src/mesa/main/compute.c

   1 /*
   2  * Copyright © 2014 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
  21  * DEALINGS IN THE SOFTWARE.
  22  */
  23
  24 #include "glheader.h"
  25 #include "bufferobj.h"
  26 #include "compute.h"
  27 #include "context.h"
  28
  29 static bool
  30 check_valid_to_compute(struct gl_context *ctx, const char *function)
  31 {
  32    if (!_mesa_has_compute_shaders(ctx)) {
  33       _mesa_error(ctx, GL_INVALID_OPERATION,
  34                   "unsupported function (%s) called",
  35                   function);
  36       return false;
  37    }
  38
  39    /* From the OpenGL 4.3 Core Specification, Chapter 19, Compute Shaders:
  40     *
  41     * "An INVALID_OPERATION error is generated if there is no active program
  42     *  for the compute shader stage."
  43     */
  44    if (ctx->_Shader->CurrentProgram[MESA_SHADER_COMPUTE] == NULL) {
  45       _mesa_error(ctx, GL_INVALID_OPERATION,
  46                   "%s(no active compute shader)",
  47                   function);
  48       return false;
  49    }
  50
  51    return true;
  52 }
  53
  54 static bool
  55 validate_DispatchCompute(struct gl_context *ctx, const GLuint *num_groups)
  56 {
  57    if (!check_valid_to_compute(ctx, "glDispatchCompute"))
  58       return GL_FALSE;
  59
  60    for (int i = 0; i < 3; i++) {
  61       /* From the OpenGL 4.3 Core Specification, Chapter 19, Compute Shaders:
  62        *
  63        * "An INVALID_VALUE error is generated if any of num_groups_x,
  64        *  num_groups_y and num_groups_z are greater than or equal to the
  65        *  maximum work group count for the corresponding dimension."
  66        *
  67        * However, the "or equal to" portions appears to be a specification
  68        * bug. In all other areas, the specification appears to indicate that
  69        * the number of workgroups can match the MAX_COMPUTE_WORK_GROUP_COUNT
  70        * value. For example, under DispatchComputeIndirect:
  71        *
  72        * "If any of num_groups_x, num_groups_y or num_groups_z is greater than
  73        *  the value of MAX_COMPUTE_WORK_GROUP_COUNT for the corresponding
  74        *  dimension then the results are undefined."
  75        *
  76        * Additionally, the OpenGLES 3.1 specification does not contain "or
  77        * equal to" as an error condition.
  78        */
  79       if (num_groups[i] > ctx->Const.MaxComputeWorkGroupCount[i]) {
  80          _mesa_error(ctx, GL_INVALID_VALUE,
  81                      "glDispatchCompute(num_groups_%c)", 'x' + i);
  82          return GL_FALSE;
  83       }
  84    }
  85
  86    /* The ARB_compute_variable_group_size spec says:
  87     *
  88     * "An INVALID_OPERATION error is generated by DispatchCompute if the active
  89     *  program for the compute shader stage has a variable work group size."
  90     */
  91    struct gl_program *prog = ctx->_Shader->CurrentProgram[MESA_SHADER_COMPUTE];
  92    if (prog->info.cs.local_size_variable) {
  93       _mesa_error(ctx, GL_INVALID_OPERATION,
  94                   "glDispatchCompute(variable work group size forbidden)");
  95       return GL_FALSE;
  96    }
  97
  98    return GL_TRUE;
  99 }
 100
 101 static bool
 102 validate_DispatchComputeGroupSizeARB(struct gl_context *ctx,
 103                                      const GLuint *num_groups,
 104                                      const GLuint *group_size)
 105 {
 106    if (!check_valid_to_compute(ctx, "glDispatchComputeGroupSizeARB"))
 107       return GL_FALSE;
 108
 109    /* The ARB_compute_variable_group_size spec says:
 110     *
 111     * "An INVALID_OPERATION error is generated by
 112     *  DispatchComputeGroupSizeARB if the active program for the compute
 113     *  shader stage has a fixed work group size."
