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
62(Trunc16to8 (ZeroExt8to16 x)) => x
63(Trunc32to8 (ZeroExt8to32 x)) => x
64(Trunc32to16 (ZeroExt8to32 x)) => (ZeroExt8to16 x)
65(Trunc32to16 (ZeroExt16to32 x)) => x
66(Trunc64to8 (ZeroExt8to64 x)) => x
67(Trunc64to16 (ZeroExt8to64 x)) => (ZeroExt8to16 x)
68(Trunc64to16 (ZeroExt16to64 x)) => x
69(Trunc64to32 (ZeroExt8to64 x)) => (ZeroExt8to32 x)
70(Trunc64to32 (ZeroExt16to64 x)) => (ZeroExt16to32 x)
71(Trunc64to32 (ZeroExt32to64 x)) => x
72(Trunc16to8 (SignExt8to16 x)) => x
73(Trunc32to8 (SignExt8to32 x)) => x
74(Trunc32to16 (SignExt8to32 x)) => (SignExt8to16 x)
75(Trunc32to16 (SignExt16to32 x)) => x
76(Trunc64to8 (SignExt8to64 x)) => x
77(Trunc64to16 (SignExt8to64 x)) => (SignExt8to16 x)
78(Trunc64to16 (SignExt16to64 x)) => x
79(Trunc64to32 (SignExt8to64 x)) => (SignExt8to32 x)
80(Trunc64to32 (SignExt16to64 x)) => (SignExt16to32 x)
81(Trunc64to32 (SignExt32to64 x)) => x
82
83(ZeroExt8to16 (Const8 [c])) => (Const16 [int16( uint8(c))])
84(ZeroExt8to32 (Const8 [c])) => (Const32 [int32( uint8(c))])
85(ZeroExt8to64 (Const8 [c])) => (Const64 [int64( uint8(c))])
86(ZeroExt16to32 (Const16 [c])) => (Const32 [int32(uint16(c))])
87(ZeroExt16to64 (Const16 [c])) => (Const64 [int64(uint16(c))])
88(ZeroExt32to64 (Const32 [c])) => (Const64 [int64(uint32(c))])
89(SignExt8to16 (Const8 [c])) => (Const16 [int16(c)])
90(SignExt8to32 (Const8 [c])) => (Const32 [int32(c)])
91(SignExt8to64 (Const8 [c])) => (Const64 [int64(c)])
92(SignExt16to32 (Const16 [c])) => (Const32 [int32(c)])
93(SignExt16to64 (Const16 [c])) => (Const64 [int64(c)])
94(SignExt32to64 (Const32 [c])) => (Const64 [int64(c)])
95
96(Neg8 (Const8 [c])) => (Const8 [-c])
97(Neg16 (Const16 [c])) => (Const16 [-c])
98(Neg32 (Const32 [c])) => (Const32 [-c])
99(Neg64 (Const64 [c])) => (Const64 [-c])
100(Neg32F (Const32F [c])) && c != 0 => (Const32F [-c])
101(Neg64F (Const64F [c])) && c != 0 => (Const64F [-c])
102
103(Add8 (Const8 [c]) (Const8 [d])) => (Const8 [c+d])
104(Add16 (Const16 [c]) (Const16 [d])) => (Const16 [c+d])
105(Add32 (Const32 [c]) (Const32 [d])) => (Const32 [c+d])
106(Add64 (Const64 [c]) (Const64 [d])) => (Const64 [c+d])
107(Add32F (Const32F [c]) (Const32F [d])) && c+d == c+d => (Const32F [c+d])
108(Add64F (Const64F [c]) (Const64F [d])) && c+d == c+d => (Const64F [c+d])
109(AddPtr <t> x (Const64 [c])) => (OffPtr <t> x [c])
110(AddPtr <t> x (Const32 [c])) => (OffPtr <t> x [int64(c)])
111
112(Sub8 (Const8 [c]) (Const8 [d])) => (Const8 [c-d])
113(Sub16 (Const16 [c]) (Const16 [d])) => (Const16 [c-d])
114(Sub32 (Const32 [c]) (Const32 [d])) => (Const32 [c-d])
115(Sub64 (Const64 [c]) (Const64 [d])) => (Const64 [c-d])
116(Sub32F (Const32F [c]) (Const32F [d])) && c-d == c-d => (Const32F [c-d])
117(Sub64F (Const64F [c]) (Const64F [d])) && c-d == c-d => (Const64F [c-d])
118
119(Mul8 (Const8 [c]) (Const8 [d])) => (Const8 [c*d])
120(Mul16 (Const16 [c]) (Const16 [d])) => (Const16 [c*d])
121(Mul32 (Const32 [c]) (Const32 [d])) => (Const32 [c*d])
122(Mul64 (Const64 [c]) (Const64 [d])) => (Const64 [c*d])
123(Mul32F (Const32F [c]) (Const32F [d])) && c*d == c*d => (Const32F [c*d])
124(Mul64F (Const64F [c]) (Const64F [d])) && c*d == c*d => (Const64F [c*d])
125
126(And8 (Const8 [c]) (Const8 [d])) => (Const8 [c&d])
127(And16 (Const16 [c]) (Const16 [d])) => (Const16 [c&d])
128(And32 (Const32 [c]) (Const32 [d])) => (Const32 [c&d])
129(And64 (Const64 [c]) (Const64 [d])) => (Const64 [c&d])
130
131(Or8 (Const8 [c]) (Const8 [d])) => (Const8 [c|d])
132(Or16 (Const16 [c]) (Const16 [d])) => (Const16 [c|d])
133(Or32 (Const32 [c]) (Const32 [d])) => (Const32 [c|d])
134(Or64 (Const64 [c]) (Const64 [d])) => (Const64 [c|d])
135
136(Xor8 (Const8 [c]) (Const8 [d])) => (Const8 [c^d])
137(Xor16 (Const16 [c]) (Const16 [d])) => (Const16 [c^d])
138(Xor32 (Const32 [c]) (Const32 [d])) => (Const32 [c^d])
139(Xor64 (Const64 [c]) (Const64 [d])) => (Const64 [c^d])
140
141(Ctz64 (Const64 [c])) && config.PtrSize == 4 => (Const32 [int32(ntz64(c))])
142(Ctz32 (Const32 [c])) && config.PtrSize == 4 => (Const32 [int32(ntz32(c))])
143(Ctz16 (Const16 [c])) && config.PtrSize == 4 => (Const32 [int32(ntz16(c))])
144(Ctz8 (Const8 [c])) && config.PtrSize == 4 => (Const32 [int32(ntz8(c))])
145
146(Ctz64 (Const64 [c])) && config.PtrSize == 8 => (Const64 [int64(ntz64(c))])
147(Ctz32 (Const32 [c])) && config.PtrSize == 8 => (Const64 [int64(ntz32(c))])
148(Ctz16 (Const16 [c])) && config.PtrSize == 8 => (Const64 [int64(ntz16(c))])
149(Ctz8 (Const8 [c])) && config.PtrSize == 8 => (Const64 [int64(ntz8(c))])
150
151(Div8 (Const8 [c]) (Const8 [d])) && d != 0 => (Const8 [c/d])
152(Div16 (Const16 [c]) (Const16 [d])) && d != 0 => (Const16 [c/d])
153(Div32 (Const32 [c]) (Const32 [d])) && d != 0 => (Const32 [c/d])
154(Div64 (Const64 [c]) (Const64 [d])) && d != 0 => (Const64 [c/d])
155(Div8u (Const8 [c]) (Const8 [d])) && d != 0 => (Const8 [int8(uint8(c)/uint8(d))])
156(Div16u (Const16 [c]) (Const16 [d])) && d != 0 => (Const16 [int16(uint16(c)/uint16(d))])
157(Div32u (Const32 [c]) (Const32 [d])) && d != 0 => (Const32 [int32(uint32(c)/uint32(d))])
158(Div64u (Const64 [c]) (Const64 [d])) && d != 0 => (Const64 [int64(uint64(c)/uint64(d))])
159(Div32F (Const32F [c]) (Const32F [d])) && c/d == c/d => (Const32F [c/d])
160(Div64F (Const64F [c]) (Const64F [d])) && c/d == c/d => (Const64F [c/d])
161(Select0 (Div128u (Const64 [0]) lo y)) => (Div64u lo y)
162(Select1 (Div128u (Const64 [0]) lo y)) => (Mod64u lo y)
163
164(Not (ConstBool [c])) => (ConstBool [!c])
165
166(Floor (Const64F [c])) => (Const64F [math.Floor(c)])
167(Ceil (Const64F [c])) => (Const64F [math.Ceil(c)])
168(Trunc (Const64F [c])) => (Const64F [math.Trunc(c)])
169(RoundToEven (Const64F [c])) => (Const64F [math.RoundToEven(c)])
170
171// Convert x * 1 to x.
172(Mul(8|16|32|64) (Const(8|16|32|64) [1]) x) => x
173(Select0 (Mul(32|64)uover (Const(32|64) [1]) x)) => x
174(Select1 (Mul(32|64)uover (Const(32|64) [1]) x)) => (ConstBool [false])
175
176// Convert x * -1 to -x.
177(Mul(8|16|32|64) (Const(8|16|32|64) [-1]) x) => (Neg(8|16|32|64) x)
178
179// DeMorgan's Laws
180(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))
181(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))
182
183// Convert multiplication by a power of two to a shift.
184(Mul8 <t> n (Const8 [c])) && isPowerOfTwo8(c) => (Lsh8x64 <t> n (Const64 <typ.UInt64> [log8(c)]))
185(Mul16 <t> n (Const16 [c])) && isPowerOfTwo16(c) => (Lsh16x64 <t> n (Const64 <typ.UInt64> [log16(c)]))
186(Mul32 <t> n (Const32 [c])) && isPowerOfTwo32(c) => (Lsh32x64 <t> n (Const64 <typ.UInt64> [log32(c)]))
187(Mul64 <t> n (Const64 [c])) && isPowerOfTwo64(c) => (Lsh64x64 <t> n (Const64 <typ.UInt64> [log64(c)]))
188(Mul8 <t> n (Const8 [c])) && t.IsSigned() && isPowerOfTwo8(-c) => (Neg8 (Lsh8x64 <t> n (Const64 <typ.UInt64> [log8(-c)])))
189(Mul16 <t> n (Const16 [c])) && t.IsSigned() && isPowerOfTwo16(-c) => (Neg16 (Lsh16x64 <t> n (Const64 <typ.UInt64> [log16(-c)])))
190(Mul32 <t> n (Const32 [c])) && t.IsSigned() && isPowerOfTwo32(-c) => (Neg32 (Lsh32x64 <t> n (Const64 <typ.UInt64> [log32(-c)])))
191(Mul64 <t> n (Const64 [c])) && t.IsSigned() && isPowerOfTwo64(-c) => (Neg64 (Lsh64x64 <t> n (Const64 <typ.UInt64> [log64(-c)])))
192
193(Mod8 (Const8 [c]) (Const8 [d])) && d != 0 => (Const8 [c % d])
194(Mod16 (Const16 [c]) (Const16 [d])) && d != 0 => (Const16 [c % d])
195(Mod32 (Const32 [c]) (Const32 [d])) && d != 0 => (Const32 [c % d])
196(Mod64 (Const64 [c]) (Const64 [d])) && d != 0 => (Const64 [c % d])
197
198(Mod8u (Const8 [c]) (Const8 [d])) && d != 0 => (Const8 [int8(uint8(c) % uint8(d))])
199(Mod16u (Const16 [c]) (Const16 [d])) && d != 0 => (Const16 [int16(uint16(c) % uint16(d))])
200(Mod32u (Const32 [c]) (Const32 [d])) && d != 0 => (Const32 [int32(uint32(c) % uint32(d))])
201(Mod64u (Const64 [c]) (Const64 [d])) && d != 0 => (Const64 [int64(uint64(c) % uint64(d))])
202
203(Lsh64x64 (Const64 [c]) (Const64 [d])) => (Const64 [c << uint64(d)])
204(Rsh64x64 (Const64 [c]) (Const64 [d])) => (Const64 [c >> uint64(d)])
205(Rsh64Ux64 (Const64 [c]) (Const64 [d])) => (Const64 [int64(uint64(c) >> uint64(d))])
206(Lsh32x64 (Const32 [c]) (Const64 [d])) => (Const32 [c << uint64(d)])
207(Rsh32x64 (Const32 [c]) (Const64 [d])) => (Const32 [c >> uint64(d)])
208(Rsh32Ux64 (Const32 [c]) (Const64 [d])) => (Const32 [int32(uint32(c) >> uint64(d))])
209(Lsh16x64 (Const16 [c]) (Const64 [d])) => (Const16 [c << uint64(d)])
210(Rsh16x64 (Const16 [c]) (Const64 [d])) => (Const16 [c >> uint64(d)])
211(Rsh16Ux64 (Const16 [c]) (Const64 [d])) => (Const16 [int16(uint16(c) >> uint64(d))])
212(Lsh8x64 (Const8 [c]) (Const64 [d])) => (Const8 [c << uint64(d)])
213(Rsh8x64 (Const8 [c]) (Const64 [d])) => (Const8 [c >> uint64(d)])
214(Rsh8Ux64 (Const8 [c]) (Const64 [d])) => (Const8 [int8(uint8(c) >> uint64(d))])
215
216// Fold IsInBounds when the range of the index cannot exceed the limit.
217(IsInBounds (ZeroExt8to32 _) (Const32 [c])) && (1 << 8) <= c => (ConstBool [true])
218(IsInBounds (ZeroExt8to64 _) (Const64 [c])) && (1 << 8) <= c => (ConstBool [true])
219(IsInBounds (ZeroExt16to32 _) (Const32 [c])) && (1 << 16) <= c => (ConstBool [true])
220(IsInBounds (ZeroExt16to64 _) (Const64 [c])) && (1 << 16) <= c => (ConstBool [true])
221(IsInBounds x x) => (ConstBool [false])
222(IsInBounds (And8 (Const8 [c]) _) (Const8 [d])) && 0 <= c && c < d => (ConstBool [true])
223(IsInBounds (ZeroExt8to16 (And8 (Const8 [c]) _)) (Const16 [d])) && 0 <= c && int16(c) < d => (ConstBool [true])
224(IsInBounds (ZeroExt8to32 (And8 (Const8 [c]) _)) (Const32 [d])) && 0 <= c && int32(c) < d => (ConstBool [true])
225(IsInBounds (ZeroExt8to64 (And8 (Const8 [c]) _)) (Const64 [d])) && 0 <= c && int64(c) < d => (ConstBool [true])
226(IsInBounds (And16 (Const16 [c]) _) (Const16 [d])) && 0 <= c && c < d => (ConstBool [true])
227(IsInBounds (ZeroExt16to32 (And16 (Const16 [c]) _)) (Const32 [d])) && 0 <= c && int32(c) < d => (ConstBool [true])
228(IsInBounds (ZeroExt16to64 (And16 (Const16 [c]) _)) (Const64 [d])) && 0 <= c && int64(c) < d => (ConstBool [true])
229(IsInBounds (And32 (Const32 [c]) _) (Const32 [d])) && 0 <= c && c < d => (ConstBool [true])
230(IsInBounds (ZeroExt32to64 (And32 (Const32 [c]) _)) (Const64 [d])) && 0 <= c && int64(c) < d => (ConstBool [true])
231(IsInBounds (And64 (Const64 [c]) _) (Const64 [d])) && 0 <= c && c < d => (ConstBool [true])
232(IsInBounds (Const32 [c]) (Const32 [d])) => (ConstBool [0 <= c && c < d])
233(IsInBounds (Const64 [c]) (Const64 [d])) => (ConstBool [0 <= c && c < d])
234// (Mod64u x y) is always between 0 (inclusive) and y (exclusive).
235(IsInBounds (Mod32u _ y) y) => (ConstBool [true])
236(IsInBounds (Mod64u _ y) y) => (ConstBool [true])
237// Right shifting an unsigned number limits its value.
238(IsInBounds (ZeroExt8to64 (Rsh8Ux64 _ (Const64 [c]))) (Const64 [d])) && 0 < c && c < 8 && 1<<uint( 8-c)-1 < d => (ConstBool [true])
239(IsInBounds (ZeroExt8to32 (Rsh8Ux64 _ (Const64 [c]))) (Const32 [d])) && 0 < c && c < 8 && 1<<uint( 8-c)-1 < d => (ConstBool [true])
240(IsInBounds (ZeroExt8to16 (Rsh8Ux64 _ (Const64 [c]))) (Const16 [d])) && 0 < c && c < 8 && 1<<uint( 8-c)-1 < d => (ConstBool [true])
241(IsInBounds (Rsh8Ux64 _ (Const64 [c])) (Const64 [d])) && 0 < c && c < 8 && 1<<uint( 8-c)-1 < d => (ConstBool [true])
242(IsInBounds (ZeroExt16to64 (Rsh16Ux64 _ (Const64 [c]))) (Const64 [d])) && 0 < c && c < 16 && 1<<uint(16-c)-1 < d => (ConstBool [true])
243(IsInBounds (ZeroExt16to32 (Rsh16Ux64 _ (Const64 [c]))) (Const64 [d])) && 0 < c && c < 16 && 1<<uint(16-c)-1 < d => (ConstBool [true])
244(IsInBounds (Rsh16Ux64 _ (Const64 [c])) (Const64 [d])) && 0 < c && c < 16 && 1<<uint(16-c)-1 < d => (ConstBool [true])
245(IsInBounds (ZeroExt32to64 (Rsh32Ux64 _ (Const64 [c]))) (Const64 [d])) && 0 < c && c < 32 && 1<<uint(32-c)-1 < d => (ConstBool [true])
246(IsInBounds (Rsh32Ux64 _ (Const64 [c])) (Const64 [d])) && 0 < c && c < 32 && 1<<uint(32-c)-1 < d => (ConstBool [true])
247(IsInBounds (Rsh64Ux64 _ (Const64 [c])) (Const64 [d])) && 0 < c && c < 64 && 1<<uint(64-c)-1 < d => (ConstBool [true])
248
249(IsSliceInBounds x x) => (ConstBool [true])
250(IsSliceInBounds (And32 (Const32 [c]) _) (Const32 [d])) && 0 <= c && c <= d => (ConstBool [true])
251(IsSliceInBounds (And64 (Const64 [c]) _) (Const64 [d])) && 0 <= c && c <= d => (ConstBool [true])
252(IsSliceInBounds (Const32 [0]) _) => (ConstBool [true])
253(IsSliceInBounds (Const64 [0]) _) => (ConstBool [true])
254(IsSliceInBounds (Const32 [c]) (Const32 [d])) => (ConstBool [0 <= c && c <= d])
255(IsSliceInBounds (Const64 [c]) (Const64 [d])) => (ConstBool [0 <= c && c <= d])
256(IsSliceInBounds (SliceLen x) (SliceCap x)) => (ConstBool [true])
257
258(Eq(64|32|16|8) x x) => (ConstBool [true])
259(EqB (ConstBool [c]) (ConstBool [d])) => (ConstBool [c == d])
260(EqB (ConstBool [false]) x) => (Not x)
261(EqB (ConstBool [true]) x) => x
262
263(Neq(64|32|16|8) x x) => (ConstBool [false])
264(NeqB (ConstBool [c]) (ConstBool [d])) => (ConstBool [c != d])
265(NeqB (ConstBool [false]) x) => x
266(NeqB (ConstBool [true]) x) => (Not x)
267(NeqB (Not x) (Not y)) => (NeqB x y)
268
269(Eq64 (Const64 <t> [c]) (Add64 (Const64 <t> [d]) x)) => (Eq64 (Const64 <t> [c-d]) x)
270(Eq32 (Const32 <t> [c]) (Add32 (Const32 <t> [d]) x)) => (Eq32 (Const32 <t> [c-d]) x)
271(Eq16 (Const16 <t> [c]) (Add16 (Const16 <t> [d]) x)) => (Eq16 (Const16 <t> [c-d]) x)
272(Eq8 (Const8 <t> [c]) (Add8 (Const8 <t> [d]) x)) => (Eq8 (Const8 <t> [c-d]) x)
273
274(Neq64 (Const64 <t> [c]) (Add64 (Const64 <t> [d]) x)) => (Neq64 (Const64 <t> [c-d]) x)
275(Neq32 (Const32 <t> [c]) (Add32 (Const32 <t> [d]) x)) => (Neq32 (Const32 <t> [c-d]) x)
276(Neq16 (Const16 <t> [c]) (Add16 (Const16 <t> [d]) x)) => (Neq16 (Const16 <t> [c-d]) x)
277(Neq8 (Const8 <t> [c]) (Add8 (Const8 <t> [d]) x)) => (Neq8 (Const8 <t> [c-d]) x)
278
279// signed integer range: ( c <= x && x (<|<=) d ) -> ( unsigned(x-c) (<|<=) unsigned(d-c) )
280(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]))
281(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]))
282(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]))
283(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]))
284
285// signed integer range: ( c < x && x (<|<=) d ) -> ( unsigned(x-(c+1)) (<|<=) unsigned(d-(c+1)) )
286(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]))
287(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]))
288(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]))
289(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]))
290
291// unsigned integer range: ( c <= x && x (<|<=) d ) -> ( x-c (<|<=) d-c )
292(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]))
293(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]))
294(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]))
295(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]))
296
297// unsigned integer range: ( c < x && x (<|<=) d ) -> ( x-(c+1) (<|<=) d-(c+1) )
298(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]))
299(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]))
300(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]))
301(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]))
302
303// signed integer range: ( c (<|<=) x || x < d ) -> ( unsigned(c-d) (<|<=) unsigned(x-d) )
304(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])))
305(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])))
306(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])))
307(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])))
308
309// signed integer range: ( c (<|<=) x || x <= d ) -> ( unsigned(c-(d+1)) (<|<=) unsigned(x-(d+1)) )
310(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])))
311(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])))
312(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])))
313(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])))
314
315// unsigned integer range: ( c (<|<=) x || x < d ) -> ( c-d (<|<=) x-d )
316(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])))
317(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])))
318(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])))
319(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])))
320
321// unsigned integer range: ( c (<|<=) x || x <= d ) -> ( c-(d+1) (<|<=) x-(d+1) )
322(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])))
323(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])))
324(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])))
325(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])))
326
327// Canonicalize x-const to x+(-const)
328(Sub64 x (Const64 <t> [c])) && x.Op != OpConst64 => (Add64 (Const64 <t> [-c]) x)
329(Sub32 x (Const32 <t> [c])) && x.Op != OpConst32 => (Add32 (Const32 <t> [-c]) x)
330(Sub16 x (Const16 <t> [c])) && x.Op != OpConst16 => (Add16 (Const16 <t> [-c]) x)
331(Sub8 x (Const8 <t> [c])) && x.Op != OpConst8 => (Add8 (Const8 <t> [-c]) x)
332
333// fold negation into comparison operators
334(Not (Eq(64|32|16|8|B|Ptr|64F|32F) x y)) => (Neq(64|32|16|8|B|Ptr|64F|32F) x y)
335(Not (Neq(64|32|16|8|B|Ptr|64F|32F) x y)) => (Eq(64|32|16|8|B|Ptr|64F|32F) x y)
336
337(Not (Less(64|32|16|8) x y)) => (Leq(64|32|16|8) y x)
338(Not (Less(64|32|16|8)U x y)) => (Leq(64|32|16|8)U y x)
339(Not (Leq(64|32|16|8) x y)) => (Less(64|32|16|8) y x)
340(Not (Leq(64|32|16|8)U x y)) => (Less(64|32|16|8)U y x)
341
342// Distribute multiplication c * (d+x) -> c*d + c*x. Useful for:
343// a[i].b = ...; a[i+1].b = ...
