| // |
| // Copyright (c) 2017 The Khronos Group Inc. |
| // |
| // Licensed under the Apache License, Version 2.0 (the "License"); |
| // you may not use this file except in compliance with the License. |
| // You may obtain a copy of the License at |
| // |
| // http://www.apache.org/licenses/LICENSE-2.0 |
| // |
| // Unless required by applicable law or agreed to in writing, software |
| // distributed under the License is distributed on an "AS IS" BASIS, |
| // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| // See the License for the specific language governing permissions and |
| // limitations under the License. |
| // |
| |
| #include "function_list.h" |
| #include "test_functions.h" |
| #include "utility.h" |
| |
| #include <cstring> |
| |
| static int BuildKernel(const char *name, int vectorSize, cl_kernel *k, |
| cl_program *p, bool relaxedMode) |
| { |
| const char *c[] = { "__kernel void math_kernel", |
| sizeNames[vectorSize], |
| "( __global float", |
| sizeNames[vectorSize], |
| "* out, __global float", |
| sizeNames[vectorSize], |
| "* out2, __global float", |
| sizeNames[vectorSize], |
| "* in )\n" |
| "{\n" |
| " size_t i = get_global_id(0);\n" |
| " out[i] = ", |
| name, |
| "( in[i], out2 + i );\n" |
| "}\n" }; |
| |
| const char *c3[] = { |
| "__kernel void math_kernel", |
| sizeNames[vectorSize], |
| "( __global float* out, __global float* out2, __global float* in)\n" |
| "{\n" |
| " size_t i = get_global_id(0);\n" |
| " if( i + 1 < get_global_size(0) )\n" |
| " {\n" |
| " float3 f0 = vload3( 0, in + 3 * i );\n" |
| " float3 iout = NAN;\n" |
| " f0 = ", |
| name, |
| "( f0, &iout );\n" |
| " vstore3( f0, 0, out + 3*i );\n" |
| " vstore3( iout, 0, out2 + 3*i );\n" |
| " }\n" |
| " else\n" |
| " {\n" |
| " size_t parity = i & 1; // Figure out how many elements are " |
| "left over after BUFFER_SIZE % (3*sizeof(float)). Assume power of two " |
| "buffer size \n" |
| " float3 iout = NAN;\n" |
| " float3 f0;\n" |
| " switch( parity )\n" |
| " {\n" |
| " case 1:\n" |
| " f0 = (float3)( in[3*i], NAN, NAN ); \n" |
| " break;\n" |
| " case 0:\n" |
| " f0 = (float3)( in[3*i], in[3*i+1], NAN ); \n" |
| " break;\n" |
| " }\n" |
| " f0 = ", |
| name, |
| "( f0, &iout );\n" |
| " switch( parity )\n" |
| " {\n" |
| " case 0:\n" |
| " out[3*i+1] = f0.y; \n" |
| " out2[3*i+1] = iout.y; \n" |
| " // fall through\n" |
| " case 1:\n" |
| " out[3*i] = f0.x; \n" |
| " out2[3*i] = iout.x; \n" |
| " break;\n" |
| " }\n" |
| " }\n" |
| "}\n" |
| }; |
| |
| const char **kern = c; |
| size_t kernSize = sizeof(c) / sizeof(c[0]); |
| |
| if (sizeValues[vectorSize] == 3) |
| { |
| kern = c3; |
| kernSize = sizeof(c3) / sizeof(c3[0]); |
| } |
| |
| char testName[32]; |
| snprintf(testName, sizeof(testName) - 1, "math_kernel%s", |
| sizeNames[vectorSize]); |
| |
| return MakeKernel(kern, (cl_uint)kernSize, testName, k, p, relaxedMode); |
| } |
