1//===-- VEInstrPatternsVec.td - VEC_-type SDNodes and isel for VE Target --===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8//
9// This file describes the VEC_* prefixed intermediate SDNodes and their
10// isel patterns.
11//
12//===----------------------------------------------------------------------===//
13
14//===----------------------------------------------------------------------===//
15// Instruction format superclass
16//===----------------------------------------------------------------------===//
17
18// Sub-register replication for packed broadcast.
19def: Pat<(i64 (repl_f32 f32:$val)),
20            (ORrr
21              (SRLri (f2l $val), 32),
22              (zero_i32 (f2l $val)))>;
23def: Pat<(i64 (repl_i32 i32:$val)),
24            (ORrr
25              (zero_f32 (i2l $val)),
26              (SLLri (i2l $val), 32))>;
27
28///// Mask Load & Store /////
29
30// Store for v256i1, v512i1 are implemented in 2 ways.  These STVM/STVM512
31// pseudo instruction is used for frameindex related load/store instructions.
32// Custom Lowering is used for other load/store instructions.
33
34def : Pat<(v256i1 (load ADDRrii:$addr)),
35          (LDVMrii ADDRrii:$addr)>;
36def : Pat<(v512i1 (load ADDRrii:$addr)),
37          (LDVM512rii ADDRrii:$addr)>;
38def : Pat<(store v256i1:$vx, ADDRrii:$addr),
39          (STVMrii ADDRrii:$addr, $vx)>;
40def : Pat<(store v512i1:$vx, ADDRrii:$addr),
41          (STVM512rii ADDRrii:$addr, $vx)>;
42
43multiclass vbrd_elem32<ValueType v32, ValueType s32, SDPatternOperator ImmOp,
44                       SDNodeXForm ImmCast, OutPatFrag SuperRegCast> {
45  // VBRDil
46  def : Pat<(v32 (vec_broadcast (s32 ImmOp:$sy), i32:$vl)),
47            (VBRDil (ImmCast $sy), i32:$vl)>;
48
49  // VBRDrl
50  def : Pat<(v32 (vec_broadcast s32:$sy, i32:$vl)),
51            (VBRDrl (SuperRegCast $sy), i32:$vl)>;
52}
53
54multiclass vbrd_elem64<ValueType v64, ValueType s64,
55                       SDPatternOperator ImmOp, SDNodeXForm ImmCast> {
56  // VBRDil
57  def : Pat<(v64 (vec_broadcast (s64 ImmOp:$sy), i32:$vl)),
58            (VBRDil (ImmCast $sy), i32:$vl)>;
59
60  // VBRDrl
61  def : Pat<(v64 (vec_broadcast s64:$sy, i32:$vl)),
62            (VBRDrl s64:$sy, i32:$vl)>;
63}
64
65multiclass extract_insert_elem32<ValueType v32, ValueType s32,
66                                 OutPatFrag SubRegCast,
67                                 OutPatFrag SuperRegCast> {
68  // LVSvi
69  def: Pat<(s32 (extractelt v32:$vec, uimm7:$idx)),
70           (SubRegCast (LVSvi v32:$vec, (ULO7 $idx)))>;
71  // LVSvr
72  def: Pat<(s32 (extractelt v32:$vec, i64:$idx)),
73           (SubRegCast (LVSvr v32:$vec, $idx))>;
74
75  // LSVir
76  def: Pat<(v32 (insertelt v32:$vec, s32:$val, uimm7:$idx)),
77           (LSVir_v (ULO7 $idx), (SuperRegCast $val), $vec)>;
78  // LSVrr
79  def: Pat<(v32 (insertelt v32:$vec, s32:$val, i64:$idx)),
80           (LSVrr_v $idx, (SuperRegCast $val), $vec)>;
81}
82
83multiclass extract_insert_elem64<ValueType v64, ValueType s64> {
84  // LVSvi
85  def: Pat<(s64 (extractelt v64:$vec, uimm7:$idx)),
86           (LVSvi v64:$vec, (ULO7 $idx))>;
87  // LVSvr
