// // 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 "testBase.h" #include "harness/typeWrappers.h" #include "harness/testHarness.h" #include #include const char *sample_single_param_kernel[] = { "__kernel void sample_test(__global int *src)\n" "{\n" " size_t tid = get_global_id(0);\n" "\n" "}\n" }; const char *sample_read_image_kernel_pattern[] = { "__kernel void sample_test( __global float *result, ", " )\n" "{\n" " sampler_t sampler = CLK_NORMALIZED_COORDS_FALSE | CLK_ADDRESS_CLAMP | " "CLK_FILTER_NEAREST;\n" " size_t tid = get_global_id(0);\n" " result[0] = 0.0f;\n", "\n" "}\n" }; const char *sample_write_image_kernel_pattern[] = { "__kernel void sample_test( ", " )\n" "{\n" " size_t tid = get_global_id(0);\n", "\n" "}\n" }; const char *sample_large_parmam_kernel_pattern[] = { "__kernel void sample_test(%s, __global long *result)\n" "{\n" "result[0] = 0;\n" "%s" "\n" "}\n" }; const char *sample_large_int_parmam_kernel_pattern[] = { "__kernel void sample_test(%s, __global int *result)\n" "{\n" "result[0] = 0;\n" "%s" "\n" "}\n" }; const char *sample_sampler_kernel_pattern[] = { "__kernel void sample_test( read_only image2d_t src, __global int4 *dst", ", sampler_t sampler%d", ")\n" "{\n" " size_t tid = get_global_id(0);\n", " dst[ 0 ] = read_imagei( src, sampler%d, (int2)( 0, 0 ) );\n", "\n" "}\n" }; const char *sample_const_arg_kernel[] = { "__kernel void sample_test(__constant int *src1, __global int *dst)\n" "{\n" " size_t tid = get_global_id(0);\n" "\n" " dst[tid] = src1[tid];\n" "\n" "}\n" }; const char *sample_local_arg_kernel[] = { "__kernel void sample_test(__local int *src1, __global int *global_src, " "__global int *dst)\n" "{\n" " size_t tid = get_global_id(0);\n" "\n" " src1[tid] = global_src[tid];\n" " barrier(CLK_GLOBAL_MEM_FENCE);\n" " dst[tid] = src1[tid];\n" "\n" "}\n" }; const char *sample_const_max_arg_kernel_pattern = "__kernel void sample_test(__constant int *src1 %s, __global int *dst)\n" "{\n" " int tid = get_global_id(0);\n" "\n" " dst[tid] = src1[tid];\n" "%s" "\n" "}\n"; int test_min_max_thread_dimensions(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int error, retVal; unsigned int maxThreadDim, threadDim, i; clProgramWrapper program; clKernelWrapper kernel; clMemWrapper streams[1]; size_t *threads, *localThreads; cl_event event; cl_int event_status; /* Get the max thread dimensions */ error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS, sizeof(maxThreadDim), &maxThreadDim, NULL); test_error(error, "Unable to get max work item dimensions from device"); if (maxThreadDim < 3) { log_error("ERROR: Reported max work item dimensions is less than " "required! (%d)\n", maxThreadDim); return -1; } log_info("Reported max thread dimensions of %d.\n", maxThreadDim); /* Create a kernel to test with */ if (create_single_kernel_helper(context, &program, &kernel, 1, sample_single_param_kernel, "sample_test") != 0) { return -1; } /* Create some I/O streams */ streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeof(cl_int) * 100, NULL, &error); if (streams[0] == NULL) { log_error("ERROR: Creating test array failed!\n"); return -1; } /* Set the arguments */ error = clSetKernelArg(kernel, 0, sizeof(streams[0]), &streams[0]); test_error(error, "Unable to set kernel arguments"); retVal = 0; /* Now try running the kernel with up to that many threads */ for (threadDim = 1; threadDim <= maxThreadDim; threadDim++) { threads = (size_t *)malloc(sizeof(size_t) * maxThreadDim); localThreads = (size_t *)malloc(sizeof(size_t) * maxThreadDim); for (i = 0; i < maxThreadDim; i++) { threads[i] = 1; localThreads[i] = 1; } error = clEnqueueNDRangeKernel(queue, kernel, maxThreadDim, NULL, threads, localThreads, 0, NULL, &event); test_error(error, "Failed clEnqueueNDRangeKernel"); // Verify that the event does not return an error from the execution error = clWaitForEvents(1, &event); test_error(error, "clWaitForEvent failed"); error = clGetEventInfo(event, CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof(event_status), &event_status, NULL); test_error( error, "clGetEventInfo for CL_EVENT_COMMAND_EXECUTION_STATUS failed"); clReleaseEvent(event); if (event_status < 0) test_error(error, "Kernel execution event returned error"); /* All done */ free(threads); free(localThreads); } return retVal; } int test_min_max_work_items_sizes(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int error; size_t *deviceMaxWorkItemSize; unsigned int maxWorkItemDim; /* Get the max work item dimensions */ error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS, sizeof(maxWorkItemDim), &maxWorkItemDim, NULL); test_error(error, "Unable to get max work item dimensions from device"); log_info("CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS returned %d\n", maxWorkItemDim); deviceMaxWorkItemSize = (size_t *)malloc(sizeof(size_t) * maxWorkItemDim); error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_WORK_ITEM_SIZES, sizeof(size_t) * maxWorkItemDim, deviceMaxWorkItemSize, NULL); test_error(error, "clDeviceInfo for CL_DEVICE_MAX_WORK_ITEM_SIZES failed"); unsigned int i; int errors = 0; for (i = 0; i < maxWorkItemDim; i++) { if (deviceMaxWorkItemSize[i] < 1) { log_error("MAX_WORK_ITEM_SIZE in dimension %d is invalid: %lu\n", i, deviceMaxWorkItemSize[i]); errors++; } else { log_info("Dimension %d has max work item size %lu\n", i, deviceMaxWorkItemSize[i]); } } free(deviceMaxWorkItemSize); if (errors) return -1; return 0; } int test_min_max_work_group_size(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int error; size_t deviceMaxThreadSize; /* Get the max thread dimensions */ error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_WORK_GROUP_SIZE, sizeof(deviceMaxThreadSize), &deviceMaxThreadSize, NULL); test_error(error, "Unable to get max work group size from device"); log_info("Reported %ld max device work group size.\n", deviceMaxThreadSize); if (deviceMaxThreadSize == 0) { log_error("ERROR: Max work group size is reported as zero!\n"); return -1; } return 0; } int test_min_max_read_image_args(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int error; unsigned int maxReadImages, i; unsigned int deviceAddressSize; clProgramWrapper program; char readArgLine[128], *programSrc; const char *readArgPattern = ", read_only image2d_t srcimg%d"; clKernelWrapper kernel; clMemWrapper *streams, result; size_t threads[2]; cl_image_format image_format_desc; size_t maxParameterSize; cl_event event; cl_int event_status; cl_float image_data[4 * 4]; float image_result = 0.0f; float actual_image_result; cl_uint minRequiredReadImages = gIsEmbedded ? 8 : 128; cl_device_type deviceType; PASSIVE_REQUIRE_IMAGE_SUPPORT(deviceID) image_format_desc.image_channel_order = CL_RGBA; image_format_desc.image_channel_data_type = CL_FLOAT; /* Get the max read image arg count */ error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_READ_IMAGE_ARGS, sizeof(maxReadImages), &maxReadImages, NULL); test_error(error, "Unable to get max read image arg count from device"); if (maxReadImages < minRequiredReadImages) { log_error("ERROR: Reported max read image arg count is less than " "required! (%d)\n", maxReadImages); return -1; } log_info("Reported %d max read image args.\n", maxReadImages); error = clGetDeviceInfo(deviceID, CL_DEVICE_ADDRESS_BITS, sizeof(deviceAddressSize), &deviceAddressSize, NULL); test_error(error, "Unable to query CL_DEVICE_ADDRESS_BITS for device"); deviceAddressSize /= 8; // convert from bits to bytes error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_PARAMETER_SIZE, sizeof(maxParameterSize), &maxParameterSize, NULL); test_error(error, "Unable to get max parameter size from device"); if (!gIsEmbedded && maxReadImages >= 128 && maxParameterSize == 1024) { error = clGetDeviceInfo(deviceID, CL_DEVICE_TYPE, sizeof(deviceType), &deviceType, NULL); test_error(error, "Unable to get device type from device"); if (deviceType != CL_DEVICE_TYPE_CUSTOM) { maxReadImages = 127; } } // Subtract the size of the result maxParameterSize -= deviceAddressSize; // Calculate the number we can use if (maxParameterSize / deviceAddressSize < maxReadImages) { log_info("WARNING: Max parameter size of %d bytes limits test to %d " "max image arguments.