src/gallium/winsys/r600/drm/radeon_ctx.c

   1 /*
   2  * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
   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  * on the rights to use, copy, modify, merge, publish, distribute, sub
   8  * license, and/or sell copies of the Software, and to permit persons to whom
   9  * the 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 NON-INFRINGEMENT. IN NO EVENT SHALL
  18  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  19  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  20  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  21  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  22  *
  23  * Authors:
  24  *      Jerome Glisse
  25  */
  26 #include <stdio.h>
  27 #include <stdlib.h>
  28 #include <string.h>
  29 #include "radeon_priv.h"
  30 #include "radeon_drm.h"
  31 #include "bof.h"
  32
  33 int radeon_ctx_set_bo_new(struct radeon_ctx *ctx, struct radeon_bo *bo)
  34 {
  35         void *ptr;
  36
  37         ptr = realloc(ctx->bo, sizeof(struct radeon_bo) * (ctx->nbo + 1));
  38         if (ptr == NULL) {
  39                 return -ENOMEM;
  40         }
  41         ctx->bo = ptr;
  42         ctx->bo[ctx->nbo] = bo;
  43         ctx->nbo++;
  44         return 0;
  45 }
  46
  47 struct radeon_bo *radeon_ctx_get_bo(struct radeon_ctx *ctx, unsigned reloc)
  48 {
  49         struct radeon_cs_reloc *greloc;
  50         unsigned i;
  51
  52         greloc = (void *)(((u8 *)ctx->reloc) + reloc * 4);
  53         for (i = 0; i < ctx->nbo; i++) {
  54                 if (ctx->bo[i]->handle == greloc->handle) {
  55                         return radeon_bo_incref(ctx->radeon, ctx->bo[i]);
  56                 }
  57         }
  58         fprintf(stderr, "%s no bo for reloc[%d 0x%08X] %d\n", __func__, reloc, greloc->handle, ctx->nbo);
  59         return NULL;
  60 }
  61
  62 void radeon_ctx_get_placement(struct radeon_ctx *ctx, unsigned reloc, u32 *placement)
  63 {
  64         struct radeon_cs_reloc *greloc;
  65         unsigned i;
  66
  67         placement[0] = 0;
  68         placement[1] = 0;
  69         greloc = (void *)(((u8 *)ctx->reloc) + reloc * 4);
  70         for (i = 0; i < ctx->nbo; i++) {
  71                 if (ctx->bo[i]->handle == greloc->handle) {
  72                         placement[0] = greloc->read_domain | greloc->write_domain;
  73                         placement[1] = placement[0];
  74                         return;
  75                 }
  76         }
  77 }
  78
  79 struct radeon_ctx *radeon_ctx(struct radeon *radeon)
  80 {
  81         struct radeon_ctx *ctx;
  82
  83         if (radeon == NULL)
  84                 return NULL;
  85         ctx = calloc(1, sizeof(*ctx));
  86         if (ctx == NULL)
  87                 return NULL;
  88         ctx->radeon = radeon_incref(radeon);
  89         return ctx;
  90 }
  91
  92 struct radeon_ctx *radeon_ctx_incref(struct radeon_ctx *ctx)
  93 {
  94         ctx->refcount++;
  95         return ctx;
  96 }
  97
  98 struct radeon_ctx *radeon_ctx_decref(struct radeon_ctx *ctx)
  99 {
 100         unsigned i;
 101
 102         if (ctx == NULL)
 103                 return NULL;
 104         if (--ctx->refcount > 0) {
 105                 return NULL;
 106         }
 107
 108         for (i = 0; i < ctx->ndraw; i++) {
 109                 ctx->draw[i] = radeon_draw_decref(ctx->draw[i]);
 110         }
 111         for (i = 0; i < ctx->nbo; i++) {
 112                 ctx->bo[i] = radeon_bo_decref(ctx->radeon, ctx->bo[i]);
 113         }
 114         ctx->radeon = radeon_decref(ctx->radeon);
 115         free(ctx->draw);
 116         free(ctx->bo);
 117         free(ctx->pm4);
 118         free(ctx->reloc);
