src/mesa/drivers/dri/i965/gen6_constant_state.c

   1 /*
   2  * Copyright © 2015 Intel Corporation
   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 "brw_context.h"
  25 #include "brw_state.h"
  26 #include "brw_defines.h"
  27 #include "brw_program.h"
  28 #include "intel_batchbuffer.h"
  29 #include "intel_buffer_objects.h"
  30 #include "program/prog_parameter.h"
  31
  32 static uint32_t
  33 f_as_u32(float f)
  34 {
  35    return *(uint32_t *)&f;
  36 }
  37
  38 static uint32_t
  39 brw_param_value(struct brw_context *brw,
  40                 const struct gl_program *prog,
  41                 const struct brw_stage_state *stage_state,
  42                 uint32_t param)
  43 {
  44    struct gl_context *ctx = &brw->ctx;
  45
  46    switch (BRW_PARAM_DOMAIN(param)) {
  47    case BRW_PARAM_DOMAIN_BUILTIN:
  48       if (param == BRW_PARAM_BUILTIN_ZERO) {
  49          return 0;
  50       } else if (BRW_PARAM_BUILTIN_IS_CLIP_PLANE(param)) {
  51          gl_clip_plane *clip_planes = brw_select_clip_planes(ctx);
  52          unsigned idx = BRW_PARAM_BUILTIN_CLIP_PLANE_IDX(param);
  53          unsigned comp = BRW_PARAM_BUILTIN_CLIP_PLANE_COMP(param);
  54          return ((uint32_t *)clip_planes[idx])[comp];
  55       } else if (param >= BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_X &&
  56                  param <= BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_W) {
  57          unsigned i = param - BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_X;
  58          return f_as_u32(ctx->TessCtrlProgram.patch_default_outer_level[i]);
  59       } else if (param == BRW_PARAM_BUILTIN_TESS_LEVEL_INNER_X) {
  60          return f_as_u32(ctx->TessCtrlProgram.patch_default_inner_level[0]);
  61       } else if (param == BRW_PARAM_BUILTIN_TESS_LEVEL_INNER_Y) {
  62          return f_as_u32(ctx->TessCtrlProgram.patch_default_inner_level[1]);
  63       } else {
  64          unreachable("Invalid param builtin");
  65       }
  66
  67    case BRW_PARAM_DOMAIN_PARAMETER: {
  68       unsigned idx = BRW_PARAM_PARAMETER_IDX(param);
  69       unsigned comp = BRW_PARAM_PARAMETER_COMP(param);
  70       assert(idx < prog->Parameters->NumParameters);
  71       return prog->Parameters->ParameterValues[idx][comp].u;
  72    }
  73
  74    case BRW_PARAM_DOMAIN_UNIFORM: {
  75       unsigned idx = BRW_PARAM_UNIFORM_IDX(param);
  76       assert(idx < prog->sh.data->NumUniformDataSlots);
  77       return prog->sh.data->UniformDataSlots[idx].u;
  78    }
  79
  80    case BRW_PARAM_DOMAIN_IMAGE: {
  81       unsigned idx = BRW_PARAM_IMAGE_IDX(param);
  82       unsigned offset = BRW_PARAM_IMAGE_OFFSET(param);
  83       assert(offset < ARRAY_SIZE(stage_state->image_param));
  84       return ((uint32_t *)&stage_state->image_param[idx])[offset];
  85    }
  86
  87    default:
  88       unreachable("Invalid param domain");
  89    }
  90 }
  91
  92
  93 void
  94 brw_populate_constant_data(struct brw_context *brw,
  95                            const struct gl_program *prog,
  96                            const struct brw_stage_state *stage_state,
  97                            void *void_dst,
  98                            const uint32_t *param,
  99                            unsigned nr_params)
 100 {
 101    uint32_t *dst = void_dst;
 102    for (unsigned i = 0; i < nr_params; i++)
 103       dst[i] = brw_param_value(brw, prog, stage_state, param[i]);
 104 }
 105
 106
 107 /**
 108  * Creates a streamed BO containing the push constants for the VS or GS on
 109  * gen6+.
 110  *
 111  * Push constants are constant values (such as GLSL uniforms) that are
 112  * pre-loaded into a shader stage's register space at thread spawn time.
 113  *
 114  * Not all GLSL uniforms will be uploaded as push constants: The hardware has
 115  * a limitation of 32 or 64 EU registers (256 or 512 floats) per stage to be
 116  * uploaded as push constants, while GL 4.4 requires at least 1024 components
 117  * to be usable for the VS.  Plus, currently we always use pull constants
 118  * instead of push constants when doing variable-index array access.
 119  *
 120  * See brw_curbe.c for the equivalent gen4/5 code.
