1 ; Check that if option prefer-no-gather/scatter can disable gather/scatter instructions.
2 ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx2,+fast-gather %s -o - | FileCheck %s --check-prefixes=GATHER
3 ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx2,+fast-gather,+prefer-no-gather %s -o - | FileCheck %s --check-prefixes=NO-GATHER
4 ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512vl,+avx512dq < %s | FileCheck %s --check-prefix=SCATTER
5 ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512vl,+avx512dq,+prefer-no-gather < %s | FileCheck %s --check-prefix=SCATTER-NO-GATHER
6 ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512vl,+avx512dq,+prefer-no-scatter < %s | FileCheck %s --check-prefix=GATHER-NO-SCATTER
7 ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512vl,+avx512dq,+prefer-no-gather,+prefer-no-scatter < %s | FileCheck %s --check-prefix=NO-SCATTER-GATHER
9 @A = global [1024 x i8] zeroinitializer, align 128
10 @B = global [1024 x i64] zeroinitializer, align 128
11 @C = global [1024 x i64] zeroinitializer, align 128
13 ; This tests the function that if prefer-no-gather can disable lowerMGather
14 define void @test() #0 {
18 ; NO-GATHER-LABEL: test:
19 ; NO-GATHER-NOT: vpgatherdq
21 ; GATHER-NO-SCATTER-LABEL: test:
22 ; GATHER-NO-SCATTER: vpgatherdq
24 ; NO-SCATTER-GATHER-LABEL: test:
25 ; NO-SCATTER-GATHER-NOT: vpgatherdq
27 br i1 false, label %vec.epilog.scalar.ph, label %vector.main.loop.iter.check
29 vector.main.loop.iter.check: ; preds = %iter.check
30 br i1 false, label %vec.epilog.ph, label %vector.ph
32 vector.ph: ; preds = %vector.main.loop.iter.check
35 vector.body: ; preds = %vector.body, %vector.ph
36 %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
37 %0 = add i64 %index, 0
38 %1 = getelementptr inbounds [1024 x i8], ptr @A, i64 0, i64 %0
39 %2 = getelementptr inbounds i8, ptr %1, i32 0
40 %wide.load = load <32 x i8>, ptr %2, align 1
41 %3 = sext <32 x i8> %wide.load to <32 x i64>
42 %4 = getelementptr inbounds [1024 x i64], ptr @B, i64 0, <32 x i64> %3
43 %wide.masked.gather = call <32 x i64> @llvm.masked.gather.v32i64.v32p0(<32 x ptr> %4, i32 8, <32 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <32 x i64> poison)
44 %5 = getelementptr inbounds [1024 x i64], ptr @C, i64 0, i64 %0
45 %6 = getelementptr inbounds i64, ptr %5, i32 0
46 store <32 x i64> %wide.masked.gather, ptr %6, align 8
47 %index.next = add nuw i64 %index, 32
48 %7 = icmp eq i64 %index.next, 1024
49 br i1 %7, label %middle.block, label %vector.body, !llvm.loop !0
51 middle.block: ; preds = %vector.body
52 %cmp.n = icmp eq i64 1024, 1024
53 br i1 %cmp.n, label %for.cond.cleanup, label %vec.epilog.iter.check
55 vec.epilog.iter.check: ; preds = %middle.block
56 br i1 true, label %vec.epilog.scalar.ph, label %vec.epilog.ph
58 vec.epilog.ph: ; preds = %vector.main.loop.iter.check, %vec.epilog.iter.check
59 %vec.epilog.resume.val = phi i64 [ 1024, %vec.epilog.iter.check ], [ 0, %vector.main.loop.iter.check ]
60 br label %vec.epilog.vector.body
62 vec.epilog.vector.body: ; preds = %vec.epilog.vector.body, %vec.epilog.ph
63 %index2 = phi i64 [ %vec.epilog.resume.val, %vec.epilog.ph ], [ %index.next5, %vec.epilog.vector.body ]
