2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
4 develop this 3D driver.
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:
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.
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.
26 **********************************************************************/
29 * Keith Whitwell <keith@tungstengraphics.com>
33 #include "main/compiler.h"
34 #include "brw_context.h"
37 #include "brw_state.h"
38 #include "program/prog_print.h"
39 #include "program/prog_parameter.h"
41 #include "glsl/ralloc.h"
43 static inline void assign_vue_slot(struct brw_vue_map *vue_map,
46 /* Make sure this vert_result hasn't been assigned a slot already */
47 assert (vue_map->vert_result_to_slot[vert_result] == -1);
49 vue_map->vert_result_to_slot[vert_result] = vue_map->num_slots;
50 vue_map->slot_to_vert_result[vue_map->num_slots++] = vert_result;
54 * Compute the VUE map for vertex shader program.
57 brw_compute_vue_map(struct brw_vue_map *vue_map,
58 const struct intel_context *intel,
60 GLbitfield64 outputs_written)
64 vue_map->num_slots = 0;
65 for (i = 0; i < BRW_VERT_RESULT_MAX; ++i) {
66 vue_map->vert_result_to_slot[i] = -1;
67 vue_map->slot_to_vert_result[i] = BRW_VERT_RESULT_MAX;
70 /* VUE header: format depends on chip generation and whether clipping is
75 /* There are 8 dwords in VUE header pre-Ironlake:
76 * dword 0-3 is indices, point width, clip flags.
77 * dword 4-7 is ndc position
78 * dword 8-11 is the first vertex data.
80 assign_vue_slot(vue_map, VERT_RESULT_PSIZ);
81 assign_vue_slot(vue_map, BRW_VERT_RESULT_NDC);
82 assign_vue_slot(vue_map, VERT_RESULT_HPOS);
85 /* There are 20 DWs (D0-D19) in VUE header on Ironlake:
86 * dword 0-3 of the header is indices, point width, clip flags.
87 * dword 4-7 is the ndc position
88 * dword 8-11 of the vertex header is the 4D space position
89 * dword 12-19 of the vertex header is the user clip distance.
90 * dword 20-23 is a pad so that the vertex element data is aligned
91 * dword 24-27 is the first vertex data we fill.
93 * Note: future pipeline stages expect 4D space position to be
94 * contiguous with the other vert_results, so we make dword 24-27 a
95 * duplicate copy of the 4D space position.
97 assign_vue_slot(vue_map, VERT_RESULT_PSIZ);
98 assign_vue_slot(vue_map, BRW_VERT_RESULT_NDC);
99 assign_vue_slot(vue_map, BRW_VERT_RESULT_HPOS_DUPLICATE);
100 assign_vue_slot(vue_map, VERT_RESULT_CLIP_DIST0);
101 assign_vue_slot(vue_map, VERT_RESULT_CLIP_DIST1);
102 assign_vue_slot(vue_map, BRW_VERT_RESULT_PAD);
103 assign_vue_slot(vue_map, VERT_RESULT_HPOS);
107 /* There are 8 or 16 DWs (D0-D15) in VUE header on Sandybridge:
108 * dword 0-3 of the header is indices, point width, clip flags.
109 * dword 4-7 is the 4D space position
110 * dword 8-15 of the vertex header is the user clip distance if
112 * dword 8-11 or 16-19 is the first vertex element data we fill.
