xref: /llvm-project/llvm/test/CodeGen/RISCV/machine-sink-load-immediate.ll (revision 6657d4bd70523e6852f07f64711fb15bdf7b347a) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
2; RUN: llc -mtriple=riscv64 < %s | FileCheck %s
3
4define i1 @sink_li(ptr %text, ptr %text.addr.0) nounwind {
5; CHECK-LABEL: sink_li:
6; CHECK:       # %bb.0: # %entry
7; CHECK-NEXT:    addi sp, sp, -32
8; CHECK-NEXT:    sd ra, 24(sp) # 8-byte Folded Spill
9; CHECK-NEXT:    sd s0, 16(sp) # 8-byte Folded Spill
10; CHECK-NEXT:    sd s1, 8(sp) # 8-byte Folded Spill
11; CHECK-NEXT:    sd s2, 0(sp) # 8-byte Folded Spill
12; CHECK-NEXT:    mv s1, a1
13; CHECK-NEXT:    mv s0, a0
14; CHECK-NEXT:    call toupper
15; CHECK-NEXT:    li a1, 0
16; CHECK-NEXT:    beqz s0, .LBB0_26
17; CHECK-NEXT:  # %bb.1: # %while.body.preheader
18; CHECK-NEXT:    li a2, 1
19; CHECK-NEXT:    li a3, 9
20; CHECK-NEXT:    li a4, 32
21; CHECK-NEXT:  .LBB0_2: # %while.body
22; CHECK-NEXT:    # =>This Inner Loop Header: Depth=1
23; CHECK-NEXT:    bnez a2, .LBB0_4
24; CHECK-NEXT:  # %bb.3: # %while.body
25; CHECK-NEXT:    # in Loop: Header=BB0_2 Depth=1
26; CHECK-NEXT:    bne a2, a3, .LBB0_16
27; CHECK-NEXT:  .LBB0_4: # %while.body.1
28; CHECK-NEXT:    # in Loop: Header=BB0_2 Depth=1
29; CHECK-NEXT:    bnez a2, .LBB0_6
30; CHECK-NEXT:  # %bb.5: # %while.body.1
31; CHECK-NEXT:    # in Loop: Header=BB0_2 Depth=1
32; CHECK-NEXT:    bne a2, a3, .LBB0_17
33; CHECK-NEXT:  .LBB0_6: # %while.body.3
34; CHECK-NEXT:    # in Loop: Header=BB0_2 Depth=1
35; CHECK-NEXT:    bnez a2, .LBB0_8
36; CHECK-NEXT:  # %bb.7: # %while.body.3
37; CHECK-NEXT:    # in Loop: Header=BB0_2 Depth=1
38; CHECK-NEXT:    bne a2, a4, .LBB0_19
39; CHECK-NEXT:  .LBB0_8: # %while.body.4
40; CHECK-NEXT:    # in Loop: Header=BB0_2 Depth=1
41; CHECK-NEXT:    bnez a2, .LBB0_10
42; CHECK-NEXT:  # %bb.9: # %while.body.4
43; CHECK-NEXT:    # in Loop: Header=BB0_2 Depth=1
44; CHECK-NEXT:    bne a2, a4, .LBB0_21
45; CHECK-NEXT:  .LBB0_10: # %while.body.5
46; CHECK-NEXT:    # in Loop: Header=BB0_2 Depth=1
47; CHECK-NEXT:    bnez a2, .LBB0_12
48; CHECK-NEXT:  # %bb.11: # %while.body.5
49; CHECK-NEXT:    # in Loop: Header=BB0_2 Depth=1
50; CHECK-NEXT:    bne a2, a3, .LBB0_23
51; CHECK-NEXT:  .LBB0_12: # %while.body.6
52; CHECK-NEXT:    # in Loop: Header=BB0_2 Depth=1
53; CHECK-NEXT:    bnez a2, .LBB0_2
54; CHECK-NEXT:  # %bb.13: # %while.body.6
55; CHECK-NEXT:    # in Loop: Header=BB0_2 Depth=1
56; CHECK-NEXT:    beq a2, a3, .LBB0_2
57; CHECK-NEXT:  # %bb.14: # %while.body.6
58; CHECK-NEXT:    beqz a2, .LBB0_24
59; CHECK-NEXT:  # %bb.15: # %strdup.exit.split.loop.exit126
60; CHECK-NEXT:    addi s0, s1, 7
61; CHECK-NEXT:    j .LBB0_25
62; CHECK-NEXT:  .LBB0_16: # %while.body
63; CHECK-NEXT:    beqz a2, .LBB0_26
64; CHECK-NEXT:    j .LBB0_18
65; CHECK-NEXT:  .LBB0_17: # %while.body.1
66; CHECK-NEXT:    beqz a2, .LBB0_24
67; CHECK-NEXT:  .LBB0_18: # %strdup.exit.loopexit
68; CHECK-NEXT:    li s0, 0
