OpenCL-CTS/test_conformance/math_brute_force/mad_double.cpp

//
// 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 "common.h"
#include "function_list.h"
#include "test_functions.h"
#include "utility.h"

#include <cstring>

namespace {

int BuildKernel(const char *name, int vectorSize, cl_kernel *k, cl_program *p,
                bool relaxedMode)
{
    auto kernel_name = GetKernelName(vectorSize);
    auto source = GetTernaryKernel(kernel_name, name, ParameterType::Double,
                                   ParameterType::Double, ParameterType::Double,
                                   ParameterType::Double, vectorSize);
    std::array<const char *, 1> sources{ source.c_str() };
    return MakeKernel(sources.data(), sources.size(), kernel_name.c_str(), k, p,
                      relaxedMode);
}

using Kernels = std::array<clKernelWrapper, VECTOR_SIZE_COUNT>;

struct BuildKernelInfo2
{
    Kernels &kernels;
    Programs &programs;
    const char *nameInCode;
    bool relaxedMode; // Whether to build with -cl-fast-relaxed-math.
};

cl_int BuildKernelFn(cl_uint job_id, cl_uint thread_id UNUSED, void *p)
{
    BuildKernelInfo2 *info = (BuildKernelInfo2 *)p;
    cl_uint vectorSize = gMinVectorSizeIndex + job_id;
    return BuildKernel(info->nameInCode, vectorSize,
                       &(info->kernels[vectorSize]),
                       &(info->programs[vectorSize]), info->relaxedMode);
}

} // anonymous namespace

int TestFunc_mad_Double(const Func *f, MTdata d, bool relaxedMode)
{
    int error;
    Programs programs;
    Kernels kernels;
    float maxError = 0.0f;
    double maxErrorVal = 0.0f;
    double maxErrorVal2 = 0.0f;
    double maxErrorVal3 = 0.0f;
    uint64_t step = getTestStep(sizeof(double), BUFFER_SIZE);

    logFunctionInfo(f->name, sizeof(cl_double), relaxedMode);

    // Init the kernels
    BuildKernelInfo2 build_info{ kernels, programs, f->nameInCode,
                                 relaxedMode };
    if ((error = ThreadPool_Do(BuildKernelFn,
                               gMaxVectorSizeIndex - gMinVectorSizeIndex,
                               &build_info)))
        return error;

    for (uint64_t i = 0; i < (1ULL << 32); i += step)
    {
        // Init input array
        double *p = (double *)gIn;
        double *p2 = (double *)gIn2;
        double *p3 = (double *)gIn3;
        for (size_t j = 0; j < BUFFER_SIZE / sizeof(double); j++)
        {
            p[j] = DoubleFromUInt32(genrand_int32(d));
            p2[j] = DoubleFromUInt32(genrand_int32(d));
            p3[j] = DoubleFromUInt32(genrand_int32(d));
        }

        if ((error = clEnqueueWriteBuffer(gQueue, gInBuffer, CL_FALSE, 0,
                                          BUFFER_SIZE, gIn, 0, NULL, NULL)))
        {
            vlog_error("\n*** Error %d in clEnqueueWriteBuffer ***\n", error);
            return error;
        }

        if ((error = clEnqueueWriteBuffer(gQueue, gInBuffer2, CL_FALSE, 0,
                                          BUFFER_SIZE, gIn2, 0, NULL, NULL)))
        {
            vlog_error("\n*** Error %d in clEnqueueWriteBuffer2 ***\n", error);
            return error;
        }

        if ((error = clEnqueueWriteBuffer(gQueue, gInBuffer3, CL_FALSE, 0,
                                          BUFFER_SIZE, gIn3, 0, NULL, NULL)))
        {
            vlog_error("\n*** Error %d in clEnqueueWriteBuffer3 ***\n", error);
            return error;
        }

