etnaviv: Use write combine instead of unached mappings for shader bo
[mesa.git] / src / gallium / drivers / etnaviv / etnaviv_shader.c
1 /*
2 * Copyright (c) 2012-2015 Etnaviv Project
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, sub license,
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
12 * next paragraph) shall be included in all copies or substantial portions
13 * of the 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 NON-INFRINGEMENT. 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
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21 * DEALINGS IN THE SOFTWARE.
22 *
23 * Authors:
24 * Wladimir J. van der Laan <laanwj@gmail.com>
25 */
26
27 #include "etnaviv_shader.h"
28
29 #include "etnaviv_compiler.h"
30 #include "etnaviv_context.h"
31 #include "etnaviv_debug.h"
32 #include "etnaviv_screen.h"
33 #include "etnaviv_util.h"
34
35 #include "tgsi/tgsi_parse.h"
36 #include "util/u_math.h"
37 #include "util/u_memory.h"
38
39 /* Upload shader code to bo, if not already done */
40 static bool etna_icache_upload_shader(struct etna_context *ctx, struct etna_shader_variant *v)
41 {
42 if (v->bo)
43 return true;
44 v->bo = etna_bo_new(ctx->screen->dev, v->code_size*4, DRM_ETNA_GEM_CACHE_WC);
45 if (!v->bo)
46 return false;
47
48 void *buf = etna_bo_map(v->bo);
49 etna_bo_cpu_prep(v->bo, DRM_ETNA_PREP_WRITE);
50 memcpy(buf, v->code, v->code_size*4);
51 etna_bo_cpu_fini(v->bo);
52 DBG("Uploaded %s of %u words to bo %p", v->processor == PIPE_SHADER_FRAGMENT ? "fs":"vs", v->code_size, v->bo);
53 return true;
54 }
55
56 /* Link vs and fs together: fill in shader_state from vs and fs
57 * as this function is called every time a new fs or vs is bound, the goal is to
58 * do little processing as possible here, and to precompute as much as possible in
59 * the vs/fs shader_object.
60 *
61 * XXX we could cache the link result for a certain set of VS/PS; usually a pair
62 * of VS and PS will be used together anyway.
63 */
64 static bool
65 etna_link_shaders(struct etna_context *ctx, struct compiled_shader_state *cs,
66 struct etna_shader_variant *vs, struct etna_shader_variant *fs)
67 {
68 struct etna_shader_link_info link = { };
69
70 assert(vs->processor == PIPE_SHADER_VERTEX);
71 assert(fs->processor == PIPE_SHADER_FRAGMENT);
72
73 #ifdef DEBUG
74 if (DBG_ENABLED(ETNA_DBG_DUMP_SHADERS)) {
75 etna_dump_shader(vs);
76 etna_dump_shader(fs);
77 }
78 #endif
79
80 if (etna_link_shader(&link, vs, fs)) {
81 /* linking failed: some fs inputs do not have corresponding
82 * vs outputs */
83 assert(0);
84
85 return false;
86 }
87
88 if (DBG_ENABLED(ETNA_DBG_LINKER_MSGS)) {
89 debug_printf("link result:\n");
90 debug_printf(" vs -> fs comps use pa_attr\n");
91
92 for (int idx = 0; idx < link.num_varyings; ++idx)
93 debug_printf(" t%-2u -> t%-2u %-5.*s %u,%u,%u,%u 0x%08x\n",
94 link.varyings[idx].reg, idx + 1,
95 link.varyings[idx].num_components, "xyzw",
96 link.varyings[idx].use[0], link.varyings[idx].use[1],
97 link.varyings[idx].use[2], link.varyings[idx].use[3],
98 link.varyings[idx].pa_attributes);
99 }
100
101 /* set last_varying_2x flag if the last varying has 1 or 2 components */
102 bool last_varying_2x = false;
103 if (link.num_varyings > 0 && link.varyings[link.num_varyings - 1].num_components <= 2)
104 last_varying_2x = true;
105
