// // 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 #include #include #include #include #include "procs.h" int hi_offset( int index, int vectorSize) { return index + vectorSize / 2; } int lo_offset( int index, int vectorSize) { return index; } int even_offset( int index, int vectorSize ) { return index * 2; } int odd_offset( int index, int vectorSize ) { return index * 2 + 1; } typedef int (*OffsetFunc)( int index, int vectorSize ); static const OffsetFunc offsetFuncs[4] = { hi_offset, lo_offset, even_offset, odd_offset }; typedef int (*verifyFunc)( const void *, const void *, const void *, int n, const char *sizeName ); static const char *operatorToUse_names[] = { "hi", "lo", "even", "odd" }; static const char *test_str_names[] = { "char", "uchar", "short", "ushort", "int", "uint", "long", "ulong", "float", "double" }; static const unsigned int vector_sizes[] = { 1, 2, 3, 4, 8, 16}; static const unsigned int vector_aligns[] = { 1, 2, 4, 4, 8, 16}; static const unsigned int out_vector_idx[] = { 0, 0, 1, 1, 3, 4}; // if input is size vector_sizes[i], output is size // vector_sizes[out_vector_idx[i]] // input type name is strcat(gentype, vector_size_names[i]); // and output type name is // strcat(gentype, vector_size_names[out_vector_idx[i]]); static const int size_to_idx[] = {-1,0,1,2,3,-1,-1,-1,4, -1,-1,-1,-1,-1,-1,-1,5}; static const char *vector_size_names[] = { "", "2", "3", "4", "8", "16"}; static const size_t kSizes[] = { 1, 1, 2, 2, 4, 4, 8, 8, 4, 8 }; static int CheckResults( void *in, void *out, size_t elementCount, int type, int vectorSize, int operatorToUse ); int test_hiloeo(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems) { cl_int *input_ptr, *output_ptr, *p; int err; cl_uint i; int hasDouble = is_extension_available( device, "cl_khr_fp64" ); cl_uint vectorSize, operatorToUse; cl_uint type; MTdata d; int expressionMode; int numExpressionModes = 2; size_t length = sizeof(cl_int) * 4 * n_elems; input_ptr = (cl_int*)malloc(length); output_ptr = (cl_int*)malloc(length); p = input_ptr; d = init_genrand( gRandomSeed ); for (i=0; i<4 * (cl_uint) n_elems; i++) p[i] = genrand_int32(d); free_mtdata(d); d = NULL; for( type = 0; type < sizeof( test_str_names ) / sizeof( test_str_names[0] ); type++ ) { // Note: restrict the element count here so we don't end up overrunning the output buffer if we're compensating for 32-bit writes size_t elementCount = length / kSizes[type]; cl_mem streams[2]; // skip double if unavailable if( !hasDouble && ( 0 == strcmp( test_str_names[type], "double" ))) continue; if( !gHasLong && (( 0 == strcmp( test_str_names[type], "long" )) || ( 0 == strcmp( test_str_names[type], "ulong" )))) continue; log_info( "%s", test_str_names[type] ); fflush( stdout ); // Set up data streams for the type streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE, length, NULL, NULL); if (!streams[0]) { log_error("clCreateBuffer failed\n"); return -1; } streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE, length, NULL, NULL); if (!streams[1]) { log_error("clCreateBuffer failed\n"); return -1; } err = clEnqueueWriteBuffer(queue, streams[0], CL_TRUE, 0, length, input_ptr, 0, NULL, NULL); if (err != CL_SUCCESS) { log_error("clEnqueueWriteBuffer failed\n"); return -1; } for( operatorToUse = 0; operatorToUse < sizeof( operatorToUse_names ) / sizeof( operatorToUse_names[0] ); operatorToUse++ ) { log_info( " %s", operatorToUse_names[ operatorToUse ] ); fflush( stdout ); for( vectorSize = 1; vectorSize < sizeof( vector_size_names ) / sizeof( vector_size_names[0] ); vectorSize++ ) { for(expressionMode = 0; expressionMode < numExpressionModes; ++expressionMode) { cl_program program = NULL; cl_kernel kernel = NULL; cl_uint outVectorSize = out_vector_idx[vectorSize]; char