[PEI][WebAssembly] Switch to backwards frame index elimination

[foad]

Summary

Backwards frame index elimination uses backwards register scavenging,
which is preferred because it does not rely on accurate kill flags.

[tlively]

Would it be possible to add a test that shows benefit from this? All the test changes in this diff seem to merely renumber virtual registers.

[arsenm]

In D156691#4555665, @tlively wrote:

Would it be possible to add a test that shows benefit from this? All the test changes in this diff seem to merely renumber virtual registers.

Probably not, any improvement implies there's a bug somewhere from not setting a kill flag. This avoids that maintenance cost

[tlively]

LGTM, then. Is there somewhere I can read more about the context for this change?

[foad]

In D156691#4556091, @tlively wrote:

LGTM, then. Is there somewhere I can read more about the context for this change?

Thanks. I am not aware of a good overall summary, but there are various comments scattered around like:

include/llvm/CodeGen/MachinePassRegistry.def
71-// LiveVariables currently requires pure SSA form.
72:// FIXME: Once TwoAddressInstruction pass no longer uses kill flags,
73-// LiveVariables can be removed completely, and LiveIntervals can be directly
74-// computed. (We still either need to regenerate kill flags after regalloc, or
75-// preferably fix the scavenger to not depend on them).

llvm/lib/Target/WebAssembly/WebAssemblyRegisterInfo.h

public:
 
   // Code Generation virtual methods.
   const MCPhysReg *getCalleeSavedRegs(const MachineFunction *MF) const override;
   BitVector getReservedRegs(const MachineFunction &MF) const override;
   bool eliminateFrameIndex(MachineBasicBlock::iterator MI, int SPAdj,
                            unsigned FIOperandNum,
                            RegScavenger *RS = nullptr) const override;
 
+  bool supportsBackwardScavenger() const override { return true; }
+
   // Debug information queries.
   Register getFrameRegister(const MachineFunction &MF) const override;
 
   const TargetRegisterClass *
   getPointerRegClass(const MachineFunction &MF,
                      unsigned Kind = 0) const override;
   // This does not apply to wasm.
   const uint32_t *getNoPreservedMask() const override { return nullptr; }
 };
 
 } // end namespace llvm
 
 #endif

llvm/test/CodeGen/WebAssembly/i128.ll

 ; CHECK-NEXT:    .local i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    global.get $push11=, __stack_pointer
 ; CHECK-NEXT:    i32.const $push12=, 32
 ; CHECK-NEXT:    i32.sub $push23=, $pop11, $pop12
 ; CHECK-NEXT:    local.tee $push22=, 5, $pop23
 ; CHECK-NEXT:    global.set __stack_pointer, $pop22
 ; CHECK-NEXT:    local.get $push24=, 5
-; CHECK-NEXT:    i32.const $push15=, 16
-; CHECK-NEXT:    i32.add $push16=, $pop24, $pop15
+; CHECK-NEXT:    i32.const $push17=, 16
+; CHECK-NEXT:    i32.add $push18=, $pop24, $pop17
 ; CHECK-NEXT:    local.get $push27=, 1
 ; CHECK-NEXT:    local.get $push26=, 2
 ; CHECK-NEXT:    local.get $push25=, 3
 ; CHECK-NEXT:    i32.wrap_i64 $push21=, $pop25
 ; CHECK-NEXT:    local.tee $push20=, 6, $pop21
-; CHECK-NEXT:    call __ashlti3, $pop16, $pop27, $pop26, $pop20
+; CHECK-NEXT:    call __ashlti3, $pop18, $pop27, $pop26, $pop20
 ; CHECK-NEXT:    local.get $push31=, 5
 ; CHECK-NEXT:    local.get $push30=, 1
 ; CHECK-NEXT:    local.get $push29=, 2
 ; CHECK-NEXT:    i32.const $push0=, 128
 ; CHECK-NEXT:    local.get $push28=, 6
 ; CHECK-NEXT:    i32.sub $push1=, $pop0, $pop28
 ; CHECK-NEXT:    call __lshrti3, $pop31, $pop30, $pop29, $pop1
 ; CHECK-NEXT:    local.get $push34=, 0
 ; CHECK-NEXT:    local.get $push32=, 5
-; CHECK-NEXT:    i32.const $push17=, 16
-; CHECK-NEXT:    i32.add $push18=, $pop32, $pop17
+; CHECK-NEXT:    i32.const $push15=, 16
+; CHECK-NEXT:    i32.add $push16=, $pop32, $pop15
 ; CHECK-NEXT:    i32.const $push2=, 8
-; CHECK-NEXT:    i32.add $push3=, $pop18, $pop2
+; CHECK-NEXT:    i32.add $push3=, $pop16, $pop2
 ; CHECK-NEXT:    i64.load $push4=, 0($pop3)
 ; CHECK-NEXT:    local.get $push33=, 5
 ; CHECK-NEXT:    i32.const $push19=, 8
 ; CHECK-NEXT:    i32.add $push5=, $pop33, $pop19
 ; CHECK-NEXT:    i64.load $push6=, 0($pop5)
 ; CHECK-NEXT:    i64.or $push7=, $pop4, $pop6
 ; CHECK-NEXT:    i64.store 8($pop34), $pop7
 ; CHECK-NEXT:    local.get $push37=, 0

