// // 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 "common.h" const char *shared_sub_buffers_test_kernel[] = { "typedef struct Node {\n" " int global_id;\n" " int position_in_list;\n" " __global struct Node* pNext;\n" "} Node;\n" // create linked lists that use nodes from 2 different buffers "__global Node* allocate_node(__global Node* pNodes1, __global Node* pNodes2, volatile __global int* allocation_index, size_t i)\n" "{\n" // mix things up, adjacent work items will allocate from different buffers " if(i & 0x1)\n" " return &pNodes1[atomic_inc(allocation_index)];\n" " else\n" " return &pNodes2[atomic_inc(allocation_index)];\n" "}\n" // The allocation_index parameter must be initialized on the host to N work-items // The first N nodes in pNodes will be the heads of the lists. // This tests passing 4 different sub-buffers that come from two parent buffers. // Note that we have arguments that appear to be unused, but they are required so that system knows to get all the sub-buffers on to the device "__kernel void create_linked_lists(__global Node* pNodes_sub1, __global Node* pNodes2_sub1, __global Node* pNodes_sub2, __global Node* pNodes2_sub2, volatile __global int* allocation_index, int list_length) \n" "{\n" " size_t i = get_global_id(0);\n" " __global Node *pNode = &pNodes_sub1[i];\n" " pNode->global_id = i;\n" " pNode->position_in_list = 0;\n" " __global Node *pNew;\n" " for(int j=1; j < list_length; j++) {\n" " pNew = allocate_node(pNodes_sub1, pNodes2_sub1, allocation_index, i);\n" " pNew->global_id = i;\n" " pNew->position_in_list = j;\n" " pNode->pNext = pNew; // link new node onto end of list\n" " pNode = pNew; // move to end of list\n" " }\n" "}\n" // Note that we have arguments that appear to be unused, but they are required so that system knows to get all the sub-buffers on to the device "__kernel void verify_linked_lists(__global Node* pNodes_sub1, __global Node* pNodes2_sub1, __global Node* pNodes_sub2, __global Node* pNodes2_sub2, volatile __global uint* num_correct, int list_length)\n" "{\n" " size_t i = get_global_id(0);\n" " __global Node *pNode = &pNodes_sub1[i];\n" " for(int j=0; j < list_length; j++) {\n" " if( pNode->global_id == i && pNode->position_in_list == j)\n" " atomic_inc(num_correct);\n" " else \n" " break;\n" " pNode = pNode->pNext;\n" " }\n" "}\n" }; // Creates linked list using host code. cl_int create_linked_lists_on_host_sb(cl_command_queue cmdq, cl_mem nodes, cl_mem nodes2, cl_int ListLength, size_t numLists ) { cl_int error = CL_SUCCESS; log_info("SVM: creating linked list on host "); Node *pNodes = (Node*) clEnqueueMapBuffer(cmdq, nodes, CL_TRUE, CL_MAP_READ | CL_MAP_WRITE, 0, sizeof(Node)*ListLength*numLists, 0, NULL,NULL, &error); test_error2(error, pNodes, "clEnqueueMapBuffer failed"); Node *pNodes2 = (Node*) clEnqueueMapBuffer(cmdq, nodes2, CL_TRUE, CL_MAP_READ | CL_MAP_WRITE, 0, sizeof(Node)*ListLength*numLists, 0, NULL,NULL, &error); test_error2(error, pNodes2, "clEnqueueMapBuffer failed"); create_linked_lists(pNodes, numLists, ListLength); error = clEnqueueUnmapMemObject(cmdq, nodes, pNodes, 0,NULL,NULL); test_error(error, "clEnqueueUnmapMemObject failed"); error = clEnqueueUnmapMemObject(cmdq, nodes2, pNodes2, 0,NULL,NULL); test_error(error, "clEnqueueUnmapMemObject failed"); error = clFinish(cmdq); test_error(error, "clFinish failed"); return error; } // Verify correctness of the linked list using host code. cl_int