i965: Store image_param in brw_context instead of prog_data
[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 /* Allocate the references to the uniforms that will end up in the
182 * prog_data associated with the compiled program, and which will be freed
183 * by the state cache.
184 *
185 * Note: param_count needs to be num_uniform_components * 4, since we add
186 * padding around uniform values below vec4 size, so the worst case is that
187 * every uniform is a float which gets padded to the size of a vec4.
188 */
189 int param_count = nir->num_uniforms / 4;
190
191 prog_data.base.base.param = rzalloc_array(NULL, uint32_t, param_count);
192 prog_data.base.base.pull_param = rzalloc_array(NULL, uint32_t, param_count);
193 prog_data.base.base.nr_params = param_count;
194
195 if (tcp) {
196 brw_assign_common_binding_table_offsets(devinfo, &tcp->program,
197 &prog_data.base.base, 0);
198
199 brw_nir_setup_glsl_uniforms(nir, &tcp->program, &prog_data.base.base,
200 compiler->scalar_stage[MESA_SHADER_TESS_CTRL]);
201 brw_nir_analyze_ubo_ranges(compiler, tcp->program.nir,
202 prog_data.base.base.ubo_ranges);
203 } else {
204 /* Upload the Patch URB Header as the first two uniforms.
205 * Do the annoying scrambling so the shader doesn't have to.
206 */
207 uint32_t *param = prog_data.base.base.param;
208 for (int i = 0; i < 8; i++)
209 param[i] = BRW_PARAM_BUILTIN_ZERO;
210
211 if (key->tes_primitive_mode == GL_QUADS) {
212 for (int i = 0; i < 4; i++)
213 param[7 - i] = BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_X + i;
214
215 param[3] = BRW_PARAM_BUILTIN_TESS_LEVEL_INNER_X;
216 param[2] = BRW_PARAM_BUILTIN_TESS_LEVEL_INNER_Y;
217 } else if (key->tes_primitive_mode == GL_TRIANGLES) {
218 for (int i = 0; i < 3; i++)
219 param[7 - i] = BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_X + i;
220
221 param[4] = BRW_PARAM_BUILTIN_TESS_LEVEL_INNER_X;
222 } else {
223 assert(key->tes_primitive_mode == GL_ISOLINES);
224 param[7] = BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_Y;
225 param[6] = BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_X;
226 }
227 }
228
229 int st_index = -1;
230 if (unlikely((INTEL_DEBUG & DEBUG_SHADER_TIME) && tep))
231 st_index = brw_get_shader_time_index(brw, &tep->program, ST_TCS, true);
232
233 if (unlikely(brw->perf_debug)) {
234 start_busy = brw->batch.last_bo && brw_bo_busy(brw->batch.last_bo);
235 start_time = get_time();
236 }
237
238 unsigned program_size;
239 char *error_str;
240 const unsigned *program =
241 brw_compile_tcs(compiler, brw, mem_ctx, key, &prog_data, nir, st_index,
242 &program_size, &error_str);
243 if (program == NULL) {
244 if (tep) {
245 tep->program.sh.data->LinkStatus = linking_failure;
246 ralloc_strcat(&tep->program.sh.data->InfoLog, error_str);
247 }
248
249 _mesa_problem(NULL, "Failed to compile tessellation control shader: "
250 "%s\n", error_str);
251
252 ralloc_free(mem_ctx);
253 return false;
254 }
255
256 if (unlikely(brw->perf_debug)) {
257 if (tcp) {
258 if (tcp->compiled_once) {
259 brw_tcs_debug_recompile(brw, &tcp->program, key);
260 }
261 tcp->compiled_once = true;
262 }
263
264 if (start_busy && !brw_bo_busy(brw->batch.last_bo)) {
265 perf_debug("TCS compile took %.03f ms and stalled the GPU\n",
266 (get_time() - start_time) * 1000);
267 }
268 }
269
270 /* Scratch space is used for register spilling */
271 brw_alloc_stage_scratch(brw, stage_state,
272 prog_data.base.base.total_scratch,
273 devinfo->max_tcs_threads);
274
275 brw_upload_cache(&brw->cache, BRW_CACHE_TCS_PROG,
276 key, sizeof(*key),
277 program, program_size,
278 &prog_data, sizeof(prog_data),
279 &stage_state->prog_offset, &brw->tcs.base.prog_data);
280 ralloc_free(mem_ctx);
281
282 return true;
283 }
284
285 void
286 brw_tcs_populate_key(struct brw_context *brw,
287 struct brw_tcs_prog_key *key)
288 {
289 const struct gen_device_info *devinfo = &brw->screen->devinfo;
290 struct brw_program *tcp =
291 (struct brw_program *) brw->programs[MESA_SHADER_TESS_CTRL];
292 struct brw_program *tep =
293 (struct brw_program *) brw->programs[MESA_SHADER_TESS_EVAL];
294 struct gl_program *tes_prog = &tep->program;
295
296 uint64_t per_vertex_slots = tes_prog->info.inputs_read;
297 uint32_t per_patch_slots = tes_prog->info.patch_inputs_read;
298
299 memset(key, 0, sizeof(*key));
300
301 if (tcp) {
302 struct gl_program *prog = &tcp->program;
303 per_vertex_slots |= prog->info.outputs_written;
304 per_patch_slots |= prog->info.patch_outputs_written;
305 }
306
307 if (devinfo->gen < 8 || !tcp)
308 key->input_vertices = brw->ctx.TessCtrlProgram.patch_vertices;
309 key->outputs_written = per_vertex_slots;
310 key->patch_outputs_written = per_patch_slots;
311
312 /* We need to specialize our code generation for tessellation levels
313 * based on the domain the DS is expecting to tessellate.
