granicus.if.org Git - imagemagick/blob - MagickCore/accelerate.c

   1 /*
   2 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
   3 %                                                                             %
   4 %                                                                             %
   5 %                                                                             %
   6 %     AAA     CCCC    CCCC  EEEEE  L      EEEEE  RRRR    AAA   TTTTT  EEEEE   %
   7 %    A   A   C       C      E      L      E      R   R  A   A    T    E       %
   8 %    AAAAA   C       C      EEE    L      EEE    RRRR   AAAAA    T    EEE     %
   9 %    A   A   C       C      E      L      E      R R    A   A    T    E       %
  10 %    A   A    CCCC    CCCC  EEEEE  LLLLL  EEEEE  R  R   A   A    T    EEEEE   %
  11 %                                                                             %
  12 %                                                                             %
  13 %                       MagickCore Acceleration Methods                       %
  14 %                                                                             %
  15 %                              Software Design                                %
  16 %                               John Cristy                                   %
  17 %                               January 2010                                  %
  18 %                                                                             %
  19 %                                                                             %
  20 %  Copyright 1999-2011 ImageMagick Studio LLC, a non-profit organization      %
  21 %  dedicated to making software imaging solutions freely available.           %
  22 %                                                                             %
  23 %  You may not use this file except in compliance with the License.  You may  %
  24 %  obtain a copy of the License at                                            %
  25 %                                                                             %
  26 %    http://www.imagemagick.org/script/license.php                            %
  27 %                                                                             %
  28 %  Unless required by applicable law or agreed to in writing, software        %
  29 %  distributed under the License is distributed on an "AS IS" BASIS,          %
  30 %  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.   %
  31 %  See the License for the specific language governing permissions and        %
  32 %  limitations under the License.                                             %
  33 %                                                                             %
  34 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
  35 %
  36 % Morphology is the the application of various kernals, of any size and even
  37 % shape, to a image in various ways (typically binary, but not always).
  38 %
  39 % Convolution (weighted sum or average) is just one specific type of
  40 % accelerate. Just one that is very common for image bluring and sharpening
  41 % effects.  Not only 2D Gaussian blurring, but also 2-pass 1D Blurring.
  42 %
  43 % This module provides not only a general accelerate function, and the ability
  44 % to apply more advanced or iterative morphologies, but also functions for the
  45 % generation of many different types of kernel arrays from user supplied
  46 % arguments. Prehaps even the generation of a kernel from a small image.
  47 */
  48 \f
  49 /*
  50   Include declarations.
  51 */
  52 #include "MagickCore/studio.h"
  53 #include "MagickCore/accelerate.h"
  54 #include "MagickCore/artifact.h"
  55 #include "MagickCore/cache.h"
  56 #include "MagickCore/cache-private.h"
  57 #include "MagickCore/cache-view.h"
  58 #include "MagickCore/color-private.h"
  59 #include "MagickCore/enhance.h"
  60 #include "MagickCore/exception.h"
  61 #include "MagickCore/exception-private.h"
  62 #include "MagickCore/gem.h"
  63 #include "MagickCore/hashmap.h"
  64 #include "MagickCore/image.h"
  65 #include "MagickCore/image-private.h"
  66 #include "MagickCore/list.h"
  67 #include "MagickCore/memory_.h"
  68 #include "MagickCore/monitor-private.h"
  69 #include "MagickCore/accelerate.h"
  70 #include "MagickCore/option.h"
  71 #include "MagickCore/pixel-accessor.h"
  72 #include "MagickCore/prepress.h"
  73 #include "MagickCore/quantize.h"
  74 #include "MagickCore/registry.h"
  75 #include "MagickCore/semaphore.h"
  76 #include "MagickCore/splay-tree.h"
  77 #include "MagickCore/statistic.h"
  78 #include "MagickCore/string_.h"
  79 #include "MagickCore/string-private.h"
  80 #include "MagickCore/token.h"
  81 \f
  82 /*
  83 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
  84 %                                                                             %
  85 %                                                                             %
  86 %                                                                             %
  87 %     A c c e l e r a t e C o n v o l v e I m a g e                           %
  88 %                                                                             %
  89 %                                                                             %
  90 %                                                                             %
  91 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
  92 %
  93 %  AccelerateConvolveImage() applies a custom convolution kernel to the image.
  94 %  It is accelerated by taking advantage of speed-ups offered by executing in
  95 %  concert across heterogeneous platforms consisting of CPUs, GPUs, and other
  96 %  processors.
