1 ; FIXME: The si scheduler crashes if when lane mask tracking is enabled, so
2 ; we need to disable this when the si scheduler is being used.
3 ; The only way the subtarget knows that the si machine scheduler is being used
4 ; is to specify -mattr=si-scheduler. If we just pass --misched=si, the backend
5 ; won't know what scheduler we are using.
6 ; RUN: llc -march=amdgcn -mcpu=tahiti --misched=si -mattr=si-scheduler < %s | FileCheck %s
7 ; RUN: llc -march=amdgcn -mcpu=gfx1010 --misched=si -mattr=si-scheduler < %s | FileCheck %s
9 ; The test checks the "si" machine scheduler pass works correctly.
11 ; CHECK-LABEL: {{^}}main:
13 ; CHECK: s_load_dwordx8
14 ; CHECK: s_load_dwordx4
15 ; CHECK: s_waitcnt lgkmcnt(0)
17 ; CHECK: s_waitcnt vmcnt(0)
20 define amdgpu_ps void @main(ptr addrspace(4) inreg %arg, ptr addrspace(4) inreg %arg1, ptr addrspace(4) inreg %arg2, ptr addrspace(4) inreg %arg3, float inreg %arg4, i32 inreg %arg5, <2 x i32> %arg6, <2 x i32> %arg7, <2 x i32> %arg8, <3 x i32> %arg9, <2 x i32> %arg10, <2 x i32> %arg11, <2 x i32> %arg12, float %arg13, float %arg14, float %arg15, float %arg16, float %arg17, float %arg18, i32 %arg19, float %arg20, float %arg21) #0 {
22 %tmp22 = load <32 x i8>, ptr addrspace(4) %arg3, align 32, !tbaa !0
23 %tmp24 = load <16 x i8>, ptr addrspace(4) %arg2, align 16, !tbaa !0
24 %i.i = extractelement <2 x i32> %arg11, i32 0
25 %j.i = extractelement <2 x i32> %arg11, i32 1
26 %i.f.i = bitcast i32 %i.i to float
27 %j.f.i = bitcast i32 %j.i to float
28 %p1.i = call float @llvm.amdgcn.interp.p1(float %i.f.i, i32 0, i32 0, i32 %arg5) #1
29 %p2.i = call float @llvm.amdgcn.interp.p2(float %p1.i, float %j.f.i, i32 0, i32 0, i32 %arg5) #1
30 %i.i1 = extractelement <2 x i32> %arg11, i32 0
31 %j.i2 = extractelement <2 x i32> %arg11, i32 1
32 %i.f.i3 = bitcast i32 %i.i1 to float
33 %j.f.i4 = bitcast i32 %j.i2 to float
34 %p1.i5 = call float @llvm.amdgcn.interp.p1(float %i.f.i3, i32 1, i32 0, i32 %arg5) #1
35 %p2.i6 = call float @llvm.amdgcn.interp.p2(float %p1.i5, float %j.f.i4, i32 1, i32 0, i32 %arg5) #1
36 %tmp22.bc = bitcast <32 x i8> %tmp22 to <8 x i32>
37 %tmp24.bc = bitcast <16 x i8> %tmp24 to <4 x i32>
38 %tmp31 = call <4 x float> @llvm.amdgcn.image.sample.2d.v4f32.f32(i32 15, float %p2.i, float %p2.i6, <8 x i32> %tmp22.bc, <4 x i32> %tmp24.bc, i1 0, i32 0, i32 0)
40 %tmp32 = extractelement <4 x float> %tmp31, i32 0
41 %tmp33 = extractelement <4 x float> %tmp31, i32 1
42 %tmp34 = extractelement <4 x float> %tmp31, i32 2
43 %tmp35 = extractelement <4 x float> %tmp31, i32 3
44 %tmp36 = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float %tmp32, float %tmp33)
45 %tmp38 = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float %tmp34, float %tmp35)
46 call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 15, <2 x half> %tmp36, <2 x half> %tmp38, i1 true, i1 false) #0
50 declare float @llvm.amdgcn.interp.p1(float, i32, i32, i32) #1
51 declare float @llvm.amdgcn.interp.p2(float, float, i32, i32, i32) #1
52 declare void @llvm.amdgcn.exp.compr.v2f16(i32, i32, <2 x half>, <2 x half>, i1, i1) #0
53 declare <2 x half> @llvm.amdgcn.cvt.pkrtz(float, float) #1
54 declare <4 x float> @llvm.amdgcn.image.sample.2d.v4f32.f32(i32, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #2
56 attributes #0 = { nounwind }
57 attributes #1 = { nounwind readnone }
58 attributes #2 = { nounwind readonly }
60 !0 = !{!1, !1, i64 0, i32 1}
65 ; CHECK-LABEL: amdgpu_ps_main:
66 ; CHECK: s_buffer_load_dword
67 define amdgpu_ps void @_amdgpu_ps_main(i32 %arg) local_unnamed_addr {
69 %tmp = insertelement <2 x i32> zeroinitializer, i32 %arg, i32 0
70 %tmp1 = bitcast <2 x i32> %tmp to i64
71 %tmp2 = inttoptr i64 %tmp1 to ptr addrspace(4)
72 %tmp3 = load <4 x i32>, ptr addrspace(4) %tmp2, align 16
73 %tmp4 = tail call i32 @llvm.amdgcn.s.buffer.load.i32(<4 x i32> %tmp3, i32 0, i32 0) #0
74 switch i32 %tmp4, label %bb [
82 bb5: ; preds = %.entry
85 bb6: ; preds = %.entry
89 ; Function Attrs: nounwind readnone
90 declare i32 @llvm.amdgcn.s.buffer.load.i32(<4 x i32>, i32, i32 immarg) #1