1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -mtriple=riscv32 -mattr=+f -verify-machineinstrs < %s \
3; RUN:   -target-abi=ilp32f | FileCheck -check-prefix=RV32IF %s
4; RUN: llc -mtriple=riscv64 -mattr=+f -verify-machineinstrs < %s \
5; RUN:   -target-abi=lp64f | FileCheck -check-prefix=RV64IF %s
6
7define dso_local float @flw(float *%a) nounwind {
8; RV32IF-LABEL: flw:
9; RV32IF:       # %bb.0:
10; RV32IF-NEXT:    flw ft0, 0(a0)
11; RV32IF-NEXT:    flw ft1, 12(a0)
12; RV32IF-NEXT:    fadd.s fa0, ft0, ft1
13; RV32IF-NEXT:    ret
14;
15; RV64IF-LABEL: flw:
16; RV64IF:       # %bb.0:
17; RV64IF-NEXT:    flw ft0, 0(a0)
18; RV64IF-NEXT:    flw ft1, 12(a0)
19; RV64IF-NEXT:    fadd.s fa0, ft0, ft1
20; RV64IF-NEXT:    ret
21  %1 = load float, float* %a
22  %2 = getelementptr float, float* %a, i32 3
23  %3 = load float, float* %2
24; Use both loaded values in an FP op to ensure an flw is used, even for the
25; soft float ABI
26  %4 = fadd float %1, %3
27  ret float %4
28}
29
30define dso_local void @fsw(float *%a, float %b, float %c) nounwind {
31; Use %b and %c in an FP op to ensure floating point registers are used, even
32; for the soft float ABI
33; RV32IF-LABEL: fsw:
34; RV32IF:       # %bb.0:
35; RV32IF-NEXT:    fadd.s ft0, fa0, fa1
36; RV32IF-NEXT:    fsw ft0, 0(a0)
37; RV32IF-NEXT:    fsw ft0, 32(a0)
38; RV32IF-NEXT:    ret
39;
40; RV64IF-LABEL: fsw:
41; RV64IF:       # %bb.0:
42; RV64IF-NEXT:    fadd.s ft0, fa0, fa1
43; RV64IF-NEXT:    fsw ft0, 0(a0)
44; RV64IF-NEXT:    fsw ft0, 32(a0)
45; RV64IF-NEXT:    ret
46  %1 = fadd float %b, %c
47  store float %1, float* %a
48  %2 = getelementptr float, float* %a, i32 8
49  store float %1, float* %2
50  ret void
51}
52
53; Check load and store to a global
54@G = dso_local global float 0.0
55
56define dso_local float @flw_fsw_global(float %a, float %b) nounwind {
57; Use %a and %b in an FP op to ensure floating point registers are used, even
58; for the soft float ABI
59; RV32IF-LABEL: flw_fsw_global:
60; RV32IF:       # %bb.0:
61; RV32IF-NEXT:    fadd.s fa0, fa0, fa1
62; RV32IF-NEXT:    lui a0, %hi(G)
63; RV32IF-NEXT:    flw ft0, %lo(G)(a0)
64; RV32IF-NEXT:    fsw fa0, %lo(G)(a0)
65; RV32IF-NEXT:    addi a0, a0, %lo(G)
66; RV32IF-NEXT:    flw ft0, 36(a0)
67; RV32IF-NEXT:    fsw fa0, 36(a0)
68; RV32IF-NEXT:    ret
69;
70; RV64IF-LABEL: flw_fsw_global:
71; RV64IF:       # %bb.0:
72; RV64IF-NEXT:    fadd.s fa0, fa0, fa1
73; RV64IF-NEXT:    lui a0, %hi(G)
74; RV64IF-NEXT:    flw ft0, %lo(G)(a0)
75; RV64IF-NEXT:    fsw fa0, %lo(G)(a0)
76; RV64IF-NEXT:    addi a0, a0, %lo(G)
77; RV64IF-NEXT:    flw ft0, 36(a0)
78; RV64IF-NEXT:    fsw fa0, 36(a0)
79; RV64IF-NEXT:    ret
80  %1 = fadd float %a, %b
81  %2 = load volatile float, float* @G
82  store float %1, float* @G
83  %3 = getelementptr float, float* @G, i32 9
84  %4 = load volatile float, float* %3
85  store float %1, float* %3
86  ret float %1
87}
88
89; Ensure that 1 is added to the high 20 bits if bit 11 of the low part is 1
90define dso_local float @flw_fsw_constant(float %a) nounwind {
91; RV32IF-LABEL: flw_fsw_constant:
92; RV32IF:       # %bb.0:
93; RV32IF-NEXT:    lui a0, 912092
94; RV32IF-NEXT:    flw ft0, -273(a0)
