projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
gallivm: optimize soa linear clamp to edge wrap mode a bit
[mesa.git] / src / gallium / auxiliary / gallivm / lp_bld_sample_soa.c
1 /**************************************************************************
2 *
3 * Copyright 2009 VMware, Inc.
4 * All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
13 *
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
16 * of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25 *
26 **************************************************************************/
27
28 /**
29 * @file
30 * Texture sampling -- SoA.
31 *
32 * @author Jose Fonseca <jfonseca@vmware.com>
33 * @author Brian Paul <brianp@vmware.com>
34 */
35
36 #include "pipe/p_defines.h"
37 #include "pipe/p_state.h"
38 #include "util/u_debug.h"
39 #include "util/u_dump.h"
40 #include "util/u_memory.h"
41 #include "util/u_math.h"
42 #include "util/u_format.h"
43 #include "lp_bld_debug.h"
44 #include "lp_bld_type.h"
45 #include "lp_bld_const.h"
46 #include "lp_bld_conv.h"
47 #include "lp_bld_arit.h"
48 #include "lp_bld_bitarit.h"
49 #include "lp_bld_logic.h"
50 #include "lp_bld_printf.h"
51 #include "lp_bld_swizzle.h"
52 #include "lp_bld_flow.h"
53 #include "lp_bld_gather.h"
54 #include "lp_bld_format.h"
55 #include "lp_bld_sample.h"
56 #include "lp_bld_sample_aos.h"
57 #include "lp_bld_struct.h"
58 #include "lp_bld_quad.h"
59
60
61 /**
62 * Generate code to fetch a texel from a texture at int coords (x, y, z).
63 * The computation depends on whether the texture is 1D, 2D or 3D.
64 * The result, texel, will be float vectors:
65 * texel[0] = red values
66 * texel[1] = green values
67 * texel[2] = blue values
68 * texel[3] = alpha values
69 */
70 static void
71 lp_build_sample_texel_soa(struct lp_build_sample_context *bld,
72 unsigned unit,
73 LLVMValueRef width,
74 LLVMValueRef height,
75 LLVMValueRef depth,
76 LLVMValueRef x,
77 LLVMValueRef y,
78 LLVMValueRef z,
79 LLVMValueRef y_stride,
80 LLVMValueRef z_stride,
81 LLVMValueRef data_ptr,
82 LLVMValueRef texel_out[4])
83 {
84 const struct lp_sampler_static_state *static_state = bld->static_state;
85 const unsigned dims = bld->dims;
86 struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
87 LLVMValueRef offset;
88 LLVMValueRef i, j;
89 LLVMValueRef use_border = NULL;
90
91 /* use_border = x < 0 || x >= width || y < 0 || y >= height */
92 if (lp_sampler_wrap_mode_uses_border_color(static_state->wrap_s,
93 static_state->min_img_filter,
94 static_state->mag_img_filter)) {
95 LLVMValueRef b1, b2;
96 b1 = lp_build_cmp(int_coord_bld, PIPE_FUNC_LESS, x, int_coord_bld->zero);
97 b2 = lp_build_cmp(int_coord_bld, PIPE_FUNC_GEQUAL, x, width);
98 use_border = LLVMBuildOr(bld->builder, b1, b2, "b1_or_b2");
99 }
100
101 if (dims >= 2 &&
102 lp_sampler_wrap_mode_uses_border_color(static_state->wrap_t,
103 static_state->min_img_filter,
104 static_state->mag_img_filter)) {
105 LLVMValueRef b1, b2;
106 b1 = lp_build_cmp(int_coord_bld, PIPE_FUNC_LESS, y, int_coord_bld->zero);
107 b2 = lp_build_cmp(int_coord_bld, PIPE_FUNC_GEQUAL, y, height);
108 if (use_border) {
109 use_border = LLVMBuildOr(bld->builder, use_border, b1, "ub_or_b1");
110 use_border = LLVMBuildOr(bld->builder, use_border, b2, "ub_or_b2");
111 }
112 else {
113 use_border = LLVMBuildOr(bld->builder, b1, b2, "b1_or_b2");
114 }
115 }
116
117 if (dims == 3 &&
118 lp_sampler_wrap_mode_uses_border_color(static_state->wrap_r,
119 static_state->min_img_filter,
120 static_state->mag_img_filter)) {
121 LLVMValueRef b1, b2;
122 b1 = lp_build_cmp(int_coord_bld, PIPE_FUNC_LESS, z, int_coord_bld->zero);
123 b2 = lp_build_cmp(int_coord_bld, PIPE_FUNC_GEQUAL, z, depth);
124 if (use_border) {
125 use_border = LLVMBuildOr(bld->builder, use_border, b1, "ub_or_b1");
126 use_border = LLVMBuildOr(bld->builder, use_border, b2, "ub_or_b2");
127 }
128 else {
129 use_border = LLVMBuildOr(bld->builder, b1, b2, "b1_or_b2");
130 }
131 }
132
133 /* convert x,y,z coords to linear offset from start of texture, in bytes */
134 lp_build_sample_offset(&bld->uint_coord_bld,
135 bld->format_desc,
136 x, y, z, y_stride, z_stride,
137 &offset, &i, &j);
138
139 if (use_border) {
140 /* If we can sample the border color, it means that texcoords may
141 * lie outside the bounds of the texture image. We need to do
142 * something to prevent reading out of bounds and causing a segfault.
143 *
144 * Simply AND the texture coords with !use_border. This will cause
145 * coords which are out of bounds to become zero. Zero's guaranteed
146 * to be inside the texture image.
147 */
148 offset = lp_build_andnot(&bld->uint_coord_bld, offset, use_border);
149 }
150
151 lp_build_fetch_rgba_soa(bld->builder,
152 bld->format_desc,
153 bld->texel_type,
154 data_ptr, offset,
155 i, j,
156 texel_out);
157
158 /*
159 * Note: if we find an app which frequently samples the texture border
160 * we might want to implement a true conditional here to avoid sampling
161 * the texture whenever possible (since that's quite a bit of code).
162 * Ex:
163 * if (use_border) {
164 * texel = border_color;
165 * }
166 * else {
167 * texel = sample_texture(coord);
168 * }
169 * As it is now, we always sample the texture, then selectively replace
170 * the texel color results with the border color.