| |
| typedef struct BuildKernelInfo |
| { |
| cl_uint offset; // the first vector size to build |
| cl_kernel *kernels; |
| cl_program *programs; |
| const char *nameInCode; |
| bool relaxedMode; // Whether to build with -cl-fast-relaxed-math. |
| } BuildKernelInfo; |
| |
| static cl_int BuildKernel_FloatFn(cl_uint job_id, cl_uint thread_id UNUSED, |
| void *p) |
| { |
| BuildKernelInfo *info = (BuildKernelInfo *)p; |
| cl_uint i = info->offset + job_id; |
| return BuildKernel(info->nameInCode, i, info->kernels + i, |
| info->programs + i, info->relaxedMode); |
| } |
| |
| int TestFunc_Float2_Float(const Func *f, MTdata d, bool relaxedMode) |
| { |
| uint64_t i; |
| uint32_t j, k; |
| uint32_t l; |
| int error; |
| char const *testing_mode; |
| cl_program programs[VECTOR_SIZE_COUNT]; |
| cl_kernel kernels[VECTOR_SIZE_COUNT]; |
| float maxError0 = 0.0f; |
| float maxError1 = 0.0f; |
| int ftz = f->ftz || gForceFTZ || 0 == (CL_FP_DENORM & gFloatCapabilities); |
| float maxErrorVal0 = 0.0f; |
| float maxErrorVal1 = 0.0f; |
| size_t bufferSize = (gWimpyMode) ? gWimpyBufferSize : BUFFER_SIZE; |
| uint64_t step = getTestStep(sizeof(float), bufferSize); |
| int scale = (int)((1ULL << 32) / (16 * bufferSize / sizeof(float)) + 1); |
| cl_uchar overflow[BUFFER_SIZE / sizeof(float)]; |
| int isFract = 0 == strcmp("fract", f->nameInCode); |
| int skipNanInf = isFract && !gInfNanSupport; |
| |
| logFunctionInfo(f->name, sizeof(cl_float), relaxedMode); |
| |
| float float_ulps = getAllowedUlpError(f, relaxedMode); |
| // Init the kernels |
| { |
| BuildKernelInfo build_info = { gMinVectorSizeIndex, kernels, programs, |
| f->nameInCode, relaxedMode }; |
| if ((error = ThreadPool_Do(BuildKernel_FloatFn, |
| gMaxVectorSizeIndex - gMinVectorSizeIndex, |
| &build_info))) |
| return error; |
| } |
| |
| for (i = 0; i < (1ULL << 32); i += step) |
| { |
| // Init input array |
| uint32_t *p = (uint32_t *)gIn; |
| if (gWimpyMode) |
| { |
| for (j = 0; j < bufferSize / sizeof(float); j++) |
| { |
| p[j] = (uint32_t)i + j * scale; |
| if (relaxedMode && strcmp(f->name, "sincos") == 0) |
| { |
| float pj = *(float *)&p[j]; |
| if (fabs(pj) > M_PI) ((float *)p)[j] = NAN; |
| } |
| } |
| } |
| else |
| { |
| for (j = 0; j < bufferSize / sizeof(float); j++) |
| { |
| p[j] = (uint32_t)i + j; |
| if (relaxedMode && strcmp(f->name, "sincos") == 0) |
| { |
| float pj = *(float *)&p[j]; |
| if (fabs(pj) > M_PI) ((float *)p)[j] = NAN; |
| } |
| } |
| } |
| |
| if ((error = clEnqueueWriteBuffer(gQueue, gInBuffer, CL_FALSE, 0, |
| bufferSize, gIn, 0, NULL, NULL))) |
| { |
| vlog_error("\n*** Error %d in clEnqueueWriteBuffer ***\n", error); |
| return error; |
| } |
| |
| // write garbage into output arrays |
| for (j = gMinVectorSizeIndex; j < gMaxVectorSizeIndex; j++) |
| { |
| uint32_t pattern = 0xffffdead; |
| memset_pattern4(gOut[j], &pattern, bufferSize); |
| if ((error = |
