使用500 RAM的10 RAM图像

Question

我的Vulkan应用程序有内存问题。加载10 is PNG图像使用的是几乎500 is的ram。

TextureObject* createTextureImage(const char* File) {

    auto Tex = Textures.emplace_back(new TextureObject(_Driver));

    //decode
    unsigned error = lodepng::decode(Tex->Pixels, Tex->Width, Tex->Height, File);

    //if there's an error, display it
    if (error) printf("PNG Decoder error: (%i) %s", error, lodepng_error_text(error));

    Tex->Empty = false;

    VkDeviceSize imageSize = Tex->Width * Tex->Height * 4;

    //
    //  Image Staging Buffer
    VkBufferCreateInfo stagingBufferInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO };
    stagingBufferInfo.size = imageSize;
    stagingBufferInfo.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT;

    VmaAllocationCreateInfo allocInfo = {};
    allocInfo.usage = VMA_MEMORY_USAGE_CPU_ONLY;
    allocInfo.flags = VMA_ALLOCATION_CREATE_MAPPED_BIT;

    VkBuffer stagingImageBuffer = VK_NULL_HANDLE;
    VmaAllocation stagingImageBufferAlloc = VK_NULL_HANDLE;
    vmaCreateBuffer(_Driver->allocator, &stagingBufferInfo, &allocInfo, &stagingImageBuffer, &stagingImageBufferAlloc, nullptr);

    memcpy(stagingImageBufferAlloc->GetMappedData(), Tex->Pixels.data(), static_cast<size_t>(imageSize));
    Tex->Pixels.clear();

    VkImageCreateInfo imageInfo = { VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO };
    imageInfo.imageType = VK_IMAGE_TYPE_2D;
    imageInfo.extent.width = static_cast<uint32_t>(Tex->Width);
    imageInfo.extent.height = static_cast<uint32_t>(Tex->Height);
    imageInfo.extent.depth = 1;
    imageInfo.mipLevels = 1;
    imageInfo.arrayLayers = 1;
    imageInfo.format = VK_FORMAT_B8G8R8A8_SRGB;
    imageInfo.tiling = VK_IMAGE_TILING_OPTIMAL;
    imageInfo.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
    imageInfo.usage = VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_SAMPLED_BIT;
    imageInfo.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
    imageInfo.samples = VK_SAMPLE_COUNT_1_BIT;

    allocInfo.usage = VMA_MEMORY_USAGE_GPU_ONLY;

    VmaAllocationInfo imageBufferAllocInfo = {};
    vmaCreateImage(_Driver->allocator, &imageInfo, &allocInfo, &Tex->Image, &Tex->Allocation, nullptr);
    //
    //  CPU->GPU Copy
    VkCommandBuffer commandBuffer = _Driver->_SceneGraph->beginSingleTimeCommands();
    VkImageMemoryBarrier imgMemBarrier = { VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER };
    imgMemBarrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
    imgMemBarrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
    imgMemBarrier.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
    imgMemBarrier.subresourceRange.baseMipLevel = 0;
    imgMemBarrier.subresourceRange.levelCount = 1;
    imgMemBarrier.subresourceRange.baseArrayLayer = 0;
    imgMemBarrier.subresourceRange.layerCount = 1;
    imgMemBarrier.oldLayout = VK_IMAGE_LAYOUT_UNDEFINED;
    imgMemBarrier.newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL;
    imgMemBarrier.image = Tex->Image;
    imgMemBarrier.srcAccessMask = 0;
    imgMemBarrier.dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;

    vkCmdPipelineBarrier(
        commandBuffer,
        VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT,
        VK_PIPELINE_STAGE_TRANSFER_BIT,
        0,
        0, nullptr,
        0, nullptr,
        1, &imgMemBarrier);

    VkBufferImageCopy region = {};
    region.imageSubresource.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
    region.imageSubresource.layerCount = 1;
    region.imageExtent.width = static_cast<uint32_t>(Tex->Width);
    region.imageExtent.height = static_cast<uint32_t>(Tex->Height);
    region.imageExtent.depth = 1;

    vkCmdCopyBufferToImage(commandBuffer, stagingImageBuffer, Tex->Image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &region);

    imgMemBarrier.oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL;
    imgMemBarrier.newLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
    imgMemBarrier.image = Tex->Image;
    imgMemBarrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
    imgMemBarrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT;

    vkCmdPipelineBarrier(
        commandBuffer,
        VK_PIPELINE_STAGE_TRANSFER_BIT,
        VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT,
        0,
        0, nullptr,
        0, nullptr,
        1, &imgMemBarrier);