 119         memset(ctx, 0, sizeof(*ctx));
 120         free(ctx);
 121         return NULL;
 122 }
 123
 124 static int radeon_ctx_state_bo(struct radeon_ctx *ctx, struct radeon_state *state)
 125 {
 126         unsigned i, j;
 127         int r;
 128
 129         if (state == NULL)
 130                 return 0;
 131         for (i = 0; i < state->nbo; i++) {
 132                 for (j = 0; j < ctx->nbo; j++) {
 133                         if (state->bo[i] == ctx->bo[j])
 134                                 break;
 135                 }
 136                 if (j == ctx->nbo) {
 137                         radeon_bo_incref(ctx->radeon, state->bo[i]);
 138                         r = radeon_ctx_set_bo_new(ctx, state->bo[i]);
 139                         if (r)
 140                                 return r;
 141                 }
 142         }
 143         return 0;
 144 }
 145
 146
 147 int radeon_ctx_submit(struct radeon_ctx *ctx)
 148 {
 149         struct drm_radeon_cs drmib;
 150         struct drm_radeon_cs_chunk chunks[2];
 151         uint64_t chunk_array[2];
 152         int r = 0;
 153
 154 #if 0
 155         for (r = 0; r < ctx->cpm4; r++) {
 156                 fprintf(stderr, "0x%08X\n", ctx->pm4[r]);
 157         }
 158 #endif
 159         drmib.num_chunks = 2;
 160         drmib.chunks = (uint64_t)(uintptr_t)chunk_array;
 161         chunks[0].chunk_id = RADEON_CHUNK_ID_IB;
 162         chunks[0].length_dw = ctx->cpm4;
 163         chunks[0].chunk_data = (uint64_t)(uintptr_t)ctx->pm4;
 164         chunks[1].chunk_id = RADEON_CHUNK_ID_RELOCS;
 165         chunks[1].length_dw = ctx->nreloc * sizeof(struct radeon_cs_reloc) / 4;
 166         chunks[1].chunk_data = (uint64_t)(uintptr_t)ctx->reloc;
 167         chunk_array[0] = (uint64_t)(uintptr_t)&chunks[0];
 168         chunk_array[1] = (uint64_t)(uintptr_t)&chunks[1];
 169 #if 1
 170         r = drmCommandWriteRead(ctx->radeon->fd, DRM_RADEON_CS, &drmib,
 171                                 sizeof(struct drm_radeon_cs));
 172 #endif
 173         return r;
 174 }
 175
 176 static int radeon_ctx_reloc(struct radeon_ctx *ctx, struct radeon_bo *bo,
 177                         unsigned id, unsigned *placement)
 178 {
 179         unsigned i;
 180         struct radeon_cs_reloc *ptr;
 181
 182         for (i = 0; i < ctx->nreloc; i++) {
 183                 if (ctx->reloc[i].handle == bo->handle) {
 184                         ctx->pm4[id] = i * sizeof(struct radeon_cs_reloc) / 4;
 185                         return 0;
 186                 }
 187         }
 188         ptr = realloc(ctx->reloc, sizeof(struct radeon_cs_reloc) * (ctx->nreloc + 1));
 189         if (ptr == NULL)
 190                 return -ENOMEM;
 191         ctx->reloc = ptr;
 192         ptr[ctx->nreloc].handle = bo->handle;
 193         ptr[ctx->nreloc].read_domain = placement[0] | placement [1];
 194         ptr[ctx->nreloc].write_domain = placement[0] | placement [1];
 195         ptr[ctx->nreloc].flags = 0;
 196         ctx->pm4[id] = ctx->nreloc * sizeof(struct radeon_cs_reloc) / 4;
 197         ctx->nreloc++;
 198         return 0;
 199 }
 200
 201 static int radeon_ctx_state_schedule(struct radeon_ctx *ctx, struct radeon_state *state)
 202 {
 203         unsigned i, rid, bid, cid;
 204         int r;
 205
 206         if (state == NULL)
 207                 return 0;
 208         memcpy(&ctx->pm4[ctx->id], state->pm4, state->cpm4 * 4);
 209         for (i = 0; i < state->nreloc; i++) {
 210                 rid = state->reloc_pm4_id[i];
 211                 bid = state->reloc_bo_id[i];
 212                 cid = ctx->id + rid;
 213                 r = radeon_ctx_reloc(ctx, state->bo[bid], cid,
 214                                         &state->placement[bid * 2]);