verify_linked_lists_on_host_sb(int ci, cl_command_queue cmdq, cl_mem nodes, cl_mem nodes2, cl_int ListLength, size_t numLists ) { cl_int error = CL_SUCCESS; //log_info(" and verifying on host "); Node *pNodes = (Node*) clEnqueueMapBuffer(cmdq, nodes, CL_TRUE, CL_MAP_READ | CL_MAP_WRITE, 0, sizeof(Node)*ListLength * numLists, 0, NULL,NULL, &error); test_error2(error, pNodes, "clEnqueueMapBuffer failed"); Node *pNodes2 = (Node*) clEnqueueMapBuffer(cmdq, nodes2, CL_TRUE, CL_MAP_READ | CL_MAP_WRITE, 0, sizeof(Node)*ListLength * numLists, 0, NULL,NULL, &error); test_error2(error, pNodes, "clEnqueueMapBuffer failed"); error = verify_linked_lists(pNodes, numLists, ListLength); if(error) return -1; error = clEnqueueUnmapMemObject(cmdq, nodes, pNodes, 0,NULL,NULL); test_error(error, "clEnqueueUnmapMemObject failed"); error = clEnqueueUnmapMemObject(cmdq, nodes2, pNodes2, 0,NULL,NULL); test_error(error, "clEnqueueUnmapMemObject failed"); error = clFinish(cmdq); test_error(error, "clFinish failed"); return error; } // This tests that shared sub-buffers can be created and that they inherit the flags from the parent buffer when no flags are specified. // This tests that passing only the sub-buffers to a kernel works. // The test is derived from the cross-buffer pointers test which // tests that shared buffers are able to contain pointers that point to other shared buffers. // This tests that all devices and the host share a common address space; using only the coarse-grain features. // This is done by creating a linked list on a device and then verifying the correctness of the list // on another device or the host. // The linked list nodes are allocated from two different buffers this is done to ensure that cross buffer pointers work correctly. // This basic test is performed for all combinations of devices and the host. int test_svm_shared_sub_buffers(cl_device_id deviceID, cl_context context2, cl_command_queue queue, int num_elements) { clContextWrapper context = NULL; clProgramWrapper program = NULL; cl_uint num_devices = 0; cl_int error = CL_SUCCESS; clCommandQueueWrapper queues[MAXQ]; error = create_cl_objects(deviceID, &shared_sub_buffers_test_kernel[0], &context, &program, &queues[0], &num_devices, CL_DEVICE_SVM_COARSE_GRAIN_BUFFER); if(error) return -1; size_t numLists = num_elements; if(numLists & 0x1) numLists++; // force even size, so we can easily create two sub-buffers of same size. cl_int ListLength = 32; clKernelWrapper kernel_create_lists = clCreateKernel(program, "create_linked_lists", &error); test_error(error, "clCreateKernel failed"); clKernelWrapper kernel_verify_lists = clCreateKernel(program, "verify_linked_lists", &error); test_error(error, "clCreateKernel failed"); size_t nodes_bufsize = sizeof(Node)*ListLength*numLists; Node* pNodes = (Node*) clSVMAlloc(context, CL_MEM_READ_WRITE, nodes_bufsize, 0); Node* pNodes2 = (Node*) clSVMAlloc(context, CL_MEM_READ_WRITE, nodes_bufsize, 0); { // this buffer holds some of the linked list nodes. clMemWrapper nodes = clCreateBuffer(context, CL_MEM_USE_HOST_PTR, nodes_bufsize, pNodes, &error); test_error(error, "clCreateBuffer failed."); cl_buffer_region r; r.origin = 0; r.size = nodes_bufsize / 2; // this should inherit the flag settings from nodes buffer clMemWrapper nodes_sb1 = clCreateSubBuffer(nodes, 0, CL_BUFFER_CREATE_TYPE_REGION, (void*)&r, &error); test_error(error, "clCreateSubBuffer"); r.origin = nodes_bufsize / 