summaryrefslogtreecommitdiffstats
path: root/vendor/golang.org/x/image/vector/gen.go
blob: 28b298b5ea00db349e287ac03e268194423cbbd0 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
// Copyright 2016 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.

// +build ignore

package main

import (
	"bytes"
	"io/ioutil"
	"log"
	"strings"
	"text/template"
)

const (
	copyright = "" +
		"// Copyright 2016 The Go Authors. All rights reserved.\n" +
		"// Use of this source code is governed by a BSD-style\n" +
		"// license that can be found in the LICENSE file.\n"

	doNotEdit = "// generated by go run gen.go; DO NOT EDIT\n"

	dashDashDash = "// --------"
)

func main() {
	tmpl, err := ioutil.ReadFile("gen_acc_amd64.s.tmpl")
	if err != nil {
		log.Fatalf("ReadFile: %v", err)
	}
	if !bytes.HasPrefix(tmpl, []byte(copyright)) {
		log.Fatal("source template did not start with the copyright header")
	}
	tmpl = tmpl[len(copyright):]

	preamble := []byte(nil)
	if i := bytes.Index(tmpl, []byte(dashDashDash)); i < 0 {
		log.Fatalf("source template did not contain %q", dashDashDash)
	} else {
		preamble, tmpl = tmpl[:i], tmpl[i:]
	}

	t, err := template.New("").Parse(string(tmpl))
	if err != nil {
		log.Fatalf("Parse: %v", err)
	}

	out := bytes.NewBuffer(nil)
	out.WriteString(doNotEdit)
	out.Write(preamble)

	for i, v := range instances {
		if i != 0 {
			out.WriteString("\n")
		}
		if strings.Contains(v.LoadArgs, "{{.ShortName}}") {
			v.LoadArgs = strings.Replace(v.LoadArgs, "{{.ShortName}}", v.ShortName, -1)
		}
		if err := t.Execute(out, v); err != nil {
			log.Fatalf("Execute(%q): %v", v.ShortName, err)
		}
	}

	if err := ioutil.WriteFile("acc_amd64.s", out.Bytes(), 0666); err != nil {
		log.Fatalf("WriteFile: %v", err)
	}
}

