1 ; Test 32-bit square root.
3 ; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s
5 declare float @llvm.sqrt.f32(float %f)
7 ; Check register square root.
8 define float @f1(float %val) {
10 ; CHECK: sqebr %f0, %f0
12 %res = call float @llvm.sqrt.f32(float %val)
16 ; Check the low end of the SQEB range.
17 define float @f2(float *%ptr) {
19 ; CHECK: sqeb %f0, 0(%r2)
21 %val = load float *%ptr
22 %res = call float @llvm.sqrt.f32(float %val)
26 ; Check the high end of the aligned SQEB range.
27 define float @f3(float *%base) {
29 ; CHECK: sqeb %f0, 4092(%r2)
31 %ptr = getelementptr float *%base, i64 1023
32 %val = load float *%ptr
33 %res = call float @llvm.sqrt.f32(float %val)
37 ; Check the next word up, which needs separate address logic.
38 ; Other sequences besides this one would be OK.
39 define float @f4(float *%base) {
41 ; CHECK: aghi %r2, 4096
42 ; CHECK: sqeb %f0, 0(%r2)
44 %ptr = getelementptr float *%base, i64 1024
45 %val = load float *%ptr
46 %res = call float @llvm.sqrt.f32(float %val)
50 ; Check negative displacements, which also need separate address logic.
51 define float @f5(float *%base) {
54 ; CHECK: sqeb %f0, 0(%r2)
56 %ptr = getelementptr float *%base, i64 -1
57 %val = load float *%ptr
58 %res = call float @llvm.sqrt.f32(float %val)
62 ; Check that SQEB allows indices.
63 define float @f6(float *%base, i64 %index) {
65 ; CHECK: sllg %r1, %r3, 2
66 ; CHECK: sqeb %f0, 400(%r1,%r2)
68 %ptr1 = getelementptr float *%base, i64 %index
69 %ptr2 = getelementptr float *%ptr1, i64 100
70 %val = load float *%ptr2
71 %res = call float @llvm.sqrt.f32(float %val)