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         static const uint32_t fs_uni[] = { 0 };
  94         uint32_t vs_uni[] = {
  95                 fui(vc4->framebuffer.width * 16.0f / 2.0f),
  96                 fui(vc4->framebuffer.height * 16.0f / 2.0f),
  97         };
  98         uint32_t cs_uni[] = {
  99                 fui(vc4->framebuffer.width * 16.0f / 2.0f),
 100                 fui(vc4->framebuffer.height * 16.0f / 2.0f),
 101         };
 102         struct vc4_bo *fs_ubo = vc4_bo_alloc_mem(vc4->screen, fs_uni,
 103                                                  sizeof(fs_uni), "fs_ubo");
 104         struct vc4_bo *vs_ubo = vc4_bo_alloc_mem(vc4->screen, vs_uni,
 105                                                  sizeof(vs_uni), "vs_ubo");
 106         struct vc4_bo *cs_ubo = vc4_bo_alloc_mem(vc4->screen, cs_uni,
 107                                                  sizeof(cs_uni), "cs_ubo");
 108
 109         vc4->needs_flush = true;
 110
 111         //   Tile state data is 48 bytes per tile, I think it can be thrown away
 112         //   as soon as binning is finished.
 113         cl_start_reloc(&vc4->bcl, 2);
 114         cl_u8(&vc4->bcl, VC4_PACKET_TILE_BINNING_MODE_CONFIG);
 115         cl_reloc(vc4, &vc4->bcl, tile_alloc, 0);
 116         cl_u32(&vc4->bcl, 0x8000); /* tile allocation memory size */
 117         cl_reloc(vc4, &vc4->bcl, tile_state, 0);
 118         cl_u8(&vc4->bcl, tilew);
 119         cl_u8(&vc4->bcl, tileh);
 120         cl_u8(&vc4->bcl, VC4_BIN_CONFIG_AUTO_INIT_TSDA);
 121
 122         cl_u8(&vc4->bcl, VC4_PACKET_START_TILE_BINNING);
 123
 124         cl_u8(&vc4->bcl, VC4_PACKET_PRIMITIVE_LIST_FORMAT);
 125         cl_u8(&vc4->bcl, 0x12); // 16 bit triangle
 126
 127         vc4_emit_state(pctx);
 128
 129         /* the actual draw call. */
 130         uint32_t nr_attributes = 1;
 131         cl_u8(&vc4->bcl, VC4_PACKET_GL_SHADER_STATE);
 132 #ifndef USE_VC4_SIMULATOR
 133         cl_u32(&vc4->bcl, nr_attributes & 0x7); /* offset into shader_rec */
 134 #else
 135         cl_u32(&vc4->bcl, simpenrose_hw_addr(vc4->shader_rec.next) |
 136                (nr_attributes & 0x7));
 137 #endif
 138
 139         cl_start_reloc(&vc4->bcl, 1);
 140         cl_u8(&vc4->bcl, VC4_PACKET_GL_INDEXED_PRIMITIVE);
 141         cl_u8(&vc4->bcl, 0x04); // 8bit index, trinagles
 142         cl_u32(&vc4->bcl, 3); // Length
 143         cl_reloc(vc4, &vc4->bcl, ibo, 0);
 144         cl_u32(&vc4->bcl, 2); // Maximum index
 145
 146         cl_u8(&vc4->bcl, VC4_PACKET_FLUSH_ALL);
 147         cl_u8(&vc4->bcl, VC4_PACKET_NOP);
 148         cl_u8(&vc4->bcl, VC4_PACKET_HALT);
 149
 150 // Shader Record
 151
 152         cl_start_shader_reloc(&vc4->shader_rec, 7);
 153         cl_u16(&vc4->shader_rec, VC4_SHADER_FLAG_ENABLE_CLIPPING);
 154         cl_u8(&vc4->shader_rec, 0); /* fs num uniforms (unused) */
 155         cl_u8(&vc4->shader_rec, 0); /* fs num varyings */
 156         cl_reloc(vc4, &vc4->shader_rec, vc4->prog.fs->bo, 0);
 157         cl_reloc(vc4, &vc4->shader_rec, fs_ubo, 0);
 158
 159         cl_u16(&vc4->shader_rec, 0); /* vs num uniforms */
 160         cl_u8(&vc4->shader_rec, 1); /* vs attribute array bitfield */
 161         cl_u8(&vc4->shader_rec, 16); /* vs total attribute size */
 162         cl_reloc(vc4, &vc4->shader_rec, vc4->prog.vs->bo, 0);
 163         cl_reloc(vc4, &vc4->shader_rec, vs_ubo, 0);
 164
