i965: Split out per-stage dirty-bit checking into separate functions
[mesa.git] / src / mesa / drivers / dri / i965 / brw_ff_gs.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 #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_ff_gs.h"
45
46 #include "util/ralloc.h"
47
48 static void compile_ff_gs_prog(struct brw_context *brw,
49 struct brw_ff_gs_prog_key *key)
50 {
51 struct brw_ff_gs_compile c;
52 const GLuint *program;
53 void *mem_ctx;
54 GLuint program_size;
55
56 memset(&c, 0, sizeof(c));
57
58 c.key = *key;
59 c.vue_map = brw->vs.prog_data->base.vue_map;
60 c.nr_regs = (c.vue_map.num_slots + 1)/2;
61
62 mem_ctx = ralloc_context(NULL);
63
64 /* Begin the compilation:
65 */
66 brw_init_compile(brw, &c.func, mem_ctx);
67
68 c.func.single_program_flow = 1;
69
70 /* For some reason the thread is spawned with only 4 channels
71 * unmasked.
72 */
73 brw_set_default_mask_control(&c.func, BRW_MASK_DISABLE);
74
75 if (brw->gen >= 6) {
76 unsigned num_verts;
77 bool check_edge_flag;
78 /* On Sandybridge, we use the GS for implementing transform feedback
79 * (called "Stream Out" in the PRM).
80 */
81 switch (key->primitive) {
82 case _3DPRIM_POINTLIST:
83 num_verts = 1;
84 check_edge_flag = false;
85 break;
86 case _3DPRIM_LINELIST:
87 case _3DPRIM_LINESTRIP:
88 case _3DPRIM_LINELOOP:
89 num_verts = 2;
90 check_edge_flag = false;
91 break;
92 case _3DPRIM_TRILIST:
93 case _3DPRIM_TRIFAN:
94 case _3DPRIM_TRISTRIP:
95 case _3DPRIM_RECTLIST:
96 num_verts = 3;
97 check_edge_flag = false;
98 break;
99 case _3DPRIM_QUADLIST:
100 case _3DPRIM_QUADSTRIP:
101 case _3DPRIM_POLYGON:
102 num_verts = 3;
103 check_edge_flag = true;
104 break;
105 default:
106 unreachable("Unexpected primitive type in Gen6 SOL program.");
107 }
108 gen6_sol_program(&c, key, num_verts, check_edge_flag);
109 } else {
110 /* On Gen4-5, we use the GS to decompose certain types of primitives.
111 * Note that primitives which don't require a GS program have already
112 * been weeded out by now.
113 */
114 switch (key->primitive) {
115 case _3DPRIM_QUADLIST:
116 brw_ff_gs_quads( &c, key );
117 break;
118 case _3DPRIM_QUADSTRIP:
119 brw_ff_gs_quad_strip( &c, key );
120 break;
121 case _3DPRIM_LINELOOP:
122 brw_ff_gs_lines( &c );
123 break;
124 default:
125 ralloc_free(mem_ctx);
126 return;
127 }
128 }
129
130 brw_compact_instructions(&c.func, 0, 0, NULL);
131
132 /* get the program
133 */
134 program = brw_get_program(&c.func, &program_size);
135
136 if (unlikely(INTEL_DEBUG & DEBUG_GS)) {
137 fprintf(stderr, "gs:\n");
138 brw_disassemble(brw, c.func.store, 0, program_size, stderr);
139 fprintf(stderr, "\n");
140 }
141
142 brw_upload_cache(&brw->cache, BRW_CACHE_FF_GS_PROG,
143 &c.key, sizeof(c.key),
144 program, program_size,
145 &c.prog_data, sizeof(c.prog_data),
146 &brw->ff_gs.prog_offset, &brw->ff_gs.prog_data);
147 ralloc_free(mem_ctx);
148 }
149
150 static bool
151 brw_ff_gs_state_dirty(struct brw_context *brw)
152 {
153 return brw_state_dirty(brw,
154 _NEW_LIGHT,
155 BRW_NEW_PRIMITIVE |
156 BRW_NEW_TRANSFORM_FEEDBACK |