var instances = []struct {
	LongName       string
	ShortName      string
	FrameSize      string
	ArgsSize       string
	Args           string
	DstElemSize1   int
	DstElemSize4   int
	XMM3           string
	XMM4           string
	XMM5           string
	XMM6           string
	XMM8           string
	XMM9           string
	XMM10          string
	LoadArgs       string
	Setup          string
	LoadXMMRegs    string
	Add            string
	ClampAndScale  string
	ConvertToInt32 string
	Store4         string
	Store1         string
}{{
	LongName:       "fixedAccumulateOpOver",
	ShortName:      "fxAccOpOver",
	FrameSize:      fxFrameSize,
	ArgsSize:       twoArgArgsSize,
	Args:           "dst []uint8, src []uint32",
	DstElemSize1:   1 * sizeOfUint8,
	DstElemSize4:   4 * sizeOfUint8,
	XMM3:           fxXMM3,
	XMM4:           fxXMM4,
	XMM5:           fxXMM5,
	XMM6:           opOverXMM6,
	XMM8:           opOverXMM8,
	XMM9:           opOverXMM9,
	XMM10:          opOverXMM10,
	LoadArgs:       twoArgLoadArgs,
	Setup:          fxSetup,
	LoadXMMRegs:    fxLoadXMMRegs + "\n" + opOverLoadXMMRegs,
	Add:            fxAdd,
	ClampAndScale:  fxClampAndScale,
	ConvertToInt32: fxConvertToInt32,
	Store4:         opOverStore4,
	Store1:         opOverStore1,
}, {
	LongName:       "fixedAccumulateOpSrc",
	ShortName:      "fxAccOpSrc",
	FrameSize:      fxFrameSize,
	ArgsSize:       twoArgArgsSize,
	Args:           "dst []uint8, src []uint32",
	DstElemSize1:   1 * sizeOfUint8,
	DstElemSize4:   4 * sizeOfUint8,
	XMM3:           fxXMM3,
	XMM4:           fxXMM4,
	XMM5:           fxXMM5,
	XMM6:           opSrcXMM6,
	XMM8:           opSrcXMM8,
	XMM9:           opSrcXMM9,
	XMM10:          opSrcXMM10,
	LoadArgs:       twoArgLoadArgs,
	Setup:          fxSetup,
	LoadXMMRegs:    fxLoadXMMRegs + "\n" + opSrcLoadXMMRegs,
	Add:            fxAdd,
	ClampAndScale:  fxClampAndScale,
	ConvertToInt32: fxConvertToInt32,
	Store4:         opSrcStore4,
	Store1:         opSrcStore1,
}, {
	LongName:       "fixedAccumulateMask",
	ShortName:      "fxAccMask",
	FrameSize:      fxFrameSize,
	ArgsSize:       oneArgArgsSize,
	Args:           "buf []uint32",
	DstElemSize1:   1 * sizeOfUint32,
	DstElemSize4:   4 * sizeOfUint32,
	XMM3:           fxXMM3,
	XMM4:           fxXMM4,
	XMM5:           fxXMM5,
	XMM6:           maskXMM6,
	XMM8:           maskXMM8,
	XMM9:           maskXMM9,
	XMM10:          maskXMM10,
	LoadArgs:       oneArgLoadArgs,
	Setup:          fxSetup,
	LoadXMMRegs:    fxLoadXMMRegs + "\n" + maskLoadXMMRegs,
	Add:            fxAdd,
	ClampAndScale:  fxClampAndScale,
	ConvertToInt32: fxConvertToInt32,
	Store4:         maskStore4,
	Store1:         maskStore1,
}, {
	LongName:       "floatingAccumulateOpOver",
	ShortName:      "flAccOpOver",
	FrameSize:      flFrameSize,
	ArgsSize:       twoArgArgsSize,
	Args:           "dst []uint8, src []float32",
	DstElemSize1:   1 * sizeOfUint8,
	DstElemSize4:   4 * sizeOfUint8,
	XMM3:           flXMM3,
	XMM4:           flXMM4,
	XMM5:           flXMM5,
	XMM6:           opOverXMM6,
	XMM8:           opOverXMM8,
	XMM9:           opOverXMM9,
	XMM10:          opOverXMM10,
	LoadArgs:       twoArgLoadArgs,
	Setup:          flSetup,
	LoadXMMRegs:    flLoadXMMRegs + "\n" + opOverLoadXMMRegs,
	Add:            flAdd,
	ClampAndScale:  flClampAndScale,
	ConvertToInt32: flConvertToInt32,
	Store4:         opOverStore4,
	Store1:         opOverStore1,
}, {
	LongName:       "floatingAccumulateOpSrc",
	ShortName:      "flAccOpSrc",
	FrameSize:      flFrameSize,
	ArgsSize:       twoArgArgsSize,
	Args:           "dst []uint8, src []float32",
	DstElemSize1:   1 * sizeOfUint8,
	DstElemSize4:   4 * sizeOfUint8,
	XMM3:           flXMM3,
	XMM4:           flXMM4,
	XMM5:           flXMM5,
	XMM6:           opSrcXMM6,
	XMM8:           opSrcXMM8,
	XMM9:           opSrcXMM9,
	XMM10:          opSrcXMM10,
	LoadArgs:       twoArgLoadArgs,
	Setup:          flSetup,
	LoadXMMRegs:    flLoadXMMRegs + "\n" + opSrcLoadXMMRegs,
	Add:            flAdd,
	ClampAndScale:  flClampAndScale,
	ConvertToInt32: flConvertToInt32,
	Store4:         opSrcStore4,
	Store1:         opSrcStore1,
}, {
	LongName:       "floatingAccumulateMask",
	ShortName:      "flAccMask",
	FrameSize:      flFrameSize,
	ArgsSize:       twoArgArgsSize,
	Args:           "dst []uint32, src []float32",
	DstElemSize1:   1 * sizeOfUint32,
	DstElemSize4:   4 * sizeOfUint32,
	XMM3:           flXMM3,
	XMM4:           flXMM4,
	XMM5:           flXMM5,
	XMM6:           maskXMM6,
	XMM8:           maskXMM8,
	XMM9:           maskXMM9,
	XMM10:          maskXMM10,
	LoadArgs:       twoArgLoadArgs,
	Setup:          flSetup,
	LoadXMMRegs:    flLoadXMMRegs + "\n" + maskLoadXMMRegs,
	Add:            flAdd,
	ClampAndScale:  flClampAndScale,
	ConvertToInt32: flConvertToInt32,
	Store4:         maskStore4,
	Store1:         maskStore1,
}}

