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 extern int gWimpyMode;
28 
29 #ifdef CL_VERSION_2_0
30 
31 static const char* enqueue_block_first_kernel[] =
32 {
33     NL, "void block_fn(uint num, __global int* res)"
34     NL, "{"
35     NL, "    size_t tid = get_global_id(0);"
36     NL, ""
37     NL, "    for(int i = 1 ; i < tid ; i++)"
38     NL, "    {"
39     NL, "      for(int j = 0 ; j < num ; j++)"
40     NL, "        atomic_add(res+tid, 1);"
41     NL, "    }"
42     NL, "}"
43     NL, ""
44     NL, "kernel void enqueue_block_first_kernel(uint num, __global int* res)"
45     NL, "{"
46     NL, "  void (^kernelBlock)(void) = ^{ block_fn(num, res); };"
47     NL, ""
48     NL, "  ndrange_t ndrange = ndrange_1D(num, 1);"
49     NL, ""
50     NL, "  int enq_res = enqueue_kernel(get_default_queue(), CLK_ENQUEUE_FLAGS_NO_WAIT, ndrange, kernelBlock);"
51     NL, "  if(enq_res != CLK_SUCCESS) { res[0] = -1; return; }"
52     NL, ""
53     NL, "}"
54     NL
55 };
56 
57 static const char* enqueue_block_second_kernel[] =
58 {
59     NL, "void block_fn(uint num, __global int* res)"
60     NL, "{"
61     NL, "    for(int i = 2 ; i < num ; i++)"
62     NL, "    {"
63     NL, "      res[i] = res[i]/num - (i-1);"
64     NL, "    }"
65     NL, "}"
66     NL, ""
67     NL, "kernel void enqueue_block_second_kernel(uint num, __global int* res)"
68     NL, "{"
69     NL, "  void (^kernelBlock)(void) = ^{ block_fn(num, res); };"
70     NL, ""
71     NL, "  ndrange_t ndrange = ndrange_1D(1);"
72     NL, ""
73     NL, "  int enq_res = enqueue_kernel(get_default_queue(), CLK_ENQUEUE_FLAGS_WAIT_KERNEL, ndrange, kernelBlock);"
74     NL, "  if(enq_res != CLK_SUCCESS) { res[0] = -1; return; }"
75     NL, ""
76     NL, "}"
77     NL
78 };
79 
check_kernel_results(cl_int * results,cl_int len)80 static int check_kernel_results(cl_int* results, cl_int len)
81 {
82     for(cl_int i = 0; i < len; ++i)
83     {
84         if(results[i] != 0) return i;
85     }
86     return -1;
87 }
88 
89 /*
90     Test checks kernel block execution order in case of two different kernels with enqueue block submitted to one ordered host queue.
91 */
test_host_queue_order(cl_device_id device,cl_context context,cl_command_queue queue,int num_elements)92 int test_host_queue_order(cl_device_id device, cl_context context, cl_command_queue queue, int num_elements)
93 {
94     cl_int k, err_ret, res = 0;
95     clCommandQueueWrapper dev_queue;
96     cl_int kernel_results[MAX_GWS] = {0};
97 
98     size_t ret_len;
99     cl_uint max_queues = 1;
100     cl_uint maxQueueSize = 0;
101     err_ret = clGetDeviceInfo(device, CL_DEVICE_QUEUE_ON_DEVICE_MAX_SIZE, sizeof(maxQueueSize), &maxQueueSize, 0);
102     test_error(err_ret, "clGetDeviceInfo(CL_DEVICE_QUEUE_ON_DEVICE_MAX_SIZE) failed");
103 
104     err_ret = clGetDeviceInfo(device, CL_DEVICE_MAX_ON_DEVICE_QUEUES, sizeof(max_queues), &max_queues, &ret_len);
105     test_error(err_ret, "clGetDeviceInfo(CL_DEVICE_MAX_ON_DEVICE_QUEUES) failed");
106 
107     size_t max_local_size = 1;
108     err_ret = clGetDeviceInfo(device, CL_DEVICE_MAX_WORK_GROUP_SIZE, sizeof(max_local_size), &max_local_size, &ret_len);
109     test_error(err_ret, "clGetDeviceInfo(CL_DEVICE_MAX_WORK_GROUP_SIZE) failed");
110 
111     cl_queue_properties queue_prop_def[] =
112     {
113         CL_QUEUE_PROPERTIES, CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE|CL_QUEUE_ON_DEVICE|CL_QUEUE_ON_DEVICE_DEFAULT,
114         CL_QUEUE_SIZE, maxQueueSize,
