Commit 1a9fbf61 authored by Simon Pilgrim's avatar Simon Pilgrim
Browse files

[X86] combineLoad - reuse an existing VBROADCAST_LOAD constant for a smaller... 

[X86] combineLoad - reuse an existing VBROADCAST_LOAD constant for a smaller vector load of the same constant

Extends the existing code that performed something similar for SUBV_BROADCAST_LOAD, but this is just for cases where AVX2 targets loads full width 128-bit constant vectors but broadcasts the equivalent 256-bit constant vector

Fixes AVX2 case for Issue #70947
parent 9e618e5e

llvm/lib/Target/X86/X86ISelLowering.cpp

+33 −11
Original line number Diff line number Diff line
@@ -49785,26 +49785,48 @@ static SDValue combineLoad(SDNode *N, SelectionDAG &DAG, 
    }

  }













  // If we also broadcast this as a subvector to a wider type, then just extract













  // the lowest subvector.













  // If we also broadcast this to a wider type, then just extract the lowest













  // subvector.















  if (Ext == ISD::NON_EXTLOAD && Subtarget.hasAVX() && Ld->isSimple() &&















      (RegVT.is128BitVector() || RegVT.is256BitVector())) {















    SDValue Ptr = Ld->getBasePtr();















    SDValue Chain = Ld->getChain();













    for (SDNode *User : Ptr->uses()) {













      if (User != N && User->getOpcode() == X86ISD::SUBV_BROADCAST_LOAD &&













          cast<MemIntrinsicSDNode>(User)->getBasePtr() == Ptr &&













    for (SDNode *User : Chain->uses()) {













      if (User != N &&













          (User->getOpcode() == X86ISD::SUBV_BROADCAST_LOAD ||













           User->getOpcode() == X86ISD::VBROADCAST_LOAD) &&















          cast<MemIntrinsicSDNode>(User)->getChain() == Chain &&













          cast<MemIntrinsicSDNode>(User)->getMemoryVT().getSizeInBits() ==













              MemVT.getSizeInBits() &&















          !User->hasAnyUseOfValue(1) &&















          User->getValueSizeInBits(0).getFixedValue() >















              RegVT.getFixedSizeInBits()) {













        if (User->getOpcode() == X86ISD::SUBV_BROADCAST_LOAD &&













            cast<MemIntrinsicSDNode>(User)->getBasePtr() == Ptr &&













            cast<MemIntrinsicSDNode>(User)->getMemoryVT().getSizeInBits() ==













                MemVT.getSizeInBits()) {















          SDValue Extract = extractSubVector(SDValue(User, 0), 0, DAG, SDLoc(N),















                                             RegVT.getSizeInBits());















          Extract = DAG.getBitcast(RegVT, Extract);















          return DCI.CombineTo(N, Extract, SDValue(User, 1));















        }













        if (User->getOpcode() == X86ISD::VBROADCAST_LOAD &&













            getTargetConstantFromBasePtr(Ptr)) {













          // See if we are loading a constant that has also been broadcast.













          APInt Undefs, UserUndefs;













          SmallVector<APInt> Bits, UserBits;













          if (getTargetConstantBitsFromNode(SDValue(N, 0), 8, Undefs, Bits) &&













              getTargetConstantBitsFromNode(SDValue(User, 0), 8, UserUndefs,













                                            UserBits)) {













            UserUndefs = UserUndefs.trunc(Undefs.getBitWidth());













            UserBits.truncate(Bits.size());













            if (Bits == UserBits && UserUndefs.isSubsetOf(Undefs)) {













              SDValue Extract = extractSubVector(













                  SDValue(User, 0), 0, DAG, SDLoc(N), RegVT.getSizeInBits());













              Extract = DAG.getBitcast(RegVT, Extract);













              return DCI.CombineTo(N, Extract, SDValue(User, 1));













            }













          }













        }













      }















    }















  }







































llvm/test/CodeGen/X86/vec_fabs.ll



+6
−8






























Original line number
Diff line number
Diff line








@@ -332,10 +332,9 @@ define void @PR70947(ptr %src, ptr %dst) {













; X86-AVX2:       # %bb.0:















; X86-AVX2-NEXT:    movl {{[0-9]+}}(%esp), %eax















; X86-AVX2-NEXT:    movl {{[0-9]+}}(%esp), %ecx













; X86-AVX2-NEXT:    vmovups 32(%ecx), %xmm0













; X86-AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm1 = [NaN,NaN,NaN,NaN]













; X86-AVX2-NEXT:    vandps (%ecx), %ymm1, %ymm1













; X86-AVX2-NEXT:    vandps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0













; X86-AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm0 = [NaN,NaN,NaN,NaN]













; X86-AVX2-NEXT:    vandps (%ecx), %ymm0, %ymm1













; X86-AVX2-NEXT:    vandps 32(%ecx), %xmm0, %xmm0















; X86-AVX2-NEXT:    vmovups %ymm1, (%eax)















; X86-AVX2-NEXT:    vmovups %xmm0, 16(%eax)















; X86-AVX2-NEXT:    vzeroupper










@@ -378,10 +377,9 @@ define void @PR70947(ptr %src, ptr %dst) {













;















; X64-AVX2-LABEL: PR70947:















; X64-AVX2:       # %bb.0:













; X64-AVX2-NEXT:    vmovups 32(%rdi), %xmm0













; X64-AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm1 = [NaN,NaN,NaN,NaN]













; X64-AVX2-NEXT:    vandps (%rdi), %ymm1, %ymm1













; X64-AVX2-NEXT:    vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0













; X64-AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm0 = [NaN,NaN,NaN,NaN]













; X64-AVX2-NEXT:    vandps (%rdi), %ymm0, %ymm1













; X64-AVX2-NEXT:    vandps 32(%rdi), %xmm0, %xmm0















; X64-AVX2-NEXT:    vmovups %ymm1, (%rsi)















; X64-AVX2-NEXT:    vmovups %xmm0, 16(%rsi)















; X64-AVX2-NEXT:    vzeroupper