Rendering a cube in Vulkan vs OpenGL

0
votes

I wrote a simple OpenGL program which merely renders a cube from an angle. It's as simple as you can get: vertex buffer only (no index buffer), a vertex shader which only multiplies the vertices by an MVP matrix from a uniform buffer, and a static fragment shader which just returns red. More recently, I have tried writing this same program in Vulkan, but I have run into some issues.

I started by following the Intel API without secrets tutorial to setup a simple 2d texture rendering program, but when I took the leap into 3d, I started having issues. In order to debug this, I simplified the program to match my older OpenGL program (removed texturing and some other extra stuff I did in Vulkan), and even went as far as to use the exact same vertex and MVP data. However, I just can't get the cube to render correctly in Vulkan.

I am aware that OpenGL coordinates do not map directly to Vulkan coordinates, as the Y coordinate is flipped, but if anything that should just flip the image upside down, and I already tried switching the Y values in the MVP. I feel like there is some other detail I am missing here with coordinates, but I just can't figure it out searching around and looking at guides about converting OpenGL code bases to Vulkan.

I'm including the data I am uploading to the shaders, and some of the core code from the Vulkan code base. The Vulkan code is in D, so it's similar to C++, but a little different. With the library I'm using for wrapping Vulkan (erupted), the device level functions are loaded into a device dispatch (access as device.dispatch in the code), and when they are called on the dispatch without the vk prefix, the device and command buffer (which is assigned to the dispatch in code) arguments of the function are auto populated.

Vertex Data:

[ [1, 1, 1, 1],
  [1, 1, -1, 1],
  [-1, 1, -1, 1],
  [1, 1, 1, 1],
  [-1, 1, -1, 1],
  [-1, 1, 1, 1],
  [1, 1, 1, 1],
  [1, -1, 1, 1],
  [1, -1, -1, 1],
  [1, 1, 1, 1],
  [1, -1, -1, 1],
  [1, 1, -1, 1],
  [1, 1, -1, 1],
  [1, -1, -1, 1],
  [-1, -1, -1, 1],
  [1, 1, -1, 1],
  [-1, -1, -1, 1],
  [-1, 1, -1, 1],
  [-1, 1, -1, 1],
  [-1, -1, -1, 1],
  [-1, -1, 1, 1],
  [-1, 1, -1, 1],
  [-1, -1, 1, 1],
  [-1, 1, 1, 1],
  [-1, 1, 1, 1],
  [-1, -1, 1, 1],
  [1, -1, 1, 1],
  [-1, 1, 1, 1],
  [1, -1, 1, 1],
  [1, 1, 1, 1],
  [1, -1, 1, 1],
  [1, -1, -1, 1],
  [-1, -1, -1, 1],
  [1, -1, 1, 1],
  [-1, -1, -1, 1],
  [-1, -1, 1, 1] ]

MVP:

[ [-1.0864, -0.993682, -0.687368, -0.685994],
  [0, 2.07017, 0.515526, -0.514496],
  [-1.44853, 0.745262, 0.515526, 0.514496],
  [-8.04095e-16, 0, 5.64243, 5.83095] ]

Graphics Pipeline Setup:

VkPipelineShaderStageCreateInfo[] shader_stage_infos = [
  {
    stage: VK_SHADER_STAGE_VERTEX_BIT,
    _module: vertex_shader,
    pName: "main"
  },
  {
    stage: VK_SHADER_STAGE_FRAGMENT_BIT,
    _module: fragment_shader,
    pName: "main"
  }
];
VkVertexInputBindingDescription[] vertex_binding_descriptions = [
  {
    binding: 0,
    stride: VertexData.sizeof,
    inputRate: VK_VERTEX_INPUT_RATE_VERTEX
  }
];
VkVertexInputAttributeDescription[] vertex_attribute_descriptions = [
  {
    location: 0,
    binding: vertex_binding_descriptions[0].binding,
    format: VK_FORMAT_R32G32B32A32_SFLOAT,
    offset: VertexData.x.offsetof
  },
  {
    location: 1,
    binding: vertex_binding_descriptions[0].binding,
    format: VK_FORMAT_R32G32_SFLOAT,
    offset: VertexData.u.offsetof
  }
];
VkPipelineVertexInputStateCreateInfo vertex_input_state_info = {
  vertexBindingDescriptionCount: vertex_binding_descriptions.length.to!uint,
  pVertexBindingDescriptions: vertex_binding_descriptions.ptr,
  vertexAttributeDescriptionCount: vertex_attribute_descriptions.length.to!uint,
  pVertexAttributeDescriptions: vertex_attribute_descriptions.ptr
};
VkPipelineInputAssemblyStateCreateInfo input_assembly_state_info = {
  topology: VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST,
  primitiveRestartEnable: VK_FALSE
};
VkPipelineViewportStateCreateInfo viewport_state_info = {
  viewportCount: 1,
  pViewports: null,
  scissorCount: 1,
  pScissors: null
};
VkPipelineRasterizationStateCreateInfo rasterization_state_info = {
  depthBiasClamp: 0.0,
  polygonMode: VK_POLYGON_MODE_FILL,
  cullMode: VK_CULL_MODE_FRONT_AND_BACK,
  frontFace: VK_FRONT_FACE_COUNTER_CLOCKWISE,
  lineWidth: 1
};
VkPipelineMultisampleStateCreateInfo multisample_state_info = {
  rasterizationSamples: VK_SAMPLE_COUNT_1_BIT,
  minSampleShading: 1
};
VkPipelineColorBlendAttachmentState[] color_blend_attachment_states = [
  {
    blendEnable: VK_FALSE,
    srcColorBlendFactor: VK_BLEND_FACTOR_ONE,
    dstColorBlendFactor: VK_BLEND_FACTOR_ZERO,
    colorBlendOp: VK_BLEND_OP_ADD,
    srcAlphaBlendFactor: VK_BLEND_FACTOR_ONE,
    dstAlphaBlendFactor: VK_BLEND_FACTOR_ZERO,
    alphaBlendOp: VK_BLEND_OP_ADD,
    colorWriteMask:
      VK_COLOR_COMPONENT_R_BIT |
      VK_COLOR_COMPONENT_G_BIT |
      VK_COLOR_COMPONENT_B_BIT |
      VK_COLOR_COMPONENT_A_BIT
  }
];
VkPipelineColorBlendStateCreateInfo color_blend_state_info = {
  logicOpEnable: VK_FALSE,
  logicOp: VK_LOGIC_OP_COPY,
  attachmentCount: color_blend_attachment_states.length.to!uint,
  pAttachments: color_blend_attachment_states.ptr,
  blendConstants: [ 0, 0, 0, 0 ]
};
VkDynamicState[] dynamic_states = [
  VK_DYNAMIC_STATE_VIEWPORT,
  VK_DYNAMIC_STATE_SCISSOR
];
VkPipelineDynamicStateCreateInfo dynamic_state_info = {
  dynamicStateCount: dynamic_states.length.to!uint,
  pDynamicStates: dynamic_states.ptr
};
VkGraphicsPipelineCreateInfo pipeline_info = {
  stageCount: shader_stage_infos.length.to!uint,
  pStages: shader_stage_infos.ptr,
  pVertexInputState: &vertex_input_state_info,
  pInputAssemblyState: &input_assembly_state_info,
  pTessellationState: null,
  pViewportState: &viewport_state_info,
  pRasterizationState: &rasterization_state_info,
  pMultisampleState: &multisample_state_info,
  pDepthStencilState: null,
  pColorBlendState: &color_blend_state_info,
  pDynamicState: &dynamic_state_info,
  layout: pipeline_layout,
  renderPass: render_pass,
  subpass: 0,
  basePipelineHandle: VK_NULL_HANDLE,
  basePipelineIndex: -1
};
VkPipeline[1] pipelines;
checkVk(device.dispatch.CreateGraphicsPipelines(VK_NULL_HANDLE, 1, [pipeline_info].ptr, pipelines.ptr));
pipeline = pipelines[0];

