i965/icl: Update the assert in brw_memory_barrier()
[mesa.git] / src / mesa / drivers / dri / i965 / brw_tcs.c
1 /*
2 * Copyright © 2014 Intel Corporation
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
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21 * DEALINGS IN THE SOFTWARE.
22 */
23
24 /**
25 * \file brw_tcs.c
26 *
27 * Tessellation control shader state upload code.
28 */
29
30 #include "brw_context.h"
31 #include "compiler/brw_nir.h"
32 #include "brw_program.h"
33 #include "brw_state.h"
34 #include "program/prog_parameter.h"
35 #include "nir_builder.h"
36
37 static nir_shader *
38 create_passthrough_tcs(void *mem_ctx, const struct brw_compiler *compiler,
39 const nir_shader_compiler_options *options,
40 const struct brw_tcs_prog_key *key)
41 {
42 nir_builder b;
43 nir_builder_init_simple_shader(&b, mem_ctx, MESA_SHADER_TESS_CTRL,
44 options);
45 nir_shader *nir = b.shader;
46 nir_variable *var;
47 nir_intrinsic_instr *load;
48 nir_intrinsic_instr *store;
49 nir_ssa_def *zero = nir_imm_int(&b, 0);
50 nir_ssa_def *invoc_id =
51 nir_load_system_value(&b, nir_intrinsic_load_invocation_id, 0);
52
53 nir->info.inputs_read = key->outputs_written &
54 ~(VARYING_BIT_TESS_LEVEL_INNER | VARYING_BIT_TESS_LEVEL_OUTER);
55 nir->info.outputs_written = key->outputs_written;
56 nir->info.tess.tcs_vertices_out = key->input_vertices;
57 nir->info.name = ralloc_strdup(nir, "passthrough");
58 nir->num_uniforms = 8 * sizeof(uint32_t);
59
60 var = nir_variable_create(nir, nir_var_uniform, glsl_vec4_type(), "hdr_0");
61 var->data.location = 0;
62 var = nir_variable_create(nir, nir_var_uniform, glsl_vec4_type(), "hdr_1");
63 var->data.location = 1;
64
65 /* Write the patch URB header. */
66 for (int i = 0; i <= 1; i++) {
67 load = nir_intrinsic_instr_create(nir, nir_intrinsic_load_uniform);
68 load->num_components = 4;
69 load->src[0] = nir_src_for_ssa(zero);
70 nir_ssa_dest_init(&load->instr, &load->dest, 4, 32, NULL);
71 nir_intrinsic_set_base(load, i * 4 * sizeof(uint32_t));
72 nir_builder_instr_insert(&b, &load->instr);
73
74 store = nir_intrinsic_instr_create(nir, nir_intrinsic_store_output);
75 store->num_components = 4;
76 store->src[0] = nir_src_for_ssa(&load->dest.ssa);
77 store->src[1] = nir_src_for_ssa(zero);
78 nir_intrinsic_set_base(store, VARYING_SLOT_TESS_LEVEL_INNER - i);
79 nir_intrinsic_set_write_mask(store, WRITEMASK_XYZW);
80 nir_builder_instr_insert(&b, &store->instr);
81 }
82
83 /* Copy inputs to outputs. */
84 uint64_t varyings = nir->info.inputs_read;
85
86 while (varyings != 0) {
87 const int varying = ffsll(varyings) - 1;
88
89 load = nir_intrinsic_instr_create(nir,
90 nir_intrinsic_load_per_vertex_input);
91 load->num_components = 4;
92 load->src[0] = nir_src_for_ssa(invoc_id);
93 load->src[1] = nir_src_for_ssa(zero);
94 nir_ssa_dest_init(&load->instr, &load->dest, 4, 32, NULL);
95 nir_intrinsic_set_base(load, varying);
96 nir_builder_instr_insert(&b, &load->instr);
97
98 store = nir_intrinsic_instr_create(nir,
99 nir_intrinsic_store_per_vertex_output);
100 store->num_components = 4;
101 store->src[0] = nir_src_for_ssa(&load->dest.ssa);
