1 ; RUN: opt -loop-unroll -S -mtriple=amdgcn-- -mcpu=tahiti %s | FileCheck %s
3 ; This IR comes from this OpenCL C code:
6 ; for (int i = 0; i < 4; i++, b++) {
14 ; This test is meant to check that this loop isn't unrolled into more than
15 ; four iterations. The loop unrolling preferences we currently use cause this
16 ; loop to not be unrolled at all, but that may change in the future.
19 ; CHECK: store i8 0, i8 addrspace(1)*
20 ; CHECK-NOT: store i8 0, i8 addrspace(1)*
22 define amdgpu_kernel void @test(i8 addrspace(1)* nocapture %dst, i32 %a, i32 %b, i32 %c) {
24 %add = add nsw i32 %b, 4
25 %cmp = icmp sgt i32 %add, %a
26 br i1 %cmp, label %for.cond.preheader, label %if.end7
28 for.cond.preheader: ; preds = %entry
29 %cmp313 = icmp slt i32 %b, %a
30 br i1 %cmp313, label %if.then4.lr.ph, label %if.end7.loopexit
32 if.then4.lr.ph: ; preds = %for.cond.preheader
33 %0 = sext i32 %c to i64
36 if.then4: ; preds = %if.then4.lr.ph, %if.then4
37 %i.015 = phi i32 [ 0, %if.then4.lr.ph ], [ %inc, %if.then4 ]
38 %b.addr.014 = phi i32 [ %b, %if.then4.lr.ph ], [ %add2, %if.then4 ]
39 %add2 = add nsw i32 %b.addr.014, 1
40 %1 = sext i32 %b.addr.014 to i64
41 %add.ptr.sum = add nsw i64 %1, %0
42 %add.ptr5 = getelementptr inbounds i8, i8 addrspace(1)* %dst, i64 %add.ptr.sum
43 store i8 0, i8 addrspace(1)* %add.ptr5, align 1
44 %inc = add nsw i32 %i.015, 1
45 %cmp1 = icmp slt i32 %inc, 4
46 %cmp3 = icmp slt i32 %add2, %a
47 %or.cond = and i1 %cmp3, %cmp1
48 br i1 %or.cond, label %if.then4, label %for.cond.if.end7.loopexit_crit_edge
50 for.cond.if.end7.loopexit_crit_edge: ; preds = %if.then4
51 br label %if.end7.loopexit
53 if.end7.loopexit: ; preds = %for.cond.if.end7.loopexit_crit_edge, %for.cond.preheader
56 if.end7: ; preds = %if.end7.loopexit, %entry