//
// 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 "testBase.h"
#include "gl/setup.h"
#include "harness/genericThread.h"

#if defined( __APPLE__ )
#include <OpenGL/glu.h>
#else
#include <GL/glu.h>
#include <CL/cl_gl.h>
#if !defined(_WIN32) && !defined(__ANDROID__)
#include <GL/glx.h>
#endif
#endif

#ifndef GLsync
// For OpenGL before 3.2, we look for the ARB_sync extension and try to use that
#if !defined(_WIN32)
#include <inttypes.h>
#endif // !_WIN32
typedef int64_t GLint64;
typedef uint64_t GLuint64;
typedef struct __GLsync *GLsync;

#ifndef APIENTRY
#define APIENTRY
#endif

typedef GLsync (APIENTRY *glFenceSyncPtr)(GLenum condition,GLbitfield flags);
glFenceSyncPtr glFenceSyncFunc;

typedef bool (APIENTRY *glIsSyncPtr)(GLsync sync);
glIsSyncPtr glIsSyncFunc;

typedef void (APIENTRY *glDeleteSyncPtr)(GLsync sync);
glDeleteSyncPtr glDeleteSyncFunc;

typedef GLenum (APIENTRY *glClientWaitSyncPtr)(GLsync sync,GLbitfield flags,GLuint64 timeout);
glClientWaitSyncPtr glClientWaitSyncFunc;

typedef void (APIENTRY *glWaitSyncPtr)(GLsync sync,GLbitfield flags,GLuint64 timeout);
glWaitSyncPtr glWaitSyncFunc;

typedef void (APIENTRY *glGetInteger64vPtr)(GLenum pname, GLint64 *params);
glGetInteger64vPtr glGetInteger64vFunc;

typedef void (APIENTRY *glGetSyncivPtr)(GLsync sync,GLenum pname,GLsizei bufSize,GLsizei *length,
                               GLint *values);
glGetSyncivPtr glGetSyncivFunc;

#define CHK_GL_ERR() printf("%s\n", gluErrorString(glGetError()))

static void InitSyncFns( void )
{
    glFenceSyncFunc = (glFenceSyncPtr)glutGetProcAddress( "glFenceSync" );
    glIsSyncFunc = (glIsSyncPtr)glutGetProcAddress( "glIsSync" );
    glDeleteSyncFunc = (glDeleteSyncPtr)glutGetProcAddress( "glDeleteSync" );
    glClientWaitSyncFunc = (glClientWaitSyncPtr)glutGetProcAddress( "glClientWaitSync" );
    glWaitSyncFunc = (glWaitSyncPtr)glutGetProcAddress( "glWaitSync" );
    glGetInteger64vFunc = (glGetInteger64vPtr)glutGetProcAddress( "glGetInteger64v" );
    glGetSyncivFunc = (glGetSyncivPtr)glutGetProcAddress( "glGetSynciv" );
}
#ifndef GL_ARB_sync
#define GL_MAX_SERVER_WAIT_TIMEOUT        0x9111

#define GL_OBJECT_TYPE            0x9112
#define GL_SYNC_CONDITION            0x9113
#define GL_SYNC_STATUS            0x9114
#define GL_SYNC_FLAGS            0x9115

#define GL_SYNC_FENCE            0x9116

#define GL_SYNC_GPU_COMMANDS_COMPLETE    0x9117

#define GL_UNSIGNALED            0x9118
#define GL_SIGNALED            0x9119

#define GL_SYNC_FLUSH_COMMANDS_BIT        0x00000001

#define GL_TIMEOUT_IGNORED            0xFFFFFFFFFFFFFFFFull

#define GL_ALREADY_SIGNALED        0x911A
#define GL_TIMEOUT_EXPIRED            0x911B
#define GL_CONDITION_SATISFIED        0x911C
#define GL_WAIT_FAILED            0x911D
#endif

