• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
17 #include "common.h"
18 #include "function_list.h"
19 #include "test_functions.h"
20 #include "utility.h"
21 
22 #include <cinttypes>
23 #include <cstring>
24 
25 namespace {
26 
BuildKernelFn(cl_uint job_id,cl_uint thread_id UNUSED,void * p)27 cl_int BuildKernelFn(cl_uint job_id, cl_uint thread_id UNUSED, void *p)
28 {
29     BuildKernelInfo &info = *(BuildKernelInfo *)p;
30     auto generator = [](const std::string &kernel_name, const char *builtin,
31                         cl_uint vector_size_index) {
32         return GetUnaryKernel(kernel_name, builtin, ParameterType::Double,
33                               ParameterType::ULong, vector_size_index);
34     };
35     return BuildKernels(info, job_id, generator);
36 }
37 
random64(MTdata d)38 cl_ulong random64(MTdata d)
39 {
40     return (cl_ulong)genrand_int32(d) | ((cl_ulong)genrand_int32(d) << 32);
41 }
42 
43 } // anonymous namespace
44 
TestFunc_Double_ULong(const Func * f,MTdata d,bool relaxedMode)45 int TestFunc_Double_ULong(const Func *f, MTdata d, bool relaxedMode)
46 {
47     int error;
48     Programs programs;
49     const unsigned thread_id = 0; // Test is currently not multithreaded.
50     KernelMatrix kernels;
51     float maxError = 0.0f;
52     int ftz = f->ftz || gForceFTZ;
53     double maxErrorVal = 0.0f;
54     uint64_t step = getTestStep(sizeof(cl_double), BUFFER_SIZE);
55 
56     logFunctionInfo(f->name, sizeof(cl_double), relaxedMode);
57 
58     Force64BitFPUPrecision();
59 
60     // Init the kernels
61     BuildKernelInfo build_info{ 1, kernels, programs, f->nameInCode,
62                                 relaxedMode };
63     if ((error = ThreadPool_Do(BuildKernelFn,
64                                gMaxVectorSizeIndex - gMinVectorSizeIndex,
65                                &build_info)))
66         return error;
67 
68     for (uint64_t i = 0; i < (1ULL << 32); i += step)
69     {
70         // Init input array
71         cl_ulong *p = (cl_ulong *)gIn;
72         for (size_t j = 0; j < BUFFER_SIZE / sizeof(cl_ulong); j++)
73             p[j] = random64(d);
74 
75         if ((error = clEnqueueWriteBuffer(gQueue, gInBuffer, CL_FALSE, 0,
76                                           BUFFER_SIZE, gIn, 0, NULL, NULL)))
77         {
78             vlog_error("\n*** Error %d in clEnqueueWriteBuffer ***\n", error);
79             return error;
80         }
81 
82         // Write garbage into output arrays
83         for (auto j = gMinVectorSizeIndex; j < gMaxVectorSizeIndex; j++)
84         {
85             uint32_t pattern = 0xffffdead;
86             if (gHostFill)
87             {
88                 memset_pattern4(gOut[j], &pattern, BUFFER_SIZE);
89                 if ((error = clEnqueueWriteBuffer(gQueue, gOutBuffer[j],
90                                                   CL_FALSE, 0, BUFFER_SIZE,
91                                                   gOut[j], 0, NULL, NULL)))
92                 {
93                     vlog_error(
94                         "\n*** Error %d in clEnqueueWriteBuffer2(%d) ***\n",
95                         error, j);
96                     return error;
97                 }
98             }
99             else
100             {
101                 if ((error = clEnqueueFillBuffer(gQueue, gOutBuffer[j],
102                                                  &pattern, sizeof(pattern), 0,
103                                                  BUFFER_SIZE, 0, NULL, NULL)))
104                 {
105                     vlog_error("Error: clEnqueueFillBuffer failed! err: %d\n",
106                                error);
107                     return error;
108                 }
109             }
110         }
111 
112         // Run the kernels
113         for (auto j = gMinVectorSizeIndex; j < gMaxVectorSizeIndex; j++)
114         {
115             size_t vectorSize = sizeValues[j] * sizeof(cl_double);
116             size_t localCount = (BUFFER_SIZE + vectorSize - 1) / vectorSize;
117             if ((error = clSetKernelArg(kernels[j][thread_id], 0,
118                                         sizeof(gOutBuffer[j]), &gOutBuffer[j])))
119             {
120                 LogBuildError(programs[j]);
121                 return error;
122             }
