1 ; RUN: llc < %s -march=r600 -mcpu=verde -verify-machineinstrs | FileCheck -check-prefix=SI %s 2 3 ; SI-LABEL: @vector_umax 4 ; SI: V_MAX_U32_e32 5 define void @vector_umax(i32 %p0, i32 %p1, i32 addrspace(1)* %in) #0 { 6 main_body: 7 %load = load i32 addrspace(1)* %in, align 4 8 %max = call i32 @llvm.AMDGPU.umax(i32 %p0, i32 %load) 9 %bc = bitcast i32 %max to float 10 call void @llvm.SI.export(i32 15, i32 1, i32 1, i32 0, i32 0, float %bc, float %bc, float %bc, float %bc) 11 ret void 12 } 13 14 ; SI-LABEL: @scalar_umax 15 ; SI: S_MAX_U32 16 define void @scalar_umax(i32 %p0, i32 %p1) #0 { 17 entry: 18 %max = call i32 @llvm.AMDGPU.umax(i32 %p0, i32 %p1) 19 %bc = bitcast i32 %max to float 20 call void @llvm.SI.export(i32 15, i32 1, i32 1, i32 0, i32 0, float %bc, float %bc, float %bc, float %bc) 21 ret void 22 } 23 24 ; SI-LABEL: @trunc_zext_umax 25 ; SI: BUFFER_LOAD_UBYTE [[VREG:v[0-9]+]], 26 ; SI: V_MAX_U32_e32 [[RESULT:v[0-9]+]], 0, [[VREG]] 27 ; SI-NOT: AND 28 ; SI: BUFFER_STORE_SHORT [[RESULT]], 29 define void @trunc_zext_umax(i16 addrspace(1)* nocapture %out, i8 addrspace(1)* nocapture %src) nounwind { 30 %tmp5 = load i8 addrspace(1)* %src, align 1 31 %tmp2 = zext i8 %tmp5 to i32 32 %tmp3 = tail call i32 @llvm.AMDGPU.umax(i32 %tmp2, i32 0) nounwind readnone 33 %tmp4 = trunc i32 %tmp3 to i8 34 %tmp6 = zext i8 %tmp4 to i16 35 store i16 %tmp6, i16 addrspace(1)* %out, align 2 36 ret void 37 } 38 39 ; Function Attrs: readnone 40 declare i32 @llvm.AMDGPU.umax(i32, i32) #1 41 42 declare void @llvm.SI.export(i32, i32, i32, i32, i32, float, float, float, float) 43 44 attributes #0 = { nounwind } 45 attributes #1 = { nounwind readnone } 46 47 !0 = metadata !{metadata !"const", null, i32 1} 48
