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 (BitLen(16|8) x) => (BitLen64 (ZeroExt(16|8)to64 x))
107
108 (Bswap64 ...) => (REV ...)
109 (Bswap32 ...) => (REVW ...)
110 (Bswap16 ...) => (REV16W ...)
111
112 (BitRev64 ...) => (RBIT ...)
113 (BitRev32 ...) => (RBITW ...)
114 (BitRev16 x) => (SRLconst [48] (RBIT <typ.UInt64> x))
115 (BitRev8 x) => (SRLconst [56] (RBIT <typ.UInt64> x))
116
117 // In fact, UMOD will be translated into UREM instruction, and UREM is originally translated into
118 // UDIV and MSUB instructions. But if there is already an identical UDIV instruction just before or
119 // after UREM (case like quo, rem := z/y, z%y), then the second UDIV instruction becomes redundant.
120 // The purpose of this rule is to have this extra UDIV instruction removed in CSE pass.
121 (UMOD <typ.UInt64> x y) => (MSUB <typ.UInt64> x y (UDIV <typ.UInt64> x y))
122 (UMODW <typ.UInt32> x y) => (MSUBW <typ.UInt32> x y (UDIVW <typ.UInt32> x y))
123
124 // 64-bit addition with carry.
125 (Select0 (Add64carry x y c)) => (Select0 <typ.UInt64> (ADCSflags x y (Select1 <types.TypeFlags> (ADDSconstflags [-1] c))))
126 (Select1 (Add64carry x y c)) => (ADCzerocarry <typ.UInt64> (Select1 <types.TypeFlags> (ADCSflags x y (Select1 <types.TypeFlags> (ADDSconstflags [-1] c)))))
127
128 // 64-bit subtraction with borrowing.
129 (Select0 (Sub64borrow x y bo)) => (Select0 <typ.UInt64> (SBCSflags x y (Select1 <types.TypeFlags> (NEGSflags bo))))
130 (Select1 (Sub64borrow x y bo)) => (NEG <typ.UInt64> (NGCzerocarry <typ.UInt64> (Select1 <types.TypeFlags> (SBCSflags x y (Select1 <types.TypeFlags> (NEGSflags bo))))))
131
132 // boolean ops -- booleans are represented with 0=false, 1=true
133 (AndB ...) => (AND ...)
134 (OrB ...) => (OR ...)
135 (EqB x y) => (XOR (MOVDconst [1]) (XOR <typ.Bool> x y))
136 (NeqB ...) => (XOR ...)
137 (Not x) => (XOR (MOVDconst [1]) x)
138
139 // shifts
140 // hardware instruction uses only the low 6 bits of the shift
141 // we compare to 64 to ensure Go semantics for large shifts
142 // Rules about rotates with non-const shift are based on the following rules,
143 // if the following rules change, please also modify the rules based on them.
144
145 // check shiftIsBounded first, if shift value is proved to be valid then we
146 // can do the shift directly.
147 // left shift
148 (Lsh(64|32|16|8)x64 <t> x y) && shiftIsBounded(v) => (SLL <t> x y)
149 (Lsh(64|32|16|8)x32 <t> x y) && shiftIsBounded(v) => (SLL <t> x y)
150 (Lsh(64|32|16|8)x16 <t> x y) && shiftIsBounded(v) => (SLL <t> x y)
151 (Lsh(64|32|16|8)x8 <t> x y) && shiftIsBounded(v) => (SLL <t> x y)
152
153 // signed right shift
154 (Rsh64x(64|32|16|8) <t> x y) && shiftIsBounded(v) => (SRA <t> x y)
155 (Rsh32x(64|32|16|8) <t> x y) && shiftIsBounded(v) => (SRA <t> (SignExt32to64 x) y)
156 (Rsh16x(64|32|16|8) <t> x y) && shiftIsBounded(v) => (SRA <t> (SignExt16to64 x) y)
157 (Rsh8x(64|32|16|8) <t> x y) && shiftIsBounded(v) => (SRA <t> (SignExt8to64 x) y)
158
159 // unsigned right shift
160 (Rsh64Ux(64|32|16|8) <t> x y) && shiftIsBounded(v) => (SRL <t> x y)
161 (Rsh32Ux(64|32|16|8) <t> x y) && shiftIsBounded(v) => (SRL <t> (ZeroExt32to64 x) y)
162 (Rsh16Ux(64|32|16|8) <t> x y) && shiftIsBounded(v) => (SRL <t> (ZeroExt16to64 x) y)
163 (Rsh8Ux(64|32|16|8) <t> x y) && shiftIsBounded(v) => (SRL <t> (ZeroExt8to64 x) y)
164
165 // shift value may be out of range, use CMP + CSEL instead
166 (Lsh64x64 <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SLL <t> x y) (Const64 <t> [0]) (CMPconst [64] y))
167 (Lsh64x(32|16|8) <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SLL <t> x y) (Const64 <t> [0]) (CMPconst [64] ((ZeroExt32to64|ZeroExt16to64|ZeroExt8to64) y)))
168
169 (Lsh32x64 <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SLL <t> x y) (Const64 <t> [0]) (CMPconst [64] y))
170 (Lsh32x(32|16|8) <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SLL <t> x y) (Const64 <t> [0]) (CMPconst [64] ((ZeroExt32to64|ZeroExt16to64|ZeroExt8to64) y)))
171
172 (Lsh16x64 <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SLL <t> x y) (Const64 <t> [0]) (CMPconst [64] y))
173 (Lsh16x(32|16|8) <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SLL <t> x y) (Const64 <t> [0]) (CMPconst [64] ((ZeroExt32to64|ZeroExt16to64|ZeroExt8to64) y)))
174
175 (Lsh8x64 <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SLL <t> x y) (Const64 <t> [0]) (CMPconst [64] y))
176 (Lsh8x(32|16|8) <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SLL <t> x y) (Const64 <t> [0]) (CMPconst [64] ((ZeroExt32to64|ZeroExt16to64|ZeroExt8to64) y)))
177
178 (Rsh64Ux64 <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SRL <t> x y) (Const64 <t> [0]) (CMPconst [64] y))
179 (Rsh64Ux(32|16|8) <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SRL <t> x y) (Const64 <t> [0]) (CMPconst [64] ((ZeroExt32to64|ZeroExt16to64|ZeroExt8to64) y)))
180
181 (Rsh32Ux64 <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SRL <t> (ZeroExt32to64 x) y) (Const64 <t> [0]) (CMPconst [64] y))
182 (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)))
183
184 (Rsh16Ux64 <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SRL <t> (ZeroExt16to64 x) y) (Const64 <t> [0]) (CMPconst [64] y))
185 (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)))
186
187 (Rsh8Ux64 <t> x y) && !shiftIsBounded(v) => (CSEL [OpARM64LessThanU] (SRL <t> (ZeroExt8to64 x) y) (Const64 <t> [0]) (CMPconst [64] y))
188 (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)))
189
190 (Rsh64x64 x y) && !shiftIsBounded(v) => (SRA x (CSEL [OpARM64LessThanU] <y.Type> y (Const64 <y.Type> [63]) (CMPconst [64] y)))
191 (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))))
192
193 (Rsh32x64 x y) && !shiftIsBounded(v) => (SRA (SignExt32to64 x) (CSEL [OpARM64LessThanU] <y.Type> y (Const64 <y.Type> [63]) (CMPconst [64] y)))
194 (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))))
195
196 (Rsh16x64 x y) && !shiftIsBounded(v) => (SRA (SignExt16to64 x) (CSEL [OpARM64LessThanU] <y.Type> y (Const64 <y.Type> [63]) (CMPconst [64] y)))
197 (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))))
198
199 (Rsh8x64 x y) && !shiftIsBounded(v) => (SRA (SignExt8to64 x) (CSEL [OpARM64LessThanU] <y.Type> y (Const64 <y.Type> [63]) (CMPconst [64] y)))
200 (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))))
201
202 // constants
203 (Const(64|32|16|8) [val]) => (MOVDconst [int64(val)])
204 (Const(32|64)F [val]) => (FMOV(S|D)const [float64(val)])
205 (ConstNil) => (MOVDconst [0])
206 (ConstBool [t]) => (MOVDconst [b2i(t)])
207
208 (Slicemask <t> x) => (SRAconst (NEG <t> x) [63])
209
210 // truncations
211 // Because we ignore high parts of registers, truncates are just copies.
212 (Trunc16to8 ...) => (Copy ...)
213 (Trunc32to8 ...) => (Copy ...)
214 (Trunc32to16 ...) => (Copy ...)
215 (Trunc64to8 ...) => (Copy ...)
216 (Trunc64to16 ...) => (Copy ...)
217 (Trunc64to32 ...) => (Copy ...)
218
219 // Zero-/Sign-extensions
220 (ZeroExt8to16 ...) => (MOVBUreg ...)
221 (ZeroExt8to32 ...) => (MOVBUreg ...)
222 (ZeroExt16to32 ...) => (MOVHUreg ...)
223 (ZeroExt8to64 ...) => (MOVBUreg ...)
224 (ZeroExt16to64 ...) => (MOVHUreg ...)
225 (ZeroExt32to64 ...) => (MOVWUreg ...)
226
227 (SignExt8to16 ...) => (MOVBreg ...)
228 (SignExt8to32 ...) => (MOVBreg ...)
229 (SignExt16to32 ...) => (MOVHreg ...)
230 (SignExt8to64 ...) => (MOVBreg ...)
231 (SignExt16to64 ...) => (MOVHreg ...)
232 (SignExt32to64 ...) => (MOVWreg ...)
233
234 // float <=> int conversion
235 (Cvt32to32F ...) => (SCVTFWS ...)
236 (Cvt32to64F ...) => (SCVTFWD ...)
237 (Cvt64to32F ...) => (SCVTFS ...)
238 (Cvt64to64F ...) => (SCVTFD ...)
239 (Cvt32Uto32F ...) => (UCVTFWS ...)
240 (Cvt32Uto64F ...) => (UCVTFWD ...)
241 (Cvt64Uto32F ...) => (UCVTFS ...)
242 (Cvt64Uto64F ...) => (UCVTFD ...)
243 (Cvt32Fto32 ...) => (FCVTZSSW ...)
244 (Cvt64Fto32 ...) => (FCVTZSDW ...)
245 (Cvt32Fto64 ...) => (FCVTZSS ...)
246 (Cvt64Fto64 ...) => (FCVTZSD ...)
247 (Cvt32Fto32U ...) => (FCVTZUSW ...)
248 (Cvt64Fto32U ...) => (FCVTZUDW ...)
249 (Cvt32Fto64U ...) => (FCVTZUS ...)
250 (Cvt64Fto64U ...) => (FCVTZUD ...)
251 (Cvt32Fto64F ...) => (FCVTSD ...)
252 (Cvt64Fto32F ...) => (FCVTDS ...)
253
254 (CvtBoolToUint8 ...) => (Copy ...)
255
256 (Round32F ...) => (LoweredRound32F ...)
257 (Round64F ...) => (LoweredRound64F ...)
258
259 // comparisons
260 (Eq8 x y) => (Equal (CMPW (ZeroExt8to32 x) (ZeroExt8to32 y)))
261 (Eq16 x y) => (Equal (CMPW (ZeroExt16to32 x) (ZeroExt16to32 y)))
262 (Eq32 x y) => (Equal (CMPW x y))
263 (Eq64 x y) => (Equal (CMP x y))
264 (EqPtr x y) => (Equal (CMP x y))
265 (Eq32F x y) => (Equal (FCMPS x y))
266 (Eq64F x y) => (Equal (FCMPD x y))
267
268 (Neq8 x y) => (NotEqual (CMPW (ZeroExt8to32 x) (ZeroExt8to32 y)))
269 (Neq16 x y) => (NotEqual (CMPW (ZeroExt16to32 x) (ZeroExt16to32 y)))
270 (Neq32 x y) => (NotEqual (CMPW x y))
271 (Neq64 x y) => (NotEqual (CMP x y))
272 (NeqPtr x y) => (NotEqual (CMP x y))
273 (Neq(32|64)F x y) => (NotEqual (FCMP(S|D) x y))
274
275 (Less(8|16) x y) => (LessThan (CMPW (SignExt(8|16)to32 x) (SignExt(8|16)to32 y)))
276 (Less32 x y) => (LessThan (CMPW x y))
277 (Less64 x y) => (LessThan (CMP x y))
278
279 // Set condition flags for floating-point comparisons "x < y"
280 // and "x <= y". Because if either or both of the operands are
281 // NaNs, all three of (x < y), (x == y) and (x > y) are false,
282 // and ARM Manual says FCMP instruction sets PSTATE.<N,Z,C,V>
283 // of this case to (0, 0, 1, 1).
284 (Less32F x y) => (LessThanF (FCMPS x y))
285 (Less64F x y) => (LessThanF (FCMPD x y))
286
287 // For an unsigned integer x, the following rules are useful when combining branch
288 // 0 < x => x != 0
289 // x <= 0 => x == 0
290 // x < 1 => x == 0
291 // 1 <= x => x != 0
292 (Less(8U|16U|32U|64U) zero:(MOVDconst [0]) x) => (Neq(8|16|32|64) zero x)
293 (Leq(8U|16U|32U|64U) x zero:(MOVDconst [0])) => (Eq(8|16|32|64) x zero)
294 (Less(8U|16U|32U|64U) x (MOVDconst [1])) => (Eq(8|16|32|64) x (MOVDconst [0]))
295 (Leq(8U|16U|32U|64U) (MOVDconst [1]) x) => (Neq(8|16|32|64) (MOVDconst [0]) x)
296
297 (Less8U x y) => (LessThanU (CMPW (ZeroExt8to32 x) (ZeroExt8to32 y)))
298 (Less16U x y) => (LessThanU (CMPW (ZeroExt16to32 x) (ZeroExt16to32 y)))
299 (Less32U x y) => (LessThanU (CMPW x y))
300 (Less64U x y) => (LessThanU (CMP x y))
301
302 (Leq8 x y) => (LessEqual (CMPW (SignExt8to32 x) (SignExt8to32 y)))
303 (Leq16 x y) => (LessEqual (CMPW (SignExt16to32 x) (SignExt16to32 y)))
304 (Leq32 x y) => (LessEqual (CMPW x y))
305 (Leq64 x y) => (LessEqual (CMP x y))
306
307 // Refer to the comments for op Less64F above.
308 (Leq32F x y) => (LessEqualF (FCMPS x y))
309 (Leq64F x y) => (LessEqualF (FCMPD x y))
310
311 (Leq8U x y) => (LessEqualU (CMPW (ZeroExt8to32 x) (ZeroExt8to32 y)))
312 (Leq16U x y) => (LessEqualU (CMPW (ZeroExt16to32 x) (ZeroExt16to32 y)))
313 (Leq32U x y) => (LessEqualU (CMPW x y))
314 (Leq64U x y) => (LessEqualU (CMP x y))
315
316 // Optimize comparison between a floating-point value and 0.0 with "FCMP $(0.0), Fn"
317 (FCMPS x (FMOVSconst [0])) => (FCMPS0 x)
318 (FCMPS (FMOVSconst [0]) x) => (InvertFlags (FCMPS0 x))
319 (FCMPD x (FMOVDconst [0])) => (FCMPD0 x)
320 (FCMPD (FMOVDconst [0]) x) => (InvertFlags (FCMPD0 x))
321
322 // CSEL needs a flag-generating argument. Synthesize a TSTW if necessary.
