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 "harness/compat.h"
17
18 #include <stdio.h>
19 #include <string.h>
20 #include <sys/types.h>
21 #include <sys/stat.h>
22
23 #include "procs.h"
24
25 static const char *fmin_kernel_code =
26 "__kernel void test_fmin(__global float *srcA, __global float *srcB, __global float *dst)\n"
27 "{\n"
28 " int tid = get_global_id(0);\n"
29 " dst[tid] = fmin(srcA[tid], srcB[tid]);\n"
30 "}\n";
31
32 static const char *fmin2_kernel_code =
33 "__kernel void test_fmin2(__global float2 *srcA, __global float *srcB, __global float2 *dst)\n"
34 "{\n"
35 " int tid = get_global_id(0);\n"
36 " dst[tid] = fmin(srcA[tid], srcB[tid]);\n"
37 "}\n";
38
39 static const char *fmin4_kernel_code =
40 "__kernel void test_fmin4(__global float4 *srcA, __global float *srcB, __global float4 *dst)\n"
41 "{\n"
42 " int tid = get_global_id(0);\n"
43 " dst[tid] = fmin(srcA[tid], srcB[tid]);\n"
44 "}\n";
45
46 static const char *fmin8_kernel_code =
47 "__kernel void test_fmin8(__global float8 *srcA, __global float *srcB, __global float8 *dst)\n"
48 "{\n"
49 " int tid = get_global_id(0);\n"
50 " dst[tid] = fmin(srcA[tid], srcB[tid]);\n"
51 "}\n";
52
53 static const char *fmin16_kernel_code =
54 "__kernel void test_fmin16(__global float16 *srcA, __global float *srcB, __global float16 *dst)\n"
55 "{\n"
56 " int tid = get_global_id(0);\n"
57 " dst[tid] = fmin(srcA[tid], srcB[tid]);\n"
58 "}\n";
59
60 static const char *fmin3_kernel_code =
61 "__kernel void test_fmin3(__global float *srcA, __global float *srcB, __global float *dst)\n"
62 "{\n"
63 " int tid = get_global_id(0);\n"
64 " vstore3(fmin(vload3(tid,srcA), srcB[tid]),tid,dst);\n"
65 "}\n";
66
67 static int
verify_fmin(float * inptrA,float * inptrB,float * outptr,int n,int veclen)68 verify_fmin(float *inptrA, float *inptrB, float *outptr, int n, int veclen)
69 {
70 float r;
71 int i, j;
72
73 for (i=0; i<n; ) {
74 int ii = i/veclen;
75 for (j=0; j<veclen && i<n; ++j, ++i) {
76 r = (inptrA[i] > inptrB[ii]) ? inptrB[ii] : inptrA[i];
77 if (r != outptr[i])
78 return -1;
79 }
80 }
81
82 return 0;
83 }
84
85 int
test_fminf(cl_device_id device,cl_context context,cl_command_queue queue,int n_elems)86 test_fminf(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
87 {
88 cl_mem streams[3];
89 cl_float *input_ptr[2], *output_ptr, *p;
90 cl_program *program;
91 cl_kernel *kernel;
92 void *values[3];
93 size_t threads[1];
94 int num_elements;
95 int err;
96 int i;
97 MTdata d;
98
99 program = (cl_program*)malloc(sizeof(cl_program)*kTotalVecCount);
100 kernel = (cl_kernel*)malloc(sizeof(cl_kernel)*kTotalVecCount);
101
102 num_elements = n_elems * (1 << (kTotalVecCount-1));
103
104 input_ptr[0] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
105 input_ptr[1] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
106 output_ptr = (cl_float*)malloc(sizeof(cl_float) * num_elements);
107 streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE,
108 sizeof(cl_float) * num_elements, NULL, NULL);
109 if (!streams[0])
110 {
111 log_error("clCreateBuffer failed\n");
112 return -1;
113 }
114 streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE,
115 sizeof(cl_float) * num_elements, NULL, NULL);
116 if (!streams[1])
117 {
118 log_error("clCreateBuffer failed\n");
119 return -1;
120 }
121 streams[2] = clCreateBuffer(context, CL_MEM_READ_WRITE,
122 sizeof(cl_float) * num_elements, NULL, NULL);
