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
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
|
/**************************************************************************
*
* Copyright 2009 VMware, Inc.
* All Rights Reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the
* "Software"), to deal in the Software without restriction, including
* without limitation the rights to use, copy, modify, merge, publish,
* distribute, sub license, and/or sell copies of the Software, and to
* permit persons to whom the Software is furnished to do so, subject to
* the following conditions:
*
* The above copyright notice and this permission notice (including the
* next paragraph) shall be included in all copies or substantial portions
* of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
* OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
* IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
* ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
*
**************************************************************************/
/**
* @file
* Helper functions for swizzling/shuffling.
*
* @author Jose Fonseca <jfonseca@vmware.com>
*/
#include "util/u_debug.h"
#include "lp_bld_type.h"
#include "lp_bld_const.h"
#include "lp_bld_init.h"
#include "lp_bld_logic.h"
#include "lp_bld_swizzle.h"
LLVMValueRef
lp_build_broadcast(struct gallivm_state *gallivm,
LLVMTypeRef vec_type,
LLVMValueRef scalar)
{
LLVMValueRef res;
if (LLVMGetTypeKind(vec_type) != LLVMVectorTypeKind) {
/* scalar */
assert(vec_type == LLVMTypeOf(scalar));
res = scalar;
} else {
LLVMBuilderRef builder = gallivm->builder;
const unsigned length = LLVMGetVectorSize(vec_type);
LLVMValueRef undef = LLVMGetUndef(vec_type);
LLVMTypeRef i32_type = LLVMInt32TypeInContext(gallivm->context);
assert(LLVMGetElementType(vec_type) == LLVMTypeOf(scalar));
if (HAVE_LLVM >= 0x207) {
/* The shuffle vector is always made of int32 elements */
LLVMTypeRef i32_vec_type = LLVMVectorType(i32_type, length);
res = LLVMBuildInsertElement(builder, undef, scalar, LLVMConstNull(i32_type), "");
res = LLVMBuildShuffleVector(builder, res, undef, LLVMConstNull(i32_vec_type), "");
} else {
/* XXX: The above path provokes a bug in LLVM 2.6 */
unsigned i;
res = undef;
for(i = 0; i < length; ++i) {
LLVMValueRef index = lp_build_const_int32(gallivm, i);
res = LLVMBuildInsertElement(builder, res, scalar, index, "");
}
}
}
return res;
}
/**
* Broadcast
*/
LLVMValueRef
lp_build_broadcast_scalar(struct lp_build_context *bld,
LLVMValueRef scalar)
{
assert(lp_check_elem_type(bld->type, LLVMTypeOf(scalar)));
return lp_build_broadcast(bld->gallivm, bld->vec_type, scalar);
}
/**
* Combined extract and broadcast (or a mere shuffle when the two types match)
*/
LLVMValueRef
lp_build_extract_broadcast(struct gallivm_state *gallivm,
struct lp_type src_type,
struct lp_type dst_type,
LLVMValueRef vector,
LLVMValueRef index)
{
LLVMTypeRef i32t = LLVMInt32TypeInContext(gallivm->context);
LLVMValueRef res;
assert(src_type.floating == dst_type.floating);
assert(src_type.width == dst_type.width);
assert(lp_check_value(src_type, vector));
assert(LLVMTypeOf(index) == i32t);
if (src_type.length == 1) {
if (dst_type.length == 1) {
/*
* Trivial scalar -> scalar.
*/
res = vector;
}
else {
/*
* Broadcast scalar -> vector.
*/
res = lp_build_broadcast(gallivm,
lp_build_vec_type(gallivm, dst_type),
vector);
}
}
else {
if (dst_type.length == src_type.length) {
/*
* Special shuffle of the same size.
*/
LLVMValueRef shuffle;
shuffle = lp_build_broadcast(gallivm,
LLVMVectorType(i32t, dst_type.length),
index);
res = LLVMBuildShuffleVector(gallivm->builder, vector,
LLVMGetUndef(lp_build_vec_type(gallivm, dst_type)),
shuffle, "");
}
else {
LLVMValueRef scalar;
scalar = LLVMBuildExtractElement(gallivm->builder, vector, index, "");
if (dst_type.length == 1) {
/*
* Trivial extract scalar from vector.
*/
res = scalar;
}
else {
/*
* General case of different sized vectors.
*/
res = lp_build_broadcast(gallivm,
lp_build_vec_type(gallivm, dst_type),
vector);
}
}
}
return res;
}
/**
* Swizzle one channel into all other three channels.
*/
LLVMValueRef
lp_build_swizzle_scalar_aos(struct lp_build_context *bld,
LLVMValueRef a,
unsigned channel)
{
LLVMBuilderRef builder = bld->gallivm->builder;
const struct lp_type type = bld->type;
const unsigned n = type.length;
unsigned i, j;
if(a == bld->undef || a == bld->zero || a == bld->one)
return a;
/* XXX: SSE3 has PSHUFB which should be better than bitmasks, but forcing
* using shuffles here actually causes worst results. More investigation is
* needed. */
if (type.width >= 16) {
/*
* Shuffle.