95; RV32IF-NEXT:    fadd.s fa0, fa0, ft0
96; RV32IF-NEXT:    fsw fa0, -273(a0)
97; RV32IF-NEXT:    ret
98;
99; RV64IF-LABEL: flw_fsw_constant:
100; RV64IF:       # %bb.0:
101; RV64IF-NEXT:    lui a0, 228023
102; RV64IF-NEXT:    slli a0, a0, 2
103; RV64IF-NEXT:    flw ft0, -273(a0)
104; RV64IF-NEXT:    fadd.s fa0, fa0, ft0
105; RV64IF-NEXT:    fsw fa0, -273(a0)
106; RV64IF-NEXT:    ret
107  %1 = inttoptr i32 3735928559 to float*
108  %2 = load volatile float, float* %1
109  %3 = fadd float %a, %2
110  store float %3, float* %1
111  ret float %3
112}
113
114declare void @notdead(i8*)
115
116define dso_local float @flw_stack(float %a) nounwind {
117; RV32IF-LABEL: flw_stack:
118; RV32IF:       # %bb.0:
119; RV32IF-NEXT:    addi sp, sp, -16
120; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
121; RV32IF-NEXT:    fsw fs0, 8(sp) # 4-byte Folded Spill
122; RV32IF-NEXT:    fmv.s fs0, fa0
123; RV32IF-NEXT:    addi a0, sp, 4
124; RV32IF-NEXT:    call notdead@plt
125; RV32IF-NEXT:    flw ft0, 4(sp)
126; RV32IF-NEXT:    fadd.s fa0, ft0, fs0
127; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
128; RV32IF-NEXT:    flw fs0, 8(sp) # 4-byte Folded Reload
129; RV32IF-NEXT:    addi sp, sp, 16
130; RV32IF-NEXT:    ret
131;
132; RV64IF-LABEL: flw_stack:
133; RV64IF:       # %bb.0:
134; RV64IF-NEXT:    addi sp, sp, -16
135; RV64IF-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
136; RV64IF-NEXT:    fsw fs0, 4(sp) # 4-byte Folded Spill
137; RV64IF-NEXT:    fmv.s fs0, fa0
138; RV64IF-NEXT:    mv a0, sp
139; RV64IF-NEXT:    call notdead@plt
140; RV64IF-NEXT:    flw ft0, 0(sp)
141; RV64IF-NEXT:    fadd.s fa0, ft0, fs0
142; RV64IF-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
143; RV64IF-NEXT:    flw fs0, 4(sp) # 4-byte Folded Reload
144; RV64IF-NEXT:    addi sp, sp, 16
145; RV64IF-NEXT:    ret
146  %1 = alloca float, align 4
147  %2 = bitcast float* %1 to i8*
148  call void @notdead(i8* %2)
149  %3 = load float, float* %1
150  %4 = fadd float %3, %a ; force load in to FPR32
151  ret float %4
152}
153
154define dso_local void @fsw_stack(float %a, float %b) nounwind {
155; RV32IF-LABEL: fsw_stack:
156; RV32IF:       # %bb.0:
157; RV32IF-NEXT:    addi sp, sp, -16
158; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
159; RV32IF-NEXT:    fadd.s ft0, fa0, fa1
160; RV32IF-NEXT:    fsw ft0, 8(sp)
161; RV32IF-NEXT:    addi a0, sp, 8
162; RV32IF-NEXT:    call notdead@plt
163; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
164; RV32IF-NEXT:    addi sp, sp, 16
165; RV32IF-NEXT:    ret
166;
167; RV64IF-LABEL: fsw_stack:
168; RV64IF:       # %bb.0:
169; RV64IF-NEXT:    addi sp, sp, -16
170; RV64IF-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
171; RV64IF-NEXT:    fadd.s ft0, fa0, fa1
172; RV64IF-NEXT:    fsw ft0, 4(sp)
173; RV64IF-NEXT:    addi a0, sp, 4
174; RV64IF-NEXT:    call notdead@plt
175; RV64IF-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
176; RV64IF-NEXT:    addi sp, sp, 16
177; RV64IF-NEXT:    ret
178  %1 = fadd float %a, %b ; force store from FPR32
179  %2 = alloca float, align 4
180  store float %1, float* %2
181  %3 = bitcast float* %2 to i8*
182  call void @notdead(i8* %3)
183  ret void
184}
185