 ; CHECK-NEXT:    .local i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    global.get $push13=, __stack_pointer
 ; CHECK-NEXT:    i32.const $push14=, 32
 ; CHECK-NEXT:    i32.sub $push25=, $pop13, $pop14
 ; CHECK-NEXT:    local.tee $push24=, 5, $pop25
 ; CHECK-NEXT:    global.set __stack_pointer, $pop24
 ; CHECK-NEXT:    local.get $push26=, 5
-; CHECK-NEXT:    i32.const $push17=, 16
-; CHECK-NEXT:    i32.add $push18=, $pop26, $pop17
+; CHECK-NEXT:    i32.const $push19=, 16
+; CHECK-NEXT:    i32.add $push20=, $pop26, $pop19
 ; CHECK-NEXT:    local.get $push29=, 1
 ; CHECK-NEXT:    local.get $push28=, 2
 ; CHECK-NEXT:    local.get $push27=, 3
 ; CHECK-NEXT:    i32.wrap_i64 $push0=, $pop27
 ; CHECK-NEXT:    i32.const $push1=, 127
 ; CHECK-NEXT:    i32.and $push23=, $pop0, $pop1
 ; CHECK-NEXT:    local.tee $push22=, 6, $pop23
-; CHECK-NEXT:    call __ashlti3, $pop18, $pop29, $pop28, $pop22
+; CHECK-NEXT:    call __ashlti3, $pop20, $pop29, $pop28, $pop22
 ; CHECK-NEXT:    local.get $push33=, 5
 ; CHECK-NEXT:    local.get $push32=, 1
 ; CHECK-NEXT:    local.get $push31=, 2
 ; CHECK-NEXT:    i32.const $push2=, 128
 ; CHECK-NEXT:    local.get $push30=, 6
 ; CHECK-NEXT:    i32.sub $push3=, $pop2, $pop30
 ; CHECK-NEXT:    call __lshrti3, $pop33, $pop32, $pop31, $pop3
 ; CHECK-NEXT:    local.get $push36=, 0
 ; CHECK-NEXT:    local.get $push34=, 5
-; CHECK-NEXT:    i32.const $push19=, 16
-; CHECK-NEXT:    i32.add $push20=, $pop34, $pop19
+; CHECK-NEXT:    i32.const $push17=, 16
+; CHECK-NEXT:    i32.add $push18=, $pop34, $pop17
 ; CHECK-NEXT:    i32.const $push4=, 8
-; CHECK-NEXT:    i32.add $push5=, $pop20, $pop4
+; CHECK-NEXT:    i32.add $push5=, $pop18, $pop4
 ; CHECK-NEXT:    i64.load $push6=, 0($pop5)
 ; CHECK-NEXT:    local.get $push35=, 5
 ; CHECK-NEXT:    i32.const $push21=, 8
 ; CHECK-NEXT:    i32.add $push7=, $pop35, $pop21
 ; CHECK-NEXT:    i64.load $push8=, 0($pop7)
 ; CHECK-NEXT:    i64.or $push9=, $pop6, $pop8
 ; CHECK-NEXT:    i64.store 8($pop36), $pop9
 ; CHECK-NEXT:    local.get $push39=, 0

 ; CHECK-NEXT:    .local i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    global.get $push11=, __stack_pointer
 ; CHECK-NEXT:    i32.const $push12=, 32
 ; CHECK-NEXT:    i32.sub $push23=, $pop11, $pop12
 ; CHECK-NEXT:    local.tee $push22=, 5, $pop23
 ; CHECK-NEXT:    global.set __stack_pointer, $pop22
 ; CHECK-NEXT:    local.get $push24=, 5
-; CHECK-NEXT:    i32.const $push15=, 16
-; CHECK-NEXT:    i32.add $push16=, $pop24, $pop15
+; CHECK-NEXT:    i32.const $push17=, 16
+; CHECK-NEXT:    i32.add $push18=, $pop24, $pop17
 ; CHECK-NEXT:    local.get $push27=, 1
 ; CHECK-NEXT:    local.get $push26=, 2
 ; CHECK-NEXT:    local.get $push25=, 3
 ; CHECK-NEXT:    i32.wrap_i64 $push21=, $pop25
 ; CHECK-NEXT:    local.tee $push20=, 6, $pop21
-; CHECK-NEXT:    call __lshrti3, $pop16, $pop27, $pop26, $pop20
+; CHECK-NEXT:    call __lshrti3, $pop18, $pop27, $pop26, $pop20
 ; CHECK-NEXT:    local.get $push31=, 5
 ; CHECK-NEXT:    local.get $push30=, 1
 ; CHECK-NEXT:    local.get $push29=, 2
 ; CHECK-NEXT:    i32.const $push0=, 128
 ; CHECK-NEXT:    local.get $push28=, 6
 ; CHECK-NEXT:    i32.sub $push1=, $pop0, $pop28
 ; CHECK-NEXT:    call __ashlti3, $pop31, $pop30, $pop29, $pop1
 ; CHECK-NEXT:    local.get $push34=, 0
 ; CHECK-NEXT:    local.get $push32=, 5
-; CHECK-NEXT:    i32.const $push17=, 16
-; CHECK-NEXT:    i32.add $push18=, $pop32, $pop17
+; CHECK-NEXT:    i32.const $push15=, 16
+; CHECK-NEXT:    i32.add $push16=, $pop32, $pop15
 ; CHECK-NEXT:    i32.const $push2=, 8
-; CHECK-NEXT:    i32.add $push3=, $pop18, $pop2
+; CHECK-NEXT:    i32.add $push3=, $pop16, $pop2
 ; CHECK-NEXT:    i64.load $push4=, 0($pop3)
 ; CHECK-NEXT:    local.get $push33=, 5
 ; CHECK-NEXT:    i32.const $push19=, 8
 ; CHECK-NEXT:    i32.add $push5=, $pop33, $pop19
 ; CHECK-NEXT:    i64.load $push6=, 0($pop5)
 ; CHECK-NEXT:    i64.or $push7=, $pop4, $pop6
 ; CHECK-NEXT:    i64.store 8($pop34), $pop7
 ; CHECK-NEXT:    local.get $push37=, 0

