xref: /aosp_15_r20/external/llvm/test/CodeGen/NVPTX/nounroll.ll (revision 9880d6810fe72a1726cb53787c6711e909410d58)
1*9880d681SAndroid Build Coastguard Worker; RUN: llc < %s -march=nvptx64 -mcpu=sm_20 | FileCheck %s
2*9880d681SAndroid Build Coastguard Worker
3*9880d681SAndroid Build Coastguard Workertarget datalayout = "e-i64:64-v16:16-v32:32-n16:32:64"
4*9880d681SAndroid Build Coastguard Workertarget triple = "nvptx64-unknown-unknown"
5*9880d681SAndroid Build Coastguard Worker
6*9880d681SAndroid Build Coastguard Worker; Compiled from the following CUDA code:
7*9880d681SAndroid Build Coastguard Worker;
8*9880d681SAndroid Build Coastguard Worker;   #pragma nounroll
9*9880d681SAndroid Build Coastguard Worker;   for (int i = 0; i < 2; ++i)
10*9880d681SAndroid Build Coastguard Worker;     output[i] = input[i];
11*9880d681SAndroid Build Coastguard Workerdefine void @nounroll(float* %input, float* %output) {
12*9880d681SAndroid Build Coastguard Worker; CHECK-LABEL: .visible .func nounroll(
13*9880d681SAndroid Build Coastguard Workerentry:
14*9880d681SAndroid Build Coastguard Worker  br label %for.body
15*9880d681SAndroid Build Coastguard Worker
16*9880d681SAndroid Build Coastguard Workerfor.body:
17*9880d681SAndroid Build Coastguard Worker; CHECK: .pragma "nounroll"
18*9880d681SAndroid Build Coastguard Worker  %i.06 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
19*9880d681SAndroid Build Coastguard Worker  %idxprom = sext i32 %i.06 to i64
20*9880d681SAndroid Build Coastguard Worker  %arrayidx = getelementptr inbounds float, float* %input, i64 %idxprom
21*9880d681SAndroid Build Coastguard Worker  %0 = load float, float* %arrayidx, align 4
22*9880d681SAndroid Build Coastguard Worker; CHECK: ld.f32
23*9880d681SAndroid Build Coastguard Worker  %arrayidx2 = getelementptr inbounds float, float* %output, i64 %idxprom
24*9880d681SAndroid Build Coastguard Worker  store float %0, float* %arrayidx2, align 4
25*9880d681SAndroid Build Coastguard Worker; CHECK: st.f32
26*9880d681SAndroid Build Coastguard Worker  %inc = add nuw nsw i32 %i.06, 1
27*9880d681SAndroid Build Coastguard Worker  %exitcond = icmp eq i32 %inc, 2
28*9880d681SAndroid Build Coastguard Worker  br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
29*9880d681SAndroid Build Coastguard Worker; CHECK-NOT: ld.f32
30*9880d681SAndroid Build Coastguard Worker; CHECK-NOT: st.f32
31*9880d681SAndroid Build Coastguard Worker
32*9880d681SAndroid Build Coastguard Workerfor.end:
33*9880d681SAndroid Build Coastguard Worker  ret void
34*9880d681SAndroid Build Coastguard Worker}
35*9880d681SAndroid Build Coastguard Worker
36*9880d681SAndroid Build Coastguard Worker!0 = distinct !{!0, !1}
37*9880d681SAndroid Build Coastguard Worker!1 = !{!"llvm.loop.unroll.disable"}
38