You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
669 lines
22 KiB
669 lines
22 KiB
//
|
|
// Copyright (c) 2017 The Khronos Group Inc.
|
|
//
|
|
// Licensed under the Apache License, Version 2.0 (the "License");
|
|
// you may not use this file except in compliance with the License.
|
|
// You may obtain a copy of the License at
|
|
//
|
|
// http://www.apache.org/licenses/LICENSE-2.0
|
|
//
|
|
// Unless required by applicable law or agreed to in writing, software
|
|
// distributed under the License is distributed on an "AS IS" BASIS,
|
|
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
// See the License for the specific language governing permissions and
|
|
// limitations under the License.
|
|
//
|
|
#include "harness/compat.h"
|
|
|
|
#include <stdio.h>
|
|
#include <string.h>
|
|
#include <sys/types.h>
|
|
#include <sys/stat.h>
|
|
|
|
#include "procs.h"
|
|
#include "harness/testHarness.h"
|
|
#include "harness/errorHelpers.h"
|
|
|
|
//--- the code for the kernel executables
|
|
static const char *readKernelCode[] = {
|
|
"__kernel void testReadf(read_only image2d_t srcimg, __global float4 *dst)\n"
|
|
"{\n"
|
|
" int tid_x = get_global_id(0);\n"
|
|
" int tid_y = get_global_id(1);\n"
|
|
" int indx = tid_y * get_image_width(srcimg) + tid_x;\n"
|
|
" float4 color;\n"
|
|
"\n"
|
|
" const sampler_t sampler = CLK_ADDRESS_CLAMP_TO_EDGE | CLK_FILTER_NEAREST | CLK_NORMALIZED_COORDS_FALSE;\n"
|
|
" color = read_imagef(srcimg, sampler, (int2)(tid_x, tid_y));\n"
|
|
" dst[indx].x = color.x;\n"
|
|
" dst[indx].y = color.y;\n"
|
|
" dst[indx].z = color.z;\n"
|
|
" dst[indx].w = color.w;\n"
|
|
"\n"
|
|
"}\n",
|
|
|
|
"__kernel void testReadi(read_only image2d_t srcimg, __global uchar4 *dst)\n"
|
|
"{\n"
|
|
" int tid_x = get_global_id(0);\n"
|
|
" int tid_y = get_global_id(1);\n"
|
|
" int indx = tid_y * get_image_width(srcimg) + tid_x;\n"
|
|
" int4 color;\n"
|
|
"\n"
|
|
" const sampler_t sampler = CLK_ADDRESS_CLAMP_TO_EDGE | CLK_FILTER_NEAREST | CLK_NORMALIZED_COORDS_FALSE;\n"
|
|
" color = read_imagei(srcimg, sampler, (int2)(tid_x, tid_y));\n"
|
|
" uchar4 dst_write;\n"
|
|
" dst_write.x = (uchar)color.x;\n"
|
|
" dst_write.y = (uchar)color.y;\n"
|
|
" dst_write.z = (uchar)color.z;\n"
|
|
" dst_write.w = (uchar)color.w;\n"
|
|
" dst[indx] = dst_write;\n"
|
|
"\n"
|
|
"}\n",
|
|
|
|
"__kernel void testReadui(read_only image2d_t srcimg, __global uchar4 *dst)\n"
|
|
"{\n"
|
|
" int tid_x = get_global_id(0);\n"
|
|
" int tid_y = get_global_id(1);\n"
|
|
" int indx = tid_y * get_image_width(srcimg) + tid_x;\n"
|
|
" uint4 color;\n"
|
|
"\n"
|
|
" const sampler_t sampler = CLK_ADDRESS_CLAMP_TO_EDGE | CLK_FILTER_NEAREST | CLK_NORMALIZED_COORDS_FALSE;\n"
