svga: fix draw elements with 8-bits indices
[mesa.git] / src / gallium / drivers / svga / svga_draw_elements.c
1 /**********************************************************
2 * Copyright 2008-2009 VMware, Inc. All rights reserved.
3 *
4 * Permission is hereby granted, free of charge, to any person
5 * obtaining a copy of this software and associated documentation
6 * files (the "Software"), to deal in the Software without
7 * restriction, including without limitation the rights to use, copy,
8 * modify, merge, publish, distribute, sublicense, and/or sell copies
9 * of the Software, and to permit persons to whom the Software is
10 * furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice shall be
13 * included in all copies or substantial portions of the Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
16 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
17 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
18 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
19 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
20 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
21 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 *
24 **********************************************************/
25
26 #include "util/u_inlines.h"
27 #include "util/u_prim.h"
28 #include "util/u_upload_mgr.h"
29 #include "indices/u_indices.h"
30
31 #include "svga_cmd.h"
32 #include "svga_draw.h"
33 #include "svga_draw_private.h"
34 #include "svga_resource_buffer.h"
35 #include "svga_winsys.h"
36 #include "svga_context.h"
37 #include "svga_hw_reg.h"
38
39
40 /**
41 * Return a new index buffer which contains a translation of the original
42 * index buffer. An example of a translation is converting from QUAD
43 * primitives to TRIANGLE primitives. Each set of four indexes for a quad
44 * will be converted to six indices for two triangles.
45 *
46 * Before generating the new index buffer we'll check if the incoming
47 * buffer already has a translated buffer that can be re-used.
48 * This benefits demos like Cinebench R15 which has many
49 * glDrawElements(GL_QUADS) commands (we can't draw quads natively).
50 *
51 * \param offset offset in bytes to first index to translate in src buffer
52 * \param orig_prim original primitive type (like PIPE_PRIM_QUADS)
53 * \param gen_prim new/generated primitive type (like PIPE_PRIM_TRIANGLES)
54 * \param orig_nr number of indexes to translate in source buffer
55 * \param gen_nr number of indexes to write into new/dest buffer
56 * \param index_size bytes per index (2 or 4)
57 * \param translate the translation function from the u_translate module
58 * \param out_buf returns the new/translated index buffer
59 * \return error code to indicate success failure
60 */
61 static enum pipe_error
62 translate_indices(struct svga_hwtnl *hwtnl,
63 const struct pipe_draw_info *info,
64 enum pipe_prim_type gen_prim,
65 unsigned orig_nr, unsigned gen_nr,
66 unsigned gen_size,
67 u_translate_func translate,
68 struct pipe_resource **out_buf,
69 unsigned *out_offset)
70 {
71 struct pipe_context *pipe = &hwtnl->svga->pipe;
72 struct svga_screen *screen = svga_screen(pipe->screen);
73 struct svga_buffer *src_sbuf = NULL;
74 struct pipe_transfer *src_transfer = NULL;
75 struct pipe_transfer *dst_transfer = NULL;
76 const unsigned size = gen_size * gen_nr;
77 const unsigned offset = info->start * info->index_size;
78 const void *src_map = NULL;
79 struct pipe_resource *dst = NULL;
80 void *dst_map = NULL;
81
82 assert(gen_size == 2 || gen_size == 4);
83 if (!info->has_user_indices)
84 src_sbuf = svga_buffer(info->index.resource);
85
86 /* If the draw_info provides us with a buffer rather than a
87 * user pointer, Check to see if we've already translated that buffer
88 */
89 if (src_sbuf && !screen->debug.no_cache_index_buffers) {
90 /* Check if we already have a translated index buffer */
91 if (src_sbuf->translated_indices.buffer &&
92 src_sbuf->translated_indices.orig_prim == info->mode &&
93 src_sbuf->translated_indices.new_prim == gen_prim &&
94 src_sbuf->translated_indices.offset == offset &&
95 src_sbuf->translated_indices.count == orig_nr &&
96 src_sbuf->translated_indices.index_size == gen_size) {
97 pipe_resource_reference(out_buf, src_sbuf->translated_indices.buffer);
98 return PIPE_OK;
99 }
100 }
101
102 /* Need to trim vertex count to make sure we don't write too much data
103 * to the dst buffer in the translate() call.
