Repository navigation
Sync S11: merge upstream PDFium up to 7c55203ed (Oct 15 2026) - #78
Merged
Merged
Conversation
Make it more obvious the code is instantiating a null FooPtr. Change-Id: Ief54445d91ecc0b90a65009429289455ff1c5bab Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160490 Reviewed-by: Tom Sepez <tsepez@chromium.org> Commit-Queue: Lei Zhang <thestig@chromium.org>
Avoid manually reimplementing it. Change-Id: I3a5510221a3ee1575e994aacec42d75f7cc6b221 Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160491 Reviewed-by: Tom Sepez <tsepez@chromium.org> Commit-Queue: Lei Zhang <thestig@chromium.org>
Adding arbitrary string data to the ByteStringPool from CPDF_String is unlikely to generate much data sharing. Instead, only use ByteStringPool for CPDF_Name, where values like "Type" and "Page" are often repeated. Change-Id: I261b16b93a9a180c86e6c78b1dfd92a5d7bd1097 Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160492 Reviewed-by: Tom Sepez <tsepez@chromium.org> Commit-Queue: Lei Zhang <thestig@chromium.org>
Per recommendations in Chromium's base/containers/README.md, switch from std::unordered_set to absl::flat_hash_set. Change-Id: I0e4dfeea9fe888b0ea2765afa42d165bc69e2e4f Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160493 Commit-Queue: Lei Zhang <thestig@chromium.org> Reviewed-by: Tom Sepez <tsepez@chromium.org>
When PDF_USE_LIBYUV is enabled, accelerate lossless pixel format conversions, alpha plane extraction/injection, horizontal mirroring, and 90/270-degree rotation / transposition in CFX_DIBBase and CFX_DIBitmap: - ConvertBuffer_RgbToArgbPremul, CFX_DIBitmap::ConvertFormat (kBgr -> kBgra), and 24-bpp CFX_DIBBase::RealizeSkImage: use libyuv::RGB24ToARGB. - ConvertBuffer_32bppRgbToRgb24: use libyuv::ARGBToRGB24. - CFX_DIBBase::CloneAlphaMask: use libyuv::ARGBExtractAlpha. - CFX_DIBitmap::MultiplyAlphaMask (kBgrx): use libyuv::ARGBCopyYToAlpha. - CFX_DIBBase::FlipImage: use libyuv::MirrorPlane, libyuv::RGB24Mirror, and libyuv::ARGBMirror. - CFX_DIBBase::SwapXY: use libyuv::RotatePlane90, libyuv::RotatePlane270, libyuv::TransposePlane, and libyuv::ARGBRotate for 8-bpp and 32-bpp bitmaps via RealizeIfNeeded(). When the source bitmap has a contiguous buffer, invoke the full 2D libyuv plane functions directly instead of row-by-row. Speedup vs Before (1275x1650, US Letter 150 DPI): | Function | Zen 5 | GNR | Grace | Wasm | RVV | |------------------|--------|--------|--------|--------|--------| | RGB24ToARGB | 6.08x | 2.53x | 5.65x | 5.54x | 31.04x | | ARGBToRGB24 | 5.88x | 1.98x | 2.35x | 1.55x | 28.07x | | ARGBExtractAlpha | 5.95x | 1.90x | 4.13x | 4.33x | 17.05x | | ARGBCopyYToAlpha | 7.61x | 3.02x | 4.71x | 3.28x | 7.22x | | MirrorPlane | 9.59x | 5.28x | 5.84x | 6.04x | 1.33x | | RGB24Mirror | 8.32x | 3.02x | 5.40x | 4.53x | 1.00x | | ARGBMirror | 4.17x | 1.33x | 2.57x | 3.20x | 21.06x | | RotatePlane90 | 10.86x | 6.60x | 7.23x | 8.50x | 1.84x | | RotatePlane270 | 10.75x | 6.59x | 6.97x | 9.31x | 1.84x | | TransposePlane | 10.92x | 6.53x | 7.13x | 8.64x | 1.84x | | ARGBRotate | 5.40x | 3.01x | 1.65x | 2.74x | 13.24x | (Zen 5: AMD EPYC Turin AVX-512; GNR: Intel Granite Rapids AVX-512; Grace: NVIDIA Grace Neoverse V2 SVE2; Wasm: V8 SIMD128; RVV: Coral NPU) Bug: 42271060, 549230311 Change-Id: I223e13be29cb4546c7d0e53c9dda340dc4375944 Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160093 Reviewed-by: Lei Zhang <thestig@chromium.org> Commit-Queue: Lei Zhang <thestig@chromium.org>
Introduce EmbedderTest::CompareDisplayList() and helper functions to compare rendered display lists against expectation files. Add VerifySavedRenderingWithDisplayList() and VerifySavedDocumentWithDisplayList() to support testing saved document rendering with display lists. Migrate FPDFViewEmbedderTest.RenderPageToDisplayList to use CompareDisplayList() with rectangles_expected.dump.txt instead of an inline string literal. Change-Id: Ie055ae896d535ff87d1894ddc871b3b81eba87b2 Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160550 Commit-Queue: Tom Sepez <tsepez@chromium.org> Reviewed-by: Lei Zhang <thestig@chromium.org>
