79 lines · plain
1# RUN: llc -mtriple=amdgcn -mcpu=tonga -verify-machineinstrs -run-pass machine-scheduler -o - %s | FileCheck -check-prefix=GCN %s2# RUN: llc -mtriple=amdgcn -mcpu=tonga -passes=machine-scheduler -o - %s | FileCheck -check-prefix=GCN %s3 4# GCN-LABEL: name: flat_load_clustering5# GCN: FLAT_LOAD_DWORD6# GCN-NEXT: FLAT_LOAD_DWORD7--- |8 define amdgpu_kernel void @flat_load_clustering(ptr addrspace(1) nocapture %arg, ptr addrspace(4) nocapture readonly %arg1) {9 bb:10 %tid = tail call i32 @llvm.amdgcn.workitem.id.x()11 %idxprom = sext i32 %tid to i6412 %gep1 = getelementptr inbounds i32, ptr addrspace(4) %arg1, i64 %idxprom13 %load1 = load i32, ptr addrspace(4) %gep1, align 414 %gep2 = getelementptr inbounds i32, ptr addrspace(1) %arg, i64 %idxprom15 %gep34 = getelementptr inbounds i32, ptr addrspace(4) %gep1, i64 416 %load2 = load i32, ptr addrspace(4) %gep34, align 417 %gep4 = getelementptr inbounds i32, ptr addrspace(1) %gep2, i64 418 store i32 %load1, ptr addrspace(1) %gep2, align 419 store i32 %load2, ptr addrspace(1) %gep4, align 420 ret void21 }22 23 declare i32 @llvm.amdgcn.workitem.id.x()24 25...26---27name: flat_load_clustering28alignment: 129exposesReturnsTwice: false30legalized: false31regBankSelected: false32selected: false33tracksRegLiveness: true34registers:35 - { id: 0, class: vgpr_32 }36 - { id: 1, class: sgpr_64 }37 - { id: 2, class: vgpr_32 }38 - { id: 3, class: sreg_64_xexec }39 - { id: 4, class: sreg_64_xexec }40 - { id: 5, class: vgpr_32 }41 - { id: 6, class: vgpr_32 }42 - { id: 7, class: vgpr_32 }43 - { id: 8, class: vgpr_32 }44 - { id: 9, class: vreg_64 }45 - { id: 10, class: vreg_64 }46 - { id: 11, class: vgpr_32 }47 - { id: 12, class: vreg_64 }48 - { id: 13, class: vreg_64 }49liveins:50 - { reg: '$vgpr0', virtual-reg: '%0' }51 - { reg: '$sgpr4_sgpr5', virtual-reg: '%1' }52body: |53 bb.0.bb:54 liveins: $vgpr0, $sgpr4_sgpr555 56 %1 = COPY $sgpr4_sgpr557 %0 = COPY $vgpr058 %3 = S_LOAD_DWORDX2_IMM %1, 0, 0 :: (non-temporal dereferenceable invariant load (s64) from `ptr addrspace(4) poison`)59 %4 = S_LOAD_DWORDX2_IMM %1, 8, 0 :: (non-temporal dereferenceable invariant load (s64) from `ptr addrspace(4) poison`)60 %7 = V_LSHLREV_B32_e32 2, %0, implicit $exec61 %2 = V_MOV_B32_e32 0, implicit $exec62 undef %12.sub0 = V_ADD_CO_U32_e32 %4.sub0, %7, implicit-def $vcc, implicit $exec63 %11 = COPY %4.sub164 %12.sub1 = V_ADDC_U32_e32 %11, %2, implicit-def dead $vcc, implicit killed $vcc, implicit $exec65 %5 = FLAT_LOAD_DWORD %12, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %ir.gep1)66 undef %9.sub0 = V_ADD_CO_U32_e32 %3.sub0, %7, implicit-def $vcc, implicit $exec67 %8 = COPY %3.sub168 %9.sub1 = V_ADDC_U32_e32 %8, %2, implicit-def dead $vcc, implicit killed $vcc, implicit $exec69 undef %13.sub0 = V_ADD_CO_U32_e32 16, %12.sub0, implicit-def $vcc, implicit $exec70 %13.sub1 = V_ADDC_U32_e32 %12.sub1, %2, implicit-def dead $vcc, implicit killed $vcc, implicit $exec71 %6 = FLAT_LOAD_DWORD %13, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %ir.gep34)72 undef %10.sub0 = V_ADD_CO_U32_e32 16, %9.sub0, implicit-def $vcc, implicit $exec73 %10.sub1 = V_ADDC_U32_e32 %9.sub1, %2, implicit-def dead $vcc, implicit killed $vcc, implicit $exec74 FLAT_STORE_DWORD %9, %5, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %ir.gep2)75 FLAT_STORE_DWORD %10, %6, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %ir.gep4)76 S_ENDPGM 077 78...79