1 ; Test high-part i64->i128 multiplications.
3 ; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s
7 ; Check zero-extended multiplication in which only the high part is used.
8 define i64 @f1(i64 %dummy, i64 %a, i64 %b) {
10 ; CHECK-NOT: {{%r[234]}}
11 ; CHECK: mlgr %r2, %r4
13 %ax = zext i64 %a to i128
14 %bx = zext i64 %b to i128
15 %mulx = mul i128 %ax, %bx
16 %highx = lshr i128 %mulx, 64
17 %high = trunc i128 %highx to i64
21 ; Check sign-extended multiplication in which only the high part is used.
22 ; This needs a rather convoluted sequence.
23 define i64 @f2(i64 %dummy, i64 %a, i64 %b) {
29 %ax = sext i64 %a to i128
30 %bx = sext i64 %b to i128
31 %mulx = mul i128 %ax, %bx
32 %highx = lshr i128 %mulx, 64
33 %high = trunc i128 %highx to i64
37 ; Check zero-extended multiplication in which only part of the high half
39 define i64 @f3(i64 %dummy, i64 %a, i64 %b) {
41 ; CHECK-NOT: {{%r[234]}}
42 ; CHECK: mlgr %r2, %r4
43 ; CHECK: srlg %r2, %r2, 3
45 %ax = zext i64 %a to i128
46 %bx = zext i64 %b to i128
47 %mulx = mul i128 %ax, %bx
48 %highx = lshr i128 %mulx, 67
49 %high = trunc i128 %highx to i64
53 ; Check zero-extended multiplication in which the result is split into
54 ; high and low halves.
55 define i64 @f4(i64 %dummy, i64 %a, i64 %b) {
57 ; CHECK-NOT: {{%r[234]}}
58 ; CHECK: mlgr %r2, %r4
61 %ax = zext i64 %a to i128
62 %bx = zext i64 %b to i128
63 %mulx = mul i128 %ax, %bx
64 %highx = lshr i128 %mulx, 64
65 %high = trunc i128 %highx to i64
66 %low = trunc i128 %mulx to i64
67 %or = or i64 %high, %low
71 ; Check division by a constant, which should use multiplication instead.
72 define i64 @f5(i64 %dummy, i64 %a) {
75 ; CHECK: srlg %r2, %r2,
77 %res = udiv i64 %a, 1234
81 ; Check MLG with no displacement.
82 define i64 @f6(i64 %dummy, i64 %a, i64 *%src) {
84 ; CHECK-NOT: {{%r[234]}}
85 ; CHECK: mlg %r2, 0(%r4)
88 %ax = zext i64 %a to i128
89 %bx = zext i64 %b to i128
90 %mulx = mul i128 %ax, %bx
91 %highx = lshr i128 %mulx, 64
92 %high = trunc i128 %highx to i64
96 ; Check the high end of the aligned MLG range.
97 define i64 @f7(i64 %dummy, i64 %a, i64 *%src) {
99 ; CHECK: mlg %r2, 524280(%r4)
101 %ptr = getelementptr i64 *%src, i64 65535
103 %ax = zext i64 %a to i128
104 %bx = zext i64 %b to i128
105 %mulx = mul i128 %ax, %bx
106 %highx = lshr i128 %mulx, 64
107 %high = trunc i128 %highx to i64
111 ; Check the next doubleword up, which requires separate address logic.
112 ; Other sequences besides this one would be OK.
113 define i64 @f8(i64 %dummy, i64 %a, i64 *%src) {
115 ; CHECK: agfi %r4, 524288
116 ; CHECK: mlg %r2, 0(%r4)
118 %ptr = getelementptr i64 *%src, i64 65536
120 %ax = zext i64 %a to i128
121 %bx = zext i64 %b to i128
122 %mulx = mul i128 %ax, %bx
123 %highx = lshr i128 %mulx, 64
124 %high = trunc i128 %highx to i64
128 ; Check the high end of the negative aligned MLG range.
129 define i64 @f9(i64 %dummy, i64 %a, i64 *%src) {
131 ; CHECK: mlg %r2, -8(%r4)
133 %ptr = getelementptr i64 *%src, i64 -1
135 %ax = zext i64 %a to i128
136 %bx = zext i64 %b to i128
137 %mulx = mul i128 %ax, %bx
138 %highx = lshr i128 %mulx, 64
139 %high = trunc i128 %highx to i64
143 ; Check the low end of the MLG range.