Add support for an optional 4th parameter to the fuzzy action in SUPPRESSIONS, allowing tests to specify an SSIM threshold in [0.0, 1.0]. Update Suppressor._ParseFuzzyAction to accept up to 4 parameters and validate that the SSIM threshold does not exceed 1.0. Update PNGDiffer._ParseTolerances to extract the SSIM threshold, defaulting to 1.0. In PNGDiffer.EvaluateMatch, accept comparisons if either the existing MSE criteria pass or the mean SSIM strictly exceeds the configured threshold. Defaulting to 1.0 preserves existing behavior since mean SSIM cannot exceed 1.0. Change-Id: Icf84478c7c825c4c49c01a3c21168f991ea8879d Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160651 Reviewed-by: Lei Zhang <thestig@chromium.org> Commit-Queue: Tom Sepez <tsepez@chromium.org>
Add unit test coverage for CFX_DIBBase::TransformTo() in cfx_dibbase_unittest.cpp, testing empty output bounds, 8-bit mask rotation/scaling, and BGRA scaling/translation. TAG=agy CONV=913b2a22-b8f8-4d2f-a2ad-8d7cc152a129 Change-Id: I51272780d37aae5361e5439cda7de2bd996e6b7f Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160611 Reviewed-by: Tom Sepez <tsepez@chromium.org> Commit-Queue: Lei Zhang <thestig@chromium.org>
Add unit test coverage for CFX_DIBBase::StretchTo() in cfx_dibbase_unittest.cpp, testing empty output bounds, 8-bit mask stretching with horizontal flipping and clipping, 8-bit grayscale stretching with vertical flipping, and BGRA stretching. TAG=agy CONV=913b2a22-b8f8-4d2f-a2ad-8d7cc152a129 Change-Id: Ie252bbe08915d93d136dd154b6560893287b2557 Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160630 Reviewed-by: Tom Sepez <tsepez@chromium.org> Commit-Queue: Lei Zhang <thestig@chromium.org>
For instance, pdf_enable_fontations and pdf_enable_freetype blocks now live closer to each other. Change-Id: Id85bfeeed07033916ee5de8c7eabd3733bb1bc8d Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160650 Commit-Queue: Lei Zhang <thestig@chromium.org> Reviewed-by: Tom Sepez <tsepez@chromium.org>
Combine columns 2 (V8 support) and 3 (XFA support) into a single engine column (column 2) accepting nov8, v8, and xfa. Because XFA requires V8, the combination nov8 with XFA is impossible, and nov8 without XFA is redundant with nov8. Combining these into a single 3-state engine column simplifies the suppression table from 7 columns to 6 columns. Add convert_suppressions.py to perform the transformation on SUPPRESSIONS. Update Suppressor._VALID_COLUMN_VALUES, _SuppressionSpecificity, and _MatchSuppression to match against the unified engine column. Change-Id: Ib4a892f2a56a9dc26e5a860bb67410a0051a093b Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160670 Reviewed-by: Lei Zhang <thestig@chromium.org> Commit-Queue: Tom Sepez <tsepez@chromium.org>
Add some TODOs as a first step towards reverting a behavior change pdfium-review.googlesource.com/c/80912 made. When a path only forms 3 sides of a rectangle and it has not been closed, that should not count as a rectangle. Bug: 42270691, 572902676 Change-Id: I5cac1867a77b7dbb351d50f3d1df5fee78e1cac6 Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160750 Commit-Queue: Lei Zhang <thestig@chromium.org> Reviewed-by: Andy Phan <andyphan@chromium.org>
Revert a behavior change pdfium-review.googlesource.com/c/80912 made, since a path that only forms 3 sides of a rectangle and has not been closed is not a rectangle. Bug: 42270691, 572902676 Change-Id: I31da35b950e281f0815bfb17b40befd14f21b5cb Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160673 Commit-Queue: Lei Zhang <thestig@chromium.org> Reviewed-by: Andy Phan <andyphan@chromium.org>
