1; RUN: llc -verify-machineinstrs -mtriple=powerpc64le-unknown-linux-gnu \
2; RUN:   -mcpu=future -ppc-asm-full-reg-names < %s \
3; RUN:   | FileCheck %s --check-prefixes=CHECK-S,CHECK-ALL
4; RUN: llc -verify-machineinstrs -mtriple=powerpc64le-unknown-linux-gnu \
5; RUN:   -mcpu=pwr9 -ppc-asm-full-reg-names < %s \
6; RUN:   | FileCheck %s --check-prefixes=CHECK-P9,CHECK-ALL
7; RUN: llc -verify-machineinstrs -mtriple=powerpc64le-unknown-linux-gnu \
8; RUN:   -mcpu=pwr9 --code-model=large -ppc-asm-full-reg-names < %s \
9; RUN:   | FileCheck %s --check-prefixes=CHECK-LARGE,CHECK-ALL
10
11@global_int = common dso_local local_unnamed_addr global i32 0, align 4
12
13define dso_local signext i32 @NoTOC() local_unnamed_addr {
14; CHECK-ALL-LABEL: NoTOC:
15; CHECK-S-NOT:     .localentry
16; CHECK-S:         li r3, 42
17; CHECK-S-NEXT:    blr
18entry:
19  ret i32 42
20}
21
22define dso_local signext i32 @AsmClobberX2(i32 signext %a, i32 signext %b) local_unnamed_addr {
23; CHECK-ALL-LABEL: AsmClobberX2:
24; CHECK-S:         .localentry AsmClobberX2, 1
25; CHECK-S:         add r3, r4, r3
26; CHECK-S:         #APP
27; CHECK-S-NEXT:    nop
28; CHECK-S-NEXT:    #NO_APP
29; CHECK-S:         blr
30entry:
31  %add = add nsw i32 %b, %a
32  tail call void asm sideeffect "nop", "~{r2}"()
33  ret i32 %add
34}
35
36; FIXME: This is actually a test case that shows a bug. On power9 and earlier
37;        this test should not compile. On later CPUs (like this test) the @toc
38;        should be replaced with @pcrel and we won't need R2 and so the problem
39;        goes away.
40define dso_local signext i32 @AsmClobberX2WithTOC(i32 signext %a, i32 signext %b) local_unnamed_addr {
41; CHECK-ALL-LABEL: AsmClobberX2WithTOC:
42; CHECK-S:         addis r2, r12, .TOC.-.Lfunc_gep2@ha
43; CHECK-S-NEXT:    addi r2, r2, .TOC.-.Lfunc_gep2@l
44; CHECK-LARGE:     ld r2, .Lfunc_toc2-.Lfunc_gep2(r12)
45; CHECK-LARGE:     add r2, r2, r12
46; CHECK-S:         .localentry     AsmClobberX2WithTOC, .Lfunc_lep2-.Lfunc_gep2
47; CHECK-S:         #APP
48; CHECK-S-NEXT:    li r2, 0
49; CHECK-S-NEXT:    #NO_APP
50; CHECK-S-NEXT:    addis r5, r2, global_int@toc@ha
51; CHECK-S-NEXT:    lwz r5, global_int@toc@l(r5)
52; CHECK-S-NEXT:    add r3, r4, r3
53; CHECK-S-NEXT:    add r3, r3, r5
54; CHECK-S-NEXT:    extsw r3, r3
55; CHECK-S-NEXT:    blr
56entry:
57  %add = add nsw i32 %b, %a
58  tail call void asm sideeffect "li 2, 0", "~{r2}"()
59  %0 = load i32, i32* @global_int, align 4
60  %add1 = add nsw i32 %add, %0
61  ret i32 %add1
62}
63
64define dso_local signext i32 @AsmClobberX5(i32 signext %a, i32 signext %b) local_unnamed_addr {
65; CHECK-ALL-LABEL: AsmClobberX5:
66; CHECK-S:         .localentry AsmClobberX5, 1
67; CHECK-P9-NOT:    .localentry
68; CHECK-ALL:       # %bb.0: # %entry
69; CHECK-S-NEXT:    add r3, r4, r3
70; CHECK-S-NEXT:    extsw r3, r3
71; CHECK-S-NEXT:    #APP
72; CHECK-S-NEXT:    nop
73; CHECK-S-NEXT:    #NO_APP
74; CHECK-S-NEXT:    blr
75entry:
76  %add = add nsw i32 %b, %a
77  tail call void asm sideeffect "nop", "~{r5}"()
78  ret i32 %add
79}
80
81; Clobber all GPRs except R2.
82define dso_local signext i32 @AsmClobberNotR2(i32 signext %a, i32 signext %b) local_unnamed_addr {
83; CHECK-ALL-LABEL: AsmClobberNotR2:
84; CHECK-S:         .localentry AsmClobberNotR2, 1
85; CHECK-P9-NOT:    .localentry
86; CHECK-S:         add r3, r4, r3
87; CHECK-S:         stw r3, -148(r1) # 4-byte Folded Spill
88; CHECK-S-NEXT:    #APP
89; CHECK-S-NEXT:    nop
90; CHECK-S-NEXT:    #NO_APP
91; CHECK-S-NEXT:    lwz r3, -148(r1) # 4-byte Folded Reload
92; CHECK-S:    blr
93entry:
94  %add = add nsw i32 %b, %a
95  tail call void asm sideeffect "nop", "~{r0},~{r1},~{r3},~{r4},~{r5},~{r6},~{r7},~{r8},~{r9},~{r10},~{r11},~{r12},~{r13},~{r14},~{r15},~{r16},~{r17},~{r18},~{r19},~{r20},~{r21},~{r22},~{r23},~{r24},~{r25},~{r26},~{r27},~{r28},~{r29},~{r30},~{r31}"()
96  ret i32 %add
97}
98
99; Increase register pressure enough to force the register allocator to
100; make use of R2.
