The Mode member of the structure has this enumeration type. That member specifies the alpha-fill mode for the input stream identified by the StreamNumber member of the same structure. To set the alpha-fill mode, call .

To find out which modes the device supports, call the method. If the device sets the flag in the FeatureCaps member of the structure, the DXVA-HD device supports any of the modes listed here. Otherwise, the alpha-fill mode must be .

If the Enable member is TRUE, the device downsamples the composed target rectangle to the size given in the Size member, and then scales it back to the size of the target rectangle.

The width and height of Size must be greater than zero. If the size is larger than the target rectangle, downsampling does not occur.

To use this state, the device must support downsampling, indicated by the capability flag. To query for this capability, call . If the device supports downsampling, it sets the flag in the FeatureCaps member of the structure.

If the device does not support downsampling, the method fails for this state.

Downsampling is sometimes used to reduce the quality of premium content when other forms of content protection are not available.

The graphics driver uses one of these enumeration constants as a hint when it creates the DXVA-HD device.

The multiplier enables the filter range to have a fractional step value.

For example, a hue filter might have an actual range of [-180.0 ... +180.0] with a step size of 0.25. The device would report the following range and multiplier:

• Minimum: -720
• Maximum: +720
• Multiplier: 0.25

In this case, a filter value of 2 would be interpreted by the device as 0.50 (or 2 ? 0.25).

The device should use a multiplier that can be represented exactly as a base-2 fraction.

These flags define video processing capabilities that are usually not needed, and therefore are not required for DXVA-HD devices to support.

The first three flags relate to RGB support for functions that are normally applied to YCbCr video: deinterlacing, color adjustment, and luma keying. A DXVA-HD device that supports these functions for YCbCr is not required to support them for RGB input. Supporting RGB input for these functions is an additional capability, reflected by these constants. The driver might convert the input to another color space, perform the indicated function, and then convert the result back to RGB.

Similarly, a device that supports de-interlacing is not required to support deinterlacing of palettized formats. This capability is indicated by the flag.

For YUV colors, these flags specify how to convert between Y'CbCr and Y'PbPr. The Y'PbPr color space has a range of [0..1] for Y' (luma) and [-0.5...0.5] for Pb/Pr (chroma).

?

For RGB colors, the flags differentiate various RGB spaces.

?

Video data might contain values above or below the nominal range.

Note??The values named and are a potential source of confusion. Wide refers to the possible range of analog values that can be represented, by mapping the nominal range [0...1] into a narrower range of digital values. Because the meaning of wide in this context is ambiguous, the equivalent values named and are preferred. These names explicitly convey the meaning of the enumeration, and are less likely to be misinterpreted.

This enumeration is equivalent to the DXVA_NominalRange enumeration used in DXVA 1.0, although it defines additional values.

If you are using the interface to describe the video format, the nominal range is specified in the attribute.

This enumeration is equivalent to the DXVA_SampleFormat enumeration used in DXVA 1.0.

The following table shows the mapping from enumeration values, which are used in Media Foundation media types, to values.

?

With the exception of , each pair of corresponding enumeration values has the same numeric value.

The value has no equivalent in the enumeration.

To use this state, the device must support luma keying, indicated by the capability flag. To query for this capability, call . If the device supports luma keying, it sets the flag in the FeatureCaps member of the structure.

If the device does not support luma keying, the method fails for this state.

If the input format is RGB, the device must also support the capability. This capability flag is set in the InputFormatCaps member of the structure. If the flag is not present, the device ignores the luma key value for RGB input.

The values of Lower and Upper give the lower and upper bounds of the luma key, using a nominal range of [0...1]. Given a format with n bits per channel, these values are converted to luma values as follows:

val = f * ((1 << n)-1)

Any pixel whose luma value falls within the upper and lower bounds (inclusive) is treated as transparent.

For example, if the pixel format uses 8-bit luma, the upper bound is calculated as follows:

BYTE Y = BYTE(max(min(1.0, Upper), 0.0) * 255.0)

Note that the value is clamped to the range [0...1] before multiplying by 255.

