Home | History | Annotate | Download | only in AMDGPU
      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: (Id != S.end() && T != S.end() && T->valno == Id->valno), function pruneSubRegValues, file ../lib/CodeGen/RegisterCoalescer.cpp, line 2870.
      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   ; Function Attrs: nounwind
     13   define amdgpu_cs void @_amdgpu_cs_main(<3 x i32> %arg) #0 {
     14     ret void
     15   }
     16 
     17   attributes #0 = { nounwind "target-cpu"="gfx803" }
     18 ...
     19 
     20 ---
     21 name:            _amdgpu_cs_main
     22 tracksRegLiveness: true
     23 liveins:
     24   - { reg: '$vgpr0', virtual-reg: '%0' }
     25 body: |
     26   bb.0:
     27     successors: %bb.1(0x40000000), %bb.21(0x40000000)
     28     liveins: $vgpr0, $vgpr1, $vgpr2
     29     %0:vgpr_32 = COPY killed $vgpr0
     30     S_CBRANCH_SCC1 %bb.21, implicit undef $scc
     31 
     32   bb.1:
     33     successors: %bb.2(0x40000000), %bb.17(0x40000000)
     34     S_CBRANCH_SCC1 %bb.17, implicit undef $scc
     35 
     36   bb.2:
     37     successors: %bb.4(0x40000000), %bb.3(0x40000000)
     38     %1:sreg_32_xm0 = S_MOV_B32 0
     39     %2:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
     40     undef %3.sub0:vreg_128 = COPY killed %0
     41     %3.sub2:vreg_128 = COPY killed %2
     42     undef %4.sub0:sreg_256 = COPY %1
     43     %4.sub1:sreg_256 = COPY %1
     44     %4.sub2:sreg_256 = COPY %1
     45     %4.sub3:sreg_256 = COPY %1
     46     %4.sub4:sreg_256 = COPY %1
     47     %4.sub5:sreg_256 = COPY %1
     48     %4.sub6:sreg_256 = COPY %1
     49     %4.sub7:sreg_256 = COPY killed %1
     50     %5:vgpr_32 = IMAGE_LOAD_V1_V4 killed %3, killed %4, 1, -1, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 16 from constant-pool, addrspace 4)
     51     %6:vgpr_32 = V_MAD_F32 0, killed %5, 0, 0, 0, 0, 0, 0, implicit $exec
     52     %7:vgpr_32 = V_RCP_F32_e32 killed %6, implicit $exec
     53     %8:vgpr_32 = V_MUL_F32_e32 0, killed %7, implicit $exec
     54     %9:vgpr_32 = V_MAD_F32 0, killed %8, 0, 0, 0, 0, 0, 0, implicit $exec
     55     dead %10:vgpr_32 = V_MAC_F32_e32 undef %11:vgpr_32, undef %12:vgpr_32, undef %10, implicit $exec
     56     undef %13.sub0:vreg_128 = COPY %9
     57     %14:vgpr_32 = V_MOV_B32_e32 -1, implicit $exec
     58     S_CBRANCH_SCC0 %bb.4, implicit undef $scc
     59 
     60   bb.3:
     61     successors: %bb.6(0x80000000)
     62     %15:vreg_128 = IMPLICIT_DEF
     63     %16:vreg_1 = COPY killed %14
     64     S_BRANCH %bb.6
     65 
     66   bb.4:
     67     successors: %bb.5(0x40000000), %bb.7(0x40000000)
     68     %17:vgpr_32 = V_MAD_F32 0, killed %9, 0, 0, 0, 0, 0, 0, implicit $exec
     69     %18:vgpr_32 = V_MIN_F32_e32 1065353216, killed %17, implicit $exec
     70     %19:sreg_64_xexec = V_CMP_NEQ_F32_e64 0, 1065353216, 0, killed %18, 0, implicit $exec
     71     %20:vgpr_32 = V_MOV_B32_e32 2143289344, implicit $exec
     72     %21:vgpr_32 = V_CNDMASK_B32_e64 0, killed %20, killed %19, implicit $exec
     73     %22:sreg_64 = V_CMP_LT_F32_e64 0, 0, 0, killed %21, 0, implicit $exec
     74     %23:sreg_64 = COPY $exec, implicit-def $exec
     75     %24:sreg_64 = S_AND_B64 %23, %22, implicit-def dead $scc
     76     $exec = S_MOV_B64_term killed %24
     77     SI_MASK_BRANCH %bb.7, implicit $exec
     78     S_BRANCH %bb.5
     79 
     80   bb.5:
     81     successors: %bb.7(0x80000000)
     82     S_BRANCH %bb.7
     83 
     84   bb.6:
     85     successors: %bb.8(0x40000000), %bb.10(0x40000000)
     86     %25:vreg_1 = COPY killed %16
     87     %26:vreg_128 = COPY killed %15
     88     %27:sreg_64 = V_CMP_NE_U32_e64 0, killed %25, implicit $exec
     89     %28:sreg_64 = S_AND_B64 $exec, killed %27, implicit-def dead $scc
     90     $vcc = COPY killed %28
     91     %29:vreg_128 = COPY killed %26
     92     S_CBRANCH_VCCNZ %bb.8, implicit killed $vcc
     93     S_BRANCH %bb.10
     94 
     95   bb.7:
     96     successors: %bb.6(0x80000000)
     97     $exec = S_OR_B64 $exec, killed %23, implicit-def $scc
     98     %30:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
     99     %15:vreg_128 = COPY %13
    100     %16:vreg_1 = COPY killed %30
    101     S_BRANCH %bb.6
    102 
    103   bb.8:
    104     successors: %bb.9(0x40000000), %bb.11(0x40000000)
    105     %31:vreg_128 = COPY killed %13
    106     S_CBRANCH_SCC1 %bb.11, implicit undef $scc
    107     S_BRANCH %bb.9
    108 
    109   bb.9:
    110     successors: %bb.11(0x80000000)
    111     %32:sreg_32_xm0 = S_MOV_B32 0
    112     undef %33.sub0:sreg_128 = COPY %32
    113     %33.sub1:sreg_128 = COPY %32
    114     %33.sub2:sreg_128 = COPY %32
    115     %33.sub3:sreg_128 = COPY killed %32
    116     %34:sreg_128 = COPY killed %33
    117     %35:vreg_128 = COPY killed %34
    118     %31:vreg_128 = COPY killed %35
    119     S_BRANCH %bb.11
    120 
    121   bb.10:
    122     successors: %bb.14(0x80000000)
    123     %36:vreg_128 = COPY killed %29
    124     S_BRANCH %bb.14
    125 
    126   bb.11:
    127     successors: %bb.13(0x40000000), %bb.12(0x40000000)
    128     %37:vreg_128 = COPY killed %31
    129     S_CBRANCH_SCC0 %bb.13, implicit undef $scc
    130 
    131   bb.12:
    132     successors: %bb.10(0x80000000)
    133     %29:vreg_128 = COPY killed %37
    134     S_BRANCH %bb.10
    135 
    136   bb.13:
    137     successors: %bb.10(0x80000000)
    138     %29:vreg_128 = COPY killed %37
    139     S_BRANCH %bb.10
    140 
    141   bb.14:
    142     successors: %bb.15(0x40000000), %bb.16(0x40000000)
    143     %38:vgpr_32 = V_MAD_F32 0, killed %36.sub0, 0, target-flags(amdgpu-gotprel) 0, 0, 0, 0, 0, implicit $exec
    144     %39:vgpr_32 = V_MAD_F32 0, killed %38, 0, 0, 0, 0, 0, 0, implicit $exec
    145     %40:vgpr_32 = V_MAD_F32 0, killed %39, 0, -1090519040, 0, 1056964608, 0, 0, implicit $exec
    146     %41:vgpr_32 = V_MAD_F32 0, killed %40, 0, 0, 0, -1090519040, 0, 0, implicit $exec
    147     %42:vgpr_32 = V_CVT_I32_F32_e32 killed %41, implicit $exec
    148     %43:sreg_32_xm0_xexec = S_BUFFER_LOAD_DWORD_IMM undef %44:sreg_128, 12, 0 :: (dereferenceable invariant load 4)
    149     %45:vgpr_32 = V_MUL_LO_I32 killed %42, killed %43, implicit $exec
    150     %46:vgpr_32 = V_LSHLREV_B32_e32 2, killed %45, implicit $exec
    151     %47:vgpr_32 = BUFFER_LOAD_FORMAT_X_IDXEN killed %46, undef %48:sreg_128, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4 from constant-pool, align 1, addrspace 4)
    152     %49:sreg_64 = V_CMP_NE_U32_e64 0, killed %47, implicit $exec
    153     %50:sreg_64 = COPY $exec, implicit-def $exec
    154     %51:sreg_64 = S_AND_B64 %50, %49, implicit-def dead $scc
    155     $exec = S_MOV_B64_term killed %51
    156     SI_MASK_BRANCH %bb.16, implicit $exec
    157     S_BRANCH %bb.15
    158 
    159   bb.15:
    160     successors: %bb.16(0x80000000)
    161 
    162   bb.16:
    163     successors: %bb.17(0x80000000)
    164     $exec = S_OR_B64 $exec, killed %50, implicit-def $scc
    165     S_BRANCH %bb.17
    166 
    167   bb.17:
    168     successors: %bb.21(0x40000000), %bb.18(0x40000000)
    169     S_CBRANCH_SCC1 %bb.21, implicit undef $scc
    170 
    171   bb.18:
    172     successors: %bb.19(0x40000000), %bb.20(0x40000000)
    173     S_CBRANCH_SCC1 %bb.19, implicit undef $scc
    174     S_BRANCH %bb.20
    175 
    176   bb.19:
    177     successors: %bb.20(0x80000000)
    178 
    179   bb.20:
    180     successors: %bb.21(0x80000000)
    181 
    182   bb.21:
    183     S_ENDPGM
    184 ...
    185