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(ptr %i_ptr, ptr %o_ptr, ptr nocapture %vT0ptr, ptr nocapture %vT1ptr) nounwind {
22 %0 = getelementptr inbounds %struct.int32x4_t, ptr %vT0ptr, i32 0, i32 0 ; <ptr> [#uses=1]
23 %1 = load <4 x i32>, ptr %0, align 16 ; <<4 x i32>> [#uses=1]
24 %2 = getelementptr inbounds %struct.int32x4_t, ptr %vT1ptr, i32 0, i32 0 ; <ptr> [#uses=1]
25 %3 = load <4 x i32>, ptr %2, align 16 ; <<4 x i32>> [#uses=1]
26 %4 = tail call <8 x i16> @llvm.arm.neon.vld1.v8i16.p0(ptr %i_ptr, i32 1) ; <<8 x i16>> [#uses=1]
27 %5 = bitcast <8 x i16> %4 to <2 x double> ; <<2 x double>> [#uses=2]
28 %6 = extractelement <2 x double> %5, i32 0 ; <double> [#uses=1]
29 %7 = bitcast double %6 to <4 x i16> ; <<4 x i16>> [#uses=1]
30 %8 = sext <4 x i16> %7 to <4 x i32> ; <<4 x i32>> [#uses=1]
31 %9 = extractelement <2 x double> %5, i32 1 ; <double> [#uses=1]
32 %10 = bitcast double %9 to <4 x i16> ; <<4 x i16>> [#uses=1]
33 %11 = sext <4 x i16> %10 to <4 x i32> ; <<4 x i32>> [#uses=1]
34 %12 = mul <4 x i32> %1, %8 ; <<4 x i32>> [#uses=1]
35 %13 = mul <4 x i32> %3, %11 ; <<4 x i32>> [#uses=1]
36 %14 = lshr <4 x i32> %12, <i32 12, i32 12, i32 12, i32 12>
37 %trunc_15 = trunc <4 x i32> %14 to <4 x i16>
38 %15 = lshr <4 x i32> %13, <i32 12, i32 12, i32 12, i32 12>
39 %trunc_16 = trunc <4 x i32> %15 to <4 x i16>
40 %16 = 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]
41 tail call void @llvm.arm.neon.vst1.p0.v8i16(ptr %o_ptr, <8 x i16> %16, i32 1)
45 define void @t2(ptr %i_ptr, ptr %o_ptr, ptr nocapture %vT0ptr, ptr nocapture %vT1ptr) nounwind {
56 %0 = getelementptr inbounds %struct.int16x8_t, ptr %vT0ptr, i32 0, i32 0 ; <ptr> [#uses=1]
57 %1 = load <8 x i16>, ptr %0, align 16 ; <<8 x i16>> [#uses=1]
58 %2 = getelementptr inbounds %struct.int16x8_t, ptr %vT1ptr, i32 0, i32 0 ; <ptr> [#uses=1]
59 %3 = load <8 x i16>, ptr %2, align 16 ; <<8 x i16>> [#uses=1]
60 %4 = tail call <8 x i16> @llvm.arm.neon.vld1.v8i16.p0(ptr %i_ptr, i32 1) ; <<8 x i16>> [#uses=1]
61 %5 = getelementptr inbounds i16, ptr %i_ptr, i32 8 ; <ptr> [#uses=1]
62 %6 = tail call <8 x i16> @llvm.arm.neon.vld1.v8i16.p0(ptr %5, i32 1) ; <<8 x i16>> [#uses=1]
63 %7 = mul <8 x i16> %1, %4 ; <<8 x i16>> [#uses=1]
64 %8 = mul <8 x i16> %3, %6 ; <<8 x i16>> [#uses=1]
65 tail call void @llvm.arm.neon.vst1.p0.v8i16(ptr %o_ptr, <8 x i16> %7, i32 1)
66 %9 = getelementptr inbounds i16, ptr %o_ptr, i32 8 ; <ptr> [#uses=1]
