1// Copyright 2015 The Go Authors. All rights reserved.
2// Use of this source code is governed by a BSD-style
3// license that can be found in the LICENSE file.
4
5// Simplifications that apply to all backend architectures. As an example, this
6// Go source code
7//
8// y := 0 * x
9//
10// can be translated into y := 0 without losing any information, which saves a
11// pointless multiplication instruction. Other .rules files in this directory
12// (for example AMD64.rules) contain rules specific to the architecture in the
13// filename. The rules here apply to every architecture.
14//
15// The code for parsing this file lives in rulegen.go; this file generates
16// ssa/rewritegeneric.go.
17
18// values are specified using the following format:
19// (op <type> [auxint] {aux} arg0 arg1 ...)
20// the type, aux, and auxint fields are optional
21// on the matching side
22// - the type, aux, and auxint fields must match if they are specified.
23// - the first occurrence of a variable defines that variable. Subsequent
24// uses must match (be == to) the first use.
25// - v is defined to be the value matched.
26// - an additional conditional can be provided after the match pattern with "&&".
27// on the generated side
28// - the type of the top-level expression is the same as the one on the left-hand side.
29// - the type of any subexpressions must be specified explicitly (or
30// be specified in the op's type field).
31// - auxint will be 0 if not specified.
32// - aux will be nil if not specified.
33
34// blocks are specified using the following format:
35// (kind controlvalue succ0 succ1 ...)
36// controlvalue must be "nil" or a value expression
37// succ* fields must be variables
38// For now, the generated successors must be a permutation of the matched successors.
39
40// constant folding
41(Trunc16to8 (Const16 [c])) => (Const8 [int8(c)])
42(Trunc32to8 (Const32 [c])) => (Const8 [int8(c)])
43(Trunc32to16 (Const32 [c])) => (Const16 [int16(c)])
44(Trunc64to8 (Const64 [c])) => (Const8 [int8(c)])
45(Trunc64to16 (Const64 [c])) => (Const16 [int16(c)])
46(Trunc64to32 (Const64 [c])) => (Const32 [int32(c)])
47(Cvt64Fto32F (Const64F [c])) => (Const32F [float32(c)])
48(Cvt32Fto64F (Const32F [c])) => (Const64F [float64(c)])
49(Cvt32to32F (Const32 [c])) => (Const32F [float32(c)])
50(Cvt32to64F (Const32 [c])) => (Const64F [float64(c)])
51(Cvt64to32F (Const64 [c])) => (Const32F [float32(c)])
52(Cvt64to64F (Const64 [c])) => (Const64F [float64(c)])
53(Cvt32Fto32 (Const32F [c])) => (Const32 [int32(c)])
54(Cvt32Fto64 (Const32F [c])) => (Const64 [int64(c)])
55(Cvt64Fto32 (Const64F [c])) => (Const32 [int32(c)])
56(Cvt64Fto64 (Const64F [c])) => (Const64 [int64(c)])
57(Round32F x:(Const32F)) => x
58(Round64F x:(Const64F)) => x
59(CvtBoolToUint8 (ConstBool [false])) => (Const8 [0])
60(CvtBoolToUint8 (ConstBool [true])) => (Const8 [1])
61(BitLen64 (Const64 [c])) && config.PtrSize == 8 => (Const64 [int64(bits.Len64(uint64(c)))])
62(BitLen32 (Const32 [c])) && config.PtrSize == 8 => (Const64 [int64(bits.Len32(uint32(c)))])
63(BitLen16 (Const16 [c])) && config.PtrSize == 8 => (Const64 [int64(bits.Len16(uint16(c)))])
64(BitLen8 (Const8 [c])) && config.PtrSize == 8 => (Const64 [int64(bits.Len8(uint8(c)))])
65(BitLen64 (Const64 [c])) && config.PtrSize == 4 => (Const32 [int32(bits.Len64(uint64(c)))])
66(BitLen32 (Const32 [c])) && config.PtrSize == 4 => (Const32 [int32(bits.Len32(uint32(c)))])
67(BitLen16 (Const16 [c])) && config.PtrSize == 4 => (Const32 [int32(bits.Len16(uint16(c)))])
68(BitLen8 (Const8 [c])) && config.PtrSize == 4 => (Const32 [int32(bits.Len8(uint8(c)))])
69(PopCount64 (Const64 [c])) && config.PtrSize == 8 => (Const64 [int64(bits.OnesCount64(uint64(c)))])
70(PopCount32 (Const32 [c])) && config.PtrSize == 8 => (Const64 [int64(bits.OnesCount32(uint32(c)))])
71(PopCount16 (Const16 [c])) && config.PtrSize == 8 => (Const64 [int64(bits.OnesCount16(uint16(c)))])
72(PopCount8 (Const8 [c])) && config.PtrSize == 8 => (Const64 [int64(bits.OnesCount8(uint8(c)))])
73(PopCount64 (Const64 [c])) && config.PtrSize == 4 => (Const32 [int32(bits.OnesCount64(uint64(c)))])
74(PopCount32 (Const32 [c])) && config.PtrSize == 4 => (Const32 [int32(bits.OnesCount32(uint32(c)))])
75(PopCount16 (Const16 [c])) && config.PtrSize == 4 => (Const32 [int32(bits.OnesCount16(uint16(c)))])
76(PopCount8 (Const8 [c])) && config.PtrSize == 4 => (Const32 [int32(bits.OnesCount8(uint8(c)))])
77(Add64carry (Const64 <t> [x]) (Const64 [y]) (Const64 [c])) && c >= 0 && c <= 1 => (MakeTuple (Const64 <t> [bitsAdd64(x, y, c).sum]) (Const64 <t> [bitsAdd64(x, y, c).carry]))
78
79(Trunc16to8 (ZeroExt8to16 x)) => x
80(Trunc32to8 (ZeroExt8to32 x)) => x
81(Trunc32to16 (ZeroExt8to32 x)) => (ZeroExt8to16 x)
82(Trunc32to16 (ZeroExt16to32 x)) => x
83(Trunc64to8 (ZeroExt8to64 x)) => x
84(Trunc64to16 (ZeroExt8to64 x)) => (ZeroExt8to16 x)
85(Trunc64to16 (ZeroExt16to64 x)) => x
86(Trunc64to32 (ZeroExt8to64 x)) => (ZeroExt8to32 x)
87(Trunc64to32 (ZeroExt16to64 x)) => (ZeroExt16to32 x)
88(Trunc64to32 (ZeroExt32to64 x)) => x
89(Trunc16to8 (SignExt8to16 x)) => x
90(Trunc32to8 (SignExt8to32 x)) => x
91(Trunc32to16 (SignExt8to32 x)) => (SignExt8to16 x)
92(Trunc32to16 (SignExt16to32 x)) => x
93(Trunc64to8 (SignExt8to64 x)) => x
94(Trunc64to16 (SignExt8to64 x)) => (SignExt8to16 x)
95(Trunc64to16 (SignExt16to64 x)) => x
96(Trunc64to32 (SignExt8to64 x)) => (SignExt8to32 x)
97(Trunc64to32 (SignExt16to64 x)) => (SignExt16to32 x)
98(Trunc64to32 (SignExt32to64 x)) => x
99
100(ZeroExt8to16 (Const8 [c])) => (Const16 [int16( uint8(c))])
101(ZeroExt8to32 (Const8 [c])) => (Const32 [int32( uint8(c))])
102(ZeroExt8to64 (Const8 [c])) => (Const64 [int64( uint8(c))])
103(ZeroExt16to32 (Const16 [c])) => (Const32 [int32(uint16(c))])
104(ZeroExt16to64 (Const16 [c])) => (Const64 [int64(uint16(c))])
105(ZeroExt32to64 (Const32 [c])) => (Const64 [int64(uint32(c))])
106(SignExt8to16 (Const8 [c])) => (Const16 [int16(c)])
107(SignExt8to32 (Const8 [c])) => (Const32 [int32(c)])
108(SignExt8to64 (Const8 [c])) => (Const64 [int64(c)])
109(SignExt16to32 (Const16 [c])) => (Const32 [int32(c)])
110(SignExt16to64 (Const16 [c])) => (Const64 [int64(c)])
111(SignExt32to64 (Const32 [c])) => (Const64 [int64(c)])
112
113(Neg8 (Const8 [c])) => (Const8 [-c])
114(Neg16 (Const16 [c])) => (Const16 [-c])
115(Neg32 (Const32 [c])) => (Const32 [-c])
116(Neg64 (Const64 [c])) => (Const64 [-c])
117(Neg32F (Const32F [c])) && c != 0 => (Const32F [-c])
118(Neg64F (Const64F [c])) && c != 0 => (Const64F [-c])
119
120(Add8 (Const8 [c]) (Const8 [d])) => (Const8 [c+d])
121(Add16 (Const16 [c]) (Const16 [d])) => (Const16 [c+d])
122(Add32 (Const32 [c]) (Const32 [d])) => (Const32 [c+d])
123(Add64 (Const64 [c]) (Const64 [d])) => (Const64 [c+d])
124(Add32F (Const32F [c]) (Const32F [d])) && c+d == c+d => (Const32F [c+d])
125(Add64F (Const64F [c]) (Const64F [d])) && c+d == c+d => (Const64F [c+d])
126(AddPtr <t> x (Const64 [c])) => (OffPtr <t> x [c])
127(AddPtr <t> x (Const32 [c])) => (OffPtr <t> x [int64(c)])
128
129(Sub8 (Const8 [c]) (Const8 [d])) => (Const8 [c-d])
130(Sub16 (Const16 [c]) (Const16 [d])) => (Const16 [c-d])
131(Sub32 (Const32 [c]) (Const32 [d])) => (Const32 [c-d])
132(Sub64 (Const64 [c]) (Const64 [d])) => (Const64 [c-d])
133(Sub32F (Const32F [c]) (Const32F [d])) && c-d == c-d => (Const32F [c-d])
134(Sub64F (Const64F [c]) (Const64F [d])) && c-d == c-d => (Const64F [c-d])
135
136(Mul8 (Const8 [c]) (Const8 [d])) => (Const8 [c*d])
137(Mul16 (Const16 [c]) (Const16 [d])) => (Const16 [c*d])
138(Mul32 (Const32 [c]) (Const32 [d])) => (Const32 [c*d])
139(Mul64 (Const64 [c]) (Const64 [d])) => (Const64 [c*d])
140(Mul32F (Const32F [c]) (Const32F [d])) && c*d == c*d => (Const32F [c*d])
141(Mul64F (Const64F [c]) (Const64F [d])) && c*d == c*d => (Const64F [c*d])
142(Mul32uhilo (Const32 [c]) (Const32 [d])) => (MakeTuple (Const32 <typ.UInt32> [bitsMulU32(c, d).hi]) (Const32 <typ.UInt32> [bitsMulU32(c,d).lo]))
143(Mul64uhilo (Const64 [c]) (Const64 [d])) => (MakeTuple (Const64 <typ.UInt64> [bitsMulU64(c, d).hi]) (Const64 <typ.UInt64> [bitsMulU64(c,d).lo]))
144(Mul32uover (Const32 [c]) (Const32 [d])) => (MakeTuple (Const32 <typ.UInt32> [bitsMulU32(c, d).lo]) (ConstBool <typ.Bool> [bitsMulU32(c,d).hi != 0]))
145(Mul64uover (Const64 [c]) (Const64 [d])) => (MakeTuple (Const64 <typ.UInt64> [bitsMulU64(c, d).lo]) (ConstBool <typ.Bool> [bitsMulU64(c,d).hi != 0]))
146
147(And8 (Const8 [c]) (Const8 [d])) => (Const8 [c&d])
148(And16 (Const16 [c]) (Const16 [d])) => (Const16 [c&d])
149(And32 (Const32 [c]) (Const32 [d])) => (Const32 [c&d])
150(And64 (Const64 [c]) (Const64 [d])) => (Const64 [c&d])
151
152(Or8 (Const8 [c]) (Const8 [d])) => (Const8 [c|d])
153(Or16 (Const16 [c]) (Const16 [d])) => (Const16 [c|d])
154(Or32 (Const32 [c]) (Const32 [d])) => (Const32 [c|d])
155(Or64 (Const64 [c]) (Const64 [d])) => (Const64 [c|d])
156
157(Xor8 (Const8 [c]) (Const8 [d])) => (Const8 [c^d])
158(Xor16 (Const16 [c]) (Const16 [d])) => (Const16 [c^d])
159(Xor32 (Const32 [c]) (Const32 [d])) => (Const32 [c^d])
160(Xor64 (Const64 [c]) (Const64 [d])) => (Const64 [c^d])
161
162(Ctz64 (Const64 [c])) && config.PtrSize == 4 => (Const32 [int32(ntz64(c))])
163(Ctz32 (Const32 [c])) && config.PtrSize == 4 => (Const32 [int32(ntz32(c))])
164(Ctz16 (Const16 [c])) && config.PtrSize == 4 => (Const32 [int32(ntz16(c))])
165(Ctz8 (Const8 [c])) && config.PtrSize == 4 => (Const32 [int32(ntz8(c))])
166
167(Ctz64 (Const64 [c])) && config.PtrSize == 8 => (Const64 [int64(ntz64(c))])
168(Ctz32 (Const32 [c])) && config.PtrSize == 8 => (Const64 [int64(ntz32(c))])
169(Ctz16 (Const16 [c])) && config.PtrSize == 8 => (Const64 [int64(ntz16(c))])
170(Ctz8 (Const8 [c])) && config.PtrSize == 8 => (Const64 [int64(ntz8(c))])
171
172(Div8 (Const8 [c]) (Const8 [d])) && d != 0 => (Const8 [c/d])
173(Div16 (Const16 [c]) (Const16 [d])) && d != 0 => (Const16 [c/d])
174(Div32 (Const32 [c]) (Const32 [d])) && d != 0 => (Const32 [c/d])
175(Div64 (Const64 [c]) (Const64 [d])) && d != 0 => (Const64 [c/d])
176(Div8u (Const8 [c]) (Const8 [d])) && d != 0 => (Const8 [int8(uint8(c)/uint8(d))])
177(Div16u (Const16 [c]) (Const16 [d])) && d != 0 => (Const16 [int16(uint16(c)/uint16(d))])
178(Div32u (Const32 [c]) (Const32 [d])) && d != 0 => (Const32 [int32(uint32(c)/uint32(d))])
179(Div64u (Const64 [c]) (Const64 [d])) && d != 0 => (Const64 [int64(uint64(c)/uint64(d))])
180(Div32F (Const32F [c]) (Const32F [d])) && c/d == c/d => (Const32F [c/d])
181(Div64F (Const64F [c]) (Const64F [d])) && c/d == c/d => (Const64F [c/d])
182(Div128u <t> (Const64 [0]) lo y) => (MakeTuple (Div64u <t.FieldType(0)> lo y) (Mod64u <t.FieldType(1)> lo y))
183
184(Not (ConstBool [c])) => (ConstBool [!c])
185
186(Floor (Const64F [c])) => (Const64F [math.Floor(c)])
187(Ceil (Const64F [c])) => (Const64F [math.Ceil(c)])
188(Trunc (Const64F [c])) => (Const64F [math.Trunc(c)])
189(RoundToEven (Const64F [c])) => (Const64F [math.RoundToEven(c)])
190
191// Convert x * 1 to x.
192(Mul(8|16|32|64) (Const(8|16|32|64) [1]) x) => x
193(Mul(32|64)uover <t> (Const(32|64) [1]) x) => (MakeTuple x (ConstBool <t.FieldType(1)> [false]))
194
195// Convert x * -1 to -x.
196(Mul(8|16|32|64) (Const(8|16|32|64) [-1]) x) => (Neg(8|16|32|64) x)
197
198// DeMorgan's Laws
199(And(8|16|32|64) <t> (Com(8|16|32|64) x) (Com(8|16|32|64) y)) => (Com(8|16|32|64) (Or(8|16|32|64) <t> x y))
200(Or(8|16|32|64) <t> (Com(8|16|32|64) x) (Com(8|16|32|64) y)) => (Com(8|16|32|64) (And(8|16|32|64) <t> x y))
201
202// Convert multiplication by a power of two to a shift.
203(Mul8 <t> n (Const8 [c])) && isPowerOfTwo(c) => (Lsh8x64 <t> n (Const64 <typ.UInt64> [log8(c)]))
204(Mul16 <t> n (Const16 [c])) && isPowerOfTwo(c) => (Lsh16x64 <t> n (Const64 <typ.UInt64> [log16(c)]))
205(Mul32 <t> n (Const32 [c])) && isPowerOfTwo(c) => (Lsh32x64 <t> n (Const64 <typ.UInt64> [log32(c)]))
206(Mul64 <t> n (Const64 [c])) && isPowerOfTwo(c) => (Lsh64x64 <t> n (Const64 <typ.UInt64> [log64(c)]))
207(Mul8 <t> n (Const8 [c])) && t.IsSigned() && isPowerOfTwo(-c) => (Neg8 (Lsh8x64 <t> n (Const64 <typ.UInt64> [log8(-c)])))
208(Mul16 <t> n (Const16 [c])) && t.IsSigned() && isPowerOfTwo(-c) => (Neg16 (Lsh16x64 <t> n (Const64 <typ.UInt64> [log16(-c)])))
209(Mul32 <t> n (Const32 [c])) && t.IsSigned() && isPowerOfTwo(-c) => (Neg32 (Lsh32x64 <t> n (Const64 <typ.UInt64> [log32(-c)])))
210(Mul64 <t> n (Const64 [c])) && t.IsSigned() && isPowerOfTwo(-c) => (Neg64 (Lsh64x64 <t> n (Const64 <typ.UInt64> [log64(-c)])))
211
212(Mod8 (Const8 [c]) (Const8 [d])) && d != 0 => (Const8 [c % d])
213(Mod16 (Const16 [c]) (Const16 [d])) && d != 0 => (Const16 [c % d])
214(Mod32 (Const32 [c]) (Const32 [d])) && d != 0 => (Const32 [c % d])
215(Mod64 (Const64 [c]) (Const64 [d])) && d != 0 => (Const64 [c % d])
216
217(Mod8u (Const8 [c]) (Const8 [d])) && d != 0 => (Const8 [int8(uint8(c) % uint8(d))])
218(Mod16u (Const16 [c]) (Const16 [d])) && d != 0 => (Const16 [int16(uint16(c) % uint16(d))])
219(Mod32u (Const32 [c]) (Const32 [d])) && d != 0 => (Const32 [int32(uint32(c) % uint32(d))])
220(Mod64u (Const64 [c]) (Const64 [d])) && d != 0 => (Const64 [int64(uint64(c) % uint64(d))])
221
222(Lsh64x64 (Const64 [c]) (Const64 [d])) => (Const64 [c << uint64(d)])
223(Rsh64x64 (Const64 [c]) (Const64 [d])) => (Const64 [c >> uint64(d)])
224(Rsh64Ux64 (Const64 [c]) (Const64 [d])) => (Const64 [int64(uint64(c) >> uint64(d))])
225(Lsh32x64 (Const32 [c]) (Const64 [d])) => (Const32 [c << uint64(d)])
226(Rsh32x64 (Const32 [c]) (Const64 [d])) => (Const32 [c >> uint64(d)])
227(Rsh32Ux64 (Const32 [c]) (Const64 [d])) => (Const32 [int32(uint32(c) >> uint64(d))])
228(Lsh16x64 (Const16 [c]) (Const64 [d])) => (Const16 [c << uint64(d)])
229(Rsh16x64 (Const16 [c]) (Const64 [d])) => (Const16 [c >> uint64(d)])
230(Rsh16Ux64 (Const16 [c]) (Const64 [d])) => (Const16 [int16(uint16(c) >> uint64(d))])
231(Lsh8x64 (Const8 [c]) (Const64 [d])) => (Const8 [c << uint64(d)])
232(Rsh8x64 (Const8 [c]) (Const64 [d])) => (Const8 [c >> uint64(d)])
233(Rsh8Ux64 (Const8 [c]) (Const64 [d])) => (Const8 [int8(uint8(c) >> uint64(d))])
234
235// Fold IsInBounds when the range of the index cannot exceed the limit.
236(IsInBounds (ZeroExt8to32 _) (Const32 [c])) && (1 << 8) <= c => (ConstBool [true])
237(IsInBounds (ZeroExt8to64 _) (Const64 [c])) && (1 << 8) <= c => (ConstBool [true])
238(IsInBounds (ZeroExt16to32 _) (Const32 [c])) && (1 << 16) <= c => (ConstBool [true])
239(IsInBounds (ZeroExt16to64 _) (Const64 [c])) && (1 << 16) <= c => (ConstBool [true])
240(IsInBounds x x) => (ConstBool [false])
241(IsInBounds (And8 (Const8 [c]) _) (Const8 [d])) && 0 <= c && c < d => (ConstBool [true])
242(IsInBounds (ZeroExt8to16 (And8 (Const8 [c]) _)) (Const16 [d])) && 0 <= c && int16(c) < d => (ConstBool [true])
243(IsInBounds (ZeroExt8to32 (And8 (Const8 [c]) _)) (Const32 [d])) && 0 <= c && int32(c) < d => (ConstBool [true])
244(IsInBounds (ZeroExt8to64 (And8 (Const8 [c]) _)) (Const64 [d])) && 0 <= c && int64(c) < d => (ConstBool [true])
245(IsInBounds (And16 (Const16 [c]) _) (Const16 [d])) && 0 <= c && c < d => (ConstBool [true])
246(IsInBounds (ZeroExt16to32 (And16 (Const16 [c]) _)) (Const32 [d])) && 0 <= c && int32(c) < d => (ConstBool [true])
247(IsInBounds (ZeroExt16to64 (And16 (Const16 [c]) _)) (Const64 [d])) && 0 <= c && int64(c) < d => (ConstBool [true])
248(IsInBounds (And32 (Const32 [c]) _) (Const32 [d])) && 0 <= c && c < d => (ConstBool [true])
249(IsInBounds (ZeroExt32to64 (And32 (Const32 [c]) _)) (Const64 [d])) && 0 <= c && int64(c) < d => (ConstBool [true])
250(IsInBounds (And64 (Const64 [c]) _) (Const64 [d])) && 0 <= c && c < d => (ConstBool [true])
251(IsInBounds (Const32 [c]) (Const32 [d])) => (ConstBool [0 <= c && c < d])
252(IsInBounds (Const64 [c]) (Const64 [d])) => (ConstBool [0 <= c && c < d])
253// (Mod64u x y) is always between 0 (inclusive) and y (exclusive).
254(IsInBounds (Mod32u _ y) y) => (ConstBool [true])
255(IsInBounds (Mod64u _ y) y) => (ConstBool [true])
256// Right shifting an unsigned number limits its value.