  97 %
  98 %  The format of the AccelerateConvolveImage method is:
  99 %
 100 %      Image *AccelerateConvolveImage(const Image *image,
 101 %        const KernelInfo *kernel,Image *convolve_image,
 102 %        ExceptionInfo *exception)
 103 %
 104 %  A description of each parameter follows:
 105 %
 106 %    o image: the image.
 107 %
 108 %    o kernel: the convolution kernel.
 109 %
 110 %    o convole_image: the convoleed image.
 111 %
 112 %    o exception: return any errors or warnings in this structure.
 113 %
 114 */
 115
 116 #if defined(MAGICKCORE_OPENCL_SUPPORT)
 117
 118 #if defined(MAGICKCORE_HDRI_SUPPORT)
 119 #define CLOptions "-DMAGICKCORE_HDRI_SUPPORT=1 -DCLQuantum=float " \
 120   "-DCLPixelType=float4 -DQuantumRange=%g -DMagickEpsilon=%g"
 121 #define CLPixelPacket  cl_float4
 122 #else
 123 #if (MAGICKCORE_QUANTUM_DEPTH == 8)
 124 #define CLOptions "-DCLQuantum=uchar -DCLPixelType=uchar4 " \
 125   "-DQuantumRange=%g -DMagickEpsilon=%g"
 126 #define CLPixelPacket  cl_uchar4
 127 #elif (MAGICKCORE_QUANTUM_DEPTH == 16)
 128 #define CLOptions "-DCLQuantum=ushort -DCLPixelType=ushort4 " \
 129   "-DQuantumRange=%g -DMagickEpsilon=%g"
 130 #define CLPixelPacket  cl_ushort4
 131 #elif (MAGICKCORE_QUANTUM_DEPTH == 32)
 132 #define CLOptions "-DCLQuantum=uint -DCLPixelType=uint4 " \
 133   "-DQuantumRange=%g -DMagickEpsilon=%g"
 134 #define CLPixelPacket  cl_uint4
 135 #elif (MAGICKCORE_QUANTUM_DEPTH == 64)
 136 #define CLOptions "-DCLQuantum=ussize_t -DCLPixelType=ussize_t4 " \
 137   "-DQuantumRange=%g -DMagickEpsilon=%g"
 138 #define CLPixelPacket  cl_ulong4
 139 #endif
 140 #endif
 141
 142 typedef struct _ConvolveInfo
 143 {
 144   cl_context
 145     context;
 146
 147   cl_device_id
 148     *devices;
 149
 150   cl_command_queue
 151     command_queue;
 152
 153   cl_kernel
 154     kernel;
 155
 156   cl_program
 157     program;
 158
 159   cl_mem
 160     pixels,
 161     convolve_pixels;
 162
 163   cl_ulong
 164     width,
 165     height;
 166
 167   cl_bool
 168     matte;
 169
 170   cl_mem
 171     filter;
 172 } ConvolveInfo;
 173
 174 static char
 175   *ConvolveKernel =
 176     "static inline long ClampToCanvas(const long offset,const unsigned long range)\n"
 177     "{\n"
 178     "  if (offset < 0L)\n"
 179     "    return(0L);\n"
 180     "  if (offset >= range)\n"
 181     "    return((long) (range-1L));\n"
 182     "  return(offset);\n"
 183     "}\n"
 184     "\n"
 185     "static inline CLQuantum ClampToQuantum(const double value)\n"
 186     "{\n"
 187     "#if defined(MAGICKCORE_HDRI_SUPPORT)\n"
 188     "  return((CLQuantum) value)\n"
 189     "#else\n"
 190     "  if (value < 0.0)\n"
 191     "    return((CLQuantum) 0);\n"
 192     "  if (value >= (double) QuantumRange)\n"
 193     "    return((CLQuantum) QuantumRange);\n"
 194     "  return((CLQuantum) (value+0.5));\n"
 195     "#endif\n"
 196     "}\n"
 197     "\n"
 198     "__kernel void Convolve(const __global CLPixelType *input,\n"
 199     "  __constant double *filter,const unsigned long width,const unsigned long height,\n"
 200     "  const bool matte,__global CLPixelType *output)\n"
 201     "{\n"
 202     "  const unsigned long columns = get_global_size(0);\n"
 203     "  const unsigned long rows = get_global_size(1);\n"
 204     "\n"
 205     "  const long x = get_global_id(0);\n"
 206     "  const long y = get_global_id(1);\n"
 207     "\n"
 208     "  const double scale = (1.0/QuantumRange);\n"
 209     "  const long mid_width = (width-1)/2;\n"
 210     "  const long mid_height = (height-1)/2;\n"