#define USING_ARB_sync 1
#endif

typedef cl_event (CL_API_CALL *clCreateEventFromGLsyncKHR_fn)( cl_context context, GLsync sync, cl_int *errCode_ret) ;

clCreateEventFromGLsyncKHR_fn clCreateEventFromGLsyncKHR_ptr;


static const char *updateBuffersKernel[] = {
    "__kernel void update( __global float4 * vertices, __global float4 *colors, int horizWrap, int rowIdx )\n"
    "{\n"
    "    size_t tid = get_global_id(0);\n"
    "\n"
    "    size_t xVal = ( tid & ( horizWrap - 1 ) );\n"
    "    vertices[ tid * 2 + 0 ] = (float4)( xVal, rowIdx*16.f, 0.0f, 1.f );\n"
    "    vertices[ tid * 2 + 1 ] = (float4)( xVal, rowIdx*16.f + 4.0f, 0.0f, 1.f );\n"
    "\n"
    "    int rowV = rowIdx + 1;\n"
    "    colors[ tid * 2 + 0 ] = (float4)( ( rowV & 1 ) / 255.f, ( ( rowV & 2 ) >> 1 ) / 255.f, ( ( rowV & 4 ) >> 2 ) / 255.f, 1.f );\n"
    "    //colors[ tid * 2 + 0 ] = (float4)( (float)xVal/(float)horizWrap, 1.0f, 1.0f, 1.0f );\n"
    "    colors[ tid * 2 + 1 ] = colors[ tid * 2 + 0 ];\n"
    "}\n" };

//Passthrough VertexShader
static const char *vertexshader =
"#version 150\n"
"uniform mat4 projMatrix;\n"
"in vec4 inPosition;\n"
"in vec4 inColor;\n"
"out vec4 vertColor;\n"
"void main (void) {\n"
"    gl_Position = projMatrix*inPosition;\n"
"   vertColor = inColor;\n"
"}\n";

//Passthrough FragmentShader
static const char *fragmentshader =
"#version 150\n"
"in vec4 vertColor;\n"
"out vec4 outColor;\n"
"void main (void) {\n"
"    outColor = vertColor;\n"
"}\n";

GLuint createShaderProgram(GLint *posLoc, GLint *colLoc)
{
    GLint  logLength, status;
    GLuint program = glCreateProgram();
    GLuint vpShader;

    vpShader = glCreateShader(GL_VERTEX_SHADER);
    glShaderSource(vpShader, 1, (const GLchar **)&vertexshader, NULL);
    glCompileShader(vpShader);
    glGetShaderiv(vpShader, GL_INFO_LOG_LENGTH, &logLength);
    if (logLength > 0) {
        GLchar *log = (GLchar*) malloc(logLength);
        glGetShaderInfoLog(vpShader, logLength, &logLength, log);
        log_info("Vtx Shader compile log:\n%s", log);
        free(log);
    }

    glGetShaderiv(vpShader, GL_COMPILE_STATUS, &status);
    if (status == 0)
    {
        log_error("Failed to compile vtx shader:\n");
        return 0;
    }

    glAttachShader(program, vpShader);

    GLuint fpShader;
    fpShader = glCreateShader(GL_FRAGMENT_SHADER);
    glShaderSource(fpShader, 1, (const GLchar **)&fragmentshader, NULL);
    glCompileShader(fpShader);

    glGetShaderiv(fpShader, GL_INFO_LOG_LENGTH, &logLength);
    if (logLength > 0) {
        GLchar *log = (GLchar*)malloc(logLength);
        glGetShaderInfoLog(fpShader, logLength, &logLength, log);
        log_info("Frag Shader compile log:\n%s", log);
        free(log);
    }

    glAttachShader(program, fpShader);
    glGetShaderiv(fpShader, GL_COMPILE_STATUS, &status);
    if (status == 0)
    {
        log_error("Failed to compile frag shader:\n\n");
        return 0;
    }

    glLinkProgram(program);
    glGetProgramiv(program, GL_INFO_LOG_LENGTH, &logLength);
    if (logLength > 0) {
        GLchar *log = (GLchar*)malloc(logLength);
        glGetProgramInfoLog(program, logLength, &logLength, log);
        log_info("Program link log:\n%s", log);
        free(log);
    }

    glGetProgramiv(program, GL_LINK_STATUS, &status);
    if (status == 0)
    {
        log_error("Failed to link program\n");
        return 0;
    }

    *posLoc = glGetAttribLocation(program, "inPosition");
    *colLoc = glGetAttribLocation(program, "inColor");

    return program;
}

void destroyShaderProgram(GLuint program)
{
    GLuint shaders[2];
    GLsizei count;
    glUseProgram(0);
    glGetAttachedShaders(program, 2, &count, shaders);
    int i;
    for(i = 0; i < count; i++)
    {
        glDetachShader(program, shaders[i]);
        glDeleteShader(shaders[i]);
    }
    glDeleteProgram(program);
}