257(IsInBounds (ZeroExt8to64 (Rsh8Ux64 _ (Const64 [c]))) (Const64 [d])) && 0 < c && c < 8 && 1<<uint( 8-c)-1 < d => (ConstBool [true])
258(IsInBounds (ZeroExt8to32 (Rsh8Ux64 _ (Const64 [c]))) (Const32 [d])) && 0 < c && c < 8 && 1<<uint( 8-c)-1 < d => (ConstBool [true])
259(IsInBounds (ZeroExt8to16 (Rsh8Ux64 _ (Const64 [c]))) (Const16 [d])) && 0 < c && c < 8 && 1<<uint( 8-c)-1 < d => (ConstBool [true])
260(IsInBounds (Rsh8Ux64 _ (Const64 [c])) (Const64 [d])) && 0 < c && c < 8 && 1<<uint( 8-c)-1 < d => (ConstBool [true])
261(IsInBounds (ZeroExt16to64 (Rsh16Ux64 _ (Const64 [c]))) (Const64 [d])) && 0 < c && c < 16 && 1<<uint(16-c)-1 < d => (ConstBool [true])
262(IsInBounds (ZeroExt16to32 (Rsh16Ux64 _ (Const64 [c]))) (Const64 [d])) && 0 < c && c < 16 && 1<<uint(16-c)-1 < d => (ConstBool [true])
263(IsInBounds (Rsh16Ux64 _ (Const64 [c])) (Const64 [d])) && 0 < c && c < 16 && 1<<uint(16-c)-1 < d => (ConstBool [true])
264(IsInBounds (ZeroExt32to64 (Rsh32Ux64 _ (Const64 [c]))) (Const64 [d])) && 0 < c && c < 32 && 1<<uint(32-c)-1 < d => (ConstBool [true])
265(IsInBounds (Rsh32Ux64 _ (Const64 [c])) (Const64 [d])) && 0 < c && c < 32 && 1<<uint(32-c)-1 < d => (ConstBool [true])
266(IsInBounds (Rsh64Ux64 _ (Const64 [c])) (Const64 [d])) && 0 < c && c < 64 && 1<<uint(64-c)-1 < d => (ConstBool [true])
267
268(IsSliceInBounds x x) => (ConstBool [true])
269(IsSliceInBounds (And32 (Const32 [c]) _) (Const32 [d])) && 0 <= c && c <= d => (ConstBool [true])
270(IsSliceInBounds (And64 (Const64 [c]) _) (Const64 [d])) && 0 <= c && c <= d => (ConstBool [true])
271(IsSliceInBounds (Const32 [0]) _) => (ConstBool [true])
272(IsSliceInBounds (Const64 [0]) _) => (ConstBool [true])
273(IsSliceInBounds (Const32 [c]) (Const32 [d])) => (ConstBool [0 <= c && c <= d])
274(IsSliceInBounds (Const64 [c]) (Const64 [d])) => (ConstBool [0 <= c && c <= d])
275(IsSliceInBounds (SliceLen x) (SliceCap x)) => (ConstBool [true])
276
277(Eq(64|32|16|8) x x) => (ConstBool [true])
278(EqB (ConstBool [c]) (ConstBool [d])) => (ConstBool [c == d])
279(EqB (ConstBool [false]) x) => (Not x)
280(EqB (ConstBool [true]) x) => x
281
282(Neq(64|32|16|8) x x) => (ConstBool [false])
283(NeqB (ConstBool [c]) (ConstBool [d])) => (ConstBool [c != d])
284(NeqB (ConstBool [false]) x) => x
285(NeqB (ConstBool [true]) x) => (Not x)
286(NeqB (Not x) (Not y)) => (NeqB x y)
287
288(Eq64 (Const64 <t> [c]) (Add64 (Const64 <t> [d]) x)) => (Eq64 (Const64 <t> [c-d]) x)
289(Eq32 (Const32 <t> [c]) (Add32 (Const32 <t> [d]) x)) => (Eq32 (Const32 <t> [c-d]) x)
290(Eq16 (Const16 <t> [c]) (Add16 (Const16 <t> [d]) x)) => (Eq16 (Const16 <t> [c-d]) x)
291(Eq8 (Const8 <t> [c]) (Add8 (Const8 <t> [d]) x)) => (Eq8 (Const8 <t> [c-d]) x)
292
293(Neq64 (Const64 <t> [c]) (Add64 (Const64 <t> [d]) x)) => (Neq64 (Const64 <t> [c-d]) x)
294(Neq32 (Const32 <t> [c]) (Add32 (Const32 <t> [d]) x)) => (Neq32 (Const32 <t> [c-d]) x)
295(Neq16 (Const16 <t> [c]) (Add16 (Const16 <t> [d]) x)) => (Neq16 (Const16 <t> [c-d]) x)
296(Neq8 (Const8 <t> [c]) (Add8 (Const8 <t> [d]) x)) => (Neq8 (Const8 <t> [c-d]) x)
297
298// signed integer range: ( c <= x && x (<|<=) d ) -> ( unsigned(x-c) (<|<=) unsigned(d-c) )
299(AndB (Leq64 (Const64 [c]) x) ((Less|Leq)64 x (Const64 [d]))) && d >= c => ((Less|Leq)64U (Sub64 <x.Type> x (Const64 <x.Type> [c])) (Const64 <x.Type> [d-c]))
300(AndB (Leq32 (Const32 [c]) x) ((Less|Leq)32 x (Const32 [d]))) && d >= c => ((Less|Leq)32U (Sub32 <x.Type> x (Const32 <x.Type> [c])) (Const32 <x.Type> [d-c]))
301(AndB (Leq16 (Const16 [c]) x) ((Less|Leq)16 x (Const16 [d]))) && d >= c => ((Less|Leq)16U (Sub16 <x.Type> x (Const16 <x.Type> [c])) (Const16 <x.Type> [d-c]))
302(AndB (Leq8 (Const8 [c]) x) ((Less|Leq)8 x (Const8 [d]))) && d >= c => ((Less|Leq)8U (Sub8 <x.Type> x (Const8 <x.Type> [c])) (Const8 <x.Type> [d-c]))
303
304// signed integer range: ( c < x && x (<|<=) d ) -> ( unsigned(x-(c+1)) (<|<=) unsigned(d-(c+1)) )
305(AndB (Less64 (Const64 [c]) x) ((Less|Leq)64 x (Const64 [d]))) && d >= c+1 && c+1 > c => ((Less|Leq)64U (Sub64 <x.Type> x (Const64 <x.Type> [c+1])) (Const64 <x.Type> [d-c-1]))
306(AndB (Less32 (Const32 [c]) x) ((Less|Leq)32 x (Const32 [d]))) && d >= c+1 && c+1 > c => ((Less|Leq)32U (Sub32 <x.Type> x (Const32 <x.Type> [c+1])) (Const32 <x.Type> [d-c-1]))
307(AndB (Less16 (Const16 [c]) x) ((Less|Leq)16 x (Const16 [d]))) && d >= c+1 && c+1 > c => ((Less|Leq)16U (Sub16 <x.Type> x (Const16 <x.Type> [c+1])) (Const16 <x.Type> [d-c-1]))
308(AndB (Less8 (Const8 [c]) x) ((Less|Leq)8 x (Const8 [d]))) && d >= c+1 && c+1 > c => ((Less|Leq)8U (Sub8 <x.Type> x (Const8 <x.Type> [c+1])) (Const8 <x.Type> [d-c-1]))
309
310// unsigned integer range: ( c <= x && x (<|<=) d ) -> ( x-c (<|<=) d-c )
311(AndB (Leq64U (Const64 [c]) x) ((Less|Leq)64U x (Const64 [d]))) && uint64(d) >= uint64(c) => ((Less|Leq)64U (Sub64 <x.Type> x (Const64 <x.Type> [c])) (Const64 <x.Type> [d-c]))
312(AndB (Leq32U (Const32 [c]) x) ((Less|Leq)32U x (Const32 [d]))) && uint32(d) >= uint32(c) => ((Less|Leq)32U (Sub32 <x.Type> x (Const32 <x.Type> [c])) (Const32 <x.Type> [d-c]))
313(AndB (Leq16U (Const16 [c]) x) ((Less|Leq)16U x (Const16 [d]))) && uint16(d) >= uint16(c) => ((Less|Leq)16U (Sub16 <x.Type> x (Const16 <x.Type> [c])) (Const16 <x.Type> [d-c]))
314(AndB (Leq8U (Const8 [c]) x) ((Less|Leq)8U x (Const8 [d]))) && uint8(d) >= uint8(c) => ((Less|Leq)8U (Sub8 <x.Type> x (Const8 <x.Type> [c])) (Const8 <x.Type> [d-c]))
315
316// unsigned integer range: ( c < x && x (<|<=) d ) -> ( x-(c+1) (<|<=) d-(c+1) )
317(AndB (Less64U (Const64 [c]) x) ((Less|Leq)64U x (Const64 [d]))) && uint64(d) >= uint64(c+1) && uint64(c+1) > uint64(c) => ((Less|Leq)64U (Sub64 <x.Type> x (Const64 <x.Type> [c+1])) (Const64 <x.Type> [d-c-1]))
318(AndB (Less32U (Const32 [c]) x) ((Less|Leq)32U x (Const32 [d]))) && uint32(d) >= uint32(c+1) && uint32(c+1) > uint32(c) => ((Less|Leq)32U (Sub32 <x.Type> x (Const32 <x.Type> [c+1])) (Const32 <x.Type> [d-c-1]))
319(AndB (Less16U (Const16 [c]) x) ((Less|Leq)16U x (Const16 [d]))) && uint16(d) >= uint16(c+1) && uint16(c+1) > uint16(c) => ((Less|Leq)16U (Sub16 <x.Type> x (Const16 <x.Type> [c+1])) (Const16 <x.Type> [d-c-1]))
320(AndB (Less8U (Const8 [c]) x) ((Less|Leq)8U x (Const8 [d]))) && uint8(d) >= uint8(c+1) && uint8(c+1) > uint8(c) => ((Less|Leq)8U (Sub8 <x.Type> x (Const8 <x.Type> [c+1])) (Const8 <x.Type> [d-c-1]))
321
322// signed integer range: ( c (<|<=) x || x < d ) -> ( unsigned(c-d) (<|<=) unsigned(x-d) )
323(OrB ((Less|Leq)64 (Const64 [c]) x) (Less64 x (Const64 [d]))) && c >= d => ((Less|Leq)64U (Const64 <x.Type> [c-d]) (Sub64 <x.Type> x (Const64 <x.Type> [d])))
324(OrB ((Less|Leq)32 (Const32 [c]) x) (Less32 x (Const32 [d]))) && c >= d => ((Less|Leq)32U (Const32 <x.Type> [c-d]) (Sub32 <x.Type> x (Const32 <x.Type> [d])))
325(OrB ((Less|Leq)16 (Const16 [c]) x) (Less16 x (Const16 [d]))) && c >= d => ((Less|Leq)16U (Const16 <x.Type> [c-d]) (Sub16 <x.Type> x (Const16 <x.Type> [d])))
326(OrB ((Less|Leq)8 (Const8 [c]) x) (Less8 x (Const8 [d]))) && c >= d => ((Less|Leq)8U (Const8 <x.Type> [c-d]) (Sub8 <x.Type> x (Const8 <x.Type> [d])))
327
328// signed integer range: ( c (<|<=) x || x <= d ) -> ( unsigned(c-(d+1)) (<|<=) unsigned(x-(d+1)) )
329(OrB ((Less|Leq)64 (Const64 [c]) x) (Leq64 x (Const64 [d]))) && c >= d+1 && d+1 > d => ((Less|Leq)64U (Const64 <x.Type> [c-d-1]) (Sub64 <x.Type> x (Const64 <x.Type> [d+1])))
330(OrB ((Less|Leq)32 (Const32 [c]) x) (Leq32 x (Const32 [d]))) && c >= d+1 && d+1 > d => ((Less|Leq)32U (Const32 <x.Type> [c-d-1]) (Sub32 <x.Type> x (Const32 <x.Type> [d+1])))
331(OrB ((Less|Leq)16 (Const16 [c]) x) (Leq16 x (Const16 [d]))) && c >= d+1 && d+1 > d => ((Less|Leq)16U (Const16 <x.Type> [c-d-1]) (Sub16 <x.Type> x (Const16 <x.Type> [d+1])))
332(OrB ((Less|Leq)8 (Const8 [c]) x) (Leq8 x (Const8 [d]))) && c >= d+1 && d+1 > d => ((Less|Leq)8U (Const8 <x.Type> [c-d-1]) (Sub8 <x.Type> x (Const8 <x.Type> [d+1])))
333
334// unsigned integer range: ( c (<|<=) x || x < d ) -> ( c-d (<|<=) x-d )
335(OrB ((Less|Leq)64U (Const64 [c]) x) (Less64U x (Const64 [d]))) && uint64(c) >= uint64(d) => ((Less|Leq)64U (Const64 <x.Type> [c-d]) (Sub64 <x.Type> x (Const64 <x.Type> [d])))
336(OrB ((Less|Leq)32U (Const32 [c]) x) (Less32U x (Const32 [d]))) && uint32(c) >= uint32(d) => ((Less|Leq)32U (Const32 <x.Type> [c-d]) (Sub32 <x.Type> x (Const32 <x.Type> [d])))
337(OrB ((Less|Leq)16U (Const16 [c]) x) (Less16U x (Const16 [d]))) && uint16(c) >= uint16(d) => ((Less|Leq)16U (Const16 <x.Type> [c-d]) (Sub16 <x.Type> x (Const16 <x.Type> [d])))
338(OrB ((Less|Leq)8U (Const8 [c]) x) (Less8U x (Const8 [d]))) && uint8(c) >= uint8(d) => ((Less|Leq)8U (Const8 <x.Type> [c-d]) (Sub8 <x.Type> x (Const8 <x.Type> [d])))
339
340// unsigned integer range: ( c (<|<=) x || x <= d ) -> ( c-(d+1) (<|<=) x-(d+1) )
341(OrB ((Less|Leq)64U (Const64 [c]) x) (Leq64U x (Const64 [d]))) && uint64(c) >= uint64(d+1) && uint64(d+1) > uint64(d) => ((Less|Leq)64U (Const64 <x.Type> [c-d-1]) (Sub64 <x.Type> x (Const64 <x.Type> [d+1])))
342(OrB ((Less|Leq)32U (Const32 [c]) x) (Leq32U x (Const32 [d]))) && uint32(c) >= uint32(d+1) && uint32(d+1) > uint32(d) => ((Less|Leq)32U (Const32 <x.Type> [c-d-1]) (Sub32 <x.Type> x (Const32 <x.Type> [d+1])))
343(OrB ((Less|Leq)16U (Const16 [c]) x) (Leq16U x (Const16 [d]))) && uint16(c) >= uint16(d+1) && uint16(d+1) > uint16(d) => ((Less|Leq)16U (Const16 <x.Type> [c-d-1]) (Sub16 <x.Type> x (Const16 <x.Type> [d+1])))
344(OrB ((Less|Leq)8U (Const8 [c]) x) (Leq8U x (Const8 [d]))) && uint8(c) >= uint8(d+1) && uint8(d+1) > uint8(d) => ((Less|Leq)8U (Const8 <x.Type> [c-d-1]) (Sub8 <x.Type> x (Const8 <x.Type> [d+1])))
345
346// Canonicalize x-const to x+(-const)
347(Sub64 x (Const64 <t> [c])) && x.Op != OpConst64 => (Add64 (Const64 <t> [-c]) x)
348(Sub32 x (Const32 <t> [c])) && x.Op != OpConst32 => (Add32 (Const32 <t> [-c]) x)
349(Sub16 x (Const16 <t> [c])) && x.Op != OpConst16 => (Add16 (Const16 <t> [-c]) x)
350(Sub8 x (Const8 <t> [c])) && x.Op != OpConst8 => (Add8 (Const8 <t> [-c]) x)
351
352// fold negation into comparison operators
353(Not (Eq(64|32|16|8|B|Ptr|64F|32F) x y)) => (Neq(64|32|16|8|B|Ptr|64F|32F) x y)
354(Not (Neq(64|32|16|8|B|Ptr|64F|32F) x y)) => (Eq(64|32|16|8|B|Ptr|64F|32F) x y)
355
356(Not (Less(64|32|16|8) x y)) => (Leq(64|32|16|8) y x)
357(Not (Less(64|32|16|8)U x y)) => (Leq(64|32|16|8)U y x)
358(Not (Leq(64|32|16|8) x y)) => (Less(64|32|16|8) y x)
359(Not (Leq(64|32|16|8)U x y)) => (Less(64|32|16|8)U y x)
360
361// Distribute multiplication c * (d+x) -> c*d + c*x. Useful for:
362// a[i].b = ...; a[i+1].b = ...
363(Mul64 (Const64 <t> [c]) (Add64 <t> (Const64 <t> [d]) x)) =>
364 (Add64 (Const64 <t> [c*d]) (Mul64 <t> (Const64 <t> [c]) x))
365(Mul32 (Const32 <t> [c]) (Add32 <t> (Const32 <t> [d]) x)) =>
366 (Add32 (Const32 <t> [c*d]) (Mul32 <t> (Const32 <t> [c]) x))
367(Mul16 (Const16 <t> [c]) (Add16 <t> (Const16 <t> [d]) x)) =>
368 (Add16 (Const16 <t> [c*d]) (Mul16 <t> (Const16 <t> [c]) x))
369(Mul8 (Const8 <t> [c]) (Add8 <t> (Const8 <t> [d]) x)) =>
370 (Add8 (Const8 <t> [c*d]) (Mul8 <t> (Const8 <t> [c]) x))
371
372// Rewrite x*y ± x*z to x*(y±z)
373(Add(64|32|16|8) <t> (Mul(64|32|16|8) x y) (Mul(64|32|16|8) x z))
374 => (Mul(64|32|16|8) x (Add(64|32|16|8) <t> y z))
375(Sub(64|32|16|8) <t> (Mul(64|32|16|8) x y) (Mul(64|32|16|8) x z))
376 => (Mul(64|32|16|8) x (Sub(64|32|16|8) <t> y z))
377
378// rewrite shifts of 8/16/32 bit consts into 64 bit consts to reduce
379// the number of the other rewrite rules for const shifts
380(Lsh64x32 <t> x (Const32 [c])) => (Lsh64x64 x (Const64 <t> [int64(uint32(c))]))
381(Lsh64x16 <t> x (Const16 [c])) => (Lsh64x64 x (Const64 <t> [int64(uint16(c))]))
382(Lsh64x8 <t> x (Const8 [c])) => (Lsh64x64 x (Const64 <t> [int64(uint8(c))]))
383(Rsh64x32 <t> x (Const32 [c])) => (Rsh64x64 x (Const64 <t> [int64(uint32(c))]))
384(Rsh64x16 <t> x (Const16 [c])) => (Rsh64x64 x (Const64 <t> [int64(uint16(c))]))
385(Rsh64x8 <t> x (Const8 [c])) => (Rsh64x64 x (Const64 <t> [int64(uint8(c))]))
386(Rsh64Ux32 <t> x (Const32 [c])) => (Rsh64Ux64 x (Const64 <t> [int64(uint32(c))]))
387(Rsh64Ux16 <t> x (Const16 [c])) => (Rsh64Ux64 x (Const64 <t> [int64(uint16(c))]))
388(Rsh64Ux8 <t> x (Const8 [c])) => (Rsh64Ux64 x (Const64 <t> [int64(uint8(c))]))
389
390(Lsh32x32 <t> x (Const32 [c])) => (Lsh32x64 x (Const64 <t> [int64(uint32(c))]))
391(Lsh32x16 <t> x (Const16 [c])) => (Lsh32x64 x (Const64 <t> [int64(uint16(c))]))
392(Lsh32x8 <t> x (Const8 [c])) => (Lsh32x64 x (Const64 <t> [int64(uint8(c))]))
393(Rsh32x32 <t> x (Const32 [c])) => (Rsh32x64 x (Const64 <t> [int64(uint32(c))]))
394(Rsh32x16 <t> x (Const16 [c])) => (Rsh32x64 x (Const64 <t> [int64(uint16(c))]))
395(Rsh32x8 <t> x (Const8 [c])) => (Rsh32x64 x (Const64 <t> [int64(uint8(c))]))
396(Rsh32Ux32 <t> x (Const32 [c])) => (Rsh32Ux64 x (Const64 <t> [int64(uint32(c))]))
397(Rsh32Ux16 <t> x (Const16 [c])) => (Rsh32Ux64 x (Const64 <t> [int64(uint16(c))]))
398(Rsh32Ux8 <t> x (Const8 [c])) => (Rsh32Ux64 x (Const64 <t> [int64(uint8(c))]))
399
400(Lsh16x32 <t> x (Const32 [c])) => (Lsh16x64 x (Const64 <t> [int64(uint32(c))]))
401(Lsh16x16 <t> x (Const16 [c])) => (Lsh16x64 x (Const64 <t> [int64(uint16(c))]))
402(Lsh16x8 <t> x (Const8 [c])) => (Lsh16x64 x (Const64 <t> [int64(uint8(c))]))
403(Rsh16x32 <t> x (Const32 [c])) => (Rsh16x64 x (Const64 <t> [int64(uint32(c))]))
404(Rsh16x16 <t> x (Const16 [c])) => (Rsh16x64 x (Const64 <t> [int64(uint16(c))]))
405(Rsh16x8 <t> x (Const8 [c])) => (Rsh16x64 x (Const64 <t> [int64(uint8(c))]))
406(Rsh16Ux32 <t> x (Const32 [c])) => (Rsh16Ux64 x (Const64 <t> [int64(uint32(c))]))
407(Rsh16Ux16 <t> x (Const16 [c])) => (Rsh16Ux64 x (Const64 <t> [int64(uint16(c))]))
408(Rsh16Ux8 <t> x (Const8 [c])) => (Rsh16Ux64 x (Const64 <t> [int64(uint8(c))]))
409
410(Lsh8x32 <t> x (Const32 [c])) => (Lsh8x64 x (Const64 <t> [int64(uint32(c))]))
411(Lsh8x16 <t> x (Const16 [c])) => (Lsh8x64 x (Const64 <t> [int64(uint16(c))]))
412(Lsh8x8 <t> x (Const8 [c])) => (Lsh8x64 x (Const64 <t> [int64(uint8(c))]))
413(Rsh8x32 <t> x (Const32 [c])) => (Rsh8x64 x (Const64 <t> [int64(uint32(c))]))
414(Rsh8x16 <t> x (Const16 [c])) => (Rsh8x64 x (Const64 <t> [int64(uint16(c))]))
415(Rsh8x8 <t> x (Const8 [c])) => (Rsh8x64 x (Const64 <t> [int64(uint8(c))]))
416(Rsh8Ux32 <t> x (Const32 [c])) => (Rsh8Ux64 x (Const64 <t> [int64(uint32(c))]))
417(Rsh8Ux16 <t> x (Const16 [c])) => (Rsh8Ux64 x (Const64 <t> [int64(uint16(c))]))
418(Rsh8Ux8 <t> x (Const8 [c])) => (Rsh8Ux64 x (Const64 <t> [int64(uint8(c))]))
419
420// shifts by zero
421(Lsh(64|32|16|8)x64 x (Const64 [0])) => x
422(Rsh(64|32|16|8)x64 x (Const64 [0])) => x
423(Rsh(64|32|16|8)Ux64 x (Const64 [0])) => x
424
425// rotates by multiples of register width
426(RotateLeft64 x (Const64 [c])) && c%64 == 0 => x
427(RotateLeft32 x (Const32 [c])) && c%32 == 0 => x
428(RotateLeft16 x (Const16 [c])) && c%16 == 0 => x
429(RotateLeft8 x (Const8 [c])) && c%8 == 0 => x
430
431// zero shifted
432(Lsh64x(64|32|16|8) (Const64 [0]) _) => (Const64 [0])
433(Rsh64x(64|32|16|8) (Const64 [0]) _) => (Const64 [0])
434(Rsh64Ux(64|32|16|8) (Const64 [0]) _) => (Const64 [0])
435(Lsh32x(64|32|16|8) (Const32 [0]) _) => (Const32 [0])
436(Rsh32x(64|32|16|8) (Const32 [0]) _) => (Const32 [0])
437(Rsh32Ux(64|32|16|8) (Const32 [0]) _) => (Const32 [0])
438(Lsh16x(64|32|16|8) (Const16 [0]) _) => (Const16 [0])
439(Rsh16x(64|32|16|8) (Const16 [0]) _) => (Const16 [0])
440(Rsh16Ux(64|32|16|8) (Const16 [0]) _) => (Const16 [0])
441(Lsh8x(64|32|16|8) (Const8 [0]) _) => (Const8 [0])
442(Rsh8x(64|32|16|8) (Const8 [0]) _) => (Const8 [0])
443(Rsh8Ux(64|32|16|8) (Const8 [0]) _) => (Const8 [0])
444
445// large left shifts of all values, and right shifts of unsigned values
446((Lsh64|Rsh64U)x64 _ (Const64 [c])) && uint64(c) >= 64 => (Const64 [0])
447((Lsh32|Rsh32U)x64 _ (Const64 [c])) && uint64(c) >= 32 => (Const32 [0])
448((Lsh16|Rsh16U)x64 _ (Const64 [c])) && uint64(c) >= 16 => (Const16 [0])
449((Lsh8|Rsh8U)x64 _ (Const64 [c])) && uint64(c) >= 8 => (Const8 [0])
450
451// combine const shifts
452(Lsh64x64 <t> (Lsh64x64 x (Const64 [c])) (Const64 [d])) && !uaddOvf(c,d) => (Lsh64x64 x (Const64 <t> [c+d]))
453(Lsh32x64 <t> (Lsh32x64 x (Const64 [c])) (Const64 [d])) && !uaddOvf(c,d) => (Lsh32x64 x (Const64 <t> [c+d]))
454(Lsh16x64 <t> (Lsh16x64 x (Const64 [c])) (Const64 [d])) && !uaddOvf(c,d) => (Lsh16x64 x (Const64 <t> [c+d]))
455(Lsh8x64 <t> (Lsh8x64 x (Const64 [c])) (Const64 [d])) && !uaddOvf(c,d) => (Lsh8x64 x (Const64 <t> [c+d]))
456
457(Rsh64x64 <t> (Rsh64x64 x (Const64 [c])) (Const64 [d])) && !uaddOvf(c,d) => (Rsh64x64 x (Const64 <t> [c+d]))
458(Rsh32x64 <t> (Rsh32x64 x (Const64 [c])) (Const64 [d])) && !uaddOvf(c,d) => (Rsh32x64 x (Const64 <t> [c+d]))
459(Rsh16x64 <t> (Rsh16x64 x (Const64 [c])) (Const64 [d])) && !uaddOvf(c,d) => (Rsh16x64 x (Const64 <t> [c+d]))
460(Rsh8x64 <t> (Rsh8x64 x (Const64 [c])) (Const64 [d])) && !uaddOvf(c,d) => (Rsh8x64 x (Const64 <t> [c+d]))
461
462(Rsh64Ux64 <t> (Rsh64Ux64 x (Const64 [c])) (Const64 [d])) && !uaddOvf(c,d) => (Rsh64Ux64 x (Const64 <t> [c+d]))
463(Rsh32Ux64 <t> (Rsh32Ux64 x (Const64 [c])) (Const64 [d])) && !uaddOvf(c,d) => (Rsh32Ux64 x (Const64 <t> [c+d]))
464(Rsh16Ux64 <t> (Rsh16Ux64 x (Const64 [c])) (Const64 [d])) && !uaddOvf(c,d) => (Rsh16Ux64 x (Const64 <t> [c+d]))
465(Rsh8Ux64 <t> (Rsh8Ux64 x (Const64 [c])) (Const64 [d])) && !uaddOvf(c,d) => (Rsh8Ux64 x (Const64 <t> [c+d]))
466
467// Remove signed right shift before an unsigned right shift that extracts the sign bit.
