xref: /llvm-project/llvm/test/Transforms/LoopVectorize/scalable-trunc-min-bitwidth.ll (revision a861ed411a359b7cb2b58d642e4a7f2fdcf92057)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -passes=loop-vectorize,instsimplify -scalable-vectorization=on -force-target-supports-scalable-vectors -S | FileCheck %s
3
4define void @trunc_minimal_bitwidth(ptr %bptr, ptr noalias %hptr, i32 %val, i64 %N) {
5; CHECK-LABEL: @trunc_minimal_bitwidth(
6; CHECK-NEXT:  entry:
7; CHECK-NEXT:    [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
8; CHECK-NEXT:    [[TMP1:%.*]] = mul i64 [[TMP0]], 4
9; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], [[TMP1]]
10; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
11; CHECK:       vector.ph:
12; CHECK-NEXT:    [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
13; CHECK-NEXT:    [[TMP3:%.*]] = mul i64 [[TMP2]], 4
14; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP3]]
15; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
16; CHECK-NEXT:    [[TMP6:%.*]] = call i64 @llvm.vscale.i64()
17; CHECK-NEXT:    [[TMP7:%.*]] = mul i64 [[TMP6]], 4
18; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[VAL:%.*]], i64 0
19; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[BROADCAST_SPLATINSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
20; CHECK-NEXT:    [[TMP4:%.*]] = trunc <vscale x 4 x i32> [[BROADCAST_SPLAT]] to <vscale x 4 x i16>
21; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
22; CHECK:       vector.body:
23; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
24; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i16, ptr [[HPTR:%.*]], i64 [[INDEX]]
25; CHECK-NEXT:    store <vscale x 4 x i16> [[TMP4]], ptr [[TMP5]], align 2
26; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP7]]
27; CHECK-NEXT:    [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
28; CHECK-NEXT:    br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
29; CHECK:       middle.block:
30; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
31; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
32; CHECK:       scalar.ph:
33; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
34; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
35; CHECK:       for.body:
36; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
37; CHECK-NEXT:    [[CONV21:%.*]] = trunc i32 [[VAL]] to i16
38; CHECK-NEXT:    [[ARRAYIDX23:%.*]] = getelementptr inbounds i16, ptr [[HPTR]], i64 [[INDVARS_IV]]
39; CHECK-NEXT:    store i16 [[CONV21]], ptr [[ARRAYIDX23]], align 2
40; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
41; CHECK-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[N]]
42; CHECK-NEXT:    br i1 [[EXITCOND_NOT]], label [[FOR_EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
43; CHECK:       for.exit:
44; CHECK-NEXT:    ret void
45;
46entry:
47  br label %for.body
48
49for.body:
50  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
51  %0 = load i8, ptr %bptr, align 1
52  %conv = zext i8 %0 to i32
53  %conv21 = trunc i32 %val to i16
54  %arrayidx23 = getelementptr inbounds i16, ptr %hptr, i64 %indvars.iv
55  store i16 %conv21, ptr %arrayidx23, align 2
56  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
57  %exitcond.not = icmp eq i64 %indvars.iv.next, %N
58  br i1 %exitcond.not, label %for.exit, label %for.body, !llvm.loop !0
59
60for.exit:
61  ret void
62}
63
64define void @trunc_minimal_bitwidths_shufflevector (ptr %p, i32 %arg1, i64 %len) {
65; CHECK-LABEL: @trunc_minimal_bitwidths_shufflevector(
66; CHECK-NEXT:  entry:
67; CHECK-NEXT:    [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
68; CHECK-NEXT:    [[TMP1:%.*]] = mul i64 [[TMP0]], 4
69; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[LEN:%.*]], [[TMP1]]
70; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
71; CHECK:       vector.ph:
72; CHECK-NEXT:    [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
73; CHECK-NEXT:    [[TMP3:%.*]] = mul i64 [[TMP2]], 4
74; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[LEN]], [[TMP3]]
75; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[LEN]], [[N_MOD_VF]]
76; CHECK-NEXT:    [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
77; CHECK-NEXT:    [[TMP9:%.*]] = mul i64 [[TMP8]], 4
78; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[ARG1:%.*]], i64 0
79; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[BROADCAST_SPLATINSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
80; CHECK-NEXT:    [[TMP4:%.*]] = trunc <vscale x 4 x i32> [[BROADCAST_SPLAT]] to <vscale x 4 x i8>
81; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
82; CHECK:       vector.body:
83; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
84; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i8, ptr [[P:%.*]], i64 [[INDEX]]
85; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <vscale x 4 x i8>, ptr [[TMP5]], align 1
86; CHECK-NEXT:    [[TMP6:%.*]] = xor <vscale x 4 x i8> [[WIDE_LOAD]], [[TMP4]]
87; CHECK-NEXT:    [[TMP7:%.*]] = mul <vscale x 4 x i8> [[TMP6]], [[WIDE_LOAD]]
88; CHECK-NEXT:    store <vscale x 4 x i8> [[TMP7]], ptr [[TMP5]], align 1
89; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP9]]
90; CHECK-NEXT:    [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
91; CHECK-NEXT:    br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
92; CHECK:       middle.block:
93; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[LEN]], [[N_VEC]]
94; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
95; CHECK:       scalar.ph:
96; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
97; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
98; CHECK:       for.body:
99; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
100; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i8, ptr [[P]], i64 [[INDVARS_IV]]
101; CHECK-NEXT:    [[TMP11:%.*]] = load i8, ptr [[ARRAYIDX]], align 1
102; CHECK-NEXT:    [[CONV:%.*]] = zext i8 [[TMP11]] to i32
103; CHECK-NEXT:    [[CONV17:%.*]] = xor i32 [[CONV]], [[ARG1]]
104; CHECK-NEXT:    [[MUL18:%.*]] = mul nuw nsw i32 [[CONV17]], [[CONV]]
105; CHECK-NEXT:    [[CONV19:%.*]] = trunc i32 [[MUL18]] to i8
106; CHECK-NEXT:    store i8 [[CONV19]], ptr [[ARRAYIDX]], align 1
107; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
108; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[LEN]]
109; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
110; CHECK:       for.exit:
111; CHECK-NEXT:    ret void
112;
113entry:
114  br label %for.body
115
116for.body:                                         ; preds = %entry
117  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
118  %arrayidx = getelementptr inbounds i8, ptr %p, i64 %indvars.iv
119  %0 = load i8, ptr %arrayidx
120  %conv = zext i8 %0 to i32
121  %conv17 = xor i32 %conv, %arg1
122  %mul18 = mul nuw nsw i32 %conv17, %conv
123  %conv19 = trunc i32 %mul18 to i8
124  store i8 %conv19, ptr %arrayidx
125  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
126  %exitcond = icmp eq i64 %indvars.iv.next, %len
127  br i1 %exitcond, label %for.exit, label %for.body, !llvm.loop !0
128
129for.exit:                                 ; preds = %for.body
130  ret void
131}
132!0 = !{!0, !1, !2}
133!1 = !{!"llvm.loop.vectorize.width", i32 4}
134!2 = !{!"llvm.loop.vectorize.scalable.enable", i1 true}
135