src/gallium/drivers/cell/ppu/cell_state_emit.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sub license, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28 #include "util/u_inlines.h"
  29 #include "util/u_memory.h"
  30 #include "util/u_math.h"
  31 #include "util/u_format.h"
  32 #include "cell_context.h"
  33 #include "cell_gen_fragment.h"
  34 #include "cell_state.h"
  35 #include "cell_state_emit.h"
  36 #include "cell_batch.h"
  37 #include "cell_texture.h"
  38 #include "draw/draw_context.h"
  39 #include "draw/draw_private.h"
  40
  41
  42 /**
  43  * Find/create a cell_command_fragment_ops object corresponding to the
  44  * current blend/stencil/z/colormask/etc. state.
  45  */
  46 static struct cell_command_fragment_ops *
  47 lookup_fragment_ops(struct cell_context *cell)
  48 {
  49    struct cell_fragment_ops_key key;
  50    struct cell_command_fragment_ops *ops;
  51
  52    /*
  53     * Build key
  54     */
  55    memset(&key, 0, sizeof(key));
  56    key.blend = *cell->blend;
  57    key.blend_color = cell->blend_color;
  58    key.dsa = *cell->depth_stencil;
  59
  60    if (cell->framebuffer.cbufs[0])
  61       key.color_format = cell->framebuffer.cbufs[0]->format;
  62    else
  63       key.color_format = PIPE_FORMAT_NONE;
  64
  65    if (cell->framebuffer.zsbuf)
  66       key.zs_format = cell->framebuffer.zsbuf->format;
  67    else
  68       key.zs_format = PIPE_FORMAT_NONE;
  69
  70    /*
  71     * Look up key in cache.
  72     */
  73    ops = (struct cell_command_fragment_ops *)
  74       util_keymap_lookup(cell->fragment_ops_cache, &key);
  75
  76    /*
  77     * If not found, create/save new fragment ops command.
  78     */
  79    if (!ops) {
  80       struct spe_function spe_code_front, spe_code_back;
  81       unsigned int facing_dependent, total_code_size;
  82
  83       if (0)
  84          debug_printf("**** Create New Fragment Ops\n");
  85
  86       /* Prepare the buffer that will hold the generated code.  The
  87        * "0" passed in for the size means that the SPE code will
  88        * use a default size.
  89        */
  90       spe_init_func(&spe_code_front, 0);
  91       spe_init_func(&spe_code_back, 0);
  92
  93       /* Generate new code.  Always generate new code for both front-facing
  94        * and back-facing fragments, even if it's the same code in both
  95        * cases.
  96        */
  97       cell_gen_fragment_function(cell, CELL_FACING_FRONT, &spe_code_front);
  98       cell_gen_fragment_function(cell, CELL_FACING_BACK, &spe_code_back);
  99
 100       /* Make sure the code is a multiple of 8 bytes long; this is
 101        * required to ensure that the dual pipe instruction alignment
 102        * is correct.  It's also important for the SPU unpacking,
 103        * which assumes 8-byte boundaries.
 104        */
 105       unsigned int front_code_size = spe_code_size(&spe_code_front);
 106       while (front_code_size % 8 != 0) {
 107          spe_lnop(&spe_code_front);
 108          front_code_size = spe_code_size(&spe_code_front);
 109       }
 110       unsigned int back_code_size = spe_code_size(&spe_code_back);
 111       while (back_code_size % 8 != 0) {
 112          spe_lnop(&spe_code_back);
 113          back_code_size = spe_code_size(&spe_code_back);
 114       }
 115
 116       /* Determine whether the code we generated is facing-dependent, by
 117        * determining whether the generated code is different for the front-
 118        * and back-facing fragments.
 119        */
 120       if (front_code_size == back_code_size && memcmp(spe_code_front.store, spe_code_back.store, front_code_size) == 0) {
 121          /* Code is identical; only need one copy. */
 122          facing_dependent = 0;
 123          total_code_size = front_code_size;
 124       }
 125       else {
 126          /* Code is different for front-facing and back-facing fragments.
 127           * Need to send both copies.
 128           */
 129          facing_dependent = 1;
 130          total_code_size = front_code_size + back_code_size;
 131       }
 132
 133       /* alloc new fragment ops command.  Note that this structure
 134        * has variant length based on the total code size required.
 135        */
 136       ops = CALLOC_VARIANT_LENGTH_STRUCT(cell_command_fragment_ops, total_code_size);
 137       /* populate the new cell_command_fragment_ops object */
 138       ops->opcode[0] = CELL_CMD_STATE_FRAGMENT_OPS;
 139       ops->total_code_size = total_code_size;
 140       ops->front_code_index = 0;
 141       memcpy(ops->code, spe_code_front.store, front_code_size);
 142       if (facing_dependent) {
 143         /* We have separate front- and back-facing code.  Append the
 144          * back-facing code to the buffer.  Be careful because the code
 145          * size is in bytes, but the buffer is of unsigned elements.