\n", (int)maxParameterSize, (int)(maxParameterSize / deviceAddressSize)); maxReadImages = (unsigned int)(maxParameterSize / deviceAddressSize); } /* Create a program with that many read args */ programSrc = (char *)malloc(strlen(sample_read_image_kernel_pattern[0]) + (strlen(readArgPattern) + 6) * (maxReadImages) + strlen(sample_read_image_kernel_pattern[1]) + 1 + 40240); strcpy(programSrc, sample_read_image_kernel_pattern[0]); strcat(programSrc, "read_only image2d_t srcimg0"); for (i = 0; i < maxReadImages - 1; i++) { sprintf(readArgLine, readArgPattern, i + 1); strcat(programSrc, readArgLine); } strcat(programSrc, sample_read_image_kernel_pattern[1]); for (i = 0; i < maxReadImages; i++) { sprintf( readArgLine, "\tresult[0] += read_imagef( srcimg%d, sampler, (int2)(0,0)).x;\n", i); strcat(programSrc, readArgLine); } strcat(programSrc, sample_read_image_kernel_pattern[2]); error = create_single_kernel_helper(context, &program, &kernel, 1, (const char **)&programSrc, "sample_test"); test_error(error, "Failed to create the program and kernel."); free(programSrc); result = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeof(cl_float), NULL, &error); test_error(error, "clCreateBufer failed"); /* Create some I/O streams */ streams = new clMemWrapper[maxReadImages + 1]; for (i = 0; i < maxReadImages; i++) { image_data[0] = i; image_result += image_data[0]; streams[i] = create_image_2d(context, CL_MEM_READ_WRITE | CL_MEM_COPY_HOST_PTR, &image_format_desc, 4, 4, 0, image_data, &error); test_error(error, "Unable to allocate test image"); } error = clSetKernelArg(kernel, 0, sizeof(result), &result); test_error(error, "Unable to set kernel arguments"); /* Set the arguments */ for (i = 1; i < maxReadImages + 1; i++) { error = clSetKernelArg(kernel, i, sizeof(streams[i - 1]), &streams[i - 1]); test_error(error, "Unable to set kernel arguments"); } /* Now try running the kernel */ threads[0] = threads[1] = 1; error = clEnqueueNDRangeKernel(queue, kernel, 2, NULL, threads, NULL, 0, NULL, &event); test_error(error, "clEnqueueNDRangeKernel failed"); // Verify that the event does not return an error from the execution error = clWaitForEvents(1, &event); test_error(error, "clWaitForEvent failed"); error = clGetEventInfo(event, CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof(event_status), &event_status, NULL); test_error(error, "clGetEventInfo for CL_EVENT_COMMAND_EXECUTION_STATUS failed"); clReleaseEvent(event); if (event_status < 0) test_error(error, "Kernel execution event returned error"); error = clEnqueueReadBuffer(queue, result, CL_TRUE, 0, sizeof(cl_float), &actual_image_result, 0, NULL, NULL); test_error(error, "clEnqueueReadBuffer failed"); delete[] streams; if (actual_image_result != image_result) { log_error("Result failed to verify. Got %g, expected %g.\n", actual_image_result, image_result); return 1; } return 0; } int test_min_max_write_image_args(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int error; unsigned int maxWriteImages, i; clProgramWrapper program; char writeArgLine[128], *programSrc; const char *writeArgPattern = ", write_only image2d_t dstimg%d"; clKernelWrapper kernel; clMemWrapper *streams; size_t threads[2]; cl_image_format image_format_desc; size_t maxParameterSize; cl_event event; cl_int event_status; cl_uint minRequiredWriteImages = gIsEmbedded ? 1 : 8; PASSIVE_REQUIRE_IMAGE_SUPPORT(deviceID) image_format_desc.image_channel_order = CL_RGBA; image_format_desc.image_channel_data_type = CL_UNORM_INT8; /* Get the max read image arg count */ error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_WRITE_IMAGE_ARGS, sizeof(maxWriteImages), &maxWriteImages, NULL); test_error(error, "Unable to get max write image arg count from device"); if (maxWriteImages == 0) { log_info( "WARNING: Device reports 0 for a max write image arg count (write " "image arguments unsupported). Skipping test (implicitly passes). " "This is only valid if the number of image formats is also 0.\n"); return 0; } if (maxWriteImages < minRequiredWriteImages) { log_error("ERROR: Reported max write image arg count is less than " "required! (%d)\n", maxWriteImages); return -1; } log_info("Reported %d max write image args.\n", maxWriteImages); error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_PARAMETER_SIZE, sizeof(maxParameterSize), &maxParameterSize, NULL); test_error(error, "Unable to get max parameter size from device"); // Calculate the number we can use if (maxParameterSize / sizeof(cl_mem) < maxWriteImages) { log_info("WARNING: Max parameter size of %d bytes limits test to %d " "max image arguments.\n", (int)maxParameterSize, (int)(maxParameterSize / sizeof(cl_mem))); maxWriteImages = (unsigned int)(maxParameterSize / sizeof(cl_mem)); } /* Create a program with that many write args + 1 */ programSrc = (char *)malloc( strlen(sample_write_image_kernel_pattern[0]) + (strlen(writeArgPattern) + 6) * (maxWriteImages + 1) + strlen(sample_write_image_kernel_pattern[1]) + 1 + 40240); strcpy(programSrc, sample_write_image_kernel_pattern[0]); strcat(programSrc, "write_only image2d_t dstimg0"); for (i = 1; i < maxWriteImages; i++) { sprintf(writeArgLine, writeArgPattern, i); strcat(programSrc, writeArgLine); } strcat(programSrc, sample_write_image_kernel_pattern[1]); for (i = 0; i < maxWriteImages; i++) { sprintf(writeArgLine, "\twrite_imagef( dstimg%d, (int2)(0,0), (float4)(0,0,0,0));\n", i); strcat(programSrc, writeArgLine); } strcat(programSrc, sample_write_image_kernel_pattern[2]); error = create_single_kernel_helper(context, &program, &kernel, 1, (const char **)&programSrc, "sample_test"); test_error(error, "Failed to create the program and kernel."); free(programSrc); /* Create some I/O streams */ streams = new clMemWrapper[maxWriteImages + 1]; for (i = 0; i < maxWriteImages; i++) { streams[i] = create_image_2d(context, CL_MEM_READ_WRITE, &image_format_desc, 16, 16, 0, NULL, &error); test_error(error, "Unable to allocate test image"); } /* Set the arguments */ for (i = 0; i < maxWriteImages; i++) { error = clSetKernelArg(kernel, i, sizeof(streams[i]), &streams[i]); test_error(error, "Unable to set kernel arguments"); } /* Now try running the kernel */ threads[0] = threads[1] = 16; error = clEnqueueNDRangeKernel(queue, kernel, 2, NULL, threads, NULL, 0, NULL, &event); test_error(error, "clEnqueueNDRangeKernel failed."); // Verify that the event does not return an error from the execution error = clWaitForEvents(1, &event); test_error(error, "clWaitForEvent failed"); error = clGetEventInfo(event, CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof(event_status), &event_status, NULL); test_error(error, "clGetEventInfo for CL_EVENT_COMMAND_EXECUTION_STATUS failed"); clReleaseEvent(event); if (event_status < 0) test_error(error, "Kernel execution event returned error"); /* All done */ delete[] streams; return 0; } int test_min_max_mem_alloc_size(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int error; cl_ulong maxAllocSize, memSize, minSizeToTry; clMemWrapper memHdl; cl_ulong requiredAllocSize; if (gIsEmbedded) requiredAllocSize = 1 * 1024 * 1024; else requiredAllocSize = 128 * 1024 * 1024; /* Get the max mem alloc size */ error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_MEM_ALLOC_SIZE, sizeof(maxAllocSize), &maxAllocSize, NULL); test_error(error, "Unable to get max mem alloc size from device"); error = clGetDeviceInfo(deviceID, CL_DEVICE_GLOBAL_MEM_SIZE, sizeof(memSize), &memSize, NULL); test_error(error, "Unable to get global memory size from device"); if (memSize > (cl_ulong)SIZE_MAX) { memSize = (cl_ulong)SIZE_MAX; } if (maxAllocSize < requiredAllocSize) { log_error("ERROR: Reported max allocation size is less than required " "%lldMB! (%llu or %lluMB, from a total mem size of %lldMB)\n", (requiredAllocSize / 1024) / 1024, maxAllocSize, (maxAllocSize / 1024) / 1024, (memSize / 1024) / 1024); return -1; } requiredAllocSize = ((memSize / 4) > (1024 * 1024 * 1024)) ? 1024 * 1024 * 1024 : memSize / 4; if (gIsEmbedded) requiredAllocSize = (requiredAllocSize < 1 * 1024 * 1024) ? 1 * 1024 * 1024 : requiredAllocSize; else requiredAllocSize = (requiredAllocSize < 128 * 1024 * 1024) ? 128 * 1024 * 1024 : requiredAllocSize; if (maxAllocSize < requiredAllocSize) { log_error( "ERROR: Reported max allocation size is less than required of " "total memory! (%llu or %lluMB, from a total mem size of %lluMB)\n", maxAllocSize, (maxAllocSize / 1024) / 1024, (requiredAllocSize / 1024) / 1024); return -1; } log_info("Reported max allocation size of %lld bytes (%gMB) and global mem " "size of %lld bytes (%gMB).\n", maxAllocSize, maxAllocSize / (1024.0 * 1024.0), requiredAllocSize, requiredAllocSize / (1024.0 * 1024.0)); if (memSize < maxAllocSize) { log_info("Global memory size is less than max allocation size, using " "that.