171 */
172
173 if (use_border) {
174 /* select texel color or border color depending on use_border */
175 LLVMValueRef border_color_ptr =
176 bld->dynamic_state->border_color(bld->dynamic_state,
177 bld->builder, unit);
178 int chan;
179 for (chan = 0; chan < 4; chan++) {
180 LLVMValueRef border_chan =
181 lp_build_array_get(bld->builder, border_color_ptr,
182 lp_build_const_int32(chan));
183 LLVMValueRef border_chan_vec =
184 lp_build_broadcast_scalar(&bld->float_vec_bld, border_chan);
185 texel_out[chan] = lp_build_select(&bld->texel_bld, use_border,
186 border_chan_vec, texel_out[chan]);
187 }
188 }
189
190 apply_sampler_swizzle(bld, texel_out);
191 }
192
193
194 /**
195 * Helper to compute the mirror function for the PIPE_WRAP_MIRROR modes.
196 */
197 static LLVMValueRef
198 lp_build_coord_mirror(struct lp_build_sample_context *bld,
199 LLVMValueRef coord)
200 {
201 struct lp_build_context *coord_bld = &bld->coord_bld;
202 struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
203 LLVMValueRef fract, flr, isOdd;
204
205 lp_build_ifloor_fract(coord_bld, coord, &flr, &fract);
206
207 /* isOdd = flr & 1 */
208 isOdd = LLVMBuildAnd(bld->builder, flr, int_coord_bld->one, "");
209
210 /* make coord positive or negative depending on isOdd */
211 coord = lp_build_set_sign(coord_bld, fract, isOdd);
212
213 /* convert isOdd to float */
214 isOdd = lp_build_int_to_float(coord_bld, isOdd);
215
216 /* add isOdd to coord */
217 coord = lp_build_add(coord_bld, coord, isOdd);
218
219 return coord;
220 }
221
222
223 /**
224 * Build LLVM code for texture wrap mode for linear filtering.
225 * \param x0_out returns first integer texcoord
226 * \param x1_out returns second integer texcoord
227 * \param weight_out returns linear interpolation weight
228 */
229 static void
230 lp_build_sample_wrap_linear(struct lp_build_sample_context *bld,
231 LLVMValueRef coord,
232 LLVMValueRef length,
233 boolean is_pot,
234 unsigned wrap_mode,
235 LLVMValueRef *x0_out,
236 LLVMValueRef *x1_out,
237 LLVMValueRef *weight_out)
238 {
239 struct lp_build_context *coord_bld = &bld->coord_bld;
240 struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
241 struct lp_build_context *uint_coord_bld = &bld->uint_coord_bld;
242 LLVMValueRef half = lp_build_const_vec(coord_bld->type, 0.5);
243 LLVMValueRef length_f = lp_build_int_to_float(coord_bld, length);
244 LLVMValueRef length_minus_one = lp_build_sub(uint_coord_bld, length, uint_coord_bld->one);
245 LLVMValueRef coord0, coord1, weight;
246
247 switch(wrap_mode) {
248 case PIPE_TEX_WRAP_REPEAT:
249 /* mul by size and subtract 0.5 */
250 coord = lp_build_mul(coord_bld, coord, length_f);
251 coord = lp_build_sub(coord_bld, coord, half);
252 /* convert to int, compute lerp weight */
253 lp_build_ifloor_fract(coord_bld, coord, &coord0, &weight);
254 /* repeat wrap */
255 if (is_pot) {
256 coord1 = lp_build_add(uint_coord_bld, coord0, uint_coord_bld->one);
257 coord0 = LLVMBuildAnd(bld->builder, coord0, length_minus_one, "");
258 coord1 = LLVMBuildAnd(bld->builder, coord1, length_minus_one, "");
259 }
260 else {
261 /* Add a bias to the texcoord to handle negative coords */
262 LLVMValueRef bias = lp_build_mul_imm(uint_coord_bld, length, 1024);
263 LLVMValueRef mask;
264 coord0 = LLVMBuildAdd(bld->builder, coord0, bias, "");
265 coord0 = LLVMBuildURem(bld->builder, coord0, length, "");
266 mask = lp_build_compare(bld->builder, int_coord_bld->type,
267 PIPE_FUNC_NOTEQUAL, coord0, length_minus_one);
268 coord1 = LLVMBuildAnd(bld->builder,
269 lp_build_add(uint_coord_bld, coord0, uint_coord_bld->one),
270 mask, "");
271 }
272 break;
273
274 case PIPE_TEX_WRAP_CLAMP:
275 if (bld->static_state->normalized_coords) {
276 /* scale coord to length */
277 coord = lp_build_mul(coord_bld, coord, length_f);
278 }
279
280 /* clamp to [0, length] */
281 coord = lp_build_clamp(coord_bld, coord, coord_bld->zero, length_f);
282
283 coord = lp_build_sub(coord_bld, coord, half);
284
285 /* convert to int, compute lerp weight */
286 lp_build_ifloor_fract(coord_bld, coord, &coord0, &weight);
287 coord1 = lp_build_add(int_coord_bld, coord0, int_coord_bld->one);
288 break;
289
290 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
291 {
292 struct lp_build_context abs_coord_bld = bld->coord_bld;
293 abs_coord_bld.type.sign = FALSE;
294
295 if (bld->static_state->normalized_coords) {
296 /* mul by tex size */
297 coord = lp_build_mul(coord_bld, coord, length_f);
298 /* clamp to length max */
299 coord = lp_build_min(coord_bld, coord, length_f);
300 /* subtract 0.5 */
301 coord = lp_build_sub(coord_bld, coord, half);
302 /* clamp to [0, length - 0.5] */
303 coord = lp_build_max(coord_bld, coord, coord_bld->zero);
304 }
305 /* XXX this is odd normalized ranges from 0 to length-0.5 after denorm
306 but non-normalized ranges from to 0.5 to length-0.5 after clamp.
