| # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 4 |
| # RUN: llc -global-isel -mtriple=amdgcn-mesa-amdpal -mcpu=gfx1010 -run-pass=amdgpu-global-isel-divergence-lowering %s -o - | FileCheck -check-prefix=GFX10 %s |
| |
| --- |
| name: divergent_i1_phi_used_outside_loop |
| legalized: true |
| tracksRegLiveness: true |
| body: | |
| ; GFX10-LABEL: name: divergent_i1_phi_used_outside_loop |
| ; GFX10: bb.0: |
| ; GFX10-NEXT: successors: %bb.1(0x80000000) |
| ; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 |
| ; GFX10-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 |
| ; GFX10-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2 |
| ; GFX10-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3 |
| ; GFX10-NEXT: [[MV:%[0-9]+]]:_(p0) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32) |
| ; GFX10-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0 |
| ; GFX10-NEXT: [[C1:%[0-9]+]]:_(s32) = G_FCONSTANT float 1.000000e+00 |
| ; GFX10-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ogt), [[COPY1]](s32), [[C1]] |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.1: |
| ; GFX10-NEXT: successors: %bb.2(0x04000000), %bb.1(0x7c000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[PHI:%[0-9]+]]:_(s32) = G_PHI %9(s32), %bb.1, [[C]](s32), %bb.0 |
| ; GFX10-NEXT: [[PHI1:%[0-9]+]]:_(s32) = G_PHI [[C]](s32), %bb.0, %11(s32), %bb.1 |
| ; GFX10-NEXT: [[PHI2:%[0-9]+]]:_(s1) = G_PHI [[FCMP]](s1), %bb.0, %13(s1), %bb.1 |
| ; GFX10-NEXT: [[C2:%[0-9]+]]:_(s1) = G_CONSTANT i1 true |
| ; GFX10-NEXT: [[XOR:%[0-9]+]]:_(s1) = G_XOR [[PHI2]], [[C2]] |
| ; GFX10-NEXT: [[UITOFP:%[0-9]+]]:_(s32) = G_UITOFP [[PHI1]](s32) |
| ; GFX10-NEXT: [[FCMP1:%[0-9]+]]:_(s1) = G_FCMP floatpred(ogt), [[UITOFP]](s32), [[COPY]] |
| ; GFX10-NEXT: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 1 |
| ; GFX10-NEXT: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[PHI1]], [[C3]] |
| ; GFX10-NEXT: [[INTRINSIC_CONVERGENT:%[0-9]+]]:sreg_32_xm0_xexec(s32) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.if.break), [[FCMP1]](s1), [[PHI]](s32) |
| ; GFX10-NEXT: SI_LOOP [[INTRINSIC_CONVERGENT]](s32), %bb.1, implicit-def $exec, implicit-def $scc, implicit $exec |
| ; GFX10-NEXT: G_BR %bb.2 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.2: |
| ; GFX10-NEXT: [[PHI3:%[0-9]+]]:_(s1) = G_PHI [[PHI2]](s1), %bb.1 |
| ; GFX10-NEXT: [[PHI4:%[0-9]+]]:_(s32) = G_PHI [[INTRINSIC_CONVERGENT]](s32), %bb.1 |
| ; GFX10-NEXT: G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), [[PHI4]](s32) |
| ; GFX10-NEXT: [[C4:%[0-9]+]]:_(s32) = G_FCONSTANT float 0.000000e+00 |
| ; GFX10-NEXT: [[C5:%[0-9]+]]:_(s32) = G_FCONSTANT float 1.000000e+00 |
| ; GFX10-NEXT: [[SELECT:%[0-9]+]]:_(s32) = G_SELECT [[PHI3]](s1), [[C5]], [[C4]] |
| ; GFX10-NEXT: G_STORE [[SELECT]](s32), [[MV]](p0) :: (store (s32)) |
| ; GFX10-NEXT: SI_RETURN |
| bb.0: |
| successors: %bb.1(0x80000000) |
| liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3 |
| |
| %0:_(s32) = COPY $vgpr0 |
| %1:_(s32) = COPY $vgpr1 |
| %2:_(s32) = COPY $vgpr2 |
| %3:_(s32) = COPY $vgpr3 |
| %4:_(p0) = G_MERGE_VALUES %2(s32), %3(s32) |
| %5:_(s32) = G_CONSTANT i32 0 |
| %6:_(s32) = G_FCONSTANT float 1.000000e+00 |
| %7:_(s1) = G_FCMP floatpred(ogt), %1(s32), %6 |
| |
| bb.1: |
| successors: %bb.2(0x04000000), %bb.1(0x7c000000) |
| |
| %8:_(s32) = G_PHI %9(s32), %bb.1, %5(s32), %bb.0 |
| %10:_(s32) = G_PHI %5(s32), %bb.0, %11(s32), %bb.1 |
| %12:_(s1) = G_PHI %7(s1), %bb.0, %13(s1), %bb.1 |
| %14:_(s1) = G_CONSTANT i1 true |
| %13:_(s1) = G_XOR %12, %14 |
| %15:_(s32) = G_UITOFP %10(s32) |
| %16:_(s1) = G_FCMP floatpred(ogt), %15(s32), %0 |
| %17:_(s32) = G_CONSTANT i32 1 |
| %11:_(s32) = G_ADD %10, %17 |
| %9:sreg_32_xm0_xexec(s32) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.if.break), %16(s1), %8(s32) |
| SI_LOOP %9(s32), %bb.1, implicit-def $exec, implicit-def $scc, implicit $exec |
| G_BR %bb.2 |
| |
| bb.2: |
| %18:_(s1) = G_PHI %12(s1), %bb.1 |
| %19:_(s32) = G_PHI %9(s32), %bb.1 |
| G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), %19(s32) |
| %20:_(s32) = G_FCONSTANT float 0.000000e+00 |
| %21:_(s32) = G_FCONSTANT float 1.000000e+00 |
| %22:_(s32) = G_SELECT %18(s1), %21, %20 |
| G_STORE %22(s32), %4(p0) :: (store (s32)) |
| SI_RETURN |
| ... |
| |
| --- |
| name: divergent_i1_phi_used_outside_loop_larger_loop_body |
| legalized: true |
| tracksRegLiveness: true |
| body: | |
| ; GFX10-LABEL: name: divergent_i1_phi_used_outside_loop_larger_loop_body |
| ; GFX10: bb.0: |
| ; GFX10-NEXT: successors: %bb.1(0x80000000) |
| ; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr1 |
| ; GFX10-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr2 |
| ; GFX10-NEXT: [[MV:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32) |
| ; GFX10-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr3 |
| ; GFX10-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr4 |
| ; GFX10-NEXT: [[MV1:%[0-9]+]]:_(p0) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32) |
| ; GFX10-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 -1 |
| ; GFX10-NEXT: [[C1:%[0-9]+]]:_(s1) = G_CONSTANT i1 true |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.1: |
| ; GFX10-NEXT: successors: %bb.2(0x40000000), %bb.3(0x40000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[PHI:%[0-9]+]]:_(s32) = G_PHI [[C]](s32), %bb.0, %9(s32), %bb.3 |
