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::Float,
33 ParameterType::UInt, vector_size_index);
34 };
35 return BuildKernels(info, job_id, generator);
36 }
37
38 } // anonymous namespace
39
TestFunc_Float_UInt(const Func * f,MTdata d,bool relaxedMode)40 int TestFunc_Float_UInt(const Func *f, MTdata d, bool relaxedMode)
41 {
42 int error;
43 Programs programs;
44 KernelMatrix kernels;
45 const unsigned thread_id = 0; // Test is currently not multithreaded.
46 float maxError = 0.0f;
47 int ftz = f->ftz || gForceFTZ || 0 == (CL_FP_DENORM & gFloatCapabilities);
48 float maxErrorVal = 0.0f;
49 uint64_t step = getTestStep(sizeof(float), BUFFER_SIZE);
50 int scale = (int)((1ULL << 32) / (16 * BUFFER_SIZE / sizeof(double)) + 1);
51
52 logFunctionInfo(f->name, sizeof(cl_float), relaxedMode);
53
54 float float_ulps;
55 if (gIsEmbedded)
56 float_ulps = f->float_embedded_ulps;
57 else
58 float_ulps = f->float_ulps;
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 uint32_t *p = (uint32_t *)gIn;
72 if (gWimpyMode)
73 {
74 for (size_t j = 0; j < BUFFER_SIZE / sizeof(float); j++)
75 p[j] = (uint32_t)i + j * scale;
76 }
77 else
78 {
79 for (size_t j = 0; j < BUFFER_SIZE / sizeof(float); j++)
80 p[j] = (uint32_t)i + j;
81 }
82 if ((error = clEnqueueWriteBuffer(gQueue, gInBuffer, CL_FALSE, 0,
83 BUFFER_SIZE, gIn, 0, NULL, NULL)))
84 {
85 vlog_error("\n*** Error %d in clEnqueueWriteBuffer ***\n", error);
86 return error;
87 }
88
89 // Write garbage into output arrays
90 for (auto j = gMinVectorSizeIndex; j < gMaxVectorSizeIndex; j++)
91 {
92 uint32_t pattern = 0xffffdead;
93 if (gHostFill)
94 {
95 memset_pattern4(gOut[j], &pattern, BUFFER_SIZE);
96 if ((error = clEnqueueWriteBuffer(gQueue, gOutBuffer[j],
97 CL_FALSE, 0, BUFFER_SIZE,
98 gOut[j], 0, NULL, NULL)))
99 {
100 vlog_error(
101 "\n*** Error %d in clEnqueueWriteBuffer2(%d) ***\n",
102 error, j);
103 return error;
104 }
105 }
106 else
107 {
108 if ((error = clEnqueueFillBuffer(gQueue, gOutBuffer[j],
109 &pattern, sizeof(pattern), 0,
110 BUFFER_SIZE, 0, NULL, NULL)))
111 {
112 vlog_error("Error: clEnqueueFillBuffer failed! err: %d\n",
113 error);
114 return error;
115 }
116 }
117 }
118
119 // Run the kernels
120 for (auto j = gMinVectorSizeIndex; j < gMaxVectorSizeIndex; j++)
121 {
122 size_t vectorSize = sizeValues[j] * sizeof(cl_float);
123 size_t localCount = (BUFFER_SIZE + vectorSize - 1) / vectorSize;
124 if ((error = clSetKernelArg(kernels[j][thread_id], 0,
125 sizeof(gOutBuffer[j]), &gOutBuffer[j])))
126 {
127 LogBuildError(programs[j]);
128 return error;
129 }
130 if ((error = clSetKernelArg(kernels[j][thread_id], 1,
131 sizeof(gInBuffer), &gInBuffer)))
132 {
133 LogBuildError(programs[j]);
134 return error;
135 }
136
137 if ((error = clEnqueueNDRangeKernel(gQueue, kernels[j][thread_id],
138 1, NULL, &localCount, NULL, 0,
139 NULL, NULL)))
140 {
141 vlog_error("FAILED -- could not execute kernel\n");
142 return error;
143 }
144 }
145
146 // Get that moving
147 if ((error = clFlush(gQueue))) vlog("clFlush failed\n");
148
149 // Calculate the correctly rounded reference result
150 float *r = (float *)gOut_Ref;
151 cl_uint *s = (cl_uint *)gIn;
152 for (size_t j = 0; j < BUFFER_SIZE / sizeof(float); j++)
153 r[j] = (float)f->func.f_u(s[j]);
154
155 // Read the data back
156 for (auto j = gMinVectorSizeIndex; j < gMaxVectorSizeIndex; j++)
157 {
158 if ((error =
159 clEnqueueReadBuffer(gQueue, gOutBuffer[j], CL_TRUE, 0,
160 BUFFER_SIZE, gOut[j], 0, NULL, NULL)))
161 {
162 vlog_error("ReadArray failed %d\n", error);
163 return error;
164 }
165 }
166
167 if (gSkipCorrectnessTesting) break;
168
169 // Verify data
170 uint32_t *t = (uint32_t *)gOut_Ref;
171 for (size_t j = 0; j < BUFFER_SIZE / sizeof(float); j++)
172 {
173 for (auto k = gMinVectorSizeIndex; k < gMaxVectorSizeIndex; k++)
174 {
175 uint32_t *q = (uint32_t *)(gOut[k]);
176
177 // If we aren't getting the correctly rounded result
178 if (t[j] != q[j])
179 {
180 float test = ((float *)q)[j];
181 double correct = f->func.f_u(s[j]);
182 float err = Ulp_Error(test, correct);
183 int fail = !(fabsf(err) <= float_ulps);
184
185 if (fail)
186 {
187 if (ftz || relaxedMode)
188 {
189 // retry per section 6.5.3.2
190 if (IsFloatResultSubnormal(correct, float_ulps))
191 {
192 fail = fail && (test != 0.0f);
193 if (!fail) err = 0.0f;
194 }
195 }
196 }
197 if (fabsf(err) > maxError)
198 {
199 maxError = fabsf(err);
200 maxErrorVal = s[j];
201 }
202 if (fail)
203 {
204 vlog_error(
205 "\n%s%s: %f ulp error at 0x%8.8x: *%a vs. %a\n",
206 f->name, sizeNames[k], err, ((uint32_t *)gIn)[j],
207 ((float *)gOut_Ref)[j], test);
208 return -1;
209 }
210 }
211 }
212 }
213
214 if (0 == (i & 0x0fffffff))
215 {
216 if (gVerboseBruteForce)
217 {
218 vlog("base:%14" PRIu64 " step:%10" PRIu64
219 " bufferSize:%10d \n",
220 i, step, BUFFER_SIZE);
221 }
222 else
223 {
224 vlog(".");
225 }
226 fflush(stdout);
227 }
228 }
229
230 if (!gSkipCorrectnessTesting)
231 {
232 if (gWimpyMode)
233 vlog("Wimp pass");
234 else
235 vlog("passed");
236
237 vlog("\t%8.2f @ %a", maxError, maxErrorVal);
238 }
239
240 vlog("\n");
241
242 return CL_SUCCESS;
243 }
244