1 // Copyright 2016 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 (Add(Ptr|64|32|16|8) ...) => (ADD ...)
6 (Add(32|64)F ...) => (FADD(S|D) ...)
7
8 (Sub(Ptr|64|32|16|8) ...) => (SUB ...)
9 (Sub(32|64)F ...) => (FSUB(S|D) ...)
10
11 (Mul64 ...) => (MUL ...)
12 (Mul(32|16|8) ...) => (MULW ...)
13 (Mul(32|64)F ...) => (FMUL(S|D) ...)
14
15 (Hmul64 ...) => (MULH ...)
16 (Hmul64u ...) => (UMULH ...)
17 (Hmul32 x y) => (SRAconst (MULL <typ.Int64> x y) [32])
18 (Hmul32u x y) => (SRAconst (UMULL <typ.UInt64> x y) [32])
19 (Select0 (Mul64uhilo x y)) => (UMULH x y)
20 (Select1 (Mul64uhilo x y)) => (MUL x y)
21
22 (Div64 [false] x y) => (DIV x y)
23 (Div32 [false] x y) => (DIVW x y)
24 (Div16 [false] x y) => (DIVW (SignExt16to32 x) (SignExt16to32 y))
25 (Div16u x y) => (UDIVW (ZeroExt16to32 x) (ZeroExt16to32 y))
26 (Div8 x y) => (DIVW (SignExt8to32 x) (SignExt8to32 y))
27 (Div8u x y) => (UDIVW (ZeroExt8to32 x) (ZeroExt8to32 y))
28 (Div64u ...) => (UDIV ...)
29 (Div32u ...) => (UDIVW ...)
30 (Div32F ...) => (FDIVS ...)
31 (Div64F ...) => (FDIVD ...)
32
33 (Mod64 x y) => (MOD x y)
34 (Mod32 x y) => (MODW x y)
35 (Mod64u ...) => (UMOD ...)
36 (Mod32u ...) => (UMODW ...)
37 (Mod(16|8) x y) => (MODW (SignExt(16|8)to32 x) (SignExt(16|8)to32 y))
38 (Mod(16|8)u x y) => (UMODW (ZeroExt(16|8)to32 x) (ZeroExt(16|8)to32 y))
39
40 // (x + y) / 2 with x>=y => (x - y) / 2 + y
41 (Avg64u <t> x y) => (ADD (SRLconst <t> (SUB <t> x y) [1]) y)
42
43 (And(64|32|16|8) ...) => (AND ...)
44 (Or(64|32|16|8) ...) => (OR ...)
45 (Xor(64|32|16|8) ...) => (XOR ...)
46
47 // unary ops
48 (Neg(64|32|16|8) ...) => (NEG ...)
49 (Neg(32|64)F ...) => (FNEG(S|D) ...)
50 (Com(64|32|16|8) ...) => (MVN ...)
51
52 // math package intrinsics
53 (Abs ...) => (FABSD ...)
54 (Sqrt ...) => (FSQRTD ...)
55 (Ceil ...) => (FRINTPD ...)
56 (Floor ...) => (FRINTMD ...)
57 (Round ...) => (FRINTAD ...)
58 (RoundToEven ...) => (FRINTND ...)
59 (Trunc ...) => (FRINTZD ...)
60 (FMA x y z) => (FMADDD z x y)
61
62 (Sqrt32 ...) => (FSQRTS ...)
63
64 (Min(64|32)F ...) => (FMIN(D|S) ...)
65 (Max(64|32)F ...) => (FMAX(D|S) ...)
66
67 // lowering rotates
68 // we do rotate detection in generic rules, if the following rules need to be changed, check generic rules first.
69 (RotateLeft8 <t> x (MOVDconst [c])) => (Or8 (Lsh8x64 <t> x (MOVDconst [c&7])) (Rsh8Ux64 <t> x (MOVDconst [-c&7])))
70 (RotateLeft8 <t> x y) => (OR <t> (SLL <t> x (ANDconst <typ.Int64> [7] y)) (SRL <t> (ZeroExt8to64 x) (ANDconst <typ.Int64> [7] (NEG <typ.Int64> y))))
71 (RotateLeft16 <t> x (MOVDconst [c])) => (Or16 (Lsh16x64 <t> x (MOVDconst [c&15])) (Rsh16Ux64 <t> x (MOVDconst [-c&15])))
72 (RotateLeft16 <t> x y) => (RORW <t> (ORshiftLL <typ.UInt32> (ZeroExt16to32 x) (ZeroExt16to32 x) [16]) (NEG <typ.Int64> y))
73 (RotateLeft32 x y) => (RORW x (NEG <y.Type> y))
74 (RotateLeft64 x y) => (ROR x (NEG <y.Type> y))
75
76 (Ctz(64|32|16|8)NonZero ...) => (Ctz(64|32|32|32) ...)
77
78 (Ctz64 <t> x) => (CLZ (RBIT <t> x))
79 (Ctz32 <t> x) => (CLZW (RBITW <t> x))
80 (Ctz16 <t> x) => (CLZW <t> (RBITW <typ.UInt32> (ORconst <typ.UInt32> [0x10000] x)))
81 (Ctz8 <t> x) => (CLZW <t> (RBITW <typ.UInt32> (ORconst <typ.UInt32> [0x100] x)))
82
83 (PopCount64 <t> x) => (FMOVDfpgp <t> (VUADDLV <typ.Float64> (VCNT <typ.Float64> (FMOVDgpfp <typ.Float64> x))))
84 (PopCount32 <t> x) => (FMOVDfpgp <t> (VUADDLV <typ.Float64> (VCNT <typ.Float64> (FMOVDgpfp <typ.Float64> (ZeroExt32to64 x)))))
85 (PopCount16 <t> x) => (FMOVDfpgp <t> (VUADDLV <typ.Float64> (VCNT <typ.Float64> (FMOVDgpfp <typ.Float64> (ZeroExt16to64 x)))))
86
87 // Load args directly into the register class where it will be used.
88 (FMOVDgpfp <t> (Arg [off] {sym})) => @b.Func.Entry (Arg <t> [off] {sym})
89 (FMOVDfpgp <t> (Arg [off] {sym})) => @b.Func.Entry (Arg <t> [off] {sym})
90
91 // Similarly for stores, if we see a store after FPR <=> GPR move, then redirect store to use the other register set.
92 (MOVDstore [off] {sym} ptr (FMOVDfpgp val) mem) => (FMOVDstore [off] {sym} ptr val mem)
93 (FMOVDstore [off] {sym} ptr (FMOVDgpfp val) mem) => (MOVDstore [off] {sym} ptr val mem)
94 (MOVWstore [off] {sym} ptr (FMOVSfpgp val) mem) => (FMOVSstore [off] {sym} ptr val mem)
95 (FMOVSstore [off] {sym} ptr (FMOVSgpfp val) mem) => (MOVWstore [off] {sym} ptr val mem)
96
97 // float <=> int register moves, with no conversion.
98 // These come up when compiling math.{Float64bits, Float64frombits, Float32bits, Float32frombits}.
99 (MOVDload [off] {sym} ptr (FMOVDstore [off] {sym} ptr val _)) => (FMOVDfpgp val)
100 (FMOVDload [off] {sym} ptr (MOVDstore [off] {sym} ptr val _)) => (FMOVDgpfp val)
101 (MOVWUload [off] {sym} ptr (FMOVSstore [off] {sym} ptr val _)) => (FMOVSfpgp val)
102 (FMOVSload [off] {sym} ptr (MOVWstore [off] {sym} ptr val _)) => (FMOVSgpfp val)
103
104 (BitLen64 x) => (SUB (MOVDconst [64]) (CLZ <typ.Int> x))
105 (BitLen32 x) => (SUB (MOVDconst [32]) (CLZW <typ.Int> x))
106
107 (Bswap64 ...) => (REV ...)
108 (Bswap32 ...) => (REVW ...)
109 (Bswap16 ...) => (REV16W ...)
110
111 (BitRev64 ...) => (RBIT ...)
112 (BitRev32 ...) => (RBITW ...)
113 (BitRev16 x) => (SRLconst [48] (RBIT <typ.UInt64> x))
114 (BitRev8 x) => (SRLconst [56] (RBIT <typ.UInt64> x))
115
116 // In fact, UMOD will be translated into UREM instruction, and UREM is originally translated into
117 // UDIV and MSUB instructions. But if there is already an identical UDIV instruction just before or
118 // after UREM (case like quo, rem := z/y, z%y), then the second UDIV instruction becomes redundant.
119 // The purpose of this rule is to have this extra UDIV instruction removed in CSE pass.
120 (UMOD <typ.UInt64> x y) => (MSUB <typ.UInt64> x y (UDIV <typ.UInt64> x y))
121 (UMODW <typ.UInt32> x y) => (MSUBW <typ.UInt32> x y (UDIVW <typ.UInt32> x y))
122
123 // 64-bit addition with carry.
124 (Select0 (Add64carry x y c)) => (Select0 <typ.UInt64> (ADCSflags x y (Select1 <types.TypeFlags> (ADDSconstflags [-1] c))))
125 (Select1 (Add64carry x y c)) => (ADCzerocarry <typ.UInt64> (Select1 <types.TypeFlags> (ADCSflags x y (Select1 <types.TypeFlags> (ADDSconstflags [-1] c)))))
126
127 // 64-bit subtraction with borrowing.
128 (Select0 (Sub64borrow x y bo)) => (Select0 <typ.UInt64> (SBCSflags x y (Select1 <types.TypeFlags> (NEGSflags bo))))
129 (Select1 (Sub64borrow x y bo)) => (NEG <typ.UInt64> (NGCzerocarry <typ.UInt64> (Select1 <types.TypeFlags> (SBCSflags x y (Select1 <types.TypeFlags> (NEGSflags bo))))))
130
131 // boolean ops -- booleans are represented with 0=false, 1=true
132 (AndB ...) => (AND ...)
133 (OrB ...) => (OR ...)
134 (EqB x y) => (XOR (MOVDconst [1]) (XOR <typ.Bool> x y))
135 (NeqB ...) => (XOR ...)
136 (Not x) => (XOR (MOVDconst [1]) x)
137
138 // shifts
139 // hardware instruction uses only the low 6 bits of the shift
140 // we compare to 64 to ensure Go semantics for large shifts
141 // Rules about rotates with non-const shift are based on the following rules,
142 // if the following rules change, please also modify the rules based on them.
143
144 // check shiftIsBounded first, if shift value is proved to be valid then we
145 // can do the shift directly.
146 // left shift
147 (Lsh(64|32|16|8)x64 <t> x y) && shiftIsBounded(v) => (SLL <t> x y)
148 (Lsh(64|32|16|8)x32 <t> x y) && shiftIsBounded(v) => (SLL <t> x y)
149 (Lsh(64|32|16|8)x16 <t> x y) && shiftIsBounded(v) => (SLL <t> x y)
150 (Lsh(64|32|16|8)x8 <t> x y) && shiftIsBounded(v) => (SLL <t> x y)
151
152 // signed right shift
153 (Rsh64x(64|32|16|8) <t> x y) && shiftIsBounded(v) => (SRA <t> x y)
154 (Rsh32x(64|32|16|8) <t> x y) && shiftIsBounded(v) => (SRA <t> (SignExt32to64 x) y)
155 (Rsh16x(64|32|16|8) <t> x y) && shiftIsBounded(v) => (SRA <t> (SignExt16to64 x) y)
156 (Rsh8x(64|32|16|8) <t> x y) && shiftIsBounded(v) => (SRA <t> (SignExt8to64 x) y)
157
158 // unsigned right shift
159 (Rsh64Ux(64|32|16|8) <t> x y) && shiftIsBounded(v) => (SRL <t> x y)
160 (Rsh32Ux(64|32|16|8) <t> x y) && shiftIsBounded(v) => (SRL <t> (ZeroExt32to64 x) y)
161 (Rsh16Ux(64|32|16|8) <t> x y) && shiftIsBounded(v) => (SRL <t> (ZeroExt16to64 x) y)
162 (Rsh8Ux(64|32|16|8) <t> x y) && shiftIsBounded(v) => (SRL <t> (ZeroExt8to64 x) y)
163
164 // shift value may be out of range, use CMP + CSEL instead
165 (Lsh64x64 <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SLL <t> x y) (Const64 <t> [0]) (CMPconst [64] y))
166 (Lsh64x(32|16|8) <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SLL <t> x y) (Const64 <t> [0]) (CMPconst [64] ((ZeroExt32to64|ZeroExt16to64|ZeroExt8to64) y)))
167
168 (Lsh32x64 <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SLL <t> x y) (Const64 <t> [0]) (CMPconst [64] y))
169 (Lsh32x(32|16|8) <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SLL <t> x y) (Const64 <t> [0]) (CMPconst [64] ((ZeroExt32to64|ZeroExt16to64|ZeroExt8to64) y)))
170
171 (Lsh16x64 <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SLL <t> x y) (Const64 <t> [0]) (CMPconst [64] y))
172 (Lsh16x(32|16|8) <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SLL <t> x y) (Const64 <t> [0]) (CMPconst [64] ((ZeroExt32to64|ZeroExt16to64|ZeroExt8to64) y)))
173
174 (Lsh8x64 <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SLL <t> x y) (Const64 <t> [0]) (CMPconst [64] y))
175 (Lsh8x(32|16|8) <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SLL <t> x y) (Const64 <t> [0]) (CMPconst [64] ((ZeroExt32to64|ZeroExt16to64|ZeroExt8to64) y)))
176
177 (Rsh64Ux64 <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SRL <t> x y) (Const64 <t> [0]) (CMPconst [64] y))
178 (Rsh64Ux(32|16|8) <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SRL <t> x y) (Const64 <t> [0]) (CMPconst [64] ((ZeroExt32to64|ZeroExt16to64|ZeroExt8to64) y)))
179
180 (Rsh32Ux64 <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SRL <t> (ZeroExt32to64 x) y) (Const64 <t> [0]) (CMPconst [64] y))
181 (Rsh32Ux(32|16|8) <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SRL <t> (ZeroExt32to64 x) y) (Const64 <t> [0]) (CMPconst [64] ((ZeroExt32to64|ZeroExt16to64|ZeroExt8to64) y)))
182
183 (Rsh16Ux64 <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SRL <t> (ZeroExt16to64 x) y) (Const64 <t> [0]) (CMPconst [64] y))
184 (Rsh16Ux(32|16|8) <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SRL <t> (ZeroExt16to64 x) y) (Const64 <t> [0]) (CMPconst [64] ((ZeroExt32to64|ZeroExt16to64|ZeroExt8to64) y)))
185
186 (Rsh8Ux64 <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SRL <t> (ZeroExt8to64 x) y) (Const64 <t> [0]) (CMPconst [64] y))
187 (Rsh8Ux(32|16|8) <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SRL <t> (ZeroExt8to64 x) y) (Const64 <t> [0]) (CMPconst [64] ((ZeroExt32to64|ZeroExt16to64|ZeroExt8to64) y)))
188
189 (Rsh64x64 x y) && !shiftIsBounded(v) => (SRA x (CSEL [OpARM64LessThanU] <y.Type> y (Const64 <y.Type> [63]) (CMPconst [64] y)))
190 (Rsh64x(32|16|8) x y) && !shiftIsBounded(v) => (SRA x (CSEL [OpARM64LessThanU] <y.Type> y (Const64 <y.Type> [63]) (CMPconst [64] ((ZeroExt32to64|ZeroExt16to64|ZeroExt8to64) y))))
191
192 (Rsh32x64 x y) && !shiftIsBounded(v) => (SRA (SignExt32to64 x) (CSEL [OpARM64LessThanU] <y.Type> y (Const64 <y.Type> [63]) (CMPconst [64] y)))
193 (Rsh32x(32|16|8) x y) && !shiftIsBounded(v) => (SRA (SignExt32to64 x) (CSEL [OpARM64LessThanU] <y.Type> y (Const64 <y.Type> [63]) (CMPconst [64] ((ZeroExt32to64|ZeroExt16to64|ZeroExt8to64) y))))
194
195 (Rsh16x64 x y) && !shiftIsBounded(v) => (SRA (SignExt16to64 x) (CSEL [OpARM64LessThanU] <y.Type> y (Const64 <y.Type> [63]) (CMPconst [64] y)))
196 (Rsh16x(32|16|8) x y) && !shiftIsBounded(v) => (SRA (SignExt16to64 x) (CSEL [OpARM64LessThanU] <y.Type> y (Const64 <y.Type> [63]) (CMPconst [64] ((ZeroExt32to64|ZeroExt16to64|ZeroExt8to64) y))))
197
198 (Rsh8x64 x y) && !shiftIsBounded(v) => (SRA (SignExt8to64 x) (CSEL [OpARM64LessThanU] <y.Type> y (Const64 <y.Type> [63]) (CMPconst [64] y)))
199 (Rsh8x(32|16|8) x y) && !shiftIsBounded(v) => (SRA (SignExt8to64 x) (CSEL [OpARM64LessThanU] <y.Type> y (Const64 <y.Type> [63]) (CMPconst [64] ((ZeroExt32to64|ZeroExt16to64|ZeroExt8to64) y))))
200
201 // constants
202 (Const(64|32|16|8) [val]) => (MOVDconst [int64(val)])
203 (Const(32|64)F [val]) => (FMOV(S|D)const [float64(val)])
204 (ConstNil) => (MOVDconst [0])
205 (ConstBool [t]) => (MOVDconst [b2i(t)])
206
207 (Slicemask <t> x) => (SRAconst (NEG <t> x) [63])
208
209 // truncations
210 // Because we ignore high parts of registers, truncates are just copies.
211 (Trunc16to8 ...) => (Copy ...)
212 (Trunc32to8 ...) => (Copy ...)
213 (Trunc32to16 ...) => (Copy ...)
214 (Trunc64to8 ...) => (Copy ...)
215 (Trunc64to16 ...) => (Copy ...)
216 (Trunc64to32 ...) => (Copy ...)
217
218 // Zero-/Sign-extensions
219 (ZeroExt8to16 ...) => (MOVBUreg ...)
220 (ZeroExt8to32 ...) => (MOVBUreg ...)
221 (ZeroExt16to32 ...) => (MOVHUreg ...)
222 (ZeroExt8to64 ...) => (MOVBUreg ...)
223 (ZeroExt16to64 ...) => (MOVHUreg ...)
224 (ZeroExt32to64 ...) => (MOVWUreg ...)
225
226 (SignExt8to16 ...) => (MOVBreg ...)
227 (SignExt8to32 ...) => (MOVBreg ...)
228 (SignExt16to32 ...) => (MOVHreg ...)
229 (SignExt8to64 ...) => (MOVBreg ...)
230 (SignExt16to64 ...) => (MOVHreg ...)
231 (SignExt32to64 ...) => (MOVWreg ...)
232
233 // float <=> int conversion
234 (Cvt32to32F ...) => (SCVTFWS ...)
235 (Cvt32to64F ...) => (SCVTFWD ...)
236 (Cvt64to32F ...) => (SCVTFS ...)
237 (Cvt64to64F ...) => (SCVTFD ...)
238 (Cvt32Uto32F ...) => (UCVTFWS ...)
239 (Cvt32Uto64F ...) => (UCVTFWD ...)
240 (Cvt64Uto32F ...) => (UCVTFS ...)
241 (Cvt64Uto64F ...) => (UCVTFD ...)
242 (Cvt32Fto32 ...) => (FCVTZSSW ...)
243 (Cvt64Fto32 ...) => (FCVTZSDW ...)
244 (Cvt32Fto64 ...) => (FCVTZSS ...)
245 (Cvt64Fto64 ...) => (FCVTZSD ...)
246 (Cvt32Fto32U ...) => (FCVTZUSW ...)
247 (Cvt64Fto32U ...) => (FCVTZUDW ...)
248 (Cvt32Fto64U ...) => (FCVTZUS ...)
249 (Cvt64Fto64U ...) => (FCVTZUD ...)
250 (Cvt32Fto64F ...) => (FCVTSD ...)
251 (Cvt64Fto32F ...) => (FCVTDS ...)
252
253 (CvtBoolToUint8 ...) => (Copy ...)
254
255 (Round32F ...) => (LoweredRound32F ...)
256 (Round64F ...) => (LoweredRound64F ...)
257
258 // comparisons
259 (Eq8 x y) => (Equal (CMPW (ZeroExt8to32 x) (ZeroExt8to32 y)))
260 (Eq16 x y) => (Equal (CMPW (ZeroExt16to32 x) (ZeroExt16to32 y)))
261 (Eq32 x y) => (Equal (CMPW x y))
262 (Eq64 x y) => (Equal (CMP x y))
263 (EqPtr x y) => (Equal (CMP x y))
264 (Eq32F x y) => (Equal (FCMPS x y))
265 (Eq64F x y) => (Equal (FCMPD x y))
266
267 (Neq8 x y) => (NotEqual (CMPW (ZeroExt8to32 x) (ZeroExt8to32 y)))
268 (Neq16 x y) => (NotEqual (CMPW (ZeroExt16to32 x) (ZeroExt16to32 y)))
269 (Neq32 x y) => (NotEqual (CMPW x y))
270 (Neq64 x y) => (NotEqual (CMP x y))
271 (NeqPtr x y) => (NotEqual (CMP x y))
272 (Neq(32|64)F x y) => (NotEqual (FCMP(S|D) x y))
273
274 (Less(8|16) x y) => (LessThan (CMPW (SignExt(8|16)to32 x) (SignExt(8|16)to32 y)))
275 (Less32 x y) => (LessThan (CMPW x y))
276 (Less64 x y) => (LessThan (CMP x y))
277
278 // Set condition flags for floating-point comparisons "x < y"
279 // and "x <= y". Because if either or both of the operands are
280 // NaNs, all three of (x < y), (x == y) and (x > y) are false,
281 // and ARM Manual says FCMP instruction sets PSTATE.<N,Z,C,V>
282 // of this case to (0, 0, 1, 1).