468(Rsh8Ux64 (Rsh8x64 x _) (Const64 <t> [7] )) => (Rsh8Ux64 x (Const64 <t> [7] ))
469(Rsh16Ux64 (Rsh16x64 x _) (Const64 <t> [15])) => (Rsh16Ux64 x (Const64 <t> [15]))
470(Rsh32Ux64 (Rsh32x64 x _) (Const64 <t> [31])) => (Rsh32Ux64 x (Const64 <t> [31]))
471(Rsh64Ux64 (Rsh64x64 x _) (Const64 <t> [63])) => (Rsh64Ux64 x (Const64 <t> [63]))
472
473// Convert x>>c<<c to x&^(1<<c-1)
474(Lsh64x64 i:(Rsh(64|64U)x64 x (Const64 [c])) (Const64 [c])) && c >= 0 && c < 64 && i.Uses == 1 => (And64 x (Const64 <v.Type> [int64(-1) << c]))
475(Lsh32x64 i:(Rsh(32|32U)x64 x (Const64 [c])) (Const64 [c])) && c >= 0 && c < 32 && i.Uses == 1 => (And32 x (Const32 <v.Type> [int32(-1) << c]))
476(Lsh16x64 i:(Rsh(16|16U)x64 x (Const64 [c])) (Const64 [c])) && c >= 0 && c < 16 && i.Uses == 1 => (And16 x (Const16 <v.Type> [int16(-1) << c]))
477(Lsh8x64 i:(Rsh(8|8U)x64 x (Const64 [c])) (Const64 [c])) && c >= 0 && c < 8 && i.Uses == 1 => (And8 x (Const8 <v.Type> [int8(-1) << c]))
478// similarly for x<<c>>c
479(Rsh64Ux64 i:(Lsh64x64 x (Const64 [c])) (Const64 [c])) && c >= 0 && c < 64 && i.Uses == 1 => (And64 x (Const64 <v.Type> [int64(^uint64(0)>>c)]))
480(Rsh32Ux64 i:(Lsh32x64 x (Const64 [c])) (Const64 [c])) && c >= 0 && c < 32 && i.Uses == 1 => (And32 x (Const32 <v.Type> [int32(^uint32(0)>>c)]))
481(Rsh16Ux64 i:(Lsh16x64 x (Const64 [c])) (Const64 [c])) && c >= 0 && c < 16 && i.Uses == 1 => (And16 x (Const16 <v.Type> [int16(^uint16(0)>>c)]))
482(Rsh8Ux64 i:(Lsh8x64 x (Const64 [c])) (Const64 [c])) && c >= 0 && c < 8 && i.Uses == 1 => (And8 x (Const8 <v.Type> [int8 (^uint8 (0)>>c)]))
483
484// ((x >> c1) << c2) >> c3
485(Rsh(64|32|16|8)Ux64 (Lsh(64|32|16|8)x64 (Rsh(64|32|16|8)Ux64 x (Const64 [c1])) (Const64 [c2])) (Const64 [c3]))
486 && uint64(c1) >= uint64(c2) && uint64(c3) >= uint64(c2) && !uaddOvf(c1-c2, c3)
487 => (Rsh(64|32|16|8)Ux64 x (Const64 <typ.UInt64> [c1-c2+c3]))
488
489// ((x << c1) >> c2) << c3
490(Lsh(64|32|16|8)x64 (Rsh(64|32|16|8)Ux64 (Lsh(64|32|16|8)x64 x (Const64 [c1])) (Const64 [c2])) (Const64 [c3]))
491 && uint64(c1) >= uint64(c2) && uint64(c3) >= uint64(c2) && !uaddOvf(c1-c2, c3)
492 => (Lsh(64|32|16|8)x64 x (Const64 <typ.UInt64> [c1-c2+c3]))
493
494// (x >> c) & uppermask = 0
495(And64 (Const64 [m]) (Rsh64Ux64 _ (Const64 [c]))) && c >= int64(64-ntz64(m)) => (Const64 [0])
496(And32 (Const32 [m]) (Rsh32Ux64 _ (Const64 [c]))) && c >= int64(32-ntz32(m)) => (Const32 [0])
497(And16 (Const16 [m]) (Rsh16Ux64 _ (Const64 [c]))) && c >= int64(16-ntz16(m)) => (Const16 [0])
498(And8 (Const8 [m]) (Rsh8Ux64 _ (Const64 [c]))) && c >= int64(8-ntz8(m)) => (Const8 [0])
499
500// (x << c) & lowermask = 0
501(And64 (Const64 [m]) (Lsh64x64 _ (Const64 [c]))) && c >= int64(64-nlz64(m)) => (Const64 [0])
502(And32 (Const32 [m]) (Lsh32x64 _ (Const64 [c]))) && c >= int64(32-nlz32(m)) => (Const32 [0])
503(And16 (Const16 [m]) (Lsh16x64 _ (Const64 [c]))) && c >= int64(16-nlz16(m)) => (Const16 [0])
504(And8 (Const8 [m]) (Lsh8x64 _ (Const64 [c]))) && c >= int64(8-nlz8(m)) => (Const8 [0])
505
506// replace shifts with zero extensions
507(Rsh16Ux64 (Lsh16x64 x (Const64 [8])) (Const64 [8])) => (ZeroExt8to16 (Trunc16to8 <typ.UInt8> x))
508(Rsh32Ux64 (Lsh32x64 x (Const64 [24])) (Const64 [24])) => (ZeroExt8to32 (Trunc32to8 <typ.UInt8> x))
509(Rsh64Ux64 (Lsh64x64 x (Const64 [56])) (Const64 [56])) => (ZeroExt8to64 (Trunc64to8 <typ.UInt8> x))
510(Rsh32Ux64 (Lsh32x64 x (Const64 [16])) (Const64 [16])) => (ZeroExt16to32 (Trunc32to16 <typ.UInt16> x))
511(Rsh64Ux64 (Lsh64x64 x (Const64 [48])) (Const64 [48])) => (ZeroExt16to64 (Trunc64to16 <typ.UInt16> x))
512(Rsh64Ux64 (Lsh64x64 x (Const64 [32])) (Const64 [32])) => (ZeroExt32to64 (Trunc64to32 <typ.UInt32> x))
513
514// replace shifts with sign extensions
515(Rsh16x64 (Lsh16x64 x (Const64 [8])) (Const64 [8])) => (SignExt8to16 (Trunc16to8 <typ.Int8> x))
516(Rsh32x64 (Lsh32x64 x (Const64 [24])) (Const64 [24])) => (SignExt8to32 (Trunc32to8 <typ.Int8> x))
517(Rsh64x64 (Lsh64x64 x (Const64 [56])) (Const64 [56])) => (SignExt8to64 (Trunc64to8 <typ.Int8> x))
518(Rsh32x64 (Lsh32x64 x (Const64 [16])) (Const64 [16])) => (SignExt16to32 (Trunc32to16 <typ.Int16> x))
519(Rsh64x64 (Lsh64x64 x (Const64 [48])) (Const64 [48])) => (SignExt16to64 (Trunc64to16 <typ.Int16> x))
520(Rsh64x64 (Lsh64x64 x (Const64 [32])) (Const64 [32])) => (SignExt32to64 (Trunc64to32 <typ.Int32> x))
521
522// ((x >> c) & d) << e
523(Lsh64x64 (And64 (Rsh(64|64U)x64 <t> x (Const64 <t2> [c])) (Const64 [d])) (Const64 [e])) && c >= e => (And64 (Rsh(64|64U)x64 <t> x (Const64 <t2> [c-e])) (Const64 <t> [d<<e]))
524(Lsh32x64 (And32 (Rsh(32|32U)x64 <t> x (Const64 <t2> [c])) (Const32 [d])) (Const64 [e])) && c >= e => (And32 (Rsh(32|32U)x64 <t> x (Const64 <t2> [c-e])) (Const32 <t> [d<<e]))
525(Lsh16x64 (And16 (Rsh(16|16U)x64 <t> x (Const64 <t2> [c])) (Const16 [d])) (Const64 [e])) && c >= e => (And16 (Rsh(16|16U)x64 <t> x (Const64 <t2> [c-e])) (Const16 <t> [d<<e]))
526(Lsh8x64 (And8 (Rsh(8|8U)x64 <t> x (Const64 <t2> [c])) (Const8 [d])) (Const64 [e])) && c >= e => (And8 (Rsh(8|8U)x64 <t> x (Const64 <t2> [c-e])) (Const8 <t> [d<<e]))
527(Lsh64x64 (And64 (Rsh(64|64U)x64 <t> x (Const64 <t2> [c])) (Const64 [d])) (Const64 [e])) && c < e => (And64 (Lsh64x64 <t> x (Const64 <t2> [e-c])) (Const64 <t> [d<<e]))
528(Lsh32x64 (And32 (Rsh(32|32U)x64 <t> x (Const64 <t2> [c])) (Const32 [d])) (Const64 [e])) && c < e => (And32 (Lsh32x64 <t> x (Const64 <t2> [e-c])) (Const32 <t> [d<<e]))
529(Lsh16x64 (And16 (Rsh(16|16U)x64 <t> x (Const64 <t2> [c])) (Const16 [d])) (Const64 [e])) && c < e => (And16 (Lsh16x64 <t> x (Const64 <t2> [e-c])) (Const16 <t> [d<<e]))
530(Lsh8x64 (And8 (Rsh(8|8U)x64 <t> x (Const64 <t2> [c])) (Const8 [d])) (Const64 [e])) && c < e => (And8 (Lsh8x64 <t> x (Const64 <t2> [e-c])) (Const8 <t> [d<<e]))
531
532// constant comparisons
533(Eq(64|32|16|8) (Const(64|32|16|8) [c]) (Const(64|32|16|8) [d])) => (ConstBool [c == d])
534(Neq(64|32|16|8) (Const(64|32|16|8) [c]) (Const(64|32|16|8) [d])) => (ConstBool [c != d])
535(Less(64|32|16|8) (Const(64|32|16|8) [c]) (Const(64|32|16|8) [d])) => (ConstBool [c < d])
536(Leq(64|32|16|8) (Const(64|32|16|8) [c]) (Const(64|32|16|8) [d])) => (ConstBool [c <= d])
537
538(Less64U (Const64 [c]) (Const64 [d])) => (ConstBool [uint64(c) < uint64(d)])
539(Less32U (Const32 [c]) (Const32 [d])) => (ConstBool [uint32(c) < uint32(d)])
540(Less16U (Const16 [c]) (Const16 [d])) => (ConstBool [uint16(c) < uint16(d)])
541(Less8U (Const8 [c]) (Const8 [d])) => (ConstBool [ uint8(c) < uint8(d)])
542
543(Leq64U (Const64 [c]) (Const64 [d])) => (ConstBool [uint64(c) <= uint64(d)])
544(Leq32U (Const32 [c]) (Const32 [d])) => (ConstBool [uint32(c) <= uint32(d)])
545(Leq16U (Const16 [c]) (Const16 [d])) => (ConstBool [uint16(c) <= uint16(d)])
546(Leq8U (Const8 [c]) (Const8 [d])) => (ConstBool [ uint8(c) <= uint8(d)])
547
548(Leq8 (Const8 [0]) (And8 _ (Const8 [c]))) && c >= 0 => (ConstBool [true])
549(Leq16 (Const16 [0]) (And16 _ (Const16 [c]))) && c >= 0 => (ConstBool [true])
550(Leq32 (Const32 [0]) (And32 _ (Const32 [c]))) && c >= 0 => (ConstBool [true])
551(Leq64 (Const64 [0]) (And64 _ (Const64 [c]))) && c >= 0 => (ConstBool [true])
552
553(Leq8 (Const8 [0]) (Rsh8Ux64 _ (Const64 [c]))) && c > 0 => (ConstBool [true])
554(Leq16 (Const16 [0]) (Rsh16Ux64 _ (Const64 [c]))) && c > 0 => (ConstBool [true])
555(Leq32 (Const32 [0]) (Rsh32Ux64 _ (Const64 [c]))) && c > 0 => (ConstBool [true])
556(Leq64 (Const64 [0]) (Rsh64Ux64 _ (Const64 [c]))) && c > 0 => (ConstBool [true])
557
558// prefer equalities with zero
559(Less(64|32|16|8) (Const(64|32|16|8) <t> [0]) x) && isNonNegative(x) => (Neq(64|32|16|8) (Const(64|32|16|8) <t> [0]) x)
560(Less(64|32|16|8) x (Const(64|32|16|8) <t> [1])) && isNonNegative(x) => (Eq(64|32|16|8) (Const(64|32|16|8) <t> [0]) x)
561(Less(64|32|16|8)U x (Const(64|32|16|8) <t> [1])) => (Eq(64|32|16|8) (Const(64|32|16|8) <t> [0]) x)
562(Leq(64|32|16|8)U (Const(64|32|16|8) <t> [1]) x) => (Neq(64|32|16|8) (Const(64|32|16|8) <t> [0]) x)
563
564// prefer comparisons with zero
565(Less(64|32|16|8) x (Const(64|32|16|8) <t> [1])) => (Leq(64|32|16|8) x (Const(64|32|16|8) <t> [0]))
566(Leq(64|32|16|8) x (Const(64|32|16|8) <t> [-1])) => (Less(64|32|16|8) x (Const(64|32|16|8) <t> [0]))
567(Leq(64|32|16|8) (Const(64|32|16|8) <t> [1]) x) => (Less(64|32|16|8) (Const(64|32|16|8) <t> [0]) x)
568(Less(64|32|16|8) (Const(64|32|16|8) <t> [-1]) x) => (Leq(64|32|16|8) (Const(64|32|16|8) <t> [0]) x)
569
570// constant floating point comparisons
571(Eq32F (Const32F [c]) (Const32F [d])) => (ConstBool [c == d])
572(Eq64F (Const64F [c]) (Const64F [d])) => (ConstBool [c == d])
573(Neq32F (Const32F [c]) (Const32F [d])) => (ConstBool [c != d])
574(Neq64F (Const64F [c]) (Const64F [d])) => (ConstBool [c != d])
575(Less32F (Const32F [c]) (Const32F [d])) => (ConstBool [c < d])
576(Less64F (Const64F [c]) (Const64F [d])) => (ConstBool [c < d])
577(Leq32F (Const32F [c]) (Const32F [d])) => (ConstBool [c <= d])
578(Leq64F (Const64F [c]) (Const64F [d])) => (ConstBool [c <= d])
579
580// simplifications
581(Or(64|32|16|8) x x) => x
582(Or(64|32|16|8) (Const(64|32|16|8) [0]) x) => x
583(Or(64|32|16|8) (Const(64|32|16|8) [-1]) _) => (Const(64|32|16|8) [-1])
584(Or(64|32|16|8) (Com(64|32|16|8) x) x) => (Const(64|32|16|8) [-1])
585
586(And(64|32|16|8) x x) => x
587(And(64|32|16|8) (Const(64|32|16|8) [-1]) x) => x
588(And(64|32|16|8) (Const(64|32|16|8) [0]) _) => (Const(64|32|16|8) [0])
589(And(64|32|16|8) (Com(64|32|16|8) x) x) => (Const(64|32|16|8) [0])
590
591(Xor(64|32|16|8) x x) => (Const(64|32|16|8) [0])
592(Xor(64|32|16|8) (Const(64|32|16|8) [0]) x) => x
593(Xor(64|32|16|8) (Com(64|32|16|8) x) x) => (Const(64|32|16|8) [-1])
594
595(Add(64|32|16|8) (Const(64|32|16|8) [0]) x) => x
596(Sub(64|32|16|8) x x) => (Const(64|32|16|8) [0])
597(Mul(64|32|16|8) (Const(64|32|16|8) [0]) _) => (Const(64|32|16|8) [0])
598(Mul(64|32)uover <t> (Const(64|32) [0]) x) => (MakeTuple (Const(64|32) <t.FieldType(0)> [0]) (ConstBool <t.FieldType(1)> [false]))
599
600(Com(64|32|16|8) (Com(64|32|16|8) x)) => x
601(Com(64|32|16|8) (Const(64|32|16|8) [c])) => (Const(64|32|16|8) [^c])
602
603(Neg(64|32|16|8) (Sub(64|32|16|8) x y)) => (Sub(64|32|16|8) y x)
604(Add(64|32|16|8) x (Neg(64|32|16|8) y)) => (Sub(64|32|16|8) x y)
605
606(Xor(64|32|16|8) (Const(64|32|16|8) [-1]) x) => (Com(64|32|16|8) x)
607
608(Sub(64|32|16|8) (Neg(64|32|16|8) x) (Com(64|32|16|8) x)) => (Const(64|32|16|8) [1])
609(Sub(64|32|16|8) (Com(64|32|16|8) x) (Neg(64|32|16|8) x)) => (Const(64|32|16|8) [-1])
610(Add(64|32|16|8) (Com(64|32|16|8) x) x) => (Const(64|32|16|8) [-1])
611
612// Simplification when involving common integer
613// (t + x) - (t + y) == x - y
614// (t + x) - (y + t) == x - y
615// (x + t) - (y + t) == x - y
616// (x + t) - (t + y) == x - y
617// (x - t) + (t + y) == x + y
618// (x - t) + (y + t) == x + y
619(Sub(64|32|16|8) (Add(64|32|16|8) t x) (Add(64|32|16|8) t y)) => (Sub(64|32|16|8) x y)
620(Add(64|32|16|8) (Sub(64|32|16|8) x t) (Add(64|32|16|8) t y)) => (Add(64|32|16|8) x y)
621
622// ^(x-1) == ^x+1 == -x
623(Add(64|32|16|8) (Const(64|32|16|8) [1]) (Com(64|32|16|8) x)) => (Neg(64|32|16|8) x)
624(Com(64|32|16|8) (Add(64|32|16|8) (Const(64|32|16|8) [-1]) x)) => (Neg(64|32|16|8) x)
625
626// -(-x) == x
627(Neg(64|32|16|8) (Neg(64|32|16|8) x)) => x
628
629// -^x == x+1
630(Neg(64|32|16|8) <t> (Com(64|32|16|8) x)) => (Add(64|32|16|8) (Const(64|32|16|8) <t> [1]) x)
631
632(And(64|32|16|8) x (And(64|32|16|8) x y)) => (And(64|32|16|8) x y)
633(Or(64|32|16|8) x (Or(64|32|16|8) x y)) => (Or(64|32|16|8) x y)
634(Xor(64|32|16|8) x (Xor(64|32|16|8) x y)) => y
635
636// Fold comparisons with numeric bounds
637(Less(64|32|16|8)U _ (Const(64|32|16|8) [0])) => (ConstBool [false])
638(Leq(64|32|16|8)U (Const(64|32|16|8) [0]) _) => (ConstBool [true])
639(Less(64|32|16|8)U (Const(64|32|16|8) [-1]) _) => (ConstBool [false])
640(Leq(64|32|16|8)U _ (Const(64|32|16|8) [-1])) => (ConstBool [true])
641(Less64 _ (Const64 [math.MinInt64])) => (ConstBool [false])
642(Less32 _ (Const32 [math.MinInt32])) => (ConstBool [false])
643(Less16 _ (Const16 [math.MinInt16])) => (ConstBool [false])
644(Less8 _ (Const8 [math.MinInt8 ])) => (ConstBool [false])
645(Leq64 (Const64 [math.MinInt64]) _) => (ConstBool [true])
646(Leq32 (Const32 [math.MinInt32]) _) => (ConstBool [true])
647(Leq16 (Const16 [math.MinInt16]) _) => (ConstBool [true])
648(Leq8 (Const8 [math.MinInt8 ]) _) => (ConstBool [true])
649(Less64 (Const64 [math.MaxInt64]) _) => (ConstBool [false])
650(Less32 (Const32 [math.MaxInt32]) _) => (ConstBool [false])
651(Less16 (Const16 [math.MaxInt16]) _) => (ConstBool [false])
652(Less8 (Const8 [math.MaxInt8 ]) _) => (ConstBool [false])
653(Leq64 _ (Const64 [math.MaxInt64])) => (ConstBool [true])
654(Leq32 _ (Const32 [math.MaxInt32])) => (ConstBool [true])
655(Leq16 _ (Const16 [math.MaxInt16])) => (ConstBool [true])
656(Leq8 _ (Const8 [math.MaxInt8 ])) => (ConstBool [true])
657
658// Canonicalize <= on numeric bounds and < near numeric bounds to ==
659(Leq(64|32|16|8)U x c:(Const(64|32|16|8) [0])) => (Eq(64|32|16|8) x c)
660(Leq(64|32|16|8)U c:(Const(64|32|16|8) [-1]) x) => (Eq(64|32|16|8) x c)
661(Less(64|32|16|8)U x (Const(64|32|16|8) <t> [1])) => (Eq(64|32|16|8) x (Const(64|32|16|8) <t> [0]))
662(Less(64|32|16|8)U (Const(64|32|16|8) <t> [-2]) x) => (Eq(64|32|16|8) x (Const(64|32|16|8) <t> [-1]))
663(Leq64 x c:(Const64 [math.MinInt64])) => (Eq64 x c)
664(Leq32 x c:(Const32 [math.MinInt32])) => (Eq32 x c)
665(Leq16 x c:(Const16 [math.MinInt16])) => (Eq16 x c)
666(Leq8 x c:(Const8 [math.MinInt8 ])) => (Eq8 x c)
667(Leq64 c:(Const64 [math.MaxInt64]) x) => (Eq64 x c)
668(Leq32 c:(Const32 [math.MaxInt32]) x) => (Eq32 x c)
669(Leq16 c:(Const16 [math.MaxInt16]) x) => (Eq16 x c)
670(Leq8 c:(Const8 [math.MaxInt8 ]) x) => (Eq8 x c)
671(Less64 x (Const64 <t> [math.MinInt64+1])) => (Eq64 x (Const64 <t> [math.MinInt64]))
672(Less32 x (Const32 <t> [math.MinInt32+1])) => (Eq32 x (Const32 <t> [math.MinInt32]))
673(Less16 x (Const16 <t> [math.MinInt16+1])) => (Eq16 x (Const16 <t> [math.MinInt16]))
674(Less8 x (Const8 <t> [math.MinInt8 +1])) => (Eq8 x (Const8 <t> [math.MinInt8 ]))
675(Less64 (Const64 <t> [math.MaxInt64-1]) x) => (Eq64 x (Const64 <t> [math.MaxInt64]))
676(Less32 (Const32 <t> [math.MaxInt32-1]) x) => (Eq32 x (Const32 <t> [math.MaxInt32]))
677(Less16 (Const16 <t> [math.MaxInt16-1]) x) => (Eq16 x (Const16 <t> [math.MaxInt16]))
678(Less8 (Const8 <t> [math.MaxInt8 -1]) x) => (Eq8 x (Const8 <t> [math.MaxInt8 ]))
679
680// Ands clear bits. Ors set bits.
681// If a subsequent Or will set all the bits
682// that an And cleared, we can skip the And.
683// This happens in bitmasking code like:
684// x &^= 3 << shift // clear two old bits
685// x |= v << shift // set two new bits
686// when shift is a small constant and v ends up a constant 3.
687(Or8 (And8 x (Const8 [c2])) (Const8 <t> [c1])) && ^(c1 | c2) == 0 => (Or8 (Const8 <t> [c1]) x)
688(Or16 (And16 x (Const16 [c2])) (Const16 <t> [c1])) && ^(c1 | c2) == 0 => (Or16 (Const16 <t> [c1]) x)
689(Or32 (And32 x (Const32 [c2])) (Const32 <t> [c1])) && ^(c1 | c2) == 0 => (Or32 (Const32 <t> [c1]) x)
690(Or64 (And64 x (Const64 [c2])) (Const64 <t> [c1])) && ^(c1 | c2) == 0 => (Or64 (Const64 <t> [c1]) x)
691
692(Trunc64to8 (And64 (Const64 [y]) x)) && y&0xFF == 0xFF => (Trunc64to8 x)
693(Trunc64to16 (And64 (Const64 [y]) x)) && y&0xFFFF == 0xFFFF => (Trunc64to16 x)
694(Trunc64to32 (And64 (Const64 [y]) x)) && y&0xFFFFFFFF == 0xFFFFFFFF => (Trunc64to32 x)
695(Trunc32to8 (And32 (Const32 [y]) x)) && y&0xFF == 0xFF => (Trunc32to8 x)
696(Trunc32to16 (And32 (Const32 [y]) x)) && y&0xFFFF == 0xFFFF => (Trunc32to16 x)
697(Trunc16to8 (And16 (Const16 [y]) x)) && y&0xFF == 0xFF => (Trunc16to8 x)
698
699(ZeroExt8to64 (Trunc64to8 x:(Rsh64Ux64 _ (Const64 [s])))) && s >= 56 => x
700(ZeroExt16to64 (Trunc64to16 x:(Rsh64Ux64 _ (Const64 [s])))) && s >= 48 => x
701(ZeroExt32to64 (Trunc64to32 x:(Rsh64Ux64 _ (Const64 [s])))) && s >= 32 => x
702(ZeroExt8to32 (Trunc32to8 x:(Rsh32Ux64 _ (Const64 [s])))) && s >= 24 => x
703(ZeroExt16to32 (Trunc32to16 x:(Rsh32Ux64 _ (Const64 [s])))) && s >= 16 => x
704(ZeroExt8to16 (Trunc16to8 x:(Rsh16Ux64 _ (Const64 [s])))) && s >= 8 => x
705
706(SignExt8to64 (Trunc64to8 x:(Rsh64x64 _ (Const64 [s])))) && s >= 56 => x
707(SignExt16to64 (Trunc64to16 x:(Rsh64x64 _ (Const64 [s])))) && s >= 48 => x
708(SignExt32to64 (Trunc64to32 x:(Rsh64x64 _ (Const64 [s])))) && s >= 32 => x
709(SignExt8to32 (Trunc32to8 x:(Rsh32x64 _ (Const64 [s])))) && s >= 24 => x
710(SignExt16to32 (Trunc32to16 x:(Rsh32x64 _ (Const64 [s])))) && s >= 16 => x
711(SignExt8to16 (Trunc16to8 x:(Rsh16x64 _ (Const64 [s])))) && s >= 8 => x
712
713(Slicemask (Const32 [x])) && x > 0 => (Const32 [-1])
714(Slicemask (Const32 [0])) => (Const32 [0])
715(Slicemask (Const64 [x])) && x > 0 => (Const64 [-1])
716(Slicemask (Const64 [0])) => (Const64 [0])
717
718// simplifications often used for lengths. e.g. len(s[i:i+5])==5
719(Sub(64|32|16|8) (Add(64|32|16|8) x y) x) => y
720(Sub(64|32|16|8) (Add(64|32|16|8) x y) y) => x
721(Sub(64|32|16|8) (Sub(64|32|16|8) x y) x) => (Neg(64|32|16|8) y)
722(Sub(64|32|16|8) x (Add(64|32|16|8) x y)) => (Neg(64|32|16|8) y)
723(Add(64|32|16|8) x (Sub(64|32|16|8) y x)) => y
724(Add(64|32|16|8) x (Add(64|32|16|8) y (Sub(64|32|16|8) z x))) => (Add(64|32|16|8) y z)
725
726// basic phi simplifications
727(Phi (Const8 [c]) (Const8 [c])) => (Const8 [c])
728(Phi (Const16 [c]) (Const16 [c])) => (Const16 [c])
729(Phi (Const32 [c]) (Const32 [c])) => (Const32 [c])
730(Phi (Const64 [c]) (Const64 [c])) => (Const64 [c])
731
732// slice and interface comparisons
733// The frontend ensures that we can only compare against nil,
734// so we need only compare the first word (interface type or slice ptr).
735(EqInter x y) => (EqPtr (ITab x) (ITab y))
736(NeqInter x y) => (NeqPtr (ITab x) (ITab y))
737(EqSlice x y) => (EqPtr (SlicePtr x) (SlicePtr y))
738(NeqSlice x y) => (NeqPtr (SlicePtr x) (SlicePtr y))
739
740// Load of store of same address, with compatibly typed value and same size
741(Load <t1> p1 (Store {t2} p2 x _))
742 && isSamePtr(p1, p2)
743 && copyCompatibleType(t1, x.Type)
744 && t1.Size() == t2.Size()
745 => x
746(Load <t1> p1 (Store {t2} p2 _ (Store {t3} p3 x _)))
747 && isSamePtr(p1, p3)
748 && copyCompatibleType(t1, x.Type)
749 && t1.Size() == t3.Size()
750 && disjoint(p3, t3.Size(), p2, t2.Size())
751 => x
752(Load <t1> p1 (Store {t2} p2 _ (Store {t3} p3 _ (Store {t4} p4 x _))))
753 && isSamePtr(p1, p4)
754 && copyCompatibleType(t1, x.Type)
755 && t1.Size() == t4.Size()
756 && disjoint(p4, t4.Size(), p2, t2.Size())
757 && disjoint(p4, t4.Size(), p3, t3.Size())
758 => x
759(Load <t1> p1 (Store {t2} p2 _ (Store {t3} p3 _ (Store {t4} p4 _ (Store {t5} p5 x _)))))
760 && isSamePtr(p1, p5)
761 && copyCompatibleType(t1, x.Type)
762 && t1.Size() == t5.Size()
763 && disjoint(p5, t5.Size(), p2, t2.Size())
764 && disjoint(p5, t5.Size(), p3, t3.Size())
765 && disjoint(p5, t5.Size(), p4, t4.Size())
766 => x
767
768// Pass constants through math.Float{32,64}bits and math.Float{32,64}frombits
769(Load <t1> p1 (Store {t2} p2 (Const64 [x]) _)) && isSamePtr(p1,p2) && t2.Size() == 8 && is64BitFloat(t1) && !math.IsNaN(math.Float64frombits(uint64(x))) => (Const64F [math.Float64frombits(uint64(x))])
770(Load <t1> p1 (Store {t2} p2 (Const32 [x]) _)) && isSamePtr(p1,p2) && t2.Size() == 4 && is32BitFloat(t1) && !math.IsNaN(float64(math.Float32frombits(uint32(x)))) => (Const32F [math.Float32frombits(uint32(x))])
771(Load <t1> p1 (Store {t2} p2 (Const64F [x]) _)) && isSamePtr(p1,p2) && t2.Size() == 8 && is64BitInt(t1) => (Const64 [int64(math.Float64bits(x))])
772(Load <t1> p1 (Store {t2} p2 (Const32F [x]) _)) && isSamePtr(p1,p2) && t2.Size() == 4 && is32BitInt(t1) => (Const32 [int32(math.Float32bits(x))])
773
774// Float Loads up to Zeros so they can be constant folded.