 114     */
 115    struct gl_program *prog = ctx->_Shader->CurrentProgram[MESA_SHADER_COMPUTE];
 116    if (!prog->info.cs.local_size_variable) {
 117       _mesa_error(ctx, GL_INVALID_OPERATION,
 118                   "glDispatchComputeGroupSizeARB(fixed work group size "
 119                   "forbidden)");
 120       return GL_FALSE;
 121    }
 122
 123    for (int i = 0; i < 3; i++) {
 124       /* The ARB_compute_variable_group_size spec says:
 125        *
 126        * "An INVALID_VALUE error is generated if any of num_groups_x,
 127        *  num_groups_y and num_groups_z are greater than or equal to the
 128        *  maximum work group count for the corresponding dimension."
 129        */
 130       if (num_groups[i] > ctx->Const.MaxComputeWorkGroupCount[i]) {
 131          _mesa_error(ctx, GL_INVALID_VALUE,
 132                      "glDispatchComputeGroupSizeARB(num_groups_%c)", 'x' + i);
 133          return GL_FALSE;
 134       }
 135
 136       /* The ARB_compute_variable_group_size spec says:
 137        *
 138        * "An INVALID_VALUE error is generated by DispatchComputeGroupSizeARB if
 139        *  any of <group_size_x>, <group_size_y>, or <group_size_z> is less than
 140        *  or equal to zero or greater than the maximum local work group size
 141        *  for compute shaders with variable group size
 142        *  (MAX_COMPUTE_VARIABLE_GROUP_SIZE_ARB) in the corresponding
 143        *  dimension."
 144        *
 145        * However, the "less than" is a spec bug because they are declared as
 146        * unsigned integers.
 147        */
 148       if (group_size[i] == 0 ||
 149           group_size[i] > ctx->Const.MaxComputeVariableGroupSize[i]) {
 150          _mesa_error(ctx, GL_INVALID_VALUE,
 151                      "glDispatchComputeGroupSizeARB(group_size_%c)", 'x' + i);
 152          return GL_FALSE;
 153       }
 154    }
 155
 156    /* The ARB_compute_variable_group_size spec says:
 157     *
 158     * "An INVALID_VALUE error is generated by DispatchComputeGroupSizeARB if
 159     *  the product of <group_size_x>, <group_size_y>, and <group_size_z> exceeds
 160     *  the implementation-dependent maximum local work group invocation count
 161     *  for compute shaders with variable group size
 162     *  (MAX_COMPUTE_VARIABLE_GROUP_INVOCATIONS_ARB)."
 163     */
 164    uint64_t total_invocations = group_size[0] * group_size[1];
 165    if (total_invocations <= UINT32_MAX) {
 166       /* Only bother multiplying the third value if total still fits in
 167        * 32-bit, since MaxComputeVariableGroupInvocations is also 32-bit.
 168        */
 169       total_invocations *= group_size[2];
 170    }
 171    if (total_invocations > ctx->Const.MaxComputeVariableGroupInvocations) {
 172       _mesa_error(ctx, GL_INVALID_VALUE,
 173                   "glDispatchComputeGroupSizeARB(product of local_sizes "
 174                   "exceeds MAX_COMPUTE_VARIABLE_GROUP_INVOCATIONS_ARB "
 175                   "(%u * %u * %u > %u))",
 176                   group_size[0], group_size[1], group_size[2],
 177                   ctx->Const.MaxComputeVariableGroupInvocations);
 178       return GL_FALSE;
 179    }
 180
 181    /* The NV_compute_shader_derivatives spec says:
 182     *
 183     * "An INVALID_VALUE error is generated by DispatchComputeGroupSizeARB if
 184     *  the active program for the compute shader stage has a compute shader
 185     *  using the "derivative_group_quadsNV" layout qualifier and
 186     *  <group_size_x> or <group_size_y> is not a multiple of two.
 187     *
 188     *  An INVALID_VALUE error is generated by DispatchComputeGroupSizeARB if
 189     *  the active program for the compute shader stage has a compute shader
 190     *  using the "derivative_group_linearNV" layout qualifier and the product
 191     *  of <group_size_x>, <group_size_y>, and <group_size_z> is not a multiple
 192     *  of four."
