|
150
|
1 ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
|
|
|
2 ; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
|
|
|
3
|
|
|
4 ; Test that we correctly commute a sub instruction
|
|
|
5 ; FUNC-LABEL: {{^}}sub_rev:
|
|
|
6 ; SI-NOT: v_sub_i32_e32 v{{[0-9]+}}, vcc, s
|
|
|
7 ; SI: v_subrev_{{[iu]}}32_e32 v{{[0-9]+}}, vcc, s
|
|
|
8
|
|
|
9 ; ModuleID = 'vop-shrink.ll'
|
|
|
10
|
|
|
11 define amdgpu_kernel void @sub_rev(i32 addrspace(1)* %out, <4 x i32> %sgpr, i32 %cond) {
|
|
|
12 entry:
|
|
|
13 %vgpr = call i32 @llvm.amdgcn.workitem.id.x() #1
|
|
|
14 %tmp = icmp eq i32 %cond, 0
|
|
|
15 br i1 %tmp, label %if, label %else
|
|
|
16
|
|
|
17 if: ; preds = %entry
|
|
|
18 %tmp1 = getelementptr i32, i32 addrspace(1)* %out, i32 1
|
|
|
19 %tmp2 = extractelement <4 x i32> %sgpr, i32 1
|
|
|
20 store i32 %tmp2, i32 addrspace(1)* %out
|
|
|
21 br label %endif
|
|
|
22
|
|
|
23 else: ; preds = %entry
|
|
|
24 %tmp3 = extractelement <4 x i32> %sgpr, i32 2
|
|
|
25 %tmp4 = sub i32 %vgpr, %tmp3
|
|
|
26 store i32 %tmp4, i32 addrspace(1)* %out
|
|
|
27 br label %endif
|
|
|
28
|
|
|
29 endif: ; preds = %else, %if
|
|
|
30 ret void
|
|
|
31 }
|
|
|
32
|
|
|
33 ; Test that we fold an immediate that was illegal for a 64-bit op into the
|
|
|
34 ; 32-bit op when we shrink it.
|
|
|
35
|
|
|
36 ; FUNC-LABEL: {{^}}add_fold:
|
|
|
37 ; SI: v_add_f32_e32 v{{[0-9]+}}, 0x44800000
|
|
|
38 define amdgpu_kernel void @add_fold(float addrspace(1)* %out) {
|
|
|
39 entry:
|
|
|
40 %tmp = call i32 @llvm.amdgcn.workitem.id.x()
|
|
|
41 %tmp1 = uitofp i32 %tmp to float
|
|
|
42 %tmp2 = fadd float %tmp1, 1.024000e+03
|
|
|
43 store float %tmp2, float addrspace(1)* %out
|
|
|
44 ret void
|
|
|
45 }
|
|
|
46
|
|
|
47 ; Function Attrs: nounwind readnone
|
|
|
48 declare i32 @llvm.amdgcn.workitem.id.x() #0
|
|
|
49
|
|
|
50 attributes #0 = { nounwind readnone }
|
|
|
51 attributes #1 = { readnone }
|
