i965/drm: Use our internal libdrm (drm_bacon) rather than the real one.
[mesa.git] / src / mesa / drivers / dri / i965 / brw_vs.c
1 /*
2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics to
4 develop this 3D driver.
5
6 Permission is hereby granted, free of charge, to any person obtaining
7 a 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, sublicense, 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
16 portions of the Software.
17
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25
26 **********************************************************************/
27 /*
28 * Authors:
29 * Keith Whitwell <keithw@vmware.com>
30 */
31
32
33 #include "main/compiler.h"
34 #include "main/context.h"
35 #include "brw_context.h"
36 #include "brw_vs.h"
37 #include "brw_util.h"
38 #include "brw_state.h"
39 #include "program/prog_print.h"
40 #include "program/prog_parameter.h"
41 #include "compiler/brw_nir.h"
42 #include "brw_program.h"
43
44 #include "util/ralloc.h"
45
46 /**
47 * Decide which set of clip planes should be used when clipping via
48 * gl_Position or gl_ClipVertex.
49 */
50 gl_clip_plane *
51 brw_select_clip_planes(struct gl_context *ctx)
52 {
53 if (ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX]) {
54 /* There is currently a GLSL vertex shader, so clip according to GLSL
55 * rules, which means compare gl_ClipVertex (or gl_Position, if
56 * gl_ClipVertex wasn't assigned) against the eye-coordinate clip planes
57 * that were stored in EyeUserPlane at the time the clip planes were
58 * specified.
59 */
60 return ctx->Transform.EyeUserPlane;
61 } else {
62 /* Either we are using fixed function or an ARB vertex program. In
63 * either case the clip planes are going to be compared against
64 * gl_Position (which is in clip coordinates) so we have to clip using
65 * _ClipUserPlane, which was transformed into clip coordinates by Mesa
66 * core.
67 */
68 return ctx->Transform._ClipUserPlane;
69 }
70 }
71
72 GLbitfield64
73 brw_vs_outputs_written(struct brw_context *brw, struct brw_vs_prog_key *key,
74 GLbitfield64 user_varyings)
75 {
76 GLbitfield64 outputs_written = user_varyings;
77
78 if (key->copy_edgeflag) {
79 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_EDGE);
80 }
81
82 if (brw->gen < 6) {
83 /* Put dummy slots into the VUE for the SF to put the replaced
84 * point sprite coords in. We shouldn't need these dummy slots,
85 * which take up precious URB space, but it would mean that the SF
86 * doesn't get nice aligned pairs of input coords into output
87 * coords, which would be a pain to handle.
88 */
89 for (unsigned i = 0; i < 8; i++) {
90 if (key->point_coord_replace & (1 << i))
91 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_TEX0 + i);
92 }
93
94 /* if back colors are written, allocate slots for front colors too */
95 if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC0))
96 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL0);
97 if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC1))
98 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL1);
99 }
100
101 /* In order for legacy clipping to work, we need to populate the clip
102 * distance varying slots whenever clipping is enabled, even if the vertex
103 * shader doesn't write to gl_ClipDistance.
104 */
105 if (key->nr_userclip_plane_consts > 0) {
106 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0);
107 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1);
108 }
109
110 return outputs_written;
111 }
112
113 static void
114 brw_vs_debug_recompile(struct brw_context *brw, struct gl_program *prog,
115 const struct brw_vs_prog_key *key)
116 {
117 perf_debug("Recompiling vertex shader for program %d\n", prog->Id);
118
119 bool found = false;
120 const struct brw_vs_prog_key *old_key =
121 brw_find_previous_compile(&brw->cache, BRW_CACHE_VS_PROG,
122 key->program_string_id);
123
124 if (!old_key) {
125 perf_debug(" Didn't find previous compile in the shader cache for "
126 "debug\n");
127 return;
128 }
129
130 for (unsigned int i = 0; i < VERT_ATTRIB_MAX; i++) {
131 found |= key_debug(brw, "Vertex attrib w/a flags",
132 old_key->gl_attrib_wa_flags[i],
133 key->gl_attrib_wa_flags[i]);
134 }
135
136 found |= key_debug(brw, "legacy user clipping",
137 old_key->nr_userclip_plane_consts,
138 key->nr_userclip_plane_consts);
139
140 found |= key_debug(brw, "copy edgeflag",
141 old_key->copy_edgeflag, key->copy_edgeflag);
142 found |= key_debug(brw, "PointCoord replace",
143 old_key->point_coord_replace, key->point_coord_replace);
144 found |= key_debug(brw, "vertex color clamping",
145 old_key->clamp_vertex_color, key->clamp_vertex_color);
146
147 found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
148
149 if (!found) {
150 perf_debug(" Something else\n");
151 }
152 }
153
154 static bool
155 brw_codegen_vs_prog(struct brw_context *brw,
156 struct brw_program *vp,
157 struct brw_vs_prog_key *key)
158 {
159 const struct brw_compiler *compiler = brw->screen->compiler;
160 const struct gen_device_info *devinfo = &brw->screen->devinfo;
161 GLuint program_size;
162 const GLuint *program;
163 struct brw_vs_prog_data prog_data;
164 struct brw_stage_prog_data *stage_prog_data = &prog_data.base.base;
165 void *mem_ctx;
166 bool start_busy = false;
167 double start_time = 0;
168
169 memset(&prog_data, 0, sizeof(prog_data));
170
171 /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
172 if (vp->program.is_arb_asm)
173 stage_prog_data->use_alt_mode = true;
174
175 mem_ctx = ralloc_context(NULL);
176
177 brw_assign_common_binding_table_offsets(devinfo, &vp->program,
178 &prog_data.base.base, 0);
179
180 /* Allocate the references to the uniforms that will end up in the
181 * prog_data associated with the compiled program, and which will be freed
182 * by the state cache.