106 cs->RA_CONTROL = VIVS_RA_CONTROL_UNK0 |
107 COND(last_varying_2x, VIVS_RA_CONTROL_LAST_VARYING_2X);
108
109 cs->PA_ATTRIBUTE_ELEMENT_COUNT = VIVS_PA_ATTRIBUTE_ELEMENT_COUNT_COUNT(link.num_varyings);
110 for (int idx = 0; idx < link.num_varyings; ++idx)
111 cs->PA_SHADER_ATTRIBUTES[idx] = link.varyings[idx].pa_attributes;
112
113 cs->VS_END_PC = vs->code_size / 4;
114 cs->VS_OUTPUT_COUNT = 1 + link.num_varyings; /* position + varyings */
115
116 /* vs outputs (varyings) */
117 DEFINE_ETNA_BITARRAY(vs_output, 16, 8) = {0};
118 int varid = 0;
119 etna_bitarray_set(vs_output, 8, varid++, vs->vs_pos_out_reg);
120 for (int idx = 0; idx < link.num_varyings; ++idx)
121 etna_bitarray_set(vs_output, 8, varid++, link.varyings[idx].reg);
122 if (vs->vs_pointsize_out_reg >= 0)
123 etna_bitarray_set(vs_output, 8, varid++, vs->vs_pointsize_out_reg); /* pointsize is last */
124
125 for (int idx = 0; idx < ARRAY_SIZE(cs->VS_OUTPUT); ++idx)
126 cs->VS_OUTPUT[idx] = vs_output[idx];
127
128 if (vs->vs_pointsize_out_reg != -1) {
129 /* vertex shader outputs point coordinate, provide extra output and make
130 * sure PA config is
131 * not masked */
132 cs->PA_CONFIG = ~0;
133 cs->VS_OUTPUT_COUNT_PSIZE = cs->VS_OUTPUT_COUNT + 1;
134 } else {
135 /* vertex shader does not output point coordinate, make sure thate
136 * POINT_SIZE_ENABLE is masked
137 * and no extra output is given */
138 cs->PA_CONFIG = ~VIVS_PA_CONFIG_POINT_SIZE_ENABLE;
139 cs->VS_OUTPUT_COUNT_PSIZE = cs->VS_OUTPUT_COUNT;
140 }
141
142 cs->VS_LOAD_BALANCING = vs->vs_load_balancing;
143 cs->VS_START_PC = 0;
144
145 cs->PS_END_PC = fs->code_size / 4;
146 cs->PS_OUTPUT_REG = fs->ps_color_out_reg;
147 cs->PS_INPUT_COUNT =
148 VIVS_PS_INPUT_COUNT_COUNT(link.num_varyings + 1) | /* Number of inputs plus position */
149 VIVS_PS_INPUT_COUNT_UNK8(fs->input_count_unk8);
150 cs->PS_TEMP_REGISTER_CONTROL =
151 VIVS_PS_TEMP_REGISTER_CONTROL_NUM_TEMPS(MAX2(fs->num_temps, link.num_varyings + 1));
152 cs->PS_CONTROL = VIVS_PS_CONTROL_UNK1; /* XXX when can we set BYPASS? */
153 cs->PS_START_PC = 0;
154
155 /* Precompute PS_INPUT_COUNT and TEMP_REGISTER_CONTROL in the case of MSAA
156 * mode, avoids some fumbling in sync_context. */
157 cs->PS_INPUT_COUNT_MSAA =
158 VIVS_PS_INPUT_COUNT_COUNT(link.num_varyings + 2) | /* MSAA adds another input */
159 VIVS_PS_INPUT_COUNT_UNK8(fs->input_count_unk8);
160 cs->PS_TEMP_REGISTER_CONTROL_MSAA =
161 VIVS_PS_TEMP_REGISTER_CONTROL_NUM_TEMPS(MAX2(fs->num_temps, link.num_varyings + 2));
162
163 uint32_t total_components = 0;
164 DEFINE_ETNA_BITARRAY(num_components, ETNA_NUM_VARYINGS, 4) = {0};
165 DEFINE_ETNA_BITARRAY(component_use, 4 * ETNA_NUM_VARYINGS, 2) = {0};
166 for (int idx = 0; idx < link.num_varyings; ++idx) {
167 const struct etna_varying *varying = &link.varyings[idx];
168
169 etna_bitarray_set(num_components, 4, idx, varying->num_components);
170 for (int comp = 0; comp < varying->num_components; ++comp) {
171 etna_bitarray_set(component_use, 2, total_components, varying->use[comp]);
172 total_components += 1;
173 }
174 }
175
176 cs->GL_VARYING_TOTAL_COMPONENTS =
177 VIVS_GL_VARYING_TOTAL_COMPONENTS_NUM(align(total_components, 2));
178 cs->GL_VARYING_NUM_COMPONENTS = num_components[0];
179 cs->GL_VARYING_COMPONENT_USE[0] = component_use[0];
180 cs->GL_VARYING_COMPONENT_USE[1] = component_use[1];