123 if (!streams[2])
124 {
125 log_error("clCreateBuffer failed\n");
126 return -1;
127 }
128
129 d = init_genrand( gRandomSeed );
130 p = input_ptr[0];
131 for (i=0; i<num_elements; i++)
132 {
133 p[i] = get_random_float(-0x20000000, 0x20000000, d);
134 }
135 p = input_ptr[1];
136 for (i=0; i<num_elements; i++)
137 {
138 p[i] = get_random_float(-0x20000000, 0x20000000, d);
139 }
140 free_mtdata(d); d = NULL;
141
142 err = clEnqueueWriteBuffer( queue, streams[0], true, 0, sizeof(cl_float)*num_elements,
143 (void *)input_ptr[0], 0, NULL, NULL );
144 if (err != CL_SUCCESS)
145 {
146 log_error("clWriteArray failed\n");
147 return -1;
148 }
149 err = clEnqueueWriteBuffer( queue, streams[1], true, 0, sizeof(cl_float)*num_elements,
150 (void *)input_ptr[1], 0, NULL, NULL );
151 if (err != CL_SUCCESS)
152 {
153 log_error("clWriteArray failed\n");
154 return -1;
155 }
156
157 err = create_single_kernel_helper( context, &program[0], &kernel[0], 1, &fmin_kernel_code, "test_fmin" );
158 if (err)
159 return -1;
160 err = create_single_kernel_helper( context, &program[1], &kernel[1], 1, &fmin2_kernel_code, "test_fmin2" );
161 if (err)
162 return -1;
163 err = create_single_kernel_helper( context, &program[2], &kernel[2], 1, &fmin4_kernel_code, "test_fmin4" );
164 if (err)
165 return -1;
166 err = create_single_kernel_helper( context, &program[3], &kernel[3], 1, &fmin8_kernel_code, "test_fmin8" );
167 if (err)
168 return -1;
169 err = create_single_kernel_helper( context, &program[4], &kernel[4], 1, &fmin16_kernel_code, "test_fmin16" );
170 if (err)
171 return -1;
172 err = create_single_kernel_helper( context, &program[5], &kernel[5], 1, &fmin3_kernel_code, "test_fmin3" );
173 if (err)
174 return -1;
175
176 values[0] = streams[0];
177 values[1] = streams[1];
178 values[2] = streams[2];
179 for (i=0; i < kTotalVecCount; i++)
180 {
181 err = clSetKernelArg(kernel[i], 0, sizeof streams[0], &streams[0] );
182 err |= clSetKernelArg(kernel[i], 1, sizeof streams[1], &streams[1] );
183 err |= clSetKernelArg(kernel[i], 2, sizeof streams[2], &streams[2] );
184 if (err != CL_SUCCESS)
185 {
186 log_error("clSetKernelArgs failed\n");
187 return -1;
188 }
189 }
190
191 threads[0] = (size_t)n_elems;
192 for (i=0; i < kTotalVecCount; i++)
193 {
194 err = clEnqueueNDRangeKernel( queue, kernel[i], 1, NULL, threads, NULL, 0, NULL, NULL );
195 if (err != CL_SUCCESS)
196 {
197 log_error("clEnqueueNDRangeKernel failed\n");
198 return -1;
199 }
200
201 err = clEnqueueReadBuffer( queue, streams[2], true, 0, sizeof(cl_float)*num_elements, output_ptr, 0, NULL, NULL );
202 if (err != CL_SUCCESS)
203 {
204 log_error("clEnqueueReadBuffer failed\n");
205 return -1;
206 }
207
208 if (verify_fmin(input_ptr[0], input_ptr[1], output_ptr, n_elems*((g_arrVecSizes[i])), (g_arrVecSizes[i])))
209 {
210 log_error("fmin float%d,float test failed\n", (g_arrVecSizes[i]));
211 err = -1;
212 }
213 else
214 {
215 log_info("fmin float%d,float test passed\n", (g_arrVecSizes[i]));
216 err = 0;
217 }
218
219 if (err)
220 break;
221 }
222
223 clReleaseMemObject(streams[0]);
224 clReleaseMemObject(streams[1]);
225 clReleaseMemObject(streams[2]);
226 for (i=0; i < kTotalVecCount; i++)
227 {
228 clReleaseKernel(kernel[i]);
229 clReleaseProgram(program[i]);
230 }
231 free(program);
232 free(kernel);
233 free(input_ptr[0]);
234 free(input_ptr[1]);
235 free(output_ptr);
236
237 return err;
238 }
239
240
241