3
/**************************************************************************
5
* Copyright 2009-2010 VMware, Inc.
8
* Permission is hereby granted, free of charge, to any person obtaining a
9
* copy of this software and associated documentation files (the
10
* "Software"), to deal in the Software without restriction, including
11
* without limitation the rights to use, copy, modify, merge, publish,
12
* distribute, sub license, and/or sell copies of the Software, and to
13
* permit persons to whom the Software is furnished to do so, subject to
14
* the following conditions:
16
* The above copyright notice and this permission notice (including the
17
* next paragraph) shall be included in all copies or substantial portions
20
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
21
* OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
22
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
23
* IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
24
* ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
25
* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
26
* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
28
**************************************************************************/
32
* Pixel format packing and unpacking functions.
34
* @author Jose Fonseca <jfonseca@vmware.com>
40
from u_format_parse import *
43
def inv_swizzles(swizzles):
44
'''Return an array[4] of inverse swizzle terms'''
45
'''Only pick the first matching value to avoid l8 getting blue and i8 getting alpha'''
46
inv_swizzle = [None]*4
49
if swizzle < 4 and inv_swizzle[swizzle] == None:
50
inv_swizzle[swizzle] = i
53
def print_channels(format, func):
54
if format.nr_channels() <= 1:
55
func(format.le_channels, format.le_swizzles)
57
if (format.le_channels == format.be_channels and
58
[c.shift for c in format.le_channels] ==
59
[c.shift for c in format.be_channels] and
60
format.le_swizzles == format.be_swizzles):
61
func(format.le_channels, format.le_swizzles)
63
print('#if UTIL_ARCH_BIG_ENDIAN')
64
func(format.be_channels, format.be_swizzles)
66
func(format.le_channels, format.le_swizzles)
69
def generate_format_type(format):
70
'''Generate a structure that describes the format.'''
72
assert format.layout == PLAIN
74
def generate_bitfields(channels, swizzles):
75
for channel in channels:
76
if channel.type == VOID:
78
print(' unsigned %s:%u;' % (channel.name, channel.size))
79
elif channel.type == UNSIGNED:
80
print(' unsigned %s:%u;' % (channel.name, channel.size))
81
elif channel.type in (SIGNED, FIXED):
82
print(' int %s:%u;' % (channel.name, channel.size))
83
elif channel.type == FLOAT:
84
if channel.size == 64:
85
print(' double %s;' % (channel.name))
86
elif channel.size == 32:
87
print(' float %s;' % (channel.name))
89
print(' unsigned %s:%u;' % (channel.name, channel.size))
93
def generate_full_fields(channels, swizzles):
94
for channel in channels:
95
assert channel.size % 8 == 0 and is_pot(channel.size)
96
if channel.type == VOID:
98
print(' uint%u_t %s;' % (channel.size, channel.name))
99
elif channel.type == UNSIGNED:
100
print(' uint%u_t %s;' % (channel.size, channel.name))
101
elif channel.type in (SIGNED, FIXED):
102
print(' int%u_t %s;' % (channel.size, channel.name))
103
elif channel.type == FLOAT:
104
if channel.size == 64:
105
print(' double %s;' % (channel.name))
106
elif channel.size == 32:
107
print(' float %s;' % (channel.name))
108
elif channel.size == 16:
109
print(' uint16_t %s;' % (channel.name))
115
use_bitfields = False
116
for channel in format.le_channels:
117
if channel.size % 8 or not is_pot(channel.size):
120
print('struct util_format_%s {' % format.short_name())
122
print_channels(format, generate_bitfields)
124
print_channels(format, generate_full_fields)
129
def is_format_supported(format):
130
'''Determines whether we actually have the plumbing necessary to generate the
131
to read/write to/from this format.'''
133
# FIXME: Ideally we would support any format combination here.
135
if format.layout != PLAIN:
139
channel = format.le_channels[i]
140
if channel.type not in (VOID, UNSIGNED, SIGNED, FLOAT, FIXED):
142
if channel.type == FLOAT and channel.size not in (16, 32, 64):
147
def native_type(format):
148
'''Get the native appropriate for a format.'''