// This function queues up and runs the above CL kernel that writes the vertex data
cl_int run_cl_kernel( cl_kernel kernel, cl_command_queue queue, cl_mem stream0, cl_mem stream1,
                     cl_int rowIdx, cl_event fenceEvent, size_t numThreads )
{
    cl_int error = clSetKernelArg( kernel, 3, sizeof( rowIdx ), &rowIdx );
    test_error( error, "Unable to set kernel arguments" );

    clEventWrapper acqEvent1, acqEvent2, kernEvent, relEvent1, relEvent2;
    int numEvents = ( fenceEvent != NULL ) ? 1 : 0;
    cl_event *fence_evt = ( fenceEvent != NULL ) ? &fenceEvent : NULL;

    error = (*clEnqueueAcquireGLObjects_ptr)( queue, 1, &stream0, numEvents, fence_evt, &acqEvent1 );
    test_error( error, "Unable to acquire GL obejcts");
    error = (*clEnqueueAcquireGLObjects_ptr)( queue, 1, &stream1, numEvents, fence_evt, &acqEvent2 );
    test_error( error, "Unable to acquire GL obejcts");

    cl_event evts[ 2 ] = { acqEvent1, acqEvent2 };

    error = clEnqueueNDRangeKernel( queue, kernel, 1, NULL, &numThreads, NULL, 2, evts, &kernEvent );
    test_error( error, "Unable to execute test kernel" );

    error = (*clEnqueueReleaseGLObjects_ptr)( queue, 1, &stream0, 1, &kernEvent, &relEvent1 );
    test_error(error, "clEnqueueReleaseGLObjects failed");
    error = (*clEnqueueReleaseGLObjects_ptr)( queue, 1, &stream1, 1, &kernEvent, &relEvent2 );
    test_error(error, "clEnqueueReleaseGLObjects failed");

    evts[ 0 ] = relEvent1;
    evts[ 1 ] = relEvent2;
    error = clWaitForEvents( 2, evts );
    test_error( error, "Unable to wait for release events" );

    return 0;
}

class RunThread : public genericThread
{
public:

    cl_kernel mKernel;
    cl_command_queue mQueue;
    cl_mem mStream0, mStream1;
    cl_int mRowIdx;
    cl_event mFenceEvent;
    size_t mNumThreads;

    RunThread( cl_kernel kernel, cl_command_queue queue, cl_mem stream0, cl_mem stream1, size_t numThreads )
    : mKernel( kernel ), mQueue( queue ), mStream0( stream0 ), mStream1( stream1 ), mNumThreads( numThreads )
    {
    }

    void SetRunData( cl_int rowIdx, cl_event fenceEvent )
    {
        mRowIdx = rowIdx;
        mFenceEvent = fenceEvent;
    }

    virtual void * IRun( void )
    {
        cl_int error = run_cl_kernel( mKernel, mQueue, mStream0, mStream1, mRowIdx, mFenceEvent, mNumThreads );
        return (void *)(uintptr_t)error;
    }
};


int test_fence_sync_single( cl_device_id device, cl_context context, cl_command_queue queue, bool separateThreads, GLint rend_vs, GLint read_vs, cl_device_id rend_device )
{
    int error;
    const int framebufferSize = 512;


    if( !is_extension_available( device, "cl_khr_gl_event" ) )
    {
        log_info( "NOTE: cl_khr_gl_event extension not present on this device; skipping fence sync test\n" );
        return 0;
    }

    // Ask OpenCL for the platforms.  Warn if more than one platform found,
    // since this might not be the platform we want.  By default, we simply
    // use the first returned platform.

    cl_uint nplatforms;
    cl_platform_id platform;
    clGetPlatformIDs(0, NULL, &nplatforms);
    clGetPlatformIDs(1, &platform, NULL);

    if (nplatforms > 1) {
        log_info("clGetPlatformIDs returned multiple values.  This is not "
            "an error, but might result in obtaining incorrect function "
            "pointers if you do not want the first returned platform.\n");

        // Show them the platform name, in case it is a problem.