323 (CondSelect x y boolval) && flagArg(boolval) != nil => (CSEL [boolval.Op] x y flagArg(boolval))
324 (CondSelect x y boolval) && flagArg(boolval) == nil => (CSEL [OpARM64NotEqual] x y (TSTWconst [1] boolval))
325
326 (OffPtr [off] ptr:(SP)) && is32Bit(off) => (MOVDaddr [int32(off)] ptr)
327 (OffPtr [off] ptr) => (ADDconst [off] ptr)
328
329 (Addr {sym} base) => (MOVDaddr {sym} base)
330 (LocalAddr <t> {sym} base mem) && t.Elem().HasPointers() => (MOVDaddr {sym} (SPanchored base mem))
331 (LocalAddr <t> {sym} base _) && !t.Elem().HasPointers() => (MOVDaddr {sym} base)
332
333 // loads
334 (Load <t> ptr mem) && t.IsBoolean() => (MOVBUload ptr mem)
335 (Load <t> ptr mem) && (is8BitInt(t) && t.IsSigned()) => (MOVBload ptr mem)
336 (Load <t> ptr mem) && (is8BitInt(t) && !t.IsSigned()) => (MOVBUload ptr mem)
337 (Load <t> ptr mem) && (is16BitInt(t) && t.IsSigned()) => (MOVHload ptr mem)
338 (Load <t> ptr mem) && (is16BitInt(t) && !t.IsSigned()) => (MOVHUload ptr mem)
339 (Load <t> ptr mem) && (is32BitInt(t) && t.IsSigned()) => (MOVWload ptr mem)
340 (Load <t> ptr mem) && (is32BitInt(t) && !t.IsSigned()) => (MOVWUload ptr mem)
341 (Load <t> ptr mem) && (is64BitInt(t) || isPtr(t)) => (MOVDload ptr mem)
342 (Load <t> ptr mem) && is32BitFloat(t) => (FMOVSload ptr mem)
343 (Load <t> ptr mem) && is64BitFloat(t) => (FMOVDload ptr mem)
344
345 // stores
346 (Store {t} ptr val mem) && t.Size() == 1 => (MOVBstore ptr val mem)
347 (Store {t} ptr val mem) && t.Size() == 2 => (MOVHstore ptr val mem)
348 (Store {t} ptr val mem) && t.Size() == 4 && !t.IsFloat() => (MOVWstore ptr val mem)
349 (Store {t} ptr val mem) && t.Size() == 8 && !t.IsFloat() => (MOVDstore ptr val mem)
350 (Store {t} ptr val mem) && t.Size() == 4 && t.IsFloat() => (FMOVSstore ptr val mem)
351 (Store {t} ptr val mem) && t.Size() == 8 && t.IsFloat() => (FMOVDstore ptr val mem)
352
353 // zeroing
354 (Zero [0] _ mem) => mem
355 (Zero [1] ptr mem) => (MOVBstore ptr (MOVDconst [0]) mem)
356 (Zero [2] ptr mem) => (MOVHstore ptr (MOVDconst [0]) mem)
357 (Zero [4] ptr mem) => (MOVWstore ptr (MOVDconst [0]) mem)
358 (Zero [3] ptr mem) =>
359 (MOVBstore [2] ptr (MOVDconst [0])
360 (MOVHstore ptr (MOVDconst [0]) mem))
361 (Zero [5] ptr mem) =>
362 (MOVBstore [4] ptr (MOVDconst [0])
363 (MOVWstore ptr (MOVDconst [0]) mem))
364 (Zero [6] ptr mem) =>
365 (MOVHstore [4] ptr (MOVDconst [0])
366 (MOVWstore ptr (MOVDconst [0]) mem))
367 (Zero [7] ptr mem) =>
368 (MOVWstore [3] ptr (MOVDconst [0])
369 (MOVWstore ptr (MOVDconst [0]) mem))
370 (Zero [8] ptr mem) => (MOVDstore ptr (MOVDconst [0]) mem)
371 (Zero [9] ptr mem) =>
372 (MOVBstore [8] ptr (MOVDconst [0])
373 (MOVDstore ptr (MOVDconst [0]) mem))
374 (Zero [10] ptr mem) =>
375 (MOVHstore [8] ptr (MOVDconst [0])
376 (MOVDstore ptr (MOVDconst [0]) mem))
377 (Zero [11] ptr mem) =>
378 (MOVDstore [3] ptr (MOVDconst [0])
379 (MOVDstore ptr (MOVDconst [0]) mem))
380 (Zero [12] ptr mem) =>
381 (MOVWstore [8] ptr (MOVDconst [0])
382 (MOVDstore ptr (MOVDconst [0]) mem))
383 (Zero [13] ptr mem) =>
384 (MOVDstore [5] ptr (MOVDconst [0])
385 (MOVDstore ptr (MOVDconst [0]) mem))
386 (Zero [14] ptr mem) =>
387 (MOVDstore [6] ptr (MOVDconst [0])
388 (MOVDstore ptr (MOVDconst [0]) mem))
389 (Zero [15] ptr mem) =>
390 (MOVDstore [7] ptr (MOVDconst [0])
391 (MOVDstore ptr (MOVDconst [0]) mem))
392 (Zero [16] ptr mem) =>
393 (STP [0] ptr (MOVDconst [0]) (MOVDconst [0]) mem)
394
395 (Zero [s] ptr mem) && s > 16 && s < 192 => (LoweredZero [s] ptr mem)
396 (Zero [s] ptr mem) && s >= 192 => (LoweredZeroLoop [s] ptr mem)
397
398 // moves
399 (Move [0] _ _ mem) => mem
400 (Move [1] dst src mem) => (MOVBstore dst (MOVBUload src mem) mem)
401 (Move [2] dst src mem) => (MOVHstore dst (MOVHUload src mem) mem)
402 (Move [3] dst src mem) =>
403 (MOVBstore [2] dst (MOVBUload [2] src mem)
404 (MOVHstore dst (MOVHUload src mem) mem))
405 (Move [4] dst src mem) => (MOVWstore dst (MOVWUload src mem) mem)
406 (Move [5] dst src mem) =>
407 (MOVBstore [4] dst (MOVBUload [4] src mem)
408 (MOVWstore dst (MOVWUload src mem) mem))
409 (Move [6] dst src mem) =>
410 (MOVHstore [4] dst (MOVHUload [4] src mem)
411 (MOVWstore dst (MOVWUload src mem) mem))
412 (Move [7] dst src mem) =>
413 (MOVWstore [3] dst (MOVWUload [3] src mem)
414 (MOVWstore dst (MOVWUload src mem) mem))
415 (Move [8] dst src mem) => (MOVDstore dst (MOVDload src mem) mem)
416 (Move [9] dst src mem) =>
417 (MOVBstore [8] dst (MOVBUload [8] src mem)
418 (MOVDstore dst (MOVDload src mem) mem))
419 (Move [10] dst src mem) =>
420 (MOVHstore [8] dst (MOVHUload [8] src mem)
421 (MOVDstore dst (MOVDload src mem) mem))
422 (Move [11] dst src mem) =>
423 (MOVDstore [3] dst (MOVDload [3] src mem)
424 (MOVDstore dst (MOVDload src mem) mem))
425 (Move [12] dst src mem) =>
426 (MOVWstore [8] dst (MOVWUload [8] src mem)
427 (MOVDstore dst (MOVDload src mem) mem))
428 (Move [13] dst src mem) =>
429 (MOVDstore [5] dst (MOVDload [5] src mem)
430 (MOVDstore dst (MOVDload src mem) mem))
431 (Move [14] dst src mem) =>
432 (MOVDstore [6] dst (MOVDload [6] src mem)
433 (MOVDstore dst (MOVDload src mem) mem))
434 (Move [15] dst src mem) =>
435 (MOVDstore [7] dst (MOVDload [7] src mem)
436 (MOVDstore dst (MOVDload src mem) mem))
437 (Move [16] dst src mem) => (FMOVQstore dst (FMOVQload src mem) mem)
438
439 (Move [s] dst src mem) && s > 16 && s <= 24 =>
440 (MOVDstore [int32(s-8)] dst (MOVDload [int32(s-8)] src mem)
441 (FMOVQstore dst (FMOVQload src mem) mem))
442 (Move [s] dst src mem) && s > 24 && s < 32 =>
443 (FMOVQstore [int32(s-16)] dst (FMOVQload [int32(s-16)] src mem)
444 (FMOVQstore dst (FMOVQload src mem) mem))
445 (Move [32] dst src mem) =>
446 (FSTPQ dst (Select0 <typ.Vec128> (FLDPQ src mem)) (Select1 <typ.Vec128> (FLDPQ src mem)) mem)
447 (Move [s] dst src mem) && s > 32 && s <= 40 =>
448 (MOVDstore [int32(s-8)] dst (MOVDload [int32(s-8)] src mem)
449 (FSTPQ dst (Select0 <typ.Vec128> (FLDPQ src mem)) (Select1 <typ.Vec128> (FLDPQ src mem)) mem))
450 (Move [s] dst src mem) && s > 40 && s <= 48 =>
451 (FMOVQstore [int32(s-16)] dst (FMOVQload [int32(s-16)] src mem)
452 (FSTPQ dst (Select0 <typ.Vec128> (FLDPQ src mem)) (Select1 <typ.Vec128> (FLDPQ src mem)) mem))
453 (Move [s] dst src mem) && s > 48 && s <= 64 =>
454 (FSTPQ [int32(s-32)] dst (Select0 <typ.Vec128> (FLDPQ [int32(s-32)] src mem)) (Select1 <typ.Vec128> (FLDPQ [int32(s-32)] src mem))
455 (FSTPQ dst (Select0 <typ.Vec128> (FLDPQ src mem)) (Select1 <typ.Vec128> (FLDPQ src mem)) mem))
456
457 (Move [s] dst src mem) && s > 64 && s < 192 && logLargeCopy(v, s) => (LoweredMove [s] dst src mem)
458 (Move [s] dst src mem) && s >= 192 && logLargeCopy(v, s) => (LoweredMoveLoop [s] dst src mem)
459
460 // calls
461 (StaticCall ...) => (CALLstatic ...)
462 (ClosureCall ...) => (CALLclosure ...)
463 (InterCall ...) => (CALLinter ...)
464 (TailCall ...) => (CALLtail ...)
465 (TailCallInter ...) => (CALLtailinter ...)
466
467 // checks
468 (NilCheck ...) => (LoweredNilCheck ...)
469 (IsNonNil ptr) => (NotEqual (CMPconst [0] ptr))
470 (IsInBounds idx len) => (LessThanU (CMP idx len))
471 (IsSliceInBounds idx len) => (LessEqualU (CMP idx len))
472
473 // pseudo-ops
474 (GetClosurePtr ...) => (LoweredGetClosurePtr ...)
475 (GetCallerSP ...) => (LoweredGetCallerSP ...)
476 (GetCallerPC ...) => (LoweredGetCallerPC ...)
477 (MemEq ...) => (LoweredMemEq ...)
478
479 // Absorb pseudo-ops into blocks.
480 (If (Equal cc) yes no) => (EQ cc yes no)
481 (If (NotEqual cc) yes no) => (NE cc yes no)
482 (If (LessThan cc) yes no) => (LT cc yes no)
483 (If (LessThanU cc) yes no) => (ULT cc yes no)
484 (If (LessEqual cc) yes no) => (LE cc yes no)
485 (If (LessEqualU cc) yes no) => (ULE cc yes no)
486 (If (GreaterThan cc) yes no) => (GT cc yes no)
487 (If (GreaterThanU cc) yes no) => (UGT cc yes no)
488 (If (GreaterEqual cc) yes no) => (GE cc yes no)
489 (If (GreaterEqualU cc) yes no) => (UGE cc yes no)
490 (If (LessThanF cc) yes no) => (FLT cc yes no)
491 (If (LessEqualF cc) yes no) => (FLE cc yes no)
492 (If (GreaterThanF cc) yes no) => (FGT cc yes no)
493 (If (GreaterEqualF cc) yes no) => (FGE cc yes no)
494
495 (If cond yes no) => (TBNZ [0] cond yes no)
496
497 (JumpTable idx) => (JUMPTABLE {makeJumpTableSym(b)} idx (MOVDaddr <typ.Uintptr> {makeJumpTableSym(b)} (SB)))
498
499 // atomic intrinsics
500 // Note: these ops do not accept offset.
501 (AtomicLoad8 ...) => (LDARB ...)
502 (AtomicLoad32 ...) => (LDARW ...)
503 (AtomicLoad64 ...) => (LDAR ...)
504 (AtomicLoadPtr ...) => (LDAR ...)
505
506 (AtomicStore8 ...) => (STLRB ...)
507 (AtomicStore32 ...) => (STLRW ...)
508 (AtomicStore64 ...) => (STLR ...)
509 (AtomicStorePtrNoWB ...) => (STLR ...)
510
511 (AtomicExchange(8|32|64) ...) => (LoweredAtomicExchange(8|32|64) ...)
512 (AtomicAdd(32|64) ...) => (LoweredAtomicAdd(32|64) ...)
513 (AtomicCompareAndSwap(32|64) ...) => (LoweredAtomicCas(32|64) ...)
514
515 (AtomicAdd(32|64)Variant ...) => (LoweredAtomicAdd(32|64)Variant ...)
516 (AtomicExchange(8|32|64)Variant ...) => (LoweredAtomicExchange(8|32|64)Variant ...)
517 (AtomicCompareAndSwap(32|64)Variant ...) => (LoweredAtomicCas(32|64)Variant ...)
518
519 // Return old contents.
520 (AtomicAnd(64|32|8)value ...) => (LoweredAtomicAnd(64|32|8) ...)
521 (AtomicOr(64|32|8)value ...) => (LoweredAtomicOr(64|32|8) ...)
522 (AtomicAnd(64|32|8)valueVariant ...) => (LoweredAtomicAnd(64|32|8)Variant ...)
523 (AtomicOr(64|32|8)valueVariant ...) => (LoweredAtomicOr(64|32|8)Variant ...)
524
525 // Write barrier.
526 (WB ...) => (LoweredWB ...)
527
528 // Publication barrier (0xe is ST option)
529 (PubBarrier mem) => (DMB [0xe] mem)
530
531 (PanicBounds ...) => (LoweredPanicBoundsRR ...)
532 (LoweredPanicBoundsRR [kind] x (MOVDconst [c]) mem) => (LoweredPanicBoundsRC [kind] x {PanicBoundsC{C:c}} mem)
533 (LoweredPanicBoundsRR [kind] (MOVDconst [c]) y mem) => (LoweredPanicBoundsCR [kind] {PanicBoundsC{C:c}} y mem)
534 (LoweredPanicBoundsRC [kind] {p} (MOVDconst [c]) mem) => (LoweredPanicBoundsCC [kind] {PanicBoundsCC{Cx:c, Cy:p.C}} mem)
535 (LoweredPanicBoundsCR [kind] {p} (MOVDconst [c]) mem) => (LoweredPanicBoundsCC [kind] {PanicBoundsCC{Cx:p.C, Cy:c}} mem)
536
537 // Optimizations
538
539 // Replace widen -> wide_unop -> narrow with narrow_unop when one exists.
540 (FCVTDS (F(ABS|SQRT|RINTP|RINTM|RINTA|RINTN|RINTZ)D (FCVTSD x))) =>
541 (F(ABS|SQRT|RINTP|RINTM|RINTA|RINTN|RINTZ)S x)
542
543
544 // Absorb boolean tests into block
545 (NZ (Equal cc) yes no) => (EQ cc yes no)
546 (NZ (NotEqual cc) yes no) => (NE cc yes no)
547 (NZ (LessThan cc) yes no) => (LT cc yes no)
548 (NZ (LessThanU cc) yes no) => (ULT cc yes no)
549 (NZ (LessEqual cc) yes no) => (LE cc yes no)
550 (NZ (LessEqualU cc) yes no) => (ULE cc yes no)
551 (NZ (GreaterThan cc) yes no) => (GT cc yes no)
552 (NZ (GreaterThanU cc) yes no) => (UGT cc yes no)
553 (NZ (GreaterEqual cc) yes no) => (GE cc yes no)
554 (NZ (GreaterEqualU cc) yes no) => (UGE cc yes no)
555 (NZ (LessThanF cc) yes no) => (FLT cc yes no)
556 (NZ (LessEqualF cc) yes no) => (FLE cc yes no)
557 (NZ (GreaterThanF cc) yes no) => (FGT cc yes no)
558 (NZ (GreaterEqualF cc) yes no) => (FGE cc yes no)
559
560 (TBNZ [0] (Equal cc) yes no) => (EQ cc yes no)
561 (TBNZ [0] (NotEqual cc) yes no) => (NE cc yes no)
562 (TBNZ [0] (LessThan cc) yes no) => (LT cc yes no)
563 (TBNZ [0] (LessThanU cc) yes no) => (ULT cc yes no)
564 (TBNZ [0] (LessEqual cc) yes no) => (LE cc yes no)
565 (TBNZ [0] (LessEqualU cc) yes no) => (ULE cc yes no)
566 (TBNZ [0] (GreaterThan cc) yes no) => (GT cc yes no)
567 (TBNZ [0] (GreaterThanU cc) yes no) => (UGT cc yes no)
568 (TBNZ [0] (GreaterEqual cc) yes no) => (GE cc yes no)
569 (TBNZ [0] (GreaterEqualU cc) yes no) => (UGE cc yes no)
570 (TBNZ [0] (LessThanF cc) yes no) => (FLT cc yes no)
571 (TBNZ [0] (LessEqualF cc) yes no) => (FLE cc yes no)
572 (TBNZ [0] (GreaterThanF cc) yes no) => (FGT cc yes no)
573 (TBNZ [0] (GreaterEqualF cc) yes no) => (FGE cc yes no)
574
575 (TB(Z|NZ) [0] (XORconst [1] x) yes no) => (TB(NZ|Z) [0] x yes no)
576
577 ((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)
578 ((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)
579 ((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)
580 ((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)
581
582 // For conditional instructions such as CSET, CSEL.
583 ((Equal|NotEqual|LessThan|LessEqual|GreaterThan|GreaterEqual) (CMPconst [0] z:(AND x y))) && z.Uses == 1 =>
584 ((Equal|NotEqual|LessThan|LessEqual|GreaterThan|GreaterEqual) (TST x y))
585 ((Equal|NotEqual|LessThan|LessEqual|GreaterThan|GreaterEqual) (CMPWconst [0] x:(ANDconst [c] y))) && x.Uses == 1 =>
586 ((Equal|NotEqual|LessThan|LessEqual|GreaterThan|GreaterEqual) (TSTWconst [int32(c)] y))
587 ((Equal|NotEqual|LessThan|LessEqual|GreaterThan|GreaterEqual) (CMPWconst [0] z:(AND x y))) && z.Uses == 1 =>
588 ((Equal|NotEqual|LessThan|LessEqual|GreaterThan|GreaterEqual) (TSTW x y))
589 ((Equal|NotEqual|LessThan|LessEqual|GreaterThan|GreaterEqual) (CMPconst [0] x:(ANDconst [c] y))) && x.Uses == 1 =>
590 ((Equal|NotEqual|LessThan|LessEqual|GreaterThan|GreaterEqual) (TSTconst [c] y))
591
592 ((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)
593 ((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)
594 ((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)
595 ((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)
596
597 // CMP(x,-y) -> CMN(x,y) is only valid for unordered comparison, if y can be -1<<63
598 ((EQ|NE) (CMP x z:(NEG y)) yes no) && z.Uses == 1 => ((EQ|NE) (CMN x y) yes no)
599 ((Equal|NotEqual) (CMP x z:(NEG y))) && z.Uses == 1 => ((Equal|NotEqual) (CMN x y))
600
601 // CMPW(x,-y) -> CMNW(x,y) is only valid for unordered comparison, if y can be -1<<31
602 ((EQ|NE) (CMPW x z:(NEG y)) yes no) && z.Uses == 1 => ((EQ|NE) (CMNW x y) yes no)
603 ((Equal|NotEqual) (CMPW x z:(NEG y))) && z.Uses == 1 => ((Equal|NotEqual) (CMNW x y))
604
605 // For conditional instructions such as CSET, CSEL.
606 // TODO: add support for LE, GT, overflow needs to be considered.