114 assign_vue_slot(vue_map, VERT_RESULT_PSIZ);
115 assign_vue_slot(vue_map, VERT_RESULT_HPOS);
116 if (userclip_active) {
117 assign_vue_slot(vue_map, VERT_RESULT_CLIP_DIST0);
118 assign_vue_slot(vue_map, VERT_RESULT_CLIP_DIST1);
120 /* front and back colors need to be consecutive so that we can use
121 * ATTRIBUTE_SWIZZLE_INPUTATTR_FACING to swizzle them when doing
124 if (outputs_written & BITFIELD64_BIT(VERT_RESULT_COL0))
125 assign_vue_slot(vue_map, VERT_RESULT_COL0);
126 if (outputs_written & BITFIELD64_BIT(VERT_RESULT_BFC0))
127 assign_vue_slot(vue_map, VERT_RESULT_BFC0);
128 if (outputs_written & BITFIELD64_BIT(VERT_RESULT_COL1))
129 assign_vue_slot(vue_map, VERT_RESULT_COL1);
130 if (outputs_written & BITFIELD64_BIT(VERT_RESULT_BFC1))
131 assign_vue_slot(vue_map, VERT_RESULT_BFC1);
134 assert (!"VUE map not known for this chip generation");
138 /* The hardware doesn't care about the rest of the vertex outputs, so just
139 * assign them contiguously. Don't reassign outputs that already have a
142 * Also, don't assign a slot for VERT_RESULT_CLIP_VERTEX, since it is
143 * unsupported in pre-GEN6, and in GEN6+ the vertex shader converts it into
146 for (int i = 0; i < VERT_RESULT_MAX; ++i) {
147 if ((outputs_written & BITFIELD64_BIT(i)) &&
148 vue_map->vert_result_to_slot[i] == -1 &&
149 i != VERT_RESULT_CLIP_VERTEX) {
150 assign_vue_slot(vue_map, i);
157 * Decide which set of clip planes should be used when clipping via
158 * gl_Position or gl_ClipVertex.
160 gl_clip_plane *brw_select_clip_planes(struct gl_context *ctx)
162 if (ctx->Shader.CurrentVertexProgram) {
163 /* There is currently a GLSL vertex shader, so clip according to GLSL
164 * rules, which means compare gl_ClipVertex (or gl_Position, if
165 * gl_ClipVertex wasn't assigned) against the eye-coordinate clip planes
166 * that were stored in EyeUserPlane at the time the clip planes were
169 return ctx->Transform.EyeUserPlane;
171 /* Either we are using fixed function or an ARB vertex program. In
172 * either case the clip planes are going to be compared against
173 * gl_Position (which is in clip coordinates) so we have to clip using
174 * _ClipUserPlane, which was transformed into clip coordinates by Mesa
177 return ctx->Transform._ClipUserPlane;
183 do_vs_prog(struct brw_context *brw,
184 struct gl_shader_program *prog,
185 struct brw_vertex_program *vp,
186 struct brw_vs_prog_key *key)
188 struct gl_context *ctx = &brw->intel.ctx;
189 struct intel_context *intel = &brw->intel;
191 const GLuint *program;
192 struct brw_vs_compile c;
197 memset(&c, 0, sizeof(c));
198 memcpy(&c.key, key, sizeof(*key));
200 mem_ctx = ralloc_context(NULL);
202 brw_init_compile(brw, &c.func, mem_ctx);
205 c.prog_data.outputs_written = vp->program.Base.OutputsWritten;
206 c.prog_data.inputs_read = vp->program.Base.InputsRead;
208 if (c.key.copy_edgeflag) {
209 c.prog_data.outputs_written |= BITFIELD64_BIT(VERT_RESULT_EDGE);
210 c.prog_data.inputs_read |= 1<<VERT_ATTRIB_EDGEFLAG;
213 /* Put dummy slots into the VUE for the SF to put the replaced
214 * point sprite coords in. We shouldn't need these dummy slots,
215 * which take up precious URB space, but it would mean that the SF
216 * doesn't get nice aligned pairs of input coords into output
217 * coords, which would be a pain to handle.
219 for (i = 0; i < 8; i++) {
220 if (c.key.point_coord_replace & (1 << i))
221 c.prog_data.outputs_written |= BITFIELD64_BIT(VERT_RESULT_TEX0 + i);
225 _mesa_fprint_program_opt(stdout, &c.vp->program.Base, PROG_PRINT_DEBUG,
231 if (brw->new_vs_backend && prog) {
232 if (!brw_vs_emit(prog, &c)) {
233 ralloc_free(mem_ctx);
240 /* Scratch space is used for register spilling */
241 if (c.last_scratch) {
242 c.prog_data.total_scratch = brw_get_scratch_size(c.last_scratch);
244 brw_get_scratch_bo(intel, &brw->vs.scratch_bo,
245 c.prog_data.total_scratch * brw->max_vs_threads);
250 program = brw_get_program(&c.func, &program_size);
252 /* We upload from &c.prog_data including the constant_map assuming
253 * they're packed together. It would be nice to have a
254 * compile-time assert macro here.