283 (Less32F x y) => (LessThanF (FCMPS x y))
284 (Less64F x y) => (LessThanF (FCMPD x y))
285
286 // For an unsigned integer x, the following rules are useful when combining branch
287 // 0 < x => x != 0
288 // x <= 0 => x == 0
289 // x < 1 => x == 0
290 // 1 <= x => x != 0
291 (Less(8U|16U|32U|64U) zero:(MOVDconst [0]) x) => (Neq(8|16|32|64) zero x)
292 (Leq(8U|16U|32U|64U) x zero:(MOVDconst [0])) => (Eq(8|16|32|64) x zero)
293 (Less(8U|16U|32U|64U) x (MOVDconst [1])) => (Eq(8|16|32|64) x (MOVDconst [0]))
294 (Leq(8U|16U|32U|64U) (MOVDconst [1]) x) => (Neq(8|16|32|64) (MOVDconst [0]) x)
295
296 (Less8U x y) => (LessThanU (CMPW (ZeroExt8to32 x) (ZeroExt8to32 y)))
297 (Less16U x y) => (LessThanU (CMPW (ZeroExt16to32 x) (ZeroExt16to32 y)))
298 (Less32U x y) => (LessThanU (CMPW x y))
299 (Less64U x y) => (LessThanU (CMP x y))
300
301 (Leq8 x y) => (LessEqual (CMPW (SignExt8to32 x) (SignExt8to32 y)))
302 (Leq16 x y) => (LessEqual (CMPW (SignExt16to32 x) (SignExt16to32 y)))
303 (Leq32 x y) => (LessEqual (CMPW x y))
304 (Leq64 x y) => (LessEqual (CMP x y))
305
306 // Refer to the comments for op Less64F above.
307 (Leq32F x y) => (LessEqualF (FCMPS x y))
308 (Leq64F x y) => (LessEqualF (FCMPD x y))
309
310 (Leq8U x y) => (LessEqualU (CMPW (ZeroExt8to32 x) (ZeroExt8to32 y)))
311 (Leq16U x y) => (LessEqualU (CMPW (ZeroExt16to32 x) (ZeroExt16to32 y)))
312 (Leq32U x y) => (LessEqualU (CMPW x y))
313 (Leq64U x y) => (LessEqualU (CMP x y))
314
315 // Optimize comparison between a floating-point value and 0.0 with "FCMP $(0.0), Fn"
316 (FCMPS x (FMOVSconst [0])) => (FCMPS0 x)
317 (FCMPS (FMOVSconst [0]) x) => (InvertFlags (FCMPS0 x))
318 (FCMPD x (FMOVDconst [0])) => (FCMPD0 x)
319 (FCMPD (FMOVDconst [0]) x) => (InvertFlags (FCMPD0 x))
320
321 // CSEL needs a flag-generating argument. Synthesize a TSTW if necessary.
322 (CondSelect x y boolval) && flagArg(boolval) != nil => (CSEL [boolval.Op] x y flagArg(boolval))
323 (CondSelect x y boolval) && flagArg(boolval) == nil => (CSEL [OpARM64NotEqual] x y (TSTWconst [1] boolval))
324
325 (OffPtr [off] ptr:(SP)) && is32Bit(off) => (MOVDaddr [int32(off)] ptr)
326 (OffPtr [off] ptr) => (ADDconst [off] ptr)
327
328 (Addr {sym} base) => (MOVDaddr {sym} base)
329 (LocalAddr <t> {sym} base mem) && t.Elem().HasPointers() => (MOVDaddr {sym} (SPanchored base mem))
330 (LocalAddr <t> {sym} base _) && !t.Elem().HasPointers() => (MOVDaddr {sym} base)
331
332 // loads
333 (Load <t> ptr mem) && t.IsBoolean() => (MOVBUload ptr mem)
334 (Load <t> ptr mem) && (is8BitInt(t) && t.IsSigned()) => (MOVBload ptr mem)
335 (Load <t> ptr mem) && (is8BitInt(t) && !t.IsSigned()) => (MOVBUload ptr mem)
336 (Load <t> ptr mem) && (is16BitInt(t) && t.IsSigned()) => (MOVHload ptr mem)
337 (Load <t> ptr mem) && (is16BitInt(t) && !t.IsSigned()) => (MOVHUload ptr mem)
338 (Load <t> ptr mem) && (is32BitInt(t) && t.IsSigned()) => (MOVWload ptr mem)
339 (Load <t> ptr mem) && (is32BitInt(t) && !t.IsSigned()) => (MOVWUload ptr mem)
340 (Load <t> ptr mem) && (is64BitInt(t) || isPtr(t)) => (MOVDload ptr mem)
341 (Load <t> ptr mem) && is32BitFloat(t) => (FMOVSload ptr mem)
342 (Load <t> ptr mem) && is64BitFloat(t) => (FMOVDload ptr mem)
343
344 // stores
345 (Store {t} ptr val mem) && t.Size() == 1 => (MOVBstore ptr val mem)
346 (Store {t} ptr val mem) && t.Size() == 2 => (MOVHstore ptr val mem)
347 (Store {t} ptr val mem) && t.Size() == 4 && !t.IsFloat() => (MOVWstore ptr val mem)
348 (Store {t} ptr val mem) && t.Size() == 8 && !t.IsFloat() => (MOVDstore ptr val mem)
349 (Store {t} ptr val mem) && t.Size() == 4 && t.IsFloat() => (FMOVSstore ptr val mem)
350 (Store {t} ptr val mem) && t.Size() == 8 && t.IsFloat() => (FMOVDstore ptr val mem)
351
352 // zeroing
353 (Zero [0] _ mem) => mem
354 (Zero [1] ptr mem) => (MOVBstore ptr (MOVDconst [0]) mem)
355 (Zero [2] ptr mem) => (MOVHstore ptr (MOVDconst [0]) mem)
356 (Zero [4] ptr mem) => (MOVWstore ptr (MOVDconst [0]) mem)
357 (Zero [3] ptr mem) =>
358 (MOVBstore [2] ptr (MOVDconst [0])
359 (MOVHstore ptr (MOVDconst [0]) mem))
360 (Zero [5] ptr mem) =>
361 (MOVBstore [4] ptr (MOVDconst [0])
362 (MOVWstore ptr (MOVDconst [0]) mem))
363 (Zero [6] ptr mem) =>
364 (MOVHstore [4] ptr (MOVDconst [0])
365 (MOVWstore ptr (MOVDconst [0]) mem))
366 (Zero [7] ptr mem) =>
367 (MOVWstore [3] ptr (MOVDconst [0])
368 (MOVWstore ptr (MOVDconst [0]) mem))
369 (Zero [8] ptr mem) => (MOVDstore ptr (MOVDconst [0]) mem)
370 (Zero [9] ptr mem) =>
371 (MOVBstore [8] ptr (MOVDconst [0])
372 (MOVDstore ptr (MOVDconst [0]) mem))
373 (Zero [10] ptr mem) =>
374 (MOVHstore [8] ptr (MOVDconst [0])
375 (MOVDstore ptr (MOVDconst [0]) mem))
376 (Zero [11] ptr mem) =>
377 (MOVDstore [3] ptr (MOVDconst [0])
378 (MOVDstore ptr (MOVDconst [0]) mem))
379 (Zero [12] ptr mem) =>
380 (MOVWstore [8] ptr (MOVDconst [0])
381 (MOVDstore ptr (MOVDconst [0]) mem))
382 (Zero [13] ptr mem) =>
383 (MOVDstore [5] ptr (MOVDconst [0])
384 (MOVDstore ptr (MOVDconst [0]) mem))
385 (Zero [14] ptr mem) =>
386 (MOVDstore [6] ptr (MOVDconst [0])
387 (MOVDstore ptr (MOVDconst [0]) mem))
388 (Zero [15] ptr mem) =>
389 (MOVDstore [7] ptr (MOVDconst [0])
390 (MOVDstore ptr (MOVDconst [0]) mem))
391 (Zero [16] ptr mem) =>
392 (STP [0] ptr (MOVDconst [0]) (MOVDconst [0]) mem)
393
394 (Zero [32] ptr mem) =>
395 (STP [16] ptr (MOVDconst [0]) (MOVDconst [0])
396 (STP [0] ptr (MOVDconst [0]) (MOVDconst [0]) mem))
397
398 (Zero [48] ptr mem) =>
399 (STP [32] ptr (MOVDconst [0]) (MOVDconst [0])
400 (STP [16] ptr (MOVDconst [0]) (MOVDconst [0])
401 (STP [0] ptr (MOVDconst [0]) (MOVDconst [0]) mem)))
402
403 (Zero [64] ptr mem) =>
404 (STP [48] ptr (MOVDconst [0]) (MOVDconst [0])
405 (STP [32] ptr (MOVDconst [0]) (MOVDconst [0])
406 (STP [16] ptr (MOVDconst [0]) (MOVDconst [0])
407 (STP [0] ptr (MOVDconst [0]) (MOVDconst [0]) mem))))
408
409 // strip off fractional word zeroing
410 (Zero [s] ptr mem) && s%16 != 0 && s%16 <= 8 && s > 16 =>
411 (Zero [8]
412 (OffPtr <ptr.Type> ptr [s-8])
413 (Zero [s-s%16] ptr mem))
414 (Zero [s] ptr mem) && s%16 != 0 && s%16 > 8 && s > 16 =>
415 (Zero [16]
416 (OffPtr <ptr.Type> ptr [s-16])
417 (Zero [s-s%16] ptr mem))
418
419 // medium zeroing uses a duff device
420 // 4, 16, and 64 are magic constants, see runtime/mkduff.go
421 (Zero [s] ptr mem)
422 && s%16 == 0 && s > 64 && s <= 16*64
423 && !config.noDuffDevice =>
424 (DUFFZERO [4 * (64 - s/16)] ptr mem)
425
426 // large zeroing uses a loop
427 (Zero [s] ptr mem)
428 && s%16 == 0 && (s > 16*64 || config.noDuffDevice) =>
429 (LoweredZero
430 ptr
431 (ADDconst <ptr.Type> [s-16] ptr)
432 mem)
433
434 // moves
435 (Move [0] _ _ mem) => mem
436 (Move [1] dst src mem) => (MOVBstore dst (MOVBUload src mem) mem)
437 (Move [2] dst src mem) => (MOVHstore dst (MOVHUload src mem) mem)
438 (Move [3] dst src mem) =>
439 (MOVBstore [2] dst (MOVBUload [2] src mem)
440 (MOVHstore dst (MOVHUload src mem) mem))
441 (Move [4] dst src mem) => (MOVWstore dst (MOVWUload src mem) mem)
442 (Move [5] dst src mem) =>
443 (MOVBstore [4] dst (MOVBUload [4] src mem)
444 (MOVWstore dst (MOVWUload src mem) mem))
445 (Move [6] dst src mem) =>
446 (MOVHstore [4] dst (MOVHUload [4] src mem)
447 (MOVWstore dst (MOVWUload src mem) mem))
448 (Move [7] dst src mem) =>
449 (MOVWstore [3] dst (MOVWUload [3] src mem)
450 (MOVWstore dst (MOVWUload src mem) mem))
451 (Move [8] dst src mem) => (MOVDstore dst (MOVDload src mem) mem)
452 (Move [9] dst src mem) =>
453 (MOVBstore [8] dst (MOVBUload [8] src mem)
454 (MOVDstore dst (MOVDload src mem) mem))
455 (Move [10] dst src mem) =>
456 (MOVHstore [8] dst (MOVHUload [8] src mem)
457 (MOVDstore dst (MOVDload src mem) mem))
458 (Move [11] dst src mem) =>
459 (MOVDstore [3] dst (MOVDload [3] src mem)
460 (MOVDstore dst (MOVDload src mem) mem))
461 (Move [12] dst src mem) =>
462 (MOVWstore [8] dst (MOVWUload [8] src mem)
463 (MOVDstore dst (MOVDload src mem) mem))
464 (Move [13] dst src mem) =>
465 (MOVDstore [5] dst (MOVDload [5] src mem)
466 (MOVDstore dst (MOVDload src mem) mem))
467 (Move [14] dst src mem) =>
468 (MOVDstore [6] dst (MOVDload [6] src mem)
469 (MOVDstore dst (MOVDload src mem) mem))
470 (Move [15] dst src mem) =>
471 (MOVDstore [7] dst (MOVDload [7] src mem)
472 (MOVDstore dst (MOVDload src mem) mem))
473 (Move [16] dst src mem) =>
474 (STP dst (Select0 <typ.UInt64> (LDP src mem)) (Select1 <typ.UInt64> (LDP src mem)) mem)
475 (Move [32] dst src mem) =>
476 (STP [16] dst (Select0 <typ.UInt64> (LDP [16] src mem)) (Select1 <typ.UInt64> (LDP [16] src mem))
477 (STP dst (Select0 <typ.UInt64> (LDP src mem)) (Select1 <typ.UInt64> (LDP src mem)) mem))
478 (Move [48] dst src mem) =>
479 (STP [32] dst (Select0 <typ.UInt64> (LDP [32] src mem)) (Select1 <typ.UInt64> (LDP [32] src mem))
480 (STP [16] dst (Select0 <typ.UInt64> (LDP [16] src mem)) (Select1 <typ.UInt64> (LDP [16] src mem))
481 (STP dst (Select0 <typ.UInt64> (LDP src mem)) (Select1 <typ.UInt64> (LDP src mem)) mem)))
482 (Move [64] dst src mem) =>
483 (STP [48] dst (Select0 <typ.UInt64> (LDP [48] src mem)) (Select1 <typ.UInt64> (LDP [48] src mem))
484 (STP [32] dst (Select0 <typ.UInt64> (LDP [32] src mem)) (Select1 <typ.UInt64> (LDP [32] src mem))
485 (STP [16] dst (Select0 <typ.UInt64> (LDP [16] src mem)) (Select1 <typ.UInt64> (LDP [16] src mem))
486 (STP dst (Select0 <typ.UInt64> (LDP src mem)) (Select1 <typ.UInt64> (LDP src mem)) mem))))
487
488 (MOVDstorezero {s} [i] ptr x:(MOVDstorezero {s} [i+8] ptr mem)) && x.Uses == 1 && setPos(v, x.Pos) && clobber(x) => (MOVQstorezero {s} [i] ptr mem)
489 (MOVDstorezero {s} [i] ptr x:(MOVDstorezero {s} [i-8] ptr mem)) && x.Uses == 1 && setPos(v, x.Pos) && clobber(x) => (MOVQstorezero {s} [i-8] ptr mem)
490
491 // strip off fractional word move
492 (Move [s] dst src mem) && s%16 != 0 && s%16 <= 8 && s > 16 =>
493 (Move [8]
494 (OffPtr <dst.Type> dst [s-8])
495 (OffPtr <src.Type> src [s-8])
496 (Move [s-s%16] dst src mem))
497 (Move [s] dst src mem) && s%16 != 0 && s%16 > 8 && s > 16 =>
498 (Move [16]
499 (OffPtr <dst.Type> dst [s-16])
500 (OffPtr <src.Type> src [s-16])
501 (Move [s-s%16] dst src mem))
502
503 // medium move uses a duff device
504 (Move [s] dst src mem)
505 && s > 64 && s <= 16*64 && s%16 == 0
506 && !config.noDuffDevice && logLargeCopy(v, s) =>
507 (DUFFCOPY [8 * (64 - s/16)] dst src mem)
508 // 8 is the number of bytes to encode:
509 //
510 // LDP.P 16(R16), (R26, R27)
511 // STP.P (R26, R27), 16(R17)
512 //
513 // 64 is number of these blocks. See runtime/duff_arm64.s:duffcopy
514
515 // large move uses a loop
516 (Move [s] dst src mem)
517 && s%16 == 0 && (s > 16*64 || config.noDuffDevice)
518 && logLargeCopy(v, s) =>
519 (LoweredMove
520 dst
521 src
522 (ADDconst <src.Type> src [s-16])
523 mem)
524
525 // calls
526 (StaticCall ...) => (CALLstatic ...)
527 (ClosureCall ...) => (CALLclosure ...)
528 (InterCall ...) => (CALLinter ...)
529 (TailCall ...) => (CALLtail ...)
530
531 // checks
532 (NilCheck ...) => (LoweredNilCheck ...)
533 (IsNonNil ptr) => (NotEqual (CMPconst [0] ptr))
534 (IsInBounds idx len) => (LessThanU (CMP idx len))
535 (IsSliceInBounds idx len) => (LessEqualU (CMP idx len))
536
537 // pseudo-ops
538 (GetClosurePtr ...) => (LoweredGetClosurePtr ...)
539 (GetCallerSP ...) => (LoweredGetCallerSP ...)
540 (GetCallerPC ...) => (LoweredGetCallerPC ...)
541
542 // Absorb pseudo-ops into blocks.
543 (If (Equal cc) yes no) => (EQ cc yes no)
544 (If (NotEqual cc) yes no) => (NE cc yes no)
545 (If (LessThan cc) yes no) => (LT cc yes no)
546 (If (LessThanU cc) yes no) => (ULT cc yes no)
547 (If (LessEqual cc) yes no) => (LE cc yes no)
548 (If (LessEqualU cc) yes no) => (ULE cc yes no)
549 (If (GreaterThan cc) yes no) => (GT cc yes no)
550 (If (GreaterThanU cc) yes no) => (UGT cc yes no)
551 (If (GreaterEqual cc) yes no) => (GE cc yes no)
552 (If (GreaterEqualU cc) yes no) => (UGE cc yes no)
553 (If (LessThanF cc) yes no) => (FLT cc yes no)
554 (If (LessEqualF cc) yes no) => (FLE cc yes no)
555 (If (GreaterThanF cc) yes no) => (FGT cc yes no)
556 (If (GreaterEqualF cc) yes no) => (FGE cc yes no)
557
558 (If cond yes no) => (TBNZ [0] cond yes no)
559
560 (JumpTable idx) => (JUMPTABLE {makeJumpTableSym(b)} idx (MOVDaddr <typ.Uintptr> {makeJumpTableSym(b)} (SB)))
561
562 // atomic intrinsics
563 // Note: these ops do not accept offset.
564 (AtomicLoad8 ...) => (LDARB ...)
565 (AtomicLoad32 ...) => (LDARW ...)
566 (AtomicLoad64 ...) => (LDAR ...)
567 (AtomicLoadPtr ...) => (LDAR ...)
568
569 (AtomicStore8 ...) => (STLRB ...)
570 (AtomicStore32 ...) => (STLRW ...)
571 (AtomicStore64 ...) => (STLR ...)
572 (AtomicStorePtrNoWB ...) => (STLR ...)
573
574 (AtomicExchange(8|32|64) ...) => (LoweredAtomicExchange(8|32|64) ...)
575 (AtomicAdd(32|64) ...) => (LoweredAtomicAdd(32|64) ...)
576 (AtomicCompareAndSwap(32|64) ...) => (LoweredAtomicCas(32|64) ...)
577
578 (AtomicAdd(32|64)Variant ...) => (LoweredAtomicAdd(32|64)Variant ...)
579 (AtomicExchange(8|32|64)Variant ...) => (LoweredAtomicExchange(8|32|64)Variant ...)
580 (AtomicCompareAndSwap(32|64)Variant ...) => (LoweredAtomicCas(32|64)Variant ...)
581
582 // Return old contents.
583 (AtomicAnd(64|32|8)value ...) => (LoweredAtomicAnd(64|32|8) ...)
584 (AtomicOr(64|32|8)value ...) => (LoweredAtomicOr(64|32|8) ...)
585 (AtomicAnd(64|32|8)valueVariant ...) => (LoweredAtomicAnd(64|32|8)Variant ...)
586 (AtomicOr(64|32|8)valueVariant ...) => (LoweredAtomicOr(64|32|8)Variant ...)
587
588 // Write barrier.
589 (WB ...) => (LoweredWB ...)
590
591 // Publication barrier (0xe is ST option)
592 (PubBarrier mem) => (DMB [0xe] mem)
593
594 (PanicBounds [kind] x y mem) && boundsABI(kind) == 0 => (LoweredPanicBoundsA [kind] x y mem)
595 (PanicBounds [kind] x y mem) && boundsABI(kind) == 1 => (LoweredPanicBoundsB [kind] x y mem)
596 (PanicBounds [kind] x y mem) && boundsABI(kind) == 2 => (LoweredPanicBoundsC [kind] x y mem)
597
598 // Optimizations
599
600 // Absorb boolean tests into block
601 (NZ (Equal cc) yes no) => (EQ cc yes no)
602 (NZ (NotEqual cc) yes no) => (NE cc yes no)
603 (NZ (LessThan cc) yes no) => (LT cc yes no)
604 (NZ (LessThanU cc) yes no) => (ULT cc yes no)
605 (NZ (LessEqual cc) yes no) => (LE cc yes no)
606 (NZ (LessEqualU cc) yes no) => (ULE cc yes no)
607 (NZ (GreaterThan cc) yes no) => (GT cc yes no)
608 (NZ (GreaterThanU cc) yes no) => (UGT cc yes no)
609 (NZ (GreaterEqual cc) yes no) => (GE cc yes no)
610 (NZ (GreaterEqualU cc) yes no) => (UGE cc yes no)
611 (NZ (LessThanF cc) yes no) => (FLT cc yes no)
612 (NZ (LessEqualF cc) yes no) => (FLE cc yes no)
613 (NZ (GreaterThanF cc) yes no) => (FGT cc yes no)
614 (NZ (GreaterEqualF cc) yes no) => (FGE cc yes no)
615
616 (TBNZ [0] (Equal cc) yes no) => (EQ cc yes no)
617 (TBNZ [0] (NotEqual cc) yes no) => (NE cc yes no)
618 (TBNZ [0] (LessThan cc) yes no) => (LT cc yes no)
619 (TBNZ [0] (LessThanU cc) yes no) => (ULT cc yes no)
620 (TBNZ [0] (LessEqual cc) yes no) => (LE cc yes no)
621 (TBNZ [0] (LessEqualU cc) yes no) => (ULE cc yes no)
622 (TBNZ [0] (GreaterThan cc) yes no) => (GT cc yes no)
623 (TBNZ [0] (GreaterThanU cc) yes no) => (UGT cc yes no)
624 (TBNZ [0] (GreaterEqual cc) yes no) => (GE cc yes no)
625 (TBNZ [0] (GreaterEqualU cc) yes no) => (UGE cc yes no)
626 (TBNZ [0] (LessThanF cc) yes no) => (FLT cc yes no)
627 (TBNZ [0] (LessEqualF cc) yes no) => (FLE cc yes no)
628 (TBNZ [0] (GreaterThanF cc) yes no) => (FGT cc yes no)
629 (TBNZ [0] (GreaterEqualF cc) yes no) => (FGE cc yes no)
630
631 ((EQ|NE|LT|LE|GT|GE) (CMPconst [0] z:(AND x y)) yes no) && z.Uses == 1 => ((EQ|NE|LT|LE|GT|GE) (TST x y) yes no)
632 ((EQ|NE|LT|LE|GT|GE) (CMPconst [0] x:(ANDconst [c] y)) yes no) && x.Uses == 1 => ((EQ|NE|LT|LE|GT|GE) (TSTconst [c] y) yes no)
633 ((EQ|NE|LT|LE|GT|GE) (CMPWconst [0] z:(AND x y)) yes no) && z.Uses == 1 => ((EQ|NE|LT|LE|GT|GE) (TSTW x y) yes no)
634 ((EQ|NE|LT|LE|GT|GE) (CMPWconst [0] x:(ANDconst [c] y)) yes no) && x.Uses == 1 => ((EQ|NE|LT|LE|GT|GE) (TSTWconst [int32(c)] y) yes no)
635
636 // For conditional instructions such as CSET, CSEL.