101define dso_local signext i32 @X2IsCallerSaved(i32 signext %a, i32 signext %b, i32 signext %c, i32 signext %d, i32 signext %e, i32 signext %f, i32 signext %g, i32 signext %h) local_unnamed_addr {
102; CHECK-ALL-LABEL: X2IsCallerSaved:
103; CHECK-S:         .localentry X2IsCallerSaved, 1
104; CHECK-P9-NOT:    .localentry
105; CHECK-ALL:       # %bb.0: # %entry
106; CHECK-S-NEXT:    std r29, -24(r1) # 8-byte Folded Spill
107; CHECK-S-NEXT:    add r11, r4, r3
108; CHECK-S-NEXT:    subf r29, r9, r8
109; CHECK-S-NEXT:    add r9, r10, r9
110; CHECK-S-NEXT:    subf r10, r3, r10
111; CHECK-S-NEXT:    mullw r3, r4, r3
112; CHECK-S-NEXT:    mullw r3, r3, r11
113; CHECK-S-NEXT:    mullw r3, r3, r5
114; CHECK-S-NEXT:    subf r12, r5, r4
115; CHECK-S-NEXT:    mullw r3, r3, r6
116; CHECK-S-NEXT:    add r0, r6, r5
117; CHECK-S-NEXT:    mullw r3, r3, r12
118; CHECK-S-NEXT:    mullw r3, r3, r0
119; CHECK-S-NEXT:    mullw r3, r3, r7
120; CHECK-S-NEXT:    subf r2, r7, r6
121; CHECK-S-NEXT:    mullw r3, r3, r8
122; CHECK-S-NEXT:    std r30, -16(r1) # 8-byte Folded Spill
123; CHECK-S-NEXT:    add r30, r8, r7
124; CHECK-S-NEXT:    mullw r3, r3, r2
125; CHECK-S-NEXT:    mullw r3, r3, r30
126; CHECK-S-NEXT:    mullw r3, r3, r29
127; CHECK-S-NEXT:    mullw r3, r3, r9
128; CHECK-S-NEXT:    ld r30, -16(r1) # 8-byte Folded Reload
129; CHECK-S-NEXT:    ld r29, -24(r1) # 8-byte Folded Reload
130; CHECK-S-NEXT:    mullw r3, r3, r10
131; CHECK-S-NEXT:    extsw r3, r3
132; CHECK-S-NEXT:    blr
133entry:
134  %add = add nsw i32 %b, %a
135  %sub = sub nsw i32 %b, %c
136  %add1 = add nsw i32 %d, %c
137  %sub2 = sub nsw i32 %d, %e
138  %add3 = add nsw i32 %f, %e
139  %sub4 = sub nsw i32 %f, %g
140  %add5 = add nsw i32 %h, %g
141  %sub6 = sub nsw i32 %h, %a
142  %mul = mul i32 %b, %a
143  %mul7 = mul i32 %mul, %add
144  %mul8 = mul i32 %mul7, %c
145  %mul9 = mul i32 %mul8, %d
146  %mul10 = mul i32 %mul9, %sub
147  %mul11 = mul i32 %mul10, %add1
148  %mul12 = mul i32 %mul11, %e
149  %mul13 = mul i32 %mul12, %f
150  %mul14 = mul i32 %mul13, %sub2
151  %mul15 = mul i32 %mul14, %add3
152  %mul16 = mul i32 %mul15, %sub4
153  %mul17 = mul i32 %mul16, %add5
154  %mul18 = mul i32 %mul17, %sub6
155  ret i32 %mul18
156}
157
158
159define dso_local signext i32 @UsesX2AsTOC() local_unnamed_addr {
160; CHECK-ALL-LABEL: UsesX2AsTOC:
161; CHECK-S:         addis r2, r12, .TOC.-.Lfunc_gep6@ha
162; CHECK-S-NEXT:    addi r2, r2, .TOC.-.Lfunc_gep6@l
163; CHECK-LARGE:     ld r2, .Lfunc_toc6-.Lfunc_gep6(r12)
164; CHECK-LARGE:     add r2, r2, r12
165; CHECK-S:       .localentry     UsesX2AsTOC, .Lfunc_lep6-.Lfunc_gep6
166; CHECK-ALL:       # %bb.0: # %entry
167; CHECK-S-NEXT:    addis r3, r2, global_int@toc@ha
168; CHECK-S-NEXT:    lwa r3, global_int@toc@l(r3)
169; CHECK-S-NEXT:    blr
170entry:
171  %0 = load i32, i32* @global_int, align 4
172  ret i32 %0
173}
174
175
176define dso_local double @UsesX2AsConstPoolTOC() local_unnamed_addr {
177; CHECK-ALL-LABEL: UsesX2AsConstPoolTOC:
178; CHECK-LARGE:     ld r2, .Lfunc_toc7-.Lfunc_gep7(r12)
179; CHECK-LARGE:     add r2, r2, r12
180; CHECK-S-NOT:       .localentry
181; CHECK-ALL:       # %bb.0: # %entry
182; CHECK-S-NEXT:    plfd f1, .LCPI7_0@PCREL(0), 1
183; CHECK-S-NEXT:    blr
184entry:
185  ret double 0x404124A4EBDD334C
186}
187
188
189