150
if format.name == 'PIPE_FORMAT_R11G11B10_FLOAT':
152
if format.name == 'PIPE_FORMAT_R9G9B9E5_FLOAT':
155
if format.layout == PLAIN:
156
if not format.is_array():
157
# For arithmetic pixel formats return the integer type that matches the whole pixel
158
return 'uint%u_t' % format.block_size()
160
# For array pixel formats return the integer type that matches the color channel
161
channel = format.array_element()
162
if channel.type in (UNSIGNED, VOID):
163
return 'uint%u_t' % channel.size
164
elif channel.type in (SIGNED, FIXED):
165
return 'int%u_t' % channel.size
166
elif channel.type == FLOAT:
167
if channel.size == 16:
169
elif channel.size == 32:
171
elif channel.size == 64:
181
def intermediate_native_type(bits, sign):
182
'''Find a native type adequate to hold intermediate results of the request bit size.'''
184
bytes = 4 # don't use anything smaller than 32bits
185
while bytes * 8 < bits:
190
return 'int%u_t' % bits
192
return 'uint%u_t' % bits
195
def get_one_shift(type):
196
'''Get the number of the bit that matches unity for this type.'''
197
if type.type == 'FLOAT':
201
if type.type == UNSIGNED:
203
if type.type == SIGNED:
205
if type.type == FIXED:
210
def truncate_mantissa(x, bits):
211
'''Truncate an integer so it can be represented exactly with a floating
214
assert isinstance(x, int)
221
# We can represent integers up to mantissa + 1 bits exactly
222
mask = (1 << (bits + 1)) - 1
224
# Slide the mask until the MSB matches
226
while (x >> shift) & ~mask:
234
def value_to_native(type, value):
235
'''Get the value of unity for this type.'''
236
if type.type == FLOAT:
238
and isinstance(value, int):
239
return truncate_mantissa(value, 23)
241
if type.type == FIXED:
242
return int(value * (1 << (type.size // 2)))
245
if type.type == UNSIGNED:
246
return int(value * ((1 << type.size) - 1))
247
if type.type == SIGNED:
248
return int(value * ((1 << (type.size - 1)) - 1))
252
def native_to_constant(type, value):
253
'''Get the value of unity for this type.'''
254
if type.type == FLOAT:
256
return "%.1ff" % float(value)
258
return "%.1f" % float(value)
260
return str(int(value))
264
'''Get the value of unity for this type.'''
265
return value_to_native(type, 1)
268
def clamp_expr(src_channel, dst_channel, dst_native_type, value):
269
'''Generate the expression to clamp the value in the source type to the
270
destination type range.'''
272
if src_channel == dst_channel:
275
src_min = src_channel.min()
276
src_max = src_channel.max()
277
dst_min = dst_channel.min()
278
dst_max = dst_channel.max()
280
# Translate the destination range to the src native value
281
dst_min_native = native_to_constant(src_channel, value_to_native(src_channel, dst_min))
282
dst_max_native = native_to_constant(src_channel, value_to_native(src_channel, dst_max))
284
if src_min < dst_min and src_max > dst_max:
285
return 'CLAMP(%s, %s, %s)' % (value, dst_min_native, dst_max_native)
287
if src_max > dst_max:
288
return 'MIN2(%s, %s)' % (value, dst_max_native)
290
if src_min < dst_min:
291
return 'MAX2(%s, %s)' % (value, dst_min_native)
296
def conversion_expr(src_channel,
297
dst_channel, dst_native_type,
300
src_colorspace = RGB,
301
dst_colorspace = RGB):
302
'''Generate the expression to convert a value between two types.'''
304
if src_colorspace != dst_colorspace:
305
if src_colorspace == SRGB:
306
assert src_channel.type == UNSIGNED
307
assert src_channel.norm
308
assert src_channel.size <= 8
309
assert src_channel.size >= 4
310
assert dst_colorspace == RGB
311
if src_channel.size < 8:
312
value = '%s << %x | %s >> %x' % (value, 8 - src_channel.size, value, 2 * src_channel.size - 8)
313
if dst_channel.type == FLOAT:
314
return 'util_format_srgb_8unorm_to_linear_float(%s)' % value
316
assert dst_channel.type == UNSIGNED
317
assert dst_channel.norm
318
assert dst_channel.size == 8
319
return 'util_format_srgb_to_linear_8unorm(%s)' % value
320
elif dst_colorspace == SRGB:
321
assert dst_channel.type == UNSIGNED
322
assert dst_channel.norm
323
assert dst_channel.size <= 8
324
assert src_colorspace == RGB
325
if src_channel.type == FLOAT:
326
value = 'util_format_linear_float_to_srgb_8unorm(%s)' % value
328
assert src_channel.type == UNSIGNED
329
assert src_channel.norm
330
assert src_channel.size == 8
331
value = 'util_format_linear_to_srgb_8unorm(%s)' % value
332
# XXX rounding is all wrong.
333
if dst_channel.size < 8:
334
return '%s >> %x' % (value, 8 - dst_channel.size)
337
elif src_colorspace == ZS:
339
elif dst_colorspace == ZS:
344
if src_channel == dst_channel:
347
src_type = src_channel.type
348
src_size = src_channel.size
349
src_norm = src_channel.norm
350
src_pure = src_channel.pure
352
# Promote half to float
353
if src_type == FLOAT and src_size == 16:
354
value = '_mesa_half_to_float(%s)' % value
357
# Special case for float <-> ubytes for more accurate results
358
# Done before clamping since these functions already take care of that
359
if src_type == UNSIGNED and src_norm and src_size == 8 and dst_channel.type == FLOAT and dst_channel.size == 32:
360
return 'ubyte_to_float(%s)' % value
361
if src_type == FLOAT and src_size == 32 and dst_channel.type == UNSIGNED and dst_channel.norm and dst_channel.size == 8:
362
return 'float_to_ubyte(%s)' % value
365
if dst_channel.type != FLOAT or src_type != FLOAT:
366
value = clamp_expr(src_channel, dst_channel, dst_native_type, value)
368
if src_type in (SIGNED, UNSIGNED) and dst_channel.type in (SIGNED, UNSIGNED):
369
if not src_norm and not dst_channel.norm:
370
# neither is normalized -- just cast
371
return '(%s)%s' % (dst_native_type, value)
373
if src_norm and dst_channel.norm:
374
return "_mesa_%snorm_to_%snorm(%s, %d, %d)" % ("s" if src_type == SIGNED else "u",
375
"s" if dst_channel.type == SIGNED else "u",
376
value, src_channel.size, dst_channel.size)
378
# We need to rescale using an intermediate type big enough to hold the multiplication of both
379
src_one = get_one(src_channel)
380
dst_one = get_one(dst_channel)
381
tmp_native_type = intermediate_native_type(src_size + dst_channel.size, src_channel.sign and dst_channel.sign)
382
value = '((%s)%s)' % (tmp_native_type, value)
383
value = '(%s)(%s * 0x%x / 0x%x)' % (dst_native_type, value, dst_one, src_one)
387
# Promote to either float or double
388
if src_type != FLOAT:
389
if src_norm or src_type == FIXED:
390
one = get_one(src_channel)
392
value = '(%s * (1.0f/0x%x))' % (value, one)
393
if dst_channel.size <= 32:
394
value = '(float)%s' % value
397
# bigger than single precision mantissa, use double
398
value = '(%s * (1.0/0x%x))' % (value, one)
402
if src_size <= 23 or dst_channel.size <= 32:
403
value = '(float)%s' % value
406
# bigger than single precision mantissa, use double
407
value = '(double)%s' % value
411
# Convert double or float to non-float
412
if dst_channel.type != FLOAT:
413
if dst_channel.norm or dst_channel.type == FIXED:
414
dst_one = get_one(dst_channel)
415
if dst_channel.size <= 23:
416
value = 'util_iround(%s * 0x%x)' % (value, dst_one)
418
# bigger than single precision mantissa, use double
419
value = '(%s * (double)0x%x)' % (value, dst_one)
420
value = '(%s)%s' % (dst_native_type, value)
422
# Cast double to float when converting to either half or float
423
if dst_channel.size <= 32 and src_size > 32:
424
value = '(float)%s' % value
427
if dst_channel.size == 16:
428
value = '_mesa_float_to_float16_rtz(%s)' % value
429
elif dst_channel.size == 64 and src_size < 64:
430
value = '(double)%s' % value
435
def generate_unpack_kernel(format, dst_channel, dst_native_type):
437
if not is_format_supported(format):
440
assert format.layout == PLAIN
442
def unpack_from_bitmask(channels, swizzles):
443
depth = format.block_size()
444
print(' uint%u_t value = *(const uint%u_t *)src;' % (depth, depth))
446
# Compute the intermediate unshifted values
447
for i in range(format.nr_channels()):
448
src_channel = channels[i]
450
shift = src_channel.shift
451
if src_channel.type == UNSIGNED:
453
value = '%s >> %u' % (value, shift)
454
if shift + src_channel.size < depth:
455
value = '(%s) & 0x%x' % (value, (1 << src_channel.size) - 1)
456