If the DXVA-HD device is a software plug-in and the surface type is , the device can support format types that are not supported natively by the graphics driver. For example, if the application requests an AYUV surface, the device could allocate a surface with a surface type of .

These flags are used with the attribute.

For more information about these values, see the remarks for the enumeration, which is the DirectX Video Acceleration (DXVA) equivalent of this enumeration.

This enumeration is equivalent to the DXVA_VideoLighting enumeration used in DXVA 1.0.

If you are using the interface to describe the video format, the video lighting is specified in the attribute.

Color primaries define how to convert RGB colors into the CIE XYZ color space, and can be used to translate colors between different RGB color spaces. An RGB color space is defined by the chromaticity coordinates (x,y) of the RGB primaries plus the white point, as listed in the following table.

?

The z coordinates can be derived from x and y as follows: z = 1 - x - y. To convert between RGB colors to CIE XYZ tristimulus values, compute a matrix T as follows:

Given T, you can use the following formulas to convert between an RGB color value and a CIE XYZ tristimulus value. These formulas assume that the RGB components are linear (not gamma corrected) and are normalized to the range [0...1].

To convert colors directly from one RGB color space to another, use the following formula, where T1 is the matrix for color space RGB1, and T2 is the matrix for color space RGB2.

For a derivation of these formulas, refer to Charles Poynton, Digital Video and HDTV Algorithms and Interfaces (Morgan Kaufmann, 2003).

This enumeration is equivalent to the DXVA_VideoPrimaries enumeration used in DXVA 1.0.

If you are using the interface to describe the video format, the color primaries are specified in the attribute.

The following table shows the formulas for the most common transfer functions. In these formulas, L is the linear value and L' is the non-linear (gamma corrected) value. These values are relative to a normalized range [0...1].

?

The following table shows the inverse formulas to obtain the original gamma-corrected values:

?

This enumeration is equivalent to the DXVA_VideoTransferFunction enumeration used in DXVA 1.0.

If you are using the interface to describe the video format, the transfer function is specified in the attribute.

The transfer matrices are defined as follows.

BT.709 transfer matrices:

BT.601 transfer matrices:

SMPTE 240M (SMPTE RP 145) transfer matrices:

This enumeration is equivalent to the DXVA_VideoTransferMatrix enumeration used in DXVA 1.0.

If you are using the interface to describe the video format, the video transfer matrix is specified in the attribute.

Use the interface to get the device capabilities, create the video processor, and allocate video surfaces.

To find out which image filters the device supports, check the FilterCaps member of the structure. Call the method to get this value.

The list of formats can include both values, such as , and FOURCC codes, such as 'NV12'. For more information, see Video FOURCCs.

To find out which image filters the device supports, check the FilterCaps member of the structure. Call the method to get this value.

Call this method to set state parameters that apply to individual input streams.

The maximum value of StreamCount is given in the MaxStreamStates member of the structure. The maximum numbr of streams that can be enabled at one time is given in the MaxInputStreams member of that structure.

This interface is exposed by the Direct3D Device Manager. To create the Direct3D device manager, call .

To get this interface from the Enhanced Video Renderer (EVR), call . The service is MR_VIDEO_ACCELERATION_SERVICE. For the DirectShow EVR filter, call GetService on the filter's pins.

The Direct3D Device Manager supports Direct3D 9 devices only. It does not support DXGI devices.

When you first create the Direct3D device manager, call this method with a reference to the Direct3D device. The device manager does not create the device; the caller must provide the device reference initially.

Also call this method if the Direct3D device becomes lost and you need to reset the device or create a new device. This occurs if returns or . For more information about lost devices, see the Direct3D documentation.

The resetToken parameter ensures that only the component which originally created the device manager can invalidate the current device.

If this method succeeds, all open device handles become invalid.

To get the Direct3D device's reference, call with the handle returned in phDevice. Close the device handle when you are done using it, by calling .

To test whether a device handle is still valid, call .

If the method returns DXVA2_E_NEW_VIDEO_DEVICE, call to close the handle and then call OpenDeviceHandle again to get a new handle. The method invalidates all open device handles.

When you are done using the Direct3D device, call to unlock to the device.