        size_t size;
        char *name;

        clGetPlatformInfo(platform, CL_PLATFORM_NAME, 0, NULL, &size);
        name = (char*)malloc(size);
        clGetPlatformInfo(platform, CL_PLATFORM_NAME, size, name, NULL);

        log_info("Using platform with name: %s \n", name);
        free(name);
    }

    clCreateEventFromGLsyncKHR_ptr = (clCreateEventFromGLsyncKHR_fn)clGetExtensionFunctionAddressForPlatform(platform, "clCreateEventFromGLsyncKHR");
    if( clCreateEventFromGLsyncKHR_ptr == NULL )
    {
        log_error( "ERROR: Unable to run fence_sync test (clCreateEventFromGLsyncKHR function not discovered!)\n" );
        clCreateEventFromGLsyncKHR_ptr = (clCreateEventFromGLsyncKHR_fn)clGetExtensionFunctionAddressForPlatform(platform, "clCreateEventFromGLsyncAPPLE");
        return -1;
    }

#ifdef USING_ARB_sync
    char *gl_version_str = (char*)glGetString( GL_VERSION );
    float glCoreVersion;
    sscanf(gl_version_str, "%f", &glCoreVersion);
    if( glCoreVersion < 3.0f )
    {
        log_info( "OpenGL version %f does not support fence/sync! Skipping test.\n", glCoreVersion );
        return 0;
    }

#ifdef __APPLE__
    CGLContextObj currCtx = CGLGetCurrentContext();
    CGLPixelFormatObj pixFmt = CGLGetPixelFormat(currCtx);
    GLint val, screen;
    CGLGetVirtualScreen(currCtx, &screen);
    CGLDescribePixelFormat(pixFmt, screen, kCGLPFAOpenGLProfile, &val);
    if(val != kCGLOGLPVersion_3_2_Core)
    {
        log_error( "OpenGL context was not created with OpenGL version >= 3.0 profile even though platform supports it"
                  "OpenGL profile %f does not support fence/sync! Skipping test.\n", glCoreVersion );
        return -1;
    }
#else
#ifdef _WIN32
    HDC hdc = wglGetCurrentDC();
    HGLRC hglrc = wglGetCurrentContext();
#else
    Display* dpy = glXGetCurrentDisplay();
    GLXDrawable drawable = glXGetCurrentDrawable();
    GLXContext ctx = glXGetCurrentContext();
#endif
#endif

    InitSyncFns();
#endif

#ifdef __APPLE__
    CGLSetVirtualScreen(CGLGetCurrentContext(), rend_vs);
#else
#ifdef _WIN32
    wglMakeCurrent(hdc, hglrc);
#else
    glXMakeCurrent(dpy, drawable, ctx);
#endif
#endif

    GLint posLoc, colLoc;
    GLuint shaderprogram = createShaderProgram(&posLoc, &colLoc);
    if(!shaderprogram)
    {
        log_error("Failed to create shader program\n");
        return -1;
    }

    float l = 0.0f; float r = framebufferSize;
    float b = 0.0f; float t = framebufferSize;

    float projMatrix[16] = { 2.0f/(r-l), 0.0f, 0.0f, 0.0f,
        0.0f, 2.0f/(t-b), 0.0f, 0.0f,
        0.0f, 0.0f, -1.0f, 0.0f,
        -(r+l)/(r-l), -(t+b)/(t-b), 0.0f, 1.0f
    };

    glUseProgram(shaderprogram);
    GLuint projMatLoc = glGetUniformLocation(shaderprogram, "projMatrix");
    glUniformMatrix4fv(projMatLoc, 1, 0, projMatrix);
    glUseProgram(0);

    // Note: the framebuffer is just the target to verify our results against, so we don't
    // really care to go through all the possible formats in this case
    glFramebufferWrapper glFramebuffer;
    glRenderbufferWrapper glRenderbuffer;
    error = CreateGLRenderbufferRaw( framebufferSize, 128, GL_COLOR_ATTACHMENT0_EXT,
                                    GL_RGBA, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV,
                                    &glFramebuffer, &glRenderbuffer );
    if( error != 0 )
        return error;

