aboutsummaryrefslogtreecommitdiff
path: root/mesalib/src/glsl/nir/nir_constant_expressions.py
blob: bf82fe533d65bb55e34f0a1f1a6f8286d94c4696 (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
#! /usr/bin/python2
template = """\
/*
 * Copyright (C) 2014 Intel Corporation
 *
 * 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, sublicense,
 * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS 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.
 *
 * Authors:
 *    Jason Ekstrand (jason@jlekstrand.net)
 */

#include <math.h>
#include "main/core.h"
#include "util/rounding.h" /* for _mesa_roundeven */
#include "nir_constant_expressions.h"

#if defined(_MSC_VER) && (_MSC_VER < 1800)
static int isnormal(double x)
{
   return _fpclass(x) == _FPCLASS_NN || _fpclass(x) == _FPCLASS_PN;
}
#elif defined(__SUNPRO_CC)
#include <ieeefp.h>
static int isnormal(double x)
{
   return fpclass(x) == FP_NORMAL;
}
#endif

#if defined(_MSC_VER)
static double copysign(double x, double y)
{
   return _copysign(x, y);
}
#endif

/**
 * Evaluate one component of packSnorm4x8.
 */
static uint8_t
pack_snorm_1x8(float x)
{
    /* From section 8.4 of the GLSL 4.30 spec:
     *
     *    packSnorm4x8
     *    ------------
     *    The conversion for component c of v to fixed point is done as
     *    follows:
     *
     *      packSnorm4x8: round(clamp(c, -1, +1) * 127.0)
     *
     * We must first cast the float to an int, because casting a negative
     * float to a uint is undefined.
     */
   return (uint8_t) (int)
          _mesa_roundevenf(CLAMP(x, -1.0f, +1.0f) * 127.0f);
}

/**
 * Evaluate one component of packSnorm2x16.
 */
static uint16_t
pack_snorm_1x16(float x)
{
    /* From section 8.4 of the GLSL ES 3.00 spec:
     *
     *    packSnorm2x16
     *    -------------
     *    The conversion for component c of v to fixed point is done as
     *    follows:
     *
     *      packSnorm2x16: round(clamp(c, -1, +1) * 32767.0)
     *
     * We must first cast the float to an int, because casting a negative
     * float to a uint is undefined.
     */
   return (uint16_t) (int)
          _mesa_roundevenf(CLAMP(x, -1.0f, +1.0f) * 32767.0f);
}

/**
 * Evaluate one component of unpackSnorm4x8.
 */
static float
unpack_snorm_1x8(uint8_t u)
{
    /* From section 8.4 of the GLSL 4.30 spec:
     *
     *    unpackSnorm4x8
     *    --------------
     *    The conversion for unpacked fixed-point value f to floating point is
     *    done as follows:
     *
     *       unpackSnorm4x8: clamp(f / 127.0, -1, +1)
     */
   return CLAMP((int8_t) u / 127.0f, -1.0f, +1.0f);
}

/**
 * Evaluate one component of unpackSnorm2x16.
 */
static float
unpack_snorm_1x16(uint16_t u)
{
    /* From section 8.4 of the GLSL ES 3.00 spec:
     *
     *    unpackSnorm2x16
     *    ---------------
     *    The conversion for unpacked fixed-point value f to floating point is
     *    done as follows:
     *
     *       unpackSnorm2x16: clamp(f / 32767.0, -1, +1)
     */
   return CLAMP((int16_t) u / 32767.0f, -1.0f, +1.0f);
}

/**
 * Evaluate one component packUnorm4x8.
 */
static uint8_t
pack_unorm_1x8(float x)
{
    /* From section 8.4 of the GLSL 4.30 spec:
     *
     *    packUnorm4x8
     *    ------------
     *    The conversion for component c of v to fixed point is done as
     *    follows:
     *
     *       packUnorm4x8: round(clamp(c, 0, +1) * 255.0)
     */
   return (uint8_t) (int)
          _mesa_roundevenf(CLAMP(x, 0.0f, 1.0f) * 255.0f);
}

/**
 * Evaluate one component packUnorm2x16.
 */
static uint16_t
pack_unorm_1x16(float x)
{
    /* From section 8.4 of the GLSL ES 3.00 spec:
     *
     *    packUnorm2x16
     *    -------------
     *    The conversion for component c of v to fixed point is done as
     *    follows:
     *
     *       packUnorm2x16: round(clamp(c, 0, +1) * 65535.0)
     */
   return (uint16_t) (int)
          _mesa_roundevenf(CLAMP(x, 0.0f, 1.0f) * 65535.0f);
}

/**
 * Evaluate one component of unpackUnorm4x8.
 */
static float
unpack_unorm_1x8(uint8_t u)
{
    /* From section 8.4 of the GLSL 4.30 spec:
     *
     *    unpackUnorm4x8
     *    --------------
     *    The conversion for unpacked fixed-point value f to floating point is
     *    done as follows:
     *
     *       unpackUnorm4x8: f / 255.0
     */
   return (float) u / 255.0f;
}