|
|
" color = read_imageui(srcimg, sampler, (int2)(tid_x, tid_y));\n"
|
|
" uchar4 dst_write;\n"
|
|
" dst_write.x = (uchar)color.x;\n"
|
|
" dst_write.y = (uchar)color.y;\n"
|
|
" dst_write.z = (uchar)color.z;\n"
|
|
" dst_write.w = (uchar)color.w;\n"
|
|
" dst[indx] = dst_write;\n"
|
|
"\n"
|
|
"}\n",
|
|
|
|
"__kernel void testWritef(__global uchar *src, write_only image2d_t dstimg)\n"
|
|
"{\n"
|
|
" int tid_x = get_global_id(0);\n"
|
|
" int tid_y = get_global_id(1);\n"
|
|
" int indx = tid_y * get_image_width(dstimg) + tid_x;\n"
|
|
" float4 color;\n"
|
|
"\n"
|
|
" indx *= 4;\n"
|
|
" color = (float4)((float)src[indx+0], (float)src[indx+1], (float)src[indx+2], (float)src[indx+3]);\n"
|
|
" color /= (float4)(255.f, 255.f, 255.f, 255.f);\n"
|
|
" write_imagef(dstimg, (int2)(tid_x, tid_y), color);\n"
|
|
"\n"
|
|
"}\n",
|
|
|
|
"__kernel void testWritei(__global char *src, write_only image2d_t dstimg)\n"
|
|
"{\n"
|
|
" int tid_x = get_global_id(0);\n"
|
|
" int tid_y = get_global_id(1);\n"
|
|
" int indx = tid_y * get_image_width(dstimg) + tid_x;\n"
|
|
" int4 color;\n"
|
|
"\n"
|
|
" indx *= 4;\n"
|
|
" color.x = (int)src[indx+0];\n"
|
|
" color.y = (int)src[indx+1];\n"
|
|
" color.z = (int)src[indx+2];\n"
|
|
" color.w = (int)src[indx+3];\n"
|
|
" write_imagei(dstimg, (int2)(tid_x, tid_y), color);\n"
|
|
"\n"
|
|
"}\n",
|
|
|
|
"__kernel void testWriteui(__global uchar *src, write_only image2d_t dstimg)\n"
|
|
"{\n"
|
|
" int tid_x = get_global_id(0);\n"
|
|
" int tid_y = get_global_id(1);\n"
|
|
" int indx = tid_y * get_image_width(dstimg) + tid_x;\n"
|
|
" uint4 color;\n"
|
|
"\n"
|
|
" indx *= 4;\n"
|
|
" color.x = (uint)src[indx+0];\n"
|
|
" color.y = (uint)src[indx+1];\n"
|
|
" color.z = (uint)src[indx+2];\n"
|
|
" color.w = (uint)src[indx+3];\n"
|
|
" write_imageui(dstimg, (int2)(tid_x, tid_y), color);\n"
|
|
"\n"
|
|
"}\n",
|
|
|
|
"__kernel void testReadWriteff(read_only image2d_t srcimg, write_only image2d_t dstimg)\n"
|
|
"{\n"
|
|
" int tid_x = get_global_id(0);\n"
|
|
" int tid_y = get_global_id(1);\n"
|
|
" float4 color;\n"
|
|
"\n"
|
|
" color = read_imagef(srcimg, CLK_DEFAULT_SAMPLER, (int2)(tid_x, tid_y));\n"
|
|
" write_imagef(dstimg, (int2)(tid_x, tid_y), color);\n"
|
|
"\n"
|
|
"}\n",
|
|
|
|
"__kernel void testReadWriteii(read_only image2d_t srcimg, write_only image2d_t dstimg)\n"
|
|
"{\n"
|
|
" int tid_x = get_global_id(0);\n"
|
|
" int tid_y = get_global_id(1);\n"
|
|
" int4 color;\n"
|
|
"\n"
|
|
" color = read_imagei(srcimg, CLK_DEFAULT_SAMPLER, (int2)(tid_x, tid_y));\n"
|
|
" write_imagei(dstimg, (int2)(tid_x, tid_y), color);\n"
|
|
"\n"
|
|
"}\n",
|
|
|
|