    GLuint vao;
    glGenVertexArrays(1, &vao);
    glBindVertexArray(vao);

    glBufferWrapper vtxBuffer, colorBuffer;
    glGenBuffers( 1, &vtxBuffer );
    glGenBuffers( 1, &colorBuffer );

    const int numHorizVertices = ( framebufferSize * 64 ) + 1;

    glBindBuffer( GL_ARRAY_BUFFER, vtxBuffer );
    glBufferData( GL_ARRAY_BUFFER, sizeof( GLfloat ) * numHorizVertices * 2 * 4, NULL, GL_STATIC_DRAW );

    glBindBuffer( GL_ARRAY_BUFFER, colorBuffer );
    glBufferData( GL_ARRAY_BUFFER, sizeof( GLfloat ) * numHorizVertices * 2 * 4, NULL, GL_STATIC_DRAW );

    // Now that the requisite objects are bound, we can attempt program
    // validation:

    glValidateProgram(shaderprogram);

    GLint logLength, status;
    glGetProgramiv(shaderprogram, GL_INFO_LOG_LENGTH, &logLength);
    if (logLength > 0) {
        GLchar *log = (GLchar*)malloc(logLength);
        glGetProgramInfoLog(shaderprogram, logLength, &logLength, log);
        log_info("Program validate log:\n%s", log);
        free(log);
    }

    glGetProgramiv(shaderprogram, GL_VALIDATE_STATUS, &status);
    if (status == 0)
    {
        log_error("Failed to validate program\n");
        return 0;
    }

    clProgramWrapper program;
    clKernelWrapper kernel;
    clMemWrapper streams[ 2 ];

    if( create_single_kernel_helper( context, &program, &kernel, 1, updateBuffersKernel, "update" ) )
        return -1;

    streams[ 0 ] = (*clCreateFromGLBuffer_ptr)( context, CL_MEM_READ_WRITE, vtxBuffer, &error );
    test_error( error, "Unable to create CL buffer from GL vertex buffer" );

    streams[ 1 ] = (*clCreateFromGLBuffer_ptr)( context, CL_MEM_READ_WRITE, colorBuffer, &error );
    test_error( error, "Unable to create CL buffer from GL color buffer" );

    error = clSetKernelArg( kernel, 0, sizeof( streams[ 0 ] ), &streams[ 0 ] );
    test_error( error, "Unable to set kernel arguments" );

    error = clSetKernelArg( kernel, 1, sizeof( streams[ 1 ] ), &streams[ 1 ] );
    test_error( error, "Unable to set kernel arguments" );

    cl_int horizWrap = (cl_int)framebufferSize;
    error = clSetKernelArg( kernel, 2, sizeof( horizWrap ), &horizWrap );
    test_error( error, "Unable to set kernel arguments" );

    glViewport( 0, 0, framebufferSize, framebufferSize );
    glClearColor( 0, 0, 0, 0 );
    glClear( GL_COLOR_BUFFER_BIT );
    glClear( GL_DEPTH_BUFFER_BIT );
    glDisable( GL_DEPTH_TEST );
    glEnable( GL_BLEND );
    glBlendFunc( GL_ONE, GL_ONE );

    clEventWrapper fenceEvent;
    GLsync glFence = 0;

    // Do a loop through 8 different horizontal stripes against the framebuffer
    RunThread thread( kernel, queue, streams[ 0 ], streams[ 1 ], (size_t)numHorizVertices );

    for( int i = 0; i < 8; i++ )
    {
        // if current rendering device is not the compute device and
        // separateThreads == false which means compute is going on same
        // thread and we are using implicit synchronization (no GLSync obj used)
        // then glFlush by clEnqueueAcquireGLObject is not sufficient ... we need
        // to wait for rendering to finish on other device before CL can start
        // writing to CL/GL shared mem objects. When separateThreads is true i.e.
        // we are using GLSync obj to synchronize then we dont need to call glFinish
        // here since CL should wait for rendering on other device before this
        // GLSync object to finish before it starts writing to shared mem object.
        // Also rend_device == compute_device no need to call glFinish
        if(rend_device != device && !separateThreads)
            glFinish();

        if( separateThreads )
        {
            if (fenceEvent != NULL)
            {
                clReleaseEvent(fenceEvent);
                glDeleteSyncFunc(glFence);
            }

            glFence = glFenceSyncFunc(GL_SYNC_GPU_COMMANDS_COMPLETE, 0);
            fenceEvent = clCreateEventFromGLsyncKHR_ptr(context, glFence, &error);
            test_error(error, "Unable to create CL event from GL fence");

