src/gallium/drivers/radeonsi/si_pipe.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 #include <stdio.h>
  24 #include <errno.h>
  25 #include "pipe/p_defines.h"
  26 #include "pipe/p_state.h"
  27 #include "pipe/p_context.h"
  28 #include "tgsi/tgsi_scan.h"
  29 #include "tgsi/tgsi_parse.h"
  30 #include "tgsi/tgsi_util.h"
  31 #include "util/u_blitter.h"
  32 #include "util/u_double_list.h"
  33 #include "util/u_format.h"
  34 #include "util/u_transfer.h"
  35 #include "util/u_surface.h"
  36 #include "util/u_pack_color.h"
  37 #include "util/u_memory.h"
  38 #include "util/u_inlines.h"
  39 #include "util/u_simple_shaders.h"
  40 #include "util/u_upload_mgr.h"
  41 #include "vl/vl_decoder.h"
  42 #include "vl/vl_video_buffer.h"
  43 #include "os/os_time.h"
  44 #include "pipebuffer/pb_buffer.h"
  45 #include "si_pipe.h"
  46 #include "radeon/radeon_uvd.h"
  47 #include "si.h"
  48 #include "sid.h"
  49 #include "si_resource.h"
  50 #include "si_pipe.h"
  51 #include "si_state.h"
  52 #include "../radeon/r600_cs.h"
  53
  54 /*
  55  * pipe_context
  56  */
  57 void si_flush(struct pipe_context *ctx, struct pipe_fence_handle **fence,
  58               unsigned flags)
  59 {
  60         struct si_context *sctx = (struct si_context *)ctx;
  61         struct pipe_query *render_cond = NULL;
  62         boolean render_cond_cond = FALSE;
  63         unsigned render_cond_mode = 0;
  64
  65         if (fence) {
  66                 *fence = sctx->b.ws->cs_create_fence(sctx->b.rings.gfx.cs);
  67         }
  68
  69         /* Disable render condition. */
  70         if (sctx->b.current_render_cond) {
  71                 render_cond = sctx->b.current_render_cond;
  72                 render_cond_cond = sctx->b.current_render_cond_cond;
  73                 render_cond_mode = sctx->b.current_render_cond_mode;
  74                 ctx->render_condition(ctx, NULL, FALSE, 0);
  75         }
  76
  77         si_context_flush(sctx, flags);
  78
  79         /* Re-enable render condition. */
  80         if (render_cond) {
  81                 ctx->render_condition(ctx, render_cond, render_cond_cond, render_cond_mode);
  82         }
  83 }
  84
  85 static void si_flush_from_st(struct pipe_context *ctx,
  86                              struct pipe_fence_handle **fence,
  87                              unsigned flags)
  88 {
  89         si_flush(ctx, fence,
  90                  flags & PIPE_FLUSH_END_OF_FRAME ? RADEON_FLUSH_END_OF_FRAME : 0);
  91 }
  92
  93 static void si_flush_from_winsys(void *ctx, unsigned flags)
  94 {
  95         si_flush((struct pipe_context*)ctx, NULL, flags);
  96 }
  97
  98 static void si_destroy_context(struct pipe_context *context)
  99 {
 100         struct si_context *sctx = (struct si_context *)context;
 101
 102         si_release_all_descriptors(sctx);
 103
 104         pipe_resource_reference(&sctx->null_const_buf.buffer, NULL);
 105         r600_resource_reference(&sctx->border_color_table, NULL);
 106
 107         if (sctx->dummy_pixel_shader) {
 108                 sctx->b.b.delete_fs_state(&sctx->b.b, sctx->dummy_pixel_shader);
 109         }
 110         for (int i = 0; i < 8; i++) {
 111                 sctx->b.b.delete_depth_stencil_alpha_state(&sctx->b.b, sctx->custom_dsa_flush_depth_stencil[i]);
 112                 sctx->b.b.delete_depth_stencil_alpha_state(&sctx->b.b, sctx->custom_dsa_flush_depth[i]);
 113                 sctx->b.b.delete_depth_stencil_alpha_state(&sctx->b.b, sctx->custom_dsa_flush_stencil[i]);
 114         }
 115         sctx->b.b.delete_depth_stencil_alpha_state(&sctx->b.b, sctx->custom_dsa_flush_inplace);
 116         sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_resolve);
 117         sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_decompress);
 118         util_unreference_framebuffer_state(&sctx->framebuffer);
 119
 120         util_blitter_destroy(sctx->blitter);
 121
 122         r600_common_context_cleanup(&sctx->b);
 123         FREE(sctx);
 124 }
 125
 126 static struct pipe_context *si_create_context(struct pipe_screen *screen, void *priv)
 127 {
 128         struct si_context *sctx = CALLOC_STRUCT(si_context);
 129         struct si_screen* sscreen = (struct si_screen *)screen;
 130         int shader, i;
 131
 132         if (sctx == NULL)
 133                 return NULL;
 134
 135         sctx->b.b.screen = screen; /* this must be set first */
 136         sctx->b.b.priv = priv;
 137         sctx->b.b.destroy = si_destroy_context;
 138         sctx->b.b.flush = si_flush_from_st;
 139         sctx->screen = sscreen; /* Easy accessing of screen/winsys. */
 140
 141         if (!r600_common_context_init(&sctx->b, &sscreen->b))
 142                 goto fail;
 143
 144         si_init_blit_functions(sctx);
 145         si_init_compute_functions(sctx);
 146
 147         if (sscreen->b.info.has_uvd) {
