1 ; RUN: llc < %s -mtriple=arm-apple-ios -mcpu=cortex-a8 -arm-atomic-cfg-tidy=0 | FileCheck %s
2 ; RUN: llc < %s -mtriple=arm-apple-ios -mcpu=cortex-a8 -arm-atomic-cfg-tidy=0 -regalloc=basic | FileCheck %s
3 ; Implementing vld / vst as REG_SEQUENCE eliminates the extra vmov's.
5 %struct.int16x8_t = type { <8 x i16> }
6 %struct.int32x4_t = type { <4 x i32> }
7 %struct.__neon_int8x8x2_t = type { <8 x i8>, <8 x i8> }
8 %struct.__neon_int8x8x3_t = type { <8 x i8>, <8 x i8>, <8 x i8> }
9 %struct.__neon_int16x8x2_t = type { <8 x i16>, <8 x i16> }
10 %struct.__neon_int32x4x2_t = type { <4 x i32>, <4 x i32> }
12 define void @t1(i16* %i_ptr, i16* %o_ptr, %struct.int32x4_t* nocapture %vT0ptr, %struct.int32x4_t* nocapture %vT1ptr) nounwind {
22 %0 = getelementptr inbounds %struct.int32x4_t, %struct.int32x4_t* %vT0ptr, i32 0, i32 0 ; <<4 x i32>*> [#uses=1]
23 %1 = load <4 x i32>, <4 x i32>* %0, align 16 ; <<4 x i32>> [#uses=1]
24 %2 = getelementptr inbounds %struct.int32x4_t, %struct.int32x4_t* %vT1ptr, i32 0, i32 0 ; <<4 x i32>*> [#uses=1]
25 %3 = load <4 x i32>, <4 x i32>* %2, align 16 ; <<4 x i32>> [#uses=1]
26 %4 = bitcast i16* %i_ptr to i8* ; <i8*> [#uses=1]
27 %5 = tail call <8 x i16> @llvm.arm.neon.vld1.v8i16.p0i8(i8* %4, i32 1) ; <<8 x i16>> [#uses=1]
28 %6 = bitcast <8 x i16> %5 to <2 x double> ; <<2 x double>> [#uses=2]
29 %7 = extractelement <2 x double> %6, i32 0 ; <double> [#uses=1]
30 %8 = bitcast double %7 to <4 x i16> ; <<4 x i16>> [#uses=1]
31 %9 = sext <4 x i16> %8 to <4 x i32> ; <<4 x i32>> [#uses=1]
32 %10 = extractelement <2 x double> %6, i32 1 ; <double> [#uses=1]
33 %11 = bitcast double %10 to <4 x i16> ; <<4 x i16>> [#uses=1]
34 %12 = sext <4 x i16> %11 to <4 x i32> ; <<4 x i32>> [#uses=1]
35 %13 = mul <4 x i32> %1, %9 ; <<4 x i32>> [#uses=1]
36 %14 = mul <4 x i32> %3, %12 ; <<4 x i32>> [#uses=1]
37 %15 = lshr <4 x i32> %13, <i32 12, i32 12, i32 12, i32 12>
38 %trunc_15 = trunc <4 x i32> %15 to <4 x i16>
39 %16 = lshr <4 x i32> %14, <i32 12, i32 12, i32 12, i32 12>
40 %trunc_16 = trunc <4 x i32> %16 to <4 x i16>
41 %17 = shufflevector <4 x i16> %trunc_15, <4 x i16> %trunc_16, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7> ; <<8 x i16>> [#uses=1]
42 %18 = bitcast i16* %o_ptr to i8* ; <i8*> [#uses=1]
43 tail call void @llvm.arm.neon.vst1.p0i8.v8i16(i8* %18, <8 x i16> %17, i32 1)
47 define void @t2(i16* %i_ptr, i16* %o_ptr, %struct.int16x8_t* nocapture %vT0ptr, %struct.int16x8_t* nocapture %vT1ptr) nounwind {
58 %0 = getelementptr inbounds %struct.int16x8_t, %struct.int16x8_t* %vT0ptr, i32 0, i32 0 ; <<8 x i16>*> [#uses=1]