607 ((Equal|NotEqual|LessThan|GreaterEqual) (CMPconst [0] x:(ADDconst [c] y))) && x.Uses == 1 => ((Equal|NotEqual|LessThanNoov|GreaterEqualNoov) (CMNconst [c] y))
608 ((Equal|NotEqual|LessThan|GreaterEqual) (CMPWconst [0] x:(ADDconst [c] y))) && x.Uses == 1 => ((Equal|NotEqual|LessThanNoov|GreaterEqualNoov) (CMNWconst [int32(c)] y))
609 ((Equal|NotEqual|LessThan|GreaterEqual) (CMPconst [0] z:(ADD x y))) && z.Uses == 1 => ((Equal|NotEqual|LessThanNoov|GreaterEqualNoov) (CMN x y))
610 ((Equal|NotEqual|LessThan|GreaterEqual) (CMPWconst [0] z:(ADD x y))) && z.Uses == 1 => ((Equal|NotEqual|LessThanNoov|GreaterEqualNoov) (CMNW x y))
611 ((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)))
612 ((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)))
613 ((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)))
614 ((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)))
615
616 ((CMPconst|CMNconst) [c] y) && c < 0 && c != -1<<63 => ((CMNconst|CMPconst) [-c] y)
617 ((CMPWconst|CMNWconst) [c] y) && c < 0 && c != -1<<31 => ((CMNWconst|CMPWconst) [-c] y)
618
619 ((EQ|NE) (CMPconst [0] x) yes no) => ((Z|NZ) x yes no)
620 ((EQ|NE) (CMPWconst [0] x) yes no) => ((ZW|NZW) x yes no)
621
622 ((ULE|UGT) (CMPconst [0] x)) => ((EQ|NE) (CMPconst [0] x))
623 ((ULE|UGT) (CMPWconst [0] x)) => ((EQ|NE) (CMPWconst [0] x))
624
625 ((Z|NZ) sub:(SUB x y)) && sub.Uses == 1 => ((EQ|NE) (CMP x y))
626 ((ZW|NZW) sub:(SUB x y)) && sub.Uses == 1 => ((EQ|NE) (CMPW x y))
627 ((Z|NZ) sub:(SUBconst [c] y)) && sub.Uses == 1 => ((EQ|NE) (CMPconst [c] y))
628 ((ZW|NZW) sub:(SUBconst [c] y)) && sub.Uses == 1 => ((EQ|NE) (CMPWconst [int32(c)] y))
629
630 ((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)
631 ((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)
632 ((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)
633 ((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)
634
635 // Absorb bit-tests into block
636 (Z (ANDconst [c] x) yes no) && oneBit(c) => (TBZ [int64(ntz64(c))] x yes no)
637 (NZ (ANDconst [c] x) yes no) && oneBit(c) => (TBNZ [int64(ntz64(c))] x yes no)
638 (ZW (ANDconst [c] x) yes no) && oneBit(int64(uint32(c))) => (TBZ [int64(ntz64(int64(uint32(c))))] x yes no)
639 (NZW (ANDconst [c] x) yes no) && oneBit(int64(uint32(c))) => (TBNZ [int64(ntz64(int64(uint32(c))))] x yes no)
640 (EQ (TSTconst [c] x) yes no) && oneBit(c) => (TBZ [int64(ntz64(c))] x yes no)
641 (NE (TSTconst [c] x) yes no) && oneBit(c) => (TBNZ [int64(ntz64(c))] x yes no)
642 (EQ (TSTWconst [c] x) yes no) && oneBit(int64(uint32(c))) => (TBZ [int64(ntz64(int64(uint32(c))))] x yes no)
643 (NE (TSTWconst [c] x) yes no) && oneBit(int64(uint32(c))) => (TBNZ [int64(ntz64(int64(uint32(c))))] x yes no)
644
645 ((Z|NZ) s:(SRLconst [63] x) yes no) && s.Uses == 1 => (TB(Z|NZ) [63] x yes no)
646 ((Z|NZ) s:(SRAconst [63] x) yes no) && s.Uses == 1 => (TB(Z|NZ) [63] x yes no)
647
648 // Merge more operations into TBZ & TBNZ
649 (TB(Z|NZ) [t] sv:(SRLconst [s] x) yes no) && t+s < 64 && sv.Uses == 1 => (TB(Z|NZ) [t+s] x yes no )
650 (TBZ [t] (SRLconst [s] x) yes no) && t+s >= 64 => (First yes no )
651 (TBNZ [t] (SRLconst [s] x) yes no) && t+s >= 64 => (First no yes)
652
653 (TB(Z|NZ) [t] sv:(SLLconst [s] x) yes no) && t-s >= 0 && sv.Uses == 1 => (TB(Z|NZ) [t-s] x yes no )
654 (TBZ [t] (SLLconst [s] x) yes no) && t-s < 0 => (First yes no )
655 (TBNZ [t] (SLLconst [s] x) yes no) && t-s < 0 => (First no yes)
656
657 (TB(Z|NZ) [t] rv:(RORconst [r] x) yes no) && rv.Uses == 1 => (TB(Z|NZ) [int64(uint64(t+r)%64)] x yes no)
658
659 (TB(Z|NZ) [t] sv:(SRAconst [s] x) yes no) && t+s < 64 && sv.Uses == 1 => (TB(Z|NZ) [t+s] x yes no)
660 (TB(Z|NZ) [t] sv:(SRAconst [s] x) yes no) && t+s >= 64 && sv.Uses == 1 => (TB(Z|NZ) [63 ] x yes no)
661
662 // Test sign-bit for signed comparisons against zero
663 (GE (CMPWconst [0] x) yes no) => (TBZ [31] x yes no)
664 (GE (CMPconst [0] x) yes no) => (TBZ [63] x yes no)
665 (LT (CMPWconst [0] x) yes no) => (TBNZ [31] x yes no)
666 (LT (CMPconst [0] x) yes no) => (TBNZ [63] x yes no)
667
668 // fold offset into address
669 (ADDconst [off1] (MOVDaddr [off2] {sym} ptr)) && is32Bit(off1+int64(off2)) =>
670 (MOVDaddr [int32(off1)+off2] {sym} ptr)
671
672 // fold address into load/store.
673 // Do not fold global variable access in -dynlink mode, where it will
674 // be rewritten to use the GOT via REGTMP, which currently cannot handle
675 // large offset.
676 (MOVBload [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
677 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
678 (MOVBload [off1+int32(off2)] {sym} ptr mem)
679 (MOVBUload [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
680 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
681 (MOVBUload [off1+int32(off2)] {sym} ptr mem)
682 (MOVHload [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
683 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
684 (MOVHload [off1+int32(off2)] {sym} ptr mem)
685 (MOVHUload [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
686 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
687 (MOVHUload [off1+int32(off2)] {sym} ptr mem)
688 (MOVWload [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
689 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
690 (MOVWload [off1+int32(off2)] {sym} ptr mem)
691 (MOVWUload [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
692 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
693 (MOVWUload [off1+int32(off2)] {sym} ptr mem)
694 (MOVDload [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
695 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
696 (MOVDload [off1+int32(off2)] {sym} ptr mem)
697 (LDP [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
698 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
699 (LDP [off1+int32(off2)] {sym} ptr mem)
700 (FMOVSload [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
701 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
702 (FMOVSload [off1+int32(off2)] {sym} ptr mem)
703 (FMOVDload [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
704 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
705 (FMOVDload [off1+int32(off2)] {sym} ptr mem)
706 (FMOVQload [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
707 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
708 (FMOVQload [off1+int32(off2)] {sym} ptr mem)
709 (FLDPQ [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
710 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
711 (FLDPQ [off1+int32(off2)] {sym} ptr mem)
712
713 // register indexed load
714 (MOVDload [off] {sym} (ADD ptr idx) mem) && off == 0 && sym == nil => (MOVDloadidx ptr idx mem)
715 (MOVWUload [off] {sym} (ADD ptr idx) mem) && off == 0 && sym == nil => (MOVWUloadidx ptr idx mem)
716 (MOVWload [off] {sym} (ADD ptr idx) mem) && off == 0 && sym == nil => (MOVWloadidx ptr idx mem)
717 (MOVHUload [off] {sym} (ADD ptr idx) mem) && off == 0 && sym == nil => (MOVHUloadidx ptr idx mem)
718 (MOVHload [off] {sym} (ADD ptr idx) mem) && off == 0 && sym == nil => (MOVHloadidx ptr idx mem)
719 (MOVBUload [off] {sym} (ADD ptr idx) mem) && off == 0 && sym == nil => (MOVBUloadidx ptr idx mem)
720 (MOVBload [off] {sym} (ADD ptr idx) mem) && off == 0 && sym == nil => (MOVBloadidx ptr idx mem)
721 (FMOVSload [off] {sym} (ADD ptr idx) mem) && off == 0 && sym == nil => (FMOVSloadidx ptr idx mem)
722 (FMOVDload [off] {sym} (ADD ptr idx) mem) && off == 0 && sym == nil => (FMOVDloadidx ptr idx mem)
723
724 (MOVDloadidx ptr (MOVDconst [c]) mem) && is32Bit(c) => (MOVDload [int32(c)] ptr mem)
725 (MOVDloadidx (MOVDconst [c]) ptr mem) && is32Bit(c) => (MOVDload [int32(c)] ptr mem)
726 (MOVWUloadidx ptr (MOVDconst [c]) mem) && is32Bit(c) => (MOVWUload [int32(c)] ptr mem)
727 (MOVWUloadidx (MOVDconst [c]) ptr mem) && is32Bit(c) => (MOVWUload [int32(c)] ptr mem)
728 (MOVWloadidx ptr (MOVDconst [c]) mem) && is32Bit(c) => (MOVWload [int32(c)] ptr mem)
729 (MOVWloadidx (MOVDconst [c]) ptr mem) && is32Bit(c) => (MOVWload [int32(c)] ptr mem)
730 (MOVHUloadidx ptr (MOVDconst [c]) mem) && is32Bit(c) => (MOVHUload [int32(c)] ptr mem)
731 (MOVHUloadidx (MOVDconst [c]) ptr mem) && is32Bit(c) => (MOVHUload [int32(c)] ptr mem)
732 (MOVHloadidx ptr (MOVDconst [c]) mem) && is32Bit(c) => (MOVHload [int32(c)] ptr mem)
733 (MOVHloadidx (MOVDconst [c]) ptr mem) && is32Bit(c) => (MOVHload [int32(c)] ptr mem)
734 (MOVBUloadidx ptr (MOVDconst [c]) mem) && is32Bit(c) => (MOVBUload [int32(c)] ptr mem)
735 (MOVBUloadidx (MOVDconst [c]) ptr mem) && is32Bit(c) => (MOVBUload [int32(c)] ptr mem)
736 (MOVBloadidx ptr (MOVDconst [c]) mem) && is32Bit(c) => (MOVBload [int32(c)] ptr mem)
737 (MOVBloadidx (MOVDconst [c]) ptr mem) && is32Bit(c) => (MOVBload [int32(c)] ptr mem)
738 (FMOVSloadidx ptr (MOVDconst [c]) mem) && is32Bit(c) => (FMOVSload [int32(c)] ptr mem)
739 (FMOVSloadidx (MOVDconst [c]) ptr mem) && is32Bit(c) => (FMOVSload [int32(c)] ptr mem)
740 (FMOVDloadidx ptr (MOVDconst [c]) mem) && is32Bit(c) => (FMOVDload [int32(c)] ptr mem)
741 (FMOVDloadidx (MOVDconst [c]) ptr mem) && is32Bit(c) => (FMOVDload [int32(c)] ptr mem)
742
743 // shifted register indexed load
744 (MOVDload [off] {sym} (ADDshiftLL [3] ptr idx) mem) && off == 0 && sym == nil => (MOVDloadidx8 ptr idx mem)
745 (MOVWUload [off] {sym} (ADDshiftLL [2] ptr idx) mem) && off == 0 && sym == nil => (MOVWUloadidx4 ptr idx mem)
746 (MOVWload [off] {sym} (ADDshiftLL [2] ptr idx) mem) && off == 0 && sym == nil => (MOVWloadidx4 ptr idx mem)
747 (MOVHUload [off] {sym} (ADDshiftLL [1] ptr idx) mem) && off == 0 && sym == nil => (MOVHUloadidx2 ptr idx mem)
748 (MOVHload [off] {sym} (ADDshiftLL [1] ptr idx) mem) && off == 0 && sym == nil => (MOVHloadidx2 ptr idx mem)
749 (MOVDloadidx ptr (SLLconst [3] idx) mem) => (MOVDloadidx8 ptr idx mem)
750 (MOVWloadidx ptr (SLLconst [2] idx) mem) => (MOVWloadidx4 ptr idx mem)
751 (MOVWUloadidx ptr (SLLconst [2] idx) mem) => (MOVWUloadidx4 ptr idx mem)
752 (MOVHloadidx ptr (SLLconst [1] idx) mem) => (MOVHloadidx2 ptr idx mem)
753 (MOVHUloadidx ptr (SLLconst [1] idx) mem) => (MOVHUloadidx2 ptr idx mem)
754 (MOVHloadidx ptr (ADD idx idx) mem) => (MOVHloadidx2 ptr idx mem)
755 (MOVHUloadidx ptr (ADD idx idx) mem) => (MOVHUloadidx2 ptr idx mem)
756 (MOVDloadidx (SLLconst [3] idx) ptr mem) => (MOVDloadidx8 ptr idx mem)
757 (MOVWloadidx (SLLconst [2] idx) ptr mem) => (MOVWloadidx4 ptr idx mem)
758 (MOVWUloadidx (SLLconst [2] idx) ptr mem) => (MOVWUloadidx4 ptr idx mem)
759 (MOVHloadidx (ADD idx idx) ptr mem) => (MOVHloadidx2 ptr idx mem)
760 (MOVHUloadidx (ADD idx idx) ptr mem) => (MOVHUloadidx2 ptr idx mem)
761 (MOVDloadidx8 ptr (MOVDconst [c]) mem) && is32Bit(c<<3) => (MOVDload [int32(c)<<3] ptr mem)
762 (MOVWUloadidx4 ptr (MOVDconst [c]) mem) && is32Bit(c<<2) => (MOVWUload [int32(c)<<2] ptr mem)
763 (MOVWloadidx4 ptr (MOVDconst [c]) mem) && is32Bit(c<<2) => (MOVWload [int32(c)<<2] ptr mem)
764 (MOVHUloadidx2 ptr (MOVDconst [c]) mem) && is32Bit(c<<1) => (MOVHUload [int32(c)<<1] ptr mem)
765 (MOVHloadidx2 ptr (MOVDconst [c]) mem) && is32Bit(c<<1) => (MOVHload [int32(c)<<1] ptr mem)
766
767 (FMOVDload [off] {sym} (ADDshiftLL [3] ptr idx) mem) && off == 0 && sym == nil => (FMOVDloadidx8 ptr idx mem)
768 (FMOVSload [off] {sym} (ADDshiftLL [2] ptr idx) mem) && off == 0 && sym == nil => (FMOVSloadidx4 ptr idx mem)
769 (FMOVDloadidx ptr (SLLconst [3] idx) mem) => (FMOVDloadidx8 ptr idx mem)
770 (FMOVSloadidx ptr (SLLconst [2] idx) mem) => (FMOVSloadidx4 ptr idx mem)
771 (FMOVDloadidx (SLLconst [3] idx) ptr mem) => (FMOVDloadidx8 ptr idx mem)
772 (FMOVSloadidx (SLLconst [2] idx) ptr mem) => (FMOVSloadidx4 ptr idx mem)
773 (FMOVDloadidx8 ptr (MOVDconst [c]) mem) && is32Bit(c<<3) => (FMOVDload ptr [int32(c)<<3] mem)
774 (FMOVSloadidx4 ptr (MOVDconst [c]) mem) && is32Bit(c<<2) => (FMOVSload ptr [int32(c)<<2] mem)
775
776 (MOVBstore [off1] {sym} (ADDconst [off2] ptr) val mem) && is32Bit(int64(off1)+off2)
777 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
778 (MOVBstore [off1+int32(off2)] {sym} ptr val mem)
779 (MOVHstore [off1] {sym} (ADDconst [off2] ptr) val mem) && is32Bit(int64(off1)+off2)
780 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
781 (MOVHstore [off1+int32(off2)] {sym} ptr val mem)
782 (MOVWstore [off1] {sym} (ADDconst [off2] ptr) val mem) && is32Bit(int64(off1)+off2)
783 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
784 (MOVWstore [off1+int32(off2)] {sym} ptr val mem)
785 (MOVDstore [off1] {sym} (ADDconst [off2] ptr) val mem) && is32Bit(int64(off1)+off2)
786 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
787 (MOVDstore [off1+int32(off2)] {sym} ptr val mem)
788 (STP [off1] {sym} (ADDconst [off2] ptr) val1 val2 mem) && is32Bit(int64(off1)+off2)
789 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
790 (STP [off1+int32(off2)] {sym} ptr val1 val2 mem)
791 (FMOVSstore [off1] {sym} (ADDconst [off2] ptr) val mem) && is32Bit(int64(off1)+off2)
792 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
793 (FMOVSstore [off1+int32(off2)] {sym} ptr val mem)
794 (FMOVDstore [off1] {sym} (ADDconst [off2] ptr) val mem) && is32Bit(int64(off1)+off2)
795 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
796 (FMOVDstore [off1+int32(off2)] {sym} ptr val mem)
797 (FMOVQstore [off1] {sym} (ADDconst [off2] ptr) val mem) && is32Bit(int64(off1)+off2)
798 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
799 (FMOVQstore [off1+int32(off2)] {sym} ptr val mem)
800 (FSTPQ [off1] {sym} (ADDconst [off2] ptr) val1 val2 mem) && is32Bit(int64(off1)+off2)
801 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
802 (FSTPQ [off1+int32(off2)] {sym} ptr val1 val2 mem)
803
804 // register indexed store
805 (MOVDstore [off] {sym} (ADD ptr idx) val mem) && off == 0 && sym == nil => (MOVDstoreidx ptr idx val mem)
806 (MOVWstore [off] {sym} (ADD ptr idx) val mem) && off == 0 && sym == nil => (MOVWstoreidx ptr idx val mem)
807 (MOVHstore [off] {sym} (ADD ptr idx) val mem) && off == 0 && sym == nil => (MOVHstoreidx ptr idx val mem)
808 (MOVBstore [off] {sym} (ADD ptr idx) val mem) && off == 0 && sym == nil => (MOVBstoreidx ptr idx val mem)
809 (FMOVDstore [off] {sym} (ADD ptr idx) val mem) && off == 0 && sym == nil => (FMOVDstoreidx ptr idx val mem)
810 (FMOVSstore [off] {sym} (ADD ptr idx) val mem) && off == 0 && sym == nil => (FMOVSstoreidx ptr idx val mem)
811 (MOVDstoreidx ptr (MOVDconst [c]) val mem) && is32Bit(c) => (MOVDstore [int32(c)] ptr val mem)
812 (MOVDstoreidx (MOVDconst [c]) idx val mem) && is32Bit(c) => (MOVDstore [int32(c)] idx val mem)