 ; CHECK-NEXT:    .local i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    global.get $push13=, __stack_pointer
 ; CHECK-NEXT:    i32.const $push14=, 32
 ; CHECK-NEXT:    i32.sub $push25=, $pop13, $pop14
 ; CHECK-NEXT:    local.tee $push24=, 5, $pop25
 ; CHECK-NEXT:    global.set __stack_pointer, $pop24
 ; CHECK-NEXT:    local.get $push26=, 5
-; CHECK-NEXT:    i32.const $push17=, 16
-; CHECK-NEXT:    i32.add $push18=, $pop26, $pop17
+; CHECK-NEXT:    i32.const $push19=, 16
+; CHECK-NEXT:    i32.add $push20=, $pop26, $pop19
 ; CHECK-NEXT:    local.get $push29=, 1
 ; CHECK-NEXT:    local.get $push28=, 2
 ; CHECK-NEXT:    local.get $push27=, 3
 ; CHECK-NEXT:    i32.wrap_i64 $push0=, $pop27
 ; CHECK-NEXT:    i32.const $push1=, 127
 ; CHECK-NEXT:    i32.and $push23=, $pop0, $pop1
 ; CHECK-NEXT:    local.tee $push22=, 6, $pop23
-; CHECK-NEXT:    call __lshrti3, $pop18, $pop29, $pop28, $pop22
+; CHECK-NEXT:    call __lshrti3, $pop20, $pop29, $pop28, $pop22
 ; CHECK-NEXT:    local.get $push33=, 5
 ; CHECK-NEXT:    local.get $push32=, 1
 ; CHECK-NEXT:    local.get $push31=, 2
 ; CHECK-NEXT:    i32.const $push2=, 128
 ; CHECK-NEXT:    local.get $push30=, 6
 ; CHECK-NEXT:    i32.sub $push3=, $pop2, $pop30
 ; CHECK-NEXT:    call __ashlti3, $pop33, $pop32, $pop31, $pop3
 ; CHECK-NEXT:    local.get $push36=, 0
 ; CHECK-NEXT:    local.get $push34=, 5
-; CHECK-NEXT:    i32.const $push19=, 16
-; CHECK-NEXT:    i32.add $push20=, $pop34, $pop19
+; CHECK-NEXT:    i32.const $push17=, 16
+; CHECK-NEXT:    i32.add $push18=, $pop34, $pop17
 ; CHECK-NEXT:    i32.const $push4=, 8
-; CHECK-NEXT:    i32.add $push5=, $pop20, $pop4
+; CHECK-NEXT:    i32.add $push5=, $pop18, $pop4
 ; CHECK-NEXT:    i64.load $push6=, 0($pop5)
 ; CHECK-NEXT:    local.get $push35=, 5
 ; CHECK-NEXT:    i32.const $push21=, 8
 ; CHECK-NEXT:    i32.add $push7=, $pop35, $pop21
 ; CHECK-NEXT:    i64.load $push8=, 0($pop7)
 ; CHECK-NEXT:    i64.or $push9=, $pop6, $pop8
 ; CHECK-NEXT:    i64.store 8($pop36), $pop9
 ; CHECK-NEXT:    local.get $push39=, 0

