1 ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=GCN %s
2 ; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=VI -check-prefix=GCN %s
4 ; GCN-LABEL: {{^}}mac_vvv:
5 ; GCN: buffer_load_dword [[A:v[0-9]+]], s[{{[0-9]+:[0-9]+}}], 0{{$}}
6 ; GCN: buffer_load_dword [[B:v[0-9]+]], s[{{[0-9]+:[0-9]+}}], 0 offset:4
7 ; GCN: buffer_load_dword [[C:v[0-9]+]], s[{{[0-9]+:[0-9]+}}], 0 offset:8
8 ; GCN: v_mac_f32_e32 [[C]], [[B]], [[A]]
9 ; GCN: buffer_store_dword [[C]]
10 define void @mac_vvv(float addrspace(1)* %out, float addrspace(1)* %in) {
12 %b_ptr = getelementptr float, float addrspace(1)* %in, i32 1
13 %c_ptr = getelementptr float, float addrspace(1)* %in, i32 2
15 %a = load float, float addrspace(1)* %in
16 %b = load float, float addrspace(1)* %b_ptr
17 %c = load float, float addrspace(1)* %c_ptr
19 %tmp0 = fmul float %a, %b
20 %tmp1 = fadd float %tmp0, %c
21 store float %tmp1, float addrspace(1)* %out
25 ; GCN-LABEL: {{^}}mad_inline_sgpr_inline:
27 ; GCN: v_mad_f32 v{{[0-9]}}, 0.5, s{{[0-9]+}}, 0.5
28 define void @mad_inline_sgpr_inline(float addrspace(1)* %out, float %in) {
30 %tmp0 = fmul float 0.5, %in
31 %tmp1 = fadd float %tmp0, 0.5
32 store float %tmp1, float addrspace(1)* %out
36 ; GCN-LABEL: {{^}}mad_vvs:
38 ; GCN: v_mad_f32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, s{{[0-9]+}}
39 define void @mad_vvs(float addrspace(1)* %out, float addrspace(1)* %in, float %c) {
41 %b_ptr = getelementptr float, float addrspace(1)* %in, i32 1
43 %a = load float, float addrspace(1)* %in
44 %b = load float, float addrspace(1)* %b_ptr
46 %tmp0 = fmul float %a, %b
47 %tmp1 = fadd float %tmp0, %c
48 store float %tmp1, float addrspace(1)* %out
52 ; GCN-LABEL: {{^}}mac_ssv:
53 ; GCN: v_mac_f32_e64 v{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
54 define void @mac_ssv(float addrspace(1)* %out, float addrspace(1)* %in, float %a) {
56 %c = load float, float addrspace(1)* %in
58 %tmp0 = fmul float %a, %a
59 %tmp1 = fadd float %tmp0, %c
60 store float %tmp1, float addrspace(1)* %out
64 ; GCN-LABEL: {{^}}mac_mad_same_add:
65 ; GCN: v_mad_f32 v{{[0-9]}}, v{{[0-9]+}}, v{{[0-9]+}}, [[ADD:v[0-9]+]]
66 ; GCN: v_mac_f32_e32 [[ADD]], v{{[0-9]+}}, v{{[0-9]+}}
67 define void @mac_mad_same_add(float addrspace(1)* %out, float addrspace(1)* %in) {
69 %b_ptr = getelementptr float, float addrspace(1)* %in, i32 1
70 %c_ptr = getelementptr float, float addrspace(1)* %in, i32 2
71 %d_ptr = getelementptr float, float addrspace(1)* %in, i32 3
72 %e_ptr = getelementptr float, float addrspace(1)* %in, i32 4
74 %a = load float, float addrspace(1)* %in
75 %b = load float, float addrspace(1)* %b_ptr
76 %c = load float, float addrspace(1)* %c_ptr
77 %d = load float, float addrspace(1)* %d_ptr
78 %e = load float, float addrspace(1)* %e_ptr
80 %tmp0 = fmul float %a, %b
81 %tmp1 = fadd float %tmp0, %c
83 %tmp2 = fmul float %d, %e
84 %tmp3 = fadd float %tmp2, %c
86 %out1 = getelementptr float, float addrspace(1)* %out, i32 1
87 store float %tmp1, float addrspace(1)* %out
88 store float %tmp3, float addrspace(1)* %out1
92 ; There is no advantage to using v_mac when one of the operands is negated
93 ; and v_mad accepts more operand types.
95 ; GCN-LABEL: {{^}}mad_neg_src0:
97 ; GCN: v_mad_f32 v{{[0-9]+}}, -v{{[0-9]+}}, v{{[0-9]+}}, v{{[-0-9]}}
98 define void @mad_neg_src0(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
100 %b_ptr = getelementptr float, float addrspace(1)* %in, i32 1
101 %c_ptr = getelementptr float, float addrspace(1)* %in, i32 2
103 %a = load float, float addrspace(1)* %in
104 %b = load float, float addrspace(1)* %b_ptr
105 %c = load float, float addrspace(1)* %c_ptr
107 %neg_a = fsub float 0.0, %a
108 %tmp0 = fmul float %neg_a, %b
109 %tmp1 = fadd float %tmp0, %c
111 store float %tmp1, float addrspace(1)* %out
115 ; GCN-LABEL: {{^}}mad_neg_src1:
117 ; GCN: v_mad_f32 v{{[0-9]+}}, -v{{[0-9]+}}, v{{[0-9]+}}, v{{[-0-9]}}
118 define void @mad_neg_src1(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
120 %b_ptr = getelementptr float, float addrspace(1)* %in, i32 1
121 %c_ptr = getelementptr float, float addrspace(1)* %in, i32 2
123 %a = load float, float addrspace(1)* %in
124 %b = load float, float addrspace(1)* %b_ptr
125 %c = load float, float addrspace(1)* %c_ptr
127 %neg_b = fsub float 0.0, %b
128 %tmp0 = fmul float %a, %neg_b
129 %tmp1 = fadd float %tmp0, %c
131 store float %tmp1, float addrspace(1)* %out
135 ; GCN-LABEL: {{^}}mad_neg_src2:
137 ; GCN: v_mad_f32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, -v{{[-0-9]}}
138 define void @mad_neg_src2(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
140 %b_ptr = getelementptr float, float addrspace(1)* %in, i32 1
141 %c_ptr = getelementptr float, float addrspace(1)* %in, i32 2
143 %a = load float, float addrspace(1)* %in
144 %b = load float, float addrspace(1)* %b_ptr
145 %c = load float, float addrspace(1)* %c_ptr
147 %neg_c = fsub float 0.0, %c
148 %tmp0 = fmul float %a, %b
149 %tmp1 = fadd float %tmp0, %neg_c
151 store float %tmp1, float addrspace(1)* %out
155 attributes #0 = { "true" "unsafe-fp-math"="true" }