| clEnqueueWriteBuffer(gQueue, gOutBuffer[j], CL_FALSE, 0, |
| bufferSize, gOut[j], 0, NULL, NULL))) |
| { |
| vlog_error("\n*** Error %d in clEnqueueWriteBuffer2(%d) ***\n", |
| error, j); |
| goto exit; |
| } |
| |
| memset_pattern4(gOut2[j], &pattern, bufferSize); |
| if ((error = |
| clEnqueueWriteBuffer(gQueue, gOutBuffer2[j], CL_FALSE, 0, |
| bufferSize, gOut2[j], 0, NULL, NULL))) |
| { |
| vlog_error("\n*** Error %d in clEnqueueWriteBuffer2b(%d) ***\n", |
| error, j); |
| goto exit; |
| } |
| } |
| |
| // Run the kernels |
| for (j = gMinVectorSizeIndex; j < gMaxVectorSizeIndex; j++) |
| { |
| size_t vectorSize = sizeValues[j] * sizeof(cl_float); |
| size_t localCount = (bufferSize + vectorSize - 1) / vectorSize; |
| if ((error = clSetKernelArg(kernels[j], 0, sizeof(gOutBuffer[j]), |
| &gOutBuffer[j]))) |
| { |
| LogBuildError(programs[j]); |
| goto exit; |
| } |
| if ((error = clSetKernelArg(kernels[j], 1, sizeof(gOutBuffer2[j]), |
| &gOutBuffer2[j]))) |
| { |
| LogBuildError(programs[j]); |
| goto exit; |
| } |
| if ((error = clSetKernelArg(kernels[j], 2, sizeof(gInBuffer), |
| &gInBuffer))) |
| { |
| LogBuildError(programs[j]); |
| goto exit; |
| } |
| |
| if ((error = |
| clEnqueueNDRangeKernel(gQueue, kernels[j], 1, NULL, |
| &localCount, NULL, 0, NULL, NULL))) |
| { |
| vlog_error("FAILED -- could not execute kernel\n"); |
| goto exit; |
| } |
| } |
| |
| // Get that moving |
| if ((error = clFlush(gQueue))) vlog("clFlush failed\n"); |
| |
| FPU_mode_type oldMode; |
| RoundingMode oldRoundMode = kRoundToNearestEven; |
| if (isFract) |
| { |
| // Calculate the correctly rounded reference result |
| memset(&oldMode, 0, sizeof(oldMode)); |
| if (ftz) ForceFTZ(&oldMode); |
| |
| // Set the rounding mode to match the device |
| if (gIsInRTZMode) |
| oldRoundMode = set_round(kRoundTowardZero, kfloat); |
| } |
| |
| // Calculate the correctly rounded reference result |
| float *r = (float *)gOut_Ref; |
| float *r2 = (float *)gOut_Ref2; |
| float *s = (float *)gIn; |
| |
| if (skipNanInf) |
| { |
| for (j = 0; j < bufferSize / sizeof(float); j++) |
| { |
| double dd; |
| feclearexcept(FE_OVERFLOW); |
| |
| if (relaxedMode) |
| r[j] = (float)f->rfunc.f_fpf(s[j], &dd); |
| else |
| r[j] = (float)f->func.f_fpf(s[j], &dd); |
| |
| r2[j] = (float)dd; |
| overflow[j] = |
| FE_OVERFLOW == (FE_OVERFLOW & fetestexcept(FE_OVERFLOW)); |
| } |
| } |
| else |
| { |
| for (j = 0; j < bufferSize / sizeof(float); j++) |
| { |
| double dd; |
| if (relaxedMode) |
| r[j] = (float)f->rfunc.f_fpf(s[j], &dd); |
| else |
| r[j] = (float)f->func.f_fpf(s[j], &dd); |
| |
| r2[j] = (float)dd; |
| } |
| } |
| |
| if (isFract && ftz) RestoreFPState(&oldMode); |
| |
| // Read the data back |
| for (j = gMinVectorSizeIndex; j < gMaxVectorSizeIndex; j++) |
| { |
| if ((error = |
| clEnqueueReadBuffer(gQueue, gOutBuffer[j], CL_TRUE, 0, |
| bufferSize, gOut[j], 0, NULL, NULL))) |
