xref: /llvm-project/llvm/test/CodeGen/ARM/load-combine.ll (revision bdf3c5af6adf7d30cee7c6c1d9fb3d0b33df3789)
1; RUN: llc < %s -mtriple=arm-unknown | FileCheck %s
2; RUN: llc < %s -mtriple=armv6-unknown | FileCheck %s --check-prefix=CHECK-ARMv6
3
4; i8* p; // p is 1 byte aligned
5; (i32) p[0] | ((i32) p[1] << 8) | ((i32) p[2] << 16) | ((i32) p[3] << 24)
6define i32 @load_i32_by_i8_unaligned(i32* %arg) {
7; CHECK-LABEL: load_i32_by_i8_unaligned:
8; CHECK: ldrb{{.*}}r0
9; CHECK: ldrb{{.*}}r0
10; CHECK: ldrb{{.*}}r0
11; CHECK: ldrb{{.*}}r0
12; CHECK: orr
13; CHECK: mov pc, lr
14
15; CHECK-ARMv6-LABEL: load_i32_by_i8_unaligned:
16; CHECK-ARMv6: ldrb{{.*}}r0
17; CHECK-ARMv6: ldrb{{.*}}r0
18; CHECK-ARMv6: ldrb{{.*}}r0
19; CHECK-ARMv6: ldrb{{.*}}r0
20; CHECK-ARMv6: orr
21; CHECK-ARMv6: bx lr
22  %tmp = bitcast i32* %arg to i8*
23  %tmp1 = getelementptr inbounds i8, i8* %tmp, i32 0
24  %tmp2 = load i8, i8* %tmp, align 1
25  %tmp3 = zext i8 %tmp2 to i32
26  %tmp4 = getelementptr inbounds i8, i8* %tmp, i32 1
27  %tmp5 = load i8, i8* %tmp4, align 1
28  %tmp6 = zext i8 %tmp5 to i32
29  %tmp7 = shl nuw nsw i32 %tmp6, 8
30  %tmp8 = or i32 %tmp7, %tmp3
31  %tmp9 = getelementptr inbounds i8, i8* %tmp, i32 2
32  %tmp10 = load i8, i8* %tmp9, align 1
33  %tmp11 = zext i8 %tmp10 to i32
34  %tmp12 = shl nuw nsw i32 %tmp11, 16
35  %tmp13 = or i32 %tmp8, %tmp12
36  %tmp14 = getelementptr inbounds i8, i8* %tmp, i32 3
37  %tmp15 = load i8, i8* %tmp14, align 1
38  %tmp16 = zext i8 %tmp15 to i32
39  %tmp17 = shl nuw nsw i32 %tmp16, 24
40  %tmp18 = or i32 %tmp13, %tmp17
41  ret i32 %tmp18
42}
43
44; i8* p; // p is 4 byte aligned
45; (i32) p[0] | ((i32) p[1] << 8) | ((i32) p[2] << 16) | ((i32) p[3] << 24)
46define i32 @load_i32_by_i8_aligned(i32* %arg) {
47; CHECK-LABEL: load_i32_by_i8_aligned:
48; CHECK: ldr  r0, [r0]
49; CHECK-NEXT: mov pc, lr
50
51; CHECK-ARMv6-LABEL: load_i32_by_i8_aligned:
52; CHECK-ARMv6: ldr  r0, [r0]
53; CHECK-ARMv6-NEXT: bx lr
54  %tmp = bitcast i32* %arg to i8*
55  %tmp1 = getelementptr inbounds i8, i8* %tmp, i32 0
56  %tmp2 = load i8, i8* %tmp, align 4
57  %tmp3 = zext i8 %tmp2 to i32
58  %tmp4 = getelementptr inbounds i8, i8* %tmp, i32 1
59  %tmp5 = load i8, i8* %tmp4, align 1
60  %tmp6 = zext i8 %tmp5 to i32
61  %tmp7 = shl nuw nsw i32 %tmp6, 8
62  %tmp8 = or i32 %tmp7, %tmp3
63  %tmp9 = getelementptr inbounds i8, i8* %tmp, i32 2
64  %tmp10 = load i8, i8* %tmp9, align 1
65  %tmp11 = zext i8 %tmp10 to i32
66  %tmp12 = shl nuw nsw i32 %tmp11, 16
67  %tmp13 = or i32 %tmp8, %tmp12
68  %tmp14 = getelementptr inbounds i8, i8* %tmp, i32 3