If the method returns DXVA2_E_NEW_VIDEO_DEVICE, call to close the handle and then call OpenDeviceHandle again to get a new handle. The method invalidates all open device handles.

If fBlock is TRUE, this method can potentially deadlock. For example, it will deadlock if a thread calls LockDevice and then waits on another thread that calls LockDevice. It will also deadlock if a thread calls LockDevice twice without calling UnlockDevice in between.

If the method returns DXVA2_E_NEW_VIDEO_DEVICE, call to close the handle and then call OpenDeviceHandle again to get a new handle. The method invalidates all open device handles.

This is the base interface for DXVA services. The Direct3D device can support any of the following DXVA services, which derive from :

• Video decoding:
• Video processing:

If the method returns E_FAIL, try calling to reset the DirectX Video Acceleration Manager.

The methods make calls to the Direct3D device. Therefore, the flags that you specify when creating the device can affect the behavior of this interface. For example, if you specify the flag, the Direct3D global critical section will be held during decode operations.

You can set any parameter to null if you are not interested in the result. At least one parameter must be non-null.

If you specify a non-null value for pppDecoderRenderTargets (to receive the render target surfaces), then pNumSurfaces cannot be null, because it receives the size of the array returned in pppDecoderRenderTargets.

The method locks the Direct3D surface that contains the buffer. When you are done using the buffer, call to unlock the surface.

This method might block if too many operations have been queued on the GPU. The method unblocks when a free buffer becomes available.

After this method is called, call to perform decoding operations. When all decoding operations have been executed, call .

Each call to BeginFrame must have a matching call to EndFrame, and BeginFrame calls cannot be nested.

DXVA 1.0 migration note: Unlike the IAMVideoAccelerator::BeginFrame method, which specifies the buffer as an index, this method takes a reference directly to the uncompressed buffer.

The surface pointed to by pRenderTarget must be created by calling with the value for DxvaType.

You must call before calling this method.

The following decoder GUIDs are defined. Some of these GUIDs have alternate names, shown in parentheses.

?

By calling this method, the caller agrees to create surfaces of the type specified in the dwType parameter.

In DirectShow, during pin connection, a video decoder that supports DVXA 2.0 should call SetSurface with the value . This notifies the video renderer that the decoder will provide the allocator and will create the Direct3D surfaces for decoding. For more information, see Supporting DXVA 2.0 in DirectShow.

The only way to undo the setting is to break the pin connection.

By calling this method, the caller agrees to create surfaces of the type specified in the dwType parameter.

In DirectShow, during pin connection, a video decoder that supports DVXA 2.0 should call SetSurface with the value . This notifies the video renderer that the decoder will provide the allocator and will create the Direct3D surfaces for decoding. For more information, see Supporting DXVA 2.0 in DirectShow.

The only way to undo the setting is to break the pin connection.

You can set any parameter to null if you are not interested in the result. At least one parameter must be non-null.

You can set any parameter to null if you are not interested in the result. At least one parameter must be non-null.

When the method returns, the operation might not be complete.

If the method returns E_INVALIDARG, check for the following:

• The number of input samples (NumSamples) must be less than or equal to MAX_DEINTERLACE_SURFACES.
• The Direct3D surface must be a valid target for VideoProcessBlt. See the description of the pRT parameter for details.
• The presentation time (TargetFrame) given in pBltParams must match the start and end times for the current picture from the primary stream. Specifically, it must be less than the end time and greater than or equal to the start time. Note that the first sample in pSamples might not be the current picture, if the pSamples array contains backward reference pictures. For more information, see Input Sample Order.
• The target rectangle (TargetRect) given in pBltParams cannot be larger than the destination surface (pRT).
• The constriction size (ConstrictionSize) given in pBltParams cannot be less than zero or larger than the target rectangle.
• The alpha component of the background color must be opqaue.
• The ProcAmp values given in pBltParams must be valid. For any ProcAmp settings that are supported by the driver, these values must fall within the ranges returned by the method.
• The noise and detail filters given in pBltParams must be valid. For any filters that are supported by the driver, these values must fall within the ranges returned by the method.
• The alpha value given in pBltParams must be in the range [0...1] inclusive.
• For each input sample given in pSamples:
  • The start time cannot be greater than the end time.
  • A valid reference must be provided.
  • The source rectangle cannot be larger than the input surface.
  • The destination rectangle cannot be larger than than the destination surface.
  • The planar alpha must be in the range [0...1] inclusive.
• For all rectangles (source, destination, and target), the rectangle cannot be inverted (left > right or top > bottom) or have negative values.