88  def: Pat<(s64 (extractelt v64:$vec, i64:$idx)),
89           (LVSvr v64:$vec, $idx)>;
90
91  // LSVir
92  def: Pat<(v64 (insertelt v64:$vec, s64:$val, uimm7:$idx)),
93           (LSVir_v (ULO7 $idx), $val, $vec)>;
94  // LSVrr
95  def: Pat<(v64 (insertelt v64:$vec, s64:$val, i64:$idx)),
96           (LSVrr_v $idx, $val, $vec)>;
97}
98
99multiclass patterns_elem32<ValueType v32, ValueType s32,
100                           SDPatternOperator ImmOp, SDNodeXForm ImmCast,
101                           OutPatFrag SubRegCast, OutPatFrag SuperRegCast> {
102  defm : vbrd_elem32<v32, s32, ImmOp, ImmCast, SuperRegCast>;
103  defm : extract_insert_elem32<v32, s32, SubRegCast, SuperRegCast>;
104}
105
106multiclass patterns_elem64<ValueType v64, ValueType s64,
107                           SDPatternOperator ImmOp, SDNodeXForm ImmCast> {
108  defm : vbrd_elem64<v64, s64, ImmOp, ImmCast>;
109  defm : extract_insert_elem64<v64, s64>;
110}
111
112defm : patterns_elem32<v256i32, i32, simm7, LO7, l2i, i2l>;
113defm : patterns_elem32<v256f32, f32, simm7fp, LO7FP, l2f, f2l>;
114
115defm : patterns_elem64<v256i64, i64, simm7, LO7>;
116defm : patterns_elem64<v256f64, f64, simm7fp, LO7FP>;
117
118defm : vbrd_elem64<v512i32, i64, simm7, LO7>;
119defm : vbrd_elem64<v512f32, i64, simm7, LO7>;
120defm : vbrd_elem64<v512i32, f64, simm7fp, LO7FP>;
121defm : vbrd_elem64<v512f32, f64, simm7fp, LO7FP>;
122
123class Mask_Binary<ValueType MaskVT, SDPatternOperator MaskOp, string InstName> :
124  Pat<(MaskVT (MaskOp MaskVT:$ma, MaskVT:$mb)), (!cast<Instruction>(InstName#"mm") $ma, $mb)>;
125
126def: Mask_Binary<v256i1, and, "ANDM">;
127def: Mask_Binary<v256i1, or,  "ORM">;
128def: Mask_Binary<v256i1, xor, "XORM">;
129
130///// Packing support /////
131
132// v256i1 <> v512i1
133def : Pat<(v256i1 (vec_unpack_lo v512i1:$vm, (i32 srcvalue))),
134          (EXTRACT_SUBREG $vm, sub_vm_odd)>;
135def : Pat<(v256i1 (vec_unpack_hi v512i1:$vm, (i32 srcvalue))),
136          (EXTRACT_SUBREG $vm, sub_vm_even)>;
137def : Pat<(v512i1 (vec_pack v256i1:$vlo, v256i1:$vhi, (i32 srcvalue))),
138          (INSERT_SUBREG (INSERT_SUBREG
139                         (v512i1 (IMPLICIT_DEF)),
140                         $vlo, sub_vm_odd),
141                         $vhi, sub_vm_even)>;
142
143// v256.32 <> v512.32
144multiclass Packing<ValueType PackVT> {
145  // no-op unpacks
146  def : Pat<(v256i32 (vec_unpack_lo PackVT:$vp, (i32 srcvalue))),
147            (COPY_TO_REGCLASS $vp, V64)>;
148  def : Pat<(v256f32 (vec_unpack_hi PackVT:$vp, (i32 srcvalue))),
149            (COPY_TO_REGCLASS $vp, V64)>;
150
151  // shuffle unpacks
152  def : Pat<(v256f32 (vec_unpack_lo PackVT:$vp, i32:$avl)),
153            (VSHFvvil $vp, $vp, 4, $avl)>; // always pick lo
154  def : Pat<(v256i32 (vec_unpack_hi PackVT:$vp, i32:$avl)),
155            (VSHFvvil $vp, $vp, 0, $avl)>; // always pick hi
156}
157
158defm : Packing<v512i32>;
159defm : Packing<v512f32>;
160
161def : Pat<(v512i32 (vec_pack v256i32:$vlo, v256i32:$vhi, i32:$avl)),
162          (VSHFvvil $vlo, $vhi, 13, $avl)>;
163def : Pat<(v512f32 (vec_pack v256f32:$vlo, v256f32:$vhi, i32:$avl)),
164          (VSHFvvil $vlo, $vhi, 8, $avl)>;
165