69; CHECK-NEXT:    j .LBB0_25
70; CHECK-NEXT:  .LBB0_19: # %while.body.3
71; CHECK-NEXT:    beqz a2, .LBB0_24
72; CHECK-NEXT:  # %bb.20: # %strdup.exit.split.loop.exit120
73; CHECK-NEXT:    addi s0, s1, 4
74; CHECK-NEXT:    j .LBB0_25
75; CHECK-NEXT:  .LBB0_21: # %while.body.4
76; CHECK-NEXT:    beqz a2, .LBB0_24
77; CHECK-NEXT:  # %bb.22: # %strdup.exit.split.loop.exit122
78; CHECK-NEXT:    addi s0, s1, 5
79; CHECK-NEXT:    j .LBB0_25
80; CHECK-NEXT:  .LBB0_23: # %while.body.5
81; CHECK-NEXT:    bnez a2, .LBB0_25
82; CHECK-NEXT:  .LBB0_24:
83; CHECK-NEXT:    li a1, 0
84; CHECK-NEXT:    j .LBB0_26
85; CHECK-NEXT:  .LBB0_25: # %strdup.exit
86; CHECK-NEXT:    li s1, 0
87; CHECK-NEXT:    mv s2, a0
88; CHECK-NEXT:    li a0, 0
89; CHECK-NEXT:    mv a1, s0
90; CHECK-NEXT:    jalr s1
91; CHECK-NEXT:    li a0, 0
92; CHECK-NEXT:    mv a1, s2
93; CHECK-NEXT:    li a2, 0
94; CHECK-NEXT:    jalr s1
95; CHECK-NEXT:    li a1, 1
96; CHECK-NEXT:  .LBB0_26: # %return
97; CHECK-NEXT:    mv a0, a1
98; CHECK-NEXT:    ld ra, 24(sp) # 8-byte Folded Reload
99; CHECK-NEXT:    ld s0, 16(sp) # 8-byte Folded Reload
100; CHECK-NEXT:    ld s1, 8(sp) # 8-byte Folded Reload
101; CHECK-NEXT:    ld s2, 0(sp) # 8-byte Folded Reload
102; CHECK-NEXT:    addi sp, sp, 32
103; CHECK-NEXT:    ret
104entry:
105  %call = call i32 @toupper()
106  %tobool.not = icmp eq ptr %text, null
107  br i1 %tobool.not, label %return, label %while.body
108
109while.body:                                       ; preds = %while.body.6, %while.body.6, %entry
110  switch i8 1, label %strdup.exit.split.loop.exit114 [
111    i8 1, label %while.body.1
112    i8 9, label %while.body.1
113    i8 0, label %return
114  ]
115
116while.body.1:                                     ; preds = %while.body, %while.body
117  switch i8 1, label %strdup.exit [
118    i8 1, label %while.body.3
119    i8 9, label %while.body.3
120    i8 0, label %return
121  ]
122
123while.body.3:                                     ; preds = %while.body.1, %while.body.1
124  switch i8 1, label %strdup.exit.split.loop.exit120 [
125    i8 32, label %while.body.4
126    i8 1, label %while.body.4
127    i8 0, label %return
128  ]
129
130while.body.4:                                     ; preds = %while.body.3, %while.body.3
131  switch i8 1, label %strdup.exit.split.loop.exit122 [
132    i8 32, label %while.body.5
133    i8 1, label %while.body.5
134    i8 0, label %return
135  ]
136
137while.body.5:                                     ; preds = %while.body.4, %while.body.4
138  switch i8 1, label %strdup.exit.split.loop.exit124 [
139    i8 1, label %while.body.6
140    i8 9, label %while.body.6
141    i8 0, label %return
142  ]
143
144while.body.6:                                     ; preds = %while.body.5, %while.body.5
145  switch i8 1, label %strdup.exit.split.loop.exit126 [
146    i8 1, label %while.body
147    i8 9, label %while.body
148    i8 0, label %return
149  ]
150
