[llvm][LICM] Limit multi-use BOAssociation to FP and Vector (#149829)

Limit the re-association of BOps with multiple users to FP and Vector
arithmetic.

Author: thetheodor

llvm/lib/Transforms/Scalar/LICM.cpp

@@ -2867,7 +2867,7 @@ static bool hoistBOAssociation(Instruction &I, Loop &L,
   bool LVInRHS = L.isLoopInvariant(BO->getOperand(0));
   auto *BO0 = dyn_cast<BinaryOperator>(BO->getOperand(LVInRHS));
   if (!BO0 || BO0->getOpcode() != Opcode || !BO0->isAssociative() ||
-      BO0->hasNUsesOrMore(3))
+      BO0->hasNUsesOrMore(BO0->getType()->isIntegerTy() ? 2 : 3))
     return false;
 
   Value *LV = BO0->getOperand(0);

llvm/test/CodeGen/PowerPC/p10-spill-crlt.ll

@@ -30,16 +30,14 @@ define dso_local void @P10_Spill_CR_LT() local_unnamed_addr {
 ; CHECK-NEXT:    mflr r0
 ; CHECK-NEXT:    std r0, 16(r1)
 ; CHECK-NEXT:    stw r12, 8(r1)
-; CHECK-NEXT:    stdu r1, -64(r1)
-; CHECK-NEXT:    .cfi_def_cfa_offset 64
+; CHECK-NEXT:    stdu r1, -48(r1)
+; CHECK-NEXT:    .cfi_def_cfa_offset 48
 ; CHECK-NEXT:    .cfi_offset lr, 16
-; CHECK-NEXT:    .cfi_offset r29, -24
 ; CHECK-NEXT:    .cfi_offset r30, -16
 ; CHECK-NEXT:    .cfi_offset cr2, 8
 ; CHECK-NEXT:    .cfi_offset cr3, 8
 ; CHECK-NEXT:    .cfi_offset cr4, 8
-; CHECK-NEXT:    std r29, 40(r1) # 8-byte Folded Spill
-; CHECK-NEXT:    std r30, 48(r1) # 8-byte Folded Spill
+; CHECK-NEXT:    std r30, 32(r1) # 8-byte Folded Spill
 ; CHECK-NEXT:    bl call_2@notoc
 ; CHECK-NEXT:    bc 12, 4*cr5+lt, .LBB0_13
 ; CHECK-NEXT:  # %bb.1: # %bb
@@ -67,11 +65,10 @@ define dso_local void @P10_Spill_CR_LT() local_unnamed_addr {
 ; CHECK-NEXT:    bc 12, 4*cr3+eq, .LBB0_11
 ; CHECK-NEXT:  # %bb.6: # %bb32
 ; CHECK-NEXT:    #
+; CHECK-NEXT:    rlwinm r30, r30, 0, 24, 22
 ; CHECK-NEXT:    andi. r3, r30, 2
-; CHECK-NEXT:    rlwinm r29, r30, 0, 24, 22
 ; CHECK-NEXT:    mcrf cr2, cr0
 ; CHECK-NEXT:    bl call_4@notoc
-; CHECK-NEXT:    mr r30, r29
 ; CHECK-NEXT:    beq+ cr2, .LBB0_3
 ; CHECK-NEXT:  # %bb.7: # %bb37
 ; CHECK-NEXT:  .LBB0_8: # %bb22
@@ -92,13 +89,11 @@ define dso_local void @P10_Spill_CR_LT() local_unnamed_addr {
 ; CHECK-BE-NEXT:    stdu r1, -144(r1)
 ; CHECK-BE-NEXT:    .cfi_def_cfa_offset 144
 ; CHECK-BE-NEXT:    .cfi_offset lr, 16
-; CHECK-BE-NEXT:    .cfi_offset r28, -32
 ; CHECK-BE-NEXT:    .cfi_offset r29, -24
 ; CHECK-BE-NEXT:    .cfi_offset r30, -16
 ; CHECK-BE-NEXT:    .cfi_offset cr2, 8
 ; CHECK-BE-NEXT:    .cfi_offset cr2, 8
 ; CHECK-BE-NEXT:    .cfi_offset cr2, 8
-; CHECK-BE-NEXT:    std r28, 112(r1) # 8-byte Folded Spill
 ; CHECK-BE-NEXT:    std r29, 120(r1) # 8-byte Folded Spill
 ; CHECK-BE-NEXT:    std r30, 128(r1) # 8-byte Folded Spill
 ; CHECK-BE-NEXT:    bl call_2
@@ -131,12 +126,11 @@ define dso_local void @P10_Spill_CR_LT() local_unnamed_addr {
 ; CHECK-BE-NEXT:    bc 12, 4*cr3+eq, .LBB0_11
 ; CHECK-BE-NEXT:  # %bb.6: # %bb32
 ; CHECK-BE-NEXT:    #
+; CHECK-BE-NEXT:    rlwinm r29, r29, 0, 24, 22
 ; CHECK-BE-NEXT:    andi. r3, r29, 2
-; CHECK-BE-NEXT:    rlwinm r28, r29, 0, 24, 22
 ; CHECK-BE-NEXT:    mcrf cr2, cr0
 ; CHECK-BE-NEXT:    bl call_4
 ; CHECK-BE-NEXT:    nop
-; CHECK-BE-NEXT:    mr r29, r28
 ; CHECK-BE-NEXT:    beq+ cr2, .LBB0_3
 ; CHECK-BE-NEXT:  # %bb.7: # %bb37
 ; CHECK-BE-NEXT:  .LBB0_8: # %bb22