344(Mul64 (Const64 <t> [c]) (Add64 <t> (Const64 <t> [d]) x)) =>
345 (Add64 (Const64 <t> [c*d]) (Mul64 <t> (Const64 <t> [c]) x))
346(Mul32 (Const32 <t> [c]) (Add32 <t> (Const32 <t> [d]) x)) =>
347 (Add32 (Const32 <t> [c*d]) (Mul32 <t> (Const32 <t> [c]) x))
348
349// Rewrite x*y ± x*z to x*(y±z)
350(Add(64|32|16|8) <t> (Mul(64|32|16|8) x y) (Mul(64|32|16|8) x z))
351 => (Mul(64|32|16|8) x (Add(64|32|16|8) <t> y z))
352(Sub(64|32|16|8) <t> (Mul(64|32|16|8) x y) (Mul(64|32|16|8) x z))
353 => (Mul(64|32|16|8) x (Sub(64|32|16|8) <t> y z))
354
355// rewrite shifts of 8/16/32 bit consts into 64 bit consts to reduce
356// the number of the other rewrite rules for const shifts
357(Lsh64x32 <t> x (Const32 [c])) => (Lsh64x64 x (Const64 <t> [int64(uint32(c))]))
358(Lsh64x16 <t> x (Const16 [c])) => (Lsh64x64 x (Const64 <t> [int64(uint16(c))]))
359(Lsh64x8 <t> x (Const8 [c])) => (Lsh64x64 x (Const64 <t> [int64(uint8(c))]))
360(Rsh64x32 <t> x (Const32 [c])) => (Rsh64x64 x (Const64 <t> [int64(uint32(c))]))
361(Rsh64x16 <t> x (Const16 [c])) => (Rsh64x64 x (Const64 <t> [int64(uint16(c))]))
362(Rsh64x8 <t> x (Const8 [c])) => (Rsh64x64 x (Const64 <t> [int64(uint8(c))]))
363(Rsh64Ux32 <t> x (Const32 [c])) => (Rsh64Ux64 x (Const64 <t> [int64(uint32(c))]))
364(Rsh64Ux16 <t> x (Const16 [c])) => (Rsh64Ux64 x (Const64 <t> [int64(uint16(c))]))
365(Rsh64Ux8 <t> x (Const8 [c])) => (Rsh64Ux64 x (Const64 <t> [int64(uint8(c))]))
366
367(Lsh32x32 <t> x (Const32 [c])) => (Lsh32x64 x (Const64 <t> [int64(uint32(c))]))
368(Lsh32x16 <t> x (Const16 [c])) => (Lsh32x64 x (Const64 <t> [int64(uint16(c))]))
369(Lsh32x8 <t> x (Const8 [c])) => (Lsh32x64 x (Const64 <t> [int64(uint8(c))]))
370(Rsh32x32 <t> x (Const32 [c])) => (Rsh32x64 x (Const64 <t> [int64(uint32(c))]))
371(Rsh32x16 <t> x (Const16 [c])) => (Rsh32x64 x (Const64 <t> [int64(uint16(c))]))
372(Rsh32x8 <t> x (Const8 [c])) => (Rsh32x64 x (Const64 <t> [int64(uint8(c))]))
373(Rsh32Ux32 <t> x (Const32 [c])) => (Rsh32Ux64 x (Const64 <t> [int64(uint32(c))]))
374(Rsh32Ux16 <t> x (Const16 [c])) => (Rsh32Ux64 x (Const64 <t> [int64(uint16(c))]))
375(Rsh32Ux8 <t> x (Const8 [c])) => (Rsh32Ux64 x (Const64 <t> [int64(uint8(c))]))
376
377(Lsh16x32 <t> x (Const32 [c])) => (Lsh16x64 x (Const64 <t> [int64(uint32(c))]))
378(Lsh16x16 <t> x (Const16 [c])) => (Lsh16x64 x (Const64 <t> [int64(uint16(c))]))
379(Lsh16x8 <t> x (Const8 [c])) => (Lsh16x64 x (Const64 <t> [int64(uint8(c))]))
380(Rsh16x32 <t> x (Const32 [c])) => (Rsh16x64 x (Const64 <t> [int64(uint32(c))]))
381(Rsh16x16 <t> x (Const16 [c])) => (Rsh16x64 x (Const64 <t> [int64(uint16(c))]))
382(Rsh16x8 <t> x (Const8 [c])) => (Rsh16x64 x (Const64 <t> [int64(uint8(c))]))
383(Rsh16Ux32 <t> x (Const32 [c])) => (Rsh16Ux64 x (Const64 <t> [int64(uint32(c))]))
384(Rsh16Ux16 <t> x (Const16 [c])) => (Rsh16Ux64 x (Const64 <t> [int64(uint16(c))]))
385(Rsh16Ux8 <t> x (Const8 [c])) => (Rsh16Ux64 x (Const64 <t> [int64(uint8(c))]))
386
387(Lsh8x32 <t> x (Const32 [c])) => (Lsh8x64 x (Const64 <t> [int64(uint32(c))]))
388(Lsh8x16 <t> x (Const16 [c])) => (Lsh8x64 x (Const64 <t> [int64(uint16(c))]))
389(Lsh8x8 <t> x (Const8 [c])) => (Lsh8x64 x (Const64 <t> [int64(uint8(c))]))
390(Rsh8x32 <t> x (Const32 [c])) => (Rsh8x64 x (Const64 <t> [int64(uint32(c))]))
391(Rsh8x16 <t> x (Const16 [c])) => (Rsh8x64 x (Const64 <t> [int64(uint16(c))]))
392(Rsh8x8 <t> x (Const8 [c])) => (Rsh8x64 x (Const64 <t> [int64(uint8(c))]))
393(Rsh8Ux32 <t> x (Const32 [c])) => (Rsh8Ux64 x (Const64 <t> [int64(uint32(c))]))
394(Rsh8Ux16 <t> x (Const16 [c])) => (Rsh8Ux64 x (Const64 <t> [int64(uint16(c))]))
395(Rsh8Ux8 <t> x (Const8 [c])) => (Rsh8Ux64 x (Const64 <t> [int64(uint8(c))]))
396
397// shifts by zero
398(Lsh(64|32|16|8)x64 x (Const64 [0])) => x
399(Rsh(64|32|16|8)x64 x (Const64 [0])) => x
400(Rsh(64|32|16|8)Ux64 x (Const64 [0])) => x
401
402// rotates by multiples of register width
403(RotateLeft64 x (Const64 [c])) && c%64 == 0 => x
404(RotateLeft32 x (Const32 [c])) && c%32 == 0 => x
405(RotateLeft16 x (Const16 [c])) && c%16 == 0 => x
406(RotateLeft8 x (Const8 [c])) && c%8 == 0 => x
407
408// zero shifted
409(Lsh64x(64|32|16|8) (Const64 [0]) _) => (Const64 [0])
410(Rsh64x(64|32|16|8) (Const64 [0]) _) => (Const64 [0])
411(Rsh64Ux(64|32|16|8) (Const64 [0]) _) => (Const64 [0])
412(Lsh32x(64|32|16|8) (Const32 [0]) _) => (Const32 [0])
413(Rsh32x(64|32|16|8) (Const32 [0]) _) => (Const32 [0])
414(Rsh32Ux(64|32|16|8) (Const32 [0]) _) => (Const32 [0])
415(Lsh16x(64|32|16|8) (Const16 [0]) _) => (Const16 [0])
416(Rsh16x(64|32|16|8) (Const16 [0]) _) => (Const16 [0])
417(Rsh16Ux(64|32|16|8) (Const16 [0]) _) => (Const16 [0])
418(Lsh8x(64|32|16|8) (Const8 [0]) _) => (Const8 [0])
419(Rsh8x(64|32|16|8) (Const8 [0]) _) => (Const8 [0])
420(Rsh8Ux(64|32|16|8) (Const8 [0]) _) => (Const8 [0])
421
422// large left shifts of all values, and right shifts of unsigned values
423((Lsh64|Rsh64U)x64 _ (Const64 [c])) && uint64(c) >= 64 => (Const64 [0])
424((Lsh32|Rsh32U)x64 _ (Const64 [c])) && uint64(c) >= 32 => (Const32 [0])
425((Lsh16|Rsh16U)x64 _ (Const64 [c])) && uint64(c) >= 16 => (Const16 [0])
426((Lsh8|Rsh8U)x64 _ (Const64 [c])) && uint64(c) >= 8 => (Const8 [0])
427
428// combine const shifts
429(Lsh64x64 <t> (Lsh64x64 x (Const64 [c])) (Const64 [d])) && !uaddOvf(c,d) => (Lsh64x64 x (Const64 <t> [c+d]))
430(Lsh32x64 <t> (Lsh32x64 x (Const64 [c])) (Const64 [d])) && !uaddOvf(c,d) => (Lsh32x64 x (Const64 <t> [c+d]))
431(Lsh16x64 <t> (Lsh16x64 x (Const64 [c])) (Const64 [d])) && !uaddOvf(c,d) => (Lsh16x64 x (Const64 <t> [c+d]))
432(Lsh8x64 <t> (Lsh8x64 x (Const64 [c])) (Const64 [d])) && !uaddOvf(c,d) => (Lsh8x64 x (Const64 <t> [c+d]))
433
434(Rsh64x64 <t> (Rsh64x64 x (Const64 [c])) (Const64 [d])) && !uaddOvf(c,d) => (Rsh64x64 x (Const64 <t> [c+d]))
435(Rsh32x64 <t> (Rsh32x64 x (Const64 [c])) (Const64 [d])) && !uaddOvf(c,d) => (Rsh32x64 x (Const64 <t> [c+d]))
436(Rsh16x64 <t> (Rsh16x64 x (Const64 [c])) (Const64 [d])) && !uaddOvf(c,d) => (Rsh16x64 x (Const64 <t> [c+d]))
437(Rsh8x64 <t> (Rsh8x64 x (Const64 [c])) (Const64 [d])) && !uaddOvf(c,d) => (Rsh8x64 x (Const64 <t> [c+d]))
438
439(Rsh64Ux64 <t> (Rsh64Ux64 x (Const64 [c])) (Const64 [d])) && !uaddOvf(c,d) => (Rsh64Ux64 x (Const64 <t> [c+d]))
440(Rsh32Ux64 <t> (Rsh32Ux64 x (Const64 [c])) (Const64 [d])) && !uaddOvf(c,d) => (Rsh32Ux64 x (Const64 <t> [c+d]))
441(Rsh16Ux64 <t> (Rsh16Ux64 x (Const64 [c])) (Const64 [d])) && !uaddOvf(c,d) => (Rsh16Ux64 x (Const64 <t> [c+d]))
442(Rsh8Ux64 <t> (Rsh8Ux64 x (Const64 [c])) (Const64 [d])) && !uaddOvf(c,d) => (Rsh8Ux64 x (Const64 <t> [c+d]))
443
444// Remove signed right shift before an unsigned right shift that extracts the sign bit.
445(Rsh8Ux64 (Rsh8x64 x _) (Const64 <t> [7] )) => (Rsh8Ux64 x (Const64 <t> [7] ))
446(Rsh16Ux64 (Rsh16x64 x _) (Const64 <t> [15])) => (Rsh16Ux64 x (Const64 <t> [15]))
447(Rsh32Ux64 (Rsh32x64 x _) (Const64 <t> [31])) => (Rsh32Ux64 x (Const64 <t> [31]))
448(Rsh64Ux64 (Rsh64x64 x _) (Const64 <t> [63])) => (Rsh64Ux64 x (Const64 <t> [63]))
449
450// Convert x>>c<<c to x&^(1<<c-1)
451(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]))
452(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]))
453(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]))
454(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]))
455// similarly for x<<c>>c
456(Rsh64Ux64 i:(Lsh64x64 x (Const64 [c])) (Const64 [c])) && c >= 0 && c < 64 && i.Uses == 1 => (And64 x (Const64 <v.Type> [int64(^uint64(0)>>c)]))
457(Rsh32Ux64 i:(Lsh32x64 x (Const64 [c])) (Const64 [c])) && c >= 0 && c < 32 && i.Uses == 1 => (And32 x (Const32 <v.Type> [int32(^uint32(0)>>c)]))
458(Rsh16Ux64 i:(Lsh16x64 x (Const64 [c])) (Const64 [c])) && c >= 0 && c < 16 && i.Uses == 1 => (And16 x (Const16 <v.Type> [int16(^uint16(0)>>c)]))
459(Rsh8Ux64 i:(Lsh8x64 x (Const64 [c])) (Const64 [c])) && c >= 0 && c < 8 && i.Uses == 1 => (And8 x (Const8 <v.Type> [int8 (^uint8 (0)>>c)]))
460
461// ((x >> c1) << c2) >> c3
462(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]))
463 && uint64(c1) >= uint64(c2) && uint64(c3) >= uint64(c2) && !uaddOvf(c1-c2, c3)
464 => (Rsh(64|32|16|8)Ux64 x (Const64 <typ.UInt64> [c1-c2+c3]))
465
466// ((x << c1) >> c2) << c3
467(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]))
468 && uint64(c1) >= uint64(c2) && uint64(c3) >= uint64(c2) && !uaddOvf(c1-c2, c3)
469 => (Lsh(64|32|16|8)x64 x (Const64 <typ.UInt64> [c1-c2+c3]))
470
471// (x >> c) & uppermask = 0
472(And64 (Const64 [m]) (Rsh64Ux64 _ (Const64 [c]))) && c >= int64(64-ntz64(m)) => (Const64 [0])
473(And32 (Const32 [m]) (Rsh32Ux64 _ (Const64 [c]))) && c >= int64(32-ntz32(m)) => (Const32 [0])
474(And16 (Const16 [m]) (Rsh16Ux64 _ (Const64 [c]))) && c >= int64(16-ntz16(m)) => (Const16 [0])
475(And8 (Const8 [m]) (Rsh8Ux64 _ (Const64 [c]))) && c >= int64(8-ntz8(m)) => (Const8 [0])
476
477// (x << c) & lowermask = 0
478(And64 (Const64 [m]) (Lsh64x64 _ (Const64 [c]))) && c >= int64(64-nlz64(m)) => (Const64 [0])
479(And32 (Const32 [m]) (Lsh32x64 _ (Const64 [c]))) && c >= int64(32-nlz32(m)) => (Const32 [0])
480(And16 (Const16 [m]) (Lsh16x64 _ (Const64 [c]))) && c >= int64(16-nlz16(m)) => (Const16 [0])
481(And8 (Const8 [m]) (Lsh8x64 _ (Const64 [c]))) && c >= int64(8-nlz8(m)) => (Const8 [0])
482
483// replace shifts with zero extensions
484(Rsh16Ux64 (Lsh16x64 x (Const64 [8])) (Const64 [8])) => (ZeroExt8to16 (Trunc16to8 <typ.UInt8> x))
485(Rsh32Ux64 (Lsh32x64 x (Const64 [24])) (Const64 [24])) => (ZeroExt8to32 (Trunc32to8 <typ.UInt8> x))
486(Rsh64Ux64 (Lsh64x64 x (Const64 [56])) (Const64 [56])) => (ZeroExt8to64 (Trunc64to8 <typ.UInt8> x))
487(Rsh32Ux64 (Lsh32x64 x (Const64 [16])) (Const64 [16])) => (ZeroExt16to32 (Trunc32to16 <typ.UInt16> x))
488(Rsh64Ux64 (Lsh64x64 x (Const64 [48])) (Const64 [48])) => (ZeroExt16to64 (Trunc64to16 <typ.UInt16> x))
489(Rsh64Ux64 (Lsh64x64 x (Const64 [32])) (Const64 [32])) => (ZeroExt32to64 (Trunc64to32 <typ.UInt32> x))
490
491// replace shifts with sign extensions
492(Rsh16x64 (Lsh16x64 x (Const64 [8])) (Const64 [8])) => (SignExt8to16 (Trunc16to8 <typ.Int8> x))
493(Rsh32x64 (Lsh32x64 x (Const64 [24])) (Const64 [24])) => (SignExt8to32 (Trunc32to8 <typ.Int8> x))
494(Rsh64x64 (Lsh64x64 x (Const64 [56])) (Const64 [56])) => (SignExt8to64 (Trunc64to8 <typ.Int8> x))
495(Rsh32x64 (Lsh32x64 x (Const64 [16])) (Const64 [16])) => (SignExt16to32 (Trunc32to16 <typ.Int16> x))
496(Rsh64x64 (Lsh64x64 x (Const64 [48])) (Const64 [48])) => (SignExt16to64 (Trunc64to16 <typ.Int16> x))
497(Rsh64x64 (Lsh64x64 x (Const64 [32])) (Const64 [32])) => (SignExt32to64 (Trunc64to32 <typ.Int32> x))
498
499// constant comparisons
500(Eq(64|32|16|8) (Const(64|32|16|8) [c]) (Const(64|32|16|8) [d])) => (ConstBool [c == d])
501(Neq(64|32|16|8) (Const(64|32|16|8) [c]) (Const(64|32|16|8) [d])) => (ConstBool [c != d])
502(Less(64|32|16|8) (Const(64|32|16|8) [c]) (Const(64|32|16|8) [d])) => (ConstBool [c < d])
503(Leq(64|32|16|8) (Const(64|32|16|8) [c]) (Const(64|32|16|8) [d])) => (ConstBool [c <= d])
504
505(Less64U (Const64 [c]) (Const64 [d])) => (ConstBool [uint64(c) < uint64(d)])
506(Less32U (Const32 [c]) (Const32 [d])) => (ConstBool [uint32(c) < uint32(d)])
507(Less16U (Const16 [c]) (Const16 [d])) => (ConstBool [uint16(c) < uint16(d)])
508(Less8U (Const8 [c]) (Const8 [d])) => (ConstBool [ uint8(c) < uint8(d)])
509
510(Leq64U (Const64 [c]) (Const64 [d])) => (ConstBool [uint64(c) <= uint64(d)])
511(Leq32U (Const32 [c]) (Const32 [d])) => (ConstBool [uint32(c) <= uint32(d)])
512(Leq16U (Const16 [c]) (Const16 [d])) => (ConstBool [uint16(c) <= uint16(d)])
513(Leq8U (Const8 [c]) (Const8 [d])) => (ConstBool [ uint8(c) <= uint8(d)])
514
515(Leq8 (Const8 [0]) (And8 _ (Const8 [c]))) && c >= 0 => (ConstBool [true])
516(Leq16 (Const16 [0]) (And16 _ (Const16 [c]))) && c >= 0 => (ConstBool [true])
517(Leq32 (Const32 [0]) (And32 _ (Const32 [c]))) && c >= 0 => (ConstBool [true])
518(Leq64 (Const64 [0]) (And64 _ (Const64 [c]))) && c >= 0 => (ConstBool [true])
519
520(Leq8 (Const8 [0]) (Rsh8Ux64 _ (Const64 [c]))) && c > 0 => (ConstBool [true])
521(Leq16 (Const16 [0]) (Rsh16Ux64 _ (Const64 [c]))) && c > 0 => (ConstBool [true])
522(Leq32 (Const32 [0]) (Rsh32Ux64 _ (Const64 [c]))) && c > 0 => (ConstBool [true])
523(Leq64 (Const64 [0]) (Rsh64Ux64 _ (Const64 [c]))) && c > 0 => (ConstBool [true])
524
525// prefer equalities with zero
526(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)
527(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)
528(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)
529(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)
530
531// prefer comparisons with zero
532(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]))
533(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]))
534(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)
535(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)
536
537// constant floating point comparisons
538(Eq32F (Const32F [c]) (Const32F [d])) => (ConstBool [c == d])
539(Eq64F (Const64F [c]) (Const64F [d])) => (ConstBool [c == d])
540(Neq32F (Const32F [c]) (Const32F [d])) => (ConstBool [c != d])
541(Neq64F (Const64F [c]) (Const64F [d])) => (ConstBool [c != d])
542(Less32F (Const32F [c]) (Const32F [d])) => (ConstBool [c < d])
543(Less64F (Const64F [c]) (Const64F [d])) => (ConstBool [c < d])
544(Leq32F (Const32F [c]) (Const32F [d])) => (ConstBool [c <= d])
545(Leq64F (Const64F [c]) (Const64F [d])) => (ConstBool [c <= d])
546
547// simplifications
548(Or(64|32|16|8) x x) => x
549(Or(64|32|16|8) (Const(64|32|16|8) [0]) x) => x
550(Or(64|32|16|8) (Const(64|32|16|8) [-1]) _) => (Const(64|32|16|8) [-1])
551(Or(64|32|16|8) (Com(64|32|16|8) x) x) => (Const(64|32|16|8) [-1])
552
553(And(64|32|16|8) x x) => x
554(And(64|32|16|8) (Const(64|32|16|8) [-1]) x) => x
555(And(64|32|16|8) (Const(64|32|16|8) [0]) _) => (Const(64|32|16|8) [0])
556(And(64|32|16|8) (Com(64|32|16|8) x) x) => (Const(64|32|16|8) [0])
557
558(Xor(64|32|16|8) x x) => (Const(64|32|16|8) [0])
559(Xor(64|32|16|8) (Const(64|32|16|8) [0]) x) => x
560(Xor(64|32|16|8) (Com(64|32|16|8) x) x) => (Const(64|32|16|8) [-1])
561
562(Add(64|32|16|8) (Const(64|32|16|8) [0]) x) => x
563(Sub(64|32|16|8) x x) => (Const(64|32|16|8) [0])
564(Mul(64|32|16|8) (Const(64|32|16|8) [0]) _) => (Const(64|32|16|8) [0])
565(Select0 (Mul(64|32)uover (Const(64|32) [0]) x)) => (Const(64|32) [0])
566(Select1 (Mul(64|32)uover (Const(64|32) [0]) x)) => (ConstBool [false])
567
568(Com(64|32|16|8) (Com(64|32|16|8) x)) => x
569(Com(64|32|16|8) (Const(64|32|16|8) [c])) => (Const(64|32|16|8) [^c])
570
571(Neg(64|32|16|8) (Sub(64|32|16|8) x y)) => (Sub(64|32|16|8) y x)
572(Add(64|32|16|8) x (Neg(64|32|16|8) y)) => (Sub(64|32|16|8) x y)
573
574(Xor(64|32|16|8) (Const(64|32|16|8) [-1]) x) => (Com(64|32|16|8) x)
575
576(Sub(64|32|16|8) (Neg(64|32|16|8) x) (Com(64|32|16|8) x)) => (Const(64|32|16|8) [1])
577(Sub(64|32|16|8) (Com(64|32|16|8) x) (Neg(64|32|16|8) x)) => (Const(64|32|16|8) [-1])
578(Add(64|32|16|8) (Com(64|32|16|8) x) x) => (Const(64|32|16|8) [-1])
579
580// Simplification when involving common integer
581// (t + x) - (t + y) == x - y
582// (t + x) - (y + t) == x - y
583// (x + t) - (y + t) == x - y
584// (x + t) - (t + y) == x - y
585// (x - t) + (t + y) == x + y
586// (x - t) + (y + t) == x + y
587(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)
588(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)
589
590// ^(x-1) == ^x+1 == -x
591(Add(64|32|16|8) (Const(64|32|16|8) [1]) (Com(64|32|16|8) x)) => (Neg(64|32|16|8) x)
592(Com(64|32|16|8) (Add(64|32|16|8) (Const(64|32|16|8) [-1]) x)) => (Neg(64|32|16|8) x)
593
594// -(-x) == x
595(Neg(64|32|16|8) (Neg(64|32|16|8) x)) => x
596
597// -^x == x+1
598(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)
599
600(And(64|32|16|8) x (And(64|32|16|8) x y)) => (And(64|32|16|8) x y)
601(Or(64|32|16|8) x (Or(64|32|16|8) x y)) => (Or(64|32|16|8) x y)
602(Xor(64|32|16|8) x (Xor(64|32|16|8) x y)) => y
603
604// Unsigned comparisons to zero.