"__kernel void testReadWriteuiui(read_only image2d_t srcimg, write_only image2d_t dstimg)\n"
|
|
"{\n"
|
|
" int tid_x = get_global_id(0);\n"
|
|
" int tid_y = get_global_id(1);\n"
|
|
" uint4 color;\n"
|
|
"\n"
|
|
" color = read_imageui(srcimg, CLK_DEFAULT_SAMPLER, (int2)(tid_x, tid_y));\n"
|
|
" write_imageui(dstimg, (int2)(tid_x, tid_y), color);\n"
|
|
"\n"
|
|
"}\n",
|
|
|
|
"__kernel void testReadWritefi(read_only image2d_t srcimg, write_only image2d_t dstimg)\n"
|
|
"{\n"
|
|
" int tid_x = get_global_id(0);\n"
|
|
" int tid_y = get_global_id(1);\n"
|
|
" float4 colorf;\n"
|
|
" int4 colori;\n"
|
|
"\n"
|
|
" const sampler_t sampler = CLK_ADDRESS_CLAMP_TO_EDGE | CLK_FILTER_NEAREST | CLK_NORMALIZED_COORDS_FALSE;\n"
|
|
" colorf = read_imagef(srcimg, sampler, (int2)(tid_x, tid_y));\n"
|
|
// since we are going from unsigned to signed, be sure to convert
|
|
// values greater 0.5 to negative values
|
|
" if( colorf.x >= 0.5f )\n"
|
|
" colori.x = (int)( ( colorf.x - 1.f ) * 255.f );\n"
|
|
" else\n"
|
|
" colori.x = (int)( colorf.x * 255.f );\n"
|
|
" if( colorf.y >= 0.5f )\n"
|
|
" colori.y = (int)( ( colorf.y - 1.f ) * 255.f );\n"
|
|
" else\n"
|
|
" colori.y = (int)( colorf.y * 255.f );\n"
|
|
" if( colorf.z >= 0.5f )\n"
|
|
" colori.z = (int)( ( colorf.z - 1.f ) * 255.f );\n"
|
|
" else\n"
|
|
" colori.z = (int)( colorf.z * 255.f );\n"
|
|
" if( colorf.w >= 0.5f )\n"
|
|
" colori.w = (int)( ( colorf.w - 1.f ) * 255.f );\n"
|
|
" else\n"
|
|
" colori.w = (int)( colorf.w * 255.f );\n"
|
|
" write_imagei(dstimg, (int2)(tid_x, tid_y), colori);\n"
|
|
"\n"
|
|
"}\n",
|
|
|
|
"__kernel void testReadWritefui(read_only image2d_t srcimg, write_only image2d_t dstimg)\n"
|
|
"{\n"
|
|
" int tid_x = get_global_id(0);\n"
|
|
" int tid_y = get_global_id(1);\n"
|
|
" float4 colorf;\n"
|
|
" uint4 colorui;\n"
|
|
"\n"
|
|
" const sampler_t sampler = CLK_ADDRESS_CLAMP_TO_EDGE | CLK_FILTER_NEAREST | CLK_NORMALIZED_COORDS_FALSE;\n"
|
|
" colorf = read_imagef(srcimg, sampler, (int2)(tid_x, tid_y));\n"
|
|
" colorui.x = (uint)( colorf.x * 255.f );\n"
|
|
" colorui.y = (uint)( colorf.y * 255.f );\n"
|
|
" colorui.z = (uint)( colorf.z * 255.f );\n"
|
|
" colorui.w = (uint)( colorf.w * 255.f );\n"
|
|
" write_imageui(dstimg, (int2)(tid_x, tid_y), colorui);\n"
|
|
"\n"
|
|
"}\n",
|
|
|
|
"__kernel void testReadWriteif(read_only image2d_t srcimg, write_only image2d_t dstimg)\n"
|
|
"{\n"
|
|
" int tid_x = get_global_id(0);\n"
|
|
" int tid_y = get_global_id(1);\n"
|
|
" int4 colori;\n"
|
|
" float4 colorf;\n"
|
|
"\n"
|
|
// since we are going from signed to unsigned, we need to adjust the rgba values from
|
|
// from the signed image to add 256 to the signed image values less than 0.
