icmp eq i32 %shl.us, 0
  br i1 %tobool157.not.us, label %if.end.us, label %if.then.us

if.then.us:                                       ; preds = %cond.end154.us
  store i32 0, ptr %arr_4, align 4
  br label %if.end.us

if.end.us:                                        ; preds = %if.then.us, %cond.end154.us
  %22 = phi i32 [ 0, %if.then.us ], [ %10, %cond.end154.us ]
  store i8 %1, ptr %arr_4, align 1
  call void @llvm.assume(i1 true)
  %23 = load i8, ptr %arrayidx149.us, align 1
  %conv150.us.2 = sext i8 %23 to i32
  %shl.us.2 = shl i32 18984, %conv150.us.2
  %tobool157.not.us.2 = icmp eq i32 %shl.us.2, 0
  br i1 %tobool157.not.us.2, label %if.then.us.5, label %if.then.us.2

if.then.us.2:                                     ; preds = %if.end.us
  %.sroa.speculated826.us.2 = tail call i32 @llvm.smin.i32(i32 %10, i32 0)
  store i8 0, ptr %arr_4, align 1
  br label %if.then.us.5

if.then.us.5:                                     ; preds = %if.then.us.2, %if.end.us
  %24 = phi i32 [ 0, %if.then.us.2 ], [ %22, %if.end.us ]
  %.sroa.speculated826.us.5 = tail call i32 @llvm.smin.i32(i32 %24, i32 1410036665)
  br i1 %var_0, label %cond.end154.us.7, label %cond.true146.us.7

cond.true146.us.7:                                ; preds = %if.then.us.5
  %25 = load i8, ptr %arrayidx149.us, align 1
  %conv150.us.7 = sext i8 %25 to i32
  br label %cond.end154.us.7

cond.end154.us.7:                                 ; preds = %cond.true146.us.7, %if.then.us.5
  %cond155.us.7 = phi i32 [ %conv150.us.7, %cond.true146.us.7 ], [ 0, %if.then.us.5 ]
  %shl.us.7 = shl i32 18984, %cond155.us.7
  %tobool157.not.us.7 = icmp eq i32 %shl.us.7, 0
  br i1 %tobool157.not.us.7, label %if.end.us.7, label %if.then.us.7

if.then.us.7:                                     ; preds = %cond.end154.us.7
  store i32 0, ptr %arr_3, align 4
  br label %if.end.us.7

if.end.us.7:                                      ; preds = %if.then.us.7, %cond.end154.us.7
  %26 = phi i32 [ 0, %if.then.us.7 ], [ %.sroa.speculated826.us.5, %cond.end154.us.7 ]
  %arrayidx259.us = getelementptr i16, ptr %arrayidx257.us, i64 %conv96880.us
  br label %for.body194.us

for.body113.us:                                   ; preds = %for.body113.us, %for.body99.us
  br i1 %var_0, label %for.body113.us, label %for.cond131.preheader.us

for.cond510.preheader.us:                         ; preds = %for.inc505.us
  %cmp97.us = icmp slt i16 %add538, 0
  br i1 %cmp97.us, label %for.body99.us, label %for.cond.cleanup98.us

for.cond376.preheader.us:                         ; preds = %for.inc371.us
  %arrayidx384.us = getelementptr i16, ptr null, i64 %conv96880.us
  %gep876.us = getelementptr [24 x [24 x i16]], ptr %invariant.gep875.us, i64 %conv96880.us
  br label %for.body380.us

for.cond131.preheader.us:                         ; preds = %for.body113.us
  store i8 %var_3, ptr %arr_4, align 1
  %27 = load i16, ptr %gep884.us, align 2
  %28 = mul i16 18984, %27
  %div.us = zext i16 %28 to i32
  %tobool145.not.us = icmp eq i8 0, 0
  br i1 %tobool145.not.us, label %cond.end154.us, label %cond.true146.us

