[DAGCombiner] Extend visitAND to include EXTRACT_SUBVECTOR

[david-arm]

Summary

Eliminate an AND by redefining an anyext|sext|zext.

   (and (extract_subvector (anyext|sext|zext v) _) iN_mask)
=> (extract_subvector (zeroext_iN v))

[paulwalker-arm]

A couple of naming issues but otherwise looks good.

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

6356–6357

With the enhanced combine the original variable names are not strictly correct.

llvm/test/CodeGen/AArch64/extract-subvec-combine.ll

23

sext?

49

sext?

llvm/include/llvm/CodeGen/ISDOpcodes.h

 inline unsigned getUnorderedFlavor(CondCode Cond) {
   return ((int)Cond >> 3) & 3;
 }
 
 /// Return the operation corresponding to !(X op Y), where 'op' is a valid
 /// SetCC operation.
 CondCode getSetCCInverse(CondCode Operation, EVT Type);
 
+inline bool isExtOpcode(unsigned Opcode) {
+  return Opcode == ISD::ANY_EXTEND || Opcode == ISD::ZERO_EXTEND ||
+         Opcode == ISD::SIGN_EXTEND;
+}
+
 namespace GlobalISel {
 /// Return the operation corresponding to !(X op Y), where 'op' is a valid
 /// SetCC operation. The U bit of the condition code has different meanings
 /// between floating point and integer comparisons and LLT's don't provide
 /// this distinction. As such we need to be told whether the comparison is
 /// floating point or integer-like. Pointers should use integer-like
 /// comparisons.
 CondCode getSetCCInverse(CondCode Operation, bool isIntegerLike);

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

if (B && Constant.isAllOnes()) {
           CombineTo(Load, NewLoad.getValue(0), NewLoad.getValue(1));
         }
       }
 
       return SDValue(N, 0); // Return N so it doesn't get rechecked!
     }
   }
 
+  if (N0.getOpcode() == ISD::EXTRACT_SUBVECTOR && N0.hasOneUse() && N1C &&
+      ISD::isExtOpcode(N0.getOperand(0).getOpcode())) {
+    SDValue Ext = N0.getOperand(0);
+    EVT ExtVT = Ext->getValueType(0);
+    SDValue Extendee = Ext->getOperand(0);
+
+    unsigned ScalarWidth = Extendee.getValueType().getScalarSizeInBits();
+    if (N1C->getAPIntValue().isMask(ScalarWidth)) {
+      //    (and (extract_subvector (zext|anyext|sext v) _) iN_mask)
+      // => (extract_subvector (iN_zeroext v))
+      SDValue ZeroExtExtendee =
+          DAG.getNode(ISD::ZERO_EXTEND, SDLoc(N), ExtVT, Extendee);
+
+      return DAG.getNode(ISD::EXTRACT_SUBVECTOR, SDLoc(N), VT, ZeroExtExtendee,
+                         N0.getOperand(1));
+    }
+  }
+
   // fold (and (masked_gather x)) -> (zext_masked_gather x)
   if (auto *GN0 = dyn_cast<MaskedGatherSDNode>(N0)) {
     EVT MemVT = GN0->getMemoryVT();
     EVT ScalarVT = MemVT.getScalarType();
 
     if (SDValue(GN0, 0).hasOneUse() &&
         isConstantSplatVectorMaskForType(N1.getNode(), ScalarVT) &&
         TLI.isVectorLoadExtDesirable(SDValue(SDValue(GN0, 0)))) {