|
|
" const sampler_t sampler = CLK_ADDRESS_CLAMP_TO_EDGE | CLK_FILTER_NEAREST | CLK_NORMALIZED_COORDS_FALSE;\n"
|
|
" colori = read_imagei(srcimg, sampler, (int2)(tid_x, tid_y));\n"
|
|
" if( colori.x < 0 )\n"
|
|
" colorf.x = ( (float)colori.x + 256.f ) / 255.f;\n"
|
|
" else\n"
|
|
" colorf.x = (float)colori.x / 255.f;\n"
|
|
" if( colori.y < 0 )\n"
|
|
" colorf.y = ( (float)colori.y + 256.f ) / 255.f;\n"
|
|
" else\n"
|
|
" colorf.y = (float)colori.y / 255.f;\n"
|
|
" if( colori.z < 0 )\n"
|
|
" colorf.z = ( (float)colori.z + 256.f ) / 255.f;\n"
|
|
" else\n"
|
|
" colorf.z = (float)colori.z / 255.f;\n"
|
|
" if( colori.w < 0 )\n"
|
|
" colorf.w = ( (float)colori.w + 256.f ) / 255.f;\n"
|
|
" else\n"
|
|
" colorf.w = (float)colori.w / 255.f;\n"
|
|
" write_imagef(dstimg, (int2)(tid_x, tid_y), colorf);\n"
|
|
"\n"
|
|
"}\n",
|
|
|
|
"__kernel void testReadWriteiui(read_only image2d_t srcimg, write_only image2d_t dstimg)\n"
|
|
"{\n"
|
|
" int tid_x = get_global_id(0);\n"
|
|
" int tid_y = get_global_id(1);\n"
|
|
" int4 colori;\n"
|
|
" uint4 colorui;\n"
|
|
"\n"
|
|
" const sampler_t sampler = CLK_ADDRESS_CLAMP_TO_EDGE | CLK_FILTER_NEAREST | CLK_NORMALIZED_COORDS_FALSE;\n"
|
|
" colori = read_imagei(srcimg, sampler, (int2)(tid_x, tid_y));\n"
|
|
// since we are going from signed to unsigned, we need to adjust the rgba values from
|
|
// from the signed image to add 256 to the signed image values less than 0.
|
|
" if( colori.x < 0 )\n"
|
|
" colorui.x = (uint)( colori.x + 256 );\n"
|
|
" else\n"
|
|
" colorui.x = (uint)colori.x;\n"
|
|
" if( colori.y < 0 )\n"
|
|
" colorui.y = (uint)( colori.y + 256 );\n"
|
|
" else\n"
|
|
" colorui.y = (uint)colori.y;\n"
|
|
" if( colori.z < 0 )\n"
|
|
" colorui.z = (uint)( colori.z + 256 );\n"
|
|
" else\n"
|
|
" colorui.z = (uint)colori.z;\n"
|
|
" if( colori.w < 0 )\n"
|
|
" colorui.w = (uint)( colori.w + 256 );\n"
|
|
" else\n"
|
|
" colorui.w = (uint)colori.w;\n"
|
|
" write_imageui(dstimg, (int2)(tid_x, tid_y), colorui);\n"
|
|
"\n"
|
|
"}\n",
|
|
|
|
"__kernel void testReadWriteuif(read_only image2d_t srcimg, write_only image2d_t dstimg)\n"
|
|
"{\n"
|
|
" int tid_x = get_global_id(0);\n"
|
|
" int tid_y = get_global_id(1);\n"
|
|
" uint4 colorui;\n"
|
|
" float4 colorf;\n"
|
|
"\n"
|
|
" const sampler_t sampler = CLK_ADDRESS_CLAMP_TO_EDGE | CLK_FILTER_NEAREST | CLK_NORMALIZED_COORDS_FALSE;\n"
|
|
" colorui = read_imageui(srcimg, sampler, (int2)(tid_x, tid_y));\n"
|
|
" colorf.x = (float)colorui.x / 255.f;\n"
|
|
" colorf.y = (float)colorui.y / 255.f;\n"
|
|
" colorf.z = (float)colorui.z / 255.f;\n"
|
|
" colorf.w = (float)colorui.w / 255.f;\n"
|
|
" write_imagef(dstimg, (int2)(tid_x, tid_y), colorf);\n"
|
|
"\n"
|
|
"}\n",
|
|
|
|
"__kernel void testReadWriteuii(read_only image2d_t srcimg, write_only image2d_t dstimg)\n"
|
|