64 %8 = add i64 %index2, 0
65 %9 = getelementptr inbounds [1024 x i8], ptr @A, i64 0, i64 %8
66 %10 = getelementptr inbounds i8, ptr %9, i32 0
67 %wide.load3 = load <16 x i8>, ptr %10, align 1
68 %11 = sext <16 x i8> %wide.load3 to <16 x i64>
69 %12 = getelementptr inbounds [1024 x i64], ptr @B, i64 0, <16 x i64> %11
70 %wide.masked.gather4 = call <16 x i64> @llvm.masked.gather.v16i64.v16p0(<16 x ptr> %12, i32 8, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i64> poison)
71 %13 = getelementptr inbounds [1024 x i64], ptr @C, i64 0, i64 %8
72 %14 = getelementptr inbounds i64, ptr %13, i32 0
73 store <16 x i64> %wide.masked.gather4, ptr %14, align 8
74 %index.next5 = add nuw i64 %index2, 16
75 %15 = icmp eq i64 %index.next5, 1024
76 br i1 %15, label %vec.epilog.middle.block, label %vec.epilog.vector.body, !llvm.loop !2
78 vec.epilog.middle.block: ; preds = %vec.epilog.vector.body
79 %cmp.n1 = icmp eq i64 1024, 1024
80 br i1 %cmp.n1, label %for.cond.cleanup, label %vec.epilog.scalar.ph
82 vec.epilog.scalar.ph: ; preds = %iter.check, %vec.epilog.iter.check, %vec.epilog.middle.block
83 %bc.resume.val = phi i64 [ 1024, %vec.epilog.middle.block ], [ 1024, %vec.epilog.iter.check ], [ 0, %iter.check ]
86 for.body: ; preds = %for.body, %vec.epilog.scalar.ph
87 %iv = phi i64 [ %bc.resume.val, %vec.epilog.scalar.ph ], [ %iv.next, %for.body ]
88 %inA = getelementptr inbounds [1024 x i8], ptr @A, i64 0, i64 %iv
89 %valA = load i8, ptr %inA, align 1
90 %valA.ext = sext i8 %valA to i64
91 %inB = getelementptr inbounds [1024 x i64], ptr @B, i64 0, i64 %valA.ext
92 %valB = load i64, ptr %inB, align 8
93 %out = getelementptr inbounds [1024 x i64], ptr @C, i64 0, i64 %iv
94 store i64 %valB, ptr %out, align 8
95 %iv.next = add nuw nsw i64 %iv, 1
96 %cmp = icmp ult i64 %iv.next, 1024
97 br i1 %cmp, label %for.body, label %for.cond.cleanup, !llvm.loop !4
99 for.cond.cleanup: ; preds = %vec.epilog.middle.block, %middle.block, %for.body
103 declare <32 x i64> @llvm.masked.gather.v32i64.v32p0(<32 x ptr>, i32 immarg, <32 x i1>, <32 x i64>) #1
105 declare <16 x i64> @llvm.masked.gather.v16i64.v16p0(<16 x ptr>, i32 immarg, <16 x i1>, <16 x i64>) #1
106 !0 = distinct !{!0, !1}
107 !1 = !{!"llvm.loop.isvectorized", i32 1}
108 !2 = distinct !{!2, !1, !3}
109 !3 = !{!"llvm.loop.unroll.runtime.disable"}
110 !4 = distinct !{!4, !3, !1}
112 ; This tests the function that if prefer-no-gather can disable ScalarizeMaskedGather
113 define <4 x float> @gather_v4f32_ptr_v4i32(<4 x ptr> %ptr, <4 x i32> %trigger, <4 x float> %passthru) {
114 ; GATHER-LABEL: gather_v4f32_ptr_v4i32:
117 ; NO-GATHER-LABEL: gather_v4f32_ptr_v4i32:
118 ; NO-GATHER-NOT: vgatherqps
120 ; GATHER-NO-SCATTER-LABEL: gather_v4f32_ptr_v4i32:
121 ; GATHER-NO-SCATTER: vgatherqps
123 ; NO-SCATTER-GATHER-LABEL: gather_v4f32_ptr_v4i32:
124 ; NO-SCATTER-GATHER-NOT: vgatherqps
125 %mask = icmp eq <4 x i32> %trigger, zeroinitializer
126 %res = call <4 x float> @llvm.masked.gather.v4f32.v4p0(<4 x ptr> %ptr, i32 4, <4 x i1> %mask, <4 x float> %passthru)
130 declare <4 x float> @llvm.masked.gather.v4f32.v4p0(<4 x ptr>, i32, <4 x i1>, <4 x float>)
132 %struct.a = type { [4 x i32], [4 x i8], %struct.b, i32 }
133 %struct.b = type { i32, i32 }