 146          */
 147         ops->back_code_index = front_code_size / sizeof(spe_code_front.store[0]);
 148         memcpy(ops->code + ops->back_code_index, spe_code_back.store, back_code_size);
 149       }
 150       else {
 151         /* Use the same code for front- and back-facing fragments */
 152         ops->back_code_index = ops->front_code_index;
 153       }
 154
 155       /* Set the fields for the fallback case.  Note that these fields
 156        * (and the whole fallback case) will eventually go away.
 157        */
 158       ops->dsa = *cell->depth_stencil;
 159       ops->blend = *cell->blend;
 160       ops->blend_color = cell->blend_color;
 161
 162       /* insert cell_command_fragment_ops object into keymap/cache */
 163       util_keymap_insert(cell->fragment_ops_cache, &key, ops, NULL);
 164
 165       /* release rtasm buffer */
 166       spe_release_func(&spe_code_front);
 167       spe_release_func(&spe_code_back);
 168    }
 169    else {
 170       if (0)
 171          debug_printf("**** Re-use Fragment Ops\n");
 172    }
 173
 174    return ops;
 175 }
 176
 177
 178
 179 static void
 180 emit_state_cmd(struct cell_context *cell, uint cmd,
 181                const void *state, uint state_size)
 182 {
 183    uint32_t *dst = (uint32_t *)
 184        cell_batch_alloc16(cell, ROUNDUP16(sizeof(opcode_t) + state_size));
 185    *dst = cmd;
 186    memcpy(dst + 4, state, state_size);
 187 }
 188
 189
 190 /**
 191  * For state marked as 'dirty', construct a state-update command block
 192  * and insert it into the current batch buffer.
 193  */
 194 void
 195 cell_emit_state(struct cell_context *cell)
 196 {
 197    if (cell->dirty & CELL_NEW_FRAMEBUFFER) {
 198       struct pipe_surface *cbuf = cell->framebuffer.cbufs[0];
 199       struct pipe_surface *zbuf = cell->framebuffer.zsbuf;
 200       STATIC_ASSERT(sizeof(struct cell_command_framebuffer) % 16 == 0);
 201       struct cell_command_framebuffer *fb
 202          = cell_batch_alloc16(cell, sizeof(*fb));
 203       fb->opcode[0] = CELL_CMD_STATE_FRAMEBUFFER;
 204       fb->color_start = cell->cbuf_map[0];
 205       fb->color_format = cbuf->format;
 206       fb->depth_start = cell->zsbuf_map;
 207       fb->depth_format = zbuf ? zbuf->format : PIPE_FORMAT_NONE;
 208       fb->width = cell->framebuffer.width;
 209       fb->height = cell->framebuffer.height;
 210 #if 0
 211       printf("EMIT color format %s\n", util_format_name(fb->color_format));
 212       printf("EMIT depth format %s\n", util_format_name(fb->depth_format));
 213 #endif
 214    }
 215
 216    if (cell->dirty & (CELL_NEW_RASTERIZER)) {
 217       STATIC_ASSERT(sizeof(struct cell_command_rasterizer) % 16 == 0);
 218       struct cell_command_rasterizer *rast =
 219          cell_batch_alloc16(cell, sizeof(*rast));
 220       rast->opcode[0] = CELL_CMD_STATE_RASTERIZER;
 221       rast->rasterizer = *cell->rasterizer;
 222    }
 223
 224    if (cell->dirty & (CELL_NEW_FS)) {
 225       /* Send new fragment program to SPUs */
 226       STATIC_ASSERT(sizeof(struct cell_command_fragment_program) % 16 == 0);
 227       struct cell_command_fragment_program *fp
 228             = cell_batch_alloc16(cell, sizeof(*fp));
 229       fp->opcode[0] = CELL_CMD_STATE_FRAGMENT_PROGRAM;
 230       fp->num_inst = cell->fs->code.num_inst;
 231       memcpy(&fp->code, cell->fs->code.store,
 232              SPU_MAX_FRAGMENT_PROGRAM_INSTS * SPE_INST_SIZE);
 233       if (0) {
 234          int i;
 235          printf("PPU Emit CELL_CMD_STATE_FRAGMENT_PROGRAM:\n");
 236          for (i = 0; i < fp->num_inst; i++) {
 237             printf(" %3d: 0x%08x\n", i, fp->code[i]);
 238          }
 239       }
 240    }
 241
 242    if (cell->dirty & (CELL_NEW_FS_CONSTANTS)) {
 243       const uint shader = PIPE_SHADER_FRAGMENT;
 244       const uint num_const = cell->constants[shader]->width0 / sizeof(float);
 245       uint i, j;
 246       float *buf = cell_batch_alloc16(cell, ROUNDUP16(32 + num_const * sizeof(float)));
 247       uint32_t *ibuf = (uint32_t *) buf;
 248       const float *constants = cell->mapped_constants[shader];
