49 lines · plain
1; RUN: llc -mtriple=amdgcn -amdgpu-scalar-ir-passes=false < %s | FileCheck -check-prefix=GCN %s2; RUN: llc -mtriple=amdgcn -mcpu=tonga -mattr=-flat-for-global -amdgpu-scalar-ir-passes=false < %s | FileCheck -check-prefix=GCN %s3; RUN: llc -mtriple=amdgcn -mcpu=gfx900 -mattr=-flat-for-global -amdgpu-scalar-ir-passes=false < %s | FileCheck -check-prefix=GCN %s4 5; These tests are split out from umed3.ll and smed3.ll and use the6; -amdgpu-scalar-ir-passes=false flag, because InstSimplify would constant7; fold these functions otherwise.8 9declare i32 @llvm.amdgcn.workitem.id.x() nounwind readnone10 11; GCN-LABEL: {{^}}v_test_umed3_r_i_i_constant_order_i32:12; GCN: v_max_u32_e32 v{{[0-9]+}}, 17, v{{[0-9]+}}13; GCN: v_min_u32_e32 v{{[0-9]+}}, 12, v{{[0-9]+}}14define amdgpu_kernel void @v_test_umed3_r_i_i_constant_order_i32(ptr addrspace(1) %out, ptr addrspace(1) %aptr) #1 {15 %tid = call i32 @llvm.amdgcn.workitem.id.x()16 %gep0 = getelementptr i32, ptr addrspace(1) %aptr, i32 %tid17 %outgep = getelementptr i32, ptr addrspace(1) %out, i32 %tid18 %a = load i32, ptr addrspace(1) %gep019 20 %icmp0 = icmp ugt i32 %a, 1721 %i0 = select i1 %icmp0, i32 %a, i32 1722 23 %icmp1 = icmp ult i32 %i0, 1224 %i1 = select i1 %icmp1, i32 %i0, i32 1225 26 store i32 %i1, ptr addrspace(1) %outgep27 ret void28}29 30; GCN-LABEL: {{^}}v_test_smed3_r_i_i_constant_order_i32:31; GCN: v_max_i32_e32 v{{[0-9]+}}, 17, v{{[0-9]+}}32; GCN: v_min_i32_e32 v{{[0-9]+}}, 12, v{{[0-9]+}}33define amdgpu_kernel void @v_test_smed3_r_i_i_constant_order_i32(ptr addrspace(1) %out, ptr addrspace(1) %aptr) #1 {34 %tid = call i32 @llvm.amdgcn.workitem.id.x()35 %gep0 = getelementptr i32, ptr addrspace(1) %aptr, i32 %tid36 %outgep = getelementptr i32, ptr addrspace(1) %out, i32 %tid37 %a = load i32, ptr addrspace(1) %gep038 39 %icmp0 = icmp sgt i32 %a, 1740 %i0 = select i1 %icmp0, i32 %a, i32 1741 42 %icmp1 = icmp slt i32 %i0, 1243 %i1 = select i1 %icmp1, i32 %i0, i32 1244 45 store i32 %i1, ptr addrspace(1) %outgep46 ret void47}48 49