src/gallium/drivers/vc4/vc4_context.c

   1 /*
   2  * Copyright © 2014 Broadcom
   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  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21  * IN THE SOFTWARE.
  22  */
  23
  24 #include <xf86drm.h>
  25 #include <err.h>
  26 #include <stdio.h>
  27
  28 #include "pipe/p_defines.h"
  29 #include "util/u_inlines.h"
  30 #include "util/u_memory.h"
  31 #include "util/u_blitter.h"
  32 #include "indices/u_primconvert.h"
  33 #include "pipe/p_screen.h"
  34
  35 #include "vc4_screen.h"
  36 #include "vc4_context.h"
  37 #include "vc4_resource.h"
  38
  39 static void
  40 dump_fbo(struct vc4_context *vc4, struct vc4_bo *fbo)
  41 {
  42 #ifndef USE_VC4_SIMULATOR
  43         uint32_t *map = vc4_bo_map(fbo);
  44         uint32_t width = vc4->framebuffer.width;
  45         uint32_t height = vc4->framebuffer.height;
  46         uint32_t chunk_w = width / 79;
  47         uint32_t chunk_h = height / 40;
  48         uint32_t found_colors[10];
  49         uint32_t num_found_colors = 0;
  50
  51         for (int by = 0; by < height; by += chunk_h) {
  52                 for (int bx = 0; bx < width; bx += chunk_w) {
  53                         bool on = false, black = false;
  54
  55                         for (int y = by; y < MIN2(height, by + chunk_h); y++) {
  56                                 for (int x = bx; x < MIN2(width, bx + chunk_w); x++) {
  57                                         uint32_t pix = map[y * width + x];
  58                                         on |= pix != 0;
  59                                         black |= pix == 0xff000000;
  60
  61                                         int i;
  62                                         for (i = 0; i < num_found_colors; i++) {
  63                                                 if (pix == found_colors[i])
  64                                                         break;
  65                                         }
  66                                         if (i == num_found_colors &&
  67                                             num_found_colors < Elements(found_colors))
  68                                                 found_colors[num_found_colors++] = pix;
  69                                 }
  70                         }
  71                         if (black)
  72                                 fprintf(stderr, "O");
  73                         else if (on)
  74                                 fprintf(stderr, "X");
  75                         else
  76                                 fprintf(stderr, ".");
  77                 }
  78                 fprintf(stderr, "\n");
  79         }
  80
  81         for (int i = 0; i < num_found_colors; i++) {
  82                 fprintf(stderr, "color %d: 0x%08x\n", i, found_colors[i]);
  83         }
  84 #endif
  85 }
  86
  87 static void
  88 vc4_setup_rcl(struct vc4_context *vc4)
  89 {
  90         struct vc4_surface *csurf = vc4_surface(vc4->framebuffer.cbufs[0]);
  91         struct vc4_resource *ctex = vc4_resource(csurf->base.texture);
  92         uint32_t resolve_uncleared = vc4->resolve & ~vc4->cleared;
  93         uint32_t width = vc4->framebuffer.width;
  94         uint32_t height = vc4->framebuffer.height;
  95         uint32_t xtiles = align(width, 64) / 64;
  96         uint32_t ytiles = align(height, 64) / 64;
  97
  98 #if 0
  99         fprintf(stderr, "RCL: resolve 0x%x clear 0x%x resolve uncleared 0x%x\n",
 100                 vc4->resolve,
 101                 vc4->cleared,
 102                 resolve_uncleared);
 103 #endif
 104
 105         cl_u8(&vc4->rcl, VC4_PACKET_CLEAR_COLORS);
 106         cl_u32(&vc4->rcl, vc4->clear_color[0]);
 107         cl_u32(&vc4->rcl, vc4->clear_color[1]);
 108         cl_u32(&vc4->rcl, 0);
 109         cl_u8(&vc4->rcl, 0);
 110
 111         cl_start_reloc(&vc4->rcl, 1);
 112         cl_u8(&vc4->rcl, VC4_PACKET_TILE_RENDERING_MODE_CONFIG);
 113         cl_reloc(vc4, &vc4->rcl, ctex->bo, csurf->offset);
 114         cl_u16(&vc4->rcl, width);
 115         cl_u16(&vc4->rcl, height);
 116         cl_u8(&vc4->rcl, (VC4_RENDER_CONFIG_MEMORY_FORMAT_LINEAR |
 117                           VC4_RENDER_CONFIG_FORMAT_RGBA8888));
 118         cl_u8(&vc4->rcl, 0);
 119
 120         /* The tile buffer normally gets cleared when the previous tile is
 121          * stored.  If the clear values changed between frames, then the tile
 122          * buffer has stale clear values in it, so we have to do a store in
 123          * None mode (no writes) so that we trigger the tile buffer clear.