 215                 if (r) {
 216                         fprintf(stderr, "%s state %d failed to reloc\n", __func__, state->type);
 217                         return r;
 218                 }
 219         }
 220         ctx->id += state->cpm4;
 221         return 0;
 222 }
 223
 224 int radeon_ctx_set_draw_new(struct radeon_ctx *ctx, struct radeon_draw *draw)
 225 {
 226         struct radeon_draw *pdraw = NULL;
 227         struct radeon_draw **ndraw;
 228         struct radeon_state *nstate, *ostate;
 229         unsigned cpm4, i, cstate;
 230         void *tmp;
 231         int r = 0;
 232
 233         ndraw = realloc(ctx->draw, sizeof(void*) * (ctx->ndraw + 1));
 234         if (ndraw == NULL)
 235                 return -ENOMEM;
 236         ctx->draw = ndraw;
 237         for (i = 0; i < draw->nstate; i++) {
 238                 r = radeon_ctx_state_bo(ctx, draw->state[i]);
 239                 if (r)
 240                         return r;
 241         }
 242         r = radeon_draw_check(draw);
 243         if (r)
 244                 return r;
 245         if (draw->cpm4 >= RADEON_CTX_MAX_PM4) {
 246                 fprintf(stderr, "%s single draw too big %d, max %d\n",
 247                         __func__, draw->cpm4, RADEON_CTX_MAX_PM4);
 248                 return -EINVAL;
 249         }
 250         tmp = realloc(ctx->state, (ctx->nstate + draw->nstate) * sizeof(void*));
 251         if (tmp == NULL)
 252                 return -ENOMEM;
 253         ctx->state = tmp;
 254         pdraw = ctx->cdraw;
 255         for (i = 0, cpm4 = 0, cstate = ctx->nstate; i < draw->nstate - 1; i++) {
 256                 nstate = draw->state[i];
 257                 if (nstate) {
 258                         if (pdraw && pdraw->state[i]) {
 259                                 ostate = pdraw->state[i];
 260                                 if (ostate->pm4_crc != nstate->pm4_crc) {
 261                                         ctx->state[cstate++] = nstate;
 262                                         cpm4 += nstate->cpm4;
 263                                 }
 264                         } else {
 265                                 ctx->state[cstate++] = nstate;
 266                                 cpm4 += nstate->cpm4;
 267                         }
 268                 }
 269         }
 270         /* The last state is the draw state always add it */
 271         if (draw->state[i] == NULL) {
 272                 fprintf(stderr, "%s no draw command\n", __func__);
 273                 return -EINVAL;
 274         }
 275         ctx->state[cstate++] = draw->state[i];
 276         cpm4 += draw->state[i]->cpm4;
 277         if ((ctx->draw_cpm4 + cpm4) > RADEON_CTX_MAX_PM4) {
 278                 /* need to flush */
 279                 return -EBUSY;
 280         }
 281         ctx->draw_cpm4 += cpm4;
 282         ctx->nstate = cstate;
 283         ctx->draw[ctx->ndraw++] = draw;
 284         ctx->cdraw = draw;
 285         return 0;
 286 }
 287
 288 int radeon_ctx_set_draw(struct radeon_ctx *ctx, struct radeon_draw *draw)
 289 {
 290         int r;
 291
 292         radeon_draw_incref(draw);
 293         r = radeon_ctx_set_draw_new(ctx, draw);
 294         if (r)
 295                 radeon_draw_decref(draw);
 296         return r;
 297 }
 298
 299 int radeon_ctx_pm4(struct radeon_ctx *ctx)
 300 {
 301         unsigned i;
 302         int r;
 303
 304         free(ctx->pm4);
 305         ctx->cpm4 = 0;
 306         ctx->pm4 = malloc(ctx->draw_cpm4 * 4);
 307         if (ctx->pm4 == NULL)
 308                 return -EINVAL;
 309         for (i = 0, ctx->id = 0; i < ctx->nstate; i++) {
 310                 r = radeon_ctx_state_schedule(ctx, ctx->state[i]);