637 ((Equal|NotEqual|LessThan|LessEqual|GreaterThan|GreaterEqual) (CMPconst [0] z:(AND x y))) && z.Uses == 1 =>
638 ((Equal|NotEqual|LessThan|LessEqual|GreaterThan|GreaterEqual) (TST x y))
639 ((Equal|NotEqual|LessThan|LessEqual|GreaterThan|GreaterEqual) (CMPWconst [0] x:(ANDconst [c] y))) && x.Uses == 1 =>
640 ((Equal|NotEqual|LessThan|LessEqual|GreaterThan|GreaterEqual) (TSTWconst [int32(c)] y))
641 ((Equal|NotEqual|LessThan|LessEqual|GreaterThan|GreaterEqual) (CMPWconst [0] z:(AND x y))) && z.Uses == 1 =>
642 ((Equal|NotEqual|LessThan|LessEqual|GreaterThan|GreaterEqual) (TSTW x y))
643 ((Equal|NotEqual|LessThan|LessEqual|GreaterThan|GreaterEqual) (CMPconst [0] x:(ANDconst [c] y))) && x.Uses == 1 =>
644 ((Equal|NotEqual|LessThan|LessEqual|GreaterThan|GreaterEqual) (TSTconst [c] y))
645
646 ((EQ|NE|LT|LE|GT|GE) (CMPconst [0] x:(ADDconst [c] y)) yes no) && x.Uses == 1 => ((EQ|NE|LTnoov|LEnoov|GTnoov|GEnoov) (CMNconst [c] y) yes no)
647 ((EQ|NE|LT|LE|GT|GE) (CMPWconst [0] x:(ADDconst [c] y)) yes no) && x.Uses == 1 => ((EQ|NE|LTnoov|LEnoov|GTnoov|GEnoov) (CMNWconst [int32(c)] y) yes no)
648 ((EQ|NE|LT|LE|GT|GE) (CMPconst [0] z:(ADD x y)) yes no) && z.Uses == 1 => ((EQ|NE|LTnoov|LEnoov|GTnoov|GEnoov) (CMN x y) yes no)
649 ((EQ|NE|LT|LE|GT|GE) (CMPWconst [0] z:(ADD x y)) yes no) && z.Uses == 1 => ((EQ|NE|LTnoov|LEnoov|GTnoov|GEnoov) (CMNW x y) yes no)
650
651 // CMP(x,-y) -> CMN(x,y) is only valid for unordered comparison, if y can be -1<<63
652 ((EQ|NE) (CMP x z:(NEG y)) yes no) && z.Uses == 1 => ((EQ|NE) (CMN x y) yes no)
653 ((Equal|NotEqual) (CMP x z:(NEG y))) && z.Uses == 1 => ((Equal|NotEqual) (CMN x y))
654
655 // CMPW(x,-y) -> CMNW(x,y) is only valid for unordered comparison, if y can be -1<<31
656 ((EQ|NE) (CMPW x z:(NEG y)) yes no) && z.Uses == 1 => ((EQ|NE) (CMNW x y) yes no)
657 ((Equal|NotEqual) (CMPW x z:(NEG y))) && z.Uses == 1 => ((Equal|NotEqual) (CMNW x y))
658
659 // For conditional instructions such as CSET, CSEL.
660 // TODO: add support for LE, GT, overflow needs to be considered.
661 ((Equal|NotEqual|LessThan|GreaterEqual) (CMPconst [0] x:(ADDconst [c] y))) && x.Uses == 1 => ((Equal|NotEqual|LessThanNoov|GreaterEqualNoov) (CMNconst [c] y))
662 ((Equal|NotEqual|LessThan|GreaterEqual) (CMPWconst [0] x:(ADDconst [c] y))) && x.Uses == 1 => ((Equal|NotEqual|LessThanNoov|GreaterEqualNoov) (CMNWconst [int32(c)] y))
663 ((Equal|NotEqual|LessThan|GreaterEqual) (CMPconst [0] z:(ADD x y))) && z.Uses == 1 => ((Equal|NotEqual|LessThanNoov|GreaterEqualNoov) (CMN x y))
664 ((Equal|NotEqual|LessThan|GreaterEqual) (CMPWconst [0] z:(ADD x y))) && z.Uses == 1 => ((Equal|NotEqual|LessThanNoov|GreaterEqualNoov) (CMNW x y))
665 ((Equal|NotEqual|LessThan|GreaterEqual) (CMPconst [0] z:(MADD a x y))) && z.Uses == 1 => ((Equal|NotEqual|LessThanNoov|GreaterEqualNoov) (CMN a (MUL <x.Type> x y)))
666 ((Equal|NotEqual|LessThan|GreaterEqual) (CMPconst [0] z:(MSUB a x y))) && z.Uses == 1 => ((Equal|NotEqual|LessThanNoov|GreaterEqualNoov) (CMP a (MUL <x.Type> x y)))
667 ((Equal|NotEqual|LessThan|GreaterEqual) (CMPWconst [0] z:(MADDW a x y))) && z.Uses == 1 => ((Equal|NotEqual|LessThanNoov|GreaterEqualNoov) (CMNW a (MULW <x.Type> x y)))
668 ((Equal|NotEqual|LessThan|GreaterEqual) (CMPWconst [0] z:(MSUBW a x y))) && z.Uses == 1 => ((Equal|NotEqual|LessThanNoov|GreaterEqualNoov) (CMPW a (MULW <x.Type> x y)))
669
670 ((CMPconst|CMNconst) [c] y) && c < 0 && c != -1<<63 => ((CMNconst|CMPconst) [-c] y)
671 ((CMPWconst|CMNWconst) [c] y) && c < 0 && c != -1<<31 => ((CMNWconst|CMPWconst) [-c] y)
672
673 ((EQ|NE) (CMPconst [0] x) yes no) => ((Z|NZ) x yes no)
674 ((EQ|NE) (CMPWconst [0] x) yes no) => ((ZW|NZW) x yes no)
675
676 ((EQ|NE|LT|LE|GT|GE) (CMPconst [0] z:(MADD a x y)) yes no) && z.Uses==1 => ((EQ|NE|LTnoov|LEnoov|GTnoov|GEnoov) (CMN a (MUL <x.Type> x y)) yes no)
677 ((EQ|NE|LT|LE|GT|GE) (CMPconst [0] z:(MSUB a x y)) yes no) && z.Uses==1 => ((EQ|NE|LTnoov|LEnoov|GTnoov|GEnoov) (CMP a (MUL <x.Type> x y)) yes no)
678 ((EQ|NE|LT|LE|GT|GE) (CMPWconst [0] z:(MADDW a x y)) yes no) && z.Uses==1 => ((EQ|NE|LTnoov|LEnoov|GTnoov|GEnoov) (CMNW a (MULW <x.Type> x y)) yes no)
679 ((EQ|NE|LT|LE|GT|GE) (CMPWconst [0] z:(MSUBW a x y)) yes no) && z.Uses==1 => ((EQ|NE|LTnoov|LEnoov|GTnoov|GEnoov) (CMPW a (MULW <x.Type> x y)) yes no)
680
681 // Absorb bit-tests into block
682 (Z (ANDconst [c] x) yes no) && oneBit(c) => (TBZ [int64(ntz64(c))] x yes no)
683 (NZ (ANDconst [c] x) yes no) && oneBit(c) => (TBNZ [int64(ntz64(c))] x yes no)
684 (ZW (ANDconst [c] x) yes no) && oneBit(int64(uint32(c))) => (TBZ [int64(ntz64(int64(uint32(c))))] x yes no)
685 (NZW (ANDconst [c] x) yes no) && oneBit(int64(uint32(c))) => (TBNZ [int64(ntz64(int64(uint32(c))))] x yes no)
686 (EQ (TSTconst [c] x) yes no) && oneBit(c) => (TBZ [int64(ntz64(c))] x yes no)
687 (NE (TSTconst [c] x) yes no) && oneBit(c) => (TBNZ [int64(ntz64(c))] x yes no)
688 (EQ (TSTWconst [c] x) yes no) && oneBit(int64(uint32(c))) => (TBZ [int64(ntz64(int64(uint32(c))))] x yes no)
689 (NE (TSTWconst [c] x) yes no) && oneBit(int64(uint32(c))) => (TBNZ [int64(ntz64(int64(uint32(c))))] x yes no)
690
691 // Test sign-bit for signed comparisons against zero
692 (GE (CMPWconst [0] x) yes no) => (TBZ [31] x yes no)
693 (GE (CMPconst [0] x) yes no) => (TBZ [63] x yes no)
694 (LT (CMPWconst [0] x) yes no) => (TBNZ [31] x yes no)
695 (LT (CMPconst [0] x) yes no) => (TBNZ [63] x yes no)
696
697 // fold offset into address
698 (ADDconst [off1] (MOVDaddr [off2] {sym} ptr)) && is32Bit(off1+int64(off2)) =>
699 (MOVDaddr [int32(off1)+off2] {sym} ptr)
700
701 // fold address into load/store.
702 // Do not fold global variable access in -dynlink mode, where it will
703 // be rewritten to use the GOT via REGTMP, which currently cannot handle
704 // large offset.
705 (MOVBload [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
706 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
707 (MOVBload [off1+int32(off2)] {sym} ptr mem)
708 (MOVBUload [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
709 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
710 (MOVBUload [off1+int32(off2)] {sym} ptr mem)
711 (MOVHload [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
712 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
713 (MOVHload [off1+int32(off2)] {sym} ptr mem)
714 (MOVHUload [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
715 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
716 (MOVHUload [off1+int32(off2)] {sym} ptr mem)
717 (MOVWload [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
718 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
719 (MOVWload [off1+int32(off2)] {sym} ptr mem)
720 (MOVWUload [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
721 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
722 (MOVWUload [off1+int32(off2)] {sym} ptr mem)
723 (MOVDload [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
724 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
725 (MOVDload [off1+int32(off2)] {sym} ptr mem)
726 (LDP [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
727 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
728 (LDP [off1+int32(off2)] {sym} ptr mem)
729 (FMOVSload [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
730 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
731 (FMOVSload [off1+int32(off2)] {sym} ptr mem)
732 (FMOVDload [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
733 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
734 (FMOVDload [off1+int32(off2)] {sym} ptr mem)
735
736 // register indexed load
737 (MOVDload [off] {sym} (ADD ptr idx) mem) && off == 0 && sym == nil => (MOVDloadidx ptr idx mem)
738 (MOVWUload [off] {sym} (ADD ptr idx) mem) && off == 0 && sym == nil => (MOVWUloadidx ptr idx mem)
739 (MOVWload [off] {sym} (ADD ptr idx) mem) && off == 0 && sym == nil => (MOVWloadidx ptr idx mem)
740 (MOVHUload [off] {sym} (ADD ptr idx) mem) && off == 0 && sym == nil => (MOVHUloadidx ptr idx mem)
741 (MOVHload [off] {sym} (ADD ptr idx) mem) && off == 0 && sym == nil => (MOVHloadidx ptr idx mem)
742 (MOVBUload [off] {sym} (ADD ptr idx) mem) && off == 0 && sym == nil => (MOVBUloadidx ptr idx mem)
743 (MOVBload [off] {sym} (ADD ptr idx) mem) && off == 0 && sym == nil => (MOVBloadidx ptr idx mem)
744 (FMOVSload [off] {sym} (ADD ptr idx) mem) && off == 0 && sym == nil => (FMOVSloadidx ptr idx mem)
745 (FMOVDload [off] {sym} (ADD ptr idx) mem) && off == 0 && sym == nil => (FMOVDloadidx ptr idx mem)
746
747 (MOVDloadidx ptr (MOVDconst [c]) mem) && is32Bit(c) => (MOVDload [int32(c)] ptr mem)
748 (MOVDloadidx (MOVDconst [c]) ptr mem) && is32Bit(c) => (MOVDload [int32(c)] ptr mem)
749 (MOVWUloadidx ptr (MOVDconst [c]) mem) && is32Bit(c) => (MOVWUload [int32(c)] ptr mem)
750 (MOVWUloadidx (MOVDconst [c]) ptr mem) && is32Bit(c) => (MOVWUload [int32(c)] ptr mem)
751 (MOVWloadidx ptr (MOVDconst [c]) mem) && is32Bit(c) => (MOVWload [int32(c)] ptr mem)
752 (MOVWloadidx (MOVDconst [c]) ptr mem) && is32Bit(c) => (MOVWload [int32(c)] ptr mem)
753 (MOVHUloadidx ptr (MOVDconst [c]) mem) && is32Bit(c) => (MOVHUload [int32(c)] ptr mem)
754 (MOVHUloadidx (MOVDconst [c]) ptr mem) && is32Bit(c) => (MOVHUload [int32(c)] ptr mem)
755 (MOVHloadidx ptr (MOVDconst [c]) mem) && is32Bit(c) => (MOVHload [int32(c)] ptr mem)
756 (MOVHloadidx (MOVDconst [c]) ptr mem) && is32Bit(c) => (MOVHload [int32(c)] ptr mem)
757 (MOVBUloadidx ptr (MOVDconst [c]) mem) && is32Bit(c) => (MOVBUload [int32(c)] ptr mem)
758 (MOVBUloadidx (MOVDconst [c]) ptr mem) && is32Bit(c) => (MOVBUload [int32(c)] ptr mem)
759 (MOVBloadidx ptr (MOVDconst [c]) mem) && is32Bit(c) => (MOVBload [int32(c)] ptr mem)
760 (MOVBloadidx (MOVDconst [c]) ptr mem) && is32Bit(c) => (MOVBload [int32(c)] ptr mem)
761 (FMOVSloadidx ptr (MOVDconst [c]) mem) && is32Bit(c) => (FMOVSload [int32(c)] ptr mem)
762 (FMOVSloadidx (MOVDconst [c]) ptr mem) && is32Bit(c) => (FMOVSload [int32(c)] ptr mem)
763 (FMOVDloadidx ptr (MOVDconst [c]) mem) && is32Bit(c) => (FMOVDload [int32(c)] ptr mem)
764 (FMOVDloadidx (MOVDconst [c]) ptr mem) && is32Bit(c) => (FMOVDload [int32(c)] ptr mem)
765
766 // shifted register indexed load
767 (MOVDload [off] {sym} (ADDshiftLL [3] ptr idx) mem) && off == 0 && sym == nil => (MOVDloadidx8 ptr idx mem)
768 (MOVWUload [off] {sym} (ADDshiftLL [2] ptr idx) mem) && off == 0 && sym == nil => (MOVWUloadidx4 ptr idx mem)
769 (MOVWload [off] {sym} (ADDshiftLL [2] ptr idx) mem) && off == 0 && sym == nil => (MOVWloadidx4 ptr idx mem)
770 (MOVHUload [off] {sym} (ADDshiftLL [1] ptr idx) mem) && off == 0 && sym == nil => (MOVHUloadidx2 ptr idx mem)
771 (MOVHload [off] {sym} (ADDshiftLL [1] ptr idx) mem) && off == 0 && sym == nil => (MOVHloadidx2 ptr idx mem)
772 (MOVDloadidx ptr (SLLconst [3] idx) mem) => (MOVDloadidx8 ptr idx mem)
773 (MOVWloadidx ptr (SLLconst [2] idx) mem) => (MOVWloadidx4 ptr idx mem)
774 (MOVWUloadidx ptr (SLLconst [2] idx) mem) => (MOVWUloadidx4 ptr idx mem)
775 (MOVHloadidx ptr (SLLconst [1] idx) mem) => (MOVHloadidx2 ptr idx mem)
776 (MOVHUloadidx ptr (SLLconst [1] idx) mem) => (MOVHUloadidx2 ptr idx mem)
777 (MOVHloadidx ptr (ADD idx idx) mem) => (MOVHloadidx2 ptr idx mem)
778 (MOVHUloadidx ptr (ADD idx idx) mem) => (MOVHUloadidx2 ptr idx mem)
779 (MOVDloadidx (SLLconst [3] idx) ptr mem) => (MOVDloadidx8 ptr idx mem)
780 (MOVWloadidx (SLLconst [2] idx) ptr mem) => (MOVWloadidx4 ptr idx mem)
781 (MOVWUloadidx (SLLconst [2] idx) ptr mem) => (MOVWUloadidx4 ptr idx mem)
782 (MOVHloadidx (ADD idx idx) ptr mem) => (MOVHloadidx2 ptr idx mem)
783 (MOVHUloadidx (ADD idx idx) ptr mem) => (MOVHUloadidx2 ptr idx mem)
784 (MOVDloadidx8 ptr (MOVDconst [c]) mem) && is32Bit(c<<3) => (MOVDload [int32(c)<<3] ptr mem)
785 (MOVWUloadidx4 ptr (MOVDconst [c]) mem) && is32Bit(c<<2) => (MOVWUload [int32(c)<<2] ptr mem)
786 (MOVWloadidx4 ptr (MOVDconst [c]) mem) && is32Bit(c<<2) => (MOVWload [int32(c)<<2] ptr mem)
787 (MOVHUloadidx2 ptr (MOVDconst [c]) mem) && is32Bit(c<<1) => (MOVHUload [int32(c)<<1] ptr mem)
788 (MOVHloadidx2 ptr (MOVDconst [c]) mem) && is32Bit(c<<1) => (MOVHload [int32(c)<<1] ptr mem)
789
790 (FMOVDload [off] {sym} (ADDshiftLL [3] ptr idx) mem) && off == 0 && sym == nil => (FMOVDloadidx8 ptr idx mem)
791 (FMOVSload [off] {sym} (ADDshiftLL [2] ptr idx) mem) && off == 0 && sym == nil => (FMOVSloadidx4 ptr idx mem)
792 (FMOVDloadidx ptr (SLLconst [3] idx) mem) => (FMOVDloadidx8 ptr idx mem)
793 (FMOVSloadidx ptr (SLLconst [2] idx) mem) => (FMOVSloadidx4 ptr idx mem)
794 (FMOVDloadidx (SLLconst [3] idx) ptr mem) => (FMOVDloadidx8 ptr idx mem)
795 (FMOVSloadidx (SLLconst [2] idx) ptr mem) => (FMOVSloadidx4 ptr idx mem)
796 (FMOVDloadidx8 ptr (MOVDconst [c]) mem) && is32Bit(c<<3) => (FMOVDload ptr [int32(c)<<3] mem)
797 (FMOVSloadidx4 ptr (MOVDconst [c]) mem) && is32Bit(c<<2) => (FMOVSload ptr [int32(c)<<2] mem)
798
799 (MOVBstore [off1] {sym} (ADDconst [off2] ptr) val mem) && is32Bit(int64(off1)+off2)
800 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
801 (MOVBstore [off1+int32(off2)] {sym} ptr val mem)
802 (MOVHstore [off1] {sym} (ADDconst [off2] ptr) val mem) && is32Bit(int64(off1)+off2)
803 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
804 (MOVHstore [off1+int32(off2)] {sym} ptr val mem)
805 (MOVWstore [off1] {sym} (ADDconst [off2] ptr) val mem) && is32Bit(int64(off1)+off2)
806 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
807 (MOVWstore [off1+int32(off2)] {sym} ptr val mem)
808 (MOVDstore [off1] {sym} (ADDconst [off2] ptr) val mem) && is32Bit(int64(off1)+off2)
809 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
810 (MOVDstore [off1+int32(off2)] {sym} ptr val mem)
811 (STP [off1] {sym} (ADDconst [off2] ptr) val1 val2 mem) && is32Bit(int64(off1)+off2)
812 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
813 (STP [off1+int32(off2)] {sym} ptr val1 val2 mem)
814 (FMOVSstore [off1] {sym} (ADDconst [off2] ptr) val mem) && is32Bit(int64(off1)+off2)
815 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
816 (FMOVSstore [off1+int32(off2)] {sym} ptr val mem)
817 (FMOVDstore [off1] {sym} (ADDconst [off2] ptr) val mem) && is32Bit(int64(off1)+off2)
818 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
819 (FMOVDstore [off1+int32(off2)] {sym} ptr val mem)
820 (MOVBstorezero [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
821 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
822 (MOVBstorezero [off1+int32(off2)] {sym} ptr mem)
823 (MOVHstorezero [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
824 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
825 (MOVHstorezero [off1+int32(off2)] {sym} ptr mem)
826 (MOVWstorezero [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
827 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
828 (MOVWstorezero [off1+int32(off2)] {sym} ptr mem)
829 (MOVDstorezero [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
830 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
831 (MOVDstorezero [off1+int32(off2)] {sym} ptr mem)
832 (MOVQstorezero [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
833 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
834 (MOVQstorezero [off1+int32(off2)] {sym} ptr mem)
835
836 // register indexed store
837 (MOVDstore [off] {sym} (ADD ptr idx) val mem) && off == 0 && sym == nil => (MOVDstoreidx ptr idx val mem)
838 (MOVWstore [off] {sym} (ADD ptr idx) val mem) && off == 0 && sym == nil => (MOVWstoreidx ptr idx val mem)
839 (MOVHstore [off] {sym} (ADD ptr idx) val mem) && off == 0 && sym == nil => (MOVHstoreidx ptr idx val mem)
840 (MOVBstore [off] {sym} (ADD ptr idx) val mem) && off == 0 && sym == nil => (MOVBstoreidx ptr idx val mem)
841 (FMOVDstore [off] {sym} (ADD ptr idx) val mem) && off == 0 && sym == nil => (FMOVDstoreidx ptr idx val mem)
842 (FMOVSstore [off] {sym} (ADD ptr idx) val mem) && off == 0 && sym == nil => (FMOVSstoreidx ptr idx val mem)
843 (MOVDstoreidx ptr (MOVDconst [c]) val mem) && is32Bit(c) => (MOVDstore [int32(c)] ptr val mem)
844 (MOVDstoreidx (MOVDconst [c]) idx val mem) && is32Bit(c) => (MOVDstore [int32(c)] idx val mem)
845 (MOVWstoreidx ptr (MOVDconst [c]) val mem) && is32Bit(c) => (MOVWstore [int32(c)] ptr val mem)
846 (MOVWstoreidx (MOVDconst [c]) idx val mem) && is32Bit(c) => (MOVWstore [int32(c)] idx val mem)
847 (MOVHstoreidx ptr (MOVDconst [c]) val mem) && is32Bit(c) => (MOVHstore [int32(c)] ptr val mem)
848 (MOVHstoreidx (MOVDconst [c]) idx val mem) && is32Bit(c) => (MOVHstore [int32(c)] idx val mem)
849 (MOVBstoreidx ptr (MOVDconst [c]) val mem) && is32Bit(c) => (MOVBstore [int32(c)] ptr val mem)
850 (MOVBstoreidx (MOVDconst [c]) idx val mem) && is32Bit(c) => (MOVBstore [int32(c)] idx val mem)
851 (FMOVDstoreidx ptr (MOVDconst [c]) val mem) && is32Bit(c) => (FMOVDstore [int32(c)] ptr val mem)
852 (FMOVDstoreidx (MOVDconst [c]) idx val mem) && is32Bit(c) => (FMOVDstore [int32(c)] idx val mem)
853 (FMOVSstoreidx ptr (MOVDconst [c]) val mem) && is32Bit(c) => (FMOVSstore [int32(c)] ptr val mem)
854 (FMOVSstoreidx (MOVDconst [c]) idx val mem) && is32Bit(c) => (FMOVSstore [int32(c)] idx val mem)
855
856 // shifted register indexed store
857 (MOVDstore [off] {sym} (ADDshiftLL [3] ptr idx) val mem) && off == 0 && sym == nil => (MOVDstoreidx8 ptr idx val mem)
858 (MOVWstore [off] {sym} (ADDshiftLL [2] ptr idx) val mem) && off == 0 && sym == nil => (MOVWstoreidx4 ptr idx val mem)
859 (MOVHstore [off] {sym} (ADDshiftLL [1] ptr idx) val mem) && off == 0 && sym == nil => (MOVHstoreidx2 ptr idx val mem)
860 (MOVDstoreidx ptr (SLLconst [3] idx) val mem) => (MOVDstoreidx8 ptr idx val mem)
861 (MOVWstoreidx ptr (SLLconst [2] idx) val mem) => (MOVWstoreidx4 ptr idx val mem)