 148                 sctx->b.b.create_video_codec = si_uvd_create_decoder;
 149                 sctx->b.b.create_video_buffer = si_video_buffer_create;
 150         } else {
 151                 sctx->b.b.create_video_codec = vl_create_decoder;
 152                 sctx->b.b.create_video_buffer = vl_video_buffer_create;
 153         }
 154
 155         sctx->b.rings.gfx.cs = sctx->b.ws->cs_create(sctx->b.ws, RING_GFX, NULL);
 156         sctx->b.rings.gfx.flush = si_flush_from_winsys;
 157
 158         si_init_all_descriptors(sctx);
 159
 160         /* Initialize cache_flush. */
 161         sctx->cache_flush = si_atom_cache_flush;
 162         sctx->atoms.cache_flush = &sctx->cache_flush;
 163
 164         sctx->atoms.streamout_begin = &sctx->b.streamout.begin_atom;
 165
 166         switch (sctx->b.chip_class) {
 167         case SI:
 168         case CIK:
 169                 si_init_state_functions(sctx);
 170                 si_init_config(sctx);
 171                 break;
 172         default:
 173                 R600_ERR("Unsupported chip class %d.\n", sctx->b.chip_class);
 174                 goto fail;
 175         }
 176
 177         sctx->b.ws->cs_set_flush_callback(sctx->b.rings.gfx.cs, si_flush_from_winsys, sctx);
 178
 179         sctx->blitter = util_blitter_create(&sctx->b.b);
 180         if (sctx->blitter == NULL)
 181                 goto fail;
 182
 183         sctx->dummy_pixel_shader =
 184                 util_make_fragment_cloneinput_shader(&sctx->b.b, 0,
 185                                                      TGSI_SEMANTIC_GENERIC,
 186                                                      TGSI_INTERPOLATE_CONSTANT);
 187         sctx->b.b.bind_fs_state(&sctx->b.b, sctx->dummy_pixel_shader);
 188
 189         /* these must be last */
 190         si_begin_new_cs(sctx);
 191         r600_query_init_backend_mask(&sctx->b); /* this emits commands and must be last */
 192
 193         /* CIK cannot unbind a constant buffer (S_BUFFER_LOAD is buggy
 194          * with a NULL buffer). We need to use a dummy buffer instead. */
 195         if (sctx->b.chip_class == CIK) {
 196                 sctx->null_const_buf.buffer = pipe_buffer_create(screen, PIPE_BIND_CONSTANT_BUFFER,
 197                                                                  PIPE_USAGE_STATIC, 16);
 198                 sctx->null_const_buf.buffer_size = sctx->null_const_buf.buffer->width0;
 199
 200                 for (shader = 0; shader < SI_NUM_SHADERS; shader++) {
 201                         for (i = 0; i < NUM_CONST_BUFFERS; i++) {
 202                                 sctx->b.b.set_constant_buffer(&sctx->b.b, shader, i,
 203                                                               &sctx->null_const_buf);
 204                         }
 205                 }
 206
 207                 /* Clear the NULL constant buffer, because loads should return zeros. */
 208                 sctx->b.clear_buffer(&sctx->b.b, sctx->null_const_buf.buffer, 0,
 209                                      sctx->null_const_buf.buffer->width0, 0);
 210         }
 211
 212         return &sctx->b.b;
 213 fail:
 214         si_destroy_context(&sctx->b.b);
 215         return NULL;
 216 }
 217
 218 /*
 219  * pipe_screen
 220  */
 221
 222 static int si_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
 223 {
 224         struct si_screen *sscreen = (struct si_screen *)pscreen;
 225
 226         switch (param) {
 227         /* Supported features (boolean caps). */
 228         case PIPE_CAP_TWO_SIDED_STENCIL:
 229         case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
 230         case PIPE_CAP_ANISOTROPIC_FILTER:
 231         case PIPE_CAP_POINT_SPRITE:
 232         case PIPE_CAP_OCCLUSION_QUERY:
 233         case PIPE_CAP_TEXTURE_SHADOW_MAP:
 234         case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
 235         case PIPE_CAP_BLEND_EQUATION_SEPARATE:
 236         case PIPE_CAP_TEXTURE_SWIZZLE:
 237         case PIPE_CAP_DEPTH_CLIP_DISABLE:
 238         case PIPE_CAP_SHADER_STENCIL_EXPORT:
 239         case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
 240         case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
 241         case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
 242         case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
 243         case PIPE_CAP_SM3:
 244         case PIPE_CAP_SEAMLESS_CUBE_MAP:
 245         case PIPE_CAP_PRIMITIVE_RESTART:
 246         case PIPE_CAP_CONDITIONAL_RENDER:
 247         case PIPE_CAP_TEXTURE_BARRIER:
 248         case PIPE_CAP_INDEP_BLEND_ENABLE:
 249         case PIPE_CAP_INDEP_BLEND_FUNC:
 250         case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
 251         case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
 252         case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
 253         case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
