Vulkan-Docs/doc/specs/vulkan/chapters/copies.txt

// Copyright (c) 2015-2016 The Khronos Group Inc.
// Copyright notice at https://www.khronos.org/registry/speccopyright.html

[[copies]]
= Copy Commands

An application can: copy buffer and image data using several methods
depending on the type of data transfer. Data can: be copied between buffer
objects with fname:vkCmdCopyBuffer and a portion of an image can: be copied
to another image with fname:vkCmdCopyImage. Image data can: also be
copied to and from buffer memory using fname:vkCmdCopyImageToBuffer and
fname:vkCmdCopyBufferToImage. Image data can: be blitted (with or without
scaling and filtering) with fname:vkCmdBlitImage. Multisampled images can:
be resolved to a non-multisampled image with fname:vkCmdResolveImage.


== Common Operation

Some rules for valid operation are common to all copy commands:

  * Copy commands must: be recorded outside of a render pass instance.
  * For non-sparse resources, the union of the source regions in a given
    buffer or image must: not overlap the union of the destination regions
    in the same buffer or image.
  * For sparse resources, the set of bytes used by all the source regions
    must: not intersect the set of bytes used by all the destination
    regions.
  * Copy regions must: be non-empty.
  * Regions must: not extend outside the bounds of the buffer or image
    level, except that regions of compressed images can: extend as far as
    the dimension of the image level rounded up to a complete block.
  * Source image subresources must: be in either the
    ename:VK_IMAGE_LAYOUT_GENERAL or
    ename:VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL layout. Destination image
    subresources must: be in either the ename:VK_IMAGE_LAYOUT_GENERAL or
    ename:VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL layout. As a consequence, if
    an image subresource is used as both source and destination of a copy,
    it must: be in the ename:VK_IMAGE_LAYOUT_GENERAL layout.
  * Source images must: have been created with the
    ename:VK_IMAGE_USAGE_TRANSFER_SRC_BIT usage bit enabled and destination
    images must: have been created with the
    ename:VK_IMAGE_USAGE_TRANSFER_DST_BIT usage bit enabled.
  * Source buffers must: have been created with the
    ename:VK_BUFFER_USAGE_TRANSFER_SRC_BIT usage bit enabled and destination
    buffers must: have been created with the
    ename:VK_BUFFER_USAGE_TRANSFER_DST_BIT usage bit enabled.

All copy commands are treated as ``transfer'' operations for the purposes of
synchronization barriers.


[[copies-buffers]]
== Copying Data Between Buffers

To copy data between buffer objects, call:

include::../protos/vkCmdCopyBuffer.txt[]

  * pname:commandBuffer is the command buffer into which the command will be
    recorded.
  * pname:srcBuffer is the source buffer.
  * pname:dstBuffer is the destination buffer.
  * pname:regionCount is the number of regions to copy.
  * pname:pRegions is a pointer to an array of slink:VkBufferCopy structures
    specifying the regions to copy.

Each region in pname:pRegions is copied from the source buffer to the same
region of the destination buffer. pname:srcBuffer and pname:dstBuffer can:
be the same buffer or alias the same memory, but the result is undefined if
the copy regions overlap in memory.

include::../validity/protos/vkCmdCopyBuffer.txt[]

Each element of pname:pRegions is a structure defined as:

include::../structs/VkBufferCopy.txt[]

  * pname:srcOffset is the starting offset in bytes from the start of
    pname:srcBuffer.
  * pname:dstOffset is the starting offset in bytes from the start of
    pname:dstBuffer.
  * pname:size is the number of bytes to copy.

include::../validity/structs/VkBufferCopy.txt[]


[[copies-images]]
== Copying Data Between Images

fname:vkCmdCopyImage performs image copies in a similar manner to a host
memcpy. It does not perform general-purpose conversions such as scaling,
resizing, blending, color-space conversion, or format conversions.
Rather, it simply copies raw image data. fname:vkCmdCopyImage can: copy
between images with different formats, provided the formats are compatible
as defined below.

To copy data between image objects, call:

include::../protos/vkCmdCopyImage.txt[]

  * pname:commandBuffer is the command buffer into which the command will be
    recorded.
  * pname:srcImage is the source image.
  * pname:srcImageLayout is the current layout of the source image
    subresource.
  * pname:dstImage is the destination image.
  * pname:dstImageLayout is  the current layout of the destination image
    subresource.
  * pname:regionCount is the number of regions to copy.
  * pname:pRegions is a pointer to an array of slink:VkImageCopy structures
    specifying the regions to copy.