862 (MOVHstoreidx ptr (SLLconst [1] idx) val mem) => (MOVHstoreidx2 ptr idx val mem)
863 (MOVHstoreidx ptr (ADD idx idx) val mem) => (MOVHstoreidx2 ptr idx val mem)
864 (MOVDstoreidx (SLLconst [3] idx) ptr val mem) => (MOVDstoreidx8 ptr idx val mem)
865 (MOVWstoreidx (SLLconst [2] idx) ptr val mem) => (MOVWstoreidx4 ptr idx val mem)
866 (MOVHstoreidx (SLLconst [1] idx) ptr val mem) => (MOVHstoreidx2 ptr idx val mem)
867 (MOVHstoreidx (ADD idx idx) ptr val mem) => (MOVHstoreidx2 ptr idx val mem)
868 (MOVDstoreidx8 ptr (MOVDconst [c]) val mem) && is32Bit(c<<3) => (MOVDstore [int32(c)<<3] ptr val mem)
869 (MOVWstoreidx4 ptr (MOVDconst [c]) val mem) && is32Bit(c<<2) => (MOVWstore [int32(c)<<2] ptr val mem)
870 (MOVHstoreidx2 ptr (MOVDconst [c]) val mem) && is32Bit(c<<1) => (MOVHstore [int32(c)<<1] ptr val mem)
871
872 (FMOVDstore [off] {sym} (ADDshiftLL [3] ptr idx) val mem) && off == 0 && sym == nil => (FMOVDstoreidx8 ptr idx val mem)
873 (FMOVSstore [off] {sym} (ADDshiftLL [2] ptr idx) val mem) && off == 0 && sym == nil => (FMOVSstoreidx4 ptr idx val mem)
874 (FMOVDstoreidx ptr (SLLconst [3] idx) val mem) => (FMOVDstoreidx8 ptr idx val mem)
875 (FMOVSstoreidx ptr (SLLconst [2] idx) val mem) => (FMOVSstoreidx4 ptr idx val mem)
876 (FMOVDstoreidx (SLLconst [3] idx) ptr val mem) => (FMOVDstoreidx8 ptr idx val mem)
877 (FMOVSstoreidx (SLLconst [2] idx) ptr val mem) => (FMOVSstoreidx4 ptr idx val mem)
878 (FMOVDstoreidx8 ptr (MOVDconst [c]) val mem) && is32Bit(c<<3) => (FMOVDstore [int32(c)<<3] ptr val mem)
879 (FMOVSstoreidx4 ptr (MOVDconst [c]) val mem) && is32Bit(c<<2) => (FMOVSstore [int32(c)<<2] ptr val mem)
880
881 (MOVBload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
882 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
883 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
884 (MOVBload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
885 (MOVBUload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
886 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
887 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
888 (MOVBUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
889 (MOVHload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
890 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
891 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
892 (MOVHload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
893 (MOVHUload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
894 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
895 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
896 (MOVHUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
897 (MOVWload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
898 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
899 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
900 (MOVWload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
901 (MOVWUload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
902 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
903 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
904 (MOVWUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
905 (MOVDload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
906 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
907 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
908 (MOVDload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
909 (LDP [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
910 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
911 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
912 (LDP [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
913 (FMOVSload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
914 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
915 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
916 (FMOVSload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
917 (FMOVDload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
918 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
919 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
920 (FMOVDload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
921
922 (MOVBstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem)
923 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
924 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
925 (MOVBstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
926 (MOVHstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem)
927 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
928 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
929 (MOVHstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
930 (MOVWstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem)
931 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
932 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
933 (MOVWstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
934 (MOVDstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem)
935 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
936 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
937 (MOVDstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
938 (STP [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val1 val2 mem)
939 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
940 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
941 (STP [off1+off2] {mergeSym(sym1,sym2)} ptr val1 val2 mem)
942 (FMOVSstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem)
943 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
944 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
945 (FMOVSstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
946 (FMOVDstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem)
947 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
948 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
949 (FMOVDstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
950 (MOVBstorezero [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
951 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
952 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
953 (MOVBstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
954 (MOVHstorezero [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
955 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
956 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
957 (MOVHstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
958 (MOVWstorezero [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
959 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
960 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
961 (MOVWstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
962 (MOVDstorezero [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
963 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
964 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
965 (MOVDstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
966 (MOVQstorezero [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
967 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
968 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
969 (MOVQstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
970
971 // store zero
972 (MOVBstore [off] {sym} ptr (MOVDconst [0]) mem) => (MOVBstorezero [off] {sym} ptr mem)
973 (MOVHstore [off] {sym} ptr (MOVDconst [0]) mem) => (MOVHstorezero [off] {sym} ptr mem)
974 (MOVWstore [off] {sym} ptr (MOVDconst [0]) mem) => (MOVWstorezero [off] {sym} ptr mem)
975 (MOVDstore [off] {sym} ptr (MOVDconst [0]) mem) => (MOVDstorezero [off] {sym} ptr mem)
976 (STP [off] {sym} ptr (MOVDconst [0]) (MOVDconst [0]) mem) => (MOVQstorezero [off] {sym} ptr mem)
977
978 // register indexed store zero
979 (MOVDstorezero [off] {sym} (ADD ptr idx) mem) && off == 0 && sym == nil => (MOVDstorezeroidx ptr idx mem)
980 (MOVWstorezero [off] {sym} (ADD ptr idx) mem) && off == 0 && sym == nil => (MOVWstorezeroidx ptr idx mem)
981 (MOVHstorezero [off] {sym} (ADD ptr idx) mem) && off == 0 && sym == nil => (MOVHstorezeroidx ptr idx mem)
982 (MOVBstorezero [off] {sym} (ADD ptr idx) mem) && off == 0 && sym == nil => (MOVBstorezeroidx ptr idx mem)
983 (MOVDstoreidx ptr idx (MOVDconst [0]) mem) => (MOVDstorezeroidx ptr idx mem)
984 (MOVWstoreidx ptr idx (MOVDconst [0]) mem) => (MOVWstorezeroidx ptr idx mem)
985 (MOVHstoreidx ptr idx (MOVDconst [0]) mem) => (MOVHstorezeroidx ptr idx mem)
986 (MOVBstoreidx ptr idx (MOVDconst [0]) mem) => (MOVBstorezeroidx ptr idx mem)
987 (MOVDstorezeroidx ptr (MOVDconst [c]) mem) && is32Bit(c) => (MOVDstorezero [int32(c)] ptr mem)
988 (MOVDstorezeroidx (MOVDconst [c]) idx mem) && is32Bit(c) => (MOVDstorezero [int32(c)] idx mem)
989 (MOVWstorezeroidx ptr (MOVDconst [c]) mem) && is32Bit(c) => (MOVWstorezero [int32(c)] ptr mem)
990 (MOVWstorezeroidx (MOVDconst [c]) idx mem) && is32Bit(c) => (MOVWstorezero [int32(c)] idx mem)
991 (MOVHstorezeroidx ptr (MOVDconst [c]) mem) && is32Bit(c) => (MOVHstorezero [int32(c)] ptr mem)
992 (MOVHstorezeroidx (MOVDconst [c]) idx mem) && is32Bit(c) => (MOVHstorezero [int32(c)] idx mem)
993 (MOVBstorezeroidx ptr (MOVDconst [c]) mem) && is32Bit(c) => (MOVBstorezero [int32(c)] ptr mem)
994 (MOVBstorezeroidx (MOVDconst [c]) idx mem) && is32Bit(c) => (MOVBstorezero [int32(c)] idx mem)
995
996 // shifted register indexed store zero
997 (MOVDstorezero [off] {sym} (ADDshiftLL [3] ptr idx) mem) && off == 0 && sym == nil => (MOVDstorezeroidx8 ptr idx mem)
998 (MOVWstorezero [off] {sym} (ADDshiftLL [2] ptr idx) mem) && off == 0 && sym == nil => (MOVWstorezeroidx4 ptr idx mem)
999 (MOVHstorezero [off] {sym} (ADDshiftLL [1] ptr idx) mem) && off == 0 && sym == nil => (MOVHstorezeroidx2 ptr idx mem)
1000 (MOVDstorezeroidx ptr (SLLconst [3] idx) mem) => (MOVDstorezeroidx8 ptr idx mem)
1001 (MOVWstorezeroidx ptr (SLLconst [2] idx) mem) => (MOVWstorezeroidx4 ptr idx mem)
1002 (MOVHstorezeroidx ptr (SLLconst [1] idx) mem) => (MOVHstorezeroidx2 ptr idx mem)
1003 (MOVHstorezeroidx ptr (ADD idx idx) mem) => (MOVHstorezeroidx2 ptr idx mem)
1004 (MOVDstorezeroidx (SLLconst [3] idx) ptr mem) => (MOVDstorezeroidx8 ptr idx mem)
1005 (MOVWstorezeroidx (SLLconst [2] idx) ptr mem) => (MOVWstorezeroidx4 ptr idx mem)
1006 (MOVHstorezeroidx (SLLconst [1] idx) ptr mem) => (MOVHstorezeroidx2 ptr idx mem)
1007 (MOVHstorezeroidx (ADD idx idx) ptr mem) => (MOVHstorezeroidx2 ptr idx mem)
1008 (MOVDstoreidx8 ptr idx (MOVDconst [0]) mem) => (MOVDstorezeroidx8 ptr idx mem)
1009 (MOVWstoreidx4 ptr idx (MOVDconst [0]) mem) => (MOVWstorezeroidx4 ptr idx mem)
1010 (MOVHstoreidx2 ptr idx (MOVDconst [0]) mem) => (MOVHstorezeroidx2 ptr idx mem)
1011 (MOVDstorezeroidx8 ptr (MOVDconst [c]) mem) && is32Bit(c<<3) => (MOVDstorezero [int32(c<<3)] ptr mem)
1012 (MOVWstorezeroidx4 ptr (MOVDconst [c]) mem) && is32Bit(c<<2) => (MOVWstorezero [int32(c<<2)] ptr mem)
1013 (MOVHstorezeroidx2 ptr (MOVDconst [c]) mem) && is32Bit(c<<1) => (MOVHstorezero [int32(c<<1)] ptr mem)
1014
1015 // replace load from same location as preceding store with zero/sign extension (or copy in case of full width)
1016 // these seem to have bad interaction with other rules, resulting in slower code
1017 //(MOVBload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVBreg x)
1018 //(MOVBUload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVBUreg x)
1019 //(MOVHload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVHreg x)
1020 //(MOVHUload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVHUreg x)
1021 //(MOVWload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVWreg x)
1022 //(MOVWUload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVWUreg x)
1023 //(MOVDload [off] {sym} ptr (MOVDstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
1024 //(FMOVSload [off] {sym} ptr (FMOVSstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
1025 //(FMOVDload [off] {sym} ptr (FMOVDstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
1026 //(LDP [off] {sym} ptr (STP [off2] {sym2} ptr2 x y _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x y
1027
1028 (MOVBload [off] {sym} ptr (MOVBstorezero [off2] {sym2} ptr2 _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) => (MOVDconst [0])
1029 (MOVBUload [off] {sym} ptr (MOVBstorezero [off2] {sym2} ptr2 _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) => (MOVDconst [0])
1030 (MOVHload [off] {sym} ptr (MOVHstorezero [off2] {sym2} ptr2 _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) => (MOVDconst [0])
1031 (MOVHUload [off] {sym} ptr (MOVHstorezero [off2] {sym2} ptr2 _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) => (MOVDconst [0])
1032 (MOVWload [off] {sym} ptr (MOVWstorezero [off2] {sym2} ptr2 _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) => (MOVDconst [0])
1033 (MOVWUload [off] {sym} ptr (MOVWstorezero [off2] {sym2} ptr2 _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) => (MOVDconst [0])
1034 (MOVDload [off] {sym} ptr (MOVDstorezero [off2] {sym2} ptr2 _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) => (MOVDconst [0])
1035
1036 (MOVBloadidx ptr idx (MOVBstorezeroidx ptr2 idx2 _))
1037 && (isSamePtr(ptr, ptr2) && isSamePtr(idx, idx2) || isSamePtr(ptr, idx2) && isSamePtr(idx, ptr2)) => (MOVDconst [0])
1038 (MOVBUloadidx ptr idx (MOVBstorezeroidx ptr2 idx2 _))
1039 && (isSamePtr(ptr, ptr2) && isSamePtr(idx, idx2) || isSamePtr(ptr, idx2) && isSamePtr(idx, ptr2)) => (MOVDconst [0])
1040 (MOVHloadidx ptr idx (MOVHstorezeroidx ptr2 idx2 _))
1041 && (isSamePtr(ptr, ptr2) && isSamePtr(idx, idx2) || isSamePtr(ptr, idx2) && isSamePtr(idx, ptr2)) => (MOVDconst [0])
1042 (MOVHUloadidx ptr idx (MOVHstorezeroidx ptr2 idx2 _))
1043 && (isSamePtr(ptr, ptr2) && isSamePtr(idx, idx2) || isSamePtr(ptr, idx2) && isSamePtr(idx, ptr2)) => (MOVDconst [0])
1044 (MOVWloadidx ptr idx (MOVWstorezeroidx ptr2 idx2 _))
1045 && (isSamePtr(ptr, ptr2) && isSamePtr(idx, idx2) || isSamePtr(ptr, idx2) && isSamePtr(idx, ptr2)) => (MOVDconst [0])
1046 (MOVWUloadidx ptr idx (MOVWstorezeroidx ptr2 idx2 _))
1047 && (isSamePtr(ptr, ptr2) && isSamePtr(idx, idx2) || isSamePtr(ptr, idx2) && isSamePtr(idx, ptr2)) => (MOVDconst [0])
1048 (MOVDloadidx ptr idx (MOVDstorezeroidx ptr2 idx2 _))
1049 && (isSamePtr(ptr, ptr2) && isSamePtr(idx, idx2) || isSamePtr(ptr, idx2) && isSamePtr(idx, ptr2)) => (MOVDconst [0])
1050
1051 (MOVHloadidx2 ptr idx (MOVHstorezeroidx2 ptr2 idx2 _)) && isSamePtr(ptr, ptr2) && isSamePtr(idx, idx2) => (MOVDconst [0])
1052 (MOVHUloadidx2 ptr idx (MOVHstorezeroidx2 ptr2 idx2 _)) && isSamePtr(ptr, ptr2) && isSamePtr(idx, idx2) => (MOVDconst [0])
1053 (MOVWloadidx4 ptr idx (MOVWstorezeroidx4 ptr2 idx2 _)) && isSamePtr(ptr, ptr2) && isSamePtr(idx, idx2) => (MOVDconst [0])
1054 (MOVWUloadidx4 ptr idx (MOVWstorezeroidx4 ptr2 idx2 _)) && isSamePtr(ptr, ptr2) && isSamePtr(idx, idx2) => (MOVDconst [0])
1055 (MOVDloadidx8 ptr idx (MOVDstorezeroidx8 ptr2 idx2 _)) && isSamePtr(ptr, ptr2) && isSamePtr(idx, idx2) => (MOVDconst [0])
1056
1057 // don't extend before store
1058 (MOVBstore [off] {sym} ptr (MOVBreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
1059 (MOVBstore [off] {sym} ptr (MOVBUreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
1060 (MOVBstore [off] {sym} ptr (MOVHreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
1061 (MOVBstore [off] {sym} ptr (MOVHUreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
1062 (MOVBstore [off] {sym} ptr (MOVWreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
1063 (MOVBstore [off] {sym} ptr (MOVWUreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
1064 (MOVHstore [off] {sym} ptr (MOVHreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
1065 (MOVHstore [off] {sym} ptr (MOVHUreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
1066 (MOVHstore [off] {sym} ptr (MOVWreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
1067 (MOVHstore [off] {sym} ptr (MOVWUreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
1068 (MOVWstore [off] {sym} ptr (MOVWreg x) mem) => (MOVWstore [off] {sym} ptr x mem)
1069 (MOVWstore [off] {sym} ptr (MOVWUreg x) mem) => (MOVWstore [off] {sym} ptr x mem)
1070 (MOVBstoreidx ptr idx (MOVBreg x) mem) => (MOVBstoreidx ptr idx x mem)
1071 (MOVBstoreidx ptr idx (MOVBUreg x) mem) => (MOVBstoreidx ptr idx x mem)
1072 (MOVBstoreidx ptr idx (MOVHreg x) mem) => (MOVBstoreidx ptr idx x mem)
1073 (MOVBstoreidx ptr idx (MOVHUreg x) mem) => (MOVBstoreidx ptr idx x mem)
1074 (MOVBstoreidx ptr idx (MOVWreg x) mem) => (MOVBstoreidx ptr idx x mem)
1075 (MOVBstoreidx ptr idx (MOVWUreg x) mem) => (MOVBstoreidx ptr idx x mem)
1076 (MOVHstoreidx ptr idx (MOVHreg x) mem) => (MOVHstoreidx ptr idx x mem)
1077 (MOVHstoreidx ptr idx (MOVHUreg x) mem) => (MOVHstoreidx ptr idx x mem)
1078 (MOVHstoreidx ptr idx (MOVWreg x) mem) => (MOVHstoreidx ptr idx x mem)
1079 (MOVHstoreidx ptr idx (MOVWUreg x) mem) => (MOVHstoreidx ptr idx x mem)
1080 (MOVWstoreidx ptr idx (MOVWreg x) mem) => (MOVWstoreidx ptr idx x mem)
1081 (MOVWstoreidx ptr idx (MOVWUreg x) mem) => (MOVWstoreidx ptr idx x mem)
1082 (MOVHstoreidx2 ptr idx (MOVHreg x) mem) => (MOVHstoreidx2 ptr idx x mem)
1083 (MOVHstoreidx2 ptr idx (MOVHUreg x) mem) => (MOVHstoreidx2 ptr idx x mem)
1084 (MOVHstoreidx2 ptr idx (MOVWreg x) mem) => (MOVHstoreidx2 ptr idx x mem)
1085 (MOVHstoreidx2 ptr idx (MOVWUreg x) mem) => (MOVHstoreidx2 ptr idx x mem)
1086 (MOVWstoreidx4 ptr idx (MOVWreg x) mem) => (MOVWstoreidx4 ptr idx x mem)
1087 (MOVWstoreidx4 ptr idx (MOVWUreg x) mem) => (MOVWstoreidx4 ptr idx x mem)
1088
1089 // if a register move has only 1 use, just use the same register without emitting instruction
1090 // MOVDnop doesn't emit instruction, only for ensuring the type.