*/
LLVMTypeRef elem_type = LLVMInt32TypeInContext(bld->gallivm->context);
LLVMValueRef shuffles[LP_MAX_VECTOR_LENGTH];
for(j = 0; j < n; j += 4)
for(i = 0; i < 4; ++i)
shuffles[j + i] = LLVMConstInt(elem_type, j + channel, 0);
return LLVMBuildShuffleVector(builder, a, bld->undef, LLVMConstVector(shuffles, n), "");
}
else {
/*
* Bit mask and recursive shifts
*
* XYZW XYZW .... XYZW <= input
* 0Y00 0Y00 .... 0Y00
* YY00 YY00 .... YY00
* YYYY YYYY .... YYYY <= output
*/
struct lp_type type4;
const char shifts[4][2] = {
{ 1, 2},
{-1, 2},
{ 1, -2},
{-1, -2}
};
unsigned i;
a = LLVMBuildAnd(builder, a,
lp_build_const_mask_aos(bld->gallivm,
type, 1 << channel), "");
/*
* Build a type where each element is an integer that cover the four
* channels.
*/
type4 = type;
type4.floating = FALSE;
type4.width *= 4;
type4.length /= 4;
a = LLVMBuildBitCast(builder, a, lp_build_vec_type(bld->gallivm, type4), "");
for(i = 0; i < 2; ++i) {
LLVMValueRef tmp = NULL;
int shift = shifts[channel][i];
#ifdef PIPE_ARCH_LITTLE_ENDIAN
shift = -shift;
#endif
if(shift > 0)
tmp = LLVMBuildLShr(builder, a, lp_build_const_int_vec(bld->gallivm, type4, shift*type.width), "");
if(shift < 0)
tmp = LLVMBuildShl(builder, a, lp_build_const_int_vec(bld->gallivm, type4, -shift*type.width), "");
assert(tmp);
if(tmp)
a = LLVMBuildOr(builder, a, tmp, "");
}
return LLVMBuildBitCast(builder, a, lp_build_vec_type(bld->gallivm, type), "");
}
}
LLVMValueRef
lp_build_swizzle_aos(struct lp_build_context *bld,
LLVMValueRef a,
const unsigned char swizzles[4])
{
LLVMBuilderRef builder = bld->gallivm->builder;
const struct lp_type type = bld->type;
const unsigned n = type.length;
unsigned i, j;
if (swizzles[0] == PIPE_SWIZZLE_RED &&
swizzles[1] == PIPE_SWIZZLE_GREEN &&
swizzles[2] == PIPE_SWIZZLE_BLUE &&
swizzles[3] == PIPE_SWIZZLE_ALPHA) {
return a;
}
if (swizzles[0] == swizzles[1] &&
swizzles[1] == swizzles[2] &&
swizzles[2] == swizzles[3]) {
switch (swizzles[0]) {
case PIPE_SWIZZLE_RED:
case PIPE_SWIZZLE_GREEN:
case PIPE_SWIZZLE_BLUE:
case PIPE_SWIZZLE_ALPHA:
return lp_build_swizzle_scalar_aos(bld, a, swizzles[0]);
case PIPE_SWIZZLE_ZERO:
return bld->zero;
case PIPE_SWIZZLE_ONE:
return bld->one;
default:
assert(0);
return bld->undef;
}
}
if (type.width >= 16) {
/*
* Shuffle.
*/
LLVMValueRef undef = LLVMGetUndef(lp_build_elem_type(bld->gallivm, type));
LLVMTypeRef i32t = LLVMInt32TypeInContext(bld->gallivm->context);
LLVMValueRef shuffles[LP_MAX_VECTOR_LENGTH];
LLVMValueRef aux[LP_MAX_VECTOR_LENGTH];
memset(aux, 0, sizeof aux);
for(j = 0; j < n; j += 4) {
for(i = 0; i < 4; ++i) {
unsigned shuffle;
switch (swizzles[i]) {
default:
assert(0);
/* fall through */
case PIPE_SWIZZLE_RED:
case PIPE_SWIZZLE_GREEN:
case PIPE_SWIZZLE_BLUE:
case PIPE_SWIZZLE_ALPHA:
shuffle = j + swizzles[i];
break;
case PIPE_SWIZZLE_ZERO:
shuffle = type.length + 0;
if (!aux[0]) {
aux[0] = lp_build_const_elem(bld->gallivm, type, 0.0);
}
break;
case PIPE_SWIZZLE_ONE:
shuffle = type.length + 1;
if (!aux[1]) {
aux[1] = lp_build_const_elem(bld->gallivm, type, 1.0);
}
break;
}
shuffles[j + i] = LLVMConstInt(i32t, shuffle, 0);
}
}
for (i = 0; i < n; ++i) {
if (!aux[i]) {
aux[i] = undef;
}
}
return LLVMBuildShuffleVector(builder, a,
LLVMConstVector(aux, n),
LLVMConstVector(shuffles, n), "");
} else {
/*
* Bit mask and shifts.