 249       ibuf[0] = CELL_CMD_STATE_FS_CONSTANTS;
 250       ibuf[4] = num_const;
 251       j = 8;
 252       for (i = 0; i < num_const; i++) {
 253          buf[j++] = constants[i];
 254       }
 255    }
 256
 257    if (cell->dirty & (CELL_NEW_FRAMEBUFFER |
 258                       CELL_NEW_DEPTH_STENCIL |
 259                       CELL_NEW_BLEND)) {
 260       struct cell_command_fragment_ops *fops, *fops_cmd;
 261       /* Note that cell_command_fragment_ops is a variant-sized record */
 262       fops = lookup_fragment_ops(cell);
 263       fops_cmd = cell_batch_alloc16(cell, ROUNDUP16(sizeof(*fops_cmd) + fops->total_code_size));
 264       memcpy(fops_cmd, fops, sizeof(*fops) + fops->total_code_size);
 265    }
 266
 267    if (cell->dirty & CELL_NEW_SAMPLER) {
 268       uint i;
 269       for (i = 0; i < CELL_MAX_SAMPLERS; i++) {
 270          if (cell->dirty_samplers & (1 << i)) {
 271             if (cell->sampler[i]) {
 272                STATIC_ASSERT(sizeof(struct cell_command_sampler) % 16 == 0);
 273                struct cell_command_sampler *sampler
 274                   = cell_batch_alloc16(cell, sizeof(*sampler));
 275                sampler->opcode[0] = CELL_CMD_STATE_SAMPLER;
 276                sampler->unit = i;
 277                sampler->state = *cell->sampler[i];
 278             }
 279          }
 280       }
 281       cell->dirty_samplers = 0x0;
 282    }
 283
 284    if (cell->dirty & CELL_NEW_TEXTURE) {
 285       uint i;
 286       for (i = 0;i < CELL_MAX_SAMPLERS; i++) {
 287          if (cell->dirty_textures & (1 << i)) {
 288             STATIC_ASSERT(sizeof(struct cell_command_texture) % 16 == 0);
 289             struct cell_command_texture *texture =
 290                (struct cell_command_texture *)
 291                cell_batch_alloc16(cell, sizeof(*texture));
 292
 293             texture->opcode[0] = CELL_CMD_STATE_TEXTURE;
 294             texture->unit = i;
 295             if (cell->texture[i]) {
 296                struct cell_resource *ct = cell->texture[i];
 297                uint level;
 298                for (level = 0; level < CELL_MAX_TEXTURE_LEVELS; level++) {
 299                   texture->start[level] = (ct->mapped +
 300                                            ct->level_offset[level]);
 301                   texture->width[level] = u_minify(ct->base.width0, level);
 302                   texture->height[level] = u_minify(ct->base.height0, level);
 303                   texture->depth[level] = u_minify(ct->base.depth0, level);
 304                }
 305                texture->target = ct->base.target;
 306             }
 307             else {
 308                uint level;
 309                for (level = 0; level < CELL_MAX_TEXTURE_LEVELS; level++) {
 310                   texture->start[level] = NULL;
 311                   texture->width[level] = 0;
 312                   texture->height[level] = 0;
 313                   texture->depth[level] = 0;
 314                }
 315                texture->target = 0;
 316             }
 317          }
 318       }
 319       cell->dirty_textures = 0x0;
 320    }
 321
 322    if (cell->dirty & CELL_NEW_VERTEX_INFO) {
 323       emit_state_cmd(cell, CELL_CMD_STATE_VERTEX_INFO,
 324                      &cell->vertex_info, sizeof(struct vertex_info));
 325    }
 326
 327 #if 0
 328    if (cell->dirty & CELL_NEW_VS) {
 329       const struct draw_context *const draw = cell->draw;
 330       struct cell_shader_info info;
 331
 332       info.num_outputs = draw_num_shader_outputs(draw);
 333       info.declarations = (uintptr_t) draw->vs.machine.Declarations;
 334       info.num_declarations = draw->vs.machine.NumDeclarations;
 335       info.instructions = (uintptr_t) draw->vs.machine.Instructions;
 336       info.num_instructions = draw->vs.machine.NumInstructions;
 337       info.immediates = (uintptr_t) draw->vs.machine.Imms;
 338       info.num_immediates = draw->vs.machine.ImmLimit / 4;
 339
 340       emit_state_cmd(cell, CELL_CMD_STATE_BIND_VS, &info, sizeof(info));
 341    }
 342 #endif
 343 }