775(Load <t1> op:(OffPtr [o1] p1)
776 (Store {t2} p2 _
777 mem:(Zero [n] p3 _)))
778 && o1 >= 0 && o1+t1.Size() <= n && isSamePtr(p1, p3)
779 && CanSSA(t1)
780 && disjoint(op, t1.Size(), p2, t2.Size())
781 => @mem.Block (Load <t1> (OffPtr <op.Type> [o1] p3) mem)
782(Load <t1> op:(OffPtr [o1] p1)
783 (Store {t2} p2 _
784 (Store {t3} p3 _
785 mem:(Zero [n] p4 _))))
786 && o1 >= 0 && o1+t1.Size() <= n && isSamePtr(p1, p4)
787 && CanSSA(t1)
788 && disjoint(op, t1.Size(), p2, t2.Size())
789 && disjoint(op, t1.Size(), p3, t3.Size())
790 => @mem.Block (Load <t1> (OffPtr <op.Type> [o1] p4) mem)
791(Load <t1> op:(OffPtr [o1] p1)
792 (Store {t2} p2 _
793 (Store {t3} p3 _
794 (Store {t4} p4 _
795 mem:(Zero [n] p5 _)))))
796 && o1 >= 0 && o1+t1.Size() <= n && isSamePtr(p1, p5)
797 && CanSSA(t1)
798 && disjoint(op, t1.Size(), p2, t2.Size())
799 && disjoint(op, t1.Size(), p3, t3.Size())
800 && disjoint(op, t1.Size(), p4, t4.Size())
801 => @mem.Block (Load <t1> (OffPtr <op.Type> [o1] p5) mem)
802(Load <t1> op:(OffPtr [o1] p1)
803 (Store {t2} p2 _
804 (Store {t3} p3 _
805 (Store {t4} p4 _
806 (Store {t5} p5 _
807 mem:(Zero [n] p6 _))))))
808 && o1 >= 0 && o1+t1.Size() <= n && isSamePtr(p1, p6)
809 && CanSSA(t1)
810 && disjoint(op, t1.Size(), p2, t2.Size())
811 && disjoint(op, t1.Size(), p3, t3.Size())
812 && disjoint(op, t1.Size(), p4, t4.Size())
813 && disjoint(op, t1.Size(), p5, t5.Size())
814 => @mem.Block (Load <t1> (OffPtr <op.Type> [o1] p6) mem)
815
816// Zero to Load forwarding.
817(Load <t1> (OffPtr [o] p1) (Zero [n] p2 _))
818 && t1.IsBoolean()
819 && isSamePtr(p1, p2)
820 && n >= o + 1
821 => (ConstBool [false])
822(Load <t1> (OffPtr [o] p1) (Zero [n] p2 _))
823 && is8BitInt(t1)
824 && isSamePtr(p1, p2)
825 && n >= o + 1
826 => (Const8 [0])
827(Load <t1> (OffPtr [o] p1) (Zero [n] p2 _))
828 && is16BitInt(t1)
829 && isSamePtr(p1, p2)
830 && n >= o + 2
831 => (Const16 [0])
832(Load <t1> (OffPtr [o] p1) (Zero [n] p2 _))
833 && is32BitInt(t1)
834 && isSamePtr(p1, p2)
835 && n >= o + 4
836 => (Const32 [0])
837(Load <t1> (OffPtr [o] p1) (Zero [n] p2 _))
838 && is64BitInt(t1)
839 && isSamePtr(p1, p2)
840 && n >= o + 8
841 => (Const64 [0])
842(Load <t1> (OffPtr [o] p1) (Zero [n] p2 _))
843 && is32BitFloat(t1)
844 && isSamePtr(p1, p2)
845 && n >= o + 4
846 => (Const32F [0])
847(Load <t1> (OffPtr [o] p1) (Zero [n] p2 _))
848 && is64BitFloat(t1)
849 && isSamePtr(p1, p2)
850 && n >= o + 8
851 => (Const64F [0])
852
853// Eliminate stores of values that have just been loaded from the same location.
854// We also handle the common case where there are some intermediate stores.
855(Store {t1} p1 (Load <t2> p2 mem) mem)
856 && isSamePtr(p1, p2)
857 && t2.Size() == t1.Size()
858 => mem
859(Store {t1} p1 (Load <t2> p2 oldmem) mem:(Store {t3} p3 _ oldmem))
860 && isSamePtr(p1, p2)
861 && t2.Size() == t1.Size()
862 && disjoint(p1, t1.Size(), p3, t3.Size())
863 => mem
864(Store {t1} p1 (Load <t2> p2 oldmem) mem:(Store {t3} p3 _ (Store {t4} p4 _ oldmem)))
865 && isSamePtr(p1, p2)
866 && t2.Size() == t1.Size()
867 && disjoint(p1, t1.Size(), p3, t3.Size())
868 && disjoint(p1, t1.Size(), p4, t4.Size())
869 => mem
870(Store {t1} p1 (Load <t2> p2 oldmem) mem:(Store {t3} p3 _ (Store {t4} p4 _ (Store {t5} p5 _ oldmem))))
871 && isSamePtr(p1, p2)
872 && t2.Size() == t1.Size()
873 && disjoint(p1, t1.Size(), p3, t3.Size())
874 && disjoint(p1, t1.Size(), p4, t4.Size())
875 && disjoint(p1, t1.Size(), p5, t5.Size())
876 => mem
877
878// Don't Store zeros to cleared variables.
879(Store {t} (OffPtr [o] p1) x mem:(Zero [n] p2 _))
880 && isConstZero(x)
881 && o >= 0 && t.Size() + o <= n && isSamePtr(p1, p2)
882 => mem
883(Store {t1} op:(OffPtr [o1] p1) x mem:(Store {t2} p2 _ (Zero [n] p3 _)))
884 && isConstZero(x)
885 && o1 >= 0 && t1.Size() + o1 <= n && isSamePtr(p1, p3)
886 && disjoint(op, t1.Size(), p2, t2.Size())
887 => mem
888(Store {t1} op:(OffPtr [o1] p1) x mem:(Store {t2} p2 _ (Store {t3} p3 _ (Zero [n] p4 _))))
889 && isConstZero(x)
890 && o1 >= 0 && t1.Size() + o1 <= n && isSamePtr(p1, p4)
891 && disjoint(op, t1.Size(), p2, t2.Size())
892 && disjoint(op, t1.Size(), p3, t3.Size())
893 => mem
894(Store {t1} op:(OffPtr [o1] p1) x mem:(Store {t2} p2 _ (Store {t3} p3 _ (Store {t4} p4 _ (Zero [n] p5 _)))))
895 && isConstZero(x)
896 && o1 >= 0 && t1.Size() + o1 <= n && isSamePtr(p1, p5)
897 && disjoint(op, t1.Size(), p2, t2.Size())
898 && disjoint(op, t1.Size(), p3, t3.Size())
899 && disjoint(op, t1.Size(), p4, t4.Size())
900 => mem
901
902// Collapse OffPtr
903(OffPtr (OffPtr p [y]) [x]) => (OffPtr p [x+y])
904(OffPtr p [0]) && v.Type.Compare(p.Type) == types.CMPeq => p
905
906// indexing operations
907// Note: bounds check has already been done
908(PtrIndex <t> ptr idx) && config.PtrSize == 4 && is32Bit(t.Elem().Size()) => (AddPtr ptr (Mul32 <typ.Int> idx (Const32 <typ.Int> [int32(t.Elem().Size())])))
909(PtrIndex <t> ptr idx) && config.PtrSize == 8 => (AddPtr ptr (Mul64 <typ.Int> idx (Const64 <typ.Int> [t.Elem().Size()])))
910
911// struct operations
912(StructSelect [i] x:(StructMake ___)) => x.Args[i]
913(Load <t> _ _) && t.IsStruct() && CanSSA(t) => rewriteStructLoad(v)
914(Store _ (StructMake ___) _) => rewriteStructStore(v)
915
916(StructSelect [i] x:(Load <t> ptr mem)) && !CanSSA(t) =>
917 @x.Block (Load <v.Type> (OffPtr <v.Type.PtrTo()> [t.FieldOff(int(i))] ptr) mem)
918
919// Putting struct{*byte} and similar into direct interfaces.
920(IMake _typ (StructMake val)) => (IMake _typ val)
921(StructSelect [0] (IData x)) => (IData x)
922
923// un-SSAable values use mem->mem copies
924(Store {t} dst (Load src mem) mem) && !CanSSA(t) =>
925 (Move {t} [t.Size()] dst src mem)
926(Store {t} dst (Load src mem) (VarDef {x} mem)) && !CanSSA(t) =>
927 (Move {t} [t.Size()] dst src (VarDef {x} mem))
928
929// array ops
930(ArraySelect (ArrayMake1 x)) => x
931
932(Load <t> _ _) && t.IsArray() && t.NumElem() == 0 =>
933 (ArrayMake0)
934
935(Load <t> ptr mem) && t.IsArray() && t.NumElem() == 1 && CanSSA(t) =>
936 (ArrayMake1 (Load <t.Elem()> ptr mem))
937
938(Store _ (ArrayMake0) mem) => mem
939(Store dst (ArrayMake1 e) mem) => (Store {e.Type} dst e mem)
940
941// Putting [1]*byte and similar into direct interfaces.
942(IMake _typ (ArrayMake1 val)) => (IMake _typ val)
943(ArraySelect [0] (IData x)) => (IData x)
944
945// string ops
946// Decomposing StringMake and lowering of StringPtr and StringLen
947// happens in a later pass, dec, so that these operations are available
948// to other passes for optimizations.
949(StringPtr (StringMake (Addr <t> {s} base) _)) => (Addr <t> {s} base)
950(StringLen (StringMake _ (Const64 <t> [c]))) => (Const64 <t> [c])
951(ConstString {str}) && config.PtrSize == 4 && str == "" =>
952 (StringMake (ConstNil) (Const32 <typ.Int> [0]))
953(ConstString {str}) && config.PtrSize == 8 && str == "" =>
954 (StringMake (ConstNil) (Const64 <typ.Int> [0]))
955(ConstString {str}) && config.PtrSize == 4 && str != "" =>
956 (StringMake
957 (Addr <typ.BytePtr> {fe.StringData(str)}
958 (SB))
959 (Const32 <typ.Int> [int32(len(str))]))
960(ConstString {str}) && config.PtrSize == 8 && str != "" =>
961 (StringMake
962 (Addr <typ.BytePtr> {fe.StringData(str)}
963 (SB))
964 (Const64 <typ.Int> [int64(len(str))]))
965
966// slice ops
967// Only a few slice rules are provided here. See dec.rules for
968// a more comprehensive set.
969(SliceLen (SliceMake _ (Const64 <t> [c]) _)) => (Const64 <t> [c])
970(SliceCap (SliceMake _ _ (Const64 <t> [c]))) => (Const64 <t> [c])
971(SliceLen (SliceMake _ (Const32 <t> [c]) _)) => (Const32 <t> [c])
972(SliceCap (SliceMake _ _ (Const32 <t> [c]))) => (Const32 <t> [c])
973(SlicePtr (SliceMake (SlicePtr x) _ _)) => (SlicePtr x)
974(SliceLen (SliceMake _ (SliceLen x) _)) => (SliceLen x)
975(SliceCap (SliceMake _ _ (SliceCap x))) => (SliceCap x)
976(SliceCap (SliceMake _ _ (SliceLen x))) => (SliceLen x)
977(ConstSlice) && config.PtrSize == 4 =>
978 (SliceMake
979 (ConstNil <v.Type.Elem().PtrTo()>)
980 (Const32 <typ.Int> [0])
981 (Const32 <typ.Int> [0]))
982(ConstSlice) && config.PtrSize == 8 =>
983 (SliceMake
984 (ConstNil <v.Type.Elem().PtrTo()>)
985 (Const64 <typ.Int> [0])
986 (Const64 <typ.Int> [0]))
987
988// interface ops
989(ConstInterface) =>
990 (IMake
991 (ConstNil <typ.Uintptr>)
992 (ConstNil <typ.BytePtr>))
993
994(NilCheck ptr:(GetG mem) mem) => ptr
995
996(If (Not cond) yes no) => (If cond no yes)
997(If (ConstBool [c]) yes no) && c => (First yes no)
998(If (ConstBool [c]) yes no) && !c => (First no yes)
999
1000(Phi <t> nx:(Not x) ny:(Not y)) && nx.Uses == 1 && ny.Uses == 1 => (Not (Phi <t> x y))
1001
1002// Get rid of Convert ops for pointer arithmetic on unsafe.Pointer.
1003(Convert (Add(64|32) (Convert ptr mem) off) mem) => (AddPtr ptr off)
1004(Convert (Convert ptr mem) mem) => ptr
1005// Note: it is important that the target rewrite is ptr+(off1+off2), not (ptr+off1)+off2.
1006// We must ensure that no intermediate computations are invalid pointers.
1007(Convert a:(Add(64|32) (Add(64|32) (Convert ptr mem) off1) off2) mem) => (AddPtr ptr (Add(64|32) <a.Type> off1 off2))
1008
1009// strength reduction of divide by a constant.
1010// See ../magic.go for a detailed description of these algorithms.
1011
1012// Unsigned divide by power of 2. Strength reduce to a shift.
1013(Div8u n (Const8 [c])) && isPowerOfTwo(c) => (Rsh8Ux64 n (Const64 <typ.UInt64> [log8(c)]))
1014(Div16u n (Const16 [c])) && isPowerOfTwo(c) => (Rsh16Ux64 n (Const64 <typ.UInt64> [log16(c)]))
1015(Div32u n (Const32 [c])) && isPowerOfTwo(c) => (Rsh32Ux64 n (Const64 <typ.UInt64> [log32(c)]))
1016(Div64u n (Const64 [c])) && isPowerOfTwo(c) => (Rsh64Ux64 n (Const64 <typ.UInt64> [log64(c)]))
1017(Div64u n (Const64 [-1<<63])) => (Rsh64Ux64 n (Const64 <typ.UInt64> [63]))
1018
1019// Signed non-negative divide by power of 2.
1020(Div8 n (Const8 [c])) && isNonNegative(n) && isPowerOfTwo(c) => (Rsh8Ux64 n (Const64 <typ.UInt64> [log8(c)]))
1021(Div16 n (Const16 [c])) && isNonNegative(n) && isPowerOfTwo(c) => (Rsh16Ux64 n (Const64 <typ.UInt64> [log16(c)]))
1022(Div32 n (Const32 [c])) && isNonNegative(n) && isPowerOfTwo(c) => (Rsh32Ux64 n (Const64 <typ.UInt64> [log32(c)]))
1023(Div64 n (Const64 [c])) && isNonNegative(n) && isPowerOfTwo(c) => (Rsh64Ux64 n (Const64 <typ.UInt64> [log64(c)]))
1024(Div64 n (Const64 [-1<<63])) && isNonNegative(n) => (Const64 [0])
1025
1026// Unsigned divide, not a power of 2. Strength reduce to a multiply.
1027// For 8-bit divides, we just do a direct 9-bit by 8-bit multiply.
1028(Div8u x (Const8 [c])) && umagicOK8(c) =>
1029 (Trunc32to8
1030 (Rsh32Ux64 <typ.UInt32>
1031 (Mul32 <typ.UInt32>
1032 (Const32 <typ.UInt32> [int32(1<<8+umagic8(c).m)])
1033 (ZeroExt8to32 x))
1034 (Const64 <typ.UInt64> [8+umagic8(c).s])))
1035
1036// For 16-bit divides on 64-bit machines, we do a direct 17-bit by 16-bit multiply.
1037(Div16u x (Const16 [c])) && umagicOK16(c) && config.RegSize == 8 =>
1038 (Trunc64to16
1039 (Rsh64Ux64 <typ.UInt64>
1040 (Mul64 <typ.UInt64>
1041 (Const64 <typ.UInt64> [int64(1<<16+umagic16(c).m)])
1042 (ZeroExt16to64 x))
1043 (Const64 <typ.UInt64> [16+umagic16(c).s])))
1044
1045// For 16-bit divides on 32-bit machines
1046(Div16u x (Const16 [c])) && umagicOK16(c) && config.RegSize == 4 && umagic16(c).m&1 == 0 =>
1047 (Trunc32to16
1048 (Rsh32Ux64 <typ.UInt32>
1049 (Mul32 <typ.UInt32>
1050 (Const32 <typ.UInt32> [int32(1<<15+umagic16(c).m/2)])
1051 (ZeroExt16to32 x))
1052 (Const64 <typ.UInt64> [16+umagic16(c).s-1])))
1053(Div16u x (Const16 [c])) && umagicOK16(c) && config.RegSize == 4 && c&1 == 0 =>
1054 (Trunc32to16
1055 (Rsh32Ux64 <typ.UInt32>
1056 (Mul32 <typ.UInt32>
1057 (Const32 <typ.UInt32> [int32(1<<15+(umagic16(c).m+1)/2)])
1058 (Rsh32Ux64 <typ.UInt32> (ZeroExt16to32 x) (Const64 <typ.UInt64> [1])))
1059 (Const64 <typ.UInt64> [16+umagic16(c).s-2])))
1060(Div16u x (Const16 [c])) && umagicOK16(c) && config.RegSize == 4 && config.useAvg =>
1061 (Trunc32to16
1062 (Rsh32Ux64 <typ.UInt32>
1063 (Avg32u
1064 (Lsh32x64 <typ.UInt32> (ZeroExt16to32 x) (Const64 <typ.UInt64> [16]))
1065 (Mul32 <typ.UInt32>
1066 (Const32 <typ.UInt32> [int32(umagic16(c).m)])
1067 (ZeroExt16to32 x)))
1068 (Const64 <typ.UInt64> [16+umagic16(c).s-1])))
1069
1070// For 32-bit divides on 32-bit machines
1071(Div32u x (Const32 [c])) && umagicOK32(c) && config.RegSize == 4 && umagic32(c).m&1 == 0 && config.useHmul =>
1072 (Rsh32Ux64 <typ.UInt32>
1073 (Hmul32u <typ.UInt32>
1074 (Const32 <typ.UInt32> [int32(1<<31+umagic32(c).m/2)])
1075 x)
1076 (Const64 <typ.UInt64> [umagic32(c).s-1]))
1077(Div32u x (Const32 [c])) && umagicOK32(c) && config.RegSize == 4 && c&1 == 0 && config.useHmul =>
1078 (Rsh32Ux64 <typ.UInt32>
1079 (Hmul32u <typ.UInt32>
1080 (Const32 <typ.UInt32> [int32(1<<31+(umagic32(c).m+1)/2)])
1081 (Rsh32Ux64 <typ.UInt32> x (Const64 <typ.UInt64> [1])))
1082 (Const64 <typ.UInt64> [umagic32(c).s-2]))
1083(Div32u x (Const32 [c])) && umagicOK32(c) && config.RegSize == 4 && config.useAvg && config.useHmul =>
1084 (Rsh32Ux64 <typ.UInt32>
1085 (Avg32u
1086 x
1087 (Hmul32u <typ.UInt32>
1088 (Const32 <typ.UInt32> [int32(umagic32(c).m)])
1089 x))
1090 (Const64 <typ.UInt64> [umagic32(c).s-1]))
1091
1092// For 32-bit divides on 64-bit machines
1093// We'll use a regular (non-hi) multiply for this case.
1094(Div32u x (Const32 [c])) && umagicOK32(c) && config.RegSize == 8 && umagic32(c).m&1 == 0 =>
1095 (Trunc64to32
1096 (Rsh64Ux64 <typ.UInt64>
1097 (Mul64 <typ.UInt64>
1098 (Const64 <typ.UInt64> [int64(1<<31+umagic32(c).m/2)])
1099 (ZeroExt32to64 x))
1100 (Const64 <typ.UInt64> [32+umagic32(c).s-1])))
1101(Div32u x (Const32 [c])) && umagicOK32(c) && config.RegSize == 8 && c&1 == 0 =>
1102 (Trunc64to32
1103 (Rsh64Ux64 <typ.UInt64>
1104 (Mul64 <typ.UInt64>
1105 (Const64 <typ.UInt64> [int64(1<<31+(umagic32(c).m+1)/2)])
1106 (Rsh64Ux64 <typ.UInt64> (ZeroExt32to64 x) (Const64 <typ.UInt64> [1])))
1107 (Const64 <typ.UInt64> [32+umagic32(c).s-2])))
1108(Div32u x (Const32 [c])) && umagicOK32(c) && config.RegSize == 8 && config.useAvg =>
1109 (Trunc64to32
1110 (Rsh64Ux64 <typ.UInt64>
1111 (Avg64u
1112 (Lsh64x64 <typ.UInt64> (ZeroExt32to64 x) (Const64 <typ.UInt64> [32]))
1113 (Mul64 <typ.UInt64>
1114 (Const64 <typ.UInt32> [int64(umagic32(c).m)])
1115 (ZeroExt32to64 x)))
1116 (Const64 <typ.UInt64> [32+umagic32(c).s-1])))
1117
1118// For unsigned 64-bit divides on 32-bit machines,
1119// if the constant fits in 16 bits (so that the last term
1120// fits in 32 bits), convert to three 32-bit divides by a constant.
1121//
1122// If 1<<32 = Q * c + R
1123// and x = hi << 32 + lo
1124//
1125// Then x = (hi/c*c + hi%c) << 32 + lo
1126// = hi/c*c<<32 + hi%c<<32 + lo
1127// = hi/c*c<<32 + (hi%c)*(Q*c+R) + lo/c*c + lo%c
1128// = hi/c*c<<32 + (hi%c)*Q*c + lo/c*c + (hi%c*R+lo%c)
1129// and x / c = (hi/c)<<32 + (hi%c)*Q + lo/c + (hi%c*R+lo%c)/c
1130(Div64u x (Const64 [c])) && c > 0 && c <= 0xFFFF && umagicOK32(int32(c)) && config.RegSize == 4 && config.useHmul =>
1131 (Add64
1132 (Add64 <typ.UInt64>
1133 (Add64 <typ.UInt64>
1134 (Lsh64x64 <typ.UInt64>
1135 (ZeroExt32to64
1136 (Div32u <typ.UInt32>
1137 (Trunc64to32 <typ.UInt32> (Rsh64Ux64 <typ.UInt64> x (Const64 <typ.UInt64> [32])))
1138 (Const32 <typ.UInt32> [int32(c)])))
1139 (Const64 <typ.UInt64> [32]))
1140 (ZeroExt32to64 (Div32u <typ.UInt32> (Trunc64to32 <typ.UInt32> x) (Const32 <typ.UInt32> [int32(c)]))))
1141 (Mul64 <typ.UInt64>
1142 (ZeroExt32to64 <typ.UInt64>
1143 (Mod32u <typ.UInt32>
1144 (Trunc64to32 <typ.UInt32> (Rsh64Ux64 <typ.UInt64> x (Const64 <typ.UInt64> [32])))
1145 (Const32 <typ.UInt32> [int32(c)])))
1146 (Const64 <typ.UInt64> [int64((1<<32)/c)])))
1147 (ZeroExt32to64
1148 (Div32u <typ.UInt32>
1149 (Add32 <typ.UInt32>
1150 (Mod32u <typ.UInt32> (Trunc64to32 <typ.UInt32> x) (Const32 <typ.UInt32> [int32(c)]))
1151 (Mul32 <typ.UInt32>
1152 (Mod32u <typ.UInt32>
1153 (Trunc64to32 <typ.UInt32> (Rsh64Ux64 <typ.UInt64> x (Const64 <typ.UInt64> [32])))
1154 (Const32 <typ.UInt32> [int32(c)]))
1155 (Const32 <typ.UInt32> [int32((1<<32)%c)])))
1156 (Const32 <typ.UInt32> [int32(c)]))))
1157
1158// For 64-bit divides on 64-bit machines
1159// (64-bit divides on 32-bit machines are lowered to a runtime call by the walk pass.)
1160(Div64u x (Const64 [c])) && umagicOK64(c) && config.RegSize == 8 && umagic64(c).m&1 == 0 && config.useHmul =>
1161 (Rsh64Ux64 <typ.UInt64>
1162 (Hmul64u <typ.UInt64>
1163 (Const64 <typ.UInt64> [int64(1<<63+umagic64(c).m/2)])
1164 x)
1165 (Const64 <typ.UInt64> [umagic64(c).s-1]))
1166(Div64u x (Const64 [c])) && umagicOK64(c) && config.RegSize == 8 && c&1 == 0 && config.useHmul =>
1167 (Rsh64Ux64 <typ.UInt64>
1168 (Hmul64u <typ.UInt64>
1169 (Const64 <typ.UInt64> [int64(1<<63+(umagic64(c).m+1)/2)])
1170 (Rsh64Ux64 <typ.UInt64> x (Const64 <typ.UInt64> [1])))
1171 (Const64 <typ.UInt64> [umagic64(c).s-2]))
1172(Div64u x (Const64 [c])) && umagicOK64(c) && config.RegSize == 8 && config.useAvg && config.useHmul =>
1173 (Rsh64Ux64 <typ.UInt64>
1174 (Avg64u
1175 x
1176 (Hmul64u <typ.UInt64>
1177 (Const64 <typ.UInt64> [int64(umagic64(c).m)])
1178 x))
1179 (Const64 <typ.UInt64> [umagic64(c).s-1]))
1180
1181// Signed divide by a negative constant. Rewrite to divide by a positive constant.
1182(Div8 <t> n (Const8 [c])) && c < 0 && c != -1<<7 => (Neg8 (Div8 <t> n (Const8 <t> [-c])))
1183(Div16 <t> n (Const16 [c])) && c < 0 && c != -1<<15 => (Neg16 (Div16 <t> n (Const16 <t> [-c])))
1184(Div32 <t> n (Const32 [c])) && c < 0 && c != -1<<31 => (Neg32 (Div32 <t> n (Const32 <t> [-c])))
1185(Div64 <t> n (Const64 [c])) && c < 0 && c != -1<<63 => (Neg64 (Div64 <t> n (Const64 <t> [-c])))
1186
1187// Dividing by the most-negative number. Result is always 0 except
1188// if the input is also the most-negative number.
1189// We can detect that using the sign bit of x & -x.
1190(Div8 <t> x (Const8 [-1<<7 ])) => (Rsh8Ux64 (And8 <t> x (Neg8 <t> x)) (Const64 <typ.UInt64> [7 ]))
1191(Div16 <t> x (Const16 [-1<<15])) => (Rsh16Ux64 (And16 <t> x (Neg16 <t> x)) (Const64 <typ.UInt64> [15]))
1192(Div32 <t> x (Const32 [-1<<31])) => (Rsh32Ux64 (And32 <t> x (Neg32 <t> x)) (Const64 <typ.UInt64> [31]))
1193(Div64 <t> x (Const64 [-1<<63])) => (Rsh64Ux64 (And64 <t> x (Neg64 <t> x)) (Const64 <typ.UInt64> [63]))
1194
1195// Signed divide by power of 2.
1196// n / c = n >> log(c) if n >= 0
1197// = (n+c-1) >> log(c) if n < 0
1198// We conditionally add c-1 by adding n>>63>>(64-log(c)) (first shift signed, second shift unsigned).