102 store->src[1] = nir_src_for_ssa(invoc_id);
103 store->src[2] = nir_src_for_ssa(zero);
104 nir_intrinsic_set_base(store, varying);
105 nir_intrinsic_set_write_mask(store, WRITEMASK_XYZW);
106 nir_builder_instr_insert(&b, &store->instr);
107
108 varyings &= ~BITFIELD64_BIT(varying);
109 }
110
111 nir_validate_shader(nir);
112
113 nir = brw_preprocess_nir(compiler, nir);
114
115 return nir;
116 }
117
118 static void
119 brw_tcs_debug_recompile(struct brw_context *brw, struct gl_program *prog,
120 const struct brw_tcs_prog_key *key)
121 {
122 perf_debug("Recompiling tessellation control shader for program %d\n",
123 prog->Id);
124
125 bool found = false;
126 const struct brw_tcs_prog_key *old_key =
127 brw_find_previous_compile(&brw->cache, BRW_CACHE_TCS_PROG,
128 key->program_string_id);
129
130 if (!old_key) {
131 perf_debug(" Didn't find previous compile in the shader cache for "
132 "debug\n");
133 return;
134 }
135
136 found |= key_debug(brw, "input vertices", old_key->input_vertices,
137 key->input_vertices);
138 found |= key_debug(brw, "outputs written", old_key->outputs_written,
139 key->outputs_written);
140 found |= key_debug(brw, "patch outputs written", old_key->patch_outputs_written,
141 key->patch_outputs_written);
142 found |= key_debug(brw, "TES primitive mode", old_key->tes_primitive_mode,
143 key->tes_primitive_mode);
144 found |= key_debug(brw, "quads and equal_spacing workaround",
145 old_key->quads_workaround, key->quads_workaround);
146 found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
147
148 if (!found) {
149 perf_debug(" Something else\n");
150 }
151 }
152
153 static bool
154 brw_codegen_tcs_prog(struct brw_context *brw, struct brw_program *tcp,
155 struct brw_program *tep, struct brw_tcs_prog_key *key)
156 {
157 struct gl_context *ctx = &brw->ctx;
158 const struct brw_compiler *compiler = brw->screen->compiler;
159 const struct gen_device_info *devinfo = compiler->devinfo;
160 struct brw_stage_state *stage_state = &brw->tcs.base;
161 nir_shader *nir;
162 struct brw_tcs_prog_data prog_data;
163 bool start_busy = false;
164 double start_time = 0;
165
166 void *mem_ctx = ralloc_context(NULL);
167 if (tcp) {
168 nir = tcp->program.nir;
169 } else {
170 /* Create a dummy nir_shader. We won't actually use NIR code to
171 * generate assembly (it's easier to generate assembly directly),
172 * but the whole compiler assumes one of these exists.
173 */
174 const nir_shader_compiler_options *options =
175 ctx->Const.ShaderCompilerOptions[MESA_SHADER_TESS_CTRL].NirOptions;
176 nir = create_passthrough_tcs(mem_ctx, compiler, options, key);
177 }
178
179 memset(&prog_data, 0, sizeof(prog_data));
180
181 if (tcp) {
182 brw_assign_common_binding_table_offsets(devinfo, &tcp->program,
183 &prog_data.base.base, 0);
184
185 brw_nir_setup_glsl_uniforms(mem_ctx, nir, &tcp->program,
186 &prog_data.base.base,
187 compiler->scalar_stage[MESA_SHADER_TESS_CTRL]);
188 brw_nir_analyze_ubo_ranges(compiler, tcp->program.nir,
189 prog_data.base.base.ubo_ranges);
190 } else {
191 /* Upload the Patch URB Header as the first two uniforms.
192 * Do the annoying scrambling so the shader doesn't have to.
