src/mesa/drivers/dri/intel/intel_batchbuffer.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * 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, sub license, 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 portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28 #include "intel_context.h"
  29 #include "intel_batchbuffer.h"
  30 #include "intel_decode.h"
  31 #include "intel_reg.h"
  32 #include "intel_bufmgr.h"
  33 #include "intel_buffers.h"
  34
  35 void
  36 intel_batchbuffer_reset(struct intel_batchbuffer *batch)
  37 {
  38    struct intel_context *intel = batch->intel;
  39
  40    if (batch->buf != NULL) {
  41       dri_bo_unreference(batch->buf);
  42       batch->buf = NULL;
  43    }
  44
  45    batch->buf = dri_bo_alloc(intel->bufmgr, "batchbuffer",
  46                              intel->maxBatchSize, 4096);
  47    batch->map = batch->buffer;
  48    batch->size = intel->maxBatchSize;
  49    batch->ptr = batch->map;
  50    batch->reserved_space = BATCH_RESERVED;
  51    batch->dirty_state = ~0;
  52 }
  53
  54 struct intel_batchbuffer *
  55 intel_batchbuffer_alloc(struct intel_context *intel)
  56 {
  57    struct intel_batchbuffer *batch = calloc(sizeof(*batch), 1);
  58
  59    batch->intel = intel;
  60    batch->buffer = malloc(intel->maxBatchSize);
  61    intel_batchbuffer_reset(batch);
  62
  63    return batch;
  64 }
  65
  66 void
  67 intel_batchbuffer_free(struct intel_batchbuffer *batch)
  68 {
  69    free (batch->buffer);
  70    dri_bo_unreference(batch->buf);
  71    batch->buf = NULL;
  72    free(batch);
  73 }
  74
  75
  76
  77 /* TODO: Push this whole function into bufmgr.
  78  */
  79 static void
  80 do_flush_locked(struct intel_batchbuffer *batch, GLuint used)
  81 {
  82    struct intel_context *intel = batch->intel;
  83    int ret = 0;
  84    int x_off = 0, y_off = 0;
  85
  86    dri_bo_subdata (batch->buf, 0, used, batch->buffer);
  87
  88    batch->ptr = NULL;
  89
  90    if (!intel->no_hw)
  91       dri_bo_exec(batch->buf, used, NULL, 0, (x_off & 0xffff) | (y_off << 16));
  92
  93    if (INTEL_DEBUG & DEBUG_BATCH) {
  94       dri_bo_map(batch->buf, GL_FALSE);
  95       intel_decode(batch->buf->virtual, used / 4, batch->buf->offset,
  96                    intel->intelScreen->deviceID);
  97       dri_bo_unmap(batch->buf);
  98
  99       if (intel->vtbl.debug_batch != NULL)
 100          intel->vtbl.debug_batch(intel);
 101    }
 102
 103    if (ret != 0) {
 104       exit(1);
 105    }
 106    intel->vtbl.new_batch(intel);
 107 }
 108
 109 void
 110 _intel_batchbuffer_flush(struct intel_batchbuffer *batch, const char *file,
 111                          int line)
 112 {
 113    struct intel_context *intel = batch->intel;
 114    GLuint used = batch->ptr - batch->map;
 115
 116    if (intel->first_post_swapbuffers_batch == NULL) {
 117       intel->first_post_swapbuffers_batch = intel->batch->buf;
 118       drm_intel_bo_reference(intel->first_post_swapbuffers_batch);
 119    }
 120
 121    if (used == 0)
 122       return;
 123
 124    if (INTEL_DEBUG & DEBUG_BATCH)
 125       fprintf(stderr, "%s:%d: Batchbuffer flush with %db used\n", file, line,
 126               used);
 127
 128    batch->reserved_space = 0;
 129    /* Emit a flush if the bufmgr doesn't do it for us. */
 130    if (intel->always_flush_cache) {
 131       intel_batchbuffer_emit_mi_flush(batch);
 132       used = batch->ptr - batch->map;
 133    }
 134
 135    /* Round batchbuffer usage to 2 DWORDs. */
 136
 137    if ((used & 4) == 0) {
 138       *(GLuint *) (batch->ptr) = 0; /* noop */
 139       batch->ptr += 4;
 140       used = batch->ptr - batch->map;
 141    }
 142
 143    /* Mark the end of the buffer. */
 144    *(GLuint *) (batch->ptr) = MI_BATCH_BUFFER_END;
 145    batch->ptr += 4;
 146    used = batch->ptr - batch->map;
 147    assert (used <= batch->buf->size);
 148
 149    /* Workaround for recursive batchbuffer flushing: If the window is
 150     * moved, we can get into a case where we try to flush during a
 151     * flush.  What happens is that when we try to grab the lock for
 152     * the first flush, we detect that the window moved which then
 153     * causes another flush (from the intel_draw_buffer() call in
 154     * intelUpdatePageFlipping()).  To work around this we reset the
 155     * batchbuffer tail pointer before trying to get the lock.  This
 156     * prevent the nested buffer flush, but a better fix would be to
 157     * avoid that in the first place. */
 158    batch->ptr = batch->map;
 159
 160    if (intel->vtbl.finish_batch)
 161       intel->vtbl.finish_batch(intel);
 162
 163    /* Check that we didn't just wrap our batchbuffer at a bad time. */
 164    assert(!intel->no_batch_wrap);
 165
 166    /* TODO: Just pass the relocation list and dma buffer up to the
 167     * kernel.