314 */
315 key->tes_primitive_mode = tep->program.info.tess.primitive_mode;
316 key->quads_workaround = devinfo->gen < 9 &&
317 tep->program.info.tess.primitive_mode == GL_QUADS &&
318 tep->program.info.tess.spacing == TESS_SPACING_EQUAL;
319
320 if (tcp) {
321 key->program_string_id = tcp->id;
322
323 /* _NEW_TEXTURE */
324 brw_populate_sampler_prog_key_data(&brw->ctx, &tcp->program, &key->tex);
325 }
326 }
327
328 void
329 brw_upload_tcs_prog(struct brw_context *brw)
330 {
331 struct brw_stage_state *stage_state = &brw->tcs.base;
332 struct brw_tcs_prog_key key;
333 /* BRW_NEW_TESS_PROGRAMS */
334 struct brw_program *tcp =
335 (struct brw_program *) brw->programs[MESA_SHADER_TESS_CTRL];
336 MAYBE_UNUSED struct brw_program *tep =
337 (struct brw_program *) brw->programs[MESA_SHADER_TESS_EVAL];
338 assert(tep);
339
340 if (!brw_state_dirty(brw,
341 _NEW_TEXTURE,
342 BRW_NEW_PATCH_PRIMITIVE |
343 BRW_NEW_TESS_PROGRAMS))
344 return;
345
346 brw_tcs_populate_key(brw, &key);
347
348 if (!brw_search_cache(&brw->cache, BRW_CACHE_TCS_PROG,
349 &key, sizeof(key),
350 &stage_state->prog_offset,
351 &brw->tcs.base.prog_data)) {
352 bool success = brw_codegen_tcs_prog(brw, tcp, tep, &key);
353 assert(success);
354 (void)success;
355 }
356 }
357
358
359 bool
360 brw_tcs_precompile(struct gl_context *ctx,
361 struct gl_shader_program *shader_prog,
362 struct gl_program *prog)
363 {
364 struct brw_context *brw = brw_context(ctx);
365 struct brw_tcs_prog_key key;
366 uint32_t old_prog_offset = brw->tcs.base.prog_offset;
367 struct brw_stage_prog_data *old_prog_data = brw->tcs.base.prog_data;
368 bool success;
369
370 struct brw_program *btcp = brw_program(prog);
371 const struct gl_linked_shader *tes =
372 shader_prog->_LinkedShaders[MESA_SHADER_TESS_EVAL];
373 const struct gen_device_info *devinfo = &brw->screen->devinfo;
374
375 memset(&key, 0, sizeof(key));
376
377 key.program_string_id = btcp->id;
378 brw_setup_tex_for_precompile(brw, &key.tex, prog);
379
380 /* Guess that the input and output patches have the same dimensionality. */
381 if (devinfo->gen < 8)
382 key.input_vertices = prog->info.tess.tcs_vertices_out;
383
384 struct brw_program *btep;
385 if (tes) {
386 btep = brw_program(tes->Program);
387 key.tes_primitive_mode = tes->Program->info.tess.primitive_mode;
388 key.quads_workaround = devinfo->gen < 9 &&
389 tes->Program->info.tess.primitive_mode == GL_QUADS &&
390 tes->Program->info.tess.spacing == TESS_SPACING_EQUAL;
391 } else {
392 btep = NULL;
393 key.tes_primitive_mode = GL_TRIANGLES;
394 }
395
396 key.outputs_written = prog->nir->info.outputs_written;
397 key.patch_outputs_written = prog->nir->info.patch_outputs_written;
398
399 success = brw_codegen_tcs_prog(brw, btcp, btep, &key);
400
401 brw->tcs.base.prog_offset = old_prog_offset;
402 brw->tcs.base.prog_data = old_prog_data;
403
404 return success;
405 }