 211     "  double4 sum = { 0.0, 0.0, 0.0, 0.0 };\n"
 212     "  double gamma = 0.0;\n"
 213     "  register unsigned long i = 0;\n"
 214     "\n"
 215     "  int method = 0;\n"
 216     "  if (matte != false)\n"
 217     "    method=1;\n"
 218     "  if ((x >= width) && (x < (columns-width-1)) &&\n"
 219     "      (y >= height) && (y < (rows-height-1)))\n"
 220     "    {\n"
 221     "      method=2;\n"
 222     "      if (matte != false)\n"
 223     "        method=3;\n"
 224     "    }\n"
 225     "  switch (method)\n"
 226     "  {\n"
 227     "    case 0:\n"
 228     "    {\n"
 229     "      for (long v=(-mid_height); v <= mid_height; v++)\n"
 230     "      {\n"
 231     "        for (long u=(-mid_width); u <= mid_width; u++)\n"
 232     "        {\n"
 233     "          const long index=ClampToCanvas(y+v,rows)*columns+\n"
 234     "            ClampToCanvas(x+u,columns);\n"
 235     "          sum.x+=filter[i]*input[index].x;\n"
 236     "          sum.y+=filter[i]*input[index].y;\n"
 237     "          sum.z+=filter[i]*input[index].z;\n"
 238     "          gamma+=filter[i];\n"
 239     "          i++;\n"
 240     "        }\n"
 241     "      }\n"
 242     "      break;\n"
 243     "    }\n"
 244     "    case 1:\n"
 245     "    {\n"
 246     "      for (long v=(-mid_height); v <= mid_height; v++)\n"
 247     "      {\n"
 248     "        for (long u=(-mid_width); u <= mid_width; u++)\n"
 249     "        {\n"
 250     "          const unsigned long index=ClampToCanvas(y+v,rows)*columns+\n"
 251     "            ClampToCanvas(x+u,columns);\n"
 252     "          const double alpha=scale*input[index].w;\n"
 253     "          sum.x+=alpha*filter[i]*input[index].x;\n"
 254     "          sum.y+=alpha*filter[i]*input[index].y;\n"
 255     "          sum.z+=alpha*filter[i]*input[index].z;\n"
 256     "          sum.w+=filter[i]*input[index].w;\n"
 257     "          gamma+=alpha*filter[i];\n"
 258     "          i++;\n"
 259     "        }\n"
 260     "      }\n"
 261     "      break;\n"
 262     "    }\n"
 263     "    case 2:\n"
 264     "    {\n"
 265     "      for (long v=(-mid_height); v <= mid_height; v++)\n"
 266     "      {\n"
 267     "        for (long u=(-mid_width); u <= mid_width; u++)\n"
 268     "        {\n"
 269     "          const unsigned long index=(y+v)*columns+(x+u);\n"
 270     "          sum.x+=filter[i]*input[index].x;\n"
 271     "          sum.y+=filter[i]*input[index].y;\n"
 272     "          sum.z+=filter[i]*input[index].z;\n"
 273     "          gamma+=filter[i];\n"
 274     "          i++;\n"
 275     "        }\n"
 276     "      }\n"
 277     "      break;\n"
 278     "    }\n"
 279     "    case 3:\n"
 280     "    {\n"
 281     "      for (long v=(-mid_height); v <= mid_height; v++)\n"
 282     "      {\n"
 283     "        for (long u=(-mid_width); u <= mid_width; u++)\n"
 284     "        {\n"
 285     "          const unsigned long index=(y+v)*columns+(x+u);\n"
 286     "          const double alpha=scale*input[index].w;\n"
 287     "          sum.x+=alpha*filter[i]*input[index].x;\n"
 288     "          sum.y+=alpha*filter[i]*input[index].y;\n"
 289     "          sum.z+=alpha*filter[i]*input[index].z;\n"
 290     "          sum.w+=filter[i]*input[index].w;\n"
 291     "          gamma+=alpha*filter[i];\n"
 292     "          i++;\n"
 293     "        }\n"
 294     "      }\n"
 295     "      break;\n"
 296     "    }\n"
 297     "  }\n"
 298     "  gamma=1.0/(fabs(gamma) <= MagickEpsilon ? 1.0 : gamma);\n"
 299     "  const unsigned long index = y*columns+x;\n"