307 Is this missing the sub 0.5? */
308 else {
309 LLVMValueRef min, max;
310 /* clamp to [0.5, length - 0.5] */
311 min = half;
312 max = lp_build_sub(coord_bld, length_f, min);
313 coord = lp_build_clamp(coord_bld, coord, min, max);
314 }
315 /* convert to int, compute lerp weight */
316 lp_build_ifloor_fract(&abs_coord_bld, coord, &coord0, &weight);
317 coord1 = lp_build_add(int_coord_bld, coord0, int_coord_bld->one);
318 /* coord1 = min(coord1, length-1) */
319 coord1 = lp_build_min(int_coord_bld, coord1, length_minus_one);
320 break;
321 }
322
323 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
324 {
325 LLVMValueRef min;
326 if (bld->static_state->normalized_coords) {
327 /* scale coord to length */
328 coord = lp_build_mul(coord_bld, coord, length_f);
329 }
330 /* was: clamp to [-0.5, length + 0.5], then sub 0.5 */
331 coord = lp_build_sub(coord_bld, coord, half);
332 min = lp_build_const_vec(coord_bld->type, -1.0F);
333 coord = lp_build_clamp(coord_bld, coord, min, length_f);
334 /* convert to int, compute lerp weight */
335 lp_build_ifloor_fract(coord_bld, coord, &coord0, &weight);
336 coord1 = lp_build_add(int_coord_bld, coord0, int_coord_bld->one);
337 }
338 break;
339
340 case PIPE_TEX_WRAP_MIRROR_REPEAT:
341 /* compute mirror function */
342 coord = lp_build_coord_mirror(bld, coord);
343
344 /* scale coord to length */
345 coord = lp_build_mul(coord_bld, coord, length_f);
346 coord = lp_build_sub(coord_bld, coord, half);
347
348 /* convert to int, compute lerp weight */
349 lp_build_ifloor_fract(coord_bld, coord, &coord0, &weight);
350 coord1 = lp_build_add(int_coord_bld, coord0, int_coord_bld->one);
351
352 /* coord0 = max(coord0, 0) */
353 coord0 = lp_build_max(int_coord_bld, coord0, int_coord_bld->zero);
354 /* coord1 = min(coord1, length-1) */
355 coord1 = lp_build_min(int_coord_bld, coord1, length_minus_one);
356 break;
357
358 case PIPE_TEX_WRAP_MIRROR_CLAMP:
359 coord = lp_build_abs(coord_bld, coord);
360
361 if (bld->static_state->normalized_coords) {
362 /* scale coord to length */
363 coord = lp_build_mul(coord_bld, coord, length_f);
364 }
365
366 /* clamp to [0, length] */
367 coord = lp_build_min(coord_bld, coord, length_f);
368
369 coord = lp_build_sub(coord_bld, coord, half);
370
371 /* convert to int, compute lerp weight */
372 lp_build_ifloor_fract(coord_bld, coord, &coord0, &weight);
373 coord1 = lp_build_add(int_coord_bld, coord0, int_coord_bld->one);
374 break;
375
376 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
377 {
378 LLVMValueRef min, max;
379 struct lp_build_context abs_coord_bld = bld->coord_bld;
380 abs_coord_bld.type.sign = FALSE;
381 coord = lp_build_abs(coord_bld, coord);
382
383 if (bld->static_state->normalized_coords) {
384 /* scale coord to length */
385 coord = lp_build_mul(coord_bld, coord, length_f);
386 }
387
388 /* clamp to [0.5, length - 0.5] */
389 min = half;
390 max = lp_build_sub(coord_bld, length_f, min);
391 coord = lp_build_clamp(coord_bld, coord, min, max);
392
393 coord = lp_build_sub(coord_bld, coord, half);
394
395 /* convert to int, compute lerp weight */
396 lp_build_ifloor_fract(&abs_coord_bld, coord, &coord0, &weight);
397 coord1 = lp_build_add(int_coord_bld, coord0, int_coord_bld->one);
398 }
399 break;
400
401 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
402 {
403 coord = lp_build_abs(coord_bld, coord);
404
405 if (bld->static_state->normalized_coords) {
406 /* scale coord to length */
407 coord = lp_build_mul(coord_bld, coord, length_f);
408 }
409
410 /* was: clamp to [-0.5, length + 0.5] then sub 0.5 */
411 /* skip -0.5 clamp (always positive), do sub first */
412 coord = lp_build_sub(coord_bld, coord, half);
413 coord = lp_build_min(coord_bld, coord, length_f);
414
415 /* convert to int, compute lerp weight */
416 lp_build_ifloor_fract(coord_bld, coord, &coord0, &weight);
417 coord1 = lp_build_add(int_coord_bld, coord0, int_coord_bld->one);
418 }
419 break;
420
421 default:
422 assert(0);
423 coord0 = NULL;
424 coord1 = NULL;
425 weight = NULL;
426 }
427
428 *x0_out = coord0;
429 *x1_out = coord1;
430 *weight_out = weight;
431 }
432
433
434 /**
435 * Build LLVM code for texture wrap mode for nearest filtering.
