1 //
2 // Copyright (c) 2017 The Khronos Group Inc.
3 //
4 // Licensed under the Apache License, Version 2.0 (the "License");
5 // you may not use this file except in compliance with the License.
6 // You may obtain a copy of the License at
7 //
8 // http://www.apache.org/licenses/LICENSE-2.0
9 //
10 // Unless required by applicable law or agreed to in writing, software
11 // distributed under the License is distributed on an "AS IS" BASIS,
12 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 // See the License for the specific language governing permissions and
14 // limitations under the License.
15 //
16 #include <stdio.h>
17 #include <string.h>
18 #include "harness/testHarness.h"
19 #include "harness/typeWrappers.h"
20
21 #include <vector>
22
23 #include "procs.h"
24 #include "utils.h"
25 #include <time.h>
26
27
28
29
30 #ifdef CL_VERSION_2_0
31 extern int gWimpyMode;
32 static const char enqueue_block_multi_queue[] =
33 NL "#define BLOCK_COMPLETED 0"
34 NL "#define BLOCK_SUBMITTED 1"
35 NL ""
36 NL "kernel void enqueue_block_multi_queue(__global int* res, __global int* buff %s)"
37 NL "{"
38 NL " uint i, n = %d;"
39 NL " clk_event_t block_evt[%d];"
40 NL " queue_t q[] = { %s };"
41 NL " queue_t *queue = q;"
42 NL ""
43 NL " clk_event_t user_evt = create_user_event();"
44 NL " queue_t def_q = get_default_queue();"
45 NL " size_t tid = get_global_id(0);"
46 NL " res[tid] = -1;"
47 NL " __global int* b = buff + tid*n;"
48 NL " for(i=0; i<n; ++i) b[i] = -1;"
49 NL ""
50 NL " ndrange_t ndrange = ndrange_1D(1);"
51 NL " for(i = 0; i < n; ++i)"
52 NL " {"
53 NL " b[i] = BLOCK_SUBMITTED;"
54 NL " int enq_res = enqueue_kernel(queue[i], CLK_ENQUEUE_FLAGS_NO_WAIT, ndrange, 1, &user_evt, &block_evt[i], "
55 NL " ^{"
56 NL " b[i] = BLOCK_COMPLETED;"
57 NL " });"
58 NL " if(enq_res != CLK_SUCCESS) { res[tid] = -2; return; }"
59 NL " }"
60 NL ""
61 NL " // check blocks are not started"
62 NL " for(i = 0; i < n; ++i)"
63 NL " {"
64 NL " if(b[i] != BLOCK_SUBMITTED) { res[tid] = -5; return; }"
65 NL " }"
66 NL ""
67 NL " res[tid] = BLOCK_SUBMITTED;"
68 NL " int enq_res = enqueue_kernel(def_q, CLK_ENQUEUE_FLAGS_NO_WAIT, ndrange, n, block_evt, NULL, "
69 NL " ^{"
70 NL " uint k;"
71 NL " // check blocks are finished"
72 NL " for(k = 0; k < n; ++k)"
73 NL " {"
74 NL " if(b[k] != BLOCK_COMPLETED) { res[tid] = -3; return; }"
75 NL " }"
76 NL " res[tid] = BLOCK_COMPLETED;"
77 NL " });"
78 NL " for(i = 0; i < n; ++i)"
79 NL " {"
80 NL " release_event(block_evt[i]);"
81 NL " }"
82 NL " if(enq_res != CLK_SUCCESS) { res[tid] = -4; return; }"
83 NL ""
84 NL " set_user_event_status(user_evt, CL_COMPLETE);"
85 NL " release_event(user_evt);"
86 NL "}";
87
88
check_kernel_results(cl_int * results,cl_int len)89 static int check_kernel_results(cl_int* results, cl_int len)
90 {
91 for(cl_int i = 0; i < len; ++i)
92 {
93 if(results[i] != 0) return i;
94 }
95 return -1;
96 }
97
test_enqueue_multi_queue(cl_device_id device,cl_context context,cl_command_queue queue,int num_elements)98 int test_enqueue_multi_queue(cl_device_id device, cl_context context, cl_command_queue queue, int num_elements)
99 {
100 cl_uint i;
101 cl_int k, err_ret, res = 0;
102 clCommandQueueWrapper dev_queue;
103 cl_int kernel_results[MAX_GWS] = {0};
104
105 size_t ret_len;
106 cl_uint n, max_queues = 1;
107 cl_uint maxQueueSize = 0;
108 err_ret = clGetDeviceInfo(device, CL_DEVICE_QUEUE_ON_DEVICE_MAX_SIZE, sizeof(maxQueueSize), &maxQueueSize, 0);
109 test_error(err_ret, "clGetDeviceInfo(CL_DEVICE_QUEUE_ON_DEVICE_MAX_SIZE) failed");