for.cond95.preheader.us:                          ; preds = %for.cond.cleanup98.us, %for.body67.us
  %indvars.iv1021 = phi i64 [ 1, %for.cond.cleanup98.us ], [ 0, %for.body67.us ]
  %29 = phi i8 [ %16, %for.cond.cleanup98.us ], [ %4, %for.body67.us ]
  %add413978.us = phi i16 [ %var_4, %for.cond.cleanup98.us ], [ %add413977.us, %for.body67.us ]
  %30 = phi i8 [ %.sroa.speculated829.us, %for.cond.cleanup98.us ], [ %5, %for.body67.us ]
  %mul354904.us = phi i64 [ 0, %for.cond.cleanup98.us ], [ %mul354903918.us, %for.body67.us ]
  %gep884.us = getelementptr [24 x [24 x i16]], ptr %gep889.us, i64 %indvars.iv1021
  %arrayidx321.us = getelementptr i8, ptr %arrayidx319.us, i64 %indvars.iv1021
  %arrayidx257.us = getelementptr [24 x i16], ptr null, i64 %indvars.iv1021
  br label %for.body99.us

for.cond563.preheader:                            ; preds = %for.body41, %entry
  ret void

for.body41:                                       ; preds = %for.body41.lr.ph
  store i8 0, ptr %arr_12, align 1
  store i8 0, ptr %arr_3, align 1
  br label %for.cond563.preheader

; uselistorder directives
  uselistorder i16 %add413.us17, { 1, 0 }
  uselistorder i16 %mul287.us, { 1, 0 }
}

; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
declare i8 @llvm.smin.i8(i8, i8) #1

; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
declare i64 @llvm.umin.i64(i64, i64) #1

; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
declare i8 @llvm.smax.i8(i8, i8) #1

; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
declare i32 @llvm.smin.i32(i32, i32) #1

; Function Attrs: nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: write)
declare void @llvm.assume(i1 noundef) #2

; uselistorder directives
uselistorder ptr @llvm.smax.i8, { 1, 0 }
uselistorder ptr @llvm.smin.i32, { 1, 0 }

attributes #0 = { "frame-pointer"="non-leaf" "target-cpu"="grace" }
attributes #1 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
attributes #2 = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: write) }

[llvmbot]

@llvm/issue-subscribers-backend-aarch64

Author: Sjoerd Meijer (sjoerdmeijer)

We are running into this verifier error:

*** Bad machine code: Illegal virtual register for instruction ***
- function:    _Z4testbtxasbiisjaytyajxiPA23_iPA24_aPtPA24_S1_PA24_A24_sPA24_A24_bS5_PA24_A24_xPjPsS8_PA24_A24_hS8_S8_PA24_A24_tS8_SM_SE_SJ_SE_
- basic block: %bb.2 for.body41.us.preheader (0x411fd580) [272B;1328B)
- instruction: 1288B	STRXui %387:gpr64sp, %stack.12, 0 :: (store (s64) into %stack.12)
- operand 0:   %387:gpr64sp
Expected a GPR64 register, but got a GPR64sp register
LLVM ERROR: Found 1 machine code errors.

See https://godbolt.org/z/Pq9G41Eqq

I've tried reducing the input, but it's still quite big. Copying it here for completeness:

target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128-Fn32"
target triple = "aarch64-unknown-linux-gnu"

@<!-- -->var_32 = external global i16
@<!-- -->var_35 = external global i64
@<!-- -->var_39 = external global i64
@<!-- -->var_46 = external global i64
@<!-- -->var_50 = external global i32

define void @<!-- -->_Z4testbtxasbiisjaytyajxiPA23_iPA24_aPtPA24_S1_PA24_A24_sPA24_A24_bS5_PA24_A24_xPjPsS8_PA24_A24_hS8_S8_PA24_A24_tS8_SM_SE_SJ_SE_(i1 %var_0, i16 %var_1, i64 %var_2, i8 %var_3, i16 %var_4, i1 %var_5, i32 %var_6, i32 %var_7, i8 %var_10, i64 %var_11, i8 %var_14, i32 %var_15, i64 %var_16, ptr %arr_3, ptr %arr_4, ptr %arr_6, ptr %arr_7, ptr %arr_12, ptr %arr_13, ptr %arr_19, i64 %mul, i64 %conv35, i64 %idxprom138.us16, i8 %0, i8 %1, ptr %invariant.gep875.us) #<!-- -->0 {
entry:
  br i1 %var_5, label %for.body41.lr.ph, label %for.cond563.preheader