436 * \param coord the incoming texcoord (nominally in [0,1])
437 * \param length the texture size along one dimension, as int vector
438 * \param is_pot if TRUE, length is a power of two
439 * \param wrap_mode one of PIPE_TEX_WRAP_x
440 */
441 static LLVMValueRef
442 lp_build_sample_wrap_nearest(struct lp_build_sample_context *bld,
443 LLVMValueRef coord,
444 LLVMValueRef length,
445 boolean is_pot,
446 unsigned wrap_mode)
447 {
448 struct lp_build_context *coord_bld = &bld->coord_bld;
449 struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
450 struct lp_build_context *uint_coord_bld = &bld->uint_coord_bld;
451 LLVMValueRef length_f = lp_build_int_to_float(coord_bld, length);
452 LLVMValueRef length_minus_one = lp_build_sub(uint_coord_bld, length, uint_coord_bld->one);
453 LLVMValueRef icoord;
454
455 switch(wrap_mode) {
456 case PIPE_TEX_WRAP_REPEAT:
457 coord = lp_build_mul(coord_bld, coord, length_f);
458 icoord = lp_build_ifloor(coord_bld, coord);
459 if (is_pot)
460 icoord = LLVMBuildAnd(bld->builder, icoord, length_minus_one, "");
461 else {
462 /* Add a bias to the texcoord to handle negative coords */
463 LLVMValueRef bias = lp_build_mul_imm(uint_coord_bld, length, 1024);
464 icoord = LLVMBuildAdd(bld->builder, icoord, bias, "");
465 icoord = LLVMBuildURem(bld->builder, icoord, length, "");
466 }
467 break;
468
469 case PIPE_TEX_WRAP_CLAMP:
470 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
471 if (bld->static_state->normalized_coords) {
472 /* scale coord to length */
473 coord = lp_build_mul(coord_bld, coord, length_f);
474 }
475
476 /* floor */
477 /* use itrunc instead since we clamp to 0 anyway */
478 icoord = lp_build_itrunc(coord_bld, coord);
479
480 /* clamp to [0, length - 1]. */
481 icoord = lp_build_clamp(int_coord_bld, icoord, int_coord_bld->zero,
482 length_minus_one);
483 break;
484
485 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
486 /* Note: this is the same as CLAMP_TO_EDGE, except min = -min */
487 {
488 LLVMValueRef min, max;
489
490 if (bld->static_state->normalized_coords) {
491 /* scale coord to length */
492 coord = lp_build_mul(coord_bld, coord, length_f);
493 }
494
495 icoord = lp_build_ifloor(coord_bld, coord);
496
497 /* clamp to [-1, length] */
498 min = lp_build_negate(int_coord_bld, int_coord_bld->one);
499 max = length;
500 icoord = lp_build_clamp(int_coord_bld, icoord, min, max);
501 }
502 break;
503
504 case PIPE_TEX_WRAP_MIRROR_REPEAT:
505 /* compute mirror function */
506 coord = lp_build_coord_mirror(bld, coord);
507
508 /* scale coord to length */
509 assert(bld->static_state->normalized_coords);
510 coord = lp_build_mul(coord_bld, coord, length_f);
511
512 /* itrunc == ifloor here */
513 icoord = lp_build_itrunc(coord_bld, coord);
514
515 /* clamp to [0, length - 1] */
516 icoord = lp_build_min(int_coord_bld, icoord, length_minus_one);
517 break;
518
519 case PIPE_TEX_WRAP_MIRROR_CLAMP:
520 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
521 coord = lp_build_abs(coord_bld, coord);
522
523 if (bld->static_state->normalized_coords) {
524 /* scale coord to length */
525 coord = lp_build_mul(coord_bld, coord, length_f);
526 }
527
528 /* itrunc == ifloor here */
529 icoord = lp_build_itrunc(coord_bld, coord);
530
531 /* clamp to [0, length - 1] */
532 icoord = lp_build_min(int_coord_bld, icoord, length_minus_one);
533 break;
534
535 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
536 coord = lp_build_abs(coord_bld, coord);
537
538 if (bld->static_state->normalized_coords) {
539 /* scale coord to length */
540 coord = lp_build_mul(coord_bld, coord, length_f);
541 }
542
543 /* itrunc == ifloor here */
544 icoord = lp_build_itrunc(coord_bld, coord);
545
546 /* clamp to [0, length] */
547 icoord = lp_build_min(int_coord_bld, icoord, length);
548 break;
549
550 default:
551 assert(0);
552 icoord = NULL;
553 }
554
555 return icoord;
556 }
557
558
559 /**
560 * Generate code to sample a mipmap level with nearest filtering.
561 * If sampling a cube texture, r = cube face in [0,5].
562 */
563 static void
564 lp_build_sample_image_nearest(struct lp_build_sample_context *bld,
565 unsigned unit,
566 LLVMValueRef width_vec,
567 LLVMValueRef height_vec,
568 LLVMValueRef depth_vec,
569 LLVMValueRef row_stride_vec,
570 LLVMValueRef img_stride_vec,
571 LLVMValueRef data_ptr,
572 LLVMValueRef s,
573 LLVMValueRef t,
574 LLVMValueRef r,
575 LLVMValueRef colors_out[4])
576 {
577 const unsigned dims = bld->dims;
578 LLVMValueRef x, y, z;
579
580 /*
581 * Compute integer texcoords.
582 */
583 x = lp_build_sample_wrap_nearest(bld, s, width_vec,
584 bld->static_state->pot_width,
585 bld->static_state->wrap_s);
586 lp_build_name(x, "tex.x.wrapped");
587
588 if (dims >= 2) {
589 y = lp_build_sample_wrap_nearest(bld, t, height_vec,
590 bld->static_state->pot_height,
591 bld->static_state->wrap_t);
592 lp_build_name(y, "tex.y.wrapped");
593
594 if (dims == 3) {
595 z = lp_build_sample_wrap_nearest(bld, r, depth_vec,
596 bld->static_state->pot_depth,
597 bld->static_state->wrap_r);
598 lp_build_name(z, "tex.z.wrapped");
599 }
600 else if (bld->static_state->target == PIPE_TEXTURE_CUBE) {
601 z = r;
602 }
603 else {
604 z = NULL;
605 }
606 }
607 else {
608 y = z = NULL;
609 }
610
611 /*
612 * Get texture colors.
613 */
614 lp_build_sample_texel_soa(bld, unit,
615 width_vec, height_vec, depth_vec,
616 x, y, z,
617 row_stride_vec, img_stride_vec,
618 data_ptr, colors_out);
619 }
620
621
622 /**
623 * Generate code to sample a mipmap level with linear filtering.
624 * If sampling a cube texture, r = cube face in [0,5].
625 */
626 static void
627 lp_build_sample_image_linear(struct lp_build_sample_context *bld,
628 unsigned unit,
629 LLVMValueRef width_vec,
630 LLVMValueRef height_vec,
631 LLVMValueRef depth_vec,
632 LLVMValueRef row_stride_vec,
633 LLVMValueRef img_stride_vec,
634 LLVMValueRef data_ptr,
635 LLVMValueRef s,
636 LLVMValueRef t,
637 LLVMValueRef r,
638 LLVMValueRef colors_out[4])
639 {
640 const unsigned dims = bld->dims;
641 LLVMValueRef x0, y0, z0, x1, y1, z1;
642 LLVMValueRef s_fpart, t_fpart, r_fpart;
643 LLVMValueRef neighbors[2][2][4];
644 int chan;
645
646 /*
647 * Compute integer texcoords.