183 */
184 int param_count = vp->program.nir->num_uniforms / 4;
185
186 prog_data.base.base.nr_image_params = vp->program.info.num_images;
187
188 /* vec4_visitor::setup_uniform_clipplane_values() also uploads user clip
189 * planes as uniforms.
190 */
191 param_count += key->nr_userclip_plane_consts * 4;
192
193 stage_prog_data->param =
194 rzalloc_array(NULL, const gl_constant_value *, param_count);
195 stage_prog_data->pull_param =
196 rzalloc_array(NULL, const gl_constant_value *, param_count);
197 stage_prog_data->image_param =
198 rzalloc_array(NULL, struct brw_image_param,
199 stage_prog_data->nr_image_params);
200 stage_prog_data->nr_params = param_count;
201
202 if (!vp->program.is_arb_asm) {
203 brw_nir_setup_glsl_uniforms(vp->program.nir, &vp->program,
204 &prog_data.base.base,
205 compiler->scalar_stage[MESA_SHADER_VERTEX]);
206 } else {
207 brw_nir_setup_arb_uniforms(vp->program.nir, &vp->program,
208 &prog_data.base.base);
209 }
210
211 uint64_t outputs_written =
212 brw_vs_outputs_written(brw, key, vp->program.info.outputs_written);
213 prog_data.inputs_read = vp->program.info.inputs_read;
214 prog_data.double_inputs_read = vp->program.info.double_inputs_read;
215
216 if (key->copy_edgeflag) {
217 prog_data.inputs_read |= VERT_BIT_EDGEFLAG;
218 }
219
220 brw_compute_vue_map(devinfo,
221 &prog_data.base.vue_map, outputs_written,
222 vp->program.nir->info->separate_shader);
223
224 if (0) {
225 _mesa_fprint_program_opt(stderr, &vp->program, PROG_PRINT_DEBUG, true);
226 }
227
228 if (unlikely(brw->perf_debug)) {
229 start_busy = (brw->batch.last_bo &&
230 drm_bacon_bo_busy(brw->batch.last_bo));
231 start_time = get_time();
232 }
233
234 if (unlikely(INTEL_DEBUG & DEBUG_VS)) {
235 if (vp->program.is_arb_asm)
236 brw_dump_arb_asm("vertex", &vp->program);
237 }
238
239 int st_index = -1;
240 if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
241 st_index = brw_get_shader_time_index(brw, &vp->program, ST_VS,
242 !vp->program.is_arb_asm);
243 }
244
245 /* Emit GEN4 code.
246 */
247 char *error_str;
248 program = brw_compile_vs(compiler, brw, mem_ctx, key, &prog_data,
249 vp->program.nir,
250 brw_select_clip_planes(&brw->ctx),
251 !_mesa_is_gles3(&brw->ctx),
252 st_index, &program_size, &error_str);
253 if (program == NULL) {
254 if (!vp->program.is_arb_asm) {
255 vp->program.sh.data->LinkStatus = linking_failure;
256 ralloc_strcat(&vp->program.sh.data->InfoLog, error_str);
257 }
258
259 _mesa_problem(NULL, "Failed to compile vertex shader: %s\n", error_str);
260
261 ralloc_free(mem_ctx);
262 return false;
263 }
264
265 if (unlikely(brw->perf_debug)) {
266 if (vp->compiled_once) {
267 brw_vs_debug_recompile(brw, &vp->program, key);
268 }
269 if (start_busy && !drm_bacon_bo_busy(brw->batch.last_bo)) {
270 perf_debug("VS compile took %.03f ms and stalled the GPU\n",
271 (get_time() - start_time) * 1000);
272 }
273 vp->compiled_once = true;
274 }
275
276 /* Scratch space is used for register spilling */
277 brw_alloc_stage_scratch(brw, &brw->vs.base,
278 prog_data.base.base.total_scratch,
279 devinfo->max_vs_threads);
280
281 brw_upload_cache(&brw->cache, BRW_CACHE_VS_PROG,
282 key, sizeof(struct brw_vs_prog_key),
283 program, program_size,
284 &prog_data, sizeof(prog_data),
285 &brw->vs.base.prog_offset, &brw->vs.base.prog_data);
286 ralloc_free(mem_ctx);
287
288 return true;
289 }
290
291 static bool
292 brw_vs_state_dirty(const struct brw_context *brw)
293 {
294 return brw_state_dirty(brw,
295 _NEW_BUFFERS |
296 _NEW_LIGHT |
297 _NEW_POINT |
298 _NEW_POLYGON |
299 _NEW_TEXTURE |
300 _NEW_TRANSFORM,
301 BRW_NEW_VERTEX_PROGRAM |
302 BRW_NEW_VS_ATTRIB_WORKAROUNDS);
303 }
304
305 void
306 brw_vs_populate_key(struct brw_context *brw,
307 struct brw_vs_prog_key *key)
308 {
309 struct gl_context *ctx = &brw->ctx;
310 /* BRW_NEW_VERTEX_PROGRAM */
311 struct brw_program *vp = (struct brw_program *)brw->vertex_program;
312 struct gl_program *prog = (struct gl_program *) brw->vertex_program;
313
314 memset(key, 0, sizeof(*key));
315
316 /* Just upload the program verbatim for now. Always send it all
317 * the inputs it asks for, whether they are varying or not.