69  %tmp15 = load i8, i8* %tmp14, align 1
70  %tmp16 = zext i8 %tmp15 to i32
71  %tmp17 = shl nuw nsw i32 %tmp16, 24
72  %tmp18 = or i32 %tmp13, %tmp17
73  ret i32 %tmp18
74}
75
76; i8* p; // p is 4 byte aligned
77; ((i32) p[0] << 24) | ((i32) p[1] << 16) | ((i32) p[2] << 8) | (i32) p[3]
78define i32 @load_i32_by_i8_bswap(i32* %arg) {
79; BSWAP is not supported by 32 bit target
80; CHECK-LABEL: load_i32_by_i8_bswap:
81; CHECK: ldr r0, [r0]
82; CHECK: and
83; CHECK-NEXT: and
84; CHECK-NEXT: orr
85; CHECK-NEXT: orr
86; CHECK-NEXT: orr
87; CHECK: mov pc, lr
88
89; CHECK-ARMv6-LABEL: load_i32_by_i8_bswap:
90; CHECK-ARMv6: ldr  r0, [r0]
91; CHECK-ARMv6-NEXT: rev  r0, r0
92; CHECK-ARMv6-NEXT: bx lr
93  %tmp = bitcast i32* %arg to i8*
94  %tmp1 = load i8, i8* %tmp, align 4
95  %tmp2 = zext i8 %tmp1 to i32
96  %tmp3 = shl nuw nsw i32 %tmp2, 24
97  %tmp4 = getelementptr inbounds i8, i8* %tmp, i32 1
98  %tmp5 = load i8, i8* %tmp4, align 1
99  %tmp6 = zext i8 %tmp5 to i32
100  %tmp7 = shl nuw nsw i32 %tmp6, 16
101  %tmp8 = or i32 %tmp7, %tmp3
102  %tmp9 = getelementptr inbounds i8, i8* %tmp, i32 2
103  %tmp10 = load i8, i8* %tmp9, align 1
104  %tmp11 = zext i8 %tmp10 to i32
105  %tmp12 = shl nuw nsw i32 %tmp11, 8
106  %tmp13 = or i32 %tmp8, %tmp12
107  %tmp14 = getelementptr inbounds i8, i8* %tmp, i32 3
108  %tmp15 = load i8, i8* %tmp14, align 1
109  %tmp16 = zext i8 %tmp15 to i32
110  %tmp17 = or i32 %tmp13, %tmp16
111  ret i32 %tmp17
112}
113
114; i8* p; // p is 8 byte aligned
115; (i64) p[0] | ((i64) p[1] << 8) | ((i64) p[2] << 16) | ((i64) p[3] << 24) | ((i64) p[4] << 32) | ((i64) p[5] << 40) | ((i64) p[6] << 48) | ((i64) p[7] << 56)
116define i64 @load_i64_by_i8(i64* %arg) {
117; CHECK-LABEL: load_i64_by_i8:
118; CHECK: ldr r2, [r0]
119; CHECK-NEXT: ldr r1, [r0, #4]
120; CHECK-NEXT: mov r0, r2
121; CHECK-NEXT: mov pc, lr
122
123; CHECK-ARMv6-LABEL: load_i64_by_i8:
124; CHECK-ARMv6: ldrd  r0, r1, [r0]
125; CHECK-ARMv6: bx  lr
126  %tmp = bitcast i64* %arg to i8*
127  %tmp1 = load i8, i8* %tmp, align 8
128  %tmp2 = zext i8 %tmp1 to i64
129  %tmp3 = getelementptr inbounds i8, i8* %tmp, i64 1
130  %tmp4 = load i8, i8* %tmp3, align 1
131  %tmp5 = zext i8 %tmp4 to i64
132  %tmp6 = shl nuw nsw i64 %tmp5, 8
133  %tmp7 = or i64 %tmp6, %tmp2
134  %tmp8 = getelementptr inbounds i8, i8* %tmp, i64 2
135  %tmp9 = load i8, i8* %tmp8, align 1
136  %tmp10 = zext i8 %tmp9 to i64
137  %tmp11 = shl nuw nsw i64 %tmp10, 16
138  %tmp12 = or i64 %tmp7, %tmp11
139  %tmp13 = getelementptr inbounds i8, i8* %tmp, i64 3
140  %tmp14 = load i8, i8* %tmp13, align 1