Each region in pname:pRegions is copied from the source image to the same
region of the destination image. pname:srcImage and pname:dstImage can: be
the same image or alias the same memory.

include::../validity/protos/vkCmdCopyImage.txt[]

Each element of pname:pRegions is a structure defined as:

include::../structs/VkImageCopy.txt[]

  * pname:srcSubresource and pname:dstSubresource are
    slink:VkImageSubresourceLayers structures specifying the subresources of
    the images used for the source and destination image data, respectively.
  * pname:srcOffset and pname:dstOffset select the initial x, y, and z
    offsets in texels of the sub-regions of the source and destination image
    data.
  * pname:extent is the size in texels of the source image to copy in
    pname:width, pname:height and pname:depth. 1D images use only pname:x
    and pname:width. 2D images use pname:x, pname:y, pname:width and
    pname:height. 3D images use pname:x, pname:y, pname:z, pname:width,
    pname:height and pname:depth.

include::../validity/structs/VkImageCopy.txt[]

The sname:VkImageSubresourceLayers structure is defined as:

include::../structs/VkImageSubresourceLayers.txt[]

  * pname:aspectMask is a combination of elink:VkImageAspectFlagBits,
    selecting the color, depth and/or stencil aspects to be copied.
  * pname:mipLevel is the mipmap level to copy from.
  * pname:baseArrayLayer and pname:layerCount are the starting layer and
    number of layers to copy.

include::../validity/structs/VkImageSubresourceLayers.txt[]

Copies are done layer by layer starting with pname:baseArrayLayer member of
pname:srcSubresource for the source and pname:dstSubresource for the
destination. pname:layerCount layers are copied to the destination image.

[[copies-images-format-compatibility]]
The formats of pname:srcImage and pname:dstImage must: be compatible.
Formats are considered compatible if their texel size in bytes is the same
between both formats. For example, ename:VK_FORMAT_R8G8B8A8_UNORM is
compatible with ename:VK_FORMAT_R32_UINT because because both texels are 4
bytes in size. Depth/stencil formats must: match exactly.

fname:vkCmdCopyImage allows copying between size-compatible compressed
and uncompressed internal formats. Formats are size-compatible if the texel
size of the uncompressed format is equal to the block size in bytes of the
compressed format. Such a copy does not perform on-the-fly compression or
decompression. When copying from an uncompressed format to a compressed
format, each texel of uncompressed data becomes a single block of compressed
data. When copying from a compressed format to an uncompressed format, a
block of compressed data becomes a single texel of uncompressed data. Thus,
for example, it is legal to copy between a 128-bit uncompressed format and a
compressed format which uses 8-bit/texel 4x4 blocks, or between a 64-bit
uncompressed format and a compressed format which uses 4-bit/texel 4x4
blocks.

When copying between compressed and uncompressed formats the pname:extent
members represent the texel dimensions of the source image and not the
destination. When copying from a compressed image to an uncompressed image
the image texel dimensions written to the uncompressed image will be source
extent divided by the block size. When copying from an uncompressed image to
a compressed image the image texel dimensions written to the compressed
image will be the source extent multiplied by the block size. In both cases
the number of bytes read and the number of bytes written will be identical.

Copying to or from block-compressed images is typically done in multiples of
the block. For this reason the pname:extent must: be a multiple of the
block dimension. There is one exception to this rule which is required: to
handle compressed images created with dimensions that are not a multiple
of the block dimensions. If the pname:srcImage is compressed and if
pname:extent.width is not a multiple of the block width then
(pname:extent.width + pname:srcOffset.x) must: equal the subresource width,
if pname:extent.height is not a multiple of the block height then
(pname:extent.height + pname:srcOffset.y) must: equal
the subresource height and if pname:extent.depth is not a multiple of the
block depth then (pname:extent.depth + pname:srcOffset.z) must: equal
the subresource depth. Similarily if the pname:dstImage is compressed and
if pname:extent.width is not a multiple of the block width then
(pname:extent.width + pname:dstOffset.x) must: equal the subresource width,
if pname:extent.height is not a multiple of the block height then
(pname:extent.height + pname:dstOffset.y) must: equal
the subresource height and if pname:extent.depth is not a multiple of the
block depth then (pname:extent.depth + pname:dstOffset.z) must: equal
the subresource depth. This allows the last block of the image in each
non-multiple dimension to be included as a source or target of the copy.

fname:vkCmdCopyImage can: be used to copy image data between multisample
images, but both images must: have the same number of samples.


