1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
|
//
// Copyright (c) 2017 The Khronos Group Inc.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
//
#include "harness/compat.h"
#include <stdio.h>
#include <string.h>
#include <assert.h>
#include <sys/types.h>
#include <sys/stat.h>
#include "procs.h"
#include "harness/errorHelpers.h"
#define TEST_INT_VALUE 100
const char* pipe_subgroups_kernel_code = {
"#pragma OPENCL EXTENSION cl_khr_subgroups : enable\n"
"__kernel void test_pipe_subgroups_divergence_write(__global int *src, __write_only pipe int out_pipe, __global int *active_work_item_buffer)\n"
"{\n"
" int gid = get_global_id(0);\n"
" reserve_id_t res_id; \n"
"\n"
" if(get_sub_group_id() % 2 == 0)\n"
" {\n"
" active_work_item_buffer[gid] = 1;\n"
" res_id = sub_group_reserve_write_pipe(out_pipe, get_sub_group_size());\n"
" if(is_valid_reserve_id(res_id))\n"
" {\n"
" write_pipe(out_pipe, res_id, get_sub_group_local_id(), &src[gid]);\n"
" sub_group_commit_write_pipe(out_pipe, res_id);\n"
" }\n"
" }\n"
"}\n"
"\n"
"__kernel void test_pipe_subgroups_divergence_read(__read_only pipe int in_pipe, __global int *dst)\n"
"{\n"
" int gid = get_global_id(0);\n"
" reserve_id_t res_id; \n"
"\n"
" if(get_sub_group_id() % 2 == 0)\n"
" {\n"
" res_id = sub_group_reserve_read_pipe(in_pipe, get_sub_group_size());\n"
" if(is_valid_reserve_id(res_id))\n"
" {\n"
" read_pipe(in_pipe, res_id, get_sub_group_local_id(), &dst[gid]);\n"
" sub_group_commit_read_pipe(in_pipe, res_id);\n"
" }\n"
" }\n"
"}\n"
};
static int verify_result(void *ptr1, void *ptr2, int n)
{
int i;
int sum_input = 0, sum_output = 0;
cl_int *inptr = (cl_int *)ptr1;
cl_int *outptr = (cl_int *)ptr2;
for(i = 0; i < n; i++)
{
sum_input += inptr[i];
}
sum_input *= TEST_INT_VALUE;
for(i = 0; i < n; i++)
{
if(outptr[i] == TEST_INT_VALUE){
sum_output += outptr[i];
}
}
if(sum_input != sum_output){
return -1;
}
return 0;
}
int test_pipe_subgroups_divergence(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements)
{
clMemWrapper pipe;
clMemWrapper buffers[3];
cl_int *outptr;
cl_int *inptr;
cl_int *active_work_item_buffer;
clProgramWrapper program;
clKernelWrapper kernel[2];
size_t global_work_size[3];
size_t local_work_size[3];
cl_int err;
cl_int size;
int i;
size_t subgroup_count;
clEventWrapper producer_sync_event = NULL;
clEventWrapper consumer_sync_event = NULL;
BufferOwningPtr<cl_int> BufferInPtr;
BufferOwningPtr<cl_int> BufferOutPtr;
const char *kernelName[] = { "test_pipe_subgroups_divergence_write",
"test_pipe_subgroups_divergence_read" };
size_t min_alignment = get_min_alignment(context);
global_work_size[0] = (cl_uint)num_elements;
if (!is_extension_available(deviceID, "cl_khr_subgroups"))
{
log_info("cl_khr_subgroups is not supported on this platform. Skipping "
"test.\n");
return CL_SUCCESS;
}
size = sizeof(int) * num_elements;
inptr = (cl_int *)align_malloc(size, min_alignment);
outptr = (cl_int *)align_malloc(size, min_alignment);
active_work_item_buffer = (cl_int *)align_malloc(size, min_alignment);
for(i = 0; i < num_elements; i++){
inptr[i] = TEST_INT_VALUE;
outptr[i] = 0;
active_work_item_buffer[i] = 0;
}
BufferInPtr.reset(inptr, nullptr, 0, size, true);
BufferOutPtr.reset(outptr, nullptr, 0, size, true);
