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 *smoothstep_kernel_code =
26 "__kernel void test_smoothstep(__global float *edge0, __global float *edge1, __global float *x, __global float *dst)\n"
27 "{\n"
28 " int tid = get_global_id(0);\n"
29 "\n"
30 " dst[tid] = smoothstep(edge0[tid], edge1[tid], x[tid]);\n"
31 "}\n";
32
33 static const char *smoothstep2_kernel_code =
34 "__kernel void test_smoothstep2(__global float2 *edge0, __global float2 *edge1, __global float2 *x, __global float2 *dst)\n"
35 "{\n"
36 " int tid = get_global_id(0);\n"
37 "\n"
38 " dst[tid] = smoothstep(edge0[tid], edge1[tid], x[tid]);\n"
39 "}\n";
40
41 static const char *smoothstep4_kernel_code =
42 "__kernel void test_smoothstep4(__global float4 *edge0, __global float4 *edge1, __global float4 *x, __global float4 *dst)\n"
43 "{\n"
44 " int tid = get_global_id(0);\n"
45 "\n"
46 " dst[tid] = smoothstep(edge0[tid], edge1[tid], x[tid]);\n"
47 "}\n";
48
49 static const char *smoothstep8_kernel_code =
50 "__kernel void test_smoothstep8(__global float8 *edge0, __global float8 *edge1, __global float8 *x, __global float8 *dst)\n"
51 "{\n"
52 " int tid = get_global_id(0);\n"
53 "\n"
54 " dst[tid] = smoothstep(edge0[tid], edge1[tid], x[tid]);\n"
55 "}\n";
56
57 static const char *smoothstep16_kernel_code =
58 "__kernel void test_smoothstep16(__global float16 *edge0, __global float16 *edge1, __global float16 *x, __global float16 *dst)\n"
59 "{\n"
60 " int tid = get_global_id(0);\n"
61 "\n"
62 " dst[tid] = smoothstep(edge0[tid], edge1[tid], x[tid]);\n"
63 "}\n";
64
65 static const char *smoothstep3_kernel_code =
66 "__kernel void test_smoothstep3(__global float *edge0, __global float *edge1, __global float *x, __global float *dst)\n"
67 "{\n"
68 " int tid = get_global_id(0);\n"
69 "\n"
70 " vstore3(smoothstep(vload3(tid,edge0),vload3(tid,edge1),vload3(tid,x)), tid, dst);\n"
71 "}\n";
72
73 #define MAX_ERR (1e-5f)
74
75 static float
verify_smoothstep(float * edge0,float * edge1,float * x,float * outptr,int n)76 verify_smoothstep(float *edge0, float *edge1, float *x, float *outptr, int n)
77 {
78 float r, t, delta, max_err = 0.0f;
79 int i;
80
81 for (i=0; i<n; i++)
82 {
83 t = (x[i] - edge0[i]) / (edge1[i] - edge0[i]);
84 if (t < 0.0f)
85 t = 0.0f;
86 else if (t > 1.0f)
87 t = 1.0f;
88 r = t * t * (3.0f - 2.0f * t);
89 delta = (float)fabs(r - outptr[i]);
90 if (delta > max_err)
91 max_err = delta;
92 }
93
94 return max_err;
95 }
96
97 const static char *fn_names[] = { "SMOOTHSTEP float", "SMOOTHSTEP float2", "SMOOTHSTEP float4", "SMOOTHSTEP float8", "SMOOTHSTEP float16", "SMOOTHSTEP float3" };
98
99 int
test_smoothstep(cl_device_id device,cl_context context,cl_command_queue queue,int n_elems)100 test_smoothstep(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
101 {
102 cl_mem streams[4];
103 cl_float *input_ptr[3], *output_ptr, *p, *p_edge0;
104 cl_program program[kTotalVecCount];
105 cl_kernel kernel[kTotalVecCount];
106 size_t threads[1];
107 float max_err;
108 int num_elements;
109 int err;
110 int i;
111 MTdata d;
112
113 num_elements = n_elems * 16;
114
115 input_ptr[0] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
116 input_ptr[1] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
117 input_ptr[2] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
118 output_ptr = (cl_float*)malloc(sizeof(cl_float) * num_elements);
119 streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE,
120 sizeof(cl_float) * num_elements, NULL, NULL);
121 if (!streams[0])
122 {
123 log_error("clCreateBuffer failed\n");
124 return -1;
125 }
126 streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE,
127 sizeof(cl_float) * num_elements, NULL, NULL);
128 if (!streams[1])
129 {
130 log_error("clCreateBuffer failed\n");
131 return -1;
132 }
133 streams[2] = clCreateBuffer(context, CL_MEM_READ_WRITE,
134 sizeof(cl_float) * num_elements, NULL, NULL);
135 if (!streams[2])
136 {
137 log_error("clCreateBuffer failed\n");
138 return -1;
139 }
140
141 streams[3] = clCreateBuffer(context, CL_MEM_READ_WRITE,
