Vulkan without Pipelines using VK_EXT_shader_object

[JeGX]

Today, Khronos® is releasing a new multi-vendor Vulkan® extension that aims to radically simplify how applications specify shaders and shader state while maintaining Vulkan's ethos of being a highly performant "API without secrets."

This extension is VK_EXT_shader_object. It introduces a new VkShaderEXT object type which represents a single compiled shader stage, along with 4 new functions to manipulate and use VkShaderEXT objects.

Shader objects serve a role similar to pipelines but expose a simpler, more flexible interface designed to empower applications whose architectures and/or needs for "dynamism" might have previously made the use of pipelines difficult. In environments where VK_EXT_shader_object is supported, applications can choose to use only pipelines, only shader objects, or an arbitrary mix of the two.

VK_EXT_shader_object has been introduced with Vulkan 1.3.246 today and is available in latest NVIDIA Vulkan driver 541.54.

Complete article:
https://www.khronos.org/blog/you-can-use-vulkan-without-pipelines-today


Example: creation of a linked pair of vertex and fragment shaders.

Code Select Expand

// Logical device created with the shaderObject feature enabled
VkDevice device;

// SPIR-V shader code for a vertex shader, along with its size in bytes
void* pVertexSpirv;
size_t vertexSpirvSize;

// SPIR-V shader code for a fragment shader, along with its size in bytes
void* pFragmentSpirv;
size_t fragmentSpirvSize;

// Descriptor set layout compatible with the shaders
VkDescriptorSetLayout descriptorSetLayout;

VkShaderCreateInfoEXT shaderCreateInfos[2] =
{
    {
        .sType = VK_STRUCTURE_TYPE_SHADER_CREATE_INFO_EXT,
        .pNext = NULL,
        .flags = VK_SHADER_CREATE_LINK_STAGE_BIT_EXT,
        .stage = VK_SHADER_STAGE_VERTEX_BIT,
        .nextStage = VK_SHADER_STAGE_FRAGMENT_BIT,
        .codeType = VK_SHADER_CODE_TYPE_SPIRV_EXT,
        .codeSize = vertexSpirvSize,
        .pCode = pVertexSpirv,
        .pName = "main",
        .setLayoutCount = 1,
        .pSetLayouts = &descriptorSetLayout;
        .pushConstantRangeCount = 0,
        .pPushConstantRanges = NULL,
        .pSpecializationInfo = NULL
    },
    {
        .sType = VK_STRUCTURE_TYPE_SHADER_CREATE_INFO_EXT,
        .pNext = NULL,
        .flags = VK_SHADER_CREATE_LINK_STAGE_BIT_EXT,
        .stage = VK_SHADER_STAGE_FRAGMENT_BIT,
        .nextStage = 0,
        .codeType = VK_SHADER_CODE_TYPE_SPIRV_EXT,
        .codeSize = fragmentSpirvSize,
        .pCode = pFragmentSpirv,
        .pName = "main",
        .setLayoutCount = 1,
        .pSetLayouts = &descriptorSetLayout;
        .pushConstantRangeCount = 0,
        .pPushConstantRanges = NULL,
        .pSpecializationInfo = NULL
    }
};

VkResult result;
VkShaderEXT shaders[2];

result = vkCmdCreateShadersEXT(device, 2, &shaderCreateInfos, NULL, shaders);
if (result != VK_SUCCESS)
{
    // Handle error
}


And how to bind this shader object:

Code Select Expand

// Command buffer in the recording state
VkCommandBuffer commandBuffer;

// Vertex and fragment shader objects created above
VkShaderEXT shaders[2];

// Assume vertex buffers, descriptor sets, etc. have been bound, and existing
// state setting commands have been called to set all required state

const VkShaderStageFlagBits stages[2] =
{
    VK_SHADER_STAGE_VERTEX_BIT,
    VK_SHADER_STAGE_FRAGMENT_BIT
};

// Bind linked shaders
vkCmdBindShadersEXT(commandBuffer, 2, stages, shaders);

// Equivalent to the previous line. Linked shaders can be bound one at a time,
// in any order:
vkCmdBindShadersEXT(commandBuffer, 1, &stages[1], &shaders[1]);
vkCmdBindShadersEXT(commandBuffer, 1, &stages[0], &shaders[0]);

// The above is sufficient to draw if the device was created with the
// tessellationShader and geometryShader features disabled. Otherwise, since
// those stages should not execute, vkCmdBindShadersEXT() must be called at
// least once with each of their stages in pStages before drawing:

const VkShaderStageFlagBits unusedStages[3] =
{
    VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT,
    VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT,
    VK_SHADER_STAGE_GEOMETRY_BIT
};

// NULL pShaders is equivalent to an array of stageCount VK_NULL_HANDLE values,
// meaning no shaders are bound to those stages, and that any previously bound
// shaders are unbound
vkCmdBindShadersEXT(commandBuffer, 3, unusedStages, NULL);

// Draw a triangle
vkCmdDraw(commandBuffer, 3, 1, 0, 0);


[CecilHarvey]

I'm a little confused. Do you not need to define a renderpass or the settings for the non-programmable stages? If not, how would you know what the default values are?

Feels like a move more towards OpenGL.