1091 (MOVDreg x) && x.Uses == 1 => (MOVDnop x)
1092
1093 // TODO: we should be able to get rid of MOVDnop all together.
1094 // But for now, this is enough to get rid of lots of them.
1095 (MOVDnop (MOVDconst [c])) => (MOVDconst [c])
1096
1097 // fold constant into arithmetic ops
1098 (ADD x (MOVDconst <t> [c])) && !t.IsPtr() => (ADDconst [c] x)
1099 (SUB x (MOVDconst [c])) => (SUBconst [c] x)
1100 (AND x (MOVDconst [c])) => (ANDconst [c] x)
1101 (OR x (MOVDconst [c])) => (ORconst [c] x)
1102 (XOR x (MOVDconst [c])) => (XORconst [c] x)
1103 (TST x (MOVDconst [c])) => (TSTconst [c] x)
1104 (TSTW x (MOVDconst [c])) => (TSTWconst [int32(c)] x)
1105 (CMN x (MOVDconst [c])) => (CMNconst [c] x)
1106 (CMNW x (MOVDconst [c])) => (CMNWconst [int32(c)] x)
1107 (BIC x (MOVDconst [c])) => (ANDconst [^c] x)
1108 (EON x (MOVDconst [c])) => (XORconst [^c] x)
1109 (ORN x (MOVDconst [c])) => (ORconst [^c] x)
1110
1111 (SLL x (MOVDconst [c])) => (SLLconst x [c&63])
1112 (SRL x (MOVDconst [c])) => (SRLconst x [c&63])
1113 (SRA x (MOVDconst [c])) => (SRAconst x [c&63])
1114 (SLL x (ANDconst [63] y)) => (SLL x y)
1115 (SRL x (ANDconst [63] y)) => (SRL x y)
1116 (SRA x (ANDconst [63] y)) => (SRA x y)
1117
1118 (CMP x (MOVDconst [c])) => (CMPconst [c] x)
1119 (CMP (MOVDconst [c]) x) => (InvertFlags (CMPconst [c] x))
1120 (CMPW x (MOVDconst [c])) => (CMPWconst [int32(c)] x)
1121 (CMPW (MOVDconst [c]) x) => (InvertFlags (CMPWconst [int32(c)] x))
1122
1123 (ROR x (MOVDconst [c])) => (RORconst x [c&63])
1124 (RORW x (MOVDconst [c])) => (RORWconst x [c&31])
1125
1126 (ADDSflags x (MOVDconst [c])) => (ADDSconstflags [c] x)
1127
1128 (ADDconst [c] y) && c < 0 => (SUBconst [-c] y)
1129
1130 // Canonicalize the order of arguments to comparisons - helps with CSE.
1131 ((CMP|CMPW) x y) && canonLessThan(x,y) => (InvertFlags ((CMP|CMPW) y x))
1132
1133 // mul-neg => mneg
1134 (NEG (MUL x y)) => (MNEG x y)
1135 (NEG (MULW x y)) && v.Type.Size() <= 4 => (MNEGW x y)
1136 (MUL (NEG x) y) => (MNEG x y)
1137 (MULW (NEG x) y) => (MNEGW x y)
1138
1139 // madd/msub
1140 (ADD a l:(MUL x y)) && l.Uses==1 && clobber(l) => (MADD a x y)
1141 (SUB a l:(MUL x y)) && l.Uses==1 && clobber(l) => (MSUB a x y)
1142 (ADD a l:(MNEG x y)) && l.Uses==1 && clobber(l) => (MSUB a x y)
1143 (SUB a l:(MNEG x y)) && l.Uses==1 && clobber(l) => (MADD a x y)
1144
1145 (ADD a l:(MULW x y)) && v.Type.Size() <= 4 && l.Uses==1 && clobber(l) => (MADDW a x y)
1146 (SUB a l:(MULW x y)) && v.Type.Size() <= 4 && l.Uses==1 && clobber(l) => (MSUBW a x y)
1147 (ADD a l:(MNEGW x y)) && v.Type.Size() <= 4 && l.Uses==1 && clobber(l) => (MSUBW a x y)
1148 (SUB a l:(MNEGW x y)) && v.Type.Size() <= 4 && l.Uses==1 && clobber(l) => (MADDW a x y)
1149
1150 // madd/msub can't take constant arguments, so do a bit of reordering if a non-constant is available.
1151 (ADD a p:(ADDconst [c] m:((MUL|MULW|MNEG|MNEGW) _ _))) && p.Uses==1 && m.Uses==1 => (ADDconst [c] (ADD <v.Type> a m))
1152 (ADD a p:(SUBconst [c] m:((MUL|MULW|MNEG|MNEGW) _ _))) && p.Uses==1 && m.Uses==1 => (SUBconst [c] (ADD <v.Type> a m))
1153 (SUB a p:(ADDconst [c] m:((MUL|MULW|MNEG|MNEGW) _ _))) && p.Uses==1 && m.Uses==1 => (SUBconst [c] (SUB <v.Type> a m))
1154 (SUB a p:(SUBconst [c] m:((MUL|MULW|MNEG|MNEGW) _ _))) && p.Uses==1 && m.Uses==1 => (ADDconst [c] (SUB <v.Type> a m))
1155
1156 // optimize ADCSflags, SBCSflags and friends
1157 (ADCSflags x y (Select1 <types.TypeFlags> (ADDSconstflags [-1] (ADCzerocarry <typ.UInt64> c)))) => (ADCSflags x y c)
1158 (ADCSflags x y (Select1 <types.TypeFlags> (ADDSconstflags [-1] (MOVDconst [0])))) => (ADDSflags x y)
1159 (SBCSflags x y (Select1 <types.TypeFlags> (NEGSflags (NEG <typ.UInt64> (NGCzerocarry <typ.UInt64> bo))))) => (SBCSflags x y bo)
1160 (SBCSflags x y (Select1 <types.TypeFlags> (NEGSflags (MOVDconst [0])))) => (SUBSflags x y)
1161
1162 // mul by constant
1163 (MUL x (MOVDconst [-1])) => (NEG x)
1164 (MUL _ (MOVDconst [0])) => (MOVDconst [0])
1165 (MUL x (MOVDconst [1])) => x
1166 (MUL x (MOVDconst [c])) && isPowerOfTwo(c) => (SLLconst [log64(c)] x)
1167 (MUL x (MOVDconst [c])) && isPowerOfTwo(c-1) && c >= 3 => (ADDshiftLL x x [log64(c-1)])
1168 (MUL x (MOVDconst [c])) && isPowerOfTwo(c+1) && c >= 7 => (ADDshiftLL (NEG <x.Type> x) x [log64(c+1)])
1169 (MUL x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo(c/3) => (SLLconst [log64(c/3)] (ADDshiftLL <x.Type> x x [1]))
1170 (MUL x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo(c/5) => (SLLconst [log64(c/5)] (ADDshiftLL <x.Type> x x [2]))
1171 (MUL x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo(c/7) => (SLLconst [log64(c/7)] (ADDshiftLL <x.Type> (NEG <x.Type> x) x [3]))
1172 (MUL x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo(c/9) => (SLLconst [log64(c/9)] (ADDshiftLL <x.Type> x x [3]))
1173
1174 (MULW x (MOVDconst [c])) && int32(c)==-1 => (MOVWUreg (NEG <x.Type> x))
1175 (MULW _ (MOVDconst [c])) && int32(c)==0 => (MOVDconst [0])
1176 (MULW x (MOVDconst [c])) && int32(c)==1 => (MOVWUreg x)
1177 (MULW x (MOVDconst [c])) && isPowerOfTwo(c) => (MOVWUreg (SLLconst <x.Type> [log64(c)] x))
1178 (MULW x (MOVDconst [c])) && isPowerOfTwo(c-1) && int32(c) >= 3 => (MOVWUreg (ADDshiftLL <x.Type> x x [log64(c-1)]))
1179 (MULW x (MOVDconst [c])) && isPowerOfTwo(c+1) && int32(c) >= 7 => (MOVWUreg (ADDshiftLL <x.Type> (NEG <x.Type> x) x [log64(c+1)]))
1180 (MULW x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo(c/3) && is32Bit(c) => (MOVWUreg (SLLconst <x.Type> [log64(c/3)] (ADDshiftLL <x.Type> x x [1])))
1181 (MULW x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo(c/5) && is32Bit(c) => (MOVWUreg (SLLconst <x.Type> [log64(c/5)] (ADDshiftLL <x.Type> x x [2])))
1182 (MULW x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo(c/7) && is32Bit(c) => (MOVWUreg (SLLconst <x.Type> [log64(c/7)] (ADDshiftLL <x.Type> (NEG <x.Type> x) x [3])))
1183 (MULW x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo(c/9) && is32Bit(c) => (MOVWUreg (SLLconst <x.Type> [log64(c/9)] (ADDshiftLL <x.Type> x x [3])))
1184
1185 // mneg by constant
1186 (MNEG x (MOVDconst [-1])) => x
1187 (MNEG _ (MOVDconst [0])) => (MOVDconst [0])
1188 (MNEG x (MOVDconst [1])) => (NEG x)
1189 (MNEG x (MOVDconst [c])) && isPowerOfTwo(c) => (NEG (SLLconst <x.Type> [log64(c)] x))
1190 (MNEG x (MOVDconst [c])) && isPowerOfTwo(c-1) && c >= 3 => (NEG (ADDshiftLL <x.Type> x x [log64(c-1)]))
1191 (MNEG x (MOVDconst [c])) && isPowerOfTwo(c+1) && c >= 7 => (NEG (ADDshiftLL <x.Type> (NEG <x.Type> x) x [log64(c+1)]))
1192 (MNEG x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo(c/3) => (SLLconst <x.Type> [log64(c/3)] (SUBshiftLL <x.Type> x x [2]))
1193 (MNEG x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo(c/5) => (NEG (SLLconst <x.Type> [log64(c/5)] (ADDshiftLL <x.Type> x x [2])))
1194 (MNEG x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo(c/7) => (SLLconst <x.Type> [log64(c/7)] (SUBshiftLL <x.Type> x x [3]))
1195 (MNEG x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo(c/9) => (NEG (SLLconst <x.Type> [log64(c/9)] (ADDshiftLL <x.Type> x x [3])))
1196
1197
1198 (MNEGW x (MOVDconst [c])) && int32(c)==-1 => (MOVWUreg x)
1199 (MNEGW _ (MOVDconst [c])) && int32(c)==0 => (MOVDconst [0])
1200 (MNEGW x (MOVDconst [c])) && int32(c)==1 => (MOVWUreg (NEG <x.Type> x))
1201 (MNEGW x (MOVDconst [c])) && isPowerOfTwo(c) => (NEG (SLLconst <x.Type> [log64(c)] x))
1202 (MNEGW x (MOVDconst [c])) && isPowerOfTwo(c-1) && int32(c) >= 3 => (MOVWUreg (NEG <x.Type> (ADDshiftLL <x.Type> x x [log64(c-1)])))
1203 (MNEGW x (MOVDconst [c])) && isPowerOfTwo(c+1) && int32(c) >= 7 => (MOVWUreg (NEG <x.Type> (ADDshiftLL <x.Type> (NEG <x.Type> x) x [log64(c+1)])))
1204 (MNEGW x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo(c/3) && is32Bit(c) => (MOVWUreg (SLLconst <x.Type> [log64(c/3)] (SUBshiftLL <x.Type> x x [2])))
1205 (MNEGW x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo(c/5) && is32Bit(c) => (MOVWUreg (NEG <x.Type> (SLLconst <x.Type> [log64(c/5)] (ADDshiftLL <x.Type> x x [2]))))
1206 (MNEGW x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo(c/7) && is32Bit(c) => (MOVWUreg (SLLconst <x.Type> [log64(c/7)] (SUBshiftLL <x.Type> x x [3])))
1207 (MNEGW x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo(c/9) && is32Bit(c) => (MOVWUreg (NEG <x.Type> (SLLconst <x.Type> [log64(c/9)] (ADDshiftLL <x.Type> x x [3]))))
1208
1209
1210 (MADD a x (MOVDconst [-1])) => (SUB a x)
1211 (MADD a _ (MOVDconst [0])) => a
1212 (MADD a x (MOVDconst [1])) => (ADD a x)
1213 (MADD a x (MOVDconst [c])) && isPowerOfTwo(c) => (ADDshiftLL a x [log64(c)])
1214 (MADD a x (MOVDconst [c])) && isPowerOfTwo(c-1) && c>=3 => (ADD a (ADDshiftLL <x.Type> x x [log64(c-1)]))
1215 (MADD a x (MOVDconst [c])) && isPowerOfTwo(c+1) && c>=7 => (SUB a (SUBshiftLL <x.Type> x x [log64(c+1)]))
1216 (MADD a x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo(c/3) => (SUBshiftLL a (SUBshiftLL <x.Type> x x [2]) [log64(c/3)])
1217 (MADD a x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo(c/5) => (ADDshiftLL a (ADDshiftLL <x.Type> x x [2]) [log64(c/5)])
1218 (MADD a x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo(c/7) => (SUBshiftLL a (SUBshiftLL <x.Type> x x [3]) [log64(c/7)])
1219 (MADD a x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo(c/9) => (ADDshiftLL a (ADDshiftLL <x.Type> x x [3]) [log64(c/9)])
1220
1221 (MADD a (MOVDconst [-1]) x) => (SUB a x)
1222 (MADD a (MOVDconst [0]) _) => a
1223 (MADD a (MOVDconst [1]) x) => (ADD a x)
1224 (MADD a (MOVDconst [c]) x) && isPowerOfTwo(c) => (ADDshiftLL a x [log64(c)])
1225 (MADD a (MOVDconst [c]) x) && isPowerOfTwo(c-1) && c>=3 => (ADD a (ADDshiftLL <x.Type> x x [log64(c-1)]))
1226 (MADD a (MOVDconst [c]) x) && isPowerOfTwo(c+1) && c>=7 => (SUB a (SUBshiftLL <x.Type> x x [log64(c+1)]))
1227 (MADD a (MOVDconst [c]) x) && c%3 == 0 && isPowerOfTwo(c/3) => (SUBshiftLL a (SUBshiftLL <x.Type> x x [2]) [log64(c/3)])
1228 (MADD a (MOVDconst [c]) x) && c%5 == 0 && isPowerOfTwo(c/5) => (ADDshiftLL a (ADDshiftLL <x.Type> x x [2]) [log64(c/5)])
1229 (MADD a (MOVDconst [c]) x) && c%7 == 0 && isPowerOfTwo(c/7) => (SUBshiftLL a (SUBshiftLL <x.Type> x x [3]) [log64(c/7)])
1230 (MADD a (MOVDconst [c]) x) && c%9 == 0 && isPowerOfTwo(c/9) => (ADDshiftLL a (ADDshiftLL <x.Type> x x [3]) [log64(c/9)])
1231
1232 (MADDW a x (MOVDconst [c])) && int32(c)==-1 => (MOVWUreg (SUB <a.Type> a x))
1233 (MADDW a _ (MOVDconst [c])) && int32(c)==0 => (MOVWUreg a)
1234 (MADDW a x (MOVDconst [c])) && int32(c)==1 => (MOVWUreg (ADD <a.Type> a x))
1235 (MADDW a x (MOVDconst [c])) && isPowerOfTwo(c) => (MOVWUreg (ADDshiftLL <a.Type> a x [log64(c)]))
1236 (MADDW a x (MOVDconst [c])) && isPowerOfTwo(c-1) && int32(c)>=3 => (MOVWUreg (ADD <a.Type> a (ADDshiftLL <x.Type> x x [log64(c-1)])))
1237 (MADDW a x (MOVDconst [c])) && isPowerOfTwo(c+1) && int32(c)>=7 => (MOVWUreg (SUB <a.Type> a (SUBshiftLL <x.Type> x x [log64(c+1)])))
1238 (MADDW a x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo(c/3) && is32Bit(c) => (MOVWUreg (SUBshiftLL <a.Type> a (SUBshiftLL <x.Type> x x [2]) [log64(c/3)]))
1239 (MADDW a x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo(c/5) && is32Bit(c) => (MOVWUreg (ADDshiftLL <a.Type> a (ADDshiftLL <x.Type> x x [2]) [log64(c/5)]))
1240 (MADDW a x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo(c/7) && is32Bit(c) => (MOVWUreg (SUBshiftLL <a.Type> a (SUBshiftLL <x.Type> x x [3]) [log64(c/7)]))
1241 (MADDW a x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo(c/9) && is32Bit(c) => (MOVWUreg (ADDshiftLL <a.Type> a (ADDshiftLL <x.Type> x x [3]) [log64(c/9)]))
1242
1243 (MADDW a (MOVDconst [c]) x) && int32(c)==-1 => (MOVWUreg (SUB <a.Type> a x))
1244 (MADDW a (MOVDconst [c]) _) && int32(c)==0 => (MOVWUreg a)
1245 (MADDW a (MOVDconst [c]) x) && int32(c)==1 => (MOVWUreg (ADD <a.Type> a x))
1246 (MADDW a (MOVDconst [c]) x) && isPowerOfTwo(c) => (MOVWUreg (ADDshiftLL <a.Type> a x [log64(c)]))
1247 (MADDW a (MOVDconst [c]) x) && isPowerOfTwo(c-1) && int32(c)>=3 => (MOVWUreg (ADD <a.Type> a (ADDshiftLL <x.Type> x x [log64(c-1)])))
1248 (MADDW a (MOVDconst [c]) x) && isPowerOfTwo(c+1) && int32(c)>=7 => (MOVWUreg (SUB <a.Type> a (SUBshiftLL <x.Type> x x [log64(c+1)])))
1249 (MADDW a (MOVDconst [c]) x) && c%3 == 0 && isPowerOfTwo(c/3) && is32Bit(c) => (MOVWUreg (SUBshiftLL <a.Type> a (SUBshiftLL <x.Type> x x [2]) [log64(c/3)]))
1250 (MADDW a (MOVDconst [c]) x) && c%5 == 0 && isPowerOfTwo(c/5) && is32Bit(c) => (MOVWUreg (ADDshiftLL <a.Type> a (ADDshiftLL <x.Type> x x [2]) [log64(c/5)]))
1251 (MADDW a (MOVDconst [c]) x) && c%7 == 0 && isPowerOfTwo(c/7) && is32Bit(c) => (MOVWUreg (SUBshiftLL <a.Type> a (SUBshiftLL <x.Type> x x [3]) [log64(c/7)]))
1252 (MADDW a (MOVDconst [c]) x) && c%9 == 0 && isPowerOfTwo(c/9) && is32Bit(c) => (MOVWUreg (ADDshiftLL <a.Type> a (ADDshiftLL <x.Type> x x [3]) [log64(c/9)]))
1253
1254 (MSUB a x (MOVDconst [-1])) => (ADD a x)
1255 (MSUB a _ (MOVDconst [0])) => a
1256 (MSUB a x (MOVDconst [1])) => (SUB a x)
1257 (MSUB a x (MOVDconst [c])) && isPowerOfTwo(c) => (SUBshiftLL a x [log64(c)])
1258 (MSUB a x (MOVDconst [c])) && isPowerOfTwo(c-1) && c>=3 => (SUB a (ADDshiftLL <x.Type> x x [log64(c-1)]))
1259 (MSUB a x (MOVDconst [c])) && isPowerOfTwo(c+1) && c>=7 => (ADD a (SUBshiftLL <x.Type> x x [log64(c+1)]))
1260 (MSUB a x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo(c/3) => (ADDshiftLL a (SUBshiftLL <x.Type> x x [2]) [log64(c/3)])
1261 (MSUB a x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo(c/5) => (SUBshiftLL a (ADDshiftLL <x.Type> x x [2]) [log64(c/5)])
1262 (MSUB a x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo(c/7) => (ADDshiftLL a (SUBshiftLL <x.Type> x x [3]) [log64(c/7)])
1263 (MSUB a x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo(c/9) => (SUBshiftLL a (ADDshiftLL <x.Type> x x [3]) [log64(c/9)])
1264
1265 (MSUB a (MOVDconst [-1]) x) => (ADD a x)
1266 (MSUB a (MOVDconst [0]) _) => a
1267 (MSUB a (MOVDconst [1]) x) => (SUB a x)
1268 (MSUB a (MOVDconst [c]) x) && isPowerOfTwo(c) => (SUBshiftLL a x [log64(c)])
1269 (MSUB a (MOVDconst [c]) x) && isPowerOfTwo(c-1) && c>=3 => (SUB a (ADDshiftLL <x.Type> x x [log64(c-1)]))
1270 (MSUB a (MOVDconst [c]) x) && isPowerOfTwo(c+1) && c>=7 => (ADD a (SUBshiftLL <x.Type> x x [log64(c+1)]))
1271 (MSUB a (MOVDconst [c]) x) && c%3 == 0 && isPowerOfTwo(c/3) => (ADDshiftLL a (SUBshiftLL <x.Type> x x [2]) [log64(c/3)])
1272 (MSUB a (MOVDconst [c]) x) && c%5 == 0 && isPowerOfTwo(c/5) => (SUBshiftLL a (ADDshiftLL <x.Type> x x [2]) [log64(c/5)])
1273 (MSUB a (MOVDconst [c]) x) && c%7 == 0 && isPowerOfTwo(c/7) => (ADDshiftLL a (SUBshiftLL <x.Type> x x [3]) [log64(c/7)])
1274 (MSUB a (MOVDconst [c]) x) && c%9 == 0 && isPowerOfTwo(c/9) => (SUBshiftLL a (ADDshiftLL <x.Type> x x [3]) [log64(c/9)])
1275
1276 (MSUBW a x (MOVDconst [c])) && int32(c)==-1 => (MOVWUreg (ADD <a.Type> a x))
1277 (MSUBW a _ (MOVDconst [c])) && int32(c)==0 => (MOVWUreg a)
1278 (MSUBW a x (MOVDconst [c])) && int32(c)==1 => (MOVWUreg (SUB <a.Type> a x))
1279 (MSUBW a x (MOVDconst [c])) && isPowerOfTwo(c) => (MOVWUreg (SUBshiftLL <a.Type> a x [log64(c)]))