            // in case of explicit synchronization, we just wait for the sync object to complete
            // in clEnqueueAcquireGLObject but we dont flush. Its application's responsibility
            // to flush on the context on which glSync is created
            glFlush();

            thread.SetRunData( (cl_int)i, fenceEvent );
            thread.Start();

            error = (cl_int)(size_t)thread.Join();
        }
        else
        {
            error = run_cl_kernel( kernel, queue, streams[ 0 ], streams[ 1 ], (cl_int)i, fenceEvent, (size_t)numHorizVertices );
        }
        test_error( error, "Unable to run CL kernel" );

        glUseProgram(shaderprogram);
        glEnableVertexAttribArray(posLoc);
        glEnableVertexAttribArray(colLoc);
        glBindBuffer( GL_ARRAY_BUFFER, vtxBuffer );
        glVertexAttribPointer(posLoc, 4, GL_FLOAT, GL_FALSE, 4*sizeof(GLfloat), 0);
        glBindBuffer( GL_ARRAY_BUFFER, colorBuffer );
        glVertexAttribPointer(colLoc, 4, GL_FLOAT, GL_FALSE, 4*sizeof(GLfloat), 0);
        glBindBuffer( GL_ARRAY_BUFFER, 0 );

        glDrawArrays( GL_TRIANGLE_STRIP, 0, numHorizVertices * 2 );

        glDisableVertexAttribArray(posLoc);
        glDisableVertexAttribArray(colLoc);
        glUseProgram(0);

        if( separateThreads )
        {
            // If we're on the same thread, then we're testing implicit syncing, so we
            // don't need the actual fence code
            if( fenceEvent != NULL )
            {
                clReleaseEvent( fenceEvent );
                glDeleteSyncFunc( glFence );
            }

            glFence = glFenceSyncFunc( GL_SYNC_GPU_COMMANDS_COMPLETE, 0 );
            fenceEvent = clCreateEventFromGLsyncKHR_ptr( context, glFence, &error );
            test_error( error, "Unable to create CL event from GL fence" );

            // in case of explicit synchronization, we just wait for the sync object to complete
            // in clEnqueueAcquireGLObject but we dont flush. Its application's responsibility
            // to flush on the context on which glSync is created
            glFlush();
        }
        else
            glFinish();
    }

    if( glFence != 0 )
        // Don't need the final release for fenceEvent, because the wrapper will take care of that
        glDeleteSyncFunc( glFence );

#ifdef __APPLE__
    CGLSetVirtualScreen(CGLGetCurrentContext(), read_vs);
#else
#ifdef _WIN32
    wglMakeCurrent(hdc, hglrc);
#else
    glXMakeCurrent(dpy, drawable, ctx);
#endif
#endif
    // Grab the contents of the final framebuffer
    BufferOwningPtr<char> resultData( ReadGLRenderbuffer( glFramebuffer, glRenderbuffer,
                                                         GL_COLOR_ATTACHMENT0_EXT,
                                                         GL_RGBA, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV, kUChar,
                                                         framebufferSize, 128 ) );

    // Check the contents now. We should end up with solid color bands 32 pixels high and the
    // full width of the framebuffer, at values (128,128,128) due to the additive blending
    for( int i = 0; i < 8; i++ )
    {
        for( int y = 0; y < 4; y++ )
        {
            // Note: coverage will be double because the 63-0 triangle overwrites again at the end of the pass
            cl_uchar valA = ( ( ( i + 1 ) & 1 )      ) * numHorizVertices * 2 / framebufferSize;
            cl_uchar valB = ( ( ( i + 1 ) & 2 ) >> 1 ) * numHorizVertices * 2 / framebufferSize;
            cl_uchar valC = ( ( ( i + 1 ) & 4 ) >> 2 ) * numHorizVertices * 2 / framebufferSize;