\n"); maxAllocSize = memSize; } minSizeToTry = maxAllocSize / 16; while (maxAllocSize > (maxAllocSize / 4)) { log_info("Trying to create a buffer of size of %lld bytes (%gMB).\n", maxAllocSize, (double)maxAllocSize / (1024.0 * 1024.0)); memHdl = clCreateBuffer(context, CL_MEM_READ_ONLY, (size_t)maxAllocSize, NULL, &error); if (error == CL_MEM_OBJECT_ALLOCATION_FAILURE || error == CL_OUT_OF_RESOURCES || error == CL_OUT_OF_HOST_MEMORY) { log_info("\tAllocation failed at size of %lld bytes (%gMB).\n", maxAllocSize, (double)maxAllocSize / (1024.0 * 1024.0)); maxAllocSize -= minSizeToTry; continue; } test_error(error, "clCreateBuffer failed for maximum sized buffer."); return 0; } log_error("Failed to allocate even %lld bytes (%gMB).\n", maxAllocSize, (double)maxAllocSize / (1024.0 * 1024.0)); return -1; } int test_min_max_image_2d_width(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int error; size_t maxDimension; clMemWrapper streams[1]; cl_image_format image_format_desc; cl_ulong maxAllocSize; cl_uint minRequiredDimension; PASSIVE_REQUIRE_IMAGE_SUPPORT(deviceID) auto version = get_device_cl_version(deviceID); if (version == Version(1, 0)) { minRequiredDimension = gIsEmbedded ? 2048 : 4096; } else { minRequiredDimension = gIsEmbedded ? 2048 : 8192; } /* Just get any ol format to test with */ error = get_8_bit_image_format(context, CL_MEM_OBJECT_IMAGE2D, CL_MEM_READ_WRITE, 0, &image_format_desc); test_error(error, "Unable to obtain suitable image format to test with!"); /* Get the max 2d image width */ error = clGetDeviceInfo(deviceID, CL_DEVICE_IMAGE2D_MAX_WIDTH, sizeof(maxDimension), &maxDimension, NULL); test_error(error, "Unable to get max image 2d width from device"); if (maxDimension < minRequiredDimension) { log_error( "ERROR: Reported max image 2d width is less than required! (%d)\n", (int)maxDimension); return -1; } log_info("Max reported width is %ld.\n", maxDimension); /* Verify we can use the format */ image_format_desc.image_channel_data_type = CL_UNORM_INT8; image_format_desc.image_channel_order = CL_RGBA; if (!is_image_format_supported(context, CL_MEM_READ_ONLY, CL_MEM_OBJECT_IMAGE2D, &image_format_desc)) { log_error("CL_UNORM_INT8 CL_RGBA not supported. Can not test."); return -1; } /* Verify that we can actually allocate an image that large */ error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_MEM_ALLOC_SIZE, sizeof(maxAllocSize), &maxAllocSize, NULL); test_error(error, "Unable to get CL_DEVICE_MAX_MEM_ALLOC_SIZE."); if ((cl_ulong)maxDimension * 1 * 4 > maxAllocSize) { log_error("Can not allocate a large enough image (min size: %lld " "bytes, max allowed: %lld bytes) to test.\n", (cl_ulong)maxDimension * 1 * 4, maxAllocSize); return -1; } log_info("Attempting to create an image of size %d x 1 = %gMB.\n", (int)maxDimension, ((float)maxDimension * 4 / 1024.0 / 1024.0)); /* Try to allocate a very big image */ streams[0] = create_image_2d(context, CL_MEM_READ_ONLY, &image_format_desc, maxDimension, 1, 0, NULL, &error); if ((streams[0] == NULL) || (error != CL_SUCCESS)) { print_error(error, "Image 2D creation failed for maximum width"); return -1; } return 0; } int test_min_max_image_2d_height(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int error; size_t maxDimension; clMemWrapper streams[1]; cl_image_format image_format_desc; cl_ulong maxAllocSize; cl_uint minRequiredDimension; PASSIVE_REQUIRE_IMAGE_SUPPORT(deviceID) auto version = get_device_cl_version(deviceID); if (version == Version(1, 0)) { minRequiredDimension = gIsEmbedded ? 2048 : 4096; } else { minRequiredDimension = gIsEmbedded ? 2048 : 8192; } /* Just get any ol format to test with */ error = get_8_bit_image_format(context, CL_MEM_OBJECT_IMAGE2D, CL_MEM_READ_WRITE, 0, &image_format_desc); test_error(error, "Unable to obtain suitable image format to test with!"); /* Get the max 2d image width */ error = clGetDeviceInfo(deviceID, CL_DEVICE_IMAGE2D_MAX_HEIGHT, sizeof(maxDimension), &maxDimension, NULL); test_error(error, "Unable to get max image 2d height from device"); if (maxDimension < minRequiredDimension) { log_error( "ERROR: Reported max image 2d height is less than required! (%d)\n", (int)maxDimension); return -1; } log_info("Max reported height is %ld.\n", maxDimension); /* Verify we can use the format */ image_format_desc.image_channel_data_type = CL_UNORM_INT8; image_format_desc.image_channel_order = CL_RGBA; if (!is_image_format_supported(context, CL_MEM_READ_ONLY, CL_MEM_OBJECT_IMAGE2D, &image_format_desc)) { log_error("CL_UNORM_INT8 CL_RGBA not supported. Can not test."); return -1; } /* Verify that we can actually allocate an image that large */ error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_MEM_ALLOC_SIZE, sizeof(maxAllocSize), &maxAllocSize, NULL); test_error(error, "Unable to get CL_DEVICE_MAX_MEM_ALLOC_SIZE."); if ((cl_ulong)maxDimension * 1 * 4 > maxAllocSize) { log_error("Can not allocate a large enough image (min size: %lld " "bytes, max allowed: %lld bytes) to test.\n", (cl_ulong)maxDimension * 1 * 4, maxAllocSize); return -1; } log_info("Attempting to create an image of size 1 x %d = %gMB.\n", (int)maxDimension, ((float)maxDimension * 4 / 1024.0 / 1024.0)); /* Try to allocate a very big image */ streams[0] = create_image_2d(context, CL_MEM_READ_ONLY, &image_format_desc, 1, maxDimension, 0, NULL, &error); if ((streams[0] == NULL) || (error != CL_SUCCESS)) { print_error(error, "Image 2D creation failed for maximum height"); return -1; } return 0; } int test_min_max_image_3d_width(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int error; size_t maxDimension; clMemWrapper streams[1]; cl_image_format image_format_desc; cl_ulong maxAllocSize; PASSIVE_REQUIRE_3D_IMAGE_SUPPORT(deviceID) /* Just get any ol format to test with */ error = get_8_bit_image_format(context, CL_MEM_OBJECT_IMAGE3D, CL_MEM_READ_ONLY, 0, &image_format_desc); test_error(error, "Unable to obtain suitable image format to test with!"); /* Get the max 2d image width */ error = clGetDeviceInfo(deviceID, CL_DEVICE_IMAGE3D_MAX_WIDTH, sizeof(maxDimension), &maxDimension, NULL); test_error(error, "Unable to get max image 3d width from device"); if (maxDimension < 2048) { log_error( "ERROR: Reported max image 3d width is less than required! (%d)\n", (int)maxDimension); return -1; } log_info("Max reported width is %ld.\n", maxDimension); /* Verify we can use the format */ image_format_desc.image_channel_data_type = CL_UNORM_INT8; image_format_desc.image_channel_order = CL_RGBA; if (!is_image_format_supported(context, CL_MEM_READ_ONLY, CL_MEM_OBJECT_IMAGE3D, &image_format_desc)) { log_error("CL_UNORM_INT8 CL_RGBA not supported. Can not test."); return -1; } /* Verify that we can actually allocate an image that large */ error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_MEM_ALLOC_SIZE, sizeof(maxAllocSize), &maxAllocSize, NULL); test_error(error, "Unable to get CL_DEVICE_MAX_MEM_ALLOC_SIZE."); if ((cl_ulong)maxDimension * 2 * 4 > maxAllocSize) { log_error("Can not allocate a large enough image (min size: %lld " "bytes, max allowed: %lld bytes) to test.\n", (cl_ulong)maxDimension * 2 * 4, maxAllocSize); return -1; } log_info("Attempting to create an image of size %d x 1 x 2 = %gMB.\n", (int)maxDimension, (2 * (float)maxDimension * 4 / 1024.0 / 1024.0)); /* Try to allocate a very big image */ streams[0] = create_image_3d(context, CL_MEM_READ_ONLY, &image_format_desc, maxDimension, 1, 2, 0, 0, NULL, &error); if ((streams[0] == NULL) || (error != CL_SUCCESS)) { print_error(error, "Image 3D creation failed for maximum width"); return -1; } return 0; } int test_min_max_image_3d_height(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int error; size_t maxDimension; clMemWrapper streams[1]; cl_image_format image_format_desc; cl_ulong maxAllocSize; PASSIVE_REQUIRE_3D_IMAGE_SUPPORT(deviceID) /* Just get any ol format to test with */ error = get_8_bit_image_format(context, CL_MEM_OBJECT_IMAGE3D, CL_MEM_READ_ONLY, 0, &image_format_desc); test_error(error, "Unable to obtain suitable image format to test with!"); /* Get the max 2d image width */ error = clGetDeviceInfo(deviceID, CL_DEVICE_IMAGE3D_MAX_HEIGHT, sizeof(maxDimension), &maxDimension, NULL); test_error(error, "Unable to get max image 3d height from device"); if (maxDimension < 2048) { log_error( "ERROR: Reported max image 3d height is less than required! (%d)\n", (int)maxDimension); return -1; } log_info("Max reported height is %ld.