Drawing:

if(device.dispatch.WaitForFences(1, [fence].ptr, VK_FALSE, 1000000000) != VK_SUCCESS)
  throw new StringException("timed out waiting for fence");
device.dispatch.ResetFences(1, [fence].ptr);

uint image_index;
switch(device.dispatch.AcquireNextImageKHR(swapchain.swapchain, uint64_t.max, image_available_semaphore, VK_NULL_HANDLE, &image_index)) {
  case VK_SUCCESS:
  case VK_SUBOPTIMAL_KHR:
    break;
  case VK_ERROR_OUT_OF_DATE_KHR:
    on_window_size_changed();
    break;
  default:
    throw new StringException("unhandled vk result on swapchain image acquisition");
}

if(framebuffer != VK_NULL_HANDLE) device.dispatch.DestroyFramebuffer(framebuffer);

VkFramebufferCreateInfo framebuffer_info = {
  renderPass: swapchain.render_pass,
  attachmentCount: 1,
  pAttachments: [swapchain.image_resources[image_index].image_view].ptr,
  width: swapchain.extent.width,
  height: swapchain.extent.height,
  layers: 1
};
checkVk(device.dispatch.CreateFramebuffer(&framebuffer_info, &framebuffer));

VkCommandBufferBeginInfo cmd_begin_info = { flags: VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT };
VkImageSubresourceRange image_subresource_range = {
  aspectMask: VK_IMAGE_ASPECT_COLOR_BIT,
  baseMipLevel: 0,
  levelCount: 1,
  baseArrayLayer: 0,
  layerCount: 1,
};
VkImageMemoryBarrier barrier_from_present_to_draw = {
  srcAccessMask: VK_ACCESS_MEMORY_READ_BIT,
  dstAccessMask: VK_ACCESS_MEMORY_READ_BIT,
  oldLayout: VK_IMAGE_LAYOUT_PRESENT_SRC_KHR,
  newLayout: VK_IMAGE_LAYOUT_PRESENT_SRC_KHR,
  srcQueueFamilyIndex: device.present_queue.family_index,
  dstQueueFamilyIndex: device.graphics_queue.family_index,
  image: swapchain.image_resources[image_index].image,
  subresourceRange: image_subresource_range
};
VkImageMemoryBarrier barrier_from_draw_to_present = {
  srcAccessMask: VK_ACCESS_MEMORY_READ_BIT,
  dstAccessMask: VK_ACCESS_MEMORY_READ_BIT,
  oldLayout: VK_IMAGE_LAYOUT_PRESENT_SRC_KHR,
  newLayout: VK_IMAGE_LAYOUT_PRESENT_SRC_KHR,
  srcQueueFamilyIndex: device.graphics_queue.family_index,
  dstQueueFamilyIndex: device.present_queue.family_index,
  image: swapchain.image_resources[image_index].image,
  subresourceRange: image_subresource_range
};
VkViewport viewport = {
  x: 0,
  y: 0,
  width: swapchain.extent.width,
  height: swapchain.extent.height,
  minDepth: 0,
  maxDepth: 1
};
VkRect2D scissor = {
  offset: {
    x: 0,
    y: 0
  },
  extent: swapchain.extent
};
VkClearValue[] clear_values = [
  { color: { [ 1.0, 0.8, 0.4, 0.0 ] } }
];
VkRenderPassBeginInfo render_pass_begin_info = {
  renderPass: swapchain.render_pass,
  framebuffer: framebuffer,
  renderArea: {
    offset: {
      x: 0,
      y: 0
    },
    extent: swapchain.extent
  },
  clearValueCount: clear_values.length.to!uint,
  pClearValues: clear_values.ptr
};

