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 *fmax_kernel_code =
26 "__kernel void test_fmax(__global float *srcA, __global float *srcB, __global float *dst)\n"
27 "{\n"
28 " int tid = get_global_id(0);\n"
29 " dst[tid] = fmax(srcA[tid], srcB[tid]);\n"
30 "}\n";
31
32 static const char *fmax2_kernel_code =
33 "__kernel void test_fmax2(__global float2 *srcA, __global float *srcB, __global float2 *dst)\n"
34 "{\n"
35 " int tid = get_global_id(0);\n"
36 " dst[tid] = fmax(srcA[tid], srcB[tid]);\n"
37 "}\n";
38
39 static const char *fmax4_kernel_code =
40 "__kernel void test_fmax4(__global float4 *srcA, __global float *srcB, __global float4 *dst)\n"
41 "{\n"
42 " int tid = get_global_id(0);\n"
43 " dst[tid] = fmax(srcA[tid], srcB[tid]);\n"
44 "}\n";
45
46 static const char *fmax8_kernel_code =
47 "__kernel void test_fmax8(__global float8 *srcA, __global float *srcB, __global float8 *dst)\n"
48 "{\n"
49 " int tid = get_global_id(0);\n"
50 " dst[tid] = fmax(srcA[tid], srcB[tid]);\n"
51 "}\n";
52
53 static const char *fmax16_kernel_code =
54 "__kernel void test_fmax16(__global float16 *srcA, __global float *srcB, __global float16 *dst)\n"
55 "{\n"
56 " int tid = get_global_id(0);\n"
57 " dst[tid] = fmax(srcA[tid], srcB[tid]);\n"
58 "}\n";
59
60 static const char *fmax3_kernel_code =
61 "__kernel void test_fmax3(__global float *srcA, __global float *srcB, __global float *dst)\n"
62 "{\n"
63 " int tid = get_global_id(0);\n"
64 " vstore3(fmax(vload3(tid,srcA), srcB[tid]),tid,dst);\n"
65 "}\n";
66
67 static int
verify_fmax(float * inptrA,float * inptrB,float * outptr,int n,int veclen)68 verify_fmax(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]) ? inptrA[i] : inptrB[ii];
77 if (r != outptr[i]) {
78 log_info("Verify noted discrepancy at %d (of %d) (vec %d, pos %d)\n",
79 i,n,ii,j);
80 log_info("SHould be %f, is %f\n", r, outptr[i]);
81 log_info("Taking max of (%f,%f)\n", inptrA[i], inptrB[i]);
82 return -1;
83 }
84 }
85 }
86
87 return 0;
88 }
89
90 int
test_fmaxf(cl_device_id device,cl_context context,cl_command_queue queue,int n_elems)91 test_fmaxf(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
92 {
93 cl_mem streams[3];
94 cl_float *input_ptr[2], *output_ptr, *p;
95 cl_program *program;
96 cl_kernel *kernel;
97 void *values[3];
98 size_t threads[1];
99 int num_elements;
100 int err;
101 int i;
102 MTdata d;
103
104 program = (cl_program*)malloc(sizeof(cl_program)*kTotalVecCount);
105 kernel = (cl_kernel*)malloc(sizeof(cl_kernel)*kTotalVecCount);
106
107 num_elements = n_elems * (1 << (kTotalVecCount-1));
108
109 input_ptr[0] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
110 input_ptr[1] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
111 output_ptr = (cl_float*)malloc(sizeof(cl_float) * num_elements);
112 streams[0] = clCreateBuffer( context, (cl_mem_flags)(CL_MEM_READ_WRITE), sizeof(cl_float) * num_elements, NULL, NULL );
113 if (!streams[0])
114 {
115 log_error("clCreateBuffer failed\n");
116 return -1;
117 }
118 streams[1] = clCreateBuffer( context, (cl_mem_flags)(CL_MEM_READ_WRITE), sizeof(cl_float) * num_elements, NULL, NULL );
119 if (!streams[1])
120 {
121 log_error("clCreateBuffer failed\n");
122 return -1;
123 }
124 streams[2] = clCreateBuffer( context, (cl_mem_flags)(CL_MEM_READ_WRITE), sizeof(cl_float) * num_elements, NULL, NULL );
125 if (!streams[2])
126 {