| { |
| vlog_error("ReadArray failed %d\n", error); |
| goto exit; |
| } |
| if ((error = |
| clEnqueueReadBuffer(gQueue, gOutBuffer2[j], CL_TRUE, 0, |
| bufferSize, gOut2[j], 0, NULL, NULL))) |
| { |
| vlog_error("ReadArray2 failed %d\n", error); |
| goto exit; |
| } |
| } |
| |
| if (gSkipCorrectnessTesting) |
| { |
| if (isFract && gIsInRTZMode) (void)set_round(oldRoundMode, kfloat); |
| break; |
| } |
| |
| // Verify data |
| uint32_t *t = (uint32_t *)gOut_Ref; |
| uint32_t *t2 = (uint32_t *)gOut_Ref2; |
| for (j = 0; j < bufferSize / sizeof(float); j++) |
| { |
| for (k = gMinVectorSizeIndex; k < gMaxVectorSizeIndex; k++) |
| { |
| uint32_t *q = (uint32_t *)gOut[k]; |
| uint32_t *q2 = (uint32_t *)gOut2[k]; |
| |
| // If we aren't getting the correctly rounded result |
| if (t[j] != q[j] || t2[j] != q2[j]) |
| { |
| double correct, correct2; |
| float err, err2; |
| float test = ((float *)q)[j]; |
| float test2 = ((float *)q2)[j]; |
| |
| if (relaxedMode) |
| correct = f->rfunc.f_fpf(s[j], &correct2); |
| else |
| correct = f->func.f_fpf(s[j], &correct2); |
| |
| // Per section 10 paragraph 6, accept any result if an input |
| // or output is a infinity or NaN or overflow |
| if (relaxedMode || skipNanInf) |
| { |
| if (skipNanInf && overflow[j]) continue; |
| // Note: no double rounding here. Reference functions |
| // calculate in single precision. |
| if (IsFloatInfinity(correct) || IsFloatNaN(correct) |
| || IsFloatInfinity(correct2) || IsFloatNaN(correct2) |
| || IsFloatInfinity(s[j]) || IsFloatNaN(s[j])) |
| continue; |
| } |
| |
| typedef int (*CheckForSubnormal)( |
| double, float); // If we are in fast relaxed math, we |
| // have a different calculation for the |
| // subnormal threshold. |
| CheckForSubnormal isFloatResultSubnormalPtr; |
| if (relaxedMode) |
| { |
| err = Abs_Error(test, correct); |
| err2 = Abs_Error(test2, correct2); |
| isFloatResultSubnormalPtr = |
| &IsFloatResultSubnormalAbsError; |
| } |
| else |
| { |
| err = Ulp_Error(test, correct); |
| err2 = Ulp_Error(test2, correct2); |
| isFloatResultSubnormalPtr = &IsFloatResultSubnormal; |
| } |
| int fail = !(fabsf(err) <= float_ulps |
| && fabsf(err2) <= float_ulps); |
| |
| if (ftz) |
| { |
| // retry per section 6.5.3.2 |
| if ((*isFloatResultSubnormalPtr)(correct, float_ulps)) |
| { |
| if ((*isFloatResultSubnormalPtr)(correct2, |
| float_ulps)) |
| { |
| fail = fail && !(test == 0.0f && test2 == 0.0f); |
| if (!fail) |
| { |
| err = 0.0f; |
| err2 = 0.0f; |
| } |
| } |
| else |
| { |
| fail = fail |
| && !(test == 0.0f |
| && fabsf(err2) <= float_ulps); |
| if (!fail) err = 0.0f; |
| } |
| } |
| else if ((*isFloatResultSubnormalPtr)(correct2, |
| float_ulps)) |
| { |
| fail = fail |
| && !(test2 == 0.0f && fabsf(err) <= float_ulps); |
| if (!fail) err2 = 0.0f; |
| } |
| |
| |
| // retry per section 6.5.3.3 |
| if (IsFloatSubnormal(s[j])) |