605(Less(64U|32U|16U|8U) _ (Const(64|32|16|8) [0])) => (ConstBool [false])
606(Leq(64U|32U|16U|8U) (Const(64|32|16|8) [0]) _) => (ConstBool [true])
607
608// Ands clear bits. Ors set bits.
609// If a subsequent Or will set all the bits
610// that an And cleared, we can skip the And.
611// This happens in bitmasking code like:
612// x &^= 3 << shift // clear two old bits
613// x |= v << shift // set two new bits
614// when shift is a small constant and v ends up a constant 3.
615(Or8 (And8 x (Const8 [c2])) (Const8 <t> [c1])) && ^(c1 | c2) == 0 => (Or8 (Const8 <t> [c1]) x)
616(Or16 (And16 x (Const16 [c2])) (Const16 <t> [c1])) && ^(c1 | c2) == 0 => (Or16 (Const16 <t> [c1]) x)
617(Or32 (And32 x (Const32 [c2])) (Const32 <t> [c1])) && ^(c1 | c2) == 0 => (Or32 (Const32 <t> [c1]) x)
618(Or64 (And64 x (Const64 [c2])) (Const64 <t> [c1])) && ^(c1 | c2) == 0 => (Or64 (Const64 <t> [c1]) x)
619
620(Trunc64to8 (And64 (Const64 [y]) x)) && y&0xFF == 0xFF => (Trunc64to8 x)
621(Trunc64to16 (And64 (Const64 [y]) x)) && y&0xFFFF == 0xFFFF => (Trunc64to16 x)
622(Trunc64to32 (And64 (Const64 [y]) x)) && y&0xFFFFFFFF == 0xFFFFFFFF => (Trunc64to32 x)
623(Trunc32to8 (And32 (Const32 [y]) x)) && y&0xFF == 0xFF => (Trunc32to8 x)
624(Trunc32to16 (And32 (Const32 [y]) x)) && y&0xFFFF == 0xFFFF => (Trunc32to16 x)
625(Trunc16to8 (And16 (Const16 [y]) x)) && y&0xFF == 0xFF => (Trunc16to8 x)
626
627(ZeroExt8to64 (Trunc64to8 x:(Rsh64Ux64 _ (Const64 [s])))) && s >= 56 => x
628(ZeroExt16to64 (Trunc64to16 x:(Rsh64Ux64 _ (Const64 [s])))) && s >= 48 => x
629(ZeroExt32to64 (Trunc64to32 x:(Rsh64Ux64 _ (Const64 [s])))) && s >= 32 => x
630(ZeroExt8to32 (Trunc32to8 x:(Rsh32Ux64 _ (Const64 [s])))) && s >= 24 => x
631(ZeroExt16to32 (Trunc32to16 x:(Rsh32Ux64 _ (Const64 [s])))) && s >= 16 => x
632(ZeroExt8to16 (Trunc16to8 x:(Rsh16Ux64 _ (Const64 [s])))) && s >= 8 => x
633
634(SignExt8to64 (Trunc64to8 x:(Rsh64x64 _ (Const64 [s])))) && s >= 56 => x
635(SignExt16to64 (Trunc64to16 x:(Rsh64x64 _ (Const64 [s])))) && s >= 48 => x
636(SignExt32to64 (Trunc64to32 x:(Rsh64x64 _ (Const64 [s])))) && s >= 32 => x
637(SignExt8to32 (Trunc32to8 x:(Rsh32x64 _ (Const64 [s])))) && s >= 24 => x
638(SignExt16to32 (Trunc32to16 x:(Rsh32x64 _ (Const64 [s])))) && s >= 16 => x
639(SignExt8to16 (Trunc16to8 x:(Rsh16x64 _ (Const64 [s])))) && s >= 8 => x
640
641(Slicemask (Const32 [x])) && x > 0 => (Const32 [-1])
642(Slicemask (Const32 [0])) => (Const32 [0])
643(Slicemask (Const64 [x])) && x > 0 => (Const64 [-1])
644(Slicemask (Const64 [0])) => (Const64 [0])
645
646// simplifications often used for lengths. e.g. len(s[i:i+5])==5
647(Sub(64|32|16|8) (Add(64|32|16|8) x y) x) => y
648(Sub(64|32|16|8) (Add(64|32|16|8) x y) y) => x
649(Sub(64|32|16|8) (Sub(64|32|16|8) x y) x) => (Neg(64|32|16|8) y)
650(Sub(64|32|16|8) x (Add(64|32|16|8) x y)) => (Neg(64|32|16|8) y)
651(Add(64|32|16|8) x (Sub(64|32|16|8) y x)) => y
652(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)
653
654// basic phi simplifications
655(Phi (Const8 [c]) (Const8 [c])) => (Const8 [c])
656(Phi (Const16 [c]) (Const16 [c])) => (Const16 [c])
657(Phi (Const32 [c]) (Const32 [c])) => (Const32 [c])
658(Phi (Const64 [c]) (Const64 [c])) => (Const64 [c])
659
660// slice and interface comparisons
661// The frontend ensures that we can only compare against nil,
662// so we need only compare the first word (interface type or slice ptr).
663(EqInter x y) => (EqPtr (ITab x) (ITab y))
664(NeqInter x y) => (NeqPtr (ITab x) (ITab y))
665(EqSlice x y) => (EqPtr (SlicePtr x) (SlicePtr y))
666(NeqSlice x y) => (NeqPtr (SlicePtr x) (SlicePtr y))
667
668// Load of store of same address, with compatibly typed value and same size
669(Load <t1> p1 (Store {t2} p2 x _))
670 && isSamePtr(p1, p2)
671 && t1.Compare(x.Type) == types.CMPeq
672 && t1.Size() == t2.Size()
673 => x
674(Load <t1> p1 (Store {t2} p2 _ (Store {t3} p3 x _)))
675 && isSamePtr(p1, p3)
676 && t1.Compare(x.Type) == types.CMPeq
677 && t1.Size() == t2.Size()
678 && disjoint(p3, t3.Size(), p2, t2.Size())
679 => x
680(Load <t1> p1 (Store {t2} p2 _ (Store {t3} p3 _ (Store {t4} p4 x _))))
681 && isSamePtr(p1, p4)
682 && t1.Compare(x.Type) == types.CMPeq
683 && t1.Size() == t2.Size()
684 && disjoint(p4, t4.Size(), p2, t2.Size())
685 && disjoint(p4, t4.Size(), p3, t3.Size())
686 => x
687(Load <t1> p1 (Store {t2} p2 _ (Store {t3} p3 _ (Store {t4} p4 _ (Store {t5} p5 x _)))))
688 && isSamePtr(p1, p5)
689 && t1.Compare(x.Type) == types.CMPeq
690 && t1.Size() == t2.Size()
691 && disjoint(p5, t5.Size(), p2, t2.Size())
692 && disjoint(p5, t5.Size(), p3, t3.Size())
693 && disjoint(p5, t5.Size(), p4, t4.Size())
694 => x
695
696// Pass constants through math.Float{32,64}bits and math.Float{32,64}frombits
697 (Load <t1> p1 (Store {t2} p2 (Const64 [x]) _)) && isSamePtr(p1,p2) && sizeof(t2) == 8 && is64BitFloat(t1) && !math.IsNaN(math.Float64frombits(uint64(x))) => (Const64F [math.Float64frombits(uint64(x))])
698 (Load <t1> p1 (Store {t2} p2 (Const32 [x]) _)) && isSamePtr(p1,p2) && sizeof(t2) == 4 && is32BitFloat(t1) && !math.IsNaN(float64(math.Float32frombits(uint32(x)))) => (Const32F [math.Float32frombits(uint32(x))])
699(Load <t1> p1 (Store {t2} p2 (Const64F [x]) _)) && isSamePtr(p1,p2) && sizeof(t2) == 8 && is64BitInt(t1) => (Const64 [int64(math.Float64bits(x))])
700(Load <t1> p1 (Store {t2} p2 (Const32F [x]) _)) && isSamePtr(p1,p2) && sizeof(t2) == 4 && is32BitInt(t1) => (Const32 [int32(math.Float32bits(x))])
701
702// Float Loads up to Zeros so they can be constant folded.
703(Load <t1> op:(OffPtr [o1] p1)
704 (Store {t2} p2 _
705 mem:(Zero [n] p3 _)))
706 && o1 >= 0 && o1+t1.Size() <= n && isSamePtr(p1, p3)
707 && CanSSA(t1)
708 && disjoint(op, t1.Size(), p2, t2.Size())
709 => @mem.Block (Load <t1> (OffPtr <op.Type> [o1] p3) mem)
710(Load <t1> op:(OffPtr [o1] p1)
711 (Store {t2} p2 _
712 (Store {t3} p3 _
713 mem:(Zero [n] p4 _))))
714 && o1 >= 0 && o1+t1.Size() <= n && isSamePtr(p1, p4)
715 && CanSSA(t1)
716 && disjoint(op, t1.Size(), p2, t2.Size())
717 && disjoint(op, t1.Size(), p3, t3.Size())
718 => @mem.Block (Load <t1> (OffPtr <op.Type> [o1] p4) mem)
719(Load <t1> op:(OffPtr [o1] p1)
720 (Store {t2} p2 _
721 (Store {t3} p3 _
722 (Store {t4} p4 _
723 mem:(Zero [n] p5 _)))))
724 && o1 >= 0 && o1+t1.Size() <= n && isSamePtr(p1, p5)
725 && CanSSA(t1)
726 && disjoint(op, t1.Size(), p2, t2.Size())
727 && disjoint(op, t1.Size(), p3, t3.Size())
728 && disjoint(op, t1.Size(), p4, t4.Size())
729 => @mem.Block (Load <t1> (OffPtr <op.Type> [o1] p5) mem)
730(Load <t1> op:(OffPtr [o1] p1)
731 (Store {t2} p2 _
732 (Store {t3} p3 _
733 (Store {t4} p4 _
734 (Store {t5} p5 _
735 mem:(Zero [n] p6 _))))))
736 && o1 >= 0 && o1+t1.Size() <= n && isSamePtr(p1, p6)
737 && CanSSA(t1)
738 && disjoint(op, t1.Size(), p2, t2.Size())
739 && disjoint(op, t1.Size(), p3, t3.Size())
740 && disjoint(op, t1.Size(), p4, t4.Size())
741 && disjoint(op, t1.Size(), p5, t5.Size())
742 => @mem.Block (Load <t1> (OffPtr <op.Type> [o1] p6) mem)
743
744// Zero to Load forwarding.
745(Load <t1> (OffPtr [o] p1) (Zero [n] p2 _))
746 && t1.IsBoolean()
747 && isSamePtr(p1, p2)
748 && n >= o + 1
749 => (ConstBool [false])
750(Load <t1> (OffPtr [o] p1) (Zero [n] p2 _))
751 && is8BitInt(t1)
752 && isSamePtr(p1, p2)
753 && n >= o + 1
754 => (Const8 [0])
755(Load <t1> (OffPtr [o] p1) (Zero [n] p2 _))
756 && is16BitInt(t1)
757 && isSamePtr(p1, p2)
758 && n >= o + 2
759 => (Const16 [0])
760(Load <t1> (OffPtr [o] p1) (Zero [n] p2 _))
761 && is32BitInt(t1)
762 && isSamePtr(p1, p2)
763 && n >= o + 4
764 => (Const32 [0])
765(Load <t1> (OffPtr [o] p1) (Zero [n] p2 _))
766 && is64BitInt(t1)
767 && isSamePtr(p1, p2)
768 && n >= o + 8
769 => (Const64 [0])
770(Load <t1> (OffPtr [o] p1) (Zero [n] p2 _))
771 && is32BitFloat(t1)
772 && isSamePtr(p1, p2)
773 && n >= o + 4
774 => (Const32F [0])
775(Load <t1> (OffPtr [o] p1) (Zero [n] p2 _))
776 && is64BitFloat(t1)
777 && isSamePtr(p1, p2)
778 && n >= o + 8
779 => (Const64F [0])
780
781// Eliminate stores of values that have just been loaded from the same location.
782// We also handle the common case where there are some intermediate stores.
783(Store {t1} p1 (Load <t2> p2 mem) mem)
784 && isSamePtr(p1, p2)
785 && t2.Size() == t1.Size()
786 => mem
787(Store {t1} p1 (Load <t2> p2 oldmem) mem:(Store {t3} p3 _ oldmem))
788 && isSamePtr(p1, p2)
789 && t2.Size() == t1.Size()
790 && disjoint(p1, t1.Size(), p3, t3.Size())
791 => mem
792(Store {t1} p1 (Load <t2> p2 oldmem) mem:(Store {t3} p3 _ (Store {t4} p4 _ oldmem)))
793 && isSamePtr(p1, p2)
794 && t2.Size() == t1.Size()
795 && disjoint(p1, t1.Size(), p3, t3.Size())
796 && disjoint(p1, t1.Size(), p4, t4.Size())
797 => mem
798(Store {t1} p1 (Load <t2> p2 oldmem) mem:(Store {t3} p3 _ (Store {t4} p4 _ (Store {t5} p5 _ oldmem))))
799 && isSamePtr(p1, p2)
800 && t2.Size() == t1.Size()
801 && disjoint(p1, t1.Size(), p3, t3.Size())
802 && disjoint(p1, t1.Size(), p4, t4.Size())
803 && disjoint(p1, t1.Size(), p5, t5.Size())
804 => mem
805
806// Don't Store zeros to cleared variables.
807(Store {t} (OffPtr [o] p1) x mem:(Zero [n] p2 _))
808 && isConstZero(x)
809 && o >= 0 && t.Size() + o <= n && isSamePtr(p1, p2)
810 => mem
811(Store {t1} op:(OffPtr [o1] p1) x mem:(Store {t2} p2 _ (Zero [n] p3 _)))
812 && isConstZero(x)
813 && o1 >= 0 && t1.Size() + o1 <= n && isSamePtr(p1, p3)
814 && disjoint(op, t1.Size(), p2, t2.Size())
815 => mem
816(Store {t1} op:(OffPtr [o1] p1) x mem:(Store {t2} p2 _ (Store {t3} p3 _ (Zero [n] p4 _))))
817 && isConstZero(x)
818 && o1 >= 0 && t1.Size() + o1 <= n && isSamePtr(p1, p4)
819 && disjoint(op, t1.Size(), p2, t2.Size())
820 && disjoint(op, t1.Size(), p3, t3.Size())
821 => mem
822(Store {t1} op:(OffPtr [o1] p1) x mem:(Store {t2} p2 _ (Store {t3} p3 _ (Store {t4} p4 _ (Zero [n] p5 _)))))
823 && isConstZero(x)
824 && o1 >= 0 && t1.Size() + o1 <= n && isSamePtr(p1, p5)
825 && disjoint(op, t1.Size(), p2, t2.Size())
826 && disjoint(op, t1.Size(), p3, t3.Size())
827 && disjoint(op, t1.Size(), p4, t4.Size())
828 => mem
829
830// Collapse OffPtr
831(OffPtr (OffPtr p [y]) [x]) => (OffPtr p [x+y])
832(OffPtr p [0]) && v.Type.Compare(p.Type) == types.CMPeq => p
833
834// indexing operations
835// Note: bounds check has already been done
836(PtrIndex <t> ptr idx) && config.PtrSize == 4 && is32Bit(t.Elem().Size()) => (AddPtr ptr (Mul32 <typ.Int> idx (Const32 <typ.Int> [int32(t.Elem().Size())])))
837(PtrIndex <t> ptr idx) && config.PtrSize == 8 => (AddPtr ptr (Mul64 <typ.Int> idx (Const64 <typ.Int> [t.Elem().Size()])))
838
839// struct operations
840(StructSelect (StructMake1 x)) => x
841(StructSelect [0] (StructMake2 x _)) => x
842(StructSelect [1] (StructMake2 _ x)) => x
843(StructSelect [0] (StructMake3 x _ _)) => x
844(StructSelect [1] (StructMake3 _ x _)) => x
845(StructSelect [2] (StructMake3 _ _ x)) => x
846(StructSelect [0] (StructMake4 x _ _ _)) => x
847(StructSelect [1] (StructMake4 _ x _ _)) => x
848(StructSelect [2] (StructMake4 _ _ x _)) => x
849(StructSelect [3] (StructMake4 _ _ _ x)) => x
850
851(Load <t> _ _) && t.IsStruct() && t.NumFields() == 0 && CanSSA(t) =>
852 (StructMake0)
853(Load <t> ptr mem) && t.IsStruct() && t.NumFields() == 1 && CanSSA(t) =>
854 (StructMake1
855 (Load <t.FieldType(0)> (OffPtr <t.FieldType(0).PtrTo()> [0] ptr) mem))
856(Load <t> ptr mem) && t.IsStruct() && t.NumFields() == 2 && CanSSA(t) =>
857 (StructMake2
858 (Load <t.FieldType(0)> (OffPtr <t.FieldType(0).PtrTo()> [0] ptr) mem)
859 (Load <t.FieldType(1)> (OffPtr <t.FieldType(1).PtrTo()> [t.FieldOff(1)] ptr) mem))
860(Load <t> ptr mem) && t.IsStruct() && t.NumFields() == 3 && CanSSA(t) =>
861 (StructMake3
862 (Load <t.FieldType(0)> (OffPtr <t.FieldType(0).PtrTo()> [0] ptr) mem)
863 (Load <t.FieldType(1)> (OffPtr <t.FieldType(1).PtrTo()> [t.FieldOff(1)] ptr) mem)
864 (Load <t.FieldType(2)> (OffPtr <t.FieldType(2).PtrTo()> [t.FieldOff(2)] ptr) mem))
865(Load <t> ptr mem) && t.IsStruct() && t.NumFields() == 4 && CanSSA(t) =>
866 (StructMake4
867 (Load <t.FieldType(0)> (OffPtr <t.FieldType(0).PtrTo()> [0] ptr) mem)
868 (Load <t.FieldType(1)> (OffPtr <t.FieldType(1).PtrTo()> [t.FieldOff(1)] ptr) mem)
869 (Load <t.FieldType(2)> (OffPtr <t.FieldType(2).PtrTo()> [t.FieldOff(2)] ptr) mem)
870 (Load <t.FieldType(3)> (OffPtr <t.FieldType(3).PtrTo()> [t.FieldOff(3)] ptr) mem))
871
872(StructSelect [i] x:(Load <t> ptr mem)) && !CanSSA(t) =>
873 @x.Block (Load <v.Type> (OffPtr <v.Type.PtrTo()> [t.FieldOff(int(i))] ptr) mem)
874
875(Store _ (StructMake0) mem) => mem
876(Store dst (StructMake1 <t> f0) mem) =>
877 (Store {t.FieldType(0)} (OffPtr <t.FieldType(0).PtrTo()> [0] dst) f0 mem)
878(Store dst (StructMake2 <t> f0 f1) mem) =>
879 (Store {t.FieldType(1)}
880 (OffPtr <t.FieldType(1).PtrTo()> [t.FieldOff(1)] dst)
881 f1
882 (Store {t.FieldType(0)}
883 (OffPtr <t.FieldType(0).PtrTo()> [0] dst)
884 f0 mem))
885(Store dst (StructMake3 <t> f0 f1 f2) mem) =>
886 (Store {t.FieldType(2)}
887 (OffPtr <t.FieldType(2).PtrTo()> [t.FieldOff(2)] dst)
888 f2
889 (Store {t.FieldType(1)}
890 (OffPtr <t.FieldType(1).PtrTo()> [t.FieldOff(1)] dst)
891 f1
892 (Store {t.FieldType(0)}
893 (OffPtr <t.FieldType(0).PtrTo()> [0] dst)
894 f0 mem)))
895(Store dst (StructMake4 <t> f0 f1 f2 f3) mem) =>
896 (Store {t.FieldType(3)}
897 (OffPtr <t.FieldType(3).PtrTo()> [t.FieldOff(3)] dst)
898 f3
899 (Store {t.FieldType(2)}
900 (OffPtr <t.FieldType(2).PtrTo()> [t.FieldOff(2)] dst)
901 f2
902 (Store {t.FieldType(1)}
903 (OffPtr <t.FieldType(1).PtrTo()> [t.FieldOff(1)] dst)
904 f1
905 (Store {t.FieldType(0)}
906 (OffPtr <t.FieldType(0).PtrTo()> [0] dst)
907 f0 mem))))
908
909// Putting struct{*byte} and similar into direct interfaces.
910(IMake _typ (StructMake1 val)) => (IMake _typ val)
911(StructSelect [0] (IData x)) => (IData x)
912
913// un-SSAable values use mem->mem copies
914(Store {t} dst (Load src mem) mem) && !CanSSA(t) =>
915 (Move {t} [t.Size()] dst src mem)
916(Store {t} dst (Load src mem) (VarDef {x} mem)) && !CanSSA(t) =>
917 (Move {t} [t.Size()] dst src (VarDef {x} mem))
918
919// array ops
920(ArraySelect (ArrayMake1 x)) => x
921
922(Load <t> _ _) && t.IsArray() && t.NumElem() == 0 =>
923 (ArrayMake0)
924
925(Load <t> ptr mem) && t.IsArray() && t.NumElem() == 1 && CanSSA(t) =>
926 (ArrayMake1 (Load <t.Elem()> ptr mem))
927
928(Store _ (ArrayMake0) mem) => mem
929(Store dst (ArrayMake1 e) mem) => (Store {e.Type} dst e mem)
930
931// Putting [1]*byte and similar into direct interfaces.
932(IMake _typ (ArrayMake1 val)) => (IMake _typ val)
933(ArraySelect [0] (IData x)) => (IData x)
934
935// string ops
936// Decomposing StringMake and lowering of StringPtr and StringLen
937// happens in a later pass, dec, so that these operations are available
938// to other passes for optimizations.
