freedreno: Make the slice pitch be bytes, not pixels.
[mesa.git] / src / gallium / drivers / freedreno / a5xx / fd5_texture.c
1 /*
2 * Copyright (C) 2016 Rob Clark <robclark@freedesktop.org>
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21 * SOFTWARE.
22 *
23 * Authors:
24 * Rob Clark <robclark@freedesktop.org>
25 */
26
27 #include "pipe/p_state.h"
28 #include "util/u_string.h"
29 #include "util/u_memory.h"
30 #include "util/u_inlines.h"
31 #include "util/format/u_format.h"
32
33 #include "fd5_texture.h"
34 #include "fd5_format.h"
35
36 static enum a5xx_tex_clamp
37 tex_clamp(unsigned wrap, bool clamp_to_edge, bool *needs_border)
38 {
39 /* Hardware does not support _CLAMP, but we emulate it: */
40 if (wrap == PIPE_TEX_WRAP_CLAMP) {
41 wrap = (clamp_to_edge) ?
42 PIPE_TEX_WRAP_CLAMP_TO_EDGE : PIPE_TEX_WRAP_CLAMP_TO_BORDER;
43 }
44
45 switch (wrap) {
46 case PIPE_TEX_WRAP_REPEAT:
47 return A5XX_TEX_REPEAT;
48 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
49 return A5XX_TEX_CLAMP_TO_EDGE;
50 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
51 *needs_border = true;
52 return A5XX_TEX_CLAMP_TO_BORDER;
53 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
54 /* only works for PoT.. need to emulate otherwise! */
55 return A5XX_TEX_MIRROR_CLAMP;
56 case PIPE_TEX_WRAP_MIRROR_REPEAT:
57 return A5XX_TEX_MIRROR_REPEAT;
58 case PIPE_TEX_WRAP_MIRROR_CLAMP:
59 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
60 /* these two we could perhaps emulate, but we currently
61 * just don't advertise PIPE_CAP_TEXTURE_MIRROR_CLAMP
62 */
63 default:
64 DBG("invalid wrap: %u", wrap);
65 return 0;
66 }
67 }
68
69 static enum a5xx_tex_filter
70 tex_filter(unsigned filter, bool aniso)
71 {
72 switch (filter) {
73 case PIPE_TEX_FILTER_NEAREST:
74 return A5XX_TEX_NEAREST;
75 case PIPE_TEX_FILTER_LINEAR:
76 return aniso ? A5XX_TEX_ANISO : A5XX_TEX_LINEAR;
77 default:
78 DBG("invalid filter: %u", filter);
79 return 0;
80 }
81 }
82
83 static void *
84 fd5_sampler_state_create(struct pipe_context *pctx,
85 const struct pipe_sampler_state *cso)
86 {
87 struct fd5_sampler_stateobj *so = CALLOC_STRUCT(fd5_sampler_stateobj);
88 unsigned aniso = util_last_bit(MIN2(cso->max_anisotropy >> 1, 8));
89 bool miplinear = false;
90 bool clamp_to_edge;
91
92 if (!so)
93 return NULL;
94
95 so->base = *cso;
96
97 if (cso->min_mip_filter == PIPE_TEX_MIPFILTER_LINEAR)
98 miplinear = true;
99
100 /*
101 * For nearest filtering, _CLAMP means _CLAMP_TO_EDGE; for linear
102 * filtering, _CLAMP means _CLAMP_TO_BORDER while additionally
103 * clamping the texture coordinates to [0.0, 1.0].
104 *
105 * The clamping will be taken care of in the shaders. There are two
106 * filters here, but let the minification one has a say.
107 */
108 clamp_to_edge = (cso->min_img_filter == PIPE_TEX_FILTER_NEAREST);
109 if (!clamp_to_edge) {
110 so->saturate_s = (cso->wrap_s == PIPE_TEX_WRAP_CLAMP);
111 so->saturate_t = (cso->wrap_t == PIPE_TEX_WRAP_CLAMP);
112 so->saturate_r = (cso->wrap_r == PIPE_TEX_WRAP_CLAMP);
113 }
114
115 so->needs_border = false;
116 so->texsamp0 =
117 COND(miplinear, A5XX_TEX_SAMP_0_MIPFILTER_LINEAR_NEAR) |
118 A5XX_TEX_SAMP_0_XY_MAG(tex_filter(cso->mag_img_filter, aniso)) |
119 A5XX_TEX_SAMP_0_XY_MIN(tex_filter(cso->min_img_filter, aniso)) |
120 A5XX_TEX_SAMP_0_ANISO(aniso) |
121 A5XX_TEX_SAMP_0_WRAP_S(tex_clamp(cso->wrap_s, clamp_to_edge, &so->needs_border)) |
122 A5XX_TEX_SAMP_0_WRAP_T(tex_clamp(cso->wrap_t, clamp_to_edge, &so->needs_border)) |
123 A5XX_TEX_SAMP_0_WRAP_R(tex_clamp(cso->wrap_r, clamp_to_edge, &so->needs_border));
124
125 so->texsamp1 =
126 COND(!cso->seamless_cube_map, A5XX_TEX_SAMP_1_CUBEMAPSEAMLESSFILTOFF) |
127 COND(!cso->normalized_coords, A5XX_TEX_SAMP_1_UNNORM_COORDS);
128
129 so->texsamp0 |= A5XX_TEX_SAMP_0_LOD_BIAS(cso->lod_bias);
130
131 if (cso->min_mip_filter != PIPE_TEX_MIPFILTER_NONE) {
132 so->texsamp1 |=
133 A5XX_TEX_SAMP_1_MIN_LOD(cso->min_lod) |
134 A5XX_TEX_SAMP_1_MAX_LOD(cso->max_lod);
135 } else {
136 /* If we're not doing mipmap filtering, we still need a slightly > 0
137 * LOD clamp so the HW can decide between min and mag filtering of
138 * level 0.