141  %tmp15 = zext i8 %tmp14 to i64
142  %tmp16 = shl nuw nsw i64 %tmp15, 24
143  %tmp17 = or i64 %tmp12, %tmp16
144  %tmp18 = getelementptr inbounds i8, i8* %tmp, i64 4
145  %tmp19 = load i8, i8* %tmp18, align 1
146  %tmp20 = zext i8 %tmp19 to i64
147  %tmp21 = shl nuw nsw i64 %tmp20, 32
148  %tmp22 = or i64 %tmp17, %tmp21
149  %tmp23 = getelementptr inbounds i8, i8* %tmp, i64 5
150  %tmp24 = load i8, i8* %tmp23, align 1
151  %tmp25 = zext i8 %tmp24 to i64
152  %tmp26 = shl nuw nsw i64 %tmp25, 40
153  %tmp27 = or i64 %tmp22, %tmp26
154  %tmp28 = getelementptr inbounds i8, i8* %tmp, i64 6
155  %tmp29 = load i8, i8* %tmp28, align 1
156  %tmp30 = zext i8 %tmp29 to i64
157  %tmp31 = shl nuw nsw i64 %tmp30, 48
158  %tmp32 = or i64 %tmp27, %tmp31
159  %tmp33 = getelementptr inbounds i8, i8* %tmp, i64 7
160  %tmp34 = load i8, i8* %tmp33, align 1
161  %tmp35 = zext i8 %tmp34 to i64
162  %tmp36 = shl nuw i64 %tmp35, 56
163  %tmp37 = or i64 %tmp32, %tmp36
164  ret i64 %tmp37
165}
166
167; i8* p; // p is 8 byte aligned
168; ((i64) p[0] << 56) | ((i64) p[1] << 48) | ((i64) p[2] << 40) | ((i64) p[3] << 32) | ((i64) p[4] << 24) | ((i64) p[5] << 16) | ((i64) p[6] << 8) | (i64) p[7]
169define i64 @load_i64_by_i8_bswap(i64* %arg) {
170; CHECK-LABEL: load_i64_by_i8_bswap:
171; CHECK: ldr{{.*}}r0
172; CHECK: ldr{{.*}}r0
173; CHECK: and
174; CHECK-NEXT: and
175; CHECK-NEXT: orr
176; CHECK-NEXT: orr
177; CHECK-NEXT: and
178; CHECK-NEXT: orr
179; CHECK-NEXT: and
180; CHECK-NEXT: orr
181; CHECK-NEXT: orr
182; CHECK-NEXT: orr
183; CHECK: mov pc, lr
184
185; CHECK-ARMv6-LABEL: load_i64_by_i8_bswap:
186; CHECK-ARMv6: ldrd  r2, r3, [r0]
187; CHECK-ARMv6: rev r0, r3
188; CHECK-ARMv6: rev r1, r2
189; CHECK-ARMv6: bx  lr
190  %tmp = bitcast i64* %arg to i8*
191  %tmp1 = load i8, i8* %tmp, align 8
192  %tmp2 = zext i8 %tmp1 to i64
193  %tmp3 = shl nuw i64 %tmp2, 56
194  %tmp4 = getelementptr inbounds i8, i8* %tmp, i64 1
195  %tmp5 = load i8, i8* %tmp4, align 1
196  %tmp6 = zext i8 %tmp5 to i64
197  %tmp7 = shl nuw nsw i64 %tmp6, 48
198  %tmp8 = or i64 %tmp7, %tmp3
199  %tmp9 = getelementptr inbounds i8, i8* %tmp, i64 2
200  %tmp10 = load i8, i8* %tmp9, align 1
201  %tmp11 = zext i8 %tmp10 to i64
202  %tmp12 = shl nuw nsw i64 %tmp11, 40
203  %tmp13 = or i64 %tmp8, %tmp12
204  %tmp14 = getelementptr inbounds i8, i8* %tmp, i64 3
205  %tmp15 = load i8, i8* %tmp14, align 1
206  %tmp16 = zext i8 %tmp15 to i64
207  %tmp17 = shl nuw nsw i64 %tmp16, 32
208  %tmp18 = or i64 %tmp13, %tmp17
209  %tmp19 = getelementptr inbounds i8, i8* %tmp, i64 4
