aboutsummaryrefslogtreecommitdiff
path: root/lib/Target/AMDGPU/AMDGPUGISel.td
blob: cad4c2ef404c78b4d892f668b3f666fda9ba3176 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
//===-- AMDGPUGIsel.td - AMDGPU GlobalISel Patterns---------*- tablegen -*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
// This files contains patterns that should only be used by GlobalISel.  For
// example patterns for V_* instructions that have S_* equivalents.
// SelectionDAG does not support selecting V_* instructions.
//===----------------------------------------------------------------------===//

include "AMDGPU.td"

def p0 : PtrValueType<i64, 0>;
def p1 : PtrValueType<i64, 1>;
def p4 : PtrValueType<i64, 4>;

def sd_vsrc0 : ComplexPattern<i32, 1, "">;
def gi_vsrc0 :
    GIComplexOperandMatcher<s32, "selectVSRC0">,
    GIComplexPatternEquiv<sd_vsrc0>;

def sd_vcsrc : ComplexPattern<i32, 1, "">;
def gi_vcsrc :
    GIComplexOperandMatcher<s32, "selectVCSRC">,
    GIComplexPatternEquiv<sd_vcsrc>;

def gi_vop3mods0 :
    GIComplexOperandMatcher<s32, "selectVOP3Mods0">,
    GIComplexPatternEquiv<VOP3Mods0>;

def gi_vop3mods :
    GIComplexOperandMatcher<s32, "selectVOP3Mods">,
    GIComplexPatternEquiv<VOP3Mods>;

def gi_vop3omods :
    GIComplexOperandMatcher<s32, "selectVOP3OMods">,
    GIComplexPatternEquiv<VOP3OMods>;

def gi_smrd_imm :
    GIComplexOperandMatcher<s64, "selectSmrdImm">,
    GIComplexPatternEquiv<SMRDImm>;

def gi_smrd_imm32 :
    GIComplexOperandMatcher<s64, "selectSmrdImm32">,
    GIComplexPatternEquiv<SMRDImm32>;

def gi_smrd_sgpr :
    GIComplexOperandMatcher<s64, "selectSmrdSgpr">,
    GIComplexPatternEquiv<SMRDSgpr>;

def gi_flat_offset :
    GIComplexOperandMatcher<s64, "selectFlatOffset">,
    GIComplexPatternEquiv<FLATOffset>;
def gi_flat_offset_signed :
    GIComplexOperandMatcher<s64, "selectFlatOffsetSigned">,
    GIComplexPatternEquiv<FLATOffsetSigned>;

def gi_mubuf_scratch_offset :
    GIComplexOperandMatcher<s32, "selectMUBUFScratchOffset">,
    GIComplexPatternEquiv<MUBUFScratchOffset>;
def gi_mubuf_scratch_offen :
    GIComplexOperandMatcher<s32, "selectMUBUFScratchOffen">,
    GIComplexPatternEquiv<MUBUFScratchOffen>;


class GISelSop2Pat <
  SDPatternOperator node,
  Instruction inst,
  ValueType dst_vt,
  ValueType src0_vt = dst_vt, ValueType src1_vt = src0_vt>   : GCNPat <

  (dst_vt (node (src0_vt SReg_32:$src0), (src1_vt SReg_32:$src1))),
  (inst src0_vt:$src0, src1_vt:$src1)
>;

class GISelVop2Pat <
  SDPatternOperator node,
  Instruction inst,
  ValueType dst_vt,
  ValueType src0_vt = dst_vt, ValueType src1_vt = src0_vt>   : GCNPat <

  (dst_vt (node (src0_vt (sd_vsrc0 src0_vt:$src0)), (src1_vt VGPR_32:$src1))),
  (inst src0_vt:$src0, src1_vt:$src1)
>;

class GISelVop2CommutePat <
  SDPatternOperator node,
  Instruction inst,
  ValueType dst_vt,
  ValueType src0_vt = dst_vt, ValueType src1_vt = src0_vt>   : GCNPat <

  (dst_vt (node (src1_vt VGPR_32:$src1), (src0_vt (sd_vsrc0 src0_vt:$src0)))),
  (inst src0_vt:$src0, src1_vt:$src1)
>;

class GISelVop3Pat2 <
  SDPatternOperator node,
  Instruction inst,
  ValueType dst_vt,
  ValueType src0_vt = dst_vt, ValueType src1_vt = src0_vt>   : GCNPat <

  (dst_vt (node (src0_vt (sd_vcsrc src0_vt:$src0)), (src1_vt (sd_vcsrc src1_vt:$src1)))),
  (inst src0_vt:$src0, src1_vt:$src1)
>;

class GISelVop3Pat2CommutePat <
  SDPatternOperator node,
  Instruction inst,
  ValueType dst_vt,
  ValueType src0_vt = dst_vt, ValueType src1_vt = src0_vt>   : GCNPat <

  (dst_vt (node (src0_vt (sd_vcsrc src0_vt:$src0)), (src1_vt (sd_vcsrc src1_vt:$src1)))),
  (inst src0_vt:$src1, src1_vt:$src0)
>;

class GISelVop3Pat2ModsPat <
  SDPatternOperator node,
  Instruction inst,
  ValueType dst_vt,
  ValueType src0_vt = dst_vt, ValueType src1_vt = src0_vt> : GCNPat <

  (dst_vt (node (src0_vt (VOP3Mods0 src0_vt:$src0, i32:$src0_modifiers, i1:$clamp, i32:$omods)),
                (src1_vt (VOP3Mods src1_vt:$src1, i32:$src1_modifiers)))),
  (inst i32:$src0_modifiers, src0_vt:$src0,
        i32:$src1_modifiers, src1_vt:$src1, $clamp, $omods)
>;

multiclass GISelVop2IntrPat <
  SDPatternOperator node, Instruction inst,
  ValueType dst_vt, ValueType src_vt = dst_vt> {

  def : GISelVop2Pat <node, inst, dst_vt, src_vt>;

  // FIXME: Intrinsics aren't marked as commutable, so we need to add an explcit
  // pattern to handle commuting.  This is another reason why legalizing to a
  // generic machine instruction may be better that matching the intrinsic
  // directly.
  def : GISelVop2CommutePat <node, inst, dst_vt, src_vt>;
}

def : GISelSop2Pat <or, S_OR_B32, i32>;
def : GISelVop2Pat <or, V_OR_B32_e32, i32>;

// FIXME: We can't re-use SelectionDAG patterns here because they match
// against a custom SDNode and we would need to create a generic machine
// instruction that is equivalent to the custom SDNode.  This would also require
// us to custom legalize the intrinsic to the new generic machine instruction,
// but I can't get custom legalizing of intrinsic to work and I'm not sure if
// this is even supported yet.
def : GISelVop3Pat2ModsPat <
  int_amdgcn_cvt_pkrtz, V_CVT_PKRTZ_F16_F32_e64, v2f16, f32>;

defm : GISelVop2IntrPat <int_maxnum, V_MAX_F32_e32, f32>;
def : GISelVop3Pat2ModsPat <int_maxnum, V_MAX_F64, f64>;
defm : GISelVop2IntrPat <int_minnum, V_MIN_F32_e32, f32>;
def : GISelVop3Pat2ModsPat <int_minnum, V_MIN_F64, f64>;

// Since GlobalISel is more flexible then SelectionDAG, I think we can get
// away with adding patterns for integer types and not legalizing all
// loads and stores to vector types.  This should help simplify the load/store
// legalization.
foreach Ty = [i64, p0, p1, p4] in {
  defm : SMRD_Pattern <"S_LOAD_DWORDX2",  Ty>;
}