134 @c = external dso_local global %struct.a, align 4
136 ; This tests the function that if prefer-no-gather can disable ScalarizeMaskedGather
137 define <8 x i32> @gather_v8i32_v8i32(<8 x i32> %trigger) {
138 ; GATHER-LABEL: gather_v8i32_v8i32:
141 ; NO-GATHER-LABEL: gather_v8i32_v8i32:
142 ; NO-GATHER-NOT: vpgatherdd
144 ; NO-SCATTER-GATHER-LABEL: gather_v8i32_v8i32:
145 ; NO-SCATTER-GATHER-NOT: vpgatherdd
146 %1 = icmp eq <8 x i32> %trigger, zeroinitializer
147 %2 = call <8 x i32> @llvm.masked.gather.v8i32.v8p0(<8 x ptr> getelementptr (%struct.a, <8 x ptr> <ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c>, <8 x i64> zeroinitializer, i32 0, <8 x i64> <i64 3, i64 3, i64 3, i64 3, i64 3, i64 3, i64 3, i64 3>), i32 4, <8 x i1> %1, <8 x i32> undef)
148 %3 = call <8 x i32> @llvm.masked.gather.v8i32.v8p0(<8 x ptr> getelementptr (%struct.a, <8 x ptr> <ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c>, <8 x i64> zeroinitializer, i32 3), i32 4, <8 x i1> %1, <8 x i32> undef)
149 %4 = add <8 x i32> %2, %3
150 %5 = call <8 x i32> @llvm.masked.gather.v8i32.v8p0(<8 x ptr> getelementptr (%struct.a, <8 x ptr> <ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c>, <8 x i64> zeroinitializer, i32 3), i32 4, <8 x i1> %1, <8 x i32> undef)
151 %6 = add <8 x i32> %4, %5
155 declare <8 x i32> @llvm.masked.gather.v8i32.v8p0(<8 x ptr>, i32, <8 x i1>, <8 x i32>)
158 define void @scatter_test1(ptr %base, <16 x i32> %ind, i16 %mask, <16 x i32>%val) {
159 ; SCATTER-LABEL: scatter_test1:
160 ; SCATTER: vpscatterdd
162 ; SCATTER-NO-GATHER-LABEL: scatter_test1:
163 ; SCATTER-NO-GATHER: vpscatterdd
165 ; GATHER-NO-SCATTER-LABEL: scatter_test1:
166 ; GATHER-NO-SCATTER-NOT: vpscatterdd
168 ; NO-SCATTER-GATHER-LABEL: scatter_test1:
169 ; NO-SCATTER-GATHER-NOT: vpscatterdd
170 %broadcast.splatinsert = insertelement <16 x ptr> undef, ptr %base, i32 0
171 %broadcast.splat = shufflevector <16 x ptr> %broadcast.splatinsert, <16 x ptr> undef, <16 x i32> zeroinitializer
173 %gep.random = getelementptr i32, <16 x ptr> %broadcast.splat, <16 x i32> %ind
174 %imask = bitcast i16 %mask to <16 x i1>
175 call void @llvm.masked.scatter.v16i32.v16p0(<16 x i32>%val, <16 x ptr> %gep.random, i32 4, <16 x i1> %imask)
176 call void @llvm.masked.scatter.v16i32.v16p0(<16 x i32>%val, <16 x ptr> %gep.random, i32 4, <16 x i1> %imask)
180 declare void @llvm.masked.scatter.v8i32.v8p0(<8 x i32> , <8 x ptr> , i32 , <8 x i1> )
181 declare void @llvm.masked.scatter.v16i32.v16p0(<16 x i32> , <16 x ptr> , i32 , <16 x i1> )
183 define <8 x i32> @scatter_test2(<8 x i32>%a1, <8 x ptr> %ptr) {
184 ; SCATTER-LABEL: scatter_test2:
185 ; SCATTER: vpscatterqd
187 ; SCATTER-NO-GATHER-LABEL: scatter_test2:
188 ; SCATTER-NO-GATHER: vpscatterqd
190 ; GATHER-NO-SCATTER-LABEL: scatter_test2:
191 ; GATHER-NO-SCATTER-NOT: vpscatterqd
193 ; NO-SCATTER-GATHER-LABEL: scatter_test2:
194 ; NO-SCATTER-GATHER-NOT: vpscatterqd
195 %a = call <8 x i32> @llvm.masked.gather.v8i32.v8p0(<8 x ptr> %ptr, i32 4, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <8 x i32> undef)
197 call void @llvm.masked.scatter.v8i32.v8p0(<8 x i32> %a1, <8 x ptr> %ptr, i32 4, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)