"{\n"
|
|
" int tid_x = get_global_id(0);\n"
|
|
" int tid_y = get_global_id(1);\n"
|
|
" uint4 colorui;\n"
|
|
" int4 colori;\n"
|
|
"\n"
|
|
" const sampler_t sampler = CLK_ADDRESS_CLAMP_TO_EDGE | CLK_FILTER_NEAREST | CLK_NORMALIZED_COORDS_FALSE;\n"
|
|
" colorui = read_imageui(srcimg, sampler, (int2)(tid_x, tid_y));\n"
|
|
// since we are going from unsigned to signed, be sure to convert
|
|
// values greater 0.5 to negative values
|
|
" if( colorui.x >= 128U )\n"
|
|
" colori.x = (int)colorui.x - 256;\n"
|
|
" else\n"
|
|
" colori.x = (int)colorui.x;\n"
|
|
" if( colorui.y >= 128U )\n"
|
|
" colori.y = (int)colorui.y - 256;\n"
|
|
" else\n"
|
|
" colori.y = (int)colorui.y;\n"
|
|
" if( colorui.z >= 128U )\n"
|
|
" colori.z = (int)colorui.z - 256;\n"
|
|
" else\n"
|
|
" colori.z = (int)colorui.z;\n"
|
|
" if( colorui.w >= 128U )\n"
|
|
" colori.w = (int)colorui.w - 256;\n"
|
|
" else\n"
|
|
" colori.w = (int)colorui.w;\n"
|
|
" write_imagei(dstimg, (int2)(tid_x, tid_y), colori);\n"
|
|
"\n"
|
|
"}\n" };
|
|
|
|
static const char *readKernelName[] = { "testReadf", "testReadi", "testReadui", "testWritef", "testWritei", "testWriteui",
|
|
"testReadWriteff", "testReadWriteii", "testReadWriteuiui", "testReadWritefi",
|
|
"testReadWritefui", "testReadWriteif", "testReadWriteiui", "testReadWriteuif",
|
|
"testReadWriteuii" };
|
|
|
|
|
|
static cl_uchar *generateImage( int n, MTdata d )
|
|
{
|
|
cl_uchar *ptr = (cl_uchar *)malloc( n * sizeof( cl_uchar ) );
|
|
int i;
|
|
|
|
for( i = 0; i < n; i++ ){
|
|
ptr[i] = (cl_uchar)genrand_int32(d);
|
|
}
|
|
|
|
return ptr;
|
|
|
|
}
|
|
|
|
|
|
static char *generateSignedImage( int n, MTdata d )
|
|
{
|
|
char *ptr = (char *)malloc( n * sizeof( char ) );
|
|
int i;
|
|
|
|
for( i = 0; i < n; i++ ){
|
|
ptr[i] = (char)genrand_int32(d);
|
|
}
|
|
|
|
return ptr;
|
|
|
|
}
|
|
|
|
|
|
static int verifyImage( cl_uchar *image, cl_uchar *outptr, int w, int h )
|
|
{
|
|
int i;
|
|
|
|
for( i = 0; i < w * h * 4; i++ ){
|
|
if( outptr[i] != image[i] ){
|
|
log_error("Image verification failed at offset %d. Actual value=%d, expected value=%d\n", i, outptr[i], image[i]);
|
|
return -1;
|
|
}
|
|
}
|
|
|
|
return 0;
|
|
}
|
|
|
|
static int verifyImageFloat ( cl_double *refptr, cl_float *outptr, int w, int h )
|
|
{
|
|
int i;
|
|
|
|
for (i=0; i<w*h*4; i++)
|
|
{
|
|
if (outptr[i] != (float)refptr[i])
|
|
{
|
|
float ulps = Ulp_Error( outptr[i], refptr[i]);
|
|
|
|
if(! (fabsf(ulps) < 1.5f) )
|
|
{
|
|
log_error( "ERROR: Data sample %d does not validate! Expected (%a), got (%a), ulp %f\n",
|
|
(int)i, refptr[i], outptr[ i ], ulps );
|
|
return -1;
|
|
}
|
|
}
|
|
}
|
|
|
|
return 0;
|
|
}
|
|
|
|
static double *prepareReference( cl_uchar *inptr, int w, int h)
|
|
{
|
|
int i;
|
|
double *refptr = (double *)malloc( w * h * 4*sizeof( double ) );
|