llvm/test/CodeGen/WebAssembly/libcalls.ll

 ; CHECK-NEXT:    .local i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    global.get $push28=, __stack_pointer
 ; CHECK-NEXT:    i32.const $push29=, 144
 ; CHECK-NEXT:    i32.sub $push73=, $pop28, $pop29
 ; CHECK-NEXT:    local.tee $push72=, 6, $pop73
 ; CHECK-NEXT:    global.set __stack_pointer, $pop72
 ; CHECK-NEXT:    local.get $push74=, 6
-; CHECK-NEXT:    i32.const $push32=, 128
-; CHECK-NEXT:    i32.add $push33=, $pop74, $pop32
+; CHECK-NEXT:    i32.const $push62=, 128
+; CHECK-NEXT:    i32.add $push63=, $pop74, $pop62
 ; CHECK-NEXT:    local.get $push78=, 1
 ; CHECK-NEXT:    local.get $push77=, 2
 ; CHECK-NEXT:    local.get $push76=, 3
 ; CHECK-NEXT:    local.get $push75=, 4
-; CHECK-NEXT:    call __addtf3, $pop33, $pop78, $pop77, $pop76, $pop75
+; CHECK-NEXT:    call __addtf3, $pop63, $pop78, $pop77, $pop76, $pop75
 ; CHECK-NEXT:    local.get $push79=, 6
-; CHECK-NEXT:    i32.const $push36=, 112
-; CHECK-NEXT:    i32.add $push37=, $pop79, $pop36
+; CHECK-NEXT:    i32.const $push58=, 112
+; CHECK-NEXT:    i32.add $push59=, $pop79, $pop58
 ; CHECK-NEXT:    local.get $push80=, 6
 ; CHECK-NEXT:    i64.load $push3=, 128($pop80)
 ; CHECK-NEXT:    local.get $push81=, 6
-; CHECK-NEXT:    i32.const $push34=, 128
-; CHECK-NEXT:    i32.add $push35=, $pop81, $pop34
+; CHECK-NEXT:    i32.const $push60=, 128
+; CHECK-NEXT:    i32.add $push61=, $pop81, $pop60
 ; CHECK-NEXT:    i32.const $push0=, 8
-; CHECK-NEXT:    i32.add $push1=, $pop35, $pop0
+; CHECK-NEXT:    i32.add $push1=, $pop61, $pop0
 ; CHECK-NEXT:    i64.load $push2=, 0($pop1)
 ; CHECK-NEXT:    local.get $push83=, 3
 ; CHECK-NEXT:    local.get $push82=, 4
-; CHECK-NEXT:    call __multf3, $pop37, $pop3, $pop2, $pop83, $pop82
+; CHECK-NEXT:    call __multf3, $pop59, $pop3, $pop2, $pop83, $pop82
 ; CHECK-NEXT:    local.get $push84=, 6
-; CHECK-NEXT:    i32.const $push40=, 96
-; CHECK-NEXT:    i32.add $push41=, $pop84, $pop40
+; CHECK-NEXT:    i32.const $push54=, 96
+; CHECK-NEXT:    i32.add $push55=, $pop84, $pop54
 ; CHECK-NEXT:    local.get $push85=, 6
 ; CHECK-NEXT:    i64.load $push6=, 112($pop85)
 ; CHECK-NEXT:    local.get $push86=, 6
-; CHECK-NEXT:    i32.const $push38=, 112
-; CHECK-NEXT:    i32.add $push39=, $pop86, $pop38
+; CHECK-NEXT:    i32.const $push56=, 112
+; CHECK-NEXT:    i32.add $push57=, $pop86, $pop56
 ; CHECK-NEXT:    i32.const $push71=, 8
-; CHECK-NEXT:    i32.add $push4=, $pop39, $pop71
+; CHECK-NEXT:    i32.add $push4=, $pop57, $pop71
 ; CHECK-NEXT:    i64.load $push5=, 0($pop4)
 ; CHECK-NEXT:    local.get $push88=, 3
 ; CHECK-NEXT:    local.get $push87=, 4
-; CHECK-NEXT:    call __divtf3, $pop41, $pop6, $pop5, $pop88, $pop87
+; CHECK-NEXT:    call __divtf3, $pop55, $pop6, $pop5, $pop88, $pop87
 ; CHECK-NEXT:    local.get $push89=, 6
-; CHECK-NEXT:    i32.const $push44=, 80
-; CHECK-NEXT:    i32.add $push45=, $pop89, $pop44
+; CHECK-NEXT:    i32.const $push50=, 80
+; CHECK-NEXT:    i32.add $push51=, $pop89, $pop50
 ; CHECK-NEXT:    local.get $push90=, 6
 ; CHECK-NEXT:    i64.load $push9=, 96($pop90)
 ; CHECK-NEXT:    local.get $push91=, 6
-; CHECK-NEXT:    i32.const $push42=, 96
-; CHECK-NEXT:    i32.add $push43=, $pop91, $pop42
+; CHECK-NEXT:    i32.const $push52=, 96
+; CHECK-NEXT:    i32.add $push53=, $pop91, $pop52
 ; CHECK-NEXT:    i32.const $push70=, 8
-; CHECK-NEXT:    i32.add $push7=, $pop43, $pop70
+; CHECK-NEXT:    i32.add $push7=, $pop53, $pop70
 ; CHECK-NEXT:    i64.load $push8=, 0($pop7)
-; CHECK-NEXT:    call sqrtl, $pop45, $pop9, $pop8
+; CHECK-NEXT:    call sqrtl, $pop51, $pop9, $pop8
 ; CHECK-NEXT:    local.get $push92=, 6
-; CHECK-NEXT:    i32.const $push48=, 64
-; CHECK-NEXT:    i32.add $push49=, $pop92, $pop48
+; CHECK-NEXT:    i32.const $push46=, 64
+; CHECK-NEXT:    i32.add $push47=, $pop92, $pop46
 ; CHECK-NEXT:    local.get $push93=, 6
 ; CHECK-NEXT:    i64.load $push12=, 80($pop93)
 ; CHECK-NEXT:    local.get $push94=, 6
-; CHECK-NEXT:    i32.const $push46=, 80
