1# RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx803 -run-pass=simple-register-coalescing -o - %s | FileCheck -check-prefix=GCN %s 2 3# With one version of the D48102 fix, this test failed with 4# Assertion failed: (ValNo && "CopyMI input register not live"), function reMaterializeTrivialDef, file ../lib/CodeGen/RegisterCoalescer.cpp, line 1107. 5 6# GCN: {{^body}} 7 8--- | 9 target datalayout = "e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-S32-A5" 10 target triple = "amdgcn--amdpal" 11 12 define amdgpu_cs void @_amdgpu_cs_main(<3 x i32>) #0 { 13 ret void 14 } 15 16 attributes #0 = { nounwind "target-cpu"="gfx803" } 17... 18 19--- 20name: _amdgpu_cs_main 21tracksRegLiveness: true 22body: | 23 bb.0: 24 successors: %bb.1(0x40000000), %bb.2(0x40000000) 25 %0:vgpr_32 = V_MUL_F32_e32 0, undef %1:vgpr_32, implicit $exec 26 %2:vgpr_32 = V_CVT_U32_F32_e32 killed %0, implicit $exec 27 %3:vgpr_32 = V_CVT_F32_I32_e32 killed %2, implicit $exec 28 %4:vgpr_32 = V_CVT_U32_F32_e32 killed %3, implicit $exec 29 %5:vgpr_32 = V_LSHRREV_B32_e32 4, killed %4, implicit $exec 30 S_CBRANCH_SCC0 %bb.2, implicit undef $scc 31 32 bb.1: 33 successors: %bb.5(0x80000000) 34 undef %6.sub1:vreg_128 = COPY killed %5 35 %7:vreg_128 = COPY killed %6 36 S_BRANCH %bb.5 37 38 bb.2: 39 successors: %bb.3(0x40000000), %bb.4(0x40000000) 40 S_CBRANCH_SCC0 %bb.4, implicit undef $scc 41 42 bb.3: 43 successors: %bb.5(0x80000000) 44 %8:sreg_32_xm0 = S_MOV_B32 0 45 undef %9.sub0:sreg_128 = COPY %8 46 %9.sub1:sreg_128 = COPY %8 47 %9.sub2:sreg_128 = COPY %8 48 %9.sub3:sreg_128 = COPY killed %8 49 %10:vreg_128 = COPY killed %9 50 %7:vreg_128 = COPY killed %10 51 S_BRANCH %bb.5 52 53 bb.4: 54 successors: %bb.5(0x80000000) 55 %11:sreg_32_xm0 = S_MOV_B32 0 56 undef %12.sub0:sreg_128 = COPY %11 57 %12.sub1:sreg_128 = COPY %11 58 %12.sub2:sreg_128 = COPY %11 59 %12.sub3:sreg_128 = COPY killed %11 60 %13:sreg_128 = COPY killed %12 61 %14:vreg_128 = COPY killed %13 62 %7:vreg_128 = COPY killed %14 63 64 bb.5: 65 successors: %bb.8(0x40000000), %bb.6(0x40000000) 66 %15:vreg_128 = COPY killed %7 67 S_CBRANCH_SCC0 %bb.8, implicit undef $scc 68 69 bb.6: 70 successors: %bb.7(0x80000000) 71 %16:vreg_128 = COPY killed %15 72 73 bb.7: 74 successors: %bb.14(0x80000000) 75 %17:vreg_128 = COPY killed %16 76 S_BRANCH %bb.14 77 78 bb.8: 79 successors: %bb.9(0x40000000), %bb.11(0x40000000) 80 %18:vgpr_32 = V_MUL_LO_I32 %15.sub1, target-flags(amdgpu-gotprel32-lo) 7, implicit $exec 81 S_CBRANCH_SCC1 %bb.11, implicit undef $scc 82 S_BRANCH %bb.9 83 84 bb.9: 85 successors: %bb.10(0x80000000) 86 %19:vreg_128 = BUFFER_LOAD_FORMAT_XYZW_IDXEN killed %18, undef %20:sreg_128, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 16 from constant-pool, align 1, addrspace 4) 87 %21:sreg_64 = V_CMP_NE_U32_e64 target-flags(amdgpu-gotprel) 0, killed %19.sub0, implicit $exec 88 %22:sreg_64 = COPY $exec, implicit-def $exec 89 %23:sreg_64 = S_AND_B64 %22, %21, implicit-def dead $scc 90 $exec = S_MOV_B64_term killed %23 91 92 bb.10: 93 successors: %bb.12(0x80000000) 94 $exec = S_OR_B64 $exec, killed %22, implicit-def $scc 95 S_BRANCH %bb.12 96 97 bb.11: 98 successors: %bb.13(0x80000000) 99 %24:vreg_128 = COPY killed %15 100 %24.sub0:vreg_128 = COPY undef %18 101 S_BRANCH %bb.13 102 103 bb.12: 104 successors: %bb.11(0x80000000) 105 S_BRANCH %bb.11 106 107 bb.13: 108 successors: %bb.7(0x80000000) 109 %16:vreg_128 = COPY killed %24 110 S_BRANCH %bb.7 111 112 bb.14: 113 successors: %bb.15(0x80000000) 114 S_CBRANCH_SCC1 %bb.15, implicit undef $scc 115 S_BRANCH %bb.15 116 117 bb.15: 118 undef %25.sub2:vreg_128 = COPY killed %17.sub2 119 %26:sreg_32_xm0 = S_MOV_B32 0 120 undef %27.sub0:sreg_256 = COPY %26 121 %27.sub1:sreg_256 = COPY %26 122 %27.sub2:sreg_256 = COPY %26 123 %27.sub3:sreg_256 = COPY %26 124 %27.sub4:sreg_256 = COPY %26 125 %27.sub5:sreg_256 = COPY %26 126 %27.sub6:sreg_256 = COPY %26 127 %27.sub7:sreg_256 = COPY killed %26 128 %28:vgpr_32 = IMAGE_LOAD_V1_V4 killed %25, killed %27, 2, -1, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 16 from constant-pool, addrspace 4) 129 %29:vgpr_32 = V_ADD_F32_e32 0, killed %28, implicit $exec 130 $m0 = S_MOV_B32 -1 131 DS_WRITE_B32 undef %30:vgpr_32, killed %29, 0, 0, implicit $m0, implicit $exec :: (store 4 into `i32 addrspace(3)* undef`, addrspace 3) 132 S_ENDPGM 133... 134