 300     "  output[index].x=ClampToQuantum(gamma*sum.x);\n"
 301     "  output[index].y=ClampToQuantum(gamma*sum.y);\n"
 302     "  output[index].z=ClampToQuantum(gamma*sum.z);\n"
 303     "  if (matte == false)\n"
 304     "    output[index].w=input[index].w;\n"
 305     "  else\n"
 306     "    output[index].w=ClampToQuantum(sum.w);\n"
 307     "}\n";
 308
 309 static void ConvolveNotify(const char *message,const void *data,size_t length,
 310   void *user_context)
 311 {
 312   ExceptionInfo
 313     *exception;
 314
 315   (void) data;
 316   (void) length;
 317   exception=(ExceptionInfo *) user_context;
 318   (void) ThrowMagickException(exception,GetMagickModule(),DelegateWarning,
 319     "DelegateFailed","`%s'",message);
 320 }
 321
 322 static MagickBooleanType BindConvolveParameters(ConvolveInfo *convolve_info,
 323   const Image *image,const void *pixels,double *filter,
 324   const size_t width,const size_t height,void *convolve_pixels)
 325 {
 326   cl_int
 327     status;
 328
 329   register cl_uint
 330     i;
 331
 332   size_t
 333     length;
 334
 335   /*
 336     Allocate OpenCL buffers.
 337   */
 338   length=image->columns*image->rows;
 339   convolve_info->pixels=clCreateBuffer(convolve_info->context,(cl_mem_flags)
 340     (CL_MEM_READ_ONLY | CL_MEM_USE_HOST_PTR),length*sizeof(CLPixelPacket),
 341     (void *) pixels,&status);
 342   if ((convolve_info->pixels == (cl_mem) NULL) || (status != CL_SUCCESS))
 343     return(MagickFalse);
 344   length=width*height;
 345   convolve_info->filter=clCreateBuffer(convolve_info->context,(cl_mem_flags)
 346     (CL_MEM_READ_ONLY | CL_MEM_USE_HOST_PTR),length*sizeof(cl_double),filter,
 347     &status);
 348   if ((convolve_info->filter == (cl_mem) NULL) || (status != CL_SUCCESS))
 349     return(MagickFalse);
 350   length=image->columns*image->rows;
 351   convolve_info->convolve_pixels=clCreateBuffer(convolve_info->context,
 352     (cl_mem_flags) (CL_MEM_WRITE_ONLY | CL_MEM_USE_HOST_PTR),length*
 353     sizeof(CLPixelPacket),convolve_pixels,&status);
 354   if ((convolve_info->convolve_pixels == (cl_mem) NULL) ||
 355       (status != CL_SUCCESS))
 356     return(MagickFalse);
 357   /*
 358     Bind OpenCL buffers.
 359   */
 360   i=0;
 361   status=clSetKernelArg(convolve_info->kernel,i++,sizeof(cl_mem),(void *)
 362     &convolve_info->pixels);
 363   if (status != CL_SUCCESS)
 364     return(MagickFalse);
 365   status=clSetKernelArg(convolve_info->kernel,i++,sizeof(cl_mem),(void *)
 366     &convolve_info->filter);
 367   if (status != CL_SUCCESS)
 368     return(MagickFalse);
 369   convolve_info->width=(cl_ulong) width;
 370   status=clSetKernelArg(convolve_info->kernel,i++,sizeof(cl_ulong),(void *)
 371     &convolve_info->width);
 372   if (status != CL_SUCCESS)
 373     return(MagickFalse);
 374   convolve_info->height=(cl_ulong) height;
 375   status=clSetKernelArg(convolve_info->kernel,i++,sizeof(cl_ulong),(void *)
 376     &convolve_info->height);
 377   if (status != CL_SUCCESS)
 378     return(MagickFalse);
 379   convolve_info->matte=(cl_bool) image->matte;
 380   status=clSetKernelArg(convolve_info->kernel,i++,sizeof(cl_bool),(void *)
 381     &convolve_info->matte);
 382   if (status != CL_SUCCESS)
 383     return(MagickFalse);
 384   status=clSetKernelArg(convolve_info->kernel,i++,sizeof(cl_mem),(void *)
 385     &convolve_info->convolve_pixels);
 386   if (status != CL_SUCCESS)
 387     return(MagickFalse);
 388   status=clFinish(convolve_info->command_queue);
 389   if (status != CL_SUCCESS)
 390     return(MagickFalse);
 391   return(MagickTrue);