59 %1 = load <8 x i16>, <8 x i16>* %0, align 16 ; <<8 x i16>> [#uses=1]
60 %2 = getelementptr inbounds %struct.int16x8_t, %struct.int16x8_t* %vT1ptr, i32 0, i32 0 ; <<8 x i16>*> [#uses=1]
61 %3 = load <8 x i16>, <8 x i16>* %2, align 16 ; <<8 x i16>> [#uses=1]
62 %4 = bitcast i16* %i_ptr to i8* ; <i8*> [#uses=1]
63 %5 = tail call <8 x i16> @llvm.arm.neon.vld1.v8i16.p0i8(i8* %4, i32 1) ; <<8 x i16>> [#uses=1]
64 %6 = getelementptr inbounds i16, i16* %i_ptr, i32 8 ; <i16*> [#uses=1]
65 %7 = bitcast i16* %6 to i8* ; <i8*> [#uses=1]
66 %8 = tail call <8 x i16> @llvm.arm.neon.vld1.v8i16.p0i8(i8* %7, i32 1) ; <<8 x i16>> [#uses=1]
67 %9 = mul <8 x i16> %1, %5 ; <<8 x i16>> [#uses=1]
68 %10 = mul <8 x i16> %3, %8 ; <<8 x i16>> [#uses=1]
69 %11 = bitcast i16* %o_ptr to i8* ; <i8*> [#uses=1]
70 tail call void @llvm.arm.neon.vst1.p0i8.v8i16(i8* %11, <8 x i16> %9, i32 1)
71 %12 = getelementptr inbounds i16, i16* %o_ptr, i32 8 ; <i16*> [#uses=1]
72 %13 = bitcast i16* %12 to i8* ; <i8*> [#uses=1]
73 tail call void @llvm.arm.neon.vst1.p0i8.v8i16(i8* %13, <8 x i16> %10, i32 1)
77 define <8 x i8> @t3(i8* %A, i8* %B) nounwind {
84 %tmp1 = call %struct.__neon_int8x8x3_t @llvm.arm.neon.vld3.v8i8.p0i8(i8* %A, i32 1) ; <%struct.__neon_int8x8x3_t> [#uses=2]
85 %tmp2 = extractvalue %struct.__neon_int8x8x3_t %tmp1, 0 ; <<8 x i8>> [#uses=1]
86 %tmp3 = extractvalue %struct.__neon_int8x8x3_t %tmp1, 2 ; <<8 x i8>> [#uses=1]
87 %tmp4 = extractvalue %struct.__neon_int8x8x3_t %tmp1, 1 ; <<8 x i8>> [#uses=1]
88 %tmp5 = sub <8 x i8> %tmp3, %tmp4
89 %tmp6 = add <8 x i8> %tmp2, %tmp3 ; <<8 x i8>> [#uses=1]
90 %tmp7 = mul <8 x i8> %tmp4, %tmp2
91 tail call void @llvm.arm.neon.vst3.p0i8.v8i8(i8* %B, <8 x i8> %tmp5, <8 x i8> %tmp6, <8 x i8> %tmp7, i32 1)
95 define void @t4(i32* %in, i32* %out) nounwind {
103 %tmp1 = bitcast i32* %in to i8* ; <i8*> [#uses=1]
104 %tmp2 = tail call %struct.__neon_int32x4x2_t @llvm.arm.neon.vld2.v4i32.p0i8(i8* %tmp1, i32 1) ; <%struct.__neon_int32x4x2_t> [#uses=2]
105 %tmp3 = getelementptr inbounds i32, i32* %in, i32 8 ; <i32*> [#uses=1]
106 %tmp4 = bitcast i32* %tmp3 to i8* ; <i8*> [#uses=1]
107 %tmp5 = tail call %struct.__neon_int32x4x2_t @llvm.arm.neon.vld2.v4i32.p0i8(i8* %tmp4, i32 1) ; <%struct.__neon_int32x4x2_t> [#uses=2]
108 %tmp8 = bitcast i32* %out to i8* ; <i8*> [#uses=1]
109 br i1 undef, label %return1, label %return2
114 ; CHECK-NEXT: vadd.i32
115 ; CHECK-NEXT: vadd.i32
116 ; CHECK-NEXT: vst2.32
117 %tmp52 = extractvalue %struct.__neon_int32x4x2_t %tmp2, 0 ; <<4 x i32>> [#uses=1]
118 %tmp57 = extractvalue %struct.__neon_int32x4x2_t %tmp2, 1 ; <<4 x i32>> [#uses=1]