The following video processor GUIDs are predefined.

?

The graphics device may define additional vendor-specific GUIDs. The driver provides the list of GUIDs in descending quality order. The mode with the highest quality is first in the list. To get the capabilities of each mode, call and pass in the for the mode.

For AES-CTR encyption, the pvPVPState member of the structure points to a structure.

The D3DAES_CTR_IV structure and the structure are equivalent.

If the IDirect3DDevice9Video::GetContentProtectionCaps method returns the D3DCPCAPS_SEQUENTIAL_CTR_IV flag, the caller should keep IV unchanged when submitting multiple buffers for the same video frame, and Count should be in sequential order of the previous submission for the frame.

Example: Suppose the software decoder submits three buffers for a single frame, and that each buffer contains three 128-bit blocks. For the first buffer, IV can be any value. For the next two buffers, the same value of IV must be used. The value of Count starts at 1. For the second buffer, Count equals 4 (1 + 3 blocks from the first submission). For the third buffer, Count equals 7 (4 + 3 blocks from the second submission).

When the D3DCPCAPS_SEQUENTIAL_CTR_IV capability is present, it is recommended to submit data in 128-bit blocks.

For each pixel, the destination color value is computed as follows:

Cd = Cs * (As * Ap * Ae) + Cd * (1.0 - As * Ap * Ae)

where

• Cd = Color value of the destination pixel.
• Cs = Color value of source pixel.
• As = Per-pixel source alpha.
• Ap = Planar alpha value.
• Ae = Palette-entry alpha value, or 1.0 (see Note).

Note??Palette-entry alpha values apply only to palettized color formats, and only when the device supports the capability. Otherwise, this factor equals 1.0.

The destination alpha value is computed according to the state. For more information, see .

To get the device capabilities, call and check the FeatureCaps member of the structure.

If the Enable member is TRUE, the device downsamples the composed target rectangle to the size given in the Size member, and then scales it back to the size of the target rectangle.

The width and height of Size must be greater than zero. If the size is larger than the target rectangle, downsampling does not occur.

To use this state, the device must support downsampling, indicated by the capability flag. To query for this capability, call . If the device supports downsampling, it sets the flag in the FeatureCaps member of the structure.

If the device does not support downsampling, the method fails for this state.

Downsampling is sometimes used to reduce the quality of premium content when other forms of content protection are not available.

The RGB_Range member applies to RGB output, while the YCbCr_Matrix and YCbCr_xvYCC members apply to YCbCr (YUV) output. If the device performs color-space conversion on the background color, it uses the values that apply to both color spaces.

Extended YCbCr can be used with either transfer matrix. Extended YCbCr does not change the black point or white point?the black point is still 16 and the white point is still 235. However, extended YCbCr explicitly allows blacker-than-black values in the range 1?15, and whiter-than-white values in the range 236?254. When extended YCbCr is used, the driver should not clip the luma values to the nominal 16?235 range.

If the device supports extended YCbCr, it sets the capability flag in the DeviceCaps member of the structure. Otherwise, the device ignores the value of the YCbCr_xvYCC member and treats all YCbCr output as conventional YCbCr. To get the device's capabilities, call .

If the output format is a wide-gamut RGB format, output might fall outside the nominal [0...1] range of sRGB. This is particularly true if one or more input streams use extended YCbCr.

Values have a nominal range of [0...1]. Given a format with n bits per channel, the value of each color component is calculated as follows:

val = f * ((1 << n)-1)

For example, for 8-bit YUV formats, val = BYTE(f * 255.0).

Reference black is (0.0625, 0.5, 0.5), which corresponds to (16, 128, 128) in an 8-bit representation.