115         0
116     };
117 
118     dev_queue = clCreateCommandQueueWithProperties(context, device, queue_prop_def, &err_ret);
119     test_error(err_ret, "clCreateCommandQueueWithProperties(CL_QUEUE_DEVICE|CL_QUEUE_DEFAULT) failed");
120 
121     cl_int status;
122     size_t size = 1;
123     cl_int result[MAX_GWS] = { 0 };
124     cl_uint num = arr_size(result);
125     if( gWimpyMode )
126     {
127         num = MAX(num / 16, 4);
128     }
129 
130     clMemWrapper res_mem;
131     clProgramWrapper program1, program2;
132     clKernelWrapper kernel1, kernel2;
133 
134     cl_event kernel_event;
135 
136     err_ret = create_single_kernel_helper(
137         context, &program1, &kernel1, arr_size(enqueue_block_first_kernel),
138         enqueue_block_first_kernel, "enqueue_block_first_kernel");
139     if(check_error(err_ret, "Create single kernel failed")) return -1;
140 
141     err_ret = create_single_kernel_helper(
142         context, &program2, &kernel2, arr_size(enqueue_block_second_kernel),
143         enqueue_block_second_kernel, "enqueue_block_second_kernel");
144     if(check_error(err_ret, "Create single kernel failed")) return -1;
145 
146     res_mem = clCreateBuffer(context, CL_MEM_READ_WRITE|CL_MEM_COPY_HOST_PTR, sizeof(kernel_results), kernel_results, &err_ret);
147     test_error(err_ret, "clCreateBuffer() failed");
148 
149     // Enqueue first kernel
150     err_ret = clSetKernelArg(kernel1, 0, sizeof(num), &num);
151     test_error(err_ret, "clSetKernelArg(0) failed");
152     err_ret = clSetKernelArg(kernel1, 1, sizeof(cl_mem), &res_mem);
153     test_error(err_ret, "clSetKernelArg(1) failed");
154 
155     cl_event event1 = clCreateUserEvent(context, &err_ret);
156     if(check_error(err_ret, "Create user event failed")) return -1;
157 
158     err_ret = clEnqueueNDRangeKernel(queue, kernel1, 1, NULL, &size, &size, 1, &event1, NULL);
159     test_error(err_ret, "clEnqueueNDRangeKernel('enqueue_block_first_kernel') failed");
160 
161     // Enqueue second kernel
162     err_ret = clSetKernelArg(kernel2, 0, sizeof(num), &num);
163     test_error(err_ret, "clSetKernelArg(0) failed");
164     err_ret = clSetKernelArg(kernel2, 1, sizeof(cl_mem), &res_mem);
165     test_error(err_ret, "clSetKernelArg(1) failed");
166 
167     err_ret = clEnqueueNDRangeKernel(queue, kernel2, 1, NULL, &size, &size, 0, NULL, &kernel_event);
168     test_error(err_ret, "clEnqueueNDRangeKernel('enqueue_block_second_kernel') failed");
169 
170     //Triger execution of first kernel
171     err_ret = clSetUserEventStatus(event1, CL_COMPLETE);
172     test_error(err_ret, "clSetUserEventStatus() failed");
173 
174     // Collect resulsts
175     err_ret = clEnqueueReadBuffer(queue, res_mem, CL_TRUE, 0, sizeof(result), result, 0, NULL, NULL);
176     test_error(err_ret, "clEnqueueReadBuffer() failed");
177 
178     err_ret = clGetEventInfo(kernel_event, CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof(status), &status, &ret_len);
179     test_error(err_ret, "clGetEventInfo() failed");
180 
181     if(check_error(status, "Kernel execution status %d", status)) return status;
182 
183     if((k = check_kernel_results(result, num)) >= 0 && check_error(-1, "'%s' results validation failed: [%d] returned %d expected 0", "test_host_queue_order", k, result[k])) res = -1;
184 
185     clReleaseEvent(kernel_event);
186     clReleaseEvent(event1);
187 
188     return res;
189 }
190 
191 #endif
192 
193