\n", maxDimension); /* Verify we can use the format */ image_format_desc.image_channel_data_type = CL_UNORM_INT8; image_format_desc.image_channel_order = CL_RGBA; if (!is_image_format_supported(context, CL_MEM_READ_ONLY, CL_MEM_OBJECT_IMAGE3D, &image_format_desc)) { log_error("CL_UNORM_INT8 CL_RGBA not supported. Can not test."); return -1; } /* Verify that we can actually allocate an image that large */ error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_MEM_ALLOC_SIZE, sizeof(maxAllocSize), &maxAllocSize, NULL); test_error(error, "Unable to get CL_DEVICE_MAX_MEM_ALLOC_SIZE."); if ((cl_ulong)maxDimension * 2 * 4 > maxAllocSize) { log_error("Can not allocate a large enough image (min size: %lld " "bytes, max allowed: %lld bytes) to test.\n", (cl_ulong)maxDimension * 2 * 4, maxAllocSize); return -1; } log_info("Attempting to create an image of size 1 x %d x 2 = %gMB.\n", (int)maxDimension, (2 * (float)maxDimension * 4 / 1024.0 / 1024.0)); /* Try to allocate a very big image */ streams[0] = create_image_3d(context, CL_MEM_READ_ONLY, &image_format_desc, 1, maxDimension, 2, 0, 0, NULL, &error); if ((streams[0] == NULL) || (error != CL_SUCCESS)) { print_error(error, "Image 3D creation failed for maximum height"); return -1; } return 0; } int test_min_max_image_3d_depth(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int error; size_t maxDimension; clMemWrapper streams[1]; cl_image_format image_format_desc; cl_ulong maxAllocSize; PASSIVE_REQUIRE_3D_IMAGE_SUPPORT(deviceID) /* Just get any ol format to test with */ error = get_8_bit_image_format(context, CL_MEM_OBJECT_IMAGE3D, CL_MEM_READ_ONLY, 0, &image_format_desc); test_error(error, "Unable to obtain suitable image format to test with!"); /* Get the max 2d image width */ error = clGetDeviceInfo(deviceID, CL_DEVICE_IMAGE3D_MAX_DEPTH, sizeof(maxDimension), &maxDimension, NULL); test_error(error, "Unable to get max image 3d depth from device"); if (maxDimension < 2048) { log_error( "ERROR: Reported max image 3d depth is less than required! (%d)\n", (int)maxDimension); return -1; } log_info("Max reported depth is %ld.\n", maxDimension); /* Verify we can use the format */ image_format_desc.image_channel_data_type = CL_UNORM_INT8; image_format_desc.image_channel_order = CL_RGBA; if (!is_image_format_supported(context, CL_MEM_READ_ONLY, CL_MEM_OBJECT_IMAGE3D, &image_format_desc)) { log_error("CL_UNORM_INT8 CL_RGBA not supported. Can not test."); return -1; } /* Verify that we can actually allocate an image that large */ error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_MEM_ALLOC_SIZE, sizeof(maxAllocSize), &maxAllocSize, NULL); test_error(error, "Unable to get CL_DEVICE_MAX_MEM_ALLOC_SIZE."); if ((cl_ulong)maxDimension * 1 * 4 > maxAllocSize) { log_error("Can not allocate a large enough image (min size: %lld " "bytes, max allowed: %lld bytes) to test.\n", (cl_ulong)maxDimension * 1 * 4, maxAllocSize); return -1; } log_info("Attempting to create an image of size 1 x 1 x %d = %gMB.\n", (int)maxDimension, ((float)maxDimension * 4 / 1024.0 / 1024.0)); /* Try to allocate a very big image */ streams[0] = create_image_3d(context, CL_MEM_READ_ONLY, &image_format_desc, 1, 1, maxDimension, 0, 0, NULL, &error); if ((streams[0] == NULL) || (error != CL_SUCCESS)) { print_error(error, "Image 3D creation failed for maximum depth"); return -1; } return 0; } int test_min_max_image_array_size(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int error; size_t maxDimension; clMemWrapper streams[1]; cl_image_format image_format_desc; cl_ulong maxAllocSize; size_t minRequiredDimension = gIsEmbedded ? 256 : 2048; PASSIVE_REQUIRE_IMAGE_SUPPORT(deviceID); /* Just get any ol format to test with */ error = get_8_bit_image_format(context, CL_MEM_OBJECT_IMAGE2D_ARRAY, CL_MEM_READ_WRITE, 0, &image_format_desc); test_error(error, "Unable to obtain suitable image format to test with!"); /* Get the max image array width */ error = clGetDeviceInfo(deviceID, CL_DEVICE_IMAGE_MAX_ARRAY_SIZE, sizeof(maxDimension), &maxDimension, NULL); test_error(error, "Unable to get max image array size from device"); if (maxDimension < minRequiredDimension) { log_error("ERROR: Reported max image array size is less than required! " "(%d)\n", (int)maxDimension); return -1; } log_info("Max reported image array size is %ld.\n", maxDimension); /* Verify we can use the format */ image_format_desc.image_channel_data_type = CL_UNORM_INT8; image_format_desc.image_channel_order = CL_RGBA; if (!is_image_format_supported(context, CL_MEM_READ_ONLY, CL_MEM_OBJECT_IMAGE2D_ARRAY, &image_format_desc)) { log_error("CL_UNORM_INT8 CL_RGBA not supported. Can not test."); return -1; } /* Verify that we can actually allocate an image that large */ error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_MEM_ALLOC_SIZE, sizeof(maxAllocSize), &maxAllocSize, NULL); test_error(error, "Unable to get CL_DEVICE_MAX_MEM_ALLOC_SIZE."); if ((cl_ulong)maxDimension * 1 * 4 > maxAllocSize) { log_error("Can not allocate a large enough image (min size: %lld " "bytes, max allowed: %lld bytes) to test.\n", (cl_ulong)maxDimension * 1 * 4, maxAllocSize); return -1; } log_info("Attempting to create an image of size 1 x 1 x %d = %gMB.\n", (int)maxDimension, ((float)maxDimension * 4 / 1024.0 / 1024.0)); /* Try to allocate a very big image */ streams[0] = create_image_2d_array(context, CL_MEM_READ_ONLY, &image_format_desc, 1, 1, maxDimension, 0, 0, NULL, &error); if ((streams[0] == NULL) || (error != CL_SUCCESS)) { print_error(error, "2D Image Array creation failed for maximum array size"); return -1; } return 0; } int test_min_max_image_buffer_size(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int error; size_t maxDimensionPixels; clMemWrapper streams[2]; cl_image_format image_format_desc = { 0 }; cl_ulong maxAllocSize; size_t minRequiredDimension = gIsEmbedded ? 2048 : 65536; unsigned int i = 0; size_t pixelBytes = 0; PASSIVE_REQUIRE_IMAGE_SUPPORT(deviceID); /* Get the max memory allocation size */ error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_MEM_ALLOC_SIZE, sizeof(maxAllocSize), &maxAllocSize, NULL); test_error(error, "Unable to get CL_DEVICE_MAX_MEM_ALLOC_SIZE."); /* Get the max image array width */ error = clGetDeviceInfo(deviceID, CL_DEVICE_IMAGE_MAX_BUFFER_SIZE, sizeof(maxDimensionPixels), &maxDimensionPixels, NULL); test_error(error, "Unable to get max image buffer size from device"); if (maxDimensionPixels < minRequiredDimension) { log_error("ERROR: Reported max image buffer size is less than " "required! (%d)\n", (int)maxDimensionPixels); return -1; } log_info("Max reported image buffer size is %ld pixels.\n", maxDimensionPixels); pixelBytes = maxAllocSize / maxDimensionPixels; if (pixelBytes == 0) { log_error("Value of CL_DEVICE_IMAGE_MAX_BUFFER_SIZE is greater than " "CL_MAX_MEM_ALLOC_SIZE so there is no way to allocate image " "of maximum size!\n"); return -1; } error = -1; for (i = pixelBytes; i > 0; --i) { error = get_8_bit_image_format(context, CL_MEM_OBJECT_IMAGE1D, CL_MEM_READ_ONLY, i, &image_format_desc); if (error == CL_SUCCESS) { pixelBytes = i; break; } } test_error(error, "Device does not support format to be used to allocate image of " "CL_DEVICE_IMAGE_MAX_BUFFER_SIZE\n"); log_info("Attempting to create an 1D image with channel order %s from " "buffer of size %d = %gMB.\n", GetChannelOrderName(image_format_desc.image_channel_order), (int)maxDimensionPixels, ((float)maxDimensionPixels * pixelBytes / 1024.0 / 1024.0)); /* Try to allocate a buffer */ streams[0] = clCreateBuffer(context, CL_MEM_READ_ONLY, maxDimensionPixels * pixelBytes, NULL, &error); if ((streams[0] == NULL) || (error != CL_SUCCESS)) { print_error(error, "Buffer creation failed for maximum image buffer size"); return -1; } /* Try to allocate a 1D image array from buffer */ streams[1] = create_image_1d(context, CL_MEM_READ_ONLY, &image_format_desc, maxDimensionPixels, 0, NULL, streams[0], &error); if ((streams[0] == NULL) || (error != CL_SUCCESS)) { print_error(error, "1D Image from buffer creation failed for maximum image " "buffer size"); return -1; } return 0; } int test_min_max_parameter_size(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int error, i; size_t maxSize; char *programSrc; char *ptr; size_t numberExpected; long numberOfIntParametersToTry; char *argumentLine, *codeLines; void *data; cl_long long_result, expectedResult; cl_int int_result; size_t decrement; cl_event event; cl_int event_status; bool embeddedNoLong = gIsEmbedded && !gHasLong; /* Get the max param size */ error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_PARAMETER_SIZE, sizeof(maxSize), &maxSize, NULL); test_error(error, "Unable to get max parameter size from device"); if (((!gIsEmbedded) && (maxSize < 1024)) || ((gIsEmbedded) && (maxSize < 256))) { log_error( "ERROR: Reported max parameter size is less than required! (%d)\n", (int)maxSize); return -1; } /* The embedded profile without cles_khr_int64 extension does not require * longs, so use ints */ if (embeddedNoLong) numberOfIntParametersToTry = numberExpected = (maxSize - sizeof(cl_mem)) / sizeof(cl_int); else numberOfIntParametersToTry = numberExpected = (maxSize - sizeof(cl_mem)) / sizeof(cl_long); decrement = (size_t)(numberOfIntParametersToTry / 8); if (decrement < 1) decrement = 1; log_info("Reported max parameter size of %d bytes.