        // Write garbage into output arrays
        for (auto j = gMinVectorSizeIndex; j < gMaxVectorSizeIndex; j++)
        {
            uint32_t pattern = 0xffffdead;
            if (gHostFill)
            {
                memset_pattern4(gOut[j], &pattern, BUFFER_SIZE);
                if ((error = clEnqueueWriteBuffer(gQueue, gOutBuffer[j],
                                                  CL_FALSE, 0, BUFFER_SIZE,
                                                  gOut[j], 0, NULL, NULL)))
                {
                    vlog_error(
                        "\n*** Error %d in clEnqueueWriteBuffer2(%d) ***\n",
                        error, j);
                    return error;
                }
            }
            else
            {
                if ((error = clEnqueueFillBuffer(gQueue, gOutBuffer[j],
                                                 &pattern, sizeof(pattern), 0,
                                                 BUFFER_SIZE, 0, NULL, NULL)))
                {
                    vlog_error("Error: clEnqueueFillBuffer failed! err: %d\n",
                               error);
                    return error;
                }
            }
        }

        // Run the kernels
        for (auto j = gMinVectorSizeIndex; j < gMaxVectorSizeIndex; j++)
        {
            size_t vectorSize = sizeof(cl_double) * sizeValues[j];
            size_t localCount = (BUFFER_SIZE + vectorSize - 1)
                / vectorSize; // BUFFER_SIZE / vectorSize  rounded up
            if ((error = clSetKernelArg(kernels[j], 0, sizeof(gOutBuffer[j]),
                                        &gOutBuffer[j])))
            {
                LogBuildError(programs[j]);
                return error;
            }
            if ((error = clSetKernelArg(kernels[j], 1, sizeof(gInBuffer),
                                        &gInBuffer)))
            {
                LogBuildError(programs[j]);
                return error;
            }
            if ((error = clSetKernelArg(kernels[j], 2, sizeof(gInBuffer2),
                                        &gInBuffer2)))
            {
                LogBuildError(programs[j]);
                return error;
            }
            if ((error = clSetKernelArg(kernels[j], 3, sizeof(gInBuffer3),
                                        &gInBuffer3)))
            {
                LogBuildError(programs[j]);
                return error;
            }

            if ((error =
                     clEnqueueNDRangeKernel(gQueue, kernels[j], 1, NULL,
                                            &localCount, NULL, 0, NULL, NULL)))
            {
                vlog_error("FAILED -- could not execute kernel\n");
                return error;
            }
        }

        // Get that moving
        if ((error = clFlush(gQueue))) vlog("clFlush failed\n");

        // Calculate the correctly rounded reference result
        double *r = (double *)gOut_Ref;
        double *s = (double *)gIn;
        double *s2 = (double *)gIn2;
        double *s3 = (double *)gIn3;
        for (size_t j = 0; j < BUFFER_SIZE / sizeof(double); j++)
            r[j] = (double)f->dfunc.f_fff(s[j], s2[j], s3[j]);

        // Read the data back
        for (auto j = gMinVectorSizeIndex; j < gMaxVectorSizeIndex; j++)
        {
            if ((error =
                     clEnqueueReadBuffer(gQueue, gOutBuffer[j], CL_TRUE, 0,
                                         BUFFER_SIZE, gOut[j], 0, NULL, NULL)))
            {
                vlog_error("ReadArray failed %d\n", error);
                return error;
            }
        }

        if (gSkipCorrectnessTesting) break;

        // Verify data -- No verification possible.
        // MAD is a random number generator.
        if (0 == (i & 0x0fffffff))
        {
            vlog(".");
            fflush(stdout);
        }
    }

    if (!gSkipCorrectnessTesting)
    {
        if (gWimpyMode)
            vlog("Wimp pass");
        else
            vlog("passed");

        vlog("\t%8.2f @ {%a, %a, %a}", maxError, maxErrorVal, maxErrorVal2,
             maxErrorVal3);
    }

    vlog("\n");

    return CL_SUCCESS;
}