1 /*
2 * Copyright © 2011 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 DEALINGS
21 * IN THE SOFTWARE.
22 */
23
24 /** \file gen6_sol.c
25 *
26 * Code to initialize the binding table entries used by transform feedback.
27 */
28
29 #include "main/bufferobj.h"
30 #include "main/macros.h"
31 #include "brw_context.h"
32 #include "intel_batchbuffer.h"
33 #include "brw_defines.h"
34 #include "brw_state.h"
35 #include "main/transformfeedback.h"
36
37 static void
gen6_update_sol_surfaces(struct brw_context * brw)38 gen6_update_sol_surfaces(struct brw_context *brw)
39 {
40 struct gl_context *ctx = &brw->ctx;
41 bool xfb_active = _mesa_is_xfb_active_and_unpaused(ctx);
42 struct gl_transform_feedback_object *xfb_obj;
43 const struct gl_transform_feedback_info *linked_xfb_info = NULL;
44
45 if (xfb_active) {
46 /* BRW_NEW_TRANSFORM_FEEDBACK */
47 xfb_obj = ctx->TransformFeedback.CurrentObject;
48 linked_xfb_info = xfb_obj->program->sh.LinkedTransformFeedback;
49 }
50
51 for (int i = 0; i < BRW_MAX_SOL_BINDINGS; ++i) {
52 const int surf_index = SURF_INDEX_GEN6_SOL_BINDING(i);
53 if (xfb_active && i < linked_xfb_info->NumOutputs) {
54 unsigned buffer = linked_xfb_info->Outputs[i].OutputBuffer;
55 unsigned buffer_offset =
56 xfb_obj->Offset[buffer] / 4 +
57 linked_xfb_info->Outputs[i].DstOffset;
58 if (brw->geometry_program) {
59 brw_update_sol_surface(
60 brw, xfb_obj->Buffers[buffer],
61 &brw->gs.base.surf_offset[surf_index],
62 linked_xfb_info->Outputs[i].NumComponents,
63 linked_xfb_info->Buffers[buffer].Stride, buffer_offset);
64 } else {
65 brw_update_sol_surface(
66 brw, xfb_obj->Buffers[buffer],
67 &brw->ff_gs.surf_offset[surf_index],
68 linked_xfb_info->Outputs[i].NumComponents,
69 linked_xfb_info->Buffers[buffer].Stride, buffer_offset);
70 }
71 } else {
72 if (!brw->geometry_program)
73 brw->ff_gs.surf_offset[surf_index] = 0;
74 else
75 brw->gs.base.surf_offset[surf_index] = 0;
76 }
77 }
78
79 brw->ctx.NewDriverState |= BRW_NEW_SURFACES;
80 }
81
82 const struct brw_tracked_state gen6_sol_surface = {
83 .dirty = {
84 .mesa = 0,
85 .brw = BRW_NEW_BATCH |
86 BRW_NEW_BLORP |
87 BRW_NEW_TRANSFORM_FEEDBACK,
88 },
89 .emit = gen6_update_sol_surfaces,
90 };
91
92 /**
93 * Constructs the binding table for the WM surface state, which maps unit
94 * numbers to surface state objects.
95 */
96 static void
brw_gs_upload_binding_table(struct brw_context * brw)97 brw_gs_upload_binding_table(struct brw_context *brw)
98 {
99 uint32_t *bind;
100 struct gl_context *ctx = &brw->ctx;
101 const struct gl_shader_program *shaderprog;
102 bool need_binding_table = false;
103
104 /* We have two scenarios here:
105 * 1) We are using a geometry shader only to implement transform feedback
106 * for a vertex shader (brw->geometry_program == NULL). In this case, we
107 * only need surfaces for transform feedback in the GS stage.
108 * 2) We have a user-provided geometry shader. In this case we may need
109 * surfaces for transform feedback and/or other stuff, like textures,
110 * in the GS stage.