67 tail call void @llvm.arm.neon.vst1.p0.v8i16(ptr %9, <8 x i16> %8, i32 1)
71 define <8 x i8> @t3(ptr %A, ptr %B) nounwind {
78 %tmp1 = call %struct.__neon_int8x8x3_t @llvm.arm.neon.vld3.v8i8.p0(ptr %A, i32 1) ; <%struct.__neon_int8x8x3_t> [#uses=2]
79 %tmp2 = extractvalue %struct.__neon_int8x8x3_t %tmp1, 0 ; <<8 x i8>> [#uses=1]
80 %tmp3 = extractvalue %struct.__neon_int8x8x3_t %tmp1, 2 ; <<8 x i8>> [#uses=1]
81 %tmp4 = extractvalue %struct.__neon_int8x8x3_t %tmp1, 1 ; <<8 x i8>> [#uses=1]
82 %tmp5 = sub <8 x i8> %tmp3, %tmp4
83 %tmp6 = add <8 x i8> %tmp2, %tmp3 ; <<8 x i8>> [#uses=1]
84 %tmp7 = mul <8 x i8> %tmp4, %tmp2
85 tail call void @llvm.arm.neon.vst3.p0.v8i8(ptr %B, <8 x i8> %tmp5, <8 x i8> %tmp6, <8 x i8> %tmp7, i32 1)
89 define void @t4(ptr %in, ptr %out) nounwind {
97 %tmp2 = tail call %struct.__neon_int32x4x2_t @llvm.arm.neon.vld2.v4i32.p0(ptr %in, i32 1) ; <%struct.__neon_int32x4x2_t> [#uses=2]
98 %tmp3 = getelementptr inbounds i32, ptr %in, i32 8 ; <ptr> [#uses=1]
99 %tmp5 = tail call %struct.__neon_int32x4x2_t @llvm.arm.neon.vld2.v4i32.p0(ptr %tmp3, i32 1) ; <%struct.__neon_int32x4x2_t> [#uses=2]
100 br i1 undef, label %return1, label %return2
105 ; CHECK-NEXT: vadd.i32
106 ; CHECK-NEXT: vadd.i32
107 ; CHECK-NEXT: vst2.32
108 %tmp52 = extractvalue %struct.__neon_int32x4x2_t %tmp2, 0 ; <<4 x i32>> [#uses=1]
109 %tmp57 = extractvalue %struct.__neon_int32x4x2_t %tmp2, 1 ; <<4 x i32>> [#uses=1]
110 %tmp = extractvalue %struct.__neon_int32x4x2_t %tmp5, 0 ; <<4 x i32>> [#uses=1]
111 %tmp39 = extractvalue %struct.__neon_int32x4x2_t %tmp5, 1 ; <<4 x i32>> [#uses=1]
112 %tmp6 = add <4 x i32> %tmp52, %tmp ; <<4 x i32>> [#uses=1]
113 %tmp7 = add <4 x i32> %tmp57, %tmp39 ; <<4 x i32>> [#uses=1]
114 tail call void @llvm.arm.neon.vst2.p0.v4i32(ptr %out, <4 x i32> %tmp6, <4 x i32> %tmp7, i32 1)
121 ; CHECK: vst2.32 {d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}}
122 %tmp100 = extractvalue %struct.__neon_int32x4x2_t %tmp2, 0 ; <<4 x i32>> [#uses=1]
123 %tmp101 = extractvalue %struct.__neon_int32x4x2_t %tmp5, 1 ; <<4 x i32>> [#uses=1]
124 %tmp102 = add <4 x i32> %tmp100, %tmp101 ; <<4 x i32>> [#uses=1]
125 tail call void @llvm.arm.neon.vst2.p0.v4i32(ptr %out, <4 x i32> %tmp102, <4 x i32> %tmp101, i32 1)
126 call void @llvm.trap()
130 define <8 x i16> @t5(ptr %A, ptr %B) nounwind {
133 ; How can FileCheck match Q and D registers? We need a lisp interpreter.