| ; GFX10-NEXT: [[PHI1:%[0-9]+]]:_(p1) = G_PHI [[MV]](p1), %bb.0, %11(p1), %bb.3 |
| ; GFX10-NEXT: [[PHI2:%[0-9]+]]:sreg_32_xm0_xexec(s1) = G_PHI [[C1]](s1), %bb.0, %13(s1), %bb.3 |
| ; GFX10-NEXT: [[SI_IF:%[0-9]+]]:sreg_32_xm0_xexec(s32) = SI_IF [[PHI2]](s1), %bb.3, implicit-def $exec, implicit-def $scc, implicit $exec |
| ; GFX10-NEXT: G_BR %bb.2 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.2: |
| ; GFX10-NEXT: successors: %bb.3(0x80000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[PHI1]](p1) :: (load (s32), addrspace 1) |
| ; GFX10-NEXT: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 0 |
| ; GFX10-NEXT: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(eq), [[LOAD]](s32), [[C2]] |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.3: |
| ; GFX10-NEXT: successors: %bb.4(0x04000000), %bb.1(0x7c000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[PHI3:%[0-9]+]]:_(s1) = G_PHI [[ICMP]](s1), %bb.2, [[PHI2]](s1), %bb.1 |
| ; GFX10-NEXT: G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), [[SI_IF]](s32) |
| ; GFX10-NEXT: [[C3:%[0-9]+]]:_(s64) = G_CONSTANT i64 4 |
| ; GFX10-NEXT: [[PTR_ADD:%[0-9]+]]:_(p1) = G_PTR_ADD [[PHI1]], [[C3]](s64) |
| ; GFX10-NEXT: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 1 |
| ; GFX10-NEXT: [[ADD:%[0-9]+]]:_(s32) = nsw G_ADD [[PHI]], [[C4]] |
| ; GFX10-NEXT: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 10 |
| ; GFX10-NEXT: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(sge), [[ADD]](s32), [[C5]] |
| ; GFX10-NEXT: G_BRCOND [[ICMP1]](s1), %bb.1 |
| ; GFX10-NEXT: G_BR %bb.4 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.4: |
| ; GFX10-NEXT: [[PHI4:%[0-9]+]]:_(s1) = G_PHI [[PHI2]](s1), %bb.3 |
| ; GFX10-NEXT: [[C6:%[0-9]+]]:_(s32) = G_FCONSTANT float 0.000000e+00 |
| ; GFX10-NEXT: [[C7:%[0-9]+]]:_(s32) = G_FCONSTANT float 1.000000e+00 |
| ; GFX10-NEXT: [[SELECT:%[0-9]+]]:_(s32) = G_SELECT [[PHI4]](s1), [[C7]], [[C6]] |
| ; GFX10-NEXT: G_STORE [[SELECT]](s32), [[MV1]](p0) :: (store (s32)) |
| ; GFX10-NEXT: SI_RETURN |
| bb.0: |
| successors: %bb.1(0x80000000) |
| liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4 |
| |
| %0:_(s32) = COPY $vgpr1 |
| %1:_(s32) = COPY $vgpr2 |
| %2:_(p1) = G_MERGE_VALUES %0(s32), %1(s32) |
| %3:_(s32) = COPY $vgpr3 |
| %4:_(s32) = COPY $vgpr4 |
| %5:_(p0) = G_MERGE_VALUES %3(s32), %4(s32) |
| %6:_(s32) = G_CONSTANT i32 -1 |
| %7:_(s1) = G_CONSTANT i1 true |
| |
| bb.1: |
| successors: %bb.2(0x40000000), %bb.3(0x40000000) |
| |
| %8:_(s32) = G_PHI %6(s32), %bb.0, %9(s32), %bb.3 |
| %10:_(p1) = G_PHI %2(p1), %bb.0, %11(p1), %bb.3 |
| %12:sreg_32_xm0_xexec(s1) = G_PHI %7(s1), %bb.0, %13(s1), %bb.3 |
| %14:sreg_32_xm0_xexec(s32) = SI_IF %12(s1), %bb.3, implicit-def $exec, implicit-def $scc, implicit $exec |
| G_BR %bb.2 |
| |
| bb.2: |
| successors: %bb.3(0x80000000) |
| |
| %15:_(s32) = G_LOAD %10(p1) :: (load (s32), addrspace 1) |
| %16:_(s32) = G_CONSTANT i32 0 |
| %17:_(s1) = G_ICMP intpred(eq), %15(s32), %16 |
| |
| bb.3: |
| successors: %bb.4(0x04000000), %bb.1(0x7c000000) |
| |
| %13:_(s1) = G_PHI %17(s1), %bb.2, %12(s1), %bb.1 |
| G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), %14(s32) |
| %18:_(s64) = G_CONSTANT i64 4 |
| %11:_(p1) = G_PTR_ADD %10, %18(s64) |
| %19:_(s32) = G_CONSTANT i32 1 |
| %9:_(s32) = nsw G_ADD %8, %19 |
| %20:_(s32) = G_CONSTANT i32 10 |
| %21:_(s1) = G_ICMP intpred(sge), %9(s32), %20 |
| G_BRCOND %21(s1), %bb.1 |
| G_BR %bb.4 |
| |
| bb.4: |
| %22:_(s1) = G_PHI %12(s1), %bb.3 |
| %23:_(s32) = G_FCONSTANT float 0.000000e+00 |
| %24:_(s32) = G_FCONSTANT float 1.000000e+00 |
| %25:_(s32) = G_SELECT %22(s1), %24, %23 |
| G_STORE %25(s32), %5(p0) :: (store (s32)) |
| SI_RETURN |
| ... |
| |
| --- |
| name: divergent_i1_xor_used_outside_loop |
| legalized: true |
| tracksRegLiveness: true |
| body: | |
| ; GFX10-LABEL: name: divergent_i1_xor_used_outside_loop |
| ; GFX10: bb.0: |
| ; GFX10-NEXT: successors: %bb.1(0x80000000) |
| ; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 |
| ; GFX10-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 |
| ; GFX10-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2 |
| ; GFX10-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3 |
| ; GFX10-NEXT: [[MV:%[0-9]+]]:_(p0) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32) |
| ; GFX10-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0 |
| ; GFX10-NEXT: [[C1:%[0-9]+]]:_(s32) = G_FCONSTANT float 1.000000e+00 |
| ; GFX10-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ogt), [[COPY1]](s32), [[C1]] |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.1: |
| ; GFX10-NEXT: successors: %bb.2(0x04000000), %bb.1(0x7c000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[PHI:%[0-9]+]]:_(s32) = G_PHI %9(s32), %bb.1, [[C]](s32), %bb.0 |
| ; GFX10-NEXT: [[PHI1:%[0-9]+]]:_(s32) = G_PHI [[C]](s32), %bb.0, %11(s32), %bb.1 |
| ; GFX10-NEXT: [[PHI2:%[0-9]+]]:_(s1) = G_PHI [[FCMP]](s1), %bb.0, %13(s1), %bb.1 |
| ; GFX10-NEXT: [[C2:%[0-9]+]]:_(s1) = G_CONSTANT i1 true |
| ; GFX10-NEXT: [[XOR:%[0-9]+]]:_(s1) = G_XOR [[PHI2]], [[C2]] |
| ; GFX10-NEXT: [[UITOFP:%[0-9]+]]:_(s32) = G_UITOFP [[PHI1]](s32) |
| ; GFX10-NEXT: [[FCMP1:%[0-9]+]]:_(s1) = G_FCMP floatpred(ogt), [[UITOFP]](s32), [[COPY]] |
| ; GFX10-NEXT: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 1 |