|
if ( !refptr )
|
|
{
|
|
log_error( "Unable to allocate refptr at %d x %d\n", (int)w, (int)h );
|
|
return 0;
|
|
}
|
|
for( i = 0; i < w * h * 4; i++ ) {
|
|
refptr[i] = ((double)inptr[i])/255;
|
|
}
|
|
return refptr;
|
|
}
|
|
|
|
//----- the test functions
|
|
int write_image( cl_device_id device, cl_context context, cl_command_queue queue, int numElements, const char *code,
|
|
const char *name, cl_image_format image_format_desc, int readFloat )
|
|
{
|
|
cl_mem memobjs[2];
|
|
cl_program program[1];
|
|
void *inptr;
|
|
double *refptr = NULL;
|
|
void *dst = NULL;
|
|
cl_kernel kernel[1];
|
|
cl_event writeEvent;
|
|
cl_ulong queueStart, submitStart, writeStart, writeEnd;
|
|
size_t threads[2];
|
|
int err;
|
|
int w = 64, h = 64;
|
|
cl_mem_flags flags;
|
|
size_t element_nbytes;
|
|
size_t num_bytes;
|
|
size_t channel_nbytes = sizeof( cl_uchar );
|
|
MTdata d;
|
|
|
|
|
|
PASSIVE_REQUIRE_IMAGE_SUPPORT( device )
|
|
|
|
if (readFloat)
|
|
channel_nbytes = sizeof( cl_float );
|
|
|
|
element_nbytes = channel_nbytes * get_format_channel_count( &image_format_desc );
|
|
num_bytes = w * h * element_nbytes;
|
|
|
|
threads[0] = (size_t)w;
|
|
threads[1] = (size_t)h;
|
|
|
|
d = init_genrand( gRandomSeed );
|
|
if( image_format_desc.image_channel_data_type == CL_SIGNED_INT8 )
|
|
inptr = (void *)generateSignedImage( w * h * 4, d );
|
|
else
|
|
inptr = (void *)generateImage( w * h * 4, d );
|
|
free_mtdata(d); d = NULL;
|
|
if( ! inptr ){
|
|
log_error("unable to allocate inptr at %d x %d\n", (int)w, (int)h );
|
|
return -1;
|
|
}
|
|
|
|
dst = malloc( num_bytes );
|
|
if( ! dst ){
|
|
free( (void *)inptr );
|
|
log_error("unable to allocate dst at %d x %d\n", (int)w, (int)h );
|
|
return -1;
|
|
}
|
|
|
|
// allocate the input and output image memory objects
|
|
flags = CL_MEM_READ_WRITE;
|
|
memobjs[0] = create_image_2d( context, flags, &image_format_desc, w, h, 0, NULL, &err );
|
|
if( memobjs[0] == (cl_mem)0 ){
|
|
free( dst );
|
|
free( (void *)inptr );
|
|
log_error("unable to create Image2D\n");
|
|
return -1;
|
|
}
|
|
|
|
memobjs[1] = clCreateBuffer(context, CL_MEM_READ_WRITE,
|
|
channel_nbytes * 4 * w * h, NULL, &err);
|
|
if( memobjs[1] == (cl_mem)0 ){
|
|
free( dst );
|
|
free( (void *)inptr );
|
|
clReleaseMemObject(memobjs[0]);
|
|
log_error("unable to create array\n");
|
|
return -1;
|
|
}
|
|
|
|
size_t origin[3] = { 0, 0, 0 };
|
|
size_t region[3] = { w, h, 1 };
|
|
err = clEnqueueWriteImage( queue, memobjs[0], false, origin, region, 0, 0, inptr, 0, NULL, &writeEvent );
|
|
if( err != CL_SUCCESS ){
|
|
clReleaseMemObject(memobjs[0]);
|
|
clReleaseMemObject(memobjs[1]);
|
|
free( dst );
|
|
free( inptr );
|
|
print_error(err, "clWriteImage failed");
|
|
return -1;
|
|
}
|
|
|
|
// This synchronization point is needed in order to assume the data is valid.
