1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -S -codegenprepare -mtriple=amdgcn--amdhsa < %s | FileCheck %s 3 4define amdgpu_kernel void @test_sink_as999_small_max_mubuf_offset(i32 addrspace(999)* %out, i8 addrspace(999)* %in) { 5; CHECK-LABEL: @test_sink_as999_small_max_mubuf_offset( 6; CHECK-NEXT: entry: 7; CHECK-NEXT: [[OUT_GEP:%.*]] = getelementptr i32, i32 addrspace(999)* [[OUT:%.*]], i32 1024 8; CHECK-NEXT: [[IN_GEP:%.*]] = getelementptr i8, i8 addrspace(999)* [[IN:%.*]], i64 4095 9; CHECK-NEXT: [[TID:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0) #1 10; CHECK-NEXT: [[TMP0:%.*]] = icmp eq i32 [[TID]], 0 11; CHECK-NEXT: br i1 [[TMP0]], label [[ENDIF:%.*]], label [[IF:%.*]] 12; CHECK: if: 13; CHECK-NEXT: [[TMP1:%.*]] = load i8, i8 addrspace(999)* [[IN_GEP]], align 1 14; CHECK-NEXT: [[TMP2:%.*]] = sext i8 [[TMP1]] to i32 15; CHECK-NEXT: br label [[ENDIF]] 16; CHECK: endif: 17; CHECK-NEXT: [[X:%.*]] = phi i32 [ [[TMP2]], [[IF]] ], [ 0, [[ENTRY:%.*]] ] 18; CHECK-NEXT: store i32 [[X]], i32 addrspace(999)* [[OUT_GEP]], align 4 19; CHECK-NEXT: br label [[DONE:%.*]] 20; CHECK: done: 21; CHECK-NEXT: ret void 22; 23entry: 24 %out.gep = getelementptr i32, i32 addrspace(999)* %out, i32 1024 25 %in.gep = getelementptr i8, i8 addrspace(999)* %in, i64 4095 26 %tid = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0) #0 27 %tmp0 = icmp eq i32 %tid, 0 28 br i1 %tmp0, label %endif, label %if 29 30if: 31 %tmp1 = load i8, i8 addrspace(999)* %in.gep 32 %tmp2 = sext i8 %tmp1 to i32 33 br label %endif 34 35endif: 36 %x = phi i32 [ %tmp2, %if ], [ 0, %entry ] 37 store i32 %x, i32 addrspace(999)* %out.gep 38 br label %done 39 40done: 41 ret void 42} 43 44declare i32 @llvm.amdgcn.mbcnt.lo(i32, i32) #0 45 46attributes #0 = { nounwind readnone } 47attributes #1 = { nounwind } 48attributes #2 = { nounwind argmemonly } 49