const (
	fxFrameSize = `0`
	flFrameSize = `8`

	oneArgArgsSize = `24`
	twoArgArgsSize = `48`

	sizeOfUint8  = 1
	sizeOfUint32 = 4

	fxXMM3 = `-`
	flXMM3 = `flSignMask`

	fxXMM4 = `-`
	flXMM4 = `flOne`

	fxXMM5 = `fxAlmost65536`
	flXMM5 = `flAlmost65536`

	oneArgLoadArgs = `
		MOVQ buf_base+0(FP), DI
		MOVQ buf_len+8(FP), BX
		MOVQ buf_base+0(FP), SI
		MOVQ buf_len+8(FP), R10
		`
	twoArgLoadArgs = `
		MOVQ dst_base+0(FP), DI
		MOVQ dst_len+8(FP), BX
		MOVQ src_base+24(FP), SI
		MOVQ src_len+32(FP), R10
		// Sanity check that len(dst) >= len(src).
		CMPQ BX, R10
		JLT  {{.ShortName}}End
		`

	fxSetup = ``
	flSetup = `
		// Prepare to set MXCSR bits 13 and 14, so that the CVTPS2PL below is
		// "Round To Zero".
		STMXCSR mxcsrOrig-8(SP)
		MOVL    mxcsrOrig-8(SP), AX
		ORL     $0x6000, AX
		MOVL    AX, mxcsrNew-4(SP)
		`

	fxLoadXMMRegs = `
		// fxAlmost65536 := XMM(0x0000ffff repeated four times) // Maximum of an uint16.
		MOVOU fxAlmost65536<>(SB), X5
		`
	flLoadXMMRegs = `
		// flSignMask    := XMM(0x7fffffff repeated four times) // All but the sign bit of a float32.
		// flOne         := XMM(0x3f800000 repeated four times) // 1 as a float32.
		// flAlmost65536 := XMM(0x477fffff repeated four times) // 255.99998 * 256 as a float32.
		MOVOU flSignMask<>(SB), X3
		MOVOU flOne<>(SB), X4
		MOVOU flAlmost65536<>(SB), X5
		`

	fxAdd = `PADDD`
	flAdd = `ADDPS`

	fxClampAndScale = `
		// y = abs(x)
		// y >>= 2 // Shift by 2*ϕ - 16.
		// y = min(y, fxAlmost65536)
		//
		// pabsd  %xmm1,%xmm2
		// psrld  $0x2,%xmm2
		// pminud %xmm5,%xmm2
		//
		// Hopefully we'll get these opcode mnemonics into the assembler for Go
		// 1.8. https://golang.org/issue/16007 isn't exactly the same thing, but
		// it's similar.
		BYTE $0x66; BYTE $0x0f; BYTE $0x38; BYTE $0x1e; BYTE $0xd1
		BYTE $0x66; BYTE $0x0f; BYTE $0x72; BYTE $0xd2; BYTE $0x02
		BYTE $0x66; BYTE $0x0f; BYTE $0x38; BYTE $0x3b; BYTE $0xd5
		`
	flClampAndScale = `
		// y = x & flSignMask
		// y = min(y, flOne)
		// y = mul(y, flAlmost65536)
		MOVOU X3, X2
		ANDPS X1, X2
		MINPS X4, X2
		MULPS X5, X2
		`

	fxConvertToInt32 = `
		// z = convertToInt32(y)
		// No-op.
		`
	flConvertToInt32 = `
		// z = convertToInt32(y)
		LDMXCSR  mxcsrNew-4(SP)
		CVTPS2PL X2, X2
		LDMXCSR  mxcsrOrig-8(SP)
		`

