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.

314 lines
11 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 <time.h>
#include <sys/types.h>
#include <sys/stat.h>
#include "procs.h"
#include "harness/testHarness.h"
#include "harness/errorHelpers.h"
static const char *read3d_kernel_code =
"\n"
"__kernel void read3d(read_only image3d_t srcimg, __global unsigned char *dst, sampler_t sampler)\n"
"{\n"
" int tid_x = get_global_id(0);\n"
" int tid_y = get_global_id(1);\n"
" int tid_z = get_global_id(2);\n"
" int indx = (tid_z * get_image_height(srcimg) + tid_y) * get_image_width(srcimg) + tid_x;\n"
" float4 color;\n"
"\n"
" color = read_imagef(srcimg, sampler, (int4)(tid_x, tid_y, tid_z, 0));\n"
" indx *= 4;\n"
" dst[indx+0] = (unsigned char)(color.x * 255.0f);\n"
" dst[indx+1] = (unsigned char)(color.y * 255.0f);\n"
" dst[indx+2] = (unsigned char)(color.z * 255.0f);\n"
" dst[indx+3] = (unsigned char)(color.w * 255.0f);\n"
"\n"
"}\n";
static cl_uchar *createImage( int elements, MTdata d )
{
int i;
cl_uchar *ptr = (cl_uchar *)malloc( elements * sizeof( cl_uchar ) );
if( ! ptr )
return NULL;
for( i = 0; i < elements; i++ ){
ptr[i] = (cl_uchar)genrand_int32(d);
}
return ptr;
} // end createImage()
static int verifyImages( cl_uchar *ptr0, cl_uchar *ptr1, cl_uchar tolerance, int xsize, int ysize, int zsize, int nChannels )
{
int x, y, z, c;
cl_uchar *p0 = ptr0;
cl_uchar *p1 = ptr1;
for( z = 0; z < zsize; z++ ){
for( y = 0; y < ysize; y++ ){
for( x = 0; x < xsize; x++ ){
for( c = 0; c < nChannels; c++ ){
if( (cl_uchar)abs( (int)( *p0++ - *p1++ ) ) > tolerance ){
log_error( " images differ at x,y,z = %d,%d,%d channel = %d, %d to %d\n",
x, y, z, c, (int)p0[-1], (int)p1[-1] );
return -1;
}
}
}
}
}
return 0;
} // end verifyImages()
static int run_kernel( cl_device_id device, cl_context context, cl_command_queue queue,
int w, int h, int d, int nChannels, cl_uchar *inptr, cl_uchar *outptr )
{
cl_program program[1];
cl_kernel kernel[1];
cl_mem memobjs[2];
cl_image_format image_format_desc = { CL_RGBA, CL_UNORM_INT8 };
cl_event executeEvent = NULL;
cl_ulong queueStart, submitStart, writeStart, writeEnd;
size_t threads[3];
size_t localThreads[3];
int err = 0;
// set thread dimensions
threads[0] = w;
threads[1] = h;
threads[2] = d;
err = clGetDeviceInfo( device, CL_DEVICE_MAX_WORK_ITEM_SIZES, sizeof( cl_uint ), (size_t*)localThreads, NULL );
if (err)
{
localThreads[0] = 256; localThreads[1] = 1; localThreads[2] = 1;
err = 0;
}
if( localThreads[0] > threads[0] )
localThreads[0] = threads[0];
if( localThreads[1] > threads[1] )
localThreads[1] = threads[1];
cl_sampler sampler = clCreateSampler( context, CL_FALSE, CL_ADDRESS_CLAMP_TO_EDGE, CL_FILTER_NEAREST, &err );
if( err ){
log_error( " clCreateSampler failed.\n" );
return -1;
}
// allocate the input and output image memory objects
memobjs[0] =
create_image_3d(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR,
&image_format_desc, w, h, d, 0, 0, inptr, &err);
if( memobjs[0] == (cl_mem)0 ){
log_error( " unable to create 2D image using create_image_2d\n" );
return -1;
}
// allocate an array memory object to load the filter weights
memobjs[1] =
clCreateBuffer(context, CL_MEM_READ_WRITE,
sizeof(cl_float) * w * h * d * nChannels, NULL, &err);
if( memobjs[1] == (cl_mem)0 ){
log_error( " unable to create array using clCreateBuffer\n" );
clReleaseMemObject( memobjs[0] );
return -1;
}
// create the compute program
err = create_single_kernel_helper( context, &program[0], &kernel[0], 1, &read3d_kernel_code, "read3d" );
if( err ){
clReleaseMemObject( memobjs[1] );
clReleaseMemObject( memobjs[0] );
return -1;
}
// create kernel args object and set arg values.
// set the args values
err |= clSetKernelArg( kernel[0], 0, sizeof( cl_mem ), (void *)&memobjs[0] );
err |= clSetKernelArg( kernel[0], 1, sizeof( cl_mem ), (void *)&memobjs[1] );
err |= clSetKernelArg(kernel[0], 2, sizeof sampler, &sampler);
if( err != CL_SUCCESS ){
print_error( err, "clSetKernelArg failed\n" );
clReleaseKernel( kernel[0] );
clReleaseProgram( program[0] );
clReleaseMemObject( memobjs[1] );
clReleaseMemObject( memobjs[0] );
return -1;
}
err = clEnqueueNDRangeKernel( queue, kernel[0], 3, NULL, threads, localThreads, 0, NULL, &executeEvent );
if( err != CL_SUCCESS ){
print_error( err, "clEnqueueNDRangeKernel failed\n" );
clReleaseKernel( kernel[0] );
clReleaseProgram( program[0] );
clReleaseMemObject( memobjs[1] );
clReleaseMemObject( memobjs[0] );
return -1;
}
if (executeEvent) {
// This synchronization point is needed in order to assume the data is valid.