 392 }
 393
 394 static void DestroyConvolveBuffers(ConvolveInfo *convolve_info)
 395 {
 396   cl_int
 397     status;
 398
 399   (void) status;
 400   if (convolve_info->convolve_pixels != (cl_mem) NULL)
 401     status=clReleaseMemObject(convolve_info->convolve_pixels);
 402   if (convolve_info->pixels != (cl_mem) NULL)
 403     status=clReleaseMemObject(convolve_info->pixels);
 404   if (convolve_info->filter != (cl_mem) NULL)
 405     status=clReleaseMemObject(convolve_info->filter);
 406 }
 407
 408 static ConvolveInfo *DestroyConvolveInfo(ConvolveInfo *convolve_info)
 409 {
 410   cl_int
 411     status;
 412
 413   (void) status;
 414   if (convolve_info->kernel != (cl_kernel) NULL)
 415     status=clReleaseKernel(convolve_info->kernel);
 416   if (convolve_info->program != (cl_program) NULL)
 417     status=clReleaseProgram(convolve_info->program);
 418   if (convolve_info->command_queue != (cl_command_queue) NULL)
 419     status=clReleaseCommandQueue(convolve_info->command_queue);
 420   if (convolve_info->context != (cl_context) NULL)
 421     status=clReleaseContext(convolve_info->context);
 422   convolve_info=(ConvolveInfo *) RelinquishMagickMemory(convolve_info);
 423   return(convolve_info);
 424 }
 425
 426 static MagickBooleanType EnqueueConvolveKernel(ConvolveInfo *convolve_info,
 427   const Image *image,const void *pixels,double *filter,
 428   const size_t width,const size_t height,void *convolve_pixels)
 429 {
 430   cl_int
 431     status;
 432
 433   size_t
 434     global_work_size[2],
 435     length;
 436
 437   length=image->columns*image->rows;
 438   status=clEnqueueWriteBuffer(convolve_info->command_queue,
 439     convolve_info->pixels,CL_TRUE,0,length*sizeof(CLPixelPacket),pixels,0,NULL,
 440     NULL);
 441   length=width*height;
 442   status=clEnqueueWriteBuffer(convolve_info->command_queue,
 443     convolve_info->filter,CL_TRUE,0,length*sizeof(cl_double),filter,0,NULL,
 444     NULL);
 445   if (status != CL_SUCCESS)
 446     return(MagickFalse);
 447   global_work_size[0]=image->columns;
 448   global_work_size[1]=image->rows;
 449   status=clEnqueueNDRangeKernel(convolve_info->command_queue,
 450     convolve_info->kernel,2,NULL,global_work_size,NULL,0,NULL,NULL);
 451   if (status != CL_SUCCESS)
 452     return(MagickFalse);
 453   length=image->columns*image->rows;
 454   status=clEnqueueReadBuffer(convolve_info->command_queue,
 455     convolve_info->convolve_pixels,CL_TRUE,0,length*sizeof(CLPixelPacket),
 456     convolve_pixels,0,NULL,NULL);
 457   if (status != CL_SUCCESS)
 458     return(MagickFalse);
 459   status=clFinish(convolve_info->command_queue);
 460   if (status != CL_SUCCESS)
 461     return(MagickFalse);
 462   return(MagickTrue);
 463 }
 464
 465 static ConvolveInfo *GetConvolveInfo(const Image *image,const char *name,
 466   const char *source,ExceptionInfo *exception)
 467 {
 468   char
 469     options[MaxTextExtent];
 470
 471   cl_int
 472     status;
 473
 474   ConvolveInfo
 475     *convolve_info;
 476
 477   size_t
 478     length,
 479     lengths[] = { strlen(source) };
 480
 481   /*
 482     Create OpenCL info.
 483   */
 484   convolve_info=(ConvolveInfo *) AcquireMagickMemory(sizeof(*convolve_info));
 485   if (convolve_info == (ConvolveInfo *) NULL)
 486     {
 487       (void) ThrowMagickException(exception,GetMagickModule(),
 488         ResourceLimitError,"MemoryAllocationFailed","`%s'",image->filename);
 489       return((ConvolveInfo *) NULL);
 490     }
 491   (void) ResetMagickMemory(convolve_info,0,sizeof(*convolve_info));