*
* For example, this will convert BGRA to RGBA by doing
*
* rgba = (bgra & 0x00ff0000) >> 16
* | (bgra & 0xff00ff00)
* | (bgra & 0x000000ff) << 16
*
* This is necessary not only for faster cause, but because X86 backend
* will refuse shuffles of <4 x i8> vectors
*/
LLVMValueRef res;
struct lp_type type4;
unsigned cond = 0;
unsigned chan;
int shift;
/*
* Start with a mixture of 1 and 0.
*/
for (chan = 0; chan < 4; ++chan) {
if (swizzles[chan] == PIPE_SWIZZLE_ONE) {
cond |= 1 << chan;
}
}
res = lp_build_select_aos(bld, cond, bld->one, bld->zero);
/*
* Build a type where each element is an integer that cover the four
* channels.
*/
type4 = type;
type4.floating = FALSE;
type4.width *= 4;
type4.length /= 4;
a = LLVMBuildBitCast(builder, a, lp_build_vec_type(bld->gallivm, type4), "");
res = LLVMBuildBitCast(builder, res, lp_build_vec_type(bld->gallivm, type4), "");
/*
* Mask and shift the channels, trying to group as many channels in the
* same shift as possible
*/
for (shift = -3; shift <= 3; ++shift) {
unsigned long long mask = 0;
assert(type4.width <= sizeof(mask)*8);
for (chan = 0; chan < 4; ++chan) {
/* FIXME: big endian */
if (swizzles[chan] < 4 &&
chan - swizzles[chan] == shift) {
mask |= ((1ULL << type.width) - 1) << (swizzles[chan] * type.width);
}
}
if (mask) {
LLVMValueRef masked;
LLVMValueRef shifted;
if (0)
debug_printf("shift = %i, mask = 0x%08llx\n", shift, mask);
masked = LLVMBuildAnd(builder, a,
lp_build_const_int_vec(bld->gallivm, type4, mask), "");
if (shift > 0) {
shifted = LLVMBuildShl(builder, masked,
lp_build_const_int_vec(bld->gallivm, type4, shift*type.width), "");
} else if (shift < 0) {
shifted = LLVMBuildLShr(builder, masked,
lp_build_const_int_vec(bld->gallivm, type4, -shift*type.width), "");
} else {
shifted = masked;
}
res = LLVMBuildOr(builder, res, shifted, "");
}
}
return LLVMBuildBitCast(builder, res,
lp_build_vec_type(bld->gallivm, type), "");
}
}
/**
* Extended swizzle of a single channel of a SoA vector.
*
* @param bld building context
* @param unswizzled array with the 4 unswizzled values
* @param swizzle one of the PIPE_SWIZZLE_*
*
* @return the swizzled value.
*/
LLVMValueRef
lp_build_swizzle_soa_channel(struct lp_build_context *bld,
const LLVMValueRef *unswizzled,
unsigned swizzle)
{
switch (swizzle) {
case PIPE_SWIZZLE_RED:
case PIPE_SWIZZLE_GREEN:
case PIPE_SWIZZLE_BLUE:
case PIPE_SWIZZLE_ALPHA:
return unswizzled[swizzle];
case PIPE_SWIZZLE_ZERO:
return bld->zero;
case PIPE_SWIZZLE_ONE:
return bld->one;
default:
assert(0);
return bld->undef;
}
}
/**
* Extended swizzle of a SoA vector.
*
* @param bld building context
* @param unswizzled array with the 4 unswizzled values
* @param swizzles array of PIPE_SWIZZLE_*
* @param swizzled output swizzled values
*/
void
lp_build_swizzle_soa(struct lp_build_context *bld,
const LLVMValueRef *unswizzled,
const unsigned char swizzles[4],
LLVMValueRef *swizzled)
{
unsigned chan;
for (chan = 0; chan < 4; ++chan) {
swizzled[chan] = lp_build_swizzle_soa_channel(bld, unswizzled,
swizzles[chan]);
}
}
/**
* Do an extended swizzle of a SoA vector inplace.
*
* @param bld building context
* @param values intput/output array with the 4 values
* @param swizzles array of PIPE_SWIZZLE_*
*/
void
lp_build_swizzle_soa_inplace(struct lp_build_context *bld,
LLVMValueRef *values,
const unsigned char swizzles[4])
{
LLVMValueRef unswizzled[4];
unsigned chan;
for (chan = 0; chan < 4; ++chan) {
unswizzled[chan] = values[chan];
}
lp_build_swizzle_soa(bld, unswizzled, swizzles, values);
}
|