print(' uint%u_t %s = %s;' % (depth, src_channel.name, value))
457
elif src_channel.type == SIGNED:
458
if shift + src_channel.size < depth:
460
lshift = depth - (shift + src_channel.size)
461
value = '%s << %u' % (value, lshift)
463
value = '(int%u_t)(%s) ' % (depth, value)
464
if src_channel.size < depth:
466
rshift = depth - src_channel.size
467
value = '(%s) >> %u' % (value, rshift)
468
print(' int%u_t %s = %s;' % (depth, src_channel.name, value))
472
# Convert, swizzle, and store final values
474
swizzle = swizzles[i]
476
src_channel = channels[swizzle]
477
src_colorspace = format.colorspace
478
if src_colorspace == SRGB and i == 3:
479
# Alpha channel is linear
481
value = src_channel.name
482
value = conversion_expr(src_channel,
483
dst_channel, dst_native_type,
485
src_colorspace = src_colorspace)
486
elif swizzle == SWIZZLE_0:
488
elif swizzle == SWIZZLE_1:
489
value = get_one(dst_channel)
490
elif swizzle == SWIZZLE_NONE:
494
print(' dst[%u] = %s; /* %s */' % (i, value, 'rgba'[i]))
496
def unpack_from_struct(channels, swizzles):
497
print(' struct util_format_%s pixel;' % format.short_name())
498
print(' memcpy(&pixel, src, sizeof pixel);')
501
swizzle = swizzles[i]
503
src_channel = channels[swizzle]
504
src_colorspace = format.colorspace
505
if src_colorspace == SRGB and i == 3:
506
# Alpha channel is linear
508
value = 'pixel.%s' % src_channel.name
509
value = conversion_expr(src_channel,
510
dst_channel, dst_native_type,
512
src_colorspace = src_colorspace)
513
elif swizzle == SWIZZLE_0:
515
elif swizzle == SWIZZLE_1:
516
value = get_one(dst_channel)
517
elif swizzle == SWIZZLE_NONE:
521
print(' dst[%u] = %s; /* %s */' % (i, value, 'rgba'[i]))
523
if format.is_bitmask():
524
print_channels(format, unpack_from_bitmask)
526
print_channels(format, unpack_from_struct)
529
def generate_pack_kernel(format, src_channel, src_native_type):
531
if not is_format_supported(format):
534
dst_native_type = native_type(format)
536
assert format.layout == PLAIN
538
def pack_into_bitmask(channels, swizzles):
539
inv_swizzle = inv_swizzles(swizzles)
541
depth = format.block_size()
542
print(' uint%u_t value = 0;' % depth)
545
dst_channel = channels[i]
546
shift = dst_channel.shift
547
if inv_swizzle[i] is not None:
548
value ='src[%u]' % inv_swizzle[i]
549
dst_colorspace = format.colorspace
550
if dst_colorspace == SRGB and inv_swizzle[i] == 3:
551
# Alpha channel is linear
553
value = conversion_expr(src_channel,
554
dst_channel, dst_native_type,
556
dst_colorspace = dst_colorspace)
557
if dst_channel.type in (UNSIGNED, SIGNED):
558
if shift + dst_channel.size < depth:
559
value = '(%s) & 0x%x' % (value, (1 << dst_channel.size) - 1)
561
value = '(uint32_t)(%s) << %u' % (value, shift)
562
if dst_channel.type == SIGNED:
564
value = '(uint%u_t)(%s) ' % (depth, value)
567
if value is not None:
568
print(' value |= %s;' % (value))
570
print(' *(uint%u_t *)dst = value;' % depth)
572
def pack_into_struct(channels, swizzles):
573
inv_swizzle = inv_swizzles(swizzles)
575
print(' struct util_format_%s pixel = {0};' % format.short_name())
578
dst_channel = channels[i]
579
width = dst_channel.size
580
if inv_swizzle[i] is None:
582
dst_colorspace = format.colorspace
583
if dst_colorspace == SRGB and inv_swizzle[i] == 3:
584
# Alpha channel is linear
586
value ='src[%u]' % inv_swizzle[i]
587
value = conversion_expr(src_channel,
588
dst_channel, dst_native_type,
590
dst_colorspace = dst_colorspace)
591
print(' pixel.%s = %s;' % (dst_channel.name, value))
593
print(' memcpy(dst, &pixel, sizeof pixel);')
595
if format.is_bitmask():
596
print_channels(format, pack_into_bitmask)
598
print_channels(format, pack_into_struct)
601
def generate_format_unpack(format, dst_channel, dst_native_type, dst_suffix):
602
'''Generate the function to unpack pixels from a particular format'''
604
name = format.short_name()
606
if "8unorm" in dst_suffix:
607
dst_proto_type = dst_native_type
609
dst_proto_type = 'void'
611
proto = 'util_format_%s_unpack_%s(%s *restrict dst_row, const uint8_t *restrict src, unsigned width)' % (
