1 //
2 // Copyright (c) 2017 The Khronos Group Inc.
3 //
4 // Licensed under the Apache License, Version 2.0 (the "License");
5 // you may not use this file except in compliance with the License.
6 // You may obtain a copy of the License at
7 //
8 // http://www.apache.org/licenses/LICENSE-2.0
9 //
10 // Unless required by applicable law or agreed to in writing, software
11 // distributed under the License is distributed on an "AS IS" BASIS,
12 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 // See the License for the specific language governing permissions and
14 // limitations under the License.
15 //
16 #include "harness/compat.h"
17
18 #include <stdio.h>
19 #include <string.h>
20 #include <time.h>
21 #include <sys/types.h>
22 #include <sys/stat.h>
23
24 #include "procs.h"
25 #include "harness/testHarness.h"
26 #include "harness/errorHelpers.h"
27
28 static const char *read3d_kernel_code =
29 "\n"
30 "__kernel void read3d(read_only image3d_t srcimg, __global unsigned char *dst, sampler_t sampler)\n"
31 "{\n"
32 " int tid_x = get_global_id(0);\n"
33 " int tid_y = get_global_id(1);\n"
34 " int tid_z = get_global_id(2);\n"
35 " int indx = (tid_z * get_image_height(srcimg) + tid_y) * get_image_width(srcimg) + tid_x;\n"
36 " float4 color;\n"
37 "\n"
38 " color = read_imagef(srcimg, sampler, (int4)(tid_x, tid_y, tid_z, 0));\n"
39 " indx *= 4;\n"
40 " dst[indx+0] = (unsigned char)(color.x * 255.0f);\n"
41 " dst[indx+1] = (unsigned char)(color.y * 255.0f);\n"
42 " dst[indx+2] = (unsigned char)(color.z * 255.0f);\n"
43 " dst[indx+3] = (unsigned char)(color.w * 255.0f);\n"
44 "\n"
45 "}\n";
46
47
createImage(int elements,MTdata d)48 static cl_uchar *createImage( int elements, MTdata d )
49 {
50 int i;
51 cl_uchar *ptr = (cl_uchar *)malloc( elements * sizeof( cl_uchar ) );
52 if( ! ptr )
53 return NULL;
54
55 for( i = 0; i < elements; i++ ){
56 ptr[i] = (cl_uchar)genrand_int32(d);
57 }
58
59 return ptr;
60
61 } // end createImage()
62
63
verifyImages(cl_uchar * ptr0,cl_uchar * ptr1,cl_uchar tolerance,int xsize,int ysize,int zsize,int nChannels)64 static int verifyImages( cl_uchar *ptr0, cl_uchar *ptr1, cl_uchar tolerance, int xsize, int ysize, int zsize, int nChannels )
65 {
66 int x, y, z, c;
67 cl_uchar *p0 = ptr0;
68 cl_uchar *p1 = ptr1;
69
70 for( z = 0; z < zsize; z++ ){
71 for( y = 0; y < ysize; y++ ){
72 for( x = 0; x < xsize; x++ ){
73 for( c = 0; c < nChannels; c++ ){
74 if( (cl_uchar)abs( (int)( *p0++ - *p1++ ) ) > tolerance ){
75 log_error( " images differ at x,y,z = %d,%d,%d channel = %d, %d to %d\n",
76 x, y, z, c, (int)p0[-1], (int)p1[-1] );
77 return -1;
78 }
79 }
80 }
81 }
82 }
83
84 return 0;
85
86 } // end verifyImages()
87
88
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)89 static int run_kernel( cl_device_id device, cl_context context, cl_command_queue queue,
90 int w, int h, int d, int nChannels, cl_uchar *inptr, cl_uchar *outptr )
91 {
92 cl_program program[1];
93 cl_kernel kernel[1];
94 cl_mem memobjs[2];
95 cl_image_format image_format_desc = { CL_RGBA, CL_UNORM_INT8 };
96 cl_event executeEvent = NULL;
97 cl_ulong queueStart, submitStart, writeStart, writeEnd;
98 size_t threads[3];
99 size_t localThreads[3];
100 int err = 0;
101
102 // set thread dimensions
103 threads[0] = w;
104 threads[1] = h;
105 threads[2] = d;
106
107 err = clGetDeviceInfo( device, CL_DEVICE_MAX_WORK_ITEM_SIZES, sizeof( cl_uint ), (size_t*)localThreads, NULL );
108 if (err)
109 {
110 localThreads[0] = 256; localThreads[1] = 1; localThreads[2] = 1;
111 err = 0;
112 }
113 if( localThreads[0] > threads[0] )