| ; GFX10-NEXT: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[PHI1]], [[C3]] |
| ; GFX10-NEXT: [[INTRINSIC_CONVERGENT:%[0-9]+]]:sreg_32_xm0_xexec(s32) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.if.break), [[FCMP1]](s1), [[PHI]](s32) |
| ; GFX10-NEXT: SI_LOOP [[INTRINSIC_CONVERGENT]](s32), %bb.1, implicit-def $exec, implicit-def $scc, implicit $exec |
| ; GFX10-NEXT: G_BR %bb.2 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.2: |
| ; GFX10-NEXT: [[PHI3:%[0-9]+]]:_(s1) = G_PHI [[XOR]](s1), %bb.1 |
| ; GFX10-NEXT: [[PHI4:%[0-9]+]]:_(s32) = G_PHI [[INTRINSIC_CONVERGENT]](s32), %bb.1 |
| ; GFX10-NEXT: G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), [[PHI4]](s32) |
| ; GFX10-NEXT: [[C4:%[0-9]+]]:_(s32) = G_FCONSTANT float 0.000000e+00 |
| ; GFX10-NEXT: [[C5:%[0-9]+]]:_(s32) = G_FCONSTANT float 1.000000e+00 |
| ; GFX10-NEXT: [[SELECT:%[0-9]+]]:_(s32) = G_SELECT [[PHI3]](s1), [[C5]], [[C4]] |
| ; GFX10-NEXT: G_STORE [[SELECT]](s32), [[MV]](p0) :: (store (s32)) |
| ; GFX10-NEXT: SI_RETURN |
| bb.0: |
| successors: %bb.1(0x80000000) |
| liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3 |
| |
| %0:_(s32) = COPY $vgpr0 |
| %1:_(s32) = COPY $vgpr1 |
| %2:_(s32) = COPY $vgpr2 |
| %3:_(s32) = COPY $vgpr3 |
| %4:_(p0) = G_MERGE_VALUES %2(s32), %3(s32) |
| %5:_(s32) = G_CONSTANT i32 0 |
| %6:_(s32) = G_FCONSTANT float 1.000000e+00 |
| %7:_(s1) = G_FCMP floatpred(ogt), %1(s32), %6 |
| |
| bb.1: |
| successors: %bb.2(0x04000000), %bb.1(0x7c000000) |
| |
| %8:_(s32) = G_PHI %9(s32), %bb.1, %5(s32), %bb.0 |
| %10:_(s32) = G_PHI %5(s32), %bb.0, %11(s32), %bb.1 |
| %12:_(s1) = G_PHI %7(s1), %bb.0, %13(s1), %bb.1 |
| %14:_(s1) = G_CONSTANT i1 true |
| %13:_(s1) = G_XOR %12, %14 |
| %15:_(s32) = G_UITOFP %10(s32) |
| %16:_(s1) = G_FCMP floatpred(ogt), %15(s32), %0 |
| %17:_(s32) = G_CONSTANT i32 1 |
| %11:_(s32) = G_ADD %10, %17 |
| %9:sreg_32_xm0_xexec(s32) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.if.break), %16(s1), %8(s32) |
| SI_LOOP %9(s32), %bb.1, implicit-def $exec, implicit-def $scc, implicit $exec |
| G_BR %bb.2 |
| |
| bb.2: |
| %18:_(s1) = G_PHI %13(s1), %bb.1 |
| %19:_(s32) = G_PHI %9(s32), %bb.1 |
| G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), %19(s32) |
| %20:_(s32) = G_FCONSTANT float 0.000000e+00 |
| %21:_(s32) = G_FCONSTANT float 1.000000e+00 |
| %22:_(s32) = G_SELECT %18(s1), %21, %20 |
| G_STORE %22(s32), %4(p0) :: (store (s32)) |
| SI_RETURN |
| ... |
| |
| --- |
| name: divergent_i1_xor_used_outside_loop_larger_loop_body |
| legalized: true |
| tracksRegLiveness: true |
| body: | |
| ; GFX10-LABEL: name: divergent_i1_xor_used_outside_loop_larger_loop_body |
| ; GFX10: bb.0: |
| ; GFX10-NEXT: successors: %bb.1(0x40000000), %bb.2(0x40000000) |
| ; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 |
| ; GFX10-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 |
| ; GFX10-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2 |
| ; GFX10-NEXT: [[MV:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[COPY1]](s32), [[COPY2]](s32) |
| ; GFX10-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3 |
| ; GFX10-NEXT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4 |
| ; GFX10-NEXT: [[MV1:%[0-9]+]]:_(p0) = G_MERGE_VALUES [[COPY3]](s32), [[COPY4]](s32) |
| ; GFX10-NEXT: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF |
| ; GFX10-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0 |
| ; GFX10-NEXT: [[ICMP:%[0-9]+]]:sreg_32_xm0_xexec(s1) = G_ICMP intpred(eq), [[COPY]](s32), [[C]] |
| ; GFX10-NEXT: [[C1:%[0-9]+]]:_(s1) = G_CONSTANT i1 true |
| ; GFX10-NEXT: [[SI_IF:%[0-9]+]]:sreg_32_xm0_xexec(s32) = SI_IF [[ICMP]](s1), %bb.2, implicit-def $exec, implicit-def $scc, implicit $exec |
| ; GFX10-NEXT: G_BR %bb.1 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.1: |
| ; GFX10-NEXT: successors: %bb.3(0x80000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 0 |
| ; GFX10-NEXT: G_BR %bb.3 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.2: |
| ; GFX10-NEXT: successors: %bb.5(0x40000000), %bb.6(0x40000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[PHI:%[0-9]+]]:sreg_32_xm0_xexec(s1) = G_PHI %14(s1), %bb.8, [[C1]](s1), %bb.0 |
| ; GFX10-NEXT: G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), [[SI_IF]](s32) |
| ; GFX10-NEXT: [[SI_IF1:%[0-9]+]]:sreg_32_xm0_xexec(s32) = SI_IF [[PHI]](s1), %bb.6, implicit-def $exec, implicit-def $scc, implicit $exec |
| ; GFX10-NEXT: G_BR %bb.5 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.3: |
| ; GFX10-NEXT: successors: %bb.4(0x40000000), %bb.7(0x40000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[PHI1:%[0-9]+]]:_(s32) = G_PHI [[C2]](s32), %bb.1, %17(s32), %bb.7 |
| ; GFX10-NEXT: [[PHI2:%[0-9]+]]:_(s32) = G_PHI %19(s32), %bb.7, [[C2]](s32), %bb.1 |
| ; GFX10-NEXT: [[C3:%[0-9]+]]:_(s1) = G_CONSTANT i1 true |
| ; GFX10-NEXT: [[SEXT:%[0-9]+]]:_(s64) = G_SEXT [[PHI2]](s32) |
| ; GFX10-NEXT: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 2 |
| ; GFX10-NEXT: [[SHL:%[0-9]+]]:_(s64) = G_SHL [[SEXT]], [[C4]](s32) |
| ; GFX10-NEXT: [[PTR_ADD:%[0-9]+]]:_(p1) = G_PTR_ADD [[MV]], [[SHL]](s64) |
| ; GFX10-NEXT: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p1) :: (load (s32), addrspace 1) |
| ; GFX10-NEXT: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 0 |