 165         cl_u16(&vc4->shader_rec, 0); /* cs num uniforms */
 166         cl_u8(&vc4->shader_rec, 1); /* cs attribute array bitfield */
 167         cl_u8(&vc4->shader_rec, 16); /* vs total attribute size */
 168         cl_reloc(vc4, &vc4->shader_rec, vc4->prog.vs->bo,
 169                 vc4->prog.vs->coord_shader_offset);
 170         cl_reloc(vc4, &vc4->shader_rec, cs_ubo, 0);
 171
 172         cl_reloc(vc4, &vc4->shader_rec, vbo, 0);
 173         cl_u8(&vc4->shader_rec, 15); /* bytes - 1 in the attribute*/
 174         cl_u8(&vc4->shader_rec, 16); /* attribute stride */
 175         cl_u8(&vc4->shader_rec, 0); /* VS VPM offset */
 176         cl_u8(&vc4->shader_rec, 0); /* CS VPM offset */
 177
 178         vc4->shader_rec_count++;
 179
 180         cl_u8(&vc4->rcl, VC4_PACKET_CLEAR_COLORS);
 181         cl_u32(&vc4->rcl, 0xff000000); // Opaque Black
 182         cl_u32(&vc4->rcl, 0xff000000); // 32 bit clear colours need to be repeated twice
 183         cl_u32(&vc4->rcl, 0);
 184         cl_u8(&vc4->rcl, 0);
 185
 186         struct vc4_surface *csurf = vc4_surface(vc4->framebuffer.cbufs[0]);
 187         struct vc4_resource *ctex = vc4_resource(csurf->base.texture);
 188
 189         cl_start_reloc(&vc4->rcl, 1);
 190         cl_u8(&vc4->rcl, VC4_PACKET_TILE_RENDERING_MODE_CONFIG);
 191         cl_reloc(vc4, &vc4->rcl, ctex->bo, csurf->offset);
 192         cl_u16(&vc4->rcl, width);
 193         cl_u16(&vc4->rcl, height);
 194         cl_u8(&vc4->rcl, (VC4_RENDER_CONFIG_MEMORY_FORMAT_LINEAR |
 195                           VC4_RENDER_CONFIG_FORMAT_RGBA8888));
 196         cl_u8(&vc4->rcl, 0);
 197
 198         // Do a store of the first tile to force the tile buffer to be cleared
 199         /* XXX: I think these two packets may be unnecessary. */
 200         cl_u8(&vc4->rcl, VC4_PACKET_TILE_COORDINATES);
 201         cl_u8(&vc4->rcl, 0);
 202         cl_u8(&vc4->rcl, 0);
 203
 204         cl_u8(&vc4->rcl, VC4_PACKET_STORE_TILE_BUFFER_GENERAL);
 205         cl_u16(&vc4->rcl, 0); // Store nothing (just clear)
 206         cl_u32(&vc4->rcl, 0); // no address is needed
 207
 208         vc4_rcl_tile_calls(vc4, tilew, tileh, tile_alloc);
 209
 210         vc4_flush(pctx);
 211 }
 212
 213 static void
 214 vc4_clear(struct pipe_context *pctx, unsigned buffers,
 215           const union pipe_color_union *color, double depth, unsigned stencil)
 216 {
 217         struct vc4_context *vc4 = vc4_context(pctx);
 218
 219         vc4->needs_flush = true;
 220 }
 221
 222 static void
 223 vc4_clear_render_target(struct pipe_context *pctx, struct pipe_surface *ps,
 224                         const union pipe_color_union *color,
 225                         unsigned x, unsigned y, unsigned w, unsigned h)
 226 {
 227         fprintf(stderr, "unimpl: clear RT\n");
 228 }
 229
 230 static void
 231 vc4_clear_depth_stencil(struct pipe_context *pctx, struct pipe_surface *ps,
 232                         unsigned buffers, double depth, unsigned stencil,
 233                         unsigned x, unsigned y, unsigned w, unsigned h)
 234 {
 235         fprintf(stderr, "unimpl: clear DS\n");
 236 }
 237
 238 void
 239 vc4_draw_init(struct pipe_context *pctx)
 240 {
 241         pctx->draw_vbo = vc4_draw_vbo;
 242         pctx->clear = vc4_clear;
 243         pctx->clear_render_target = vc4_clear_render_target;
 244         pctx->clear_depth_stencil = vc4_clear_depth_stencil;
 245 }