 193     */
 194    if (prog->info.cs.derivative_group == DERIVATIVE_GROUP_QUADS &&
 195        ((group_size[0] & 1) || (group_size[1] & 1))) {
 196       _mesa_error(ctx, GL_INVALID_VALUE,
 197                   "glDispatchComputeGroupSizeARB(derivative_group_quadsNV "
 198                   "requires group_size_x (%d) and group_size_y (%d) to be "
 199                   "divisble by 2)", group_size[0], group_size[1]);
 200       return GL_FALSE;
 201    }
 202
 203    if (prog->info.cs.derivative_group == DERIVATIVE_GROUP_LINEAR &&
 204        total_invocations & 3) {
 205       _mesa_error(ctx, GL_INVALID_VALUE,
 206                   "glDispatchComputeGroupSizeARB(derivative_group_linearNV "
 207                   "requires product of group sizes (%"PRIu64") to be divisible "
 208                   "by 4)", total_invocations);
 209       return GL_FALSE;
 210    }
 211
 212    return GL_TRUE;
 213 }
 214
 215 static bool
 216 valid_dispatch_indirect(struct gl_context *ctx,  GLintptr indirect)
 217 {
 218    GLsizei size = 3 * sizeof(GLuint);
 219    const uint64_t end = (uint64_t) indirect + size;
 220    const char *name = "glDispatchComputeIndirect";
 221
 222    if (!check_valid_to_compute(ctx, name))
 223       return GL_FALSE;
 224
 225    /* From the OpenGL 4.3 Core Specification, Chapter 19, Compute Shaders:
 226     *
 227     * "An INVALID_VALUE error is generated if indirect is negative or is not a
 228     *  multiple of four."
 229     */
 230    if (indirect & (sizeof(GLuint) - 1)) {
 231       _mesa_error(ctx, GL_INVALID_VALUE,
 232                   "%s(indirect is not aligned)", name);
 233       return GL_FALSE;
 234    }
 235
 236    if (indirect < 0) {
 237       _mesa_error(ctx, GL_INVALID_VALUE,
 238                   "%s(indirect is less than zero)", name);
 239       return GL_FALSE;
 240    }
 241
 242    /* From the OpenGL 4.3 Core Specification, Chapter 19, Compute Shaders:
 243     *
 244     * "An INVALID_OPERATION error is generated if no buffer is bound to the
 245     *  DRAW_INDIRECT_BUFFER binding, or if the command would source data
 246     *  beyond the end of the buffer object."
 247     */
 248    if (!ctx->DispatchIndirectBuffer) {
 249       _mesa_error(ctx, GL_INVALID_OPERATION,
 250                   "%s: no buffer bound to DISPATCH_INDIRECT_BUFFER", name);
 251       return GL_FALSE;
 252    }
 253
 254    if (_mesa_check_disallowed_mapping(ctx->DispatchIndirectBuffer)) {
 255       _mesa_error(ctx, GL_INVALID_OPERATION,
 256                   "%s(DISPATCH_INDIRECT_BUFFER is mapped)", name);
 257       return GL_FALSE;
 258    }
 259
 260    if (ctx->DispatchIndirectBuffer->Size < end) {
 261       _mesa_error(ctx, GL_INVALID_OPERATION,
 262                   "%s(DISPATCH_INDIRECT_BUFFER too small)", name);
 263       return GL_FALSE;
 264    }
 265
 266    /* The ARB_compute_variable_group_size spec says:
 267     *
 268     * "An INVALID_OPERATION error is generated if the active program for the
 269     *  compute shader stage has a variable work group size."