648 */
649 lp_build_sample_wrap_linear(bld, s, width_vec,
650 bld->static_state->pot_width,
651 bld->static_state->wrap_s,
652 &x0, &x1, &s_fpart);
653 lp_build_name(x0, "tex.x0.wrapped");
654 lp_build_name(x1, "tex.x1.wrapped");
655
656 if (dims >= 2) {
657 lp_build_sample_wrap_linear(bld, t, height_vec,
658 bld->static_state->pot_height,
659 bld->static_state->wrap_t,
660 &y0, &y1, &t_fpart);
661 lp_build_name(y0, "tex.y0.wrapped");
662 lp_build_name(y1, "tex.y1.wrapped");
663
664 if (dims == 3) {
665 lp_build_sample_wrap_linear(bld, r, depth_vec,
666 bld->static_state->pot_depth,
667 bld->static_state->wrap_r,
668 &z0, &z1, &r_fpart);
669 lp_build_name(z0, "tex.z0.wrapped");
670 lp_build_name(z1, "tex.z1.wrapped");
671 }
672 else if (bld->static_state->target == PIPE_TEXTURE_CUBE) {
673 z0 = z1 = r; /* cube face */
674 r_fpart = NULL;
675 }
676 else {
677 z0 = z1 = NULL;
678 r_fpart = NULL;
679 }
680 }
681 else {
682 y0 = y1 = t_fpart = NULL;
683 z0 = z1 = r_fpart = NULL;
684 }
685
686 /*
687 * Get texture colors.
688 */
689 /* get x0/x1 texels */
690 lp_build_sample_texel_soa(bld, unit,
691 width_vec, height_vec, depth_vec,
692 x0, y0, z0,
693 row_stride_vec, img_stride_vec,
694 data_ptr, neighbors[0][0]);
695 lp_build_sample_texel_soa(bld, unit,
696 width_vec, height_vec, depth_vec,
697 x1, y0, z0,
698 row_stride_vec, img_stride_vec,
699 data_ptr, neighbors[0][1]);
700
701 if (dims == 1) {
702 /* Interpolate two samples from 1D image to produce one color */
703 for (chan = 0; chan < 4; chan++) {
704 colors_out[chan] = lp_build_lerp(&bld->texel_bld, s_fpart,
705 neighbors[0][0][chan],
706 neighbors[0][1][chan]);
707 }
708 }
709 else {
710 /* 2D/3D texture */
711 LLVMValueRef colors0[4];
712
713 /* get x0/x1 texels at y1 */
714 lp_build_sample_texel_soa(bld, unit,
715 width_vec, height_vec, depth_vec,
716 x0, y1, z0,
717 row_stride_vec, img_stride_vec,
718 data_ptr, neighbors[1][0]);
719 lp_build_sample_texel_soa(bld, unit,
720 width_vec, height_vec, depth_vec,
721 x1, y1, z0,
722 row_stride_vec, img_stride_vec,
723 data_ptr, neighbors[1][1]);
724
725 /* Bilinear interpolate the four samples from the 2D image / 3D slice */
726 for (chan = 0; chan < 4; chan++) {
727 colors0[chan] = lp_build_lerp_2d(&bld->texel_bld,
728 s_fpart, t_fpart,
729 neighbors[0][0][chan],
730 neighbors[0][1][chan],
731 neighbors[1][0][chan],
732 neighbors[1][1][chan]);
733 }
734
735 if (dims == 3) {
736 LLVMValueRef neighbors1[2][2][4];
737 LLVMValueRef colors1[4];
738
739 /* get x0/x1/y0/y1 texels at z1 */
740 lp_build_sample_texel_soa(bld, unit,
741 width_vec, height_vec, depth_vec,
742 x0, y0, z1,
743 row_stride_vec, img_stride_vec,
744 data_ptr, neighbors1[0][0]);
745 lp_build_sample_texel_soa(bld, unit,
746 width_vec, height_vec, depth_vec,
747 x1, y0, z1,
748 row_stride_vec, img_stride_vec,
749 data_ptr, neighbors1[0][1]);
750 lp_build_sample_texel_soa(bld, unit,
751 width_vec, height_vec, depth_vec,
752 x0, y1, z1,
753 row_stride_vec, img_stride_vec,
754 data_ptr, neighbors1[1][0]);
755 lp_build_sample_texel_soa(bld, unit,
756 width_vec, height_vec, depth_vec,
757 x1, y1, z1,
758 row_stride_vec, img_stride_vec,
759 data_ptr, neighbors1[1][1]);
760
761 /* Bilinear interpolate the four samples from the second Z slice */
762 for (chan = 0; chan < 4; chan++) {
763 colors1[chan] = lp_build_lerp_2d(&bld->texel_bld,
764 s_fpart, t_fpart,
765 neighbors1[0][0][chan],
766 neighbors1[0][1][chan],
767 neighbors1[1][0][chan],
768 neighbors1[1][1][chan]);
769 }
770
771 /* Linearly interpolate the two samples from the two 3D slices */
772 for (chan = 0; chan < 4; chan++) {
773 colors_out[chan] = lp_build_lerp(&bld->texel_bld,
774 r_fpart,
775 colors0[chan], colors1[chan]);
776 }
777 }
778 else {
779 /* 2D tex */
780 for (chan = 0; chan < 4; chan++) {
781 colors_out[chan] = colors0[chan];
782 }
783 }
784 }
785 }
786
787
788 /**
789 * Sample the texture/mipmap using given image filter and mip filter.
790 * data0_ptr and data1_ptr point to the two mipmap levels to sample
791 * from. width0/1_vec, height0/1_vec, depth0/1_vec indicate their sizes.
792 * If we're using nearest miplevel sampling the '1' values will be null/unused.