813 (MOVWstoreidx ptr (MOVDconst [c]) val mem) && is32Bit(c) => (MOVWstore [int32(c)] ptr val mem)
814 (MOVWstoreidx (MOVDconst [c]) idx val mem) && is32Bit(c) => (MOVWstore [int32(c)] idx val mem)
815 (MOVHstoreidx ptr (MOVDconst [c]) val mem) && is32Bit(c) => (MOVHstore [int32(c)] ptr val mem)
816 (MOVHstoreidx (MOVDconst [c]) idx val mem) && is32Bit(c) => (MOVHstore [int32(c)] idx val mem)
817 (MOVBstoreidx ptr (MOVDconst [c]) val mem) && is32Bit(c) => (MOVBstore [int32(c)] ptr val mem)
818 (MOVBstoreidx (MOVDconst [c]) idx val mem) && is32Bit(c) => (MOVBstore [int32(c)] idx val mem)
819 (FMOVDstoreidx ptr (MOVDconst [c]) val mem) && is32Bit(c) => (FMOVDstore [int32(c)] ptr val mem)
820 (FMOVDstoreidx (MOVDconst [c]) idx val mem) && is32Bit(c) => (FMOVDstore [int32(c)] idx val mem)
821 (FMOVSstoreidx ptr (MOVDconst [c]) val mem) && is32Bit(c) => (FMOVSstore [int32(c)] ptr val mem)
822 (FMOVSstoreidx (MOVDconst [c]) idx val mem) && is32Bit(c) => (FMOVSstore [int32(c)] idx val mem)
823
824 // shifted register indexed store
825 (MOVDstore [off] {sym} (ADDshiftLL [3] ptr idx) val mem) && off == 0 && sym == nil => (MOVDstoreidx8 ptr idx val mem)
826 (MOVWstore [off] {sym} (ADDshiftLL [2] ptr idx) val mem) && off == 0 && sym == nil => (MOVWstoreidx4 ptr idx val mem)
827 (MOVHstore [off] {sym} (ADDshiftLL [1] ptr idx) val mem) && off == 0 && sym == nil => (MOVHstoreidx2 ptr idx val mem)
828 (MOVDstoreidx ptr (SLLconst [3] idx) val mem) => (MOVDstoreidx8 ptr idx val mem)
829 (MOVWstoreidx ptr (SLLconst [2] idx) val mem) => (MOVWstoreidx4 ptr idx val mem)
830 (MOVHstoreidx ptr (SLLconst [1] idx) val mem) => (MOVHstoreidx2 ptr idx val mem)
831 (MOVHstoreidx ptr (ADD idx idx) val mem) => (MOVHstoreidx2 ptr idx val mem)
832 (MOVDstoreidx (SLLconst [3] idx) ptr val mem) => (MOVDstoreidx8 ptr idx val mem)
833 (MOVWstoreidx (SLLconst [2] idx) ptr val mem) => (MOVWstoreidx4 ptr idx val mem)
834 (MOVHstoreidx (SLLconst [1] idx) ptr val mem) => (MOVHstoreidx2 ptr idx val mem)
835 (MOVHstoreidx (ADD idx idx) ptr val mem) => (MOVHstoreidx2 ptr idx val mem)
836 (MOVDstoreidx8 ptr (MOVDconst [c]) val mem) && is32Bit(c<<3) => (MOVDstore [int32(c)<<3] ptr val mem)
837 (MOVWstoreidx4 ptr (MOVDconst [c]) val mem) && is32Bit(c<<2) => (MOVWstore [int32(c)<<2] ptr val mem)
838 (MOVHstoreidx2 ptr (MOVDconst [c]) val mem) && is32Bit(c<<1) => (MOVHstore [int32(c)<<1] ptr val mem)
839
840 (FMOVDstore [off] {sym} (ADDshiftLL [3] ptr idx) val mem) && off == 0 && sym == nil => (FMOVDstoreidx8 ptr idx val mem)
841 (FMOVSstore [off] {sym} (ADDshiftLL [2] ptr idx) val mem) && off == 0 && sym == nil => (FMOVSstoreidx4 ptr idx val mem)
842 (FMOVDstoreidx ptr (SLLconst [3] idx) val mem) => (FMOVDstoreidx8 ptr idx val mem)
843 (FMOVSstoreidx ptr (SLLconst [2] idx) val mem) => (FMOVSstoreidx4 ptr idx val mem)
844 (FMOVDstoreidx (SLLconst [3] idx) ptr val mem) => (FMOVDstoreidx8 ptr idx val mem)
845 (FMOVSstoreidx (SLLconst [2] idx) ptr val mem) => (FMOVSstoreidx4 ptr idx val mem)
846 (FMOVDstoreidx8 ptr (MOVDconst [c]) val mem) && is32Bit(c<<3) => (FMOVDstore [int32(c)<<3] ptr val mem)
847 (FMOVSstoreidx4 ptr (MOVDconst [c]) val mem) && is32Bit(c<<2) => (FMOVSstore [int32(c)<<2] ptr val mem)
848
849 (MOVBload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
850 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
851 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
852 (MOVBload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
853 (MOVBUload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
854 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
855 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
856 (MOVBUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
857 (MOVHload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
858 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
859 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
860 (MOVHload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
861 (MOVHUload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
862 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
863 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
864 (MOVHUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
865 (MOVWload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
866 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
867 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
868 (MOVWload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
869 (MOVWUload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
870 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
871 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
872 (MOVWUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
873 (MOVDload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
874 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
875 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
876 (MOVDload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
877 (LDP [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
878 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
879 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
880 (LDP [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
881 (FMOVSload [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 (FMOVSload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
885 (FMOVDload [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 (FMOVDload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
889 (FMOVQload [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 (FMOVQload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
893 (FLDPQ [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 (FLDPQ [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
897
898 (MOVBstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem)
899 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
900 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
901 (MOVBstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
902 (MOVHstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem)
903 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
904 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
905 (MOVHstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
906 (MOVWstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem)
907 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
908 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
909 (MOVWstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
910 (MOVDstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem)
911 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
912 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
913 (MOVDstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
914 (STP [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val1 val2 mem)
915 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
916 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
917 (STP [off1+off2] {mergeSym(sym1,sym2)} ptr val1 val2 mem)
918 (FMOVSstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem)
919 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
920 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
921 (FMOVSstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
922 (FMOVDstore [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 (FMOVDstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
926 (FMOVQstore [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 (FMOVQstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
930 (FSTPQ [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val1 val2 mem)
931 && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
932 && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
933 (FSTPQ [off1+off2] {mergeSym(sym1,sym2)} ptr val1 val2 mem)
934
935 // replace load from same location as preceding store with zero/sign extension (or copy in case of full width)
936 // these seem to have bad interaction with other rules, resulting in slower code
937 //(MOVBload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVBreg x)
938 //(MOVBUload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVBUreg x)
939 //(MOVHload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVHreg x)
940 //(MOVHUload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVHUreg x)
941 //(MOVWload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVWreg x)
942 //(MOVWUload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVWUreg x)
943 //(MOVDload [off] {sym} ptr (MOVDstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
944 //(FMOVSload [off] {sym} ptr (FMOVSstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
945 //(FMOVDload [off] {sym} ptr (FMOVDstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
946 //(LDP [off] {sym} ptr (STP [off2] {sym2} ptr2 x y _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x y
947
948 // don't extend before store
949 (MOVBstore [off] {sym} ptr (MOVBreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
950 (MOVBstore [off] {sym} ptr (MOVBUreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
951 (MOVBstore [off] {sym} ptr (MOVHreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
952 (MOVBstore [off] {sym} ptr (MOVHUreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
953 (MOVBstore [off] {sym} ptr (MOVWreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
954 (MOVBstore [off] {sym} ptr (MOVWUreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
955 (MOVHstore [off] {sym} ptr (MOVHreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
956 (MOVHstore [off] {sym} ptr (MOVHUreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
957 (MOVHstore [off] {sym} ptr (MOVWreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
958 (MOVHstore [off] {sym} ptr (MOVWUreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
959 (MOVWstore [off] {sym} ptr (MOVWreg x) mem) => (MOVWstore [off] {sym} ptr x mem)
960 (MOVWstore [off] {sym} ptr (MOVWUreg x) mem) => (MOVWstore [off] {sym} ptr x mem)
961 (MOVBstoreidx ptr idx (MOVBreg x) mem) => (MOVBstoreidx ptr idx x mem)
962 (MOVBstoreidx ptr idx (MOVBUreg x) mem) => (MOVBstoreidx ptr idx x mem)
963 (MOVBstoreidx ptr idx (MOVHreg x) mem) => (MOVBstoreidx ptr idx x mem)
964 (MOVBstoreidx ptr idx (MOVHUreg x) mem) => (MOVBstoreidx ptr idx x mem)
965 (MOVBstoreidx ptr idx (MOVWreg x) mem) => (MOVBstoreidx ptr idx x mem)
966 (MOVBstoreidx ptr idx (MOVWUreg x) mem) => (MOVBstoreidx ptr idx x mem)
967 (MOVHstoreidx ptr idx (MOVHreg x) mem) => (MOVHstoreidx ptr idx x mem)
968 (MOVHstoreidx ptr idx (MOVHUreg x) mem) => (MOVHstoreidx ptr idx x mem)
969 (MOVHstoreidx ptr idx (MOVWreg x) mem) => (MOVHstoreidx ptr idx x mem)
970 (MOVHstoreidx ptr idx (MOVWUreg x) mem) => (MOVHstoreidx ptr idx x mem)
971 (MOVWstoreidx ptr idx (MOVWreg x) mem) => (MOVWstoreidx ptr idx x mem)
972 (MOVWstoreidx ptr idx (MOVWUreg x) mem) => (MOVWstoreidx ptr idx x mem)
973 (MOVHstoreidx2 ptr idx (MOVHreg x) mem) => (MOVHstoreidx2 ptr idx x mem)
974 (MOVHstoreidx2 ptr idx (MOVHUreg x) mem) => (MOVHstoreidx2 ptr idx x mem)
975 (MOVHstoreidx2 ptr idx (MOVWreg x) mem) => (MOVHstoreidx2 ptr idx x mem)
976 (MOVHstoreidx2 ptr idx (MOVWUreg x) mem) => (MOVHstoreidx2 ptr idx x mem)
977 (MOVWstoreidx4 ptr idx (MOVWreg x) mem) => (MOVWstoreidx4 ptr idx x mem)
978 (MOVWstoreidx4 ptr idx (MOVWUreg x) mem) => (MOVWstoreidx4 ptr idx x mem)
979
980 // if a register move has only 1 use, just use the same register without emitting instruction
981 // MOVDnop doesn't emit instruction, only for ensuring the type.
982 (MOVDreg x) && x.Uses == 1 => (MOVDnop x)
983
984 // TODO: we should be able to get rid of MOVDnop all together.
985 // But for now, this is enough to get rid of lots of them.
986 (MOVDnop (MOVDconst [c])) => (MOVDconst [c])
987
988 // fold constant into arithmetic ops
989 (ADD x (MOVDconst <t> [c])) && !t.IsPtr() => (ADDconst [c] x)
990 (SUB x (MOVDconst [c])) => (SUBconst [c] x)
991 (AND x (MOVDconst [c])) => (ANDconst [c] x)
992 (OR x (MOVDconst [c])) => (ORconst [c] x)
993 (XOR x (MOVDconst [c])) => (XORconst [c] x)
994 (TST x (MOVDconst [c])) => (TSTconst [c] x)
995 (TSTW x (MOVDconst [c])) => (TSTWconst [int32(c)] x)
996 (CMN x (MOVDconst [c])) => (CMNconst [c] x)
997 (CMNW x (MOVDconst [c])) => (CMNWconst [int32(c)] x)
998 (BIC x (MOVDconst [c])) => (ANDconst [^c] x)
999 (EON x (MOVDconst [c])) => (XORconst [^c] x)
1000 (ORN x (MOVDconst [c])) => (ORconst [^c] x)
1001
1002 (SLL x (MOVDconst [c])) => (SLLconst x [c&63])
1003 (SRL x (MOVDconst [c])) => (SRLconst x [c&63])
1004 (SRA x (MOVDconst [c])) => (SRAconst x [c&63])
1005 (SLL x (ANDconst [63] y)) => (SLL x y)
1006 (SRL x (ANDconst [63] y)) => (SRL x y)
1007 (SRA x (ANDconst [63] y)) => (SRA x y)
1008
1009 (CMP x (MOVDconst [c])) => (CMPconst [c] x)
1010 (CMP (MOVDconst [c]) x) => (InvertFlags (CMPconst [c] x))
1011 (CMPW x (MOVDconst [c])) => (CMPWconst [int32(c)] x)
1012 (CMPW (MOVDconst [c]) x) => (InvertFlags (CMPWconst [int32(c)] x))
1013
1014 (ROR x (MOVDconst [c])) => (RORconst x [c&63])
1015 (RORW x (MOVDconst [c])) => (RORWconst x [c&31])
1016
1017 (ADDSflags x (MOVDconst [c])) => (ADDSconstflags [c] x)
1018
1019 (ADDconst [c] y) && c < 0 => (SUBconst [-c] y)
1020
1021 // Canonicalize the order of arguments to comparisons - helps with CSE.
1022 ((CMP|CMPW) x y) && canonLessThan(x,y) => (InvertFlags ((CMP|CMPW) y x))
1023
1024 // mul-neg => mneg
1025 (NEG (MUL x y)) => (MNEG x y)
1026 (NEG (MULW x y)) && v.Type.Size() <= 4 => (MNEGW x y)
1027 (MUL (NEG x) y) => (MNEG x y)
1028 (MULW (NEG x) y) => (MNEGW x y)
1029
1030 // madd/msub
1031 (ADD a l:(MUL x y)) && l.Uses==1 && clobber(l) => (MADD a x y)
1032 (SUB a l:(MUL x y)) && l.Uses==1 && clobber(l) => (MSUB a x y)
1033 (ADD a l:(MNEG x y)) && l.Uses==1 && clobber(l) => (MSUB a x y)
1034 (SUB a l:(MNEG x y)) && l.Uses==1 && clobber(l) => (MADD a x y)
1035
1036 (ADD a l:(MULW x y)) && v.Type.Size() <= 4 && l.Uses==1 && clobber(l) => (MADDW a x y)
1037 (SUB a l:(MULW x y)) && v.Type.Size() <= 4 && l.Uses==1 && clobber(l) => (MSUBW a x y)
1038 (ADD a l:(MNEGW x y)) && v.Type.Size() <= 4 && l.Uses==1 && clobber(l) => (MSUBW a x y)
1039 (SUB a l:(MNEGW x y)) && v.Type.Size() <= 4 && l.Uses==1 && clobber(l) => (MADDW a x y)
1040
1041 // madd/msub can't take constant arguments, so do a bit of reordering if a non-constant is available.
1042 // Note: don't reorder arithmetic concerning pointers, as we must ensure that
1043 // no intermediate computations are invalid pointers.