 121  */
 122 void
 123 gen6_upload_push_constants(struct brw_context *brw,
 124                            const struct gl_program *prog,
 125                            const struct brw_stage_prog_data *prog_data,
 126                            struct brw_stage_state *stage_state)
 127 {
 128    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 129    struct gl_context *ctx = &brw->ctx;
 130
 131    if (prog_data->nr_params == 0) {
 132       stage_state->push_const_size = 0;
 133    } else {
 134       /* Updates the ParamaterValues[i] pointers for all parameters of the
 135        * basic type of PROGRAM_STATE_VAR.
 136        */
 137       /* XXX: Should this happen somewhere before to get our state flag set? */
 138       if (prog)
 139          _mesa_load_state_parameters(ctx, prog->Parameters);
 140
 141       int i;
 142       const int size = prog_data->nr_params * sizeof(gl_constant_value);
 143       gl_constant_value *param;
 144       if (devinfo->gen >= 8 || devinfo->is_haswell) {
 145          param = intel_upload_space(brw, size, 32,
 146                                     &stage_state->push_const_bo,
 147                                     &stage_state->push_const_offset);
 148       } else {
 149          param = brw_state_batch(brw, size, 32,
 150                                  &stage_state->push_const_offset);
 151       }
 152
 153       STATIC_ASSERT(sizeof(gl_constant_value) == sizeof(float));
 154
 155       /* _NEW_PROGRAM_CONSTANTS
 156        *
 157        * Also _NEW_TRANSFORM -- we may reference clip planes other than as a
 158        * side effect of dereferencing uniforms, so _NEW_PROGRAM_CONSTANTS
 159        * wouldn't be set for them.
 160        */
 161       brw_populate_constant_data(brw, prog, stage_state, param,
 162                                  prog_data->param,
 163                                  prog_data->nr_params);
 164
 165       if (0) {
 166          fprintf(stderr, "%s constants:\n",
 167                  _mesa_shader_stage_to_string(stage_state->stage));
 168          for (i = 0; i < prog_data->nr_params; i++) {
 169             if ((i & 7) == 0)
 170                fprintf(stderr, "g%d: ",
 171                        prog_data->dispatch_grf_start_reg + i / 8);
 172             fprintf(stderr, "%8f ", param[i].f);
 173             if ((i & 7) == 7)
 174                fprintf(stderr, "\n");
 175          }
 176          if ((i & 7) != 0)
 177             fprintf(stderr, "\n");
 178          fprintf(stderr, "\n");
 179       }
 180
 181       stage_state->push_const_size = ALIGN(prog_data->nr_params, 8) / 8;
 182       /* We can only push 32 registers of constants at a time. */
 183
 184       /* From the SNB PRM (vol2, part 1, section 3.2.1.4: 3DSTATE_CONSTANT_VS:
 185        *
 186        *     "The sum of all four read length fields (each incremented to
 187        *      represent the actual read length) must be less than or equal to
 188        *      32"
 189        *
 190        * From the IVB PRM (vol2, part 1, section 3.2.1.3: 3DSTATE_CONSTANT_VS:
 191        *
 192        *     "The sum of all four read length fields must be less than or
 193        *      equal to the size of 64"
 194        *
 195        * The other shader stages all match the VS's limits.
 196        */
 197       assert(stage_state->push_const_size <= 32);
 198    }
 199
 200    stage_state->push_constants_dirty = true;
 201 }
 202
 203
 204 /**
 205  * Creates a temporary BO containing the pull constant data for the shader
 206  * stage, and the SURFACE_STATE struct that points at it.
 207  *
 208  * Pull constants are GLSL uniforms (and other constant data) beyond what we
 209  * could fit as push constants, or that have variable-index array access
 210  * (which is easiest to support using pull constants, and avoids filling
 211  * register space with mostly-unused data).
 212  *
 213  * Compare this path to brw_curbe.c for gen4/5 push constants, and
 214  * gen6_vs_state.c for gen6+ push constants.
 215  */
 216 void
 217 brw_upload_pull_constants(struct brw_context *brw,
 218                           GLbitfield64 brw_new_constbuf,
 219                           const struct gl_program *prog,
 220                           struct brw_stage_state *stage_state,
 221                           const struct brw_stage_prog_data *prog_data)
 222 {
 223    unsigned i;
 224    uint32_t surf_index = prog_data->binding_table.pull_constants_start;
 225
 226    if (!prog_data->nr_pull_params) {
 227       if (stage_state->surf_offset[surf_index]) {
 228          stage_state->surf_offset[surf_index] = 0;
 229          brw->ctx.NewDriverState |= brw_new_constbuf;
 230       }
 231       return;
 232    }
 233
 234    /* Updates the ParamaterValues[i] pointers for all parameters of the
 235     * basic type of PROGRAM_STATE_VAR.
