1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -mtriple=riscv32 -verify-machineinstrs < %s \
3; RUN:   | FileCheck -check-prefix=RV32I-FPELIM %s
4; RUN: llc -mtriple=riscv32 -mattr=+f -target-abi ilp32f \
5; RUN:    -verify-machineinstrs < %s \
6; RUN:   | FileCheck -check-prefix=RV32I-FPELIM %s
7; RUN: llc -mtriple=riscv32 -verify-machineinstrs -frame-pointer=all < %s \
8; RUN:   | FileCheck -check-prefix=RV32I-WITHFP %s
9; RUN: llc -mtriple=riscv32 -verify-machineinstrs -frame-pointer=all \
10; RUN:   -mattr=+f -target-abi ilp32f < %s \
11; RUN:   | FileCheck -check-prefix=RV32I-WITHFP %s
12
13; This file contains tests that should have identical output for the ilp32,
14; and ilp32f. As well as calling convention details, we check that
15; ra and fp are consistently stored to fp-4 and fp-8.
16
17; Check that on RV32 ilp32[f], double is passed in a pair of registers. Unlike
18; the convention for varargs, this need not be an aligned pair.
19
20define i32 @callee_double_in_regs(i32 %a, double %b) nounwind {
21; RV32I-FPELIM-LABEL: callee_double_in_regs:
22; RV32I-FPELIM:       # %bb.0:
23; RV32I-FPELIM-NEXT:    addi sp, sp, -16
24; RV32I-FPELIM-NEXT:    sw ra, 12(sp)
25; RV32I-FPELIM-NEXT:    sw s0, 8(sp)
26; RV32I-FPELIM-NEXT:    mv s0, a0
27; RV32I-FPELIM-NEXT:    mv a0, a1
28; RV32I-FPELIM-NEXT:    mv a1, a2
29; RV32I-FPELIM-NEXT:    call __fixdfsi
30; RV32I-FPELIM-NEXT:    add a0, s0, a0
31; RV32I-FPELIM-NEXT:    lw s0, 8(sp)
32; RV32I-FPELIM-NEXT:    lw ra, 12(sp)
33; RV32I-FPELIM-NEXT:    addi sp, sp, 16
34; RV32I-FPELIM-NEXT:    ret
35;
36; RV32I-WITHFP-LABEL: callee_double_in_regs:
37; RV32I-WITHFP:       # %bb.0:
38; RV32I-WITHFP-NEXT:    addi sp, sp, -16
39; RV32I-WITHFP-NEXT:    sw ra, 12(sp)
40; RV32I-WITHFP-NEXT:    sw s0, 8(sp)
41; RV32I-WITHFP-NEXT:    sw s1, 4(sp)
42; RV32I-WITHFP-NEXT:    addi s0, sp, 16
43; RV32I-WITHFP-NEXT:    mv s1, a0
44; RV32I-WITHFP-NEXT:    mv a0, a1
45; RV32I-WITHFP-NEXT:    mv a1, a2
46; RV32I-WITHFP-NEXT:    call __fixdfsi
47; RV32I-WITHFP-NEXT:    add a0, s1, a0
48; RV32I-WITHFP-NEXT:    lw s1, 4(sp)
49; RV32I-WITHFP-NEXT:    lw s0, 8(sp)
50; RV32I-WITHFP-NEXT:    lw ra, 12(sp)
51; RV32I-WITHFP-NEXT:    addi sp, sp, 16
52; RV32I-WITHFP-NEXT:    ret
53  %b_fptosi = fptosi double %b to i32
54  %1 = add i32 %a, %b_fptosi
55  ret i32 %1
56}
57
58define i32 @caller_double_in_regs() nounwind {
59; RV32I-FPELIM-LABEL: caller_double_in_regs:
60; RV32I-FPELIM:       # %bb.0:
61; RV32I-FPELIM-NEXT:    addi sp, sp, -16
62; RV32I-FPELIM-NEXT:    sw ra, 12(sp)
63; RV32I-FPELIM-NEXT:    addi a0, zero, 1
64; RV32I-FPELIM-NEXT:    lui a2, 262144
65; RV32I-FPELIM-NEXT:    mv a1, zero
66; RV32I-FPELIM-NEXT:    call callee_double_in_regs
67; RV32I-FPELIM-NEXT:    lw ra, 12(sp)
68; RV32I-FPELIM-NEXT:    addi sp, sp, 16
69; RV32I-FPELIM-NEXT:    ret
70;
71; RV32I-WITHFP-LABEL: caller_double_in_regs:
72; RV32I-WITHFP:       # %bb.0:
73; RV32I-WITHFP-NEXT:    addi sp, sp, -16
74; RV32I-WITHFP-NEXT:    sw ra, 12(sp)
75; RV32I-WITHFP-NEXT:    sw s0, 8(sp)
76; RV32I-WITHFP-NEXT:    addi s0, sp, 16
77; RV32I-WITHFP-NEXT:    addi a0, zero, 1
78; RV32I-WITHFP-NEXT:    lui a2, 262144
79; RV32I-WITHFP-NEXT:    mv a1, zero
80; RV32I-WITHFP-NEXT:    call callee_double_in_regs
81; RV32I-WITHFP-NEXT:    lw s0, 8(sp)
82; RV32I-WITHFP-NEXT:    lw ra, 12(sp)
83; RV32I-WITHFP-NEXT:    addi sp, sp, 16
84; RV32I-WITHFP-NEXT:    ret
85  %1 = call i32 @callee_double_in_regs(i32 1, double 2.0)
86  ret i32 %1
87}
88
89; Check 2x*xlen values are aligned appropriately when passed on the stack
90; Must keep define on a single line due to an update_llc_test_checks.py limitation
91define i32 @callee_aligned_stack(i32 %a, i32 %b, fp128 %c, i32 %d, i32 %e, i64 %f, i32 %g, i32 %h, double %i, i32 %j, [2 x i32] %k) nounwind {
92; The double should be 8-byte aligned on the stack, but the two-element array
93; should only be 4-byte aligned
94; RV32I-FPELIM-LABEL: callee_aligned_stack:
95; RV32I-FPELIM:       # %bb.0:
96; RV32I-FPELIM-NEXT:    lw a0, 0(a2)
97; RV32I-FPELIM-NEXT:    lw a1, 20(sp)
98; RV32I-FPELIM-NEXT:    lw a2, 0(sp)
99; RV32I-FPELIM-NEXT:    lw a3, 8(sp)
100; RV32I-FPELIM-NEXT:    lw a4, 16(sp)
101; RV32I-FPELIM-NEXT:    add a0, a0, a7
102; RV32I-FPELIM-NEXT:    add a0, a0, a2
103; RV32I-FPELIM-NEXT:    add a0, a0, a3
104; RV32I-FPELIM-NEXT:    add a0, a0, a4
105; RV32I-FPELIM-NEXT:    add a0, a0, a1
106; RV32I-FPELIM-NEXT:    ret
107;
108; RV32I-WITHFP-LABEL: callee_aligned_stack:
109; RV32I-WITHFP:       # %bb.0:
110; RV32I-WITHFP-NEXT:    addi sp, sp, -16
111; RV32I-WITHFP-NEXT:    sw ra, 12(sp)
112; RV32I-WITHFP-NEXT:    sw s0, 8(sp)
113; RV32I-WITHFP-NEXT:    addi s0, sp, 16
114; RV32I-WITHFP-NEXT:    lw a0, 0(a2)
115; RV32I-WITHFP-NEXT:    lw a1, 20(s0)
116; RV32I-WITHFP-NEXT:    lw a2, 0(s0)
117; RV32I-WITHFP-NEXT:    lw a3, 8(s0)
118; RV32I-WITHFP-NEXT:    lw a4, 16(s0)
119; RV32I-WITHFP-NEXT:    add a0, a0, a7
120; RV32I-WITHFP-NEXT:    add a0, a0, a2
121; RV32I-WITHFP-NEXT:    add a0, a0, a3
122; RV32I-WITHFP-NEXT:    add a0, a0, a4
123; RV32I-WITHFP-NEXT:    add a0, a0, a1
124; RV32I-WITHFP-NEXT:    lw s0, 8(sp)
125; RV32I-WITHFP-NEXT:    lw ra, 12(sp)
126; RV32I-WITHFP-NEXT:    addi sp, sp, 16
127; RV32I-WITHFP-NEXT:    ret
128  %1 = bitcast fp128 %c to i128
129  %2 = trunc i128 %1 to i32
130  %3 = add i32 %2, %g
131  %4 = add i32 %3, %h
132  %5 = bitcast double %i to i64
133  %6 = trunc i64 %5 to i32
134  %7 = add i32 %4, %6
135  %8 = add i32 %7, %j
136  %9 = extractvalue [2 x i32] %k, 0
137  %10 = add i32 %8, %9
138  ret i32 %10
139}
140
141define void @caller_aligned_stack() nounwind {
142; The double should be 8-byte aligned on the stack, but the two-element array
143; should only be 4-byte aligned
144; RV32I-FPELIM-LABEL: caller_aligned_stack:
145; RV32I-FPELIM:       # %bb.0:
146; RV32I-FPELIM-NEXT:    addi sp, sp, -64
147; RV32I-FPELIM-NEXT:    sw ra, 60(sp)
148; RV32I-FPELIM-NEXT:    addi a0, zero, 18
149; RV32I-FPELIM-NEXT:    sw a0, 24(sp)
150; RV32I-FPELIM-NEXT:    addi a0, zero, 17
151; RV32I-FPELIM-NEXT:    sw a0, 20(sp)
152; RV32I-FPELIM-NEXT:    addi a0, zero, 16
153; RV32I-FPELIM-NEXT:    sw a0, 16(sp)
154; RV32I-FPELIM-NEXT:    lui a0, 262236
155; RV32I-FPELIM-NEXT:    addi a0, a0, 655
156; RV32I-FPELIM-NEXT:    sw a0, 12(sp)
157; RV32I-FPELIM-NEXT:    lui a0, 377487
158; RV32I-FPELIM-NEXT:    addi a0, a0, 1475
159; RV32I-FPELIM-NEXT:    sw a0, 8(sp)
160; RV32I-FPELIM-NEXT:    addi a0, zero, 15
161; RV32I-FPELIM-NEXT:    sw a0, 0(sp)
162; RV32I-FPELIM-NEXT:    lui a0, 262153
163; RV32I-FPELIM-NEXT:    addi a0, a0, 491
164; RV32I-FPELIM-NEXT:    sw a0, 44(sp)
165; RV32I-FPELIM-NEXT:    lui a0, 545260
166; RV32I-FPELIM-NEXT:    addi a0, a0, -1967
167; RV32I-FPELIM-NEXT:    sw a0, 40(sp)
168; RV32I-FPELIM-NEXT:    lui a0, 964690
169; RV32I-FPELIM-NEXT:    addi a0, a0, -328
170; RV32I-FPELIM-NEXT:    sw a0, 36(sp)
171; RV32I-FPELIM-NEXT:    lui a0, 335544
172; RV32I-FPELIM-NEXT:    addi t0, a0, 1311
173; RV32I-FPELIM-NEXT:    lui a0, 688509
174; RV32I-FPELIM-NEXT:    addi a5, a0, -2048
175; RV32I-FPELIM-NEXT:    addi a0, zero, 1
176; RV32I-FPELIM-NEXT:    addi a1, zero, 11
177; RV32I-FPELIM-NEXT:    addi a2, sp, 32
178; RV32I-FPELIM-NEXT:    addi a3, zero, 12
179; RV32I-FPELIM-NEXT:    addi a4, zero, 13
180; RV32I-FPELIM-NEXT:    addi a6, zero, 4
181; RV32I-FPELIM-NEXT:    addi a7, zero, 14
182; RV32I-FPELIM-NEXT:    sw t0, 32(sp)
183; RV32I-FPELIM-NEXT:    call callee_aligned_stack
184; RV32I-FPELIM-NEXT:    lw ra, 60(sp)
185; RV32I-FPELIM-NEXT:    addi sp, sp, 64
186; RV32I-FPELIM-NEXT:    ret
187;
188; RV32I-WITHFP-LABEL: caller_aligned_stack:
189; RV32I-WITHFP:       # %bb.0:
190; RV32I-WITHFP-NEXT:    addi sp, sp, -64
191; RV32I-WITHFP-NEXT:    sw ra, 60(sp)
192; RV32I-WITHFP-NEXT:    sw s0, 56(sp)
193; RV32I-WITHFP-NEXT:    addi s0, sp, 64
194; RV32I-WITHFP-NEXT:    addi a0, zero, 18
195; RV32I-WITHFP-NEXT:    sw a0, 24(sp)
196; RV32I-WITHFP-NEXT:    addi a0, zero, 17
197; RV32I-WITHFP-NEXT:    sw a0, 20(sp)
198; RV32I-WITHFP-NEXT:    addi a0, zero, 16
199; RV32I-WITHFP-NEXT:    sw a0, 16(sp)
200; RV32I-WITHFP-NEXT:    lui a0, 262236
201; RV32I-WITHFP-NEXT:    addi a0, a0, 655
202; RV32I-WITHFP-NEXT:    sw a0, 12(sp)
203; RV32I-WITHFP-NEXT:    lui a0, 377487
204; RV32I-WITHFP-NEXT:    addi a0, a0, 1475
205; RV32I-WITHFP-NEXT:    sw a0, 8(sp)
206; RV32I-WITHFP-NEXT:    addi a0, zero, 15
207; RV32I-WITHFP-NEXT:    sw a0, 0(sp)
208; RV32I-WITHFP-NEXT:    lui a0, 262153
209; RV32I-WITHFP-NEXT:    addi a0, a0, 491
210; RV32I-WITHFP-NEXT:    sw a0, -20(s0)
211; RV32I-WITHFP-NEXT:    lui a0, 545260
212; RV32I-WITHFP-NEXT:    addi a0, a0, -1967