114 localThreads[0] = threads[0];
115 if( localThreads[1] > threads[1] )
116 localThreads[1] = threads[1];
117
118 cl_sampler sampler = clCreateSampler( context, CL_FALSE, CL_ADDRESS_CLAMP_TO_EDGE, CL_FILTER_NEAREST, &err );
119 if( err ){
120 log_error( " clCreateSampler failed.\n" );
121 return -1;
122 }
123
124 // allocate the input and output image memory objects
125 memobjs[0] = create_image_3d( context, (cl_mem_flags)(CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR), &image_format_desc, w, h, d, 0, 0, inptr, &err );
126 if( memobjs[0] == (cl_mem)0 ){
127 log_error( " unable to create 2D image using create_image_2d\n" );
128 return -1;
129 }
130
131 // allocate an array memory object to load the filter weights
132 memobjs[1] = clCreateBuffer( context, (cl_mem_flags)( CL_MEM_READ_WRITE ), sizeof( cl_float ) * w*h*d*nChannels, NULL, &err );
133 if( memobjs[1] == (cl_mem)0 ){
134 log_error( " unable to create array using clCreateBuffer\n" );
135 clReleaseMemObject( memobjs[0] );
136 return -1;
137 }
138
139 // create the compute program
140 err = create_single_kernel_helper( context, &program[0], &kernel[0], 1, &read3d_kernel_code, "read3d" );
141 if( err ){
142 clReleaseMemObject( memobjs[1] );
143 clReleaseMemObject( memobjs[0] );
144 return -1;
145 }
146
147
148 // create kernel args object and set arg values.
149 // set the args values
150 err |= clSetKernelArg( kernel[0], 0, sizeof( cl_mem ), (void *)&memobjs[0] );
151 err |= clSetKernelArg( kernel[0], 1, sizeof( cl_mem ), (void *)&memobjs[1] );
152 err |= clSetKernelArg(kernel[0], 2, sizeof sampler, &sampler);
153
154 if( err != CL_SUCCESS ){
155 print_error( err, "clSetKernelArg failed\n" );
156 clReleaseKernel( kernel[0] );
157 clReleaseProgram( program[0] );
158 clReleaseMemObject( memobjs[1] );
159 clReleaseMemObject( memobjs[0] );
160 return -1;
161 }
162
163 err = clEnqueueNDRangeKernel( queue, kernel[0], 3, NULL, threads, localThreads, 0, NULL, &executeEvent );
164
165 if( err != CL_SUCCESS ){
166 print_error( err, "clEnqueueNDRangeKernel failed\n" );
167 clReleaseKernel( kernel[0] );
168 clReleaseProgram( program[0] );
169 clReleaseMemObject( memobjs[1] );
170 clReleaseMemObject( memobjs[0] );
171 return -1;
172 }
173
174 if (executeEvent) {
175
176 // This synchronization point is needed in order to assume the data is valid.
177 // Getting profiling information is not a synchronization point.
178 err = clWaitForEvents( 1, &executeEvent );
179 if( err != CL_SUCCESS )
180 {
181 print_error( err, "clWaitForEvents failed\n" );
182 clReleaseKernel( kernel[0] );
183 clReleaseProgram( program[0] );
184 clReleaseMemObject( memobjs[1] );
185 clReleaseMemObject( memobjs[0] );
186 return -1;
187 }
188
189 // test profiling
190 while( ( err = clGetEventProfilingInfo( executeEvent, CL_PROFILING_COMMAND_QUEUED, sizeof( cl_ulong ), &queueStart, NULL ) ) == CL_PROFILING_INFO_NOT_AVAILABLE );
191 if( err != CL_SUCCESS ){
192 print_error( err, "clGetEventProfilingInfo failed" );
193 clReleaseKernel( kernel[0] );
194 clReleaseProgram( program[0] );
195 clReleaseMemObject( memobjs[1] );
196 clReleaseMemObject( memobjs[0] );
197 return -1;
198 }
199
200 while( ( err = clGetEventProfilingInfo( executeEvent, CL_PROFILING_COMMAND_SUBMIT, sizeof( cl_ulong ), &submitStart, NULL ) ) == CL_PROFILING_INFO_NOT_AVAILABLE );
201 if( err != CL_SUCCESS ){
202 print_error( err, "clGetEventProfilingInfo failed" );
203 clReleaseKernel( kernel[0] );
204 clReleaseProgram( program[0] );
205 clReleaseMemObject( memobjs[1] );
206 clReleaseMemObject( memobjs[0] );