793 */
794 static void
795 lp_build_sample_mipmap(struct lp_build_sample_context *bld,
796 unsigned unit,
797 unsigned img_filter,
798 unsigned mip_filter,
799 LLVMValueRef s,
800 LLVMValueRef t,
801 LLVMValueRef r,
802 LLVMValueRef ilevel0,
803 LLVMValueRef ilevel1,
804 LLVMValueRef lod_fpart,
805 LLVMValueRef *colors_out)
806 {
807 LLVMBuilderRef builder = bld->builder;
808 LLVMValueRef width0_vec;
809 LLVMValueRef width1_vec;
810 LLVMValueRef height0_vec;
811 LLVMValueRef height1_vec;
812 LLVMValueRef depth0_vec;
813 LLVMValueRef depth1_vec;
814 LLVMValueRef row_stride0_vec;
815 LLVMValueRef row_stride1_vec;
816 LLVMValueRef img_stride0_vec;
817 LLVMValueRef img_stride1_vec;
818 LLVMValueRef data_ptr0;
819 LLVMValueRef data_ptr1;
820 LLVMValueRef colors0[4], colors1[4];
821 unsigned chan;
822
823 /* sample the first mipmap level */
824 lp_build_mipmap_level_sizes(bld, ilevel0,
825 &width0_vec, &height0_vec, &depth0_vec,
826 &row_stride0_vec, &img_stride0_vec);
827 data_ptr0 = lp_build_get_mipmap_level(bld, ilevel0);
828 if (img_filter == PIPE_TEX_FILTER_NEAREST) {
829 lp_build_sample_image_nearest(bld, unit,
830 width0_vec, height0_vec, depth0_vec,
831 row_stride0_vec, img_stride0_vec,
832 data_ptr0, s, t, r,
833 colors0);
834 }
835 else {
836 assert(img_filter == PIPE_TEX_FILTER_LINEAR);
837 lp_build_sample_image_linear(bld, unit,
838 width0_vec, height0_vec, depth0_vec,
839 row_stride0_vec, img_stride0_vec,
840 data_ptr0, s, t, r,
841 colors0);
842 }
843
844 /* Store the first level's colors in the output variables */
845 for (chan = 0; chan < 4; chan++) {
846 LLVMBuildStore(builder, colors0[chan], colors_out[chan]);
847 }
848
849 if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR) {
850 struct lp_build_flow_context *flow_ctx;
851 struct lp_build_if_state if_ctx;
852 LLVMValueRef need_lerp;
853
854 flow_ctx = lp_build_flow_create(builder);
855
856 /* need_lerp = lod_fpart > 0 */
857 need_lerp = LLVMBuildFCmp(builder, LLVMRealUGT,
858 lod_fpart,
859 bld->float_bld.zero,
860 "need_lerp");
861
862 lp_build_if(&if_ctx, flow_ctx, builder, need_lerp);
863 {
864 /* sample the second mipmap level */
865 lp_build_mipmap_level_sizes(bld, ilevel1,
866 &width1_vec, &height1_vec, &depth1_vec,
867 &row_stride1_vec, &img_stride1_vec);
868 data_ptr1 = lp_build_get_mipmap_level(bld, ilevel1);
869 if (img_filter == PIPE_TEX_FILTER_NEAREST) {
870 lp_build_sample_image_nearest(bld, unit,
871 width1_vec, height1_vec, depth1_vec,
872 row_stride1_vec, img_stride1_vec,
873 data_ptr1, s, t, r,
874 colors1);
875 }
876 else {
877 lp_build_sample_image_linear(bld, unit,
878 width1_vec, height1_vec, depth1_vec,
879 row_stride1_vec, img_stride1_vec,
880 data_ptr1, s, t, r,
881 colors1);
882 }
883
884 /* interpolate samples from the two mipmap levels */
885
886 lod_fpart = lp_build_broadcast_scalar(&bld->texel_bld, lod_fpart);
887
888 for (chan = 0; chan < 4; chan++) {
889 colors0[chan] = lp_build_lerp(&bld->texel_bld, lod_fpart,
890 colors0[chan], colors1[chan]);
891 LLVMBuildStore(builder, colors0[chan], colors_out[chan]);
892 }
893 }
894 lp_build_endif(&if_ctx);
895
896 lp_build_flow_destroy(flow_ctx);
897 }
898 }
899
900
901
902 /**
903 * General texture sampling codegen.
904 * This function handles texture sampling for all texture targets (1D,
905 * 2D, 3D, cube) and all filtering modes.
906 */
907 static void
908 lp_build_sample_general(struct lp_build_sample_context *bld,
909 unsigned unit,
910 LLVMValueRef s,
911 LLVMValueRef t,
912 LLVMValueRef r,
913 const LLVMValueRef *ddx,
914 const LLVMValueRef *ddy,
915 LLVMValueRef lod_bias, /* optional */
916 LLVMValueRef explicit_lod, /* optional */
917 LLVMValueRef *colors_out)
918 {
919 struct lp_build_context *int_bld = &bld->int_bld;
920 LLVMBuilderRef builder = bld->builder;
921 const unsigned mip_filter = bld->static_state->min_mip_filter;
922 const unsigned min_filter = bld->static_state->min_img_filter;
923 const unsigned mag_filter = bld->static_state->mag_img_filter;
924 LLVMValueRef lod_ipart = NULL, lod_fpart = NULL;
925 LLVMValueRef ilevel0, ilevel1 = NULL;
926 LLVMValueRef face_ddx[4], face_ddy[4];
927 LLVMValueRef texels[4];
928 LLVMTypeRef i32t = LLVMInt32Type();
929 LLVMValueRef i32t_zero = LLVMConstInt(i32t, 0, 0);
930 unsigned chan;
931
932 /*
933 printf("%s mip %d min %d mag %d\n", __FUNCTION__,
934 mip_filter, min_filter, mag_filter);
935 */
936
937 /*
938 * Choose cube face, recompute texcoords and derivatives for the chosen face.
939 */
940 if (bld->static_state->target == PIPE_TEXTURE_CUBE) {
941 LLVMValueRef face, face_s, face_t;
942 lp_build_cube_lookup(bld, s, t, r, &face, &face_s, &face_t);
943 s = face_s; /* vec */
944 t = face_t; /* vec */
945 /* use 'r' to indicate cube face */
946 r = lp_build_broadcast_scalar(&bld->int_coord_bld, face); /* vec */
947
948 /* recompute ddx, ddy using the new (s,t) face texcoords */
949 face_ddx[0] = lp_build_ddx(&bld->coord_bld, s);
950 face_ddx[1] = lp_build_ddx(&bld->coord_bld, t);
951 face_ddx[2] = NULL;
952 face_ddx[3] = NULL;
953 face_ddy[0] = lp_build_ddy(&bld->coord_bld, s);
954 face_ddy[1] = lp_build_ddy(&bld->coord_bld, t);
955 face_ddy[2] = NULL;
956 face_ddy[3] = NULL;
957 ddx = face_ddx;
958 ddy = face_ddy;
959 }
960
961 /*
962 * Compute the level of detail (float).