 168     */
 169    do_flush_locked(batch, used);
 170
 171    if (INTEL_DEBUG & DEBUG_SYNC) {
 172       fprintf(stderr, "waiting for idle\n");
 173       dri_bo_map(batch->buf, GL_TRUE);
 174       dri_bo_unmap(batch->buf);
 175    }
 176
 177    /* Reset the buffer:
 178     */
 179    intel_batchbuffer_reset(batch);
 180 }
 181
 182
 183 /*  This is the only way buffers get added to the validate list.
 184  */
 185 GLboolean
 186 intel_batchbuffer_emit_reloc(struct intel_batchbuffer *batch,
 187                              dri_bo *buffer,
 188                              uint32_t read_domains, uint32_t write_domain,
 189                              uint32_t delta)
 190 {
 191    int ret;
 192
 193    assert(delta < buffer->size);
 194
 195    if (batch->ptr - batch->map > batch->buf->size)
 196     printf ("bad relocation ptr %p map %p offset %d size %lu\n",
 197             batch->ptr, batch->map, batch->ptr - batch->map, batch->buf->size);
 198    ret = dri_bo_emit_reloc(batch->buf, read_domains, write_domain,
 199                            delta, batch->ptr - batch->map, buffer);
 200
 201    /*
 202     * Using the old buffer offset, write in what the right data would be, in case
 203     * the buffer doesn't move and we can short-circuit the relocation processing
 204     * in the kernel
 205     */
 206    intel_batchbuffer_emit_dword (batch, buffer->offset + delta);
 207
 208    return GL_TRUE;
 209 }
 210
 211 GLboolean
 212 intel_batchbuffer_emit_reloc_fenced(struct intel_batchbuffer *batch,
 213                                     drm_intel_bo *buffer,
 214                                     uint32_t read_domains, uint32_t write_domain,
 215                                     uint32_t delta)
 216 {
 217    int ret;
 218
 219    assert(delta < buffer->size);
 220
 221    if (batch->ptr - batch->map > batch->buf->size)
 222     printf ("bad relocation ptr %p map %p offset %d size %lu\n",
 223             batch->ptr, batch->map, batch->ptr - batch->map, batch->buf->size);
 224    ret = drm_intel_bo_emit_reloc_fence(batch->buf, batch->ptr - batch->map,
 225                                        buffer, delta,
 226                                        read_domains, write_domain);
 227
 228    /*
 229     * Using the old buffer offset, write in what the right data would
 230     * be, in case the buffer doesn't move and we can short-circuit the
 231     * relocation processing in the kernel
 232     */
 233    intel_batchbuffer_emit_dword (batch, buffer->offset + delta);
 234
 235    return GL_TRUE;
 236 }
 237
 238 void
 239 intel_batchbuffer_data(struct intel_batchbuffer *batch,
 240                        const void *data, GLuint bytes)
 241 {
 242    assert((bytes & 3) == 0);
 243    intel_batchbuffer_require_space(batch, bytes);
 244    __memcpy(batch->ptr, data, bytes);
 245    batch->ptr += bytes;
 246 }
 247
 248 /* Emit a pipelined flush to either flush render and texture cache for
 249  * reading from a FBO-drawn texture, or flush so that frontbuffer
 250  * render appears on the screen in DRI1.
 251  *
 252  * This is also used for the always_flush_cache driconf debug option.
 253  */
 254 void
 255 intel_batchbuffer_emit_mi_flush(struct intel_batchbuffer *batch)
 256 {
 257    struct intel_context *intel = batch->intel;
 258
 259    if (intel->gen >= 6) {
 260       BEGIN_BATCH(4);
 261       OUT_BATCH(_3DSTATE_PIPE_CONTROL);
 262       OUT_BATCH(PIPE_CONTROL_INSTRUCTION_FLUSH |
 263                 PIPE_CONTROL_WRITE_FLUSH |
 264                 PIPE_CONTROL_NO_WRITE);
 265       OUT_BATCH(0); /* write address */
 266       OUT_BATCH(0); /* write data */
 267       ADVANCE_BATCH();
 268    } else if (intel->gen >= 4) {
 269       BEGIN_BATCH(4);
 270       OUT_BATCH(_3DSTATE_PIPE_CONTROL |
 271                 PIPE_CONTROL_WRITE_FLUSH |
 272                 PIPE_CONTROL_NO_WRITE);
 273       OUT_BATCH(0); /* write address */
 274       OUT_BATCH(0); /* write data */
 275       OUT_BATCH(0); /* write data */
 276       ADVANCE_BATCH();
 277    } else {
 278       BEGIN_BATCH(1);
 279       OUT_BATCH(MI_FLUSH);
 280       ADVANCE_BATCH();
 281    }
 282 }