210  %tmp20 = load i8, i8* %tmp19, align 1
211  %tmp21 = zext i8 %tmp20 to i64
212  %tmp22 = shl nuw nsw i64 %tmp21, 24
213  %tmp23 = or i64 %tmp18, %tmp22
214  %tmp24 = getelementptr inbounds i8, i8* %tmp, i64 5
215  %tmp25 = load i8, i8* %tmp24, align 1
216  %tmp26 = zext i8 %tmp25 to i64
217  %tmp27 = shl nuw nsw i64 %tmp26, 16
218  %tmp28 = or i64 %tmp23, %tmp27
219  %tmp29 = getelementptr inbounds i8, i8* %tmp, i64 6
220  %tmp30 = load i8, i8* %tmp29, align 1
221  %tmp31 = zext i8 %tmp30 to i64
222  %tmp32 = shl nuw nsw i64 %tmp31, 8
223  %tmp33 = or i64 %tmp28, %tmp32
224  %tmp34 = getelementptr inbounds i8, i8* %tmp, i64 7
225  %tmp35 = load i8, i8* %tmp34, align 1
226  %tmp36 = zext i8 %tmp35 to i64
227  %tmp37 = or i64 %tmp33, %tmp36
228  ret i64 %tmp37
229}
230
231; i8* p; // p[1] is 4 byte aligned
232; (i32) p[1] | ((i32) p[2] << 8) | ((i32) p[3] << 16) | ((i32) p[4] << 24)
233define i32 @load_i32_by_i8_nonzero_offset(i32* %arg) {
234; CHECK-LABEL: load_i32_by_i8_nonzero_offset:
235; CHECK: ldrb  r1, [r0, #1]
236; CHECK-NEXT: ldrb  r2, [r0, #2]
237; CHECK-NEXT: ldrb  r3, [r0, #3]
238; CHECK-NEXT: ldrb  r0, [r0, #4]
239; CHECK-NEXT: orr r1, r1, r2, lsl #8
240; CHECK-NEXT: orr r1, r1, r3, lsl #16
241; CHECK-NEXT: orr r0, r1, r0, lsl #24
242; CHECK-NEXT: mov pc, lr
243
244; CHECK-ARMv6-LABEL: load_i32_by_i8_nonzero_offset:
245; CHECK-ARMv6: ldrb  r1, [r0, #1]
246; CHECK-ARMv6-NEXT: ldrb  r2, [r0, #2]
247; CHECK-ARMv6-NEXT: ldrb  r3, [r0, #3]
248; CHECK-ARMv6-NEXT: ldrb  r0, [r0, #4]
249; CHECK-ARMv6-NEXT: orr r1, r1, r2, lsl #8
250; CHECK-ARMv6-NEXT: orr r1, r1, r3, lsl #16
251; CHECK-ARMv6-NEXT: orr r0, r1, r0, lsl #24
252; CHECK-ARMv6-NEXT: bx  lr
253
254  %tmp = bitcast i32* %arg to i8*
255  %tmp1 = getelementptr inbounds i8, i8* %tmp, i32 1
256  %tmp2 = load i8, i8* %tmp1, align 4
257  %tmp3 = zext i8 %tmp2 to i32
258  %tmp4 = getelementptr inbounds i8, i8* %tmp, i32 2
259  %tmp5 = load i8, i8* %tmp4, align 1
260  %tmp6 = zext i8 %tmp5 to i32
261  %tmp7 = shl nuw nsw i32 %tmp6, 8
262  %tmp8 = or i32 %tmp7, %tmp3
263  %tmp9 = getelementptr inbounds i8, i8* %tmp, i32 3
264  %tmp10 = load i8, i8* %tmp9, align 1
265  %tmp11 = zext i8 %tmp10 to i32
266  %tmp12 = shl nuw nsw i32 %tmp11, 16
267  %tmp13 = or i32 %tmp8, %tmp12
268  %tmp14 = getelementptr inbounds i8, i8* %tmp, i32 4
269  %tmp15 = load i8, i8* %tmp14, align 1
270  %tmp16 = zext i8 %tmp15 to i32
271  %tmp17 = shl nuw nsw i32 %tmp16, 24
272  %tmp18 = or i32 %tmp13, %tmp17
273  ret i32 %tmp18
274}
275
276; i8* p; // p[-4] is 4 byte aligned
277; (i32) p[-4] | ((i32) p[-3] << 8) | ((i32) p[-2] << 16) | ((i32) p[-1] << 24)
