Index: lib/CodeGen/RegisterCoalescer.cpp =================================================================== --- lib/CodeGen/RegisterCoalescer.cpp +++ lib/CodeGen/RegisterCoalescer.cpp @@ -1396,11 +1396,29 @@ } else if (SrcLI.liveAt(Idx)) return false; - LLVM_DEBUG(dbgs() << "\tEliminating copy of value\n"); - - // Remove any DstReg segments starting at the instruction. + // If the undef copy defines a live-out value (i.e. an input to a PHI def), + // then replace it with an IMPLICIT_DEF. LiveInterval &DstLI = LIS->getInterval(DstReg); SlotIndex RegIndex = Idx.getRegSlot(); + LiveRange::Segment *Seg = DstLI.getSegmentContaining(RegIndex); + assert(Seg != nullptr && "No segment for defining instruction"); + if (VNInfo *V = DstLI.getVNInfoAt(Seg->end)) { + if (V->isPHIDef()) { + CopyMI->setDesc(TII->get(TargetOpcode::IMPLICIT_DEF)); + for (unsigned i = CopyMI->getNumOperands(); i != 0; --i) { + MachineOperand &MO = CopyMI->getOperand(i-1); + if (MO.isReg() && MO.isUse()) + CopyMI->RemoveOperand(i-1); + } + LLVM_DEBUG(dbgs() << "\tReplaced copy of value with an " + "implicit def\n"); + return false; + } + } + + // Remove any DstReg segments starting at the instruction. + LLVM_DEBUG(dbgs() << "\tEliminating copy of value\n"); + // Remove value or merge with previous one in case of a subregister def. if (VNInfo *PrevVNI = DstLI.getVNInfoAt(Idx)) { VNInfo *VNI = DstLI.getVNInfoAt(RegIndex); @@ -2078,6 +2096,13 @@ /// True once Pruned above has been computed. bool PrunedComputed = false; + /// True if this value is determined to be identical to OtherVNI + /// (in valuesIdentical). This is used with CR_Erase where the erased + /// copy is redundant, i.e. the source value is already the same as + /// the destination. In such cases the subranges need to be updated + /// properly. See comment at pruneSubRegValues for more info. + bool Identical = false; + Val() = default; bool isAnalyzed() const { return WriteLanes.any(); } @@ -2092,7 +2117,8 @@ LaneBitmask computeWriteLanes(const MachineInstr *DefMI, bool &Redef) const; /// Find the ultimate value that VNI was copied from. - std::pair followCopyChain(const VNInfo *VNI) const; + std::pair followCopyChain(const VNInfo *VNI, + unsigned OtherReg) const; bool valuesIdentical(VNInfo *Val0, VNInfo *Val1, const JoinVals &Other) const; @@ -2211,18 +2237,18 @@ } std::pair JoinVals::followCopyChain( - const VNInfo *VNI) const { - unsigned Reg = this->Reg; + const VNInfo *VNI, unsigned OtherReg) const { + unsigned TrackReg = Reg; while (!VNI->isPHIDef()) { SlotIndex Def = VNI->def; MachineInstr *MI = Indexes->getInstructionFromIndex(Def); assert(MI && "No defining instruction"); if (!MI->isFullCopy()) - return std::make_pair(VNI, Reg); + return std::make_pair(VNI, TrackReg); unsigned SrcReg = MI->getOperand(1).getReg(); if (!TargetRegisterInfo::isVirtualRegister(SrcReg)) - return std::make_pair(VNI, Reg); + return std::make_pair(VNI, TrackReg); const LiveInterval &LI = LIS->getInterval(SrcReg); const VNInfo *ValueIn; @@ -2243,25 +2269,41 @@ break; } } - if (ValueIn == nullptr) + if (ValueIn == nullptr) { + // Reaching undefined value is legitimate in cases where it comes from + // one of the considered registers (Reg or OtherReg). + // + // 1 undef %0.sub1 = ... ;; %0.sub0 == undef + // 2 %1 = COPY %0 ;; %1 is defined here. + // 3 %0 = COPY %1 ;; Now %0.sub0 has a definition! + // 4 ... ;; Here we should indicate that %0 at 3 is + // ;; the same as %0 at 2 (%0 at 3 == %1 at 2). + if (SrcReg == Reg || SrcReg == OtherReg) + return std::make_pair(nullptr, SrcReg); break; + } VNI = ValueIn; - Reg = SrcReg; + TrackReg = SrcReg; } - return std::make_pair(VNI, Reg); + return std::make_pair(VNI, TrackReg); } bool JoinVals::valuesIdentical(VNInfo *Value0, VNInfo *Value1, const JoinVals &Other) const { const VNInfo *Orig0; unsigned Reg0; - std::tie(Orig0, Reg0) = followCopyChain(Value0); - if (Orig0 == Value1) + std::tie(Orig0, Reg0) = followCopyChain(Value0, Other.Reg); + if (Orig0 == Value1 && Reg0 == Other.Reg) return true; const VNInfo *Orig1; unsigned Reg1; - std::tie(Orig1, Reg1) = Other.followCopyChain(Value1); + std::tie(Orig1, Reg1) = Other.followCopyChain(Value1, Reg); + // If both values are undefined, and the source registers are the same + // register, the values are identical. Filter out cases where only one + // value is defined. + if (Orig0 == nullptr || Orig1 == nullptr) + return Orig0 == Orig1 && Reg0 == Reg1; // The values are equal if they are defined at the same place and use the // same register. Note that we cannot compare VNInfos directly as some of @@ -2438,8 +2480,10 @@ // %this = COPY %ext <-- Erase this copy // if (DefMI->isFullCopy() && !CP.isPartial() - && valuesIdentical(VNI, V.OtherVNI, Other)) + && valuesIdentical(VNI, V.OtherVNI, Other)) { + V.Identical = true; return CR_Erase; + } // If the lanes written by this instruction were all undef in OtherVNI, it is // still safe to join the live ranges. This can't be done with a simple value @@ -2743,19 +2787,62 @@ } } +/// Consider the following situation when coalescing the copy between +/// %31 and %45 at 800. (The vertical lines represent live range segments.) +/// +/// Main range Subrange 0004 (sub2) +/// %31 %45 %31 %45 +/// 544 %45 = COPY %28 + + +/// | v1 | v1 +/// 560B bb.1: + + +/// 624 = %45.sub2 | v2 | v2 +/// 800 %31 = COPY %45 + + + + +/// | v0 | v0 +/// 816 %31.sub1 = ... + | +/// 880 %30 = COPY %31 | v1 + +/// 928 %45 = COPY %30 | + + +/// | | v0 | v0 <--+ +/// 992B ; backedge -> bb.1 | + + | +/// 1040 = %31.sub0 + | +/// This value must remain +/// live-out! +/// +/// Assuming that %31 is coalesced into %45, the copy at 928 becomes +/// redundant, since it copies the value from %45 back into it. The +/// conflict resolution for the main range determines that %45.v0 is +/// to be erased, which is ok since %31.v1 is identical to it. +/// The problem happens with the subrange for sub2: it has to be live +/// on exit from the block, but since 928 was actually a point of +/// definition of %45.sub2, %45.sub2 was not live immediately prior +/// to that definition. As a result, when 928 was erased, the value v0 +/// for %45.sub2 was pruned in pruneSubRegValues. Consequently, an +/// IMPLICIT_DEF was inserted as a "backedge" definition for %45.sub2, +/// providing an incorrect value to the use at 624. +/// +/// Since the main-range values %31.v1 and %45.v0 were proved to be +/// identical, the corresponding values in subranges must also be the +/// same. A redundant copy is removed because it's not needed, and not +/// because it copied an undefined value, so any liveness that originated +/// from that copy cannot disappear. When pruning a value that started +/// at the removed copy, the corresponding identical value must be +/// extended to replace it. void JoinVals::pruneSubRegValues(LiveInterval &LI, LaneBitmask &ShrinkMask) { // Look for values being erased. bool DidPrune = false; for (unsigned i = 0, e = LR.getNumValNums(); i != e; ++i) { + Val &V = Vals[i]; // We should trigger in all cases in which eraseInstrs() does something. // match what eraseInstrs() is doing, print a message so - if (Vals[i].Resolution != CR_Erase && - (Vals[i].Resolution != CR_Keep || !Vals[i].ErasableImplicitDef || - !Vals[i].Pruned)) + if (V.Resolution != CR_Erase && + (V.Resolution != CR_Keep || !V.ErasableImplicitDef || !V.Pruned)) continue; // Check subranges at the point where the copy will be removed. SlotIndex Def = LR.getValNumInfo(i)->def; + SlotIndex OtherDef; + if (V.Identical) + OtherDef = V.OtherVNI->def; + // Print message so mismatches with eraseInstrs() can be diagnosed. LLVM_DEBUG(dbgs() << "\t\tExpecting instruction removal at " << Def << '\n'); @@ -2768,10 +2855,24 @@ if (ValueOut != nullptr && Q.valueIn() == nullptr) { LLVM_DEBUG(dbgs() << "\t\tPrune sublane " << PrintLaneMask(S.LaneMask) << " at " << Def << "\n"); - LIS->pruneValue(S, Def, nullptr); + SmallVector EndPoints; + LIS->pruneValue(S, Def, &EndPoints); DidPrune = true; // Mark value number as unused. ValueOut->markUnused(); + + if (V.Identical && S.Query(OtherDef).valueOut()) { + // If V is identical to V.OtherVNI (and S was live at OtherDef), + // then we can't simply prune V from S. V needs to be replaced + // with V.OtherVNI. + LIS->extendToIndices(S, EndPoints); +#ifndef NDEBUG + // Make sure that the value at Def is really V.OtherVNI. + LiveRange::iterator Id = S.find(OtherDef); + LiveRange::iterator T = S.find(Def); + assert(Id != S.end() && T != S.end() && T->valno == Id->valno); +#endif + } continue; } // If a subrange ends at the copy, then a value was copied but only Index: test/CodeGen/AMDGPU/coalescer-extend-pruned-subrange.mir =================================================================== --- /dev/null +++ test/CodeGen/AMDGPU/coalescer-extend-pruned-subrange.mir @@ -0,0 +1,128 @@ +# RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx803 -run-pass=simple-register-coalescing -o - %s | FileCheck -check-prefix=GCN %s + +# GCN: {{^body}} + +--- +name: foo +tracksRegLiveness: true +body: | + bb.0: + successors: %bb.2 + + %0:sreg_32_xm0 = S_MOV_B32 1 + %1:vgpr_32 = COPY %0 + INLINEASM &"; %1", 1, 327690, def %1, 2147483657, %1(tied-def 3) + %2:sreg_64 = V_CMP_NE_U32_e64 0, %1, implicit $exec + undef %3.sub0:sreg_128 = COPY %0 + %3.sub1:sreg_128 = COPY %0 + %3.sub2:sreg_128 = COPY %0 + %4:sreg_128 = COPY %3 + %5:vgpr_32 = V_MOV_B32_e32 -64, implicit $exec + %6:vreg_128 = COPY %4 + %7:sreg_32_xm0 = S_AND_B32 target-flags(amdgpu-gotprel) 1, %2.sub0, implicit-def dead $scc + %8:sreg_32_xm0 = S_MOV_B32 0 + %9:vgpr_32 = COPY %5 + %10:vreg_128 = COPY %6 + S_BRANCH %bb.2 + + bb.1: + %11:vgpr_32 = V_OR_B32_e32 %12.sub0, %12.sub1, implicit $exec + %13:vgpr_32 = V_OR_B32_e32 %11, %12.sub2, implicit $exec + %14:vgpr_32 = V_AND_B32_e32 1, %13, implicit $exec + %15:sreg_64_xexec = V_CMP_EQ_U32_e64 0, %14, implicit $exec + %16:vgpr_32 = V_CNDMASK_B32_e64 0, 1, %15, implicit $exec + BUFFER_STORE_DWORD_OFFEN_exact %16, undef %17:vgpr_32, undef %18:sreg_128, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4 into constant-pool, align 1, addrspace 4) + S_ENDPGM + + bb.2: + successors: %bb.3, %bb.4 + + %19:sreg_64 = V_CMP_EQ_U32_e64 1, %7, implicit $exec + %20:sreg_64 = COPY $exec, implicit-def $exec + %21:sreg_64 = S_AND_B64 %20, %19, implicit-def dead $scc + $exec = S_MOV_B64_term %21 + SI_MASK_BRANCH %bb.4, implicit $exec + S_BRANCH %bb.3 + + bb.3: + successors: %bb.4 + + undef %22.sub0:sreg_128 = COPY %8 + %22.sub1:sreg_128 = COPY %8 + %22.sub2:sreg_128 = COPY %8 + %23:sreg_128 = COPY %22 + %24:vreg_128 = COPY %23 + %10:vreg_128 = COPY %24 + + bb.4: + successors: %bb.5 + + $exec = S_OR_B64 $exec, %20, implicit-def $scc + + bb.5: + successors: %bb.7, %bb.6 + + S_CBRANCH_SCC0 %bb.7, implicit undef $scc + + bb.6: + successors: %bb.9 + + %12:vreg_128 = COPY %10 + S_BRANCH %bb.9 + + bb.7: + successors: %bb.8, %bb.10 + + %25:vgpr_32 = V_AND_B32_e32 target-flags(amdgpu-gotprel32-hi) 1, %10.sub2, implicit $exec + %26:sreg_64 = V_CMP_EQ_U32_e64 1, %25, implicit $exec + %27:vgpr_32 = V_MOV_B32_e32 0, implicit $exec + %28:vreg_1 = COPY %27 + %29:sreg_64 = COPY $exec, implicit-def $exec + %30:sreg_64 = S_AND_B64 %29, %26, implicit-def dead $scc + $exec = S_MOV_B64_term %30 + SI_MASK_BRANCH %bb.10, implicit $exec + S_BRANCH %bb.8 + + bb.8: + successors: %bb.10 + + %31:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN undef %32:vgpr_32, undef %33:sreg_128, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4 from constant-pool, align 1, addrspace 4) + %34:sreg_64_xexec = V_CMP_NE_U32_e64 0, %31, implicit $exec + %35:vgpr_32 = V_CNDMASK_B32_e64 0, -1, %34, implicit $exec + %28:vreg_1 = COPY %35 + S_BRANCH %bb.10 + + bb.9: + successors: %bb.11 + + S_BRANCH %bb.11 + + bb.10: + successors: %bb.9 + + $exec = S_OR_B64 $exec, %29, implicit-def $scc + %36:vreg_1 = COPY %28 + %37:sreg_64_xexec = V_CMP_NE_U32_e64 0, %36, implicit $exec + %38:vgpr_32 = V_CNDMASK_B32_e64 0, 1, %37, implicit $exec + %39:vgpr_32 = V_MOV_B32_e32 0, implicit $exec + undef %40.sub0:vreg_128 = COPY %39 + %40.sub1:vreg_128 = COPY %39 + %40.sub2:vreg_128 = COPY %38 + %41:vreg_128 = COPY %40 + %12:vreg_128 = COPY %41 + S_BRANCH %bb.9 + + bb.11: + successors: %bb.2, %bb.1 + + %42:vgpr_32 = V_ADD_I32_e32 32, %9, implicit-def dead $vcc, implicit $exec + V_CMP_EQ_U32_e32 0, %42, implicit-def $vcc, implicit $exec + %43:vgpr_32 = COPY %42 + $vcc = S_AND_B64 $exec, killed $vcc, implicit-def dead $scc + %44:vreg_128 = COPY %12 + %9:vgpr_32 = COPY %43 + %10:vreg_128 = COPY %44 + S_CBRANCH_VCCNZ %bb.1, implicit killed $vcc + S_BRANCH %bb.2 + +... Index: test/CodeGen/AMDGPU/coalescer-identical-values-undef.mir =================================================================== --- /dev/null +++ test/CodeGen/AMDGPU/coalescer-identical-values-undef.mir @@ -0,0 +1,29 @@ +# RUN: llc -mtriple=amdgcn--amdpal -run-pass=simple-register-coalescing -o - %s | FileCheck %s + +# Check that this doesn't crash. Check for some legitimate output. +# CHECK: S_CBRANCH_SCC1 + +--- +name: fred +tracksRegLiveness: true +body: | + bb.0: + successors: %bb.1, %bb.2 + liveins: $sgpr4 + undef %0.sub2:sreg_128 = COPY $sgpr4 + %3 = IMPLICIT_DEF + S_CBRANCH_SCC1 %bb.2, implicit undef $scc + + bb.1: + successors: %bb.2 + %0.sub0:sreg_128 = COPY %0.sub2 + %0.sub1:sreg_128 = COPY %0.sub2 + %1:sreg_128 = COPY %0 + %2:sreg_128 = COPY %0 + %0:sreg_128 = COPY %2 + %3:sreg_128 = COPY %1 + + bb.2: + $sgpr1 = COPY %3 + $sgpr2 = COPY %0.sub2 +... Index: test/CodeGen/AMDGPU/coalescing-with-subregs-in-loop-bug.mir =================================================================== --- /dev/null +++ test/CodeGen/AMDGPU/coalescing-with-subregs-in-loop-bug.mir @@ -0,0 +1,98 @@ +# RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx803 -run-pass=simple-register-coalescing,rename-independent-subregs %s -o - | FileCheck -check-prefix=GCN %s + +# This test is for a bug where the following happens: +# +# Inside the loop, %29.sub2 is used in a V_LSHLREV