119 %tmp = extractvalue %struct.__neon_int32x4x2_t %tmp5, 0 ; <<4 x i32>> [#uses=1]
120 %tmp39 = extractvalue %struct.__neon_int32x4x2_t %tmp5, 1 ; <<4 x i32>> [#uses=1]
121 %tmp6 = add <4 x i32> %tmp52, %tmp ; <<4 x i32>> [#uses=1]
122 %tmp7 = add <4 x i32> %tmp57, %tmp39 ; <<4 x i32>> [#uses=1]
123 tail call void @llvm.arm.neon.vst2.p0i8.v4i32(i8* %tmp8, <4 x i32> %tmp6, <4 x i32> %tmp7, i32 1)
130 ; CHECK: vst2.32 {d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}}
131 %tmp100 = extractvalue %struct.__neon_int32x4x2_t %tmp2, 0 ; <<4 x i32>> [#uses=1]
132 %tmp101 = extractvalue %struct.__neon_int32x4x2_t %tmp5, 1 ; <<4 x i32>> [#uses=1]
133 %tmp102 = add <4 x i32> %tmp100, %tmp101 ; <<4 x i32>> [#uses=1]
134 tail call void @llvm.arm.neon.vst2.p0i8.v4i32(i8* %tmp8, <4 x i32> %tmp102, <4 x i32> %tmp101, i32 1)
135 call void @llvm.trap()
139 define <8 x i16> @t5(i16* %A, <8 x i16>* %B) nounwind {
142 ; How can FileCheck match Q and D registers? We need a lisp interpreter.
143 ; CHECK: vorr {{q[0-9]+}}, {{q[0-9]+}}, {{q[0-9]+}}
145 ; CHECK: vld2.16 {d{{[0-9]+}}[1], d{{[0-9]+}}[1]}, [r0]
148 %tmp0 = bitcast i16* %A to i8* ; <i8*> [#uses=1]
149 %tmp1 = load <8 x i16>, <8 x i16>* %B ; <<8 x i16>> [#uses=2]
150 %tmp2 = call %struct.__neon_int16x8x2_t @llvm.arm.neon.vld2lane.v8i16.p0i8(i8* %tmp0, <8 x i16> %tmp1, <8 x i16> %tmp1, i32 1, i32 1) ; <%struct.__neon_int16x8x2_t> [#uses=2]
151 %tmp3 = extractvalue %struct.__neon_int16x8x2_t %tmp2, 0 ; <<8 x i16>> [#uses=1]
152 %tmp4 = extractvalue %struct.__neon_int16x8x2_t %tmp2, 1 ; <<8 x i16>> [#uses=1]
153 %tmp5 = add <8 x i16> %tmp3, %tmp4 ; <<8 x i16>> [#uses=1]
157 define <8 x i8> @t6(i8* %A, <8 x i8>* %B) nounwind {
160 ; CHECK: vorr d[[D0:[0-9]+]], d[[D1:[0-9]+]]
161 ; CHECK-NEXT: vld2.8 {d[[D1]][1], d[[D0]][1]}
162 %tmp1 = load <8 x i8>, <8 x i8>* %B ; <<8 x i8>> [#uses=2]
163 %tmp2 = call %struct.__neon_int8x8x2_t @llvm.arm.neon.vld2lane.v8i8.p0i8(i8* %A, <8 x i8> %tmp1, <8 x i8> %tmp1, i32 1, i32 1) ; <%struct.__neon_int8x8x2_t> [#uses=2]
164 %tmp3 = extractvalue %struct.__neon_int8x8x2_t %tmp2, 0 ; <<8 x i8>> [#uses=1]
165 %tmp4 = extractvalue %struct.__neon_int8x8x2_t %tmp2, 1 ; <<8 x i8>> [#uses=1]
166 %tmp5 = add <8 x i8> %tmp3, %tmp4 ; <<8 x i8>> [#uses=1]
170 define void @t7(i32* %iptr, i32* %optr) nounwind {
175 ; CHECK: vld1.32 {d{{[0-9]+}}, d{{[0-9]+}}},
176 ; CHECK: vorr q[[Q0:[0-9]+]], q[[Q1:[0-9]+]], q[[Q1:[0-9]+]]
178 ; CHECK: vuzp.32 q[[Q1]], q[[Q0]]
180 %0 = bitcast i32* %iptr to i8* ; <i8*> [#uses=2]
181 %1 = tail call %struct.__neon_int32x4x2_t @llvm.arm.neon.vld2.v4i32.p0i8(i8* %0, i32 1) ; <%struct.__neon_int32x4x2_t> [#uses=2]