for.body41.lr.ph:                                 ; preds = %entry
  %arrayidx147 = getelementptr i8, ptr %arr_3, i64 120
  %tobool326.not = icmp eq i64 %var_2, 0
  %not353 = xor i64 0, -1
  %add538 = select i1 %var_0, i16 0, i16 1
  br i1 %var_0, label %for.body41.us, label %for.body41

for.body41.us:                                    ; preds = %for.cond.cleanup93.us, %for.body41.lr.ph
  %var_24.promoted9271009.us = phi i64 [ 0, %for.body41.lr.ph ], [ %6, %for.cond.cleanup93.us ]
  %var_37.promoted9301008.us = phi i64 [ 1, %for.body41.lr.ph ], [ 0, %for.cond.cleanup93.us ]
  %2 = phi i8 [ 0, %for.body41.lr.ph ], [ 1, %for.cond.cleanup93.us ]
  %add4139751001.us = phi i16 [ 0, %for.body41.lr.ph ], [ 1, %for.cond.cleanup93.us ]
  %3 = phi i8 [ 0, %for.body41.lr.ph ], [ %var_10, %for.cond.cleanup93.us ]
  store i32 %var_6, ptr %arr_3, align 4
  store i8 %var_10, ptr %arr_3, align 1
  br label %for.body67.us

for.body67.us:                                    ; preds = %for.cond.cleanup93.us, %for.body41.us
  %4 = phi i8 [ %3, %for.body41.us ], [ 0, %for.cond.cleanup93.us ]
  %add413977.us = phi i16 [ %add4139751001.us, %for.body41.us ], [ %add413.us17, %for.cond.cleanup93.us ]
  %5 = phi i8 [ %2, %for.body41.us ], [ %.sroa.speculated829.us, %for.cond.cleanup93.us ]
  %conv64922.us = phi i32 [ 1, %for.body41.us ], [ 0, %for.cond.cleanup93.us ]
  %6 = phi i64 [ %var_24.promoted9271009.us, %for.body41.us ], [ %.sroa.speculated832.us, %for.cond.cleanup93.us ]
  %mul354903918.us = phi i64 [ %var_37.promoted9301008.us, %for.body41.us ], [ 0, %for.cond.cleanup93.us ]
  %i_2.0921.us = zext i32 %var_15 to i64
  %.sroa.speculated832.us = tail call i64 @<!-- -->llvm.umin.i64(i64 %var_24.promoted9271009.us, i64 -30)
  %sext1023 = shl i64 %i_2.0921.us, 1
  %idxprom138.us162 = ashr i64 %sext1023, 1
  %gep889.us = getelementptr [24 x i16], ptr %arr_19, i64 %idxprom138.us16
  %arrayidx149.us = getelementptr i8, ptr %arrayidx147, i64 %idxprom138.us162
  %arrayidx319.us = getelementptr [24 x i8], ptr null, i64 %idxprom138.us162
  %7 = sext i32 %conv64922.us to i64
  %8 = getelementptr i32, ptr %arr_12, i64 %7
  %arrayidx226.us = getelementptr i8, ptr %8, i64 -12
  br label %for.cond95.preheader.us

for.cond.cleanup93.us:                            ; preds = %for.cond.cleanup98.us
  br i1 %var_5, label %for.body67.us, label %for.body41.us

for.cond.cleanup98.us:                            ; preds = %for.cond510.preheader.us
  br i1 %var_0, label %for.cond.cleanup93.us, label %for.cond95.preheader.us