| ; GFX10-NEXT: [[ICMP1:%[0-9]+]]:sreg_32_xm0_xexec(s1) = G_ICMP intpred(ne), [[LOAD]](s32), [[C5]] |
| ; GFX10-NEXT: [[SI_IF2:%[0-9]+]]:sreg_32_xm0_xexec(s32) = SI_IF [[ICMP1]](s1), %bb.7, implicit-def $exec, implicit-def $scc, implicit $exec |
| ; GFX10-NEXT: G_BR %bb.4 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.4: |
| ; GFX10-NEXT: successors: %bb.7(0x80000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[C6:%[0-9]+]]:_(s1) = G_CONSTANT i1 false |
| ; GFX10-NEXT: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 1 |
| ; GFX10-NEXT: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[PHI2]], [[C7]] |
| ; GFX10-NEXT: [[ICMP2:%[0-9]+]]:_(s1) = G_ICMP intpred(slt), [[PHI2]](s32), [[COPY]] |
| ; GFX10-NEXT: G_BR %bb.7 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.5: |
| ; GFX10-NEXT: successors: %bb.6(0x80000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 5 |
| ; GFX10-NEXT: G_STORE [[C8]](s32), [[MV1]](p0) :: (store (s32)) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.6: |
| ; GFX10-NEXT: G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), [[SI_IF1]](s32) |
| ; GFX10-NEXT: SI_RETURN |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.7: |
| ; GFX10-NEXT: successors: %bb.8(0x04000000), %bb.3(0x7c000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[PHI3:%[0-9]+]]:_(s32) = G_PHI [[ADD]](s32), %bb.4, [[DEF]](s32), %bb.3 |
| ; GFX10-NEXT: [[PHI4:%[0-9]+]]:_(s1) = G_PHI [[C6]](s1), %bb.4, [[C3]](s1), %bb.3 |
| ; GFX10-NEXT: [[PHI5:%[0-9]+]]:_(s1) = G_PHI [[ICMP2]](s1), %bb.4, [[C3]](s1), %bb.3 |
| ; GFX10-NEXT: G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), [[SI_IF2]](s32) |
| ; GFX10-NEXT: [[C9:%[0-9]+]]:_(s1) = G_CONSTANT i1 true |
| ; GFX10-NEXT: [[XOR:%[0-9]+]]:_(s1) = G_XOR [[PHI4]], [[C9]] |
| ; GFX10-NEXT: [[INTRINSIC_CONVERGENT:%[0-9]+]]:sreg_32_xm0_xexec(s32) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.if.break), [[PHI5]](s1), [[PHI1]](s32) |
| ; GFX10-NEXT: SI_LOOP [[INTRINSIC_CONVERGENT]](s32), %bb.3, implicit-def $exec, implicit-def $scc, implicit $exec |
| ; GFX10-NEXT: G_BR %bb.8 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.8: |
| ; GFX10-NEXT: successors: %bb.2(0x80000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[PHI6:%[0-9]+]]:_(s1) = G_PHI [[XOR]](s1), %bb.7 |
| ; GFX10-NEXT: [[PHI7:%[0-9]+]]:_(s32) = G_PHI [[INTRINSIC_CONVERGENT]](s32), %bb.7 |
| ; GFX10-NEXT: G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), [[PHI7]](s32) |
| ; GFX10-NEXT: G_BR %bb.2 |
| bb.0: |
| successors: %bb.1(0x40000000), %bb.2(0x40000000) |
| liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4 |
| |
| %0:_(s32) = COPY $vgpr0 |
| %1:_(s32) = COPY $vgpr1 |
| %2:_(s32) = COPY $vgpr2 |
| %3:_(p1) = G_MERGE_VALUES %1(s32), %2(s32) |
| %4:_(s32) = COPY $vgpr3 |
| %5:_(s32) = COPY $vgpr4 |
| %6:_(p0) = G_MERGE_VALUES %4(s32), %5(s32) |
| %7:_(s32) = G_IMPLICIT_DEF |
| %8:_(s32) = G_CONSTANT i32 0 |
| %9:sreg_32_xm0_xexec(s1) = G_ICMP intpred(eq), %0(s32), %8 |
| %10:_(s1) = G_CONSTANT i1 true |
| %11:sreg_32_xm0_xexec(s32) = SI_IF %9(s1), %bb.2, implicit-def $exec, implicit-def $scc, implicit $exec |
| G_BR %bb.1 |
| |
| bb.1: |
| successors: %bb.3(0x80000000) |
| |
| %12:_(s32) = G_CONSTANT i32 0 |
| G_BR %bb.3 |
| |
| bb.2: |
| successors: %bb.5(0x40000000), %bb.6(0x40000000) |
| |
| %13:sreg_32_xm0_xexec(s1) = G_PHI %14(s1), %bb.8, %10(s1), %bb.0 |
| G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), %11(s32) |
| %15:sreg_32_xm0_xexec(s32) = SI_IF %13(s1), %bb.6, implicit-def $exec, implicit-def $scc, implicit $exec |
| G_BR %bb.5 |
| |
| bb.3: |
| successors: %bb.4(0x40000000), %bb.7(0x40000000) |
| |
| %16:_(s32) = G_PHI %12(s32), %bb.1, %17(s32), %bb.7 |
| %18:_(s32) = G_PHI %19(s32), %bb.7, %12(s32), %bb.1 |
| %20:_(s1) = G_CONSTANT i1 true |
| %21:_(s64) = G_SEXT %18(s32) |
| %22:_(s32) = G_CONSTANT i32 2 |
| %23:_(s64) = G_SHL %21, %22(s32) |
| %24:_(p1) = G_PTR_ADD %3, %23(s64) |
| %25:_(s32) = G_LOAD %24(p1) :: (load (s32), addrspace 1) |
| %26:_(s32) = G_CONSTANT i32 0 |
| %27:sreg_32_xm0_xexec(s1) = G_ICMP intpred(ne), %25(s32), %26 |
| %28:sreg_32_xm0_xexec(s32) = SI_IF %27(s1), %bb.7, implicit-def $exec, implicit-def $scc, implicit $exec |
| G_BR %bb.4 |
| |
| bb.4: |
| successors: %bb.7(0x80000000) |
| |
| %29:_(s1) = G_CONSTANT i1 false |
| %30:_(s32) = G_CONSTANT i32 1 |
| %31:_(s32) = G_ADD %18, %30 |
| %32:_(s1) = G_ICMP intpred(slt), %18(s32), %0 |
| G_BR %bb.7 |
| |
| bb.5: |
| successors: %bb.6(0x80000000) |
| |
| %33:_(s32) = G_CONSTANT i32 5 |
| G_STORE %33(s32), %6(p0) :: (store (s32)) |
| |
| bb.6: |
| G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), %15(s32) |
| SI_RETURN |
| |
| bb.7: |
| successors: %bb.8(0x04000000), %bb.3(0x7c000000) |
| |
| %19:_(s32) = G_PHI %31(s32), %bb.4, %7(s32), %bb.3 |
| %34:_(s1) = G_PHI %29(s1), %bb.4, %20(s1), %bb.3 |
| %35:_(s1) = G_PHI %32(s1), %bb.4, %20(s1), %bb.3 |
| G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), %28(s32) |
| %36:_(s1) = G_CONSTANT i1 true |
| %37:_(s1) = G_XOR %34, %36 |
| %17:sreg_32_xm0_xexec(s32) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.if.break), %35(s1), %16(s32) |
| SI_LOOP %17(s32), %bb.3, implicit-def $exec, implicit-def $scc, implicit $exec |
| G_BR %bb.8 |
| |
| bb.8: |
| successors: %bb.2(0x80000000) |
| |
| %14:_(s1) = G_PHI %37(s1), %bb.7 |