-; CHECK-NEXT:    i32.add $push47=, $pop94, $pop46
+; CHECK-NEXT:    i32.const $push48=, 80
+; CHECK-NEXT:    i32.add $push49=, $pop94, $pop48
 ; CHECK-NEXT:    i32.const $push69=, 8
-; CHECK-NEXT:    i32.add $push10=, $pop47, $pop69
+; CHECK-NEXT:    i32.add $push10=, $pop49, $pop69
 ; CHECK-NEXT:    i64.load $push11=, 0($pop10)
-; CHECK-NEXT:    call floorl, $pop49, $pop12, $pop11
+; CHECK-NEXT:    call floorl, $pop47, $pop12, $pop11
 ; CHECK-NEXT:    local.get $push95=, 6
-; CHECK-NEXT:    i32.const $push52=, 48
-; CHECK-NEXT:    i32.add $push53=, $pop95, $pop52
+; CHECK-NEXT:    i32.const $push42=, 48
+; CHECK-NEXT:    i32.add $push43=, $pop95, $pop42
 ; CHECK-NEXT:    local.get $push96=, 6
 ; CHECK-NEXT:    i64.load $push15=, 64($pop96)
 ; CHECK-NEXT:    local.get $push97=, 6
-; CHECK-NEXT:    i32.const $push50=, 64
-; CHECK-NEXT:    i32.add $push51=, $pop97, $pop50
+; CHECK-NEXT:    i32.const $push44=, 64
+; CHECK-NEXT:    i32.add $push45=, $pop97, $pop44
 ; CHECK-NEXT:    i32.const $push68=, 8
-; CHECK-NEXT:    i32.add $push13=, $pop51, $pop68
+; CHECK-NEXT:    i32.add $push13=, $pop45, $pop68
 ; CHECK-NEXT:    i64.load $push14=, 0($pop13)
 ; CHECK-NEXT:    local.get $push99=, 3
 ; CHECK-NEXT:    local.get $push98=, 4
-; CHECK-NEXT:    call powl, $pop53, $pop15, $pop14, $pop99, $pop98
+; CHECK-NEXT:    call powl, $pop43, $pop15, $pop14, $pop99, $pop98
 ; CHECK-NEXT:    local.get $push100=, 6
-; CHECK-NEXT:    i32.const $push56=, 32
-; CHECK-NEXT:    i32.add $push57=, $pop100, $pop56
+; CHECK-NEXT:    i32.const $push38=, 32
+; CHECK-NEXT:    i32.add $push39=, $pop100, $pop38
 ; CHECK-NEXT:    local.get $push101=, 6
 ; CHECK-NEXT:    i64.load $push18=, 48($pop101)
 ; CHECK-NEXT:    local.get $push102=, 6
-; CHECK-NEXT:    i32.const $push54=, 48
-; CHECK-NEXT:    i32.add $push55=, $pop102, $pop54
+; CHECK-NEXT:    i32.const $push40=, 48
+; CHECK-NEXT:    i32.add $push41=, $pop102, $pop40
 ; CHECK-NEXT:    i32.const $push67=, 8
-; CHECK-NEXT:    i32.add $push16=, $pop55, $pop67
+; CHECK-NEXT:    i32.add $push16=, $pop41, $pop67
 ; CHECK-NEXT:    i64.load $push17=, 0($pop16)
 ; CHECK-NEXT:    local.get $push103=, 5
-; CHECK-NEXT:    call __powitf2, $pop57, $pop18, $pop17, $pop103
+; CHECK-NEXT:    call __powitf2, $pop39, $pop18, $pop17, $pop103
 ; CHECK-NEXT:    local.get $push104=, 6
-; CHECK-NEXT:    i32.const $push60=, 16
-; CHECK-NEXT:    i32.add $push61=, $pop104, $pop60
+; CHECK-NEXT:    i32.const $push34=, 16
+; CHECK-NEXT:    i32.add $push35=, $pop104, $pop34
 ; CHECK-NEXT:    local.get $push105=, 6
 ; CHECK-NEXT:    i64.load $push21=, 32($pop105)
 ; CHECK-NEXT:    local.get $push106=, 6
-; CHECK-NEXT:    i32.const $push58=, 32
-; CHECK-NEXT:    i32.add $push59=, $pop106, $pop58
+; CHECK-NEXT:    i32.const $push36=, 32
+; CHECK-NEXT:    i32.add $push37=, $pop106, $pop36
 ; CHECK-NEXT:    i32.const $push66=, 8
-; CHECK-NEXT:    i32.add $push19=, $pop59, $pop66
+; CHECK-NEXT:    i32.add $push19=, $pop37, $pop66
 ; CHECK-NEXT:    i64.load $push20=, 0($pop19)
-; CHECK-NEXT:    call truncl, $pop61, $pop21, $pop20
+; CHECK-NEXT:    call truncl, $pop35, $pop21, $pop20
 ; CHECK-NEXT:    local.get $push109=, 6
 ; CHECK-NEXT:    local.get $push107=, 6
 ; CHECK-NEXT:    i64.load $push24=, 16($pop107)
 ; CHECK-NEXT:    local.get $push108=, 6
-; CHECK-NEXT:    i32.const $push62=, 16
-; CHECK-NEXT:    i32.add $push63=, $pop108, $pop62
+; CHECK-NEXT:    i32.const $push32=, 16
+; CHECK-NEXT:    i32.add $push33=, $pop108, $pop32
 ; CHECK-NEXT:    i32.const $push65=, 8
-; CHECK-NEXT:    i32.add $push22=, $pop63, $pop65
+; CHECK-NEXT:    i32.add $push22=, $pop33, $pop65
 ; CHECK-NEXT:    i64.load $push23=, 0($pop22)
 ; CHECK-NEXT:    call nearbyintl, $pop109, $pop24, $pop23
 ; CHECK-NEXT:    local.get $push111=, 0
 ; CHECK-NEXT:    local.get $push110=, 6
 ; CHECK-NEXT:    i32.const $push64=, 8
 ; CHECK-NEXT:    i32.add $push25=, $pop110, $pop64
 ; CHECK-NEXT:    i64.load $push26=, 0($pop25)
 ; CHECK-NEXT:    i64.store 8($pop111), $pop26

