[MIPS GlobalISel] Fix mul operands

[Petar.Avramovic]

Summary

Unsigned mul high for MIPS32 is selected into two PseudoInstructions:
PseudoMULTu and PseudoMFHI that use accumulator register class ACC64 for
some of its operands. Registers in this class have appropriate hi and lo
register as subregisters: $lo0 and $hi0 are subregisters of $ac0 etc.
mul instruction implicit-defs $lo0 and $hi0 according to MipsInstrInfo.td.
In functions where mul and PseudoMULTu are present fastRegisterAllocator
will "run out of registers during register allocation" because
'calcSpillCost' for $ac0 will return spillImpossible because subregisters
$lo0 and $hi0 of $ac0 are reserved by mul instruction above. A solution is
to mark implicit-defs of $lo0 and $hi0 as dead in mul instruction.

[atanasyan]

LGTM

[Petar.Avramovic]

Reverted in https://reviews.llvm.org/rL355219
since it causes failure on bot that tests clang built with address sanitizer.

[Petar.Avramovic]

selectImpl in some cases deletes I (using eraseFromParent()) and creates new instruction that is inserted in place of I in machine basic block.
In such cases I.getOpcode() accesses memory that was freed and can return unpredictable value.
For that reason we now select G_MUL manually before call to selectImpl.

[atanasyan]

LGTM

llvm/trunk/lib/Target/Mips/MipsInstructionSelector.cpp

bool MipsInstructionSelector::select(MachineInstr &I,
 
   if (!isPreISelGenericOpcode(I.getOpcode())) {
     if (I.isCopy())
       return selectCopy(I, TII, MRI, TRI, RBI);
 
     return true;
   }
 
-  if (selectImpl(I, CoverageInfo)) {
+  if (I.getOpcode() == Mips::G_MUL) {
+    MachineInstr *Mul = BuildMI(MBB, I, I.getDebugLoc(), TII.get(Mips::MUL))
+                            .add(I.getOperand(0))
+                            .add(I.getOperand(1))
+                            .add(I.getOperand(2));
+    if (!constrainSelectedInstRegOperands(*Mul, TII, TRI, RBI))
+      return false;
+    Mul->getOperand(3).setIsDead(true);
+    Mul->getOperand(4).setIsDead(true);
+
+    I.eraseFromParent();
     return true;
   }
 
+  if (selectImpl(I, CoverageInfo))
+    return true;
+
   MachineInstr *MI = nullptr;
   using namespace TargetOpcode;
 
   switch (I.getOpcode()) {
   case G_UMULH: {
     unsigned PseudoMULTuReg = MRI.createVirtualRegister(&Mips::ACC64RegClass);
     MachineInstr *PseudoMULTu, *PseudoMove;
 

llvm/trunk/test/CodeGen/Mips/GlobalISel/instruction-select/mul.mir

 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
 # RUN: llc -O0 -mtriple=mipsel-linux-gnu -run-pass=instruction-select -verify-machineinstrs %s -o - | FileCheck %s -check-prefixes=MIPS32
 --- |
 
   define void @mul_i32(i32 %x, i32 %y) {entry: ret void}
-  define void @umul_with_overflow(i32 %lhs, i32 %rhs, i1* %pcarry_flag) { ret void }
+  define void @umul_with_overflow(i32 %lhs, i32 %rhs, i32* %pmul, i1* %pcarry_flag) { ret void }
 
 ...
 ---
 name:            mul_i32
 alignment:       2
 legalized:       true
 regBankSelected: true
 tracksRegLiveness: true
 body:             |
   bb.0.entry:
     liveins: $a0, $a1
 
     ; MIPS32-LABEL: name: mul_i32
     ; MIPS32: liveins: $a0, $a1
     ; MIPS32: [[COPY:%[0-9]+]]:gpr32 = COPY $a0
     ; MIPS32: [[COPY1:%[0-9]+]]:gpr32 = COPY $a1
-    ; MIPS32: [[MUL:%[0-9]+]]:gpr32 = MUL [[COPY]], [[COPY1]], implicit-def $hi0, implicit-def $lo0
+    ; MIPS32: [[MUL:%[0-9]+]]:gpr32 = MUL [[COPY]], [[COPY1]], implicit-def dead $hi0, implicit-def dead $lo0
     ; MIPS32: $v0 = COPY [[MUL]]
     ; MIPS32: RetRA implicit $v0
     %0:gprb(s32) = COPY $a0
     %1:gprb(s32) = COPY $a1
     %2:gprb(s32) = G_MUL %0, %1
     $v0 = COPY %2(s32)
     RetRA implicit $v0
 