 254         case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
 255         case PIPE_CAP_USER_INDEX_BUFFERS:
 256         case PIPE_CAP_USER_CONSTANT_BUFFERS:
 257         case PIPE_CAP_START_INSTANCE:
 258         case PIPE_CAP_NPOT_TEXTURES:
 259         case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
 260         case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
 261         case PIPE_CAP_TGSI_INSTANCEID:
 262         case PIPE_CAP_COMPUTE:
 263         case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
 264         case PIPE_CAP_TGSI_VS_LAYER:
 265         case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
 266                 return 1;
 267
 268         case PIPE_CAP_TEXTURE_MULTISAMPLE:
 269                 /* 2D tiling on CIK is supported since DRM 2.35.0 */
 270                 return HAVE_LLVM >= 0x0304 && (sscreen->b.chip_class < CIK ||
 271                                                sscreen->b.info.drm_minor >= 35);
 272
 273         case PIPE_CAP_TGSI_TEXCOORD:
 274                 return 0;
 275
 276         case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
 277                 return 64;
 278
 279         case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
 280                 return 256;
 281
 282         case PIPE_CAP_GLSL_FEATURE_LEVEL:
 283                 return 140;
 284
 285         case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
 286                 return 1;
 287         case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
 288                 return MIN2(sscreen->b.info.vram_size, 0xFFFFFFFF);
 289
 290         /* Unsupported features. */
 291         case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
 292         case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
 293         case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
 294         case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
 295         case PIPE_CAP_VERTEX_COLOR_CLAMPED:
 296         case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
 297         case PIPE_CAP_USER_VERTEX_BUFFERS:
 298         case PIPE_CAP_CUBE_MAP_ARRAY:
 299                 return 0;
 300
 301         case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
 302                 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_R600;
 303
 304         /* Stream output. */
 305         case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
 306                 return sscreen->b.has_streamout ? 4 : 0;
 307         case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
 308                 return sscreen->b.has_streamout ? 1 : 0;
 309         case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
 310         case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
 311                 return sscreen->b.has_streamout ? 32*4 : 0;
 312
 313         /* Texturing. */
 314         case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
 315         case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
 316         case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
 317                         return 15;
 318         case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
 319                 return 16384;
 320         case PIPE_CAP_MAX_COMBINED_SAMPLERS:
 321                 return 32;
 322
 323         /* Render targets. */
 324         case PIPE_CAP_MAX_RENDER_TARGETS:
 325                 return 8;
 326
 327         case PIPE_CAP_MAX_VIEWPORTS:
 328                 return 1;
 329
 330         /* Timer queries, present when the clock frequency is non zero. */
 331         case PIPE_CAP_QUERY_TIMESTAMP:
 332         case PIPE_CAP_QUERY_TIME_ELAPSED:
 333                 return sscreen->b.info.r600_clock_crystal_freq != 0;
 334
 335         case PIPE_CAP_MIN_TEXEL_OFFSET:
 336                 return -8;
 337
 338         case PIPE_CAP_MAX_TEXEL_OFFSET:
 339                 return 7;
 340         case PIPE_CAP_ENDIANNESS:
 341                 return PIPE_ENDIAN_LITTLE;
 342         }
 343         return 0;
 344 }
 345
 346 static int si_get_shader_param(struct pipe_screen* pscreen, unsigned shader, enum pipe_shader_cap param)
 347 {
 348         switch(shader)
 349         {
 350         case PIPE_SHADER_FRAGMENT:
 351         case PIPE_SHADER_VERTEX:
 352                 break;
 353         case PIPE_SHADER_GEOMETRY:
 354                 /* TODO: support and enable geometry programs */
 355                 return 0;
 356         case PIPE_SHADER_COMPUTE:
 357                 switch (param) {
 358                 case PIPE_SHADER_CAP_PREFERRED_IR:
 359                         return PIPE_SHADER_IR_LLVM;
 360                 default:
 361                         return 0;
 362                 }
 363         default:
 364                 /* TODO: support tessellation */
 365                 return 0;
 366         }
