src/gallium/drivers/vc4/vc4_draw.c

   1 /*
   2  * Copyright (c) 2014 Scott Mansell
   3  * Copyright © 2014 Broadcom
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the "Software"),
   7  * to deal in the Software without restriction, including without limitation
   8  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   9  * and/or sell copies of the Software, and to permit persons to whom the
  10  * Software is furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice (including the next
  13  * paragraph) shall be included in all copies or substantial portions of the
  14  * Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  21  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  22  * IN THE SOFTWARE.
  23  */
  24
  25 #include <stdio.h>
  26
  27 #include "vc4_context.h"
  28 #include "vc4_resource.h"
  29
  30 static struct vc4_bo *
  31 get_vbo(struct vc4_context *vc4, uint32_t width, uint32_t height)
  32 {
  33         struct {
  34                 float x, y, z, w;
  35         } verts[] = {
  36                 { -1, -1, 1, 1 },
  37                 {  1, -1, 1, 1 },
  38                 { -1,  1, 1, 1 },
  39         };
  40
  41         return vc4_bo_alloc_mem(vc4->screen, verts, sizeof(verts), "verts");
  42 }
  43 static struct vc4_bo *
  44 get_ibo(struct vc4_context *vc4)
  45 {
  46         static const uint8_t indices[] = { 0, 1, 2 };
  47
  48         return vc4_bo_alloc_mem(vc4->screen, indices, sizeof(indices), "indices");
  49 }
  50
  51 static void
  52 vc4_rcl_tile_calls(struct vc4_context *vc4,
  53                    uint32_t xtiles, uint32_t ytiles,
  54                    struct vc4_bo *tile_alloc)
  55 {
  56         for (int x = 0; x < xtiles; x++) {
  57                 for (int y = 0; y < ytiles; y++) {
  58                         cl_u8(&vc4->rcl, VC4_PACKET_TILE_COORDINATES);
  59                         cl_u8(&vc4->rcl, x);
  60                         cl_u8(&vc4->rcl, y);
  61
  62                         cl_start_reloc(&vc4->rcl, 1);
  63                         cl_u8(&vc4->rcl, VC4_PACKET_BRANCH_TO_SUB_LIST);
  64                         cl_reloc(vc4, &vc4->rcl, tile_alloc,
  65                                  (y * xtiles + x) * 32);
  66
  67                         if (x == xtiles - 1 && y == ytiles - 1) {
  68                                 cl_u8(&vc4->rcl,
  69                                       VC4_PACKET_STORE_MS_TILE_BUFFER_AND_EOF);
  70                         } else {
  71                                 cl_u8(&vc4->rcl,
  72                                       VC4_PACKET_STORE_MS_TILE_BUFFER);
  73                         }
  74                 }
  75         }
  76 }
  77
  78 static void
  79 vc4_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
  80 {
  81         struct vc4_context *vc4 = vc4_context(pctx);
  82         uint32_t width = vc4->framebuffer.width;
  83         uint32_t height = vc4->framebuffer.height;
  84         uint32_t tilew = align(width, 64) / 64;
  85         uint32_t tileh = align(height, 64) / 64;
  86         struct vc4_bo *tile_alloc = vc4_bo_alloc(vc4->screen,
  87                                                  32 * tilew * tileh, "tilea");
  88         struct vc4_bo *tile_state = vc4_bo_alloc(vc4->screen,
  89                                                  48 * tilew * tileh, "tilestate");
  90         struct vc4_bo *ibo = get_ibo(vc4);
  91
  92         struct vc4_bo *vbo = get_vbo(vc4, width, height);
  93
  94         vc4->needs_flush = true;
  95
  96         //   Tile state data is 48 bytes per tile, I think it can be thrown away
  97         //   as soon as binning is finished.