157 BRW_NEW_VS_PROG_DATA);
158 }
159
160 static void
161 brw_ff_gs_populate_key(struct brw_context *brw,
162 struct brw_ff_gs_prog_key *key)
163 {
164 static const unsigned swizzle_for_offset[4] = {
165 BRW_SWIZZLE4(0, 1, 2, 3),
166 BRW_SWIZZLE4(1, 2, 3, 3),
167 BRW_SWIZZLE4(2, 3, 3, 3),
168 BRW_SWIZZLE4(3, 3, 3, 3)
169 };
170
171 struct gl_context *ctx = &brw->ctx;
172
173 memset(key, 0, sizeof(*key));
174
175 /* BRW_NEW_VS_PROG_DATA (part of VUE map) */
176 key->attrs = brw->vs.prog_data->base.vue_map.slots_valid;
177
178 /* BRW_NEW_PRIMITIVE */
179 key->primitive = brw->primitive;
180
181 /* _NEW_LIGHT */
182 key->pv_first = (ctx->Light.ProvokingVertex == GL_FIRST_VERTEX_CONVENTION);
183 if (key->primitive == _3DPRIM_QUADLIST && ctx->Light.ShadeModel != GL_FLAT) {
184 /* Provide consistent primitive order with brw_set_prim's
185 * optimization of single quads to trifans.
186 */
187 key->pv_first = true;
188 }
189
190 if (brw->gen >= 7) {
191 /* On Gen7 and later, we don't use GS (yet). */
192 key->need_gs_prog = false;
193 } else if (brw->gen == 6) {
194 /* On Gen6, GS is used for transform feedback. */
195 /* BRW_NEW_TRANSFORM_FEEDBACK */
196 if (_mesa_is_xfb_active_and_unpaused(ctx)) {
197 const struct gl_shader_program *shaderprog =
198 ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX];
199 const struct gl_transform_feedback_info *linked_xfb_info =
200 &shaderprog->LinkedTransformFeedback;
201 int i;
202
203 /* Make sure that the VUE slots won't overflow the unsigned chars in
204 * key->transform_feedback_bindings[].
205 */
206 STATIC_ASSERT(BRW_VARYING_SLOT_COUNT <= 256);
207
208 /* Make sure that we don't need more binding table entries than we've
209 * set aside for use in transform feedback. (We shouldn't, since we
210 * set aside enough binding table entries to have one per component).
211 */
212 assert(linked_xfb_info->NumOutputs <= BRW_MAX_SOL_BINDINGS);
213
214 key->need_gs_prog = true;
215 key->num_transform_feedback_bindings = linked_xfb_info->NumOutputs;
216 for (i = 0; i < key->num_transform_feedback_bindings; ++i) {
217 key->transform_feedback_bindings[i] =
218 linked_xfb_info->Outputs[i].OutputRegister;
219 key->transform_feedback_swizzles[i] =
220 swizzle_for_offset[linked_xfb_info->Outputs[i].ComponentOffset];
221 }
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 void
236 brw_upload_ff_gs_prog(struct brw_context *brw)
237 {
238 struct brw_ff_gs_prog_key key;
239
240 if (!brw_ff_gs_state_dirty(brw))
241 return;
242
243 /* Populate the key:
244 */
245 brw_ff_gs_populate_key(brw, &key);
246
247 if (brw->ff_gs.prog_active != key.need_gs_prog) {
248 brw->ctx.NewDriverState |= BRW_NEW_FF_GS_PROG_DATA;
249 brw->ff_gs.prog_active = key.need_gs_prog;
250 }
251
252 if (brw->ff_gs.prog_active) {
253 if (!brw_search_cache(&brw->cache, BRW_CACHE_FF_GS_PROG,
254 &key, sizeof(key),
255 &brw->ff_gs.prog_offset, &brw->ff_gs.prog_data)) {
256 compile_ff_gs_prog( brw, &key );
257 }
258 }
259 }
260
261 void gen6_brw_upload_ff_gs_prog(struct brw_context *brw)
262 {
263 brw_upload_ff_gs_prog(brw);
264 }