for.body99.us:                                    ; preds = %for.cond95.preheader.us, %for.cond510.preheader.us
  %mul287985.us = phi i16 [ 0, %for.cond95.preheader.us ], [ %mul287.us, %for.cond510.preheader.us ]
  %9 = phi i8 [ %29, %for.cond95.preheader.us ], [ %var_14, %for.cond510.preheader.us ]
  %add413979.us = phi i16 [ %add413978.us, %for.cond95.preheader.us ], [ %add413.us17, %for.cond510.preheader.us ]
  %10 = phi i32 [ 0, %for.cond95.preheader.us ], [ %26, %for.cond510.preheader.us ]
  %mul354905.us = phi i64 [ %mul354904.us, %for.cond95.preheader.us ], [ %mul354907.us, %for.cond510.preheader.us ]
  %sub283896.us = phi i64 [ 1, %for.cond95.preheader.us ], [ %sub283.us, %for.cond510.preheader.us ]
  %conv96880.us = phi i64 [ 1, %for.cond95.preheader.us ], [ 0, %for.cond510.preheader.us ]
  %.sroa.speculated829.us = tail call i8 @<!-- -->llvm.smin.i8(i8 %30, i8 0)
  br label %for.body113.us

for.body380.us:                                   ; preds = %for.cond376.preheader.us, %for.inc505.us
  %indvars.iv1018 = phi i64 [ 0, %for.cond376.preheader.us ], [ %indvars.iv.next1019, %for.inc505.us ]
  %11 = phi i8 [ 0, %for.cond376.preheader.us ], [ %13, %for.inc505.us ]
  %add413980.us = phi i16 [ %add413979.us, %for.cond376.preheader.us ], [ 0, %for.inc505.us ]
  %12 = load i16, ptr %arrayidx384.us, align 2
  store i16 %12, ptr %invariant.gep875.us, align 2
  %add413.us17 = or i16 %add413980.us, 0
  %arrayidx416.us = getelementptr i16, ptr %arr_13, i64 %indvars.iv1018
  %conv419.us = select i1 %var_0, i32 36006, i32 -7680
  store i32 %conv419.us, ptr @<!-- -->var_50, align 4
  %tobool435.not.us = icmp eq i64 %mul, 0
  br i1 %tobool435.not.us, label %for.inc505.us, label %if.then436.us

if.then436.us:                                    ; preds = %for.body380.us
  %.sroa.speculated817.us = tail call i8 @<!-- -->llvm.smax.i8(i8 %11, i8 0)
  %cond464.in.us = load i16, ptr %gep876.us, align 2
  %tobool465.not.us = icmp eq i16 %cond464.in.us, 0
  br i1 %tobool465.not.us, label %for.inc505.us, label %if.then466.us

if.then466.us:                                    ; preds = %if.then436.us
  store i64 %conv35, ptr %arr_3, align 8
  br label %for.inc505.us

for.inc505.us:                                    ; preds = %if.then466.us, %if.then436.us, %for.body380.us
  %13 = phi i8 [ %11, %for.body380.us ], [ %.sroa.speculated817.us, %if.then466.us ], [ 0, %if.then436.us ]
  %indvars.iv.next1019 = add i64 %indvars.iv1018, 1
  %cmp378.us = icmp ult i64 %indvars.iv1018, 0
  br i1 %cmp378.us, label %for.body380.us, label %for.cond510.preheader.us

for.body194.us:                                   ; preds = %if.end.us.7, %for.inc371.us
  %indvars.iv = phi i64 [ 0, %if.end.us.7 ], [ %indvars.iv.next, %for.inc371.us ]
  %mul287986.us = phi i16 [ %mul287985.us, %if.end.us.7 ], [ %mul287.us, %for.inc371.us ]
  %14 = phi i8 [ %9, %if.end.us.7 ], [ %16, %for.inc371.us ]
  %mul354906.us = phi i64 [ %mul354905.us, %if.end.us.7 ], [ %var_11, %for.inc371.us ]
  %sub283897.us = phi i64 [ %sub283896.us, %if.end.us.7 ], [ 0, %for.inc371.us ]
  %tobool221.not.us = icmp eq i32 1, 0
  br i1 %tobool221.not.us, label %if.end239.us, label %if.then222.us

if.then222.us:                                    ; preds = %for.body194.us
  %15 = load i32, ptr %arrayidx226.us, align 4
  %conv227.us = trunc i32 %15 to i16
  store i16 %conv227.us, ptr @<!-- -->var_32, align 2
  %.sroa.speculated820.us = tail call i8 @<!-- -->llvm.smax.i8(i8 %14, i8 0)
  br label %if.end239.us

