// // 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 "procs.h" #include const char *work_offset_test[] = { "__kernel void test( __global int * outputID_A, \n" " __global int * outputID_B, __global int * outputID_C )\n" "{\n" " size_t id0 = get_local_id( 0 ) + get_group_id( 0 ) * get_local_size( 0 );\n" " size_t id1 = get_local_id( 1 ) + get_group_id( 1 ) * get_local_size( 1 );\n" " size_t id2 = get_local_id( 2 ) + get_group_id( 2 ) * get_local_size( 2 );\n" " size_t id = ( id2 * get_global_size( 0 ) * get_global_size( 1 ) ) + ( id1 * get_global_size( 0 ) ) + id0;\n" "\n" " outputID_A[ id ] = get_global_id( 0 );\n" " outputID_B[ id ] = get_global_id( 1 );\n" " outputID_C[ id ] = get_global_id( 2 );\n" "}\n" }; #define MAX_TEST_ITEMS 16 * 16 * 16 #define NUM_TESTS 16 #define MAX_OFFSET 256 #define CHECK_RANGE( v, m, c ) \ if( ( v >= (cl_int)m ) || ( v < 0 ) ) \ { \ log_error( "ERROR: ouputID_%c[%lu]: %d is < 0 or >= %lu\n", c, i, v, m ); \ return -1; \ } int check_results( size_t threads[], size_t offsets[], cl_int outputA[], cl_int outputB[], cl_int outputC[] ) { size_t offsettedSizes[ 3 ] = { threads[ 0 ] + offsets[ 0 ], threads[ 1 ] + offsets[ 1 ], threads[ 2 ] + offsets[ 2 ] }; size_t limit = threads[ 0 ] * threads[ 1 ] * threads[ 2 ]; static char counts[ MAX_OFFSET + 32 ][ MAX_OFFSET + 16 ][ MAX_OFFSET + 16 ]; memset( counts, 0, sizeof( counts ) ); for( size_t i = 0; i < limit; i++ ) { // Check ranges first CHECK_RANGE( outputA[ i ], offsettedSizes[ 0 ], 'A' ) CHECK_RANGE( outputB[ i ], offsettedSizes[ 1 ], 'B' ) CHECK_RANGE( outputC[ i ], offsettedSizes[ 2 ], 'C' ) // Now set the value in the map counts[ outputA[ i ] ][ outputB[ i ] ][ outputC[ i ] ]++; } // Now check the map int missed = 0, multiple = 0, errored = 0, corrected = 0; for( size_t x = 0; x < offsettedSizes[ 0 ]; x++ ) { for( size_t y = 0; y < offsettedSizes[ 1 ]; y++ ) { for( size_t z = 0; z < offsettedSizes[ 2 ]; z++ ) { const char * limitMsg = " (further errors of this type suppressed)"; if( ( x >= offsets[ 0 ] ) && ( y >= offsets[ 1 ] ) && ( z >= offsets[ 2 ] ) ) { if( counts[ x ][ y ][ z ] < 1 ) { if( missed < 3 ) log_error( "ERROR: Map value (%ld,%ld,%ld) was missed%s\n", x, y, z, ( missed == 2 ) ? limitMsg : "" ); missed++; } else if( counts[ x ][ y ][ z ] > 1 ) { if( multiple < 3 ) log_error( "ERROR: Map value (%ld,%ld,%ld) was returned multiple times%s\n", x, y, z, ( multiple == 2 ) ? limitMsg : "" ); multiple++; } } else { if( counts[ x ][ y ][ z ] > 0 ) { if( errored < 3 ) log_error( "ERROR: Map value (%ld,%ld,%ld) was erroneously returned%s\n", x, y, z, ( errored == 2 ) ? limitMsg : "" ); errored++; } } } } } if( missed || multiple || errored ) { size_t diffs[3] = { ( offsets[ 0 ] > threads[ 0 ] ? 0 : threads[ 0 ] - offsets[ 0 ] ), ( offsets[ 1 ] > threads[ 1 ] ? 0 : threads[ 1 ] - offsets[ 1 ] ), ( offsets[ 2 ] > threads[ 2 ] ? 