1199(Div8 <t> n (Const8 [c])) && isPowerOfTwo(c) =>
1200 (Rsh8x64
1201 (Add8 <t> n (Rsh8Ux64 <t> (Rsh8x64 <t> n (Const64 <typ.UInt64> [ 7])) (Const64 <typ.UInt64> [int64( 8-log8(c))])))
1202 (Const64 <typ.UInt64> [int64(log8(c))]))
1203(Div16 <t> n (Const16 [c])) && isPowerOfTwo(c) =>
1204 (Rsh16x64
1205 (Add16 <t> n (Rsh16Ux64 <t> (Rsh16x64 <t> n (Const64 <typ.UInt64> [15])) (Const64 <typ.UInt64> [int64(16-log16(c))])))
1206 (Const64 <typ.UInt64> [int64(log16(c))]))
1207(Div32 <t> n (Const32 [c])) && isPowerOfTwo(c) =>
1208 (Rsh32x64
1209 (Add32 <t> n (Rsh32Ux64 <t> (Rsh32x64 <t> n (Const64 <typ.UInt64> [31])) (Const64 <typ.UInt64> [int64(32-log32(c))])))
1210 (Const64 <typ.UInt64> [int64(log32(c))]))
1211(Div64 <t> n (Const64 [c])) && isPowerOfTwo(c) =>
1212 (Rsh64x64
1213 (Add64 <t> n (Rsh64Ux64 <t> (Rsh64x64 <t> n (Const64 <typ.UInt64> [63])) (Const64 <typ.UInt64> [int64(64-log64(c))])))
1214 (Const64 <typ.UInt64> [int64(log64(c))]))
1215
1216// Signed divide, not a power of 2. Strength reduce to a multiply.
1217(Div8 <t> x (Const8 [c])) && smagicOK8(c) =>
1218 (Sub8 <t>
1219 (Rsh32x64 <t>
1220 (Mul32 <typ.UInt32>
1221 (Const32 <typ.UInt32> [int32(smagic8(c).m)])
1222 (SignExt8to32 x))
1223 (Const64 <typ.UInt64> [8+smagic8(c).s]))
1224 (Rsh32x64 <t>
1225 (SignExt8to32 x)
1226 (Const64 <typ.UInt64> [31])))
1227(Div16 <t> x (Const16 [c])) && smagicOK16(c) =>
1228 (Sub16 <t>
1229 (Rsh32x64 <t>
1230 (Mul32 <typ.UInt32>
1231 (Const32 <typ.UInt32> [int32(smagic16(c).m)])
1232 (SignExt16to32 x))
1233 (Const64 <typ.UInt64> [16+smagic16(c).s]))
1234 (Rsh32x64 <t>
1235 (SignExt16to32 x)
1236 (Const64 <typ.UInt64> [31])))
1237(Div32 <t> x (Const32 [c])) && smagicOK32(c) && config.RegSize == 8 =>
1238 (Sub32 <t>
1239 (Rsh64x64 <t>
1240 (Mul64 <typ.UInt64>
1241 (Const64 <typ.UInt64> [int64(smagic32(c).m)])
1242 (SignExt32to64 x))
1243 (Const64 <typ.UInt64> [32+smagic32(c).s]))
1244 (Rsh64x64 <t>
1245 (SignExt32to64 x)
1246 (Const64 <typ.UInt64> [63])))
1247(Div32 <t> x (Const32 [c])) && smagicOK32(c) && config.RegSize == 4 && smagic32(c).m&1 == 0 && config.useHmul =>
1248 (Sub32 <t>
1249 (Rsh32x64 <t>
1250 (Hmul32 <t>
1251 (Const32 <typ.UInt32> [int32(smagic32(c).m/2)])
1252 x)
1253 (Const64 <typ.UInt64> [smagic32(c).s-1]))
1254 (Rsh32x64 <t>
1255 x
1256 (Const64 <typ.UInt64> [31])))
1257(Div32 <t> x (Const32 [c])) && smagicOK32(c) && config.RegSize == 4 && smagic32(c).m&1 != 0 && config.useHmul =>
1258 (Sub32 <t>
1259 (Rsh32x64 <t>
1260 (Add32 <t>
1261 (Hmul32 <t>
1262 (Const32 <typ.UInt32> [int32(smagic32(c).m)])
1263 x)
1264 x)
1265 (Const64 <typ.UInt64> [smagic32(c).s]))
1266 (Rsh32x64 <t>
1267 x
1268 (Const64 <typ.UInt64> [31])))
1269(Div64 <t> x (Const64 [c])) && smagicOK64(c) && smagic64(c).m&1 == 0 && config.useHmul =>
1270 (Sub64 <t>
1271 (Rsh64x64 <t>
1272 (Hmul64 <t>
1273 (Const64 <typ.UInt64> [int64(smagic64(c).m/2)])
1274 x)
1275 (Const64 <typ.UInt64> [smagic64(c).s-1]))
1276 (Rsh64x64 <t>
1277 x
1278 (Const64 <typ.UInt64> [63])))
1279(Div64 <t> x (Const64 [c])) && smagicOK64(c) && smagic64(c).m&1 != 0 && config.useHmul =>
1280 (Sub64 <t>
1281 (Rsh64x64 <t>
1282 (Add64 <t>
1283 (Hmul64 <t>
1284 (Const64 <typ.UInt64> [int64(smagic64(c).m)])
1285 x)
1286 x)
1287 (Const64 <typ.UInt64> [smagic64(c).s]))
1288 (Rsh64x64 <t>
1289 x
1290 (Const64 <typ.UInt64> [63])))
1291
1292// Unsigned mod by power of 2 constant.
1293(Mod8u <t> n (Const8 [c])) && isPowerOfTwo(c) => (And8 n (Const8 <t> [c-1]))
1294(Mod16u <t> n (Const16 [c])) && isPowerOfTwo(c) => (And16 n (Const16 <t> [c-1]))
1295(Mod32u <t> n (Const32 [c])) && isPowerOfTwo(c) => (And32 n (Const32 <t> [c-1]))
1296(Mod64u <t> n (Const64 [c])) && isPowerOfTwo(c) => (And64 n (Const64 <t> [c-1]))
1297(Mod64u <t> n (Const64 [-1<<63])) => (And64 n (Const64 <t> [1<<63-1]))
1298
1299// Signed non-negative mod by power of 2 constant.
1300(Mod8 <t> n (Const8 [c])) && isNonNegative(n) && isPowerOfTwo(c) => (And8 n (Const8 <t> [c-1]))
1301(Mod16 <t> n (Const16 [c])) && isNonNegative(n) && isPowerOfTwo(c) => (And16 n (Const16 <t> [c-1]))
1302(Mod32 <t> n (Const32 [c])) && isNonNegative(n) && isPowerOfTwo(c) => (And32 n (Const32 <t> [c-1]))
1303(Mod64 <t> n (Const64 [c])) && isNonNegative(n) && isPowerOfTwo(c) => (And64 n (Const64 <t> [c-1]))
1304(Mod64 n (Const64 [-1<<63])) && isNonNegative(n) => n
1305
1306// Signed mod by negative constant.
1307(Mod8 <t> n (Const8 [c])) && c < 0 && c != -1<<7 => (Mod8 <t> n (Const8 <t> [-c]))
1308(Mod16 <t> n (Const16 [c])) && c < 0 && c != -1<<15 => (Mod16 <t> n (Const16 <t> [-c]))
1309(Mod32 <t> n (Const32 [c])) && c < 0 && c != -1<<31 => (Mod32 <t> n (Const32 <t> [-c]))
1310(Mod64 <t> n (Const64 [c])) && c < 0 && c != -1<<63 => (Mod64 <t> n (Const64 <t> [-c]))
1311
1312// All other mods by constants, do A%B = A-(A/B*B).
1313// This implements % with two * and a bunch of ancillary ops.
1314// One of the * is free if the user's code also computes A/B.
1315(Mod8 <t> x (Const8 [c])) && x.Op != OpConst8 && (c > 0 || c == -1<<7)
1316 => (Sub8 x (Mul8 <t> (Div8 <t> x (Const8 <t> [c])) (Const8 <t> [c])))
1317(Mod16 <t> x (Const16 [c])) && x.Op != OpConst16 && (c > 0 || c == -1<<15)
1318 => (Sub16 x (Mul16 <t> (Div16 <t> x (Const16 <t> [c])) (Const16 <t> [c])))
1319(Mod32 <t> x (Const32 [c])) && x.Op != OpConst32 && (c > 0 || c == -1<<31)
1320 => (Sub32 x (Mul32 <t> (Div32 <t> x (Const32 <t> [c])) (Const32 <t> [c])))
1321(Mod64 <t> x (Const64 [c])) && x.Op != OpConst64 && (c > 0 || c == -1<<63)
1322 => (Sub64 x (Mul64 <t> (Div64 <t> x (Const64 <t> [c])) (Const64 <t> [c])))
1323(Mod8u <t> x (Const8 [c])) && x.Op != OpConst8 && c > 0 && umagicOK8( c)
1324 => (Sub8 x (Mul8 <t> (Div8u <t> x (Const8 <t> [c])) (Const8 <t> [c])))
1325(Mod16u <t> x (Const16 [c])) && x.Op != OpConst16 && c > 0 && umagicOK16(c)
1326 => (Sub16 x (Mul16 <t> (Div16u <t> x (Const16 <t> [c])) (Const16 <t> [c])))
1327(Mod32u <t> x (Const32 [c])) && x.Op != OpConst32 && c > 0 && umagicOK32(c)
1328 => (Sub32 x (Mul32 <t> (Div32u <t> x (Const32 <t> [c])) (Const32 <t> [c])))
1329(Mod64u <t> x (Const64 [c])) && x.Op != OpConst64 && c > 0 && umagicOK64(c)
1330 => (Sub64 x (Mul64 <t> (Div64u <t> x (Const64 <t> [c])) (Const64 <t> [c])))
1331
1332// For architectures without rotates on less than 32-bits, promote these checks to 32-bit.
1333(Eq8 (Mod8u x (Const8 [c])) (Const8 [0])) && x.Op != OpConst8 && udivisibleOK8(c) && !hasSmallRotate(config) =>
1334 (Eq32 (Mod32u <typ.UInt32> (ZeroExt8to32 <typ.UInt32> x) (Const32 <typ.UInt32> [int32(uint8(c))])) (Const32 <typ.UInt32> [0]))
1335(Eq16 (Mod16u x (Const16 [c])) (Const16 [0])) && x.Op != OpConst16 && udivisibleOK16(c) && !hasSmallRotate(config) =>
1336 (Eq32 (Mod32u <typ.UInt32> (ZeroExt16to32 <typ.UInt32> x) (Const32 <typ.UInt32> [int32(uint16(c))])) (Const32 <typ.UInt32> [0]))
1337(Eq8 (Mod8 x (Const8 [c])) (Const8 [0])) && x.Op != OpConst8 && sdivisibleOK8(c) && !hasSmallRotate(config) =>
1338 (Eq32 (Mod32 <typ.Int32> (SignExt8to32 <typ.Int32> x) (Const32 <typ.Int32> [int32(c)])) (Const32 <typ.Int32> [0]))
1339(Eq16 (Mod16 x (Const16 [c])) (Const16 [0])) && x.Op != OpConst16 && sdivisibleOK16(c) && !hasSmallRotate(config) =>
1340 (Eq32 (Mod32 <typ.Int32> (SignExt16to32 <typ.Int32> x) (Const32 <typ.Int32> [int32(c)])) (Const32 <typ.Int32> [0]))
1341
1342// Divisibility checks x%c == 0 convert to multiply and rotate.
1343// Note, x%c == 0 is rewritten as x == c*(x/c) during the opt pass
1344// where (x/c) is performed using multiplication with magic constants.
1345// To rewrite x%c == 0 requires pattern matching the rewritten expression
1346// and checking that the division by the same constant wasn't already calculated.
1347// This check is made by counting uses of the magic constant multiplication.
1348// Note that if there were an intermediate opt pass, this rule could be applied
1349// directly on the Div op and magic division rewrites could be delayed to late opt.
1350
1351// Unsigned divisibility checks convert to multiply and rotate.
1352(Eq8 x (Mul8 (Const8 [c])
1353 (Trunc32to8
1354 (Rsh32Ux64
1355 mul:(Mul32
1356 (Const32 [m])
1357 (ZeroExt8to32 x))
1358 (Const64 [s])))
1359 )
1360)
1361 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1362 && m == int32(1<<8+umagic8(c).m) && s == 8+umagic8(c).s
1363 && x.Op != OpConst8 && udivisibleOK8(c)
1364 => (Leq8U
1365 (RotateLeft8 <typ.UInt8>
1366 (Mul8 <typ.UInt8>
1367 (Const8 <typ.UInt8> [int8(udivisible8(c).m)])
1368 x)
1369 (Const8 <typ.UInt8> [int8(8-udivisible8(c).k)])
1370 )
1371 (Const8 <typ.UInt8> [int8(udivisible8(c).max)])
1372 )
1373
1374(Eq16 x (Mul16 (Const16 [c])
1375 (Trunc64to16
1376 (Rsh64Ux64
1377 mul:(Mul64
1378 (Const64 [m])
1379 (ZeroExt16to64 x))
1380 (Const64 [s])))
1381 )
1382)
1383 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1384 && m == int64(1<<16+umagic16(c).m) && s == 16+umagic16(c).s
1385 && x.Op != OpConst16 && udivisibleOK16(c)
1386 => (Leq16U
1387 (RotateLeft16 <typ.UInt16>
1388 (Mul16 <typ.UInt16>
1389 (Const16 <typ.UInt16> [int16(udivisible16(c).m)])
1390 x)
1391 (Const16 <typ.UInt16> [int16(16-udivisible16(c).k)])
1392 )
1393 (Const16 <typ.UInt16> [int16(udivisible16(c).max)])
1394 )
1395
1396(Eq16 x (Mul16 (Const16 [c])
1397 (Trunc32to16
1398 (Rsh32Ux64
1399 mul:(Mul32
1400 (Const32 [m])
1401 (ZeroExt16to32 x))
1402 (Const64 [s])))
1403 )
1404)
1405 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1406 && m == int32(1<<15+umagic16(c).m/2) && s == 16+umagic16(c).s-1
1407 && x.Op != OpConst16 && udivisibleOK16(c)
1408 => (Leq16U
1409 (RotateLeft16 <typ.UInt16>
1410 (Mul16 <typ.UInt16>
1411 (Const16 <typ.UInt16> [int16(udivisible16(c).m)])
1412 x)
1413 (Const16 <typ.UInt16> [int16(16-udivisible16(c).k)])
1414 )
1415 (Const16 <typ.UInt16> [int16(udivisible16(c).max)])
1416 )
1417
1418(Eq16 x (Mul16 (Const16 [c])
1419 (Trunc32to16
1420 (Rsh32Ux64
1421 mul:(Mul32
1422 (Const32 [m])
1423 (Rsh32Ux64 (ZeroExt16to32 x) (Const64 [1])))
1424 (Const64 [s])))
1425 )
1426)
1427 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1428 && m == int32(1<<15+(umagic16(c).m+1)/2) && s == 16+umagic16(c).s-2
1429 && x.Op != OpConst16 && udivisibleOK16(c)
1430 => (Leq16U
1431 (RotateLeft16 <typ.UInt16>
1432 (Mul16 <typ.UInt16>
1433 (Const16 <typ.UInt16> [int16(udivisible16(c).m)])
1434 x)
1435 (Const16 <typ.UInt16> [int16(16-udivisible16(c).k)])
1436 )
1437 (Const16 <typ.UInt16> [int16(udivisible16(c).max)])
1438 )
1439
1440(Eq16 x (Mul16 (Const16 [c])
1441 (Trunc32to16
1442 (Rsh32Ux64
1443 (Avg32u
1444 (Lsh32x64 (ZeroExt16to32 x) (Const64 [16]))
1445 mul:(Mul32
1446 (Const32 [m])
1447 (ZeroExt16to32 x)))
1448 (Const64 [s])))
1449 )
1450)
1451 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1452 && m == int32(umagic16(c).m) && s == 16+umagic16(c).s-1
1453 && x.Op != OpConst16 && udivisibleOK16(c)
1454 => (Leq16U
1455 (RotateLeft16 <typ.UInt16>
1456 (Mul16 <typ.UInt16>
1457 (Const16 <typ.UInt16> [int16(udivisible16(c).m)])
1458 x)
1459 (Const16 <typ.UInt16> [int16(16-udivisible16(c).k)])
1460 )
1461 (Const16 <typ.UInt16> [int16(udivisible16(c).max)])
1462 )
1463
1464(Eq32 x (Mul32 (Const32 [c])
1465 (Rsh32Ux64
1466 mul:(Hmul32u
1467 (Const32 [m])
1468 x)
1469 (Const64 [s]))
1470 )
1471)
1472 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1473 && m == int32(1<<31+umagic32(c).m/2) && s == umagic32(c).s-1
1474 && x.Op != OpConst32 && udivisibleOK32(c)
1475 => (Leq32U
1476 (RotateLeft32 <typ.UInt32>
1477 (Mul32 <typ.UInt32>
1478 (Const32 <typ.UInt32> [int32(udivisible32(c).m)])
1479 x)
1480 (Const32 <typ.UInt32> [int32(32-udivisible32(c).k)])
1481 )
1482 (Const32 <typ.UInt32> [int32(udivisible32(c).max)])
1483 )
1484
1485(Eq32 x (Mul32 (Const32 [c])
1486 (Rsh32Ux64
1487 mul:(Hmul32u
1488 (Const32 <typ.UInt32> [m])
1489 (Rsh32Ux64 x (Const64 [1])))
1490 (Const64 [s]))
1491 )
1492)
1493 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1494 && m == int32(1<<31+(umagic32(c).m+1)/2) && s == umagic32(c).s-2
1495 && x.Op != OpConst32 && udivisibleOK32(c)
1496 => (Leq32U
1497 (RotateLeft32 <typ.UInt32>
1498 (Mul32 <typ.UInt32>
1499 (Const32 <typ.UInt32> [int32(udivisible32(c).m)])
1500 x)
1501 (Const32 <typ.UInt32> [int32(32-udivisible32(c).k)])
1502 )
1503 (Const32 <typ.UInt32> [int32(udivisible32(c).max)])
1504 )
1505
1506(Eq32 x (Mul32 (Const32 [c])
1507 (Rsh32Ux64
1508 (Avg32u
1509 x
1510 mul:(Hmul32u
1511 (Const32 [m])
1512 x))
1513 (Const64 [s]))
1514 )
1515)
1516 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1517 && m == int32(umagic32(c).m) && s == umagic32(c).s-1
1518 && x.Op != OpConst32 && udivisibleOK32(c)
1519 => (Leq32U
1520 (RotateLeft32 <typ.UInt32>
1521 (Mul32 <typ.UInt32>
1522 (Const32 <typ.UInt32> [int32(udivisible32(c).m)])
1523 x)
1524 (Const32 <typ.UInt32> [int32(32-udivisible32(c).k)])
1525 )
1526 (Const32 <typ.UInt32> [int32(udivisible32(c).max)])
1527 )
1528
1529(Eq32 x (Mul32 (Const32 [c])
1530 (Trunc64to32
1531 (Rsh64Ux64
1532 mul:(Mul64
1533 (Const64 [m])
1534 (ZeroExt32to64 x))
1535 (Const64 [s])))
1536 )
1537)
1538 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1539 && m == int64(1<<31+umagic32(c).m/2) && s == 32+umagic32(c).s-1
1540 && x.Op != OpConst32 && udivisibleOK32(c)
1541 => (Leq32U
1542 (RotateLeft32 <typ.UInt32>
1543 (Mul32 <typ.UInt32>
1544 (Const32 <typ.UInt32> [int32(udivisible32(c).m)])
1545 x)
1546 (Const32 <typ.UInt32> [int32(32-udivisible32(c).k)])
1547 )
1548 (Const32 <typ.UInt32> [int32(udivisible32(c).max)])
1549 )
1550
1551(Eq32 x (Mul32 (Const32 [c])
1552 (Trunc64to32
1553 (Rsh64Ux64
1554 mul:(Mul64
1555 (Const64 [m])
1556 (Rsh64Ux64 (ZeroExt32to64 x) (Const64 [1])))
1557 (Const64 [s])))
1558 )
1559)
1560 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1561 && m == int64(1<<31+(umagic32(c).m+1)/2) && s == 32+umagic32(c).s-2
1562 && x.Op != OpConst32 && udivisibleOK32(c)
1563 => (Leq32U
1564 (RotateLeft32 <typ.UInt32>
1565 (Mul32 <typ.UInt32>
1566 (Const32 <typ.UInt32> [int32(udivisible32(c).m)])
1567 x)
1568 (Const32 <typ.UInt32> [int32(32-udivisible32(c).k)])
1569 )
1570 (Const32 <typ.UInt32> [int32(udivisible32(c).max)])
1571 )
1572
1573(Eq32 x (Mul32 (Const32 [c])
1574 (Trunc64to32
1575 (Rsh64Ux64
1576 (Avg64u
1577 (Lsh64x64 (ZeroExt32to64 x) (Const64 [32]))
1578 mul:(Mul64
1579 (Const64 [m])
1580 (ZeroExt32to64 x)))
1581 (Const64 [s])))
1582 )
1583)
1584 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1585 && m == int64(umagic32(c).m) && s == 32+umagic32(c).s-1
1586 && x.Op != OpConst32 && udivisibleOK32(c)
1587 => (Leq32U
1588 (RotateLeft32 <typ.UInt32>
1589 (Mul32 <typ.UInt32>
1590 (Const32 <typ.UInt32> [int32(udivisible32(c).m)])
1591 x)
1592 (Const32 <typ.UInt32> [int32(32-udivisible32(c).k)])
1593 )
1594 (Const32 <typ.UInt32> [int32(udivisible32(c).max)])
1595 )
1596
1597(Eq64 x (Mul64 (Const64 [c])
1598 (Rsh64Ux64
1599 mul:(Hmul64u
1600 (Const64 [m])
1601 x)
1602 (Const64 [s]))
1603 )
1604) && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1605 && m == int64(1<<63+umagic64(c).m/2) && s == umagic64(c).s-1
1606 && x.Op != OpConst64 && udivisibleOK64(c)
1607 => (Leq64U
1608 (RotateLeft64 <typ.UInt64>
1609 (Mul64 <typ.UInt64>
1610 (Const64 <typ.UInt64> [int64(udivisible64(c).m)])
1611 x)
1612 (Const64 <typ.UInt64> [64-udivisible64(c).k])
1613 )
1614 (Const64 <typ.UInt64> [int64(udivisible64(c).max)])
1615 )
1616(Eq64 x (Mul64 (Const64 [c])
1617 (Rsh64Ux64
1618 mul:(Hmul64u
1619 (Const64 [m])
1620 (Rsh64Ux64 x (Const64 [1])))
1621 (Const64 [s]))
1622 )
1623) && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1624 && m == int64(1<<63+(umagic64(c).m+1)/2) && s == umagic64(c).s-2
1625 && x.Op != OpConst64 && udivisibleOK64(c)
1626 => (Leq64U
1627 (RotateLeft64 <typ.UInt64>
1628 (Mul64 <typ.UInt64>
1629 (Const64 <typ.UInt64> [int64(udivisible64(c).m)])
1630 x)
1631 (Const64 <typ.UInt64> [64-udivisible64(c).k])
1632 )
1633 (Const64 <typ.UInt64> [int64(udivisible64(c).max)])
1634 )
1635(Eq64 x (Mul64 (Const64 [c])
1636 (Rsh64Ux64
1637 (Avg64u
1638 x
1639 mul:(Hmul64u
1640 (Const64 [m])
1641 x))
1642 (Const64 [s]))
1643 )
1644) && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1645 && m == int64(umagic64(c).m) && s == umagic64(c).s-1
1646 && x.Op != OpConst64 && udivisibleOK64(c)
1647 => (Leq64U
1648 (RotateLeft64 <typ.UInt64>
1649 (Mul64 <typ.UInt64>
1650 (Const64 <typ.UInt64> [int64(udivisible64(c).m)])
1651 x)
1652 (Const64 <typ.UInt64> [64-udivisible64(c).k])
1653 )
1654 (Const64 <typ.UInt64> [int64(udivisible64(c).max)])
1655 )
1656
1657// Signed divisibility checks convert to multiply, add and rotate.