 492   /*
 493     Create OpenCL context.
 494   */
 495   convolve_info->context=clCreateContextFromType((cl_context_properties *)
 496     NULL,(cl_device_type) CL_DEVICE_TYPE_GPU,ConvolveNotify,exception,&status);
 497   if ((convolve_info->context == (cl_context) NULL) || (status != CL_SUCCESS))
 498     convolve_info->context=clCreateContextFromType((cl_context_properties *)
 499       NULL,(cl_device_type) CL_DEVICE_TYPE_CPU,ConvolveNotify,exception,
 500       &status);
 501   if ((convolve_info->context == (cl_context) NULL) || (status != CL_SUCCESS))
 502     convolve_info->context=clCreateContextFromType((cl_context_properties *)
 503       NULL,(cl_device_type) CL_DEVICE_TYPE_DEFAULT,ConvolveNotify,exception,
 504       &status);
 505   if ((convolve_info->context == (cl_context) NULL) || (status != CL_SUCCESS))
 506     {
 507       (void) ThrowMagickException(exception,GetMagickModule(),DelegateWarning,
 508         "failed to create OpenCL context","`%s' (%d)",image->filename,status);
 509       convolve_info=DestroyConvolveInfo(convolve_info);
 510       return((ConvolveInfo *) NULL);
 511     }
 512   /*
 513     Detect OpenCL devices.
 514   */
 515   status=clGetContextInfo(convolve_info->context,CL_CONTEXT_DEVICES,0,NULL,
 516     &length);
 517   if ((status != CL_SUCCESS) || (length == 0))
 518     {
 519       convolve_info=DestroyConvolveInfo(convolve_info);
 520       return((ConvolveInfo *) NULL);
 521     }
 522   convolve_info->devices=(cl_device_id *) AcquireMagickMemory(length);
 523   if (convolve_info->devices == (cl_device_id *) NULL)
 524     {
 525       (void) ThrowMagickException(exception,GetMagickModule(),
 526         ResourceLimitError,"MemoryAllocationFailed","`%s'",image->filename);
 527       convolve_info=DestroyConvolveInfo(convolve_info);
 528       return((ConvolveInfo *) NULL);
 529     }
 530   status=clGetContextInfo(convolve_info->context,CL_CONTEXT_DEVICES,length,
 531     convolve_info->devices,NULL);
 532   if (status != CL_SUCCESS)
 533     {
 534       convolve_info=DestroyConvolveInfo(convolve_info);
 535       return((ConvolveInfo *) NULL);
 536     }
 537   /*
 538     Create OpenCL command queue.
 539   */
 540   convolve_info->command_queue=clCreateCommandQueue(convolve_info->context,
 541     convolve_info->devices[0],0,&status);
 542   if ((convolve_info->command_queue == (cl_command_queue) NULL) ||
 543       (status != CL_SUCCESS))
 544     {
 545       convolve_info=DestroyConvolveInfo(convolve_info);
 546       return((ConvolveInfo *) NULL);
 547     }
 548   /*
 549     Build OpenCL program.
 550   */
 551   convolve_info->program=clCreateProgramWithSource(convolve_info->context,1,
 552     &source,lengths,&status);
 553   if ((convolve_info->program == (cl_program) NULL) || (status != CL_SUCCESS))
 554     {
 555       convolve_info=DestroyConvolveInfo(convolve_info);
 556       return((ConvolveInfo *) NULL);
 557     }
 558   (void) FormatLocaleString(options,MaxTextExtent,CLOptions,(double)
 559     QuantumRange,MagickEpsilon);
 560   status=clBuildProgram(convolve_info->program,1,convolve_info->devices,options,
 561     NULL,NULL);
 562   if ((convolve_info->program == (cl_program) NULL) || (status != CL_SUCCESS))
 563     {
 564       char
 565         *log;
 566
 567       status=clGetProgramBuildInfo(convolve_info->program,
 568         convolve_info->devices[0],CL_PROGRAM_BUILD_LOG,0,NULL,&length);
 569       log=(char *) AcquireMagickMemory(length);
 570       if (log == (char *) NULL)
 571         {
 572           convolve_info=DestroyConvolveInfo(convolve_info);
 573           return((ConvolveInfo *) NULL);
 574         }
 575       status=clGetProgramBuildInfo(convolve_info->program,
 576         convolve_info->devices[0],CL_PROGRAM_BUILD_LOG,length,log,&length);
 577       (void) ThrowMagickException(exception,GetMagickModule(),DelegateWarning,
 578         "failed to build OpenCL program","`%s' (%s)",image->filename,log);
 579       log=DestroyString(log);
 580       convolve_info=DestroyConvolveInfo(convolve_info);
 581       return((ConvolveInfo *) NULL);
