1 ; RUN: llc -march=mips -mcpu=mips32r2 -mattr=dsp < %s | FileCheck %s -check-prefix=R1
2 ; RUN: llc -march=mips -mcpu=mips32r2 -mattr=dspr2 < %s | FileCheck %s -check-prefix=R2
7 define zeroext i8 @test_lbux(i8* nocapture %b, i32 %i) {
9 %add.ptr = getelementptr inbounds i8, i8* %b, i32 %i
10 %0 = load i8, i8* %add.ptr, align 1
17 define signext i16 @test_lhx(i16* nocapture %b, i32 %i) {
19 %add.ptr = getelementptr inbounds i16, i16* %b, i32 %i
20 %0 = load i16, i16* %add.ptr, align 2
27 define i32 @test_lwx(i32* nocapture %b, i32 %i) {
29 %add.ptr = getelementptr inbounds i32, i32* %b, i32 %i
30 %0 = load i32, i32* %add.ptr, align 4
34 ; R1-LABEL: test_add_v2q15_:
35 ; R1: addq.ph ${{[0-9]+}}
37 define { i32 } @test_add_v2q15_(i32 %a.coerce, i32 %b.coerce) {
39 %0 = bitcast i32 %a.coerce to <2 x i16>
40 %1 = bitcast i32 %b.coerce to <2 x i16>
41 %add = add <2 x i16> %0, %1
42 %2 = bitcast <2 x i16> %add to i32
43 %.fca.0.insert = insertvalue { i32 } undef, i32 %2, 0
44 ret { i32 } %.fca.0.insert
47 ; R1-LABEL: test_sub_v2q15_:
48 ; R1: subq.ph ${{[0-9]+}}
50 define { i32 } @test_sub_v2q15_(i32 %a.coerce, i32 %b.coerce) {
52 %0 = bitcast i32 %a.coerce to <2 x i16>
53 %1 = bitcast i32 %b.coerce to <2 x i16>
54 %sub = sub <2 x i16> %0, %1
55 %2 = bitcast <2 x i16> %sub to i32
56 %.fca.0.insert = insertvalue { i32 } undef, i32 %2, 0
57 ret { i32 } %.fca.0.insert
60 ; R2-LABEL: test_mul_v2q15_:
61 ; R2: mul.ph ${{[0-9]+}}
63 ; mul.ph is an R2 instruction. Check that multiply node gets expanded.
64 ; R1-LABEL: test_mul_v2q15_:
68 define { i32 } @test_mul_v2q15_(i32 %a.coerce, i32 %b.coerce) {
70 %0 = bitcast i32 %a.coerce to <2 x i16>
71 %1 = bitcast i32 %b.coerce to <2 x i16>
72 %mul = mul <2 x i16> %0, %1
73 %2 = bitcast <2 x i16> %mul to i32
74 %.fca.0.insert = insertvalue { i32 } undef, i32 %2, 0
75 ret { i32 } %.fca.0.insert
78 ; R1-LABEL: test_add_v4i8_:
79 ; R1: addu.qb ${{[0-9]+}}
81 define { i32 } @test_add_v4i8_(i32 %a.coerce, i32 %b.coerce) {
83 %0 = bitcast i32 %a.coerce to <4 x i8>
84 %1 = bitcast i32 %b.coerce to <4 x i8>
85 %add = add <4 x i8> %0, %1
86 %2 = bitcast <4 x i8> %add to i32
87 %.fca.0.insert = insertvalue { i32 } undef, i32 %2, 0
88 ret { i32 } %.fca.0.insert
91 ; R1-LABEL: test_sub_v4i8_:
92 ; R1: subu.qb ${{[0-9]+}}
94 define { i32 } @test_sub_v4i8_(i32 %a.coerce, i32 %b.coerce) {
96 %0 = bitcast i32 %a.coerce to <4 x i8>
97 %1 = bitcast i32 %b.coerce to <4 x i8>
98 %sub = sub <4 x i8> %0, %1
99 %2 = bitcast <4 x i8> %sub to i32
100 %.fca.0.insert = insertvalue { i32 } undef, i32 %2, 0
101 ret { i32 } %.fca.0.insert
104 ; DSP-ASE doesn't have a v4i8 multiply instruction. Check that multiply node gets expanded.