 367
 368         switch (param) {
 369         case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 370         case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 371         case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 372         case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 373                 return 16384;
 374         case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 375                 return 32;
 376         case PIPE_SHADER_CAP_MAX_INPUTS:
 377                 return 32;
 378         case PIPE_SHADER_CAP_MAX_TEMPS:
 379                 return 256; /* Max native temporaries. */
 380         case PIPE_SHADER_CAP_MAX_ADDRS:
 381                 /* FIXME Isn't this equal to TEMPS? */
 382                 return 1; /* Max native address registers */
 383         case PIPE_SHADER_CAP_MAX_CONSTS:
 384                 return 4096; /* actually only memory limits this */
 385         case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 386                 return NUM_PIPE_CONST_BUFFERS;
 387         case PIPE_SHADER_CAP_MAX_PREDS:
 388                 return 0; /* FIXME */
 389         case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 390                 return 1;
 391         case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
 392                 return 0;
 393         case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 394         case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 395         case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 396         case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 397                 return 1;
 398         case PIPE_SHADER_CAP_INTEGERS:
 399                 return 1;
 400         case PIPE_SHADER_CAP_SUBROUTINES:
 401                 return 0;
 402         case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 403         case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
 404                 return 16;
 405         case PIPE_SHADER_CAP_PREFERRED_IR:
 406                 return PIPE_SHADER_IR_TGSI;
 407         }
 408         return 0;
 409 }
 410
 411 static void si_destroy_screen(struct pipe_screen* pscreen)
 412 {
 413         struct si_screen *sscreen = (struct si_screen *)pscreen;
 414
 415         if (sscreen == NULL)
 416                 return;
 417
 418         if (!radeon_winsys_unref(sscreen->b.ws))
 419                 return;
 420
 421         r600_common_screen_cleanup(&sscreen->b);
 422
 423 #if SI_TRACE_CS
 424         if (sscreen->b.trace_bo) {
 425                 sscreen->ws->buffer_unmap(sscreen->b.trace_bo->cs_buf);
 426                 pipe_resource_reference((struct pipe_resource**)&sscreen->b.trace_bo, NULL);
 427         }
 428 #endif
 429
 430         sscreen->b.ws->destroy(sscreen->b.ws);
 431         FREE(sscreen);
 432 }
 433
 434 struct pipe_screen *radeonsi_screen_create(struct radeon_winsys *ws)
 435 {
 436         struct si_screen *sscreen = CALLOC_STRUCT(si_screen);
 437         if (sscreen == NULL) {
 438                 return NULL;
 439         }
 440
 441         /* Set functions first. */
 442         sscreen->b.b.context_create = si_create_context;
 443         sscreen->b.b.destroy = si_destroy_screen;
 444         sscreen->b.b.get_param = si_get_param;
 445         sscreen->b.b.get_shader_param = si_get_shader_param;
 446         sscreen->b.b.is_format_supported = si_is_format_supported;
 447
 448         if (!r600_common_screen_init(&sscreen->b, ws)) {
 449                 FREE(sscreen);
 450                 return NULL;
 451         }
 452
 453         sscreen->b.has_cp_dma = true;
 454         sscreen->b.has_streamout = HAVE_LLVM >= 0x0304;
 455
 456         if (debug_get_bool_option("RADEON_DUMP_SHADERS", FALSE))
 457                 sscreen->b.debug_flags |= DBG_FS | DBG_VS | DBG_GS | DBG_PS | DBG_CS;
 458
 459 #if SI_TRACE_CS
 460         sscreen->b.cs_count = 0;
 461         if (sscreen->info.drm_minor >= 28) {
 462                 sscreen->b.trace_bo = (struct r600_resource*)pipe_buffer_create(&sscreen->screen,
 463                                                                                 PIPE_BIND_CUSTOM,
 464                                                                                 PIPE_USAGE_STAGING,
 465                                                                                 4096);
 466                 if (sscreen->b.trace_bo) {
 467                         sscreen->b.trace_ptr = sscreen->ws->buffer_map(sscreen->b.trace_bo->cs_buf, NULL,
 468                                                                         PIPE_TRANSFER_UNSYNCHRONIZED);
 469                 }
 470         }
 471 #endif
 472
 473         /* Create the auxiliary context. This must be done last. */
 474         sscreen->b.aux_context = sscreen->b.b.context_create(&sscreen->b.b, NULL);
 475
 476         return &sscreen->b.b;
 477 }