expression[1024]; const char *source[] = { "", // optional pragma string "__kernel void test_", operatorToUse_names[ operatorToUse ], "_", test_str_names[type], vector_size_names[vectorSize], "(__global ", test_str_names[type], vector_size_names[vectorSize], " *srcA, __global ", test_str_names[type], vector_size_names[outVectorSize], " *dst)\n" "{\n" " int tid = get_global_id(0);\n" "\n" " ", test_str_names[type], vector_size_names[out_vector_idx[vectorSize]], " tmp = ", expression, ".", operatorToUse_names[ operatorToUse ], ";\n" " dst[tid] = tmp;\n" "}\n" }; if(expressionMode == 0) { sprintf(expression, "srcA[tid]"); } else if(expressionMode == 1) { switch(vector_sizes[vectorSize]) { case 16: sprintf(expression, "((%s16)(srcA[tid].s0, srcA[tid].s1, srcA[tid].s2, srcA[tid].s3, srcA[tid].s4, srcA[tid].s5, srcA[tid].s6, srcA[tid].s7, srcA[tid].s8, srcA[tid].s9, srcA[tid].sA, srcA[tid].sB, srcA[tid].sC, srcA[tid].sD, srcA[tid].sE, srcA[tid].sf))", test_str_names[type] ); break; case 8: sprintf(expression, "((%s8)(srcA[tid].s0, srcA[tid].s1, srcA[tid].s2, srcA[tid].s3, srcA[tid].s4, srcA[tid].s5, srcA[tid].s6, srcA[tid].s7))", test_str_names[type] ); break; case 4: sprintf(expression, "((%s4)(srcA[tid].s0, srcA[tid].s1, srcA[tid].s2, srcA[tid].s3))", test_str_names[type] ); break; case 3: sprintf(expression, "((%s3)(srcA[tid].s0, srcA[tid].s1, srcA[tid].s2))", test_str_names[type] ); break; case 2: sprintf(expression, "((%s2)(srcA[tid].s0, srcA[tid].s1))", test_str_names[type] ); break; default : sprintf(expression, "srcA[tid]"); log_info("Default\n"); } } else { sprintf(expression, "srcA[tid]"); } if (0 == strcmp( test_str_names[type], "double" )) source[0] = "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"; char kernelName[128]; snprintf( kernelName, sizeof( kernelName ), "test_%s_%s%s", operatorToUse_names[ operatorToUse ], test_str_names[type], vector_size_names[vectorSize] ); err = create_single_kernel_helper(context, &program, &kernel, sizeof( source ) / sizeof( source[0] ), source, kernelName ); if (err) return -1; err = clSetKernelArg(kernel, 0, sizeof streams[0], &streams[0]); err |= clSetKernelArg(kernel, 1, sizeof streams[1], &streams[1]); if (err != CL_SUCCESS) { log_error("clSetKernelArgs failed\n"); return -1; } //Wipe the output buffer clean uint32_t pattern = 0xdeadbeef; memset_pattern4( output_ptr, &pattern, length ); err = clEnqueueWriteBuffer(queue, streams[1], CL_TRUE, 0, length, output_ptr, 0, NULL, NULL); if (err != CL_SUCCESS) { log_error("clEnqueueWriteBuffer failed\n"); return -1; } size_t size = elementCount / (vector_aligns[vectorSize]); err = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &size, NULL, 0, NULL, NULL); if (err != CL_SUCCESS) { log_error("clEnqueueNDRangeKernel failed\n"); return -1; } err = clEnqueueReadBuffer(queue, streams[1], CL_TRUE, 0, length, output_ptr, 0, NULL, NULL); if (err != CL_SUCCESS) { log_error("clEnqueueReadBuffer failed\n"); return -1; } char *inP = (char *)input_ptr; char *outP = (char *)output_ptr; outP += kSizes[type] * ( ( vector_sizes[outVectorSize] ) - ( vector_sizes[ out_vector_idx[vectorSize] ] ) ); // was outP += kSizes[type] * ( ( 1 << outVectorSize ) - ( 1 << ( vectorSize - 1 ) ) ); for( size_t e = 0; e < size; e++ ) { if( CheckResults( inP, outP, 1, type, vectorSize, operatorToUse ) ) { log_info("e is %d\n", (int)e); fflush(stdout); // break; return -1; } inP += kSizes[type] * ( vector_aligns[vectorSize] ); outP += kSizes[type] * ( vector_aligns[outVectorSize] ); } clReleaseKernel( kernel ); clReleaseProgram( program ); log_info( "." ); fflush( stdout ); } } } clReleaseMemObject( streams[0] ); clReleaseMemObject( streams[1] ); log_info( "done\n" ); } log_info("HiLoEO test passed\n"); free(input_ptr); free(output_ptr); return