\n", (int)maxSize); while (numberOfIntParametersToTry > 0) { // These need to be inside to be deallocated automatically on each loop // iteration. clProgramWrapper program; clMemWrapper mem; clKernelWrapper kernel; if (embeddedNoLong) { log_info( "Trying a kernel with %ld int arguments (%ld bytes) and one " "cl_mem (%ld bytes) for %ld bytes total.\n", numberOfIntParametersToTry, sizeof(cl_int) * numberOfIntParametersToTry, sizeof(cl_mem), sizeof(cl_mem) + numberOfIntParametersToTry * sizeof(cl_int)); } else { log_info( "Trying a kernel with %ld long arguments (%ld bytes) and one " "cl_mem (%ld bytes) for %ld bytes total.\n", numberOfIntParametersToTry, sizeof(cl_long) * numberOfIntParametersToTry, sizeof(cl_mem), sizeof(cl_mem) + numberOfIntParametersToTry * sizeof(cl_long)); } // Allocate memory for the program storage data = malloc(sizeof(cl_long) * numberOfIntParametersToTry); argumentLine = (char *)malloc(sizeof(char) * numberOfIntParametersToTry * 32); codeLines = (char *)malloc(sizeof(char) * numberOfIntParametersToTry * 32); programSrc = (char *)malloc(sizeof(char) * (numberOfIntParametersToTry * 64 + 1024)); argumentLine[0] = '\0'; codeLines[0] = '\0'; programSrc[0] = '\0'; // Generate our results expectedResult = 0; for (i = 0; i < (int)numberOfIntParametersToTry; i++) { if (gHasLong) { ((cl_long *)data)[i] = i; expectedResult += i; } else { ((cl_int *)data)[i] = i; expectedResult += i; } } // Build the program if (gHasLong) sprintf(argumentLine, "%s", "long arg0"); else sprintf(argumentLine, "%s", "int arg0"); sprintf(codeLines, "%s", "result[0] += arg0;"); for (i = 1; i < (int)numberOfIntParametersToTry; i++) { if (gHasLong) sprintf(argumentLine + strlen(argumentLine), ", long arg%d", i); else sprintf(argumentLine + strlen(argumentLine), ", int arg%d", i); sprintf(codeLines + strlen(codeLines), "\nresult[0] += arg%d;", i); } /* Create a kernel to test with */ sprintf(programSrc, gHasLong ? sample_large_parmam_kernel_pattern[0] : sample_large_int_parmam_kernel_pattern[0], argumentLine, codeLines); ptr = programSrc; if (create_single_kernel_helper(context, &program, &kernel, 1, (const char **)&ptr, "sample_test") != 0) { log_info("Create program failed, decrementing number of parameters " "to try.\n"); numberOfIntParametersToTry -= decrement; continue; } /* Try to set a large argument to the kernel */ mem = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeof(cl_long), NULL, &error); test_error(error, "clCreateBuffer failed"); for (i = 0; i < (int)numberOfIntParametersToTry; i++) { if (gHasLong) error = clSetKernelArg(kernel, i, sizeof(cl_long), &(((cl_long *)data)[i])); else error = clSetKernelArg(kernel, i, sizeof(cl_int), &(((cl_int *)data)[i])); if (error != CL_SUCCESS) { log_info("clSetKernelArg failed (%s), decrementing number of " "parameters to try.\n", IGetErrorString(error)); numberOfIntParametersToTry -= decrement; break; } } if (error != CL_SUCCESS) continue; error = clSetKernelArg(kernel, i, sizeof(cl_mem), &mem); if (error != CL_SUCCESS) { log_info("clSetKernelArg failed (%s), decrementing number of " "parameters to try.\n", IGetErrorString(error)); numberOfIntParametersToTry -= decrement; continue; } size_t globalDim[3] = { 1, 1, 1 }, localDim[3] = { 1, 1, 1 }; error = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, globalDim, localDim, 0, NULL, &event); if (error != CL_SUCCESS) { log_info("clEnqueueNDRangeKernel failed (%s), decrementing number " "of parameters to try.\n", IGetErrorString(error)); numberOfIntParametersToTry -= decrement; continue; } // Verify that the event does not return an error from the execution error = clWaitForEvents(1, &event); test_error(error, "clWaitForEvent failed"); error = clGetEventInfo(event, CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof(event_status), &event_status, NULL); test_error( error, "clGetEventInfo for CL_EVENT_COMMAND_EXECUTION_STATUS failed"); clReleaseEvent(event); if (event_status < 0) test_error(error, "Kernel execution event returned error"); if (gHasLong) error = clEnqueueReadBuffer(queue, mem, CL_TRUE, 0, sizeof(cl_long), &long_result, 0, NULL, NULL); else error = clEnqueueReadBuffer(queue, mem, CL_TRUE, 0, sizeof(cl_int), &int_result, 0, NULL, NULL); test_error(error, "clEnqueueReadBuffer failed") free(data); free(argumentLine); free(codeLines); free(programSrc); if (gHasLong) { if (long_result != expectedResult) { log_error("Expected result (%lld) does not equal actual result " "(%lld).\n", expectedResult, long_result); numberOfIntParametersToTry -= decrement; continue; } else { log_info("Results verified at %ld bytes of arguments.\n", sizeof(cl_mem) + numberOfIntParametersToTry * sizeof(cl_long)); break; } } else { if (int_result != expectedResult) { log_error("Expected result (%lld) does not equal actual result " "(%d).\n", expectedResult, int_result); numberOfIntParametersToTry -= decrement; continue; } else { log_info("Results verified at %ld bytes of arguments.\n", sizeof(cl_mem) + numberOfIntParametersToTry * sizeof(cl_int)); break; } } } if (numberOfIntParametersToTry == (long)numberExpected) return 0; return -1; } int test_min_max_samplers(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int error; cl_uint maxSamplers, i; clProgramWrapper program; clKernelWrapper kernel; char *programSrc, samplerLine[1024]; size_t maxParameterSize; cl_event event; cl_int event_status; cl_uint minRequiredSamplers = gIsEmbedded ? 8 : 16; PASSIVE_REQUIRE_IMAGE_SUPPORT(deviceID) /* Get the max value */ error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_SAMPLERS, sizeof(maxSamplers), &maxSamplers, NULL); test_error(error, "Unable to get max sampler count from device"); if (maxSamplers < minRequiredSamplers) { log_error( "ERROR: Reported max sampler count is less than required! (%d)\n", (int)maxSamplers); return -1; } log_info("Reported max %d samplers.\n", maxSamplers); error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_PARAMETER_SIZE, sizeof(maxParameterSize), &maxParameterSize, NULL); test_error(error, "Unable to get max parameter size from device"); // Subtract the size of the result maxParameterSize -= 2 * sizeof(cl_mem); // Calculate the number we can use if (maxParameterSize / sizeof(cl_sampler) < maxSamplers) { log_info("WARNING: Max parameter size of %d bytes limits test to %d " "max sampler arguments.\n", (int)maxParameterSize, (int)(maxParameterSize / sizeof(cl_sampler))); maxSamplers = (unsigned int)(maxParameterSize / sizeof(cl_sampler)); } /* Create a kernel to test with */ programSrc = (char *)malloc( (strlen(sample_sampler_kernel_pattern[1]) + 8) * (maxSamplers) + strlen(sample_sampler_kernel_pattern[0]) + strlen(sample_sampler_kernel_pattern[2]) + (strlen(sample_sampler_kernel_pattern[3]) + 8) * maxSamplers + strlen(sample_sampler_kernel_pattern[4])); strcpy(programSrc, sample_sampler_kernel_pattern[0]); for (i = 0; i < maxSamplers; i++) { sprintf(samplerLine, sample_sampler_kernel_pattern[1], i); strcat(programSrc, samplerLine); } strcat(programSrc, sample_sampler_kernel_pattern[2]); for (i = 0; i < maxSamplers; i++) { sprintf(samplerLine, sample_sampler_kernel_pattern[3], i); strcat(programSrc, samplerLine); } strcat(programSrc, sample_sampler_kernel_pattern[4]); error = create_single_kernel_helper(context, &program, &kernel, 1, (const char **)&programSrc, "sample_test"); test_error(error, "Failed to create the program and kernel."); // We have to set up some fake parameters so it'll work clSamplerWrapper *samplers = new clSamplerWrapper[maxSamplers]; cl_image_format format = { CL_RGBA, CL_SIGNED_INT8 }; clMemWrapper image = create_image_2d(context, CL_MEM_READ_WRITE, &format, 