| %38:_(s32) = G_PHI %17(s32), %bb.7 |
| G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), %38(s32) |
| G_BR %bb.2 |
| ... |
| |
| --- |
| name: divergent_i1_icmp_used_outside_loop |
| legalized: true |
| tracksRegLiveness: true |
| body: | |
| ; GFX10-LABEL: name: divergent_i1_icmp_used_outside_loop |
| ; GFX10: bb.0: |
| ; GFX10-NEXT: successors: %bb.1(0x80000000) |
| ; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 |
| ; GFX10-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 |
| ; GFX10-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2 |
| ; GFX10-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3 |
| ; GFX10-NEXT: [[MV:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32) |
| ; GFX10-NEXT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr6 |
| ; GFX10-NEXT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr7 |
| ; GFX10-NEXT: [[MV1:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32) |
| ; GFX10-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0 |
| ; GFX10-NEXT: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.1: |
| ; GFX10-NEXT: successors: %bb.2(0x80000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[PHI:%[0-9]+]]:_(s32) = G_PHI %11(s32), %bb.6, [[C]](s32), %bb.0 |
| ; GFX10-NEXT: [[PHI1:%[0-9]+]]:_(s32) = G_PHI [[C]](s32), %bb.0, %13(s32), %bb.6 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.2: |
| ; GFX10-NEXT: successors: %bb.3(0x40000000), %bb.4(0x40000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[ICMP:%[0-9]+]]:sreg_32_xm0_xexec(s1) = G_ICMP intpred(eq), [[COPY]](s32), [[PHI1]] |
| ; GFX10-NEXT: [[SI_IF:%[0-9]+]]:sreg_32_xm0_xexec(s32) = SI_IF [[ICMP]](s1), %bb.4, implicit-def $exec, implicit-def $scc, implicit $exec |
| ; GFX10-NEXT: G_BR %bb.3 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.3: |
| ; GFX10-NEXT: successors: %bb.4(0x80000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[SEXT:%[0-9]+]]:_(s64) = G_SEXT [[PHI1]](s32) |
| ; GFX10-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2 |
| ; GFX10-NEXT: [[SHL:%[0-9]+]]:_(s64) = G_SHL [[SEXT]], [[C1]](s32) |
| ; GFX10-NEXT: [[PTR_ADD:%[0-9]+]]:_(p1) = G_PTR_ADD [[MV]], [[SHL]](s64) |
| ; GFX10-NEXT: G_STORE [[PHI1]](s32), [[PTR_ADD]](p1) :: (store (s32), addrspace 1) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.4: |
| ; GFX10-NEXT: successors: %bb.5(0x40000000), %bb.6(0x40000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[C2:%[0-9]+]]:_(s1) = G_CONSTANT i1 true |
| ; GFX10-NEXT: G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), [[SI_IF]](s32) |
| ; GFX10-NEXT: [[ICMP1:%[0-9]+]]:sreg_32_xm0_xexec(s1) = G_ICMP intpred(ne), [[COPY1]](s32), [[PHI1]] |
| ; GFX10-NEXT: [[SI_IF1:%[0-9]+]]:sreg_32_xm0_xexec(s32) = SI_IF [[ICMP1]](s1), %bb.6, implicit-def $exec, implicit-def $scc, implicit $exec |
| ; GFX10-NEXT: G_BR %bb.5 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.5: |
| ; GFX10-NEXT: successors: %bb.6(0x80000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[C3:%[0-9]+]]:_(s1) = G_CONSTANT i1 false |
| ; GFX10-NEXT: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 1 |
| ; GFX10-NEXT: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[PHI1]], [[C4]] |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.6: |
| ; GFX10-NEXT: successors: %bb.7(0x04000000), %bb.1(0x7c000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[PHI2:%[0-9]+]]:_(s32) = G_PHI [[ADD]](s32), %bb.5, [[DEF]](s32), %bb.4 |
| ; GFX10-NEXT: [[PHI3:%[0-9]+]]:_(s1) = G_PHI [[C3]](s1), %bb.5, [[C2]](s1), %bb.4 |
| ; GFX10-NEXT: G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), [[SI_IF1]](s32) |
| ; GFX10-NEXT: [[INTRINSIC_CONVERGENT:%[0-9]+]]:sreg_32_xm0_xexec(s32) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.if.break), [[PHI3]](s1), [[PHI]](s32) |
| ; GFX10-NEXT: SI_LOOP [[INTRINSIC_CONVERGENT]](s32), %bb.1, implicit-def $exec, implicit-def $scc, implicit $exec |
| ; GFX10-NEXT: G_BR %bb.7 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.7: |
| ; GFX10-NEXT: successors: %bb.8(0x40000000), %bb.9(0x40000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[PHI4:%[0-9]+]]:_(s32) = G_PHI [[INTRINSIC_CONVERGENT]](s32), %bb.6 |
| ; GFX10-NEXT: [[PHI5:%[0-9]+]]:sreg_32_xm0_xexec(s1) = G_PHI [[ICMP]](s1), %bb.6 |
| ; GFX10-NEXT: [[PHI6:%[0-9]+]]:_(s32) = G_PHI [[PHI1]](s32), %bb.6 |
| ; GFX10-NEXT: G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), [[PHI4]](s32) |
| ; GFX10-NEXT: [[SI_IF2:%[0-9]+]]:sreg_32_xm0_xexec(s32) = SI_IF [[PHI5]](s1), %bb.9, implicit-def $exec, implicit-def $scc, implicit $exec |
| ; GFX10-NEXT: G_BR %bb.8 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.8: |
| ; GFX10-NEXT: successors: %bb.9(0x80000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: G_STORE [[PHI6]](s32), [[MV1]](p1) :: (store (s32), addrspace 1) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.9: |
| ; GFX10-NEXT: G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), [[SI_IF2]](s32) |
| ; GFX10-NEXT: SI_RETURN |
| bb.0: |
| successors: %bb.1(0x80000000) |
| liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9 |
| |
| %0:_(s32) = COPY $vgpr0 |
| %1:_(s32) = COPY $vgpr1 |
| %2:_(s32) = COPY $vgpr2 |
| %3:_(s32) = COPY $vgpr3 |
| %4:_(p1) = G_MERGE_VALUES %2(s32), %3(s32) |
| %5:_(s32) = COPY $vgpr6 |
| %6:_(s32) = COPY $vgpr7 |
