Assembly good first issues

30 open Assembly issues currently match the default 10+ star filter. Scored difficulty: 5,946 at 1/5, 8,786 at 2/5, 12,699 at 3/5, 12,118 at 4/5, 857 at 5/5. This hub uses the same feed as the homepage, limited to repositories whose GitHub language is Assembly.

All languages · Browse projects · How to find good first issues

Active projects

Unique repos with open help-wanted issues updated in the last 30 days

Showing 30 of ~30 matching filters (Assembly · 10+ stars)

List

[BUG] frontend drops the RAS call-push / return-pop when a call or return in the upper fetch slot follows a not-taken branch in the lower slot

AI summary

A bug in the frontend's prediction logic causes the RAS (Return Address Stack) push/pop operations to be incorrectly cleared when a not-taken branch in a lower fetch slot is followed by a call or return in an upper fetch slot. This leads to mispredictions for return addresses, corrupting the stack's state.

Component:RTLType:BugGood First IssueStatus:In ProgressCV32A65X
3.1K
Difficulty
3/5

UCSD p-System IV.0: floating-point support appears missing from SYSTEM.INTERP

AI summary

The user is encountering an 'Unimplemented instruction' error when running compiled BASIC and FORTRAN programs on a UCSD p-System IV.0 distribution. This error appears to be caused by missing floating-point support modules (specifically FP4.CODE) required by the interpreter (SYSTEM.INTERP). Attempts to manually link the interpreter with a recovered FP4.CODE module failed to integrate it correctly.

help wanted
384
Difficulty
4/5

[Feature]: Implement relaxed affine tensor shape requirements for remaining BatchNorm frontend operations in hipDNN

AI summary

This issue proposes extending relaxed affine tensor shape requirements for BatchNorm operations within hipDNN. Currently, this feature is only implemented for the base BatchNorm operation, and needs to be applied to BatchNormBackwardNode, BatchNormInferenceNode, and BatchNormInferenceNodeVarianceExt to provide full user benefit.

good first issueproject: hipdnn
381
Difficulty
3/5

RFC: Policy on bot/vibe-coded/stochastic/tainted/non-human contributions

AI summary

This is an open Request for Comments (RFC) proposing a policy for AI-generated or AI-assisted contributions across gbdev projects. The policy aims to define acceptable levels of AI involvement in all community interactions, from code to discussions, and to communicate the community's stance on such contributions.

help wantedquestion
34
Difficulty
3/5

[Python bindings] Expose Graph lifecycle and plan APIs

AI summary

This issue proposes exposing a comprehensive set of hipDNN Graph lifecycle, plan management, inspection, and serialization APIs to Python. The goal is to make these functionalities accessible for users to build, inspect, execute, and serialize compiled graph plans, including support for runtime shape overrides. This will involve adding new Python bindings and ensuring compatibility with existing behaviors and error reporting mechanisms.

good first issueproject: hipdnn
381
Difficulty
4/5

[Python bindings] Complete TensorAttributes metadata bindings

AI summary

This issue aims to complete the Python bindings for hipDNN's TensorAttributes, exposing all public tensor metadata and pass-by-value APIs. This includes adding support for scalar tensor construction, various getter/setter methods for metadata like pass-by-value and compile-time constants, and handling ragged offsets and alignment. The goal is to achieve full frontend parity and enable DLPack integration without implementing the DLPack parsing itself.

good first issueproject: hipdnn
381
Difficulty
4/5

[hipDNN] Harden SDPA backward gradient property inference and validation

AI summary

This issue addresses gaps in hipDNN's handling of SDPA backward gradient property inference and validation. Currently, hipDNN infers gradient dimensions correctly but accepts caller-supplied gradient metadata too loosely, potentially allowing incorrect shapes to pass validation. It also lacks proper validation for caller-provided strides, which can lead to backend failures.

good first issueproject: hipdnn
381
Difficulty
3/5

Everything hangs after calling hbios CIOOUT

AI summary

The user is experiencing a system hang after calling a BIOS function (hbios CIOOUT) which leads to a `RLCA` instruction. This hang occurs both during single-stepping and when running from a floppy, with the floppy drive remaining active. The user suspects a bug or a misunderstanding of how `cbios` should be used, as standard CPM output works fine.