1044 (ADD <t> a p:(ADDconst [c] m:((MUL|MULW|MNEG|MNEGW) _ _))) && p.Uses==1 && m.Uses==1 && !t.IsPtrShaped() => (ADDconst [c] (ADD <v.Type> a m))
1045 (ADD <t> a p:(SUBconst [c] m:((MUL|MULW|MNEG|MNEGW) _ _))) && p.Uses==1 && m.Uses==1 && !t.IsPtrShaped() => (SUBconst [c] (ADD <v.Type> a m))
1046 (SUB <t> a p:(ADDconst [c] m:((MUL|MULW|MNEG|MNEGW) _ _))) && p.Uses==1 && m.Uses==1 && !t.IsPtrShaped() => (SUBconst [c] (SUB <v.Type> a m))
1047 (SUB <t> a p:(SUBconst [c] m:((MUL|MULW|MNEG|MNEGW) _ _))) && p.Uses==1 && m.Uses==1 && !t.IsPtrShaped() => (ADDconst [c] (SUB <v.Type> a m))
1048
1049 // optimize ADCSflags, SBCSflags and friends
1050 (ADCSflags x y (Select1 <types.TypeFlags> (ADDSconstflags [-1] (ADCzerocarry <typ.UInt64> c)))) => (ADCSflags x y c)
1051 (ADCSflags x y (Select1 <types.TypeFlags> (ADDSconstflags [-1] (MOVDconst [0])))) => (ADDSflags x y)
1052 (SBCSflags x y (Select1 <types.TypeFlags> (NEGSflags (NEG <typ.UInt64> (NGCzerocarry <typ.UInt64> bo))))) => (SBCSflags x y bo)
1053 (SBCSflags x y (Select1 <types.TypeFlags> (NEGSflags (MOVDconst [0])))) => (SUBSflags x y)
1054
1055 // mul by constant
1056 (MUL _ (MOVDconst [0])) => (MOVDconst [0])
1057 (MUL x (MOVDconst [1])) => x
1058
1059 (MULW _ (MOVDconst [c])) && int32(c)==0 => (MOVDconst [0])
1060 (MULW x (MOVDconst [c])) && int32(c)==1 => (MOVWUreg x)
1061
1062 (MUL x (MOVDconst [c])) && canMulStrengthReduce(config, c) => {mulStrengthReduce(v, x, c)}
1063 (MULW x (MOVDconst [c])) && v.Type.Size() <= 4 && canMulStrengthReduce32(config, int32(c)) => {mulStrengthReduce32(v, x, int32(c))}
1064
1065 // mneg by constant
1066 (MNEG x (MOVDconst [-1])) => x
1067 (MNEG _ (MOVDconst [0])) => (MOVDconst [0])
1068 (MNEG x (MOVDconst [1])) => (NEG x)
1069 (MNEG x (MOVDconst [c])) && isPowerOfTwo(c) => (NEG (SLLconst <x.Type> [log64(c)] x))
1070 (MNEG x (MOVDconst [c])) && isPowerOfTwo(c-1) && c >= 3 => (NEG (ADDshiftLL <x.Type> x x [log64(c-1)]))
1071 (MNEG x (MOVDconst [c])) && isPowerOfTwo(c+1) && c >= 7 => (NEG (ADDshiftLL <x.Type> (NEG <x.Type> x) x [log64(c+1)]))
1072 (MNEG x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo(c/3) => (SLLconst <x.Type> [log64(c/3)] (SUBshiftLL <x.Type> x x [2]))
1073 (MNEG x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo(c/5) => (NEG (SLLconst <x.Type> [log64(c/5)] (ADDshiftLL <x.Type> x x [2])))
1074 (MNEG x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo(c/7) => (SLLconst <x.Type> [log64(c/7)] (SUBshiftLL <x.Type> x x [3]))
1075 (MNEG x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo(c/9) => (NEG (SLLconst <x.Type> [log64(c/9)] (ADDshiftLL <x.Type> x x [3])))
1076
1077
1078 (MNEGW x (MOVDconst [c])) && int32(c)==-1 => (MOVWUreg x)
1079 (MNEGW _ (MOVDconst [c])) && int32(c)==0 => (MOVDconst [0])
1080 (MNEGW x (MOVDconst [c])) && int32(c)==1 => (MOVWUreg (NEG <x.Type> x))
1081 (MNEGW x (MOVDconst [c])) && isPowerOfTwo(c) => (NEG (SLLconst <x.Type> [log64(c)] x))
1082 (MNEGW x (MOVDconst [c])) && isPowerOfTwo(c-1) && int32(c) >= 3 => (MOVWUreg (NEG <x.Type> (ADDshiftLL <x.Type> x x [log64(c-1)])))
1083 (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)])))
1084 (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])))
1085 (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]))))
1086 (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])))
1087 (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]))))
1088
1089
1090 (MADD a x (MOVDconst [-1])) => (SUB a x)
1091 (MADD a _ (MOVDconst [0])) => a
1092 (MADD a x (MOVDconst [1])) => (ADD a x)
1093 (MADD a x (MOVDconst [c])) && isPowerOfTwo(c) => (ADDshiftLL a x [log64(c)])
1094 (MADD a x (MOVDconst [c])) && isPowerOfTwo(c-1) && c>=3 => (ADD a (ADDshiftLL <x.Type> x x [log64(c-1)]))
1095 (MADD a x (MOVDconst [c])) && isPowerOfTwo(c+1) && c>=7 => (SUB a (SUBshiftLL <x.Type> x x [log64(c+1)]))
1096 (MADD a x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo(c/3) => (SUBshiftLL a (SUBshiftLL <x.Type> x x [2]) [log64(c/3)])
1097 (MADD a x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo(c/5) => (ADDshiftLL a (ADDshiftLL <x.Type> x x [2]) [log64(c/5)])
1098 (MADD a x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo(c/7) => (SUBshiftLL a (SUBshiftLL <x.Type> x x [3]) [log64(c/7)])
1099 (MADD a x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo(c/9) => (ADDshiftLL a (ADDshiftLL <x.Type> x x [3]) [log64(c/9)])
1100
1101 (MADD a (MOVDconst [-1]) x) => (SUB a x)
1102 (MADD a (MOVDconst [0]) _) => a
1103 (MADD a (MOVDconst [1]) x) => (ADD a x)
1104 (MADD a (MOVDconst [c]) x) && isPowerOfTwo(c) => (ADDshiftLL a x [log64(c)])
1105 (MADD a (MOVDconst [c]) x) && isPowerOfTwo(c-1) && c>=3 => (ADD a (ADDshiftLL <x.Type> x x [log64(c-1)]))
1106 (MADD a (MOVDconst [c]) x) && isPowerOfTwo(c+1) && c>=7 => (SUB a (SUBshiftLL <x.Type> x x [log64(c+1)]))
1107 (MADD a (MOVDconst [c]) x) && c%3 == 0 && isPowerOfTwo(c/3) => (SUBshiftLL a (SUBshiftLL <x.Type> x x [2]) [log64(c/3)])
1108 (MADD a (MOVDconst [c]) x) && c%5 == 0 && isPowerOfTwo(c/5) => (ADDshiftLL a (ADDshiftLL <x.Type> x x [2]) [log64(c/5)])
1109 (MADD a (MOVDconst [c]) x) && c%7 == 0 && isPowerOfTwo(c/7) => (SUBshiftLL a (SUBshiftLL <x.Type> x x [3]) [log64(c/7)])
1110 (MADD a (MOVDconst [c]) x) && c%9 == 0 && isPowerOfTwo(c/9) => (ADDshiftLL a (ADDshiftLL <x.Type> x x [3]) [log64(c/9)])
1111
1112 (MADDW a x (MOVDconst [c])) && int32(c)==-1 => (MOVWUreg (SUB <a.Type> a x))
1113 (MADDW a _ (MOVDconst [c])) && int32(c)==0 => (MOVWUreg a)
1114 (MADDW a x (MOVDconst [c])) && int32(c)==1 => (MOVWUreg (ADD <a.Type> a x))
1115 (MADDW a x (MOVDconst [c])) && isPowerOfTwo(c) => (MOVWUreg (ADDshiftLL <a.Type> a x [log64(c)]))
1116 (MADDW a x (MOVDconst [c])) && isPowerOfTwo(c-1) && int32(c)>=3 => (MOVWUreg (ADD <a.Type> a (ADDshiftLL <x.Type> x x [log64(c-1)])))
1117 (MADDW a x (MOVDconst [c])) && isPowerOfTwo(c+1) && int32(c)>=7 => (MOVWUreg (SUB <a.Type> a (SUBshiftLL <x.Type> x x [log64(c+1)])))
1118 (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)]))
1119 (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)]))
1120 (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)]))
1121 (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)]))
1122
1123 (MADDW a (MOVDconst [c]) x) && int32(c)==-1 => (MOVWUreg (SUB <a.Type> a x))
1124 (MADDW a (MOVDconst [c]) _) && int32(c)==0 => (MOVWUreg a)
1125 (MADDW a (MOVDconst [c]) x) && int32(c)==1 => (MOVWUreg (ADD <a.Type> a x))
1126 (MADDW a (MOVDconst [c]) x) && isPowerOfTwo(c) => (MOVWUreg (ADDshiftLL <a.Type> a x [log64(c)]))
1127 (MADDW a (MOVDconst [c]) x) && isPowerOfTwo(c-1) && int32(c)>=3 => (MOVWUreg (ADD <a.Type> a (ADDshiftLL <x.Type> x x [log64(c-1)])))
1128 (MADDW a (MOVDconst [c]) x) && isPowerOfTwo(c+1) && int32(c)>=7 => (MOVWUreg (SUB <a.Type> a (SUBshiftLL <x.Type> x x [log64(c+1)])))
1129 (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)]))
1130 (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)]))
1131 (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)]))
1132 (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)]))
1133
1134 (MSUB a x (MOVDconst [-1])) => (ADD a x)
1135 (MSUB a _ (MOVDconst [0])) => a
1136 (MSUB a x (MOVDconst [1])) => (SUB a x)
1137 (MSUB a x (MOVDconst [c])) && isPowerOfTwo(c) => (SUBshiftLL a x [log64(c)])
1138 (MSUB a x (MOVDconst [c])) && isPowerOfTwo(c-1) && c>=3 => (SUB a (ADDshiftLL <x.Type> x x [log64(c-1)]))
1139 (MSUB a x (MOVDconst [c])) && isPowerOfTwo(c+1) && c>=7 => (ADD a (SUBshiftLL <x.Type> x x [log64(c+1)]))
1140 (MSUB a x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo(c/3) => (ADDshiftLL a (SUBshiftLL <x.Type> x x [2]) [log64(c/3)])
1141 (MSUB a x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo(c/5) => (SUBshiftLL a (ADDshiftLL <x.Type> x x [2]) [log64(c/5)])
1142 (MSUB a x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo(c/7) => (ADDshiftLL a (SUBshiftLL <x.Type> x x [3]) [log64(c/7)])
1143 (MSUB a x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo(c/9) => (SUBshiftLL a (ADDshiftLL <x.Type> x x [3]) [log64(c/9)])
1144
1145 (MSUB a (MOVDconst [-1]) x) => (ADD a x)
1146 (MSUB a (MOVDconst [0]) _) => a
1147 (MSUB a (MOVDconst [1]) x) => (SUB a x)
1148 (MSUB a (MOVDconst [c]) x) && isPowerOfTwo(c) => (SUBshiftLL a x [log64(c)])
1149 (MSUB a (MOVDconst [c]) x) && isPowerOfTwo(c-1) && c>=3 => (SUB a (ADDshiftLL <x.Type> x x [log64(c-1)]))
1150 (MSUB a (MOVDconst [c]) x) && isPowerOfTwo(c+1) && c>=7 => (ADD a (SUBshiftLL <x.Type> x x [log64(c+1)]))
1151 (MSUB a (MOVDconst [c]) x) && c%3 == 0 && isPowerOfTwo(c/3) => (ADDshiftLL a (SUBshiftLL <x.Type> x x [2]) [log64(c/3)])
1152 (MSUB a (MOVDconst [c]) x) && c%5 == 0 && isPowerOfTwo(c/5) => (SUBshiftLL a (ADDshiftLL <x.Type> x x [2]) [log64(c/5)])
1153 (MSUB a (MOVDconst [c]) x) && c%7 == 0 && isPowerOfTwo(c/7) => (ADDshiftLL a (SUBshiftLL <x.Type> x x [3]) [log64(c/7)])
1154 (MSUB a (MOVDconst [c]) x) && c%9 == 0 && isPowerOfTwo(c/9) => (SUBshiftLL a (ADDshiftLL <x.Type> x x [3]) [log64(c/9)])
1155
1156 (MSUBW a x (MOVDconst [c])) && int32(c)==-1 => (MOVWUreg (ADD <a.Type> a x))
1157 (MSUBW a _ (MOVDconst [c])) && int32(c)==0 => (MOVWUreg a)
1158 (MSUBW a x (MOVDconst [c])) && int32(c)==1 => (MOVWUreg (SUB <a.Type> a x))
1159 (MSUBW a x (MOVDconst [c])) && isPowerOfTwo(c) => (MOVWUreg (SUBshiftLL <a.Type> a x [log64(c)]))
1160 (MSUBW a x (MOVDconst [c])) && isPowerOfTwo(c-1) && int32(c)>=3 => (MOVWUreg (SUB <a.Type> a (ADDshiftLL <x.Type> x x [log64(c-1)])))
1161 (MSUBW a x (MOVDconst [c])) && isPowerOfTwo(c+1) && int32(c)>=7 => (MOVWUreg (ADD <a.Type> a (SUBshiftLL <x.Type> x x [log64(c+1)])))
1162 (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)]))
1163 (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)]))
1164 (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)]))
1165 (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)]))
1166
1167 (MSUBW a (MOVDconst [c]) x) && int32(c)==-1 => (MOVWUreg (ADD <a.Type> a x))
1168 (MSUBW a (MOVDconst [c]) _) && int32(c)==0 => (MOVWUreg a)
1169 (MSUBW a (MOVDconst [c]) x) && int32(c)==1 => (MOVWUreg (SUB <a.Type> a x))
1170 (MSUBW a (MOVDconst [c]) x) && isPowerOfTwo(c) => (MOVWUreg (SUBshiftLL <a.Type> a x [log64(c)]))
1171 (MSUBW a (MOVDconst [c]) x) && isPowerOfTwo(c-1) && int32(c)>=3 => (MOVWUreg (SUB <a.Type> a (ADDshiftLL <x.Type> x x [log64(c-1)])))
1172 (MSUBW a (MOVDconst [c]) x) && isPowerOfTwo(c+1) && int32(c)>=7 => (MOVWUreg (ADD <a.Type> a (SUBshiftLL <x.Type> x x [log64(c+1)])))
1173 (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)]))
1174 (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)]))
1175 (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)]))
1176 (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)]))
1177
1178 // div by constant
1179 (UDIV x (MOVDconst [1])) => x
1180 (UDIV x (MOVDconst [c])) && isPowerOfTwo(c) => (SRLconst [log64(c)] x)
1181 (UDIVW x (MOVDconst [c])) && uint32(c)==1 => (MOVWUreg x)
1182 (UDIVW x (MOVDconst [c])) && isPowerOfTwo(c) && is32Bit(c) => (SRLconst [log64(c)] (MOVWUreg <v.Type> x))
1183 (UMOD _ (MOVDconst [1])) => (MOVDconst [0])
1184 (UMOD x (MOVDconst [c])) && isPowerOfTwo(c) => (ANDconst [c-1] x)
1185 (UMODW _ (MOVDconst [c])) && uint32(c)==1 => (MOVDconst [0])
1186 (UMODW x (MOVDconst [c])) && isPowerOfTwo(c) && is32Bit(c) => (ANDconst [c-1] x)
1187
1188 // generic simplifications
1189 (ADD x (NEG y)) => (SUB x y)
1190 (SUB x (NEG y)) => (ADD x y)
1191 (SUB x x) => (MOVDconst [0])
1192 (AND x x) => x
1193 (OR x x) => x
1194 (XOR x x) => (MOVDconst [0])
1195 (BIC x x) => (MOVDconst [0])
1196 (EON x x) => (MOVDconst [-1])
1197 (ORN x x) => (MOVDconst [-1])
1198 (AND x (MVN y)) => (BIC x y)
1199 (XOR x (MVN y)) => (EON x y)
1200 (OR x (MVN y)) => (ORN x y)
1201 (MVN (XOR x y)) => (EON x y)
1202 (NEG (SUB x y)) => (SUB y x)
1203 (NEG (NEG x)) => x
1204
1205 (CSEL [cc] (MOVDconst [-1]) (MOVDconst [0]) flag) => (CSETM [cc] flag)
1206 (CSEL [cc] (MOVDconst [0]) (MOVDconst [-1]) flag) => (CSETM [arm64Negate(cc)] flag)
1207 (CSEL [cc] x (MOVDconst [0]) flag) => (CSEL0 [cc] x flag)
1208 (CSEL [cc] (MOVDconst [0]) y flag) => (CSEL0 [arm64Negate(cc)] y flag)
1209 (CSEL [cc] x (ADDconst [1] a) flag) => (CSINC [cc] x a flag)
1210 (CSEL [cc] (ADDconst [1] a) x flag) => (CSINC [arm64Negate(cc)] x a flag)
1211 (CSEL [cc] x (MVN a) flag) => (CSINV [cc] x a flag)
1212 (CSEL [cc] (MVN a) x flag) => (CSINV [arm64Negate(cc)] x a flag)
1213 (CSEL [cc] x (NEG a) flag) => (CSNEG [cc] x a flag)
1214 (CSEL [cc] (NEG a) x flag) => (CSNEG [arm64Negate(cc)] x a flag)
1215
1216 (SUB x (SUB y z)) => (SUB (ADD <v.Type> x z) y)
1217 (SUB (SUB x y) z) => (SUB x (ADD <y.Type> y z))
1218
1219 // remove redundant *const ops
1220 (ADDconst [0] x) => x
1221 (SUBconst [0] x) => x
1222 (ANDconst [0] _) => (MOVDconst [0])
1223 (ANDconst [-1] x) => x
1224 (ORconst [0] x) => x
1225 (ORconst [-1] _) => (MOVDconst [-1])
1226 (XORconst [0] x) => x
1227 (XORconst [-1] x) => (MVN x)
1228
1229 // generic constant folding
1230 (ADDconst [c] (MOVDconst [d])) => (MOVDconst [c+d])
1231 (ADDconst [c] (ADDconst [d] x)) => (ADDconst [c+d] x)
1232 (ADDconst [c] (SUBconst [d] x)) => (ADDconst [c-d] x)
1233 (SUBconst [c] (MOVDconst [d])) => (MOVDconst [d-c])
1234 (SUBconst [c] (SUBconst [d] x)) => (ADDconst [-c-d] x)
1235 (SUBconst [c] (ADDconst [d] x)) => (ADDconst [-c+d] x)