[[copies-buffers-images]]
== Copying Data Between Buffers and Images

To copy data from a buffer object to an image object, call:

include::../protos/vkCmdCopyBufferToImage.txt[]

  * pname:commandBuffer is the command buffer into which the command will be
    recorded.
  * pname:srcBuffer is the source buffer.
  * pname:dstImage is the destination image.
  * pname:dstImageLayout is the layout of the destination image subresources
    for the copy.
  * pname:regionCount is the number of regions to copy.
  * pname:pRegions is a pointer to an array of slink:VkBufferImageCopy
    structures specifying the regions to copy.

Each region in pname:pRegions is copied from the specified region of the
source buffer to the specified region of the destination image.

include::../validity/protos/vkCmdCopyBufferToImage.txt[]

To copy data from an image object to a buffer object, call:

include::../protos/vkCmdCopyImageToBuffer.txt[]

  * pname:commandBuffer is the command buffer into which the command will be
    recorded.
  * pname:srcImage is the source image.
  * pname:srcImageLayout is the layout of the source image subresources for
    the copy.
  * pname:dstBuffer is the destination buffer.
  * pname:regionCount is the number of regions to copy.
  * pname:pRegions is a pointer to an array of slink:VkBufferImageCopy
    structures specifying the regions to copy.

Each region in pname:pRegions is copied from the specified region of the
source buffer to the specified region of the destination image.

include::../validity/protos/vkCmdCopyImageToBuffer.txt[]

For both flink:vkCmdCopyBufferToImage and flink:vkCmdCopyImageToBuffer, each
element of pname:pRegions is a structure defined as:

include::../structs/VkBufferImageCopy.txt[]

  * pname:bufferOffset is the offset in bytes from the start of the
    buffer object where the image data is copied from or to.
  * pname:bufferRowLength and pname:bufferImageHeight specify the
    data in buffer memory as a subregion of a larger two- or
    three-dimensional image, and control the addressing calculations of data
    in buffer memory. If either of these values is zero, that aspect of the
    buffer memory is considered to be tightly packed according to the
    pname:imageExtent.
  * pname:imageSubresource is an slink:VkImageSubresourceLayers used to
    specify the specific subresources of the image used for the source or
    destination image data.
  * pname:imageOffset selects the initial x, y, z offsets in texels of the
    sub-region of the source or destination image data.
  * pname:imageExtent is the size in texels of the image to copy in
    pname:width, pname:height and pname:depth. 1D images use only pname:x
    and pname:width. 2D images use pname:x, pname:y, pname:width and
    pname:height. 3D images use pname:x, pname:y, pname:z, pname:width,
    pname:height and pname:depth.

When copying to or from a depth or stencil aspect, the data in buffer
memory uses a layout that is a (mostly) tightly packed representation of
the depth or stencil data. Specifically:

   * data copied to or from the stencil aspect of any depth/stencil format
     is tightly packed with one ename:VK_FORMAT_S8_UINT value per texel.
   * data copied to or from the depth aspect of a ename:VK_FORMAT_D16_UNORM
     or ename:VK_FORMAT_D16_UNORM_S8_UINT format is tightly packed with one
     ename:VK_FORMAT_D16_UNORM value per texel.
   * data copied to or from the depth aspect of a ename:VK_FORMAT_D32_SFLOAT
     or ename:VK_FORMAT_D32_SFLOAT_S8_UINT format is tightly packed with
     one ename:VK_FORMAT_D32_SFLOAT value per texel.
   * data copied to or from the depth aspect of a
     ename:VK_FORMAT_X8_D24_UNORM_PACK32 or
     ename:VK_FORMAT_D24_UNORM_S8_UINT format is packed with one 32-bit word
     per texel with the D24 value in the LSBs of the word, and undefined
     values in the eight MSBs.

[NOTE]
.Note
====
To copy both the depth and stencil aspects of a depth/stencil format, two
entries in pname:pRegions can: be used, where one specifies the depth
aspect in pname:imageSubresource, and the other specifies the stencil
aspect.
====

Because depth or stencil aspect buffer to image copies may: require format
conversions on some implementations, they are not supported on queues
that do not support graphics.