16, 16, 0, NULL, &error); test_error(error, "Unable to create a test image"); clMemWrapper stream = clCreateBuffer(context, CL_MEM_READ_WRITE, 16, NULL, &error); test_error(error, "Unable to create test buffer"); error = clSetKernelArg(kernel, 0, sizeof(cl_mem), &image); error |= clSetKernelArg(kernel, 1, sizeof(cl_mem), &stream); test_error(error, "Unable to set kernel arguments"); for (i = 0; i < maxSamplers; i++) { samplers[i] = clCreateSampler(context, CL_FALSE, CL_ADDRESS_NONE, CL_FILTER_NEAREST, &error); test_error(error, "Unable to create sampler"); error = clSetKernelArg(kernel, 2 + i, sizeof(cl_sampler), &samplers[i]); test_error(error, "Unable to set sampler argument"); } size_t globalDim[3] = { 1, 1, 1 }, localDim[3] = { 1, 1, 1 }; error = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, globalDim, localDim, 0, NULL, &event); test_error( error, "clEnqueueNDRangeKernel failed with maximum number of samplers."); // Verify that the event does not return an error from the execution error = clWaitForEvents(1, &event); test_error(error, "clWaitForEvent failed"); error = clGetEventInfo(event, CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof(event_status), &event_status, NULL); test_error(error, "clGetEventInfo for CL_EVENT_COMMAND_EXECUTION_STATUS failed"); clReleaseEvent(event); if (event_status < 0) test_error(error, "Kernel execution event returned error"); free(programSrc); delete[] samplers; return 0; } #define PASSING_FRACTION 4 int test_min_max_constant_buffer_size(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int error; clProgramWrapper program; clKernelWrapper kernel; size_t threads[1], localThreads[1]; cl_int *constantData, *resultData; cl_ulong maxSize, stepSize, currentSize, maxGlobalSize, maxAllocSize; int i; cl_event event; cl_int event_status; MTdata d; /* Verify our test buffer won't be bigger than allowed */ error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_CONSTANT_BUFFER_SIZE, sizeof(maxSize), &maxSize, 0); test_error(error, "Unable to get max constant buffer size"); if ((0 == gIsEmbedded && maxSize < 64L * 1024L) || maxSize < 1L * 1024L) { log_error("ERROR: Reported max constant buffer size less than required " "by OpenCL 1.0 (reported %d KB)\n", (int)(maxSize / 1024L)); return -1; } log_info("Reported max constant buffer size of %lld bytes.\n", maxSize); // Limit test buffer size to 1/8 of CL_DEVICE_GLOBAL_MEM_SIZE error = clGetDeviceInfo(deviceID, CL_DEVICE_GLOBAL_MEM_SIZE, sizeof(maxGlobalSize), &maxGlobalSize, 0); test_error(error, "Unable to get CL_DEVICE_GLOBAL_MEM_SIZE"); if (maxSize > maxGlobalSize / 8) maxSize = maxGlobalSize / 8; error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_MEM_ALLOC_SIZE, sizeof(maxAllocSize), &maxAllocSize, 0); test_error(error, "Unable to get CL_DEVICE_MAX_MEM_ALLOC_SIZE "); if (maxSize > maxAllocSize) maxSize = maxAllocSize; /* Create a kernel to test with */ if (create_single_kernel_helper(context, &program, &kernel, 1, sample_const_arg_kernel, "sample_test") != 0) { return -1; } /* Try the returned max size and decrease it until we get one that works. */ stepSize = maxSize / 16; currentSize = maxSize; int allocPassed = 0; d = init_genrand(gRandomSeed); while (!allocPassed && currentSize >= maxSize / PASSING_FRACTION) { log_info("Attempting to allocate constant buffer of size %lld bytes\n", maxSize); /* Create some I/O streams */ size_t sizeToAllocate = ((size_t)currentSize / sizeof(cl_int)) * sizeof(cl_int); size_t numberOfInts = sizeToAllocate / sizeof(cl_int); constantData = (cl_int *)malloc(sizeToAllocate); if (constantData == NULL) { log_error("Failed to allocate memory for constantData!\n"); free_mtdata(d); return EXIT_FAILURE; } for (i = 0; i < (int)(numberOfInts); i++) constantData[i] = (int)genrand_int32(d); clMemWrapper streams[3]; streams[0] = clCreateBuffer(context, CL_MEM_COPY_HOST_PTR, sizeToAllocate, constantData, &error); test_error(error, "Creating test array failed"); streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeToAllocate, NULL, &error); test_error(error, "Creating test array failed"); /* Set the arguments */ error = clSetKernelArg(kernel, 0, sizeof(streams[0]), &streams[0]); test_error(error, "Unable to set indexed kernel arguments"); error = clSetKernelArg(kernel, 1, sizeof(streams[1]), &streams[1]); test_error(error, "Unable to set indexed kernel arguments"); /* Test running the kernel and verifying it */ threads[0] = numberOfInts; localThreads[0] = 1; log_info("Filling constant buffer with %d cl_ints (%d bytes).\n", (int)threads[0], (int)(threads[0] * sizeof(cl_int))); error = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, threads, localThreads, 0, NULL, &event); /* If we failed due to a resource issue, reduce the size and try again. */ if ((error == CL_OUT_OF_RESOURCES) || (error == CL_MEM_OBJECT_ALLOCATION_FAILURE) || (error == CL_OUT_OF_HOST_MEMORY)) { log_info("Kernel enqueue failed at size %lld, trying at a reduced " "size.\n", currentSize); currentSize -= stepSize; free(constantData); continue; } test_error( error, "clEnqueueNDRangeKernel with maximum constant buffer size failed."); // Verify that the event does not return an error from the execution error = clWaitForEvents(1, &event); test_error(error, "clWaitForEvent failed"); error = clGetEventInfo(event, CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof(event_status), &event_status, NULL); test_error( error, "clGetEventInfo for CL_EVENT_COMMAND_EXECUTION_STATUS failed"); clReleaseEvent(event); if (event_status < 0) { if ((event_status == CL_OUT_OF_RESOURCES) || (event_status == CL_MEM_OBJECT_ALLOCATION_FAILURE) || (event_status == CL_OUT_OF_HOST_MEMORY)) { log_info("Kernel event indicates failure at size %lld, trying " "at a reduced size.\n", currentSize); currentSize -= stepSize; free(constantData); continue; } else { test_error(error, "Kernel execution event returned error"); } } /* Otherwise we did not fail due to resource issues. */ allocPassed = 1; resultData = (cl_int *)malloc(sizeToAllocate); if (resultData == NULL) { log_error("Failed to allocate memory for resultData!\n"); free(constantData); free_mtdata(d); return EXIT_FAILURE; } error = clEnqueueReadBuffer(queue, streams[1], CL_TRUE, 0, sizeToAllocate, resultData, 0, NULL, NULL); test_error(error, "clEnqueueReadBuffer failed"); for (i = 0; i < (int)(numberOfInts); i++) if (constantData[i] != resultData[i]) { log_error("Data failed to verify: constantData[%d]=%d != " "resultData[%d]=%d\n", i, constantData[i], i, resultData[i]); free(constantData); free(resultData); free_mtdata(d); d = NULL; return -1; } free(constantData); free(resultData); } free_mtdata(d); d = NULL; if (allocPassed) { if (currentSize < maxSize / PASSING_FRACTION) { log_error("Failed to allocate at least 1/8 of the reported " "constant size.\n"); return -1; } else if (currentSize != maxSize) { log_info("Passed at reduced size. (%lld of %lld bytes)\n", currentSize, maxSize); return 0; } return 0; } return -1; } int test_min_max_constant_args(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int error; clProgramWrapper program; clKernelWrapper kernel; clMemWrapper *streams; size_t threads[1], localThreads[1]; cl_uint i, maxArgs; cl_ulong maxSize; cl_ulong maxParameterSize; size_t individualBufferSize; char *programSrc, *constArgs, *str2; char str[512]; const char *ptr; cl_event event; cl_int event_status; /* Verify our test buffer won't be bigger than allowed */ error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_CONSTANT_ARGS, sizeof(maxArgs), &maxArgs, 0); test_error(error, "Unable to get max constant arg count"); error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_PARAMETER_SIZE, sizeof(maxParameterSize), &maxParameterSize, NULL); test_error(error, "Unable to get max parameter size from device"); // Subtract the size of the result maxParameterSize -= sizeof(cl_mem); // Calculate the number we can use if (maxParameterSize / sizeof(cl_mem) < maxArgs) { log_info("WARNING: Max parameter size of %d bytes limits test to %d " "max image arguments.\n", (int)maxParameterSize, (int)(maxParameterSize / sizeof(cl_mem))); maxArgs = (unsigned int)(maxParameterSize / sizeof(cl_mem)); } if (maxArgs < (gIsEmbedded ? 