xref: /llvm-project/llvm/test/CodeGen/ARM/cortex-a57-misched-vldm-wrback.ll (revision bed1c7f061aa12417aa081e334afdba45767b938)
1; REQUIRES: asserts
2; RUN: llc < %s -mtriple=armv8r-eabi -mcpu=cortex-a57 -mattr=use-misched -verify-misched -debug-only=machine-scheduler -o - 2>&1 > /dev/null | FileCheck %s
3;
4
5@a = dso_local global double 0.0, align 4
6@b = dso_local global double 0.0, align 4
7@c = dso_local global double 0.0, align 4
8
9; CHECK:       ********** MI Scheduling **********
10; We need second, post-ra scheduling to have VLDM instruction combined from single-loads
11; CHECK:       ********** MI Scheduling **********
12; CHECK:       VLDMDIA_UPD
13; CHECK:       rdefs left
14; CHECK-NEXT:  Latency            : 6
15; CHECK:       Successors:
16; CHECK:       Data
17; CHECK-SAME:  Latency=1
18; CHECK-NEXT:  Data
19; CHECK-SAME:  Latency=1
20; CHECK-NEXT:  Data
21; CHECK-SAME:  Latency=5
22; CHECK-NEXT:  Data
23; CHECK-SAME:  Latency=0
24; CHECK-NEXT:  Data
25; CHECK-SAME:  Latency=0
26define dso_local i32 @bar(ptr %iptr) minsize optsize {
27  %1 = load double, ptr @a, align 8
28  %2 = load double, ptr @b, align 8
29  %3 = load double, ptr @c, align 8
30
31  %ptr_after = getelementptr double, ptr @a, i32 3
32
33  %ptr_new_ival = ptrtoint ptr %ptr_after to i32
34  %ptr_new = inttoptr i32 %ptr_new_ival to ptr
35
36  store i32 %ptr_new_ival, ptr %iptr, align 8
37
38  %v1 = fptoui double %1 to i32
39
40  %mul1 = mul i32 %ptr_new_ival, %v1
41
42  %v2 = fptoui double %2 to i32
43  %v3 = fptoui double %3 to i32
44
45  %mul2 = mul i32 %mul1, %v2
46  %mul3 = mul i32 %mul2, %v3
47
48  ret i32 %mul3
49}
50
51