Convert ByteStringPool into an interface with ByteStringPool::Create() and a pure virtual Intern() method, and move the absl::flat_hash_set implementation into ByteStringPoolImpl in bytestring_pool.cpp. This avoids exposing absl/container/flat_hash_set.h in bytestring_pool.h. Since bytestring_pool.h gets transitively included in many translation units, explosing flat_hash_set.h increases the Chromium compile size by ~450 MB, as seen on crrev.com/c/8563220. TAG=agy CONV=9d52233e-f437-4c0d-acb6-07f402d6540c Change-Id: Iaecae19ee53ce97e57e793f4962c2e8865b1a295 Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160690 Reviewed-by: Tom Sepez <tsepez@chromium.org> Commit-Queue: Lei Zhang <thestig@chromium.org>
CFX_Path::IsRect(), GetRect(), and GetRectOppositeCorner() recognize both 4-point and 5-point rectangles, but PathPointsNeedNormalization() and GetNormalizedPoints() currently only normalize paths that have more than 5 points and repeat the first point at the end. As a result, 4-point rectangles with duplicate or collinear edge points (such as paths closed with CFX_Path::ClosePath() instead of repeating the start point) are not recognized as rectangles. Add/update tests in cpdf_pagecontentgenerator_unittest.cpp and cfx_path_unittest.cpp to document the current behavior, along with TODOs to fix this. In particular, add one more scenario in CFXPath.FivePointRect, and remove the identical scenario from CFXPath.NotRect. TAG=agy CONV=e2fb55f5-ba6e-4820-9194-44fb5bf14b16 Bug: 572902676 Change-Id: Ife1104f4417a9f9e9a612f856e31848806cfa7d7 Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160710 Commit-Queue: Lei Zhang <thestig@chromium.org> Reviewed-by: Andy Phan <andyphan@chromium.org>
CFX_Path::IsRect(), GetRect(), and GetRectOppositeCorner() recognize both 4-point rectangles (where the start point is not repeated at the end) and 5-point rectangles (where the start point is repeated at the end). However, PathPointsNeedNormalization() and GetNormalizedPoints() only normalized paths with more than 5 points that repeated the start point at the end. As a result, 4-point rectangles with duplicate or collinear edge points, such as paths closed via CFX_Path::ClosePath() / FPDFPath_Close(), were not recognized as rectangles in memory, even though saving and reloading the PDF via CPDF_StreamContentParser::Handle_ClosePath() appended the start point and turned them into rectangles. Update PathPointsNeedNormalization() to also normalize 5-point paths where the first and last points differ, and update GetNormalizedPoints() to normalize paths with differing first and last points to 4 points while preserving `close_figure_` when deleting a final closing duplicate or collinear point. TAG=agy CONV=e2fb55f5-ba6e-4820-9194-44fb5bf14b16 Bug: 572902676 Change-Id: Ie01ae75c1441666bd488104716f8a08804361684 Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160711 Reviewed-by: Andy Phan <andyphan@chromium.org> Commit-Queue: Lei Zhang <thestig@chromium.org>
In CPDF_ClipPath::AppendPathWithAutoMerge() and CPDF_ContentParser::CheckClip(), use GetRectOppositeCorner() and avoid assuming the point at index 2 is always the opposite corner. After this, CPDF_Path::IsRect() is dead code, so delete it. Change-Id: Id2ad3fc4e7d2e7b782bedad56946f91f522608fd Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160770 Reviewed-by: Andy Phan <andyphan@chromium.org> Commit-Queue: Lei Zhang <thestig@chromium.org>
Prevent a (somewhat) misbehaving embedder from tearing down the runtime while a callback is still in-flight. Bug: 572237007 Change-Id: I003a6b67703305c465a2ed0845f4274df93ed5af Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160614 Commit-Queue: Lei Zhang <thestig@chromium.org> Reviewed-by: Lei Zhang <thestig@chromium.org>