 270     */
 271    struct gl_program *prog = ctx->_Shader->CurrentProgram[MESA_SHADER_COMPUTE];
 272    if (prog->info.cs.local_size_variable) {
 273       _mesa_error(ctx, GL_INVALID_OPERATION,
 274                   "%s(variable work group size forbidden)", name);
 275       return GL_FALSE;
 276    }
 277
 278    return GL_TRUE;
 279 }
 280
 281 static ALWAYS_INLINE void
 282 dispatch_compute(GLuint num_groups_x, GLuint num_groups_y,
 283                  GLuint num_groups_z, bool no_error)
 284 {
 285    GET_CURRENT_CONTEXT(ctx);
 286    const GLuint num_groups[3] = { num_groups_x, num_groups_y, num_groups_z };
 287
 288    FLUSH_VERTICES(ctx, 0);
 289
 290    if (MESA_VERBOSE & VERBOSE_API)
 291       _mesa_debug(ctx, "glDispatchCompute(%d, %d, %d)\n",
 292                   num_groups_x, num_groups_y, num_groups_z);
 293
 294    if (!no_error && !validate_DispatchCompute(ctx, num_groups))
 295       return;
 296
 297    if (num_groups_x == 0u || num_groups_y == 0u || num_groups_z == 0u)
 298        return;
 299
 300    ctx->Driver.DispatchCompute(ctx, num_groups);
 301 }
 302
 303 void GLAPIENTRY
 304 _mesa_DispatchCompute_no_error(GLuint num_groups_x, GLuint num_groups_y,
 305                                GLuint num_groups_z)
 306 {
 307    dispatch_compute(num_groups_x, num_groups_y, num_groups_z, true);
 308 }
 309
 310 void GLAPIENTRY
 311 _mesa_DispatchCompute(GLuint num_groups_x,
 312                       GLuint num_groups_y,
 313                       GLuint num_groups_z)
 314 {
 315    dispatch_compute(num_groups_x, num_groups_y, num_groups_z, false);
 316 }
 317
 318 static ALWAYS_INLINE void
 319 dispatch_compute_indirect(GLintptr indirect, bool no_error)
 320 {
 321    GET_CURRENT_CONTEXT(ctx);
 322
 323    FLUSH_VERTICES(ctx, 0);
 324
 325    if (MESA_VERBOSE & VERBOSE_API)
 326       _mesa_debug(ctx, "glDispatchComputeIndirect(%ld)\n", (long) indirect);
 327
 328    if (!no_error && !valid_dispatch_indirect(ctx, indirect))
 329       return;
 330
 331    ctx->Driver.DispatchComputeIndirect(ctx, indirect);
 332 }
 333
 334 extern void GLAPIENTRY
 335 _mesa_DispatchComputeIndirect_no_error(GLintptr indirect)
 336 {
 337    dispatch_compute_indirect(indirect, true);
 338 }
 339
 340 extern void GLAPIENTRY
 341 _mesa_DispatchComputeIndirect(GLintptr indirect)
 342 {
 343    dispatch_compute_indirect(indirect, false);
 344 }
 345
 346 static ALWAYS_INLINE void
 347 dispatch_compute_group_size(GLuint num_groups_x, GLuint num_groups_y,
 348                             GLuint num_groups_z, GLuint group_size_x,
 349                             GLuint group_size_y, GLuint group_size_z,
 350                             bool no_error)
 351 {
 352    GET_CURRENT_CONTEXT(ctx);
 353    const GLuint num_groups[3] = { num_groups_x, num_groups_y, num_groups_z };
 354    const GLuint group_size[3] = { group_size_x, group_size_y, group_size_z };
 355
 356    FLUSH_VERTICES(ctx, 0);
 357
 358    if (MESA_VERBOSE & VERBOSE_API)
 359       _mesa_debug(ctx,
 360                   "glDispatchComputeGroupSizeARB(%d, %d, %d, %d, %d, %d)\n",
 361                   num_groups_x, num_groups_y, num_groups_z,
 362                   group_size_x, group_size_y, group_size_z);
 363
 364    if (!no_error &&
 365        !validate_DispatchComputeGroupSizeARB(ctx, num_groups, group_size))
 366       return;
 367
 368    if (num_groups_x == 0u || num_groups_y == 0u || num_groups_z == 0u)
 369        return;
 370
 371    ctx->Driver.DispatchComputeGroupSize(ctx, num_groups, group_size);
 372 }
 373
 374 void GLAPIENTRY
 375 _mesa_DispatchComputeGroupSizeARB_no_error(GLuint num_groups_x,
 376                                            GLuint num_groups_y,
 377                                            GLuint num_groups_z,
 378                                            GLuint group_size_x,
 379                                            GLuint group_size_y,
 380                                            GLuint group_size_z)
 381 {
 382    dispatch_compute_group_size(num_groups_x, num_groups_y, num_groups_z,
 383                                group_size_x, group_size_y, group_size_z,
 384                                true);
 385 }
 386
 387 void GLAPIENTRY
 388 _mesa_DispatchComputeGroupSizeARB(GLuint num_groups_x, GLuint num_groups_y,
 389                                   GLuint num_groups_z, GLuint group_size_x,
 390                                   GLuint group_size_y, GLuint group_size_z)
 391 {
 392    dispatch_compute_group_size(num_groups_x, num_groups_y, num_groups_z,
 393                                group_size_x, group_size_y, group_size_z,
 394                                false);
 395 }