1236 (SLLconst [c] (MOVDconst [d])) => (MOVDconst [d<<uint64(c)])
1237 (SRLconst [c] (MOVDconst [d])) => (MOVDconst [int64(uint64(d)>>uint64(c))])
1238 (SRAconst [c] (MOVDconst [d])) => (MOVDconst [d>>uint64(c)])
1239 (MUL (MOVDconst [c]) (MOVDconst [d])) => (MOVDconst [c*d])
1240 (MNEG (MOVDconst [c]) (MOVDconst [d])) => (MOVDconst [-c*d])
1241 (MULW (MOVDconst [c]) (MOVDconst [d])) => (MOVDconst [int64(uint32(c*d))])
1242 (MNEGW (MOVDconst [c]) (MOVDconst [d])) => (MOVDconst [int64(uint32(-c*d))])
1243 (MADD (MOVDconst [c]) x y) => (ADDconst [c] (MUL <x.Type> x y))
1244 (MSUB (MOVDconst [c]) x y) => (ADDconst [c] (MNEG <x.Type> x y))
1245 (MADD a (MOVDconst [c]) (MOVDconst [d])) => (ADDconst [c*d] a)
1246 (MSUB a (MOVDconst [c]) (MOVDconst [d])) => (SUBconst [c*d] a)
1247 (MADDW (MOVDconst [c]) x y) => (MOVWUreg (ADDconst <x.Type> [c] (MULW <x.Type> x y)))
1248 (MSUBW (MOVDconst [c]) x y) => (MOVWUreg (ADDconst <x.Type> [c] (MNEGW <x.Type> x y)))
1249 (MADDW a (MOVDconst [c]) (MOVDconst [d])) => (MOVWUreg (ADDconst <a.Type> [c*d] a))
1250 (MSUBW a (MOVDconst [c]) (MOVDconst [d])) => (MOVWUreg (SUBconst <a.Type> [c*d] a))
1251 (DIV (MOVDconst [c]) (MOVDconst [d])) && d != 0 => (MOVDconst [c/d])
1252 (UDIV (MOVDconst [c]) (MOVDconst [d])) && d != 0 => (MOVDconst [int64(uint64(c)/uint64(d))])
1253 (DIVW (MOVDconst [c]) (MOVDconst [d])) && d != 0 => (MOVDconst [int64(uint32(int32(c)/int32(d)))])
1254 (UDIVW (MOVDconst [c]) (MOVDconst [d])) && d != 0 => (MOVDconst [int64(uint32(c)/uint32(d))])
1255 (MOD (MOVDconst [c]) (MOVDconst [d])) && d != 0 => (MOVDconst [c%d])
1256 (UMOD (MOVDconst [c]) (MOVDconst [d])) && d != 0 => (MOVDconst [int64(uint64(c)%uint64(d))])
1257 (MODW (MOVDconst [c]) (MOVDconst [d])) && d != 0 => (MOVDconst [int64(uint32(int32(c)%int32(d)))])
1258 (UMODW (MOVDconst [c]) (MOVDconst [d])) && d != 0 => (MOVDconst [int64(uint32(c)%uint32(d))])
1259 (ANDconst [c] (MOVDconst [d])) => (MOVDconst [c&d])
1260 (ANDconst [c] (ANDconst [d] x)) => (ANDconst [c&d] x)
1261 (ANDconst [c] (MOVWUreg x)) => (ANDconst [c&(1<<32-1)] x)
1262 (ANDconst [c] (MOVHUreg x)) => (ANDconst [c&(1<<16-1)] x)
1263 (ANDconst [c] (MOVBUreg x)) => (ANDconst [c&(1<<8-1)] x)
1264 (MOVWUreg (ANDconst [c] x)) => (ANDconst [c&(1<<32-1)] x)
1265 (MOVHUreg (ANDconst [c] x)) => (ANDconst [c&(1<<16-1)] x)
1266 (MOVBUreg (ANDconst [c] x)) => (ANDconst [c&(1<<8-1)] x)
1267 (ORconst [c] (MOVDconst [d])) => (MOVDconst [c|d])
1268 (ORconst [c] (ORconst [d] x)) => (ORconst [c|d] x)
1269 (XORconst [c] (MOVDconst [d])) => (MOVDconst [c^d])
1270 (XORconst [c] (XORconst [d] x)) => (XORconst [c^d] x)
1271 (MVN (MOVDconst [c])) => (MOVDconst [^c])
1272 (NEG (MOVDconst [c])) => (MOVDconst [-c])
1273 (MOVBreg (MOVDconst [c])) => (MOVDconst [int64(int8(c))])
1274 (MOVBUreg (MOVDconst [c])) => (MOVDconst [int64(uint8(c))])
1275 (MOVHreg (MOVDconst [c])) => (MOVDconst [int64(int16(c))])
1276 (MOVHUreg (MOVDconst [c])) => (MOVDconst [int64(uint16(c))])
1277 (MOVWreg (MOVDconst [c])) => (MOVDconst [int64(int32(c))])
1278 (MOVWUreg (MOVDconst [c])) => (MOVDconst [int64(uint32(c))])
1279 (MOVDreg (MOVDconst [c])) => (MOVDconst [c])
1280
1281 // constant comparisons
1282 (CMPconst (MOVDconst [x]) [y]) => (FlagConstant [subFlags64(x,y)])
1283 (CMPWconst (MOVDconst [x]) [y]) => (FlagConstant [subFlags32(int32(x),y)])
1284 (TSTconst (MOVDconst [x]) [y]) => (FlagConstant [logicFlags64(x&y)])
1285 (TSTWconst (MOVDconst [x]) [y]) => (FlagConstant [logicFlags32(int32(x)&y)])
1286 (CMNconst (MOVDconst [x]) [y]) => (FlagConstant [addFlags64(x,y)])
1287 (CMNWconst (MOVDconst [x]) [y]) => (FlagConstant [addFlags32(int32(x),y)])
1288
1289 // other known comparisons
1290 (CMPconst (MOVBUreg _) [c]) && 0xff < c => (FlagConstant [subFlags64(0,1)])
1291 (CMPconst (MOVHUreg _) [c]) && 0xffff < c => (FlagConstant [subFlags64(0,1)])
1292 (CMPconst (MOVWUreg _) [c]) && 0xffffffff < c => (FlagConstant [subFlags64(0,1)])
1293 (CMPconst (ANDconst _ [m]) [n]) && 0 <= m && m < n => (FlagConstant [subFlags64(0,1)])
1294 (CMPconst (SRLconst _ [c]) [n]) && 0 <= n && 0 < c && c <= 63 && (1<<uint64(64-c)) <= uint64(n) => (FlagConstant [subFlags64(0,1)])
1295 (CMPWconst (MOVBUreg _) [c]) && 0xff < c => (FlagConstant [subFlags64(0,1)])
1296 (CMPWconst (MOVHUreg _) [c]) && 0xffff < c => (FlagConstant [subFlags64(0,1)])
1297
1298 // absorb flag constants into branches
1299 (EQ (FlagConstant [fc]) yes no) && fc.eq() => (First yes no)
1300 (EQ (FlagConstant [fc]) yes no) && !fc.eq() => (First no yes)
1301
1302 (NE (FlagConstant [fc]) yes no) && fc.ne() => (First yes no)
1303 (NE (FlagConstant [fc]) yes no) && !fc.ne() => (First no yes)
1304
1305 (LT (FlagConstant [fc]) yes no) && fc.lt() => (First yes no)
1306 (LT (FlagConstant [fc]) yes no) && !fc.lt() => (First no yes)
1307
1308 (LE (FlagConstant [fc]) yes no) && fc.le() => (First yes no)
1309 (LE (FlagConstant [fc]) yes no) && !fc.le() => (First no yes)
1310
1311 (GT (FlagConstant [fc]) yes no) && fc.gt() => (First yes no)
1312 (GT (FlagConstant [fc]) yes no) && !fc.gt() => (First no yes)
1313
1314 (GE (FlagConstant [fc]) yes no) && fc.ge() => (First yes no)
1315 (GE (FlagConstant [fc]) yes no) && !fc.ge() => (First no yes)
1316
1317 (ULT (FlagConstant [fc]) yes no) && fc.ult() => (First yes no)
1318 (ULT (FlagConstant [fc]) yes no) && !fc.ult() => (First no yes)
1319
1320 (ULE (FlagConstant [fc]) yes no) && fc.ule() => (First yes no)
1321 (ULE (FlagConstant [fc]) yes no) && !fc.ule() => (First no yes)
1322
1323 (UGT (FlagConstant [fc]) yes no) && fc.ugt() => (First yes no)
1324 (UGT (FlagConstant [fc]) yes no) && !fc.ugt() => (First no yes)
1325
1326 (UGE (FlagConstant [fc]) yes no) && fc.uge() => (First yes no)
1327 (UGE (FlagConstant [fc]) yes no) && !fc.uge() => (First no yes)
1328
1329 (LTnoov (FlagConstant [fc]) yes no) && fc.ltNoov() => (First yes no)
1330 (LTnoov (FlagConstant [fc]) yes no) && !fc.ltNoov() => (First no yes)
1331
1332 (LEnoov (FlagConstant [fc]) yes no) && fc.leNoov() => (First yes no)
1333 (LEnoov (FlagConstant [fc]) yes no) && !fc.leNoov() => (First no yes)
1334
1335 (GTnoov (FlagConstant [fc]) yes no) && fc.gtNoov() => (First yes no)
1336 (GTnoov (FlagConstant [fc]) yes no) && !fc.gtNoov() => (First no yes)
1337
1338 (GEnoov (FlagConstant [fc]) yes no) && fc.geNoov() => (First yes no)
1339 (GEnoov (FlagConstant [fc]) yes no) && !fc.geNoov() => (First no yes)
1340
1341 (Z (MOVDconst [0]) yes no) => (First yes no)
1342 (Z (MOVDconst [c]) yes no) && c != 0 => (First no yes)
1343 (NZ (MOVDconst [0]) yes no) => (First no yes)
1344 (NZ (MOVDconst [c]) yes no) && c != 0 => (First yes no)
1345 (ZW (MOVDconst [c]) yes no) && int32(c) == 0 => (First yes no)
1346 (ZW (MOVDconst [c]) yes no) && int32(c) != 0 => (First no yes)
1347 (NZW (MOVDconst [c]) yes no) && int32(c) == 0 => (First no yes)
1348 (NZW (MOVDconst [c]) yes no) && int32(c) != 0 => (First yes no)
1349
1350 // absorb InvertFlags into branches
1351 (LT (InvertFlags cmp) yes no) => (GT cmp yes no)
1352 (GT (InvertFlags cmp) yes no) => (LT cmp yes no)
1353 (LE (InvertFlags cmp) yes no) => (GE cmp yes no)
1354 (GE (InvertFlags cmp) yes no) => (LE cmp yes no)
1355 (ULT (InvertFlags cmp) yes no) => (UGT cmp yes no)
1356 (UGT (InvertFlags cmp) yes no) => (ULT cmp yes no)
1357 (ULE (InvertFlags cmp) yes no) => (UGE cmp yes no)
1358 (UGE (InvertFlags cmp) yes no) => (ULE cmp yes no)
1359 (EQ (InvertFlags cmp) yes no) => (EQ cmp yes no)
1360 (NE (InvertFlags cmp) yes no) => (NE cmp yes no)
1361 (FLT (InvertFlags cmp) yes no) => (FGT cmp yes no)
1362 (FGT (InvertFlags cmp) yes no) => (FLT cmp yes no)
1363 (FLE (InvertFlags cmp) yes no) => (FGE cmp yes no)
1364 (FGE (InvertFlags cmp) yes no) => (FLE cmp yes no)
1365 (LTnoov (InvertFlags cmp) yes no) => (GTnoov cmp yes no)
1366 (GEnoov (InvertFlags cmp) yes no) => (LEnoov cmp yes no)
1367 (LEnoov (InvertFlags cmp) yes no) => (GEnoov cmp yes no)
1368 (GTnoov (InvertFlags cmp) yes no) => (LTnoov cmp yes no)
1369
1370 // absorb InvertFlags into conditional instructions
1371 (CSEL [cc] x y (InvertFlags cmp)) => (CSEL [arm64Invert(cc)] x y cmp)
1372 (CSEL0 [cc] x (InvertFlags cmp)) => (CSEL0 [arm64Invert(cc)] x cmp)
1373 (CSETM [cc] (InvertFlags cmp)) => (CSETM [arm64Invert(cc)] cmp)
1374 (CSINC [cc] x y (InvertFlags cmp)) => (CSINC [arm64Invert(cc)] x y cmp)
1375 (CSINV [cc] x y (InvertFlags cmp)) => (CSINV [arm64Invert(cc)] x y cmp)
1376 (CSNEG [cc] x y (InvertFlags cmp)) => (CSNEG [arm64Invert(cc)] x y cmp)
1377
1378 // absorb flag constants into boolean values
1379 (Equal (FlagConstant [fc])) => (MOVDconst [b2i(fc.eq())])
1380 (NotEqual (FlagConstant [fc])) => (MOVDconst [b2i(fc.ne())])
1381 (LessThan (FlagConstant [fc])) => (MOVDconst [b2i(fc.lt())])
1382 (LessThanU (FlagConstant [fc])) => (MOVDconst [b2i(fc.ult())])
1383 (LessEqual (FlagConstant [fc])) => (MOVDconst [b2i(fc.le())])
1384 (LessEqualU (FlagConstant [fc])) => (MOVDconst [b2i(fc.ule())])
1385 (GreaterThan (FlagConstant [fc])) => (MOVDconst [b2i(fc.gt())])
1386 (GreaterThanU (FlagConstant [fc])) => (MOVDconst [b2i(fc.ugt())])
1387 (GreaterEqual (FlagConstant [fc])) => (MOVDconst [b2i(fc.ge())])
1388 (GreaterEqualU (FlagConstant [fc])) => (MOVDconst [b2i(fc.uge())])
1389 (LessThanNoov (FlagConstant [fc])) => (MOVDconst [b2i(fc.ltNoov())])
1390 (GreaterEqualNoov (FlagConstant [fc])) => (MOVDconst [b2i(fc.geNoov())])
1391
1392 // absorb InvertFlags into boolean values
1393 (Equal (InvertFlags x)) => (Equal x)
1394 (NotEqual (InvertFlags x)) => (NotEqual x)
1395 (LessThan (InvertFlags x)) => (GreaterThan x)
1396 (LessThanU (InvertFlags x)) => (GreaterThanU x)
1397 (GreaterThan (InvertFlags x)) => (LessThan x)
1398 (GreaterThanU (InvertFlags x)) => (LessThanU x)
1399 (LessEqual (InvertFlags x)) => (GreaterEqual x)
1400 (LessEqualU (InvertFlags x)) => (GreaterEqualU x)
1401 (GreaterEqual (InvertFlags x)) => (LessEqual x)
1402 (GreaterEqualU (InvertFlags x)) => (LessEqualU x)
1403 (LessThanF (InvertFlags x)) => (GreaterThanF x)
1404 (LessEqualF (InvertFlags x)) => (GreaterEqualF x)
1405 (GreaterThanF (InvertFlags x)) => (LessThanF x)
1406 (GreaterEqualF (InvertFlags x)) => (LessEqualF x)
1407 (LessThanNoov (InvertFlags x)) => (CSEL0 [OpARM64NotEqual] (GreaterEqualNoov <typ.Bool> x) x)
1408 (GreaterEqualNoov (InvertFlags x)) => (CSINC [OpARM64NotEqual] (LessThanNoov <typ.Bool> x) (MOVDconst [0]) x)
1409
1410 // Don't bother extending if we're not using the higher bits.
1411 (MOV(B|BU)reg x) && v.Type.Size() <= 1 => x
1412 (MOV(H|HU)reg x) && v.Type.Size() <= 2 => x
1413 (MOV(W|WU)reg x) && v.Type.Size() <= 4 => x
1414
1415 // omit sign extension
1416 (MOVWreg <t> (ANDconst x [c])) && uint64(c) & uint64(0xffffffff80000000) == 0 => (ANDconst <t> x [c])
1417 (MOVHreg <t> (ANDconst x [c])) && uint64(c) & uint64(0xffffffffffff8000) == 0 => (ANDconst <t> x [c])
1418 (MOVBreg <t> (ANDconst x [c])) && uint64(c) & uint64(0xffffffffffffff80) == 0 => (ANDconst <t> x [c])
1419
1420 // absorb flag constants into conditional instructions
1421 (CSEL [cc] x _ flag) && ccARM64Eval(cc, flag) > 0 => x
1422 (CSEL [cc] _ y flag) && ccARM64Eval(cc, flag) < 0 => y
1423 (CSEL0 [cc] x flag) && ccARM64Eval(cc, flag) > 0 => x
1424 (CSEL0 [cc] _ flag) && ccARM64Eval(cc, flag) < 0 => (MOVDconst [0])
1425 (CSNEG [cc] x _ flag) && ccARM64Eval(cc, flag) > 0 => x
1426 (CSNEG [cc] _ y flag) && ccARM64Eval(cc, flag) < 0 => (NEG y)
1427 (CSINV [cc] x _ flag) && ccARM64Eval(cc, flag) > 0 => x
1428 (CSINV [cc] _ y flag) && ccARM64Eval(cc, flag) < 0 => (Not y)
1429 (CSINC [cc] x _ flag) && ccARM64Eval(cc, flag) > 0 => x
1430 (CSINC [cc] _ y flag) && ccARM64Eval(cc, flag) < 0 => (ADDconst [1] y)
1431 (CSETM [cc] flag) && ccARM64Eval(cc, flag) > 0 => (MOVDconst [-1])
1432 (CSETM [cc] flag) && ccARM64Eval(cc, flag) < 0 => (MOVDconst [0])
1433
1434 // absorb flags back into boolean CSEL
1435 (CSEL [cc] x y (CMPWconst [0] boolval)) && cc == OpARM64NotEqual && flagArg(boolval) != nil =>
1436 (CSEL [boolval.Op] x y flagArg(boolval))
1437 (CSEL [cc] x y (CMPWconst [0] boolval)) && cc == OpARM64Equal && flagArg(boolval) != nil =>
1438 (CSEL [arm64Negate(boolval.Op)] x y flagArg(boolval))
1439 (CSEL0 [cc] x (CMPWconst [0] boolval)) && cc == OpARM64NotEqual && flagArg(boolval) != nil =>
1440 (CSEL0 [boolval.Op] x flagArg(boolval))
1441 (CSEL0 [cc] x (CMPWconst [0] boolval)) && cc == OpARM64Equal && flagArg(boolval) != nil =>
1442 (CSEL0 [arm64Negate(boolval.Op)] x flagArg(boolval))
1443
1444 // absorb shifts into ops
1445 (NEG x:(SLLconst [c] y)) && clobberIfDead(x) => (NEGshiftLL [c] y)
1446 (NEG x:(SRLconst [c] y)) && clobberIfDead(x) => (NEGshiftRL [c] y)
1447 (NEG x:(SRAconst [c] y)) && clobberIfDead(x) => (NEGshiftRA [c] y)
1448 (MVN x:(SLLconst [c] y)) && clobberIfDead(x) => (MVNshiftLL [c] y)
1449 (MVN x:(SRLconst [c] y)) && clobberIfDead(x) => (MVNshiftRL [c] y)
1450 (MVN x:(SRAconst [c] y)) && clobberIfDead(x) => (MVNshiftRA [c] y)
1451 (MVN x:(RORconst [c] y)) && clobberIfDead(x) => (MVNshiftRO [c] y)
1452 (ADD x0 x1:(SLLconst [c] y)) && clobberIfDead(x1) => (ADDshiftLL x0 y [c])
1453 (ADD x0 x1:(SRLconst [c] y)) && clobberIfDead(x1) => (ADDshiftRL x0 y [c])
1454 (ADD x0 x1:(SRAconst [c] y)) && clobberIfDead(x1) => (ADDshiftRA x0 y [c])
1455 (SUB x0 x1:(SLLconst [c] y)) && clobberIfDead(x1) => (SUBshiftLL x0 y [c])
1456 (SUB x0 x1:(SRLconst [c] y)) && clobberIfDead(x1) => (SUBshiftRL x0 y [c])