Commit 21e954b changed CFX_CSSSyntaxParser::GetCurrentString() to return a string view instead of taking an out-parameter length, leaving `iLen2` in CFX_CSSStyleSelector::AppendInlineStyle() and `iValueLen` in CFX_CSSStyleSheet::LoadStyleRule() permanently zero. Replace the stale length checks with !wsName.IsEmpty() so custom CSS properties reach CFX_CSSDeclaration::AddProperty(). Update CFX_CSSDeclaration::empty() to check both properties_ and custom_properties_ so style rules containing only custom properties are not discarded. Add unit tests covering custom CSS properties in both inline styles and stylesheet rules. Bug: 573263143 Change-Id: Ib9cddf49d6687a34ab2c4195d2dc70eaf74e2aa3 Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160850 Reviewed-by: Lei Zhang <thestig@chromium.org> Commit-Queue: Lei Zhang <thestig@chromium.org>
This keeps platform-specific device text eligibility checks inside the concrete device drivers rather than in CFX_RenderDevice, removes a redundant ShouldDrawDeviceText() call from the Windows printer path, and ensures non-rasterizing drivers will still receive DrawDeviceText() calls on macOS. Instead, since it is interrogating a font, make it an Apple-only method of CFX_Font. Bug: 573199238 Change-Id: I0a107edd4a47c7f22a3e9ffea7becd3e00f05493 Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160830 Reviewed-by: Lei Zhang <thestig@chromium.org> Commit-Queue: Lei Zhang <thestig@chromium.org>
Add DestroyPlatformFont() to pair with CreatePlatformFont() on CFX_GEModule::PlatformIface under IS_APPLE. This allows CFX_Font::ReleasePlatformResource() to move from fx_apple_impl.cpp into cfx_font.cpp alongside the rest of CFX_Font. Change-Id: I01e730dcf380e36b80cf356b2e9dc530641e8bd5 Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160652 Commit-Queue: Lei Zhang <thestig@chromium.org> Reviewed-by: Lei Zhang <thestig@chromium.org>
Replace fragile golden PNG image comparisons in FPDFEditEmbedderTest text removal tests with display list text inspections. The display list approach inspects the emitted text directly, avoiding rasterization, hinting, and font engine discrepancies between FreeType and Fontations/ Skia. Delete 22 obsolete golden PNG expectation files and remove the now unused kHelloWorldRemovedPng constant. Change-Id: I1f296f748e7ea025e3a5c63d51cbd2a7e9e85bf4 Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160050 Reviewed-by: Lei Zhang <thestig@chromium.org> Commit-Queue: Lei Zhang <thestig@chromium.org>
Combine std::optional::has_value() and std::optional::value() checks on CFX_Path::GetRect() into single EXPECT_EQ() calls, and inline the rest of the `rect` checks into EXPECT_FALSE(path.GetRect(...).has_value()). TAG=agy CONV=6567e49a-585d-49fa-a2a2-cea8e1071f98 Change-Id: I774f0286072a80560b29ef2e17f7f90793b8029a Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160870 Reviewed-by: Andy Phan <andyphan@chromium.org> Commit-Queue: Lei Zhang <thestig@chromium.org>
Refactor GetNormalizedPoints() so the case where there are too many points can return slightly earlier by checking a vector size before appending to the vector. As such, the same vector can reserve one fewer element. Change-Id: I79f787c85e18bb0558042f340a8b7e48f788e086 Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160890 Commit-Queue: Lei Zhang <thestig@chromium.org> Reviewed-by: Andy Phan <andyphan@chromium.org>
This indicates a misbehaving embedder, so flag it with a CHECK() while it is still active on the stack. -- mass rename two hungarian length variables. -- consistently name AutoRestorer local `restorer` Bug: 572237007 Change-Id: Ice9212769e4072d76a427379aa13d02fa36be55d Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160615 Reviewed-by: Lei Zhang <thestig@chromium.org> Commit-Queue: Tom Sepez <tsepez@chromium.org>
Provide a new value at construction time. Then use this to simplify CPDFSDK_FormFillEnvironment. Change-Id: Iadc1af0ae0e22f441380565219a814b7b56b080f Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160910 Commit-Queue: Tom Sepez <tsepez@chromium.org> Reviewed-by: Lei Zhang <thestig@chromium.org>
Add third_party/.clang-format-ignore so clang-format does not format 3P code. It is copied from Chromium's file at the same relative location. Change-Id: I4903da2517c409a10859d8a039642612f07cce9c Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160930 Reviewed-by: Andy Phan <andyphan@chromium.org> Commit-Queue: Lei Zhang <thestig@chromium.org>
