1 ; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=CI %s 2 3 ; The memory operand was dropped from the buffer_load_dword_offset 4 ; when replaced with the addr64 during operand legalization, resulting 5 ; in the global loads not being scheduled together. 6 7 ; GCN-LABEL: {{^}}reschedule_global_load_lds_store: 8 ; GCN: buffer_load_dword 9 ; GCN: buffer_load_dword 10 ; GCN: ds_write_b32 11 ; GCN: ds_write_b32 12 ; GCN: s_endpgm 13 define void @reschedule_global_load_lds_store(i32 addrspace(1)* noalias %gptr0, i32 addrspace(1)* noalias %gptr1, i32 addrspace(3)* noalias %lptr, i32 %c) #0 { 14 entry: 15 %tid = tail call i32 @llvm.r600.read.tidig.x() #1 16 %idx = shl i32 %tid, 2 17 %gep0 = getelementptr i32, i32 addrspace(1)* %gptr0, i32 %idx 18 %gep1 = getelementptr i32, i32 addrspace(1)* %gptr1, i32 %idx 19 %gep2 = getelementptr i32, i32 addrspace(3)* %lptr, i32 %tid 20 %cmp0 = icmp eq i32 %c, 0 21 br i1 %cmp0, label %for.body, label %exit 22 23 for.body: ; preds = %for.body, %entry 24 %i = phi i32 [ 0, %entry ], [ %i.inc, %for.body ] 25 %gptr0.phi = phi i32 addrspace(1)* [ %gep0, %entry ], [ %gep0.inc, %for.body ] 26 %gptr1.phi = phi i32 addrspace(1)* [ %gep1, %entry ], [ %gep1.inc, %for.body ] 27 %lptr0.phi = phi i32 addrspace(3)* [ %gep2, %entry ], [ %gep2.inc, %for.body ] 28 %lptr1 = getelementptr i32, i32 addrspace(3)* %lptr0.phi, i32 1 29 %val0 = load i32, i32 addrspace(1)* %gep0 30 store i32 %val0, i32 addrspace(3)* %lptr0.phi 31 %val1 = load i32, i32 addrspace(1)* %gep1 32 store i32 %val1, i32 addrspace(3)* %lptr1 33 %gep0.inc = getelementptr i32, i32 addrspace(1)* %gptr0.phi, i32 4 34 %gep1.inc = getelementptr i32, i32 addrspace(1)* %gptr1.phi, i32 4 35 %gep2.inc = getelementptr i32, i32 addrspace(3)* %lptr0.phi, i32 4 36 %i.inc = add nsw i32 %i, 1 37 %cmp1 = icmp ne i32 %i, 256 38 br i1 %cmp1, label %for.body, label %exit 39 40 exit: ; preds = %for.body, %entry 41 ret void 42 } 43 44 ; Function Attrs: nounwind readnone 45 declare i32 @llvm.r600.read.tidig.x() #1 46 47 ; Function Attrs: nounwind readnone 48 declare i32 @llvm.r600.read.tgid.x() #1 49 50 attributes #0 = { nounwind } 51 attributes #1 = { nounwind readnone } 52 attributes #2 = { convergent nounwind } 53