105 ; R2-LABEL: test_mul_v4i8_:
106 ; R2: mul ${{[0-9]+}}
107 ; R2: mul ${{[0-9]+}}
108 ; R2: mul ${{[0-9]+}}
109 ; R2: mul ${{[0-9]+}}
111 define { i32 } @test_mul_v4i8_(i32 %a.coerce, i32 %b.coerce) {
113 %0 = bitcast i32 %a.coerce to <4 x i8>
114 %1 = bitcast i32 %b.coerce to <4 x i8>
115 %mul = mul <4 x i8> %0, %1
116 %2 = bitcast <4 x i8> %mul to i32
117 %.fca.0.insert = insertvalue { i32 } undef, i32 %2, 0
118 ret { i32 } %.fca.0.insert
121 ; R1-LABEL: test_addsc:
122 ; R1: addsc ${{[0-9]+}}
123 ; R1: addwc ${{[0-9]+}}
125 define i64 @test_addsc(i64 %a, i64 %b) {
127 %add = add nsw i64 %b, %a
131 ; R1-LABEL: shift1_v2i16_shl_:
132 ; R1: shll.ph ${{[0-9]+}}, ${{[0-9]+}}, 15
134 define { i32 } @shift1_v2i16_shl_(i32 %a0.coerce) {
136 %0 = bitcast i32 %a0.coerce to <2 x i16>
137 %shl = shl <2 x i16> %0, <i16 15, i16 15>
138 %1 = bitcast <2 x i16> %shl to i32
139 %.fca.0.insert = insertvalue { i32 } undef, i32 %1, 0
140 ret { i32 } %.fca.0.insert
143 ; R1-LABEL: shift1_v2i16_sra_:
144 ; R1: shra.ph ${{[0-9]+}}, ${{[0-9]+}}, 15
146 define { i32 } @shift1_v2i16_sra_(i32 %a0.coerce) {
148 %0 = bitcast i32 %a0.coerce to <2 x i16>
149 %shr = ashr <2 x i16> %0, <i16 15, i16 15>
150 %1 = bitcast <2 x i16> %shr to i32
151 %.fca.0.insert = insertvalue { i32 } undef, i32 %1, 0
152 ret { i32 } %.fca.0.insert
155 ; R1-LABEL: shift1_v2ui16_srl_:
157 ; R2-LABEL: shift1_v2ui16_srl_:
158 ; R2: shrl.ph ${{[0-9]+}}, ${{[0-9]+}}, 15
160 define { i32 } @shift1_v2ui16_srl_(i32 %a0.coerce) {
162 %0 = bitcast i32 %a0.coerce to <2 x i16>
163 %shr = lshr <2 x i16> %0, <i16 15, i16 15>
164 %1 = bitcast <2 x i16> %shr to i32
165 %.fca.0.insert = insertvalue { i32 } undef, i32 %1, 0
166 ret { i32 } %.fca.0.insert
169 ; R1-LABEL: shift1_v4i8_shl_:
170 ; R1: shll.qb ${{[0-9]+}}, ${{[0-9]+}}, 7
172 define { i32 } @shift1_v4i8_shl_(i32 %a0.coerce) {
174 %0 = bitcast i32 %a0.coerce to <4 x i8>
175 %shl = shl <4 x i8> %0, <i8 7, i8 7, i8 7, i8 7>
176 %1 = bitcast <4 x i8> %shl to i32
177 %.fca.0.insert = insertvalue { i32 } undef, i32 %1, 0
178 ret { i32 } %.fca.0.insert
181 ; R1-LABEL: shift1_v4i8_sra_:
183 ; R2-LABEL: shift1_v4i8_sra_:
184 ; R2: shra.qb ${{[0-9]+}}, ${{[0-9]+}}, 7
186 define { i32 } @shift1_v4i8_sra_(i32 %a0.coerce) {
188 %0 = bitcast i32 %a0.coerce to <4 x i8>
189 %shr = ashr <4 x i8> %0, <i8 7, i8 7, i8 7, i8 7>
190 %1 = bitcast <4 x i8> %shr to i32
191 %.fca.0.insert = insertvalue { i32 } undef, i32 %1, 0
192 ret { i32 } %.fca.0.insert
195 ; R1-LABEL: shift1_v4ui8_srl_:
196 ; R1: shrl.qb ${{[0-9]+}}, ${{[0-9]+}}, 7
198 define { i32 } @shift1_v4ui8_srl_(i32 %a0.coerce) {
200 %0 = bitcast i32 %a0.coerce to <4 x i8>
201 %shr = lshr <4 x i8> %0, <i8 7, i8 7, i8 7, i8 7>
202 %1 = bitcast <4 x i8> %shr to i32
203 %.fca.0.insert = insertvalue { i32 } undef, i32 %1, 0
204 ret { i32 } %.fca.0.insert
207 ; Check that shift node is expanded if splat element size is not 16-bit.
209 ; R1-LABEL: test_vector_splat_imm_v2q15:
212 define { i32 } @test_vector_splat_imm_v2q15(i32 %a.coerce) {
214 %0 = bitcast i32 %a.coerce to <2 x i16>
215 %shl = shl <2 x i16> %0, <i16 0, i16 2>
216 %1 = bitcast <2 x i16> %shl to i32
217 %.fca.0.insert = insertvalue { i32 } undef, i32 %1, 0
218 ret { i32 } %.fca.0.insert
221 ; Check that shift node is expanded if splat element size is not 8-bit.
223 ; R1-LABEL: test_vector_splat_imm_v4i8:
226 define { i32 } @test_vector_splat_imm_v4i8(i32 %a.coerce) {
228 %0 = bitcast i32 %a.coerce to <4 x i8>
229 %shl = shl <4 x i8> %0, <i8 0, i8 2, i8 0, i8 2>
230 %1 = bitcast <4 x i8> %shl to i32
231 %.fca.0.insert = insertvalue { i32 } undef, i32 %1, 0
232 ret { i32 } %.fca.0.insert
235 ; Check that shift node is expanded if shift amount doesn't fit in 4-bit sa field.
237 ; R1-LABEL: test_shift_amount_v2q15:
240 define { i32 } @test_shift_amount_v2q15(i32 %a.coerce) {
242 %0 = bitcast i32 %a.coerce to <2 x i16>
243 %shl = shl <2 x i16> %0, <i16 16, i16 16>
244 %1 = bitcast <2 x i16> %shl to i32
245 %.fca.0.insert = insertvalue { i32 } undef, i32 %1, 0
246 ret { i32 } %.fca.0.insert
249 ; Check that shift node is expanded if shift amount doesn't fit in 3-bit sa field.
251 ; R1-LABEL: test_shift_amount_v4i8:
254 define { i32 } @test_shift_amount_v4i8(i32 %a.coerce) {
256 %0 = bitcast i32 %a.coerce to <4 x i8>
257 %shl = shl <4 x i8> %0, <i8 8, i8 8, i8 8, i8 8>
258 %1 = bitcast <4 x i8> %shl to i32
259 %.fca.0.insert = insertvalue { i32 } undef, i32 %1, 0
260 ret { i32 } %.fca.0.insert