// // 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" #include "harness/typeWrappers.h" #include "harness/imageHelpers.h" #include "harness/conversions.h" static const char *param_kernel[] = { "__kernel void test_fn(read_only image2d_t srcimg, sampler_t sampler, __global float4 *results )\n" "{\n" " int tid_x = get_global_id(0);\n" " int tid_y = get_global_id(1);\n" " results[ tid_y * get_image_width( srcimg ) + tid_x ] = read_imagef(srcimg, sampler, (int2)(tid_x, tid_y));\n" "\n" "}\n" }; int validate_results( size_t width, size_t height, cl_image_format &format, char *inputData, cl_float *actualResults ) { for( size_t i = 0; i < width * height; i++ ) { cl_float expected[ 4 ], tolerance; switch( format.image_channel_data_type ) { case CL_UNORM_INT8: { cl_uchar *p = (cl_uchar *)inputData; expected[ 0 ] = p[ 0 ] / 255.f; expected[ 1 ] = p[ 1 ] / 255.f; expected[ 2 ] = p[ 2 ] / 255.f; expected[ 3 ] = p[ 3 ] / 255.f; tolerance = 1.f / 255.f; break; } case CL_SNORM_INT8: { cl_char *p = (cl_char *)inputData; expected[ 0 ] = fmaxf( p[ 0 ] / 127.f, -1.f ); expected[ 1 ] = fmaxf( p[ 1 ] / 127.f, -1.f ); expected[ 2 ] = fmaxf( p[ 2 ] / 127.f, -1.f ); expected[ 3 ] = fmaxf( p[ 3 ] / 127.f, -1.f ); tolerance = 1.f / 127.f; break; } case CL_UNSIGNED_INT8: { cl_uchar *p = (cl_uchar *)inputData; expected[ 0 ] = p[ 0 ]; expected[ 1 ] = p[ 1 ]; expected[ 2 ] = p[ 2 ]; expected[ 3 ] = p[ 3 ]; tolerance = 1.f / 127.f; break; } case CL_SIGNED_INT8: { cl_short *p = (cl_short *)inputData; expected[ 0 ] = p[ 0 ]; expected[ 1 ] = p[ 1 ]; expected[ 2 ] = p[ 2 ]; expected[ 3 ] = p[ 3 ]; tolerance = 1.f / 127.f; break; } case CL_UNORM_INT16: { cl_ushort *p = (cl_ushort *)inputData; expected[ 0 ] = p[ 0 ] / 65535.f; expected[ 1 ] = p[ 1 ] / 65535.f; expected[ 2 ] = p[ 2 ] / 65535.f; expected[ 3 ] = p[ 3 ] / 65535.f; tolerance = 1.f / 65535.f; break; } case CL_UNSIGNED_INT32: { cl_uint *p = (cl_uint *)inputData; expected[ 0 ] = p[ 0 ]; expected[ 1 ] = p[ 1 ]; expected[ 2 ] = p[ 2 ]; expected[ 3 ] = p[ 3 ]; tolerance = 0.0001f; break; } case CL_FLOAT: { cl_float *p = (cl_float *)inputData; expected[ 0 ] = p[ 0 ]; expected[ 1 ] = p[ 1 ]; expected[ 2 ] = p[ 2 ]; expected[ 3 ] = p[ 3 ]; tolerance = 0.0001f; break; } default: // Should never get here break; } if( format.image_channel_order == CL_BGRA ) { cl_float tmp = expected[ 0 ]; expected[ 0 ] = expected[ 2 ]; expected[ 2 ] = tmp; } // Within an error tolerance, make sure the results match cl_float error1 = fabsf( expected[ 0 ] - actualResults[ 0 ] ); cl_float error2 = fabsf( expected[ 1 ] - actualResults[ 1 ] ); cl_float error3 = fabsf( expected[ 2 ] - actualResults[ 2 ] ); cl_float error4 = fabsf( expected[ 3 ] - actualResults[ 3 ] ); if( error1 > tolerance || error2 > tolerance || error3 > tolerance || error4 > tolerance ) { log_error( "ERROR: Sample %d did not validate against expected results for %d x %d %s:%s image\n", (int)i, (int)width, (int)height, GetChannelOrderName( format.image_channel_order ), GetChannelTypeName( format.image_channel_data_type ) ); log_error( " Expected: %f %f %f %f\n", (float)expected[ 0 ], (float)expected[ 1 ], (float)expected[ 2 ], (float)expected[ 3 ] ); log_error( " Actual: %f %f %f %f\n", (float)actualResults[ 0 ], (float)actualResults[ 1 ], (float)actualResults[ 2 ], (float)actualResults[ 3 ] ); // Check real quick a special case error here cl_float error1 = fabsf( expected[ 3 ] - actualResults[ 0 ] ); cl_float error2 = fabsf( expected[ 2 ] - actualResults[ 1 ] ); cl_float error3 = fabsf( expected[ 1 ] - actualResults[ 2 ] ); cl_float error4 = fabsf( expected[ 0 ] - actualResults[ 3 ] ); if( error1 <= tolerance && error2 <= tolerance && error3 <= tolerance && error4 <= tolerance ) { log_error( "\t(Kernel did not respect change in channel order)\n" ); } return -1; } // Increment and go actualResults += 4; inputData += get_format_type_size( &format ) * 4; } return 0; } int test_image_param(cl_device_id device, cl_context context, cl_command_queue queue, int num_elements) { size_t sizes[] = { 64, 100, 128, 250, 512 }; cl_image_format