if.end239.us:                                     ; preds = %if.then222.us, %for.body194.us
  %16 = phi i8 [ %.sroa.speculated820.us, %if.then222.us ], [ 0, %for.body194.us ]
  store i8 -107, ptr %arr_7, align 1
  %tobool253.not.us = icmp eq i8 %0, 0
  br i1 %tobool253.not.us, label %if.end282.us, label %if.then254.us

if.then254.us:                                    ; preds = %if.end239.us
  %17 = load i16, ptr %arrayidx259.us, align 2
  %tobool261.not.us = icmp eq i16 %17, 0
  %conv268.us = select i1 %tobool261.not.us, i64 0, i64 %var_16
  store i64 %conv268.us, ptr @<!-- -->var_35, align 8
  %gep867.us = getelementptr [24 x [24 x i64]], ptr null, i64 %indvars.iv
  store i16 %var_1, ptr %arr_6, align 2
  br label %if.end282.us

if.end282.us:                                     ; preds = %if.then254.us, %if.end239.us
  %sub283.us = or i64 %sub283897.us, -434259939
  store i64 %sub283.us, ptr %arr_4, align 8
  %mul287.us = mul i16 %mul287986.us, -18978
  store i64 0, ptr @<!-- -->var_39, align 8
  %18 = load i8, ptr %arrayidx321.us, align 1
  %conv322.us = zext i8 %18 to i64
  store i64 %conv322.us, ptr %arr_4, align 8
  br i1 %tobool326.not, label %if.then327.us, label %for.inc371.us

if.then327.us:                                    ; preds = %if.end282.us
  %tobool330.not.us = icmp eq i32 0, 0
  br i1 %tobool330.not.us, label %cond.end345.us, label %cond.true331.us

cond.true331.us:                                  ; preds = %if.then327.us
  %19 = load i8, ptr null, align 1
  %20 = sext i8 %19 to i16
  br label %cond.end345.us

cond.end345.us:                                   ; preds = %cond.true331.us, %if.then327.us
  %cond346.us = phi i16 [ %20, %cond.true331.us ], [ 0, %if.then327.us ]
  store i16 %cond346.us, ptr %arr_4, align 2
  %mul354.us = mul i64 %mul354906.us, %not353
  store i64 %mul354.us, ptr @<!-- -->var_46, align 8
  br label %for.inc371.us

for.inc371.us:                                    ; preds = %cond.end345.us, %if.end282.us
  %mul354907.us = phi i64 [ 1, %if.end282.us ], [ 0, %cond.end345.us ]
  %indvars.iv.next = or i64 %indvars.iv, 1
  br i1 %var_0, label %for.body194.us, label %for.cond376.preheader.us

cond.true146.us:                                  ; preds = %for.cond131.preheader.us
  %21 = load i8, ptr %arrayidx149.us, align 1
  %conv150.us = sext i8 %21 to i32
  br label %cond.end154.us

cond.end154.us:                                   ; preds = %for.cond131.preheader.us, %cond.true146.us
  %cond155.us = phi i32 [ %conv150.us, %cond.true146.us ], [ 0, %for.cond131.preheader.us ]
  %shl.us = shl i32 %div.us, %cond155.us
  %tobool157.not.us = icmp eq i32 %shl.us, 0
  br i1 %tobool157.not.us, label %if.end.us, label %if.then.us

if.then.us:                                       ; preds = %cond.end154.us
  store i32 0, ptr %arr_4, align 4
  br label %if.end.us

if.end.us:                                        ; preds = %if.then.us, %cond.end154.us
  %22 = phi i32 [ 0, %if.then.us ], [ %10, %cond.end154.us ]
  store i8 %1, ptr %arr_4, align 1
  call void @<!-- -->llvm.assume(i1 true)
  %23 = load i8, ptr %arrayidx149.us, align 1
  %conv150.us.2 = sext i8 %23 to i32
  %shl.us.2 = shl i32 18984, %conv150.us.2
  %tobool157.not.us.2 = icmp eq i32 %shl.us.2, 0
  br i1 %tobool157.not.us.2, label %if.then.us.5, label %if.then.us.2