110
111 err_ret = clGetDeviceInfo(device, CL_DEVICE_MAX_ON_DEVICE_QUEUES, sizeof(max_queues), &max_queues, &ret_len);
112 test_error(err_ret, "clGetDeviceInfo(CL_DEVICE_MAX_ON_DEVICE_QUEUES) failed");
113
114 size_t max_local_size = 1;
115 err_ret = clGetDeviceInfo(device, CL_DEVICE_MAX_WORK_GROUP_SIZE, sizeof(max_local_size), &max_local_size, &ret_len);
116 test_error(err_ret, "clGetDeviceInfo(CL_DEVICE_MAX_WORK_GROUP_SIZE) failed");
117
118 cl_queue_properties queue_prop_def[] =
119 {
120 CL_QUEUE_PROPERTIES, CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE|CL_QUEUE_ON_DEVICE|CL_QUEUE_ON_DEVICE_DEFAULT,
121 CL_QUEUE_SIZE, maxQueueSize,
122 0
123 };
124
125 dev_queue = clCreateCommandQueueWithProperties(context, device, queue_prop_def, &err_ret);
126 test_error(err_ret, "clCreateCommandQueueWithProperties(CL_QUEUE_DEVICE|CL_QUEUE_DEFAULT) failed");
127
128 if(max_queues > 1)
129 {
130 n = (max_queues > MAX_QUEUES) ? MAX_QUEUES : max_queues-1;
131 clMemWrapper mem, buff, evt;
132 std::vector<clCommandQueueWrapper> queues(n);
133 std::vector<cl_command_queue> q(n);
134 cl_queue_properties queue_prop[] =
135 {
136 CL_QUEUE_PROPERTIES, CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE|CL_QUEUE_ON_DEVICE,
137 CL_QUEUE_SIZE, maxQueueSize,
138 0
139 };
140
141 for(i = 0; i < n; ++i)
142 {
143 queues[i] = clCreateCommandQueueWithProperties(context, device, queue_prop, &err_ret);
144 test_error(err_ret, "clCreateCommandQueueWithProperties(CL_QUEUE_DEVICE) failed");
145 q[i] = queues[i];
146 }
147
148 size_t global_size = MAX_GWS;
149 size_t local_size = (max_local_size > global_size/16) ? global_size/16 : max_local_size;
150 if(gWimpyMode)
151 {
152 global_size = 4;
153 local_size = 2;
154 }
155
156 evt = clCreateBuffer(context, CL_MEM_READ_WRITE, n * sizeof(cl_event), NULL, &err_ret);
157 test_error(err_ret, "clCreateBuffer() failed");
158
159 mem = clCreateBuffer(context, CL_MEM_READ_ONLY|CL_MEM_COPY_HOST_PTR, n * sizeof(cl_command_queue), &q[0], &err_ret);
160 test_error(err_ret, "clCreateBuffer() failed");
161
162 buff = clCreateBuffer(context, CL_MEM_READ_WRITE, global_size * n * sizeof(cl_int), NULL, &err_ret);
163 test_error(err_ret, "clCreateBuffer() failed");
164
165 // Prepare CL source
166 char cl[65536] = { 0 };
167 char q_args[16384] = { 0 };
168 char q_list[8192] = { 0 };
169
170 kernel_arg arg_res = { sizeof(cl_mem), &buff };
171
172 std::vector<kernel_arg> args(n+1);
173 args[0] = arg_res;
174
175 for(i = 0; i < n; ++i)
176 {
177 snprintf(q_args+strlen(q_args), sizeof(q_args)-strlen(q_args)-1, ", queue_t q%d", i);
178 snprintf(q_list+strlen(q_list), sizeof(q_list)-strlen(q_list)-1, "q%d, ", i);
179 kernel_arg arg_q = { sizeof(cl_command_queue), &q[i] };
180 args[i+1] = arg_q;
181 }
182
183 snprintf(cl, sizeof(cl)-1, enqueue_block_multi_queue, q_args, n, n, q_list);
184 const char *source = cl;
185
186 err_ret = run_n_kernel_args(context, queue, &source, 1, "enqueue_block_multi_queue", local_size, global_size, kernel_results, sizeof(kernel_results), args.size(), &args[0]);
187 if(check_error(err_ret, "'%s' kernel execution failed", "enqueue_block_multi_queue")) res = -1;
188 else if((k = check_kernel_results(kernel_results, arr_size(kernel_results))) >= 0 && check_error(-1, "'%s' kernel results validation failed: [%d] returned %d expected 0", "enqueue_block_multi_queue", k, kernel_results[k])) res = -1;
189 else log_info("'%s' kernel is OK.\n", "enqueue_block_multi_queue");
190 }
191 return res;
192 }
193
194
195
196 #endif
197
198
199