142 sizeof(cl_float) * num_elements, NULL, NULL);
143 if (!streams[3])
144 {
145 log_error("clCreateBuffer failed\n");
146 return -1;
147 }
148
149 p = input_ptr[0];
150 d = init_genrand( gRandomSeed );
151 for (i=0; i<num_elements; i++)
152 {
153 p[i] = get_random_float(-0x00400000, 0x00400000, d);
154 }
155
156 p = input_ptr[1];
157 p_edge0 = input_ptr[0];
158 for (i=0; i<num_elements; i++)
159 {
160 float edge0 = p_edge0[i];
161 float edge1;
162 do {
163 edge1 = get_random_float(-0x00400000, 0x00400000, d);
164 if (edge0 < edge1)
165 break;
166 } while (1);
167 p[i] = edge1;
168 }
169
170 p = input_ptr[2];
171 for (i=0; i<num_elements; i++)
172 {
173 p[i] = get_random_float(-0x00400000, 0x00400000, d);
174 }
175 free_mtdata(d);
176 d = NULL;
177
178 err = clEnqueueWriteBuffer( queue, streams[0], true, 0, sizeof(cl_float)*num_elements, (void *)input_ptr[0], 0, NULL, NULL );
179 if (err != CL_SUCCESS)
180 {
181 log_error("clWriteArray failed\n");
182 return -1;
183 }
184 err = clEnqueueWriteBuffer( queue, streams[1], true, 0, sizeof(cl_float)*num_elements, (void *)input_ptr[1], 0, NULL, NULL );
185 if (err != CL_SUCCESS)
186 {
187 log_error("clWriteArray failed\n");
188 return -1;
189 }
190 err = clEnqueueWriteBuffer( queue, streams[2], true, 0, sizeof(cl_float)*num_elements, (void *)input_ptr[2], 0, NULL, NULL );
191 if (err != CL_SUCCESS)
192 {
193 log_error("clWriteArray failed\n");
194 return -1;
195 }
196
197 err = create_single_kernel_helper( context, &program[0], &kernel[0], 1, &smoothstep_kernel_code, "test_smoothstep" );
198 if (err)
199 return -1;
200 err = create_single_kernel_helper( context, &program[1], &kernel[1], 1, &smoothstep2_kernel_code, "test_smoothstep2" );
201 if (err)
202 return -1;
203 err = create_single_kernel_helper( context, &program[2], &kernel[2], 1, &smoothstep4_kernel_code, "test_smoothstep4" );
204 if (err)
205 return -1;
206 err = create_single_kernel_helper( context, &program[3], &kernel[3], 1, &smoothstep8_kernel_code, "test_smoothstep8" );
207 if (err)
208 return -1;
209 err = create_single_kernel_helper( context, &program[4], &kernel[4], 1, &smoothstep16_kernel_code, "test_smoothstep16" );
210 if (err)
211 return -1;
212 err = create_single_kernel_helper( context, &program[5], &kernel[5], 1, &smoothstep3_kernel_code, "test_smoothstep3" );
213 if (err)
214 return -1;
215
216 for (i=0; i<kTotalVecCount; i++)
217 {
218 err = clSetKernelArg(kernel[i], 0, sizeof streams[0], &streams[0] );
219 err |= clSetKernelArg(kernel[i], 1, sizeof streams[1], &streams[1] );
220 err |= clSetKernelArg(kernel[i], 2, sizeof streams[2], &streams[2] );
221 err |= clSetKernelArg(kernel[i], 3, sizeof streams[3], &streams[3] );
222 if (err != CL_SUCCESS)
223 {
224 log_error("clSetKernelArgs failed\n");
225 return -1;
226 }
227 }
228
229
230 threads[0] = (size_t)n_elems;
231 for (i=0; i<kTotalVecCount; i++)
232 {
233 err = clEnqueueNDRangeKernel( queue, kernel[i], 1, NULL, threads, NULL, 0, NULL, NULL );
234 if (err != CL_SUCCESS)
235 {
236 log_error("clEnqueueNDRangeKernel failed\n");
237 return -1;
238 }
239
240
241 err = clEnqueueReadBuffer( queue, streams[3], true, 0, sizeof(cl_float)*num_elements, (void *)output_ptr, 0, NULL, NULL );
242 if (err != CL_SUCCESS)
243 {
244 log_error("clEnqueueReadBuffer failed\n");
245 return -1;
246 }
247
248 max_err = verify_smoothstep(input_ptr[0], input_ptr[1], input_ptr[2], output_ptr, n_elems * g_arrVecSizes[i]);
249
250 if (max_err > MAX_ERR)
251 {
252 log_error("%s test failed %g max err\n", fn_names[i], max_err);
253 err = -1;
254 }
255 else
256 {
257 log_info("%s test passed %g max err\n", fn_names[i], max_err);
258 err = 0;
259 }
260
261 if (err)
262 break;
263 }
264
265 clReleaseMemObject(streams[0]);
266 clReleaseMemObject(streams[1]);
267 clReleaseMemObject(streams[2]);
268 clReleaseMemObject(streams[3]);
269 for (i=0; i<kTotalVecCount; i++)
270 {
271 clReleaseKernel(kernel[i]);
272 clReleaseProgram(program[i]);
273 }
274 free(input_ptr[0]);
275 free(input_ptr[1]);
276 free(input_ptr[2]);
277 free(output_ptr);
278
279 return err;
280 }
281
282
283