 ...
 ---
 name:            umul_with_overflow
 alignment:       2
 legalized:       true
 regBankSelected: true
 tracksRegLiveness: true
 body:             |
   bb.1 (%ir-block.0):
-    liveins: $a0, $a1, $a2
+    liveins: $a0, $a1, $a2, $a3
 
     ; MIPS32-LABEL: name: umul_with_overflow
-    ; MIPS32: liveins: $a0, $a1, $a2
+    ; MIPS32: liveins: $a0, $a1, $a2, $a3
     ; MIPS32: [[COPY:%[0-9]+]]:gpr32 = COPY $a0
     ; MIPS32: [[COPY1:%[0-9]+]]:gpr32 = COPY $a1
     ; MIPS32: [[COPY2:%[0-9]+]]:gpr32 = COPY $a2
+    ; MIPS32: [[COPY3:%[0-9]+]]:gpr32 = COPY $a3
+    ; MIPS32: [[MUL:%[0-9]+]]:gpr32 = MUL [[COPY]], [[COPY1]], implicit-def dead $hi0, implicit-def dead $lo0
     ; MIPS32: [[PseudoMULTu:%[0-9]+]]:acc64 = PseudoMULTu [[COPY]], [[COPY1]]
     ; MIPS32: [[PseudoMFHI:%[0-9]+]]:gpr32 = PseudoMFHI [[PseudoMULTu]]
     ; MIPS32: [[LUi:%[0-9]+]]:gpr32 = LUi 0
     ; MIPS32: [[ORi:%[0-9]+]]:gpr32 = ORi [[LUi]], 0
     ; MIPS32: [[XOR:%[0-9]+]]:gpr32 = XOR [[PseudoMFHI]], [[ORi]]
     ; MIPS32: [[SLTu:%[0-9]+]]:gpr32 = SLTu $zero, [[XOR]]
     ; MIPS32: [[LUi1:%[0-9]+]]:gpr32 = LUi 0
     ; MIPS32: [[ORi1:%[0-9]+]]:gpr32 = ORi [[LUi1]], 1
     ; MIPS32: [[AND:%[0-9]+]]:gpr32 = AND [[SLTu]], [[ORi1]]
-    ; MIPS32: SB [[AND]], [[COPY2]], 0 :: (store 1 into %ir.pcarry_flag)
+    ; MIPS32: SB [[AND]], [[COPY3]], 0 :: (store 1 into %ir.pcarry_flag)
+    ; MIPS32: SW [[MUL]], [[COPY2]], 0 :: (store 4 into %ir.pmul)
     ; MIPS32: RetRA
     %0:gprb(s32) = COPY $a0
     %1:gprb(s32) = COPY $a1
     %2:gprb(p0) = COPY $a2
-    %6:gprb(s32) = G_UMULH %0, %1
-    %7:gprb(s32) = G_CONSTANT i32 0
-    %8:gprb(s32) = G_ICMP intpred(ne), %6(s32), %7
-    %9:gprb(s32) = G_CONSTANT i32 1
-    %10:gprb(s32) = COPY %8(s32)
-    %5:gprb(s32) = G_AND %10, %9
-    G_STORE %5(s32), %2(p0) :: (store 1 into %ir.pcarry_flag)
+    %3:gprb(p0) = COPY $a3
+    %4:gprb(s32) = G_MUL %0, %1
+    %7:gprb(s32) = G_UMULH %0, %1
+    %8:gprb(s32) = G_CONSTANT i32 0
+    %9:gprb(s32) = G_ICMP intpred(ne), %7(s32), %8
+    %10:gprb(s32) = G_CONSTANT i32 1
+    %11:gprb(s32) = COPY %9(s32)
+    %6:gprb(s32) = G_AND %11, %10
+    G_STORE %6(s32), %3(p0) :: (store 1 into %ir.pcarry_flag)
+    G_STORE %4(s32), %2(p0) :: (store 4 into %ir.pmul)
     RetRA
 
 ...

llvm/trunk/test/CodeGen/Mips/GlobalISel/legalizer/mul.mir

 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
 # RUN: llc -O0 -mtriple=mipsel-linux-gnu -run-pass=legalizer -verify-machineinstrs %s -o - | FileCheck %s -check-prefixes=MIPS32
 --- |
 
   define void @mul_i32() {entry: ret void}
   define void @mul_i8_sext() {entry: ret void}
   define void @mul_i8_zext() {entry: ret void}
   define void @mul_i8_aext() {entry: ret void}
   define void @mul_i16_sext() {entry: ret void}
   define void @mul_i16_zext() {entry: ret void}
   define void @mul_i16_aext() {entry: ret void}
-  define void @umul_with_overflow(i32 %lhs, i32 %rhs, i1* %pcarry_flag) { ret void }
+  define void @umul_with_overflow(i32 %lhs, i32 %rhs, i32* %pmul, i1* %pcarry_flag) { ret void }
 