144 define i64 @f10(i64 %dummy, i64 %a, i64 *%src) {
146 ; CHECK: mlg %r2, -524288(%r4)
148 %ptr = getelementptr i64 *%src, i64 -65536
150 %ax = zext i64 %a to i128
151 %bx = zext i64 %b to i128
152 %mulx = mul i128 %ax, %bx
153 %highx = lshr i128 %mulx, 64
154 %high = trunc i128 %highx to i64
158 ; Check the next doubleword down, which needs separate address logic.
159 ; Other sequences besides this one would be OK.
160 define i64 @f11(i64 *%dest, i64 %a, i64 *%src) {
162 ; CHECK: agfi %r4, -524296
163 ; CHECK: mlg %r2, 0(%r4)
165 %ptr = getelementptr i64 *%src, i64 -65537
167 %ax = zext i64 %a to i128
168 %bx = zext i64 %b to i128
169 %mulx = mul i128 %ax, %bx
170 %highx = lshr i128 %mulx, 64
171 %high = trunc i128 %highx to i64
175 ; Check that MLG allows an index.
176 define i64 @f12(i64 *%dest, i64 %a, i64 %src, i64 %index) {
178 ; CHECK: mlg %r2, 524287(%r5,%r4)
180 %add1 = add i64 %src, %index
181 %add2 = add i64 %add1, 524287
182 %ptr = inttoptr i64 %add2 to i64 *
184 %ax = zext i64 %a to i128
185 %bx = zext i64 %b to i128
186 %mulx = mul i128 %ax, %bx
187 %highx = lshr i128 %mulx, 64
188 %high = trunc i128 %highx to i64
192 ; Check that multiplications of spilled values can use MLG rather than MLGR.
193 define i64 @f13(i64 *%ptr0) {
195 ; CHECK: brasl %r14, foo@PLT
196 ; CHECK: mlg {{%r[0-9]+}}, 160(%r15)
198 %ptr1 = getelementptr i64 *%ptr0, i64 2
199 %ptr2 = getelementptr i64 *%ptr0, i64 4
200 %ptr3 = getelementptr i64 *%ptr0, i64 6
201 %ptr4 = getelementptr i64 *%ptr0, i64 8
202 %ptr5 = getelementptr i64 *%ptr0, i64 10
203 %ptr6 = getelementptr i64 *%ptr0, i64 12
204 %ptr7 = getelementptr i64 *%ptr0, i64 14
205 %ptr8 = getelementptr i64 *%ptr0, i64 16
206 %ptr9 = getelementptr i64 *%ptr0, i64 18
208 %val0 = load i64 *%ptr0
209 %val1 = load i64 *%ptr1
210 %val2 = load i64 *%ptr2
211 %val3 = load i64 *%ptr3
212 %val4 = load i64 *%ptr4
213 %val5 = load i64 *%ptr5
214 %val6 = load i64 *%ptr6
215 %val7 = load i64 *%ptr7
216 %val8 = load i64 *%ptr8
217 %val9 = load i64 *%ptr9
219 %ret = call i64 @foo()
221 %retx = zext i64 %ret to i128
222 %val0x = zext i64 %val0 to i128
223 %mul0d = mul i128 %retx, %val0x
224 %mul0x = lshr i128 %mul0d, 64
226 %val1x = zext i64 %val1 to i128
227 %mul1d = mul i128 %mul0x, %val1x
228 %mul1x = lshr i128 %mul1d, 64
230 %val2x = zext i64 %val2 to i128
231 %mul2d = mul i128 %mul1x, %val2x
232 %mul2x = lshr i128 %mul2d, 64
234 %val3x = zext i64 %val3 to i128
235 %mul3d = mul i128 %mul2x, %val3x
236 %mul3x = lshr i128 %mul3d, 64
238 %val4x = zext i64 %val4 to i128
239 %mul4d = mul i128 %mul3x, %val4x
240 %mul4x = lshr i128 %mul4d, 64
242 %val5x = zext i64 %val5 to i128
243 %mul5d = mul i128 %mul4x, %val5x
244 %mul5x = lshr i128 %mul5d, 64
246 %val6x = zext i64 %val6 to i128
247 %mul6d = mul i128 %mul5x, %val6x
248 %mul6x = lshr i128 %mul6d, 64
250 %val7x = zext i64 %val7 to i128
251 %mul7d = mul i128 %mul6x, %val7x
252 %mul7x = lshr i128 %mul7d, 64
254 %val8x = zext i64 %val8 to i128
255 %mul8d = mul i128 %mul7x, %val8x
256 %mul8x = lshr i128 %mul8d, 64
258 %val9x = zext i64 %val9 to i128
259 %mul9d = mul i128 %mul8x, %val9x
260 %mul9x = lshr i128 %mul9d, 64
262 %mul9 = trunc i128 %mul9x to i64