[llvm] [AArch64] SimplifyDemandedBitsForTargetNode - add AArch64ISD::BICi handling (PR #76644)

[Sizov Nikita via llvm-commits]

https://github.com/snikitav updated https://github.com/llvm/llvm-project/pull/76644

>From afa5a78d50434f7ef1219df36811adb5a0e6bcc0 Mon Sep 17 00:00:00 2001
From: Sizov Nikita <s.nikita.v at gmail.com>
Date: Sun, 31 Dec 2023 04:33:21 +0300
Subject: [PATCH] Missing AArch64ISD::BICi handling

---
 .../lib/CodeGen/SelectionDAG/SelectionDAG.cpp | 15 ++++++----
 .../Target/AArch64/AArch64ISelLowering.cpp    | 30 +++++++++++++++++++
 .../AArch64/aarch64-known-bits-hadd.ll        |  4 ---
 3 files changed, 40 insertions(+), 9 deletions(-)

diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index f7ace79e8c51d4..937e6061bf01b4 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -3416,13 +3416,18 @@ KnownBits SelectionDAG::computeKnownBits(SDValue Op, const APInt &DemandedElts,
       Known = KnownBits::mulhs(Known, Known2);
     break;
   }
-  case ISD::AVGCEILU: {
+  case ISD::AVGFLOORU:
+  case ISD::AVGCEILU:
+  case ISD::AVGFLOORS:
+  case ISD::AVGCEILS: {
+    bool IsCeil = Opcode == ISD::AVGCEILU || Opcode == ISD::AVGCEILS;
+    bool IsSigned = Opcode == ISD::AVGFLOORS || Opcode == ISD::AVGCEILS;
     Known = computeKnownBits(Op.getOperand(0), DemandedElts, Depth + 1);
     Known2 = computeKnownBits(Op.getOperand(1), DemandedElts, Depth + 1);
-    Known = Known.zext(BitWidth + 1);
-    Known2 = Known2.zext(BitWidth + 1);
-    KnownBits One = KnownBits::makeConstant(APInt(1, 1));
-    Known = KnownBits::computeForAddCarry(Known, Known2, One);
+    Known = IsSigned ? Known.sext(BitWidth + 1) : Known.zext(BitWidth + 1);
+    Known2 = IsSigned ? Known2.sext(BitWidth + 1) : Known2.zext(BitWidth + 1);
+    KnownBits Carry = KnownBits::makeConstant(APInt(1, IsCeil ? 1 : 0));
+    Known = KnownBits::computeForAddCarry(Known, Known2, Carry);
     Known = Known.extractBits(BitWidth, 1);
     break;
   }
diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index 2290223a06f8ef..2a2f4e20efff02 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -24387,6 +24387,19 @@ SDValue AArch64TargetLowering::PerformDAGCombine(SDNode *N,
     if (auto R = foldOverflowCheck(N, DAG, /* IsAdd */ false))
       return R;
     return performFlagSettingCombine(N, DCI, AArch64ISD::SBC);
+  case AArch64ISD::BICi: {
+    KnownBits Known;
+    APInt DemandedBits =
+        APInt::getAllOnes(N->getValueType(0).getScalarSizeInBits());
+    APInt DemandedElts =
+        APInt::getAllOnes(N->getValueType(0).getVectorNumElements());
+    TargetLowering::TargetLoweringOpt TLO(DAG, !DCI.isBeforeLegalize(),
+                                          !DCI.isBeforeLegalizeOps());
+    if (DAG.getTargetLoweringInfo().SimplifyDemandedBits(
+            SDValue(N, 0), DemandedBits, DemandedElts, Known, TLO))
+      return TLO.New;
