mesa: convert _NEW_RASTERIZER_DISCARD to a driver flag
[mesa.git] / src / mesa / drivers / dri / i965 / brw_gs.c
1 /*
2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) 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 <keith@tungstengraphics.com>
30 */
31
32 #include "main/glheader.h"
33 #include "main/macros.h"
34 #include "main/enums.h"
35 #include "main/transformfeedback.h"
36
37 #include "intel_batchbuffer.h"
38
39 #include "brw_defines.h"
40 #include "brw_context.h"
41 #include "brw_eu.h"
42 #include "brw_util.h"
43 #include "brw_state.h"
44 #include "brw_gs.h"
45
46 #include "glsl/ralloc.h"
47
48 static void compile_gs_prog( struct brw_context *brw,
49 struct brw_gs_prog_key *key )
50 {
51 struct intel_context *intel = &brw->intel;
52 struct brw_gs_compile c;
53 const GLuint *program;
54 void *mem_ctx;
55 GLuint program_size;
56
57 memset(&c, 0, sizeof(c));
58
59 c.key = *key;
60 c.vue_map = brw->vs.prog_data->base.vue_map;
61 c.nr_regs = (c.vue_map.num_slots + 1)/2;
62
63 mem_ctx = ralloc_context(NULL);
64
65 /* Begin the compilation:
66 */
67 brw_init_compile(brw, &c.func, mem_ctx);
68
69 c.func.single_program_flow = 1;
70
71 /* For some reason the thread is spawned with only 4 channels
72 * unmasked.
73 */
74 brw_set_mask_control(&c.func, BRW_MASK_DISABLE);
75
76 if (intel->gen >= 6) {
77 unsigned num_verts;
78 bool check_edge_flag;
79 /* On Sandybridge, we use the GS for implementing transform feedback
80 * (called "Stream Out" in the PRM).
81 */
82 switch (key->primitive) {
83 case _3DPRIM_POINTLIST:
84 num_verts = 1;
85 check_edge_flag = false;
86 break;
87 case _3DPRIM_LINELIST:
88 case _3DPRIM_LINESTRIP:
89 case _3DPRIM_LINELOOP:
90 num_verts = 2;
91 check_edge_flag = false;
92 break;
93 case _3DPRIM_TRILIST:
94 case _3DPRIM_TRIFAN:
95 case _3DPRIM_TRISTRIP:
96 case _3DPRIM_RECTLIST:
97 num_verts = 3;
98 check_edge_flag = false;
99 break;
100 case _3DPRIM_QUADLIST:
101 case _3DPRIM_QUADSTRIP:
102 case _3DPRIM_POLYGON:
103 num_verts = 3;
104 check_edge_flag = true;
105 break;
106 default:
107 assert(!"Unexpected primitive type in Gen6 SOL program.");
108 return;
109 }
110 gen6_sol_program(&c, key, num_verts, check_edge_flag);
111 } else {
112 /* On Gen4-5, we use the GS to decompose certain types of primitives.
113 * Note that primitives which don't require a GS program have already
114 * been weeded out by now.
115 */
116 switch (key->primitive) {
117 case _3DPRIM_QUADLIST:
118 brw_gs_quads( &c, key );
119 break;
120 case _3DPRIM_QUADSTRIP:
121 brw_gs_quad_strip( &c, key );
122 break;
123 case _3DPRIM_LINELOOP:
124 brw_gs_lines( &c );
125 break;
126 default:
127 ralloc_free(mem_ctx);
128 return;
129 }
130 }
131
132 /* get the program
133 */
134 program = brw_get_program(&c.func, &program_size);
135
136 if (unlikely(INTEL_DEBUG & DEBUG_GS)) {
137 int i;
138
139 printf("gs:\n");
140 for (i = 0; i < program_size / sizeof(struct brw_instruction); i++)
141 brw_disasm(stdout, &((struct brw_instruction *)program)[i],
142 intel->gen);
143 printf("\n");
144 }
145
146 brw_upload_cache(&brw->cache, BRW_GS_PROG,
147 &c.key, sizeof(c.key),
148 program, program_size,
149 &c.prog_data, sizeof(c.prog_data),
150 &brw->gs.prog_offset, &brw->gs.prog_data);
151 ralloc_free(mem_ctx);
152 }
153
154 static void populate_key( struct brw_context *brw,
155 struct brw_gs_prog_key *key )
156 {
157 static const unsigned swizzle_for_offset[4] = {
158 BRW_SWIZZLE4(0, 1, 2, 3),
159 BRW_SWIZZLE4(1, 2, 3, 3),
160 BRW_SWIZZLE4(2, 3, 3, 3),
161 BRW_SWIZZLE4(3, 3, 3, 3)
162 };
163
164 struct gl_context *ctx = &brw->intel.ctx;
165 struct intel_context *intel = &brw->intel;
166
167 memset(key, 0, sizeof(*key));
168
169 /* CACHE_NEW_VS_PROG (part of VUE map) */
170 key->attrs = brw->vs.prog_data->base.vue_map.slots_valid;
171
172 /* BRW_NEW_PRIMITIVE */
173 key->primitive = brw->primitive;
174
175 /* _NEW_LIGHT */
176 key->pv_first = (ctx->Light.ProvokingVertex == GL_FIRST_VERTEX_CONVENTION);
177 if (key->primitive == _3DPRIM_QUADLIST && ctx->Light.ShadeModel != GL_FLAT) {
178 /* Provide consistent primitive order with brw_set_prim's
179 * optimization of single quads to trifans.