|
|
// Getting profiling information is not a synchronization point.
|
|
err = clWaitForEvents( 1, &writeEvent );
|
|
if( err != CL_SUCCESS )
|
|
{
|
|
print_error( err, "clWaitForEvents failed" );
|
|
clReleaseEvent(writeEvent);
|
|
clReleaseMemObject(memobjs[0]);
|
|
clReleaseMemObject(memobjs[1]);
|
|
free( dst );
|
|
free( inptr );
|
|
return -1;
|
|
}
|
|
|
|
// test profiling
|
|
while( ( err = clGetEventProfilingInfo( writeEvent, CL_PROFILING_COMMAND_QUEUED, sizeof( cl_ulong ), &queueStart, NULL ) ) ==
|
|
CL_PROFILING_INFO_NOT_AVAILABLE );
|
|
if( err != CL_SUCCESS ){
|
|
print_error( err, "clGetEventProfilingInfo failed" );
|
|
clReleaseEvent(writeEvent);
|
|
clReleaseMemObject(memobjs[0]);
|
|
clReleaseMemObject(memobjs[1]);
|
|
free( dst );
|
|
free( inptr );
|
|
return -1;
|
|
}
|
|
|
|
while( ( err = clGetEventProfilingInfo( writeEvent, CL_PROFILING_COMMAND_SUBMIT, sizeof( cl_ulong ), &submitStart, NULL ) ) ==
|
|
CL_PROFILING_INFO_NOT_AVAILABLE );
|
|
if( err != CL_SUCCESS ){
|
|
print_error( err, "clGetEventProfilingInfo failed" );
|
|
clReleaseEvent(writeEvent);
|
|
clReleaseMemObject(memobjs[0]);
|
|
clReleaseMemObject(memobjs[1]);
|
|
free( dst );
|
|
free( inptr );
|
|
return -1;
|
|
}
|
|
|
|
err = clGetEventProfilingInfo( writeEvent, CL_PROFILING_COMMAND_START, sizeof( cl_ulong ), &writeStart, NULL );
|
|
if( err != CL_SUCCESS ){
|
|
print_error( err, "clGetEventProfilingInfo failed" );
|
|
clReleaseEvent(writeEvent);
|
|
clReleaseMemObject(memobjs[0]);
|
|
clReleaseMemObject(memobjs[1]);
|
|
free( dst );
|
|
free( inptr );
|
|
return -1;
|
|
}
|
|
|
|
err = clGetEventProfilingInfo( writeEvent, CL_PROFILING_COMMAND_END, sizeof( cl_ulong ), &writeEnd, NULL );
|
|
if( err != CL_SUCCESS ){
|
|
print_error( err, "clGetEventProfilingInfo failed" );
|
|
clReleaseEvent(writeEvent);
|
|
clReleaseMemObject(memobjs[0]);
|
|
clReleaseMemObject(memobjs[1]);
|
|
free( dst );
|
|
free( inptr );
|
|
return -1;
|
|
}
|
|
|
|
err = create_single_kernel_helper( context, &program[0], &kernel[0], 1, &code, name );
|
|
if( err ){
|
|
log_error( "Unable to create program and kernel\n" );
|
|
clReleaseEvent(writeEvent);
|
|
clReleaseMemObject(memobjs[0]);
|
|
clReleaseMemObject(memobjs[1]);
|
|
free( dst );
|
|
free( inptr );
|
|
return -1;
|
|
}
|
|
|
|
err = clSetKernelArg( kernel[0], 0, sizeof( cl_mem ), (void *)&memobjs[0] );
|
|
err |= clSetKernelArg( kernel[0], 1, sizeof( cl_mem ), (void *)&memobjs[1] );
|
|
if( err != CL_SUCCESS ){
|
|
log_error( "clSetKernelArg failed\n" );
|
|
clReleaseEvent(writeEvent);
|
|
clReleaseKernel( kernel[0] );
|
|
clReleaseProgram( program[0] );
|
|
clReleaseMemObject(memobjs[0]);
|
|
clReleaseMemObject(memobjs[1]);
|
|
free( dst );
|
|
free( inptr );
|
|
return -1;
|
|
}
|
|
|
|