if.then.us.2:                                     ; preds = %if.end.us
  %.sroa.speculated826.us.2 = tail call i32 @<!-- -->llvm.smin.i32(i32 %10, i32 0)
  store i8 0, ptr %arr_4, align 1
  br label %if.then.us.5

if.then.us.5:                                     ; preds = %if.then.us.2, %if.end.us
  %24 = phi i32 [ 0, %if.then.us.2 ], [ %22, %if.end.us ]
  %.sroa.speculated826.us.5 = tail call i32 @<!-- -->llvm.smin.i32(i32 %24, i32 1410036665)
  br i1 %var_0, label %cond.end154.us.7, label %cond.true146.us.7

cond.true146.us.7:                                ; preds = %if.then.us.5
  %25 = load i8, ptr %arrayidx149.us, align 1
  %conv150.us.7 = sext i8 %25 to i32
  br label %cond.end154.us.7

cond.end154.us.7:                                 ; preds = %cond.true146.us.7, %if.then.us.5
  %cond155.us.7 = phi i32 [ %conv150.us.7, %cond.true146.us.7 ], [ 0, %if.then.us.5 ]
  %shl.us.7 = shl i32 18984, %cond155.us.7
  %tobool157.not.us.7 = icmp eq i32 %shl.us.7, 0
  br i1 %tobool157.not.us.7, label %if.end.us.7, label %if.then.us.7

if.then.us.7:                                     ; preds = %cond.end154.us.7
  store i32 0, ptr %arr_3, align 4
  br label %if.end.us.7

if.end.us.7:                                      ; preds = %if.then.us.7, %cond.end154.us.7
  %26 = phi i32 [ 0, %if.then.us.7 ], [ %.sroa.speculated826.us.5, %cond.end154.us.7 ]
  %arrayidx259.us = getelementptr i16, ptr %arrayidx257.us, i64 %conv96880.us
  br label %for.body194.us

for.body113.us:                                   ; preds = %for.body113.us, %for.body99.us
  br i1 %var_0, label %for.body113.us, label %for.cond131.preheader.us

for.cond510.preheader.us:                         ; preds = %for.inc505.us
  %cmp97.us = icmp slt i16 %add538, 0
  br i1 %cmp97.us, label %for.body99.us, label %for.cond.cleanup98.us

for.cond376.preheader.us:                         ; preds = %for.inc371.us
  %arrayidx384.us = getelementptr i16, ptr null, i64 %conv96880.us
  %gep876.us = getelementptr [24 x [24 x i16]], ptr %invariant.gep875.us, i64 %conv96880.us
  br label %for.body380.us

for.cond131.preheader.us:                         ; preds = %for.body113.us
  store i8 %var_3, ptr %arr_4, align 1
  %27 = load i16, ptr %gep884.us, align 2
  %28 = mul i16 18984, %27
  %div.us = zext i16 %28 to i32
  %tobool145.not.us = icmp eq i8 0, 0
  br i1 %tobool145.not.us, label %cond.end154.us, label %cond.true146.us

for.cond95.preheader.us:                          ; preds = %for.cond.cleanup98.us, %for.body67.us
  %indvars.iv1021 = phi i64 [ 1, %for.cond.cleanup98.us ], [ 0, %for.body67.us ]
  %29 = phi i8 [ %16, %for.cond.cleanup98.us ], [ %4, %for.body67.us ]
  %add413978.us = phi i16 [ %var_4, %for.cond.cleanup98.us ], [ %add413977.us, %for.body67.us ]
  %30 = phi i8 [ %.sroa.speculated829.us, %for.cond.cleanup98.us ], [ %5, %for.body67.us ]
  %mul354904.us = phi i64 [ 0, %for.cond.cleanup98.us ], [ %mul354903918.us, %for.body67.us ]
  %gep884.us = getelementptr [24 x [24 x i16]], ptr %gep889.us, i64 %indvars.iv1021
  %arrayidx321.us = getelementptr i8, ptr %arrayidx319.us, i64 %indvars.iv1021
  %arrayidx257.us = getelementptr [24 x i16], ptr null, i64 %indvars.iv1021
  br label %for.body99.us

for.cond563.preheader:                            ; preds = %for.body41, %entry
  ret void

for.body41:                                       ; preds = %for.body41.lr.ph
  store i8 0, ptr %arr_12, align 1
  store i8 0, ptr %arr_3, align 1
  br label %for.cond563.preheader