1457 (SUB x0 x1:(SRAconst [c] y)) && clobberIfDead(x1) => (SUBshiftRA x0 y [c])
1458 (AND x0 x1:(SLLconst [c] y)) && clobberIfDead(x1) => (ANDshiftLL x0 y [c])
1459 (AND x0 x1:(SRLconst [c] y)) && clobberIfDead(x1) => (ANDshiftRL x0 y [c])
1460 (AND x0 x1:(SRAconst [c] y)) && clobberIfDead(x1) => (ANDshiftRA x0 y [c])
1461 (AND x0 x1:(RORconst [c] y)) && clobberIfDead(x1) => (ANDshiftRO x0 y [c])
1462 (OR x0 x1:(SLLconst [c] y)) && clobberIfDead(x1) => (ORshiftLL x0 y [c]) // useful for combined load
1463 (OR x0 x1:(SRLconst [c] y)) && clobberIfDead(x1) => (ORshiftRL x0 y [c])
1464 (OR x0 x1:(SRAconst [c] y)) && clobberIfDead(x1) => (ORshiftRA x0 y [c])
1465 (OR x0 x1:(RORconst [c] y)) && clobberIfDead(x1) => (ORshiftRO x0 y [c])
1466 (XOR x0 x1:(SLLconst [c] y)) && clobberIfDead(x1) => (XORshiftLL x0 y [c])
1467 (XOR x0 x1:(SRLconst [c] y)) && clobberIfDead(x1) => (XORshiftRL x0 y [c])
1468 (XOR x0 x1:(SRAconst [c] y)) && clobberIfDead(x1) => (XORshiftRA x0 y [c])
1469 (XOR x0 x1:(RORconst [c] y)) && clobberIfDead(x1) => (XORshiftRO x0 y [c])
1470 (BIC x0 x1:(SLLconst [c] y)) && clobberIfDead(x1) => (BICshiftLL x0 y [c])
1471 (BIC x0 x1:(SRLconst [c] y)) && clobberIfDead(x1) => (BICshiftRL x0 y [c])
1472 (BIC x0 x1:(SRAconst [c] y)) && clobberIfDead(x1) => (BICshiftRA x0 y [c])
1473 (BIC x0 x1:(RORconst [c] y)) && clobberIfDead(x1) => (BICshiftRO x0 y [c])
1474 (ORN x0 x1:(SLLconst [c] y)) && clobberIfDead(x1) => (ORNshiftLL x0 y [c])
1475 (ORN x0 x1:(SRLconst [c] y)) && clobberIfDead(x1) => (ORNshiftRL x0 y [c])
1476 (ORN x0 x1:(SRAconst [c] y)) && clobberIfDead(x1) => (ORNshiftRA x0 y [c])
1477 (ORN x0 x1:(RORconst [c] y)) && clobberIfDead(x1) => (ORNshiftRO x0 y [c])
1478 (EON x0 x1:(SLLconst [c] y)) && clobberIfDead(x1) => (EONshiftLL x0 y [c])
1479 (EON x0 x1:(SRLconst [c] y)) && clobberIfDead(x1) => (EONshiftRL x0 y [c])
1480 (EON x0 x1:(SRAconst [c] y)) && clobberIfDead(x1) => (EONshiftRA x0 y [c])
1481 (EON x0 x1:(RORconst [c] y)) && clobberIfDead(x1) => (EONshiftRO x0 y [c])
1482 (CMP x0 x1:(SLLconst [c] y)) && clobberIfDead(x1) => (CMPshiftLL x0 y [c])
1483 (CMP x0:(SLLconst [c] y) x1) && clobberIfDead(x0) => (InvertFlags (CMPshiftLL x1 y [c]))
1484 (CMP x0 x1:(SRLconst [c] y)) && clobberIfDead(x1) => (CMPshiftRL x0 y [c])
1485 (CMP x0:(SRLconst [c] y) x1) && clobberIfDead(x0) => (InvertFlags (CMPshiftRL x1 y [c]))
1486 (CMP x0 x1:(SRAconst [c] y)) && clobberIfDead(x1) => (CMPshiftRA x0 y [c])
1487 (CMP x0:(SRAconst [c] y) x1) && clobberIfDead(x0) => (InvertFlags (CMPshiftRA x1 y [c]))
1488 (CMN x0 x1:(SLLconst [c] y)) && clobberIfDead(x1) => (CMNshiftLL x0 y [c])
1489 (CMN x0 x1:(SRLconst [c] y)) && clobberIfDead(x1) => (CMNshiftRL x0 y [c])
1490 (CMN x0 x1:(SRAconst [c] y)) && clobberIfDead(x1) => (CMNshiftRA x0 y [c])
1491 (TST x0 x1:(SLLconst [c] y)) && clobberIfDead(x1) => (TSTshiftLL x0 y [c])
1492 (TST x0 x1:(SRLconst [c] y)) && clobberIfDead(x1) => (TSTshiftRL x0 y [c])
1493 (TST x0 x1:(SRAconst [c] y)) && clobberIfDead(x1) => (TSTshiftRA x0 y [c])
1494 (TST x0 x1:(RORconst [c] y)) && clobberIfDead(x1) => (TSTshiftRO x0 y [c])
1495
1496 // prefer *const ops to *shift ops
1497 (ADDshiftLL (MOVDconst [c]) x [d]) => (ADDconst [c] (SLLconst <x.Type> x [d]))
1498 (ADDshiftRL (MOVDconst [c]) x [d]) => (ADDconst [c] (SRLconst <x.Type> x [d]))
1499 (ADDshiftRA (MOVDconst [c]) x [d]) => (ADDconst [c] (SRAconst <x.Type> x [d]))
1500 (ANDshiftLL (MOVDconst [c]) x [d]) => (ANDconst [c] (SLLconst <x.Type> x [d]))
1501 (ANDshiftRL (MOVDconst [c]) x [d]) => (ANDconst [c] (SRLconst <x.Type> x [d]))
1502 (ANDshiftRA (MOVDconst [c]) x [d]) => (ANDconst [c] (SRAconst <x.Type> x [d]))
1503 (ANDshiftRO (MOVDconst [c]) x [d]) => (ANDconst [c] (RORconst <x.Type> x [d]))
1504 (ORshiftLL (MOVDconst [c]) x [d]) => (ORconst [c] (SLLconst <x.Type> x [d]))
1505 (ORshiftRL (MOVDconst [c]) x [d]) => (ORconst [c] (SRLconst <x.Type> x [d]))
1506 (ORshiftRA (MOVDconst [c]) x [d]) => (ORconst [c] (SRAconst <x.Type> x [d]))
1507 (ORshiftRO (MOVDconst [c]) x [d]) => (ORconst [c] (RORconst <x.Type> x [d]))
1508 (XORshiftLL (MOVDconst [c]) x [d]) => (XORconst [c] (SLLconst <x.Type> x [d]))
1509 (XORshiftRL (MOVDconst [c]) x [d]) => (XORconst [c] (SRLconst <x.Type> x [d]))
1510 (XORshiftRA (MOVDconst [c]) x [d]) => (XORconst [c] (SRAconst <x.Type> x [d]))
1511 (XORshiftRO (MOVDconst [c]) x [d]) => (XORconst [c] (RORconst <x.Type> x [d]))
1512 (CMPshiftLL (MOVDconst [c]) x [d]) => (InvertFlags (CMPconst [c] (SLLconst <x.Type> x [d])))
1513 (CMPshiftRL (MOVDconst [c]) x [d]) => (InvertFlags (CMPconst [c] (SRLconst <x.Type> x [d])))
1514 (CMPshiftRA (MOVDconst [c]) x [d]) => (InvertFlags (CMPconst [c] (SRAconst <x.Type> x [d])))
1515 (CMNshiftLL (MOVDconst [c]) x [d]) => (CMNconst [c] (SLLconst <x.Type> x [d]))
1516 (CMNshiftRL (MOVDconst [c]) x [d]) => (CMNconst [c] (SRLconst <x.Type> x [d]))
1517 (CMNshiftRA (MOVDconst [c]) x [d]) => (CMNconst [c] (SRAconst <x.Type> x [d]))
1518 (TSTshiftLL (MOVDconst [c]) x [d]) => (TSTconst [c] (SLLconst <x.Type> x [d]))
1519 (TSTshiftRL (MOVDconst [c]) x [d]) => (TSTconst [c] (SRLconst <x.Type> x [d]))
1520 (TSTshiftRA (MOVDconst [c]) x [d]) => (TSTconst [c] (SRAconst <x.Type> x [d]))
1521 (TSTshiftRO (MOVDconst [c]) x [d]) => (TSTconst [c] (RORconst <x.Type> x [d]))
1522
1523 // constant folding in *shift ops
1524 (MVNshiftLL (MOVDconst [c]) [d]) => (MOVDconst [^int64(uint64(c)<<uint64(d))])
1525 (MVNshiftRL (MOVDconst [c]) [d]) => (MOVDconst [^int64(uint64(c)>>uint64(d))])
1526 (MVNshiftRA (MOVDconst [c]) [d]) => (MOVDconst [^(c>>uint64(d))])
1527 (MVNshiftRO (MOVDconst [c]) [d]) => (MOVDconst [^rotateRight64(c, d)])
1528 (NEGshiftLL (MOVDconst [c]) [d]) => (MOVDconst [-int64(uint64(c)<<uint64(d))])
1529 (NEGshiftRL (MOVDconst [c]) [d]) => (MOVDconst [-int64(uint64(c)>>uint64(d))])
1530 (NEGshiftRA (MOVDconst [c]) [d]) => (MOVDconst [-(c>>uint64(d))])
1531 (ADDshiftLL x (MOVDconst [c]) [d]) => (ADDconst x [int64(uint64(c)<<uint64(d))])
1532 (ADDshiftRL x (MOVDconst [c]) [d]) => (ADDconst x [int64(uint64(c)>>uint64(d))])
1533 (ADDshiftRA x (MOVDconst [c]) [d]) => (ADDconst x [c>>uint64(d)])
1534 (SUBshiftLL x (MOVDconst [c]) [d]) => (SUBconst x [int64(uint64(c)<<uint64(d))])
1535 (SUBshiftRL x (MOVDconst [c]) [d]) => (SUBconst x [int64(uint64(c)>>uint64(d))])
1536 (SUBshiftRA x (MOVDconst [c]) [d]) => (SUBconst x [c>>uint64(d)])
1537 (ANDshiftLL x (MOVDconst [c]) [d]) => (ANDconst x [int64(uint64(c)<<uint64(d))])
1538 (ANDshiftRL x (MOVDconst [c]) [d]) => (ANDconst x [int64(uint64(c)>>uint64(d))])
1539 (ANDshiftRA x (MOVDconst [c]) [d]) => (ANDconst x [c>>uint64(d)])
1540 (ANDshiftRO x (MOVDconst [c]) [d]) => (ANDconst x [rotateRight64(c, d)])
1541 (ORshiftLL x (MOVDconst [c]) [d]) => (ORconst x [int64(uint64(c)<<uint64(d))])
1542 (ORshiftRL x (MOVDconst [c]) [d]) => (ORconst x [int64(uint64(c)>>uint64(d))])
1543 (ORshiftRA x (MOVDconst [c]) [d]) => (ORconst x [c>>uint64(d)])
1544 (ORshiftRO x (MOVDconst [c]) [d]) => (ORconst x [rotateRight64(c, d)])
1545 (XORshiftLL x (MOVDconst [c]) [d]) => (XORconst x [int64(uint64(c)<<uint64(d))])
1546 (XORshiftRL x (MOVDconst [c]) [d]) => (XORconst x [int64(uint64(c)>>uint64(d))])
1547 (XORshiftRA x (MOVDconst [c]) [d]) => (XORconst x [c>>uint64(d)])
1548 (XORshiftRO x (MOVDconst [c]) [d]) => (XORconst x [rotateRight64(c, d)])
1549 (BICshiftLL x (MOVDconst [c]) [d]) => (ANDconst x [^int64(uint64(c)<<uint64(d))])
1550 (BICshiftRL x (MOVDconst [c]) [d]) => (ANDconst x [^int64(uint64(c)>>uint64(d))])
1551 (BICshiftRA x (MOVDconst [c]) [d]) => (ANDconst x [^(c>>uint64(d))])
1552 (BICshiftRO x (MOVDconst [c]) [d]) => (ANDconst x [^rotateRight64(c, d)])
1553 (ORNshiftLL x (MOVDconst [c]) [d]) => (ORconst x [^int64(uint64(c)<<uint64(d))])
1554 (ORNshiftRL x (MOVDconst [c]) [d]) => (ORconst x [^int64(uint64(c)>>uint64(d))])
1555 (ORNshiftRA x (MOVDconst [c]) [d]) => (ORconst x [^(c>>uint64(d))])
1556 (ORNshiftRO x (MOVDconst [c]) [d]) => (ORconst x [^rotateRight64(c, d)])
1557 (EONshiftLL x (MOVDconst [c]) [d]) => (XORconst x [^int64(uint64(c)<<uint64(d))])
1558 (EONshiftRL x (MOVDconst [c]) [d]) => (XORconst x [^int64(uint64(c)>>uint64(d))])
1559 (EONshiftRA x (MOVDconst [c]) [d]) => (XORconst x [^(c>>uint64(d))])
1560 (EONshiftRO x (MOVDconst [c]) [d]) => (XORconst x [^rotateRight64(c, d)])
1561 (CMPshiftLL x (MOVDconst [c]) [d]) => (CMPconst x [int64(uint64(c)<<uint64(d))])
1562 (CMPshiftRL x (MOVDconst [c]) [d]) => (CMPconst x [int64(uint64(c)>>uint64(d))])
1563 (CMPshiftRA x (MOVDconst [c]) [d]) => (CMPconst x [c>>uint64(d)])
1564 (CMNshiftLL x (MOVDconst [c]) [d]) => (CMNconst x [int64(uint64(c)<<uint64(d))])
1565 (CMNshiftRL x (MOVDconst [c]) [d]) => (CMNconst x [int64(uint64(c)>>uint64(d))])
1566 (CMNshiftRA x (MOVDconst [c]) [d]) => (CMNconst x [c>>uint64(d)])
1567 (TSTshiftLL x (MOVDconst [c]) [d]) => (TSTconst x [int64(uint64(c)<<uint64(d))])
1568 (TSTshiftRL x (MOVDconst [c]) [d]) => (TSTconst x [int64(uint64(c)>>uint64(d))])
1569 (TSTshiftRA x (MOVDconst [c]) [d]) => (TSTconst x [c>>uint64(d)])
1570 (TSTshiftRO x (MOVDconst [c]) [d]) => (TSTconst x [rotateRight64(c, d)])
1571
1572 // simplification with *shift ops
1573 (SUBshiftLL (SLLconst x [c]) x [c]) => (MOVDconst [0])
1574 (SUBshiftRL (SRLconst x [c]) x [c]) => (MOVDconst [0])
1575 (SUBshiftRA (SRAconst x [c]) x [c]) => (MOVDconst [0])
1576 (ANDshiftLL y:(SLLconst x [c]) x [c]) => y
1577 (ANDshiftRL y:(SRLconst x [c]) x [c]) => y
1578 (ANDshiftRA y:(SRAconst x [c]) x [c]) => y
1579 (ANDshiftRO y:(RORconst x [c]) x [c]) => y
1580 (ORshiftLL y:(SLLconst x [c]) x [c]) => y
1581 (ORshiftRL y:(SRLconst x [c]) x [c]) => y
1582 (ORshiftRA y:(SRAconst x [c]) x [c]) => y
1583 (ORshiftRO y:(RORconst x [c]) x [c]) => y
1584 (XORshiftLL (SLLconst x [c]) x [c]) => (MOVDconst [0])
1585 (XORshiftRL (SRLconst x [c]) x [c]) => (MOVDconst [0])
1586 (XORshiftRA (SRAconst x [c]) x [c]) => (MOVDconst [0])
1587 (XORshiftRO (RORconst x [c]) x [c]) => (MOVDconst [0])
1588 (BICshiftLL (SLLconst x [c]) x [c]) => (MOVDconst [0])
1589 (BICshiftRL (SRLconst x [c]) x [c]) => (MOVDconst [0])
1590 (BICshiftRA (SRAconst x [c]) x [c]) => (MOVDconst [0])
1591 (BICshiftRO (RORconst x [c]) x [c]) => (MOVDconst [0])
1592 (EONshiftLL (SLLconst x [c]) x [c]) => (MOVDconst [-1])
1593 (EONshiftRL (SRLconst x [c]) x [c]) => (MOVDconst [-1])
1594 (EONshiftRA (SRAconst x [c]) x [c]) => (MOVDconst [-1])
1595 (EONshiftRO (RORconst x [c]) x [c]) => (MOVDconst [-1])
1596 (ORNshiftLL (SLLconst x [c]) x [c]) => (MOVDconst [-1])
1597 (ORNshiftRL (SRLconst x [c]) x [c]) => (MOVDconst [-1])
1598 (ORNshiftRA (SRAconst x [c]) x [c]) => (MOVDconst [-1])
1599 (ORNshiftRO (RORconst x [c]) x [c]) => (MOVDconst [-1])
1600
1601 // rev16w | rev16
1602 // ((x>>8) | (x<<8)) => (REV16W x), the type of x is uint16, "|" can also be "^" or "+".
1603 ((ADDshiftLL|ORshiftLL|XORshiftLL) <typ.UInt16> [8] (UBFX <typ.UInt16> [armBFAuxInt(8, 8)] x) x) => (REV16W x)
1604
1605 // ((x & 0xff00ff00)>>8) | ((x & 0x00ff00ff)<<8), "|" can also be "^" or "+".
1606 ((ADDshiftLL|ORshiftLL|XORshiftLL) [8] (UBFX [armBFAuxInt(8, 24)] (ANDconst [c1] x)) (ANDconst [c2] x))
1607 && uint32(c1) == 0xff00ff00 && uint32(c2) == 0x00ff00ff
1608 => (REV16W x)
1609
1610 // ((x & 0xff00ff00ff00ff00)>>8) | ((x & 0x00ff00ff00ff00ff)<<8), "|" can also be "^" or "+".
1611 ((ADDshiftLL|ORshiftLL|XORshiftLL) [8] (SRLconst [8] (ANDconst [c1] x)) (ANDconst [c2] x))
1612 && (uint64(c1) == 0xff00ff00ff00ff00 && uint64(c2) == 0x00ff00ff00ff00ff)
1613 => (REV16 x)
1614
1615 // ((x & 0xff00ff00)>>8) | ((x & 0x00ff00ff)<<8), "|" can also be "^" or "+".
1616 ((ADDshiftLL|ORshiftLL|XORshiftLL) [8] (SRLconst [8] (ANDconst [c1] x)) (ANDconst [c2] x))
1617 && (uint64(c1) == 0xff00ff00 && uint64(c2) == 0x00ff00ff)
1618 => (REV16 (ANDconst <x.Type> [0xffffffff] x))
1619
1620 // Extract from reg pair
1621 (ADDshiftLL [c] (SRLconst x [64-c]) x2) => (EXTRconst [64-c] x2 x)
1622 ( ORshiftLL [c] (SRLconst x [64-c]) x2) => (EXTRconst [64-c] x2 x)
1623 (XORshiftLL [c] (SRLconst x [64-c]) x2) => (EXTRconst [64-c] x2 x)
1624
1625 (ADDshiftLL <t> [c] (UBFX [bfc] x) x2) && c < 32 && t.Size() == 4 && bfc == armBFAuxInt(32-c, c)
1626 => (EXTRWconst [32-c] x2 x)
1627 ( ORshiftLL <t> [c] (UBFX [bfc] x) x2) && c < 32 && t.Size() == 4 && bfc == armBFAuxInt(32-c, c)
1628 => (EXTRWconst [32-c] x2 x)
1629 (XORshiftLL <t> [c] (UBFX [bfc] x) x2) && c < 32 && t.Size() == 4 && bfc == armBFAuxInt(32-c, c)
1630 => (EXTRWconst [32-c] x2 x)
1631
1632 // Rewrite special pairs of shifts to AND.
1633 // On ARM64 the bitmask can fit into an instruction.
1634 (SRLconst [c] (SLLconst [c] x)) && 0 < c && c < 64 => (ANDconst [1<<uint(64-c)-1] x) // mask out high bits
1635 (SLLconst [c] (SRLconst [c] x)) && 0 < c && c < 64 => (ANDconst [^(1<<uint(c)-1)] x) // mask out low bits
1636
1637 // Special case setting bit as 1. An example is math.Copysign(c,-1)
1638 (ORconst [c1] (ANDconst [c2] x)) && c2|c1 == ^0 => (ORconst [c1] x)
1639
1640 // If the shift amount is larger than the datasize(32, 16, 8), we can optimize to constant 0.
1641 (MOVWUreg (SLLconst [lc] x)) && lc >= 32 => (MOVDconst [0])
1642 (MOVHUreg (SLLconst [lc] x)) && lc >= 16 => (MOVDconst [0])
1643 (MOVBUreg (SLLconst [lc] x)) && lc >= 8 => (MOVDconst [0])
1644
1645 // After zero extension, the upper (64-datasize(32|16|8)) bits are zero, we can optimiza to constant 0.