| { |
| double correctp, correctn; |
| double correct2p, correct2n; |
| float errp, err2p, errn, err2n; |
| |
| if (skipNanInf) feclearexcept(FE_OVERFLOW); |
| if (relaxedMode) |
| { |
| correctp = f->rfunc.f_fpf(0.0, &correct2p); |
| correctn = f->rfunc.f_fpf(-0.0, &correct2n); |
| } |
| else |
| { |
| correctp = f->func.f_fpf(0.0, &correct2p); |
| correctn = f->func.f_fpf(-0.0, &correct2n); |
| } |
| |
| // Per section 10 paragraph 6, accept any result if |
| // an input or output is a infinity or NaN or |
| // overflow |
| if (skipNanInf) |
| { |
| if (fetestexcept(FE_OVERFLOW)) continue; |
| |
| // Note: no double rounding here. Reference |
| // functions calculate in single precision. |
| if (IsFloatInfinity(correctp) |
| || IsFloatNaN(correctp) |
| || IsFloatInfinity(correctn) |
| || IsFloatNaN(correctn) |
| || IsFloatInfinity(correct2p) |
| || IsFloatNaN(correct2p) |
| || IsFloatInfinity(correct2n) |
| || IsFloatNaN(correct2n)) |
| continue; |
| } |
| |
| if (relaxedMode) |
| { |
| errp = Abs_Error(test, correctp); |
| err2p = Abs_Error(test, correct2p); |
| errn = Abs_Error(test, correctn); |
| err2n = Abs_Error(test, correct2n); |
| } |
| else |
| { |
| errp = Ulp_Error(test, correctp); |
| err2p = Ulp_Error(test, correct2p); |
| errn = Ulp_Error(test, correctn); |
| err2n = Ulp_Error(test, correct2n); |
| } |
| |
| fail = fail |
| && ((!(fabsf(errp) <= float_ulps)) |
| && (!(fabsf(err2p) <= float_ulps)) |
| && ((!(fabsf(errn) <= float_ulps)) |
| && (!(fabsf(err2n) <= float_ulps)))); |
| if (fabsf(errp) < fabsf(err)) err = errp; |
| if (fabsf(errn) < fabsf(err)) err = errn; |
| if (fabsf(err2p) < fabsf(err2)) err2 = err2p; |
| if (fabsf(err2n) < fabsf(err2)) err2 = err2n; |
| |
| // retry per section 6.5.3.4 |
| if ((*isFloatResultSubnormalPtr)(correctp, |
| float_ulps) |
| || (*isFloatResultSubnormalPtr)(correctn, |
| float_ulps)) |
| { |
| if ((*isFloatResultSubnormalPtr)(correct2p, |
| float_ulps) |
| || (*isFloatResultSubnormalPtr)(correct2n, |
| float_ulps)) |
| { |
| fail = fail |
| && !(test == 0.0f && test2 == 0.0f); |
| if (!fail) err = err2 = 0.0f; |
| } |
| else |
| { |
| fail = fail |
| && !(test == 0.0f |
| && fabsf(err2) <= float_ulps); |
| if (!fail) err = 0.0f; |
| } |
| } |
| else if ((*isFloatResultSubnormalPtr)(correct2p, |
| float_ulps) |
| || (*isFloatResultSubnormalPtr)( |
| correct2n, float_ulps)) |
| { |
| fail = fail |
| && !(test2 == 0.0f |
| && (fabsf(err) <= float_ulps)); |
| if (!fail) err2 = 0.0f; |
| } |
| } |
| } |
| if (fabsf(err) > maxError0) |
| { |
| maxError0 = fabsf(err); |
| maxErrorVal0 = s[j]; |
| } |
| if (fabsf(err2) > maxError1) |
| { |
| maxError1 = fabsf(err2); |
| maxErrorVal1 = s[j]; |
| } |
| if (fail) |
| { |
| vlog_error("\nERROR: %s%s: {%f, %f} ulp error at %a: " |
| "*{%a, %a} vs. {%a, %a}\n", |
| f->name, sizeNames[k], err, err2, |
| ((float *)gIn)[j], ((float *)gOut_Ref)[j], |