 236     */
 237    _mesa_load_state_parameters(&brw->ctx, prog->Parameters);
 238
 239    /* BRW_NEW_*_PROG_DATA | _NEW_PROGRAM_CONSTANTS */
 240    uint32_t size = prog_data->nr_pull_params * 4;
 241    struct brw_bo *const_bo = NULL;
 242    uint32_t const_offset;
 243    gl_constant_value *constants = intel_upload_space(brw, size, 64,
 244                                                      &const_bo, &const_offset);
 245
 246    STATIC_ASSERT(sizeof(gl_constant_value) == sizeof(float));
 247
 248    brw_populate_constant_data(brw, prog, stage_state, constants,
 249                               prog_data->pull_param,
 250                               prog_data->nr_pull_params);
 251
 252    if (0) {
 253       for (i = 0; i < ALIGN(prog_data->nr_pull_params, 4) / 4; i++) {
 254          const gl_constant_value *row = &constants[i * 4];
 255          fprintf(stderr, "const surface %3d: %4.3f %4.3f %4.3f %4.3f\n",
 256                  i, row[0].f, row[1].f, row[2].f, row[3].f);
 257       }
 258    }
 259
 260    brw_create_constant_surface(brw, const_bo, const_offset, size,
 261                                &stage_state->surf_offset[surf_index]);
 262    brw_bo_unreference(const_bo);
 263
 264    brw->ctx.NewDriverState |= brw_new_constbuf;
 265 }
 266
 267 /**
 268  * Creates a region containing the push constants for the CS on gen7+.
 269  *
 270  * Push constants are constant values (such as GLSL uniforms) that are
 271  * pre-loaded into a shader stage's register space at thread spawn time.
 272  *
 273  * For other stages, see brw_curbe.c:brw_upload_constant_buffer for the
 274  * equivalent gen4/5 code and gen6_vs_state.c:gen6_upload_push_constants for
 275  * gen6+.
 276  */
 277 void
 278 brw_upload_cs_push_constants(struct brw_context *brw,
 279                              const struct gl_program *prog,
 280                              const struct brw_cs_prog_data *cs_prog_data,
 281                              struct brw_stage_state *stage_state)
 282 {
 283    struct gl_context *ctx = &brw->ctx;
 284    const struct brw_stage_prog_data *prog_data =
 285       (struct brw_stage_prog_data*) cs_prog_data;
 286
 287    /* Updates the ParamaterValues[i] pointers for all parameters of the
 288     * basic type of PROGRAM_STATE_VAR.
 289     */
 290    /* XXX: Should this happen somewhere before to get our state flag set? */
 291    _mesa_load_state_parameters(ctx, prog->Parameters);
 292
 293    if (cs_prog_data->push.total.size == 0) {
 294       stage_state->push_const_size = 0;
 295       return;
 296    }
 297
 298
 299    uint32_t *param =
 300       brw_state_batch(brw, ALIGN(cs_prog_data->push.total.size, 64),
 301                       64, &stage_state->push_const_offset);
 302    assert(param);
 303
 304    STATIC_ASSERT(sizeof(gl_constant_value) == sizeof(float));
 305
 306    if (cs_prog_data->push.cross_thread.size > 0) {
 307       uint32_t *param_copy = param;
 308       assert(cs_prog_data->thread_local_id_index < 0 ||
 309              cs_prog_data->thread_local_id_index >=
 310                 cs_prog_data->push.cross_thread.dwords);
 311       for (unsigned i = 0;
 312            i < cs_prog_data->push.cross_thread.dwords;
 313            i++) {
 314          param_copy[i] = brw_param_value(brw, prog, stage_state,
 315                                          prog_data->param[i]);
 316       }
 317    }
 318
 319    if (cs_prog_data->push.per_thread.size > 0) {
 320       for (unsigned t = 0; t < cs_prog_data->threads; t++) {
 321          unsigned dst =
 322             8 * (cs_prog_data->push.per_thread.regs * t +
 323                  cs_prog_data->push.cross_thread.regs);
 324          unsigned src = cs_prog_data->push.cross_thread.dwords;
 325          for ( ; src < prog_data->nr_params; src++, dst++) {
 326             if (src != cs_prog_data->thread_local_id_index) {
 327                param[dst] = brw_param_value(brw, prog, stage_state,
 328                                             prog_data->param[src]);
 329             } else {
 330                param[dst] = t * cs_prog_data->simd_size;
 331             }
 332          }
 333       }
 334    }
 335
 336    stage_state->push_const_size =
 337       cs_prog_data->push.cross_thread.regs +
 338       cs_prog_data->push.per_thread.regs;
 339 }