1646 (SRLconst [rc] (MOVWUreg x)) && rc >= 32 => (MOVDconst [0])
1647 (SRLconst [rc] (MOVHUreg x)) && rc >= 16 => (MOVDconst [0])
1648 (SRLconst [rc] (MOVBUreg x)) && rc >= 8 => (MOVDconst [0])
1649
1650 // Special cases for slice operations
1651 (ADD x0 x1:(ANDshiftRA x2:(SLLconst [sl] y) z [63])) && x1.Uses == 1 && x2.Uses == 1 => (ADDshiftLL x0 (ANDshiftRA <y.Type> y z [63]) [sl])
1652 (ADD x0 x1:(ANDshiftLL x2:(SRAconst [63] z) y [sl])) && x1.Uses == 1 && x2.Uses == 1 => (ADDshiftLL x0 (ANDshiftRA <y.Type> y z [63]) [sl])
1653
1654 // bitfield ops
1655
1656 // sbfiz
1657 // (x << lc) >> rc
1658 (SRAconst [rc] (SLLconst [lc] x)) && lc > rc => (SBFIZ [armBFAuxInt(lc-rc, 64-lc)] x)
1659 // int64(x << lc)
1660 (MOVWreg (SLLconst [lc] x)) && lc < 32 => (SBFIZ [armBFAuxInt(lc, 32-lc)] x)
1661 (MOVHreg (SLLconst [lc] x)) && lc < 16 => (SBFIZ [armBFAuxInt(lc, 16-lc)] x)
1662 (MOVBreg (SLLconst [lc] x)) && lc < 8 => (SBFIZ [armBFAuxInt(lc, 8-lc)] x)
1663 // int64(x) << lc
1664 (SLLconst [lc] (MOVWreg x)) => (SBFIZ [armBFAuxInt(lc, min(32, 64-lc))] x)
1665 (SLLconst [lc] (MOVHreg x)) => (SBFIZ [armBFAuxInt(lc, min(16, 64-lc))] x)
1666 (SLLconst [lc] (MOVBreg x)) => (SBFIZ [armBFAuxInt(lc, min(8, 64-lc))] x)
1667
1668 // sbfx
1669 // (x << lc) >> rc
1670 (SRAconst [rc] (SLLconst [lc] x)) && lc <= rc => (SBFX [armBFAuxInt(rc-lc, 64-rc)] x)
1671 // int64(x) >> rc
1672 (SRAconst [rc] (MOVWreg x)) && rc < 32 => (SBFX [armBFAuxInt(rc, 32-rc)] x)
1673 (SRAconst [rc] (MOVHreg x)) && rc < 16 => (SBFX [armBFAuxInt(rc, 16-rc)] x)
1674 (SRAconst [rc] (MOVBreg x)) && rc < 8 => (SBFX [armBFAuxInt(rc, 8-rc)] x)
1675 // merge sbfx and sign-extension into sbfx
1676 (MOVWreg (SBFX [bfc] x)) && bfc.width() <= 32 => (SBFX [bfc] x)
1677 (MOVHreg (SBFX [bfc] x)) && bfc.width() <= 16 => (SBFX [bfc] x)
1678 (MOVBreg (SBFX [bfc] x)) && bfc.width() <= 8 => (SBFX [bfc] x)
1679
1680 // sbfiz/sbfx combinations: merge shifts into bitfield ops
1681 (SRAconst [sc] (SBFIZ [bfc] x)) && sc < bfc.lsb()
1682 => (SBFIZ [armBFAuxInt(bfc.lsb()-sc, bfc.width())] x)
1683 (SRAconst [sc] (SBFIZ [bfc] x)) && sc >= bfc.lsb()
1684 && sc < bfc.lsb()+bfc.width()
1685 => (SBFX [armBFAuxInt(sc-bfc.lsb(), bfc.lsb()+bfc.width()-sc)] x)
1686 (SBFX [bfc] s:(SLLconst [sc] x))
1687 && s.Uses == 1
1688 && sc <= bfc.lsb()
1689 => (SBFX [armBFAuxInt(bfc.lsb() - sc, bfc.width())] x)
1690 (SBFX [bfc] s:(SLLconst [sc] x))
1691 && s.Uses == 1
1692 && sc > bfc.lsb()
1693 => (SBFIZ [armBFAuxInt(sc - bfc.lsb(), bfc.width() - (sc-bfc.lsb()))] x)
1694
1695 // ubfiz
1696 // (x << lc) >> rc
1697 (SRLconst [rc] (SLLconst [lc] x)) && lc > rc => (UBFIZ [armBFAuxInt(lc-rc, 64-lc)] x)
1698 // uint64(x) << lc
1699 (SLLconst [lc] (MOVWUreg x)) => (UBFIZ [armBFAuxInt(lc, min(32, 64-lc))] x)
1700 (SLLconst [lc] (MOVHUreg x)) => (UBFIZ [armBFAuxInt(lc, min(16, 64-lc))] x)
1701 (SLLconst [lc] (MOVBUreg x)) => (UBFIZ [armBFAuxInt(lc, min(8, 64-lc))] x)
1702 // uint64(x << lc)
1703 (MOVWUreg (SLLconst [lc] x)) && lc < 32 => (UBFIZ [armBFAuxInt(lc, 32-lc)] x)
1704 (MOVHUreg (SLLconst [lc] x)) && lc < 16 => (UBFIZ [armBFAuxInt(lc, 16-lc)] x)
1705 (MOVBUreg (SLLconst [lc] x)) && lc < 8 => (UBFIZ [armBFAuxInt(lc, 8-lc)] x)
1706
1707 // merge ANDconst into ubfiz
1708 // (x & ac) << sc
1709 (SLLconst [sc] (ANDconst [ac] x)) && isARM64BFMask(sc, ac, 0)
1710 => (UBFIZ [armBFAuxInt(sc, arm64BFWidth(ac, 0))] x)
1711 // (x << sc) & ac
1712 (ANDconst [ac] (SLLconst [sc] x)) && isARM64BFMask(sc, ac, sc)
1713 => (UBFIZ [armBFAuxInt(sc, arm64BFWidth(ac, sc))] x)
1714
1715 // ubfx
1716 // (x << lc) >> rc
1717 (SRLconst [rc] (SLLconst [lc] x)) && lc < rc => (UBFX [armBFAuxInt(rc-lc, 64-rc)] x)
1718 // uint64(x) >> rc
1719 (SRLconst [rc] (MOVWUreg x)) && rc < 32 => (UBFX [armBFAuxInt(rc, 32-rc)] x)
1720 (SRLconst [rc] (MOVHUreg x)) && rc < 16 => (UBFX [armBFAuxInt(rc, 16-rc)] x)
1721 (SRLconst [rc] (MOVBUreg x)) && rc < 8 => (UBFX [armBFAuxInt(rc, 8-rc)] x)
1722 // uint64(x >> rc)
1723 (MOVWUreg (SRLconst [rc] x)) && rc < 32 => (UBFX [armBFAuxInt(rc, 32)] x)
1724 (MOVHUreg (SRLconst [rc] x)) && rc < 16 => (UBFX [armBFAuxInt(rc, 16)] x)
1725 (MOVBUreg (SRLconst [rc] x)) && rc < 8 => (UBFX [armBFAuxInt(rc, 8)] x)
1726 // merge ANDconst into ubfx
1727 // (x >> sc) & ac
1728 (ANDconst [ac] (SRLconst [sc] x)) && isARM64BFMask(sc, ac, 0)
1729 => (UBFX [armBFAuxInt(sc, arm64BFWidth(ac, 0))] x)
1730 // (x & ac) >> sc
1731 (SRLconst [sc] (ANDconst [ac] x)) && isARM64BFMask(sc, ac, sc)
1732 => (UBFX [armBFAuxInt(sc, arm64BFWidth(ac, sc))] x)
1733 // merge ANDconst and ubfx into ubfx
1734 (ANDconst [c] (UBFX [bfc] x)) && isARM64BFMask(0, c, 0) =>
1735 (UBFX [armBFAuxInt(bfc.lsb(), min(bfc.width(), arm64BFWidth(c, 0)))] x)
1736 (UBFX [bfc] (ANDconst [c] x)) && isARM64BFMask(0, c, 0) && bfc.lsb() + bfc.width() <= arm64BFWidth(c, 0) =>
1737 (UBFX [bfc] x)
1738 // merge ubfx and zero-extension into ubfx
1739 (MOVWUreg (UBFX [bfc] x)) && bfc.width() <= 32 => (UBFX [bfc] x)
1740 (MOVHUreg (UBFX [bfc] x)) && bfc.width() <= 16 => (UBFX [bfc] x)
1741 (MOVBUreg (UBFX [bfc] x)) && bfc.width() <= 8 => (UBFX [bfc] x)
1742
1743 // Extracting bits from across a zero-extension boundary.
1744 (UBFX [bfc] e:(MOVWUreg x))
1745 && e.Uses == 1
1746 && bfc.lsb() < 32
1747 => (UBFX [armBFAuxInt(bfc.lsb(), min(bfc.width(), 32-bfc.lsb()))] x)
1748 (UBFX [bfc] e:(MOVHUreg x))
1749 && e.Uses == 1
1750 && bfc.lsb() < 16
1751 => (UBFX [armBFAuxInt(bfc.lsb(), min(bfc.width(), 16-bfc.lsb()))] x)
1752 (UBFX [bfc] e:(MOVBUreg x))
1753 && e.Uses == 1
1754 && bfc.lsb() < 8
1755 => (UBFX [armBFAuxInt(bfc.lsb(), min(bfc.width(), 8-bfc.lsb()))] x)
1756
1757 // ubfiz/ubfx combinations: merge shifts into bitfield ops
1758 (SRLconst [sc] (UBFX [bfc] x)) && sc < bfc.width()
1759 => (UBFX [armBFAuxInt(bfc.lsb()+sc, bfc.width()-sc)] x)
1760 (UBFX [bfc] (SRLconst [sc] x)) && sc+bfc.width()+bfc.lsb() < 64
1761 => (UBFX [armBFAuxInt(bfc.lsb()+sc, bfc.width())] x)
1762 (SLLconst [sc] (UBFIZ [bfc] x)) && sc+bfc.width()+bfc.lsb() < 64
1763 => (UBFIZ [armBFAuxInt(bfc.lsb()+sc, bfc.width())] x)
1764 (UBFIZ [bfc] (SLLconst [sc] x)) && sc < bfc.width()
1765 => (UBFIZ [armBFAuxInt(bfc.lsb()+sc, bfc.width()-sc)] x)
1766 // ((x << c1) >> c2) >> c3
1767 (SRLconst [sc] (UBFIZ [bfc] x)) && sc == bfc.lsb()
1768 => (ANDconst [1<<uint(bfc.width())-1] x)
1769 (SRLconst [sc] (UBFIZ [bfc] x)) && sc < bfc.lsb()
1770 => (UBFIZ [armBFAuxInt(bfc.lsb()-sc, bfc.width())] x)
1771 (SRLconst [sc] (UBFIZ [bfc] x)) && sc > bfc.lsb()
1772 && sc < bfc.lsb()+bfc.width()
1773 => (UBFX [armBFAuxInt(sc-bfc.lsb(), bfc.lsb()+bfc.width()-sc)] x)
1774 // ((x << c1) << c2) >> c3
1775 (UBFX [bfc] (SLLconst [sc] x)) && sc == bfc.lsb()
1776 => (ANDconst [1<<uint(bfc.width())-1] x)
1777 (UBFX [bfc] (SLLconst [sc] x)) && sc < bfc.lsb()
1778 => (UBFX [armBFAuxInt(bfc.lsb()-sc, bfc.width())] x)
1779 (UBFX [bfc] (SLLconst [sc] x)) && sc > bfc.lsb()
1780 && sc < bfc.lsb()+bfc.width()
1781 => (UBFIZ [armBFAuxInt(sc-bfc.lsb(), bfc.lsb()+bfc.width()-sc)] x)
1782
1783 // bfi
1784 (OR (UBFIZ [bfc] x) (ANDconst [ac] y))
1785 && ac == ^((1<<uint(bfc.width())-1) << uint(bfc.lsb()))
1786 => (BFI [bfc] y x)
1787 (ORshiftLL [s] (ANDconst [xc] x) (ANDconst [yc] y))
1788 && xc == ^(yc << s) // opposite masks
1789 && yc & (yc+1) == 0 // power of 2 minus 1
1790 && yc > 0 // not 0, not all 64 bits (there are better rewrites in that case)
1791 && s+log64(yc+1) <= 64 // shifted mask doesn't overflow
1792 => (BFI [armBFAuxInt(s, log64(yc+1))] x y)
1793 (ORshiftRL [rc] (ANDconst [ac] x) (SLLconst [lc] y))
1794 && lc > rc && ac == ^((1<<uint(64-lc)-1) << uint64(lc-rc))
1795 => (BFI [armBFAuxInt(lc-rc, 64-lc)] x y)
1796 // bfxil
1797 (OR (UBFX [bfc] x) (ANDconst [ac] y)) && ac == ^(1<<uint(bfc.width())-1)
1798 => (BFXIL [bfc] y x)
1799 (ORshiftLL [sc] (UBFX [bfc] x) (SRLconst [sc] y)) && sc == bfc.width()
1800 => (BFXIL [bfc] y x)
1801 (ORshiftRL [rc] (ANDconst [ac] y) (SLLconst [lc] x)) && lc < rc && ac == ^((1<<uint(64-rc)-1))
1802 => (BFXIL [armBFAuxInt(rc-lc, 64-rc)] y x)
1803
1804 // FP simplification
1805 (FNEGS (FMULS x y)) => (FNMULS x y)
1806 (FNEGD (FMULD x y)) => (FNMULD x y)
1807 (FMULS (FNEGS x) y) => (FNMULS x y)
1808 (FMULD (FNEGD x) y) => (FNMULD x y)
1809 (FNEGS (FNMULS x y)) => (FMULS x y)
1810 (FNEGD (FNMULD x y)) => (FMULD x y)
1811 (FNMULS (FNEGS x) y) => (FMULS x y)
1812 (FNMULD (FNEGD x) y) => (FMULD x y)
1813
1814 (FADDS a (FMULS x y)) && a.Block.Func.useFMA(v) => (FMADDS a x y)
1815 (FADDD a (FMULD x y)) && a.Block.Func.useFMA(v) => (FMADDD a x y)
1816 (FSUBS a (FMULS x y)) && a.Block.Func.useFMA(v) => (FMSUBS a x y)
1817 (FSUBD a (FMULD x y)) && a.Block.Func.useFMA(v) => (FMSUBD a x y)
1818 (FSUBS (FMULS x y) a) && a.Block.Func.useFMA(v) => (FNMSUBS a x y)
1819 (FSUBD (FMULD x y) a) && a.Block.Func.useFMA(v) => (FNMSUBD a x y)
1820 (FADDS a (FNMULS x y)) && a.Block.Func.useFMA(v) => (FMSUBS a x y)
1821 (FADDD a (FNMULD x y)) && a.Block.Func.useFMA(v) => (FMSUBD a x y)
1822 (FSUBS a (FNMULS x y)) && a.Block.Func.useFMA(v) => (FMADDS a x y)
1823 (FSUBD a (FNMULD x y)) && a.Block.Func.useFMA(v) => (FMADDD a x y)
1824 (FSUBS (FNMULS x y) a) && a.Block.Func.useFMA(v) => (FNMADDS a x y)
1825 (FSUBD (FNMULD x y) a) && a.Block.Func.useFMA(v) => (FNMADDD a x y)
1826
1827 (MOVBUload [off] {sym} (SB) _) && symIsRO(sym) => (MOVDconst [int64(read8(sym, int64(off)))])
1828 (MOVHUload [off] {sym} (SB) _) && symIsRO(sym) => (MOVDconst [int64(read16(sym, int64(off), config.ctxt.Arch.ByteOrder))])
1829 (MOVWUload [off] {sym} (SB) _) && symIsRO(sym) => (MOVDconst [int64(read32(sym, int64(off), config.ctxt.Arch.ByteOrder))])
1830 (MOVDload [off] {sym} (SB) _) && symIsRO(sym) => (MOVDconst [int64(read64(sym, int64(off), config.ctxt.Arch.ByteOrder))])
1831 (MOVBload [off] {sym} (SB) _) && symIsRO(sym) => (MOVDconst [int64(int8(read8(sym, int64(off))))])
1832 (MOVHload [off] {sym} (SB) _) && symIsRO(sym) => (MOVDconst [int64(int16(read16(sym, int64(off), config.ctxt.Arch.ByteOrder)))])
1833 (MOVWload [off] {sym} (SB) _) && symIsRO(sym) => (MOVDconst [int64(int32(read32(sym, int64(off), config.ctxt.Arch.ByteOrder)))])
1834
1835 // Prefetch instructions (aux is option: 0 - PLDL1KEEP; 1 - PLDL1STRM)
1836 (PrefetchCache addr mem) => (PRFM [0] addr mem)
1837 (PrefetchCacheStreamed addr mem) => (PRFM [1] addr mem)
1838
1839 // Arch-specific inlining for small or disjoint runtime.memmove
1840 (SelectN [0] call:(CALLstatic {sym} s1:(MOVDstore _ (MOVDconst [sz]) s2:(MOVDstore _ src s3:(MOVDstore {t} _ dst mem)))))
1841 && sz >= 0
1842 && isSameCall(sym, "runtime.memmove")
1843 && s1.Uses == 1 && s2.Uses == 1 && s3.Uses == 1
1844 && isInlinableMemmove(dst, src, sz, config)
1845 && clobber(s1, s2, s3, call)
1846 => (Move [sz] dst src mem)
1847
1848 // Match post-lowering calls, register version.
1849 (SelectN [0] call:(CALLstatic {sym} dst src (MOVDconst [sz]) mem))
1850 && sz >= 0
1851 && isSameCall(sym, "runtime.memmove")
1852 && call.Uses == 1
1853 && isInlinableMemmove(dst, src, sz, config)
1854 && clobber(call)
1855 => (Move [sz] dst src mem)
1856
1857 ((REV|REVW) ((REV|REVW) p)) => p
1858
1859 // internal/runtime/math.MulUintptr intrinsics
1860
1861 (Select0 (Mul64uover x y)) => (MUL x y)
1862 (Select1 (Mul64uover x y)) => (NotEqual (CMPconst (UMULH <typ.UInt64> x y) [0]))
1863
1864 // 32 mul 32 -> 64
1865 (MUL r:(MOVWUreg x) s:(MOVWUreg y)) && r.Uses == 1 && s.Uses == 1 => (UMULL x y)
1866 (MUL r:(MOVWreg x) s:(MOVWreg y)) && r.Uses == 1 && s.Uses == 1 => (MULL x y)
1867
View as plain text