    _Driver->_SceneGraph->endSingleTimeCommands(commandBuffer);

    vmaDestroyBuffer(_Driver->allocator, stagingImageBuffer, stagingImageBufferAlloc);

    VkImageViewCreateInfo textureImageViewInfo = { VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO };
    textureImageViewInfo.image = Tex->Image;
    textureImageViewInfo.viewType = VK_IMAGE_VIEW_TYPE_2D;
    textureImageViewInfo.format = VK_FORMAT_B8G8R8A8_SRGB;
    textureImageViewInfo.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
    textureImageViewInfo.subresourceRange.baseMipLevel = 0;
    textureImageViewInfo.subresourceRange.levelCount = 1;
    textureImageViewInfo.subresourceRange.baseArrayLayer = 0;
    textureImageViewInfo.subresourceRange.layerCount = 1;
    vkCreateImageView(_Driver->device, &textureImageViewInfo, nullptr, &Tex->ImageView);

    VkSamplerCreateInfo samplerInfo = { VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO };
    samplerInfo.magFilter = VK_FILTER_LINEAR;
    samplerInfo.minFilter = VK_FILTER_LINEAR;
    samplerInfo.addressModeU = VK_SAMPLER_ADDRESS_MODE_REPEAT;
    samplerInfo.addressModeV = VK_SAMPLER_ADDRESS_MODE_REPEAT;
    samplerInfo.addressModeW = VK_SAMPLER_ADDRESS_MODE_REPEAT;
    samplerInfo.anisotropyEnable = VK_TRUE;
    samplerInfo.maxAnisotropy = 16;
    samplerInfo.borderColor = VK_BORDER_COLOR_INT_OPAQUE_BLACK;
    samplerInfo.unnormalizedCoordinates = VK_FALSE;
    samplerInfo.compareEnable = VK_FALSE;
    samplerInfo.compareOp = VK_COMPARE_OP_ALWAYS;
    samplerInfo.mipmapMode = VK_SAMPLER_MIPMAP_MODE_LINEAR;
    samplerInfo.mipLodBias = 0.0f;
    samplerInfo.minLod = 0.0f;
    samplerInfo.maxLod = 0.0f;
    if (vkCreateSampler(_Driver->device, &samplerInfo, nullptr, &Tex->Sampler) != VK_SUCCESS) {
#ifdef _DEBUG
        throw std::runtime_error("failed to create texture sampler!");
#endif
    }

    return Tex;
}

我已经把分配范围缩小到了这个函数中。只需将从10 to加载到1kb png的映像交换，就可以大大减少此函数调用期间的内存消耗。

我是不是做错了什么，导致大量内存被分配而没有释放？

在每一行中放置断点显示在调用以下命令之后立即发生的巨大分配：

unsigned error = lodepng::decode(Tex->Pixels, Tex->Width, Tex->Height, File);

如何加载一个10 to的png文件使用500 to以上的加载？

lodepng::decode在Tex中填充以下3个变量

unsigned int Width = 0;
unsigned int Height = 0;
std::vector<unsigned char> Pixels = {};

即使是1kb的png也使用大约5mb来加载，这似乎非常高。

我还尝试使用图像加载png文件，结果也是一样的。

EDIT2: 10 16x16为4096x4096，1kb png为16x16。

EDIT2:经过一些内存分析之后，在检查一个内存尖峰之后，将分配58,000个包含563,000,000字节get的空实例，其中500,000,000多个实例位于我称为.clear()的向量中。我不知道如何才能真正解决这个问题，或者它是否有那么大的问题？

Answer 1

如何加载一个10 to的png文件使用500 to以上的加载？

它的10 of压缩数据。你在这里扔的所有缓冲区都是解压缩后的，所以每个.

一个可能的漏洞是vector.clear()不能保证重新分配，所以当您在这里调用clear()来释放对象时，您可能仍然坐在64 of的备份存储上，直到向量被销毁后才会释放。