            cl_uchar *row = (cl_uchar *)&resultData[ ( i * 16 + y ) * framebufferSize * 4 ];
            for( int x = 0; x < ( framebufferSize - 1 ) - 1; x++ )
            {
                if( ( row[ x * 4 ] != valA ) || ( row[ x * 4 + 1 ] != valB ) ||
                   ( row[ x * 4 + 2 ] != valC ) )
                {
                    log_error( "ERROR: Output framebuffer did not validate!\n" );
                    DumpGLBuffer( GL_UNSIGNED_BYTE, framebufferSize, 128, resultData );
                    log_error( "RUNS:\n" );
                    uint32_t *p = (uint32_t *)(char *)resultData;
                    size_t a = 0;
                    for( size_t t = 1; t < framebufferSize * framebufferSize; t++ )
                    {
                        if( p[ a ] != 0 )
                        {
                            if( p[ t ] == 0 )
                            {
                                log_error( "RUN: %ld to %ld (%d,%d to %d,%d) 0x%08x\n", a, t - 1,
                                          (int)( a % framebufferSize ), (int)( a / framebufferSize ),
                                          (int)( ( t - 1 ) % framebufferSize ), (int)( ( t - 1 ) / framebufferSize ),
                                          p[ a ] );
                                a = t;
                            }
                        }
                        else
                        {
                            if( p[ t ] != 0 )
                            {
                                a = t;
                            }
                        }

                    }
                    return -1;
                }
            }
        }
    }

    destroyShaderProgram(shaderprogram);
    glDeleteVertexArrays(1, &vao);
    return 0;
}

int test_fence_sync( cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
{
    GLint vs_count = 0;
    cl_device_id *device_list = NULL;

    if( !is_extension_available( device, "cl_khr_gl_event" ) )
    {
        log_info( "NOTE: cl_khr_gl_event extension not present on this device; skipping fence sync test\n" );
        return 0;
    }
#ifdef __APPLE__
    CGLContextObj ctx = CGLGetCurrentContext();
    CGLPixelFormatObj pix = CGLGetPixelFormat(ctx);
    CGLError err = CGLDescribePixelFormat(pix, 0, kCGLPFAVirtualScreenCount, &vs_count);

    device_list = (cl_device_id *) malloc(sizeof(cl_device_id)*vs_count);
    clGetGLContextInfoAPPLE(context, ctx, CL_CGL_DEVICES_FOR_SUPPORTED_VIRTUAL_SCREENS_APPLE, sizeof(cl_device_id)*vs_count, device_list, NULL);
#else
    // Need platform specific way of getting devices from CL context to which OpenGL can render
    // If not available it can be replaced with clGetContextInfo with CL_CONTEXT_DEVICES
    size_t device_cb;
    cl_int err = clGetContextInfo( context, CL_CONTEXT_DEVICES, 0, NULL, &device_cb);
    if( err != CL_SUCCESS )
    {
      print_error( err, "Unable to get device count from context" );
      return -1;
    }
    vs_count = (GLint)device_cb / sizeof(cl_device_id);

    if (vs_count < 1) {
      log_error("No devices found.\n");
      return -1;
    }

    device_list = (cl_device_id *) malloc(device_cb);
    err = clGetContextInfo( context, CL_CONTEXT_DEVICES, device_cb, device_list, NULL);
    if( err != CL_SUCCESS ) {
      free(device_list);
      print_error( err, "Unable to get device list from context" );
      return -1;
    }

#endif

    GLint rend_vs, read_vs;
    int error = 0;
    int any_failed = 0;

    // Loop through all the devices capable to OpenGL rendering
    // and set them as current rendering target
    for(rend_vs = 0; rend_vs < vs_count; rend_vs++)
    {
        // Loop through all the devices and set them as current
        // compute target
        for(read_vs = 0; read_vs < vs_count; read_vs++)
        {
            cl_device_id rend_device = device_list[rend_vs], read_device = device_list[read_vs];
            char rend_name[200], read_name[200];

            clGetDeviceInfo(rend_device, CL_DEVICE_NAME, sizeof(rend_name), rend_name, NULL);
            clGetDeviceInfo(read_device, CL_DEVICE_NAME, sizeof(read_name), read_name, NULL);

            log_info("Rendering on: %s, read back on: %s\n", rend_name, read_name);
            error = test_fence_sync_single( device, context, queue, false, rend_vs, read_vs, rend_device );
            any_failed |= error;
            if( error != 0 )
                log_error( "ERROR: Implicit syncing with GL sync events failed!\n\n" );
            else
                log_info("Implicit syncing Passed\n");

            error = test_fence_sync_single( device, context, queue, true, rend_vs, read_vs, rend_device );
            any_failed |= error;
            if( error != 0 )
                log_error( "ERROR: Explicit syncing with GL sync events failed!\n\n" );
            else
                log_info("Explicit syncing Passed\n");
        }
    }

    free(device_list);

    return any_failed;
}