278define i32 @load_i32_by_i8_neg_offset(i32* %arg) {
279; CHECK-LABEL: load_i32_by_i8_neg_offset:
280; CHECK: ldrb  r1, [r0, #-4]
281; CHECK-NEXT: ldrb  r2, [r0, #-3]
282; CHECK-NEXT: ldrb  r3, [r0, #-2]
283; CHECK-NEXT: ldrb  r0, [r0, #-1]
284; CHECK-NEXT: orr r1, r1, r2, lsl #8
285; CHECK-NEXT: orr r1, r1, r3, lsl #16
286; CHECK-NEXT: orr r0, r1, r0, lsl #24
287; CHECK-NEXT: mov pc, lr
288
289; CHECK-ARMv6-LABEL: load_i32_by_i8_neg_offset:
290; CHECK-ARMv6: ldrb  r1, [r0, #-4]
291; CHECK-ARMv6-NEXT: ldrb  r2, [r0, #-3]
292; CHECK-ARMv6-NEXT: ldrb  r3, [r0, #-2]
293; CHECK-ARMv6-NEXT: ldrb  r0, [r0, #-1]
294; CHECK-ARMv6-NEXT: orr r1, r1, r2, lsl #8
295; CHECK-ARMv6-NEXT: orr r1, r1, r3, lsl #16
296; CHECK-ARMv6-NEXT: orr r0, r1, r0, lsl #24
297; CHECK-ARMv6-NEXT: bx  lr
298
299  %tmp = bitcast i32* %arg to i8*
300  %tmp1 = getelementptr inbounds i8, i8* %tmp, i32 -4
301  %tmp2 = load i8, i8* %tmp1, align 4
302  %tmp3 = zext i8 %tmp2 to i32
303  %tmp4 = getelementptr inbounds i8, i8* %tmp, i32 -3
304  %tmp5 = load i8, i8* %tmp4, align 1
305  %tmp6 = zext i8 %tmp5 to i32
306  %tmp7 = shl nuw nsw i32 %tmp6, 8
307  %tmp8 = or i32 %tmp7, %tmp3
308  %tmp9 = getelementptr inbounds i8, i8* %tmp, i32 -2
309  %tmp10 = load i8, i8* %tmp9, align 1
310  %tmp11 = zext i8 %tmp10 to i32
311  %tmp12 = shl nuw nsw i32 %tmp11, 16
312  %tmp13 = or i32 %tmp8, %tmp12
313  %tmp14 = getelementptr inbounds i8, i8* %tmp, i32 -1
314  %tmp15 = load i8, i8* %tmp14, align 1
315  %tmp16 = zext i8 %tmp15 to i32
316  %tmp17 = shl nuw nsw i32 %tmp16, 24
317  %tmp18 = or i32 %tmp13, %tmp17
318  ret i32 %tmp18
319}
320
321; i8* p; // p[1] is 4 byte aligned
322; (i32) p[4] | ((i32) p[3] << 8) | ((i32) p[2] << 16) | ((i32) p[1] << 24)
323define i32 @load_i32_by_i8_nonzero_offset_bswap(i32* %arg) {
324; CHECK-LABEL: load_i32_by_i8_nonzero_offset_bswap:
325; CHECK: ldrb  r1, [r0, #1]
326; CHECK-NEXT: ldrb  r2, [r0, #2]
327; CHECK-NEXT: ldrb  r3, [r0, #3]
328; CHECK-NEXT: ldrb  r0, [r0, #4]
329; CHECK-NEXT: orr r0, r0, r3, lsl #8
330; CHECK-NEXT: orr r0, r0, r2, lsl #16
331; CHECK-NEXT: orr r0, r0, r1, lsl #24
332; CHECK-NEXT: mov pc, lr
333
334; CHECK-ARMv6-LABEL: load_i32_by_i8_nonzero_offset_bswap:
335; CHECK-ARMv6: ldrb  r1, [r0, #1]
336; CHECK-ARMv6-NEXT: ldrb  r2, [r0, #2]
337; CHECK-ARMv6-NEXT: ldrb  r3, [r0, #3]
338; CHECK-ARMv6-NEXT: ldrb  r0, [r0, #4]
339; CHECK-ARMv6-NEXT: orr r0, r0, r3, lsl #8
340; CHECK-ARMv6-NEXT: orr r0, r0, r2, lsl #16
341; CHECK-ARMv6-NEXT: orr r0, r0, r1, lsl #24
342; CHECK-ARMv6-NEXT: bx  lr
343
