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 {
 191    SVGA3dPrimitiveRange range;
 192    unsigned hw_prim;
 193    unsigned hw_count;
 194    unsigned index_offset = start * index_size;
 195
 196    hw_prim = svga_translate_prim(prim, count, &hw_count);
 197    if (hw_count == 0)
 198       return PIPE_OK; /* nothing to draw */
 199
 200    range.primType = hw_prim;
 201    range.primitiveCount = hw_count;
 202    range.indexArray.offset = index_offset;
 203    range.indexArray.stride = index_size;
 204    range.indexWidth = index_size;
 205    range.indexBias = index_bias;
 206
 207    return svga_hwtnl_prim(hwtnl, &range, count,
 208                           min_index, max_index, index_buffer,
 209                           start_instance, instance_count);
 210 }
 211
 212
 213 enum pipe_error
 214 svga_hwtnl_draw_range_elements(struct svga_hwtnl *hwtnl,
 215                                const struct pipe_draw_info *info,
 216                                unsigned count)
 217 {
 218    struct pipe_context *pipe = &hwtnl->svga->pipe;
 219    enum pipe_prim_type gen_prim;
 220    unsigned gen_size, gen_nr;
 221    enum indices_mode gen_type;
 222    u_translate_func gen_func;
 223    enum pipe_error ret = PIPE_OK;
 224
 225    SVGA_STATS_TIME_PUSH(svga_sws(hwtnl->svga),
 226                         SVGA_STATS_TIME_HWTNLDRAWELEMENTS);
 227
 228    if (svga_need_unfilled_fallback(hwtnl, info->mode)) {
 229       gen_type = u_unfilled_translator(info->mode,
 230                                        info->index_size,
 231                                        count,
 232                                        hwtnl->api_fillmode,
 233                                        &gen_prim,
 234                                        &gen_size, &gen_nr, &gen_func);
 235    }
 236    else {
 237       gen_type = u_index_translator(svga_hw_prims,
 238                                     info->mode,
 239                                     info->index_size,
 240                                     count,
 241                                     hwtnl->api_pv,
 242                                     hwtnl->hw_pv,
 243                                     PR_DISABLE,
 244                                     &gen_prim, &gen_size, &gen_nr, &gen_func);
 245    }
 246
 247    if (gen_type == U_TRANSLATE_MEMCPY) {
 248       /* No need for translation, just pass through to hardware:
 249        */
 250       unsigned start_offset = info->start * info->index_size;
 251       struct pipe_resource *index_buffer = NULL;
 252       unsigned index_offset;
 253
 254       if (info->has_user_indices) {
 255          u_upload_data(pipe->stream_uploader, 0, count * info->index_size,
 256                        info->index_size, (char *) info->index.user + start_offset,
 257                        &index_offset, &index_buffer);
 258          u_upload_unmap(pipe->stream_uploader);
 259          index_offset /= info->index_size;
 260       } else {
 261          pipe_resource_reference(&index_buffer, info->index.resource);
 262          index_offset = info->start;
 263       }
 264
 265       assert(index_buffer != NULL);
 266
 267       ret = svga_hwtnl_simple_draw_range_elements(hwtnl, index_buffer,
 268                                                   info->index_size,
 269                                                   info->index_bias,
 270                                                   info->min_index,
 271                                                   info->max_index,
 272                                                   gen_prim, index_offset, count,
 273                                                   info->start_instance,
 274                                                   info->instance_count);
 275       pipe_resource_reference(&index_buffer, NULL);
 276    }
 277    else {
 278       struct pipe_resource *gen_buf = NULL;
 279       unsigned gen_offset = 0;
 280
 281       /* Need to allocate a new index buffer and run the translate
 282        * func to populate it.  Could potentially cache this translated
 283        * index buffer with the original to avoid future
 284        * re-translations.  Not much point if we're just accelerating
 285        * GL though, as index buffers are typically used only once
 286        * there.
 287        */
 288       ret = translate_indices(hwtnl, info, gen_prim,
 289                               count, gen_nr, gen_size,
 290                               gen_func, &gen_buf, &gen_offset);
 291       if (ret == PIPE_OK) {
 292          gen_offset /= gen_size;
 293          ret = svga_hwtnl_simple_draw_range_elements(hwtnl,
 294                                                      gen_buf,
 295                                                      gen_size,
 296                                                      info->index_bias,
 297                                                      info->min_index,
 298                                                      info->max_index,
 299                                                      gen_prim, gen_offset,
 300                                                      gen_nr,
 301                                                      info->start_instance,
 302                                                      info->instance_count);
 303       }
 304
 305       if (gen_buf) {
 306          pipe_resource_reference(&gen_buf, NULL);
 307       }
 308    }
 309
 310    SVGA_STATS_TIME_POP(svga_sws(hwtnl->svga));
 311    return ret;
 312 }