111 */
112
113 if (!brw->geometry_program) {
114 /* BRW_NEW_VERTEX_PROGRAM */
115 shaderprog = ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX];
116 if (shaderprog) {
117 /* Skip making a binding table if we don't have anything to put in it */
118 const struct gl_transform_feedback_info *linked_xfb_info =
119 shaderprog->xfb_program->sh.LinkedTransformFeedback;
120 need_binding_table = linked_xfb_info->NumOutputs > 0;
121 }
122 if (!need_binding_table) {
123 if (brw->ff_gs.bind_bo_offset != 0) {
124 brw->ctx.NewDriverState |= BRW_NEW_BINDING_TABLE_POINTERS;
125 brw->ff_gs.bind_bo_offset = 0;
126 }
127 return;
128 }
129
130 /* Might want to calculate nr_surfaces first, to avoid taking up so much
131 * space for the binding table. Anyway, in this case we know that we only
132 * use BRW_MAX_SOL_BINDINGS surfaces at most.
133 */
134 bind = brw_state_batch(brw, AUB_TRACE_BINDING_TABLE,
135 sizeof(uint32_t) * BRW_MAX_SOL_BINDINGS,
136 32, &brw->ff_gs.bind_bo_offset);
137
138 /* BRW_NEW_SURFACES */
139 memcpy(bind, brw->ff_gs.surf_offset,
140 BRW_MAX_SOL_BINDINGS * sizeof(uint32_t));
141 } else {
142 /* BRW_NEW_GEOMETRY_PROGRAM */
143 shaderprog = ctx->_Shader->CurrentProgram[MESA_SHADER_GEOMETRY];
144 if (shaderprog) {
145 /* Skip making a binding table if we don't have anything to put in it */
146 struct brw_stage_prog_data *prog_data = brw->gs.base.prog_data;
147 const struct gl_transform_feedback_info *linked_xfb_info =
148 shaderprog->xfb_program->sh.LinkedTransformFeedback;
149 need_binding_table = linked_xfb_info->NumOutputs > 0 ||
150 prog_data->binding_table.size_bytes > 0;
151 }
152 if (!need_binding_table) {
153 if (brw->gs.base.bind_bo_offset != 0) {
154 brw->gs.base.bind_bo_offset = 0;
155 brw->ctx.NewDriverState |= BRW_NEW_BINDING_TABLE_POINTERS;
156 }
157 return;
158 }
159
160 /* Might want to calculate nr_surfaces first, to avoid taking up so much
161 * space for the binding table.
162 */
163 bind = brw_state_batch(brw, AUB_TRACE_BINDING_TABLE,
164 sizeof(uint32_t) * BRW_MAX_SURFACES,
165 32, &brw->gs.base.bind_bo_offset);
166
167 /* BRW_NEW_SURFACES */
168 memcpy(bind, brw->gs.base.surf_offset,
169 BRW_MAX_SURFACES * sizeof(uint32_t));
170 }
171
172 brw->ctx.NewDriverState |= BRW_NEW_BINDING_TABLE_POINTERS;
173 }
174
175 const struct brw_tracked_state gen6_gs_binding_table = {
176 .dirty = {
177 .mesa = 0,
178 .brw = BRW_NEW_BATCH |
179 BRW_NEW_BLORP |
180 BRW_NEW_GEOMETRY_PROGRAM |
181 BRW_NEW_VERTEX_PROGRAM |
182 BRW_NEW_SURFACES,
183 },
184 .emit = brw_gs_upload_binding_table,
185 };
186
187 struct gl_transform_feedback_object *
brw_new_transform_feedback(struct gl_context * ctx,GLuint name)188 brw_new_transform_feedback(struct gl_context *ctx, GLuint name)
189 {
190 struct brw_context *brw = brw_context(ctx);
191 struct brw_transform_feedback_object *brw_obj =
192 CALLOC_STRUCT(brw_transform_feedback_object);
193 if (!brw_obj)
194 return NULL;
195
196 _mesa_init_transform_feedback_object(&brw_obj->base, name);
197
198 brw_obj->offset_bo =
199 drm_intel_bo_alloc(brw->bufmgr, "transform feedback offsets", 16, 64);
200 brw_obj->prim_count_bo =
201 drm_intel_bo_alloc(brw->bufmgr, "xfb primitive counts", 4096, 64);
