| # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 6 |
| # RUN: llc -mtriple=amdgpu10.30 -verify-machineinstrs -run-pass prolog-epilog -o - %s | FileCheck %s |
| # s_cselect_b32 does not allow vreg & should use the sreg frameindex generated |
| # by v_readfirstlane_b32 in eliminateFrameIndex. |
| --- |
| name: test_s_cselect_b32 |
| tracksRegLiveness: true |
| stack: |
| - { id: 0, size: 12} |
| machineFunctionInfo: |
| scratchRSrcReg: '$sgpr0_sgpr1_sgpr2_sgpr3' |
| frameOffsetReg: '$sgpr33' |
| stackPtrOffsetReg: '$sgpr32' |
| body: | |
| ; CHECK-LABEL: name: test_s_cselect_b32 |
| ; CHECK: bb.0: |
| ; CHECK-NEXT: successors: %bb.1(0x80000000) |
| ; CHECK-NEXT: liveins: $vgpr41 |
| ; CHECK-NEXT: {{ $}} |
| ; CHECK-NEXT: frame-setup CFI_INSTRUCTION llvm_def_aspace_cfa $sgpr32, 0, 6 |
| ; CHECK-NEXT: frame-setup CFI_INSTRUCTION llvm_register_pair $pc_reg, $sgpr30, 32, $sgpr31, 32 |
| ; CHECK-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr4 |
| ; CHECK-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr5 |
| ; CHECK-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr6 |
| ; CHECK-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr41, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: ("amdgpu-thread-private" store (s32) into %stack.1, addrspace 5) |
| ; CHECK-NEXT: frame-setup CFI_INSTRUCTION llvm_vector_offset $vgpr41, 32, $exec_lo, 32, 0 |
| ; CHECK-NEXT: renamable $vgpr41 = V_MOV_B32_e32 0, implicit $exec |
| ; CHECK-NEXT: renamable $sgpr4 = S_MOV_B32 0 |
| ; CHECK-NEXT: renamable $sgpr5 = S_LSHR_B32 $sgpr32, 5, implicit-def dead $scc |
| ; CHECK-NEXT: renamable $sgpr54 = S_ADD_I32 $sgpr5, 8, implicit-def dead $scc |
| ; CHECK-NEXT: {{ $}} |
| ; CHECK-NEXT: bb.1: |
| ; CHECK-NEXT: successors: %bb.2(0x80000000) |
| ; CHECK-NEXT: liveins: $sgpr4, $sgpr54:0x000000000000000F |
| ; CHECK-NEXT: {{ $}} |
| ; CHECK-NEXT: S_CMP_EQ_U32 killed renamable $sgpr4, 0, implicit-def $scc |
| ; CHECK-NEXT: $vgpr0 = V_LSHRREV_B32_e64 5, $sgpr32, implicit $exec |
| ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 4, killed $vgpr0, implicit $exec |
| ; CHECK-NEXT: $sgpr5 = V_READFIRSTLANE_B32 $vgpr0, implicit $exec |
| ; CHECK-NEXT: renamable $sgpr4 = S_CSELECT_B32 0, killed $sgpr5, implicit $scc |
| ; CHECK-NEXT: $vgpr0 = V_LSHRREV_B32_e64 5, $sgpr32, implicit $exec |
| ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 4, killed $vgpr0, implicit $exec |
| ; CHECK-NEXT: $sgpr4 = V_READFIRSTLANE_B32 $vgpr0, implicit $exec |
| ; CHECK-NEXT: renamable $sgpr6 = S_CSELECT_B32 killed $sgpr4, 0, implicit $scc |
| ; CHECK-NEXT: renamable $sgpr5 = S_CSELECT_B32 4, renamable $sgpr54, implicit $scc |
| ; CHECK-NEXT: S_BRANCH %bb.2 |
| ; CHECK-NEXT: {{ $}} |
| ; CHECK-NEXT: bb.2: |
| ; CHECK-NEXT: $vgpr41 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: ("amdgpu-thread-private" load (s32) from %stack.1, addrspace 5) |
| ; CHECK-NEXT: SI_RETURN |
| bb.0: |
| renamable $vgpr41 = V_MOV_B32_e32 0, implicit $exec |
| renamable $sgpr4 = S_MOV_B32 0 |
| renamable $sgpr54 = S_ADD_I32 %stack.0, 4, implicit-def dead $scc |
