| Matt Arsenault | 3a99759 | 2014-09-26 01:09:46 +0000 | [diff] [blame] | 1 | ; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -strict-whitespace %s |
| Vincent Lejeune | d6cbede | 2013-10-13 17:56:28 +0000 | [diff] [blame] | 2 | |
| Tom Stellard | 79243d9 | 2014-10-01 17:15:17 +0000 | [diff] [blame] | 3 | ; CHECK-LABEL: {{^}}main: |
| Tom Stellard | 326d6ec | 2014-11-05 14:50:53 +0000 | [diff] [blame] | 4 | ; CHECK: s_load_dwordx4 |
| 5 | ; CHECK: s_load_dwordx4 |
| 6 | ; CHECK: s_waitcnt lgkmcnt(0){{$}} |
| 7 | ; CHECK: s_waitcnt vmcnt(0){{$}} |
| 8 | ; CHECK: s_waitcnt expcnt(0) lgkmcnt(0){{$}} |
| Matt Arsenault | c10853f | 2014-08-06 00:29:43 +0000 | [diff] [blame] | 9 | define void @main(<16 x i8> addrspace(2)* inreg %arg, <16 x i8> addrspace(2)* inreg %arg1, <32 x i8> addrspace(2)* inreg %arg2, <16 x i8> addrspace(2)* inreg %arg3, <16 x i8> addrspace(2)* inreg %arg4, i32 inreg %arg5, i32 %arg6, i32 %arg7, i32 %arg8, i32 %arg9, float addrspace(2)* inreg %constptr) #0 { |
| Vincent Lejeune | d6cbede | 2013-10-13 17:56:28 +0000 | [diff] [blame] | 10 | main_body: |
| Matt Arsenault | c10853f | 2014-08-06 00:29:43 +0000 | [diff] [blame] | 11 | %tmp = getelementptr <16 x i8> addrspace(2)* %arg3, i32 0 |
| 12 | %tmp10 = load <16 x i8> addrspace(2)* %tmp, !tbaa !0 |
| 13 | %tmp11 = call <4 x float> @llvm.SI.vs.load.input(<16 x i8> %tmp10, i32 0, i32 %arg6) |
| 14 | %tmp12 = extractelement <4 x float> %tmp11, i32 0 |
| 15 | %tmp13 = extractelement <4 x float> %tmp11, i32 1 |
| 16 | call void @llvm.AMDGPU.barrier.global() #1 |
| 17 | %tmp14 = extractelement <4 x float> %tmp11, i32 2 |
| 18 | ; %tmp15 = extractelement <4 x float> %tmp11, i32 3 |
| 19 | %tmp15 = load float addrspace(2)* %constptr, align 4 ; Force waiting for expcnt and lgkmcnt |
| 20 | %tmp16 = getelementptr <16 x i8> addrspace(2)* %arg3, i32 1 |
| 21 | %tmp17 = load <16 x i8> addrspace(2)* %tmp16, !tbaa !0 |
| 22 | %tmp18 = call <4 x float> @llvm.SI.vs.load.input(<16 x i8> %tmp17, i32 0, i32 %arg6) |
| 23 | %tmp19 = extractelement <4 x float> %tmp18, i32 0 |
| 24 | %tmp20 = extractelement <4 x float> %tmp18, i32 1 |
| 25 | %tmp21 = extractelement <4 x float> %tmp18, i32 2 |
| 26 | %tmp22 = extractelement <4 x float> %tmp18, i32 3 |
| 27 | call void @llvm.SI.export(i32 15, i32 0, i32 0, i32 32, i32 0, float %tmp19, float %tmp20, float %tmp21, float %tmp22) |
| 28 | call void @llvm.SI.export(i32 15, i32 0, i32 1, i32 12, i32 0, float %tmp12, float %tmp13, float %tmp14, float %tmp15) |
| Vincent Lejeune | d6cbede | 2013-10-13 17:56:28 +0000 | [diff] [blame] | 29 | ret void |
| 30 | } |
| 31 | |
| Matt Arsenault | c10853f | 2014-08-06 00:29:43 +0000 | [diff] [blame] | 32 | ; Function Attrs: noduplicate nounwind |
| 33 | declare void @llvm.AMDGPU.barrier.global() #1 |
| 34 | |
| Vincent Lejeune | d6cbede | 2013-10-13 17:56:28 +0000 | [diff] [blame] | 35 | ; Function Attrs: nounwind readnone |
| Matt Arsenault | c10853f | 2014-08-06 00:29:43 +0000 | [diff] [blame] | 36 | declare <4 x float> @llvm.SI.vs.load.input(<16 x i8>, i32, i32) #2 |
| Vincent Lejeune | d6cbede | 2013-10-13 17:56:28 +0000 | [diff] [blame] | 37 | |
| 38 | declare void @llvm.SI.export(i32, i32, i32, i32, i32, float, float, float, float) |
| 39 | |
| 40 | attributes #0 = { "ShaderType"="1" } |
| Matt Arsenault | c10853f | 2014-08-06 00:29:43 +0000 | [diff] [blame] | 41 | attributes #1 = { noduplicate nounwind } |
| 42 | attributes #2 = { nounwind readnone } |
| Vincent Lejeune | d6cbede | 2013-10-13 17:56:28 +0000 | [diff] [blame] | 43 | |
| Matt Arsenault | c10853f | 2014-08-06 00:29:43 +0000 | [diff] [blame] | 44 | !0 = metadata !{metadata !1, metadata !1, i64 0, i32 1} |
| 45 | !1 = metadata !{metadata !"const", null} |