134 ; CHECK: vorr {{q[0-9]+}}, {{q[0-9]+}}, {{q[0-9]+}}
136 ; CHECK: vld2.16 {d{{[0-9]+}}[1], d{{[0-9]+}}[1]}, [r0]
139 %tmp1 = load <8 x i16>, ptr %B ; <<8 x i16>> [#uses=2]
140 %tmp2 = call %struct.__neon_int16x8x2_t @llvm.arm.neon.vld2lane.v8i16.p0(ptr %A, <8 x i16> %tmp1, <8 x i16> %tmp1, i32 1, i32 1) ; <%struct.__neon_int16x8x2_t> [#uses=2]
141 %tmp3 = extractvalue %struct.__neon_int16x8x2_t %tmp2, 0 ; <<8 x i16>> [#uses=1]
142 %tmp4 = extractvalue %struct.__neon_int16x8x2_t %tmp2, 1 ; <<8 x i16>> [#uses=1]
143 %tmp5 = add <8 x i16> %tmp3, %tmp4 ; <<8 x i16>> [#uses=1]
147 define <8 x i8> @t6(ptr %A, ptr %B) nounwind {
150 ; CHECK: vorr d[[D0:[0-9]+]], d[[D1:[0-9]+]]
151 ; CHECK-NEXT: vld2.8 {d[[D1]][1], d[[D0]][1]}
152 %tmp1 = load <8 x i8>, ptr %B ; <<8 x i8>> [#uses=2]
153 %tmp2 = call %struct.__neon_int8x8x2_t @llvm.arm.neon.vld2lane.v8i8.p0(ptr %A, <8 x i8> %tmp1, <8 x i8> %tmp1, i32 1, i32 1) ; <%struct.__neon_int8x8x2_t> [#uses=2]
154 %tmp3 = extractvalue %struct.__neon_int8x8x2_t %tmp2, 0 ; <<8 x i8>> [#uses=1]
155 %tmp4 = extractvalue %struct.__neon_int8x8x2_t %tmp2, 1 ; <<8 x i8>> [#uses=1]
156 %tmp5 = add <8 x i8> %tmp3, %tmp4 ; <<8 x i8>> [#uses=1]
160 define void @t7(ptr %iptr, ptr %optr) nounwind {
165 ; CHECK: vld1.32 {d{{[0-9]+}}, d{{[0-9]+}}},
166 ; CHECK: vorr q[[Q0:[0-9]+]], q[[Q1:[0-9]+]], q[[Q1:[0-9]+]]
168 ; CHECK: vuzp.32 q[[Q1]], q[[Q0]]
170 %0 = tail call %struct.__neon_int32x4x2_t @llvm.arm.neon.vld2.v4i32.p0(ptr %iptr, i32 1) ; <%struct.__neon_int32x4x2_t> [#uses=2]
171 %tmp57 = extractvalue %struct.__neon_int32x4x2_t %0, 0 ; <<4 x i32>> [#uses=1]
172 %tmp60 = extractvalue %struct.__neon_int32x4x2_t %0, 1 ; <<4 x i32>> [#uses=1]
173 tail call void @llvm.arm.neon.vst2.p0.v4i32(ptr %optr, <4 x i32> %tmp57, <4 x i32> %tmp60, i32 1)
174 %1 = tail call <4 x i32> @llvm.arm.neon.vld1.v4i32.p0(ptr %iptr, i32 1) ; <<4 x i32>> [#uses=1]
175 %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 0, i32 2, i32 0, i32 2> ; <<4 x i32>> [#uses=1]
176 tail call void @llvm.arm.neon.vst1.p0.v4i32(ptr %optr, <4 x i32> %2, i32 1)
181 define arm_aapcs_vfpcc i32 @t8() nounwind {
183 ; CHECK: vrsqrte.f32 q8, q8
184 bb.nph55.bb.nph55.split_crit_edge:
187 bb3: ; preds = %bb3, %bb.nph55.bb.nph55.split_crit_edge
188 br i1 undef, label %bb5, label %bb3
193 bb.i25: ; preds = %bb.i25, %bb5
194 %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]
195 %1 = call <4 x float> @llvm.arm.neon.vrsqrte.v4f32(<4 x float> %0) nounwind ; <<4 x float>> [#uses=1]
196 %2 = fmul <4 x float> %1, %1 ; <<4 x float>> [#uses=1]
197 %3 = fmul <4 x float> %2, %2 ; <<4 x float>> [#uses=1]
198 %tmp26.i = bitcast <4 x float> %3 to <2 x double> ; <<2 x double>> [#uses=1]
199 %4 = extractelement <2 x double> %tmp26.i, i32 0 ; <double> [#uses=1]
200 %5 = bitcast double %4 to <2 x float> ; <<2 x float>> [#uses=1]
201 %6 = extractelement <2 x float> %5, i32 1 ; <float> [#uses=1]
202 store float %6, ptr undef, align 4
203 br i1 undef, label %bb6, label %bb.i25
205 bb6: ; preds = %bb.i25
206 br i1 undef, label %bb7, label %bb14
211 bb.i49: ; preds = %bb.i49, %bb7