4 : 8)) { log_error("ERROR: Reported max constant arg count less than required " "by OpenCL 1.0 (reported %d)\n", (int)maxArgs); return -1; } error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_CONSTANT_BUFFER_SIZE, sizeof(maxSize), &maxSize, 0); test_error(error, "Unable to get max constant buffer size"); individualBufferSize = (maxSize / 2) / maxArgs; log_info( "Reported max constant arg count of %u and max constant buffer " "size of %llu. Test will attempt to allocate half of that, or %llu " "buffers of size %zu.\n", maxArgs, maxSize, maxArgs, individualBufferSize); str2 = (char *)malloc(sizeof(char) * 32 * (maxArgs + 2)); constArgs = (char *)malloc(sizeof(char) * 32 * (maxArgs + 2)); programSrc = (char *)malloc(sizeof(char) * 32 * 2 * (maxArgs + 2) + 1024); /* Create a test program */ constArgs[0] = 0; str2[0] = 0; for (i = 0; i < maxArgs - 1; i++) { sprintf(str, ", __constant int *src%d", (int)(i + 2)); strcat(constArgs, str); sprintf(str2 + strlen(str2), "\tdst[tid] += src%d[tid];\n", (int)(i + 2)); if (strlen(str2) > (sizeof(char) * 32 * (maxArgs + 2) - 32) || strlen(constArgs) > (sizeof(char) * 32 * (maxArgs + 2) - 32)) { log_info("Limiting number of arguments tested to %d due to test " "program allocation size.\n", i); break; } } sprintf(programSrc, sample_const_max_arg_kernel_pattern, constArgs, str2); /* Create a kernel to test with */ ptr = programSrc; if (create_single_kernel_helper(context, &program, &kernel, 1, &ptr, "sample_test") != 0) { return -1; } /* Create some I/O streams */ streams = new clMemWrapper[maxArgs + 1]; for (i = 0; i < maxArgs + 1; i++) { streams[i] = clCreateBuffer(context, CL_MEM_READ_WRITE, individualBufferSize, NULL, &error); test_error(error, "Creating test array failed"); } /* Set the arguments */ for (i = 0; i < maxArgs + 1; i++) { error = clSetKernelArg(kernel, i, sizeof(streams[i]), &streams[i]); test_error(error, "Unable to set kernel argument"); } /* Test running the kernel and verifying it */ threads[0] = (size_t)10; while (threads[0] * sizeof(cl_int) > individualBufferSize) threads[0]--; error = get_max_common_work_group_size(context, kernel, threads[0], &localThreads[0]); test_error(error, "Unable to get work group size to use"); error = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, threads, localThreads, 0, NULL, &event); test_error(error, "clEnqueueNDRangeKernel failed"); // Verify that the event does not return an error from the execution error = clWaitForEvents(1, &event); test_error(error, "clWaitForEvent failed"); error = clGetEventInfo(event, CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof(event_status), &event_status, NULL); test_error(error, "clGetEventInfo for CL_EVENT_COMMAND_EXECUTION_STATUS failed"); clReleaseEvent(event); if (event_status < 0) test_error(error, "Kernel execution event returned error"); error = clFinish(queue); test_error(error, "clFinish failed."); delete[] streams; free(str2); free(constArgs); free(programSrc); return 0; } int test_min_max_compute_units(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int error; cl_uint value; error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_COMPUTE_UNITS, sizeof(value), &value, 0); test_error(error, "Unable to get compute unit count"); if (value < 1) { log_error("ERROR: Reported compute unit count less than required by " "OpenCL 1.0 (reported %d)\n", (int)value); return -1; } log_info("Reported %d max compute units.\n", value); return 0; } int test_min_max_address_bits(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int error; cl_uint value; error = clGetDeviceInfo(deviceID, CL_DEVICE_ADDRESS_BITS, sizeof(value), &value, 0); test_error(error, "Unable to get address bit count"); if (value != 32 && value != 64) { log_error("ERROR: Reported address bit count not valid by OpenCL 1.0 " "(reported %d)\n", (int)value); return -1; } log_info("Reported %d device address bits.\n", value); return 0; } int test_min_max_single_fp_config(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int error; cl_device_fp_config value; char profile[128] = ""; error = clGetDeviceInfo(deviceID, CL_DEVICE_SINGLE_FP_CONFIG, sizeof(value), &value, 0); test_error(error, "Unable to get device single fp config"); // Check to see if we are an embedded profile device if ((error = clGetDeviceInfo(deviceID, CL_DEVICE_PROFILE, sizeof(profile), profile, NULL))) { log_error("FAILURE: Unable to get CL_DEVICE_PROFILE: error %d\n", error); return error; } if (0 == strcmp(profile, "EMBEDDED_PROFILE")) { // embedded device if (0 == (value & (CL_FP_ROUND_TO_NEAREST | CL_FP_ROUND_TO_ZERO))) { log_error("FAILURE: embedded device supports neither " "CL_FP_ROUND_TO_NEAREST or CL_FP_ROUND_TO_ZERO\n"); return -1; } } else { // Full profile if ((value & (CL_FP_ROUND_TO_NEAREST | CL_FP_INF_NAN)) != (CL_FP_ROUND_TO_NEAREST | CL_FP_INF_NAN)) { log_error("ERROR: Reported single fp config doesn't meet minimum " "set by OpenCL 1.0 (reported 0x%08x)\n", (int)value); return -1; } } return 0; } int test_min_max_double_fp_config(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int error; cl_device_fp_config value; error = clGetDeviceInfo(deviceID, CL_DEVICE_DOUBLE_FP_CONFIG, sizeof(value), &value, 0); test_error(error, "Unable to get device double fp config"); if (value == 0) return 0; if ((value & (CL_FP_FMA | CL_FP_ROUND_TO_NEAREST | CL_FP_ROUND_TO_ZERO | CL_FP_ROUND_TO_INF | CL_FP_INF_NAN | CL_FP_DENORM)) != (CL_FP_FMA | CL_FP_ROUND_TO_NEAREST | CL_FP_ROUND_TO_ZERO | CL_FP_ROUND_TO_INF | CL_FP_INF_NAN | CL_FP_DENORM)) { log_error("ERROR: Reported double fp config doesn't meet minimum set " "by OpenCL 1.0 (reported 0x%08x)\n", (int)value); return -1; } return 0; } int test_min_max_local_mem_size(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int error; clProgramWrapper program; clKernelWrapper kernel; clMemWrapper streams[3]; size_t threads[1], localThreads[1]; cl_int *localData, *resultData; cl_ulong maxSize, kernelLocalUsage, min_max_local_mem_size; Version device_version; int i; int err = 0; MTdata d; /* Verify our test buffer won't be bigger than allowed */ error = clGetDeviceInfo(deviceID, CL_DEVICE_LOCAL_MEM_SIZE, sizeof(maxSize), &maxSize, 0); test_error(error, "Unable to get max local buffer size"); try { device_version = get_device_cl_version(deviceID); } catch (const std::runtime_error &e) { log_error("%s", e.what()); return -1; } if (!gIsEmbedded) { if (device_version == Version(1, 0)) min_max_local_mem_size = 16L * 1024L; else min_max_local_mem_size = 32L * 1024L; } else { min_max_local_mem_size = 1L * 1024L; } if (maxSize < min_max_local_mem_size) { const std::string version_as_string = device_version.to_string(); log_error("ERROR: Reported local mem size less than required by OpenCL " "%s (reported %d KB)\n", version_as_string.c_str(), (int)(maxSize / 1024L)); return -1; } log_info("Reported max local buffer size for device: %lld bytes.\n", maxSize); /* Create a kernel to test with */ if (create_single_kernel_helper(context, &program, &kernel, 1, sample_local_arg_kernel, "sample_test") != 0) { return -1; } error = clGetKernelWorkGroupInfo(kernel, deviceID, CL_KERNEL_LOCAL_MEM_SIZE, sizeof(kernelLocalUsage), &kernelLocalUsage, NULL); test_error(error, "clGetKernelWorkGroupInfo for CL_KERNEL_LOCAL_MEM_SIZE failed"); log_info("Reported local buffer usage for kernel " "(CL_KERNEL_LOCAL_MEM_SIZE): %lld bytes.\n", kernelLocalUsage); /* Create some I/O streams */ size_t sizeToAllocate = ((size_t)(maxSize - kernelLocalUsage) / sizeof(cl_int)) * sizeof(cl_int); size_t numberOfInts = sizeToAllocate / sizeof(cl_int); log_info("Attempting to use %zu bytes of local memory.\n", sizeToAllocate); localData = (cl_int *)malloc(sizeToAllocate); d = init_genrand(gRandomSeed); for (i = 0; i < (int)(numberOfInts); i++) localData[i] = (int)genrand_int32(d); free_mtdata(d); d = NULL; streams[0] = clCreateBuffer(context, CL_MEM_COPY_HOST_PTR, sizeToAllocate, localData, &error); test_error(error, "Creating test array failed"); streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeToAllocate, NULL, &error); test_error(error, "Creating test array failed"); /* Set the arguments */ error = clSetKernelArg(kernel, 0, sizeToAllocate, NULL); test_error(error, "Unable to set indexed kernel arguments"); error = clSetKernelArg(kernel, 1, sizeof(streams[0]), &streams[0]); test_error(error, "Unable to set indexed kernel arguments"); error = clSetKernelArg(kernel, 2, sizeof(streams[1]), &streams[1]); test_error(error, "Unable to set indexed kernel arguments"); /* Test running the kernel and verifying it */ threads[0] = numberOfInts; localThreads[0] = 1; log_info("Creating local buffer with %zu cl_ints (%zu bytes).