963 */
964 if (min_filter != mag_filter ||
965 mip_filter != PIPE_TEX_MIPFILTER_NONE) {
966 /* Need to compute lod either to choose mipmap levels or to
967 * distinguish between minification/magnification with one mipmap level.
968 */
969 lp_build_lod_selector(bld, unit, ddx, ddy,
970 lod_bias, explicit_lod,
971 mip_filter,
972 &lod_ipart, &lod_fpart);
973 } else {
974 lod_ipart = i32t_zero;
975 }
976
977 /*
978 * Compute integer mipmap level(s) to fetch texels from: ilevel0, ilevel1
979 */
980 switch (mip_filter) {
981 default:
982 assert(0 && "bad mip_filter value in lp_build_sample_soa()");
983 /* fall-through */
984 case PIPE_TEX_MIPFILTER_NONE:
985 /* always use mip level 0 */
986 if (bld->static_state->target == PIPE_TEXTURE_CUBE) {
987 /* XXX this is a work-around for an apparent bug in LLVM 2.7.
988 * We should be able to set ilevel0 = const(0) but that causes
989 * bad x86 code to be emitted.
990 */
991 assert(lod_ipart);
992 lp_build_nearest_mip_level(bld, unit, lod_ipart, &ilevel0);
993 }
994 else {
995 ilevel0 = i32t_zero;
996 }
997 break;
998 case PIPE_TEX_MIPFILTER_NEAREST:
999 assert(lod_ipart);
1000 lp_build_nearest_mip_level(bld, unit, lod_ipart, &ilevel0);
1001 break;
1002 case PIPE_TEX_MIPFILTER_LINEAR:
1003 assert(lod_ipart);
1004 assert(lod_fpart);
1005 lp_build_linear_mip_levels(bld, unit,
1006 lod_ipart, &lod_fpart,
1007 &ilevel0, &ilevel1);
1008 break;
1009 }
1010
1011 /*
1012 * Get/interpolate texture colors.
1013 */
1014
1015 for (chan = 0; chan < 4; ++chan) {
1016 texels[chan] = lp_build_alloca(builder, bld->texel_bld.vec_type, "");
1017 lp_build_name(texels[chan], "sampler%u_texel_%c_var", unit, "xyzw"[chan]);
1018 }
1019
1020 if (min_filter == mag_filter) {
1021 /* no need to distinquish between minification and magnification */
1022 lp_build_sample_mipmap(bld, unit,
1023 min_filter, mip_filter,
1024 s, t, r,
1025 ilevel0, ilevel1, lod_fpart,
1026 texels);
1027 }
1028 else {
1029 /* Emit conditional to choose min image filter or mag image filter
1030 * depending on the lod being > 0 or <= 0, respectively.
1031 */
1032 struct lp_build_flow_context *flow_ctx;
1033 struct lp_build_if_state if_ctx;
1034 LLVMValueRef minify;
1035
1036 flow_ctx = lp_build_flow_create(builder);
1037
1038 /* minify = lod >= 0.0 */
1039 minify = LLVMBuildICmp(builder, LLVMIntSGE,
1040 lod_ipart, int_bld->zero, "");
1041
1042 lp_build_if(&if_ctx, flow_ctx, builder, minify);
1043 {
1044 /* Use the minification filter */
1045 lp_build_sample_mipmap(bld, unit,
1046 min_filter, mip_filter,
1047 s, t, r,
1048 ilevel0, ilevel1, lod_fpart,
1049 texels);
1050 }
1051 lp_build_else(&if_ctx);
1052 {
1053 /* Use the magnification filter */
1054 lp_build_sample_mipmap(bld, unit,
1055 mag_filter, PIPE_TEX_MIPFILTER_NONE,
1056 s, t, r,
1057 i32t_zero, NULL, NULL,
1058 texels);
1059 }
1060 lp_build_endif(&if_ctx);
1061
1062 lp_build_flow_destroy(flow_ctx);
1063 }
1064
1065 for (chan = 0; chan < 4; ++chan) {
1066 colors_out[chan] = LLVMBuildLoad(builder, texels[chan], "");
1067 lp_build_name(colors_out[chan], "sampler%u_texel_%c", unit, "xyzw"[chan]);
1068 }
1069 }
1070
1071
1072 /**
1073 * Do shadow test/comparison.
1074 * \param p the texcoord Z (aka R, aka P) component
1075 * \param texel the texel to compare against (use the X channel)
1076 */
1077 static void
1078 lp_build_sample_compare(struct lp_build_sample_context *bld,
1079 LLVMValueRef p,
1080 LLVMValueRef texel[4])
1081 {
1082 struct lp_build_context *texel_bld = &bld->texel_bld;
1083 LLVMValueRef res;
1084 const unsigned chan = 0;
1085
1086 if (bld->static_state->compare_mode == PIPE_TEX_COMPARE_NONE)
1087 return;
1088
1089 /* debug code */
1090 if (0) {
1091 LLVMValueRef indx = lp_build_const_int32(0);
1092 LLVMValueRef coord = LLVMBuildExtractElement(bld->builder, p, indx, "");
1093 LLVMValueRef tex = LLVMBuildExtractElement(bld->builder,
1094 texel[chan], indx, "");
1095 lp_build_printf(bld->builder, "shadow compare coord %f to texture %f\n",
1096 coord, tex);
1097 }
1098
1099 /* result = (p FUNC texel) ? 1 : 0 */
1100 res = lp_build_cmp(texel_bld, bld->static_state->compare_func,
1101 p, texel[chan]);
1102 res = lp_build_select(texel_bld, res, texel_bld->one, texel_bld->zero);
1103
1104 /* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */
1105 texel[0] =
1106 texel[1] =
1107 texel[2] = res;
1108 texel[3] = texel_bld->one;
1109 }
1110
1111
1112 /**
1113 * Just set texels to white instead of actually sampling the texture.
1114 * For debugging.
1115 */
1116 void
1117 lp_build_sample_nop(struct lp_type type,
1118 LLVMValueRef texel_out[4])
1119 {
1120 LLVMValueRef one = lp_build_one(type);
1121 unsigned chan;
1122
1123 for (chan = 0; chan < 4; chan++) {
1124 texel_out[chan] = one;
1125 }
1126 }
1127
1128
1129 /**
1130 * Build texture sampling code.