151strdup.exit.split.loop.exit114:        ; preds = %while.body
152  br label %strdup.exit
153
154strdup.exit.split.loop.exit120:        ; preds = %while.body.3
155  %incdec.ptr.3.le = getelementptr i8, ptr %text.addr.0, i64 4
156  br label %strdup.exit
157
158strdup.exit.split.loop.exit122:        ; preds = %while.body.4
159  %incdec.ptr.4.le = getelementptr i8, ptr %text.addr.0, i64 5
160  br label %strdup.exit
161
162strdup.exit.split.loop.exit124:        ; preds = %while.body.5
163  br label %strdup.exit
164
165strdup.exit.split.loop.exit126:        ; preds = %while.body.6
166  %incdec.ptr.6.le = getelementptr i8, ptr %text.addr.0, i64 7
167  br label %strdup.exit
168
169strdup.exit:                           ; preds = %strdup.exit.split.loop.exit126, %strdup.exit.split.loop.exit124, %strdup.exit.split.loop.exit122, %strdup.exit.split.loop.exit120, %strdup.exit.split.loop.exit114, %while.body.1
170  %text.addr.0.lcssa = phi ptr [ null, %strdup.exit.split.loop.exit114 ], [ %incdec.ptr.3.le, %strdup.exit.split.loop.exit120 ], [ %incdec.ptr.4.le, %strdup.exit.split.loop.exit122 ], [ %text, %strdup.exit.split.loop.exit124 ], [ %incdec.ptr.6.le, %strdup.exit.split.loop.exit126 ], [ null, %while.body.1 ]
171  %call5.i = tail call ptr null(ptr null, ptr %text.addr.0.lcssa)
172  %memchr64 = tail call ptr null(ptr null, i32 %call, i64 0)
173  br label %return
174
175return:                                           ; preds = %strdup.exit, %while.body.6, %while.body.5, %while.body.4, %while.body.3, %while.body.1, %while.body, %entry
176  %retval.1 = phi i1 [ false, %entry ], [ true, %strdup.exit ], [ false, %while.body ], [ false, %while.body.1 ], [ false, %while.body.3 ], [ false, %while.body.4 ], [ false, %while.body.5 ], [ false, %while.body.6 ]
177  ret i1 %retval.1
178}
179
180declare i32 @toupper()
181
182; In this example, %arg and the return value (13) have overlapping live
183; intervals because the ABI mandidates they both be placed in a0.
184define signext i32 @overlap_live_ranges(ptr %arg, i32 signext %arg1) {
185; CHECK-LABEL: overlap_live_ranges:
186; CHECK:       # %bb.0: # %bb
187; CHECK-NEXT:    li a2, 1
188; CHECK-NEXT:    bne a1, a2, .LBB1_2
189; CHECK-NEXT:  # %bb.1: # %bb2
190; CHECK-NEXT:    lw a0, 4(a0)
191; CHECK-NEXT:    ret
192; CHECK-NEXT:  .LBB1_2:
193; CHECK-NEXT:    li a0, 13
194; CHECK-NEXT:    ret
195bb:
196  %i = icmp eq i32 %arg1, 1
197  br i1 %i, label %bb2, label %bb5
198
199bb2:                                              ; preds = %bb
200  %i3 = getelementptr inbounds nuw i8, ptr %arg, i64 4
201  %i4 = load i32, ptr %i3, align 4
202  br label %bb5
203
204bb5:                                              ; preds = %bb2, %bb
205  %i6 = phi i32 [ %i4, %bb2 ], [ 13, %bb ]
206  ret i32 %i6
207}
208
209
210; For switches, the values feeding the phi are always sunk into the
211; target blocks as the IR syntax requires the intermediate block and
212; DAG lowers it in the immediate predecessor of the phi.