212 br i1 undef, label %bb.i19, label %bb.i49
214 bb.i19: ; preds = %bb.i19, %bb.i49
215 br i1 undef, label %exit, label %bb.i19
217 exit: ; preds = %bb.i19
224 %0 = type { %1, %1, %1, %1 }
226 %2 = type { <4 x float> }
230 define arm_aapcs_vfpcc float @t9(ptr nocapture, ptr nocapture) nounwind {
232 ; CHECK: vmov.i32 d16, #0x0
233 ; CHECK-NEXT: vst1.64 {d16, d17}, [r0:128]
234 ; CHECK-NEXT: vorr d17, d16, d16
235 ; CHECK-NEXT: vst1.64 {d16, d17}, [r0:128]
236 %3 = bitcast double 0.000000e+00 to <2 x float> ; <<2 x float>> [#uses=2]
237 %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]
238 store <4 x float> %4, ptr undef, align 16
239 %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]
240 store <4 x float> %5, ptr undef, align 16
243 ; <label>:6 ; preds = %8
246 ; <label>:7 ; preds = %6
249 ; <label>:8 ; preds = %7, %2
252 ; <label>:9 ; preds = %8
255 ; <label>:10 ; preds = %6
256 ret float 9.990000e+02
260 define arm_aapcs_vfpcc i32 @t10(float %x) nounwind {
262 ; CHECK: vmov.i32 [[Q9:q[0-9]+]], #0x3f000000
263 ; CHECK: vmul.f32 [[Q8:q[0-9]+]], [[Q0:q[0-9]+]], [[Q0]]
264 ; CHECK-NEXT: vadd.f32 [[Q8]], [[Q8]], [[Q8]]
265 ; CHECK-NEXT: vadd.f32 [[Q1:q[0-9]+]], [[Q8]], [[Q8]]
266 ; CHECK-NEXT: vmul.f32 [[Q8]], [[Q9]], d1[0]
267 ; CHECK-NEXT: vmul.f32 [[Q8]], [[Q8]], [[Q8]]
268 ; CHECK-NEXT: vadd.f32 [[Q8]], [[Q8]], [[Q8]]
269 ; CHECK-NEXT: vmul.f32 [[Q8]], [[Q8]], [[Q8]]
270 ; CHECK-NEXT: vst1.32 {d17[1]}, [r0:32]
271 ; CHECK-NEXT: mov r0, #0
272 ; CHECK-NEXT: cmp r0, #0
273 ; CHECK-NEXT: bxne lr
274 ; CHECK-NEXT: LBB9_1:
277 %0 = shufflevector <4 x float> zeroinitializer, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>> [#uses=1]
278 %1 = insertelement <4 x float> %0, float %x, i32 1 ; <<4 x float>> [#uses=1]
279 %2 = insertelement <4 x float> %1, float %x, i32 2 ; <<4 x float>> [#uses=1]
280 %3 = insertelement <4 x float> %2, float %x, i32 3 ; <<4 x float>> [#uses=1]
281 %tmp54.i = bitcast <4 x float> %3 to <2 x double> ; <<2 x double>> [#uses=1]
282 %4 = extractelement <2 x double> %tmp54.i, i32 1 ; <double> [#uses=1]
283 %5 = bitcast double %4 to <2 x float> ; <<2 x float>> [#uses=1]
284 %6 = shufflevector <2 x float> %5, <2 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>> [#uses=1]
285 %7 = fmul <4 x float> %6, %6
286 %8 = fadd <4 x float> %7, %7
287 %9 = fadd <4 x float> %8, %8
288 %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]
289 %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]
290 %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]
291 %13 = shufflevector <4 x float> %12, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>> [#uses=1]
292 %14 = fmul <4 x float> %13, %13
293 %15 = fadd <4 x float> %14, %14
294 %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]
295 %17 = fmul <4 x float> %16, %16
296 %18 = extractelement <4 x float> %17, i32 2 ; <float> [#uses=1]
297 store float %18, ptr undef, align 4
298 br i1 undef, label %exit, label %bb14
300 exit: ; preds = %bb.i19
307 ; This test crashes the coalescer because live variables were not updated properly.