1280 (MSUBW a x (MOVDconst [c])) && isPowerOfTwo(c-1) && int32(c)>=3 => (MOVWUreg (SUB <a.Type> a (ADDshiftLL <x.Type> x x [log64(c-1)])))
1281 (MSUBW a x (MOVDconst [c])) && isPowerOfTwo(c+1) && int32(c)>=7 => (MOVWUreg (ADD <a.Type> a (SUBshiftLL <x.Type> x x [log64(c+1)])))
1282 (MSUBW a x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo(c/3) && is32Bit(c) => (MOVWUreg (ADDshiftLL <a.Type> a (SUBshiftLL <x.Type> x x [2]) [log64(c/3)]))
1283 (MSUBW a x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo(c/5) && is32Bit(c) => (MOVWUreg (SUBshiftLL <a.Type> a (ADDshiftLL <x.Type> x x [2]) [log64(c/5)]))
1284 (MSUBW a x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo(c/7) && is32Bit(c) => (MOVWUreg (ADDshiftLL <a.Type> a (SUBshiftLL <x.Type> x x [3]) [log64(c/7)]))
1285 (MSUBW a x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo(c/9) && is32Bit(c) => (MOVWUreg (SUBshiftLL <a.Type> a (ADDshiftLL <x.Type> x x [3]) [log64(c/9)]))
1286
1287 (MSUBW a (MOVDconst [c]) x) && int32(c)==-1 => (MOVWUreg (ADD <a.Type> a x))
1288 (MSUBW a (MOVDconst [c]) _) && int32(c)==0 => (MOVWUreg a)
1289 (MSUBW a (MOVDconst [c]) x) && int32(c)==1 => (MOVWUreg (SUB <a.Type> a x))
1290 (MSUBW a (MOVDconst [c]) x) && isPowerOfTwo(c) => (MOVWUreg (SUBshiftLL <a.Type> a x [log64(c)]))
1291 (MSUBW a (MOVDconst [c]) x) && isPowerOfTwo(c-1) && int32(c)>=3 => (MOVWUreg (SUB <a.Type> a (ADDshiftLL <x.Type> x x [log64(c-1)])))
1292 (MSUBW a (MOVDconst [c]) x) && isPowerOfTwo(c+1) && int32(c)>=7 => (MOVWUreg (ADD <a.Type> a (SUBshiftLL <x.Type> x x [log64(c+1)])))
1293 (MSUBW a (MOVDconst [c]) x) && c%3 == 0 && isPowerOfTwo(c/3) && is32Bit(c) => (MOVWUreg (ADDshiftLL <a.Type> a (SUBshiftLL <x.Type> x x [2]) [log64(c/3)]))
1294 (MSUBW a (MOVDconst [c]) x) && c%5 == 0 && isPowerOfTwo(c/5) && is32Bit(c) => (MOVWUreg (SUBshiftLL <a.Type> a (ADDshiftLL <x.Type> x x [2]) [log64(c/5)]))
1295 (MSUBW a (MOVDconst [c]) x) && c%7 == 0 && isPowerOfTwo(c/7) && is32Bit(c) => (MOVWUreg (ADDshiftLL <a.Type> a (SUBshiftLL <x.Type> x x [3]) [log64(c/7)]))
1296 (MSUBW a (MOVDconst [c]) x) && c%9 == 0 && isPowerOfTwo(c/9) && is32Bit(c) => (MOVWUreg (SUBshiftLL <a.Type> a (ADDshiftLL <x.Type> x x [3]) [log64(c/9)]))
1297
1298 // div by constant
1299 (UDIV x (MOVDconst [1])) => x
1300 (UDIV x (MOVDconst [c])) && isPowerOfTwo(c) => (SRLconst [log64(c)] x)
1301 (UDIVW x (MOVDconst [c])) && uint32(c)==1 => (MOVWUreg x)
1302 (UDIVW x (MOVDconst [c])) && isPowerOfTwo(c) && is32Bit(c) => (SRLconst [log64(c)] (MOVWUreg <v.Type> x))
1303 (UMOD _ (MOVDconst [1])) => (MOVDconst [0])
1304 (UMOD x (MOVDconst [c])) && isPowerOfTwo(c) => (ANDconst [c-1] x)
1305 (UMODW _ (MOVDconst [c])) && uint32(c)==1 => (MOVDconst [0])
1306 (UMODW x (MOVDconst [c])) && isPowerOfTwo(c) && is32Bit(c) => (ANDconst [c-1] x)
1307
1308 // generic simplifications
1309 (ADD x (NEG y)) => (SUB x y)
1310 (SUB x x) => (MOVDconst [0])
1311 (AND x x) => x
1312 (OR x x) => x
1313 (XOR x x) => (MOVDconst [0])
1314 (BIC x x) => (MOVDconst [0])
1315 (EON x x) => (MOVDconst [-1])
1316 (ORN x x) => (MOVDconst [-1])
1317 (AND x (MVN y)) => (BIC x y)
1318 (XOR x (MVN y)) => (EON x y)
1319 (OR x (MVN y)) => (ORN x y)
1320 (MVN (XOR x y)) => (EON x y)
1321 (NEG (NEG x)) => x
1322
1323 (CSEL [cc] (MOVDconst [-1]) (MOVDconst [0]) flag) => (CSETM [cc] flag)
1324 (CSEL [cc] (MOVDconst [0]) (MOVDconst [-1]) flag) => (CSETM [arm64Negate(cc)] flag)
1325 (CSEL [cc] x (MOVDconst [0]) flag) => (CSEL0 [cc] x flag)
1326 (CSEL [cc] (MOVDconst [0]) y flag) => (CSEL0 [arm64Negate(cc)] y flag)
1327 (CSEL [cc] x (ADDconst [1] a) flag) => (CSINC [cc] x a flag)
1328 (CSEL [cc] (ADDconst [1] a) x flag) => (CSINC [arm64Negate(cc)] x a flag)
1329 (CSEL [cc] x (MVN a) flag) => (CSINV [cc] x a flag)
1330 (CSEL [cc] (MVN a) x flag) => (CSINV [arm64Negate(cc)] x a flag)
1331 (CSEL [cc] x (NEG a) flag) => (CSNEG [cc] x a flag)
1332 (CSEL [cc] (NEG a) x flag) => (CSNEG [arm64Negate(cc)] x a flag)
1333
1334 (SUB x (SUB y z)) => (SUB (ADD <v.Type> x z) y)
1335 (SUB (SUB x y) z) => (SUB x (ADD <y.Type> y z))
1336
1337 // remove redundant *const ops
1338 (ADDconst [0] x) => x
1339 (SUBconst [0] x) => x
1340 (ANDconst [0] _) => (MOVDconst [0])
1341 (ANDconst [-1] x) => x
1342 (ORconst [0] x) => x
1343 (ORconst [-1] _) => (MOVDconst [-1])
1344 (XORconst [0] x) => x
1345 (XORconst [-1] x) => (MVN x)
1346
1347 // generic constant folding
1348 (ADDconst [c] (MOVDconst [d])) => (MOVDconst [c+d])
1349 (ADDconst [c] (ADDconst [d] x)) => (ADDconst [c+d] x)
1350 (ADDconst [c] (SUBconst [d] x)) => (ADDconst [c-d] x)
1351 (SUBconst [c] (MOVDconst [d])) => (MOVDconst [d-c])
1352 (SUBconst [c] (SUBconst [d] x)) => (ADDconst [-c-d] x)
1353 (SUBconst [c] (ADDconst [d] x)) => (ADDconst [-c+d] x)
1354 (SLLconst [c] (MOVDconst [d])) => (MOVDconst [d<<uint64(c)])
1355 (SRLconst [c] (MOVDconst [d])) => (MOVDconst [int64(uint64(d)>>uint64(c))])
1356 (SRAconst [c] (MOVDconst [d])) => (MOVDconst [d>>uint64(c)])
1357 (MUL (MOVDconst [c]) (MOVDconst [d])) => (MOVDconst [c*d])
1358 (MNEG (MOVDconst [c]) (MOVDconst [d])) => (MOVDconst [-c*d])
1359 (MULW (MOVDconst [c]) (MOVDconst [d])) => (MOVDconst [int64(uint32(c*d))])
1360 (MNEGW (MOVDconst [c]) (MOVDconst [d])) => (MOVDconst [int64(uint32(-c*d))])
1361 (MADD (MOVDconst [c]) x y) => (ADDconst [c] (MUL <x.Type> x y))
1362 (MSUB (MOVDconst [c]) x y) => (ADDconst [c] (MNEG <x.Type> x y))
1363 (MADD a (MOVDconst [c]) (MOVDconst [d])) => (ADDconst [c*d] a)
1364 (MSUB a (MOVDconst [c]) (MOVDconst [d])) => (SUBconst [c*d] a)
1365 (MADDW (MOVDconst [c]) x y) => (MOVWUreg (ADDconst <x.Type> [c] (MULW <x.Type> x y)))
1366 (MSUBW (MOVDconst [c]) x y) => (MOVWUreg (ADDconst <x.Type> [c] (MNEGW <x.Type> x y)))
1367 (MADDW a (MOVDconst [c]) (MOVDconst [d])) => (MOVWUreg (ADDconst <a.Type> [c*d] a))
1368 (MSUBW a (MOVDconst [c]) (MOVDconst [d])) => (MOVWUreg (SUBconst <a.Type> [c*d] a))
1369 (DIV (MOVDconst [c]) (MOVDconst [d])) && d != 0 => (MOVDconst [c/d])
1370 (UDIV (MOVDconst [c]) (MOVDconst [d])) && d != 0 => (MOVDconst [int64(uint64(c)/uint64(d))])
1371 (DIVW (MOVDconst [c]) (MOVDconst [d])) && d != 0 => (MOVDconst [int64(uint32(int32(c)/int32(d)))])
1372 (UDIVW (MOVDconst [c]) (MOVDconst [d])) && d != 0 => (MOVDconst [int64(uint32(c)/uint32(d))])
1373 (MOD (MOVDconst [c]) (MOVDconst [d])) && d != 0 => (MOVDconst [c%d])
1374 (UMOD (MOVDconst [c]) (MOVDconst [d])) && d != 0 => (MOVDconst [int64(uint64(c)%uint64(d))])
1375 (MODW (MOVDconst [c]) (MOVDconst [d])) && d != 0 => (MOVDconst [int64(uint32(int32(c)%int32(d)))])
1376 (UMODW (MOVDconst [c]) (MOVDconst [d])) && d != 0 => (MOVDconst [int64(uint32(c)%uint32(d))])
1377 (ANDconst [c] (MOVDconst [d])) => (MOVDconst [c&d])
1378 (ANDconst [c] (ANDconst [d] x)) => (ANDconst [c&d] x)
1379 (ANDconst [c] (MOVWUreg x)) => (ANDconst [c&(1<<32-1)] x)
1380 (ANDconst [c] (MOVHUreg x)) => (ANDconst [c&(1<<16-1)] x)
1381 (ANDconst [c] (MOVBUreg x)) => (ANDconst [c&(1<<8-1)] x)
1382 (MOVWUreg (ANDconst [c] x)) => (ANDconst [c&(1<<32-1)] x)
1383 (MOVHUreg (ANDconst [c] x)) => (ANDconst [c&(1<<16-1)] x)
1384 (MOVBUreg (ANDconst [c] x)) => (ANDconst [c&(1<<8-1)] x)
1385 (ORconst [c] (MOVDconst [d])) => (MOVDconst [c|d])
1386 (ORconst [c] (ORconst [d] x)) => (ORconst [c|d] x)
1387 (XORconst [c] (MOVDconst [d])) => (MOVDconst [c^d])
1388 (XORconst [c] (XORconst [d] x)) => (XORconst [c^d] x)
1389 (MVN (MOVDconst [c])) => (MOVDconst [^c])
1390 (NEG (MOVDconst [c])) => (MOVDconst [-c])
1391 (MOVBreg (MOVDconst [c])) => (MOVDconst [int64(int8(c))])
1392 (MOVBUreg (MOVDconst [c])) => (MOVDconst [int64(uint8(c))])
1393 (MOVHreg (MOVDconst [c])) => (MOVDconst [int64(int16(c))])
1394 (MOVHUreg (MOVDconst [c])) => (MOVDconst [int64(uint16(c))])
1395 (MOVWreg (MOVDconst [c])) => (MOVDconst [int64(int32(c))])
1396 (MOVWUreg (MOVDconst [c])) => (MOVDconst [int64(uint32(c))])
1397 (MOVDreg (MOVDconst [c])) => (MOVDconst [c])
1398
1399 // constant comparisons
1400 (CMPconst (MOVDconst [x]) [y]) => (FlagConstant [subFlags64(x,y)])
1401 (CMPWconst (MOVDconst [x]) [y]) => (FlagConstant [subFlags32(int32(x),y)])
1402 (TSTconst (MOVDconst [x]) [y]) => (FlagConstant [logicFlags64(x&y)])
1403 (TSTWconst (MOVDconst [x]) [y]) => (FlagConstant [logicFlags32(int32(x)&y)])
1404 (CMNconst (MOVDconst [x]) [y]) => (FlagConstant [addFlags64(x,y)])
1405 (CMNWconst (MOVDconst [x]) [y]) => (FlagConstant [addFlags32(int32(x),y)])
1406
1407 // other known comparisons
1408 (CMPconst (MOVBUreg _) [c]) && 0xff < c => (FlagConstant [subFlags64(0,1)])
1409 (CMPconst (MOVHUreg _) [c]) && 0xffff < c => (FlagConstant [subFlags64(0,1)])
1410 (CMPconst (MOVWUreg _) [c]) && 0xffffffff < c => (FlagConstant [subFlags64(0,1)])
1411 (CMPconst (ANDconst _ [m]) [n]) && 0 <= m && m < n => (FlagConstant [subFlags64(0,1)])
1412 (CMPconst (SRLconst _ [c]) [n]) && 0 <= n && 0 < c && c <= 63 && (1<<uint64(64-c)) <= uint64(n) => (FlagConstant [subFlags64(0,1)])
1413 (CMPWconst (MOVBUreg _) [c]) && 0xff < c => (FlagConstant [subFlags64(0,1)])
1414 (CMPWconst (MOVHUreg _) [c]) && 0xffff < c => (FlagConstant [subFlags64(0,1)])
1415
1416 // absorb flag constants into branches
1417 (EQ (FlagConstant [fc]) yes no) && fc.eq() => (First yes no)
1418 (EQ (FlagConstant [fc]) yes no) && !fc.eq() => (First no yes)
1419
1420 (NE (FlagConstant [fc]) yes no) && fc.ne() => (First yes no)
1421 (NE (FlagConstant [fc]) yes no) && !fc.ne() => (First no yes)
1422
1423 (LT (FlagConstant [fc]) yes no) && fc.lt() => (First yes no)
1424 (LT (FlagConstant [fc]) yes no) && !fc.lt() => (First no yes)
1425
1426 (LE (FlagConstant [fc]) yes no) && fc.le() => (First yes no)
1427 (LE (FlagConstant [fc]) yes no) && !fc.le() => (First no yes)
1428
1429 (GT (FlagConstant [fc]) yes no) && fc.gt() => (First yes no)
1430 (GT (FlagConstant [fc]) yes no) && !fc.gt() => (First no yes)
1431
1432 (GE (FlagConstant [fc]) yes no) && fc.ge() => (First yes no)
1433 (GE (FlagConstant [fc]) yes no) && !fc.ge() => (First no yes)
1434
1435 (ULT (FlagConstant [fc]) yes no) && fc.ult() => (First yes no)
1436 (ULT (FlagConstant [fc]) yes no) && !fc.ult() => (First no yes)
1437
1438 (ULE (FlagConstant [fc]) yes no) && fc.ule() => (First yes no)
1439 (ULE (FlagConstant [fc]) yes no) && !fc.ule() => (First no yes)
1440
1441 (UGT (FlagConstant [fc]) yes no) && fc.ugt() => (First yes no)
1442 (UGT (FlagConstant [fc]) yes no) && !fc.ugt() => (First no yes)
1443
1444 (UGE (FlagConstant [fc]) yes no) && fc.uge() => (First yes no)
1445 (UGE (FlagConstant [fc]) yes no) && !fc.uge() => (First no yes)
1446
1447 (LTnoov (FlagConstant [fc]) yes no) && fc.ltNoov() => (First yes no)
1448 (LTnoov (FlagConstant [fc]) yes no) && !fc.ltNoov() => (First no yes)
1449
1450 (LEnoov (FlagConstant [fc]) yes no) && fc.leNoov() => (First yes no)
1451 (LEnoov (FlagConstant [fc]) yes no) && !fc.leNoov() => (First no yes)
1452
1453 (GTnoov (FlagConstant [fc]) yes no) && fc.gtNoov() => (First yes no)
1454 (GTnoov (FlagConstant [fc]) yes no) && !fc.gtNoov() => (First no yes)
1455
1456 (GEnoov (FlagConstant [fc]) yes no) && fc.geNoov() => (First yes no)
1457 (GEnoov (FlagConstant [fc]) yes no) && !fc.geNoov() => (First no yes)
1458
1459 (Z (MOVDconst [0]) yes no) => (First yes no)
1460 (Z (MOVDconst [c]) yes no) && c != 0 => (First no yes)
1461 (NZ (MOVDconst [0]) yes no) => (First no yes)
1462 (NZ (MOVDconst [c]) yes no) && c != 0 => (First yes no)
1463 (ZW (MOVDconst [c]) yes no) && int32(c) == 0 => (First yes no)
1464 (ZW (MOVDconst [c]) yes no) && int32(c) != 0 => (First no yes)
1465 (NZW (MOVDconst [c]) yes no) && int32(c) == 0 => (First no yes)
1466 (NZW (MOVDconst [c]) yes no) && int32(c) != 0 => (First yes no)
1467
1468 // absorb InvertFlags into branches
1469 (LT (InvertFlags cmp) yes no) => (GT cmp yes no)
1470 (GT (InvertFlags cmp) yes no) => (LT cmp yes no)
1471 (LE (InvertFlags cmp) yes no) => (GE cmp yes no)
1472 (GE (InvertFlags cmp) yes no) => (LE cmp yes no)
1473 (ULT (InvertFlags cmp) yes no) => (UGT cmp yes no)
1474 (UGT (InvertFlags cmp) yes no) => (ULT cmp yes no)
1475 (ULE (InvertFlags cmp) yes no) => (UGE cmp yes no)
1476 (UGE (InvertFlags cmp) yes no) => (ULE cmp yes no)
1477 (EQ (InvertFlags cmp) yes no) => (EQ cmp yes no)
1478 (NE (InvertFlags cmp) yes no) => (NE cmp yes no)
1479 (FLT (InvertFlags cmp) yes no) => (FGT cmp yes no)
1480 (FGT (InvertFlags cmp) yes no) => (FLT cmp yes no)
1481 (FLE (InvertFlags cmp) yes no) => (FGE cmp yes no)
1482 (FGE (InvertFlags cmp) yes no) => (FLE cmp yes no)
1483 (LTnoov (InvertFlags cmp) yes no) => (GTnoov cmp yes no)
1484 (GEnoov (InvertFlags cmp) yes no) => (LEnoov cmp yes no)
1485 (LEnoov (InvertFlags cmp) yes no) => (GEnoov cmp yes no)
1486 (GTnoov (InvertFlags cmp) yes no) => (LTnoov cmp yes no)
1487
1488 // absorb InvertFlags into conditional instructions
1489 (CSEL [cc] x y (InvertFlags cmp)) => (CSEL [arm64Invert(cc)] x y cmp)
1490 (CSEL0 [cc] x (InvertFlags cmp)) => (CSEL0 [arm64Invert(cc)] x cmp)
1491 (CSETM [cc] (InvertFlags cmp)) => (CSETM [arm64Invert(cc)] cmp)
1492 (CSINC [cc] x y (InvertFlags cmp)) => (CSINC [arm64Invert(cc)] x y cmp)
1493 (CSINV [cc] x y (InvertFlags cmp)) => (CSINV [arm64Invert(cc)] x y cmp)
1494 (CSNEG [cc] x y (InvertFlags cmp)) => (CSNEG [arm64Invert(cc)] x y cmp)
1495
1496 // absorb flag constants into boolean values
1497 (Equal (FlagConstant [fc])) => (MOVDconst [b2i(fc.eq())])
1498 (NotEqual (FlagConstant [fc])) => (MOVDconst [b2i(fc.ne())])
1499 (LessThan (FlagConstant [fc])) => (MOVDconst [b2i(fc.lt())])
1500 (LessThanU (FlagConstant [fc])) => (MOVDconst [b2i(fc.ult())])
1501 (LessEqual (FlagConstant [fc])) => (MOVDconst [b2i(fc.le())])
1502 (LessEqualU (FlagConstant [fc])) => (MOVDconst [b2i(fc.ule())])
1503 (GreaterThan (FlagConstant [fc])) => (MOVDconst [b2i(fc.gt())])
1504 (GreaterThanU (FlagConstant [fc])) => (MOVDconst [b2i(fc.ugt())])
1505 (GreaterEqual (FlagConstant [fc])) => (MOVDconst [b2i(fc.ge())])
1506 (GreaterEqualU (FlagConstant [fc])) => (MOVDconst [b2i(fc.uge())])
1507
1508 // absorb InvertFlags into boolean values
1509 (Equal (InvertFlags x)) => (Equal x)
1510 (NotEqual (InvertFlags x)) => (NotEqual x)
1511 (LessThan (InvertFlags x)) => (GreaterThan x)
1512 (LessThanU (InvertFlags x)) => (GreaterThanU x)
1513 (GreaterThan (InvertFlags x)) => (LessThan x)
1514 (GreaterThanU (InvertFlags x)) => (LessThanU x)
1515 (LessEqual (InvertFlags x)) => (GreaterEqual x)
1516 (LessEqualU (InvertFlags x)) => (GreaterEqualU x)
1517 (GreaterEqual (InvertFlags x)) => (LessEqual x)
1518 (GreaterEqualU (InvertFlags x)) => (LessEqualU x)
1519 (LessThanF (InvertFlags x)) => (GreaterThanF x)
1520 (LessEqualF (InvertFlags x)) => (GreaterEqualF x)
1521 (GreaterThanF (InvertFlags x)) => (LessThanF x)
1522 (GreaterEqualF (InvertFlags x)) => (LessEqualF x)
1523 (LessThanNoov (InvertFlags x)) => (CSEL0 [OpARM64NotEqual] (GreaterEqualNoov <typ.Bool> x) x)
1524 (GreaterEqualNoov (InvertFlags x)) => (CSINC [OpARM64NotEqual] (LessThanNoov <typ.Bool> x) (MOVDconst [0]) x)
1525
1526 // Don't bother extending if we're not using the higher bits.