buffers[0] = clCreateBuffer(context, CL_MEM_COPY_HOST_PTR, size, inptr, &err);
test_error_ret(err, " clCreateBuffer failed", -1);
buffers[1] = clCreateBuffer(context, CL_MEM_COPY_HOST_PTR, size, outptr, &err);
test_error_ret(err, " clCreateBuffer failed", -1);
buffers[2] = clCreateBuffer(context, CL_MEM_COPY_HOST_PTR, size, active_work_item_buffer, &err);
test_error_ret(err, " clCreateBuffer failed", -1);
pipe = clCreatePipe(context, CL_MEM_HOST_NO_ACCESS, sizeof(int), num_elements, NULL, &err);
test_error_ret(err, " clCreatePipe failed", -1);
// Create producer kernel
err = create_single_kernel_helper(
context, &program, &kernel[0], 1,
(const char **)&pipe_subgroups_kernel_code, kernelName[0]);
test_error_ret(err, " Error creating program", -1);
//Create consumer kernel
kernel[1] = clCreateKernel(program, kernelName[1], &err);
test_error_ret(err, " Error creating kernel", -1);
err = clSetKernelArg(kernel[0], 0, sizeof(cl_mem), (void*)&buffers[0]);
err |= clSetKernelArg(kernel[0], 1, sizeof(cl_mem), (void*)&pipe);
err |= clSetKernelArg(kernel[0], 2, sizeof(cl_mem), (void*)&buffers[2]);
err |= clSetKernelArg(kernel[1], 0, sizeof(cl_mem), (void*)&pipe);
err |= clSetKernelArg(kernel[1], 1, sizeof(cl_mem), (void*)&buffers[1]);
test_error_ret(err, " clSetKernelArg failed", -1);
err = get_max_common_work_group_size( context, kernel[0], global_work_size[0], &local_work_size[0] );
test_error_ret(err, " Unable to get work group size to use", -1);
cl_platform_id platform;
err = clGetDeviceInfo(deviceID, CL_DEVICE_PLATFORM, sizeof(platform),
&platform, NULL);
test_error_ret(err, " clGetDeviceInfo failed", -1);
clGetKernelSubGroupInfoKHR_fn clGetKernelSubGroupInfoKHR =
(clGetKernelSubGroupInfoKHR_fn)clGetExtensionFunctionAddressForPlatform(
platform, "clGetKernelSubGroupInfoKHR");
err = clGetKernelSubGroupInfoKHR(kernel[0], deviceID, CL_KERNEL_SUB_GROUP_COUNT_FOR_NDRANGE_KHR, sizeof(local_work_size[0]), &local_work_size[0], sizeof(subgroup_count), &subgroup_count, NULL);
test_error_ret(err, " clGetKernelSubGroupInfoKHR failed", -1);
if(subgroup_count <= 1)
{
log_info("Only 1 subgroup per workgroup for the kernel. Hence no divergence among subgroups possible. Skipping test.\n");
return CL_SUCCESS;
}
// Launch Producer kernel
err = clEnqueueNDRangeKernel( queue, kernel[0], 1, NULL, global_work_size, local_work_size, 0, NULL, &producer_sync_event );
test_error_ret(err, " clEnqueueNDRangeKernel failed", -1);
err = clEnqueueReadBuffer(queue, buffers[2], true, 0, size, active_work_item_buffer, 1, &producer_sync_event, NULL);
test_error_ret(err, " clEnqueueReadBuffer failed", -1);
// Launch Consumer kernel
err = clEnqueueNDRangeKernel( queue, kernel[1], 1, NULL, global_work_size, local_work_size, 1, &producer_sync_event, &consumer_sync_event );
test_error_ret(err, " clEnqueueNDRangeKernel failed", -1);
err = clEnqueueReadBuffer(queue, buffers[1], true, 0, size, outptr, 1, &consumer_sync_event, NULL);
test_error_ret(err, " clEnqueueReadBuffer failed", -1);
if( verify_result( active_work_item_buffer, outptr, num_elements)){
log_error("test_pipe_subgroups_divergence failed\n");
return -1;
}
else {
log_info("test_pipe_subgroups_divergence passed\n");
}
return 0;
}
|