The RGB values have a nominal range of [0...1]. For an RGB format with n bits per channel, the value of each color component is calculated as follows:

val = f * ((1 << n)-1)

For example, for RGB-32 (8 bits per channel), val = BYTE(f * 255.0).

For full-range RGB, reference black is (0.0, 0.0, 0.0), which corresponds to (0, 0, 0) in an 8-bit representation. For limited-range RGB, reference black is (0.0625, 0.0625, 0.0625), which corresponds to (16, 16, 16) in an 8-bit representation. For wide-gamut formats, the values might fall outside of the [0...1] range.

Frame rates are expressed as ratios. For example, 30 frames per second (fps) is expressed as 30:1, and 29.97 fps is expressed as 30000/1001. For interlaced content, a frame consists of two fields, so that the frame rate is half the field rate.

If the application will composite two or more input streams, use the largest stream for the values of InputWidth and InputHeight.

For AES-CTR encyption, the pvPVPState member of the structure points to a structure.

The D3DAES_CTR_IV structure and the structure are equivalent.

If the IDirect3DDevice9Video::GetContentProtectionCaps method returns the D3DCPCAPS_SEQUENTIAL_CTR_IV flag, the caller should keep IV unchanged when submitting multiple buffers for the same video frame, and Count should be in sequential order of the previous submission for the frame.

Example: Suppose the software decoder submits three buffers for a single frame, and that each buffer contains three 128-bit blocks. For the first buffer, IV can be any value. For the next two buffers, the same value of IV must be used. The value of Count starts at 1. For the second buffer, Count equals 4 (1 + 3 blocks from the first submission). For the third buffer, Count equals 7 (4 + 3 blocks from the second submission).

When the D3DCPCAPS_SEQUENTIAL_CTR_IV capability is present, it is recommended to submit data in 128-bit blocks.

You must call before calling this method.

This structure corresponds to parameters of the IAMVideoAccelerator::Execute method in DirectX Video Acceleration (DXVA) version 1.

Most of the values in this structure can be translated directly to and from attributes. For a code example that fills in the values from an reference, see .

The multiplier enables the filter range to have a fractional step value.

For example, a hue filter might have an actual range of [-180.0 ... +180.0] with a step size of 0.25. The device would report the following range and multiplier:

• Minimum: -720
• Maximum: +720
• Multiplier: 0.25

In this case, a filter value of 2 would be interpreted by the device as 0.50 (or 2 ? 0.25).

The device should use a multiplier that can be represented exactly as a base-2 fraction.

The value 0/0 indicates an unknown frequency. Values of the form n/0, where n is not zero, are invalid. Values of the form 0/n, where n is not zero, indicate a frequency of zero.

Before calling this method, set the video processor mode. To select a video processor mode, call . Otherwise, the EVR automatically selects a mode when streaming begins.

To find out which ProcAmp settings the driver supports, call .

Pixel aspect ratios of the form 0/n and n/0 are not valid.

If the Enable member is , the device ignores the values of SourceAspectRatio and DestinationAspectRatio.

The Direct3D surfaces must be allocated in the memory pool specified by the InputPool member of the structure. The following surface types can be used:

• A video surface of type or . See .
• A decoder render-target surface of type . See .
• An off-screen plain surface.

The past and future reference frames must be placed in the arrays in temporal order, from oldest to newest. For example, if T is the current input frame, the arrays would be ordered as follows:

• ppPastSurfaces: { ?, T-3, T-2, T-1 }
• ppInputSurface: T
• ppFutureSurfaces: { T+1, T+2, T+3, ? }

The structure specifies the number of reference frames required to get the best deinterlacing quality. If the application provides fewer reference frames, the device will fall back to simpler deinterlacing algorithms. If no reference frames are provided, the device can use bob deinterlacing. Here are some cases where an application might provide fewer reference frames:

• At the beginning or end of the video sequence
• With progressive input
• During a transition between progressive and interlaced input
• For a substream that does not require high deinterlacing quality
• While dropping frames

The OutputIndex and InputFrameOrField members are used to correlate input frames or fields with output frames. The value of the OutputIndex member is cyclic and resets to zero after each cycle.