err; } static int CheckResults( void *in, void *out, size_t elementCount, int type, int vectorSize, int operatorToUse ) { cl_ulong array[8]; void *p = array; size_t halfVectorSize = vector_sizes[out_vector_idx[vectorSize]]; size_t cmpVectorSize = vector_sizes[out_vector_idx[vectorSize]]; // was 1 << (vectorSize-1); OffsetFunc f = offsetFuncs[ operatorToUse ]; size_t elementSize = kSizes[type]; if(vector_size_names[vectorSize][0] == '3') { if(operatorToUse_names[operatorToUse][0] == 'h' || operatorToUse_names[operatorToUse][0] == 'o') // hi or odd { cmpVectorSize = 1; // special case for vec3 ignored values } } switch( elementSize ) { case 1: { char *i = (char*)in; char *o = (char*)out; size_t j; cl_uint k; OffsetFunc f = offsetFuncs[ operatorToUse ]; for( k = 0; k < elementCount; k++ ) { char *o2 = (char*)p; for( j = 0; j < halfVectorSize; j++ ) o2[j] = i[ f((int)j, (int)halfVectorSize*2) ]; if( memcmp( o, o2, elementSize * cmpVectorSize ) ) { log_info( "\n%d) Failure for %s%s.%s { %d", k, test_str_names[type], vector_size_names[ vectorSize ], operatorToUse_names[ operatorToUse ], i[0] ); for( j = 1; j < halfVectorSize * 2; j++ ) log_info( ", %d", i[j] ); log_info( " } --> { %d", o[0] ); for( j = 1; j < halfVectorSize; j++ ) log_info( ", %d", o[j] ); log_info( " }\n" ); return -1; } i += 2 * halfVectorSize; o += halfVectorSize; } } break; case 2: { short *i = (short*)in; short *o = (short*)out; size_t j; cl_uint k; for( k = 0; k < elementCount; k++ ) { short *o2 = (short*)p; for( j = 0; j < halfVectorSize; j++ ) o2[j] = i[ f((int)j, (int)halfVectorSize*2) ]; if( memcmp( o, o2, elementSize * cmpVectorSize ) ) { log_info( "\n%d) Failure for %s%s.%s { %d", k, test_str_names[type], vector_size_names[ vectorSize ], operatorToUse_names[ operatorToUse ], i[0] ); for( j = 1; j < halfVectorSize * 2; j++ ) log_info( ", %d", i[j] ); log_info( " } --> { %d", o[0] ); for( j = 1; j < halfVectorSize; j++ ) log_info( ", %d", o[j] ); log_info( " }\n" ); return -1; } i += 2 * halfVectorSize; o += halfVectorSize; } } break; case 4: { int *i = (int*)in; int *o = (int*)out; size_t j; cl_uint k; for( k = 0; k < elementCount; k++ ) { int *o2 = (int *)p; for( j = 0; j < halfVectorSize; j++ ) o2[j] = i[ f((int)j, (int)halfVectorSize*2) ]; for( j = 0; j < cmpVectorSize; j++ ) { /* Allow float nans to be binary different */ if( memcmp( &o[j], &o2[j], elementSize ) && !((strcmp(test_str_names[type], "float") == 0) && isnan(((float *)o)[j]) && isnan(((float *)o2)[j]))) { log_info( "\n%d) Failure for %s%s.%s { 0x%8.8x", k, test_str_names[type], vector_size_names[ vectorSize ], operatorToUse_names[ operatorToUse ], i[0] ); for( j = 1; j < halfVectorSize * 2; j++ ) log_info( ", 0x%8.8x", i[j] ); log_info( " } --> { 0x%8.8x", o[0] ); for( j = 1; j < halfVectorSize; j++ ) log_info( ", 0x%8.8x", o[j] ); log_info( " }\n" ); return -1; } } i += 2 * halfVectorSize; o += halfVectorSize; } } break; case 8: { cl_ulong *i = (cl_ulong*)in; cl_ulong *o = (cl_ulong*)out; size_t j; cl_uint k; for( k = 0; k < elementCount; k++ ) { cl_ulong *o2 = (cl_ulong*)p; for( j = 0; j < halfVectorSize; j++ ) o2[j] = i[ f((int)j, (int)halfVectorSize*2) ]; if( memcmp( o, o2, elementSize * cmpVectorSize ) ) { log_info( "\n%d) Failure for %s%s.%s { 0x%16.16llx", k, test_str_names[type], vector_size_names[ vectorSize ], operatorToUse_names[ operatorToUse ], i[0] ); for( j = 1; j < halfVectorSize * 2; j++ ) log_info( ", 0x%16.16llx", i[j] ); log_info( " } --> { 0x%16.16llx", o[0] ); for( j = 1; j < halfVectorSize; j++ ) log_info( ", 0x%16.16llx", o[j] ); log_info( " }\n" ); return -1; } i += 2 * halfVectorSize; o += halfVectorSize; } } break; default: log_info( "Internal error. Unknown data type\n" ); return -2; } return 0; }