1527 (MOV(B|BU)reg x) && v.Type.Size() <= 1 => x
1528 (MOV(H|HU)reg x) && v.Type.Size() <= 2 => x
1529 (MOV(W|WU)reg x) && v.Type.Size() <= 4 => x
1530
1531 // omit sign extension
1532 (MOVWreg <t> (ANDconst x [c])) && uint64(c) & uint64(0xffffffff80000000) == 0 => (ANDconst <t> x [c])
1533 (MOVHreg <t> (ANDconst x [c])) && uint64(c) & uint64(0xffffffffffff8000) == 0 => (ANDconst <t> x [c])
1534 (MOVBreg <t> (ANDconst x [c])) && uint64(c) & uint64(0xffffffffffffff80) == 0 => (ANDconst <t> x [c])
1535
1536 // absorb flag constants into conditional instructions
1537 (CSEL [cc] x _ flag) && ccARM64Eval(cc, flag) > 0 => x
1538 (CSEL [cc] _ y flag) && ccARM64Eval(cc, flag) < 0 => y
1539 (CSEL0 [cc] x flag) && ccARM64Eval(cc, flag) > 0 => x
1540 (CSEL0 [cc] _ flag) && ccARM64Eval(cc, flag) < 0 => (MOVDconst [0])
1541 (CSNEG [cc] x _ flag) && ccARM64Eval(cc, flag) > 0 => x
1542 (CSNEG [cc] _ y flag) && ccARM64Eval(cc, flag) < 0 => (NEG y)
1543 (CSINV [cc] x _ flag) && ccARM64Eval(cc, flag) > 0 => x
1544 (CSINV [cc] _ y flag) && ccARM64Eval(cc, flag) < 0 => (Not y)
1545 (CSINC [cc] x _ flag) && ccARM64Eval(cc, flag) > 0 => x
1546 (CSINC [cc] _ y flag) && ccARM64Eval(cc, flag) < 0 => (ADDconst [1] y)
1547 (CSETM [cc] flag) && ccARM64Eval(cc, flag) > 0 => (MOVDconst [-1])
1548 (CSETM [cc] flag) && ccARM64Eval(cc, flag) < 0 => (MOVDconst [0])
1549
1550 // absorb flags back into boolean CSEL
1551 (CSEL [cc] x y (CMPWconst [0] boolval)) && cc == OpARM64NotEqual && flagArg(boolval) != nil =>
1552 (CSEL [boolval.Op] x y flagArg(boolval))
1553 (CSEL [cc] x y (CMPWconst [0] boolval)) && cc == OpARM64Equal && flagArg(boolval) != nil =>
1554 (CSEL [arm64Negate(boolval.Op)] x y flagArg(boolval))
1555 (CSEL0 [cc] x (CMPWconst [0] boolval)) && cc == OpARM64NotEqual && flagArg(boolval) != nil =>
1556 (CSEL0 [boolval.Op] x flagArg(boolval))
1557 (CSEL0 [cc] x (CMPWconst [0] boolval)) && cc == OpARM64Equal && flagArg(boolval) != nil =>
1558 (CSEL0 [arm64Negate(boolval.Op)] x flagArg(boolval))
1559
1560 // absorb shifts into ops
1561 (NEG x:(SLLconst [c] y)) && clobberIfDead(x) => (NEGshiftLL [c] y)
1562 (NEG x:(SRLconst [c] y)) && clobberIfDead(x) => (NEGshiftRL [c] y)
1563 (NEG x:(SRAconst [c] y)) && clobberIfDead(x) => (NEGshiftRA [c] y)
1564 (MVN x:(SLLconst [c] y)) && clobberIfDead(x) => (MVNshiftLL [c] y)
1565 (MVN x:(SRLconst [c] y)) && clobberIfDead(x) => (MVNshiftRL [c] y)
1566 (MVN x:(SRAconst [c] y)) && clobberIfDead(x) => (MVNshiftRA [c] y)
1567 (MVN x:(RORconst [c] y)) && clobberIfDead(x) => (MVNshiftRO [c] y)
1568 (ADD x0 x1:(SLLconst [c] y)) && clobberIfDead(x1) => (ADDshiftLL x0 y [c])
1569 (ADD x0 x1:(SRLconst [c] y)) && clobberIfDead(x1) => (ADDshiftRL x0 y [c])
1570 (ADD x0 x1:(SRAconst [c] y)) && clobberIfDead(x1) => (ADDshiftRA x0 y [c])
1571 (SUB x0 x1:(SLLconst [c] y)) && clobberIfDead(x1) => (SUBshiftLL x0 y [c])
1572 (SUB x0 x1:(SRLconst [c] y)) && clobberIfDead(x1) => (SUBshiftRL x0 y [c])
1573 (SUB x0 x1:(SRAconst [c] y)) && clobberIfDead(x1) => (SUBshiftRA x0 y [c])
1574 (AND x0 x1:(SLLconst [c] y)) && clobberIfDead(x1) => (ANDshiftLL x0 y [c])
1575 (AND x0 x1:(SRLconst [c] y)) && clobberIfDead(x1) => (ANDshiftRL x0 y [c])
1576 (AND x0 x1:(SRAconst [c] y)) && clobberIfDead(x1) => (ANDshiftRA x0 y [c])
1577 (AND x0 x1:(RORconst [c] y)) && clobberIfDead(x1) => (ANDshiftRO x0 y [c])
1578 (OR x0 x1:(SLLconst [c] y)) && clobberIfDead(x1) => (ORshiftLL x0 y [c]) // useful for combined load
1579 (OR x0 x1:(SRLconst [c] y)) && clobberIfDead(x1) => (ORshiftRL x0 y [c])
1580 (OR x0 x1:(SRAconst [c] y)) && clobberIfDead(x1) => (ORshiftRA x0 y [c])
1581 (OR x0 x1:(RORconst [c] y)) && clobberIfDead(x1) => (ORshiftRO x0 y [c])
1582 (XOR x0 x1:(SLLconst [c] y)) && clobberIfDead(x1) => (XORshiftLL x0 y [c])
1583 (XOR x0 x1:(SRLconst [c] y)) && clobberIfDead(x1) => (XORshiftRL x0 y [c])
1584 (XOR x0 x1:(SRAconst [c] y)) && clobberIfDead(x1) => (XORshiftRA x0 y [c])
1585 (XOR x0 x1:(RORconst [c] y)) && clobberIfDead(x1) => (XORshiftRO x0 y [c])
1586 (BIC x0 x1:(SLLconst [c] y)) && clobberIfDead(x1) => (BICshiftLL x0 y [c])
1587 (BIC x0 x1:(SRLconst [c] y)) && clobberIfDead(x1) => (BICshiftRL x0 y [c])
1588 (BIC x0 x1:(SRAconst [c] y)) && clobberIfDead(x1) => (BICshiftRA x0 y [c])
1589 (BIC x0 x1:(RORconst [c] y)) && clobberIfDead(x1) => (BICshiftRO x0 y [c])
1590 (ORN x0 x1:(SLLconst [c] y)) && clobberIfDead(x1) => (ORNshiftLL x0 y [c])
1591 (ORN x0 x1:(SRLconst [c] y)) && clobberIfDead(x1) => (ORNshiftRL x0 y [c])
1592 (ORN x0 x1:(SRAconst [c] y)) && clobberIfDead(x1) => (ORNshiftRA x0 y [c])
1593 (ORN x0 x1:(RORconst [c] y)) && clobberIfDead(x1) => (ORNshiftRO x0 y [c])
1594 (EON x0 x1:(SLLconst [c] y)) && clobberIfDead(x1) => (EONshiftLL x0 y [c])
1595 (EON x0 x1:(SRLconst [c] y)) && clobberIfDead(x1) => (EONshiftRL x0 y [c])
1596 (EON x0 x1:(SRAconst [c] y)) && clobberIfDead(x1) => (EONshiftRA x0 y [c])
1597 (EON x0 x1:(RORconst [c] y)) && clobberIfDead(x1) => (EONshiftRO x0 y [c])
1598 (CMP x0 x1:(SLLconst [c] y)) && clobberIfDead(x1) => (CMPshiftLL x0 y [c])
1599 (CMP x0:(SLLconst [c] y) x1) && clobberIfDead(x0) => (InvertFlags (CMPshiftLL x1 y [c]))
1600 (CMP x0 x1:(SRLconst [c] y)) && clobberIfDead(x1) => (CMPshiftRL x0 y [c])
1601 (CMP x0:(SRLconst [c] y) x1) && clobberIfDead(x0) => (InvertFlags (CMPshiftRL x1 y [c]))
1602 (CMP x0 x1:(SRAconst [c] y)) && clobberIfDead(x1) => (CMPshiftRA x0 y [c])
1603 (CMP x0:(SRAconst [c] y) x1) && clobberIfDead(x0) => (InvertFlags (CMPshiftRA x1 y [c]))
1604 (CMN x0 x1:(SLLconst [c] y)) && clobberIfDead(x1) => (CMNshiftLL x0 y [c])
1605 (CMN x0 x1:(SRLconst [c] y)) && clobberIfDead(x1) => (CMNshiftRL x0 y [c])
1606 (CMN x0 x1:(SRAconst [c] y)) && clobberIfDead(x1) => (CMNshiftRA x0 y [c])
1607 (TST x0 x1:(SLLconst [c] y)) && clobberIfDead(x1) => (TSTshiftLL x0 y [c])
1608 (TST x0 x1:(SRLconst [c] y)) && clobberIfDead(x1) => (TSTshiftRL x0 y [c])
1609 (TST x0 x1:(SRAconst [c] y)) && clobberIfDead(x1) => (TSTshiftRA x0 y [c])
1610 (TST x0 x1:(RORconst [c] y)) && clobberIfDead(x1) => (TSTshiftRO x0 y [c])
1611
1612 // prefer *const ops to *shift ops
1613 (ADDshiftLL (MOVDconst [c]) x [d]) => (ADDconst [c] (SLLconst <x.Type> x [d]))
1614 (ADDshiftRL (MOVDconst [c]) x [d]) => (ADDconst [c] (SRLconst <x.Type> x [d]))
1615 (ADDshiftRA (MOVDconst [c]) x [d]) => (ADDconst [c] (SRAconst <x.Type> x [d]))
1616 (ANDshiftLL (MOVDconst [c]) x [d]) => (ANDconst [c] (SLLconst <x.Type> x [d]))
1617 (ANDshiftRL (MOVDconst [c]) x [d]) => (ANDconst [c] (SRLconst <x.Type> x [d]))
1618 (ANDshiftRA (MOVDconst [c]) x [d]) => (ANDconst [c] (SRAconst <x.Type> x [d]))
1619 (ANDshiftRO (MOVDconst [c]) x [d]) => (ANDconst [c] (RORconst <x.Type> x [d]))
1620 (ORshiftLL (MOVDconst [c]) x [d]) => (ORconst [c] (SLLconst <x.Type> x [d]))
1621 (ORshiftRL (MOVDconst [c]) x [d]) => (ORconst [c] (SRLconst <x.Type> x [d]))
1622 (ORshiftRA (MOVDconst [c]) x [d]) => (ORconst [c] (SRAconst <x.Type> x [d]))
1623 (ORshiftRO (MOVDconst [c]) x [d]) => (ORconst [c] (RORconst <x.Type> x [d]))
1624 (XORshiftLL (MOVDconst [c]) x [d]) => (XORconst [c] (SLLconst <x.Type> x [d]))
1625 (XORshiftRL (MOVDconst [c]) x [d]) => (XORconst [c] (SRLconst <x.Type> x [d]))
1626 (XORshiftRA (MOVDconst [c]) x [d]) => (XORconst [c] (SRAconst <x.Type> x [d]))
1627 (XORshiftRO (MOVDconst [c]) x [d]) => (XORconst [c] (RORconst <x.Type> x [d]))
1628 (CMPshiftLL (MOVDconst [c]) x [d]) => (InvertFlags (CMPconst [c] (SLLconst <x.Type> x [d])))
1629 (CMPshiftRL (MOVDconst [c]) x [d]) => (InvertFlags (CMPconst [c] (SRLconst <x.Type> x [d])))
1630 (CMPshiftRA (MOVDconst [c]) x [d]) => (InvertFlags (CMPconst [c] (SRAconst <x.Type> x [d])))
1631 (CMNshiftLL (MOVDconst [c]) x [d]) => (CMNconst [c] (SLLconst <x.Type> x [d]))
1632 (CMNshiftRL (MOVDconst [c]) x [d]) => (CMNconst [c] (SRLconst <x.Type> x [d]))
1633 (CMNshiftRA (MOVDconst [c]) x [d]) => (CMNconst [c] (SRAconst <x.Type> x [d]))
1634 (TSTshiftLL (MOVDconst [c]) x [d]) => (TSTconst [c] (SLLconst <x.Type> x [d]))
1635 (TSTshiftRL (MOVDconst [c]) x [d]) => (TSTconst [c] (SRLconst <x.Type> x [d]))
1636 (TSTshiftRA (MOVDconst [c]) x [d]) => (TSTconst [c] (SRAconst <x.Type> x [d]))
1637 (TSTshiftRO (MOVDconst [c]) x [d]) => (TSTconst [c] (RORconst <x.Type> x [d]))
1638
1639 // constant folding in *shift ops
1640 (MVNshiftLL (MOVDconst [c]) [d]) => (MOVDconst [^int64(uint64(c)<<uint64(d))])
1641 (MVNshiftRL (MOVDconst [c]) [d]) => (MOVDconst [^int64(uint64(c)>>uint64(d))])
1642 (MVNshiftRA (MOVDconst [c]) [d]) => (MOVDconst [^(c>>uint64(d))])
1643 (MVNshiftRO (MOVDconst [c]) [d]) => (MOVDconst [^rotateRight64(c, d)])
1644 (NEGshiftLL (MOVDconst [c]) [d]) => (MOVDconst [-int64(uint64(c)<<uint64(d))])
1645 (NEGshiftRL (MOVDconst [c]) [d]) => (MOVDconst [-int64(uint64(c)>>uint64(d))])
1646 (NEGshiftRA (MOVDconst [c]) [d]) => (MOVDconst [-(c>>uint64(d))])
1647 (ADDshiftLL x (MOVDconst [c]) [d]) => (ADDconst x [int64(uint64(c)<<uint64(d))])
1648 (ADDshiftRL x (MOVDconst [c]) [d]) => (ADDconst x [int64(uint64(c)>>uint64(d))])
1649 (ADDshiftRA x (MOVDconst [c]) [d]) => (ADDconst x [c>>uint64(d)])
1650 (SUBshiftLL x (MOVDconst [c]) [d]) => (SUBconst x [int64(uint64(c)<<uint64(d))])
1651 (SUBshiftRL x (MOVDconst [c]) [d]) => (SUBconst x [int64(uint64(c)>>uint64(d))])
1652 (SUBshiftRA x (MOVDconst [c]) [d]) => (SUBconst x [c>>uint64(d)])
1653 (ANDshiftLL x (MOVDconst [c]) [d]) => (ANDconst x [int64(uint64(c)<<uint64(d))])
1654 (ANDshiftRL x (MOVDconst [c]) [d]) => (ANDconst x [int64(uint64(c)>>uint64(d))])
1655 (ANDshiftRA x (MOVDconst [c]) [d]) => (ANDconst x [c>>uint64(d)])
1656 (ANDshiftRO x (MOVDconst [c]) [d]) => (ANDconst x [rotateRight64(c, d)])
1657 (ORshiftLL x (MOVDconst [c]) [d]) => (ORconst x [int64(uint64(c)<<uint64(d))])
1658 (ORshiftRL x (MOVDconst [c]) [d]) => (ORconst x [int64(uint64(c)>>uint64(d))])
1659 (ORshiftRA x (MOVDconst [c]) [d]) => (ORconst x [c>>uint64(d)])
1660 (ORshiftRO x (MOVDconst [c]) [d]) => (ORconst x [rotateRight64(c, d)])
1661 (XORshiftLL x (MOVDconst [c]) [d]) => (XORconst x [int64(uint64(c)<<uint64(d))])
1662 (XORshiftRL x (MOVDconst [c]) [d]) => (XORconst x [int64(uint64(c)>>uint64(d))])
1663 (XORshiftRA x (MOVDconst [c]) [d]) => (XORconst x [c>>uint64(d)])
1664 (XORshiftRO x (MOVDconst [c]) [d]) => (XORconst x [rotateRight64(c, d)])
1665 (BICshiftLL x (MOVDconst [c]) [d]) => (ANDconst x [^int64(uint64(c)<<uint64(d))])
1666 (BICshiftRL x (MOVDconst [c]) [d]) => (ANDconst x [^int64(uint64(c)>>uint64(d))])
1667 (BICshiftRA x (MOVDconst [c]) [d]) => (ANDconst x [^(c>>uint64(d))])
1668 (BICshiftRO x (MOVDconst [c]) [d]) => (ANDconst x [^rotateRight64(c, d)])
1669 (ORNshiftLL x (MOVDconst [c]) [d]) => (ORconst x [^int64(uint64(c)<<uint64(d))])
1670 (ORNshiftRL x (MOVDconst [c]) [d]) => (ORconst x [^int64(uint64(c)>>uint64(d))])
1671 (ORNshiftRA x (MOVDconst [c]) [d]) => (ORconst x [^(c>>uint64(d))])
1672 (ORNshiftRO x (MOVDconst [c]) [d]) => (ORconst x [^rotateRight64(c, d)])
1673 (EONshiftLL x (MOVDconst [c]) [d]) => (XORconst x [^int64(uint64(c)<<uint64(d))])
1674 (EONshiftRL x (MOVDconst [c]) [d]) => (XORconst x [^int64(uint64(c)>>uint64(d))])
1675 (EONshiftRA x (MOVDconst [c]) [d]) => (XORconst x [^(c>>uint64(d))])
1676 (EONshiftRO x (MOVDconst [c]) [d]) => (XORconst x [^rotateRight64(c, d)])
1677 (CMPshiftLL x (MOVDconst [c]) [d]) => (CMPconst x [int64(uint64(c)<<uint64(d))])
1678 (CMPshiftRL x (MOVDconst [c]) [d]) => (CMPconst x [int64(uint64(c)>>uint64(d))])
1679 (CMPshiftRA x (MOVDconst [c]) [d]) => (CMPconst x [c>>uint64(d)])
1680 (CMNshiftLL x (MOVDconst [c]) [d]) => (CMNconst x [int64(uint64(c)<<uint64(d))])
1681 (CMNshiftRL x (MOVDconst [c]) [d]) => (CMNconst x [int64(uint64(c)>>uint64(d))])
1682 (CMNshiftRA x (MOVDconst [c]) [d]) => (CMNconst x [c>>uint64(d)])
1683 (TSTshiftLL x (MOVDconst [c]) [d]) => (TSTconst x [int64(uint64(c)<<uint64(d))])
1684 (TSTshiftRL x (MOVDconst [c]) [d]) => (TSTconst x [int64(uint64(c)>>uint64(d))])
1685 (TSTshiftRA x (MOVDconst [c]) [d]) => (TSTconst x [c>>uint64(d)])
1686 (TSTshiftRO x (MOVDconst [c]) [d]) => (TSTconst x [rotateRight64(c, d)])
1687
1688 // simplification with *shift ops
1689 (SUBshiftLL (SLLconst x [c]) x [c]) => (MOVDconst [0])
1690 (SUBshiftRL (SRLconst x [c]) x [c]) => (MOVDconst [0])
1691 (SUBshiftRA (SRAconst x [c]) x [c]) => (MOVDconst [0])
1692 (ANDshiftLL y:(SLLconst x [c]) x [c]) => y
1693 (ANDshiftRL y:(SRLconst x [c]) x [c]) => y
1694 (ANDshiftRA y:(SRAconst x [c]) x [c]) => y
1695 (ANDshiftRO y:(RORconst x [c]) x [c]) => y
1696 (ORshiftLL y:(SLLconst x [c]) x [c]) => y
1697 (ORshiftRL y:(SRLconst x [c]) x [c]) => y
1698 (ORshiftRA y:(SRAconst x [c]) x [c]) => y
1699 (ORshiftRO y:(RORconst x [c]) x [c]) => y
1700 (XORshiftLL (SLLconst x [c]) x [c]) => (MOVDconst [0])
1701 (XORshiftRL (SRLconst x [c]) x [c]) => (MOVDconst [0])
1702 (XORshiftRA (SRAconst x [c]) x [c]) => (MOVDconst [0])
1703 (XORshiftRO (RORconst x [c]) x [c]) => (MOVDconst [0])
1704 (BICshiftLL (SLLconst x [c]) x [c]) => (MOVDconst [0])
1705 (BICshiftRL (SRLconst x [c]) x [c]) => (MOVDconst [0])
1706 (BICshiftRA (SRAconst x [c]) x [c]) => (MOVDconst [0])
1707 (BICshiftRO (RORconst x [c]) x [c]) => (MOVDconst [0])
1708 (EONshiftLL (SLLconst x [c]) x [c]) => (MOVDconst [-1])
1709 (EONshiftRL (SRLconst x [c]) x [c]) => (MOVDconst [-1])
1710 (EONshiftRA (SRAconst x [c]) x [c]) => (MOVDconst [-1])
1711 (EONshiftRO (RORconst x [c]) x [c]) => (MOVDconst [-1])
1712 (ORNshiftLL (SLLconst x [c]) x [c]) => (MOVDconst [-1])
1713 (ORNshiftRL (SRLconst x [c]) x [c]) => (MOVDconst [-1])
1714 (ORNshiftRA (SRAconst x [c]) x [c]) => (MOVDconst [-1])
1715 (ORNshiftRO (RORconst x [c]) x [c]) => (MOVDconst [-1])
1716
1717 // rev16w | rev16
1718 // ((x>>8) | (x<<8)) => (REV16W x), the type of x is uint16, "|" can also be "^" or "+".