 ...
 ---
 name:            mul_i32
 alignment:       2
 tracksRegLiveness: true
 body:             |
   bb.0.entry:

 
 ...
 ---
 name:            umul_with_overflow
 alignment:       2
 tracksRegLiveness: true
 body:             |
   bb.1 (%ir-block.0):
-    liveins: $a0, $a1, $a2
+    liveins: $a0, $a1, $a2, $a3
 
     ; MIPS32-LABEL: name: umul_with_overflow
-    ; MIPS32: liveins: $a0, $a1, $a2
+    ; MIPS32: liveins: $a0, $a1, $a2, $a3
     ; MIPS32: [[COPY:%[0-9]+]]:_(s32) = COPY $a0
     ; MIPS32: [[COPY1:%[0-9]+]]:_(s32) = COPY $a1
     ; MIPS32: [[COPY2:%[0-9]+]]:_(p0) = COPY $a2
+    ; MIPS32: [[COPY3:%[0-9]+]]:_(p0) = COPY $a3
+    ; MIPS32: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[COPY]], [[COPY1]]
     ; MIPS32: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[COPY]], [[COPY1]]
     ; MIPS32: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
     ; MIPS32: [[ICMP:%[0-9]+]]:_(s32) = G_ICMP intpred(ne), [[UMULH]](s32), [[C]]
     ; MIPS32: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
-    ; MIPS32: [[COPY3:%[0-9]+]]:_(s32) = COPY [[ICMP]](s32)
-    ; MIPS32: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
-    ; MIPS32: G_STORE [[AND]](s32), [[COPY2]](p0) :: (store 1 into %ir.pcarry_flag)
+    ; MIPS32: [[COPY4:%[0-9]+]]:_(s32) = COPY [[ICMP]](s32)
+    ; MIPS32: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
+    ; MIPS32: G_STORE [[AND]](s32), [[COPY3]](p0) :: (store 1 into %ir.pcarry_flag)
+    ; MIPS32: G_STORE [[MUL]](s32), [[COPY2]](p0) :: (store 4 into %ir.pmul)
     ; MIPS32: RetRA
     %0:_(s32) = COPY $a0
     %1:_(s32) = COPY $a1
     %2:_(p0) = COPY $a2
-    %3:_(s32), %4:_(s1) = G_UMULO %0, %1
-    G_STORE %4(s1), %2(p0) :: (store 1 into %ir.pcarry_flag)
+    %3:_(p0) = COPY $a3
+    %4:_(s32), %5:_(s1) = G_UMULO %0, %1
+    G_STORE %5(s1), %3(p0) :: (store 1 into %ir.pcarry_flag)
+    G_STORE %4(s32), %2(p0) :: (store 4 into %ir.pmul)
     RetRA
 
 ...

llvm/trunk/test/CodeGen/Mips/GlobalISel/llvm-ir/mul.ll

 ; MIPS32-NEXT:    jr $ra
 ; MIPS32-NEXT:    nop
 entry:
   %mul = mul i16 %b, %a
   ret i16 %mul
 }
 
 declare { i32, i1 } @llvm.umul.with.overflow.i32(i32, i32)
-define void @umul_with_overflow(i32 %lhs, i32 %rhs, i1* %pcarry_flag) {
+define void @umul_with_overflow(i32 %lhs, i32 %rhs, i32* %pmul, i1* %pcarry_flag) {
 ; MIPS32-LABEL: umul_with_overflow:
 ; MIPS32:       # %bb.0:
+; MIPS32-NEXT:    mul $1, $4, $5
 ; MIPS32-NEXT:    multu $4, $5
 ; MIPS32-NEXT:    mfhi $4
 ; MIPS32-NEXT:    lui $5, 0
 ; MIPS32-NEXT:    ori $5, $5, 0
 ; MIPS32-NEXT:    xor $4, $4, $5
 ; MIPS32-NEXT:    sltu $4, $zero, $4
 ; MIPS32-NEXT:    lui $5, 0
 ; MIPS32-NEXT:    ori $5, $5, 1
 ; MIPS32-NEXT:    and $4, $4, $5
-; MIPS32-NEXT:    sb $4, 0($6)
+; MIPS32-NEXT:    sb $4, 0($7)
+; MIPS32-NEXT:    sw $1, 0($6)
 ; MIPS32-NEXT:    jr $ra
 ; MIPS32-NEXT:    nop
   %res = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 %lhs, i32 %rhs)
   %carry_flag = extractvalue { i32, i1 } %res, 1
+  %mul = extractvalue { i32, i1 } %res, 0
   store i1 %carry_flag, i1* %pcarry_flag
+  store i32 %mul, i32* %pmul
   ret void
 }