formats[] = { { CL_RGBA, CL_UNORM_INT8 }, { CL_RGBA, CL_UNORM_INT16 }, { CL_RGBA, CL_FLOAT }, { CL_BGRA, CL_UNORM_INT8 } }; cl_image_format *supported_formats; ExplicitType types[] = { kUChar, kUShort, kFloat, kUChar }; int error; size_t i, j, idx; size_t threads[ 2 ]; MTdata d; int supportsBGRA = 0; cl_uint numSupportedFormats = 0; const size_t numSizes = sizeof( sizes ) / sizeof( sizes[ 0 ] ); const size_t numFormats = sizeof( formats ) / sizeof( formats[ 0 ] ); const size_t numAttempts = numSizes * numFormats; clProgramWrapper program; clKernelWrapper kernel; clMemWrapper streams[ numAttempts ][ 2 ]; BufferOwningPtr inputs[ numAttempts ]; PASSIVE_REQUIRE_IMAGE_SUPPORT( device ) if(gIsEmbedded) { /* Get the supported image formats to see if BGRA is supported */ clGetSupportedImageFormats (context, CL_MEM_READ_WRITE, CL_MEM_OBJECT_IMAGE2D, 0, NULL, &numSupportedFormats); supported_formats = (cl_image_format *) malloc(sizeof(cl_image_format) * numSupportedFormats); clGetSupportedImageFormats (context, CL_MEM_READ_WRITE, CL_MEM_OBJECT_IMAGE2D, numFormats, supported_formats, NULL); for(i = 0; i < numSupportedFormats; i++) { if(supported_formats[i].image_channel_order == CL_BGRA) { supportsBGRA = 1; break; } } } else { supportsBGRA = 1; } d = init_genrand( gRandomSeed ); for( i = 0, idx = 0; i < numSizes; i++ ) { for( j = 0; j < numFormats; j++, idx++ ) { if(formats[j].image_channel_order == CL_BGRA && !supportsBGRA) continue; // For each attempt, we create a pair: an input image, whose parameters keep changing, and an output buffer // that we can read values from. The output buffer will remain consistent to ensure that any changes we // witness are due to the image changes inputs[ idx ].reset(create_random_data( types[ j ], d, sizes[ i ] * sizes[ i ] * 4 )); streams[ idx ][ 0 ] = create_image_2d( context, CL_MEM_COPY_HOST_PTR, &formats[ j ], sizes[ i ], sizes[ i ], 0, inputs[ idx ], &error ); { char err_str[256]; sprintf(err_str, "Unable to create input image for format %s order %s" , GetChannelOrderName( formats[j].image_channel_order ), GetChannelTypeName( formats[j].image_channel_data_type )); test_error( error, err_str); } streams[ idx ][ 1 ] = clCreateBuffer( context, CL_MEM_READ_WRITE, sizes[ i ] * sizes[ i ] * 4 * sizeof( cl_float ), NULL, &error ); test_error( error, "Unable to create output buffer" ); } } free_mtdata(d); d = NULL; // Create a single kernel to use for all the tests error = create_single_kernel_helper( context, &program, &kernel, 1, param_kernel, "test_fn" ); test_error( error, "Unable to create testing kernel" ); // Also create a sampler to use for all the runs clSamplerWrapper sampler = clCreateSampler(context, CL_FALSE, CL_ADDRESS_CLAMP_TO_EDGE, CL_FILTER_NEAREST, &error ); test_error( error, "clCreateSampler failed" ); // Set up the arguments for each and queue for( i = 0, idx = 0; i < numSizes; i++ ) { for( j = 0; j < numFormats; j++, idx++ ) { if(formats[j].image_channel_order == CL_BGRA && !supportsBGRA) continue; error = clSetKernelArg( kernel, 0, sizeof( streams[ idx ][ 0 ] ), &streams[ idx ][ 0 ] ); error |= clSetKernelArg( kernel, 1, sizeof( sampler ), &sampler ); error |= clSetKernelArg( kernel, 2, sizeof( streams[ idx ][ 1 ] ), &streams[ idx ][ 1 ]); test_error( error, "Unable to set kernel arguments" ); threads[ 0 ] = threads[ 1 ] = (size_t)sizes[ i ]; error = clEnqueueNDRangeKernel( queue, kernel, 2, NULL, threads, NULL, 0, NULL, NULL ); test_error( error, "clEnqueueNDRangeKernel failed" ); } } // Now go through each combo and validate the results for( i = 0, idx = 0; i < numSizes; i++ ) { for( j = 0; j < numFormats; j++, idx++ ) { if(formats[j].image_channel_order == CL_BGRA && !supportsBGRA) continue; BufferOwningPtr output(malloc(sizeof(cl_float) * sizes[ i ] * sizes[ i ] * 4 )); error = clEnqueueReadBuffer( queue, streams[ idx ][ 1 ], CL_TRUE, 0, sizes[ i ] * sizes[ i ] * 4 * sizeof( cl_float ), output, 0, NULL, NULL ); test_error( error, "Unable to read results" ); error = validate_results( sizes[ i ], sizes[ i ], formats[ j ], inputs[ idx ], output ); if( error ) return -1; } } return 0; }