213define signext i32 @switch_dispatch(i8 %a) {
214; CHECK-LABEL: switch_dispatch:
215; CHECK:       # %bb.0: # %bb
216; CHECK-NEXT:    addi sp, sp, -16
217; CHECK-NEXT:    .cfi_def_cfa_offset 16
218; CHECK-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
219; CHECK-NEXT:    sd s0, 0(sp) # 8-byte Folded Spill
220; CHECK-NEXT:    .cfi_offset ra, -8
221; CHECK-NEXT:    .cfi_offset s0, -16
222; CHECK-NEXT:    andi a0, a0, 255
223; CHECK-NEXT:    li a1, 31
224; CHECK-NEXT:    blt a1, a0, .LBB2_5
225; CHECK-NEXT:  # %bb.1: # %bb
226; CHECK-NEXT:    beqz a0, .LBB2_10
227; CHECK-NEXT:  # %bb.2: # %bb
228; CHECK-NEXT:    li a1, 12
229; CHECK-NEXT:    beq a0, a1, .LBB2_11
230; CHECK-NEXT:  # %bb.3: # %bb
231; CHECK-NEXT:    li a1, 13
232; CHECK-NEXT:    bne a0, a1, .LBB2_9
233; CHECK-NEXT:  # %bb.4: # %case.4
234; CHECK-NEXT:    li s0, 644
235; CHECK-NEXT:    j .LBB2_13
236; CHECK-NEXT:  .LBB2_5: # %bb
237; CHECK-NEXT:    li a1, 234
238; CHECK-NEXT:    beq a0, a1, .LBB2_9
239; CHECK-NEXT:  # %bb.6: # %bb
240; CHECK-NEXT:    li a1, 70
241; CHECK-NEXT:    beq a0, a1, .LBB2_12
242; CHECK-NEXT:  # %bb.7: # %bb
243; CHECK-NEXT:    li a1, 32
244; CHECK-NEXT:    bne a0, a1, .LBB2_9
245; CHECK-NEXT:  # %bb.8: # %case.0
246; CHECK-NEXT:    li s0, 13
247; CHECK-NEXT:    j .LBB2_13
248; CHECK-NEXT:  .LBB2_9: # %case.default
249; CHECK-NEXT:    li s0, 23
250; CHECK-NEXT:    j .LBB2_13
251; CHECK-NEXT:  .LBB2_10: # %case.5
252; CHECK-NEXT:    li s0, 54
253; CHECK-NEXT:    j .LBB2_13
254; CHECK-NEXT:  .LBB2_11: # %case.1
255; CHECK-NEXT:    li s0, 53
256; CHECK-NEXT:    j .LBB2_13
257; CHECK-NEXT:  .LBB2_12: # %case.2
258; CHECK-NEXT:    li s0, 33
259; CHECK-NEXT:  .LBB2_13: # %merge
260; CHECK-NEXT:    mv a0, s0
261; CHECK-NEXT:    call use
262; CHECK-NEXT:    mv a0, s0
263; CHECK-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
264; CHECK-NEXT:    ld s0, 0(sp) # 8-byte Folded Reload
265; CHECK-NEXT:    .cfi_restore ra
266; CHECK-NEXT:    .cfi_restore s0
267; CHECK-NEXT:    addi sp, sp, 16
268; CHECK-NEXT:    .cfi_def_cfa_offset 0
269; CHECK-NEXT:    ret
270bb:
271  switch i8 %a, label %case.default [
272    i8 32, label %case.0
273    i8 12, label %case.1
274    i8 70, label %case.2
275    i8 -22, label %case.3
276    i8 13, label %case.4
277    i8 0, label %case.5
278  ]
279
280case.0:
281  br label %merge
282case.1:
283  br label %merge
284case.2:
285  br label %merge
286case.3:
287  br label %merge
288case.4:
289  br label %merge
290case.5:
291  br label %merge
292case.default:
293  br label %merge
294
295merge:
296  %res = phi i32 [ 23, %case.default ], [ 13, %case.0 ], [ 53, %case.1 ], [ 33, %case.2 ], [ 23, %case.3 ], [ 644, %case.4 ], [ 54, %case.5 ]
297  call void @use(i32 %res)
298  ret i32 %res
299}
300
301; Same as for the switch, but written via manual branching.