| %7:_(p1) = G_MERGE_VALUES %5(s32), %6(s32) |
| %8:_(s32) = G_CONSTANT i32 0 |
| %9:_(s32) = G_IMPLICIT_DEF |
| |
| bb.1: |
| successors: %bb.2(0x80000000) |
| |
| %10:_(s32) = G_PHI %11(s32), %bb.6, %8(s32), %bb.0 |
| %12:_(s32) = G_PHI %8(s32), %bb.0, %13(s32), %bb.6 |
| |
| bb.2: |
| successors: %bb.3(0x40000000), %bb.4(0x40000000) |
| |
| %14:sreg_32_xm0_xexec(s1) = G_ICMP intpred(eq), %0(s32), %12 |
| %15:sreg_32_xm0_xexec(s32) = SI_IF %14(s1), %bb.4, implicit-def $exec, implicit-def $scc, implicit $exec |
| G_BR %bb.3 |
| |
| bb.3: |
| successors: %bb.4(0x80000000) |
| |
| %16:_(s64) = G_SEXT %12(s32) |
| %17:_(s32) = G_CONSTANT i32 2 |
| %18:_(s64) = G_SHL %16, %17(s32) |
| %19:_(p1) = G_PTR_ADD %4, %18(s64) |
| G_STORE %12(s32), %19(p1) :: (store (s32), addrspace 1) |
| |
| bb.4: |
| successors: %bb.5(0x40000000), %bb.6(0x40000000) |
| |
| %20:_(s1) = G_CONSTANT i1 true |
| G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), %15(s32) |
| %21:sreg_32_xm0_xexec(s1) = G_ICMP intpred(ne), %1(s32), %12 |
| %22:sreg_32_xm0_xexec(s32) = SI_IF %21(s1), %bb.6, implicit-def $exec, implicit-def $scc, implicit $exec |
| G_BR %bb.5 |
| |
| bb.5: |
| successors: %bb.6(0x80000000) |
| |
| %23:_(s1) = G_CONSTANT i1 false |
| %24:_(s32) = G_CONSTANT i32 1 |
| %25:_(s32) = G_ADD %12, %24 |
| |
| bb.6: |
| successors: %bb.7(0x04000000), %bb.1(0x7c000000) |
| |
| %13:_(s32) = G_PHI %25(s32), %bb.5, %9(s32), %bb.4 |
| %26:_(s1) = G_PHI %23(s1), %bb.5, %20(s1), %bb.4 |
| G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), %22(s32) |
| %11:sreg_32_xm0_xexec(s32) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.if.break), %26(s1), %10(s32) |
| SI_LOOP %11(s32), %bb.1, implicit-def $exec, implicit-def $scc, implicit $exec |
| G_BR %bb.7 |
| |
| bb.7: |
| successors: %bb.8(0x40000000), %bb.9(0x40000000) |
| |
| %27:_(s32) = G_PHI %11(s32), %bb.6 |
| %28:sreg_32_xm0_xexec(s1) = G_PHI %14(s1), %bb.6 |
| %29:_(s32) = G_PHI %12(s32), %bb.6 |
| G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), %27(s32) |
| %30:sreg_32_xm0_xexec(s32) = SI_IF %28(s1), %bb.9, implicit-def $exec, implicit-def $scc, implicit $exec |
| G_BR %bb.8 |
| |
| bb.8: |
| successors: %bb.9(0x80000000) |
| |
| G_STORE %29(s32), %7(p1) :: (store (s32), addrspace 1) |
| |
| bb.9: |
| G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), %30(s32) |
| SI_RETURN |
| ... |
| |
| --- |
| name: divergent_i1_freeze_used_outside_loop |
| legalized: true |
| tracksRegLiveness: true |
| body: | |
| ; GFX10-LABEL: name: divergent_i1_freeze_used_outside_loop |
| ; GFX10: bb.0: |
| ; GFX10-NEXT: successors: %bb.1(0x80000000) |
| ; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 |
| ; GFX10-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 |
| ; GFX10-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2 |
| ; GFX10-NEXT: [[MV:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[COPY1]](s32), [[COPY2]](s32) |
| ; GFX10-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3 |
| ; GFX10-NEXT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4 |
| ; GFX10-NEXT: [[MV1:%[0-9]+]]:_(p0) = G_MERGE_VALUES [[COPY3]](s32), [[COPY4]](s32) |
| ; GFX10-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0 |
| ; GFX10-NEXT: [[C1:%[0-9]+]]:_(s1) = G_CONSTANT i1 true |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.1: |
| ; GFX10-NEXT: successors: %bb.2(0x40000000), %bb.3(0x40000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[PHI:%[0-9]+]]:_(s32) = G_PHI %10(s32), %bb.3, [[C]](s32), %bb.0 |
| ; GFX10-NEXT: [[PHI1:%[0-9]+]]:_(s32) = G_PHI [[C]](s32), %bb.0, %12(s32), %bb.3 |
| ; GFX10-NEXT: [[PHI2:%[0-9]+]]:sreg_32_xm0_xexec(s1) = G_PHI [[C1]](s1), %bb.0, %14(s1), %bb.3 |
| ; GFX10-NEXT: [[SI_IF:%[0-9]+]]:sreg_32_xm0_xexec(s32) = SI_IF [[PHI2]](s1), %bb.3, implicit-def $exec, implicit-def $scc, implicit $exec |
| ; GFX10-NEXT: G_BR %bb.2 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.2: |
| ; GFX10-NEXT: successors: %bb.3(0x80000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[SEXT:%[0-9]+]]:_(s64) = G_SEXT [[PHI1]](s32) |
| ; GFX10-NEXT: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 2 |
| ; GFX10-NEXT: [[SHL:%[0-9]+]]:_(s64) = G_SHL [[SEXT]], [[C2]](s32) |
| ; GFX10-NEXT: [[PTR_ADD:%[0-9]+]]:_(p1) = G_PTR_ADD [[MV]], [[SHL]](s64) |
| ; GFX10-NEXT: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p1) :: (load (s32), addrspace 1) |
| ; GFX10-NEXT: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 0 |
| ; GFX10-NEXT: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(eq), [[LOAD]](s32), [[C3]] |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.3: |
| ; GFX10-NEXT: successors: %bb.4(0x04000000), %bb.1(0x7c000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[PHI3:%[0-9]+]]:_(s1) = G_PHI [[ICMP]](s1), %bb.2, [[PHI2]](s1), %bb.1 |
| ; GFX10-NEXT: G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), [[SI_IF]](s32) |
| ; GFX10-NEXT: [[FREEZE:%[0-9]+]]:_(s1) = G_FREEZE [[PHI3]] |
| ; GFX10-NEXT: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 1 |
| ; GFX10-NEXT: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[PHI1]], [[C4]] |
| ; GFX10-NEXT: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(slt), [[PHI1]](s32), [[COPY]] |
| ; GFX10-NEXT: [[INTRINSIC_CONVERGENT:%[0-9]+]]:sreg_32_xm0_xexec(s32) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.if.break), [[ICMP1]](s1), [[PHI]](s32) |