123             if ((error = clSetKernelArg(kernels[j][thread_id], 1,
124                                         sizeof(gInBuffer), &gInBuffer)))
125             {
126                 LogBuildError(programs[j]);
127                 return error;
128             }
129 
130             if ((error = clEnqueueNDRangeKernel(gQueue, kernels[j][thread_id],
131                                                 1, NULL, &localCount, NULL, 0,
132                                                 NULL, NULL)))
133             {
134                 vlog_error("FAILED -- could not execute kernel\n");
135                 return error;
136             }
137         }
138 
139         // Get that moving
140         if ((error = clFlush(gQueue))) vlog("clFlush failed\n");
141 
142         // Calculate the correctly rounded reference result
143         double *r = (double *)gOut_Ref;
144         cl_ulong *s = (cl_ulong *)gIn;
145         for (size_t j = 0; j < BUFFER_SIZE / sizeof(cl_double); j++)
146             r[j] = (double)f->dfunc.f_u(s[j]);
147 
148         // Read the data back
149         for (auto j = gMinVectorSizeIndex; j < gMaxVectorSizeIndex; j++)
150         {
151             if ((error =
152                      clEnqueueReadBuffer(gQueue, gOutBuffer[j], CL_TRUE, 0,
153                                          BUFFER_SIZE, gOut[j], 0, NULL, NULL)))
154             {
155                 vlog_error("ReadArray failed %d\n", error);
156                 return error;
157             }
158         }
159 
160         if (gSkipCorrectnessTesting) break;
161 
162         // Verify data
163         uint64_t *t = (uint64_t *)gOut_Ref;
164         for (size_t j = 0; j < BUFFER_SIZE / sizeof(cl_double); j++)
165         {
166             for (auto k = gMinVectorSizeIndex; k < gMaxVectorSizeIndex; k++)
167             {
168                 uint64_t *q = (uint64_t *)(gOut[k]);
169 
170                 // If we aren't getting the correctly rounded result
171                 if (t[j] != q[j])
172                 {
173                     double test = ((double *)q)[j];
174                     long double correct = f->dfunc.f_u(s[j]);
175                     float err = Bruteforce_Ulp_Error_Double(test, correct);
176                     int fail = !(fabsf(err) <= f->double_ulps);
177 
178                     if (fail)
179                     {
180                         if (ftz || relaxedMode)
181                         {
182                             // retry per section 6.5.3.2
183                             if (IsDoubleResultSubnormal(correct,
184                                                         f->double_ulps))
185                             {
186                                 fail = fail && (test != 0.0);
187                                 if (!fail) err = 0.0f;
188                             }
189                         }
190                     }
191                     if (fabsf(err) > maxError)
192                     {
193                         maxError = fabsf(err);
194                         maxErrorVal = s[j];
195                     }
196                     if (fail)
197                     {
198                         vlog_error(
199                             "\n%s%sD: %f ulp error at 0x%16.16" PRIx64 ": "
200                             "*%.13la vs. %.13la\n",
201                             f->name, sizeNames[k], err, ((uint64_t *)gIn)[j],
202                             ((double *)gOut_Ref)[j], test);
203                         return -1;
204                     }
205                 }
206             }
207         }
208 
209         if (0 == (i & 0x0fffffff))
210         {
211             if (gVerboseBruteForce)
212             {
213                 vlog("base:%14" PRIu64 " step:%10" PRIu64
214                      "  bufferSize:%10d \n",
215                      i, step, BUFFER_SIZE);
216             }
217             else
218             {
219                 vlog(".");
220             }
221             fflush(stdout);
222         }
223     }
224 
225     if (!gSkipCorrectnessTesting)
226     {
227         if (gWimpyMode)
228             vlog("Wimp pass");
229         else
230             vlog("passed");
231 
232         vlog("\t%8.2f @ %a", maxError, maxErrorVal);
233     }
234 
235     vlog("\n");
236 
237     return CL_SUCCESS;
238 }
239