939(StringPtr (StringMake (Addr <t> {s} base) _)) => (Addr <t> {s} base)
940(StringLen (StringMake _ (Const64 <t> [c]))) => (Const64 <t> [c])
941(ConstString {str}) && config.PtrSize == 4 && str == "" =>
942 (StringMake (ConstNil) (Const32 <typ.Int> [0]))
943(ConstString {str}) && config.PtrSize == 8 && str == "" =>
944 (StringMake (ConstNil) (Const64 <typ.Int> [0]))
945(ConstString {str}) && config.PtrSize == 4 && str != "" =>
946 (StringMake
947 (Addr <typ.BytePtr> {fe.StringData(str)}
948 (SB))
949 (Const32 <typ.Int> [int32(len(str))]))
950(ConstString {str}) && config.PtrSize == 8 && str != "" =>
951 (StringMake
952 (Addr <typ.BytePtr> {fe.StringData(str)}
953 (SB))
954 (Const64 <typ.Int> [int64(len(str))]))
955
956// slice ops
957// Only a few slice rules are provided here. See dec.rules for
958// a more comprehensive set.
959(SliceLen (SliceMake _ (Const64 <t> [c]) _)) => (Const64 <t> [c])
960(SliceCap (SliceMake _ _ (Const64 <t> [c]))) => (Const64 <t> [c])
961(SliceLen (SliceMake _ (Const32 <t> [c]) _)) => (Const32 <t> [c])
962(SliceCap (SliceMake _ _ (Const32 <t> [c]))) => (Const32 <t> [c])
963(SlicePtr (SliceMake (SlicePtr x) _ _)) => (SlicePtr x)
964(SliceLen (SliceMake _ (SliceLen x) _)) => (SliceLen x)
965(SliceCap (SliceMake _ _ (SliceCap x))) => (SliceCap x)
966(SliceCap (SliceMake _ _ (SliceLen x))) => (SliceLen x)
967(ConstSlice) && config.PtrSize == 4 =>
968 (SliceMake
969 (ConstNil <v.Type.Elem().PtrTo()>)
970 (Const32 <typ.Int> [0])
971 (Const32 <typ.Int> [0]))
972(ConstSlice) && config.PtrSize == 8 =>
973 (SliceMake
974 (ConstNil <v.Type.Elem().PtrTo()>)
975 (Const64 <typ.Int> [0])
976 (Const64 <typ.Int> [0]))
977
978// interface ops
979(ConstInterface) =>
980 (IMake
981 (ConstNil <typ.Uintptr>)
982 (ConstNil <typ.BytePtr>))
983
984(NilCheck ptr:(GetG mem) mem) => ptr
985
986(If (Not cond) yes no) => (If cond no yes)
987(If (ConstBool [c]) yes no) && c => (First yes no)
988(If (ConstBool [c]) yes no) && !c => (First no yes)
989
990(Phi <t> nx:(Not x) ny:(Not y)) && nx.Uses == 1 && ny.Uses == 1 => (Not (Phi <t> x y))
991
992// Get rid of Convert ops for pointer arithmetic on unsafe.Pointer.
993(Convert (Add(64|32) (Convert ptr mem) off) mem) => (AddPtr ptr off)
994(Convert (Convert ptr mem) mem) => ptr
995
996// strength reduction of divide by a constant.
997// See ../magic.go for a detailed description of these algorithms.
998
999// Unsigned divide by power of 2. Strength reduce to a shift.
1000(Div8u n (Const8 [c])) && isPowerOfTwo8(c) => (Rsh8Ux64 n (Const64 <typ.UInt64> [log8(c)]))
1001(Div16u n (Const16 [c])) && isPowerOfTwo16(c) => (Rsh16Ux64 n (Const64 <typ.UInt64> [log16(c)]))
1002(Div32u n (Const32 [c])) && isPowerOfTwo32(c) => (Rsh32Ux64 n (Const64 <typ.UInt64> [log32(c)]))
1003(Div64u n (Const64 [c])) && isPowerOfTwo64(c) => (Rsh64Ux64 n (Const64 <typ.UInt64> [log64(c)]))
1004(Div64u n (Const64 [-1<<63])) => (Rsh64Ux64 n (Const64 <typ.UInt64> [63]))
1005
1006// Signed non-negative divide by power of 2.
1007(Div8 n (Const8 [c])) && isNonNegative(n) && isPowerOfTwo8(c) => (Rsh8Ux64 n (Const64 <typ.UInt64> [log8(c)]))
1008(Div16 n (Const16 [c])) && isNonNegative(n) && isPowerOfTwo16(c) => (Rsh16Ux64 n (Const64 <typ.UInt64> [log16(c)]))
1009(Div32 n (Const32 [c])) && isNonNegative(n) && isPowerOfTwo32(c) => (Rsh32Ux64 n (Const64 <typ.UInt64> [log32(c)]))
1010(Div64 n (Const64 [c])) && isNonNegative(n) && isPowerOfTwo64(c) => (Rsh64Ux64 n (Const64 <typ.UInt64> [log64(c)]))
1011(Div64 n (Const64 [-1<<63])) && isNonNegative(n) => (Const64 [0])
1012
1013// Unsigned divide, not a power of 2. Strength reduce to a multiply.
1014// For 8-bit divides, we just do a direct 9-bit by 8-bit multiply.
1015(Div8u x (Const8 [c])) && umagicOK8(c) =>
1016 (Trunc32to8
1017 (Rsh32Ux64 <typ.UInt32>
1018 (Mul32 <typ.UInt32>
1019 (Const32 <typ.UInt32> [int32(1<<8+umagic8(c).m)])
1020 (ZeroExt8to32 x))
1021 (Const64 <typ.UInt64> [8+umagic8(c).s])))
1022
1023// For 16-bit divides on 64-bit machines, we do a direct 17-bit by 16-bit multiply.
1024(Div16u x (Const16 [c])) && umagicOK16(c) && config.RegSize == 8 =>
1025 (Trunc64to16
1026 (Rsh64Ux64 <typ.UInt64>
1027 (Mul64 <typ.UInt64>
1028 (Const64 <typ.UInt64> [int64(1<<16+umagic16(c).m)])
1029 (ZeroExt16to64 x))
1030 (Const64 <typ.UInt64> [16+umagic16(c).s])))
1031
1032// For 16-bit divides on 32-bit machines
1033(Div16u x (Const16 [c])) && umagicOK16(c) && config.RegSize == 4 && umagic16(c).m&1 == 0 =>
1034 (Trunc32to16
1035 (Rsh32Ux64 <typ.UInt32>
1036 (Mul32 <typ.UInt32>
1037 (Const32 <typ.UInt32> [int32(1<<15+umagic16(c).m/2)])
1038 (ZeroExt16to32 x))
1039 (Const64 <typ.UInt64> [16+umagic16(c).s-1])))
1040(Div16u x (Const16 [c])) && umagicOK16(c) && config.RegSize == 4 && c&1 == 0 =>
1041 (Trunc32to16
1042 (Rsh32Ux64 <typ.UInt32>
1043 (Mul32 <typ.UInt32>
1044 (Const32 <typ.UInt32> [int32(1<<15+(umagic16(c).m+1)/2)])
1045 (Rsh32Ux64 <typ.UInt32> (ZeroExt16to32 x) (Const64 <typ.UInt64> [1])))
1046 (Const64 <typ.UInt64> [16+umagic16(c).s-2])))
1047(Div16u x (Const16 [c])) && umagicOK16(c) && config.RegSize == 4 && config.useAvg =>
1048 (Trunc32to16
1049 (Rsh32Ux64 <typ.UInt32>
1050 (Avg32u
1051 (Lsh32x64 <typ.UInt32> (ZeroExt16to32 x) (Const64 <typ.UInt64> [16]))
1052 (Mul32 <typ.UInt32>
1053 (Const32 <typ.UInt32> [int32(umagic16(c).m)])
1054 (ZeroExt16to32 x)))
1055 (Const64 <typ.UInt64> [16+umagic16(c).s-1])))
1056
1057// For 32-bit divides on 32-bit machines
1058(Div32u x (Const32 [c])) && umagicOK32(c) && config.RegSize == 4 && umagic32(c).m&1 == 0 && config.useHmul =>
1059 (Rsh32Ux64 <typ.UInt32>
1060 (Hmul32u <typ.UInt32>
1061 (Const32 <typ.UInt32> [int32(1<<31+umagic32(c).m/2)])
1062 x)
1063 (Const64 <typ.UInt64> [umagic32(c).s-1]))
1064(Div32u x (Const32 [c])) && umagicOK32(c) && config.RegSize == 4 && c&1 == 0 && config.useHmul =>
1065 (Rsh32Ux64 <typ.UInt32>
1066 (Hmul32u <typ.UInt32>
1067 (Const32 <typ.UInt32> [int32(1<<31+(umagic32(c).m+1)/2)])
1068 (Rsh32Ux64 <typ.UInt32> x (Const64 <typ.UInt64> [1])))
1069 (Const64 <typ.UInt64> [umagic32(c).s-2]))
1070(Div32u x (Const32 [c])) && umagicOK32(c) && config.RegSize == 4 && config.useAvg && config.useHmul =>
1071 (Rsh32Ux64 <typ.UInt32>
1072 (Avg32u
1073 x
1074 (Hmul32u <typ.UInt32>
1075 (Const32 <typ.UInt32> [int32(umagic32(c).m)])
1076 x))
1077 (Const64 <typ.UInt64> [umagic32(c).s-1]))
1078
1079// For 32-bit divides on 64-bit machines
1080// We'll use a regular (non-hi) multiply for this case.
1081(Div32u x (Const32 [c])) && umagicOK32(c) && config.RegSize == 8 && umagic32(c).m&1 == 0 =>
1082 (Trunc64to32
1083 (Rsh64Ux64 <typ.UInt64>
1084 (Mul64 <typ.UInt64>
1085 (Const64 <typ.UInt64> [int64(1<<31+umagic32(c).m/2)])
1086 (ZeroExt32to64 x))
1087 (Const64 <typ.UInt64> [32+umagic32(c).s-1])))
1088(Div32u x (Const32 [c])) && umagicOK32(c) && config.RegSize == 8 && c&1 == 0 =>
1089 (Trunc64to32
1090 (Rsh64Ux64 <typ.UInt64>
1091 (Mul64 <typ.UInt64>
1092 (Const64 <typ.UInt64> [int64(1<<31+(umagic32(c).m+1)/2)])
1093 (Rsh64Ux64 <typ.UInt64> (ZeroExt32to64 x) (Const64 <typ.UInt64> [1])))
1094 (Const64 <typ.UInt64> [32+umagic32(c).s-2])))
1095(Div32u x (Const32 [c])) && umagicOK32(c) && config.RegSize == 8 && config.useAvg =>
1096 (Trunc64to32
1097 (Rsh64Ux64 <typ.UInt64>
1098 (Avg64u
1099 (Lsh64x64 <typ.UInt64> (ZeroExt32to64 x) (Const64 <typ.UInt64> [32]))
1100 (Mul64 <typ.UInt64>
1101 (Const64 <typ.UInt32> [int64(umagic32(c).m)])
1102 (ZeroExt32to64 x)))
1103 (Const64 <typ.UInt64> [32+umagic32(c).s-1])))
1104
1105// For unsigned 64-bit divides on 32-bit machines,
1106// if the constant fits in 16 bits (so that the last term
1107// fits in 32 bits), convert to three 32-bit divides by a constant.
1108//
1109// If 1<<32 = Q * c + R
1110// and x = hi << 32 + lo
1111//
1112// Then x = (hi/c*c + hi%c) << 32 + lo
1113// = hi/c*c<<32 + hi%c<<32 + lo
1114// = hi/c*c<<32 + (hi%c)*(Q*c+R) + lo/c*c + lo%c
1115// = hi/c*c<<32 + (hi%c)*Q*c + lo/c*c + (hi%c*R+lo%c)
1116// and x / c = (hi/c)<<32 + (hi%c)*Q + lo/c + (hi%c*R+lo%c)/c
1117(Div64u x (Const64 [c])) && c > 0 && c <= 0xFFFF && umagicOK32(int32(c)) && config.RegSize == 4 && config.useHmul =>
1118 (Add64
1119 (Add64 <typ.UInt64>
1120 (Add64 <typ.UInt64>
1121 (Lsh64x64 <typ.UInt64>
1122 (ZeroExt32to64
1123 (Div32u <typ.UInt32>
1124 (Trunc64to32 <typ.UInt32> (Rsh64Ux64 <typ.UInt64> x (Const64 <typ.UInt64> [32])))
1125 (Const32 <typ.UInt32> [int32(c)])))
1126 (Const64 <typ.UInt64> [32]))
1127 (ZeroExt32to64 (Div32u <typ.UInt32> (Trunc64to32 <typ.UInt32> x) (Const32 <typ.UInt32> [int32(c)]))))
1128 (Mul64 <typ.UInt64>
1129 (ZeroExt32to64 <typ.UInt64>
1130 (Mod32u <typ.UInt32>
1131 (Trunc64to32 <typ.UInt32> (Rsh64Ux64 <typ.UInt64> x (Const64 <typ.UInt64> [32])))
1132 (Const32 <typ.UInt32> [int32(c)])))
1133 (Const64 <typ.UInt64> [int64((1<<32)/c)])))
1134 (ZeroExt32to64
1135 (Div32u <typ.UInt32>
1136 (Add32 <typ.UInt32>
1137 (Mod32u <typ.UInt32> (Trunc64to32 <typ.UInt32> x) (Const32 <typ.UInt32> [int32(c)]))
1138 (Mul32 <typ.UInt32>
1139 (Mod32u <typ.UInt32>
1140 (Trunc64to32 <typ.UInt32> (Rsh64Ux64 <typ.UInt64> x (Const64 <typ.UInt64> [32])))
1141 (Const32 <typ.UInt32> [int32(c)]))
1142 (Const32 <typ.UInt32> [int32((1<<32)%c)])))
1143 (Const32 <typ.UInt32> [int32(c)]))))
1144
1145// For 64-bit divides on 64-bit machines
1146// (64-bit divides on 32-bit machines are lowered to a runtime call by the walk pass.)
1147(Div64u x (Const64 [c])) && umagicOK64(c) && config.RegSize == 8 && umagic64(c).m&1 == 0 && config.useHmul =>
1148 (Rsh64Ux64 <typ.UInt64>
1149 (Hmul64u <typ.UInt64>
1150 (Const64 <typ.UInt64> [int64(1<<63+umagic64(c).m/2)])
1151 x)
1152 (Const64 <typ.UInt64> [umagic64(c).s-1]))
1153(Div64u x (Const64 [c])) && umagicOK64(c) && config.RegSize == 8 && c&1 == 0 && config.useHmul =>
1154 (Rsh64Ux64 <typ.UInt64>
1155 (Hmul64u <typ.UInt64>
1156 (Const64 <typ.UInt64> [int64(1<<63+(umagic64(c).m+1)/2)])
1157 (Rsh64Ux64 <typ.UInt64> x (Const64 <typ.UInt64> [1])))
1158 (Const64 <typ.UInt64> [umagic64(c).s-2]))
1159(Div64u x (Const64 [c])) && umagicOK64(c) && config.RegSize == 8 && config.useAvg && config.useHmul =>
1160 (Rsh64Ux64 <typ.UInt64>
1161 (Avg64u
1162 x
1163 (Hmul64u <typ.UInt64>
1164 (Const64 <typ.UInt64> [int64(umagic64(c).m)])
1165 x))
1166 (Const64 <typ.UInt64> [umagic64(c).s-1]))
1167
1168// Signed divide by a negative constant. Rewrite to divide by a positive constant.
1169(Div8 <t> n (Const8 [c])) && c < 0 && c != -1<<7 => (Neg8 (Div8 <t> n (Const8 <t> [-c])))
1170(Div16 <t> n (Const16 [c])) && c < 0 && c != -1<<15 => (Neg16 (Div16 <t> n (Const16 <t> [-c])))
1171(Div32 <t> n (Const32 [c])) && c < 0 && c != -1<<31 => (Neg32 (Div32 <t> n (Const32 <t> [-c])))
1172(Div64 <t> n (Const64 [c])) && c < 0 && c != -1<<63 => (Neg64 (Div64 <t> n (Const64 <t> [-c])))
1173
1174// Dividing by the most-negative number. Result is always 0 except
1175// if the input is also the most-negative number.
1176// We can detect that using the sign bit of x & -x.
1177(Div8 <t> x (Const8 [-1<<7 ])) => (Rsh8Ux64 (And8 <t> x (Neg8 <t> x)) (Const64 <typ.UInt64> [7 ]))
1178(Div16 <t> x (Const16 [-1<<15])) => (Rsh16Ux64 (And16 <t> x (Neg16 <t> x)) (Const64 <typ.UInt64> [15]))
1179(Div32 <t> x (Const32 [-1<<31])) => (Rsh32Ux64 (And32 <t> x (Neg32 <t> x)) (Const64 <typ.UInt64> [31]))
1180(Div64 <t> x (Const64 [-1<<63])) => (Rsh64Ux64 (And64 <t> x (Neg64 <t> x)) (Const64 <typ.UInt64> [63]))
1181
1182// Signed divide by power of 2.
1183// n / c = n >> log(c) if n >= 0
1184// = (n+c-1) >> log(c) if n < 0
1185// We conditionally add c-1 by adding n>>63>>(64-log(c)) (first shift signed, second shift unsigned).
1186(Div8 <t> n (Const8 [c])) && isPowerOfTwo8(c) =>
1187 (Rsh8x64
1188 (Add8 <t> n (Rsh8Ux64 <t> (Rsh8x64 <t> n (Const64 <typ.UInt64> [ 7])) (Const64 <typ.UInt64> [int64( 8-log8(c))])))
1189 (Const64 <typ.UInt64> [int64(log8(c))]))
1190(Div16 <t> n (Const16 [c])) && isPowerOfTwo16(c) =>
1191 (Rsh16x64
1192 (Add16 <t> n (Rsh16Ux64 <t> (Rsh16x64 <t> n (Const64 <typ.UInt64> [15])) (Const64 <typ.UInt64> [int64(16-log16(c))])))
1193 (Const64 <typ.UInt64> [int64(log16(c))]))
1194(Div32 <t> n (Const32 [c])) && isPowerOfTwo32(c) =>
1195 (Rsh32x64
1196 (Add32 <t> n (Rsh32Ux64 <t> (Rsh32x64 <t> n (Const64 <typ.UInt64> [31])) (Const64 <typ.UInt64> [int64(32-log32(c))])))
1197 (Const64 <typ.UInt64> [int64(log32(c))]))
1198(Div64 <t> n (Const64 [c])) && isPowerOfTwo64(c) =>
1199 (Rsh64x64
1200 (Add64 <t> n (Rsh64Ux64 <t> (Rsh64x64 <t> n (Const64 <typ.UInt64> [63])) (Const64 <typ.UInt64> [int64(64-log64(c))])))
1201 (Const64 <typ.UInt64> [int64(log64(c))]))
1202
1203// Signed divide, not a power of 2. Strength reduce to a multiply.
1204(Div8 <t> x (Const8 [c])) && smagicOK8(c) =>
1205 (Sub8 <t>
1206 (Rsh32x64 <t>
1207 (Mul32 <typ.UInt32>
1208 (Const32 <typ.UInt32> [int32(smagic8(c).m)])
1209 (SignExt8to32 x))
1210 (Const64 <typ.UInt64> [8+smagic8(c).s]))
1211 (Rsh32x64 <t>
1212 (SignExt8to32 x)
1213 (Const64 <typ.UInt64> [31])))
1214(Div16 <t> x (Const16 [c])) && smagicOK16(c) =>
1215 (Sub16 <t>
1216 (Rsh32x64 <t>
1217 (Mul32 <typ.UInt32>
1218 (Const32 <typ.UInt32> [int32(smagic16(c).m)])
1219 (SignExt16to32 x))
1220 (Const64 <typ.UInt64> [16+smagic16(c).s]))
1221 (Rsh32x64 <t>
1222 (SignExt16to32 x)
1223 (Const64 <typ.UInt64> [31])))
1224(Div32 <t> x (Const32 [c])) && smagicOK32(c) && config.RegSize == 8 =>
1225 (Sub32 <t>
1226 (Rsh64x64 <t>
1227 (Mul64 <typ.UInt64>
1228 (Const64 <typ.UInt64> [int64(smagic32(c).m)])
1229 (SignExt32to64 x))
1230 (Const64 <typ.UInt64> [32+smagic32(c).s]))
1231 (Rsh64x64 <t>
1232 (SignExt32to64 x)
1233 (Const64 <typ.UInt64> [63])))
1234(Div32 <t> x (Const32 [c])) && smagicOK32(c) && config.RegSize == 4 && smagic32(c).m&1 == 0 && config.useHmul =>
1235 (Sub32 <t>
1236 (Rsh32x64 <t>
1237 (Hmul32 <t>
1238 (Const32 <typ.UInt32> [int32(smagic32(c).m/2)])
1239 x)
1240 (Const64 <typ.UInt64> [smagic32(c).s-1]))
1241 (Rsh32x64 <t>
1242 x
1243 (Const64 <typ.UInt64> [31])))
1244(Div32 <t> x (Const32 [c])) && smagicOK32(c) && config.RegSize == 4 && smagic32(c).m&1 != 0 && config.useHmul =>
1245 (Sub32 <t>
1246 (Rsh32x64 <t>
1247 (Add32 <t>
1248 (Hmul32 <t>
1249 (Const32 <typ.UInt32> [int32(smagic32(c).m)])
1250 x)
1251 x)
1252 (Const64 <typ.UInt64> [smagic32(c).s]))
1253 (Rsh32x64 <t>
1254 x
1255 (Const64 <typ.UInt64> [31])))
1256(Div64 <t> x (Const64 [c])) && smagicOK64(c) && smagic64(c).m&1 == 0 && config.useHmul =>
1257 (Sub64 <t>
1258 (Rsh64x64 <t>
1259 (Hmul64 <t>
1260 (Const64 <typ.UInt64> [int64(smagic64(c).m/2)])
1261 x)
1262 (Const64 <typ.UInt64> [smagic64(c).s-1]))
1263 (Rsh64x64 <t>
1264 x
1265 (Const64 <typ.UInt64> [63])))
1266(Div64 <t> x (Const64 [c])) && smagicOK64(c) && smagic64(c).m&1 != 0 && config.useHmul =>
1267 (Sub64 <t>
1268 (Rsh64x64 <t>
1269 (Add64 <t>
1270 (Hmul64 <t>
1271 (Const64 <typ.UInt64> [int64(smagic64(c).m)])
1272 x)
1273 x)
1274 (Const64 <typ.UInt64> [smagic64(c).s]))
1275 (Rsh64x64 <t>
1276 x
1277 (Const64 <typ.UInt64> [63])))
1278
1279// Unsigned mod by power of 2 constant.
1280(Mod8u <t> n (Const8 [c])) && isPowerOfTwo8(c) => (And8 n (Const8 <t> [c-1]))
1281(Mod16u <t> n (Const16 [c])) && isPowerOfTwo16(c) => (And16 n (Const16 <t> [c-1]))
1282(Mod32u <t> n (Const32 [c])) && isPowerOfTwo32(c) => (And32 n (Const32 <t> [c-1]))
1283(Mod64u <t> n (Const64 [c])) && isPowerOfTwo64(c) => (And64 n (Const64 <t> [c-1]))
1284(Mod64u <t> n (Const64 [-1<<63])) => (And64 n (Const64 <t> [1<<63-1]))
1285
1286// Signed non-negative mod by power of 2 constant.