344  %tmp = bitcast i32* %arg to i8*
345  %tmp1 = getelementptr inbounds i8, i8* %tmp, i32 4
346  %tmp2 = load i8, i8* %tmp1, align 1
347  %tmp3 = zext i8 %tmp2 to i32
348  %tmp4 = getelementptr inbounds i8, i8* %tmp, i32 3
349  %tmp5 = load i8, i8* %tmp4, align 1
350  %tmp6 = zext i8 %tmp5 to i32
351  %tmp7 = shl nuw nsw i32 %tmp6, 8
352  %tmp8 = or i32 %tmp7, %tmp3
353  %tmp9 = getelementptr inbounds i8, i8* %tmp, i32 2
354  %tmp10 = load i8, i8* %tmp9, align 1
355  %tmp11 = zext i8 %tmp10 to i32
356  %tmp12 = shl nuw nsw i32 %tmp11, 16
357  %tmp13 = or i32 %tmp8, %tmp12
358  %tmp14 = getelementptr inbounds i8, i8* %tmp, i32 1
359  %tmp15 = load i8, i8* %tmp14, align 4
360  %tmp16 = zext i8 %tmp15 to i32
361  %tmp17 = shl nuw nsw i32 %tmp16, 24
362  %tmp18 = or i32 %tmp13, %tmp17
363  ret i32 %tmp18
364}
365
366; i8* p; // p[-4] is 4 byte aligned
367; (i32) p[-1] | ((i32) p[-2] << 8) | ((i32) p[-3] << 16) | ((i32) p[-4] << 24)
368define i32 @load_i32_by_i8_neg_offset_bswap(i32* %arg) {
369; CHECK-LABEL: load_i32_by_i8_neg_offset_bswap:
370; CHECK: ldrb  r1, [r0, #-4]
371; CHECK-NEXT: ldrb  r2, [r0, #-3]
372; CHECK-NEXT: ldrb  r3, [r0, #-2]
373; CHECK-NEXT: ldrb  r0, [r0, #-1]
374; CHECK-NEXT: orr r0, r0, r3, lsl #8
375; CHECK-NEXT: orr r0, r0, r2, lsl #16
376; CHECK-NEXT: orr r0, r0, r1, lsl #24
377; CHECK-NEXT: mov pc, lr
378
379; CHECK-ARMv6-LABEL: load_i32_by_i8_neg_offset_bswap:
380; CHECK-ARMv6: ldrb  r1, [r0, #-4]
381; CHECK-ARMv6-NEXT: ldrb  r2, [r0, #-3]
382; CHECK-ARMv6-NEXT: ldrb  r3, [r0, #-2]
383; CHECK-ARMv6-NEXT: ldrb  r0, [r0, #-1]
384; CHECK-ARMv6-NEXT: orr r0, r0, r3, lsl #8
385; CHECK-ARMv6-NEXT: orr r0, r0, r2, lsl #16
386; CHECK-ARMv6-NEXT: orr r0, r0, r1, lsl #24
387; CHECK-ARMv6-NEXT: bx  lr
388
389  %tmp = bitcast i32* %arg to i8*
390  %tmp1 = getelementptr inbounds i8, i8* %tmp, i32 -1
391  %tmp2 = load i8, i8* %tmp1, align 1
392  %tmp3 = zext i8 %tmp2 to i32
393  %tmp4 = getelementptr inbounds i8, i8* %tmp, i32 -2
394  %tmp5 = load i8, i8* %tmp4, align 1
395  %tmp6 = zext i8 %tmp5 to i32
396  %tmp7 = shl nuw nsw i32 %tmp6, 8
397  %tmp8 = or i32 %tmp7, %tmp3
398  %tmp9 = getelementptr inbounds i8, i8* %tmp, i32 -3
399  %tmp10 = load i8, i8* %tmp9, align 1
400  %tmp11 = zext i8 %tmp10 to i32
401  %tmp12 = shl nuw nsw i32 %tmp11, 16
402  %tmp13 = or i32 %tmp8, %tmp12
403  %tmp14 = getelementptr inbounds i8, i8* %tmp, i32 -4
404  %tmp15 = load i8, i8* %tmp14, align 4
405  %tmp16 = zext i8 %tmp15 to i32
406  %tmp17 = shl nuw nsw i32 %tmp16, 24
407  %tmp18 = or i32 %tmp13, %tmp17
408  ret i32 %tmp18
409}
410