213; RV32I-WITHFP-NEXT:    sw a0, -24(s0)
214; RV32I-WITHFP-NEXT:    lui a0, 964690
215; RV32I-WITHFP-NEXT:    addi a0, a0, -328
216; RV32I-WITHFP-NEXT:    sw a0, -28(s0)
217; RV32I-WITHFP-NEXT:    lui a0, 335544
218; RV32I-WITHFP-NEXT:    addi t0, a0, 1311
219; RV32I-WITHFP-NEXT:    lui a0, 688509
220; RV32I-WITHFP-NEXT:    addi a5, a0, -2048
221; RV32I-WITHFP-NEXT:    addi a0, zero, 1
222; RV32I-WITHFP-NEXT:    addi a1, zero, 11
223; RV32I-WITHFP-NEXT:    addi a2, s0, -32
224; RV32I-WITHFP-NEXT:    addi a3, zero, 12
225; RV32I-WITHFP-NEXT:    addi a4, zero, 13
226; RV32I-WITHFP-NEXT:    addi a6, zero, 4
227; RV32I-WITHFP-NEXT:    addi a7, zero, 14
228; RV32I-WITHFP-NEXT:    sw t0, -32(s0)
229; RV32I-WITHFP-NEXT:    call callee_aligned_stack
230; RV32I-WITHFP-NEXT:    lw s0, 56(sp)
231; RV32I-WITHFP-NEXT:    lw ra, 60(sp)
232; RV32I-WITHFP-NEXT:    addi sp, sp, 64
233; RV32I-WITHFP-NEXT:    ret
234  %1 = call i32 @callee_aligned_stack(i32 1, i32 11,
235    fp128 0xLEB851EB851EB851F400091EB851EB851, i32 12, i32 13,
236    i64 20000000000, i32 14, i32 15, double 2.720000e+00, i32 16,
237    [2 x i32] [i32 17, i32 18])
238  ret void
239}
240
241define double @callee_small_scalar_ret() nounwind {
242; RV32I-FPELIM-LABEL: callee_small_scalar_ret:
243; RV32I-FPELIM:       # %bb.0:
244; RV32I-FPELIM-NEXT:    lui a1, 261888
245; RV32I-FPELIM-NEXT:    mv a0, zero
246; RV32I-FPELIM-NEXT:    ret
247;
248; RV32I-WITHFP-LABEL: callee_small_scalar_ret:
249; RV32I-WITHFP:       # %bb.0:
250; RV32I-WITHFP-NEXT:    addi sp, sp, -16
251; RV32I-WITHFP-NEXT:    sw ra, 12(sp)
252; RV32I-WITHFP-NEXT:    sw s0, 8(sp)
253; RV32I-WITHFP-NEXT:    addi s0, sp, 16
254; RV32I-WITHFP-NEXT:    lui a1, 261888
255; RV32I-WITHFP-NEXT:    mv a0, zero
256; RV32I-WITHFP-NEXT:    lw s0, 8(sp)
257; RV32I-WITHFP-NEXT:    lw ra, 12(sp)
258; RV32I-WITHFP-NEXT:    addi sp, sp, 16
259; RV32I-WITHFP-NEXT:    ret
260  ret double 1.0
261}
262
263define i64 @caller_small_scalar_ret() nounwind {
264; RV32I-FPELIM-LABEL: caller_small_scalar_ret:
265; RV32I-FPELIM:       # %bb.0:
266; RV32I-FPELIM-NEXT:    addi sp, sp, -16
267; RV32I-FPELIM-NEXT:    sw ra, 12(sp)
268; RV32I-FPELIM-NEXT:    call callee_small_scalar_ret
269; RV32I-FPELIM-NEXT:    lw ra, 12(sp)
270; RV32I-FPELIM-NEXT:    addi sp, sp, 16
271; RV32I-FPELIM-NEXT:    ret
272;
273; RV32I-WITHFP-LABEL: caller_small_scalar_ret:
274; RV32I-WITHFP:       # %bb.0:
275; RV32I-WITHFP-NEXT:    addi sp, sp, -16
276; RV32I-WITHFP-NEXT:    sw ra, 12(sp)
277; RV32I-WITHFP-NEXT:    sw s0, 8(sp)
278; RV32I-WITHFP-NEXT:    addi s0, sp, 16
279; RV32I-WITHFP-NEXT:    call callee_small_scalar_ret
280; RV32I-WITHFP-NEXT:    lw s0, 8(sp)
281; RV32I-WITHFP-NEXT:    lw ra, 12(sp)
282; RV32I-WITHFP-NEXT:    addi sp, sp, 16
283; RV32I-WITHFP-NEXT:    ret
284  %1 = call double @callee_small_scalar_ret()
285  %2 = bitcast double %1 to i64
286  ret i64 %2
287}
288