| bb.1: |
| liveins: $sgpr4, $sgpr54:0x000000000000000F |
| S_CMP_EQ_U32 killed renamable $sgpr4, 0, implicit-def $scc |
| renamable $sgpr4 = S_CSELECT_B32 0, %stack.0, implicit $scc |
| renamable $sgpr6 = S_CSELECT_B32 %stack.0, 0, implicit $scc |
| renamable $sgpr5 = S_CSELECT_B32 4, renamable $sgpr54, implicit $scc |
| S_BRANCH %bb.2 |
| bb.2: |
| SI_RETURN |
| ... |
| # ensure register constraints of V_CNDMASK_B32_e32 are respected by |
| # eliminateFrameIndex. |
| --- |
| name: test_v_cndmask_e32 |
| tracksRegLiveness: true |
| stack: |
| - { id: 0, size: 12} |
| machineFunctionInfo: |
| scratchRSrcReg: '$sgpr0_sgpr1_sgpr2_sgpr3' |
| frameOffsetReg: '$sgpr33' |
| stackPtrOffsetReg: '$sgpr32' |
| body: | |
| bb.0: |
| liveins: $sgpr4 |
| ; CHECK-LABEL: name: test_v_cndmask_e32 |
| ; CHECK: liveins: $sgpr4 |
| ; CHECK-NEXT: {{ $}} |
| ; CHECK-NEXT: frame-setup CFI_INSTRUCTION llvm_def_aspace_cfa $sgpr32, 0, 6 |
| ; CHECK-NEXT: frame-setup CFI_INSTRUCTION llvm_register_pair $pc_reg, $sgpr30, 32, $sgpr31, 32 |
| ; CHECK-NEXT: frame-setup CFI_INSTRUCTION undefined $vgpr0 |
| ; CHECK-NEXT: frame-setup CFI_INSTRUCTION undefined $vgpr2 |
| ; CHECK-NEXT: $vcc_lo = V_CMP_EQ_U32_e64 killed $sgpr4, 0, implicit $exec |
| ; CHECK-NEXT: renamable $vgpr2 = V_MOV_B32_e32 0, implicit $exec |
| ; CHECK-NEXT: $vgpr1 = V_LSHRREV_B32_e64 5, $sgpr32, implicit $exec |
| ; CHECK-NEXT: renamable $vgpr0 = V_CNDMASK_B32_e32 killed $vgpr1, killed $vgpr2, implicit $vcc_lo, implicit $vcc_lo, implicit $exec |
| ; CHECK-NEXT: SI_RETURN |
| V_CMP_EQ_U32_e64 def $vcc_lo, killed $sgpr4, 0, implicit $exec |
| renamable $vgpr2 = V_MOV_B32_e32 0, implicit $exec |
| renamable $vgpr0 = V_CNDMASK_B32_e32 %stack.0, killed $vgpr2, implicit $vcc_lo, implicit $vcc, implicit $exec |
| SI_RETURN |
| ... |
| # ensure register constraints of V_CNDMASK_B32_e64 are respected by |
| # eliminateFrameIndex. |
| --- |
| name: test_v_cndmask_e64 |
| tracksRegLiveness: true |
| stack: |
| - { id: 0, size: 12} |
| machineFunctionInfo: |
| scratchRSrcReg: '$sgpr0_sgpr1_sgpr2_sgpr3' |
| frameOffsetReg: '$sgpr33' |
| stackPtrOffsetReg: '$sgpr32' |
| body: | |
| bb.0: |
| liveins: $sgpr4 |
| ; CHECK-LABEL: name: test_v_cndmask_e64 |
| ; CHECK: liveins: $sgpr4 |
| ; CHECK-NEXT: {{ $}} |
| ; CHECK-NEXT: frame-setup CFI_INSTRUCTION llvm_def_aspace_cfa $sgpr32, 0, 6 |
| ; CHECK-NEXT: frame-setup CFI_INSTRUCTION llvm_register_pair $pc_reg, $sgpr30, 32, $sgpr31, 32 |
| ; CHECK-NEXT: frame-setup CFI_INSTRUCTION undefined $vgpr0 |
| ; CHECK-NEXT: frame-setup CFI_INSTRUCTION undefined $vgpr2 |
| ; CHECK-NEXT: renamable $vgpr2 = V_MOV_B32_e32 0, implicit $exec |
| ; CHECK-NEXT: $vgpr1 = V_LSHRREV_B32_e64 5, $sgpr32, implicit $exec |
| ; CHECK-NEXT: renamable $vgpr0 = V_CNDMASK_B32_e64 0, killed $vgpr1, 0, killed $vgpr2, $sgpr4, implicit $exec |
| ; CHECK-NEXT: SI_RETURN |
| renamable $vgpr2 = V_MOV_B32_e32 0, implicit $exec |
| renamable $vgpr0 = V_CNDMASK_B32_e64 0, %stack.0, 0, killed $vgpr2, $sgpr4, implicit $exec |
| SI_RETURN |
| ... |