llvm/trunk/test/CodeGen/Mips/GlobalISel/regbankselect/mul.mir

 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
 # RUN: llc -O0 -mtriple=mipsel-linux-gnu -run-pass=regbankselect -verify-machineinstrs %s -o - | FileCheck %s -check-prefixes=MIPS32
 --- |
 
   define void @mul_i32(i32 %x, i32 %y) {entry: ret void}
-  define void @umul_with_overflow(i32 %lhs, i32 %rhs, i1* %pcarry_flag) { ret void }
+  define void @umul_with_overflow(i32 %lhs, i32 %rhs, i32* %pmul, i1* %pcarry_flag) { ret void }
 
 ...
 ---
 name:            mul_i32
 alignment:       2
 legalized:       true
 tracksRegLiveness: true
 body:             |

 ...
 ---
 name:            umul_with_overflow
 alignment:       2
 legalized:       true
 tracksRegLiveness: true
 body:             |
   bb.1 (%ir-block.0):
-    liveins: $a0, $a1, $a2
+    liveins: $a0, $a1, $a2, $a3
 
     ; MIPS32-LABEL: name: umul_with_overflow
-    ; MIPS32: liveins: $a0, $a1, $a2
+    ; MIPS32: liveins: $a0, $a1, $a2, $a3
     ; MIPS32: [[COPY:%[0-9]+]]:gprb(s32) = COPY $a0
     ; MIPS32: [[COPY1:%[0-9]+]]:gprb(s32) = COPY $a1
     ; MIPS32: [[COPY2:%[0-9]+]]:gprb(p0) = COPY $a2
+    ; MIPS32: [[COPY3:%[0-9]+]]:gprb(p0) = COPY $a3
+    ; MIPS32: [[MUL:%[0-9]+]]:gprb(s32) = G_MUL [[COPY]], [[COPY1]]
     ; MIPS32: [[UMULH:%[0-9]+]]:gprb(s32) = G_UMULH [[COPY]], [[COPY1]]
     ; MIPS32: [[C:%[0-9]+]]:gprb(s32) = G_CONSTANT i32 0
     ; MIPS32: [[ICMP:%[0-9]+]]:gprb(s32) = G_ICMP intpred(ne), [[UMULH]](s32), [[C]]
     ; MIPS32: [[C1:%[0-9]+]]:gprb(s32) = G_CONSTANT i32 1
-    ; MIPS32: [[COPY3:%[0-9]+]]:gprb(s32) = COPY [[ICMP]](s32)
-    ; MIPS32: [[AND:%[0-9]+]]:gprb(s32) = G_AND [[COPY3]], [[C1]]
-    ; MIPS32: G_STORE [[AND]](s32), [[COPY2]](p0) :: (store 1 into %ir.pcarry_flag)
+    ; MIPS32: [[COPY4:%[0-9]+]]:gprb(s32) = COPY [[ICMP]](s32)
+    ; MIPS32: [[AND:%[0-9]+]]:gprb(s32) = G_AND [[COPY4]], [[C1]]
+    ; MIPS32: G_STORE [[AND]](s32), [[COPY3]](p0) :: (store 1 into %ir.pcarry_flag)
+    ; MIPS32: G_STORE [[MUL]](s32), [[COPY2]](p0) :: (store 4 into %ir.pmul)
     ; MIPS32: RetRA
     %0:_(s32) = COPY $a0
     %1:_(s32) = COPY $a1
     %2:_(p0) = COPY $a2
-    %6:_(s32) = G_UMULH %0, %1
-    %7:_(s32) = G_CONSTANT i32 0
-    %8:_(s32) = G_ICMP intpred(ne), %6(s32), %7
-    %9:_(s32) = G_CONSTANT i32 1
-    %10:_(s32) = COPY %8(s32)
-    %5:_(s32) = G_AND %10, %9
-    G_STORE %5(s32), %2(p0) :: (store 1 into %ir.pcarry_flag)
+    %3:_(p0) = COPY $a3
+    %4:_(s32) = G_MUL %0, %1
+    %7:_(s32) = G_UMULH %0, %1
+    %8:_(s32) = G_CONSTANT i32 0
+    %9:_(s32) = G_ICMP intpred(ne), %7(s32), %8
+    %10:_(s32) = G_CONSTANT i32 1
+    %11:_(s32) = COPY %9(s32)
+    %6:_(s32) = G_AND %11, %10
+    G_STORE %6(s32), %3(p0) :: (store 1 into %ir.pcarry_flag)
+    G_STORE %4(s32), %2(p0) :: (store 4 into %ir.pmul)
     RetRA
 
 ...