	opOverStore4 = `
		// Blend over the dst's prior value. SIMD for i in 0..3:
		//
		// dstA := uint32(dst[i]) * 0x101
		// maskA := z@i
		// outA := dstA*(0xffff-maskA)/0xffff + maskA
		// dst[i] = uint8(outA >> 8)
		//
		// First, set X0 to dstA*(0xfff-maskA).
		MOVL   (DI), X0
		PSHUFB X8, X0
		MOVOU  X9, X11
		PSUBL  X2, X11
		PMULLD X11, X0
		// We implement uint32 division by 0xffff as multiplication by a magic
		// constant (0x800080001) and then a shift by a magic constant (47).
		// See TestDivideByFFFF for a justification.
		//
		// That multiplication widens from uint32 to uint64, so we have to
		// duplicate and shift our four uint32s from one XMM register (X0) to
		// two XMM registers (X0 and X11).
		//
		// Move the second and fourth uint32s in X0 to be the first and third
		// uint32s in X11.
		MOVOU X0, X11
		PSRLQ $32, X11
		// Multiply by magic, shift by magic.
		//
		// pmuludq %xmm10,%xmm0
		// pmuludq %xmm10,%xmm11
		BYTE  $0x66; BYTE $0x41; BYTE $0x0f; BYTE $0xf4; BYTE $0xc2
		BYTE  $0x66; BYTE $0x45; BYTE $0x0f; BYTE $0xf4; BYTE $0xda
		PSRLQ $47, X0
		PSRLQ $47, X11
		// Merge the two registers back to one, X11, and add maskA.
		PSLLQ $32, X11
		XORPS X0, X11
		PADDD X11, X2
		// As per opSrcStore4, shuffle and copy the 4 second-lowest bytes.
		PSHUFB X6, X2
		MOVL   X2, (DI)
		`
	opSrcStore4 = `
		// z = shuffleTheSecondLowestBytesOfEach4ByteElement(z)
		// copy(dst[:4], low4BytesOf(z))
		PSHUFB X6, X2
		MOVL   X2, (DI)
		`
	maskStore4 = `
		// copy(dst[:4], z)
		MOVOU X2, (DI)
		`

	opOverStore1 = `
		// Blend over the dst's prior value.
		//
		// dstA := uint32(dst[0]) * 0x101
		// maskA := z
		// outA := dstA*(0xffff-maskA)/0xffff + maskA
		// dst[0] = uint8(outA >> 8)
		MOVBLZX (DI), R12
		IMULL   $0x101, R12
		MOVL    X2, R13
		MOVL    $0xffff, AX
		SUBL    R13, AX
		MULL    R12             // MULL's implicit arg is AX, and the result is stored in DX:AX.
		MOVL    $0x80008001, BX // Divide by 0xffff is to first multiply by a magic constant...
		MULL    BX              // MULL's implicit arg is AX, and the result is stored in DX:AX.
		SHRL    $15, DX         // ...and then shift by another magic constant (47 - 32 = 15).
		ADDL    DX, R13
		SHRL    $8, R13
		MOVB    R13, (DI)
		`
	opSrcStore1 = `
		// dst[0] = uint8(z>>8)
		MOVL X2, BX
		SHRL $8, BX
		MOVB BX, (DI)
		`
	maskStore1 = `
		// dst[0] = uint32(z)
		MOVL X2, (DI)
		`

	opOverXMM6 = `gather`
	opSrcXMM6  = `gather`
	maskXMM6   = `-`

	opOverXMM8 = `scatterAndMulBy0x101`
	opSrcXMM8  = `-`
	maskXMM8   = `-`

	opOverXMM9 = `fxAlmost65536`
	opSrcXMM9  = `-`
	maskXMM9   = `-`

	opOverXMM10 = `inverseFFFF`
	opSrcXMM10  = `-`
	maskXMM10   = `-`

	opOverLoadXMMRegs = `
		// gather               := XMM(see above)                      // PSHUFB shuffle mask.
		// scatterAndMulBy0x101 := XMM(see above)                      // PSHUFB shuffle mask.
		// fxAlmost65536        := XMM(0x0000ffff repeated four times) // 0xffff.
		// inverseFFFF          := XMM(0x80008001 repeated four times) // Magic constant for dividing by 0xffff.
		MOVOU gather<>(SB), X6
		MOVOU scatterAndMulBy0x101<>(SB), X8
		MOVOU fxAlmost65536<>(SB), X9
		MOVOU inverseFFFF<>(SB), X10
		`
	opSrcLoadXMMRegs = `
		// gather := XMM(see above) // PSHUFB shuffle mask.
		MOVOU gather<>(SB), X6
		`
	maskLoadXMMRegs = ``
)