2; clMemWrapper nodes_sb2 = clCreateSubBuffer(nodes, 0, CL_BUFFER_CREATE_TYPE_REGION, (void*)&r, &error); test_error(error, "clCreateSubBuffer"); // this buffer holds some of the linked list nodes. clMemWrapper nodes2 = clCreateBuffer(context, CL_MEM_USE_HOST_PTR, sizeof(Node)*ListLength*numLists, pNodes2, &error); test_error(error, "clCreateBuffer failed."); r.origin = 0; r.size = nodes_bufsize / 2; // this should inherit the flag settings from nodes buffer clMemWrapper nodes2_sb1 = clCreateSubBuffer(nodes2, 0, CL_BUFFER_CREATE_TYPE_REGION, (void*)&r, &error); test_error(error, "clCreateSubBuffer"); r.origin = nodes_bufsize / 2; clMemWrapper nodes2_sb2 = clCreateSubBuffer(nodes2, 0, CL_BUFFER_CREATE_TYPE_REGION,(void*)&r, &error); test_error(error, "clCreateSubBuffer"); // this buffer holds the index into the nodes buffer that is used for node allocation clMemWrapper allocator = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeof(size_t), NULL, &error); test_error(error, "clCreateBuffer failed."); // this buffer holds the count of correct nodes which is computed by the verify kernel. clMemWrapper num_correct = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeof(cl_int), NULL, &error); test_error(error, "clCreateBuffer failed."); error |= clSetKernelArg(kernel_create_lists, 0, sizeof(void*), (void *) &nodes_sb1); error |= clSetKernelArg(kernel_create_lists, 1, sizeof(void*), (void *) &nodes2_sb1); error |= clSetKernelArg(kernel_create_lists, 2, sizeof(void*), (void *) &nodes_sb2); error |= clSetKernelArg(kernel_create_lists, 3, sizeof(void*), (void *) &nodes2_sb2); error |= clSetKernelArg(kernel_create_lists, 4, sizeof(void*), (void *) &allocator); error |= clSetKernelArg(kernel_create_lists, 5, sizeof(cl_int),(void *) &ListLength); error |= clSetKernelArg(kernel_verify_lists, 0, sizeof(void*), (void *) &nodes_sb1); error |= clSetKernelArg(kernel_verify_lists, 1, sizeof(void*), (void *) &nodes2_sb1); error |= clSetKernelArg(kernel_verify_lists, 2, sizeof(void*), (void *) &nodes_sb2); error |= clSetKernelArg(kernel_verify_lists, 3, sizeof(void*), (void *) &nodes2_sb2); error |= clSetKernelArg(kernel_verify_lists, 4, sizeof(void*), (void *) &num_correct); error |= clSetKernelArg(kernel_verify_lists, 5, sizeof(cl_int),(void *) &ListLength); test_error(error, "clSetKernelArg failed"); // Create linked list on one device and verify on another device (or the host). // Do this for all possible combinations of devices and host within the platform. for (int ci=0; ci<(int)num_devices+1; ci++) // ci is CreationIndex, index of device/q to create linked list on { for (int vi=0; vi<(int)num_devices+1; vi++) // vi is VerificationIndex, index of device/q to verify linked list on { if(ci == num_devices) // last device index represents the host, note the num_device+1 above. { error = create_linked_lists_on_host_sb(queues[0], nodes, nodes2, ListLength, numLists); if(error) return -1; } else { error = create_linked_lists_on_device(ci, queues[ci], allocator, kernel_create_lists, numLists); if(error) return -1; } if(vi == num_devices) { error = verify_linked_lists_on_host_sb(vi, queues[0], nodes, nodes2, ListLength, numLists); if(error) return -1; } else { error = verify_linked_lists_on_device(vi, queues[vi], num_correct, kernel_verify_lists, ListLength, numLists); if(error) return -1; } } // inner loop, vi } // outer loop, ci } clSVMFree(context, pNodes2); clSVMFree(context, pNodes); return 0; }