; uselistorder directives
  uselistorder i16 %add413.us17, { 1, 0 }
  uselistorder i16 %mul287.us, { 1, 0 }
}

; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
declare i8 @<!-- -->llvm.smin.i8(i8, i8) #<!-- -->1

; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
declare i64 @<!-- -->llvm.umin.i64(i64, i64) #<!-- -->1

; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
declare i8 @<!-- -->llvm.smax.i8(i8, i8) #<!-- -->1

; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
declare i32 @<!-- -->llvm.smin.i32(i32, i32) #<!-- -->1

; Function Attrs: nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: write)
declare void @<!-- -->llvm.assume(i1 noundef) #<!-- -->2

; uselistorder directives
uselistorder ptr @<!-- -->llvm.smax.i8, { 1, 0 }
uselistorder ptr @<!-- -->llvm.smin.i32, { 1, 0 }

attributes #<!-- -->0 = { "frame-pointer"="non-leaf" "target-cpu"="grace" }
attributes #<!-- -->1 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
attributes #<!-- -->2 = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: write) }

[sjoerdmeijer]

Looks related to #149092, but that was a different STR.

[sjoerdmeijer]

CC: @paulwalker-arm , @davemgreen , @rj-jesus

[rj-jesus]

It looks like this bisects to #159211. We have something like this:

selectOrSplit GPR64common:%382 [752r,1288r:0) 0@752r  weight:2.670940e-04
RS_Split Cascade 0
Analyze counted 2 instrs in 1 blocks, through 0 blocks.
Inline spilling GPR64common:%382 [752r,1288r:0) 0@752r  weight:2.670940e-04
From original %116
        skipping remat of def %382:gpr64common = ADDXri %307:gpr64common, 120, 0
        remat:  1284r   %387:gpr64sp = ADDXri %307:gpr64common, 120, 0
                1288e   STRXui killed %387:gpr64sp, %stack.12, 0 :: (store (s64) into %stack.12)

With:

752B      %116:gpr64sp = ADDXri %85:gpr64common, 120, 0

The register class mismatch seems to be introduced when copying from %116 in InlineSpiller::reMaterializeFor.

I'm not sure if we should constrain the class of NewVReg to accommodate its uses or if there's a better way of fixing this?

CC @lukel97 @preames

[lukel97]

Thanks for bisecting that. If I'm understanding this correctly gpr64sp is a subclass of gpr64common? In that case I wonder why the rematerialized instruction is unnecessarily constrained to gpr64sp if STRXui needs gpr64common. I can't immediately see anything in InlineSpiller::reMaterializeFor that would cause the register class to be constrained but will take a further look.

[rj-jesus]

If I'm understanding this correctly gpr64sp is a subclass of gpr64common?

It's the other way round I believe, gpr64sp contains gpr64common and SP. IIUC the constraint is initially copied from the original def %116 in

llvm-project/llvm/lib/CodeGen/InlineSpiller.cpp

Line 722 in 253e435

Register NewVReg = Edit->createFrom(Original);

It then doesn't seem to be constrained further to comply with the use in the store.

[lukel97]

If I'm understanding this correctly gpr64sp is a subclass of gpr64common?

It's the other way round I believe, gpr64sp contains gpr64common and SP. IIUC the constraint is initially copied from the original def %116 in

llvm-project/llvm/lib/CodeGen/InlineSpiller.cpp

Line 722 in 253e435

Register NewVReg = Edit->createFrom(Original);
It then doesn't seem to be constrained further to comply with the use in the store.

Oh right. In that case just constraining the register class immediately after creating it from Original does the trick. Weirdly enough you can't just use Edit->createFrom(VirtReg.reg()), since you run into the same problem but the other way round in some tests, where the definition now has an incorrect register class. Looks like you need to take both registers into account. Will submit a patch soon.