 311                 if (r)
 312                         return r;
 313         }
 314         if (ctx->id != ctx->draw_cpm4) {
 315                 fprintf(stderr, "%s miss predicted pm4 size %d for %d\n",
 316                         __func__, ctx->draw_cpm4, ctx->id);
 317                 return -EINVAL;
 318         }
 319         ctx->cpm4 = ctx->draw_cpm4;
 320         return 0;
 321 }
 322
 323 void radeon_ctx_dump_bof(struct radeon_ctx *ctx, const char *file)
 324 {
 325         bof_t *bcs, *blob, *array, *bo, *size, *handle, *device_id, *root;
 326         unsigned i;
 327
 328         root = device_id = bcs = blob = array = bo = size = handle = NULL;
 329         root = bof_object();
 330         if (root == NULL)
 331                 goto out_err;
 332         device_id = bof_int32(ctx->radeon->device);
 333         if (device_id == NULL)
 334                 return;
 335         if (bof_object_set(root, "device_id", device_id))
 336                 goto out_err;
 337         bof_decref(device_id);
 338         device_id = NULL;
 339         /* dump relocs */
 340 printf("%d relocs\n", ctx->nreloc);
 341         blob = bof_blob(ctx->nreloc * 16, ctx->reloc);
 342         if (blob == NULL)
 343                 goto out_err;
 344         if (bof_object_set(root, "reloc", blob))
 345                 goto out_err;
 346         bof_decref(blob);
 347         blob = NULL;
 348         /* dump cs */
 349 printf("%d pm4\n", ctx->cpm4);
 350         blob = bof_blob(ctx->cpm4 * 4, ctx->pm4);
 351         if (blob == NULL)
 352                 goto out_err;
 353         if (bof_object_set(root, "pm4", blob))
 354                 goto out_err;
 355         bof_decref(blob);
 356         blob = NULL;
 357         /* dump bo */
 358         array = bof_array();
 359         if (array == NULL)
 360                 goto out_err;
 361         for (i = 0; i < ctx->nbo; i++) {
 362                 bo = bof_object();
 363                 if (bo == NULL)
 364                         goto out_err;
 365                 size = bof_int32(ctx->bo[i]->size);
 366 printf("[%d] %d bo\n", i, size);
 367                 if (size == NULL)
 368                         goto out_err;
 369                 if (bof_object_set(bo, "size", size))
 370                         goto out_err;
 371                 bof_decref(size);
 372                 size = NULL;
 373                 handle = bof_int32(ctx->bo[i]->handle);
 374                 if (handle == NULL)
 375                         goto out_err;
 376                 if (bof_object_set(bo, "handle", handle))
 377                         goto out_err;
 378                 bof_decref(handle);
 379                 handle = NULL;
 380                 radeon_bo_map(ctx->radeon, ctx->bo[i]);
 381                 blob = bof_blob(ctx->bo[i]->size, ctx->bo[i]->data);
 382                 radeon_bo_unmap(ctx->radeon, ctx->bo[i]);
 383                 if (blob == NULL)
 384                         goto out_err;
 385                 if (bof_object_set(bo, "data", blob))
 386                         goto out_err;
 387                 bof_decref(blob);
 388                 blob = NULL;
 389                 if (bof_array_append(array, bo))
 390                         goto out_err;
 391                 bof_decref(bo);
 392                 bo = NULL;
 393         }
 394         if (bof_object_set(root, "bo", array))
 395                 goto out_err;
 396         bof_dump_file(root, file);
 397 printf("done dump\n");
 398 out_err:
 399         bof_decref(blob);
 400         bof_decref(array);
 401         bof_decref(bo);
 402         bof_decref(size);
 403         bof_decref(handle);
 404         bof_decref(device_id);
 405         bof_decref(root);
 406 }