180 */
181 key->pv_first = true;
182 }
183
184 if (intel->gen >= 7) {
185 /* On Gen7 and later, we don't use GS (yet). */
186 key->need_gs_prog = false;
187 } else if (intel->gen == 6) {
188 /* On Gen6, GS is used for transform feedback. */
189 /* BRW_NEW_TRANSFORM_FEEDBACK */
190 if (_mesa_is_xfb_active_and_unpaused(ctx)) {
191 const struct gl_shader_program *shaderprog =
192 ctx->Shader.CurrentVertexProgram;
193 const struct gl_transform_feedback_info *linked_xfb_info =
194 &shaderprog->LinkedTransformFeedback;
195 int i;
196
197 /* Make sure that the VUE slots won't overflow the unsigned chars in
198 * key->transform_feedback_bindings[].
199 */
200 STATIC_ASSERT(BRW_VARYING_SLOT_COUNT <= 256);
201
202 /* Make sure that we don't need more binding table entries than we've
203 * set aside for use in transform feedback. (We shouldn't, since we
204 * set aside enough binding table entries to have one per component).
205 */
206 assert(linked_xfb_info->NumOutputs <= BRW_MAX_SOL_BINDINGS);
207
208 key->need_gs_prog = true;
209 key->num_transform_feedback_bindings = linked_xfb_info->NumOutputs;
210 for (i = 0; i < key->num_transform_feedback_bindings; ++i) {
211 key->transform_feedback_bindings[i] =
212 linked_xfb_info->Outputs[i].OutputRegister;
213 key->transform_feedback_swizzles[i] =
214 swizzle_for_offset[linked_xfb_info->Outputs[i].ComponentOffset];
215 }
216 }
217 /* On Gen6, GS is also used for rasterizer discard. */
218 /* BRW_NEW_RASTERIZER_DISCARD */
219 if (ctx->RasterDiscard) {
220 key->need_gs_prog = true;
221 key->rasterizer_discard = true;
222 }
223 } else {
224 /* Pre-gen6, GS is used to transform QUADLIST, QUADSTRIP, and LINELOOP
225 * into simpler primitives.
226 */
227 key->need_gs_prog = (brw->primitive == _3DPRIM_QUADLIST ||
228 brw->primitive == _3DPRIM_QUADSTRIP ||
229 brw->primitive == _3DPRIM_LINELOOP);
230 }
231 }
232
233 /* Calculate interpolants for triangle and line rasterization.
234 */
235 static void
236 brw_upload_gs_prog(struct brw_context *brw)
237 {
238 struct brw_gs_prog_key key;
239 /* Populate the key:
240 */
241 populate_key(brw, &key);
242
243 if (brw->gs.prog_active != key.need_gs_prog) {
244 brw->state.dirty.cache |= CACHE_NEW_GS_PROG;
245 brw->gs.prog_active = key.need_gs_prog;
246 }
247
248 if (brw->gs.prog_active) {
249 if (!brw_search_cache(&brw->cache, BRW_GS_PROG,
250 &key, sizeof(key),
251 &brw->gs.prog_offset, &brw->gs.prog_data)) {
252 compile_gs_prog( brw, &key );
253 }
254 }
255 }
256
257
258 const struct brw_tracked_state brw_gs_prog = {
259 .dirty = {
260 .mesa = (_NEW_LIGHT),
261 .brw = (BRW_NEW_PRIMITIVE |
262 BRW_NEW_TRANSFORM_FEEDBACK |
263 BRW_NEW_RASTERIZER_DISCARD),
264 .cache = CACHE_NEW_VS_PROG
265 },
266 .emit = brw_upload_gs_prog
267 };