308 define <8 x i8> @t11(ptr %A1, ptr %A2, ptr %A3, ptr %A4, ptr %A5, ptr %A6, ptr %A7, ptr %A8, ptr %B) nounwind {
309 %tmp1d = call %struct.__neon_int8x8x3_t @llvm.arm.neon.vld3.v8i8.p0(ptr %A4, i32 1) ; <%struct.__neon_int8x8x3_t> [#uses=1]
310 %tmp2d = extractvalue %struct.__neon_int8x8x3_t %tmp1d, 0 ; <<8 x i8>> [#uses=1]
311 %tmp1f = call %struct.__neon_int8x8x3_t @llvm.arm.neon.vld3.v8i8.p0(ptr %A6, i32 1) ; <%struct.__neon_int8x8x3_t> [#uses=1]
312 %tmp2f = extractvalue %struct.__neon_int8x8x3_t %tmp1f, 0 ; <<8 x i8>> [#uses=1]
313 %tmp2bd = add <8 x i8> zeroinitializer, %tmp2d ; <<8 x i8>> [#uses=1]
314 %tmp2abcd = mul <8 x i8> zeroinitializer, %tmp2bd ; <<8 x i8>> [#uses=1]
315 %tmp2ef = sub <8 x i8> zeroinitializer, %tmp2f ; <<8 x i8>> [#uses=1]
316 %tmp2efgh = mul <8 x i8> %tmp2ef, undef ; <<8 x i8>> [#uses=2]
317 call void @llvm.arm.neon.vst3.p0.v8i8(ptr %A2, <8 x i8> undef, <8 x i8> undef, <8 x i8> %tmp2efgh, i32 1)
318 %tmp2 = sub <8 x i8> %tmp2efgh, %tmp2abcd ; <<8 x i8>> [#uses=1]
319 %tmp7 = mul <8 x i8> undef, %tmp2 ; <<8 x i8>> [#uses=1]
320 tail call void @llvm.arm.neon.vst3.p0.v8i8(ptr %B, <8 x i8> undef, <8 x i8> undef, <8 x i8> %tmp7, i32 1)
324 declare <4 x i32> @llvm.arm.neon.vld1.v4i32.p0(ptr, i32) nounwind readonly
326 declare <8 x i16> @llvm.arm.neon.vld1.v8i16.p0(ptr, i32) nounwind readonly
328 declare <4 x i16> @llvm.arm.neon.vshiftn.v4i16(<4 x i32>, <4 x i32>) nounwind readnone
330 declare void @llvm.arm.neon.vst1.p0.v4i32(ptr, <4 x i32>, i32) nounwind
332 declare void @llvm.arm.neon.vst1.p0.v8i16(ptr, <8 x i16>, i32) nounwind
334 declare void @llvm.arm.neon.vst3.p0.v8i8(ptr, <8 x i8>, <8 x i8>, <8 x i8>, i32)
337 declare %struct.__neon_int8x8x3_t @llvm.arm.neon.vld3.v8i8.p0(ptr, i32) nounwind readonly
339 declare %struct.__neon_int32x4x2_t @llvm.arm.neon.vld2.v4i32.p0(ptr, i32) nounwind readonly
341 declare %struct.__neon_int8x8x2_t @llvm.arm.neon.vld2lane.v8i8.p0(ptr, <8 x i8>, <8 x i8>, i32, i32) nounwind readonly
343 declare %struct.__neon_int16x8x2_t @llvm.arm.neon.vld2lane.v8i16.p0(ptr, <8 x i16>, <8 x i16>, i32, i32) nounwind readonly
345 declare void @llvm.arm.neon.vst2.p0.v4i32(ptr, <4 x i32>, <4 x i32>, i32) nounwind
347 declare <4 x float> @llvm.arm.neon.vrsqrte.v4f32(<4 x float>) nounwind readnone
349 declare void @llvm.trap() nounwind