 124          */
 125         if (vc4->cleared & PIPE_CLEAR_COLOR0) {
 126                 cl_u8(&vc4->rcl, VC4_PACKET_TILE_COORDINATES);
 127                 cl_u8(&vc4->rcl, 0);
 128                 cl_u8(&vc4->rcl, 0);
 129
 130                 cl_u8(&vc4->rcl, VC4_PACKET_STORE_TILE_BUFFER_GENERAL);
 131                 cl_u16(&vc4->rcl, VC4_LOADSTORE_TILE_BUFFER_NONE);
 132                 cl_u32(&vc4->rcl, 0); /* no address, since we're in None mode */
 133         }
 134
 135         for (int y = 0; y < ytiles; y++) {
 136                 for (int x = 0; x < xtiles; x++) {
 137                         bool end_of_frame = (x == xtiles - 1 &&
 138                                              y == ytiles - 1);
 139
 140                         /* Note that the load doesn't actually occur until the
 141                          * tile coords packet is processed.
 142                          */
 143                         if (resolve_uncleared & PIPE_CLEAR_COLOR) {
 144                                 cl_start_reloc(&vc4->rcl, 1);
 145                                 cl_u8(&vc4->rcl, VC4_PACKET_LOAD_TILE_BUFFER_GENERAL);
 146                                 cl_u8(&vc4->rcl,
 147                                       VC4_LOADSTORE_TILE_BUFFER_COLOR |
 148                                       VC4_LOADSTORE_TILE_BUFFER_FORMAT_RASTER);
 149                                 cl_u8(&vc4->rcl,
 150                                       VC4_LOADSTORE_TILE_BUFFER_RGBA8888);
 151                                 cl_reloc(vc4, &vc4->rcl, ctex->bo,
 152                                          csurf->offset);
 153                         }
 154
 155                         cl_u8(&vc4->rcl, VC4_PACKET_TILE_COORDINATES);
 156                         cl_u8(&vc4->rcl, x);
 157                         cl_u8(&vc4->rcl, y);
 158
 159                         cl_start_reloc(&vc4->rcl, 1);
 160                         cl_u8(&vc4->rcl, VC4_PACKET_BRANCH_TO_SUB_LIST);
 161                         cl_reloc(vc4, &vc4->rcl, vc4->tile_alloc,
 162                                  (y * xtiles + x) * 32);
 163
 164                         if (vc4->resolve & PIPE_CLEAR_COLOR0) {
 165                                 if (end_of_frame) {
 166                                         cl_u8(&vc4->rcl,
 167                                               VC4_PACKET_STORE_MS_TILE_BUFFER_AND_EOF);
 168                                 } else {
 169                                         cl_u8(&vc4->rcl,
 170                                               VC4_PACKET_STORE_MS_TILE_BUFFER);
 171                                 }
 172                         } else {
 173                                 assert(!"unfinished: Need to end the frame\n");
 174                         }
 175                 }
 176         }
 177 }
 178
 179 void
 180 vc4_flush(struct pipe_context *pctx)
 181 {
 182         struct vc4_context *vc4 = vc4_context(pctx);
 183
 184         if (!vc4->needs_flush)
 185                 return;
 186
 187         cl_u8(&vc4->bcl, VC4_PACKET_FLUSH_ALL);
 188         cl_u8(&vc4->bcl, VC4_PACKET_NOP);
 189         cl_u8(&vc4->bcl, VC4_PACKET_HALT);
 190
 191         vc4_setup_rcl(vc4);
 192
 193         struct vc4_surface *csurf = vc4_surface(vc4->framebuffer.cbufs[0]);
 194         struct vc4_resource *ctex = vc4_resource(csurf->base.texture);
 195
 196         struct drm_vc4_submit_cl submit;
 197         memset(&submit, 0, sizeof(submit));
 198
 199         submit.bo_handles = vc4->bo_handles.base;
 200         submit.bo_handle_count = (vc4->bo_handles.next -
 201                                   vc4->bo_handles.base) / 4;
 202         submit.bin_cl = vc4->bcl.base;
 203         submit.bin_cl_size = vc4->bcl.next - vc4->bcl.base;
 204         submit.render_cl = vc4->rcl.base;
 205         submit.render_cl_size = vc4->rcl.next - vc4->rcl.base;