 ; CHECK-NEXT:    .local i32, i64
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    global.get $push11=, __stack_pointer
 ; CHECK-NEXT:    i32.const $push12=, 32
 ; CHECK-NEXT:    i32.sub $push23=, $pop11, $pop12
 ; CHECK-NEXT:    local.tee $push22=, 5, $pop23
 ; CHECK-NEXT:    global.set __stack_pointer, $pop22
 ; CHECK-NEXT:    local.get $push24=, 5
-; CHECK-NEXT:    i32.const $push15=, 16
-; CHECK-NEXT:    i32.add $push16=, $pop24, $pop15
+; CHECK-NEXT:    i32.const $push17=, 16
+; CHECK-NEXT:    i32.add $push18=, $pop24, $pop17
 ; CHECK-NEXT:    local.get $push26=, 1
 ; CHECK-NEXT:    local.get $push25=, 3
 ; CHECK-NEXT:    i64.add $push21=, $pop26, $pop25
 ; CHECK-NEXT:    local.tee $push20=, 6, $pop21
 ; CHECK-NEXT:    local.get $push28=, 2
 ; CHECK-NEXT:    local.get $push27=, 4
 ; CHECK-NEXT:    i64.add $push0=, $pop28, $pop27
 ; CHECK-NEXT:    local.get $push30=, 6
 ; CHECK-NEXT:    local.get $push29=, 1
 ; CHECK-NEXT:    i64.lt_u $push1=, $pop30, $pop29
 ; CHECK-NEXT:    i64.extend_i32_u $push2=, $pop1
 ; CHECK-NEXT:    i64.add $push3=, $pop0, $pop2
 ; CHECK-NEXT:    local.get $push32=, 3
 ; CHECK-NEXT:    local.get $push31=, 4
-; CHECK-NEXT:    call __multi3, $pop16, $pop20, $pop3, $pop32, $pop31
+; CHECK-NEXT:    call __multi3, $pop18, $pop20, $pop3, $pop32, $pop31
 ; CHECK-NEXT:    local.get $push37=, 5
 ; CHECK-NEXT:    local.get $push33=, 5
 ; CHECK-NEXT:    i64.load $push7=, 16($pop33)
 ; CHECK-NEXT:    local.get $push34=, 5
-; CHECK-NEXT:    i32.const $push17=, 16
-; CHECK-NEXT:    i32.add $push18=, $pop34, $pop17
+; CHECK-NEXT:    i32.const $push15=, 16
+; CHECK-NEXT:    i32.add $push16=, $pop34, $pop15
 ; CHECK-NEXT:    i32.const $push4=, 8
-; CHECK-NEXT:    i32.add $push5=, $pop18, $pop4
+; CHECK-NEXT:    i32.add $push5=, $pop16, $pop4
 ; CHECK-NEXT:    i64.load $push6=, 0($pop5)
 ; CHECK-NEXT:    local.get $push36=, 3
 ; CHECK-NEXT:    local.get $push35=, 4
 ; CHECK-NEXT:    call __umodti3, $pop37, $pop7, $pop6, $pop36, $pop35
 ; CHECK-NEXT:    local.get $push39=, 0
 ; CHECK-NEXT:    local.get $push38=, 5
 ; CHECK-NEXT:    i32.const $push19=, 8
 ; CHECK-NEXT:    i32.add $push8=, $pop38, $pop19

llvm/test/CodeGen/WebAssembly/multi-return.ll

; CHECK: local.copy $push8=, $[[L1]]
   ret i64 %t1
 }
 
 ; Test returning multiple values from @return_multi_multi.
 