| ; GFX10-NEXT: SI_LOOP [[INTRINSIC_CONVERGENT]](s32), %bb.1, implicit-def $exec, implicit-def $scc, implicit $exec |
| ; GFX10-NEXT: G_BR %bb.4 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.4: |
| ; GFX10-NEXT: [[PHI4:%[0-9]+]]:_(s1) = G_PHI [[FREEZE]](s1), %bb.3 |
| ; GFX10-NEXT: [[PHI5:%[0-9]+]]:_(s32) = G_PHI [[INTRINSIC_CONVERGENT]](s32), %bb.3 |
| ; GFX10-NEXT: G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), [[PHI5]](s32) |
| ; GFX10-NEXT: [[C5:%[0-9]+]]:_(s32) = G_FCONSTANT float 0.000000e+00 |
| ; GFX10-NEXT: [[C6:%[0-9]+]]:_(s32) = G_FCONSTANT float 1.000000e+00 |
| ; GFX10-NEXT: [[SELECT:%[0-9]+]]:_(s32) = G_SELECT [[PHI4]](s1), [[C6]], [[C5]] |
| ; GFX10-NEXT: G_STORE [[SELECT]](s32), [[MV1]](p0) :: (store (s32)) |
| ; GFX10-NEXT: S_ENDPGM 0 |
| bb.0: |
| successors: %bb.1(0x80000000) |
| liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4 |
| |
| %0:_(s32) = COPY $vgpr0 |
| %1:_(s32) = COPY $vgpr1 |
| %2:_(s32) = COPY $vgpr2 |
| %3:_(p1) = G_MERGE_VALUES %1(s32), %2(s32) |
| %4:_(s32) = COPY $vgpr3 |
| %5:_(s32) = COPY $vgpr4 |
| %6:_(p0) = G_MERGE_VALUES %4(s32), %5(s32) |
| %7:_(s32) = G_CONSTANT i32 0 |
| %8:_(s1) = G_CONSTANT i1 true |
| |
| bb.1: |
| successors: %bb.2(0x40000000), %bb.3(0x40000000) |
| |
| %9:_(s32) = G_PHI %10(s32), %bb.3, %7(s32), %bb.0 |
| %11:_(s32) = G_PHI %7(s32), %bb.0, %12(s32), %bb.3 |
| %13:sreg_32_xm0_xexec(s1) = G_PHI %8(s1), %bb.0, %14(s1), %bb.3 |
| %15:sreg_32_xm0_xexec(s32) = SI_IF %13(s1), %bb.3, implicit-def $exec, implicit-def $scc, implicit $exec |
| G_BR %bb.2 |
| |
| bb.2: |
| successors: %bb.3(0x80000000) |
| |
| %16:_(s64) = G_SEXT %11(s32) |
| %17:_(s32) = G_CONSTANT i32 2 |
| %18:_(s64) = G_SHL %16, %17(s32) |
| %19:_(p1) = G_PTR_ADD %3, %18(s64) |
| %20:_(s32) = G_LOAD %19(p1) :: (load (s32), addrspace 1) |
| %21:_(s32) = G_CONSTANT i32 0 |
| %22:_(s1) = G_ICMP intpred(eq), %20(s32), %21 |
| |
| bb.3: |
| successors: %bb.4(0x04000000), %bb.1(0x7c000000) |
| |
| %23:_(s1) = G_PHI %22(s1), %bb.2, %13(s1), %bb.1 |
| G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), %15(s32) |
| %14:_(s1) = G_FREEZE %23 |
| %24:_(s32) = G_CONSTANT i32 1 |
| %12:_(s32) = G_ADD %11, %24 |
| %25:_(s1) = G_ICMP intpred(slt), %11(s32), %0 |
| %10:sreg_32_xm0_xexec(s32) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.if.break), %25(s1), %9(s32) |
| SI_LOOP %10(s32), %bb.1, implicit-def $exec, implicit-def $scc, implicit $exec |
| G_BR %bb.4 |
| |
| bb.4: |
| %26:_(s1) = G_PHI %14(s1), %bb.3 |
| %27:_(s32) = G_PHI %10(s32), %bb.3 |
| G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), %27(s32) |
| %28:_(s32) = G_FCONSTANT float 0.000000e+00 |
| %29:_(s32) = G_FCONSTANT float 1.000000e+00 |
| %30:_(s32) = G_SELECT %26(s1), %29, %28 |
| G_STORE %30(s32), %6(p0) :: (store (s32)) |
| S_ENDPGM 0 |
| ... |
| |
| --- |
| name: loop_with_1break |
| legalized: true |
| tracksRegLiveness: true |
| body: | |
| ; GFX10-LABEL: name: loop_with_1break |
| ; GFX10: bb.0: |
| ; GFX10-NEXT: successors: %bb.1(0x80000000) |
| ; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 |
| ; GFX10-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 |
| ; GFX10-NEXT: [[MV:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32) |
| ; GFX10-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2 |
| ; GFX10-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3 |
| ; GFX10-NEXT: [[MV1:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32) |
| ; GFX10-NEXT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4 |
| ; GFX10-NEXT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5 |
| ; GFX10-NEXT: [[MV2:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32) |
| ; GFX10-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0 |
| ; GFX10-NEXT: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.1: |
| ; GFX10-NEXT: successors: %bb.3(0x40000000), %bb.5(0x40000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[PHI:%[0-9]+]]:_(s32) = G_PHI %12(s32), %bb.5, [[C]](s32), %bb.0 |
| ; GFX10-NEXT: [[PHI1:%[0-9]+]]:_(s32) = G_PHI [[C]](s32), %bb.0, %14(s32), %bb.5 |
| ; GFX10-NEXT: [[C1:%[0-9]+]]:_(s1) = G_CONSTANT i1 true |
| ; GFX10-NEXT: [[SEXT:%[0-9]+]]:_(s64) = G_SEXT [[PHI1]](s32) |
| ; GFX10-NEXT: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 2 |
| ; GFX10-NEXT: [[SHL:%[0-9]+]]:_(s64) = G_SHL [[SEXT]], [[C2]](s32) |
| ; GFX10-NEXT: [[PTR_ADD:%[0-9]+]]:_(p1) = G_PTR_ADD [[MV1]], [[SHL]](s64) |
| ; GFX10-NEXT: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p1) :: (load (s32), addrspace 1) |
| ; GFX10-NEXT: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 0 |
| ; GFX10-NEXT: [[ICMP:%[0-9]+]]:sreg_32_xm0_xexec(s1) = G_ICMP intpred(ne), [[LOAD]](s32), [[C3]] |
| ; GFX10-NEXT: [[SI_IF:%[0-9]+]]:sreg_32_xm0_xexec(s32) = SI_IF [[ICMP]](s1), %bb.5, implicit-def $exec, implicit-def $scc, implicit $exec |
| ; GFX10-NEXT: G_BR %bb.3 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.2: |
| ; GFX10-NEXT: successors: %bb.4(0x80000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 10 |
| ; GFX10-NEXT: G_STORE [[C4]](s32), [[MV2]](p1) :: (store (s32), addrspace 1) |
| ; GFX10-NEXT: G_BR %bb.4 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.3: |
| ; GFX10-NEXT: successors: %bb.5(0x80000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[C5:%[0-9]+]]:_(s1) = G_CONSTANT i1 false |
| ; GFX10-NEXT: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 2 |
| ; GFX10-NEXT: [[SHL1:%[0-9]+]]:_(s64) = G_SHL [[SEXT]], [[C6]](s32) |
| ; GFX10-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p1) = G_PTR_ADD [[MV]], [[SHL1]](s64) |
| ; GFX10-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p1) :: (load (s32), addrspace 1) |
| ; GFX10-NEXT: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 1 |
| ; GFX10-NEXT: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[LOAD1]], [[C7]] |
| ; GFX10-NEXT: G_STORE [[ADD]](s32), [[PTR_ADD1]](p1) :: (store (s32), addrspace 1) |
| ; GFX10-NEXT: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[PHI1]], [[C7]] |
| ; GFX10-NEXT: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 100 |
| ; GFX10-NEXT: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ult), [[PHI1]](s32), [[C8]] |
| ; GFX10-NEXT: G_BR %bb.5 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.4: |
| ; GFX10-NEXT: G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), %35(s32) |
| ; GFX10-NEXT: S_ENDPGM 0 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.5: |
| ; GFX10-NEXT: successors: %bb.6(0x04000000), %bb.1(0x7c000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[PHI2:%[0-9]+]]:_(s32) = G_PHI [[ADD1]](s32), %bb.3, [[DEF]](s32), %bb.1 |
| ; GFX10-NEXT: [[PHI3:%[0-9]+]]:_(s1) = G_PHI [[C5]](s1), %bb.3, [[C1]](s1), %bb.1 |
| ; GFX10-NEXT: [[PHI4:%[0-9]+]]:_(s1) = G_PHI [[ICMP1]](s1), %bb.3, [[C1]](s1), %bb.1 |
| ; GFX10-NEXT: G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), [[SI_IF]](s32) |
| ; GFX10-NEXT: [[INTRINSIC_CONVERGENT:%[0-9]+]]:sreg_32_xm0_xexec(s32) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.if.break), [[PHI4]](s1), [[PHI]](s32) |
| ; GFX10-NEXT: SI_LOOP [[INTRINSIC_CONVERGENT]](s32), %bb.1, implicit-def $exec, implicit-def $scc, implicit $exec |
| ; GFX10-NEXT: G_BR %bb.6 |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: bb.6: |
| ; GFX10-NEXT: successors: %bb.2(0x40000000), %bb.4(0x40000000) |
| ; GFX10-NEXT: {{ $}} |
| ; GFX10-NEXT: [[PHI5:%[0-9]+]]:sreg_32_xm0_xexec(s1) = G_PHI [[PHI3]](s1), %bb.5 |
| ; GFX10-NEXT: [[PHI6:%[0-9]+]]:_(s32) = G_PHI [[INTRINSIC_CONVERGENT]](s32), %bb.5 |
| ; GFX10-NEXT: G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), [[PHI6]](s32) |
| ; GFX10-NEXT: [[SI_IF1:%[0-9]+]]:sreg_32_xm0_xexec(s32) = SI_IF [[PHI5]](s1), %bb.4, implicit-def $exec, implicit-def $scc, implicit $exec |
| ; GFX10-NEXT: G_BR %bb.2 |
| bb.0: |
| successors: %bb.1(0x80000000) |
| liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5 |
| |
| %0:_(s32) = COPY $vgpr0 |
| %1:_(s32) = COPY $vgpr1 |
| %2:_(p1) = G_MERGE_VALUES %0(s32), %1(s32) |
| %3:_(s32) = COPY $vgpr2 |
| %4:_(s32) = COPY $vgpr3 |
| %5:_(p1) = G_MERGE_VALUES %3(s32), %4(s32) |
| %6:_(s32) = COPY $vgpr4 |
| %7:_(s32) = COPY $vgpr5 |
| %8:_(p1) = G_MERGE_VALUES %6(s32), %7(s32) |
| %9:_(s32) = G_CONSTANT i32 0 |
| %10:_(s32) = G_IMPLICIT_DEF |
| |
| bb.1: |
| successors: %bb.3(0x40000000), %bb.5(0x40000000) |
| |
| %11:_(s32) = G_PHI %12(s32), %bb.5, %9(s32), %bb.0 |
| %13:_(s32) = G_PHI %9(s32), %bb.0, %14(s32), %bb.5 |
| %15:_(s1) = G_CONSTANT i1 true |
| %16:_(s64) = G_SEXT %13(s32) |
| %17:_(s32) = G_CONSTANT i32 2 |
| %18:_(s64) = G_SHL %16, %17(s32) |
| %19:_(p1) = G_PTR_ADD %5, %18(s64) |
| %20:_(s32) = G_LOAD %19(p1) :: (load (s32), addrspace 1) |
| %21:_(s32) = G_CONSTANT i32 0 |
| %22:sreg_32_xm0_xexec(s1) = G_ICMP intpred(ne), %20(s32), %21 |
| %23:sreg_32_xm0_xexec(s32) = SI_IF %22(s1), %bb.5, implicit-def $exec, implicit-def $scc, implicit $exec |
| G_BR %bb.3 |
| |
| bb.2: |
| successors: %bb.4(0x80000000) |
| |
| %24:_(s32) = G_CONSTANT i32 10 |
| G_STORE %24(s32), %8(p1) :: (store (s32), addrspace 1) |
| G_BR %bb.4 |
| |
| bb.3: |
| successors: %bb.5(0x80000000) |
| |
| %25:_(s1) = G_CONSTANT i1 false |
| %26:_(s32) = G_CONSTANT i32 2 |
| %27:_(s64) = G_SHL %16, %26(s32) |
| %28:_(p1) = G_PTR_ADD %2, %27(s64) |
| %29:_(s32) = G_LOAD %28(p1) :: (load (s32), addrspace 1) |
| %30:_(s32) = G_CONSTANT i32 1 |
| %31:_(s32) = G_ADD %29, %30 |
| G_STORE %31(s32), %28(p1) :: (store (s32), addrspace 1) |
| %32:_(s32) = G_ADD %13, %30 |
| %33:_(s32) = G_CONSTANT i32 100 |
| %34:_(s1) = G_ICMP intpred(ult), %13(s32), %33 |
| G_BR %bb.5 |
| |
| bb.4: |
| G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), %35(s32) |
| S_ENDPGM 0 |
| |
| bb.5: |
| successors: %bb.6(0x04000000), %bb.1(0x7c000000) |
| |
| %14:_(s32) = G_PHI %32(s32), %bb.3, %10(s32), %bb.1 |
| %36:_(s1) = G_PHI %25(s1), %bb.3, %15(s1), %bb.1 |
| %37:_(s1) = G_PHI %34(s1), %bb.3, %15(s1), %bb.1 |
| G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), %23(s32) |
| %12:sreg_32_xm0_xexec(s32) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.if.break), %37(s1), %11(s32) |
| SI_LOOP %12(s32), %bb.1, implicit-def $exec, implicit-def $scc, implicit $exec |
| G_BR %bb.6 |
| |
| bb.6: |
| successors: %bb.2(0x40000000), %bb.4(0x40000000) |
| |
| %38:sreg_32_xm0_xexec(s1) = G_PHI %36(s1), %bb.5 |
| %39:_(s32) = G_PHI %12(s32), %bb.5 |
| G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.end.cf), %39(s32) |
| %35:sreg_32_xm0_xexec(s32) = SI_IF %38(s1), %bb.4, implicit-def $exec, implicit-def $scc, implicit $exec |
| G_BR %bb.2 |
| ... |