318 */
319 key->program_string_id = vp->id;
320
321 if (ctx->Transform.ClipPlanesEnabled != 0 &&
322 (ctx->API == API_OPENGL_COMPAT || ctx->API == API_OPENGLES) &&
323 vp->program.info.clip_distance_array_size == 0) {
324 key->nr_userclip_plane_consts =
325 _mesa_logbase2(ctx->Transform.ClipPlanesEnabled) + 1;
326 }
327
328 if (brw->gen < 6) {
329 /* _NEW_POLYGON */
330 key->copy_edgeflag = (ctx->Polygon.FrontMode != GL_FILL ||
331 ctx->Polygon.BackMode != GL_FILL);
332
333 /* _NEW_POINT */
334 if (ctx->Point.PointSprite) {
335 key->point_coord_replace = ctx->Point.CoordReplace & 0xff;
336 }
337 }
338
339 if (prog->info.outputs_written &
340 (VARYING_BIT_COL0 | VARYING_BIT_COL1 | VARYING_BIT_BFC0 |
341 VARYING_BIT_BFC1)) {
342 /* _NEW_LIGHT | _NEW_BUFFERS */
343 key->clamp_vertex_color = ctx->Light._ClampVertexColor;
344 }
345
346 /* _NEW_TEXTURE */
347 brw_populate_sampler_prog_key_data(ctx, prog, &key->tex);
348
349 /* BRW_NEW_VS_ATTRIB_WORKAROUNDS */
350 if (brw->gen < 8 && !brw->is_haswell) {
351 memcpy(key->gl_attrib_wa_flags, brw->vb.attrib_wa_flags,
352 sizeof(brw->vb.attrib_wa_flags));
353 }
354 }
355
356 void
357 brw_upload_vs_prog(struct brw_context *brw)
358 {
359 struct brw_vs_prog_key key;
360 /* BRW_NEW_VERTEX_PROGRAM */
361 struct brw_program *vp = (struct brw_program *)brw->vertex_program;
362
363 if (!brw_vs_state_dirty(brw))
364 return;
365
366 brw_vs_populate_key(brw, &key);
367
368 if (!brw_search_cache(&brw->cache, BRW_CACHE_VS_PROG,
369 &key, sizeof(key),
370 &brw->vs.base.prog_offset, &brw->vs.base.prog_data)) {
371 bool success = brw_codegen_vs_prog(brw, vp, &key);
372 (void) success;
373 assert(success);
374 }
375 }
376
377 bool
378 brw_vs_precompile(struct gl_context *ctx, struct gl_program *prog)
379 {
380 struct brw_context *brw = brw_context(ctx);
381 struct brw_vs_prog_key key;
382 uint32_t old_prog_offset = brw->vs.base.prog_offset;
383 struct brw_stage_prog_data *old_prog_data = brw->vs.base.prog_data;
384 bool success;
385
386 struct brw_program *bvp = brw_program(prog);
387
388 memset(&key, 0, sizeof(key));
389
390 brw_setup_tex_for_precompile(brw, &key.tex, prog);
391 key.program_string_id = bvp->id;
392 key.clamp_vertex_color =
393 (prog->info.outputs_written &
394 (VARYING_BIT_COL0 | VARYING_BIT_COL1 | VARYING_BIT_BFC0 |
395 VARYING_BIT_BFC1));
396
397 success = brw_codegen_vs_prog(brw, bvp, &key);
398
399 brw->vs.base.prog_offset = old_prog_offset;
400 brw->vs.base.prog_data = old_prog_data;
401
402 return success;
403 }