193 */
194 assert(nir->num_uniforms == 32);
195 prog_data.base.base.param = rzalloc_array(mem_ctx, uint32_t, 8);
196 prog_data.base.base.nr_params = 8;
197
198 uint32_t *param = prog_data.base.base.param;
199 for (int i = 0; i < 8; i++)
200 param[i] = BRW_PARAM_BUILTIN_ZERO;
201
202 if (key->tes_primitive_mode == GL_QUADS) {
203 for (int i = 0; i < 4; i++)
204 param[7 - i] = BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_X + i;
205
206 param[3] = BRW_PARAM_BUILTIN_TESS_LEVEL_INNER_X;
207 param[2] = BRW_PARAM_BUILTIN_TESS_LEVEL_INNER_Y;
208 } else if (key->tes_primitive_mode == GL_TRIANGLES) {
209 for (int i = 0; i < 3; i++)
210 param[7 - i] = BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_X + i;
211
212 param[4] = BRW_PARAM_BUILTIN_TESS_LEVEL_INNER_X;
213 } else {
214 assert(key->tes_primitive_mode == GL_ISOLINES);
215 param[7] = BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_Y;
216 param[6] = BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_X;
217 }
218 }
219
220 int st_index = -1;
221 if (unlikely((INTEL_DEBUG & DEBUG_SHADER_TIME) && tep))
222 st_index = brw_get_shader_time_index(brw, &tep->program, ST_TCS, true);
223
224 if (unlikely(brw->perf_debug)) {
225 start_busy = brw->batch.last_bo && brw_bo_busy(brw->batch.last_bo);
226 start_time = get_time();
227 }
228
229 char *error_str;
230 const unsigned *program =
231 brw_compile_tcs(compiler, brw, mem_ctx, key, &prog_data, nir, st_index,
232 &error_str);
233 if (program == NULL) {
234 if (tep) {
235 tep->program.sh.data->LinkStatus = LINKING_FAILURE;
236 ralloc_strcat(&tep->program.sh.data->InfoLog, error_str);
237 }
238
239 _mesa_problem(NULL, "Failed to compile tessellation control shader: "
240 "%s\n", error_str);
241
242 ralloc_free(mem_ctx);
243 return false;
244 }
245
246 if (unlikely(brw->perf_debug)) {
247 if (tcp) {
248 if (tcp->compiled_once) {
249 brw_tcs_debug_recompile(brw, &tcp->program, key);
250 }
251 tcp->compiled_once = true;
252 }
253
254 if (start_busy && !brw_bo_busy(brw->batch.last_bo)) {
255 perf_debug("TCS compile took %.03f ms and stalled the GPU\n",
256 (get_time() - start_time) * 1000);
257 }
258 }
259
260 /* Scratch space is used for register spilling */
261 brw_alloc_stage_scratch(brw, stage_state,
262 prog_data.base.base.total_scratch);
263
264 /* The param and pull_param arrays will be freed by the shader cache. */
265 ralloc_steal(NULL, prog_data.base.base.param);
266 ralloc_steal(NULL, prog_data.base.base.pull_param);
267 brw_upload_cache(&brw->cache, BRW_CACHE_TCS_PROG,
268 key, sizeof(*key),
269 program, prog_data.base.base.program_size,
270 &prog_data, sizeof(prog_data),
271 &stage_state->prog_offset, &brw->tcs.base.prog_data);
272 ralloc_free(mem_ctx);
273
274 return true;
275 }
276
277 void
278 brw_tcs_populate_key(struct brw_context *brw,
279 struct brw_tcs_prog_key *key)
280 {
281 const struct gen_device_info *devinfo = &brw->screen->devinfo;
282 struct brw_program *tcp =
283 (struct brw_program *) brw->programs[MESA_SHADER_TESS_CTRL];
284 struct brw_program *tep =
285 (struct brw_program *) brw->programs[MESA_SHADER_TESS_EVAL];
286 struct gl_program *tes_prog = &tep->program;
287
288 uint64_t per_vertex_slots = tes_prog->info.inputs_read;
289 uint32_t per_patch_slots = tes_prog->info.patch_inputs_read;
290
291 memset(key, 0, sizeof(*key));
292
293 if (tcp) {
294 struct gl_program *prog = &tcp->program;
295 per_vertex_slots |= prog->info.outputs_written;
296 per_patch_slots |= prog->info.patch_outputs_written;
297 }
298
299 if (devinfo->gen < 8 || !tcp)
300 key->input_vertices = brw->ctx.TessCtrlProgram.patch_vertices;
301 key->outputs_written = per_vertex_slots;
302 key->patch_outputs_written = per_patch_slots;
303
304 /* We need to specialize our code generation for tessellation levels
305 * based on the domain the DS is expecting to tessellate.