device.dispatch.commandBuffer = command_buffer;
device.dispatch.BeginCommandBuffer(&cmd_begin_info);
  if(device.graphics_queue.handle != device.present_queue.handle)
    device.dispatch.CmdPipelineBarrier(
        VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
        VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
        0, 0, null, 0, null, 1,
        &barrier_from_present_to_draw
    );
  device.dispatch.CmdBeginRenderPass(&render_pass_begin_info, VK_SUBPASS_CONTENTS_INLINE);
    device.dispatch.CmdBindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, swapchain.pipeline);
    device.dispatch.CmdSetViewport(0, 1, &viewport);
    device.dispatch.CmdSetScissor(0, 1, &scissor);
    const(ulong) vertex_buffer_offset = 0;
    device.dispatch.CmdBindVertexBuffers(0, 1, &vertex_buffer, &vertex_buffer_offset);
    device.dispatch.CmdBindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline_layout, 0, 1, &descriptor_set, 0, null);
    device.dispatch.CmdDraw(draw_count, 1, 0, 0);
  device.dispatch.CmdEndRenderPass();
  if(device.graphics_queue.handle != device.present_queue.handle)
    device.dispatch.CmdPipelineBarrier(
        VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
        VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT,
        0, 0, null, 0, null, 1,
        &barrier_from_draw_to_present
    );
checkVk(device.dispatch.EndCommandBuffer());
device.dispatch.commandBuffer = VK_NULL_HANDLE;

VkSubmitInfo submit_info = {
  waitSemaphoreCount: 1,
  pWaitSemaphores: [image_available_semaphore].ptr,
  pWaitDstStageMask: castFrom!(VkPipelineStageFlagBits*).to!(const(uint)*)([VK_PIPELINE_STAGE_TRANSFER_BIT].ptr),
  commandBufferCount: 1,
  pCommandBuffers: [command_buffer].ptr,
  signalSemaphoreCount: 1,
  pSignalSemaphores: [rendering_finished_semaphore].ptr
};
checkVk(device.dispatch.vkQueueSubmit(device.graphics_queue.handle, 1, [submit_info].ptr, fence));

VkPresentInfoKHR present_info = {
  waitSemaphoreCount: 1,
  pWaitSemaphores: [rendering_finished_semaphore].ptr,
  swapchainCount: 1,
  pSwapchains: [swapchain.swapchain].ptr,
  pImageIndices: [image_index].ptr
};
switch(device.dispatch.vkQueuePresentKHR(device.present_queue.handle, &present_info)) {
  case VK_SUCCESS:
    break;

  case VK_ERROR_OUT_OF_DATE_KHR:
  case VK_SUBOPTIMAL_KHR:
    on_window_size_changed();
    break;

  default:
    throw new StringException("unhandled vk result on presentation");
}

(I can't embed the images because my rep is too low, sorry)

Program Outputs:

OpenGL draws the cube as expected OpenGL Output

Vulkan does not render anything except for the clear color.

UPDATE:

After fixing the cull mode by changing it to VK_CULL_MODE_NONE, this is the result I get: Output after cull mode fix

2
vulkan
> device.dispatch.CmdDraw(draw_count, 1, 0, 0); Does draw_count refer to the vertex count? - jeremyong
Yes. Draw count is set to 36 in this case. - NathanH
Did you check the triangle winding order? Maybe turn off backface culling as a test. I would look at the renderdoc capture mesh output as well. - jeremyong
omg just set viewport offset y to your viewport height and set your viewport height to -(viewport height) for a simple quick fix - sansa

2 Answers

1
votes

VK_CULL_MODE_FRONT_AND_BACK

I think this is your problem :)

0
votes

After cull mode fix, seems that your problem in your vertex data layout. Vulkan expects (accordingly to your layout binding) something like

struct Vertex {
    vec4 x;
    vec2 u;
};

Vertex VertexData[] = {...};

because you set VK_VERTEX_INPUT_RATE_VERTEX in your vertex_binding_descriptions.inputRate field.

And it seems that in your case you should set VK_VERTEX_INPUT_RATE_INSTANCE instead to work with buffers after each other.

Fix: Have seen your new comment, it looks like i misunderstood your vertices layout, so it won't help.