181
182 cs->GL_HALTI5_SH_SPECIALS =
183 0x7f7f0000 | /* unknown bits, probably other PS inputs */
184 /* pointsize is last (see above) */
185 VIVS_GL_HALTI5_SH_SPECIALS_VS_PSIZE_OUT((vs->vs_pointsize_out_reg != -1) ?
186 cs->VS_OUTPUT_COUNT * 4 : 0x00) |
187 VIVS_GL_HALTI5_SH_SPECIALS_PS_PCOORD_IN((link.pcoord_varying_comp_ofs != -1) ?
188 link.pcoord_varying_comp_ofs : 0x7f);
189
190 /* reference instruction memory */
191 cs->vs_inst_mem_size = vs->code_size;
192 cs->VS_INST_MEM = vs->code;
193
194 cs->ps_inst_mem_size = fs->code_size;
195 cs->PS_INST_MEM = fs->code;
196
197 if (vs->needs_icache | fs->needs_icache) {
198 /* If either of the shaders needs ICACHE, we use it for both. It is
199 * either switched on or off for the entire shader processor.
200 */
201 if (!etna_icache_upload_shader(ctx, vs) ||
202 !etna_icache_upload_shader(ctx, fs)) {
203 assert(0);
204 return false;
205 }
206
207 cs->VS_INST_ADDR.bo = vs->bo;
208 cs->VS_INST_ADDR.offset = 0;
209 cs->VS_INST_ADDR.flags = ETNA_RELOC_READ;
210 cs->PS_INST_ADDR.bo = fs->bo;
211 cs->PS_INST_ADDR.offset = 0;
212 cs->PS_INST_ADDR.flags = ETNA_RELOC_READ;
213 } else {
214 /* clear relocs */
215 memset(&cs->VS_INST_ADDR, 0, sizeof(cs->VS_INST_ADDR));
216 memset(&cs->PS_INST_ADDR, 0, sizeof(cs->PS_INST_ADDR));
217 }
218
219 return true;
220 }
221
222 bool
223 etna_shader_link(struct etna_context *ctx)
224 {
225 if (!ctx->shader.vs || !ctx->shader.fs)
226 return false;
227
228 /* re-link vs and fs if needed */
229 return etna_link_shaders(ctx, &ctx->shader_state, ctx->shader.vs, ctx->shader.fs);
230 }
231
232 static bool
233 etna_shader_update_vs_inputs(struct etna_context *ctx,
234 struct compiled_shader_state *cs,
235 const struct etna_shader_variant *vs,
236 const struct compiled_vertex_elements_state *ves)
237 {
238 unsigned num_temps, cur_temp, num_vs_inputs;
239
240 if (!vs)
241 return false;
242
243 /* Number of vertex elements determines number of VS inputs. Otherwise,
244 * the GPU crashes. Allocate any unused vertex elements to VS temporary
245 * registers. */
246 num_vs_inputs = MAX2(ves->num_elements, vs->infile.num_reg);
247 if (num_vs_inputs != ves->num_elements) {
248 BUG("Number of elements %u does not match the number of VS inputs %zu",
249 ctx->vertex_elements->num_elements, ctx->shader.vs->infile.num_reg);
250 return false;
251 }
252
253 cur_temp = vs->num_temps;
254 num_temps = num_vs_inputs - vs->infile.num_reg + cur_temp;
255
256 cs->VS_INPUT_COUNT = VIVS_VS_INPUT_COUNT_COUNT(num_vs_inputs) |
257 VIVS_VS_INPUT_COUNT_UNK8(vs->input_count_unk8);
258 cs->VS_TEMP_REGISTER_CONTROL =
259 VIVS_VS_TEMP_REGISTER_CONTROL_NUM_TEMPS(num_temps);
260
261 /* vs inputs (attributes) */
262 DEFINE_ETNA_BITARRAY(vs_input, 16, 8) = {0};
263 for (int idx = 0; idx < num_vs_inputs; ++idx) {
264 if (idx < vs->infile.num_reg)
265 etna_bitarray_set(vs_input, 8, idx, vs->infile.reg[idx].reg);
266 else
267 etna_bitarray_set(vs_input, 8, idx, cur_temp++);
268 }
269
270 for (int idx = 0; idx < ARRAY_SIZE(cs->VS_INPUT); ++idx)
271 cs->VS_INPUT[idx] = vs_input[idx];
272
273 return true;
274 }
275
276 static inline const char *
277 etna_shader_stage(struct etna_shader_variant *shader)
278 {
279 switch (shader->processor) {
280 case PIPE_SHADER_VERTEX: return "VERT";
281 case PIPE_SHADER_FRAGMENT: return "FRAG";
282 case PIPE_SHADER_COMPUTE: return "CL";
283 default:
284 unreachable("invalid type");
285 return NULL;
286 }
287 }
288
289 static void
290 dump_shader_info(struct etna_shader_variant *v, struct pipe_debug_callback *debug)
291 {
292 if (!unlikely(etna_mesa_debug & ETNA_DBG_SHADERDB))
293 return;
294