 206         submit.shader_rec = vc4->shader_rec.base;
 207         submit.shader_rec_size = vc4->shader_rec.next - vc4->shader_rec.base;
 208         submit.shader_rec_count = vc4->shader_rec_count;
 209         submit.uniforms = vc4->uniforms.base;
 210         submit.uniforms_size = vc4->uniforms.next - vc4->uniforms.base;
 211
 212         if (!(vc4_debug & VC4_DEBUG_NORAST)) {
 213                 int ret;
 214
 215 #ifndef USE_VC4_SIMULATOR
 216                 ret = drmIoctl(vc4->fd, DRM_IOCTL_VC4_SUBMIT_CL, &submit);
 217 #else
 218                 ret = vc4_simulator_flush(vc4, &submit, csurf);
 219 #endif
 220                 if (ret)
 221                         errx(1, "VC4 submit failed\n");
 222         }
 223
 224         vc4_reset_cl(&vc4->bcl);
 225         vc4_reset_cl(&vc4->rcl);
 226         vc4_reset_cl(&vc4->shader_rec);
 227         vc4_reset_cl(&vc4->uniforms);
 228         vc4_reset_cl(&vc4->bo_handles);
 229         struct vc4_bo **referenced_bos = vc4->bo_pointers.base;
 230         for (int i = 0; i < submit.bo_handle_count; i++)
 231                 vc4_bo_unreference(&referenced_bos[i]);
 232         vc4_reset_cl(&vc4->bo_pointers);
 233         vc4->shader_rec_count = 0;
 234
 235         vc4->needs_flush = false;
 236         vc4->draw_call_queued = false;
 237         vc4->dirty = ~0;
 238         vc4->resolve = 0;
 239         vc4->cleared = 0;
 240
 241         dump_fbo(vc4, ctex->bo);
 242 }
 243
 244 static void
 245 vc4_pipe_flush(struct pipe_context *pctx, struct pipe_fence_handle **fence,
 246                unsigned flags)
 247 {
 248         vc4_flush(pctx);
 249 }
 250
 251 static void
 252 vc4_context_destroy(struct pipe_context *pctx)
 253 {
 254         struct vc4_context *vc4 = vc4_context(pctx);
 255
 256         if (vc4->blitter)
 257                 util_blitter_destroy(vc4->blitter);
 258
 259         if (vc4->primconvert)
 260                 util_primconvert_destroy(vc4->primconvert);
 261
 262         util_slab_destroy(&vc4->transfer_pool);
 263
 264         free(vc4);
 265 }
 266
 267 struct pipe_context *
 268 vc4_context_create(struct pipe_screen *pscreen, void *priv)
 269 {
 270         struct vc4_screen *screen = vc4_screen(pscreen);
 271         struct vc4_context *vc4;
 272
 273         /* Prevent dumping of the shaders built during context setup. */
 274         uint32_t saved_shaderdb_flag = vc4_debug & VC4_DEBUG_SHADERDB;
 275         vc4_debug &= ~VC4_DEBUG_SHADERDB;
 276
 277         vc4 = CALLOC_STRUCT(vc4_context);
 278         if (vc4 == NULL)
 279                 return NULL;
 280         struct pipe_context *pctx = &vc4->base;
 281
 282         vc4->screen = screen;
 283
 284         pctx->screen = pscreen;
 285         pctx->priv = priv;
 286         pctx->destroy = vc4_context_destroy;
 287         pctx->flush = vc4_pipe_flush;
 288
 289         vc4_draw_init(pctx);
 290         vc4_state_init(pctx);
 291         vc4_program_init(pctx);
 292         vc4_resource_context_init(pctx);
 293
 294         vc4_init_cl(vc4, &vc4->bcl);
 295         vc4_init_cl(vc4, &vc4->rcl);
 296         vc4_init_cl(vc4, &vc4->shader_rec);
 297         vc4_init_cl(vc4, &vc4->bo_handles);
 298
 299         vc4->dirty = ~0;
 300         vc4->fd = screen->fd;
 301
 302         util_slab_create(&vc4->transfer_pool, sizeof(struct pipe_transfer),
 303                          16, UTIL_SLAB_SINGLETHREADED);
 304         vc4->blitter = util_blitter_create(pctx);
 305         if (!vc4->blitter)
 306                 goto fail;
 307
 308         vc4->primconvert = util_primconvert_create(pctx,
 309                                                    !((1 << PIPE_PRIM_QUADS) - 1));
 310         if (!vc4->primconvert)
 311                 goto fail;
 312
 313         vc4_debug |= saved_shaderdb_flag;
 314
 315         return &vc4->base;
 316
 317 fail:
 318         pctx->destroy(pctx);
 319         return NULL;
 320 }