Here are some example patterns:

• Progressive video at normal rate. Each input frame produces one output frame.
  • Output index: 0, 0, 0, 0, ?
  • Input index: 0, 1, 2, 3, ?
• Interlaced video at normal rate. Each interlaced frame (two fields) produces two output frames.
  • Output index: 0, 1, 0, 1, ?
  • Input index: 0, 1, 2, 3, ?
• Progressive video at 2/1 output rate. Each input frame produces one output frame.
  • Output index: 0, 1, 0, 1, ?
  • Input index: 0, 1, 2, 3, ?
• Interlaced video at 1/2 output rate. Each interlaced frame produces one output frame. (The two fields are blended to create one frame.)
  • Output index: 0, 0, 0, 0, ?
  • Input index: 0, 2, 4, 6, ?

For each pixel, the destination color value is computed as follows:

Cd = Cs * (As * Ap * Ae) + Cd * (1.0 - As * Ap * Ae)

where

• Cd = Color value of the destination pixel.
• Cs = Color value of source pixel.
• As = Per-pixel source alpha.
• Ap = Planar alpha value.
• Ae = Palette-entry alpha value, or 1.0 (see Note).

Note??Palette-entry alpha values apply only to palettized color formats, and only when the device supports the capability. Otherwise, this factor equals 1.0.

The destination alpha value is computed according to the state. For more information, see .

To get the device capabilities, call and check the FeatureCaps member of the structure.

Pixel aspect ratios of the form 0/n and n/0 are not valid.

If the Enable member is , the device ignores the values of SourceAspectRatio and DestinationAspectRatio.

For a list of image filters that are defined for DXVA-HD, see . The device might not support every type of image filter. To find out whether the device supports a particular filter, call the method and check the FilterCaps member of the structure.

The RGB_Range member applies to RGB input, while the YCbCr_Matrix and YCbCr_xvYCC members apply to YCbCr (YUV) input.

In some situations, the device might perform an intermediate color conversion on the input stream. If so, it uses the flags that apply to both color spaces. For example, suppose the device converts from RGB to YCbCr. If the RGB_Range member is 0 and the YCbCr_Matrix member is 1, the device will convert from full-range RGB to BT.709 YCbCr.

If the device supports xvYCC, it returns the capability flag in the DeviceCaps member of the structure. Otherwise, the device ignores the value of YCbCr_xvYCC and treats all YCbCr input as conventional YCbCr. To get the device's capabilities, call .

To use this state, the device must support luma keying, indicated by the capability flag. To query for this capability, call . If the device supports luma keying, it sets the flag in the FeatureCaps member of the structure.

If the device does not support luma keying, the method fails for this state.

If the input format is RGB, the device must also support the capability. This capability flag is set in the InputFormatCaps member of the structure. If the flag is not present, the device ignores the luma key value for RGB input.

The values of Lower and Upper give the lower and upper bounds of the luma key, using a nominal range of [0...1]. Given a format with n bits per channel, these values are converted to luma values as follows:

val = f * ((1 << n)-1)

Any pixel whose luma value falls within the upper and lower bounds (inclusive) is treated as transparent.

For example, if the pixel format uses 8-bit luma, the upper bound is calculated as follows:

BYTE Y = BYTE(max(min(1.0, Upper), 0.0) * 255.0)

Note that the value is clamped to the range [0...1] before multiplying by 255.

The output rate might require the device to convert the frame rate of the input stream. If so, the value of RepeatFrame controls whether the device creates interpolated frames or simply repeats input frames.

This structure corresponds to parameters of the IAMVideoAccelerator::Execute method in DirectX Video Acceleration (DXVA) version 1.

The InputSampleFreq member gives the frame rate of the decoded video stream, as received by the video renderer. The OutputFrameFreq member gives the frame rate of the video that is displayed after deinterlacing. If the input video is interlaced and the samples contain interleaved fields, the output frame rate is twice the input frame rate. If the input video is progressive or contains single fields, the output frame rate is the same as the input frame rate.

Decoders should set the values of InputSampleFreq and OutputFrameFreq if the frame rate is known. Otherwise, set these members to 0/0 to indicate an unknown frame rate.