  98         cl_start_reloc(&vc4->bcl, 2);
  99         cl_u8(&vc4->bcl, VC4_PACKET_TILE_BINNING_MODE_CONFIG);
 100         cl_reloc(vc4, &vc4->bcl, tile_alloc, 0);
 101         cl_u32(&vc4->bcl, 0x8000); /* tile allocation memory size */
 102         cl_reloc(vc4, &vc4->bcl, tile_state, 0);
 103         cl_u8(&vc4->bcl, tilew);
 104         cl_u8(&vc4->bcl, tileh);
 105         cl_u8(&vc4->bcl, VC4_BIN_CONFIG_AUTO_INIT_TSDA);
 106
 107         cl_u8(&vc4->bcl, VC4_PACKET_START_TILE_BINNING);
 108
 109         cl_u8(&vc4->bcl, VC4_PACKET_PRIMITIVE_LIST_FORMAT);
 110         cl_u8(&vc4->bcl, 0x12); // 16 bit triangle
 111
 112         vc4_emit_state(pctx);
 113
 114         /* the actual draw call. */
 115         uint32_t nr_attributes = 1;
 116         cl_u8(&vc4->bcl, VC4_PACKET_GL_SHADER_STATE);
 117 #ifndef USE_VC4_SIMULATOR
 118         cl_u32(&vc4->bcl, nr_attributes & 0x7); /* offset into shader_rec */
 119 #else
 120         cl_u32(&vc4->bcl, simpenrose_hw_addr(vc4->shader_rec.next) |
 121                (nr_attributes & 0x7));
 122 #endif
 123
 124         cl_start_reloc(&vc4->bcl, 1);
 125         cl_u8(&vc4->bcl, VC4_PACKET_GL_INDEXED_PRIMITIVE);
 126         cl_u8(&vc4->bcl, 0x04); // 8bit index, trinagles
 127         cl_u32(&vc4->bcl, 3); // Length
 128         cl_reloc(vc4, &vc4->bcl, ibo, 0);
 129         cl_u32(&vc4->bcl, 2); // Maximum index
 130
 131         cl_u8(&vc4->bcl, VC4_PACKET_FLUSH_ALL);
 132         cl_u8(&vc4->bcl, VC4_PACKET_NOP);
 133         cl_u8(&vc4->bcl, VC4_PACKET_HALT);
 134
 135 // Shader Record
 136
 137         struct vc4_bo *fs_ubo, *vs_ubo, *cs_ubo;
 138         uint32_t fs_ubo_offset, vs_ubo_offset, cs_ubo_offset;
 139         vc4_get_uniform_bo(vc4, vc4->prog.fs,
 140                            &vc4->constbuf[PIPE_SHADER_FRAGMENT],
 141                            0, &fs_ubo, &fs_ubo_offset);
 142         vc4_get_uniform_bo(vc4, vc4->prog.vs,
 143                            &vc4->constbuf[PIPE_SHADER_VERTEX],
 144                            0, &vs_ubo, &vs_ubo_offset);
 145         vc4_get_uniform_bo(vc4, vc4->prog.vs,
 146                            &vc4->constbuf[PIPE_SHADER_VERTEX],
 147                            1, &cs_ubo, &cs_ubo_offset);
 148
 149         cl_start_shader_reloc(&vc4->shader_rec, 7);
 150         cl_u16(&vc4->shader_rec, VC4_SHADER_FLAG_ENABLE_CLIPPING);
 151         cl_u8(&vc4->shader_rec, 0); /* fs num uniforms (unused) */
 152         cl_u8(&vc4->shader_rec, 0); /* fs num varyings */
 153         cl_reloc(vc4, &vc4->shader_rec, vc4->prog.fs->bo, 0);
 154         cl_reloc(vc4, &vc4->shader_rec, fs_ubo, fs_ubo_offset);
 155
 156         cl_u16(&vc4->shader_rec, 0); /* vs num uniforms */
 157         cl_u8(&vc4->shader_rec, 1); /* vs attribute array bitfield */
 158         cl_u8(&vc4->shader_rec, 16); /* vs total attribute size */
 159         cl_reloc(vc4, &vc4->shader_rec, vc4->prog.vs->bo, 0);
 160         cl_reloc(vc4, &vc4->shader_rec, vs_ubo, vs_ubo_offset);
 161
 162         cl_u16(&vc4->shader_rec, 0); /* cs num uniforms */
 163         cl_u8(&vc4->shader_rec, 1); /* cs attribute array bitfield */
 164         cl_u8(&vc4->shader_rec, 16); /* vs total attribute size */