295 pipe_debug_message(debug, SHADER_INFO, "\n"
296 "SHADER-DB: %s prog %d/%d: %u instructions %u temps\n"
297 "SHADER-DB: %s prog %d/%d: %u immediates %u consts\n"
298 "SHADER-DB: %s prog %d/%d: %u loops\n",
299 etna_shader_stage(v),
300 v->shader->id, v->id,
301 v->code_size,
302 v->num_temps,
303 etna_shader_stage(v),
304 v->shader->id, v->id,
305 v->uniforms.imm_count,
306 v->uniforms.const_count,
307 etna_shader_stage(v),
308 v->shader->id, v->id,
309 v->num_loops);
310 }
311
312 bool
313 etna_shader_update_vertex(struct etna_context *ctx)
314 {
315 return etna_shader_update_vs_inputs(ctx, &ctx->shader_state, ctx->shader.vs,
316 ctx->vertex_elements);
317 }
318
319 static struct etna_shader_variant *
320 create_variant(struct etna_shader *shader, struct etna_shader_key key)
321 {
322 struct etna_shader_variant *v = CALLOC_STRUCT(etna_shader_variant);
323 int ret;
324
325 if (!v)
326 return NULL;
327
328 v->shader = shader;
329 v->key = key;
330
331 ret = etna_compile_shader(v);
332 if (!ret) {
333 debug_error("compile failed!");
334 goto fail;
335 }
336
337 v->id = ++shader->variant_count;
338
339 return v;
340
341 fail:
342 FREE(v);
343 return NULL;
344 }
345
346 struct etna_shader_variant *
347 etna_shader_variant(struct etna_shader *shader, struct etna_shader_key key,
348 struct pipe_debug_callback *debug)
349 {
350 struct etna_shader_variant *v;
351
352 for (v = shader->variants; v; v = v->next)
353 if (etna_shader_key_equal(&key, &v->key))
354 return v;
355
356 /* compile new variant if it doesn't exist already */
357 v = create_variant(shader, key);
358 if (v) {
359 v->next = shader->variants;
360 shader->variants = v;
361 dump_shader_info(v, debug);
362 }
363
364 return v;
365 }
366
367 static void *
368 etna_create_shader_state(struct pipe_context *pctx,
369 const struct pipe_shader_state *pss)
370 {
371 struct etna_context *ctx = etna_context(pctx);
372 struct etna_shader *shader = CALLOC_STRUCT(etna_shader);
373
374 if (!shader)
375 return NULL;
376
377 static uint32_t id;
378 shader->id = id++;
379 shader->specs = &ctx->specs;
380 shader->tokens = tgsi_dup_tokens(pss->tokens);
381
382 if (etna_mesa_debug & ETNA_DBG_SHADERDB) {
383 /* if shader-db run, create a standard variant immediately
384 * (as otherwise nothing will trigger the shader to be
385 * actually compiled).
386 */
387 struct etna_shader_key key = {};
388 etna_shader_variant(shader, key, &ctx->debug);
389 }
390
391 return shader;
392 }
393
394 static void
395 etna_delete_shader_state(struct pipe_context *pctx, void *ss)
396 {
397 struct etna_shader *shader = ss;
398 struct etna_shader_variant *v, *t;
399
400 v = shader->variants;
401 while (v) {
402 t = v;
403 v = v->next;
404 if (t->bo)
405 etna_bo_del(t->bo);
406 etna_destroy_shader(t);
407 }
408
409 FREE(shader->tokens);
410 FREE(shader);
411 }
412
413 static void
414 etna_bind_fs_state(struct pipe_context *pctx, void *hwcso)
415 {
416 struct etna_context *ctx = etna_context(pctx);
417
418 ctx->shader.bind_fs = hwcso;
419 ctx->dirty |= ETNA_DIRTY_SHADER;
420 }
421
422 static void
423 etna_bind_vs_state(struct pipe_context *pctx, void *hwcso)
424 {
425 struct etna_context *ctx = etna_context(pctx);
426
427 ctx->shader.bind_vs = hwcso;
428 ctx->dirty |= ETNA_DIRTY_SHADER;
429 }
430
431 void
432 etna_shader_init(struct pipe_context *pctx)
433 {
434 pctx->create_fs_state = etna_create_shader_state;
435 pctx->bind_fs_state = etna_bind_fs_state;
436 pctx->delete_fs_state = etna_delete_shader_state;
437 pctx->create_vs_state = etna_create_shader_state;
438 pctx->bind_vs_state = etna_bind_vs_state;
439 pctx->delete_vs_state = etna_delete_shader_state;
440 }