Index: llvm/lib/CodeGen/MachineSink.cpp =================================================================== --- llvm/lib/CodeGen/MachineSink.cpp +++ llvm/lib/CodeGen/MachineSink.cpp @@ -1279,10 +1279,14 @@ for (MCSubRegIterator S(DefReg, TRI, true); S.isValid(); ++S) SuccBB->removeLiveIn(*S); for (auto U : UsedOpsInCopy) { - Register Reg = MI->getOperand(U).getReg(); - if (!SuccBB->isLiveIn(Reg)) - SuccBB->addLiveIn(Reg); + Register SrcReg = MI->getOperand(U).getReg(); + LaneBitmask Mask; + for (MCSubRegIndexIterator S(SrcReg, TRI); S.isValid(); ++S) { + Mask |= TRI->getSubRegIndexLaneMask(S.getSubRegIndex()); + } + SuccBB->addLiveIn(SrcReg, Mask.any() ? Mask : LaneBitmask::getAll()); } + SuccBB->sortUniqueLiveIns(); } static bool hasRegisterDependency(MachineInstr *MI, Index: llvm/test/CodeGen/AArch64/bisect-post-ra-machine-sink.mir =================================================================== --- llvm/test/CodeGen/AArch64/bisect-post-ra-machine-sink.mir +++ llvm/test/CodeGen/AArch64/bisect-post-ra-machine-sink.mir @@ -9,7 +9,7 @@ # RUN-POSTRA: bb.0: # RUN-POSTRA-NOT: $w19 = COPY killed $w0 # RUN-POSTRA: bb.1: -# RUN-POSTRA: liveins: $w1, $w0 +# RUN-POSTRA: liveins: $w0, $w1 # RUN-POSTRA: renamable $w19 = COPY killed $w0 # Make sure the pass it not run. Index: llvm/test/CodeGen/AArch64/post-ra-machine-sink.mir =================================================================== --- llvm/test/CodeGen/AArch64/post-ra-machine-sink.mir +++ llvm/test/CodeGen/AArch64/post-ra-machine-sink.mir @@ -6,7 +6,7 @@ # CHECK-LABEL: bb.0: # CHECK-NOT: $w19 = COPY killed $w0 # CHECK-LABEL: bb.1: -# CHECK: liveins: $w1, $w0 +# CHECK: liveins: $w0, $w1 # CHECK: renamable $w19 = COPY killed $w0 name: sinkcopy1 @@ -35,7 +35,7 @@ # CHECK-LABEL: bb.0: # CHECK-NOT: renamable $w19 = COPY killed $w0 # CHECK-LABEL: bb.2: -# CHECK: liveins: $w1, $w0 +# CHECK: liveins: $w0, $w1 # CHECK: renamable $w19 = COPY killed $w0 name: sinkcopy2 tracksRegLiveness: true @@ -63,7 +63,7 @@ # CHECK-LABEL: bb.0: # CHECK-NOT: renamable $w19 = COPY killed $w0 # CHECK-LABEL: bb.1: -# CHECK: liveins: $w1, $w0 +# CHECK: liveins: $w0, $w1 # CHECK: renamable $w19 = COPY killed $w0 # CHECK: renamable $w20 = COPY killed $w1 name: sinkcopy3 @@ -89,7 +89,7 @@ # CHECK-NOT: renamable $w19 = COPY killed $w0 # CHECK-NOT: renamable $w20 = COPY killed $w1 # CHECK-LABEL: bb.1: -# CHECK: liveins: $w1, $w0 +# CHECK: liveins: $w0, $w1 # CHECK: renamable $w19 = COPY killed $w0 # CHECK-LABEL: bb.2: # CHECK: liveins: $w0, $w1 @@ -123,7 +123,7 @@ # CHECK-LABEL: bb.2: # CHECK: $w1 = ADDWrr $w1, $w0 # CHECK-LABEL: bb.3: -# CHECK: liveins: $w1, $w0 +# CHECK: liveins: $w0, $w1 # CHECK: renamable $w19 = COPY killed $w0 name: sinkcopy5 tracksRegLiveness: true @@ -155,7 +155,7 @@ # CHECK-NOT: renamable $w19 = COPY $w0 # CHECK-NOT: renamable $w20 = COPY $w0 # CHECK-LABEL: bb.2: -# CHECK: liveins: $w1, $w0 +# CHECK: liveins: $w0, $w1 # CHECK: renamable $w19 = COPY $w0 # CHECK: renamable $w20 = COPY $w19 name: sinkcopy6 @@ -213,7 +213,7 @@ # CHECK-LABEL: bb.2: # CHECK: $w1 = ADDWrr $w1, $w0, implicit $x0 # CHECK-LABEL: bb.3: -# CHECK: liveins: $x1, $w0 +# CHECK: liveins: $w0, $x1 # CHECK: renamable $w19 = COPY killed $w0, implicit-def $x19 name: sinkcopy8 tracksRegLiveness: true Index: llvm/test/CodeGen/AMDGPU/postra-machine-sink.mir =================================================================== --- /dev/null +++ llvm/test/CodeGen/AMDGPU/postra-machine-sink.mir @@ -0,0 +1,31 @@ +# RUN: llc -march=amdgcn -mcpu=gfx908 -run-pass=postra-machine-sink -verify-machineinstrs -o - %s | FileCheck %s + +# Don't sink copy that writes sub-register of another copy source register +# CHECK-LABEL: name: donotsinkcopy +# CHECK-LABEL: bb.0: +# CHECK: renamable $sgpr1 = COPY renamable $sgpr2 +# CHECK-LABEL: bb.1: +# CHECK: liveins: $sgpr0_sgpr1:0x00000003 +# CHECK: renamable $vgpr1_vgpr2 = COPY renamable $sgpr0_sgpr1 + +--- +name: donotsinkcopy +tracksRegLiveness: true +body: | + bb.0: + liveins: $sgpr0, $sgpr2, $vgpr3 + S_CMP_LT_U32 renamable $sgpr2, target-flags(amdgpu-gotprel32-lo) 2, implicit-def $scc + renamable $sgpr1 = COPY renamable $sgpr2 + renamable $vgpr1_vgpr2 = COPY renamable $sgpr0_sgpr1 + S_CBRANCH_SCC0 %bb.2, implicit $scc + + bb.1: + liveins: $vgpr2, $sgpr0_sgpr1:0x00000001 + S_BRANCH %bb.4 + + bb.2: + liveins: $sgpr1 + + bb.4: + S_ENDPGM 0 +...