| ((float *)gOut_Ref2)[j], test, test2); |
| error = -1; |
| goto exit; |
| } |
| } |
| } |
| } |
| |
| if (isFract && gIsInRTZMode) (void)set_round(oldRoundMode, kfloat); |
| |
| if (0 == (i & 0x0fffffff)) |
| { |
| if (gVerboseBruteForce) |
| { |
| vlog("base:%14u step:%10zu bufferSize:%10zd \n", i, step, |
| bufferSize); |
| } |
| else |
| { |
| vlog("."); |
| } |
| fflush(stdout); |
| } |
| } |
| |
| if (!gSkipCorrectnessTesting) |
| { |
| if (gWimpyMode) |
| vlog("Wimp pass"); |
| else |
| vlog("passed"); |
| } |
| |
| if (gMeasureTimes) |
| { |
| // Init input array |
| uint32_t *p = (uint32_t *)gIn; |
| for (j = 0; j < bufferSize / sizeof(float); j++) |
| p[j] = genrand_int32(d); |
| |
| if ((error = clEnqueueWriteBuffer(gQueue, gInBuffer, CL_FALSE, 0, |
| bufferSize, gIn, 0, NULL, NULL))) |
| { |
| vlog_error("\n*** Error %d in clEnqueueWriteBuffer ***\n", error); |
| return error; |
| } |
| |
| |
| // Run the kernels |
| for (j = gMinVectorSizeIndex; j < gMaxVectorSizeIndex; j++) |
| { |
| size_t vectorSize = sizeValues[j] * sizeof(cl_float); |
| size_t localCount = (bufferSize + vectorSize - 1) / vectorSize; |
| if ((error = clSetKernelArg(kernels[j], 0, sizeof(gOutBuffer[j]), |
| &gOutBuffer[j]))) |
| { |
| LogBuildError(programs[j]); |
| goto exit; |
| } |
| if ((error = clSetKernelArg(kernels[j], 1, sizeof(gOutBuffer2[j]), |
| &gOutBuffer2[j]))) |
| { |
| LogBuildError(programs[j]); |
| goto exit; |
| } |
| if ((error = clSetKernelArg(kernels[j], 2, sizeof(gInBuffer), |
| &gInBuffer))) |
| { |
| LogBuildError(programs[j]); |
| goto exit; |
| } |
| |
| double sum = 0.0; |
| double bestTime = INFINITY; |
| for (k = 0; k < PERF_LOOP_COUNT; k++) |
| { |
| uint64_t startTime = GetTime(); |
| if ((error = clEnqueueNDRangeKernel(gQueue, kernels[j], 1, NULL, |
| &localCount, NULL, 0, NULL, |
| NULL))) |
| { |
| vlog_error("FAILED -- could not execute kernel\n"); |
| goto exit; |
| } |
| |
| // Make sure OpenCL is done |
| if ((error = clFinish(gQueue))) |
| { |
| vlog_error("Error %d at clFinish\n", error); |
| goto exit; |
| } |
| |
| uint64_t endTime = GetTime(); |
| double time = SubtractTime(endTime, startTime); |
| sum += time; |
| if (time < bestTime) bestTime = time; |
| } |
| |
| if (gReportAverageTimes) bestTime = sum / PERF_LOOP_COUNT; |
| double clocksPerOp = bestTime * (double)gDeviceFrequency |
| * gComputeDevices * gSimdSize * 1e6 |
| / (bufferSize / sizeof(float)); |
| vlog_perf(clocksPerOp, LOWER_IS_BETTER, "clocks / element", "%sf%s", |
| f->name, sizeNames[j]); |
| } |
| } |
| |
| if (!gSkipCorrectnessTesting) |
| vlog("\t{%8.2f, %8.2f} @ {%a, %a}", maxError0, maxError1, maxErrorVal0, |
| maxErrorVal1); |
| vlog("\n"); |
| |
| exit: |
| // Release |
| for (k = gMinVectorSizeIndex; k < gMaxVectorSizeIndex; k++) |
| { |
| clReleaseKernel(kernels[k]); |
| clReleaseProgram(programs[k]); |
| } |
| |
| return error; |
| } |