1287(Mod8 <t> n (Const8 [c])) && isNonNegative(n) && isPowerOfTwo8(c) => (And8 n (Const8 <t> [c-1]))
1288(Mod16 <t> n (Const16 [c])) && isNonNegative(n) && isPowerOfTwo16(c) => (And16 n (Const16 <t> [c-1]))
1289(Mod32 <t> n (Const32 [c])) && isNonNegative(n) && isPowerOfTwo32(c) => (And32 n (Const32 <t> [c-1]))
1290(Mod64 <t> n (Const64 [c])) && isNonNegative(n) && isPowerOfTwo64(c) => (And64 n (Const64 <t> [c-1]))
1291(Mod64 n (Const64 [-1<<63])) && isNonNegative(n) => n
1292
1293// Signed mod by negative constant.
1294(Mod8 <t> n (Const8 [c])) && c < 0 && c != -1<<7 => (Mod8 <t> n (Const8 <t> [-c]))
1295(Mod16 <t> n (Const16 [c])) && c < 0 && c != -1<<15 => (Mod16 <t> n (Const16 <t> [-c]))
1296(Mod32 <t> n (Const32 [c])) && c < 0 && c != -1<<31 => (Mod32 <t> n (Const32 <t> [-c]))
1297(Mod64 <t> n (Const64 [c])) && c < 0 && c != -1<<63 => (Mod64 <t> n (Const64 <t> [-c]))
1298
1299// All other mods by constants, do A%B = A-(A/B*B).
1300// This implements % with two * and a bunch of ancillary ops.
1301// One of the * is free if the user's code also computes A/B.
1302(Mod8 <t> x (Const8 [c])) && x.Op != OpConst8 && (c > 0 || c == -1<<7)
1303 => (Sub8 x (Mul8 <t> (Div8 <t> x (Const8 <t> [c])) (Const8 <t> [c])))
1304(Mod16 <t> x (Const16 [c])) && x.Op != OpConst16 && (c > 0 || c == -1<<15)
1305 => (Sub16 x (Mul16 <t> (Div16 <t> x (Const16 <t> [c])) (Const16 <t> [c])))
1306(Mod32 <t> x (Const32 [c])) && x.Op != OpConst32 && (c > 0 || c == -1<<31)
1307 => (Sub32 x (Mul32 <t> (Div32 <t> x (Const32 <t> [c])) (Const32 <t> [c])))
1308(Mod64 <t> x (Const64 [c])) && x.Op != OpConst64 && (c > 0 || c == -1<<63)
1309 => (Sub64 x (Mul64 <t> (Div64 <t> x (Const64 <t> [c])) (Const64 <t> [c])))
1310(Mod8u <t> x (Const8 [c])) && x.Op != OpConst8 && c > 0 && umagicOK8( c)
1311 => (Sub8 x (Mul8 <t> (Div8u <t> x (Const8 <t> [c])) (Const8 <t> [c])))
1312(Mod16u <t> x (Const16 [c])) && x.Op != OpConst16 && c > 0 && umagicOK16(c)
1313 => (Sub16 x (Mul16 <t> (Div16u <t> x (Const16 <t> [c])) (Const16 <t> [c])))
1314(Mod32u <t> x (Const32 [c])) && x.Op != OpConst32 && c > 0 && umagicOK32(c)
1315 => (Sub32 x (Mul32 <t> (Div32u <t> x (Const32 <t> [c])) (Const32 <t> [c])))
1316(Mod64u <t> x (Const64 [c])) && x.Op != OpConst64 && c > 0 && umagicOK64(c)
1317 => (Sub64 x (Mul64 <t> (Div64u <t> x (Const64 <t> [c])) (Const64 <t> [c])))
1318
1319// For architectures without rotates on less than 32-bits, promote these checks to 32-bit.
1320(Eq8 (Mod8u x (Const8 [c])) (Const8 [0])) && x.Op != OpConst8 && udivisibleOK8(c) && !hasSmallRotate(config) =>
1321 (Eq32 (Mod32u <typ.UInt32> (ZeroExt8to32 <typ.UInt32> x) (Const32 <typ.UInt32> [int32(uint8(c))])) (Const32 <typ.UInt32> [0]))
1322(Eq16 (Mod16u x (Const16 [c])) (Const16 [0])) && x.Op != OpConst16 && udivisibleOK16(c) && !hasSmallRotate(config) =>
1323 (Eq32 (Mod32u <typ.UInt32> (ZeroExt16to32 <typ.UInt32> x) (Const32 <typ.UInt32> [int32(uint16(c))])) (Const32 <typ.UInt32> [0]))
1324(Eq8 (Mod8 x (Const8 [c])) (Const8 [0])) && x.Op != OpConst8 && sdivisibleOK8(c) && !hasSmallRotate(config) =>
1325 (Eq32 (Mod32 <typ.Int32> (SignExt8to32 <typ.Int32> x) (Const32 <typ.Int32> [int32(c)])) (Const32 <typ.Int32> [0]))
1326(Eq16 (Mod16 x (Const16 [c])) (Const16 [0])) && x.Op != OpConst16 && sdivisibleOK16(c) && !hasSmallRotate(config) =>
1327 (Eq32 (Mod32 <typ.Int32> (SignExt16to32 <typ.Int32> x) (Const32 <typ.Int32> [int32(c)])) (Const32 <typ.Int32> [0]))
1328
1329// Divisibility checks x%c == 0 convert to multiply and rotate.
1330// Note, x%c == 0 is rewritten as x == c*(x/c) during the opt pass
1331// where (x/c) is performed using multiplication with magic constants.
1332// To rewrite x%c == 0 requires pattern matching the rewritten expression
1333// and checking that the division by the same constant wasn't already calculated.
1334// This check is made by counting uses of the magic constant multiplication.
1335// Note that if there were an intermediate opt pass, this rule could be applied
1336// directly on the Div op and magic division rewrites could be delayed to late opt.
1337
1338// Unsigned divisibility checks convert to multiply and rotate.
1339(Eq8 x (Mul8 (Const8 [c])
1340 (Trunc32to8
1341 (Rsh32Ux64
1342 mul:(Mul32
1343 (Const32 [m])
1344 (ZeroExt8to32 x))
1345 (Const64 [s])))
1346 )
1347)
1348 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1349 && m == int32(1<<8+umagic8(c).m) && s == 8+umagic8(c).s
1350 && x.Op != OpConst8 && udivisibleOK8(c)
1351 => (Leq8U
1352 (RotateLeft8 <typ.UInt8>
1353 (Mul8 <typ.UInt8>
1354 (Const8 <typ.UInt8> [int8(udivisible8(c).m)])
1355 x)
1356 (Const8 <typ.UInt8> [int8(8-udivisible8(c).k)])
1357 )
1358 (Const8 <typ.UInt8> [int8(udivisible8(c).max)])
1359 )
1360
1361(Eq16 x (Mul16 (Const16 [c])
1362 (Trunc64to16
1363 (Rsh64Ux64
1364 mul:(Mul64
1365 (Const64 [m])
1366 (ZeroExt16to64 x))
1367 (Const64 [s])))
1368 )
1369)
1370 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1371 && m == int64(1<<16+umagic16(c).m) && s == 16+umagic16(c).s
1372 && x.Op != OpConst16 && udivisibleOK16(c)
1373 => (Leq16U
1374 (RotateLeft16 <typ.UInt16>
1375 (Mul16 <typ.UInt16>
1376 (Const16 <typ.UInt16> [int16(udivisible16(c).m)])
1377 x)
1378 (Const16 <typ.UInt16> [int16(16-udivisible16(c).k)])
1379 )
1380 (Const16 <typ.UInt16> [int16(udivisible16(c).max)])
1381 )
1382
1383(Eq16 x (Mul16 (Const16 [c])
1384 (Trunc32to16
1385 (Rsh32Ux64
1386 mul:(Mul32
1387 (Const32 [m])
1388 (ZeroExt16to32 x))
1389 (Const64 [s])))
1390 )
1391)
1392 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1393 && m == int32(1<<15+umagic16(c).m/2) && s == 16+umagic16(c).s-1
1394 && x.Op != OpConst16 && udivisibleOK16(c)
1395 => (Leq16U
1396 (RotateLeft16 <typ.UInt16>
1397 (Mul16 <typ.UInt16>
1398 (Const16 <typ.UInt16> [int16(udivisible16(c).m)])
1399 x)
1400 (Const16 <typ.UInt16> [int16(16-udivisible16(c).k)])
1401 )
1402 (Const16 <typ.UInt16> [int16(udivisible16(c).max)])
1403 )
1404
1405(Eq16 x (Mul16 (Const16 [c])
1406 (Trunc32to16
1407 (Rsh32Ux64
1408 mul:(Mul32
1409 (Const32 [m])
1410 (Rsh32Ux64 (ZeroExt16to32 x) (Const64 [1])))
1411 (Const64 [s])))
1412 )
1413)
1414 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1415 && m == int32(1<<15+(umagic16(c).m+1)/2) && s == 16+umagic16(c).s-2
1416 && x.Op != OpConst16 && udivisibleOK16(c)
1417 => (Leq16U
1418 (RotateLeft16 <typ.UInt16>
1419 (Mul16 <typ.UInt16>
1420 (Const16 <typ.UInt16> [int16(udivisible16(c).m)])
1421 x)
1422 (Const16 <typ.UInt16> [int16(16-udivisible16(c).k)])
1423 )
1424 (Const16 <typ.UInt16> [int16(udivisible16(c).max)])
1425 )
1426
1427(Eq16 x (Mul16 (Const16 [c])
1428 (Trunc32to16
1429 (Rsh32Ux64
1430 (Avg32u
1431 (Lsh32x64 (ZeroExt16to32 x) (Const64 [16]))
1432 mul:(Mul32
1433 (Const32 [m])
1434 (ZeroExt16to32 x)))
1435 (Const64 [s])))
1436 )
1437)
1438 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1439 && m == int32(umagic16(c).m) && s == 16+umagic16(c).s-1
1440 && x.Op != OpConst16 && udivisibleOK16(c)
1441 => (Leq16U
1442 (RotateLeft16 <typ.UInt16>
1443 (Mul16 <typ.UInt16>
1444 (Const16 <typ.UInt16> [int16(udivisible16(c).m)])
1445 x)
1446 (Const16 <typ.UInt16> [int16(16-udivisible16(c).k)])
1447 )
1448 (Const16 <typ.UInt16> [int16(udivisible16(c).max)])
1449 )
1450
1451(Eq32 x (Mul32 (Const32 [c])
1452 (Rsh32Ux64
1453 mul:(Hmul32u
1454 (Const32 [m])
1455 x)
1456 (Const64 [s]))
1457 )
1458)
1459 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1460 && m == int32(1<<31+umagic32(c).m/2) && s == umagic32(c).s-1
1461 && x.Op != OpConst32 && udivisibleOK32(c)
1462 => (Leq32U
1463 (RotateLeft32 <typ.UInt32>
1464 (Mul32 <typ.UInt32>
1465 (Const32 <typ.UInt32> [int32(udivisible32(c).m)])
1466 x)
1467 (Const32 <typ.UInt32> [int32(32-udivisible32(c).k)])
1468 )
1469 (Const32 <typ.UInt32> [int32(udivisible32(c).max)])
1470 )
1471
1472(Eq32 x (Mul32 (Const32 [c])
1473 (Rsh32Ux64
1474 mul:(Hmul32u
1475 (Const32 <typ.UInt32> [m])
1476 (Rsh32Ux64 x (Const64 [1])))
1477 (Const64 [s]))
1478 )
1479)
1480 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1481 && m == int32(1<<31+(umagic32(c).m+1)/2) && s == umagic32(c).s-2
1482 && x.Op != OpConst32 && udivisibleOK32(c)
1483 => (Leq32U
1484 (RotateLeft32 <typ.UInt32>
1485 (Mul32 <typ.UInt32>
1486 (Const32 <typ.UInt32> [int32(udivisible32(c).m)])
1487 x)
1488 (Const32 <typ.UInt32> [int32(32-udivisible32(c).k)])
1489 )
1490 (Const32 <typ.UInt32> [int32(udivisible32(c).max)])
1491 )
1492
1493(Eq32 x (Mul32 (Const32 [c])
1494 (Rsh32Ux64
1495 (Avg32u
1496 x
1497 mul:(Hmul32u
1498 (Const32 [m])
1499 x))
1500 (Const64 [s]))
1501 )
1502)
1503 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1504 && m == int32(umagic32(c).m) && s == umagic32(c).s-1
1505 && x.Op != OpConst32 && udivisibleOK32(c)
1506 => (Leq32U
1507 (RotateLeft32 <typ.UInt32>
1508 (Mul32 <typ.UInt32>
1509 (Const32 <typ.UInt32> [int32(udivisible32(c).m)])
1510 x)
1511 (Const32 <typ.UInt32> [int32(32-udivisible32(c).k)])
1512 )
1513 (Const32 <typ.UInt32> [int32(udivisible32(c).max)])
1514 )
1515
1516(Eq32 x (Mul32 (Const32 [c])
1517 (Trunc64to32
1518 (Rsh64Ux64
1519 mul:(Mul64
1520 (Const64 [m])
1521 (ZeroExt32to64 x))
1522 (Const64 [s])))
1523 )
1524)
1525 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1526 && m == int64(1<<31+umagic32(c).m/2) && s == 32+umagic32(c).s-1
1527 && x.Op != OpConst32 && udivisibleOK32(c)
1528 => (Leq32U
1529 (RotateLeft32 <typ.UInt32>
1530 (Mul32 <typ.UInt32>
1531 (Const32 <typ.UInt32> [int32(udivisible32(c).m)])
1532 x)
1533 (Const32 <typ.UInt32> [int32(32-udivisible32(c).k)])
1534 )
1535 (Const32 <typ.UInt32> [int32(udivisible32(c).max)])
1536 )
1537
1538(Eq32 x (Mul32 (Const32 [c])
1539 (Trunc64to32
1540 (Rsh64Ux64
1541 mul:(Mul64
1542 (Const64 [m])
1543 (Rsh64Ux64 (ZeroExt32to64 x) (Const64 [1])))
1544 (Const64 [s])))
1545 )
1546)
1547 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1548 && m == int64(1<<31+(umagic32(c).m+1)/2) && s == 32+umagic32(c).s-2
1549 && x.Op != OpConst32 && udivisibleOK32(c)
1550 => (Leq32U
1551 (RotateLeft32 <typ.UInt32>
1552 (Mul32 <typ.UInt32>
1553 (Const32 <typ.UInt32> [int32(udivisible32(c).m)])
1554 x)
1555 (Const32 <typ.UInt32> [int32(32-udivisible32(c).k)])
1556 )
1557 (Const32 <typ.UInt32> [int32(udivisible32(c).max)])
1558 )
1559
1560(Eq32 x (Mul32 (Const32 [c])
1561 (Trunc64to32
1562 (Rsh64Ux64
1563 (Avg64u
1564 (Lsh64x64 (ZeroExt32to64 x) (Const64 [32]))
1565 mul:(Mul64
1566 (Const64 [m])
1567 (ZeroExt32to64 x)))
1568 (Const64 [s])))
1569 )
1570)
1571 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1572 && m == int64(umagic32(c).m) && s == 32+umagic32(c).s-1
1573 && x.Op != OpConst32 && udivisibleOK32(c)
1574 => (Leq32U
1575 (RotateLeft32 <typ.UInt32>
1576 (Mul32 <typ.UInt32>
1577 (Const32 <typ.UInt32> [int32(udivisible32(c).m)])
1578 x)
1579 (Const32 <typ.UInt32> [int32(32-udivisible32(c).k)])
1580 )
1581 (Const32 <typ.UInt32> [int32(udivisible32(c).max)])
1582 )
1583
1584(Eq64 x (Mul64 (Const64 [c])
1585 (Rsh64Ux64
1586 mul:(Hmul64u
1587 (Const64 [m])
1588 x)
1589 (Const64 [s]))
1590 )
1591) && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1592 && m == int64(1<<63+umagic64(c).m/2) && s == umagic64(c).s-1
1593 && x.Op != OpConst64 && udivisibleOK64(c)
1594 => (Leq64U
1595 (RotateLeft64 <typ.UInt64>
1596 (Mul64 <typ.UInt64>
1597 (Const64 <typ.UInt64> [int64(udivisible64(c).m)])
1598 x)
1599 (Const64 <typ.UInt64> [64-udivisible64(c).k])
1600 )
1601 (Const64 <typ.UInt64> [int64(udivisible64(c).max)])
1602 )
1603(Eq64 x (Mul64 (Const64 [c])
1604 (Rsh64Ux64
1605 mul:(Hmul64u
1606 (Const64 [m])
1607 (Rsh64Ux64 x (Const64 [1])))
1608 (Const64 [s]))
1609 )
1610) && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1611 && m == int64(1<<63+(umagic64(c).m+1)/2) && s == umagic64(c).s-2
1612 && x.Op != OpConst64 && udivisibleOK64(c)
1613 => (Leq64U
1614 (RotateLeft64 <typ.UInt64>
1615 (Mul64 <typ.UInt64>
1616 (Const64 <typ.UInt64> [int64(udivisible64(c).m)])
1617 x)
1618 (Const64 <typ.UInt64> [64-udivisible64(c).k])
1619 )
1620 (Const64 <typ.UInt64> [int64(udivisible64(c).max)])
1621 )
1622(Eq64 x (Mul64 (Const64 [c])
1623 (Rsh64Ux64
1624 (Avg64u
1625 x
1626 mul:(Hmul64u
1627 (Const64 [m])
1628 x))
1629 (Const64 [s]))
1630 )
1631) && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1632 && m == int64(umagic64(c).m) && s == umagic64(c).s-1
1633 && x.Op != OpConst64 && udivisibleOK64(c)
1634 => (Leq64U
1635 (RotateLeft64 <typ.UInt64>
1636 (Mul64 <typ.UInt64>
1637 (Const64 <typ.UInt64> [int64(udivisible64(c).m)])
1638 x)
1639 (Const64 <typ.UInt64> [64-udivisible64(c).k])
1640 )
1641 (Const64 <typ.UInt64> [int64(udivisible64(c).max)])
1642 )
1643
1644// Signed divisibility checks convert to multiply, add and rotate.
1645(Eq8 x (Mul8 (Const8 [c])
1646 (Sub8
1647 (Rsh32x64
1648 mul:(Mul32
1649 (Const32 [m])
1650 (SignExt8to32 x))
1651 (Const64 [s]))
1652 (Rsh32x64
1653 (SignExt8to32 x)
1654 (Const64 [31])))
1655 )
1656)
1657 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1658 && m == int32(smagic8(c).m) && s == 8+smagic8(c).s
1659 && x.Op != OpConst8 && sdivisibleOK8(c)
1660 => (Leq8U
1661 (RotateLeft8 <typ.UInt8>
1662 (Add8 <typ.UInt8>
1663 (Mul8 <typ.UInt8>
1664 (Const8 <typ.UInt8> [int8(sdivisible8(c).m)])
1665 x)
1666 (Const8 <typ.UInt8> [int8(sdivisible8(c).a)])
1667 )
1668 (Const8 <typ.UInt8> [int8(8-sdivisible8(c).k)])
1669 )
1670 (Const8 <typ.UInt8> [int8(sdivisible8(c).max)])
1671 )
1672
1673(Eq16 x (Mul16 (Const16 [c])
1674 (Sub16
1675 (Rsh32x64
1676 mul:(Mul32
1677 (Const32 [m])
1678 (SignExt16to32 x))
1679 (Const64 [s]))
1680 (Rsh32x64
1681 (SignExt16to32 x)
1682 (Const64 [31])))
1683 )
1684)
1685 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1686 && m == int32(smagic16(c).m) && s == 16+smagic16(c).s
1687 && x.Op != OpConst16 && sdivisibleOK16(c)
1688 => (Leq16U
1689 (RotateLeft16 <typ.UInt16>
1690 (Add16 <typ.UInt16>
1691 (Mul16 <typ.UInt16>
1692 (Const16 <typ.UInt16> [int16(sdivisible16(c).m)])
1693 x)
1694 (Const16 <typ.UInt16> [int16(sdivisible16(c).a)])
1695 )
1696 (Const16 <typ.UInt16> [int16(16-sdivisible16(c).k)])
1697 )
1698 (Const16 <typ.UInt16> [int16(sdivisible16(c).max)])
1699 )
1700
1701(Eq32 x (Mul32 (Const32 [c])
1702 (Sub32
1703 (Rsh64x64
1704 mul:(Mul64
1705 (Const64 [m])
1706 (SignExt32to64 x))
1707 (Const64 [s]))
1708 (Rsh64x64
1709 (SignExt32to64 x)
1710 (Const64 [63])))
1711 )
1712)
1713 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1714 && m == int64(smagic32(c).m) && s == 32+smagic32(c).s
1715 && x.Op != OpConst32 && sdivisibleOK32(c)
1716 => (Leq32U
1717 (RotateLeft32 <typ.UInt32>
1718 (Add32 <typ.UInt32>
1719 (Mul32 <typ.UInt32>
1720 (Const32 <typ.UInt32> [int32(sdivisible32(c).m)])
1721 x)
1722 (Const32 <typ.UInt32> [int32(sdivisible32(c).a)])
1723 )
1724 (Const32 <typ.UInt32> [int32(32-sdivisible32(c).k)])
1725 )
1726 (Const32 <typ.UInt32> [int32(sdivisible32(c).max)])
1727 )
1728
1729(Eq32 x (Mul32 (Const32 [c])
1730 (Sub32
1731 (Rsh32x64
1732 mul:(Hmul32
1733 (Const32 [m])
1734 x)
1735 (Const64 [s]))
1736 (Rsh32x64
1737 x
1738 (Const64 [31])))
1739 )
1740)
1741 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1742 && m == int32(smagic32(c).m/2) && s == smagic32(c).s-1
1743 && x.Op != OpConst32 && sdivisibleOK32(c)
1744 => (Leq32U
1745 (RotateLeft32 <typ.UInt32>
1746 (Add32 <typ.UInt32>
1747 (Mul32 <typ.UInt32>
1748 (Const32 <typ.UInt32> [int32(sdivisible32(c).m)])
1749 x)
1750 (Const32 <typ.UInt32> [int32(sdivisible32(c).a)])
1751 )
1752 (Const32 <typ.UInt32> [int32(32-sdivisible32(c).k)])
1753 )
1754 (Const32 <typ.UInt32> [int32(sdivisible32(c).max)])
1755 )
1756
1757(Eq32 x (Mul32 (Const32 [c])
1758 (Sub32
1759 (Rsh32x64
1760 (Add32
1761 mul:(Hmul32
1762 (Const32 [m])
1763 x)
1764 x)
1765 (Const64 [s]))
1766 (Rsh32x64
1767 x
1768 (Const64 [31])))
1769 )
1770)
1771 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1772 && m == int32(smagic32(c).m) && s == smagic32(c).s
1773 && x.Op != OpConst32 && sdivisibleOK32(c)
1774 => (Leq32U
1775 (RotateLeft32 <typ.UInt32>
1776 (Add32 <typ.UInt32>
1777 (Mul32 <typ.UInt32>
1778 (Const32 <typ.UInt32> [int32(sdivisible32(c).m)])
1779 x)
1780 (Const32 <typ.UInt32> [int32(sdivisible32(c).a)])
1781 )
1782 (Const32 <typ.UInt32> [int32(32-sdivisible32(c).k)])
1783 )
1784 (Const32 <typ.UInt32> [int32(sdivisible32(c).max)])
1785 )
1786
1787(Eq64 x (Mul64 (Const64 [c])
1788 (Sub64
1789 (Rsh64x64
1790 mul:(Hmul64
1791 (Const64 [m])
1792 x)
1793 (Const64 [s]))
1794 (Rsh64x64
1795 x
1796 (Const64 [63])))
1797 )
1798)
1799 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1800 && m == int64(smagic64(c).m/2) && s == smagic64(c).s-1
1801 && x.Op != OpConst64 && sdivisibleOK64(c)
1802 => (Leq64U
1803 (RotateLeft64 <typ.UInt64>
1804 (Add64 <typ.UInt64>
1805 (Mul64 <typ.UInt64>
1806 (Const64 <typ.UInt64> [int64(sdivisible64(c).m)])
1807 x)
1808 (Const64 <typ.UInt64> [int64(sdivisible64(c).a)])
1809 )
1810 (Const64 <typ.UInt64> [64-sdivisible64(c).k])
1811 )
1812 (Const64 <typ.UInt64> [int64(sdivisible64(c).max)])
1813 )
1814
1815(Eq64 x (Mul64 (Const64 [c])
1816 (Sub64
1817 (Rsh64x64
1818 (Add64
1819 mul:(Hmul64
1820 (Const64 [m])
1821 x)
1822 x)
1823 (Const64 [s]))
1824 (Rsh64x64
1825 x
1826 (Const64 [63])))
1827 )
1828)
1829 && v.Block.Func.pass.name != "opt" && mul.Uses == 1
1830 && m == int64(smagic64(c).m) && s == smagic64(c).s
1831 && x.Op != OpConst64 && sdivisibleOK64(c)
1832 => (Leq64U
1833 (RotateLeft64 <typ.UInt64>
1834 (Add64 <typ.UInt64>
1835 (Mul64 <typ.UInt64>
1836 (Const64 <typ.UInt64> [int64(sdivisible64(c).m)])
1837 x)
1838 (Const64 <typ.UInt64> [int64(sdivisible64(c).a)])
1839 )
1840 (Const64 <typ.UInt64> [64-sdivisible64(c).k])
1841 )
1842 (Const64 <typ.UInt64> [int64(sdivisible64(c).max)])
1843 )
1844
1845// Divisibility check for signed integers for power of two constant are simple mask.