// Getting profiling information is not a synchronization point.
err = clWaitForEvents( 1, &executeEvent );
if( err != CL_SUCCESS )
{
print_error( err, "clWaitForEvents failed\n" );
clReleaseKernel( kernel[0] );
clReleaseProgram( program[0] );
clReleaseMemObject( memobjs[1] );
clReleaseMemObject( memobjs[0] );
return -1;
}
// test profiling
while( ( err = clGetEventProfilingInfo( executeEvent, CL_PROFILING_COMMAND_QUEUED, sizeof( cl_ulong ), &queueStart, NULL ) ) == CL_PROFILING_INFO_NOT_AVAILABLE );
if( err != CL_SUCCESS ){
print_error( err, "clGetEventProfilingInfo failed" );
clReleaseKernel( kernel[0] );
clReleaseProgram( program[0] );
clReleaseMemObject( memobjs[1] );
clReleaseMemObject( memobjs[0] );
return -1;
}
while( ( err = clGetEventProfilingInfo( executeEvent, CL_PROFILING_COMMAND_SUBMIT, sizeof( cl_ulong ), &submitStart, NULL ) ) == CL_PROFILING_INFO_NOT_AVAILABLE );
if( err != CL_SUCCESS ){
print_error( err, "clGetEventProfilingInfo failed" );
clReleaseKernel( kernel[0] );
clReleaseProgram( program[0] );
clReleaseMemObject( memobjs[1] );
clReleaseMemObject( memobjs[0] );
return -1;
}
err = clGetEventProfilingInfo( executeEvent, CL_PROFILING_COMMAND_START, sizeof( cl_ulong ), &writeStart, NULL );
if( err != CL_SUCCESS ){
print_error( err, "clGetEventProfilingInfo failed" );
clReleaseKernel( kernel[0] );
clReleaseProgram( program[0] );
clReleaseMemObject( memobjs[1] );
clReleaseMemObject( memobjs[0] );
return -1;
}
err = clGetEventProfilingInfo( executeEvent, CL_PROFILING_COMMAND_END, sizeof( cl_ulong ), &writeEnd, NULL );
if( err != CL_SUCCESS ){
print_error( err, "clGetEventProfilingInfo failed" );
clReleaseKernel( kernel[0] );
clReleaseProgram( program[0] );
clReleaseMemObject( memobjs[1] );
clReleaseMemObject( memobjs[0] );
return -1;
}
log_info( "Profiling info:\n" );
log_info( "Time from queue to start of clEnqueueNDRangeKernel: %f seconds\n", (double)(writeStart - queueStart) / 1000000000000.f );
log_info( "Time from start of clEnqueueNDRangeKernel to end: %f seconds\n", (double)(writeEnd - writeStart) / 1000000000000.f );
}
// read output image
err = clEnqueueReadBuffer(queue, memobjs[1], CL_TRUE, 0, w*h*d*nChannels*4, outptr, 0, NULL, NULL);
if( err != CL_SUCCESS ){
print_error( err, "clReadImage failed\n" );
clReleaseKernel( kernel[0] );
clReleaseProgram( program[0] );
clReleaseMemObject( memobjs[1] );
clReleaseMemObject( memobjs[0] );
return -1;
}
// release kernel, program, and memory objects
clReleaseKernel( kernel[0] );
clReleaseProgram( program[0] );
clReleaseMemObject( memobjs[1] );
clReleaseMemObject( memobjs[0] );
return err;
} // end run_kernel()
// The main point of this test is to exercise code that causes a multipass cld launch for a single
// kernel exec at the cl level. This is done on the gpu for 3d launches, and it's also done
// to handle gdims that excede the maximums allowed by the hardware. In this case we
// use 3d to exercise the multipass events. In the future 3d may not be multpass, in which
// case we will need to ensure that we use gdims large enough to force multipass.
int execute_multipass( cl_device_id device, cl_context context, cl_command_queue queue, int num_elements )
{
cl_uchar *inptr;
cl_uchar *outptr;
int w = 256, h = 128, d = 32;
int nChannels = 4;
int nElements = w * h * d * nChannels;
int err = 0;
MTdata mtData;
PASSIVE_REQUIRE_IMAGE_SUPPORT( device )
mtData = init_genrand( gRandomSeed );
inptr = createImage( nElements, mtData );
free_mtdata( mtData); mtData = NULL;
if( ! inptr ){
log_error( " unable to allocate %d bytes of memory for image\n", nElements );
return -1;
}
outptr = (cl_uchar *)malloc( nElements * sizeof( cl_uchar ) );
if( ! outptr ){
log_error( " unable to allocate %d bytes of memory for output image #1\n", nElements );
free( (void *)inptr );
return -1;
}
err = run_kernel( device, context, queue, w, h, d, nChannels, inptr, outptr );
if( ! err ){
// verify that the images are the same
err = verifyImages( outptr, inptr, (cl_uchar)0x1, w, h, d, nChannels );
if( err )
log_error( " images do not match\n" );
}
// clean up
free( (void *)outptr );
free( (void *)inptr );
return err;
} // end execute()