104 */
105 u_trim_pipe_prim(gen_prim, &gen_nr);
106
107 if (src_sbuf) {
108 /* If we have a source buffer, create a destination buffer in the
109 * hope that we can reuse the translated data later. If not,
110 * we'd probably be better off using the upload buffer.
111 */
112 dst = pipe_buffer_create(pipe->screen,
113 PIPE_BIND_INDEX_BUFFER, PIPE_USAGE_IMMUTABLE,
114 size);
115 if (!dst)
116 goto fail;
117
118 dst_map = pipe_buffer_map(pipe, dst, PIPE_TRANSFER_WRITE, &dst_transfer);
119 if (!dst_map)
120 goto fail;
121
122 *out_offset = 0;
123 src_map = pipe_buffer_map(pipe, info->index.resource,
124 PIPE_TRANSFER_READ |
125 PIPE_TRANSFER_UNSYNCHRONIZED,
126 &src_transfer);
127 if (!src_map)
128 goto fail;
129 } else {
130 /* Allocate upload buffer space. Align to the index size. */
131 u_upload_alloc(pipe->stream_uploader, 0, size, gen_size,
132 out_offset, &dst, &dst_map);
133 if (!dst)
134 goto fail;
135
136 src_map = info->index.user;
137 }
138
139 translate((const char *) src_map + offset, 0, 0, gen_nr, 0, dst_map);
140
141 if (src_transfer)
142 pipe_buffer_unmap(pipe, src_transfer);
143
144 if (dst_transfer)
145 pipe_buffer_unmap(pipe, dst_transfer);
146 else
147 u_upload_unmap(pipe->stream_uploader);
148
149 *out_buf = dst;
150
151 if (src_sbuf && !screen->debug.no_cache_index_buffers) {
152 /* Save the new, translated index buffer in the hope we can use it
153 * again in the future.
154 */
155 pipe_resource_reference(&src_sbuf->translated_indices.buffer, dst);
156 src_sbuf->translated_indices.orig_prim = info->mode;
157 src_sbuf->translated_indices.new_prim = gen_prim;
158 src_sbuf->translated_indices.offset = offset;
159 src_sbuf->translated_indices.count = orig_nr;
160 src_sbuf->translated_indices.index_size = gen_size;
161 }
162
163 return PIPE_OK;
164
165 fail:
166 if (src_transfer)
167 pipe_buffer_unmap(pipe, src_transfer);
168
169 if (dst_transfer)
170 pipe_buffer_unmap(pipe, dst_transfer);
171 else if (dst_map)
172 u_upload_unmap(pipe->stream_uploader);
173
174 if (dst)
175 pipe_resource_reference(&dst, NULL);
176
177 return PIPE_ERROR_OUT_OF_MEMORY;
178 }
179
180
181 enum pipe_error
182 svga_hwtnl_simple_draw_range_elements(struct svga_hwtnl *hwtnl,
183 struct pipe_resource *index_buffer,
184 unsigned index_size, int index_bias,
185 unsigned min_index, unsigned max_index,
186 enum pipe_prim_type prim, unsigned start,
187 unsigned count,
188 unsigned start_instance,
189 unsigned instance_count,
190 ubyte vertices_per_patch)
191 {
192 SVGA3dPrimitiveRange range;
193 unsigned hw_prim;
194 unsigned hw_count;
195 unsigned index_offset = start * index_size;
196
197 hw_prim = svga_translate_prim(prim, count, &hw_count, vertices_per_patch);
198 if (hw_count == 0)
199 return PIPE_OK; /* nothing to draw */
200
201 range.primType = hw_prim;
202 range.primitiveCount = hw_count;
203 range.indexArray.offset = index_offset;
204 range.indexArray.stride = index_size;
205 range.indexWidth = index_size;
206 range.indexBias = index_bias;
207
208 return svga_hwtnl_prim(hwtnl, &range, count,
209 min_index, max_index, index_buffer,
210 start_instance, instance_count,
211 NULL, NULL);
212 }
213
214
215 enum pipe_error
216 svga_hwtnl_draw_range_elements(struct svga_hwtnl *hwtnl,
217 const struct pipe_draw_info *info,
218 unsigned count)
219 {
220 struct pipe_context *pipe = &hwtnl->svga->pipe;
221 enum pipe_prim_type gen_prim;
222 unsigned gen_size, gen_nr;
223 enum indices_mode gen_type;
224 u_translate_func gen_func;
225 enum pipe_error ret = PIPE_OK;
226
227 SVGA_STATS_TIME_PUSH(svga_sws(hwtnl->svga),
228 SVGA_STATS_TIME_HWTNLDRAWELEMENTS);
229
230 if (svga_need_unfilled_fallback(hwtnl, info->mode)) {
231 gen_type = u_unfilled_translator(info->mode,
232 info->index_size,
233 count,
234 hwtnl->api_fillmode,
235 &gen_prim,
236 &gen_size, &gen_nr, &gen_func);
237 }
238 else {
239 unsigned hw_pv;
240
241 /* There is no geometry ordering with PATCH, so no need to
242 * consider provoking vertex mode for the translation.