302define signext i32 @branch_dispatch(i8 %a) {
303; CHECK-LABEL: branch_dispatch:
304; CHECK:       # %bb.0: # %case.0
305; CHECK-NEXT:    addi sp, sp, -16
306; CHECK-NEXT:    .cfi_def_cfa_offset 16
307; CHECK-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
308; CHECK-NEXT:    sd s0, 0(sp) # 8-byte Folded Spill
309; CHECK-NEXT:    .cfi_offset ra, -8
310; CHECK-NEXT:    .cfi_offset s0, -16
311; CHECK-NEXT:    .cfi_remember_state
312; CHECK-NEXT:    andi a0, a0, 255
313; CHECK-NEXT:    li a1, 32
314; CHECK-NEXT:    beq a0, a1, .LBB3_7
315; CHECK-NEXT:  # %bb.1: # %case.1
316; CHECK-NEXT:    li a1, 12
317; CHECK-NEXT:    beq a0, a1, .LBB3_8
318; CHECK-NEXT:  # %bb.2: # %case.2
319; CHECK-NEXT:    li a1, 70
320; CHECK-NEXT:    beq a0, a1, .LBB3_9
321; CHECK-NEXT:  # %bb.3: # %case.3
322; CHECK-NEXT:    li a1, 234
323; CHECK-NEXT:    li s0, 23
324; CHECK-NEXT:    beq a0, a1, .LBB3_10
325; CHECK-NEXT:  # %bb.4: # %case.4
326; CHECK-NEXT:    beqz a0, .LBB3_11
327; CHECK-NEXT:  # %bb.5: # %case.5
328; CHECK-NEXT:    li a1, 5
329; CHECK-NEXT:    bne a0, a1, .LBB3_10
330; CHECK-NEXT:  # %bb.6:
331; CHECK-NEXT:    li s0, 54
332; CHECK-NEXT:    j .LBB3_10
333; CHECK-NEXT:  .LBB3_7:
334; CHECK-NEXT:    li s0, 13
335; CHECK-NEXT:    j .LBB3_10
336; CHECK-NEXT:  .LBB3_8:
337; CHECK-NEXT:    li s0, 53
338; CHECK-NEXT:    j .LBB3_10
339; CHECK-NEXT:  .LBB3_9:
340; CHECK-NEXT:    li s0, 33
341; CHECK-NEXT:  .LBB3_10: # %merge
342; CHECK-NEXT:    mv a0, s0
343; CHECK-NEXT:    call use
344; CHECK-NEXT:    mv a0, s0
345; CHECK-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
346; CHECK-NEXT:    ld s0, 0(sp) # 8-byte Folded Reload
347; CHECK-NEXT:    .cfi_restore ra
348; CHECK-NEXT:    .cfi_restore s0
349; CHECK-NEXT:    addi sp, sp, 16
350; CHECK-NEXT:    .cfi_def_cfa_offset 0
351; CHECK-NEXT:    ret
352; CHECK-NEXT:  .LBB3_11:
353; CHECK-NEXT:    .cfi_restore_state
354; CHECK-NEXT:    li s0, 644
355; CHECK-NEXT:    j .LBB3_10
356case.0:
357  %c0 = icmp ne i8 %a, 32
358  br i1 %c0, label %case.1, label %merge
359case.1:
360  %c1 = icmp ne i8 %a, 12
361  br i1 %c1, label %case.2, label %merge
362case.2:
363  %c2 = icmp ne i8 %a, 70
364  br i1 %c2, label %case.3, label %merge
365case.3:
366  %c3 = icmp ne i8 %a, -22
367  br i1 %c3, label %case.4, label %merge
368case.4:
369  %c4 = icmp ne i8 %a, 0
370  br i1 %c4, label %case.5, label %merge
371case.5:
372  %c5 = icmp ne i8 %a, 5
373  br i1 %c5, label %case.default, label %merge
374case.default:
375  br label %merge
376
377merge:
378  %res = phi i32 [ 23, %case.default ], [ 13, %case.0 ], [ 53, %case.1 ], [ 33, %case.2 ], [ 23, %case.3 ], [ 644, %case.4 ], [ 54, %case.5 ]
379  call void @use(i32 %res)
380  ret i32 %res
381}
382
383
384declare void @use(i32)
385
386