\n", numberOfInts, sizeToAllocate); cl_event evt; cl_int evt_err; error = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, threads, localThreads, 0, NULL, &evt); test_error(error, "clEnqueueNDRangeKernel failed"); error = clFinish(queue); test_error(error, "clFinish failed"); error = clGetEventInfo(evt, CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof evt_err, &evt_err, NULL); test_error(error, "clGetEventInfo with maximum local buffer size failed."); if (evt_err != CL_COMPLETE) { print_error(evt_err, "Kernel event returned error"); clReleaseEvent(evt); return -1; } resultData = (cl_int *)malloc(sizeToAllocate); error = clEnqueueReadBuffer(queue, streams[1], CL_TRUE, 0, sizeToAllocate, resultData, 0, NULL, NULL); test_error(error, "clEnqueueReadBuffer failed"); for (i = 0; i < (int)(numberOfInts); i++) if (localData[i] != resultData[i]) { clReleaseEvent(evt); free(localData); free(resultData); log_error("Results failed to verify.\n"); return -1; } clReleaseEvent(evt); free(localData); free(resultData); return err; } int test_min_max_kernel_preferred_work_group_size_multiple( cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int err; clProgramWrapper program; clKernelWrapper kernel; size_t max_local_workgroup_size[3]; size_t max_workgroup_size = 0, preferred_workgroup_size = 0; err = create_single_kernel_helper(context, &program, &kernel, 1, sample_local_arg_kernel, "sample_test"); test_error(err, "Failed to build kernel/program."); err = clGetKernelWorkGroupInfo(kernel, deviceID, CL_KERNEL_WORK_GROUP_SIZE, sizeof(max_workgroup_size), &max_workgroup_size, NULL); test_error(err, "clGetKernelWorkgroupInfo failed."); err = clGetKernelWorkGroupInfo( kernel, deviceID, CL_KERNEL_PREFERRED_WORK_GROUP_SIZE_MULTIPLE, sizeof(preferred_workgroup_size), &preferred_workgroup_size, NULL); test_error(err, "clGetKernelWorkgroupInfo failed."); err = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_WORK_ITEM_SIZES, sizeof(max_local_workgroup_size), max_local_workgroup_size, NULL); test_error(err, "clGetDeviceInfo failed for CL_DEVICE_MAX_WORK_ITEM_SIZES"); // Since the preferred size is only a performance hint, we can only really // check that we get a sane value back log_info("size: %ld preferred: %ld max: %ld\n", max_workgroup_size, preferred_workgroup_size, max_local_workgroup_size[0]); if (preferred_workgroup_size > max_workgroup_size) { log_error("ERROR: Reported preferred workgroup multiple larger than " "max workgroup size (preferred %ld, max %ld)\n", preferred_workgroup_size, max_workgroup_size); return -1; } return 0; } int test_min_max_execution_capabilities(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int error; cl_device_exec_capabilities value; error = clGetDeviceInfo(deviceID, CL_DEVICE_EXECUTION_CAPABILITIES, sizeof(value), &value, 0); test_error(error, "Unable to get execution capabilities"); if ((value & CL_EXEC_KERNEL) != CL_EXEC_KERNEL) { log_error("ERROR: Reported execution capabilities less than required " "by OpenCL 1.0 (reported 0x%08x)\n", (int)value); return -1; } return 0; } int test_min_max_queue_properties(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { int error; cl_command_queue_properties value; error = clGetDeviceInfo(deviceID, CL_DEVICE_QUEUE_ON_HOST_PROPERTIES, sizeof(value), &value, 0); test_error(error, "Unable to get queue properties"); if ((value & CL_QUEUE_PROFILING_ENABLE) != CL_QUEUE_PROFILING_ENABLE) { log_error("ERROR: Reported queue properties less than required by " "OpenCL 1.0 (reported 0x%08x)\n", (int)value); return -1; } return 0; } int test_min_max_device_version(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { // Query for the device version. Version device_cl_version = get_device_cl_version(deviceID); log_info("Returned version %s.\n", device_cl_version.to_string().c_str()); // Make sure 2.x devices support required extensions for 2.x // note: these extensions are **not** required for devices // supporting OpenCL-3.0 const char *requiredExtensions2x[] = { "cl_khr_3d_image_writes", "cl_khr_image2d_from_buffer", "cl_khr_depth_images", }; // Make sure 1.1 devices support required extensions for 1.1 const char *requiredExtensions11[] = { "cl_khr_global_int32_base_atomics", "cl_khr_global_int32_extended_atomics", "cl_khr_local_int32_base_atomics", "cl_khr_local_int32_extended_atomics", "cl_khr_byte_addressable_store", }; if (device_cl_version >= Version(1, 1)) { log_info("Checking for required extensions for OpenCL 1.1 and later " "devices...\n"); for (size_t i = 0; i < ARRAY_SIZE(requiredExtensions11); i++) { if (!is_extension_available(deviceID, requiredExtensions11[i])) { log_error("ERROR: Required extension for 1.1 and greater " "devices is not in extension string: %s\n", requiredExtensions11[i]); return -1; } else log_info("\t%s\n", requiredExtensions11[i]); } if (device_cl_version >= Version(1, 2)) { log_info("Checking for required extensions for OpenCL 1.2 and " "later devices...\n"); // The only required extension for an OpenCL-1.2 device is // cl_khr_fp64 and it is only required if double precision is // supported. cl_device_fp_config doubles_supported; cl_int error = clGetDeviceInfo(deviceID, CL_DEVICE_DOUBLE_FP_CONFIG, sizeof(doubles_supported), &doubles_supported, 0); test_error(error, "Unable to get device double fp config"); if (doubles_supported) { if (!is_extension_available(deviceID, "cl_khr_fp64")) { log_error( "ERROR: Required extension for 1.2 and greater devices " "is not in extension string: cl_khr_fp64\n"); } else { log_info("\t%s\n", "cl_khr_fp64"); } } } if (device_cl_version >= Version(2, 0) && device_cl_version < Version(3, 0)) { log_info("Checking for required extensions for OpenCL 2.0, 2.1 and " "2.2 devices...\n"); for (size_t i = 0; i < ARRAY_SIZE(requiredExtensions2x); i++) { if (!is_extension_available(deviceID, requiredExtensions2x[i])) { log_error("ERROR: Required extension for 2.0, 2.1 and 2.2 " "devices is not in extension string: %s\n", requiredExtensions2x[i]); return -1; } else { log_info("\t%s\n", requiredExtensions2x[i]); } } } } else log_info("WARNING: skipping required extension test -- OpenCL 1.0 " "device.\n"); return 0; } int test_min_max_language_version(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements) { cl_int error; cl_char buffer[4098]; size_t length; // Device version should fit the regex "OpenCL [0-9]+\.[0-9]+ *.*" error = clGetDeviceInfo(deviceID, CL_DEVICE_OPENCL_C_VERSION, sizeof(buffer), buffer, &length); test_error(error, "Unable to get device opencl c version string"); if (memcmp(buffer, "OpenCL C ", strlen("OpenCL C ")) != 0) { log_error("ERROR: Initial part of device language version string does " "not match required format! (returned: \"%s\")\n", (char *)buffer); return -1; } log_info("Returned version \"%s\".\n", buffer); char *p1 = (char *)buffer + strlen("OpenCL C "); while (*p1 == ' ') p1++; char *p2 = p1; if (!isdigit(*p2)) { log_error("ERROR: Major revision number must follow space behind " "OpenCL C! (returned %s)\n", (char *)buffer); return -1; } while (isdigit(*p2)) p2++; if (*p2 != '.') { log_error("ERROR: Version number must contain a decimal point! " "(returned: %s)\n", (char *)buffer); return -1; } char *p3 = p2 + 1; if (!isdigit(*p3)) { log_error("ERROR: Minor revision number is missing or does not abut " "the decimal point! (returned %s)\n", (char *)buffer); return -1; } while (isdigit(*p3)) p3++; if (*p3 != ' ') { log_error("ERROR: A space must appear after the minor version! " "(returned: %s)\n", (char *)buffer); return -1; } *p2 = ' '; // Put in a space for atoi below. p2++; int major = atoi(p1); int minor = atoi(p2); int minor_revision = 2; if (major * 10 + minor < 10 + minor_revision) { // If the language version did not match, check to see if // OPENCL_1_0_DEVICE is set. if (getenv("OPENCL_1_0_DEVICE")) { log_info("WARNING: This test was run with OPENCL_1_0_DEVICE " "defined! This is not a OpenCL 1.1 or OpenCL 1.2 " "compatible device!!!\n"); } else if (getenv("OPENCL_1_1_DEVICE")) { log_info( "WARNING: This test was run with OPENCL_1_1_DEVICE defined! " "This is not a OpenCL 1.2 compatible device!!!\n"); } else { log_error("ERROR: OpenCL device language version returned is less " "than 1.%d! (Returned: %s)\n", minor_revision, (char *)buffer); return -1; } } // Sanity checks on the returned values if (length != (strlen((char *)buffer) + 1)) { log_error("ERROR: Returned length of version string does not match " "actual length (actual: %d, returned: %d)\n", (int)strlen((char *)buffer), (int)length); return -1; } return 0; }