 582     }
 583   /*
 584     Get a kernel object.
 585   */
 586   convolve_info->kernel=clCreateKernel(convolve_info->program,name,&status);
 587   if ((convolve_info->kernel == (cl_kernel) NULL) || (status != CL_SUCCESS))
 588     {
 589       convolve_info=DestroyConvolveInfo(convolve_info);
 590       return((ConvolveInfo *) NULL);
 591     }
 592   return(convolve_info);
 593 }
 594
 595 #endif
 596
 597 MagickExport MagickBooleanType AccelerateConvolveImage(const Image *image,
 598   const KernelInfo *kernel,Image *convolve_image,ExceptionInfo *exception)
 599 {
 600   assert(image != (Image *) NULL);
 601   assert(image->signature == MagickSignature);
 602   if (image->debug != MagickFalse)
 603     (void) LogMagickEvent(TraceEvent,GetMagickModule(),"%s",image->filename);
 604   assert(kernel != (KernelInfo *) NULL);
 605   assert(kernel->signature == MagickSignature);
 606   assert(convolve_image != (Image *) NULL);
 607   assert(convolve_image->signature == MagickSignature);
 608   assert(exception != (ExceptionInfo *) NULL);
 609   assert(exception->signature == MagickSignature);
 610   if ((image->storage_class != DirectClass) ||
 611       (image->colorspace == CMYKColorspace))
 612   if ((GetImageVirtualPixelMethod(image) != UndefinedVirtualPixelMethod) &&
 613       (GetImageVirtualPixelMethod(image) != EdgeVirtualPixelMethod))
 614     return(MagickFalse);
 615 #if !defined(MAGICKCORE_OPENCL_SUPPORT)
 616   return(MagickFalse);
 617 #else
 618   {
 619     const void
 620       *pixels;
 621
 622     ConvolveInfo
 623       *convolve_info;
 624
 625     MagickBooleanType
 626       status;
 627
 628     MagickSizeType
 629       length;
 630
 631     void
 632       *convolve_pixels;
 633
 634     convolve_info=GetConvolveInfo(image,"Convolve",ConvolveKernel,exception);
 635     if (convolve_info == (ConvolveInfo *) NULL)
 636       return(MagickFalse);
 637     pixels=AcquirePixelCachePixels(image,&length,exception);
 638     if (pixels == (const void *) NULL)
 639       {
 640         (void) ThrowMagickException(exception,GetMagickModule(),CacheError,
 641           "UnableToReadPixelCache","`%s'",image->filename);
 642         convolve_info=DestroyConvolveInfo(convolve_info);
 643         return(MagickFalse);
 644       }
 645     convolve_pixels=GetPixelCachePixels(convolve_image,&length,exception);
 646     if (convolve_pixels == (void *) NULL)
 647       {
 648         (void) ThrowMagickException(exception,GetMagickModule(),CacheError,
 649           "UnableToReadPixelCache","`%s'",image->filename);
 650         convolve_info=DestroyConvolveInfo(convolve_info);
 651         return(MagickFalse);
 652       }
 653     status=BindConvolveParameters(convolve_info,image,pixels,kernel->values,
 654       kernel->width,kernel->height,convolve_pixels);
 655     if (status == MagickFalse)
 656       {
 657         DestroyConvolveBuffers(convolve_info);
 658         convolve_info=DestroyConvolveInfo(convolve_info);
 659         return(MagickFalse);
 660       }
 661     status=EnqueueConvolveKernel(convolve_info,image,pixels,kernel->values,
 662       kernel->width,kernel->height,convolve_pixels);
 663     if (status == MagickFalse)
 664       {
 665         DestroyConvolveBuffers(convolve_info);
 666         convolve_info=DestroyConvolveInfo(convolve_info);
 667         return(MagickFalse);
 668       }
 669     DestroyConvolveBuffers(convolve_info);
 670     convolve_info=DestroyConvolveInfo(convolve_info);
 671     return(MagickTrue);
 672   }
 673 #endif
 674 }