202
203 return &brw_obj->base;
204 }
205
206 void
brw_delete_transform_feedback(struct gl_context * ctx,struct gl_transform_feedback_object * obj)207 brw_delete_transform_feedback(struct gl_context *ctx,
208 struct gl_transform_feedback_object *obj)
209 {
210 struct brw_transform_feedback_object *brw_obj =
211 (struct brw_transform_feedback_object *) obj;
212
213 for (unsigned i = 0; i < ARRAY_SIZE(obj->Buffers); i++) {
214 _mesa_reference_buffer_object(ctx, &obj->Buffers[i], NULL);
215 }
216
217 drm_intel_bo_unreference(brw_obj->offset_bo);
218 drm_intel_bo_unreference(brw_obj->prim_count_bo);
219
220 free(brw_obj);
221 }
222
223 void
brw_begin_transform_feedback(struct gl_context * ctx,GLenum mode,struct gl_transform_feedback_object * obj)224 brw_begin_transform_feedback(struct gl_context *ctx, GLenum mode,
225 struct gl_transform_feedback_object *obj)
226 {
227 struct brw_context *brw = brw_context(ctx);
228 const struct gl_shader_program *shaderprog;
229 const struct gl_transform_feedback_info *linked_xfb_info;
230 struct gl_transform_feedback_object *xfb_obj =
231 ctx->TransformFeedback.CurrentObject;
232
233 assert(brw->gen == 6);
234
235 if (ctx->_Shader->CurrentProgram[MESA_SHADER_GEOMETRY]) {
236 /* BRW_NEW_GEOMETRY_PROGRAM */
237 shaderprog =
238 ctx->_Shader->CurrentProgram[MESA_SHADER_GEOMETRY];
239 } else {
240 /* BRW_NEW_VERTEX_PROGRAM */
241 shaderprog =
242 ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX];
243 }
244 linked_xfb_info = shaderprog->xfb_program->sh.LinkedTransformFeedback;
245
246 /* Compute the maximum number of vertices that we can write without
247 * overflowing any of the buffers currently being used for feedback.
248 */
249 unsigned max_index
250 = _mesa_compute_max_transform_feedback_vertices(ctx, xfb_obj,
251 linked_xfb_info);
252
253 /* Initialize the SVBI 0 register to zero and set the maximum index. */
254 BEGIN_BATCH(4);
255 OUT_BATCH(_3DSTATE_GS_SVB_INDEX << 16 | (4 - 2));
256 OUT_BATCH(0); /* SVBI 0 */
257 OUT_BATCH(0); /* starting index */
258 OUT_BATCH(max_index);
259 ADVANCE_BATCH();
260
261 /* Initialize the rest of the unused streams to sane values. Otherwise,
262 * they may indicate that there is no room to write data and prevent
263 * anything from happening at all.
264 */
265 for (int i = 1; i < 4; i++) {
266 BEGIN_BATCH(4);
267 OUT_BATCH(_3DSTATE_GS_SVB_INDEX << 16 | (4 - 2));
268 OUT_BATCH(i << SVB_INDEX_SHIFT);
269 OUT_BATCH(0); /* starting index */
270 OUT_BATCH(0xffffffff);
271 ADVANCE_BATCH();
272 }
273 }
274
275 void
brw_end_transform_feedback(struct gl_context * ctx,struct gl_transform_feedback_object * obj)276 brw_end_transform_feedback(struct gl_context *ctx,
277 struct gl_transform_feedback_object *obj)
278 {
279 /* After EndTransformFeedback, it's likely that the client program will try
280 * to draw using the contents of the transform feedback buffer as vertex
281 * input. In order for this to work, we need to flush the data through at
282 * least the GS stage of the pipeline, and flush out the render cache. For
283 * simplicity, just do a full flush.
284 */
285 struct brw_context *brw = brw_context(ctx);
286 brw_emit_mi_flush(brw);
287 }
288