xref: /llvm-project/llvm/test/CodeGen/RISCV/bf16-promote.ll (revision 9122c5235ec85ce0c0ad337e862b006e7b349d84)
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -mtriple=riscv64 -mattr=+d -target-abi=lp64d < %s | FileCheck --check-prefixes=CHECK,RV64 %s
3; RUN: llc -mtriple=riscv32 -mattr=+d -target-abi=ilp32d < %s | FileCheck --check-prefixes=CHECK,RV32 %s
4
5define void @test_load_store(ptr %p, ptr %q) nounwind {
6; CHECK-LABEL: test_load_store:
7; CHECK:       # %bb.0:
8; CHECK-NEXT:    lh a0, 0(a0)
9; CHECK-NEXT:    sh a0, 0(a1)
10; CHECK-NEXT:    ret
11  %a = load bfloat, ptr %p
12  store bfloat %a, ptr %q
13  ret void
14}
15
16define float @test_fpextend_float(ptr %p) nounwind {
17; CHECK-LABEL: test_fpextend_float:
18; CHECK:       # %bb.0:
19; CHECK-NEXT:    lhu a0, 0(a0)
20; CHECK-NEXT:    slli a0, a0, 16
21; CHECK-NEXT:    fmv.w.x fa0, a0
22; CHECK-NEXT:    ret
23  %a = load bfloat, ptr %p
24  %r = fpext bfloat %a to float
25  ret float %r
26}
27
28define double @test_fpextend_double(ptr %p) nounwind {
29; CHECK-LABEL: test_fpextend_double:
30; CHECK:       # %bb.0:
31; CHECK-NEXT:    lhu a0, 0(a0)
32; CHECK-NEXT:    slli a0, a0, 16
33; CHECK-NEXT:    fmv.w.x fa5, a0
34; CHECK-NEXT:    fcvt.d.s fa0, fa5
35; CHECK-NEXT:    ret
36  %a = load bfloat, ptr %p
37  %r = fpext bfloat %a to double
38  ret double %r
39}
40
41define void @test_fptrunc_float(float %f, ptr %p) nounwind {
42; RV64-LABEL: test_fptrunc_float:
43; RV64:       # %bb.0:
44; RV64-NEXT:    addi sp, sp, -16
45; RV64-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
46; RV64-NEXT:    sd s0, 0(sp) # 8-byte Folded Spill
47; RV64-NEXT:    mv s0, a0
48; RV64-NEXT:    call __truncsfbf2
49; RV64-NEXT:    fmv.x.w a0, fa0
50; RV64-NEXT:    sh a0, 0(s0)
51; RV64-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
52; RV64-NEXT:    ld s0, 0(sp) # 8-byte Folded Reload
53; RV64-NEXT:    addi sp, sp, 16
54; RV64-NEXT:    ret
55;
56; RV32-LABEL: test_fptrunc_float:
57; RV32:       # %bb.0:
58; RV32-NEXT:    addi sp, sp, -16
59; RV32-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
60; RV32-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
61; RV32-NEXT:    mv s0, a0
62; RV32-NEXT:    call __truncsfbf2
63; RV32-NEXT:    fmv.x.w a0, fa0
64; RV32-NEXT:    sh a0, 0(s0)
65; RV32-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
66; RV32-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
67; RV32-NEXT:    addi sp, sp, 16
68; RV32-NEXT:    ret
69  %a = fptrunc float %f to bfloat
70  store bfloat %a, ptr %p
71  ret void
72}
73
74define void @test_fptrunc_double(double %d, ptr %p) nounwind {
75; RV64-LABEL: test_fptrunc_double:
76; RV64:       # %bb.0:
77; RV64-NEXT:    addi sp, sp, -16
78; RV64-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
79; RV64-NEXT:    sd s0, 0(sp) # 8-byte Folded Spill
80; RV64-NEXT:    mv s0, a0
81; RV64-NEXT:    call __truncdfbf2
82; RV64-NEXT:    fmv.x.w a0, fa0
83; RV64-NEXT:    sh a0, 0(s0)
84; RV64-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
85; RV64-NEXT:    ld s0, 0(sp) # 8-byte Folded Reload
86; RV64-NEXT:    addi sp, sp, 16
87; RV64-NEXT:    ret
88;
89; RV32-LABEL: test_fptrunc_double:
90; RV32:       # %bb.0:
91; RV32-NEXT:    addi sp, sp, -16
92; RV32-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
93; RV32-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
94; RV32-NEXT:    mv s0, a0
95; RV32-NEXT:    call __truncdfbf2
96; RV32-NEXT:    fmv.x.w a0, fa0
97; RV32-NEXT:    sh a0, 0(s0)
98; RV32-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
99; RV32-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
100; RV32-NEXT:    addi sp, sp, 16
101; RV32-NEXT:    ret
102  %a = fptrunc double %d to bfloat
103  store bfloat %a, ptr %p
104  ret void
105}
106
107define void @test_fadd(ptr %p, ptr %q) nounwind {
108; RV64-LABEL: test_fadd:
109; RV64:       # %bb.0:
110; RV64-NEXT:    addi sp, sp, -16
111; RV64-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
112; RV64-NEXT:    sd s0, 0(sp) # 8-byte Folded Spill
113; RV64-NEXT:    mv s0, a0
114; RV64-NEXT:    lhu a0, 0(a1)
115; RV64-NEXT:    lhu a1, 0(s0)
116; RV64-NEXT:    slli a0, a0, 16
117; RV64-NEXT:    slli a1, a1, 16
118; RV64-NEXT:    fmv.w.x fa5, a0
119; RV64-NEXT:    fmv.w.x fa4, a1
120; RV64-NEXT:    fadd.s fa0, fa4, fa5
121; RV64-NEXT:    call __truncsfbf2
122; RV64-NEXT:    fmv.x.w a0, fa0
123; RV64-NEXT:    sh a0, 0(s0)
124; RV64-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
125; RV64-NEXT:    ld s0, 0(sp) # 8-byte Folded Reload
126; RV64-NEXT:    addi sp, sp, 16
127; RV64-NEXT:    ret
128;
129; RV32-LABEL: test_fadd:
130; RV32:       # %bb.0:
131; RV32-NEXT:    addi sp, sp, -16
132; RV32-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
133; RV32-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
134; RV32-NEXT:    mv s0, a0
135; RV32-NEXT:    lhu a0, 0(a1)
136; RV32-NEXT:    lhu a1, 0(s0)
137; RV32-NEXT:    slli a0, a0, 16
138; RV32-NEXT:    slli a1, a1, 16
139; RV32-NEXT:    fmv.w.x fa5, a0
140; RV32-NEXT:    fmv.w.x fa4, a1
141; RV32-NEXT:    fadd.s fa0, fa4, fa5
142; RV32-NEXT:    call __truncsfbf2
143; RV32-NEXT:    fmv.x.w a0, fa0
144; RV32-NEXT:    sh a0, 0(s0)
145; RV32-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
146; RV32-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
147; RV32-NEXT:    addi sp, sp, 16
148; RV32-NEXT:    ret
149  %a = load bfloat, ptr %p
150  %b = load bfloat, ptr %q
151  %r = fadd bfloat %a, %b
152  store bfloat %r, ptr %p
153  ret void
154}
155