1846// However, we must match against the rewritten n%c == 0 -> n - c*(n/c) == 0 -> n == c*(n/c)
1847// where n/c contains fixup code to handle signed n.
1848((Eq8|Neq8) n (Lsh8x64
1849 (Rsh8x64
1850 (Add8 <t> n (Rsh8Ux64 <t> (Rsh8x64 <t> n (Const64 <typ.UInt64> [ 7])) (Const64 <typ.UInt64> [kbar])))
1851 (Const64 <typ.UInt64> [k]))
1852 (Const64 <typ.UInt64> [k]))
1853) && k > 0 && k < 7 && kbar == 8 - k
1854 => ((Eq8|Neq8) (And8 <t> n (Const8 <t> [1<<uint(k)-1])) (Const8 <t> [0]))
1855
1856((Eq16|Neq16) n (Lsh16x64
1857 (Rsh16x64
1858 (Add16 <t> n (Rsh16Ux64 <t> (Rsh16x64 <t> n (Const64 <typ.UInt64> [15])) (Const64 <typ.UInt64> [kbar])))
1859 (Const64 <typ.UInt64> [k]))
1860 (Const64 <typ.UInt64> [k]))
1861) && k > 0 && k < 15 && kbar == 16 - k
1862 => ((Eq16|Neq16) (And16 <t> n (Const16 <t> [1<<uint(k)-1])) (Const16 <t> [0]))
1863
1864((Eq32|Neq32) n (Lsh32x64
1865 (Rsh32x64
1866 (Add32 <t> n (Rsh32Ux64 <t> (Rsh32x64 <t> n (Const64 <typ.UInt64> [31])) (Const64 <typ.UInt64> [kbar])))
1867 (Const64 <typ.UInt64> [k]))
1868 (Const64 <typ.UInt64> [k]))
1869) && k > 0 && k < 31 && kbar == 32 - k
1870 => ((Eq32|Neq32) (And32 <t> n (Const32 <t> [1<<uint(k)-1])) (Const32 <t> [0]))
1871
1872((Eq64|Neq64) n (Lsh64x64
1873 (Rsh64x64
1874 (Add64 <t> n (Rsh64Ux64 <t> (Rsh64x64 <t> n (Const64 <typ.UInt64> [63])) (Const64 <typ.UInt64> [kbar])))
1875 (Const64 <typ.UInt64> [k]))
1876 (Const64 <typ.UInt64> [k]))
1877) && k > 0 && k < 63 && kbar == 64 - k
1878 => ((Eq64|Neq64) (And64 <t> n (Const64 <t> [1<<uint(k)-1])) (Const64 <t> [0]))
1879
1880(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)
1881(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)
1882
1883// Optimize bitsets
1884(Eq8 (And8 <t> x (Const8 <t> [y])) (Const8 <t> [y])) && oneBit8(y)
1885 => (Neq8 (And8 <t> x (Const8 <t> [y])) (Const8 <t> [0]))
1886(Eq16 (And16 <t> x (Const16 <t> [y])) (Const16 <t> [y])) && oneBit16(y)
1887 => (Neq16 (And16 <t> x (Const16 <t> [y])) (Const16 <t> [0]))
1888(Eq32 (And32 <t> x (Const32 <t> [y])) (Const32 <t> [y])) && oneBit32(y)
1889 => (Neq32 (And32 <t> x (Const32 <t> [y])) (Const32 <t> [0]))
1890(Eq64 (And64 <t> x (Const64 <t> [y])) (Const64 <t> [y])) && oneBit64(y)
1891 => (Neq64 (And64 <t> x (Const64 <t> [y])) (Const64 <t> [0]))
1892(Neq8 (And8 <t> x (Const8 <t> [y])) (Const8 <t> [y])) && oneBit8(y)
1893 => (Eq8 (And8 <t> x (Const8 <t> [y])) (Const8 <t> [0]))
1894(Neq16 (And16 <t> x (Const16 <t> [y])) (Const16 <t> [y])) && oneBit16(y)
1895 => (Eq16 (And16 <t> x (Const16 <t> [y])) (Const16 <t> [0]))
1896(Neq32 (And32 <t> x (Const32 <t> [y])) (Const32 <t> [y])) && oneBit32(y)
1897 => (Eq32 (And32 <t> x (Const32 <t> [y])) (Const32 <t> [0]))
1898(Neq64 (And64 <t> x (Const64 <t> [y])) (Const64 <t> [y])) && oneBit64(y)
1899 => (Eq64 (And64 <t> x (Const64 <t> [y])) (Const64 <t> [0]))
1900
1901// Reassociate expressions involving
1902// constants such that constants come first,
1903// exposing obvious constant-folding opportunities.
1904// Reassociate (op (op y C) x) to (op C (op x y)) or similar, where C
1905// is constant, which pushes constants to the outside
1906// of the expression. At that point, any constant-folding
1907// opportunities should be obvious.
1908// Note: don't include AddPtr here! In order to maintain the
1909// invariant that pointers must stay within the pointed-to object,
1910// we can't pull part of a pointer computation above the AddPtr.
1911// See issue 37881.
1912// Note: we don't need to handle any (x-C) cases because we already rewrite
1913// (x-C) to (x+(-C)).
1914
1915// x + (C + z) -> C + (x + z)
1916(Add64 (Add64 i:(Const64 <t>) z) x) && (z.Op != OpConst64 && x.Op != OpConst64) => (Add64 i (Add64 <t> z x))
1917(Add32 (Add32 i:(Const32 <t>) z) x) && (z.Op != OpConst32 && x.Op != OpConst32) => (Add32 i (Add32 <t> z x))
1918(Add16 (Add16 i:(Const16 <t>) z) x) && (z.Op != OpConst16 && x.Op != OpConst16) => (Add16 i (Add16 <t> z x))
1919(Add8 (Add8 i:(Const8 <t>) z) x) && (z.Op != OpConst8 && x.Op != OpConst8) => (Add8 i (Add8 <t> z x))
1920
1921// x + (C - z) -> C + (x - z)
1922(Add64 (Sub64 i:(Const64 <t>) z) x) && (z.Op != OpConst64 && x.Op != OpConst64) => (Add64 i (Sub64 <t> x z))
1923(Add32 (Sub32 i:(Const32 <t>) z) x) && (z.Op != OpConst32 && x.Op != OpConst32) => (Add32 i (Sub32 <t> x z))
1924(Add16 (Sub16 i:(Const16 <t>) z) x) && (z.Op != OpConst16 && x.Op != OpConst16) => (Add16 i (Sub16 <t> x z))
1925(Add8 (Sub8 i:(Const8 <t>) z) x) && (z.Op != OpConst8 && x.Op != OpConst8) => (Add8 i (Sub8 <t> x z))
1926
1927// x - (C - z) -> x + (z - C) -> (x + z) - C
1928(Sub64 x (Sub64 i:(Const64 <t>) z)) && (z.Op != OpConst64 && x.Op != OpConst64) => (Sub64 (Add64 <t> x z) i)
1929(Sub32 x (Sub32 i:(Const32 <t>) z)) && (z.Op != OpConst32 && x.Op != OpConst32) => (Sub32 (Add32 <t> x z) i)
1930(Sub16 x (Sub16 i:(Const16 <t>) z)) && (z.Op != OpConst16 && x.Op != OpConst16) => (Sub16 (Add16 <t> x z) i)
1931(Sub8 x (Sub8 i:(Const8 <t>) z)) && (z.Op != OpConst8 && x.Op != OpConst8) => (Sub8 (Add8 <t> x z) i)
1932
1933// x - (z + C) -> x + (-z - C) -> (x - z) - C
1934(Sub64 x (Add64 z i:(Const64 <t>))) && (z.Op != OpConst64 && x.Op != OpConst64) => (Sub64 (Sub64 <t> x z) i)
1935(Sub32 x (Add32 z i:(Const32 <t>))) && (z.Op != OpConst32 && x.Op != OpConst32) => (Sub32 (Sub32 <t> x z) i)
1936(Sub16 x (Add16 z i:(Const16 <t>))) && (z.Op != OpConst16 && x.Op != OpConst16) => (Sub16 (Sub16 <t> x z) i)
1937(Sub8 x (Add8 z i:(Const8 <t>))) && (z.Op != OpConst8 && x.Op != OpConst8) => (Sub8 (Sub8 <t> x z) i)
1938
1939// (C - z) - x -> C - (z + x)
1940(Sub64 (Sub64 i:(Const64 <t>) z) x) && (z.Op != OpConst64 && x.Op != OpConst64) => (Sub64 i (Add64 <t> z x))
1941(Sub32 (Sub32 i:(Const32 <t>) z) x) && (z.Op != OpConst32 && x.Op != OpConst32) => (Sub32 i (Add32 <t> z x))
1942(Sub16 (Sub16 i:(Const16 <t>) z) x) && (z.Op != OpConst16 && x.Op != OpConst16) => (Sub16 i (Add16 <t> z x))
1943(Sub8 (Sub8 i:(Const8 <t>) z) x) && (z.Op != OpConst8 && x.Op != OpConst8) => (Sub8 i (Add8 <t> z x))
1944
1945// (z + C) -x -> C + (z - x)
1946(Sub64 (Add64 z i:(Const64 <t>)) x) && (z.Op != OpConst64 && x.Op != OpConst64) => (Add64 i (Sub64 <t> z x))
1947(Sub32 (Add32 z i:(Const32 <t>)) x) && (z.Op != OpConst32 && x.Op != OpConst32) => (Add32 i (Sub32 <t> z x))
1948(Sub16 (Add16 z i:(Const16 <t>)) x) && (z.Op != OpConst16 && x.Op != OpConst16) => (Add16 i (Sub16 <t> z x))
1949(Sub8 (Add8 z i:(Const8 <t>)) x) && (z.Op != OpConst8 && x.Op != OpConst8) => (Add8 i (Sub8 <t> z x))
1950
1951// x & (C & z) -> C & (x & z)
1952(And64 (And64 i:(Const64 <t>) z) x) && (z.Op != OpConst64 && x.Op != OpConst64) => (And64 i (And64 <t> z x))
1953(And32 (And32 i:(Const32 <t>) z) x) && (z.Op != OpConst32 && x.Op != OpConst32) => (And32 i (And32 <t> z x))
1954(And16 (And16 i:(Const16 <t>) z) x) && (z.Op != OpConst16 && x.Op != OpConst16) => (And16 i (And16 <t> z x))
1955(And8 (And8 i:(Const8 <t>) z) x) && (z.Op != OpConst8 && x.Op != OpConst8) => (And8 i (And8 <t> z x))
1956
1957// x | (C | z) -> C | (x | z)
1958(Or64 (Or64 i:(Const64 <t>) z) x) && (z.Op != OpConst64 && x.Op != OpConst64) => (Or64 i (Or64 <t> z x))
1959(Or32 (Or32 i:(Const32 <t>) z) x) && (z.Op != OpConst32 && x.Op != OpConst32) => (Or32 i (Or32 <t> z x))
1960(Or16 (Or16 i:(Const16 <t>) z) x) && (z.Op != OpConst16 && x.Op != OpConst16) => (Or16 i (Or16 <t> z x))
1961(Or8 (Or8 i:(Const8 <t>) z) x) && (z.Op != OpConst8 && x.Op != OpConst8) => (Or8 i (Or8 <t> z x))
1962
1963// x ^ (C ^ z) -> C ^ (x ^ z)
1964(Xor64 (Xor64 i:(Const64 <t>) z) x) && (z.Op != OpConst64 && x.Op != OpConst64) => (Xor64 i (Xor64 <t> z x))
1965(Xor32 (Xor32 i:(Const32 <t>) z) x) && (z.Op != OpConst32 && x.Op != OpConst32) => (Xor32 i (Xor32 <t> z x))
1966(Xor16 (Xor16 i:(Const16 <t>) z) x) && (z.Op != OpConst16 && x.Op != OpConst16) => (Xor16 i (Xor16 <t> z x))
1967(Xor8 (Xor8 i:(Const8 <t>) z) x) && (z.Op != OpConst8 && x.Op != OpConst8) => (Xor8 i (Xor8 <t> z x))
1968
1969// x * (D * z) = D * (x * z)
1970(Mul64 (Mul64 i:(Const64 <t>) z) x) && (z.Op != OpConst64 && x.Op != OpConst64) => (Mul64 i (Mul64 <t> x z))
1971(Mul32 (Mul32 i:(Const32 <t>) z) x) && (z.Op != OpConst32 && x.Op != OpConst32) => (Mul32 i (Mul32 <t> x z))
1972(Mul16 (Mul16 i:(Const16 <t>) z) x) && (z.Op != OpConst16 && x.Op != OpConst16) => (Mul16 i (Mul16 <t> x z))
1973(Mul8 (Mul8 i:(Const8 <t>) z) x) && (z.Op != OpConst8 && x.Op != OpConst8) => (Mul8 i (Mul8 <t> x z))
1974
1975// C + (D + x) -> (C + D) + x
1976(Add64 (Const64 <t> [c]) (Add64 (Const64 <t> [d]) x)) => (Add64 (Const64 <t> [c+d]) x)
1977(Add32 (Const32 <t> [c]) (Add32 (Const32 <t> [d]) x)) => (Add32 (Const32 <t> [c+d]) x)
1978(Add16 (Const16 <t> [c]) (Add16 (Const16 <t> [d]) x)) => (Add16 (Const16 <t> [c+d]) x)
1979(Add8 (Const8 <t> [c]) (Add8 (Const8 <t> [d]) x)) => (Add8 (Const8 <t> [c+d]) x)
1980
1981// C + (D - x) -> (C + D) - x
1982(Add64 (Const64 <t> [c]) (Sub64 (Const64 <t> [d]) x)) => (Sub64 (Const64 <t> [c+d]) x)
1983(Add32 (Const32 <t> [c]) (Sub32 (Const32 <t> [d]) x)) => (Sub32 (Const32 <t> [c+d]) x)
1984(Add16 (Const16 <t> [c]) (Sub16 (Const16 <t> [d]) x)) => (Sub16 (Const16 <t> [c+d]) x)
1985(Add8 (Const8 <t> [c]) (Sub8 (Const8 <t> [d]) x)) => (Sub8 (Const8 <t> [c+d]) x)
1986
1987// C - (D - x) -> (C - D) + x
1988(Sub64 (Const64 <t> [c]) (Sub64 (Const64 <t> [d]) x)) => (Add64 (Const64 <t> [c-d]) x)
1989(Sub32 (Const32 <t> [c]) (Sub32 (Const32 <t> [d]) x)) => (Add32 (Const32 <t> [c-d]) x)
1990(Sub16 (Const16 <t> [c]) (Sub16 (Const16 <t> [d]) x)) => (Add16 (Const16 <t> [c-d]) x)
1991(Sub8 (Const8 <t> [c]) (Sub8 (Const8 <t> [d]) x)) => (Add8 (Const8 <t> [c-d]) x)
1992
1993// C - (D + x) -> (C - D) - x
1994(Sub64 (Const64 <t> [c]) (Add64 (Const64 <t> [d]) x)) => (Sub64 (Const64 <t> [c-d]) x)
1995(Sub32 (Const32 <t> [c]) (Add32 (Const32 <t> [d]) x)) => (Sub32 (Const32 <t> [c-d]) x)
1996(Sub16 (Const16 <t> [c]) (Add16 (Const16 <t> [d]) x)) => (Sub16 (Const16 <t> [c-d]) x)
1997(Sub8 (Const8 <t> [c]) (Add8 (Const8 <t> [d]) x)) => (Sub8 (Const8 <t> [c-d]) x)
1998
1999// C & (D & x) -> (C & D) & x
2000(And64 (Const64 <t> [c]) (And64 (Const64 <t> [d]) x)) => (And64 (Const64 <t> [c&d]) x)
2001(And32 (Const32 <t> [c]) (And32 (Const32 <t> [d]) x)) => (And32 (Const32 <t> [c&d]) x)
2002(And16 (Const16 <t> [c]) (And16 (Const16 <t> [d]) x)) => (And16 (Const16 <t> [c&d]) x)
2003(And8 (Const8 <t> [c]) (And8 (Const8 <t> [d]) x)) => (And8 (Const8 <t> [c&d]) x)
2004
2005// C | (D | x) -> (C | D) | x
2006(Or64 (Const64 <t> [c]) (Or64 (Const64 <t> [d]) x)) => (Or64 (Const64 <t> [c|d]) x)
2007(Or32 (Const32 <t> [c]) (Or32 (Const32 <t> [d]) x)) => (Or32 (Const32 <t> [c|d]) x)
2008(Or16 (Const16 <t> [c]) (Or16 (Const16 <t> [d]) x)) => (Or16 (Const16 <t> [c|d]) x)
2009(Or8 (Const8 <t> [c]) (Or8 (Const8 <t> [d]) x)) => (Or8 (Const8 <t> [c|d]) x)
2010
2011// C ^ (D ^ x) -> (C ^ D) ^ x
2012(Xor64 (Const64 <t> [c]) (Xor64 (Const64 <t> [d]) x)) => (Xor64 (Const64 <t> [c^d]) x)
2013(Xor32 (Const32 <t> [c]) (Xor32 (Const32 <t> [d]) x)) => (Xor32 (Const32 <t> [c^d]) x)
2014(Xor16 (Const16 <t> [c]) (Xor16 (Const16 <t> [d]) x)) => (Xor16 (Const16 <t> [c^d]) x)
2015(Xor8 (Const8 <t> [c]) (Xor8 (Const8 <t> [d]) x)) => (Xor8 (Const8 <t> [c^d]) x)
2016
2017// C * (D * x) = (C * D) * x
2018(Mul64 (Const64 <t> [c]) (Mul64 (Const64 <t> [d]) x)) => (Mul64 (Const64 <t> [c*d]) x)
2019(Mul32 (Const32 <t> [c]) (Mul32 (Const32 <t> [d]) x)) => (Mul32 (Const32 <t> [c*d]) x)
2020(Mul16 (Const16 <t> [c]) (Mul16 (Const16 <t> [d]) x)) => (Mul16 (Const16 <t> [c*d]) x)
2021(Mul8 (Const8 <t> [c]) (Mul8 (Const8 <t> [d]) x)) => (Mul8 (Const8 <t> [c*d]) x)
2022
2023// floating point optimizations
2024(Mul(32|64)F x (Const(32|64)F [1])) => x
2025(Mul32F x (Const32F [-1])) => (Neg32F x)
2026(Mul64F x (Const64F [-1])) => (Neg64F x)
2027(Mul32F x (Const32F [2])) => (Add32F x x)
2028(Mul64F x (Const64F [2])) => (Add64F x x)
2029
2030(Div32F x (Const32F <t> [c])) && reciprocalExact32(c) => (Mul32F x (Const32F <t> [1/c]))
2031(Div64F x (Const64F <t> [c])) && reciprocalExact64(c) => (Mul64F x (Const64F <t> [1/c]))
2032
2033// rewrite single-precision sqrt expression "float32(math.Sqrt(float64(x)))"
2034(Cvt64Fto32F sqrt0:(Sqrt (Cvt32Fto64F x))) && sqrt0.Uses==1 => (Sqrt32 x)
2035
2036(Sqrt (Const64F [c])) && !math.IsNaN(math.Sqrt(c)) => (Const64F [math.Sqrt(c)])
2037
2038// for rewriting results of some late-expanded rewrites (below)
2039(SelectN [0] (MakeResult x ___)) => x
2040(SelectN [1] (MakeResult x y ___)) => y
2041(SelectN [2] (MakeResult x y z ___)) => z
2042
2043// for late-expanded calls, recognize newobject and remove zeroing and nilchecks
2044(Zero (SelectN [0] call:(StaticLECall _ _)) mem:(SelectN [1] call))
2045 && isSameCall(call.Aux, "runtime.newobject")
2046 => mem
2047
2048(Store (SelectN [0] call:(StaticLECall _ _)) x mem:(SelectN [1] call))
2049 && isConstZero(x)
2050 && isSameCall(call.Aux, "runtime.newobject")
2051 => mem
2052
2053(Store (OffPtr (SelectN [0] call:(StaticLECall _ _))) x mem:(SelectN [1] call))
2054 && isConstZero(x)
2055 && isSameCall(call.Aux, "runtime.newobject")
2056 => mem
2057
2058(NilCheck ptr:(SelectN [0] call:(StaticLECall _ _)) _)
2059 && isSameCall(call.Aux, "runtime.newobject")
2060 && warnRule(fe.Debug_checknil(), v, "removed nil check")
2061 => ptr
2062
2063(NilCheck ptr:(OffPtr (SelectN [0] call:(StaticLECall _ _))) _)
2064 && isSameCall(call.Aux, "runtime.newobject")
2065 && warnRule(fe.Debug_checknil(), v, "removed nil check")
2066 => ptr
2067
2068// Addresses of globals are always non-nil.