139 */
140 so->texsamp1 |=
141 A5XX_TEX_SAMP_1_MIN_LOD(MIN2(cso->min_lod, 0.125)) |
142 A5XX_TEX_SAMP_1_MAX_LOD(MIN2(cso->max_lod, 0.125));
143 }
144
145 if (cso->compare_mode)
146 so->texsamp1 |= A5XX_TEX_SAMP_1_COMPARE_FUNC(cso->compare_func); /* maps 1:1 */
147
148 return so;
149 }
150
151 static void
152 fd5_sampler_states_bind(struct pipe_context *pctx,
153 enum pipe_shader_type shader, unsigned start,
154 unsigned nr, void **hwcso)
155 {
156 struct fd_context *ctx = fd_context(pctx);
157 struct fd5_context *fd5_ctx = fd5_context(ctx);
158 uint16_t saturate_s = 0, saturate_t = 0, saturate_r = 0;
159 unsigned i;
160
161 if (!hwcso)
162 nr = 0;
163
164 for (i = 0; i < nr; i++) {
165 if (hwcso[i]) {
166 struct fd5_sampler_stateobj *sampler =
167 fd5_sampler_stateobj(hwcso[i]);
168 if (sampler->saturate_s)
169 saturate_s |= (1 << i);
170 if (sampler->saturate_t)
171 saturate_t |= (1 << i);
172 if (sampler->saturate_r)
173 saturate_r |= (1 << i);
174 }
175 }
176
177 fd_sampler_states_bind(pctx, shader, start, nr, hwcso);
178
179 if (shader == PIPE_SHADER_FRAGMENT) {
180 fd5_ctx->fsaturate =
181 (saturate_s != 0) ||
182 (saturate_t != 0) ||
183 (saturate_r != 0);
184 fd5_ctx->fsaturate_s = saturate_s;
185 fd5_ctx->fsaturate_t = saturate_t;
186 fd5_ctx->fsaturate_r = saturate_r;
187 } else if (shader == PIPE_SHADER_VERTEX) {
188 fd5_ctx->vsaturate =
189 (saturate_s != 0) ||
190 (saturate_t != 0) ||
191 (saturate_r != 0);
192 fd5_ctx->vsaturate_s = saturate_s;
193 fd5_ctx->vsaturate_t = saturate_t;
194 fd5_ctx->vsaturate_r = saturate_r;
195 }
196 }
197
198 static bool
199 use_astc_srgb_workaround(struct pipe_context *pctx, enum pipe_format format)
200 {
201 return false; // TODO check if this is still needed on a5xx
202 }
203
204 static struct pipe_sampler_view *
205 fd5_sampler_view_create(struct pipe_context *pctx, struct pipe_resource *prsc,
206 const struct pipe_sampler_view *cso)
207 {
208 struct fd5_pipe_sampler_view *so = CALLOC_STRUCT(fd5_pipe_sampler_view);
209 struct fd_resource *rsc = fd_resource(prsc);
210 struct fdl_slice *slice = NULL;
211 enum pipe_format format = cso->format;
212 unsigned lvl, layers = 0;
213
214 if (!so)
215 return NULL;
216
217 if (format == PIPE_FORMAT_X32_S8X24_UINT) {
218 rsc = rsc->stencil;
219 format = rsc->base.format;
220 }
221
222 so->base = *cso;
223 pipe_reference(NULL, &prsc->reference);
224 so->base.texture = prsc;
225 so->base.reference.count = 1;
226 so->base.context = pctx;
227
228 so->texconst0 =
229 A5XX_TEX_CONST_0_FMT(fd5_pipe2tex(format)) |
230 A5XX_TEX_CONST_0_SAMPLES(fd_msaa_samples(prsc->nr_samples)) |
231 fd5_tex_swiz(format, cso->swizzle_r, cso->swizzle_g,
232 cso->swizzle_b, cso->swizzle_a);
233
234 /* NOTE: since we sample z24s8 using 8888_UINT format, the swizzle
235 * we get isn't quite right. Use SWAP(XYZW) as a cheap and cheerful
236 * way to re-arrange things so stencil component is where the swiz
237 * expects.
238 *
239 * Note that gallium expects stencil sampler to return (s,s,s,s)
240 * which isn't quite true. To make that happen we'd have to massage
241 * the swizzle. But in practice only the .x component is used.