306 */
307 key->tes_primitive_mode = tep->program.info.tess.primitive_mode;
308 key->quads_workaround = devinfo->gen < 9 &&
309 tep->program.info.tess.primitive_mode == GL_QUADS &&
310 tep->program.info.tess.spacing == TESS_SPACING_EQUAL;
311
312 if (tcp) {
313 key->program_string_id = tcp->id;
314
315 /* _NEW_TEXTURE */
316 brw_populate_sampler_prog_key_data(&brw->ctx, &tcp->program, &key->tex);
317 }
318 }
319
320 void
321 brw_upload_tcs_prog(struct brw_context *brw)
322 {
323 struct brw_stage_state *stage_state = &brw->tcs.base;
324 struct brw_tcs_prog_key key;
325 /* BRW_NEW_TESS_PROGRAMS */
326 struct brw_program *tcp =
327 (struct brw_program *) brw->programs[MESA_SHADER_TESS_CTRL];
328 MAYBE_UNUSED struct brw_program *tep =
329 (struct brw_program *) brw->programs[MESA_SHADER_TESS_EVAL];
330 assert(tep);
331
332 if (!brw_state_dirty(brw,
333 _NEW_TEXTURE,
334 BRW_NEW_PATCH_PRIMITIVE |
335 BRW_NEW_TESS_PROGRAMS))
336 return;
337
338 brw_tcs_populate_key(brw, &key);
339
340 if (brw_search_cache(&brw->cache, BRW_CACHE_TCS_PROG,
341 &key, sizeof(key),
342 &stage_state->prog_offset,
343 &brw->tcs.base.prog_data))
344 return;
345
346 if (brw_disk_cache_upload_program(brw, MESA_SHADER_TESS_CTRL))
347 return;
348
349 tcp = (struct brw_program *) brw->programs[MESA_SHADER_TESS_CTRL];
350 if (tcp)
351 tcp->id = key.program_string_id;
352
353 MAYBE_UNUSED bool success = brw_codegen_tcs_prog(brw, tcp, tep, &key);
354 assert(success);
355 }
356
357
358 bool
359 brw_tcs_precompile(struct gl_context *ctx,
360 struct gl_shader_program *shader_prog,
361 struct gl_program *prog)
362 {
363 struct brw_context *brw = brw_context(ctx);
364 struct brw_tcs_prog_key key;
365 uint32_t old_prog_offset = brw->tcs.base.prog_offset;
366 struct brw_stage_prog_data *old_prog_data = brw->tcs.base.prog_data;
367 bool success;
368
369 struct brw_program *btcp = brw_program(prog);
370 const struct gl_linked_shader *tes =
371 shader_prog->_LinkedShaders[MESA_SHADER_TESS_EVAL];
372 const struct gen_device_info *devinfo = &brw->screen->devinfo;
373
374 memset(&key, 0, sizeof(key));
375
376 key.program_string_id = btcp->id;
377 brw_setup_tex_for_precompile(brw, &key.tex, prog);
378
379 /* Guess that the input and output patches have the same dimensionality. */
380 if (devinfo->gen < 8)
381 key.input_vertices = prog->info.tess.tcs_vertices_out;
382
383 struct brw_program *btep;
384 if (tes) {
385 btep = brw_program(tes->Program);
386 key.tes_primitive_mode = tes->Program->info.tess.primitive_mode;
387 key.quads_workaround = devinfo->gen < 9 &&
388 tes->Program->info.tess.primitive_mode == GL_QUADS &&
389 tes->Program->info.tess.spacing == TESS_SPACING_EQUAL;
390 } else {
391 btep = NULL;
392 key.tes_primitive_mode = GL_TRIANGLES;
393 }
394
395 key.outputs_written = prog->nir->info.outputs_written;
396 key.patch_outputs_written = prog->nir->info.patch_outputs_written;
397
398 success = brw_codegen_tcs_prog(brw, btcp, btep, &key);
399
400 brw->tcs.base.prog_offset = old_prog_offset;
401 brw->tcs.base.prog_data = old_prog_data;
402
403 return success;
404 }