2069(NilCheck ptr:(Addr {_} (SB)) _) => ptr
2070(NilCheck ptr:(Convert (Addr {_} (SB)) _) _) => ptr
2071
2072// for late-expanded calls, recognize memequal applied to a single constant byte
2073// Support is limited by 1, 2, 4, 8 byte sizes
2074(StaticLECall {callAux} sptr (Addr {scon} (SB)) (Const64 [1]) mem)
2075 && isSameCall(callAux, "runtime.memequal")
2076 && symIsRO(scon)
2077 => (MakeResult (Eq8 (Load <typ.Int8> sptr mem) (Const8 <typ.Int8> [int8(read8(scon,0))])) mem)
2078
2079(StaticLECall {callAux} (Addr {scon} (SB)) sptr (Const64 [1]) mem)
2080 && isSameCall(callAux, "runtime.memequal")
2081 && symIsRO(scon)
2082 => (MakeResult (Eq8 (Load <typ.Int8> sptr mem) (Const8 <typ.Int8> [int8(read8(scon,0))])) mem)
2083
2084(StaticLECall {callAux} sptr (Addr {scon} (SB)) (Const64 [2]) mem)
2085 && isSameCall(callAux, "runtime.memequal")
2086 && symIsRO(scon)
2087 && canLoadUnaligned(config)
2088 => (MakeResult (Eq16 (Load <typ.Int16> sptr mem) (Const16 <typ.Int16> [int16(read16(scon,0,config.ctxt.Arch.ByteOrder))])) mem)
2089
2090(StaticLECall {callAux} (Addr {scon} (SB)) sptr (Const64 [2]) mem)
2091 && isSameCall(callAux, "runtime.memequal")
2092 && symIsRO(scon)
2093 && canLoadUnaligned(config)
2094 => (MakeResult (Eq16 (Load <typ.Int16> sptr mem) (Const16 <typ.Int16> [int16(read16(scon,0,config.ctxt.Arch.ByteOrder))])) mem)
2095
2096(StaticLECall {callAux} sptr (Addr {scon} (SB)) (Const64 [4]) mem)
2097 && isSameCall(callAux, "runtime.memequal")
2098 && symIsRO(scon)
2099 && canLoadUnaligned(config)
2100 => (MakeResult (Eq32 (Load <typ.Int32> sptr mem) (Const32 <typ.Int32> [int32(read32(scon,0,config.ctxt.Arch.ByteOrder))])) mem)
2101
2102(StaticLECall {callAux} (Addr {scon} (SB)) sptr (Const64 [4]) mem)
2103 && isSameCall(callAux, "runtime.memequal")
2104 && symIsRO(scon)
2105 && canLoadUnaligned(config)
2106 => (MakeResult (Eq32 (Load <typ.Int32> sptr mem) (Const32 <typ.Int32> [int32(read32(scon,0,config.ctxt.Arch.ByteOrder))])) mem)
2107
2108(StaticLECall {callAux} sptr (Addr {scon} (SB)) (Const64 [8]) mem)
2109 && isSameCall(callAux, "runtime.memequal")
2110 && symIsRO(scon)
2111 && canLoadUnaligned(config) && config.PtrSize == 8
2112 => (MakeResult (Eq64 (Load <typ.Int64> sptr mem) (Const64 <typ.Int64> [int64(read64(scon,0,config.ctxt.Arch.ByteOrder))])) mem)
2113
2114(StaticLECall {callAux} (Addr {scon} (SB)) sptr (Const64 [8]) mem)
2115 && isSameCall(callAux, "runtime.memequal")
2116 && symIsRO(scon)
2117 && canLoadUnaligned(config) && config.PtrSize == 8
2118 => (MakeResult (Eq64 (Load <typ.Int64> sptr mem) (Const64 <typ.Int64> [int64(read64(scon,0,config.ctxt.Arch.ByteOrder))])) mem)
2119
2120(StaticLECall {callAux} _ _ (Const64 [0]) mem)
2121 && isSameCall(callAux, "runtime.memequal")
2122 => (MakeResult (ConstBool <typ.Bool> [true]) mem)
2123
2124(Static(Call|LECall) {callAux} p q _ mem)
2125 && isSameCall(callAux, "runtime.memequal")
2126 && isSamePtr(p, q)
2127 => (MakeResult (ConstBool <typ.Bool> [true]) mem)
2128
2129// Turn known-size calls to memclrNoHeapPointers into a Zero.
2130// Note that we are using types.Types[types.TUINT8] instead of sptr.Type.Elem() - see issue 55122 and CL 431496 for more details.
2131(SelectN [0] call:(StaticCall {sym} sptr (Const(64|32) [c]) mem))
2132 && isInlinableMemclr(config, int64(c))
2133 && isSameCall(sym, "runtime.memclrNoHeapPointers")
2134 && call.Uses == 1
2135 && clobber(call)
2136 => (Zero {types.Types[types.TUINT8]} [int64(c)] sptr mem)
2137
2138// Recognise make([]T, 0) and replace it with a pointer to the zerobase
2139(StaticLECall {callAux} _ (Const(64|32) [0]) (Const(64|32) [0]) mem)
2140 && isSameCall(callAux, "runtime.makeslice")
2141 => (MakeResult (Addr <v.Type.FieldType(0)> {ir.Syms.Zerobase} (SB)) mem)
2142
2143// Evaluate constant address comparisons.
2144(EqPtr x x) => (ConstBool [true])
2145(NeqPtr x x) => (ConstBool [false])
2146(EqPtr (Addr {x} _) (Addr {y} _)) => (ConstBool [x == y])
2147(EqPtr (Addr {x} _) (OffPtr [o] (Addr {y} _))) => (ConstBool [x == y && o == 0])
2148(EqPtr (OffPtr [o1] (Addr {x} _)) (OffPtr [o2] (Addr {y} _))) => (ConstBool [x == y && o1 == o2])
2149(NeqPtr (Addr {x} _) (Addr {y} _)) => (ConstBool [x != y])
2150(NeqPtr (Addr {x} _) (OffPtr [o] (Addr {y} _))) => (ConstBool [x != y || o != 0])
2151(NeqPtr (OffPtr [o1] (Addr {x} _)) (OffPtr [o2] (Addr {y} _))) => (ConstBool [x != y || o1 != o2])
2152(EqPtr (LocalAddr {x} _ _) (LocalAddr {y} _ _)) => (ConstBool [x == y])
2153(EqPtr (LocalAddr {x} _ _) (OffPtr [o] (LocalAddr {y} _ _))) => (ConstBool [x == y && o == 0])
2154(EqPtr (OffPtr [o1] (LocalAddr {x} _ _)) (OffPtr [o2] (LocalAddr {y} _ _))) => (ConstBool [x == y && o1 == o2])
2155(NeqPtr (LocalAddr {x} _ _) (LocalAddr {y} _ _)) => (ConstBool [x != y])
2156(NeqPtr (LocalAddr {x} _ _) (OffPtr [o] (LocalAddr {y} _ _))) => (ConstBool [x != y || o != 0])
2157(NeqPtr (OffPtr [o1] (LocalAddr {x} _ _)) (OffPtr [o2] (LocalAddr {y} _ _))) => (ConstBool [x != y || o1 != o2])
2158(EqPtr (OffPtr [o1] p1) p2) && isSamePtr(p1, p2) => (ConstBool [o1 == 0])
2159(NeqPtr (OffPtr [o1] p1) p2) && isSamePtr(p1, p2) => (ConstBool [o1 != 0])
2160(EqPtr (OffPtr [o1] p1) (OffPtr [o2] p2)) && isSamePtr(p1, p2) => (ConstBool [o1 == o2])
2161(NeqPtr (OffPtr [o1] p1) (OffPtr [o2] p2)) && isSamePtr(p1, p2) => (ConstBool [o1 != o2])
2162(EqPtr (Const(32|64) [c]) (Const(32|64) [d])) => (ConstBool [c == d])
2163(NeqPtr (Const(32|64) [c]) (Const(32|64) [d])) => (ConstBool [c != d])
2164(EqPtr (Convert (Addr {x} _) _) (Addr {y} _)) => (ConstBool [x==y])
2165(NeqPtr (Convert (Addr {x} _) _) (Addr {y} _)) => (ConstBool [x!=y])
2166
2167(EqPtr (LocalAddr _ _) (Addr _)) => (ConstBool [false])
2168(EqPtr (OffPtr (LocalAddr _ _)) (Addr _)) => (ConstBool [false])
2169(EqPtr (LocalAddr _ _) (OffPtr (Addr _))) => (ConstBool [false])
2170(EqPtr (OffPtr (LocalAddr _ _)) (OffPtr (Addr _))) => (ConstBool [false])
2171(NeqPtr (LocalAddr _ _) (Addr _)) => (ConstBool [true])
2172(NeqPtr (OffPtr (LocalAddr _ _)) (Addr _)) => (ConstBool [true])
2173(NeqPtr (LocalAddr _ _) (OffPtr (Addr _))) => (ConstBool [true])
2174(NeqPtr (OffPtr (LocalAddr _ _)) (OffPtr (Addr _))) => (ConstBool [true])
2175
2176// Simplify address comparisons.
2177(EqPtr (AddPtr p1 o1) p2) && isSamePtr(p1, p2) => (Not (IsNonNil o1))
2178(NeqPtr (AddPtr p1 o1) p2) && isSamePtr(p1, p2) => (IsNonNil o1)
2179(EqPtr (Const(32|64) [0]) p) => (Not (IsNonNil p))
2180(NeqPtr (Const(32|64) [0]) p) => (IsNonNil p)
2181(EqPtr (ConstNil) p) => (Not (IsNonNil p))
2182(NeqPtr (ConstNil) p) => (IsNonNil p)
2183
2184// Evaluate constant user nil checks.
2185(IsNonNil (ConstNil)) => (ConstBool [false])
2186(IsNonNil (Const(32|64) [c])) => (ConstBool [c != 0])
2187(IsNonNil (Addr _) ) => (ConstBool [true])
2188(IsNonNil (Convert (Addr _) _)) => (ConstBool [true])
2189(IsNonNil (LocalAddr _ _)) => (ConstBool [true])
2190
2191// Inline small or disjoint runtime.memmove calls with constant length.
2192// See the comment in op Move in genericOps.go for discussion of the type.
2193//
2194// Note that we've lost any knowledge of the type and alignment requirements
2195// of the source and destination. We only know the size, and that the type
2196// contains no pointers.
2197// The type of the move is not necessarily v.Args[0].Type().Elem()!
2198// See issue 55122 for details.
2199//
2200// Because expand calls runs after prove, constants useful to this pattern may not appear.
2201// Both versions need to exist; the memory and register variants.
2202//
2203// Match post-expansion calls, memory version.
2204(SelectN [0] call:(StaticCall {sym} s1:(Store _ (Const(64|32) [sz]) s2:(Store _ src s3:(Store {t} _ dst mem)))))
2205 && sz >= 0
2206 && isSameCall(sym, "runtime.memmove")
2207 && s1.Uses == 1 && s2.Uses == 1 && s3.Uses == 1
2208 && isInlinableMemmove(dst, src, int64(sz), config)
2209 && clobber(s1, s2, s3, call)
2210 => (Move {types.Types[types.TUINT8]} [int64(sz)] dst src mem)
2211
2212// Match post-expansion calls, register version.
2213(SelectN [0] call:(StaticCall {sym} dst src (Const(64|32) [sz]) mem))
2214 && sz >= 0
2215 && call.Uses == 1 // this will exclude all calls with results
2216 && isSameCall(sym, "runtime.memmove")
2217 && isInlinableMemmove(dst, src, int64(sz), config)
2218 && clobber(call)
2219 => (Move {types.Types[types.TUINT8]} [int64(sz)] dst src mem)
2220
2221// Match pre-expansion calls.
2222(SelectN [0] call:(StaticLECall {sym} dst src (Const(64|32) [sz]) mem))
2223 && sz >= 0
2224 && call.Uses == 1 // this will exclude all calls with results
2225 && isSameCall(sym, "runtime.memmove")
2226 && isInlinableMemmove(dst, src, int64(sz), config)
2227 && clobber(call)
2228 => (Move {types.Types[types.TUINT8]} [int64(sz)] dst src mem)
2229
2230// De-virtualize late-expanded interface calls into late-expanded static calls.
2231(InterLECall [argsize] {auxCall} (Addr {fn} (SB)) ___) => devirtLECall(v, fn.(*obj.LSym))
2232
2233// Move and Zero optimizations.
2234// Move source and destination may overlap.
2235
2236// Convert Moves into Zeros when the source is known to be zeros.
2237(Move {t} [n] dst1 src mem:(Zero {t} [n] dst2 _)) && isSamePtr(src, dst2)
2238 => (Zero {t} [n] dst1 mem)
2239(Move {t} [n] dst1 src mem:(VarDef (Zero {t} [n] dst0 _))) && isSamePtr(src, dst0)
2240 => (Zero {t} [n] dst1 mem)
2241(Move {t} [n] dst (Addr {sym} (SB)) mem) && symIsROZero(sym) => (Zero {t} [n] dst mem)
2242
2243// Don't Store to variables that are about to be overwritten by Move/Zero.
2244(Zero {t1} [n] p1 store:(Store {t2} (OffPtr [o2] p2) _ mem))
2245 && isSamePtr(p1, p2) && store.Uses == 1
2246 && n >= o2 + t2.Size()
2247 && clobber(store)
2248 => (Zero {t1} [n] p1 mem)
2249(Move {t1} [n] dst1 src1 store:(Store {t2} op:(OffPtr [o2] dst2) _ mem))
2250 && isSamePtr(dst1, dst2) && store.Uses == 1
2251 && n >= o2 + t2.Size()
2252 && disjoint(src1, n, op, t2.Size())
2253 && clobber(store)
2254 => (Move {t1} [n] dst1 src1 mem)
2255
2256// Don't Move to variables that are immediately completely overwritten.
2257(Zero {t} [n] dst1 move:(Move {t} [n] dst2 _ mem))
2258 && move.Uses == 1
2259 && isSamePtr(dst1, dst2)
2260 && clobber(move)
2261 => (Zero {t} [n] dst1 mem)
2262(Move {t} [n] dst1 src1 move:(Move {t} [n] dst2 _ mem))
2263 && move.Uses == 1
2264 && isSamePtr(dst1, dst2) && disjoint(src1, n, dst2, n)
2265 && clobber(move)
2266 => (Move {t} [n] dst1 src1 mem)
2267(Zero {t} [n] dst1 vardef:(VarDef {x} move:(Move {t} [n] dst2 _ mem)))
2268 && move.Uses == 1 && vardef.Uses == 1
2269 && isSamePtr(dst1, dst2)
2270 && clobber(move, vardef)
2271 => (Zero {t} [n] dst1 (VarDef {x} mem))
2272(Move {t} [n] dst1 src1 vardef:(VarDef {x} move:(Move {t} [n] dst2 _ mem)))
2273 && move.Uses == 1 && vardef.Uses == 1
2274 && isSamePtr(dst1, dst2) && disjoint(src1, n, dst2, n)
2275 && clobber(move, vardef)
2276 => (Move {t} [n] dst1 src1 (VarDef {x} mem))
2277(Store {t1} op1:(OffPtr [o1] p1) d1
2278 m2:(Store {t2} op2:(OffPtr [0] p2) d2
2279 m3:(Move [n] p3 _ mem)))
2280 && m2.Uses == 1 && m3.Uses == 1
2281 && o1 == t2.Size()
2282 && n == t2.Size() + t1.Size()
2283 && isSamePtr(p1, p2) && isSamePtr(p2, p3)
2284 && clobber(m2, m3)
2285 => (Store {t1} op1 d1 (Store {t2} op2 d2 mem))
2286(Store {t1} op1:(OffPtr [o1] p1) d1
2287 m2:(Store {t2} op2:(OffPtr [o2] p2) d2
2288 m3:(Store {t3} op3:(OffPtr [0] p3) d3
2289 m4:(Move [n] p4 _ mem))))
2290 && m2.Uses == 1 && m3.Uses == 1 && m4.Uses == 1
2291 && o2 == t3.Size()
2292 && o1-o2 == t2.Size()
2293 && n == t3.Size() + t2.Size() + t1.Size()
2294 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4)
2295 && clobber(m2, m3, m4)
2296 => (Store {t1} op1 d1 (Store {t2} op2 d2 (Store {t3} op3 d3 mem)))
2297(Store {t1} op1:(OffPtr [o1] p1) d1
2298 m2:(Store {t2} op2:(OffPtr [o2] p2) d2
2299 m3:(Store {t3} op3:(OffPtr [o3] p3) d3
2300 m4:(Store {t4} op4:(OffPtr [0] p4) d4
2301 m5:(Move [n] p5 _ mem)))))
2302 && m2.Uses == 1 && m3.Uses == 1 && m4.Uses == 1 && m5.Uses == 1
2303 && o3 == t4.Size()
2304 && o2-o3 == t3.Size()
2305 && o1-o2 == t2.Size()
2306 && n == t4.Size() + t3.Size() + t2.Size() + t1.Size()
2307 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4) && isSamePtr(p4, p5)
2308 && clobber(m2, m3, m4, m5)
2309 => (Store {t1} op1 d1 (Store {t2} op2 d2 (Store {t3} op3 d3 (Store {t4} op4 d4 mem))))
2310
2311// Don't Zero variables that are immediately completely overwritten
2312// before being accessed.
2313(Move {t} [n] dst1 src1 zero:(Zero {t} [n] dst2 mem))
2314 && zero.Uses == 1
2315 && isSamePtr(dst1, dst2) && disjoint(src1, n, dst2, n)
2316 && clobber(zero)
2317 => (Move {t} [n] dst1 src1 mem)
2318(Move {t} [n] dst1 src1 vardef:(VarDef {x} zero:(Zero {t} [n] dst2 mem)))
2319 && zero.Uses == 1 && vardef.Uses == 1
2320 && isSamePtr(dst1, dst2) && disjoint(src1, n, dst2, n)
2321 && clobber(zero, vardef)
2322 => (Move {t} [n] dst1 src1 (VarDef {x} mem))
2323(Store {t1} op1:(OffPtr [o1] p1) d1
2324 m2:(Store {t2} op2:(OffPtr [0] p2) d2
2325 m3:(Zero [n] p3 mem)))
2326 && m2.Uses == 1 && m3.Uses == 1
2327 && o1 == t2.Size()
2328 && n == t2.Size() + t1.Size()
2329 && isSamePtr(p1, p2) && isSamePtr(p2, p3)
2330 && clobber(m2, m3)
2331 => (Store {t1} op1 d1 (Store {t2} op2 d2 mem))
2332(Store {t1} op1:(OffPtr [o1] p1) d1
2333 m2:(Store {t2} op2:(OffPtr [o2] p2) d2
2334 m3:(Store {t3} op3:(OffPtr [0] p3) d3
2335 m4:(Zero [n] p4 mem))))
2336 && m2.Uses == 1 && m3.Uses == 1 && m4.Uses == 1
2337 && o2 == t3.Size()
2338 && o1-o2 == t2.Size()
2339 && n == t3.Size() + t2.Size() + t1.Size()
2340 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4)
2341 && clobber(m2, m3, m4)
2342 => (Store {t1} op1 d1 (Store {t2} op2 d2 (Store {t3} op3 d3 mem)))
2343(Store {t1} op1:(OffPtr [o1] p1) d1
2344 m2:(Store {t2} op2:(OffPtr [o2] p2) d2
2345 m3:(Store {t3} op3:(OffPtr [o3] p3) d3
2346 m4:(Store {t4} op4:(OffPtr [0] p4) d4
2347 m5:(Zero [n] p5 mem)))))
2348 && m2.Uses == 1 && m3.Uses == 1 && m4.Uses == 1 && m5.Uses == 1
2349 && o3 == t4.Size()
2350 && o2-o3 == t3.Size()
2351 && o1-o2 == t2.Size()
2352 && n == t4.Size() + t3.Size() + t2.Size() + t1.Size()
2353 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4) && isSamePtr(p4, p5)
2354 && clobber(m2, m3, m4, m5)
2355 => (Store {t1} op1 d1 (Store {t2} op2 d2 (Store {t3} op3 d3 (Store {t4} op4 d4 mem))))
2356
2357// Don't Move from memory if the values are likely to already be
2358// in registers.
2359(Move {t1} [n] dst p1
2360 mem:(Store {t2} op2:(OffPtr <tt2> [o2] p2) d1
2361 (Store {t3} op3:(OffPtr <tt3> [0] p3) d2 _)))
2362 && isSamePtr(p1, p2) && isSamePtr(p2, p3)
2363 && t2.Alignment() <= t1.Alignment()
2364 && t3.Alignment() <= t1.Alignment()
2365 && registerizable(b, t2)
2366 && registerizable(b, t3)
2367 && o2 == t3.Size()
2368 && n == t2.Size() + t3.Size()
2369 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2370 (Store {t3} (OffPtr <tt3> [0] dst) d2 mem))
2371(Move {t1} [n] dst p1
2372 mem:(Store {t2} op2:(OffPtr <tt2> [o2] p2) d1
2373 (Store {t3} op3:(OffPtr <tt3> [o3] p3) d2
2374 (Store {t4} op4:(OffPtr <tt4> [0] p4) d3 _))))
2375 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4)
2376 && t2.Alignment() <= t1.Alignment()
2377 && t3.Alignment() <= t1.Alignment()
2378 && t4.Alignment() <= t1.Alignment()
2379 && registerizable(b, t2)
2380 && registerizable(b, t3)
2381 && registerizable(b, t4)
2382 && o3 == t4.Size()
2383 && o2-o3 == t3.Size()
2384 && n == t2.Size() + t3.Size() + t4.Size()
2385 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2386 (Store {t3} (OffPtr <tt3> [o3] dst) d2
2387 (Store {t4} (OffPtr <tt4> [0] dst) d3 mem)))
2388(Move {t1} [n] dst p1
2389 mem:(Store {t2} op2:(OffPtr <tt2> [o2] p2) d1
2390 (Store {t3} op3:(OffPtr <tt3> [o3] p3) d2
2391 (Store {t4} op4:(OffPtr <tt4> [o4] p4) d3
2392 (Store {t5} op5:(OffPtr <tt5> [0] p5) d4 _)))))
2393 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4) && isSamePtr(p4, p5)
2394 && t2.Alignment() <= t1.Alignment()
2395 && t3.Alignment() <= t1.Alignment()
2396 && t4.Alignment() <= t1.Alignment()
2397 && t5.Alignment() <= t1.Alignment()
2398 && registerizable(b, t2)
2399 && registerizable(b, t3)
2400 && registerizable(b, t4)
2401 && registerizable(b, t5)
2402 && o4 == t5.Size()
2403 && o3-o4 == t4.Size()
2404 && o2-o3 == t3.Size()
2405 && n == t2.Size() + t3.Size() + t4.Size() + t5.Size()
2406 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2407 (Store {t3} (OffPtr <tt3> [o3] dst) d2
2408 (Store {t4} (OffPtr <tt4> [o4] dst) d3
2409 (Store {t5} (OffPtr <tt5> [0] dst) d4 mem))))
2410
2411// Same thing but with VarDef in the middle.
