0

https://i.stack.imgur.com/TA9v6.png

我一直在尝试编译一个内核，该内核使用 OpenCL 通过 clEnqueueReadBuffer 函数将某些索引分配给 std::vector，但它似乎无法正常工作，因为第一个结果是 std::vector 中唯一分配的

c++中主机的源代码如下：

cl_mem originalPixelsBuffer = clCreateBuffer(p1.context, CL_MEM_READ_ONLY | CL_MEM_USE_HOST_PTR, sizeof(Color) * imageObj->SourceLength(), source, &p1.status);
        CheckErrorCode(p1.status, p1.program, p1.devices[0], "未能创建缓冲区 0");


        cl_mem targetBuffer = clCreateBuffer(p1.context, CL_MEM_READ_WRITE | CL_MEM_USE_HOST_PTR, sizeof(Color) * imageObj->OutputLength(), target, &p1.status);
        CheckErrorCode(p1.status, p1.program, p1.devices[0], "未能创建缓冲区 1");


//写缓冲区
p1.status = clEnqueueWriteBuffer(p1.commandQueue, originalPixelsBuffer, CL_FALSE, 0, sizeof(Color) * imageObj->SourceLength(), source, 0, NULL, NULL);
        CheckErrorCode(p1.status, p1.program, p1.devices[0], "写入缓冲区 0 失败");
        p1.status = clEnqueueWriteBuffer(p1.commandQueue, targetBuffer, CL_TRUE, 0, sizeof(Color) * imageObj->OutputLength(), target, 0, NULL, NULL);
        CheckErrorCode(p1.status, p1.program, p1.devices[0], "写入缓冲区 1 失败");

        size_t globalWorkSize[2] = { imageObj->originalWidth * 4, imageObj->originalHeight * 4 };
        尺寸_t localWorkSize[2]{ 64,64 };
        SetLocalWorkSize(IsDivisibleBy64(localWorkSize[0]), localWorkSize);

//执行内核
        p1.status = clEnqueueNDRangeKernel(p1.commandQueue, Kernel, 1, NULL, globalWorkSize, IsDisibibleByLocalWorkSize(globalWorkSize, localWorkSize) ? localWorkSize : NULL, 0, NULL, NULL);
        CheckErrorCode(p1.status, p1.program, p1.devices[0], "clEnqueueDRangeKernel 失败");

//读取缓冲区

        p1.status = clEnqueueReadBuffer(p1.commandQueue, targetBuffer, CL_TRUE, 0, sizeof(Color) * imageObj->OutputLength(), target, 0, NULL, NULL);
        CheckErrorCode(p1.status, p1.program, p1.devices[0], "写入缓冲区 1 失败");

内核代码：


      __kernel void interp(__global struct Color* source,__global struct Color* target,uint64 width,uint64 height,uint64 ratio,uint64 limit, uint64 originalHeight)
        {
            __private fp32 wIndex = (int64)get_global_id(0);
            __private fp32 hIndex = (int64)get_global_id(1);

            if(((int64)wIndex)%ratio==MATCH && ((int64)hIndex)%ratio ==MATCH)
            {
                __private int64  Index = (wIndex/ratio) * (originalHeight/ratio) + (hIndex/ratio);

                if(Index < limit)
                {
                
                        __private int64 tIndex  = wIndex * height + hIndex;
                        
                        target[tIndex].R = source[Index].R;
                        target[tIndex].G = source[Index].G;
                        target[tIndex].B = source[Index].B;
                        target[tIndex].A = source[Index].A;
                }
            }
        
        }```
4

0 回答 0