1 ; RUN: opt -mtriple=amdgcn--amdhsa -S -amdgpu-inline -inline-threshold=0 < %s | FileCheck %s
3 target datalayout = "e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-S32-A5"
5 define void @use_flat_ptr_arg(float* nocapture %p) {
7 %tmp1 = load float, float* %p, align 4
8 %div = fdiv float 1.000000e+00, %tmp1
9 %add0 = fadd float %div, 1.0
10 %add1 = fadd float %add0, 1.0
11 %add2 = fadd float %add1, 1.0
12 %add3 = fadd float %add2, 1.0
13 %add4 = fadd float %add3, 1.0
14 %add5 = fadd float %add4, 1.0
15 %add6 = fadd float %add5, 1.0
16 %add7 = fadd float %add6, 1.0
17 %add8 = fadd float %add7, 1.0
18 %add9 = fadd float %add8, 1.0
19 %add10 = fadd float %add9, 1.0
20 store float %add10, float* %p, align 4
24 define void @use_private_ptr_arg(float addrspace(5)* nocapture %p) {
26 %tmp1 = load float, float addrspace(5)* %p, align 4
27 %div = fdiv float 1.000000e+00, %tmp1
28 %add0 = fadd float %div, 1.0
29 %add1 = fadd float %add0, 1.0
30 %add2 = fadd float %add1, 1.0
31 %add3 = fadd float %add2, 1.0
32 %add4 = fadd float %add3, 1.0
33 %add5 = fadd float %add4, 1.0
34 %add6 = fadd float %add5, 1.0
35 %add7 = fadd float %add6, 1.0
36 %add8 = fadd float %add7, 1.0
37 %add9 = fadd float %add8, 1.0
38 %add10 = fadd float %add9, 1.0
39 store float %add10, float addrspace(5)* %p, align 4
43 ; Test that the inline threshold is boosted if called with an
44 ; addrspacecasted' alloca.
45 ; CHECK-LABEL: @test_inliner_flat_ptr(
46 ; CHECK: call i32 @llvm.amdgcn.workitem.id.x()
49 define amdgpu_kernel void @test_inliner_flat_ptr(float addrspace(1)* nocapture %a, i32 %n) {
51 %pvt_arr = alloca [64 x float], align 4, addrspace(5)
52 %tid = tail call i32 @llvm.amdgcn.workitem.id.x()
53 %arrayidx = getelementptr inbounds float, float addrspace(1)* %a, i32 %tid
54 %tmp2 = load float, float addrspace(1)* %arrayidx, align 4
55 %add = add i32 %tid, 1
56 %arrayidx2 = getelementptr inbounds float, float addrspace(1)* %a, i32 %add
57 %tmp5 = load float, float addrspace(1)* %arrayidx2, align 4
59 %arrayidx5 = getelementptr inbounds [64 x float], [64 x float] addrspace(5)* %pvt_arr, i32 0, i32 %or
60 %arrayidx7 = getelementptr inbounds [64 x float], [64 x float] addrspace(5)* %pvt_arr, i32 0, i32 %or
61 %to.flat = addrspacecast float addrspace(5)* %arrayidx7 to float*
62 call void @use_private_ptr_arg(float addrspace(5)* %arrayidx7)
63 call void @use_flat_ptr_arg(float* %to.flat)
67 declare i32 @llvm.amdgcn.workitem.id.x() #1
69 attributes #0 = { noinline }
70 attributes #1 = { nounwind readnone }