err = clEnqueueNDRangeKernel(queue, kernel[0], 2, NULL, threads, NULL, 0, NULL, NULL );
|
|
|
|
if( err != CL_SUCCESS ){
|
|
print_error( err, "clEnqueueNDRangeKernel failed" );
|
|
clReleaseEvent(writeEvent);
|
|
clReleaseKernel( kernel[0] );
|
|
clReleaseProgram( program[0] );
|
|
clReleaseMemObject(memobjs[0]);
|
|
clReleaseMemObject(memobjs[1]);
|
|
free( dst );
|
|
free( inptr );
|
|
return -1;
|
|
}
|
|
|
|
err = clEnqueueReadBuffer( queue, memobjs[1], true, 0, num_bytes, dst, 0, NULL, NULL );
|
|
if( err != CL_SUCCESS ){
|
|
print_error( err, "clEnqueueReadBuffer failed" );
|
|
clReleaseEvent(writeEvent);
|
|
clReleaseKernel( kernel[0] );
|
|
clReleaseProgram( program[0] );
|
|
clReleaseMemObject(memobjs[0]);
|
|
clReleaseMemObject(memobjs[1]);
|
|
free( dst );
|
|
free( inptr );
|
|
return -1;
|
|
}
|
|
|
|
if ( readFloat )
|
|
{
|
|
refptr = prepareReference( (cl_uchar *)inptr, w, h );
|
|
if ( refptr )
|
|
{
|
|
err = verifyImageFloat( refptr, (cl_float *)dst, w, h );
|
|
free ( refptr );
|
|
}
|
|
else
|
|
err = -1;
|
|
}
|
|
else
|
|
err = verifyImage( (cl_uchar *)inptr, (cl_uchar *)dst, w, h );
|
|
|
|
if( err )
|
|
{
|
|
log_error( "Image failed to verify.\n" );
|
|
}
|
|
else
|
|
{
|
|
log_info( "Image verified.\n" );
|
|
}
|
|
|
|
// cleanup
|
|
clReleaseEvent(writeEvent);
|
|
clReleaseKernel( kernel[0] );
|
|
clReleaseProgram( program[0] );
|
|
clReleaseMemObject(memobjs[0]);
|
|
clReleaseMemObject(memobjs[1]);
|
|
free( dst );
|
|
free( inptr );
|
|
|
|
if (check_times(queueStart, submitStart, writeStart, writeEnd, device))
|
|
err = -1;
|
|
|
|
return err;
|
|
|
|
} // end write_image()
|
|
|
|
|
|
int test_write_image_float( cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
|
|
{
|
|
cl_image_format image_format_desc = { CL_RGBA, CL_UNORM_INT8 };
|
|
PASSIVE_REQUIRE_IMAGE_SUPPORT( device )
|
|
// 0 to 255 for unsigned image data
|
|
return write_image( device, context, queue, numElements, readKernelCode[0], readKernelName[0], image_format_desc, 1 );
|
|
|
|
}
|
|
|
|
|
|
int test_write_image_char( cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
|
|
{
|
|
cl_image_format image_format_desc = { CL_RGBA, CL_SIGNED_INT8 };
|
|
PASSIVE_REQUIRE_IMAGE_SUPPORT( device )
|
|
// -128 to 127 for signed iamge data
|
|
return write_image( device, context, queue, numElements, readKernelCode[1], readKernelName[1], image_format_desc, 0 );
|
|
|
|
}
|
|
|
|
|
|
int test_write_image_uchar( cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
|
|
{
|
|
cl_image_format image_format_desc = { CL_RGBA, CL_UNSIGNED_INT8 };
|
|
PASSIVE_REQUIRE_IMAGE_SUPPORT( device )
|
|
// 0 to 255 for unsigned image data
|
|
return write_image( device, context, queue, numElements, readKernelCode[2], readKernelName[2], image_format_desc, 0 );
|
|
|
|
}
|
|
|
|
|