/**
 * Evaluate one component of unpackUnorm2x16.
 */
static float
unpack_unorm_1x16(uint16_t u)
{
    /* From section 8.4 of the GLSL ES 3.00 spec:
     *
     *    unpackUnorm2x16
     *    ---------------
     *    The conversion for unpacked fixed-point value f to floating point is
     *    done as follows:
     *
     *       unpackUnorm2x16: f / 65535.0
     */
   return (float) u / 65535.0f;
}

/**
 * Evaluate one component of packHalf2x16.
 */
static uint16_t
pack_half_1x16(float x)
{
   return _mesa_float_to_half(x);
}

/**
 * Evaluate one component of unpackHalf2x16.
 */
static float
unpack_half_1x16(uint16_t u)
{
   return _mesa_half_to_float(u);
}

/* Some typed vector structures to make things like src0.y work */
% for type in ["float", "int", "unsigned", "bool"]:
struct ${type}_vec {
   ${type} x;
   ${type} y;
   ${type} z;
   ${type} w;
};
% endfor

% for name, op in sorted(opcodes.iteritems()):
static nir_const_value
evaluate_${name}(unsigned num_components, nir_const_value *_src)
{
   nir_const_value _dst_val = { { {0, 0, 0, 0} } };

   ## For each non-per-component input, create a variable srcN that
   ## contains x, y, z, and w elements which are filled in with the
   ## appropriately-typed values.
   % for j in range(op.num_inputs):
      % if op.input_sizes[j] == 0:
         <% continue %>
      % elif "src" + str(j) not in op.const_expr:
         ## Avoid unused variable warnings
         <% continue %>
      %endif

      struct ${op.input_types[j]}_vec src${j} = {
      % for k in range(op.input_sizes[j]):
         % if op.input_types[j] == "bool":
            _src[${j}].u[${k}] != 0,
         % else:
            _src[${j}].${op.input_types[j][:1]}[${k}],
         % endif
      % endfor
      };
   % endfor

   % if op.output_size == 0:
      ## For per-component instructions, we need to iterate over the
      ## components and apply the constant expression one component
      ## at a time.
      for (unsigned _i = 0; _i < num_components; _i++) {
         ## For each per-component input, create a variable srcN that
         ## contains the value of the current (_i'th) component.
         % for j in range(op.num_inputs):
            % if op.input_sizes[j] != 0:
               <% continue %>
            % elif "src" + str(j) not in op.const_expr:
               ## Avoid unused variable warnings
               <% continue %>
            % elif op.input_types[j] == "bool":
               bool src${j} = _src[${j}].u[_i] != 0;
            % else:
               ${op.input_types[j]} src${j} = _src[${j}].${op.input_types[j][:1]}[_i];
            % endif
         % endfor

         ## Create an appropriately-typed variable dst and assign the
         ## result of the const_expr to it.  If const_expr already contains
         ## writes to dst, just include const_expr directly.
         % if "dst" in op.const_expr:
            ${op.output_type} dst;
            ${op.const_expr}
         % else:
            ${op.output_type} dst = ${op.const_expr};
         % endif

         ## Store the current component of the actual destination to the
         ## value of dst.
         % if op.output_type == "bool":
            ## Sanitize the C value to a proper NIR bool
            _dst_val.u[_i] = dst ? NIR_TRUE : NIR_FALSE;
         % else:
            _dst_val.${op.output_type[:1]}[_i] = dst;
         % endif
      }
   % else:
      ## In the non-per-component case, create a struct dst with
      ## appropriately-typed elements x, y, z, and w and assign the result
      ## of the const_expr to all components of dst, or include the
      ## const_expr directly if it writes to dst already.
      struct ${op.output_type}_vec dst;

      % if "dst" in op.const_expr:
         ${op.const_expr}
      % else:
         ## Splat the value to all components.  This way expressions which
         ## write the same value to all components don't need to explicitly
         ## write to dest.  One such example is fnoise which has a
         ## const_expr of 0.0f.
         dst.x = dst.y = dst.z = dst.w = ${op.const_expr};
      % endif

      ## For each component in the destination, copy the value of dst to
      ## the actual destination.
      % for k in range(op.output_size):
         % if op.output_type == "bool":
            ## Sanitize the C value to a proper NIR bool
            _dst_val.u[${k}] = dst.${"xyzw"[k]} ? NIR_TRUE : NIR_FALSE;
         % else:
            _dst_val.${op.output_type[:1]}[${k}] = dst.${"xyzw"[k]};
         % endif
      % endfor
   % endif

   return _dst_val;
}
% endfor

nir_const_value
nir_eval_const_opcode(nir_op op, unsigned num_components,
                      nir_const_value *src)
{
   switch (op) {
% for name in sorted(opcodes.iterkeys()):
   case nir_op_${name}: {
      return evaluate_${name}(num_components, src);
      break;
   }
% endfor
   default:
      unreachable("shouldn't get here");
   }
}"""

from nir_opcodes import opcodes
from mako.template import Template

print Template(template).render(opcodes=opcodes)