 165         cl_reloc(vc4, &vc4->shader_rec, vc4->prog.vs->bo,
 166                 vc4->prog.vs->coord_shader_offset);
 167         cl_reloc(vc4, &vc4->shader_rec, cs_ubo, cs_ubo_offset);
 168
 169         cl_reloc(vc4, &vc4->shader_rec, vbo, 0);
 170         cl_u8(&vc4->shader_rec, 15); /* bytes - 1 in the attribute*/
 171         cl_u8(&vc4->shader_rec, 16); /* attribute stride */
 172         cl_u8(&vc4->shader_rec, 0); /* VS VPM offset */
 173         cl_u8(&vc4->shader_rec, 0); /* CS VPM offset */
 174
 175         vc4->shader_rec_count++;
 176
 177         cl_u8(&vc4->rcl, VC4_PACKET_CLEAR_COLORS);
 178         cl_u32(&vc4->rcl, 0xff000000); // Opaque Black
 179         cl_u32(&vc4->rcl, 0xff000000); // 32 bit clear colours need to be repeated twice
 180         cl_u32(&vc4->rcl, 0);
 181         cl_u8(&vc4->rcl, 0);
 182
 183         struct vc4_surface *csurf = vc4_surface(vc4->framebuffer.cbufs[0]);
 184         struct vc4_resource *ctex = vc4_resource(csurf->base.texture);
 185
 186         cl_start_reloc(&vc4->rcl, 1);
 187         cl_u8(&vc4->rcl, VC4_PACKET_TILE_RENDERING_MODE_CONFIG);
 188         cl_reloc(vc4, &vc4->rcl, ctex->bo, csurf->offset);
 189         cl_u16(&vc4->rcl, width);
 190         cl_u16(&vc4->rcl, height);
 191         cl_u8(&vc4->rcl, (VC4_RENDER_CONFIG_MEMORY_FORMAT_LINEAR |
 192                           VC4_RENDER_CONFIG_FORMAT_RGBA8888));
 193         cl_u8(&vc4->rcl, 0);
 194
 195         // Do a store of the first tile to force the tile buffer to be cleared
 196         /* XXX: I think these two packets may be unnecessary. */
 197         cl_u8(&vc4->rcl, VC4_PACKET_TILE_COORDINATES);
 198         cl_u8(&vc4->rcl, 0);
 199         cl_u8(&vc4->rcl, 0);
 200
 201         cl_u8(&vc4->rcl, VC4_PACKET_STORE_TILE_BUFFER_GENERAL);
 202         cl_u16(&vc4->rcl, 0); // Store nothing (just clear)
 203         cl_u32(&vc4->rcl, 0); // no address is needed
 204
 205         vc4_rcl_tile_calls(vc4, tilew, tileh, tile_alloc);
 206
 207         vc4_flush(pctx);
 208 }
 209
 210 static void
 211 vc4_clear(struct pipe_context *pctx, unsigned buffers,
 212           const union pipe_color_union *color, double depth, unsigned stencil)
 213 {
 214         struct vc4_context *vc4 = vc4_context(pctx);
 215
 216         vc4->needs_flush = true;
 217 }
 218
 219 static void
 220 vc4_clear_render_target(struct pipe_context *pctx, struct pipe_surface *ps,
 221                         const union pipe_color_union *color,
 222                         unsigned x, unsigned y, unsigned w, unsigned h)
 223 {
 224         fprintf(stderr, "unimpl: clear RT\n");
 225 }
 226
 227 static void
 228 vc4_clear_depth_stencil(struct pipe_context *pctx, struct pipe_surface *ps,
 229                         unsigned buffers, double depth, unsigned stencil,
 230                         unsigned x, unsigned y, unsigned w, unsigned h)
 231 {
 232         fprintf(stderr, "unimpl: clear DS\n");
 233 }
 234
 235 void
 236 vc4_draw_init(struct pipe_context *pctx)
 237 {
 238         pctx->draw_vbo = vc4_draw_vbo;
 239         pctx->clear = vc4_clear;
 240         pctx->clear_render_target = vc4_clear_render_target;
 241         pctx->clear_depth_stencil = vc4_clear_depth_stencil;
 242 }