Automated code review of src/subgraph/ by Gemini for potential bugs#9726
Merged
copybara-service[bot] merged 1 commit intomasterfrom Mar 26, 2026
Merged
Automated code review of src/subgraph/ by Gemini for potential bugs#9726copybara-service[bot] merged 1 commit intomasterfrom
copybara-service[bot] merged 1 commit intomasterfrom
Conversation
copybara-service
Bot
force-pushed
the
test_885703148
branch
4 times, most recently
from
81c9b68 to
34bb300
Compare
copybara-service
Bot
force-pushed
the
test_885703148
branch
from
34bb300 to
726fd3b
Compare
4 tasks
mohammadmseet-hue
added a commit
to mohammadmseet-hue/XNNPACK
that referenced
this pull request
Mar 26, 2026
Per review feedback: in the optimizer path (optimize_common_subgraphs_static_reshapes), when combined dims exceed XNN_MAX_TENSOR_DIMS, skip the optimization and return xnn_status_success rather than xnn_status_invalid_parameter. The runtime reshape functions in copy.c already have the authoritative bounds checks — the optimizer should gracefully bail out, consistent with the pattern in google#9726.
copybara-service Bot
pushed a commit
that referenced
this pull request
Mar 26, 2026
-- 2101890 by mohammadmseet-hue <mohammadmseet@gmail.com>: Fix missing bounds checks and error handling in tensor APIs 1. Add num_dims > XNN_MAX_TENSOR_DIMS check to xnn_reshape_external_value() in runtime.c. This was the only public API entry point that did not validate num_dims, unlike all xnn_define_*() functions in tensor.c and all operator reshape functions. 2. Add missing return statement after error log in xnn_define_blockwise_quantized_tensor_value_v2() when block_size == 0. Without this, execution falls through to a division by block_size, causing a division-by-zero crash (SIGFPE). 3. Add bounds checks in subgraph optimization passes that merge static_reshape and static_expand_dims nodes. The combined num_dims could exceed XNN_MAX_TENSOR_DIMS, causing out-of-bounds writes to stack-allocated xnn_shape.dim[6] arrays. -- fab78ca by mohammadmseet-hue <mohammadmseet@gmail.com>: Fix subgraph optimizer bounds check to return success instead of error Per review feedback: in the optimizer path (optimize_common_subgraphs_static_reshapes), when combined dims exceed XNN_MAX_TENSOR_DIMS, skip the optimization and return xnn_status_success rather than xnn_status_invalid_parameter. The runtime reshape functions in copy.c already have the authoritative bounds checks — the optimizer should gracefully bail out, consistent with the pattern in #9726. FUTURE_COPYBARA_INTEGRATE_REVIEW=#9778 from mohammadmseet-hue:fix/missing-bounds-checks fab78ca PiperOrigin-RevId: 889687406
copybara-service
Bot
force-pushed
the
test_885703148
branch
from
726fd3b to
dc82713
Compare
The fixes are from a few basic categories: - Lack of validation of input tensor rank before using the shape. This is tricky because the tensor shape and rank are not expected to be defined, but we sometimes use these shapes anyways before a reshaping happens. - We mix up input `input1` and `input2` in a lot of places, in ways that expose invalid memory accesses. - We mix up `filter_id` and `bias_id` often. Sometimes only for logging/error reporting, but sometimes not... - Sometimes we check something is valid, but not before using it first. - Ignoring the return value of validation functions. It makes one questionable change: when a loop has an unsigned extent, it wants to use an unsigned index. This is a questionable choice, there are arguments both ways on this one. PiperOrigin-RevId: 890058240
copybara-service
Bot
force-pushed
the
test_885703148
branch
from
dc82713 to
d45f452
Compare
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
1 participant
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.
Automated code review of src/subgraph/ by Gemini for potential bugs
The fixes are from a few basic categories:
input1andinput2in a lot of places, in ways that expose invalid memory accesses.filter_idandbias_idoften. Sometimes only for logging/error reporting, but sometimes not...It makes one questionable change: when a loop has an unsigned extent, it wants to use an unsigned index. This is a questionable choice, there are arguments both ways on this one.