Simplify the cases where assigning a bool, enum, or integer literal. Change-Id: I3fc7104a819f232e668aa92704c145a5e9dc6fd9 Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160911 Reviewed-by: Lei Zhang <thestig@chromium.org> Commit-Queue: Tom Sepez <tsepez@chromium.org>
This is possible now that the out-of-line constructor requirement has been relaxed. Use designated initializers where possible. -- convert members to google_style. -- remove redundant assignment of members to default values. Change-Id: If5144766a3c5c5b7809328d9263413364251af52 Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160950 Reviewed-by: Lei Zhang <thestig@chromium.org> Commit-Queue: Tom Sepez <tsepez@chromium.org>
Structs default to public access, so the explicit public: access specifier is redundant and can be safely removed to clean up the codebase. -- Make aggregates where possible. -- Make struct inheriting from class into a class. Change-Id: I8f4762dbdbc380061db4d92cb1393c57483568ba Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160970 Reviewed-by: Lei Zhang <thestig@chromium.org> Commit-Queue: Tom Sepez <tsepez@chromium.org>
Replace fragile golden PNG image comparisons in FPDFSaveWithFontSubsetEmbedderTest with display list codepoint checks. The display list approach verifies the text emitted by the font subsetting and save operations directly, avoiding rasterization, hinting, and font engine discrepancies between FreeType and Fontations/Skia. Delete 12 obsolete golden PNG expectation files. Change-Id: I63bab25d46cf02f6b9e40686ff63c5dea9b53448 Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160070 Reviewed-by: Lei Zhang <thestig@chromium.org> Commit-Queue: Tom Sepez <tsepez@chromium.org>
Replace fragile golden PNG image comparisons in FPDFEditEmbedderTest for text addition, font loading, and glyph positioning tests with display list inspections. Support UTF-8 multi-byte decoding in ExtractTextLinesFromDisplayList to verify non-ASCII and CID font text directly. Delete 46 obsolete golden PNG expectation files. Change-Id: Ib10a4210ee33aeccd315d1ca0f0b54187925efaa Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160350 Reviewed-by: Lei Zhang <thestig@chromium.org> Commit-Queue: Tom Sepez <tsepez@chromium.org>
- Remove CPDF_SyntaxParser::CreateForTesting() and inline the implementation into the only caller. - Rename variable names in remaining ctors to use Google C++ style. - Add inline comment to help document what "0" is. Change-Id: I8f584ec4295bf8e0d9bb1cf1994423f6750a9f8a Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160991 Commit-Queue: Lei Zhang <thestig@chromium.org> Reviewed-by: Tom Sepez <tsepez@chromium.org>
Move struct LayoutTestData out of the anonymous namespace inside cpvt_section_unittest.cpp. The existing code causes GCC to fail with -Werror=subobject-linkage. Change-Id: If224f666c2a702d21a13d31d51e8c0e05a926d71 Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160971 Reviewed-by: Andy Phan <andyphan@chromium.org> Commit-Queue: Lei Zhang <thestig@chromium.org>
Previously, form_callbacks kept per-page data alive after a page was processed. Now, --render-repeats measures image decode time in each repeat. (When using --pages=1..10, form_fill_info_->loaded_pages keeps per-page image caches alive for all pages still, so decoding many pages needs memory for the images of all pages still. See an earlier patch set on the CL for a different approach that changes that too.) Follow-up to https://pdfium-review.googlesource.com/c/pdfium/+/132470 Bug: none Change-Id: Ief69b738ce21f9e82c0f9a95a63f4ec201a9cfc8 Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160790 Reviewed-by: Nico Weber <thakis@google.com> Reviewed-by: Lei Zhang <thestig@chromium.org> Commit-Queue: Nico Weber <thakis@chromium.org>
