1 ; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG %s -check-prefix=FUNC
2 ; RUN: llc -march=r600 -mcpu=verde -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
4 ; mul24 and mad24 are affected
6 ; FUNC-LABEL: @test_mul_v2i32
7 ; EG: MULLO_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
8 ; EG: MULLO_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
10 ; SI: V_MUL_LO_I32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
11 ; SI: V_MUL_LO_I32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
13 define void @test_mul_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) {
14 %b_ptr = getelementptr <2 x i32> addrspace(1)* %in, i32 1
15 %a = load <2 x i32> addrspace(1) * %in
16 %b = load <2 x i32> addrspace(1) * %b_ptr
17 %result = mul <2 x i32> %a, %b
18 store <2 x i32> %result, <2 x i32> addrspace(1)* %out
22 ; FUNC-LABEL: @v_mul_v4i32
23 ; EG: MULLO_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
24 ; EG: MULLO_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
25 ; EG: MULLO_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
26 ; EG: MULLO_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
28 ; SI: V_MUL_LO_I32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
29 ; SI: V_MUL_LO_I32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
30 ; SI: V_MUL_LO_I32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
31 ; SI: V_MUL_LO_I32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
33 define void @v_mul_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
34 %b_ptr = getelementptr <4 x i32> addrspace(1)* %in, i32 1
35 %a = load <4 x i32> addrspace(1) * %in
36 %b = load <4 x i32> addrspace(1) * %b_ptr
37 %result = mul <4 x i32> %a, %b
38 store <4 x i32> %result, <4 x i32> addrspace(1)* %out
42 ; FUNC-LABEL: @s_trunc_i64_mul_to_i32
46 ; SI: BUFFER_STORE_DWORD
47 define void @s_trunc_i64_mul_to_i32(i32 addrspace(1)* %out, i64 %a, i64 %b) {
49 %trunc = trunc i64 %mul to i32
50 store i32 %trunc, i32 addrspace(1)* %out, align 8
54 ; FUNC-LABEL: @v_trunc_i64_mul_to_i32
58 ; SI: BUFFER_STORE_DWORD
59 define void @v_trunc_i64_mul_to_i32(i32 addrspace(1)* %out, i64 addrspace(1)* %aptr, i64 addrspace(1)* %bptr) nounwind {
60 %a = load i64 addrspace(1)* %aptr, align 8
61 %b = load i64 addrspace(1)* %bptr, align 8
63 %trunc = trunc i64 %mul to i32
64 store i32 %trunc, i32 addrspace(1)* %out, align 8
68 ; This 64-bit multiply should just use MUL_HI and MUL_LO, since the top
69 ; 32-bits of both arguments are sign bits.
70 ; FUNC-LABEL: @mul64_sext_c
74 ; SI-DAG: V_MUL_HI_I32
75 define void @mul64_sext_c(i64 addrspace(1)* %out, i32 %in) {
77 %0 = sext i32 %in to i64
79 store i64 %1, i64 addrspace(1)* %out
83 ; FUNC-LABEL: @v_mul64_sext_c:
86 ; SI-DAG: V_MUL_LO_I32
87 ; SI-DAG: V_MUL_HI_I32
89 define void @v_mul64_sext_c(i64 addrspace(1)* %out, i32 addrspace(1)* %in) {
90 %val = load i32 addrspace(1)* %in, align 4
91 %ext = sext i32 %val to i64
92 %mul = mul i64 %ext, 80
93 store i64 %mul, i64 addrspace(1)* %out, align 8
97 ; FUNC-LABEL: @v_mul64_sext_inline_imm:
98 ; SI-DAG: V_MUL_LO_I32 v{{[0-9]+}}, 9, v{{[0-9]+}}
99 ; SI-DAG: V_MUL_HI_I32 v{{[0-9]+}}, 9, v{{[0-9]+}}
101 define void @v_mul64_sext_inline_imm(i64 addrspace(1)* %out, i32 addrspace(1)* %in) {
102 %val = load i32 addrspace(1)* %in, align 4
103 %ext = sext i32 %val to i64
104 %mul = mul i64 %ext, 9
105 store i64 %mul, i64 addrspace(1)* %out, align 8
109 ; FUNC-LABEL: @s_mul_i32:
110 ; SI: S_LOAD_DWORD [[SRC0:s[0-9]+]],
111 ; SI: S_LOAD_DWORD [[SRC1:s[0-9]+]],
112 ; SI: S_MUL_I32 [[SRESULT:s[0-9]+]], [[SRC0]], [[SRC1]]
113 ; SI: V_MOV_B32_e32 [[VRESULT:v[0-9]+]], [[SRESULT]]
114 ; SI: BUFFER_STORE_DWORD [[VRESULT]],
116 define void @s_mul_i32(i32 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
117 %mul = mul i32 %a, %b
118 store i32 %mul, i32 addrspace(1)* %out, align 4
122 ; FUNC-LABEL: @v_mul_i32
123 ; SI: V_MUL_LO_I32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
124 define void @v_mul_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
125 %b_ptr = getelementptr i32 addrspace(1)* %in, i32 1
126 %a = load i32 addrspace(1)* %in
127 %b = load i32 addrspace(1)* %b_ptr
128 %result = mul i32 %a, %b
129 store i32 %result, i32 addrspace(1)* %out
133 ; A standard 64-bit multiply. The expansion should be around 6 instructions.
134 ; It would be difficult to match the expansion correctly without writing
135 ; a really complicated list of FileCheck expressions. I don't want
136 ; to confuse people who may 'break' this test with a correct optimization,
137 ; so this test just uses FUNC-LABEL to make sure the compiler does not
138 ; crash with a 'failed to select' error.
140 ; FUNC-LABEL: @s_mul_i64:
141 define void @s_mul_i64(i64 addrspace(1)* %out, i64 %a, i64 %b) nounwind {
142 %mul = mul i64 %a, %b
143 store i64 %mul, i64 addrspace(1)* %out, align 8
147 ; FUNC-LABEL: @v_mul_i64
149 define void @v_mul_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr, i64 addrspace(1)* %bptr) {
150 %a = load i64 addrspace(1)* %aptr, align 8
151 %b = load i64 addrspace(1)* %bptr, align 8
152 %mul = mul i64 %a, %b
153 store i64 %mul, i64 addrspace(1)* %out, align 8
157 ; FUNC-LABEL: @mul32_in_branch
159 define void @mul32_in_branch(i32 addrspace(1)* %out, i32 addrspace(1)* %in, i32 %a, i32 %b, i32 %c) {
161 %0 = icmp eq i32 %a, 0
162 br i1 %0, label %if, label %else
165 %1 = load i32 addrspace(1)* %in
173 %3 = phi i32 [%1, %if], [%2, %else]
174 store i32 %3, i32 addrspace(1)* %out
178 ; FUNC-LABEL: @mul64_in_branch
179 ; SI-DAG: V_MUL_LO_I32
180 ; SI-DAG: V_MUL_HI_U32
182 define void @mul64_in_branch(i64 addrspace(1)* %out, i64 addrspace(1)* %in, i64 %a, i64 %b, i64 %c) {
184 %0 = icmp eq i64 %a, 0
185 br i1 %0, label %if, label %else
188 %1 = load i64 addrspace(1)* %in
196 %3 = phi i64 [%1, %if], [%2, %else]
197 store i64 %3, i64 addrspace(1)* %out