256 assert(c.constant_map == (int8_t *)&c.prog_data +
257 sizeof(c.prog_data));
258 assert(ctx->Const.VertexProgram.MaxNativeParameters ==
259 ARRAY_SIZE(c.constant_map));
262 aux_size = sizeof(c.prog_data);
264 aux_size += c.vp->program.Base.Parameters->NumParameters;
266 brw_upload_cache(&brw->cache, BRW_VS_PROG,
267 &c.key, sizeof(c.key),
268 program, program_size,
269 &c.prog_data, aux_size,
270 &brw->vs.prog_offset, &brw->vs.prog_data);
271 ralloc_free(mem_ctx);
277 static void brw_upload_vs_prog(struct brw_context *brw)
279 struct intel_context *intel = &brw->intel;
280 struct gl_context *ctx = &intel->ctx;
281 struct brw_vs_prog_key key;
282 struct brw_vertex_program *vp =
283 (struct brw_vertex_program *)brw->vertex_program;
286 memset(&key, 0, sizeof(key));
288 /* Just upload the program verbatim for now. Always send it all
289 * the inputs it asks for, whether they are varying or not.
291 key.program_string_id = vp->id;
292 key.userclip_active = (ctx->Transform.ClipPlanesEnabled != 0);
293 key.uses_clip_distance = vp->program.UsesClipDistance;
294 if (key.userclip_active && !key.uses_clip_distance) {
295 if (intel->gen < 6) {
296 key.nr_userclip_plane_consts
297 = _mesa_bitcount_64(ctx->Transform.ClipPlanesEnabled);
298 key.userclip_planes_enabled_gen_4_5
299 = ctx->Transform.ClipPlanesEnabled;
301 key.nr_userclip_plane_consts
302 = _mesa_logbase2(ctx->Transform.ClipPlanesEnabled) + 1;
305 key.copy_edgeflag = (ctx->Polygon.FrontMode != GL_FILL ||
306 ctx->Polygon.BackMode != GL_FILL);
308 /* _NEW_LIGHT | _NEW_BUFFERS */
309 key.clamp_vertex_color = ctx->Light._ClampVertexColor;
312 if (ctx->Point.PointSprite) {
313 for (i = 0; i < 8; i++) {
314 if (ctx->Point.CoordReplace[i])
315 key.point_coord_replace |= (1 << i);
319 /* BRW_NEW_VERTICES */
320 for (i = 0; i < VERT_ATTRIB_MAX; i++) {
321 if (vp->program.Base.InputsRead & (1 << i) &&
322 brw->vb.inputs[i].glarray->Type == GL_FIXED) {
323 key.gl_fixed_input_size[i] = brw->vb.inputs[i].glarray->Size;
327 if (!brw_search_cache(&brw->cache, BRW_VS_PROG,
329 &brw->vs.prog_offset, &brw->vs.prog_data)) {
330 bool success = do_vs_prog(brw, ctx->Shader.CurrentVertexProgram,
335 brw->vs.constant_map = ((int8_t *)brw->vs.prog_data +
336 sizeof(*brw->vs.prog_data));
341 const struct brw_tracked_state brw_vs_prog = {
343 .mesa = (_NEW_TRANSFORM | _NEW_POLYGON | _NEW_POINT | _NEW_LIGHT |
345 .brw = (BRW_NEW_VERTEX_PROGRAM |
349 .prepare = brw_upload_vs_prog
353 brw_vs_precompile(struct gl_context *ctx, struct gl_shader_program *prog)
355 struct brw_context *brw = brw_context(ctx);
356 struct brw_vs_prog_key key;
357 uint32_t old_prog_offset = brw->vs.prog_offset;
358 struct brw_vs_prog_data *old_prog_data = brw->vs.prog_data;
361 if (!prog->_LinkedShaders[MESA_SHADER_VERTEX])
364 struct gl_vertex_program *vp = (struct gl_vertex_program *)
365 prog->_LinkedShaders[MESA_SHADER_VERTEX]->Program;
366 struct brw_vertex_program *bvp = brw_vertex_program(vp);
368 memset(&key, 0, sizeof(key));
370 key.program_string_id = bvp->id;
371 key.clamp_vertex_color = true;
373 success = do_vs_prog(brw, prog, bvp, &key);
375 brw->vs.prog_offset = old_prog_offset;
376 brw->vs.prog_data = old_prog_data;