llvm/test/CodeGen/AArch64/extract-subvec-combine.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=aarch64-linux-gnu < %s | FileCheck %s --check-prefixes=CHECK
+
+define <2 x i32> @and_extract_zext_idx0(<4 x i16> %vec) nounwind {
+; CHECK-LABEL: and_extract_zext_idx0:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ushll v0.4s, v0.4h, #0
+; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-NEXT:    ret
+  %zext = zext <4 x i16> %vec to <4 x i32>
+  %extract = call <2 x i32> @llvm.vector.extract.v2i32.v4i32(<4 x i32> %zext, i64 0)
+  %and = and <2 x i32> %extract, <i32 65535, i32 65535>
+  ret <2 x i32> %and
+}
+
+define <4 x i16> @and_extract_sext_idx0(<8 x i8> %vec) nounwind {
+; CHECK-LABEL: and_extract_sext_idx0:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ushll v0.8h, v0.8b, #0
+; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-NEXT:    ret
+  %sext = sext <8 x i8> %vec to <8 x i16>
+  %extract = call <4 x i16> @llvm.vector.extract.v4i16.v8i16(<8 x i16> %sext, i64 0)
+  %and = and <4 x i16> %extract, <i16 255, i16 255, i16 255, i16 255>
+  ret <4 x i16> %and
+}
+
+define <2 x i32> @and_extract_zext_idx2(<4 x i16> %vec) nounwind {
+; CHECK-LABEL: and_extract_zext_idx2:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ushll v0.4s, v0.4h, #0
+; CHECK-NEXT:    ext v0.16b, v0.16b, v0.16b, #8
+; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-NEXT:    ret
+  %zext = zext <4 x i16> %vec to <4 x i32>
+  %extract = call <2 x i32> @llvm.vector.extract.v2i32.v4i32(<4 x i32> %zext, i64 2)
+  %and = and <2 x i32> %extract, <i32 65535, i32 65535>
+  ret <2 x i32> %and
+}
+
+define <4 x i16> @and_extract_sext_idx4(<8 x i8> %vec) nounwind {
+; CHECK-LABEL: and_extract_sext_idx4:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ushll v0.8h, v0.8b, #0
+; CHECK-NEXT:    ext v0.16b, v0.16b, v0.16b, #8
+; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-NEXT:    ret
+  %sext = sext <8 x i8> %vec to <8 x i16>
+  %extract = call <4 x i16> @llvm.vector.extract.v4i16.v8i16(<8 x i16> %sext, i64 4)
+  %and = and <4 x i16> %extract, <i16 255, i16 255, i16 255, i16 255>
+  ret <4 x i16> %and
+}
+
+declare <2 x i32> @llvm.vector.extract.v2i32.v4i32(<4 x i32>, i64)
+declare <4 x i16> @llvm.vector.extract.v4i16.v8i16(<8 x i16>, i64)

llvm/test/CodeGen/AArch64/vector-fcvt.ll

 ; CHECK-NEXT:    ret
   %1 = uitofp <16 x i8> %a to <16 x double>
   ret <16 x double> %1
 }
 
 define <8 x double> @uitofp_i16_double(<8 x i16> %a) {
 ; CHECK-LABEL: uitofp_i16_double:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    movi d1, #0x00ffff0000ffff
-; CHECK-NEXT:    ushll2 v2.4s, v0.8h, #0
+; CHECK-NEXT:    ushll2 v1.4s, v0.8h, #0
 ; CHECK-NEXT:    ushll v0.4s, v0.4h, #0
-; CHECK-NEXT:    and v3.8b, v2.8b, v1.8b
-; CHECK-NEXT:    and v4.8b, v0.8b, v1.8b
-; CHECK-NEXT:    ext v2.16b, v2.16b, v2.16b, #8
-; CHECK-NEXT:    ext v0.16b, v0.16b, v0.16b, #8
-; CHECK-NEXT:    ushll v3.2d, v3.2s, #0
-; CHECK-NEXT:    ushll v4.2d, v4.2s, #0
-; CHECK-NEXT:    and v2.8b, v2.8b, v1.8b
-; CHECK-NEXT:    and v0.8b, v0.8b, v1.8b
-; CHECK-NEXT:    ushll v5.2d, v2.2s, #0
-; CHECK-NEXT:    ucvtf v2.2d, v3.2d
-; CHECK-NEXT:    ushll v1.2d, v0.2s, #0
-; CHECK-NEXT:    ucvtf v0.2d, v4.2d
-; CHECK-NEXT:    ucvtf v1.2d, v1.2d
-; CHECK-NEXT:    ucvtf v3.2d, v5.2d
+; CHECK-NEXT:    ushll2 v2.2d, v1.4s, #0
+; CHECK-NEXT:    ushll2 v3.2d, v0.4s, #0
+; CHECK-NEXT:    ushll v4.2d, v1.2s, #0
+; CHECK-NEXT:    ushll v0.2d, v0.2s, #0
+; CHECK-NEXT:    ucvtf v1.2d, v3.2d
+; CHECK-NEXT:    ucvtf v0.2d, v0.2d
+; CHECK-NEXT:    ucvtf v3.2d, v2.2d
+; CHECK-NEXT:    ucvtf v2.2d, v4.2d
 ; CHECK-NEXT:    ret
   %1 = uitofp <8 x i16> %a to <8 x double>
   ret <8 x double> %1
 }
 
 define <8 x double> @uitofp_i32_double(<8 x i32> %a) {
 ; CHECK-LABEL: uitofp_i32_double:
 ; CHECK:       // %bb.0: