|
| 1 | +# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py |
| 2 | +# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx908 %s -o - -run-pass prologepilog -verify-machineinstrs | FileCheck -check-prefix=GFX908-PEI %s |
| 3 | +# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx908 %s -o - -run-pass prologepilog,machine-cp -verify-machineinstrs | FileCheck -check-prefix=GFX908-PEI-MACHINECP %s |
| 4 | + |
| 5 | +# When VGPRs are available for spilling, prologepilog marks the tuple implicit-def as well as implicit in the first spill instruction. |
| 6 | +# As a consequence, machine-cp would NOT delete agpr2 copy here. |
| 7 | + |
| 8 | +--- |
| 9 | +name: agpr-spill-to-vgpr-machine-cp |
| 10 | +tracksRegLiveness: true |
| 11 | +stack: |
| 12 | + - { id: 0, name: '', type: spill-slot, offset: 0, size: 128, alignment: 4 } |
| 13 | +machineFunctionInfo: |
| 14 | + scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3 |
| 15 | + stackPtrOffsetReg: '$sgpr32' |
| 16 | + hasSpilledVGPRs: true |
| 17 | +body: | |
| 18 | + bb.0: |
| 19 | + successors: |
| 20 | + liveins: $vgpr0, $vgpr1 |
| 21 | + ; GFX908-PEI-LABEL: name: agpr-spill-to-vgpr-machine-cp |
| 22 | + ; GFX908-PEI: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33 |
| 23 | + ; GFX908-PEI-NEXT: {{ $}} |
| 24 | + ; GFX908-PEI-NEXT: renamable $agpr0 = COPY renamable $vgpr0, implicit $exec |
| 25 | + ; GFX908-PEI-NEXT: renamable $agpr2 = COPY renamable $vgpr1, implicit $exec |
| 26 | + ; GFX908-PEI-NEXT: $vgpr33 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2, implicit $agpr0_agpr1_agpr2 |
| 27 | + ; GFX908-PEI-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec |
| 28 | + ; GFX908-PEI-NEXT: $vgpr31 = V_ACCVGPR_READ_B32_e64 $agpr2, implicit $exec, implicit $agpr0_agpr1_agpr2 |
| 29 | + ; GFX908-PEI-NEXT: S_ENDPGM 0 |
| 30 | + ; |
| 31 | + ; GFX908-PEI-MACHINECP-LABEL: name: agpr-spill-to-vgpr-machine-cp |
| 32 | + ; GFX908-PEI-MACHINECP: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33 |
| 33 | + ; GFX908-PEI-MACHINECP-NEXT: {{ $}} |
| 34 | + ; GFX908-PEI-MACHINECP-NEXT: renamable $agpr0 = COPY renamable $vgpr0, implicit $exec |
| 35 | + ; GFX908-PEI-MACHINECP-NEXT: renamable $agpr2 = COPY renamable $vgpr1, implicit $exec |
| 36 | + ; GFX908-PEI-MACHINECP-NEXT: $vgpr33 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2, implicit $agpr0_agpr1_agpr2 |
| 37 | + ; GFX908-PEI-MACHINECP-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec |
| 38 | + ; GFX908-PEI-MACHINECP-NEXT: $vgpr31 = V_ACCVGPR_READ_B32_e64 $agpr2, implicit $exec, implicit $agpr0_agpr1_agpr2 |
| 39 | + ; GFX908-PEI-MACHINECP-NEXT: S_ENDPGM 0 |
| 40 | + renamable $agpr0 = COPY renamable $vgpr0, implicit $exec |
| 41 | + renamable $agpr2 = COPY renamable $vgpr1, implicit $exec |
| 42 | + SI_SPILL_AV96_SAVE $agpr0_agpr1_agpr2, %stack.0, $sgpr32, 0, implicit $exec :: (store (s96) into %stack.0, align 4, addrspace 5) |
| 43 | + S_ENDPGM 0 |
| 44 | +... |
| 45 | + |
| 46 | +# When VGPRs are NOT available for spilling (stack is used), prologepilog marks the tuple implicit-def only and NOT implicit. |
| 47 | +# As a consequence, machine-cp would delete agpr2 copy here. Presently, this is incorrect behavior. |
| 48 | + |
| 49 | +--- |
| 50 | +name: agpr-spill-to-vgpr-to-stack-machine-cp |
| 51 | +tracksRegLiveness: true |
| 52 | +stack: |
| 53 | + - { id: 0, name: '', type: spill-slot, offset: 0, size: 128, alignment: 4 } |
| 54 | +machineFunctionInfo: |
| 55 | + scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3 |
| 56 | + stackPtrOffsetReg: '$sgpr32' |
| 57 | + hasSpilledVGPRs: true |
| 58 | +body: | |
| 59 | + bb.0: |
| 60 | + successors: |
| 61 | + liveins: $vgpr0, $vgpr1 |
| 62 | + ; GFX908-PEI-LABEL: name: agpr-spill-to-vgpr-to-stack-machine-cp |
| 63 | + ; GFX908-PEI: liveins: $vgpr0, $vgpr1, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33, $vgpr34, $vgpr35, $vgpr36, $vgpr37, $vgpr38, $vgpr39, $vgpr48, $vgpr49, $vgpr50, $vgpr51, $vgpr52, $vgpr53, $vgpr54, $vgpr55 |
| 64 | + ; GFX908-PEI-NEXT: {{ $}} |
| 65 | + ; GFX908-PEI-NEXT: renamable $agpr0 = COPY renamable $vgpr0, implicit $exec |
| 66 | + ; GFX908-PEI-NEXT: renamable $agpr2 = COPY renamable $vgpr1, implicit $exec |
| 67 | + ; GFX908-PEI-NEXT: $vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9 = IMPLICIT_DEF |
| 68 | + ; GFX908-PEI-NEXT: $vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17 = IMPLICIT_DEF |
| 69 | + ; GFX908-PEI-NEXT: $vgpr40 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2 |
| 70 | + ; GFX908-PEI-NEXT: BUFFER_STORE_DWORD_OFFSET $vgpr40, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0, addrspace 5) |
| 71 | + ; GFX908-PEI-NEXT: $vgpr40 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec |
| 72 | + ; GFX908-PEI-NEXT: BUFFER_STORE_DWORD_OFFSET $vgpr40, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, implicit $exec :: (store (s32) into %stack.0 + 4, addrspace 5) |
| 73 | + ; GFX908-PEI-NEXT: $vgpr55 = V_ACCVGPR_READ_B32_e64 $agpr2, implicit $exec, implicit $agpr0_agpr1_agpr2 |
| 74 | + ; GFX908-PEI-NEXT: S_ENDPGM 0 |
| 75 | + ; |
| 76 | + ; GFX908-PEI-MACHINECP-LABEL: name: agpr-spill-to-vgpr-to-stack-machine-cp |
| 77 | + ; GFX908-PEI-MACHINECP: liveins: $vgpr0, $vgpr1, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33, $vgpr34, $vgpr35, $vgpr36, $vgpr37, $vgpr38, $vgpr39, $vgpr48, $vgpr49, $vgpr50, $vgpr51, $vgpr52, $vgpr53, $vgpr54, $vgpr55 |
| 78 | + ; GFX908-PEI-MACHINECP-NEXT: {{ $}} |
| 79 | + ; GFX908-PEI-MACHINECP-NEXT: renamable $agpr0 = COPY renamable $vgpr0, implicit $exec |
| 80 | + ; GFX908-PEI-MACHINECP-NEXT: $vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9 = IMPLICIT_DEF |
| 81 | + ; GFX908-PEI-MACHINECP-NEXT: $vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17 = IMPLICIT_DEF |
| 82 | + ; GFX908-PEI-MACHINECP-NEXT: $vgpr40 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2 |
| 83 | + ; GFX908-PEI-MACHINECP-NEXT: BUFFER_STORE_DWORD_OFFSET $vgpr40, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0, addrspace 5) |
| 84 | + ; GFX908-PEI-MACHINECP-NEXT: $vgpr40 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec |
| 85 | + ; GFX908-PEI-MACHINECP-NEXT: BUFFER_STORE_DWORD_OFFSET $vgpr40, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, implicit $exec :: (store (s32) into %stack.0 + 4, addrspace 5) |
| 86 | + ; GFX908-PEI-MACHINECP-NEXT: $vgpr55 = V_ACCVGPR_READ_B32_e64 $agpr2, implicit $exec, implicit $agpr0_agpr1_agpr2 |
| 87 | + ; GFX908-PEI-MACHINECP-NEXT: S_ENDPGM 0 |
| 88 | + renamable $agpr0 = COPY renamable $vgpr0, implicit $exec |
| 89 | + renamable $agpr2 = COPY renamable $vgpr1, implicit $exec |
| 90 | + $vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9 = IMPLICIT_DEF |
| 91 | + $vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17 = IMPLICIT_DEF |
| 92 | + SI_SPILL_AV96_SAVE $agpr0_agpr1_agpr2, %stack.0, $sgpr32, 0, implicit $exec :: (store (s96) into %stack.0, align 4, addrspace 5) |
| 93 | + S_ENDPGM 0 |
| 94 | +... |
0 commit comments