Copies are done layer by layer starting with image layer
pname:baseArrayLayer member of pname:imageSubresource. pname:layerCount
layers are copied from the source image or to the destination image.

include::../validity/structs/VkBufferImageCopy.txt[]

Pseudocode for image/buffer addressing is:

[source,c]
---------------------------------------------------
rowLength = region->bufferRowLength;
if (rowLength == 0)
    rowLength = region->imageExtent.width;

imageHeight = region->bufferImageHeight;
if (imageHeight == 0)
    imageHeight = region->imageExtent.height;

texelSize = <texel size taken from the src/dstImage>;

address of (x,y,z) = region->bufferOffset + (((z * imageHeight) + y) * rowLength + x) * texelSize;

where x,y,z range from (0,0,0) to region->imageExtent.{width,height,depth}.
---------------------------------------------------

Note that pname:imageOffset does not affect addressing calculations for
buffer memory. Instead, pname:bufferOffset can: be used to
select the starting address in buffer memory.

For block-compression formats, all parameters are still specified in texels
rather than blocks, but the addressing math operates on whole blocks.
Pseudocode for compressed copy addressing is:

[source,c]
---------------------------------------------------
rowLength = region->bufferRowLength;
if (rowLength == 0)
    rowLength = region->imageExtent.width;

imageHeight = region->bufferImageHeight;
if (imageHeight == 0)
    imageHeight = region->imageExtent.height;

blockSizeInBytes = <block size taken from the src/dstImage>;
rowLength /= blockWidth;
imageHeight /= blockHeight;

address of (x,y,z) = region->bufferOffset + (((z * imageHeight) + y) * rowLength + x) * blockSizeInBytes;

where x,y,z range from (0,0,0) to region->imageExtent.{width/blockWidth,height/blockHeight,depth/blockDepth}.
---------------------------------------------------

Copying to or from block-compressed images is typically done in multiples of
the block. For this reason the pname:imageExtent must: be a multiple of the
block dimension. There is one exception to this rule which is required: to
handle compressed images created with dimensions that are not a multiple of
the block dimensions. If pname:imageExtent.width is not a multiple of the
block width then (pname:imageExtent.width + pname:imageOffset.x) must: equal
the subresource width, if pname:imageExtent.height is not a multiple of the
block height then (pname:imageExtent.height + pname:imageOffset.y) must:
equal the subresource height and if pname:imageExtent.depth is not a
multiple of the block depth then (pname:imageExtent.depth +
pname:imageOffset.z) must: equal the subresource depth. This allows the last
block of the image in each non-multiple dimension to be included as a source
or target of the copy.


[[copies-imagescaling]]
== Image Copies with Scaling

To copy regions of a source image into a destination image, potentially
performing format conversion, arbitrary scaling, and filtering, call:

include::../protos/vkCmdBlitImage.txt[]

  * pname:commandBuffer is the command buffer into which the command will be
    recorded.
  * pname:srcImage is the source image.
  * pname:srcImageLayout is the layout of the source image subresources for
    the blit.
  * pname:dstImage is the destination image.
  * pname:dstImageLayout is the layout of the destination image subresources
    for the blit.
  * pname:regionCount is the number of regions to blit.
  * pname:pRegions is a pointer to an array of slink:VkImageBlit structures
    specifying the regions to blit.
  * pname:filter is a elink:VkFilter specifying the filter to apply if the
    blits require scaling.

include::../validity/protos/vkCmdBlitImage.txt[]

fname:vkCmdBlitImage mustnot: be used for multisampled source or destination
images. Use flink:vkCmdResolveImage for this purpose.

Each element of pname:pRegions is a structure defined as:

include::../structs/VkImageBlit.txt[]

For each element of the pname:pRegions array, a blit operation is performed
between the region of pname:srcSubresource of pname:srcImage (bounded by
pname:srcOffsets[0] and pname:srcOffsets[1]) and a region of
pname:dstSubresource of pname:dstImage (bounded by pname:dstOffsets[0] and
pname:dstOffsets[1]).

include::../validity/structs/VkImageBlit.txt[]

If sizes of source and destination extents do not match, scaling is
performed by applying the filtering mode specified by pname:filter
parameter. ename:VK_FILTER_LINEAR uses bilinear interpolation, and
ename:VK_FILTER_NEAREST uses point sampling. When using
ename:VK_FILTER_LINEAR, magnifying blits may: generate texel coordinates
outside of the source region. If those coordinates are outside the bounds of
the image level, the coordinates are clamped as in
ename:VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE address mode. However, if the
coordinates are outside the source region but inside the image level, the
implementation may: clamp coordinates to the source region.