ContinueStretchHorz() resamples one destination column at a time. Its modal tap counts are 1 (no scaling in this axis) and 2 (bilinear interpolation), so a bilinear row is a long run of columns doing the same two multiply-accumulates on neighbouring source pixels, and that pass is the largest single cost of rendering image-heavy pages. Four consecutive columns sharing a tap count, whose source pixels lie within 16 or 32 bytes of each other, are now resampled together: one vectorized gather per tap pulls all four columns' source bytes into 32-bit lanes, one multiply-accumulate per tap does the arithmetic for all four, and the result is packed and stored as one vector. The arithmetic is the per-column path's, in tap order, on the same bytes with the same weights, so the output is identical bit for bit. PlanHorzGroups() runs once per stretch, after the weight table: it finds maximal spans of same-tap-count columns, and for each whole group of four within a span whose combined source window fits the gather it precomputes the gather indices and weights (StretchHorzGroup). A span is planned only if at least three in four of its groups fit; consecutive fitting groups form a run (StretchHorzRun), which is the unit the kernel handles per call. Columns not covered -- an odd tap count, a short span, a group whose window does not fit -- take the per-column path unchanged. The kernels live in cstretchengine_hwy.cpp and use Highway. Three arms are covered: - k8BppTo8Bpp and kManyBpptoManyBpp (BGR and BGRx destinations): StretchHorzRunSimd(), overloaded on the destination pixel type. 1bpp sources take the k8BppTo8Bpp arm too, on the row Expand1BppRow() produces, so the plan for them is made against that expanded row. - kManyBpptoManyBppWithAlpha: StretchHorzRunAlphaSimd(), where each tap's weight is weight * src.alpha / 255 -- a per-lane truncating divide that depends on the gathered alpha, so it is computed in the kernel with the exact reciprocal-multiply trick for division by 255 (the (x + 1 + (x >> 8)) >> 8 approximation would be wrong here: weight * alpha reaches 65536 * 255, past where it is exact). The kernels take the source row and destination as spans and process a whole run per call, so the runtime target selection and the call happen once per run (hundreds of groups for an unclipped bilinear row) rather than once per four columns. Stores are packed vectors: 4 bytes for 8bpp, 16 bytes for BGRA and for BGRx (whose 4th bytes are read back from the destination and kept, as the per-column path leaves them untouched), and 8 + 4 bytes for BGR so that nothing is written past the 12 bytes a group owns. The new unit tests compare every kernel, over a small fixed set of groups and windows chosen to cover known edges (gather positions at the start/middle/end of the window, weights at both ends and the midpoint of the fixed-point range) for each (taps, window width) variant, against a scalar re-derivation of what a group is documented to compute, including the bytes a group must not touch, and check that a run is processed in full. Bug: 559426535 Change-Id: I4f5e3e977b4f6f9b184988ae518cf70abd868a20 Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/158070 Reviewed-by: Tom Sepez <tsepez@chromium.org> Commit-Queue: Lei Zhang <thestig@chromium.org> Reviewed-by: Lei Zhang <thestig@chromium.org>
Replace bitmap comparisons in FPDFFormFillEmbedderTest with DisplayListRenderer assertions. This verifies rendered form text and rectangle highlights directly against display list output instead of platform-specific rasterized bitmaps, eliminating 42 obsolete golden PNG files. Change-Id: I47a8f536bf267559b3a4f4cf5506bb90fc2b1b07 Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/160351 Commit-Queue: Tom Sepez <tsepez@chromium.org> Reviewed-by: Lei Zhang <thestig@chromium.org>
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
6 participants
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.
No description provided.