612
name, dst_suffix, dst_proto_type)
613
print('void %s;' % proto, file=sys.stdout2)
619
if is_format_supported(format):
620
print(' %s *dst = dst_row;' % (dst_native_type))
622
' for (unsigned x = 0; x < width; x += %u) {' % (format.block_width,))
624
generate_unpack_kernel(format, dst_channel, dst_native_type)
626
print(' src += %u;' % (format.block_size() / 8,))
634
def generate_format_pack(format, src_channel, src_native_type, src_suffix):
635
'''Generate the function to pack pixels to a particular format'''
637
name = format.short_name()
640
print('util_format_%s_pack_%s(uint8_t *restrict dst_row, unsigned dst_stride, const %s *restrict src_row, unsigned src_stride, unsigned width, unsigned height)' %
641
(name, src_suffix, src_native_type))
644
print('void util_format_%s_pack_%s(uint8_t *restrict dst_row, unsigned dst_stride, const %s *restrict src_row, unsigned src_stride, unsigned width, unsigned height);' %
645
(name, src_suffix, src_native_type), file=sys.stdout2)
647
if is_format_supported(format):
648
print(' unsigned x, y;')
649
print(' for(y = 0; y < height; y += %u) {' % (format.block_height,))
650
print(' const %s *src = src_row;' % (src_native_type))
651
print(' uint8_t *dst = dst_row;')
652
print(' for(x = 0; x < width; x += %u) {' % (format.block_width,))
654
generate_pack_kernel(format, src_channel, src_native_type)
657
print(' dst += %u;' % (format.block_size() / 8,))
659
print(' dst_row += dst_stride;')
660
print(' src_row += src_stride/sizeof(*src_row);')
667
def generate_format_fetch(format, dst_channel, dst_native_type):
668
'''Generate the function to unpack pixels from a particular format'''
670
name = format.short_name()
672
proto = 'util_format_%s_fetch_rgba(void *restrict in_dst, const uint8_t *restrict src, UNUSED unsigned i, UNUSED unsigned j)' % (name)
673
print('void %s;' % proto, file=sys.stdout2)
679
print(' %s *dst = in_dst;' % dst_native_type)
681
if is_format_supported(format):
682
generate_unpack_kernel(format, dst_channel, dst_native_type)
688
def is_format_hand_written(format):
689
return format.layout != PLAIN or format.colorspace == ZS
692
def generate(formats):
694
print('#include "pipe/p_compiler.h"')
695
print('#include "util/u_math.h"')
696
print('#include "util/half_float.h"')
697
print('#include "u_format.h"')
698
print('#include "u_format_other.h"')
699
print('#include "util/format_srgb.h"')
700
print('#include "format_utils.h"')
701
print('#include "u_format_yuv.h"')
702
print('#include "u_format_zs.h"')
703
print('#include "u_format_pack.h"')
706
for format in formats:
707
if not is_format_hand_written(format):
709
if is_format_supported(format) and not format.is_bitmask():
710
generate_format_type(format)
712
if format.is_pure_unsigned():
713
native_type = 'unsigned'
715
channel = Channel(UNSIGNED, False, True, 32)
717
generate_format_unpack(format, channel, native_type, suffix)
718
generate_format_pack(format, channel, native_type, suffix)
719
generate_format_fetch(format, channel, native_type)
721
channel = Channel(SIGNED, False, True, 32)
724
generate_format_pack(format, channel, native_type, suffix)
725
elif format.is_pure_signed():
728
channel = Channel(SIGNED, False, True, 32)
730
generate_format_unpack(format, channel, native_type, suffix)
731
generate_format_pack(format, channel, native_type, suffix)
732
generate_format_fetch(format, channel, native_type)
734
native_type = 'unsigned'
736
channel = Channel(UNSIGNED, False, True, 32)
737
generate_format_pack(format, channel, native_type, suffix)
739
channel = Channel(FLOAT, False, False, 32)
740
native_type = 'float'
741
suffix = 'rgba_float'
743
generate_format_unpack(format, channel, native_type, suffix)
744
generate_format_pack(format, channel, native_type, suffix)
745
generate_format_fetch(format, channel, native_type)
747
channel = Channel(UNSIGNED, True, False, 8)
748
native_type = 'uint8_t'
749
suffix = 'rgba_8unorm'
751
generate_format_unpack(format, channel, native_type, suffix)
752
generate_format_pack(format, channel, native_type, suffix)