1131 * 'texel' will return a vector of four LLVMValueRefs corresponding to
1132 * R, G, B, A.
1133 * \param type vector float type to use for coords, etc.
1134 * \param ddx partial derivatives of (s,t,r,q) with respect to x
1135 * \param ddy partial derivatives of (s,t,r,q) with respect to y
1136 */
1137 void
1138 lp_build_sample_soa(LLVMBuilderRef builder,
1139 const struct lp_sampler_static_state *static_state,
1140 struct lp_sampler_dynamic_state *dynamic_state,
1141 struct lp_type type,
1142 unsigned unit,
1143 unsigned num_coords,
1144 const LLVMValueRef *coords,
1145 const LLVMValueRef ddx[4],
1146 const LLVMValueRef ddy[4],
1147 LLVMValueRef lod_bias, /* optional */
1148 LLVMValueRef explicit_lod, /* optional */
1149 LLVMValueRef texel_out[4])
1150 {
1151 unsigned dims = texture_dims(static_state->target);
1152 struct lp_build_sample_context bld;
1153 LLVMTypeRef i32t = LLVMInt32Type();
1154
1155 LLVMValueRef s;
1156 LLVMValueRef t;
1157 LLVMValueRef r;
1158 struct lp_type float_vec_type;
1159
1160 if (0) {
1161 enum pipe_format fmt = static_state->format;
1162 debug_printf("Sample from %s\n", util_format_name(fmt));
1163 }
1164
1165 assert(type.floating);
1166
1167 /* Setup our build context */
1168 memset(&bld, 0, sizeof bld);
1169 bld.builder = builder;
1170 bld.static_state = static_state;
1171 bld.dynamic_state = dynamic_state;
1172 bld.format_desc = util_format_description(static_state->format);
1173 bld.dims = dims;
1174
1175 bld.float_type = lp_type_float(32);
1176 bld.int_type = lp_type_int(32);
1177 bld.coord_type = type;
1178 bld.uint_coord_type = lp_uint_type(type);
1179 bld.int_coord_type = lp_int_type(type);
1180 bld.float_size_type = lp_type_float(32);
1181 bld.float_size_type.length = dims > 1 ? 4 : 1;
1182 bld.int_size_type = lp_int_type(bld.float_size_type);
1183 bld.texel_type = type;
1184
1185 float_vec_type = lp_type_float_vec(32);
1186
1187 lp_build_context_init(&bld.float_bld, builder, bld.float_type);
1188 lp_build_context_init(&bld.float_vec_bld, builder, float_vec_type);
1189 lp_build_context_init(&bld.int_bld, builder, bld.int_type);
1190 lp_build_context_init(&bld.coord_bld, builder, bld.coord_type);
1191 lp_build_context_init(&bld.uint_coord_bld, builder, bld.uint_coord_type);
1192 lp_build_context_init(&bld.int_coord_bld, builder, bld.int_coord_type);
1193 lp_build_context_init(&bld.int_size_bld, builder, bld.int_size_type);
1194 lp_build_context_init(&bld.float_size_bld, builder, bld.float_size_type);
1195 lp_build_context_init(&bld.texel_bld, builder, bld.texel_type);
1196
1197 /* Get the dynamic state */
1198 bld.width = dynamic_state->width(dynamic_state, builder, unit);
1199 bld.height = dynamic_state->height(dynamic_state, builder, unit);
1200 bld.depth = dynamic_state->depth(dynamic_state, builder, unit);
1201 bld.row_stride_array = dynamic_state->row_stride(dynamic_state, builder, unit);
1202 bld.img_stride_array = dynamic_state->img_stride(dynamic_state, builder, unit);
1203 bld.data_array = dynamic_state->data_ptr(dynamic_state, builder, unit);
1204 /* Note that data_array is an array[level] of pointers to texture images */
1205
1206 s = coords[0];
1207 t = coords[1];
1208 r = coords[2];
1209
1210 /* width, height, depth as single int vector */
1211 if (dims <= 1) {
1212 bld.int_size = bld.width;
1213 }
1214 else {
1215 bld.int_size = LLVMBuildInsertElement(builder, bld.int_size_bld.undef,
1216 bld.width, LLVMConstInt(i32t, 0, 0), "");
1217 if (dims >= 2) {
1218 bld.int_size = LLVMBuildInsertElement(builder, bld.int_size,
1219 bld.height, LLVMConstInt(i32t, 1, 0), "");
1220 if (dims >= 3) {
1221 bld.int_size = LLVMBuildInsertElement(builder, bld.int_size,
1222 bld.depth, LLVMConstInt(i32t, 2, 0), "");
1223 }
1224 }
1225 }
1226
1227 if (0) {
1228 /* For debug: no-op texture sampling */
1229 lp_build_sample_nop(bld.texel_type, texel_out);
1230 }
1231 else if (util_format_fits_8unorm(bld.format_desc) &&
1232 lp_is_simple_wrap_mode(static_state->wrap_s) &&
1233 lp_is_simple_wrap_mode(static_state->wrap_t)) {
1234 /* do sampling/filtering with fixed pt arithmetic */
1235 lp_build_sample_aos(&bld, unit, s, t, r, ddx, ddy,
1236 lod_bias, explicit_lod,
1237 texel_out);
1238 }
1239
1240 else {
1241 if ((gallivm_debug & GALLIVM_DEBUG_PERF) &&
1242 util_format_fits_8unorm(bld.format_desc)) {
1243 debug_printf("%s: using floating point linear filtering for %s\n",
1244 __FUNCTION__, bld.format_desc->short_name);
1245 debug_printf(" min_img %d mag_img %d mip %d wraps %d wrapt %d\n",
1246 static_state->min_img_filter,
1247 static_state->mag_img_filter,
1248 static_state->min_mip_filter,
1249 static_state->wrap_s,
1250 static_state->wrap_t);
1251 }
1252
1253 lp_build_sample_general(&bld, unit, s, t, r, ddx, ddy,
1254 lod_bias, explicit_lod,
1255 texel_out);
1256 }
1257
1258 lp_build_sample_compare(&bld, r, texel_out);
1259 }