1658(Eq8 x (Mul8 (Const8 [c])
1659 (Sub8
1660 (Rsh32x64
1661 mul:(Mul32
1662 (Const32 [m])
1663 (SignExt8to32 x))
1664 (Const64 [s]))
1665 (Rsh32x64
1666 (SignExt8to32 x)
1667 (Const64 [31])))
1668 )
1669)
1670 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1671 && m == int32(smagic8(c).m) && s == 8+smagic8(c).s
1672 && x.Op != OpConst8 && sdivisibleOK8(c)
1673 => (Leq8U
1674 (RotateLeft8 <typ.UInt8>
1675 (Add8 <typ.UInt8>
1676 (Mul8 <typ.UInt8>
1677 (Const8 <typ.UInt8> [int8(sdivisible8(c).m)])
1678 x)
1679 (Const8 <typ.UInt8> [int8(sdivisible8(c).a)])
1680 )
1681 (Const8 <typ.UInt8> [int8(8-sdivisible8(c).k)])
1682 )
1683 (Const8 <typ.UInt8> [int8(sdivisible8(c).max)])
1684 )
1685
1686(Eq16 x (Mul16 (Const16 [c])
1687 (Sub16
1688 (Rsh32x64
1689 mul:(Mul32
1690 (Const32 [m])
1691 (SignExt16to32 x))
1692 (Const64 [s]))
1693 (Rsh32x64
1694 (SignExt16to32 x)
1695 (Const64 [31])))
1696 )
1697)
1698 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1699 && m == int32(smagic16(c).m) && s == 16+smagic16(c).s
1700 && x.Op != OpConst16 && sdivisibleOK16(c)
1701 => (Leq16U
1702 (RotateLeft16 <typ.UInt16>
1703 (Add16 <typ.UInt16>
1704 (Mul16 <typ.UInt16>
1705 (Const16 <typ.UInt16> [int16(sdivisible16(c).m)])
1706 x)
1707 (Const16 <typ.UInt16> [int16(sdivisible16(c).a)])
1708 )
1709 (Const16 <typ.UInt16> [int16(16-sdivisible16(c).k)])
1710 )
1711 (Const16 <typ.UInt16> [int16(sdivisible16(c).max)])
1712 )
1713
1714(Eq32 x (Mul32 (Const32 [c])
1715 (Sub32
1716 (Rsh64x64
1717 mul:(Mul64
1718 (Const64 [m])
1719 (SignExt32to64 x))
1720 (Const64 [s]))
1721 (Rsh64x64
1722 (SignExt32to64 x)
1723 (Const64 [63])))
1724 )
1725)
1726 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1727 && m == int64(smagic32(c).m) && s == 32+smagic32(c).s
1728 && x.Op != OpConst32 && sdivisibleOK32(c)
1729 => (Leq32U
1730 (RotateLeft32 <typ.UInt32>
1731 (Add32 <typ.UInt32>
1732 (Mul32 <typ.UInt32>
1733 (Const32 <typ.UInt32> [int32(sdivisible32(c).m)])
1734 x)
1735 (Const32 <typ.UInt32> [int32(sdivisible32(c).a)])
1736 )
1737 (Const32 <typ.UInt32> [int32(32-sdivisible32(c).k)])
1738 )
1739 (Const32 <typ.UInt32> [int32(sdivisible32(c).max)])
1740 )
1741
1742(Eq32 x (Mul32 (Const32 [c])
1743 (Sub32
1744 (Rsh32x64
1745 mul:(Hmul32
1746 (Const32 [m])
1747 x)
1748 (Const64 [s]))
1749 (Rsh32x64
1750 x
1751 (Const64 [31])))
1752 )
1753)
1754 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1755 && m == int32(smagic32(c).m/2) && s == smagic32(c).s-1
1756 && x.Op != OpConst32 && sdivisibleOK32(c)
1757 => (Leq32U
1758 (RotateLeft32 <typ.UInt32>
1759 (Add32 <typ.UInt32>
1760 (Mul32 <typ.UInt32>
1761 (Const32 <typ.UInt32> [int32(sdivisible32(c).m)])
1762 x)
1763 (Const32 <typ.UInt32> [int32(sdivisible32(c).a)])
1764 )
1765 (Const32 <typ.UInt32> [int32(32-sdivisible32(c).k)])
1766 )
1767 (Const32 <typ.UInt32> [int32(sdivisible32(c).max)])
1768 )
1769
1770(Eq32 x (Mul32 (Const32 [c])
1771 (Sub32
1772 (Rsh32x64
1773 (Add32
1774 mul:(Hmul32
1775 (Const32 [m])
1776 x)
1777 x)
1778 (Const64 [s]))
1779 (Rsh32x64
1780 x
1781 (Const64 [31])))
1782 )
1783)
1784 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1785 && m == int32(smagic32(c).m) && s == smagic32(c).s
1786 && x.Op != OpConst32 && sdivisibleOK32(c)
1787 => (Leq32U
1788 (RotateLeft32 <typ.UInt32>
1789 (Add32 <typ.UInt32>
1790 (Mul32 <typ.UInt32>
1791 (Const32 <typ.UInt32> [int32(sdivisible32(c).m)])
1792 x)
1793 (Const32 <typ.UInt32> [int32(sdivisible32(c).a)])
1794 )
1795 (Const32 <typ.UInt32> [int32(32-sdivisible32(c).k)])
1796 )
1797 (Const32 <typ.UInt32> [int32(sdivisible32(c).max)])
1798 )
1799
1800(Eq64 x (Mul64 (Const64 [c])
1801 (Sub64
1802 (Rsh64x64
1803 mul:(Hmul64
1804 (Const64 [m])
1805 x)
1806 (Const64 [s]))
1807 (Rsh64x64
1808 x
1809 (Const64 [63])))
1810 )
1811)
1812 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1813 && m == int64(smagic64(c).m/2) && s == smagic64(c).s-1
1814 && x.Op != OpConst64 && sdivisibleOK64(c)
1815 => (Leq64U
1816 (RotateLeft64 <typ.UInt64>
1817 (Add64 <typ.UInt64>
1818 (Mul64 <typ.UInt64>
1819 (Const64 <typ.UInt64> [int64(sdivisible64(c).m)])
1820 x)
1821 (Const64 <typ.UInt64> [int64(sdivisible64(c).a)])
1822 )
1823 (Const64 <typ.UInt64> [64-sdivisible64(c).k])
1824 )
1825 (Const64 <typ.UInt64> [int64(sdivisible64(c).max)])
1826 )
1827
1828(Eq64 x (Mul64 (Const64 [c])
1829 (Sub64
1830 (Rsh64x64
1831 (Add64
1832 mul:(Hmul64
1833 (Const64 [m])
1834 x)
1835 x)
1836 (Const64 [s]))
1837 (Rsh64x64
1838 x
1839 (Const64 [63])))
1840 )
1841)
1842 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1843 && m == int64(smagic64(c).m) && s == smagic64(c).s
1844 && x.Op != OpConst64 && sdivisibleOK64(c)
1845 => (Leq64U
1846 (RotateLeft64 <typ.UInt64>
1847 (Add64 <typ.UInt64>
1848 (Mul64 <typ.UInt64>
1849 (Const64 <typ.UInt64> [int64(sdivisible64(c).m)])
1850 x)
1851 (Const64 <typ.UInt64> [int64(sdivisible64(c).a)])
1852 )
1853 (Const64 <typ.UInt64> [64-sdivisible64(c).k])
1854 )
1855 (Const64 <typ.UInt64> [int64(sdivisible64(c).max)])
1856 )
1857
1858// Divisibility check for signed integers for power of two constant are simple mask.
1859// However, we must match against the rewritten n%c == 0 -> n - c*(n/c) == 0 -> n == c*(n/c)
1860// where n/c contains fixup code to handle signed n.
1861((Eq8|Neq8) n (Lsh8x64
1862 (Rsh8x64
1863 (Add8 <t> n (Rsh8Ux64 <t> (Rsh8x64 <t> n (Const64 <typ.UInt64> [ 7])) (Const64 <typ.UInt64> [kbar])))
1864 (Const64 <typ.UInt64> [k]))
1865 (Const64 <typ.UInt64> [k]))
1866) && k > 0 && k < 7 && kbar == 8 - k
1867 => ((Eq8|Neq8) (And8 <t> n (Const8 <t> [1<<uint(k)-1])) (Const8 <t> [0]))
1868
1869((Eq16|Neq16) n (Lsh16x64
1870 (Rsh16x64
1871 (Add16 <t> n (Rsh16Ux64 <t> (Rsh16x64 <t> n (Const64 <typ.UInt64> [15])) (Const64 <typ.UInt64> [kbar])))
1872 (Const64 <typ.UInt64> [k]))
1873 (Const64 <typ.UInt64> [k]))
1874) && k > 0 && k < 15 && kbar == 16 - k
1875 => ((Eq16|Neq16) (And16 <t> n (Const16 <t> [1<<uint(k)-1])) (Const16 <t> [0]))
1876
1877((Eq32|Neq32) n (Lsh32x64
1878 (Rsh32x64
1879 (Add32 <t> n (Rsh32Ux64 <t> (Rsh32x64 <t> n (Const64 <typ.UInt64> [31])) (Const64 <typ.UInt64> [kbar])))
1880 (Const64 <typ.UInt64> [k]))
1881 (Const64 <typ.UInt64> [k]))
1882) && k > 0 && k < 31 && kbar == 32 - k
1883 => ((Eq32|Neq32) (And32 <t> n (Const32 <t> [1<<uint(k)-1])) (Const32 <t> [0]))
1884
1885((Eq64|Neq64) n (Lsh64x64
1886 (Rsh64x64
1887 (Add64 <t> n (Rsh64Ux64 <t> (Rsh64x64 <t> n (Const64 <typ.UInt64> [63])) (Const64 <typ.UInt64> [kbar])))
1888 (Const64 <typ.UInt64> [k]))
1889 (Const64 <typ.UInt64> [k]))
1890) && k > 0 && k < 63 && kbar == 64 - k
1891 => ((Eq64|Neq64) (And64 <t> n (Const64 <t> [1<<uint(k)-1])) (Const64 <t> [0]))
1892
1893(Eq(8|16|32|64) s:(Sub(8|16|32|64) x y) (Const(8|16|32|64) [0])) && s.Uses == 1 => (Eq(8|16|32|64) x y)
1894(Neq(8|16|32|64) s:(Sub(8|16|32|64) x y) (Const(8|16|32|64) [0])) && s.Uses == 1 => (Neq(8|16|32|64) x y)
1895
1896// Optimize bitsets
1897(Eq8 (And8 <t> x (Const8 <t> [y])) (Const8 <t> [y])) && oneBit8(y)
1898 => (Neq8 (And8 <t> x (Const8 <t> [y])) (Const8 <t> [0]))
1899(Eq16 (And16 <t> x (Const16 <t> [y])) (Const16 <t> [y])) && oneBit16(y)
1900 => (Neq16 (And16 <t> x (Const16 <t> [y])) (Const16 <t> [0]))
1901(Eq32 (And32 <t> x (Const32 <t> [y])) (Const32 <t> [y])) && oneBit32(y)
1902 => (Neq32 (And32 <t> x (Const32 <t> [y])) (Const32 <t> [0]))
1903(Eq64 (And64 <t> x (Const64 <t> [y])) (Const64 <t> [y])) && oneBit64(y)
1904 => (Neq64 (And64 <t> x (Const64 <t> [y])) (Const64 <t> [0]))
1905(Neq8 (And8 <t> x (Const8 <t> [y])) (Const8 <t> [y])) && oneBit8(y)
1906 => (Eq8 (And8 <t> x (Const8 <t> [y])) (Const8 <t> [0]))
1907(Neq16 (And16 <t> x (Const16 <t> [y])) (Const16 <t> [y])) && oneBit16(y)
1908 => (Eq16 (And16 <t> x (Const16 <t> [y])) (Const16 <t> [0]))
1909(Neq32 (And32 <t> x (Const32 <t> [y])) (Const32 <t> [y])) && oneBit32(y)
1910 => (Eq32 (And32 <t> x (Const32 <t> [y])) (Const32 <t> [0]))
1911(Neq64 (And64 <t> x (Const64 <t> [y])) (Const64 <t> [y])) && oneBit64(y)
1912 => (Eq64 (And64 <t> x (Const64 <t> [y])) (Const64 <t> [0]))
1913
1914// Reassociate expressions involving
1915// constants such that constants come first,
1916// exposing obvious constant-folding opportunities.
1917// Reassociate (op (op y C) x) to (op C (op x y)) or similar, where C
1918// is constant, which pushes constants to the outside
1919// of the expression. At that point, any constant-folding
1920// opportunities should be obvious.
1921// Note: don't include AddPtr here! In order to maintain the
1922// invariant that pointers must stay within the pointed-to object,
1923// we can't pull part of a pointer computation above the AddPtr.
1924// See issue 37881.
1925// Note: we don't need to handle any (x-C) cases because we already rewrite
1926// (x-C) to (x+(-C)).
1927
1928// x + (C + z) -> C + (x + z)
1929(Add64 (Add64 i:(Const64 <t>) z) x) && (z.Op != OpConst64 && x.Op != OpConst64) => (Add64 i (Add64 <t> z x))
1930(Add32 (Add32 i:(Const32 <t>) z) x) && (z.Op != OpConst32 && x.Op != OpConst32) => (Add32 i (Add32 <t> z x))
1931(Add16 (Add16 i:(Const16 <t>) z) x) && (z.Op != OpConst16 && x.Op != OpConst16) => (Add16 i (Add16 <t> z x))
1932(Add8 (Add8 i:(Const8 <t>) z) x) && (z.Op != OpConst8 && x.Op != OpConst8) => (Add8 i (Add8 <t> z x))
1933
1934// x + (C - z) -> C + (x - z)
1935(Add64 (Sub64 i:(Const64 <t>) z) x) && (z.Op != OpConst64 && x.Op != OpConst64) => (Add64 i (Sub64 <t> x z))
1936(Add32 (Sub32 i:(Const32 <t>) z) x) && (z.Op != OpConst32 && x.Op != OpConst32) => (Add32 i (Sub32 <t> x z))
1937(Add16 (Sub16 i:(Const16 <t>) z) x) && (z.Op != OpConst16 && x.Op != OpConst16) => (Add16 i (Sub16 <t> x z))
1938(Add8 (Sub8 i:(Const8 <t>) z) x) && (z.Op != OpConst8 && x.Op != OpConst8) => (Add8 i (Sub8 <t> x z))
1939
1940// x - (C - z) -> x + (z - C) -> (x + z) - C
1941(Sub64 x (Sub64 i:(Const64 <t>) z)) && (z.Op != OpConst64 && x.Op != OpConst64) => (Sub64 (Add64 <t> x z) i)
1942(Sub32 x (Sub32 i:(Const32 <t>) z)) && (z.Op != OpConst32 && x.Op != OpConst32) => (Sub32 (Add32 <t> x z) i)
1943(Sub16 x (Sub16 i:(Const16 <t>) z)) && (z.Op != OpConst16 && x.Op != OpConst16) => (Sub16 (Add16 <t> x z) i)
1944(Sub8 x (Sub8 i:(Const8 <t>) z)) && (z.Op != OpConst8 && x.Op != OpConst8) => (Sub8 (Add8 <t> x z) i)
1945
1946// x - (z + C) -> x + (-z - C) -> (x - z) - C
1947(Sub64 x (Add64 z i:(Const64 <t>))) && (z.Op != OpConst64 && x.Op != OpConst64) => (Sub64 (Sub64 <t> x z) i)
1948(Sub32 x (Add32 z i:(Const32 <t>))) && (z.Op != OpConst32 && x.Op != OpConst32) => (Sub32 (Sub32 <t> x z) i)
1949(Sub16 x (Add16 z i:(Const16 <t>))) && (z.Op != OpConst16 && x.Op != OpConst16) => (Sub16 (Sub16 <t> x z) i)
1950(Sub8 x (Add8 z i:(Const8 <t>))) && (z.Op != OpConst8 && x.Op != OpConst8) => (Sub8 (Sub8 <t> x z) i)
1951
1952// (C - z) - x -> C - (z + x)
1953(Sub64 (Sub64 i:(Const64 <t>) z) x) && (z.Op != OpConst64 && x.Op != OpConst64) => (Sub64 i (Add64 <t> z x))
1954(Sub32 (Sub32 i:(Const32 <t>) z) x) && (z.Op != OpConst32 && x.Op != OpConst32) => (Sub32 i (Add32 <t> z x))
1955(Sub16 (Sub16 i:(Const16 <t>) z) x) && (z.Op != OpConst16 && x.Op != OpConst16) => (Sub16 i (Add16 <t> z x))
1956(Sub8 (Sub8 i:(Const8 <t>) z) x) && (z.Op != OpConst8 && x.Op != OpConst8) => (Sub8 i (Add8 <t> z x))
1957
1958// (z + C) -x -> C + (z - x)
1959(Sub64 (Add64 z i:(Const64 <t>)) x) && (z.Op != OpConst64 && x.Op != OpConst64) => (Add64 i (Sub64 <t> z x))
1960(Sub32 (Add32 z i:(Const32 <t>)) x) && (z.Op != OpConst32 && x.Op != OpConst32) => (Add32 i (Sub32 <t> z x))
1961(Sub16 (Add16 z i:(Const16 <t>)) x) && (z.Op != OpConst16 && x.Op != OpConst16) => (Add16 i (Sub16 <t> z x))
1962(Sub8 (Add8 z i:(Const8 <t>)) x) && (z.Op != OpConst8 && x.Op != OpConst8) => (Add8 i (Sub8 <t> z x))
1963
1964// x & (C & z) -> C & (x & z)
1965(And64 (And64 i:(Const64 <t>) z) x) && (z.Op != OpConst64 && x.Op != OpConst64) => (And64 i (And64 <t> z x))
1966(And32 (And32 i:(Const32 <t>) z) x) && (z.Op != OpConst32 && x.Op != OpConst32) => (And32 i (And32 <t> z x))
1967(And16 (And16 i:(Const16 <t>) z) x) && (z.Op != OpConst16 && x.Op != OpConst16) => (And16 i (And16 <t> z x))
1968(And8 (And8 i:(Const8 <t>) z) x) && (z.Op != OpConst8 && x.Op != OpConst8) => (And8 i (And8 <t> z x))
1969
1970// x | (C | z) -> C | (x | z)
1971(Or64 (Or64 i:(Const64 <t>) z) x) && (z.Op != OpConst64 && x.Op != OpConst64) => (Or64 i (Or64 <t> z x))
1972(Or32 (Or32 i:(Const32 <t>) z) x) && (z.Op != OpConst32 && x.Op != OpConst32) => (Or32 i (Or32 <t> z x))
1973(Or16 (Or16 i:(Const16 <t>) z) x) && (z.Op != OpConst16 && x.Op != OpConst16) => (Or16 i (Or16 <t> z x))
1974(Or8 (Or8 i:(Const8 <t>) z) x) && (z.Op != OpConst8 && x.Op != OpConst8) => (Or8 i (Or8 <t> z x))
1975
1976// x ^ (C ^ z) -> C ^ (x ^ z)
1977(Xor64 (Xor64 i:(Const64 <t>) z) x) && (z.Op != OpConst64 && x.Op != OpConst64) => (Xor64 i (Xor64 <t> z x))
1978(Xor32 (Xor32 i:(Const32 <t>) z) x) && (z.Op != OpConst32 && x.Op != OpConst32) => (Xor32 i (Xor32 <t> z x))
1979(Xor16 (Xor16 i:(Const16 <t>) z) x) && (z.Op != OpConst16 && x.Op != OpConst16) => (Xor16 i (Xor16 <t> z x))
1980(Xor8 (Xor8 i:(Const8 <t>) z) x) && (z.Op != OpConst8 && x.Op != OpConst8) => (Xor8 i (Xor8 <t> z x))
1981
1982// x * (D * z) = D * (x * z)
1983(Mul64 (Mul64 i:(Const64 <t>) z) x) && (z.Op != OpConst64 && x.Op != OpConst64) => (Mul64 i (Mul64 <t> x z))
1984(Mul32 (Mul32 i:(Const32 <t>) z) x) && (z.Op != OpConst32 && x.Op != OpConst32) => (Mul32 i (Mul32 <t> x z))
1985(Mul16 (Mul16 i:(Const16 <t>) z) x) && (z.Op != OpConst16 && x.Op != OpConst16) => (Mul16 i (Mul16 <t> x z))
1986(Mul8 (Mul8 i:(Const8 <t>) z) x) && (z.Op != OpConst8 && x.Op != OpConst8) => (Mul8 i (Mul8 <t> x z))
1987
1988// C + (D + x) -> (C + D) + x
1989(Add64 (Const64 <t> [c]) (Add64 (Const64 <t> [d]) x)) => (Add64 (Const64 <t> [c+d]) x)
1990(Add32 (Const32 <t> [c]) (Add32 (Const32 <t> [d]) x)) => (Add32 (Const32 <t> [c+d]) x)
1991(Add16 (Const16 <t> [c]) (Add16 (Const16 <t> [d]) x)) => (Add16 (Const16 <t> [c+d]) x)
1992(Add8 (Const8 <t> [c]) (Add8 (Const8 <t> [d]) x)) => (Add8 (Const8 <t> [c+d]) x)
1993
1994// C + (D - x) -> (C + D) - x
1995(Add64 (Const64 <t> [c]) (Sub64 (Const64 <t> [d]) x)) => (Sub64 (Const64 <t> [c+d]) x)
1996(Add32 (Const32 <t> [c]) (Sub32 (Const32 <t> [d]) x)) => (Sub32 (Const32 <t> [c+d]) x)
1997(Add16 (Const16 <t> [c]) (Sub16 (Const16 <t> [d]) x)) => (Sub16 (Const16 <t> [c+d]) x)
1998(Add8 (Const8 <t> [c]) (Sub8 (Const8 <t> [d]) x)) => (Sub8 (Const8 <t> [c+d]) x)
1999
2000// C - (D - x) -> (C - D) + x
2001(Sub64 (Const64 <t> [c]) (Sub64 (Const64 <t> [d]) x)) => (Add64 (Const64 <t> [c-d]) x)
2002(Sub32 (Const32 <t> [c]) (Sub32 (Const32 <t> [d]) x)) => (Add32 (Const32 <t> [c-d]) x)
2003(Sub16 (Const16 <t> [c]) (Sub16 (Const16 <t> [d]) x)) => (Add16 (Const16 <t> [c-d]) x)
2004(Sub8 (Const8 <t> [c]) (Sub8 (Const8 <t> [d]) x)) => (Add8 (Const8 <t> [c-d]) x)
2005
2006// C - (D + x) -> (C - D) - x
2007(Sub64 (Const64 <t> [c]) (Add64 (Const64 <t> [d]) x)) => (Sub64 (Const64 <t> [c-d]) x)
2008(Sub32 (Const32 <t> [c]) (Add32 (Const32 <t> [d]) x)) => (Sub32 (Const32 <t> [c-d]) x)
2009(Sub16 (Const16 <t> [c]) (Add16 (Const16 <t> [d]) x)) => (Sub16 (Const16 <t> [c-d]) x)
2010(Sub8 (Const8 <t> [c]) (Add8 (Const8 <t> [d]) x)) => (Sub8 (Const8 <t> [c-d]) x)
2011
2012// C & (D & x) -> (C & D) & x
2013(And64 (Const64 <t> [c]) (And64 (Const64 <t> [d]) x)) => (And64 (Const64 <t> [c&d]) x)
2014(And32 (Const32 <t> [c]) (And32 (Const32 <t> [d]) x)) => (And32 (Const32 <t> [c&d]) x)
2015(And16 (Const16 <t> [c]) (And16 (Const16 <t> [d]) x)) => (And16 (Const16 <t> [c&d]) x)
2016(And8 (Const8 <t> [c]) (And8 (Const8 <t> [d]) x)) => (And8 (Const8 <t> [c&d]) x)
2017
2018// C | (D | x) -> (C | D) | x
2019(Or64 (Const64 <t> [c]) (Or64 (Const64 <t> [d]) x)) => (Or64 (Const64 <t> [c|d]) x)
2020(Or32 (Const32 <t> [c]) (Or32 (Const32 <t> [d]) x)) => (Or32 (Const32 <t> [c|d]) x)
2021(Or16 (Const16 <t> [c]) (Or16 (Const16 <t> [d]) x)) => (Or16 (Const16 <t> [c|d]) x)
2022(Or8 (Const8 <t> [c]) (Or8 (Const8 <t> [d]) x)) => (Or8 (Const8 <t> [c|d]) x)
2023
2024// C ^ (D ^ x) -> (C ^ D) ^ x
2025(Xor64 (Const64 <t> [c]) (Xor64 (Const64 <t> [d]) x)) => (Xor64 (Const64 <t> [c^d]) x)
2026(Xor32 (Const32 <t> [c]) (Xor32 (Const32 <t> [d]) x)) => (Xor32 (Const32 <t> [c^d]) x)
2027(Xor16 (Const16 <t> [c]) (Xor16 (Const16 <t> [d]) x)) => (Xor16 (Const16 <t> [c^d]) x)
2028(Xor8 (Const8 <t> [c]) (Xor8 (Const8 <t> [d]) x)) => (Xor8 (Const8 <t> [c^d]) x)
2029
2030// C * (D * x) = (C * D) * x
2031(Mul64 (Const64 <t> [c]) (Mul64 (Const64 <t> [d]) x)) => (Mul64 (Const64 <t> [c*d]) x)
2032(Mul32 (Const32 <t> [c]) (Mul32 (Const32 <t> [d]) x)) => (Mul32 (Const32 <t> [c*d]) x)
2033(Mul16 (Const16 <t> [c]) (Mul16 (Const16 <t> [d]) x)) => (Mul16 (Const16 <t> [c*d]) x)
2034(Mul8 (Const8 <t> [c]) (Mul8 (Const8 <t> [d]) x)) => (Mul8 (Const8 <t> [c*d]) x)
2035
2036// floating point optimizations
2037(Mul(32|64)F x (Const(32|64)F [1])) => x
2038(Mul32F x (Const32F [-1])) => (Neg32F x)
2039(Mul64F x (Const64F [-1])) => (Neg64F x)
2040(Mul32F x (Const32F [2])) => (Add32F x x)
2041(Mul64F x (Const64F [2])) => (Add64F x x)
2042
2043(Div32F x (Const32F <t> [c])) && reciprocalExact32(c) => (Mul32F x (Const32F <t> [1/c]))
2044(Div64F x (Const64F <t> [c])) && reciprocalExact64(c) => (Mul64F x (Const64F <t> [1/c]))
2045
2046// rewrite single-precision sqrt expression "float32(math.Sqrt(float64(x)))"
2047(Cvt64Fto32F sqrt0:(Sqrt (Cvt32Fto64F x))) && sqrt0.Uses==1 => (Sqrt32 x)
2048
2049(Sqrt (Const64F [c])) && !math.IsNaN(math.Sqrt(c)) => (Const64F [math.Sqrt(c)])
2050
2051// for rewriting constant folded math/bits ops
2052(Select0 (MakeTuple x y)) => x
2053(Select1 (MakeTuple x y)) => y
2054
2055// for rewriting results of some late-expanded rewrites (below)
2056(SelectN [0] (MakeResult x ___)) => x
2057(SelectN [1] (MakeResult x y ___)) => y
2058(SelectN [2] (MakeResult x y z ___)) => z
2059
2060// for late-expanded calls, recognize newobject and remove zeroing and nilchecks
2061(Zero (SelectN [0] call:(StaticLECall _ _)) mem:(SelectN [1] call))
2062 && isSameCall(call.Aux, "runtime.newobject")
2063 => mem
2064
2065(Store (SelectN [0] call:(StaticLECall _ _)) x mem:(SelectN [1] call))
2066 && isConstZero(x)
2067 && isSameCall(call.Aux, "runtime.newobject")
2068 => mem
2069
2070(Store (OffPtr (SelectN [0] call:(StaticLECall _ _))) x mem:(SelectN [1] call))
2071 && isConstZero(x)
2072 && isSameCall(call.Aux, "runtime.newobject")
2073 => mem
2074
2075(NilCheck ptr:(SelectN [0] call:(StaticLECall _ _)) _)
2076 && isSameCall(call.Aux, "runtime.newobject")
2077 && warnRule(fe.Debug_checknil(), v, "removed nil check")
2078 => ptr
2079
2080(NilCheck ptr:(OffPtr (SelectN [0] call:(StaticLECall _ _))) _)
2081 && isSameCall(call.Aux, "runtime.newobject")
2082 && warnRule(fe.Debug_checknil(), v, "removed nil check")
2083 => ptr
2084
2085// Addresses of globals are always non-nil.
2086(NilCheck ptr:(Addr {_} (SB)) _) => ptr
2087(NilCheck ptr:(Convert (Addr {_} (SB)) _) _) => ptr
2088
2089// Addresses of locals are always non-nil.
2090(NilCheck ptr:(LocalAddr _ _) _)
2091 && warnRule(fe.Debug_checknil(), v, "removed nil check")
2092 => ptr
2093
2094// Nil checks of nil checks are redundant.
2095// See comment at the end of https://go-review.googlesource.com/c/go/+/537775.