182 %tmp57 = extractvalue %struct.__neon_int32x4x2_t %1, 0 ; <<4 x i32>> [#uses=1]
183 %tmp60 = extractvalue %struct.__neon_int32x4x2_t %1, 1 ; <<4 x i32>> [#uses=1]
184 %2 = bitcast i32* %optr to i8* ; <i8*> [#uses=2]
185 tail call void @llvm.arm.neon.vst2.p0i8.v4i32(i8* %2, <4 x i32> %tmp57, <4 x i32> %tmp60, i32 1)
186 %3 = tail call <4 x i32> @llvm.arm.neon.vld1.v4i32.p0i8(i8* %0, i32 1) ; <<4 x i32>> [#uses=1]
187 %4 = shufflevector <4 x i32> %3, <4 x i32> undef, <4 x i32> <i32 0, i32 2, i32 0, i32 2> ; <<4 x i32>> [#uses=1]
188 tail call void @llvm.arm.neon.vst1.p0i8.v4i32(i8* %2, <4 x i32> %4, i32 1)
193 define arm_aapcs_vfpcc i32 @t8() nounwind {
195 ; CHECK: vrsqrte.f32 q8, q8
196 bb.nph55.bb.nph55.split_crit_edge:
199 bb3: ; preds = %bb3, %bb.nph55.bb.nph55.split_crit_edge
200 br i1 undef, label %bb5, label %bb3
205 bb.i25: ; preds = %bb.i25, %bb5
206 %0 = shufflevector <2 x float> undef, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; <<4 x float>> [#uses=1]
207 %1 = call <4 x float> @llvm.arm.neon.vrsqrte.v4f32(<4 x float> %0) nounwind ; <<4 x float>> [#uses=1]
208 %2 = fmul <4 x float> %1, %1 ; <<4 x float>> [#uses=1]
209 %3 = fmul <4 x float> %2, %2 ; <<4 x float>> [#uses=1]
210 %tmp26.i = bitcast <4 x float> %3 to <2 x double> ; <<2 x double>> [#uses=1]
211 %4 = extractelement <2 x double> %tmp26.i, i32 0 ; <double> [#uses=1]
212 %5 = bitcast double %4 to <2 x float> ; <<2 x float>> [#uses=1]
213 %6 = extractelement <2 x float> %5, i32 1 ; <float> [#uses=1]
214 store float %6, float* undef, align 4
215 br i1 undef, label %bb6, label %bb.i25
217 bb6: ; preds = %bb.i25
218 br i1 undef, label %bb7, label %bb14
223 bb.i49: ; preds = %bb.i49, %bb7
224 br i1 undef, label %bb.i19, label %bb.i49
226 bb.i19: ; preds = %bb.i19, %bb.i49
227 br i1 undef, label %exit, label %bb.i19
229 exit: ; preds = %bb.i19
236 %0 = type { %1, %1, %1, %1 }
238 %2 = type { <4 x float> }
242 define arm_aapcs_vfpcc float @t9(%0* nocapture, %3* nocapture) nounwind {
244 ; CHECK: vmov.i32 d16, #0x0
245 ; CHECK-NEXT: vst1.64 {d16, d17}, [r0:128]
246 ; CHECK-NEXT: vorr d17, d16, d16
247 ; CHECK-NEXT: vst1.64 {d16, d17}, [r0:128]
248 %3 = bitcast double 0.000000e+00 to <2 x float> ; <<2 x float>> [#uses=2]
249 %4 = shufflevector <2 x float> %3, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; <<4 x float>> [#uses=1]
250 store <4 x float> %4, <4 x float>* undef, align 16
251 %5 = shufflevector <2 x float> %3, <2 x float> zeroinitializer, <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; <<4 x float>> [#uses=1]
252 store <4 x float> %5, <4 x float>* undef, align 16