207 return -1;
208 }
209
210 err = clGetEventProfilingInfo( executeEvent, CL_PROFILING_COMMAND_START, sizeof( cl_ulong ), &writeStart, NULL );
211 if( err != CL_SUCCESS ){
212 print_error( err, "clGetEventProfilingInfo failed" );
213 clReleaseKernel( kernel[0] );
214 clReleaseProgram( program[0] );
215 clReleaseMemObject( memobjs[1] );
216 clReleaseMemObject( memobjs[0] );
217 return -1;
218 }
219
220 err = clGetEventProfilingInfo( executeEvent, CL_PROFILING_COMMAND_END, sizeof( cl_ulong ), &writeEnd, NULL );
221 if( err != CL_SUCCESS ){
222 print_error( err, "clGetEventProfilingInfo failed" );
223 clReleaseKernel( kernel[0] );
224 clReleaseProgram( program[0] );
225 clReleaseMemObject( memobjs[1] );
226 clReleaseMemObject( memobjs[0] );
227 return -1;
228 }
229
230 log_info( "Profiling info:\n" );
231 log_info( "Time from queue to start of clEnqueueNDRangeKernel: %f seconds\n", (double)(writeStart - queueStart) / 1000000000000.f );
232 log_info( "Time from start of clEnqueueNDRangeKernel to end: %f seconds\n", (double)(writeEnd - writeStart) / 1000000000000.f );
233 }
234
235 // read output image
236 err = clEnqueueReadBuffer(queue, memobjs[1], CL_TRUE, 0, w*h*d*nChannels*4, outptr, 0, NULL, NULL);
237 if( err != CL_SUCCESS ){
238 print_error( err, "clReadImage failed\n" );
239 clReleaseKernel( kernel[0] );
240 clReleaseProgram( program[0] );
241 clReleaseMemObject( memobjs[1] );
242 clReleaseMemObject( memobjs[0] );
243 return -1;
244 }
245
246 // release kernel, program, and memory objects
247 clReleaseKernel( kernel[0] );
248 clReleaseProgram( program[0] );
249 clReleaseMemObject( memobjs[1] );
250 clReleaseMemObject( memobjs[0] );
251
252 return err;
253
254 } // end run_kernel()
255
256
257 // The main point of this test is to exercise code that causes a multipass cld launch for a single
258 // kernel exec at the cl level. This is done on the gpu for 3d launches, and it's also done
259 // to handle gdims that excede the maximums allowed by the hardware. In this case we
260 // use 3d to exercise the multipass events. In the future 3d may not be multpass, in which
261 // case we will need to ensure that we use gdims large enough to force multipass.
262
execute_multipass(cl_device_id device,cl_context context,cl_command_queue queue,int num_elements)263 int execute_multipass( cl_device_id device, cl_context context, cl_command_queue queue, int num_elements )
264 {
265 cl_uchar *inptr;
266 cl_uchar *outptr;
267 int w = 256, h = 128, d = 32;
268 int nChannels = 4;
269 int nElements = w * h * d * nChannels;
270 int err = 0;
271 MTdata mtData;
272
273 PASSIVE_REQUIRE_IMAGE_SUPPORT( device )
274
275 mtData = init_genrand( gRandomSeed );
276 inptr = createImage( nElements, mtData );
277 free_mtdata( mtData); mtData = NULL;
278 if( ! inptr ){
279 log_error( " unable to allocate %d bytes of memory for image\n", nElements );
280 return -1;
281 }
282
283 outptr = (cl_uchar *)malloc( nElements * sizeof( cl_uchar ) );
284 if( ! outptr ){
285 log_error( " unable to allocate %d bytes of memory for output image #1\n", nElements );
286 free( (void *)inptr );
287 return -1;
288 }
289
290
291 err = run_kernel( device, context, queue, w, h, d, nChannels, inptr, outptr );
292
293 if( ! err ){
294 // verify that the images are the same
295 err = verifyImages( outptr, inptr, (cl_uchar)0x1, w, h, d, nChannels );
296 if( err )
297 log_error( " images do not match\n" );
298 }
299
300 // clean up
301 free( (void *)outptr );
302 free( (void *)inptr );
303
304 return err;
305
306 } // end execute()
307
308
309
310