 define { i64, i128 } @test5() {
 ; CHECK-LABEL: test5:
 ; CHECK: call     return_multi_multi
-; CHECK: i32.const  $push10=, 8
-; CHECK: i32.add  $push11=, $[[SP:[0-9]+]], $pop10
+; CHECK: i32.const  $push8=, 8
+; CHECK: i32.add  $push9=, $[[SP:[0-9]+]], $pop8
 ; CHECK: i32.const  $push0=, 16
-; CHECK: i32.add  $push1=, $pop11, $pop0
+; CHECK: i32.add  $push1=, $pop9, $pop0
 ; CHECK: i64.load $[[L1:[0-9]+]]=, 0($pop1)
 ; CHECK: i64.load $[[L2:[0-9]+]]=, 8($[[SP]])
 ; CHECK: i64.load $push2=, 16($[[SP]])
 ; CHECK: i64.store  8($0), $pop2
 ; CHECK: i64.store  0($0), $[[L2]]
 ; CHECK: i32.const  $push12=, 16
 ; CHECK: i32.add  $push3=, $0, $pop12
 ; CHECK: i64.store  0($pop3), $[[L1]]
; CHECK: i64.store  0($pop8), $[[L1]]
   %s0 = insertvalue { i64, i192 } undef, i64 %r0, 0
   %s1 = insertvalue { i64, i192 } %s0, i192 %r2, 1
   ret { i64, i192 } %s1
 }
 
 define { i128, i192, i128, i64 } @test8() {
 ; CHECK-LABEL: test8:
 ; CHECK: call     return_multi_multi
-; CHECK: i32.const  $push20=, 8
-; CHECK: i32.add  $push21=, $[[SP:[0-9]+]], $pop20
+; CHECK: i32.const  $push18=, 8
+; CHECK: i32.add  $push19=, $[[SP:[0-9]+]], $pop18
 ; CHECK: i32.const  $push0=, 32
-; CHECK: i32.add  $push1=, $pop21, $pop0
+; CHECK: i32.add  $push1=, $pop19, $pop0
 ; CHECK: i64.load $[[L1:[0-9]+]]=, 0($pop1)
 ; CHECK: i32.const  $push2=, 48
 ; CHECK: i32.add  $push3=, $[[SP]], $pop2
 ; CHECK: i64.load $[[L2:[0-9]+]]=, 0($pop3)
 ; CHECK: i32.const  $push4=, 24
 ; CHECK: i32.add  $push5=, $[[SP]], $pop4
 ; CHECK: i64.load $[[L3:[0-9]+]]=, 0($pop5)
 ; CHECK: i32.const  $push6=, 64

llvm/test/CodeGen/WebAssembly/umulo-128-legalisation-lowering.ll

 ; RUN: llc < %s -mtriple=wasm32 -wasm-keep-registers | FileCheck %s --check-prefixes=WASM32
 ; NOTE: did not compile on wasm64 at the time the test was created!
 
 define { i128, i8 } @muloti_test(i128 %l, i128 %r) unnamed_addr #0 {
 ; WASM32-LABEL: muloti_test
 ; WASM32: global.get      $push18=, __stack_pointer
 ; WASM32: i32.const       $push19=, 48
 ; WASM32: i32.sub         $push40=, $pop18, $pop19
 ; WASM32: local.tee       $push39=, 5, $pop40
 ; WASM32: global.set      __stack_pointer, $pop39
 ; WASM32: local.get       $push41=, 5
-; WASM32: i32.const       $push22=, 32
-; WASM32: i32.add         $push23=, $pop41, $pop22
+; WASM32: i32.const       $push24=, 32
+; WASM32: i32.add         $push25=, $pop41, $pop24
 ; WASM32: local.get       $push43=, 1
 ; WASM32: i64.const       $push0=, 0
 ; WASM32: local.get       $push42=, 3
 ; WASM32: i64.const       $push38=, 0
-; WASM32: call            __multi3, $pop23, $pop43, $pop0, $pop42, $pop38
+; WASM32: call            __multi3, $pop25, $pop43, $pop0, $pop42, $pop38
 ; WASM32: local.get       $push44=, 5
-; WASM32: i32.const       $push24=, 16
-; WASM32: i32.add         $push25=, $pop44, $pop24
+; WASM32: i32.const       $push22=, 16
+; WASM32: i32.add         $push23=, $pop44, $pop22
 ; WASM32: local.get       $push46=, 4
 ; WASM32: i64.const       $push37=, 0
 ; WASM32: local.get       $push45=, 1
 ; WASM32: i64.const       $push36=, 0
-; WASM32: call            __multi3, $pop25, $pop46, $pop37, $pop45, $pop36
+; WASM32: call            __multi3, $pop23, $pop46, $pop37, $pop45, $pop36
 ; WASM32: local.get       $push49=, 5
 ; WASM32: local.get       $push48=, 2
 ; WASM32: i64.const       $push35=, 0
 ; WASM32: local.get       $push47=, 3
 ; WASM32: i64.const       $push34=, 0
 ; WASM32: call            __multi3, $pop49, $pop48, $pop35, $pop47, $pop34
 ; WASM32: local.get       $push51=, 0
 ; WASM32: local.get       $push50=, 5