1719 ((ADDshiftLL|ORshiftLL|XORshiftLL) <typ.UInt16> [8] (UBFX <typ.UInt16> [armBFAuxInt(8, 8)] x) x) => (REV16W x)
1720
1721 // ((x & 0xff00ff00)>>8) | ((x & 0x00ff00ff)<<8), "|" can also be "^" or "+".
1722 ((ADDshiftLL|ORshiftLL|XORshiftLL) [8] (UBFX [armBFAuxInt(8, 24)] (ANDconst [c1] x)) (ANDconst [c2] x))
1723 && uint32(c1) == 0xff00ff00 && uint32(c2) == 0x00ff00ff
1724 => (REV16W x)
1725
1726 // ((x & 0xff00ff00ff00ff00)>>8) | ((x & 0x00ff00ff00ff00ff)<<8), "|" can also be "^" or "+".
1727 ((ADDshiftLL|ORshiftLL|XORshiftLL) [8] (SRLconst [8] (ANDconst [c1] x)) (ANDconst [c2] x))
1728 && (uint64(c1) == 0xff00ff00ff00ff00 && uint64(c2) == 0x00ff00ff00ff00ff)
1729 => (REV16 x)
1730
1731 // ((x & 0xff00ff00)>>8) | ((x & 0x00ff00ff)<<8), "|" can also be "^" or "+".
1732 ((ADDshiftLL|ORshiftLL|XORshiftLL) [8] (SRLconst [8] (ANDconst [c1] x)) (ANDconst [c2] x))
1733 && (uint64(c1) == 0xff00ff00 && uint64(c2) == 0x00ff00ff)
1734 => (REV16 (ANDconst <x.Type> [0xffffffff] x))
1735
1736 // Extract from reg pair
1737 (ADDshiftLL [c] (SRLconst x [64-c]) x2) => (EXTRconst [64-c] x2 x)
1738 ( ORshiftLL [c] (SRLconst x [64-c]) x2) => (EXTRconst [64-c] x2 x)
1739 (XORshiftLL [c] (SRLconst x [64-c]) x2) => (EXTRconst [64-c] x2 x)
1740
1741 (ADDshiftLL <t> [c] (UBFX [bfc] x) x2) && c < 32 && t.Size() == 4 && bfc == armBFAuxInt(32-c, c)
1742 => (EXTRWconst [32-c] x2 x)
1743 ( ORshiftLL <t> [c] (UBFX [bfc] x) x2) && c < 32 && t.Size() == 4 && bfc == armBFAuxInt(32-c, c)
1744 => (EXTRWconst [32-c] x2 x)
1745 (XORshiftLL <t> [c] (UBFX [bfc] x) x2) && c < 32 && t.Size() == 4 && bfc == armBFAuxInt(32-c, c)
1746 => (EXTRWconst [32-c] x2 x)
1747
1748 // Rewrite special pairs of shifts to AND.
1749 // On ARM64 the bitmask can fit into an instruction.
1750 (SRLconst [c] (SLLconst [c] x)) && 0 < c && c < 64 => (ANDconst [1<<uint(64-c)-1] x) // mask out high bits
1751 (SLLconst [c] (SRLconst [c] x)) && 0 < c && c < 64 => (ANDconst [^(1<<uint(c)-1)] x) // mask out low bits
1752
1753 // Special case setting bit as 1. An example is math.Copysign(c,-1)
1754 (ORconst [c1] (ANDconst [c2] x)) && c2|c1 == ^0 => (ORconst [c1] x)
1755
1756 // If the shift amount is larger than the datasize(32, 16, 8), we can optimize to constant 0.
1757 (MOVWUreg (SLLconst [lc] x)) && lc >= 32 => (MOVDconst [0])
1758 (MOVHUreg (SLLconst [lc] x)) && lc >= 16 => (MOVDconst [0])
1759 (MOVBUreg (SLLconst [lc] x)) && lc >= 8 => (MOVDconst [0])
1760
1761 // After zero extension, the upper (64-datasize(32|16|8)) bits are zero, we can optimiza to constant 0.
1762 (SRLconst [rc] (MOVWUreg x)) && rc >= 32 => (MOVDconst [0])
1763 (SRLconst [rc] (MOVHUreg x)) && rc >= 16 => (MOVDconst [0])
1764 (SRLconst [rc] (MOVBUreg x)) && rc >= 8 => (MOVDconst [0])
1765
1766 // bitfield ops
1767
1768 // sbfiz
1769 // (x << lc) >> rc
1770 (SRAconst [rc] (SLLconst [lc] x)) && lc > rc => (SBFIZ [armBFAuxInt(lc-rc, 64-lc)] x)
1771 // int64(x << lc)
1772 (MOVWreg (SLLconst [lc] x)) && lc < 32 => (SBFIZ [armBFAuxInt(lc, 32-lc)] x)
1773 (MOVHreg (SLLconst [lc] x)) && lc < 16 => (SBFIZ [armBFAuxInt(lc, 16-lc)] x)
1774 (MOVBreg (SLLconst [lc] x)) && lc < 8 => (SBFIZ [armBFAuxInt(lc, 8-lc)] x)
1775 // int64(x) << lc
1776 (SLLconst [lc] (MOVWreg x)) => (SBFIZ [armBFAuxInt(lc, min(32, 64-lc))] x)
1777 (SLLconst [lc] (MOVHreg x)) => (SBFIZ [armBFAuxInt(lc, min(16, 64-lc))] x)
1778 (SLLconst [lc] (MOVBreg x)) => (SBFIZ [armBFAuxInt(lc, min(8, 64-lc))] x)
1779
1780 // sbfx
1781 // (x << lc) >> rc
1782 (SRAconst [rc] (SLLconst [lc] x)) && lc <= rc => (SBFX [armBFAuxInt(rc-lc, 64-rc)] x)
1783 // int64(x) >> rc
1784 (SRAconst [rc] (MOVWreg x)) && rc < 32 => (SBFX [armBFAuxInt(rc, 32-rc)] x)
1785 (SRAconst [rc] (MOVHreg x)) && rc < 16 => (SBFX [armBFAuxInt(rc, 16-rc)] x)
1786 (SRAconst [rc] (MOVBreg x)) && rc < 8 => (SBFX [armBFAuxInt(rc, 8-rc)] x)
1787 // merge sbfx and sign-extension into sbfx
1788 (MOVWreg (SBFX [bfc] x)) && bfc.width() <= 32 => (SBFX [bfc] x)
1789 (MOVHreg (SBFX [bfc] x)) && bfc.width() <= 16 => (SBFX [bfc] x)
1790 (MOVBreg (SBFX [bfc] x)) && bfc.width() <= 8 => (SBFX [bfc] x)
1791
1792 // sbfiz/sbfx combinations: merge shifts into bitfield ops
1793 (SRAconst [sc] (SBFIZ [bfc] x)) && sc < bfc.lsb()
1794 => (SBFIZ [armBFAuxInt(bfc.lsb()-sc, bfc.width())] x)
1795 (SRAconst [sc] (SBFIZ [bfc] x)) && sc >= bfc.lsb()
1796 && sc < bfc.lsb()+bfc.width()
1797 => (SBFX [armBFAuxInt(sc-bfc.lsb(), bfc.lsb()+bfc.width()-sc)] x)
1798 (SBFX [bfc] s:(SLLconst [sc] x))
1799 && s.Uses == 1
1800 && sc <= bfc.lsb()
1801 => (SBFX [armBFAuxInt(bfc.lsb() - sc, bfc.width())] x)
1802 (SBFX [bfc] s:(SLLconst [sc] x))
1803 && s.Uses == 1
1804 && sc > bfc.lsb()
1805 => (SBFIZ [armBFAuxInt(sc - bfc.lsb(), bfc.width() - (sc-bfc.lsb()))] x)
1806
1807 // ubfiz
1808 // (x << lc) >> rc
1809 (SRLconst [rc] (SLLconst [lc] x)) && lc > rc => (UBFIZ [armBFAuxInt(lc-rc, 64-lc)] x)
1810 // uint64(x) << lc
1811 (SLLconst [lc] (MOVWUreg x)) => (UBFIZ [armBFAuxInt(lc, min(32, 64-lc))] x)
1812 (SLLconst [lc] (MOVHUreg x)) => (UBFIZ [armBFAuxInt(lc, min(16, 64-lc))] x)
1813 (SLLconst [lc] (MOVBUreg x)) => (UBFIZ [armBFAuxInt(lc, min(8, 64-lc))] x)
1814 // uint64(x << lc)
1815 (MOVWUreg (SLLconst [lc] x)) && lc < 32 => (UBFIZ [armBFAuxInt(lc, 32-lc)] x)
1816 (MOVHUreg (SLLconst [lc] x)) && lc < 16 => (UBFIZ [armBFAuxInt(lc, 16-lc)] x)
1817 (MOVBUreg (SLLconst [lc] x)) && lc < 8 => (UBFIZ [armBFAuxInt(lc, 8-lc)] x)
1818
1819 // merge ANDconst into ubfiz
1820 // (x & ac) << sc
1821 (SLLconst [sc] (ANDconst [ac] x)) && isARM64BFMask(sc, ac, 0)
1822 => (UBFIZ [armBFAuxInt(sc, arm64BFWidth(ac, 0))] x)
1823 // (x << sc) & ac
1824 (ANDconst [ac] (SLLconst [sc] x)) && isARM64BFMask(sc, ac, sc)
1825 => (UBFIZ [armBFAuxInt(sc, arm64BFWidth(ac, sc))] x)
1826
1827 // ubfx
1828 // (x << lc) >> rc
1829 (SRLconst [rc] (SLLconst [lc] x)) && lc < rc => (UBFX [armBFAuxInt(rc-lc, 64-rc)] x)
1830 // uint64(x) >> rc
1831 (SRLconst [rc] (MOVWUreg x)) && rc < 32 => (UBFX [armBFAuxInt(rc, 32-rc)] x)
1832 (SRLconst [rc] (MOVHUreg x)) && rc < 16 => (UBFX [armBFAuxInt(rc, 16-rc)] x)
1833 (SRLconst [rc] (MOVBUreg x)) && rc < 8 => (UBFX [armBFAuxInt(rc, 8-rc)] x)
1834 // uint64(x >> rc)
1835 (MOVWUreg (SRLconst [rc] x)) && rc < 32 => (UBFX [armBFAuxInt(rc, 32)] x)
1836 (MOVHUreg (SRLconst [rc] x)) && rc < 16 => (UBFX [armBFAuxInt(rc, 16)] x)
1837 (MOVBUreg (SRLconst [rc] x)) && rc < 8 => (UBFX [armBFAuxInt(rc, 8)] x)
1838 // merge ANDconst into ubfx
1839 // (x >> sc) & ac
1840 (ANDconst [ac] (SRLconst [sc] x)) && isARM64BFMask(sc, ac, 0)
1841 => (UBFX [armBFAuxInt(sc, arm64BFWidth(ac, 0))] x)
1842 // (x & ac) >> sc
1843 (SRLconst [sc] (ANDconst [ac] x)) && isARM64BFMask(sc, ac, sc)
1844 => (UBFX [armBFAuxInt(sc, arm64BFWidth(ac, sc))] x)
1845 // merge ANDconst and ubfx into ubfx
1846 (ANDconst [c] (UBFX [bfc] x)) && isARM64BFMask(0, c, 0) =>
1847 (UBFX [armBFAuxInt(bfc.lsb(), min(bfc.width(), arm64BFWidth(c, 0)))] x)
1848 (UBFX [bfc] (ANDconst [c] x)) && isARM64BFMask(0, c, 0) && bfc.lsb() + bfc.width() <= arm64BFWidth(c, 0) =>
1849 (UBFX [bfc] x)
1850 // merge ubfx and zero-extension into ubfx
1851 (MOVWUreg (UBFX [bfc] x)) && bfc.width() <= 32 => (UBFX [bfc] x)
1852 (MOVHUreg (UBFX [bfc] x)) && bfc.width() <= 16 => (UBFX [bfc] x)
1853 (MOVBUreg (UBFX [bfc] x)) && bfc.width() <= 8 => (UBFX [bfc] x)
1854
1855 // Extracting bits from across a zero-extension boundary.
1856 (UBFX [bfc] e:(MOVWUreg x))
1857 && e.Uses == 1
1858 && bfc.lsb() < 32
1859 => (UBFX [armBFAuxInt(bfc.lsb(), min(bfc.width(), 32-bfc.lsb()))] x)
1860 (UBFX [bfc] e:(MOVHUreg x))
1861 && e.Uses == 1
1862 && bfc.lsb() < 16
1863 => (UBFX [armBFAuxInt(bfc.lsb(), min(bfc.width(), 16-bfc.lsb()))] x)
1864 (UBFX [bfc] e:(MOVBUreg x))
1865 && e.Uses == 1
1866 && bfc.lsb() < 8
1867 => (UBFX [armBFAuxInt(bfc.lsb(), min(bfc.width(), 8-bfc.lsb()))] x)
1868
1869 // ubfiz/ubfx combinations: merge shifts into bitfield ops
1870 (SRLconst [sc] (UBFX [bfc] x)) && sc < bfc.width()
1871 => (UBFX [armBFAuxInt(bfc.lsb()+sc, bfc.width()-sc)] x)
1872 (UBFX [bfc] (SRLconst [sc] x)) && sc+bfc.width()+bfc.lsb() < 64
1873 => (UBFX [armBFAuxInt(bfc.lsb()+sc, bfc.width())] x)
1874 (SLLconst [sc] (UBFIZ [bfc] x)) && sc+bfc.width()+bfc.lsb() < 64
1875 => (UBFIZ [armBFAuxInt(bfc.lsb()+sc, bfc.width())] x)
1876 (UBFIZ [bfc] (SLLconst [sc] x)) && sc < bfc.width()
1877 => (UBFIZ [armBFAuxInt(bfc.lsb()+sc, bfc.width()-sc)] x)
1878 // ((x << c1) >> c2) >> c3
1879 (SRLconst [sc] (UBFIZ [bfc] x)) && sc == bfc.lsb()
1880 => (ANDconst [1<<uint(bfc.width())-1] x)
1881 (SRLconst [sc] (UBFIZ [bfc] x)) && sc < bfc.lsb()
1882 => (UBFIZ [armBFAuxInt(bfc.lsb()-sc, bfc.width())] x)
1883 (SRLconst [sc] (UBFIZ [bfc] x)) && sc > bfc.lsb()
1884 && sc < bfc.lsb()+bfc.width()
1885 => (UBFX [armBFAuxInt(sc-bfc.lsb(), bfc.lsb()+bfc.width()-sc)] x)
1886 // ((x << c1) << c2) >> c3
1887 (UBFX [bfc] (SLLconst [sc] x)) && sc == bfc.lsb()
1888 => (ANDconst [1<<uint(bfc.width())-1] x)
1889 (UBFX [bfc] (SLLconst [sc] x)) && sc < bfc.lsb()
1890 => (UBFX [armBFAuxInt(bfc.lsb()-sc, bfc.width())] x)
1891 (UBFX [bfc] (SLLconst [sc] x)) && sc > bfc.lsb()
1892 && sc < bfc.lsb()+bfc.width()
1893 => (UBFIZ [armBFAuxInt(sc-bfc.lsb(), bfc.lsb()+bfc.width()-sc)] x)
1894
1895 // bfi
1896 (OR (UBFIZ [bfc] x) (ANDconst [ac] y))
1897 && ac == ^((1<<uint(bfc.width())-1) << uint(bfc.lsb()))
1898 => (BFI [bfc] y x)
1899 (ORshiftLL [s] (ANDconst [xc] x) (ANDconst [yc] y))
1900 && xc == ^(yc << s) // opposite masks
1901 && yc & (yc+1) == 0 // power of 2 minus 1
1902 && yc > 0 // not 0, not all 64 bits (there are better rewrites in that case)
1903 && s+log64(yc+1) <= 64 // shifted mask doesn't overflow
1904 => (BFI [armBFAuxInt(s, log64(yc+1))] x y)
1905 (ORshiftRL [rc] (ANDconst [ac] x) (SLLconst [lc] y))
1906 && lc > rc && ac == ^((1<<uint(64-lc)-1) << uint64(lc-rc))
1907 => (BFI [armBFAuxInt(lc-rc, 64-lc)] x y)
1908 // bfxil
1909 (OR (UBFX [bfc] x) (ANDconst [ac] y)) && ac == ^(1<<uint(bfc.width())-1)
1910 => (BFXIL [bfc] y x)
1911 (ORshiftLL [sc] (UBFX [bfc] x) (SRLconst [sc] y)) && sc == bfc.width()
1912 => (BFXIL [bfc] y x)
1913 (ORshiftRL [rc] (ANDconst [ac] y) (SLLconst [lc] x)) && lc < rc && ac == ^((1<<uint(64-rc)-1))
1914 => (BFXIL [armBFAuxInt(rc-lc, 64-rc)] y x)
1915
1916 // FP simplification
1917 (FNEGS (FMULS x y)) => (FNMULS x y)
1918 (FNEGD (FMULD x y)) => (FNMULD x y)
1919 (FMULS (FNEGS x) y) => (FNMULS x y)
1920 (FMULD (FNEGD x) y) => (FNMULD x y)
1921 (FNEGS (FNMULS x y)) => (FMULS x y)
1922 (FNEGD (FNMULD x y)) => (FMULD x y)
1923 (FNMULS (FNEGS x) y) => (FMULS x y)
1924 (FNMULD (FNEGD x) y) => (FMULD x y)
1925
1926 (FADDS a (FMULS x y)) && a.Block.Func.useFMA(v) => (FMADDS a x y)
1927 (FADDD a (FMULD x y)) && a.Block.Func.useFMA(v) => (FMADDD a x y)
1928 (FSUBS a (FMULS x y)) && a.Block.Func.useFMA(v) => (FMSUBS a x y)
1929 (FSUBD a (FMULD x y)) && a.Block.Func.useFMA(v) => (FMSUBD a x y)
1930 (FSUBS (FMULS x y) a) && a.Block.Func.useFMA(v) => (FNMSUBS a x y)
1931 (FSUBD (FMULD x y) a) && a.Block.Func.useFMA(v) => (FNMSUBD a x y)
1932 (FADDS a (FNMULS x y)) && a.Block.Func.useFMA(v) => (FMSUBS a x y)
1933 (FADDD a (FNMULD x y)) && a.Block.Func.useFMA(v) => (FMSUBD a x y)
1934 (FSUBS a (FNMULS x y)) && a.Block.Func.useFMA(v) => (FMADDS a x y)
1935 (FSUBD a (FNMULD x y)) && a.Block.Func.useFMA(v) => (FMADDD a x y)
1936 (FSUBS (FNMULS x y) a) && a.Block.Func.useFMA(v) => (FNMADDS a x y)
1937 (FSUBD (FNMULD x y) a) && a.Block.Func.useFMA(v) => (FNMADDD a x y)
1938
1939 (MOVBUload [off] {sym} (SB) _) && symIsRO(sym) => (MOVDconst [int64(read8(sym, int64(off)))])
1940 (MOVHUload [off] {sym} (SB) _) && symIsRO(sym) => (MOVDconst [int64(read16(sym, int64(off), config.ctxt.Arch.ByteOrder))])
1941 (MOVWUload [off] {sym} (SB) _) && symIsRO(sym) => (MOVDconst [int64(read32(sym, int64(off), config.ctxt.Arch.ByteOrder))])
1942 (MOVDload [off] {sym} (SB) _) && symIsRO(sym) => (MOVDconst [int64(read64(sym, int64(off), config.ctxt.Arch.ByteOrder))])
1943
1944 // Prefetch instructions (aux is option: 0 - PLDL1KEEP; 1 - PLDL1STRM)
1945 (PrefetchCache addr mem) => (PRFM [0] addr mem)
1946 (PrefetchCacheStreamed addr mem) => (PRFM [1] addr mem)
1947
1948 // Arch-specific inlining for small or disjoint runtime.memmove
1949 (SelectN [0] call:(CALLstatic {sym} s1:(MOVDstore _ (MOVDconst [sz]) s2:(MOVDstore _ src s3:(MOVDstore {t} _ dst mem)))))
1950 && sz >= 0
1951 && isSameCall(sym, "runtime.memmove")
1952 && s1.Uses == 1 && s2.Uses == 1 && s3.Uses == 1
1953 && isInlinableMemmove(dst, src, sz, config)
1954 && clobber(s1, s2, s3, call)
1955 => (Move [sz] dst src mem)
1956
1957 // Match post-lowering calls, register version.
1958 (SelectN [0] call:(CALLstatic {sym} dst src (MOVDconst [sz]) mem))
1959 && sz >= 0
1960 && isSameCall(sym, "runtime.memmove")
1961 && call.Uses == 1
1962 && isInlinableMemmove(dst, src, sz, config)
1963 && clobber(call)
1964 => (Move [sz] dst src mem)
1965
1966 ((REV|REVW) ((REV|REVW) p)) => p
1967
1968 // internal/runtime/math.MulUintptr intrinsics
1969
1970 (Select0 (Mul64uover x y)) => (MUL x y)
1971 (Select1 (Mul64uover x y)) => (NotEqual (CMPconst (UMULH <typ.UInt64> x y) [0]))
1972
View as plain text