0 : threads[ 2 ] - offsets[ 2 ] ) }; int diff = (int)( ( threads[ 0 ] - diffs[ 0 ] ) * ( threads[ 1 ] - diffs[ 1 ] ) * ( threads[ 2 ] - diffs[ 2 ] ) ); if( ( multiple == 0 ) && ( missed == diff ) && ( errored == diff ) ) log_error( "ERROR: Global work offset values are not being respected by get_global_id()\n" ); else log_error( "ERROR: Global work offset values did not function as expected (%d missed, %d reported multiple times, %d erroneously hit)\n", missed, multiple, errored ); } return ( missed | multiple | errored | corrected ); } int test_global_work_offsets(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { clProgramWrapper program; clKernelWrapper kernel; clMemWrapper streams[ 7 ]; int error; size_t threads[] = {1,1,1}, localThreads[] = {1,1,1}, offsets[] = {0,0,0}; cl_int outputA[ MAX_TEST_ITEMS ], outputB[ MAX_TEST_ITEMS ], outputC[ MAX_TEST_ITEMS ]; // Create the kernel if( create_single_kernel_helper( context, &program, &kernel, 1, work_offset_test, "test" ) != 0 ) { return -1; } //// Create some output streams // Use just one output array to init them all (no need to init every single stack storage here) memset( outputA, 0xff, sizeof( outputA ) ); for( int i = 0; i < 3; i++ ) { streams[i] = clCreateBuffer(context, CL_MEM_READ_WRITE | CL_MEM_COPY_HOST_PTR, sizeof(outputA), outputA, &error); test_error( error, "Unable to create output array" ); } // Run a few different times MTdata seed = init_genrand( gRandomSeed ); for( int test = 0; test < NUM_TESTS; test++ ) { // Choose a random combination of thread size, but in total less than MAX_TEST_ITEMS threads[ 0 ] = random_in_range( 1, 32, seed ); threads[ 1 ] = random_in_range( 1, 16, seed ); threads[ 2 ] = random_in_range( 1, MAX_TEST_ITEMS / (int)( threads[ 0 ] * threads[ 1 ] ), seed ); // Make sure we get the local thread count right error = get_max_common_3D_work_group_size( context, kernel, threads, localThreads ); test_error( error, "Unable to determine local work group sizes" ); // Randomize some offsets for( int j = 0; j < 3; j++ ) offsets[ j ] = random_in_range( 0, MAX_OFFSET, seed ); log_info( "\tTesting %ld,%ld,%ld (%ld,%ld,%ld) with offsets (%ld,%ld,%ld)...\n", threads[ 0 ], threads[ 1 ], threads[ 2 ], localThreads[ 0 ], localThreads[ 1 ], localThreads[ 2 ], offsets[ 0 ], offsets[ 1 ], offsets[ 2 ] ); // Now set up and run for( int i = 0; i < 3; i++ ) { error = clSetKernelArg( kernel, i, sizeof( streams[i] ), &streams[i] ); test_error( error, "Unable to set indexed kernel arguments" ); } error = clEnqueueNDRangeKernel( queue, kernel, 3, offsets, threads, localThreads, 0, NULL, NULL ); test_error( error, "Kernel execution failed" ); // Read our results back now cl_int * resultBuffers[] = { outputA, outputB, outputC }; for( int i = 0; i < 3; i++ ) { error = clEnqueueReadBuffer( queue, streams[ i ], CL_TRUE, 0, sizeof( outputA ), resultBuffers[ i ], 0, NULL, NULL ); test_error( error, "Unable to get result data" ); } // Now we need to check the results. The outputs should have one entry for each possible ID, // but they won't be in order, so we need to construct a count map to determine what we got if( check_results( threads, offsets, outputA, outputB, outputC ) ) { log_error( "\t(Test failed for global dim %ld,%ld,%ld, local dim %ld,%ld,%ld, offsets %ld,%ld,%ld)\n", threads[ 