2412(Move {t1} [n] dst p1
2413 mem:(VarDef
2414 (Store {t2} op2:(OffPtr <tt2> [o2] p2) d1
2415 (Store {t3} op3:(OffPtr <tt3> [0] p3) d2 _))))
2416 && isSamePtr(p1, p2) && isSamePtr(p2, p3)
2417 && t2.Alignment() <= t1.Alignment()
2418 && t3.Alignment() <= t1.Alignment()
2419 && registerizable(b, t2)
2420 && registerizable(b, t3)
2421 && o2 == t3.Size()
2422 && n == t2.Size() + t3.Size()
2423 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2424 (Store {t3} (OffPtr <tt3> [0] dst) d2 mem))
2425(Move {t1} [n] dst p1
2426 mem:(VarDef
2427 (Store {t2} op2:(OffPtr <tt2> [o2] p2) d1
2428 (Store {t3} op3:(OffPtr <tt3> [o3] p3) d2
2429 (Store {t4} op4:(OffPtr <tt4> [0] p4) d3 _)))))
2430 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4)
2431 && t2.Alignment() <= t1.Alignment()
2432 && t3.Alignment() <= t1.Alignment()
2433 && t4.Alignment() <= t1.Alignment()
2434 && registerizable(b, t2)
2435 && registerizable(b, t3)
2436 && registerizable(b, t4)
2437 && o3 == t4.Size()
2438 && o2-o3 == t3.Size()
2439 && n == t2.Size() + t3.Size() + t4.Size()
2440 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2441 (Store {t3} (OffPtr <tt3> [o3] dst) d2
2442 (Store {t4} (OffPtr <tt4> [0] dst) d3 mem)))
2443(Move {t1} [n] dst p1
2444 mem:(VarDef
2445 (Store {t2} op2:(OffPtr <tt2> [o2] p2) d1
2446 (Store {t3} op3:(OffPtr <tt3> [o3] p3) d2
2447 (Store {t4} op4:(OffPtr <tt4> [o4] p4) d3
2448 (Store {t5} op5:(OffPtr <tt5> [0] p5) d4 _))))))
2449 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4) && isSamePtr(p4, p5)
2450 && t2.Alignment() <= t1.Alignment()
2451 && t3.Alignment() <= t1.Alignment()
2452 && t4.Alignment() <= t1.Alignment()
2453 && t5.Alignment() <= t1.Alignment()
2454 && registerizable(b, t2)
2455 && registerizable(b, t3)
2456 && registerizable(b, t4)
2457 && registerizable(b, t5)
2458 && o4 == t5.Size()
2459 && o3-o4 == t4.Size()
2460 && o2-o3 == t3.Size()
2461 && n == t2.Size() + t3.Size() + t4.Size() + t5.Size()
2462 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2463 (Store {t3} (OffPtr <tt3> [o3] dst) d2
2464 (Store {t4} (OffPtr <tt4> [o4] dst) d3
2465 (Store {t5} (OffPtr <tt5> [0] dst) d4 mem))))
2466
2467// Prefer to Zero and Store than to Move.
2468(Move {t1} [n] dst p1
2469 mem:(Store {t2} op2:(OffPtr <tt2> [o2] p2) d1
2470 (Zero {t3} [n] p3 _)))
2471 && isSamePtr(p1, p2) && isSamePtr(p2, p3)
2472 && t2.Alignment() <= t1.Alignment()
2473 && t3.Alignment() <= t1.Alignment()
2474 && registerizable(b, t2)
2475 && n >= o2 + t2.Size()
2476 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2477 (Zero {t1} [n] dst mem))
2478(Move {t1} [n] dst p1
2479 mem:(Store {t2} (OffPtr <tt2> [o2] p2) d1
2480 (Store {t3} (OffPtr <tt3> [o3] p3) d2
2481 (Zero {t4} [n] p4 _))))
2482 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4)
2483 && t2.Alignment() <= t1.Alignment()
2484 && t3.Alignment() <= t1.Alignment()
2485 && t4.Alignment() <= t1.Alignment()
2486 && registerizable(b, t2)
2487 && registerizable(b, t3)
2488 && n >= o2 + t2.Size()
2489 && n >= o3 + t3.Size()
2490 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2491 (Store {t3} (OffPtr <tt3> [o3] dst) d2
2492 (Zero {t1} [n] dst mem)))
2493(Move {t1} [n] dst p1
2494 mem:(Store {t2} (OffPtr <tt2> [o2] p2) d1
2495 (Store {t3} (OffPtr <tt3> [o3] p3) d2
2496 (Store {t4} (OffPtr <tt4> [o4] p4) d3
2497 (Zero {t5} [n] p5 _)))))
2498 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4) && isSamePtr(p4, p5)
2499 && t2.Alignment() <= t1.Alignment()
2500 && t3.Alignment() <= t1.Alignment()
2501 && t4.Alignment() <= t1.Alignment()
2502 && t5.Alignment() <= t1.Alignment()
2503 && registerizable(b, t2)
2504 && registerizable(b, t3)
2505 && registerizable(b, t4)
2506 && n >= o2 + t2.Size()
2507 && n >= o3 + t3.Size()
2508 && n >= o4 + t4.Size()
2509 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2510 (Store {t3} (OffPtr <tt3> [o3] dst) d2
2511 (Store {t4} (OffPtr <tt4> [o4] dst) d3
2512 (Zero {t1} [n] dst mem))))
2513(Move {t1} [n] dst p1
2514 mem:(Store {t2} (OffPtr <tt2> [o2] p2) d1
2515 (Store {t3} (OffPtr <tt3> [o3] p3) d2
2516 (Store {t4} (OffPtr <tt4> [o4] p4) d3
2517 (Store {t5} (OffPtr <tt5> [o5] p5) d4
2518 (Zero {t6} [n] p6 _))))))
2519 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4) && isSamePtr(p4, p5) && isSamePtr(p5, p6)
2520 && t2.Alignment() <= t1.Alignment()
2521 && t3.Alignment() <= t1.Alignment()
2522 && t4.Alignment() <= t1.Alignment()
2523 && t5.Alignment() <= t1.Alignment()
2524 && t6.Alignment() <= t1.Alignment()
2525 && registerizable(b, t2)
2526 && registerizable(b, t3)
2527 && registerizable(b, t4)
2528 && registerizable(b, t5)
2529 && n >= o2 + t2.Size()
2530 && n >= o3 + t3.Size()
2531 && n >= o4 + t4.Size()
2532 && n >= o5 + t5.Size()
2533 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2534 (Store {t3} (OffPtr <tt3> [o3] dst) d2
2535 (Store {t4} (OffPtr <tt4> [o4] dst) d3
2536 (Store {t5} (OffPtr <tt5> [o5] dst) d4
2537 (Zero {t1} [n] dst mem)))))
2538(Move {t1} [n] dst p1
2539 mem:(VarDef
2540 (Store {t2} op2:(OffPtr <tt2> [o2] p2) d1
2541 (Zero {t3} [n] p3 _))))
2542 && isSamePtr(p1, p2) && isSamePtr(p2, p3)
2543 && t2.Alignment() <= t1.Alignment()
2544 && t3.Alignment() <= t1.Alignment()
2545 && registerizable(b, t2)
2546 && n >= o2 + t2.Size()
2547 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2548 (Zero {t1} [n] dst mem))
2549(Move {t1} [n] dst p1
2550 mem:(VarDef
2551 (Store {t2} (OffPtr <tt2> [o2] p2) d1
2552 (Store {t3} (OffPtr <tt3> [o3] p3) d2
2553 (Zero {t4} [n] p4 _)))))
2554 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4)
2555 && t2.Alignment() <= t1.Alignment()
2556 && t3.Alignment() <= t1.Alignment()
2557 && t4.Alignment() <= t1.Alignment()
2558 && registerizable(b, t2)
2559 && registerizable(b, t3)
2560 && n >= o2 + t2.Size()
2561 && n >= o3 + t3.Size()
2562 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2563 (Store {t3} (OffPtr <tt3> [o3] dst) d2
2564 (Zero {t1} [n] dst mem)))
2565(Move {t1} [n] dst p1
2566 mem:(VarDef
2567 (Store {t2} (OffPtr <tt2> [o2] p2) d1
2568 (Store {t3} (OffPtr <tt3> [o3] p3) d2
2569 (Store {t4} (OffPtr <tt4> [o4] p4) d3
2570 (Zero {t5} [n] p5 _))))))
2571 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4) && isSamePtr(p4, p5)
2572 && t2.Alignment() <= t1.Alignment()
2573 && t3.Alignment() <= t1.Alignment()
2574 && t4.Alignment() <= t1.Alignment()
2575 && t5.Alignment() <= t1.Alignment()
2576 && registerizable(b, t2)
2577 && registerizable(b, t3)
2578 && registerizable(b, t4)
2579 && n >= o2 + t2.Size()
2580 && n >= o3 + t3.Size()
2581 && n >= o4 + t4.Size()
2582 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2583 (Store {t3} (OffPtr <tt3> [o3] dst) d2
2584 (Store {t4} (OffPtr <tt4> [o4] dst) d3
2585 (Zero {t1} [n] dst mem))))
2586(Move {t1} [n] dst p1
2587 mem:(VarDef
2588 (Store {t2} (OffPtr <tt2> [o2] p2) d1
2589 (Store {t3} (OffPtr <tt3> [o3] p3) d2
2590 (Store {t4} (OffPtr <tt4> [o4] p4) d3
2591 (Store {t5} (OffPtr <tt5> [o5] p5) d4
2592 (Zero {t6} [n] p6 _)))))))
2593 && isSamePtr(p1, p2) && isSamePtr(p2, p3) && isSamePtr(p3, p4) && isSamePtr(p4, p5) && isSamePtr(p5, p6)
2594 && t2.Alignment() <= t1.Alignment()
2595 && t3.Alignment() <= t1.Alignment()
2596 && t4.Alignment() <= t1.Alignment()
2597 && t5.Alignment() <= t1.Alignment()
2598 && t6.Alignment() <= t1.Alignment()
2599 && registerizable(b, t2)
2600 && registerizable(b, t3)
2601 && registerizable(b, t4)
2602 && registerizable(b, t5)
2603 && n >= o2 + t2.Size()
2604 && n >= o3 + t3.Size()
2605 && n >= o4 + t4.Size()
2606 && n >= o5 + t5.Size()
2607 => (Store {t2} (OffPtr <tt2> [o2] dst) d1
2608 (Store {t3} (OffPtr <tt3> [o3] dst) d2
2609 (Store {t4} (OffPtr <tt4> [o4] dst) d3
2610 (Store {t5} (OffPtr <tt5> [o5] dst) d4
2611 (Zero {t1} [n] dst mem)))))
2612
2613(SelectN [0] call:(StaticLECall {sym} a x)) && needRaceCleanup(sym, call) && clobber(call) => x
2614(SelectN [0] call:(StaticLECall {sym} x)) && needRaceCleanup(sym, call) && clobber(call) => x
2615
2616// When rewriting append to growslice, we use as the new length the result of
2617// growslice so that we don't have to spill/restore the new length around the growslice call.
2618// The exception here is that if the new length is a constant, avoiding spilling it
2619// is pointless and its constantness is sometimes useful for subsequent optimizations.
2620// See issue 56440.
2621// Note there are 2 rules here, one for the pre-decomposed []T result and one for
2622// the post-decomposed (*T,int,int) result. (The latter is generated after call expansion.)
2623(SliceLen (SelectN [0] (StaticLECall {sym} _ newLen:(Const(64|32)) _ _ _ _))) && isSameCall(sym, "runtime.growslice") => newLen
2624(SelectN [1] (StaticCall {sym} _ newLen:(Const(64|32)) _ _ _ _)) && v.Type.IsInteger() && isSameCall(sym, "runtime.growslice") => newLen
2625
2626// Collapse moving A -> B -> C into just A -> C.
2627// Later passes (deadstore, elim unread auto) will remove the A -> B move, if possible.
2628// This happens most commonly when B is an autotmp inserted earlier
2629// during compilation to ensure correctness.
2630// Take care that overlapping moves are preserved.
2631// Restrict this optimization to the stack, to avoid duplicating loads from the heap;
2632// see CL 145208 for discussion.
2633(Move {t1} [s] dst tmp1 midmem:(Move {t2} [s] tmp2 src _))
2634 && t1.Compare(t2) == types.CMPeq
2635 && isSamePtr(tmp1, tmp2)
2636 && isStackPtr(src) && !isVolatile(src)
2637 && disjoint(src, s, tmp2, s)
2638 && (disjoint(src, s, dst, s) || isInlinableMemmove(dst, src, s, config))
2639 => (Move {t1} [s] dst src midmem)
2640
2641// Same, but for large types that require VarDefs.
2642(Move {t1} [s] dst tmp1 midmem:(VarDef (Move {t2} [s] tmp2 src _)))
2643 && t1.Compare(t2) == types.CMPeq
2644 && isSamePtr(tmp1, tmp2)
2645 && isStackPtr(src) && !isVolatile(src)
2646 && disjoint(src, s, tmp2, s)
2647 && (disjoint(src, s, dst, s) || isInlinableMemmove(dst, src, s, config))
2648 => (Move {t1} [s] dst src midmem)
2649
2650// Don't zero the same bits twice.
2651(Zero {t} [s] dst1 zero:(Zero {t} [s] dst2 _)) && isSamePtr(dst1, dst2) => zero
2652(Zero {t} [s] dst1 vardef:(VarDef (Zero {t} [s] dst2 _))) && isSamePtr(dst1, dst2) => vardef
2653
2654// Elide self-moves. This only happens rarely (e.g test/fixedbugs/bug277.go).
2655// However, this rule is needed to prevent the previous rule from looping forever in such cases.
2656(Move dst src mem) && isSamePtr(dst, src) => mem
2657
2658// Constant rotate detection.
2659((Add64|Or64|Xor64) (Lsh64x64 x z:(Const64 <t> [c])) (Rsh64Ux64 x (Const64 [d]))) && c < 64 && d == 64-c && canRotate(config, 64) => (RotateLeft64 x z)
2660((Add32|Or32|Xor32) (Lsh32x64 x z:(Const64 <t> [c])) (Rsh32Ux64 x (Const64 [d]))) && c < 32 && d == 32-c && canRotate(config, 32) => (RotateLeft32 x z)
2661((Add16|Or16|Xor16) (Lsh16x64 x z:(Const64 <t> [c])) (Rsh16Ux64 x (Const64 [d]))) && c < 16 && d == 16-c && canRotate(config, 16) => (RotateLeft16 x z)
2662((Add8|Or8|Xor8) (Lsh8x64 x z:(Const64 <t> [c])) (Rsh8Ux64 x (Const64 [d]))) && c < 8 && d == 8-c && canRotate(config, 8) => (RotateLeft8 x z)
2663
2664// Non-constant rotate detection.
2665// We use shiftIsBounded to make sure that neither of the shifts are >64.
2666// Note: these rules are subtle when the shift amounts are 0/64, as Go shifts
2667// are different from most native shifts. But it works out.
2668((Add64|Or64|Xor64) left:(Lsh64x64 x y) right:(Rsh64Ux64 x (Sub64 (Const64 [64]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 64) => (RotateLeft64 x y)
2669((Add64|Or64|Xor64) left:(Lsh64x32 x y) right:(Rsh64Ux32 x (Sub32 (Const32 [64]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 64) => (RotateLeft64 x y)
2670((Add64|Or64|Xor64) left:(Lsh64x16 x y) right:(Rsh64Ux16 x (Sub16 (Const16 [64]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 64) => (RotateLeft64 x y)
2671((Add64|Or64|Xor64) left:(Lsh64x8 x y) right:(Rsh64Ux8 x (Sub8 (Const8 [64]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 64) => (RotateLeft64 x y)
2672
2673((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)
2674((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)
2675((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)
2676((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)
2677
2678((Add32|Or32|Xor32) left:(Lsh32x64 x y) right:(Rsh32Ux64 x (Sub64 (Const64 [32]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 32) => (RotateLeft32 x y)
2679((Add32|Or32|Xor32) left:(Lsh32x32 x y) right:(Rsh32Ux32 x (Sub32 (Const32 [32]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 32) => (RotateLeft32 x y)
2680((Add32|Or32|Xor32) left:(Lsh32x16 x y) right:(Rsh32Ux16 x (Sub16 (Const16 [32]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 32) => (RotateLeft32 x y)
2681((Add32|Or32|Xor32) left:(Lsh32x8 x y) right:(Rsh32Ux8 x (Sub8 (Const8 [32]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 32) => (RotateLeft32 x y)
2682
2683((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)
2684((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)
2685((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)
2686((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)
2687
2688((Add16|Or16|Xor16) left:(Lsh16x64 x y) right:(Rsh16Ux64 x (Sub64 (Const64 [16]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 16) => (RotateLeft16 x y)
2689((Add16|Or16|Xor16) left:(Lsh16x32 x y) right:(Rsh16Ux32 x (Sub32 (Const32 [16]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 16) => (RotateLeft16 x y)
2690((Add16|Or16|Xor16) left:(Lsh16x16 x y) right:(Rsh16Ux16 x (Sub16 (Const16 [16]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 16) => (RotateLeft16 x y)
2691((Add16|Or16|Xor16) left:(Lsh16x8 x y) right:(Rsh16Ux8 x (Sub8 (Const8 [16]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 16) => (RotateLeft16 x y)
2692
2693((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)
2694((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)
2695((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)
2696((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)
2697
2698((Add8|Or8|Xor8) left:(Lsh8x64 x y) right:(Rsh8Ux64 x (Sub64 (Const64 [8]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 8) => (RotateLeft8 x y)
2699((Add8|Or8|Xor8) left:(Lsh8x32 x y) right:(Rsh8Ux32 x (Sub32 (Const32 [8]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 8) => (RotateLeft8 x y)
2700((Add8|Or8|Xor8) left:(Lsh8x16 x y) right:(Rsh8Ux16 x (Sub16 (Const16 [8]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 8) => (RotateLeft8 x y)
2701((Add8|Or8|Xor8) left:(Lsh8x8 x y) right:(Rsh8Ux8 x (Sub8 (Const8 [8]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 8) => (RotateLeft8 x y)
2702
2703((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)
2704((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)
2705((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)
2706((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)
2707
2708// Rotating by y&c, with c a mask that doesn't change the bottom bits, is the same as rotating by y.
2709(RotateLeft64 x (And(64|32|16|8) y (Const(64|32|16|8) [c]))) && c&63 == 63 => (RotateLeft64 x y)
2710(RotateLeft32 x (And(64|32|16|8) y (Const(64|32|16|8) [c]))) && c&31 == 31 => (RotateLeft32 x y)
2711(RotateLeft16 x (And(64|32|16|8) y (Const(64|32|16|8) [c]))) && c&15 == 15 => (RotateLeft16 x y)
2712(RotateLeft8 x (And(64|32|16|8) y (Const(64|32|16|8) [c]))) && c&7 == 7 => (RotateLeft8 x y)
2713
2714// Rotating by -(y&c), with c a mask that doesn't change the bottom bits, is the same as rotating by -y.
2715(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))
2716(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))
2717(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))
2718(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))
2719
2720// Rotating by y+c, with c a multiple of the value width, is the same as rotating by y.
2721(RotateLeft64 x (Add(64|32|16|8) y (Const(64|32|16|8) [c]))) && c&63 == 0 => (RotateLeft64 x y)
2722(RotateLeft32 x (Add(64|32|16|8) y (Const(64|32|16|8) [c]))) && c&31 == 0 => (RotateLeft32 x y)
2723(RotateLeft16 x (Add(64|32|16|8) y (Const(64|32|16|8) [c]))) && c&15 == 0 => (RotateLeft16 x y)
2724(RotateLeft8 x (Add(64|32|16|8) y (Const(64|32|16|8) [c]))) && c&7 == 0 => (RotateLeft8 x y)
2725
2726// Rotating by c-y, with c a multiple of the value width, is the same as rotating by -y.
2727(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))
2728(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))
2729(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))
2730(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))
2731
2732// Ensure we don't do Const64 rotates in a 32-bit system.
2733(RotateLeft64 x (Const64 <t> [c])) && config.PtrSize == 4 => (RotateLeft64 x (Const32 <t> [int32(c)]))
2734(RotateLeft32 x (Const64 <t> [c])) && config.PtrSize == 4 => (RotateLeft32 x (Const32 <t> [int32(c)]))
2735(RotateLeft16 x (Const64 <t> [c])) && config.PtrSize == 4 => (RotateLeft16 x (Const32 <t> [int32(c)]))
2736(RotateLeft8 x (Const64 <t> [c])) && config.PtrSize == 4 => (RotateLeft8 x (Const32 <t> [int32(c)]))
2737
2738// Rotating by c, then by d, is the same as rotating by c+d.
2739// We're trading a rotate for an add, which seems generally a good choice. It is especially good when c and d are constants.
2740// This rule is a bit tricky as c and d might be different widths. We handle only cases where they are the same width.
2741(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))
2742(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))
2743(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))
2744(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))
2745
2746// Loading constant values from dictionaries and itabs.
2747(Load <t> (OffPtr [off] (Addr {s} sb) ) _) && t.IsUintptr() && isFixedSym(s, off) => (Addr {fixedSym(b.Func, s, off)} sb)
2748(Load <t> (OffPtr [off] (Convert (Addr {s} sb) _) ) _) && t.IsUintptr() && isFixedSym(s, off) => (Addr {fixedSym(b.Func, s, off)} sb)
2749(Load <t> (OffPtr [off] (ITab (IMake (Addr {s} sb) _))) _) && t.IsUintptr() && isFixedSym(s, off) => (Addr {fixedSym(b.Func, s, off)} sb)
2750(Load <t> (OffPtr [off] (ITab (IMake (Convert (Addr {s} sb) _) _))) _) && t.IsUintptr() && isFixedSym(s, off) => (Addr {fixedSym(b.Func, s, off)} sb)
2751
2752// Loading constant values from runtime._type.hash.
2753(Load <t> (OffPtr [off] (Addr {sym} _) ) _) && t.IsInteger() && t.Size() == 4 && isFixed32(config, sym, off) => (Const32 [fixed32(config, sym, off)])
2754(Load <t> (OffPtr [off] (Convert (Addr {sym} _) _) ) _) && t.IsInteger() && t.Size() == 4 && isFixed32(config, sym, off) => (Const32 [fixed32(config, sym, off)])
2755(Load <t> (OffPtr [off] (ITab (IMake (Addr {sym} _) _))) _) && t.IsInteger() && t.Size() == 4 && isFixed32(config, sym, off) => (Const32 [fixed32(config, sym, off)])
2756(Load <t> (OffPtr [off] (ITab (IMake (Convert (Addr {sym} _) _) _))) _) && t.IsInteger() && t.Size() == 4 && isFixed32(config, sym, off) => (Const32 [fixed32(config, sym, off)])
View as plain text