llvm/test/CodeGen/WebAssembly/userstack.ll

 ; CHECK-32-NEXT:    .local i32
 ; CHECK-32-NEXT:  # %bb.0:
 ; CHECK-32-NEXT:    global.get $push0=, __stack_pointer
 ; CHECK-32-NEXT:    i32.const $push1=, 48
 ; CHECK-32-NEXT:    i32.sub $push9=, $pop0, $pop1
 ; CHECK-32-NEXT:    local.tee $push8=, 1, $pop9
 ; CHECK-32-NEXT:    global.set __stack_pointer, $pop8
 ; CHECK-32-NEXT:    local.get $push10=, 1
-; CHECK-32-NEXT:    i32.const $push4=, 8
-; CHECK-32-NEXT:    i32.add $push5=, $pop10, $pop4
-; CHECK-32-NEXT:    call ext_func, $pop5
+; CHECK-32-NEXT:    i32.const $push6=, 8
+; CHECK-32-NEXT:    i32.add $push7=, $pop10, $pop6
+; CHECK-32-NEXT:    call ext_func, $pop7
 ; CHECK-32-NEXT:    local.get $push11=, 1
 ; CHECK-32-NEXT:    call ext_func, $pop11
 ; CHECK-32-NEXT:    local.get $push13=, 0
 ; CHECK-32-NEXT:    local.get $push12=, 1
-; CHECK-32-NEXT:    i32.const $push6=, 16
-; CHECK-32-NEXT:    i32.add $push7=, $pop12, $pop6
-; CHECK-32-NEXT:    i32.store 0($pop13), $pop7
+; CHECK-32-NEXT:    i32.const $push4=, 16
+; CHECK-32-NEXT:    i32.add $push5=, $pop12, $pop4
+; CHECK-32-NEXT:    i32.store 0($pop13), $pop5
 ; CHECK-32-NEXT:    local.get $push14=, 1
 ; CHECK-32-NEXT:    i32.const $push2=, 48
 ; CHECK-32-NEXT:    i32.add $push3=, $pop14, $pop2
 ; CHECK-32-NEXT:    global.set __stack_pointer, $pop3
 ; CHECK-32-NEXT:    return
 ;
 ; CHECK-64-LABEL: non_mem_use:
 ; CHECK-64:         .functype non_mem_use (i64) -> ()
 ; CHECK-64-NEXT:    .local i64
 ; CHECK-64-NEXT:  # %bb.0:
 ; CHECK-64-NEXT:    global.get $push0=, __stack_pointer
 ; CHECK-64-NEXT:    i64.const $push1=, 48
 ; CHECK-64-NEXT:    i64.sub $push9=, $pop0, $pop1
 ; CHECK-64-NEXT:    local.tee $push8=, 1, $pop9
 ; CHECK-64-NEXT:    global.set __stack_pointer, $pop8
 ; CHECK-64-NEXT:    local.get $push10=, 1
-; CHECK-64-NEXT:    i64.const $push4=, 8
-; CHECK-64-NEXT:    i64.add $push5=, $pop10, $pop4
-; CHECK-64-NEXT:    call ext_func, $pop5
+; CHECK-64-NEXT:    i64.const $push6=, 8
+; CHECK-64-NEXT:    i64.add $push7=, $pop10, $pop6
+; CHECK-64-NEXT:    call ext_func, $pop7
 ; CHECK-64-NEXT:    local.get $push11=, 1
 ; CHECK-64-NEXT:    call ext_func, $pop11
 ; CHECK-64-NEXT:    local.get $push13=, 0
 ; CHECK-64-NEXT:    local.get $push12=, 1
-; CHECK-64-NEXT:    i64.const $push6=, 16
-; CHECK-64-NEXT:    i64.add $push7=, $pop12, $pop6
-; CHECK-64-NEXT:    i64.store 0($pop13), $pop7
+; CHECK-64-NEXT:    i64.const $push4=, 16
+; CHECK-64-NEXT:    i64.add $push5=, $pop12, $pop4
+; CHECK-64-NEXT:    i64.store 0($pop13), $pop5
 ; CHECK-64-NEXT:    local.get $push14=, 1
 ; CHECK-64-NEXT:    i64.const $push2=, 48
 ; CHECK-64-NEXT:    i64.add $push3=, $pop14, $pop2
 ; CHECK-64-NEXT:    global.set __stack_pointer, $pop3
 ; CHECK-64-NEXT:    return
  %buf = alloca [27 x i8], align 16
  %r = alloca i64
  %r2 = alloca i64