255 ; <label>:6 ; preds = %8
258 ; <label>:7 ; preds = %6
261 ; <label>:8 ; preds = %7, %2
264 ; <label>:9 ; preds = %8
267 ; <label>:10 ; preds = %6
268 ret float 9.990000e+02
272 define arm_aapcs_vfpcc i32 @t10(float %x) nounwind {
274 ; CHECK: vdup.32 [[Q0:q[0-9]+]], d0[0]
275 ; CHECK: vmov.i32 [[Q9:q[0-9]+]], #0x3f000000
276 ; CHECK: vmul.f32 [[Q8:q[0-9]+]], [[Q0]], [[Q0]]
277 ; CHECK-NEXT: vadd.f32 [[Q8]], [[Q8]], [[Q8]]
278 ; CHECK-NEXT: vadd.f32 [[Q1:q[0-9]+]], [[Q8]], [[Q8]]
279 ; CHECK-NEXT: vmul.f32 [[Q8]], [[Q9]], d1[0]
280 ; CHECK-NEXT: vmul.f32 [[Q8]], [[Q8]], [[Q8]]
281 ; CHECK-NEXT: vadd.f32 [[Q8]], [[Q8]], [[Q8]]
282 ; CHECK-NEXT: vmul.f32 [[Q8]], [[Q8]], [[Q8]]
283 ; CHECK-NEXT: vst1.32 {d17[1]}, [r0:32]
284 ; CHECK-NEXT: mov r0, #0
285 ; CHECK-NEXT: cmp r0, #0
286 ; CHECK-NEXT: movne r0, #0
287 ; CHECK-NEXT: bxne lr
290 %0 = shufflevector <4 x float> zeroinitializer, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>> [#uses=1]
291 %1 = insertelement <4 x float> %0, float %x, i32 1 ; <<4 x float>> [#uses=1]
292 %2 = insertelement <4 x float> %1, float %x, i32 2 ; <<4 x float>> [#uses=1]
293 %3 = insertelement <4 x float> %2, float %x, i32 3 ; <<4 x float>> [#uses=1]
294 %tmp54.i = bitcast <4 x float> %3 to <2 x double> ; <<2 x double>> [#uses=1]
295 %4 = extractelement <2 x double> %tmp54.i, i32 1 ; <double> [#uses=1]
296 %5 = bitcast double %4 to <2 x float> ; <<2 x float>> [#uses=1]
297 %6 = shufflevector <2 x float> %5, <2 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>> [#uses=1]
298 %7 = fmul <4 x float> %6, %6
299 %8 = fadd <4 x float> %7, %7
300 %9 = fadd <4 x float> %8, %8
301 %10 = shufflevector <4 x float> undef, <4 x float> %9, <4 x i32> <i32 0, i32 1, i32 2, i32 7> ; <<4 x float>> [#uses=1]
302 %11 = fmul <4 x float> %10, <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01> ; <<4 x float>> [#uses=1]
303 %12 = shufflevector <4 x float> %11, <4 x float> undef, <4 x i32> <i32 3, i32 undef, i32 undef, i32 undef> ; <<4 x float>> [#uses=1]
304 %13 = shufflevector <4 x float> %12, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>> [#uses=1]
305 %14 = fmul <4 x float> %13, %13
306 %15 = fadd <4 x float> %14, %14
307 %16 = shufflevector <4 x float> undef, <4 x float> %15, <4 x i32> <i32 0, i32 1, i32 6, i32 3> ; <<4 x float>> [#uses=1]
308 %17 = fmul <4 x float> %16, %16
309 %18 = extractelement <4 x float> %17, i32 2 ; <float> [#uses=1]
310 store float %18, float* undef, align 4
311 br i1 undef, label %exit, label %bb14
313 exit: ; preds = %bb.i19
320 ; This test crashes the coalescer because live variables were not updated properly.