127 log_error("clCreateBuffer failed\n");
128 return -1;
129 }
130
131 d = init_genrand( gRandomSeed );
132 p = input_ptr[0];
133 for (i=0; i<num_elements; i++)
134 {
135 p[i] = get_random_float(-0x20000000, 0x20000000, d);
136 }
137 p = input_ptr[1];
138 for (i=0; i<num_elements; i++)
139 {
140 p[i] = get_random_float(-0x20000000, 0x20000000, d);
141 }
142 free_mtdata(d); d = NULL;
143
144 err = clEnqueueWriteBuffer( queue, streams[0], true, 0, sizeof(cl_float)*num_elements,
145 (void *)input_ptr[0], 0, NULL, NULL );
146 if (err != CL_SUCCESS)
147 {
148 log_error("clWriteArray failed\n");
149 return -1;
150 }
151 err = clEnqueueWriteBuffer( queue, streams[1], true, 0, sizeof(cl_float)*num_elements,
152 (void *)input_ptr[1], 0, NULL, NULL );
153 if (err != CL_SUCCESS)
154 {
155 log_error("clWriteArray failed\n");
156 return -1;
157 }
158
159 err = create_single_kernel_helper( context, &program[0], &kernel[0], 1, &fmax_kernel_code, "test_fmax" );
160 if (err)
161 return -1;
162 err = create_single_kernel_helper( context, &program[1], &kernel[1], 1, &fmax2_kernel_code, "test_fmax2" );
163 if (err)
164 return -1;
165 err = create_single_kernel_helper( context, &program[2], &kernel[2], 1, &fmax4_kernel_code, "test_fmax4" );
166 if (err)
167 return -1;
168 err = create_single_kernel_helper( context, &program[3], &kernel[3], 1, &fmax8_kernel_code, "test_fmax8" );
169 if (err)
170 return -1;
171 err = create_single_kernel_helper( context, &program[4], &kernel[4], 1, &fmax16_kernel_code, "test_fmax16" );
172 if (err)
173 return -1;
174 err = create_single_kernel_helper( context, &program[5], &kernel[5], 1, &fmax3_kernel_code, "test_fmax3" );
175 if (err)
176 return -1;
177
178 values[0] = streams[0];
179 values[1] = streams[1];
180 values[2] = streams[2];
181 for (i=0; i < kTotalVecCount; i++)
182 {
183 err = clSetKernelArg(kernel[i], 0, sizeof streams[0], &streams[0] );
184 err |= clSetKernelArg(kernel[i], 1, sizeof streams[1], &streams[1] );
185 err |= clSetKernelArg(kernel[i], 2, sizeof streams[2], &streams[2] );
186 if (err != CL_SUCCESS)
187 {
188 log_error("clSetKernelArgs failed\n");
189 return -1;
190 }
191 }
192
193 threads[0] = (size_t)n_elems;
194 for (i=0; i < kTotalVecCount; i++)
195 {
196 err = clEnqueueNDRangeKernel( queue, kernel[i], 1, NULL, threads, NULL, 0, NULL, NULL );
197 if (err != CL_SUCCESS)
198 {
199 log_error("clEnqueueNDRangeKernel failed\n");
200 return -1;
201 }
202
203 err = clEnqueueReadBuffer(queue, streams[2], true, 0, sizeof(cl_float)*num_elements,
204 output_ptr, 0, NULL, NULL);
205 if (err != CL_SUCCESS)
206 {
207 log_error("clEnqueueReadBuffer failed\n");
208 return -1;
209 }
210
211 if (verify_fmax(input_ptr[0], input_ptr[1], output_ptr, n_elems*((g_arrVecSizes[i])), (g_arrVecSizes[i])))
212 {
213 log_error("FMAX float%d,float test failed\n", (g_arrVecSizes[i]));
214 err = -1;
215 }
216 else
217 {
218 log_info("FMAX float%d,float test passed\n", (g_arrVecSizes[i]));
219 err = 0;
220 }
221
222 if (err)
223 break;
224 }
225
226 clReleaseMemObject(streams[0]);
227 clReleaseMemObject(streams[1]);
228 clReleaseMemObject(streams[2]);
229 for (i=0; i < kTotalVecCount; i++)
230 {
231 clReleaseKernel(kernel[i]);
232 clReleaseProgram(program[i]);
233 }
234 free(program);
235 free(kernel);
236 free(input_ptr[0]);
237 free(input_ptr[1]);
238 free(output_ptr);
239
240 return err;
241 }
242
243
244