0 ], threads[ 1 ], threads[ 2 ], localThreads[ 0 ], localThreads[ 1 ], localThreads[ 2 ], offsets[ 0 ], offsets[ 1 ], offsets[ 2 ] ); return -1; } } free_mtdata(seed); // All done! return 0; } const char *get_offset_test[] = { "__kernel void test( __global int * outOffsets )\n" "{\n" " // We use local ID here so we don't have to worry about offsets\n" " // Also note that these should be the same for ALL threads, so we won't worry about contention\n" " outOffsets[ 0 ] = (int)get_global_offset( 0 );\n" " outOffsets[ 1 ] = (int)get_global_offset( 1 );\n" " outOffsets[ 2 ] = (int)get_global_offset( 2 );\n" "}\n" }; int test_get_global_offset(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { clProgramWrapper program; clKernelWrapper kernel; clMemWrapper streams[ 1 ]; int error; size_t threads[] = {1,1,1}, localThreads[] = {1,1,1}, offsets[] = {0,0,0}; cl_int outOffsets[ 3 ]; // Create the kernel if( create_single_kernel_helper( context, &program, &kernel, 1, get_offset_test, "test" ) != 0 ) { return -1; } // Create some output streams, and storage for a single control ID memset( outOffsets, 0xff, sizeof( outOffsets ) ); streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE | CL_MEM_COPY_HOST_PTR, sizeof(outOffsets), outOffsets, &error); test_error( error, "Unable to create control ID buffer" ); // Run a few different times MTdata seed = init_genrand( gRandomSeed ); for( int test = 0; test < NUM_TESTS; test++ ) { // Choose a random combination of thread size, but in total less than MAX_TEST_ITEMS threads[ 0 ] = random_in_range( 1, 32, seed ); threads[ 1 ] = random_in_range( 1, 16, seed ); threads[ 2 ] = random_in_range( 1, MAX_TEST_ITEMS / (int)( threads[ 0 ] * threads[ 1 ] ), seed ); // Make sure we get the local thread count right error = get_max_common_3D_work_group_size( context, kernel, threads, localThreads ); test_error( error, "Unable to determine local work group sizes" ); // Randomize some offsets for( int j = 0; j < 3; j++ ) offsets[ j ] = random_in_range( 0, MAX_OFFSET, seed ); log_info( "\tTesting %ld,%ld,%ld (%ld,%ld,%ld) with offsets (%ld,%ld,%ld)...\n", threads[ 0 ], threads[ 1 ], threads[ 2 ], localThreads[ 0 ], localThreads[ 1 ], localThreads[ 2 ], offsets[ 0 ], offsets[ 1 ], offsets[ 2 ] ); // Now set up and run error = clSetKernelArg( kernel, 0, sizeof( streams[0] ), &streams[0] ); test_error( error, "Unable to set indexed kernel arguments" ); error = clEnqueueNDRangeKernel( queue, kernel, 3, offsets, threads, localThreads, 0, NULL, NULL ); test_error( error, "Kernel execution failed" ); // Read our results back now error = clEnqueueReadBuffer( queue, streams[ 0 ], CL_TRUE, 0, sizeof( outOffsets ), outOffsets, 0, NULL, NULL ); test_error( error, "Unable to get result data" ); // And check! int errors = 0; for( int j = 0; j < 3; j++ ) { if( outOffsets[ j ] != (cl_int)offsets[ j ] ) { log_error( "ERROR: get_global_offset( %d ) did not return expected value (expected %ld, got %d)\n", j, offsets[ j ], outOffsets[ j ] ); errors++; } } if( errors > 0 ) return errors; } free_mtdata(seed); // All done! return 0; }