If source and destination extents are identical, no filtering is performed.
Pixels in the axis-aligned region bounded by srcOffsets[0] and srcOffsets[1]
will be copied to the destination region bound by dstOffsets[0] and
dstOffsets[1]. In the case where dstOffsets[0].x > dstOffsets[1].x the
copied pixels are reversed in that direction. Likewise for y and z.

Blits are done layer by layer starting with the pname:baseArrayLayer member
of pname:srcSubresource for the source and pname:dstSubresource for the
destination. pname:layerCount layers are blitted to the destination image.

3D textures are blitted slice by slice. Slices in the source region bounded
by pname:srcOffsets[0].z and pname:srcOffsets[1].z are copied to slices in
the destination region bounded by pname:dstOffsets[0].z and
pname:dstOffsets[1].z. For each destination slice, a source z coordinate is
linearly interpolated between pname:srcOffsets[0].z and
pname:srcOffsets[1].z. If the pname:filter parameter is
ename:VK_FILTER_LINEAR then the value sampled from the source image is taken
by doing linear filtering using the interpolated z coordinate. If
pname:filter parameter is ename:VK_FILTER_NEAREST then value sampled from
the source image is taken from the single nearest slice (with undefined
rounding mode).

If fname:vkCmdBlitImage is used on images of different formats, the
following conversion rules apply:

  * Integer formats can: only be converted to other integer formats with the
    same signedness.
  * No format conversion is supported between depth/stencil images - the
    formats must: match.
  * Format conversions on unorm, snorm, unscaled and packed float formats of
    the copied aspect of the image are performed by first converting the
    pixels to float values.
  * In case of sRGB source format, values are converted to linear color
    space prior to filtering.
  * After filtering, the float values are first clamped and then cast to the
    destination image format. In case of sRGB destination format, values are
    converted to sRGB color space before writing the pixel to the image.

Signed and unsigned integers are converted by first clamping to the
representable range of the destination format, then casting the value.


[[copies-resolve]]
== Resolving Multisample Images

To resolve a multisample image to a non-multisample image, call:

include::../protos/vkCmdResolveImage.txt[]

  * pname:commandBuffer is the command buffer into which the command will be
    recorded.
  * pname:srcImage is the source image.
  * pname:srcImageLayout is the layout of the source image subresources for
    the blit.
  * pname:dstImage is the destination image.
  * pname:dstImageLayout is the layout of the destination image subresources
    for the blit.
  * pname:regionCount is the number of regions to blit.
  * pname:pRegions is a pointer to an array of slink:VkImageResolve
    structures specifying the regions to resolve.

include::../validity/protos/vkCmdResolveImage.txt[]

Each element of pname:pRegions is a structure defined as:

include::../structs/VkImageResolve.txt[]

  * pname:srcSubresource and pname:dstSubresource are
    slink:VkImageSubresourceLayers structures specifying the subresources of
    the images used for the source and destination image data, respectively.
    Resolve of depth/stencil images is not supported.
  * pname:srcOffset and pname:dstOffset select the initial x, y, and z
    offsets in texels of the sub-regions of the source and destination image
    data.
  * pname:extent is the size in texels of the source image to resolve in
    pname:width, pname:height and pname:depth. 1D images use only pname:x
    and pname:width. 2D images use pname:x, pname:y, pname:width and
    pname:height. 3D images use pname:x, pname:y, pname:z, pname:width,
    pname:height and pname:depth.

include::../validity/structs/VkImageResolve.txt[]

During the resolve the samples corresponding to each pixel location in the
source are converted to a single sample before being written to the
destination. If the source formats are floating-point or normalized types,
the sample values for each pixel are resolved in an implementation-dependent
manner. If the source formats are integer types, a single sample's value is
selected for each pixel.

pname:srcOffset and pname:dstOffset select the initial x, y, and z
offsets in texels of the sub-regions of the source and destination image
data. pname:extent is the size in texels of the source
image to resolve in pname:width, pname:height and pname:depth. 1D images use
only pname:x and pname:width. 2D images use pname:x, pname:y, pname:width
and pname:height. 3D images use pname:x, pname:y, pname:z, pname:width,
pname:height and pname:depth.

Resolves are done layer by layer starting with pname:baseArrayLayer member
of pname:srcSubresource for the source and pname:dstSubresource for the
destination. pname:layerCount layers are blitted to the destination image.