llvm/test/CodeGen/PowerPC/swaps-le-1.ll

@@ -187,34 +187,34 @@ define void @foo() {
 ; CHECK-P9-NEXT:    .p2align 4
 ; CHECK-P9-NEXT:  .LBB0_1: # %vector.body
 ; CHECK-P9-NEXT:    #
-; CHECK-P9-NEXT:    lxv 2, -32(6)
-; CHECK-P9-NEXT:    lxv 3, -32(5)
-; CHECK-P9-NEXT:    lxv 4, -16(5)
-; CHECK-P9-NEXT:    vadduwm 2, 3, 2
+; CHECK-P9-NEXT:    lxv 2, -32(3)
 ; CHECK-P9-NEXT:    lxv 3, -32(4)
+; CHECK-P9-NEXT:    lxv 4, -16(4)
+; CHECK-P9-NEXT:    vadduwm 2, 3, 2
+; CHECK-P9-NEXT:    lxv 3, -32(5)
 ; CHECK-P9-NEXT:    vmuluwm 2, 2, 3
-; CHECK-P9-NEXT:    lxv 3, -16(6)
-; CHECK-P9-NEXT:    vadduwm 3, 4, 3
-; CHECK-P9-NEXT:    lxv 4, 0(5)
-; CHECK-P9-NEXT:    stxv 2, -32(3)
-; CHECK-P9-NEXT:    lxv 2, -16(4)
-; CHECK-P9-NEXT:    vmuluwm 2, 3, 2
-; CHECK-P9-NEXT:    lxv 3, 0(6)
+; CHECK-P9-NEXT:    lxv 3, -16(3)
 ; CHECK-P9-NEXT:    vadduwm 3, 4, 3
-; CHECK-P9-NEXT:    lxv 4, 16(5)
-; CHECK-P9-NEXT:    addi 5, 5, 64
-; CHECK-P9-NEXT:    stxv 2, -16(3)
-; CHECK-P9-NEXT:    lxv 2, 0(4)
+; CHECK-P9-NEXT:    lxv 4, 0(4)
+; CHECK-P9-NEXT:    stxv 2, -32(6)
+; CHECK-P9-NEXT:    lxv 2, -16(5)
 ; CHECK-P9-NEXT:    vmuluwm 2, 3, 2
-; CHECK-P9-NEXT:    lxv 3, 16(6)
-; CHECK-P9-NEXT:    addi 6, 6, 64
+; CHECK-P9-NEXT:    lxv 3, 0(3)
 ; CHECK-P9-NEXT:    vadduwm 3, 4, 3
-; CHECK-P9-NEXT:    stxv 2, 0(3)
-; CHECK-P9-NEXT:    lxv 2, 16(4)
+; CHECK-P9-NEXT:    lxv 4, 16(4)
 ; CHECK-P9-NEXT:    addi 4, 4, 64
+; CHECK-P9-NEXT:    stxv 2, -16(6)
+; CHECK-P9-NEXT:    lxv 2, 0(5)
 ; CHECK-P9-NEXT:    vmuluwm 2, 3, 2
-; CHECK-P9-NEXT:    stxv 2, 16(3)
+; CHECK-P9-NEXT:    lxv 3, 16(3)
 ; CHECK-P9-NEXT:    addi 3, 3, 64
+; CHECK-P9-NEXT:    vadduwm 3, 4, 3
+; CHECK-P9-NEXT:    stxv 2, 0(6)
+; CHECK-P9-NEXT:    lxv 2, 16(5)
+; CHECK-P9-NEXT:    addi 5, 5, 64
+; CHECK-P9-NEXT:    vmuluwm 2, 3, 2
+; CHECK-P9-NEXT:    stxv 2, 16(6)
+; CHECK-P9-NEXT:    addi 6, 6, 64
 ; CHECK-P9-NEXT:    bdnz .LBB0_1
 ; CHECK-P9-NEXT:  # %bb.2: # %for.end
 ; CHECK-P9-NEXT:    blr