+    break;
+  }
   case ISD::XOR:
     return performXorCombine(N, DAG, DCI, Subtarget);
   case ISD::MUL:
@@ -27427,6 +27440,23 @@ bool AArch64TargetLowering::SimplifyDemandedBitsForTargetNode(
     // used - simplify to just Val.
     return TLO.CombineTo(Op, ShiftR->getOperand(0));
   }
+  case AArch64ISD::BICi: {
+    // Fold BICi if all destination bits already known to be zeroed
+    SDValue Op0 = Op.getOperand(0);
+    KnownBits KnownOp0 =
+        TLO.DAG.computeKnownBits(Op0, OriginalDemandedElts, Depth + 1);
+    // Op0 &= ~(ConstantOperandVal(1) << ConstantOperandVal(2))
+    uint64_t BitsToClear = Op->getConstantOperandVal(1)
+                           << Op->getConstantOperandVal(2);
+    APInt AlreadyZeroedBitsToClear = BitsToClear & KnownOp0.Zero;
+    if (APInt(Known.getBitWidth(), BitsToClear)
+            .isSubsetOf(AlreadyZeroedBitsToClear))
+      return TLO.CombineTo(Op, Op0);
+
+    Known &= KnownBits::makeConstant(APInt(Known.getBitWidth(), ~BitsToClear));
+
+    return false;
+  }
   case ISD::INTRINSIC_WO_CHAIN: {
     if (auto ElementSize = IsSVECntIntrinsic(Op)) {
       unsigned MaxSVEVectorSizeInBits = Subtarget->getMaxSVEVectorSizeInBits();
diff --git a/llvm/test/CodeGen/AArch64/aarch64-known-bits-hadd.ll b/llvm/test/CodeGen/AArch64/aarch64-known-bits-hadd.ll
index 017f382774892c..f36b8440fe4bfb 100644
--- a/llvm/test/CodeGen/AArch64/aarch64-known-bits-hadd.ll
+++ b/llvm/test/CodeGen/AArch64/aarch64-known-bits-hadd.ll
@@ -12,7 +12,6 @@ define <8 x i16> @haddu_zext(<8 x i8> %a0, <8 x i8> %a1) {
 ; CHECK-NEXT:    ushll v0.8h, v0.8b, #0
 ; CHECK-NEXT:    ushll v1.8h, v1.8b, #0
 ; CHECK-NEXT:    uhadd v0.8h, v0.8h, v1.8h
-; CHECK-NEXT:    bic v0.8h, #254, lsl #8
 ; CHECK-NEXT:    ret
   %x0 = zext <8 x i8> %a0 to <8 x i16>
   %x1 = zext <8 x i8> %a1 to <8 x i16>
@@ -27,7 +26,6 @@ define <8 x i16> @rhaddu_zext(<8 x i8> %a0, <8 x i8> %a1) {
 ; CHECK-NEXT:    ushll v0.8h, v0.8b, #0
 ; CHECK-NEXT:    ushll v1.8h, v1.8b, #0
 ; CHECK-NEXT:    urhadd v0.8h, v0.8h, v1.8h
-; CHECK-NEXT:    bic v0.8h, #254, lsl #8
 ; CHECK-NEXT:    ret
   %x0 = zext <8 x i8> %a0 to <8 x i16>
   %x1 = zext <8 x i8> %a1 to <8 x i16>
@@ -42,7 +40,6 @@ define <8 x i16> @hadds_zext(<8 x i8> %a0, <8 x i8> %a1) {
 ; CHECK-NEXT:    ushll v0.8h, v0.8b, #0
 ; CHECK-NEXT:    ushll v1.8h, v1.8b, #0
 ; CHECK-NEXT:    shadd v0.8h, v0.8h, v1.8h
-; CHECK-NEXT:    bic v0.8h, #254, lsl #8
 ; CHECK-NEXT:    ret
   %x0 = zext <8 x i8> %a0 to <8 x i16>
   %x1 = zext <8 x i8> %a1 to <8 x i16>
@@ -57,7 +54,6 @@ define <8 x i16> @shaddu_zext(<8 x i8> %a0, <8 x i8> %a1) {
 ; CHECK-NEXT:    ushll v0.8h, v0.8b, #0
 ; CHECK-NEXT:    ushll v1.8h, v1.8b, #0
 ; CHECK-NEXT:    srhadd v0.8h, v0.8h, v1.8h
-; CHECK-NEXT:    bic v0.8h, #254, lsl #8
 ; CHECK-NEXT:    ret
   %x0 = zext <8 x i8> %a0 to <8 x i16>
   %x1 = zext <8 x i8> %a1 to <8 x i16>