Nvvfx videosuperres - #797
Open
cyj98 wants to merge 4 commits into
Open
Conversation
There was a problem hiding this comment.
🟡 Changes recommended
There are confirmed logic issues in the NVVFX path (misleading global runtime buffer logs/warnings and a width/height check that can skip the NVVFX pass).
Once you've addressed the issues Copilot identified, you can request another Copilot review.
Pull request overview
Adds a new NVVFX-based resize algorithm nvvfx-videosuperres (MAXINE VideoEffects SDK 1.2 “VideoSuperRes”) to NVEnc’s NVEncCore resize pipeline, including CLI parameter plumbing and documentation/help updates.
Changes:
- Introduces
RGY_VPP_RESIZE_NVVFX_VIDEO_SUPER_RES+nvvfx-videosuperresoption and wires it through resize selection/printing. - Adds
videosuperres-quality=<int>parsing and command/help/doc generation. - Implements
NVEncFilterNvvfxVideoSuperRes, including RGBA U8 (SDK 1.2+) vs legacy BGR F32 (older runtime) buffer handling and ratio-search initialization.
File summaries
| File | Description |
|---|---|
| NVEncCore/rgy_prm.h | Adds the new resize enum + string mapping and exposes videosuperres-quality as a resize parameter. |
| NVEncCore/rgy_cmd.cpp | Extends --vpp-resize help text with videosuperres-quality semantics and constraints. |
| NVEncCore/NVEncFilterResize.cu | Adds initialization path and runtime execution support for the new NVVFX VideoSuperRes resize filter. |
| NVEncCore/NVEncFilterParam.h | Adds VppNvvfxVideoSuperRes and default quality constant to the shared VPP parameter model. |
| NVEncCore/NVEncFilterParam.cpp | Implements printing/comparison, parses videosuperres-quality, and emits it in generated command lines. |
| NVEncCore/NVEncFilterNvvfx.h | Adds VideoSuperRes effect selectors + new parameter/filter class declarations and RGBA buffer support members. |
| NVEncCore/NVEncFilterNvvfx.cpp | Implements the VideoSuperRes NVVFX effect, including buffer allocation/conversion changes. |
| NVEncCore/NVEncFilter.h | Extends resize filter params/state to carry the new NVVFX VideoSuperRes configuration and instance. |
| NVEncCore/NVEncCore.cpp | Ensures NVVFX is enabled for the new algorithm and enforces same-res creation behavior for quality 8–15. |
| NVEncC_Options.en.md | Documents nvvfx-videosuperres and videosuperres-quality usage/constraints (English). |
| NVEncC_Options.ja.md | Documents nvvfx-videosuperres and videosuperres-quality usage/constraints (Japanese). |
| NVEncC_Options.zh-cn.md | Documents nvvfx-videosuperres and videosuperres-quality usage/constraints (Simplified Chinese). |
Review details
- Files reviewed: 12/12 changed files
- Comments generated: 2
- Review effort level: Lite
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
Comment on lines
+2020
to
+2022
| if (pResizeParam->frameOut.width != ppOutputFrames[0]->width | ||
| && pResizeParam->frameOut.height != ppOutputFrames[0]->height) { | ||
| int nvvfxOutputNum = 0; |
Comment on lines
+183
to
+204
| // VFX SDK 1.2+ expects RGBA U8 interleaved buffers for VideoSuperRes, | ||
| // while pre-1.2 runtimes (0.7.x) only accept the legacy BGR F32 planar | ||
| // layout. Pick the buffer format by the loaded runtime version so that | ||
| // behavior on older runtimes stays identical to the legacy format. | ||
| { | ||
| uint32_t vfxVersion = 0; | ||
| if (err_to_rgy(NvVFX_GetVersion(&vfxVersion)) == RGY_ERR_NONE | ||
| && vfxVersion >= ((1u << 24) | (2u << 16))) { | ||
| AddMessage(RGY_LOG_DEBUG, _T("nvvfx runtime version %d.%d: using RGBA U8 buffers.\n"), | ||
| (vfxVersion >> 24) & 0xff, (vfxVersion >> 16) & 0xff); | ||
| } else { | ||
| m_bgraU8 = false; | ||
| AddMessage(RGY_LOG_WARN, _T("nvvfx runtime version %d.%d does not support RGBA U8 buffers (requires 1.2+), using legacy BGR F32 planar buffers.\n"), | ||
| (vfxVersion >> 24) & 0xff, (vfxVersion >> 16) & 0xff); | ||
| if (auto prmVsr = dynamic_cast<const NVEncFilterParamNvvfxVideoSuperRes*>(pParam.get()); | ||
| prmVsr != nullptr && prmVsr->nvvfxVideoSuperRes.quality >= 8) { | ||
| AddMessage(RGY_LOG_ERROR, _T("quality 8-19 requires VFX SDK 1.2 or later, but the loaded nvvfx runtime is %d.%d. Please install or update the NVIDIA Video Effects runtime, or use quality 1-4.\n"), | ||
| (vfxVersion >> 24) & 0xff, (vfxVersion >> 16) & 0xff); | ||
| return RGY_ERR_UNSUPPORTED; | ||
| } | ||
| } | ||
| } |
…log to VideoSuperRes - run the post-resize nvvfx pass when either output dimension differs from the target: the previous && condition (inherited from upstream) silently skipped the AI pass when only one dimension differed - perform the RGBA U8 runtime-version detection only for the VideoSuperRes effect, so other nvvfx effects no longer log misleading buffer-format messages
Owner
|
Thank you for the pull request! I don't think retaining backward compatibility is necessary in this case. Therefore, I'm sorry for the extra work, but would you please remove the current codes for SDK 1.0-1.1, and replace |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds a new resize algorithm
nvvfx-videosuperresusing theVideoSuperReseffectintroduced in NVIDIA MAXINE VideoEffects SDK 1.2 (in SDK 1.0/1.1 the same feature
was named
SuperRes).videosuperres-quality=<int>(1 - 19, except 5 - 7, default = 1):created even when no resize is requested, and an explicit
--output-resmatching the input resolution is required. Modes 1 - 4 and 16 - 19 are
upscalers and follow the normal resize rules (intermediate ratio search like
nvvfx-superres).runtimes the filter falls back to the legacy BGR F32 planar buffers and warns.
NVEncC_Options.en.md/.ja.md/.zh-cn.md).Design note
SDK 1.2 renamed SuperResolution to VideoSuperResolution and changed both the
parameter semantics (quality levels replacing mode/strength) and the required
buffer format (RGBA8). I therefore added a separate
nvvfx-videosuperresalgorithm instead of extending
nvvfx-superreswith runtime version detection,so users on SDK 1.0/1.1 are unaffected. If you would rather have a single
nvvfx-superresalgorithm that switches on the SDK version, I'm happy torework it that way.
Testing
--output-res, 16 - 19 upscaling; same-res mode without--output-rescorrectly errors out.
Notes
Related PR: ngx-vsr quality 8 - 19 support (the same SDK 1.2 dll also exposes
these modes through the NGX path)
Runtime / SDK
Install the Video Effects SDK Core,
then run
install_feature.ps1with the feature namenvvfxvideosuperres.Per the SDK docs, the effect's GPU buffers are BGRA/RGBA interleaved 8-bit
(what the filter uses on SDK 1.2+ runtimes), and the suggested minimum input
resolution is 360p.