243 * So use the same api_pv as the hw_pv.
244 */
245 hw_pv = info->mode == PIPE_PRIM_PATCHES ? hwtnl->api_pv :
246 hwtnl->hw_pv;
247 gen_type = u_index_translator(svga_hw_prims,
248 info->mode,
249 info->index_size,
250 count,
251 hwtnl->api_pv,
252 hw_pv,
253 PR_DISABLE,
254 &gen_prim, &gen_size, &gen_nr, &gen_func);
255 }
256
257 if ((gen_type == U_TRANSLATE_MEMCPY) && (info->index_size == gen_size)) {
258 /* No need for translation, just pass through to hardware:
259 */
260 unsigned start_offset = info->start * info->index_size;
261 struct pipe_resource *index_buffer = NULL;
262 unsigned index_offset;
263
264 if (info->has_user_indices) {
265 u_upload_data(pipe->stream_uploader, 0, count * info->index_size,
266 info->index_size, (char *) info->index.user + start_offset,
267 &index_offset, &index_buffer);
268 u_upload_unmap(pipe->stream_uploader);
269 index_offset /= info->index_size;
270 } else {
271 pipe_resource_reference(&index_buffer, info->index.resource);
272 index_offset = info->start;
273 }
274
275 assert(index_buffer != NULL);
276
277 ret = svga_hwtnl_simple_draw_range_elements(hwtnl, index_buffer,
278 info->index_size,
279 info->index_bias,
280 info->min_index,
281 info->max_index,
282 gen_prim, index_offset, count,
283 info->start_instance,
284 info->instance_count,
285 info->vertices_per_patch);
286 pipe_resource_reference(&index_buffer, NULL);
287 }
288 else {
289 struct pipe_resource *gen_buf = NULL;
290 unsigned gen_offset = 0;
291
292 /* Need to allocate a new index buffer and run the translate
293 * func to populate it. Could potentially cache this translated
294 * index buffer with the original to avoid future
295 * re-translations. Not much point if we're just accelerating
296 * GL though, as index buffers are typically used only once
297 * there.
298 */
299 ret = translate_indices(hwtnl, info, gen_prim,
300 count, gen_nr, gen_size,
301 gen_func, &gen_buf, &gen_offset);
302 if (ret == PIPE_OK) {
303 gen_offset /= gen_size;
304 ret = svga_hwtnl_simple_draw_range_elements(hwtnl,
305 gen_buf,
306 gen_size,
307 info->index_bias,
308 info->min_index,
309 info->max_index,
310 gen_prim, gen_offset,
311 gen_nr,
312 info->start_instance,
313 info->instance_count,
314 info->vertices_per_patch);
315 }
316
317 if (gen_buf) {
318 pipe_resource_reference(&gen_buf, NULL);
319 }
320 }
321
322 SVGA_STATS_TIME_POP(svga_sws(hwtnl->svga));
323 return ret;
324 }