321 define <8 x i8> @t11(i8* %A1, i8* %A2, i8* %A3, i8* %A4, i8* %A5, i8* %A6, i8* %A7, i8* %A8, i8* %B) nounwind {
322 %tmp1d = call %struct.__neon_int8x8x3_t @llvm.arm.neon.vld3.v8i8.p0i8(i8* %A4, i32 1) ; <%struct.__neon_int8x8x3_t> [#uses=1]
323 %tmp2d = extractvalue %struct.__neon_int8x8x3_t %tmp1d, 0 ; <<8 x i8>> [#uses=1]
324 %tmp1f = call %struct.__neon_int8x8x3_t @llvm.arm.neon.vld3.v8i8.p0i8(i8* %A6, i32 1) ; <%struct.__neon_int8x8x3_t> [#uses=1]
325 %tmp2f = extractvalue %struct.__neon_int8x8x3_t %tmp1f, 0 ; <<8 x i8>> [#uses=1]
326 %tmp2bd = add <8 x i8> zeroinitializer, %tmp2d ; <<8 x i8>> [#uses=1]
327 %tmp2abcd = mul <8 x i8> zeroinitializer, %tmp2bd ; <<8 x i8>> [#uses=1]
328 %tmp2ef = sub <8 x i8> zeroinitializer, %tmp2f ; <<8 x i8>> [#uses=1]
329 %tmp2efgh = mul <8 x i8> %tmp2ef, undef ; <<8 x i8>> [#uses=2]
330 call void @llvm.arm.neon.vst3.p0i8.v8i8(i8* %A2, <8 x i8> undef, <8 x i8> undef, <8 x i8> %tmp2efgh, i32 1)
331 %tmp2 = sub <8 x i8> %tmp2efgh, %tmp2abcd ; <<8 x i8>> [#uses=1]
332 %tmp7 = mul <8 x i8> undef, %tmp2 ; <<8 x i8>> [#uses=1]
333 tail call void @llvm.arm.neon.vst3.p0i8.v8i8(i8* %B, <8 x i8> undef, <8 x i8> undef, <8 x i8> %tmp7, i32 1)
337 declare <4 x i32> @llvm.arm.neon.vld1.v4i32.p0i8(i8*, i32) nounwind readonly
339 declare <8 x i16> @llvm.arm.neon.vld1.v8i16.p0i8(i8*, i32) nounwind readonly
341 declare <4 x i16> @llvm.arm.neon.vshiftn.v4i16(<4 x i32>, <4 x i32>) nounwind readnone
343 declare void @llvm.arm.neon.vst1.p0i8.v4i32(i8*, <4 x i32>, i32) nounwind
345 declare void @llvm.arm.neon.vst1.p0i8.v8i16(i8*, <8 x i16>, i32) nounwind
347 declare void @llvm.arm.neon.vst3.p0i8.v8i8(i8*, <8 x i8>, <8 x i8>, <8 x i8>, i32)
350 declare %struct.__neon_int8x8x3_t @llvm.arm.neon.vld3.v8i8.p0i8(i8*, i32) nounwind readonly
352 declare %struct.__neon_int32x4x2_t @llvm.arm.neon.vld2.v4i32.p0i8(i8*, i32) nounwind readonly
354 declare %struct.__neon_int8x8x2_t @llvm.arm.neon.vld2lane.v8i8.p0i8(i8*, <8 x i8>, <8 x i8>, i32, i32) nounwind readonly
356 declare %struct.__neon_int16x8x2_t @llvm.arm.neon.vld2lane.v8i16.p0i8(i8*, <8 x i16>, <8 x i16>, i32, i32) nounwind readonly
358 declare void @llvm.arm.neon.vst2.p0i8.v4i32(i8*, <4 x i32>, <4 x i32>, i32) nounwind
360 declare <4 x float> @llvm.arm.neon.vrsqrte.v4f32(<4 x float>) nounwind readnone
362 declare void @llvm.trap() nounwind