242 */
243 if (format == PIPE_FORMAT_X24S8_UINT) {
244 so->texconst0 |= A5XX_TEX_CONST_0_SWAP(XYZW);
245 }
246
247 if (util_format_is_srgb(format)) {
248 if (use_astc_srgb_workaround(pctx, format))
249 so->astc_srgb = true;
250 so->texconst0 |= A5XX_TEX_CONST_0_SRGB;
251 }
252
253 if (cso->target == PIPE_BUFFER) {
254 unsigned elements = cso->u.buf.size / util_format_get_blocksize(format);
255
256 lvl = 0;
257 so->texconst1 =
258 A5XX_TEX_CONST_1_WIDTH(elements) |
259 A5XX_TEX_CONST_1_HEIGHT(1);
260 so->texconst2 =
261 A5XX_TEX_CONST_2_FETCHSIZE(fd5_pipe2fetchsize(format)) |
262 A5XX_TEX_CONST_2_PITCH(elements * rsc->layout.cpp);
263 so->offset = cso->u.buf.offset;
264 } else {
265 unsigned miplevels;
266
267 lvl = fd_sampler_first_level(cso);
268 slice = fd_resource_slice(rsc, lvl);
269 miplevels = fd_sampler_last_level(cso) - lvl;
270 layers = cso->u.tex.last_layer - cso->u.tex.first_layer + 1;
271
272 so->texconst0 |= A5XX_TEX_CONST_0_MIPLVLS(miplevels);
273 so->texconst1 =
274 A5XX_TEX_CONST_1_WIDTH(u_minify(prsc->width0, lvl)) |
275 A5XX_TEX_CONST_1_HEIGHT(u_minify(prsc->height0, lvl));
276 so->texconst2 =
277 A5XX_TEX_CONST_2_FETCHSIZE(fd5_pipe2fetchsize(format)) |
278 A5XX_TEX_CONST_2_PITCH(slice->pitch);
279 so->offset = fd_resource_offset(rsc, lvl, cso->u.tex.first_layer);
280 }
281
282 so->texconst2 |= A5XX_TEX_CONST_2_TYPE(fd5_tex_type(cso->target));
283
284 switch (cso->target) {
285 case PIPE_TEXTURE_RECT:
286 case PIPE_TEXTURE_1D:
287 case PIPE_TEXTURE_2D:
288 so->texconst3 =
289 A5XX_TEX_CONST_3_ARRAY_PITCH(rsc->layout.layer_size);
290 so->texconst5 =
291 A5XX_TEX_CONST_5_DEPTH(1);
292 break;
293 case PIPE_TEXTURE_1D_ARRAY:
294 case PIPE_TEXTURE_2D_ARRAY:
295 so->texconst3 =
296 A5XX_TEX_CONST_3_ARRAY_PITCH(rsc->layout.layer_size);
297 so->texconst5 =
298 A5XX_TEX_CONST_5_DEPTH(layers);
299 break;
300 case PIPE_TEXTURE_CUBE:
301 case PIPE_TEXTURE_CUBE_ARRAY:
302 so->texconst3 =
303 A5XX_TEX_CONST_3_ARRAY_PITCH(rsc->layout.layer_size);
304 so->texconst5 =
305 A5XX_TEX_CONST_5_DEPTH(layers / 6);
306 break;
307 case PIPE_TEXTURE_3D:
308 so->texconst3 =
309 A5XX_TEX_CONST_3_ARRAY_PITCH(slice->size0);
310 so->texconst5 =
311 A5XX_TEX_CONST_5_DEPTH(u_minify(prsc->depth0, lvl));
312 break;
313 default:
314 so->texconst3 = 0x00000000;
315 break;
316 }
317
318 return &so->base;
319 }
320
321 static void
322 fd5_set_sampler_views(struct pipe_context *pctx, enum pipe_shader_type shader,
323 unsigned start, unsigned nr,
324 struct pipe_sampler_view **views)
325 {
326 struct fd_context *ctx = fd_context(pctx);
327 struct fd5_context *fd5_ctx = fd5_context(ctx);
328 uint16_t astc_srgb = 0;
329 unsigned i;
330
331 for (i = 0; i < nr; i++) {
332 if (views[i]) {
333 struct fd5_pipe_sampler_view *view =
334 fd5_pipe_sampler_view(views[i]);
335 if (view->astc_srgb)
336 astc_srgb |= (1 << i);
337 }
338 }
339
340 fd_set_sampler_views(pctx, shader, start, nr, views);
341
342 if (shader == PIPE_SHADER_FRAGMENT) {
343 fd5_ctx->fastc_srgb = astc_srgb;
344 } else if (shader == PIPE_SHADER_VERTEX) {
345 fd5_ctx->vastc_srgb = astc_srgb;
346 }
347 }
348
349 void
350 fd5_texture_init(struct pipe_context *pctx)
351 {
352 pctx->create_sampler_state = fd5_sampler_state_create;
353 pctx->bind_sampler_states = fd5_sampler_states_bind;
354 pctx->create_sampler_view = fd5_sampler_view_create;
355 pctx->set_sampler_views = fd5_set_sampler_views;
356 }