2096(NilCheck ptr:(NilCheck _ _) _ ) => ptr
2097
2098// for late-expanded calls, recognize memequal applied to a single constant byte
2099// Support is limited by 1, 2, 4, 8 byte sizes
2100(StaticLECall {callAux} sptr (Addr {scon} (SB)) (Const64 [1]) mem)
2101 && isSameCall(callAux, "runtime.memequal")
2102 && symIsRO(scon)
2103 => (MakeResult (Eq8 (Load <typ.Int8> sptr mem) (Const8 <typ.Int8> [int8(read8(scon,0))])) mem)
2104
2105(StaticLECall {callAux} (Addr {scon} (SB)) sptr (Const64 [1]) mem)
2106 && isSameCall(callAux, "runtime.memequal")
2107 && symIsRO(scon)
2108 => (MakeResult (Eq8 (Load <typ.Int8> sptr mem) (Const8 <typ.Int8> [int8(read8(scon,0))])) mem)
2109
2110(StaticLECall {callAux} sptr (Addr {scon} (SB)) (Const64 [2]) mem)
2111 && isSameCall(callAux, "runtime.memequal")
2112 && symIsRO(scon)
2113 && canLoadUnaligned(config)
2114 => (MakeResult (Eq16 (Load <typ.Int16> sptr mem) (Const16 <typ.Int16> [int16(read16(scon,0,config.ctxt.Arch.ByteOrder))])) mem)
2115
2116(StaticLECall {callAux} (Addr {scon} (SB)) sptr (Const64 [2]) mem)
2117 && isSameCall(callAux, "runtime.memequal")
2118 && symIsRO(scon)
2119 && canLoadUnaligned(config)
2120 => (MakeResult (Eq16 (Load <typ.Int16> sptr mem) (Const16 <typ.Int16> [int16(read16(scon,0,config.ctxt.Arch.ByteOrder))])) mem)
2121
2122(StaticLECall {callAux} sptr (Addr {scon} (SB)) (Const64 [4]) mem)
2123 && isSameCall(callAux, "runtime.memequal")
2124 && symIsRO(scon)
2125 && canLoadUnaligned(config)
2126 => (MakeResult (Eq32 (Load <typ.Int32> sptr mem) (Const32 <typ.Int32> [int32(read32(scon,0,config.ctxt.Arch.ByteOrder))])) mem)
2127
2128(StaticLECall {callAux} (Addr {scon} (SB)) sptr (Const64 [4]) mem)
2129 && isSameCall(callAux, "runtime.memequal")
2130 && symIsRO(scon)
2131 && canLoadUnaligned(config)
2132 => (MakeResult (Eq32 (Load <typ.Int32> sptr mem) (Const32 <typ.Int32> [int32(read32(scon,0,config.ctxt.Arch.ByteOrder))])) mem)
2133
2134(StaticLECall {callAux} sptr (Addr {scon} (SB)) (Const64 [8]) mem)
2135 && isSameCall(callAux, "runtime.memequal")
2136 && symIsRO(scon)
2137 && canLoadUnaligned(config) && config.PtrSize == 8
2138 => (MakeResult (Eq64 (Load <typ.Int64> sptr mem) (Const64 <typ.Int64> [int64(read64(scon,0,config.ctxt.Arch.ByteOrder))])) mem)
2139
2140(StaticLECall {callAux} (Addr {scon} (SB)) sptr (Const64 [8]) mem)
2141 && isSameCall(callAux, "runtime.memequal")
2142 && symIsRO(scon)
2143 && canLoadUnaligned(config) && config.PtrSize == 8
2144 => (MakeResult (Eq64 (Load <typ.Int64> sptr mem) (Const64 <typ.Int64> [int64(read64(scon,0,config.ctxt.Arch.ByteOrder))])) mem)
2145
2146(StaticLECall {callAux} _ _ (Const64 [0]) mem)
2147 && isSameCall(callAux, "runtime.memequal")
2148 => (MakeResult (ConstBool <typ.Bool> [true]) mem)
2149
2150(Static(Call|LECall) {callAux} p q _ mem)
2151 && isSameCall(callAux, "runtime.memequal")
2152 && isSamePtr(p, q)
2153 => (MakeResult (ConstBool <typ.Bool> [true]) mem)
2154
2155// Turn known-size calls to memclrNoHeapPointers into a Zero.
2156// Note that we are using types.Types[types.TUINT8] instead of sptr.Type.Elem() - see issue 55122 and CL 431496 for more details.
2157(SelectN [0] call:(StaticCall {sym} sptr (Const(64|32) [c]) mem))
2158 && isInlinableMemclr(config, int64(c))
2159 && isSameCall(sym, "runtime.memclrNoHeapPointers")
2160 && call.Uses == 1
2161 && clobber(call)
2162 => (Zero {types.Types[types.TUINT8]} [int64(c)] sptr mem)
2163
2164// Recognise make([]T, 0) and replace it with a pointer to the zerobase
2165(StaticLECall {callAux} _ (Const(64|32) [0]) (Const(64|32) [0]) mem)
2166 && isSameCall(callAux, "runtime.makeslice")
2167 => (MakeResult (Addr <v.Type.FieldType(0)> {ir.Syms.Zerobase} (SB)) mem)
2168
2169// Evaluate constant address comparisons.
2170(EqPtr x x) => (ConstBool [true])
2171(NeqPtr x x) => (ConstBool [false])
2172(EqPtr (Addr {x} _) (Addr {y} _)) => (ConstBool [x == y])
2173(EqPtr (Addr {x} _) (OffPtr [o] (Addr {y} _))) => (ConstBool [x == y && o == 0])
2174(EqPtr (OffPtr [o1] (Addr {x} _)) (OffPtr [o2] (Addr {y} _))) => (ConstBool [x == y && o1 == o2])
2175(NeqPtr (Addr {x} _) (Addr {y} _)) => (ConstBool [x != y])
2176(NeqPtr (Addr {x} _) (OffPtr [o] (Addr {y} _))) => (ConstBool [x != y || o != 0])
2177(NeqPtr (OffPtr [o1] (Addr {x} _)) (OffPtr [o2] (Addr {y} _))) => (ConstBool [x != y || o1 != o2])
2178(EqPtr (LocalAddr {x} _ _) (LocalAddr {y} _ _)) => (ConstBool [x == y])
2179(EqPtr (LocalAddr {x} _ _) (OffPtr [o] (LocalAddr {y} _ _))) => (ConstBool [x == y && o == 0])
2180(EqPtr (OffPtr [o1] (LocalAddr {x} _ _)) (OffPtr [o2] (LocalAddr {y} _ _))) => (ConstBool [x == y && o1 == o2])
2181(NeqPtr (LocalAddr {x} _ _) (LocalAddr {y} _ _)) => (ConstBool [x != y])
2182(NeqPtr (LocalAddr {x} _ _) (OffPtr [o] (LocalAddr {y} _ _))) => (ConstBool [x != y || o != 0])
2183(NeqPtr (OffPtr [o1] (LocalAddr {x} _ _)) (OffPtr [o2] (LocalAddr {y} _ _))) => (ConstBool [x != y || o1 != o2])
2184(EqPtr (OffPtr [o1] p1) p2) && isSamePtr(p1, p2) => (ConstBool [o1 == 0])
2185(NeqPtr (OffPtr [o1] p1) p2) && isSamePtr(p1, p2) => (ConstBool [o1 != 0])
2186(EqPtr (OffPtr [o1] p1) (OffPtr [o2] p2)) && isSamePtr(p1, p2) => (ConstBool [o1 == o2])
2187(NeqPtr (OffPtr [o1] p1) (OffPtr [o2] p2)) && isSamePtr(p1, p2) => (ConstBool [o1 != o2])
2188(EqPtr (Const(32|64) [c]) (Const(32|64) [d])) => (ConstBool [c == d])
2189(NeqPtr (Const(32|64) [c]) (Const(32|64) [d])) => (ConstBool [c != d])
2190(EqPtr (Convert (Addr {x} _) _) (Addr {y} _)) => (ConstBool [x==y])
2191(NeqPtr (Convert (Addr {x} _) _) (Addr {y} _)) => (ConstBool [x!=y])
2192
2193(EqPtr (LocalAddr _ _) (Addr _)) => (ConstBool [false])
2194(EqPtr (OffPtr (LocalAddr _ _)) (Addr _)) => (ConstBool [false])
2195(EqPtr (LocalAddr _ _) (OffPtr (Addr _))) => (ConstBool [false])
2196(EqPtr (OffPtr (LocalAddr _ _)) (OffPtr (Addr _))) => (ConstBool [false])
2197(NeqPtr (LocalAddr _ _) (Addr _)) => (ConstBool [true])
2198(NeqPtr (OffPtr (LocalAddr _ _)) (Addr _)) => (ConstBool [true])
2199(NeqPtr (LocalAddr _ _) (OffPtr (Addr _))) => (ConstBool [true])
2200(NeqPtr (OffPtr (LocalAddr _ _)) (OffPtr (Addr _))) => (ConstBool [true])
2201
2202// Simplify address comparisons.
2203(EqPtr (AddPtr p1 o1) p2) && isSamePtr(p1, p2) => (Not (IsNonNil o1))
2204(NeqPtr (AddPtr p1 o1) p2) && isSamePtr(p1, p2) => (IsNonNil o1)
2205(EqPtr (Const(32|64) [0]) p) => (Not (IsNonNil p))
2206(NeqPtr (Const(32|64) [0]) p) => (IsNonNil p)
2207(EqPtr (ConstNil) p) => (Not (IsNonNil p))
2208(NeqPtr (ConstNil) p) => (IsNonNil p)
2209
2210// Evaluate constant user nil checks.
2211(IsNonNil (ConstNil)) => (ConstBool [false])
2212(IsNonNil (Const(32|64) [c])) => (ConstBool [c != 0])
2213(IsNonNil (Addr _) ) => (ConstBool [true])
2214(IsNonNil (Convert (Addr _) _)) => (ConstBool [true])
2215(IsNonNil (LocalAddr _ _)) => (ConstBool [true])
2216
2217// Inline small or disjoint runtime.memmove calls with constant length.
2218// See the comment in op Move in genericOps.go for discussion of the type.
2219//
2220// Note that we've lost any knowledge of the type and alignment requirements
2221// of the source and destination. We only know the size, and that the type
2222// contains no pointers.
2223// The type of the move is not necessarily v.Args[0].Type().Elem()!
2224// See issue 55122 for details.
2225//
2226// Because expand calls runs after prove, constants useful to this pattern may not appear.
2227// Both versions need to exist; the memory and register variants.
2228//
2229// Match post-expansion calls, memory version.
2230(SelectN [0] call:(StaticCall {sym} s1:(Store _ (Const(64|32) [sz]) s2:(Store _ src s3:(Store {t} _ dst mem)))))
2231 && sz >= 0
2232 && isSameCall(sym, "runtime.memmove")
2233 && s1.Uses == 1 && s2.Uses == 1 && s3.Uses == 1
2234 && isInlinableMemmove(dst, src, int64(sz), config)
2235 && clobber(s1, s2, s3, call)
2236 => (Move {types.Types[types.TUINT8]} [int64(sz)] dst src mem)
2237
2238// Match post-expansion calls, register version.
2239(SelectN [0] call:(StaticCall {sym} dst src (Const(64|32) [sz]) mem))
2240 && sz >= 0
2241 && call.Uses == 1 // this will exclude all calls with results
2242 && isSameCall(sym, "runtime.memmove")
2243 && isInlinableMemmove(dst, src, int64(sz), config)
2244 && clobber(call)
2245 => (Move {types.Types[types.TUINT8]} [int64(sz)] dst src mem)
2246
2247// Match pre-expansion calls.
2248(SelectN [0] call:(StaticLECall {sym} dst src (Const(64|32) [sz]) mem))
2249 && sz >= 0
2250 && call.Uses == 1 // this will exclude all calls with results
2251 && isSameCall(sym, "runtime.memmove")
2252 && isInlinableMemmove(dst, src, int64(sz), config)
2253 && clobber(call)
2254 => (Move {types.Types[types.TUINT8]} [int64(sz)] dst src mem)
2255
2256// De-virtualize late-expanded interface calls into late-expanded static calls.
2257(InterLECall [argsize] {auxCall} (Addr {fn} (SB)) ___) => devirtLECall(v, fn.(*obj.LSym))
2258
2259// Move and Zero optimizations.
2260// Move source and destination may overlap.
2261
2262// Convert Moves into Zeros when the source is known to be zeros.
2263(Move {t} [n] dst1 src mem:(Zero {t} [n] dst2 _)) && isSamePtr(src, dst2)
2264 => (Zero {t} [n] dst1 mem)
2265(Move {t} [n] dst1 src mem:(VarDef (Zero {t} [n] dst0 _))) && isSamePtr(src, dst0)
2266 => (Zero {t} [n] dst1 mem)
2267(Move {t} [n] dst (Addr {sym} (SB)) mem) && symIsROZero(sym) => (Zero {t} [n] dst mem)
2268
2269// Don't Store to variables that are about to be overwritten by Move/Zero.
2270(Zero {t1} [n] p1 store:(Store {t2} (OffPtr [o2] p2) _ mem))
2271 && isSamePtr(p1, p2) && store.Uses == 1
2272 && n >= o2 + t2.Size()
2273 && clobber(store)
2274 => (Zero {t1} [n] p1 mem)
2275(Move {t1} [n] dst1 src1 store:(Store {t2} op:(OffPtr [o2] dst2) _ mem))
2276 && isSamePtr(dst1, dst2) && store.Uses == 1
2277 && n >= o2 + t2.Size()
2278 && disjoint(src1, n, op, t2.Size())
2279 && clobber(store)
2280 => (Move {t1} [n] dst1 src1 mem)
2281
2282// Don't Move to variables that are immediately completely overwritten.
2283(Zero {t} [n] dst1 move:(Move {t} [n] dst2 _ mem))
2284 && move.Uses == 1
2285 && isSamePtr(dst1, dst2)
2286 && clobber(move)
2287 => (Zero {t} [n] dst1 mem)
2288(Move {t} [n] dst1 src1 move:(Move {t} [n] dst2 _ mem))
2289 && move.Uses == 1
2290 && isSamePtr(dst1, dst2) && disjoint(src1, n, dst2, n)
2291 && clobber(move)
2292 => (Move {t} [n] dst1 src1 mem)
2293(Zero {t} [n] dst1 vardef:(VarDef {x} move:(Move {t} [n] dst2 _ mem)))
2294 && move.Uses == 1 && vardef.Uses == 1
2295 && isSamePtr(dst1, dst2)
2296 && clobber(move, vardef)
2297 => (Zero {t} [n] dst1 (VarDef {x} mem))
2298(Move {t} [n] dst1 src1 vardef:(VarDef {x} move:(Move {t} [n] dst2 _ mem)))
2299 && move.Uses == 1 && vardef.Uses == 1
2300 && isSamePtr(dst1, dst2) && disjoint(src1, n, dst2, n)
2301 && clobber(move, vardef)
2302 => (Move {t} [n] dst1 src1 (VarDef {x} mem))
2303(Store {t1} op1:(OffPtr [o1] p1) d1
2304 m2:(Store {t2} op2:(OffPtr [0] p2) d2
2305 m3:(Move [n] p3 _ mem)))
2306 && m2.Uses == 1 && m3.Uses == 1
2307 && o1 == t2.Size()
2308 && n == t2.Size() + t1.Size()
2309 && isSamePtr(p1, p2) && isSamePtr(p2, p3)
2310 && clobber(m2, m3)
2311 => (Store {t1} op1 d1 (Store {t2} op2 d2 mem))
2312(Store {t1} op1:(OffPtr [o1] p1) d1
2313 m2:(Store {t2} op2:(OffPtr [o2] p2) d2
2314 m3:(Store {t3} op3:(OffPtr [0] p3) d3
2315 m4:(Move [n] p4 _ mem))))
2316 && m2.Uses == 1 && m3.Uses == 1 && m4.Uses == 1
2317 && o2 == t3.Size()
2318 && o1-o2 == t2.Size()
2319 && n == t3.Size() + t2.Size() + t1.Size()
2320 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4)
2321 && clobber(m2, m3, m4)
2322 => (Store {t1} op1 d1 (Store {t2} op2 d2 (Store {t3} op3 d3 mem)))
2323(Store {t1} op1:(OffPtr [o1] p1) d1
2324 m2:(Store {t2} op2:(OffPtr [o2] p2) d2
2325 m3:(Store {t3} op3:(OffPtr [o3] p3) d3
2326 m4:(Store {t4} op4:(OffPtr [0] p4) d4
2327 m5:(Move [n] p5 _ mem)))))
2328 && m2.Uses == 1 && m3.Uses == 1 && m4.Uses == 1 && m5.Uses == 1
2329 && o3 == t4.Size()
2330 && o2-o3 == t3.Size()
2331 && o1-o2 == t2.Size()
2332 && n == t4.Size() + t3.Size() + t2.Size() + t1.Size()
2333 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4) && isSamePtr(p4, p5)
2334 && clobber(m2, m3, m4, m5)
2335 => (Store {t1} op1 d1 (Store {t2} op2 d2 (Store {t3} op3 d3 (Store {t4} op4 d4 mem))))
2336
2337// Don't Zero variables that are immediately completely overwritten
2338// before being accessed.
2339(Move {t} [n] dst1 src1 zero:(Zero {t} [n] dst2 mem))
2340 && zero.Uses == 1
2341 && isSamePtr(dst1, dst2) && disjoint(src1, n, dst2, n)
2342 && clobber(zero)
2343 => (Move {t} [n] dst1 src1 mem)
2344(Move {t} [n] dst1 src1 vardef:(VarDef {x} zero:(Zero {t} [n] dst2 mem)))
2345 && zero.Uses == 1 && vardef.Uses == 1
2346 && isSamePtr(dst1, dst2) && disjoint(src1, n, dst2, n)
2347 && clobber(zero, vardef)
2348 => (Move {t} [n] dst1 src1 (VarDef {x} mem))
2349(Store {t1} op1:(OffPtr [o1] p1) d1
2350 m2:(Store {t2} op2:(OffPtr [0] p2) d2
2351 m3:(Zero [n] p3 mem)))
2352 && m2.Uses == 1 && m3.Uses == 1
2353 && o1 == t2.Size()
2354 && n == t2.Size() + t1.Size()
2355 && isSamePtr(p1, p2) && isSamePtr(p2, p3)
2356 && clobber(m2, m3)
2357 => (Store {t1} op1 d1 (Store {t2} op2 d2 mem))
2358(Store {t1} op1:(OffPtr [o1] p1) d1
2359 m2:(Store {t2} op2:(OffPtr [o2] p2) d2
2360 m3:(Store {t3} op3:(OffPtr [0] p3) d3
2361 m4:(Zero [n] p4 mem))))
2362 && m2.Uses == 1 && m3.Uses == 1 && m4.Uses == 1
2363 && o2 == t3.Size()
2364 && o1-o2 == t2.Size()
2365 && n == t3.Size() + t2.Size() + t1.Size()
2366 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4)
2367 && clobber(m2, m3, m4)
2368 => (Store {t1} op1 d1 (Store {t2} op2 d2 (Store {t3} op3 d3 mem)))
2369(Store {t1} op1:(OffPtr [o1] p1) d1
2370 m2:(Store {t2} op2:(OffPtr [o2] p2) d2
2371 m3:(Store {t3} op3:(OffPtr [o3] p3) d3
2372 m4:(Store {t4} op4:(OffPtr [0] p4) d4
2373 m5:(Zero [n] p5 mem)))))
2374 && m2.Uses == 1 && m3.Uses == 1 && m4.Uses == 1 && m5.Uses == 1
2375 && o3 == t4.Size()
2376 && o2-o3 == t3.Size()
2377 && o1-o2 == t2.Size()
2378 && n == t4.Size() + t3.Size() + t2.Size() + t1.Size()
2379 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4) && isSamePtr(p4, p5)
2380 && clobber(m2, m3, m4, m5)
2381 => (Store {t1} op1 d1 (Store {t2} op2 d2 (Store {t3} op3 d3 (Store {t4} op4 d4 mem))))
2382
2383// Don't Move from memory if the values are likely to already be
2384// in registers.
2385(Move {t1} [n] dst p1
2386 mem:(Store {t2} op2:(OffPtr <tt2> [o2] p2) d1
2387 (Store {t3} op3:(OffPtr <tt3> [0] p3) d2 _)))
2388 && isSamePtr(p1, p2) && isSamePtr(p2, p3)
2389 && t2.Alignment() <= t1.Alignment()
2390 && t3.Alignment() <= t1.Alignment()
2391 && registerizable(b, t2)
2392 && registerizable(b, t3)
2393 && o2 == t3.Size()
2394 && n == t2.Size() + t3.Size()
2395 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2396 (Store {t3} (OffPtr <tt3> [0] dst) d2 mem))
2397(Move {t1} [n] dst p1
2398 mem:(Store {t2} op2:(OffPtr <tt2> [o2] p2) d1
2399 (Store {t3} op3:(OffPtr <tt3> [o3] p3) d2
2400 (Store {t4} op4:(OffPtr <tt4> [0] p4) d3 _))))
2401 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4)
2402 && t2.Alignment() <= t1.Alignment()
2403 && t3.Alignment() <= t1.Alignment()
2404 && t4.Alignment() <= t1.Alignment()
2405 && registerizable(b, t2)
2406 && registerizable(b, t3)
2407 && registerizable(b, t4)
2408 && o3 == t4.Size()
2409 && o2-o3 == t3.Size()
2410 && n == t2.Size() + t3.Size() + t4.Size()
2411 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2412 (Store {t3} (OffPtr <tt3> [o3] dst) d2
2413 (Store {t4} (OffPtr <tt4> [0] dst) d3 mem)))
2414(Move {t1} [n] dst p1
2415 mem:(Store {t2} op2:(OffPtr <tt2> [o2] p2) d1
2416 (Store {t3} op3:(OffPtr <tt3> [o3] p3) d2
2417 (Store {t4} op4:(OffPtr <tt4> [o4] p4) d3
2418 (Store {t5} op5:(OffPtr <tt5> [0] p5) d4 _)))))
2419 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4) && isSamePtr(p4, p5)
2420 && t2.Alignment() <= t1.Alignment()
2421 && t3.Alignment() <= t1.Alignment()
2422 && t4.Alignment() <= t1.Alignment()
2423 && t5.Alignment() <= t1.Alignment()
2424 && registerizable(b, t2)
2425 && registerizable(b, t3)
2426 && registerizable(b, t4)
2427 && registerizable(b, t5)
2428 && o4 == t5.Size()
2429 && o3-o4 == t4.Size()
2430 && o2-o3 == t3.Size()
2431 && n == t2.Size() + t3.Size() + t4.Size() + t5.Size()
2432 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2433 (Store {t3} (OffPtr <tt3> [o3] dst) d2
2434 (Store {t4} (OffPtr <tt4> [o4] dst) d3
2435 (Store {t5} (OffPtr <tt5> [0] dst) d4 mem))))
2436
2437// Same thing but with VarDef in the middle.
2438(Move {t1} [n] dst p1
2439 mem:(VarDef
2440 (Store {t2} op2:(OffPtr <tt2> [o2] p2) d1
2441 (Store {t3} op3:(OffPtr <tt3> [0] p3) d2 _))))
2442 && isSamePtr(p1, p2) && isSamePtr(p2, p3)
2443 && t2.Alignment() <= t1.Alignment()
2444 && t3.Alignment() <= t1.Alignment()
2445 && registerizable(b, t2)
2446 && registerizable(b, t3)
2447 && o2 == t3.Size()
2448 && n == t2.Size() + t3.Size()
2449 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2450 (Store {t3} (OffPtr <tt3> [0] dst) d2 mem))
2451(Move {t1} [n] dst p1
2452 mem:(VarDef
2453 (Store {t2} op2:(OffPtr <tt2> [o2] p2) d1
2454 (Store {t3} op3:(OffPtr <tt3> [o3] p3) d2
2455 (Store {t4} op4:(OffPtr <tt4> [0] p4) d3 _)))))
2456 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4)
2457 && t2.Alignment() <= t1.Alignment()
2458 && t3.Alignment() <= t1.Alignment()
2459 && t4.Alignment() <= t1.Alignment()
2460 && registerizable(b, t2)
2461 && registerizable(b, t3)
2462 && registerizable(b, t4)
2463 && o3 == t4.Size()
2464 && o2-o3 == t3.Size()
2465 && n == t2.Size() + t3.Size() + t4.Size()
2466 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2467 (Store {t3} (OffPtr <tt3> [o3] dst) d2
2468 (Store {t4} (OffPtr <tt4> [0] dst) d3 mem)))
2469(Move {t1} [n] dst p1
2470 mem:(VarDef
2471 (Store {t2} op2:(OffPtr <tt2> [o2] p2) d1
2472 (Store {t3} op3:(OffPtr <tt3> [o3] p3) d2
2473 (Store {t4} op4:(OffPtr <tt4> [o4] p4) d3
2474 (Store {t5} op5:(OffPtr <tt5> [0] p5) d4 _))))))
2475 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4) && isSamePtr(p4, p5)
2476 && t2.Alignment() <= t1.Alignment()
2477 && t3.Alignment() <= t1.Alignment()
2478 && t4.Alignment() <= t1.Alignment()
2479 && t5.Alignment() <= t1.Alignment()
2480 && registerizable(b, t2)
2481 && registerizable(b, t3)
2482 && registerizable(b, t4)
2483 && registerizable(b, t5)
2484 && o4 == t5.Size()
2485 && o3-o4 == t4.Size()
2486 && o2-o3 == t3.Size()
2487 && n == t2.Size() + t3.Size() + t4.Size() + t5.Size()
2488 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2489 (Store {t3} (OffPtr <tt3> [o3] dst) d2
2490 (Store {t4} (OffPtr <tt4> [o4] dst) d3
2491 (Store {t5} (OffPtr <tt5> [0] dst) d4 mem))))
2492
2493// Prefer to Zero and Store than to Move.
