Gitiles
Code Review Sign In
gerrit-public.fairphone.software / toolchain / llvm-project / a79ac14fa68297f9888bc70a10df5ed9b8864e38 / . / llvm / test / CodeGen / R600 / wait.ll
blob: 5cc7577cad3326386ecec362f892d62e2bb795df [file] [log] [blame]
Tom Stellard49f8bfd2015-01-06 18:00:21 +0000[diff] [blame]1; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs < %s | FileCheck -strict-whitespace %s
Marek Olsak75170772015-01-27 17:27:15 +0000[diff] [blame]2; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -strict-whitespace %s
Vincent Lejeuned6cbede2013-10-13 17:56:28 +0000[diff] [blame]3
Tom Stellard79243d92014-10-01 17:15:17 +0000[diff] [blame]4; CHECK-LABEL: {{^}}main:
Tom Stellard326d6ec2014-11-05 14:50:53 +0000[diff] [blame]5; CHECK: s_load_dwordx4
6; CHECK: s_load_dwordx4
Tom Stellard83f0bce2015-01-29 16:55:25 +0000[diff] [blame]7; CHECK: s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
8; CHECK: s_endpgm
Matt Arsenaultc10853f2014-08-06 00:29:43 +0000[diff] [blame]9define 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 Lejeuned6cbede2013-10-13 17:56:28 +0000[diff] [blame]10main_body:
David Blaikie79e6c742015-02-27 19:29:02 +0000[diff] [blame]11 %tmp = getelementptr <16 x i8>, <16 x i8> addrspace(2)* %arg3, i32 0
David Blaikiea79ac142015-02-27 21:17:42 +0000[diff] [blame^]12 %tmp10 = load <16 x i8>, <16 x i8> addrspace(2)* %tmp, !tbaa !0
Matt Arsenaultc10853f2014-08-06 00:29:43 +0000[diff] [blame]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
David Blaikiea79ac142015-02-27 21:17:42 +0000[diff] [blame^]19 %tmp15 = load float, float addrspace(2)* %constptr, align 4 ; Force waiting for expcnt and lgkmcnt
David Blaikie79e6c742015-02-27 19:29:02 +0000[diff] [blame]20 %tmp16 = getelementptr <16 x i8>, <16 x i8> addrspace(2)* %arg3, i32 1
David Blaikiea79ac142015-02-27 21:17:42 +0000[diff] [blame^]21 %tmp17 = load <16 x i8>, <16 x i8> addrspace(2)* %tmp16, !tbaa !0
Matt Arsenaultc10853f2014-08-06 00:29:43 +0000[diff] [blame]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 Lejeuned6cbede2013-10-13 17:56:28 +0000[diff] [blame]29 ret void
30}
31
Matt Arsenaultc10853f2014-08-06 00:29:43 +0000[diff] [blame]32; Function Attrs: noduplicate nounwind
33declare void @llvm.AMDGPU.barrier.global() #1
34
Vincent Lejeuned6cbede2013-10-13 17:56:28 +0000[diff] [blame]35; Function Attrs: nounwind readnone
Matt Arsenaultc10853f2014-08-06 00:29:43 +0000[diff] [blame]36declare <4 x float> @llvm.SI.vs.load.input(<16 x i8>, i32, i32) #2
Vincent Lejeuned6cbede2013-10-13 17:56:28 +0000[diff] [blame]37
38declare void @llvm.SI.export(i32, i32, i32, i32, i32, float, float, float, float)
39
40attributes #0 = { "ShaderType"="1" }
Matt Arsenaultc10853f2014-08-06 00:29:43 +0000[diff] [blame]41attributes #1 = { noduplicate nounwind }
42attributes #2 = { nounwind readnone }
Vincent Lejeuned6cbede2013-10-13 17:56:28 +0000[diff] [blame]43
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +0000[diff] [blame]44!0 = !{!1, !1, i64 0, i32 1}
45!1 = !{!"const", null}