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