2494(Move {t1} [n] dst p1
2495 mem:(Store {t2} op2:(OffPtr <tt2> [o2] p2) d1
2496 (Zero {t3} [n] p3 _)))
2497 && isSamePtr(p1, p2) && isSamePtr(p2, p3)
2498 && t2.Alignment() <= t1.Alignment()
2499 && t3.Alignment() <= t1.Alignment()
2500 && registerizable(b, t2)
2501 && n >= o2 + t2.Size()
2502 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2503 (Zero {t1} [n] dst mem))
2504(Move {t1} [n] dst p1
2505 mem:(Store {t2} (OffPtr <tt2> [o2] p2) d1
2506 (Store {t3} (OffPtr <tt3> [o3] p3) d2
2507 (Zero {t4} [n] p4 _))))
2508 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4)
2509 && t2.Alignment() <= t1.Alignment()
2510 && t3.Alignment() <= t1.Alignment()
2511 && t4.Alignment() <= t1.Alignment()
2512 && registerizable(b, t2)
2513 && registerizable(b, t3)
2514 && n >= o2 + t2.Size()
2515 && n >= o3 + t3.Size()
2516 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2517 (Store {t3} (OffPtr <tt3> [o3] dst) d2
2518 (Zero {t1} [n] dst mem)))
2519(Move {t1} [n] dst p1
2520 mem:(Store {t2} (OffPtr <tt2> [o2] p2) d1
2521 (Store {t3} (OffPtr <tt3> [o3] p3) d2
2522 (Store {t4} (OffPtr <tt4> [o4] p4) d3
2523 (Zero {t5} [n] p5 _)))))
2524 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4) && isSamePtr(p4, p5)
2525 && t2.Alignment() <= t1.Alignment()
2526 && t3.Alignment() <= t1.Alignment()
2527 && t4.Alignment() <= t1.Alignment()
2528 && t5.Alignment() <= t1.Alignment()
2529 && registerizable(b, t2)
2530 && registerizable(b, t3)
2531 && registerizable(b, t4)
2532 && n >= o2 + t2.Size()
2533 && n >= o3 + t3.Size()
2534 && n >= o4 + t4.Size()
2535 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2536 (Store {t3} (OffPtr <tt3> [o3] dst) d2
2537 (Store {t4} (OffPtr <tt4> [o4] dst) d3
2538 (Zero {t1} [n] dst mem))))
2539(Move {t1} [n] dst p1
2540 mem:(Store {t2} (OffPtr <tt2> [o2] p2) d1
2541 (Store {t3} (OffPtr <tt3> [o3] p3) d2
2542 (Store {t4} (OffPtr <tt4> [o4] p4) d3
2543 (Store {t5} (OffPtr <tt5> [o5] p5) d4
2544 (Zero {t6} [n] p6 _))))))
2545 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4) && isSamePtr(p4, p5) && isSamePtr(p5, p6)
2546 && t2.Alignment() <= t1.Alignment()
2547 && t3.Alignment() <= t1.Alignment()
2548 && t4.Alignment() <= t1.Alignment()
2549 && t5.Alignment() <= t1.Alignment()
2550 && t6.Alignment() <= t1.Alignment()
2551 && registerizable(b, t2)
2552 && registerizable(b, t3)
2553 && registerizable(b, t4)
2554 && registerizable(b, t5)
2555 && n >= o2 + t2.Size()
2556 && n >= o3 + t3.Size()
2557 && n >= o4 + t4.Size()
2558 && n >= o5 + t5.Size()
2559 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2560 (Store {t3} (OffPtr <tt3> [o3] dst) d2
2561 (Store {t4} (OffPtr <tt4> [o4] dst) d3
2562 (Store {t5} (OffPtr <tt5> [o5] dst) d4
2563 (Zero {t1} [n] dst mem)))))
2564(Move {t1} [n] dst p1
2565 mem:(VarDef
2566 (Store {t2} op2:(OffPtr <tt2> [o2] p2) d1
2567 (Zero {t3} [n] p3 _))))
2568 && isSamePtr(p1, p2) && isSamePtr(p2, p3)
2569 && t2.Alignment() <= t1.Alignment()
2570 && t3.Alignment() <= t1.Alignment()
2571 && registerizable(b, t2)
2572 && n >= o2 + t2.Size()
2573 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2574 (Zero {t1} [n] dst mem))
2575(Move {t1} [n] dst p1
2576 mem:(VarDef
2577 (Store {t2} (OffPtr <tt2> [o2] p2) d1
2578 (Store {t3} (OffPtr <tt3> [o3] p3) d2
2579 (Zero {t4} [n] p4 _)))))
2580 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4)
2581 && t2.Alignment() <= t1.Alignment()
2582 && t3.Alignment() <= t1.Alignment()
2583 && t4.Alignment() <= t1.Alignment()
2584 && registerizable(b, t2)
2585 && registerizable(b, t3)
2586 && n >= o2 + t2.Size()
2587 && n >= o3 + t3.Size()
2588 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2589 (Store {t3} (OffPtr <tt3> [o3] dst) d2
2590 (Zero {t1} [n] dst mem)))
2591(Move {t1} [n] dst p1
2592 mem:(VarDef
2593 (Store {t2} (OffPtr <tt2> [o2] p2) d1
2594 (Store {t3} (OffPtr <tt3> [o3] p3) d2
2595 (Store {t4} (OffPtr <tt4> [o4] p4) d3
2596 (Zero {t5} [n] p5 _))))))
2597 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4) && isSamePtr(p4, p5)
2598 && t2.Alignment() <= t1.Alignment()
2599 && t3.Alignment() <= t1.Alignment()
2600 && t4.Alignment() <= t1.Alignment()
2601 && t5.Alignment() <= t1.Alignment()
2602 && registerizable(b, t2)
2603 && registerizable(b, t3)
2604 && registerizable(b, t4)
2605 && n >= o2 + t2.Size()
2606 && n >= o3 + t3.Size()
2607 && n >= o4 + t4.Size()
2608 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2609 (Store {t3} (OffPtr <tt3> [o3] dst) d2
2610 (Store {t4} (OffPtr <tt4> [o4] dst) d3
2611 (Zero {t1} [n] dst mem))))
2612(Move {t1} [n] dst p1
2613 mem:(VarDef
2614 (Store {t2} (OffPtr <tt2> [o2] p2) d1
2615 (Store {t3} (OffPtr <tt3> [o3] p3) d2
2616 (Store {t4} (OffPtr <tt4> [o4] p4) d3
2617 (Store {t5} (OffPtr <tt5> [o5] p5) d4
2618 (Zero {t6} [n] p6 _)))))))
2619 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4) && isSamePtr(p4, p5) && isSamePtr(p5, p6)
2620 && t2.Alignment() <= t1.Alignment()
2621 && t3.Alignment() <= t1.Alignment()
2622 && t4.Alignment() <= t1.Alignment()
2623 && t5.Alignment() <= t1.Alignment()
2624 && t6.Alignment() <= t1.Alignment()
2625 && registerizable(b, t2)
2626 && registerizable(b, t3)
2627 && registerizable(b, t4)
2628 && registerizable(b, t5)
2629 && n >= o2 + t2.Size()
2630 && n >= o3 + t3.Size()
2631 && n >= o4 + t4.Size()
2632 && n >= o5 + t5.Size()
2633 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2634 (Store {t3} (OffPtr <tt3> [o3] dst) d2
2635 (Store {t4} (OffPtr <tt4> [o4] dst) d3
2636 (Store {t5} (OffPtr <tt5> [o5] dst) d4
2637 (Zero {t1} [n] dst mem)))))
2638
2639(SelectN [0] call:(StaticLECall {sym} a x)) && needRaceCleanup(sym, call) && clobber(call) => x
2640(SelectN [0] call:(StaticLECall {sym} x)) && needRaceCleanup(sym, call) && clobber(call) => x
2641
2642// When rewriting append to growslice, we use as the new length the result of
2643// growslice so that we don't have to spill/restore the new length around the growslice call.
2644// The exception here is that if the new length is a constant, avoiding spilling it
2645// is pointless and its constantness is sometimes useful for subsequent optimizations.
2646// See issue 56440.
2647// Note there are 2 rules here, one for the pre-decomposed []T result and one for
2648// the post-decomposed (*T,int,int) result. (The latter is generated after call expansion.)
2649(SliceLen (SelectN [0] (StaticLECall {sym} _ newLen:(Const(64|32)) _ _ _ _))) && isSameCall(sym, "runtime.growslice") => newLen
2650(SelectN [1] (StaticCall {sym} _ newLen:(Const(64|32)) _ _ _ _)) && v.Type.IsInteger() && isSameCall(sym, "runtime.growslice") => newLen
2651
2652// Collapse moving A -> B -> C into just A -> C.
2653// Later passes (deadstore, elim unread auto) will remove the A -> B move, if possible.
2654// This happens most commonly when B is an autotmp inserted earlier
2655// during compilation to ensure correctness.
2656// Take care that overlapping moves are preserved.
2657// Restrict this optimization to the stack, to avoid duplicating loads from the heap;
2658// see CL 145208 for discussion.
2659(Move {t1} [s] dst tmp1 midmem:(Move {t2} [s] tmp2 src _))
2660 && t1.Compare(t2) == types.CMPeq
2661 && isSamePtr(tmp1, tmp2)
2662 && isStackPtr(src) && !isVolatile(src)
2663 && disjoint(src, s, tmp2, s)
2664 && (disjoint(src, s, dst, s) || isInlinableMemmove(dst, src, s, config))
2665 => (Move {t1} [s] dst src midmem)
2666
2667// Same, but for large types that require VarDefs.
2668(Move {t1} [s] dst tmp1 midmem:(VarDef (Move {t2} [s] tmp2 src _)))
2669 && t1.Compare(t2) == types.CMPeq
2670 && isSamePtr(tmp1, tmp2)
2671 && isStackPtr(src) && !isVolatile(src)
2672 && disjoint(src, s, tmp2, s)
2673 && (disjoint(src, s, dst, s) || isInlinableMemmove(dst, src, s, config))
2674 => (Move {t1} [s] dst src midmem)
2675
2676// Don't zero the same bits twice.
2677(Zero {t} [s] dst1 zero:(Zero {t} [s] dst2 _)) && isSamePtr(dst1, dst2) => zero
2678(Zero {t} [s] dst1 vardef:(VarDef (Zero {t} [s] dst2 _))) && isSamePtr(dst1, dst2) => vardef
2679
2680// Elide self-moves. This only happens rarely (e.g test/fixedbugs/bug277.go).
2681// However, this rule is needed to prevent the previous rule from looping forever in such cases.
2682(Move dst src mem) && isSamePtr(dst, src) => mem
2683
2684// Constant rotate detection.
2685((Add64|Or64|Xor64) (Lsh64x64 x z:(Const64 <t> [c])) (Rsh64Ux64 x (Const64 [d]))) && c < 64 && d == 64-c && canRotate(config, 64) => (RotateLeft64 x z)
2686((Add32|Or32|Xor32) (Lsh32x64 x z:(Const64 <t> [c])) (Rsh32Ux64 x (Const64 [d]))) && c < 32 && d == 32-c && canRotate(config, 32) => (RotateLeft32 x z)
2687((Add16|Or16|Xor16) (Lsh16x64 x z:(Const64 <t> [c])) (Rsh16Ux64 x (Const64 [d]))) && c < 16 && d == 16-c && canRotate(config, 16) => (RotateLeft16 x z)
2688((Add8|Or8|Xor8) (Lsh8x64 x z:(Const64 <t> [c])) (Rsh8Ux64 x (Const64 [d]))) && c < 8 && d == 8-c && canRotate(config, 8) => (RotateLeft8 x z)
2689
2690// Non-constant rotate detection.
2691// We use shiftIsBounded to make sure that neither of the shifts are >64.
2692// Note: these rules are subtle when the shift amounts are 0/64, as Go shifts
2693// are different from most native shifts. But it works out.
2694((Add64|Or64|Xor64) left:(Lsh64x64 x y) right:(Rsh64Ux64 x (Sub64 (Const64 [64]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 64) => (RotateLeft64 x y)
2695((Add64|Or64|Xor64) left:(Lsh64x32 x y) right:(Rsh64Ux32 x (Sub32 (Const32 [64]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 64) => (RotateLeft64 x y)
2696((Add64|Or64|Xor64) left:(Lsh64x16 x y) right:(Rsh64Ux16 x (Sub16 (Const16 [64]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 64) => (RotateLeft64 x y)
2697((Add64|Or64|Xor64) left:(Lsh64x8 x y) right:(Rsh64Ux8 x (Sub8 (Const8 [64]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 64) => (RotateLeft64 x y)
2698
2699((Add64|Or64|Xor64) right:(Rsh64Ux64 x y) left:(Lsh64x64 x z:(Sub64 (Const64 [64]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 64) => (RotateLeft64 x z)
2700((Add64|Or64|Xor64) right:(Rsh64Ux32 x y) left:(Lsh64x32 x z:(Sub32 (Const32 [64]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 64) => (RotateLeft64 x z)
2701((Add64|Or64|Xor64) right:(Rsh64Ux16 x y) left:(Lsh64x16 x z:(Sub16 (Const16 [64]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 64) => (RotateLeft64 x z)
2702((Add64|Or64|Xor64) right:(Rsh64Ux8 x y) left:(Lsh64x8 x z:(Sub8 (Const8 [64]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 64) => (RotateLeft64 x z)
2703
2704((Add32|Or32|Xor32) left:(Lsh32x64 x y) right:(Rsh32Ux64 x (Sub64 (Const64 [32]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 32) => (RotateLeft32 x y)
2705((Add32|Or32|Xor32) left:(Lsh32x32 x y) right:(Rsh32Ux32 x (Sub32 (Const32 [32]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 32) => (RotateLeft32 x y)
2706((Add32|Or32|Xor32) left:(Lsh32x16 x y) right:(Rsh32Ux16 x (Sub16 (Const16 [32]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 32) => (RotateLeft32 x y)
2707((Add32|Or32|Xor32) left:(Lsh32x8 x y) right:(Rsh32Ux8 x (Sub8 (Const8 [32]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 32) => (RotateLeft32 x y)
2708
2709((Add32|Or32|Xor32) right:(Rsh32Ux64 x y) left:(Lsh32x64 x z:(Sub64 (Const64 [32]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 32) => (RotateLeft32 x z)
2710((Add32|Or32|Xor32) right:(Rsh32Ux32 x y) left:(Lsh32x32 x z:(Sub32 (Const32 [32]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 32) => (RotateLeft32 x z)
2711((Add32|Or32|Xor32) right:(Rsh32Ux16 x y) left:(Lsh32x16 x z:(Sub16 (Const16 [32]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 32) => (RotateLeft32 x z)
2712((Add32|Or32|Xor32) right:(Rsh32Ux8 x y) left:(Lsh32x8 x z:(Sub8 (Const8 [32]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 32) => (RotateLeft32 x z)
2713
2714((Add16|Or16|Xor16) left:(Lsh16x64 x y) right:(Rsh16Ux64 x (Sub64 (Const64 [16]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 16) => (RotateLeft16 x y)
2715((Add16|Or16|Xor16) left:(Lsh16x32 x y) right:(Rsh16Ux32 x (Sub32 (Const32 [16]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 16) => (RotateLeft16 x y)
2716((Add16|Or16|Xor16) left:(Lsh16x16 x y) right:(Rsh16Ux16 x (Sub16 (Const16 [16]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 16) => (RotateLeft16 x y)
2717((Add16|Or16|Xor16) left:(Lsh16x8 x y) right:(Rsh16Ux8 x (Sub8 (Const8 [16]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 16) => (RotateLeft16 x y)
2718
2719((Add16|Or16|Xor16) right:(Rsh16Ux64 x y) left:(Lsh16x64 x z:(Sub64 (Const64 [16]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 16) => (RotateLeft16 x z)
2720((Add16|Or16|Xor16) right:(Rsh16Ux32 x y) left:(Lsh16x32 x z:(Sub32 (Const32 [16]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 16) => (RotateLeft16 x z)
2721((Add16|Or16|Xor16) right:(Rsh16Ux16 x y) left:(Lsh16x16 x z:(Sub16 (Const16 [16]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 16) => (RotateLeft16 x z)
2722((Add16|Or16|Xor16) right:(Rsh16Ux8 x y) left:(Lsh16x8 x z:(Sub8 (Const8 [16]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 16) => (RotateLeft16 x z)
2723
2724((Add8|Or8|Xor8) left:(Lsh8x64 x y) right:(Rsh8Ux64 x (Sub64 (Const64 [8]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 8) => (RotateLeft8 x y)
2725((Add8|Or8|Xor8) left:(Lsh8x32 x y) right:(Rsh8Ux32 x (Sub32 (Const32 [8]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 8) => (RotateLeft8 x y)
2726((Add8|Or8|Xor8) left:(Lsh8x16 x y) right:(Rsh8Ux16 x (Sub16 (Const16 [8]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 8) => (RotateLeft8 x y)
2727((Add8|Or8|Xor8) left:(Lsh8x8 x y) right:(Rsh8Ux8 x (Sub8 (Const8 [8]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 8) => (RotateLeft8 x y)
2728
2729((Add8|Or8|Xor8) right:(Rsh8Ux64 x y) left:(Lsh8x64 x z:(Sub64 (Const64 [8]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 8) => (RotateLeft8 x z)
2730((Add8|Or8|Xor8) right:(Rsh8Ux32 x y) left:(Lsh8x32 x z:(Sub32 (Const32 [8]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 8) => (RotateLeft8 x z)
2731((Add8|Or8|Xor8) right:(Rsh8Ux16 x y) left:(Lsh8x16 x z:(Sub16 (Const16 [8]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 8) => (RotateLeft8 x z)
2732((Add8|Or8|Xor8) right:(Rsh8Ux8 x y) left:(Lsh8x8 x z:(Sub8 (Const8 [8]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 8) => (RotateLeft8 x z)
2733
2734// Rotating by y&c, with c a mask that doesn't change the bottom bits, is the same as rotating by y.
2735(RotateLeft64 x (And(64|32|16|8) y (Const(64|32|16|8) [c]))) && c&63 == 63 => (RotateLeft64 x y)
2736(RotateLeft32 x (And(64|32|16|8) y (Const(64|32|16|8) [c]))) && c&31 == 31 => (RotateLeft32 x y)
2737(RotateLeft16 x (And(64|32|16|8) y (Const(64|32|16|8) [c]))) && c&15 == 15 => (RotateLeft16 x y)
2738(RotateLeft8 x (And(64|32|16|8) y (Const(64|32|16|8) [c]))) && c&7 == 7 => (RotateLeft8 x y)
2739
2740// Rotating by -(y&c), with c a mask that doesn't change the bottom bits, is the same as rotating by -y.
2741(RotateLeft64 x (Neg(64|32|16|8) (And(64|32|16|8) y (Const(64|32|16|8) [c])))) && c&63 == 63 => (RotateLeft64 x (Neg(64|32|16|8) <y.Type> y))
2742(RotateLeft32 x (Neg(64|32|16|8) (And(64|32|16|8) y (Const(64|32|16|8) [c])))) && c&31 == 31 => (RotateLeft32 x (Neg(64|32|16|8) <y.Type> y))
2743(RotateLeft16 x (Neg(64|32|16|8) (And(64|32|16|8) y (Const(64|32|16|8) [c])))) && c&15 == 15 => (RotateLeft16 x (Neg(64|32|16|8) <y.Type> y))
2744(RotateLeft8 x (Neg(64|32|16|8) (And(64|32|16|8) y (Const(64|32|16|8) [c])))) && c&7 == 7 => (RotateLeft8 x (Neg(64|32|16|8) <y.Type> y))
2745
2746// Rotating by y+c, with c a multiple of the value width, is the same as rotating by y.
2747(RotateLeft64 x (Add(64|32|16|8) y (Const(64|32|16|8) [c]))) && c&63 == 0 => (RotateLeft64 x y)
2748(RotateLeft32 x (Add(64|32|16|8) y (Const(64|32|16|8) [c]))) && c&31 == 0 => (RotateLeft32 x y)
2749(RotateLeft16 x (Add(64|32|16|8) y (Const(64|32|16|8) [c]))) && c&15 == 0 => (RotateLeft16 x y)
2750(RotateLeft8 x (Add(64|32|16|8) y (Const(64|32|16|8) [c]))) && c&7 == 0 => (RotateLeft8 x y)
2751
2752// Rotating by c-y, with c a multiple of the value width, is the same as rotating by -y.
2753(RotateLeft64 x (Sub(64|32|16|8) (Const(64|32|16|8) [c]) y)) && c&63 == 0 => (RotateLeft64 x (Neg(64|32|16|8) <y.Type> y))
2754(RotateLeft32 x (Sub(64|32|16|8) (Const(64|32|16|8) [c]) y)) && c&31 == 0 => (RotateLeft32 x (Neg(64|32|16|8) <y.Type> y))
2755(RotateLeft16 x (Sub(64|32|16|8) (Const(64|32|16|8) [c]) y)) && c&15 == 0 => (RotateLeft16 x (Neg(64|32|16|8) <y.Type> y))
2756(RotateLeft8 x (Sub(64|32|16|8) (Const(64|32|16|8) [c]) y)) && c&7 == 0 => (RotateLeft8 x (Neg(64|32|16|8) <y.Type> y))
2757
2758// Ensure we don't do Const64 rotates in a 32-bit system.
2759(RotateLeft64 x (Const64 <t> [c])) && config.PtrSize == 4 => (RotateLeft64 x (Const32 <t> [int32(c)]))
2760(RotateLeft32 x (Const64 <t> [c])) && config.PtrSize == 4 => (RotateLeft32 x (Const32 <t> [int32(c)]))
2761(RotateLeft16 x (Const64 <t> [c])) && config.PtrSize == 4 => (RotateLeft16 x (Const32 <t> [int32(c)]))
2762(RotateLeft8 x (Const64 <t> [c])) && config.PtrSize == 4 => (RotateLeft8 x (Const32 <t> [int32(c)]))
2763
2764// Rotating by c, then by d, is the same as rotating by c+d.
2765// We're trading a rotate for an add, which seems generally a good choice. It is especially good when c and d are constants.
2766// This rule is a bit tricky as c and d might be different widths. We handle only cases where they are the same width.
2767(RotateLeft(64|32|16|8) (RotateLeft(64|32|16|8) x c) d) && c.Type.Size() == 8 && d.Type.Size() == 8 => (RotateLeft(64|32|16|8) x (Add64 <c.Type> c d))
2768(RotateLeft(64|32|16|8) (RotateLeft(64|32|16|8) x c) d) && c.Type.Size() == 4 && d.Type.Size() == 4 => (RotateLeft(64|32|16|8) x (Add32 <c.Type> c d))
2769(RotateLeft(64|32|16|8) (RotateLeft(64|32|16|8) x c) d) && c.Type.Size() == 2 && d.Type.Size() == 2 => (RotateLeft(64|32|16|8) x (Add16 <c.Type> c d))
2770(RotateLeft(64|32|16|8) (RotateLeft(64|32|16|8) x c) d) && c.Type.Size() == 1 && d.Type.Size() == 1 => (RotateLeft(64|32|16|8) x (Add8 <c.Type> c d))
2771
2772// Loading constant values from dictionaries and itabs.
2773(Load <typ.BytePtr> (OffPtr [off] (Addr {s} sb) ) _) && isFixedSym(s, off) => (Addr {fixedSym(b.Func, s, off)} sb)
2774(Load <typ.BytePtr> (OffPtr [off] (Convert (Addr {s} sb) _) ) _) && isFixedSym(s, off) => (Addr {fixedSym(b.Func, s, off)} sb)
2775(Load <typ.BytePtr> (OffPtr [off] (ITab (IMake (Addr {s} sb) _))) _) && isFixedSym(s, off) => (Addr {fixedSym(b.Func, s, off)} sb)
2776(Load <typ.BytePtr> (OffPtr [off] (ITab (IMake (Convert (Addr {s} sb) _) _))) _) && isFixedSym(s, off) => (Addr {fixedSym(b.Func, s, off)} sb)
2777(Load <typ.Uintptr> (OffPtr [off] (Addr {s} sb) ) _) && isFixedSym(s, off) => (Addr {fixedSym(b.Func, s, off)} sb)
2778(Load <typ.Uintptr> (OffPtr [off] (Convert (Addr {s} sb) _) ) _) && isFixedSym(s, off) => (Addr {fixedSym(b.Func, s, off)} sb)
2779(Load <typ.Uintptr> (OffPtr [off] (ITab (IMake (Addr {s} sb) _))) _) && isFixedSym(s, off) => (Addr {fixedSym(b.Func, s, off)} sb)
2780(Load <typ.Uintptr> (OffPtr [off] (ITab (IMake (Convert (Addr {s} sb) _) _))) _) && isFixedSym(s, off) => (Addr {fixedSym(b.Func, s, off)} sb)
2781
2782// Loading constant values from runtime._type.hash.
2783(Load <t> (OffPtr [off] (Addr {sym} _) ) _) && t.IsInteger() && t.Size() == 4 && isFixed32(config, sym, off) => (Const32 [fixed32(config, sym, off)])
2784(Load <t> (OffPtr [off] (Convert (Addr {sym} _) _) ) _) && t.IsInteger() && t.Size() == 4 && isFixed32(config, sym, off) => (Const32 [fixed32(config, sym, off)])
2785(Load <t> (OffPtr [off] (ITab (IMake (Addr {sym} _) _))) _) && t.IsInteger() && t.Size() == 4 && isFixed32(config, sym, off) => (Const32 [fixed32(config, sym, off)])
2786(Load <t> (OffPtr [off] (ITab (IMake (Convert (Addr {sym} _) _) _))) _) && t.IsInteger() && t.Size() == 4 && isFixed32(config, sym, off) => (Const32 [fixed32(config, sym, off)])
2787
2788// Calling cmpstring a second time with the same arguments in the
2789// same memory state can reuse the results of the first call.
2790// See issue 61725.
2791// Note that this could pretty easily generalize to any pure function.
2792(SelectN [0] (StaticLECall {f} x y (SelectN [1] c:(StaticLECall {g} x y mem))))
2793 && isSameCall(f, "runtime.cmpstring")
2794 && isSameCall(g, "runtime.cmpstring")
2795=> @c.Block (SelectN [0] <typ.Int> c)
2796
2797// If we don't use the result of cmpstring, might as well not call it.
2798// Note that this could pretty easily generalize to any pure function.
2799(SelectN [1] c:(StaticLECall {f} _ _ mem)) && c.Uses == 1 && isSameCall(f, "runtime.cmpstring") && clobber(c) => mem
2800
2801// We can easily compute the result of efaceeq if
2802// we know the underlying type is pointer-ish.
2803(StaticLECall {f} typ_ x y mem)
2804 && isSameCall(f, "runtime.efaceeq")
2805 && isDirectType(typ_)
2806 && clobber(v)
2807 => (MakeResult (EqPtr x y) mem)
2808
2809// We can easily compute the result of ifaceeq if
2810// we know the underlying type is pointer-ish.
2811(StaticLECall {f} itab x y mem)
2812 && isSameCall(f, "runtime.ifaceeq")
2813 && isDirectIface(itab)
2814 && clobber(v)
2815 => (MakeResult (EqPtr x y) mem)
2816
2817// If we use the result of slicebytetostring in a map lookup operation,
2818// then we don't need to actually do the []byte->string conversion.
2819// We can just use the ptr/len of the byte slice directly as a (temporary) string.
2820//
2821// Note that this does not handle some obscure cases like
2822// m[[2]string{string(b1), string(b2)}]. There is code in ../walk/order.go
2823// which handles some of those cases.
2824(StaticLECall {f} [argsize] typ_ map_ key:(SelectN [0] sbts:(StaticLECall {g} _ ptr len mem)) m:(SelectN [1] sbts))
2825 && (isSameCall(f, "runtime.mapaccess1_faststr")
2826 || isSameCall(f, "runtime.mapaccess2_faststr")
2827 || isSameCall(f, "runtime.mapdelete_faststr"))
2828 && isSameCall(g, "runtime.slicebytetostring")
2829 && key.Uses == 1
2830 && sbts.Uses == 2
2831 && resetCopy(m, mem)
2832 && clobber(sbts)
2833 && clobber(key)
2834=> (StaticLECall {f} [argsize] typ_ map_ (StringMake <typ.String> ptr len) mem)
2835
2836// Similarly to map lookups, also handle unique.Make for strings, which unique.Make will clone.
2837(StaticLECall {f} [argsize] dict_ key:(SelectN [0] sbts:(StaticLECall {g} _ ptr len mem)) m:(SelectN [1] sbts))
2838 && isSameCall(f, "unique.Make[go.shape.string]")
2839 && isSameCall(g, "runtime.slicebytetostring")
2840 && key.Uses == 1
2841 && sbts.Uses == 2
2842 && resetCopy(m, mem)
2843 && clobber(sbts)
2844 && clobber(key)
2845=> (StaticLECall {f} [argsize] dict_ (StringMake <typ.String> ptr len) mem)
View as plain text