whose result is then used +# in an LDS read. %29 is a 128 bit value that is linked by copies to +# %45 (from phi elimination), %28 (the value in the loop pre-header), +# %31 (defined and subreg-modified in the loop, and used after the loop) +# and %30: +# +# %45:vreg_128 = COPY killed %28 +# bb.1: +# %29:vreg_128 = COPY killed %45 +# %39:vgpr_32 = V_LSHLREV_B32_e32 2, %29.sub2, implicit $exec +# %31:vreg_128 = COPY killed %29 +# %31.sub1:vreg_128 = COPY %34 +# %30:vreg_128 = COPY %31 +# %45:vreg_128 = COPY killed %30 +# S_CBRANCH_EXECNZ %bb.39, implicit $exec +# S_BRANCH %bb.40 +# bb.2: +# undef %32.sub0:vreg_128 = COPY killed %31.sub0 +# +# So this coalesces together into a single 128 bit value whose sub1 is modified +# in the loop, but the sub2 used in the V_LSHLREV is not modified in the loop. +# +# The bug is that the coalesced value has a L00000004 subrange (for sub2) that +# says that it is not live up to the end of the loop block. The symptom is that +# Rename Independent Subregs separates sub2 into its own register, and it is +# not live round the loop, so that pass adds an IMPLICIT_DEF for it just before +# the loop backedge. + +# GCN: bb.1: +# GCN: V_LSHLREV_B32_e32 2, [[val:%[0-9][0-9]*]].sub2 +# GCN-NOT: [[val]]:vreg_128 = IMPLICIT_DEF + +--- +name: _amdgpu_cs_main +tracksRegLiveness: true +body: | + bb.0: + successors: %bb.1 + + %3:sgpr_32 = S_MOV_B32 0 + undef %19.sub1:vreg_128 = COPY undef %3 + %4:sgpr_32 = S_MOV_B32 1 + %5:sgpr_32 = S_MOV_B32 2 + %11:sreg_32_xm0 = S_MOV_B32 255 + undef %28.sub0:vreg_128 = COPY killed %3 + %28.sub1:vreg_128 = COPY killed %4 + %28.sub2:vreg_128 = COPY killed %11 + %28.sub3:vreg_128 = COPY killed %5 + %2:sreg_64 = S_MOV_B64 0 + %34:sreg_32 = S_MOV_B32 7 + %37:vreg_128 = COPY undef %42:vreg_128 + %43:sreg_64 = COPY killed %2 + %44:vreg_128 = COPY killed %37 + %45:vreg_128 = COPY killed %28 + + bb.1: + successors: %bb.1, %bb.2 + + %29:vreg_128 = COPY killed %45 + %36:vreg_128 = COPY killed %44 + %0:sreg_64 = COPY killed %43 + %39:vgpr_32 = V_LSHLREV_B32_e32 2, %29.sub2, implicit $exec + %41:vgpr_32 = V_ADD_I32_e32 1152, %39, implicit-def dead $vcc, implicit $exec + $m0 = S_MOV_B32 -1 + %12:vreg_64 = DS_READ2_B32 killed %41, 0, 1, 0, implicit $m0, implicit $exec + %13:vreg_64 = DS_READ2_B32 %39, -112, -111, 0, implicit $m0, implicit $exec + %14:vreg_64 = DS_READ2_B32 %39, 0, 1, 0, implicit $m0, implicit $exec + %40:vgpr_32 = V_ADD_I32_e32 1160, %39, implicit-def dead $vcc, implicit $exec + %15:vreg_64 = DS_READ2_B32 killed %40, 0, 1, 0, implicit $m0, implicit $exec + %16:vreg_64 = DS_READ2_B32 %39, -110, -109, 0, implicit $m0, implicit $exec + %17:vreg_64 = DS_READ2_B32 %39, 2, 3, 0, implicit $m0, implicit $exec + undef %35.sub1:vreg_128 = COPY undef %34 + %31:vreg_128 = COPY killed %29 + %31.sub1:vreg_128 = COPY %34 + %38:vgpr_32 = V_ADD_I32_e32 1, %36.sub0, implicit-def dead $vcc, implicit $exec + %18:sreg_64 = V_CMP_LT_I32_e64 5, %38, implicit $exec + %1:sreg_64 = S_OR_B64 killed %18, killed %0, implicit-def $scc + %30:vreg_128 = COPY %31 + %43:sreg_64 = COPY %1 + %44:vreg_128 = COPY %35 + %45:vreg_128 = COPY killed %30 + $exec = S_ANDN2_B64_term $exec, %1 + S_CBRANCH_EXECNZ %bb.1, implicit $exec + S_BRANCH %bb.2 + + bb.2: + $exec = S_OR_B64 $exec, killed %1, implicit-def $scc + %33:vgpr_32 = V_MOV_B32_e32 0, implicit $exec + undef %32.sub0:vreg_128 = COPY killed %31.sub0 + %32.sub2:vreg_128 = COPY %33 + $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %32:vreg_128 + S_ENDPGM + +...