help wanted
384
Difficulty
4/5

Use Go 1.23's new structs.HostLayout

AI summary

This issue proposes updating various platform-specific structs in the codebase to utilize Go 1.23's new `structs.HostLayout` type. This change is intended to guarantee that these structs conform to the C ABI, mitigating potential future compatibility issues if Go's internal struct layout were to change.

enhancementgood first issue
3.8K
Difficulty
3/5

[BUG] Out-of-bounds access in debug loop over unpacked arrays

AI summary

A bug has been identified in the CVA6 project where a debug loop iterates beyond the bounds of unpacked arrays. This out-of-bounds access, occurring when the loop index exceeds 10 for arrays with 11 elements (indexed 0-10), can lead to simulator warnings or undefined behavior. The suggested fix is to limit the loop to the actual size of the arrays, preferably using the `$size()` function for better maintainability.

Type:BugGood First IssueStatus:In ProgressComponent:APU
3.1K
Difficulty
1/5

[BUG] frontend/instr_scan incorrectly decodes reserved C.JR and C.EBREAK as valid RVC jump instructions

AI summary

This bug report indicates that the `frontend/instr_scan` module in CVA6 incorrectly identifies reserved or invalid compressed instructions as valid RVC jump instructions. Specifically, `c.jr` with `rs1=x0` and `c.ebreak` are being misclassified, leading to incorrect behavior in the instruction decoding pipeline.

Component:RTLStatus:ResolvedType:BugGood First Issue
3.1K
Difficulty
2/5

Give feedback when read_line input hits the length limit

AI summary

This issue proposes adding user feedback when the input buffer in `read_line` reaches its maximum length. Currently, exceeding the limit results in silent input rejection. The task is to implement a visual or auditory cue (like a PC speaker beep) to inform the user that the buffer is full, while ensuring existing functionality like Enter, Backspace, and history navigation remains unaffected.

enhancementgood first issuekernel32
35
Difficulty
2/5

Add print_dec32 in kernel16/io/print_regs.asm and use it in bios-api/memory/high.asm

AI summary

This issue proposes adding a new 32-bit decimal printing routine, `print_dec32`, to the kernel's assembly code. This is needed because the existing 16-bit routine limits the display of usable memory to 65535 MB, which is inaccurate on systems with more RAM. The new routine will allow for accurate display of larger memory values.

enhancementgood first issuekernel16
35
Difficulty
2/5

Sign GitHub releases

AI summary

This issue proposes implementing signed GitHub releases to improve security and verification. The plan involves researching best practices, setting up necessary tooling, signing the next release, and updating the Debian packaging repository to verify these signatures.

enhancementhelp wanted
449
Difficulty
3/5

Dump Saver graphics, and store signs

AI summary

This issue is a "good first issue" related to graphics and store signs for the Ruby/Sapphire game boards. It specifically points to the `data/board_data/{board}.inc` files as the area for modification.

good first issue
135
Difficulty
1/5

Dump Sapphire Ball Power Up & 'Hole' lights

AI summary

This issue requests the implementation of graphics for the 'Sapphire Ball Power Up' and 'Hole' lights within the game's data files. The developer is directed to use the 'Ruby board' as a reference for how graphics are organized and folded.

good first issue
135
Difficulty
2/5

ci: move pre-commit checkout preparation into a tested Python helper

AI summary

This issue proposes refactoring the Git history and diff calculation logic for pre-commit checks from a shell script into a Python helper. This aims to improve testability, maintainability, and safety by isolating complex Git operations and using robust Python subprocess calls.

good first issuecode qualityci
381
Difficulty
3/5

Improve deb package for Debian distro (to eventually get accepted into distro)

AI summary

This issue aims to get the sjasmplus project officially packaged for Debian and other Debian-based distributions. While a .deb package can be built and passes initial checks, constructive feedback from Debian Developers is needed to address unspecified issues and proceed with finding a mentor and sponsor for the package.

enhancementhelp wanted
449
Difficulty
4/5

[miopen-provider]: Add integration tests to verify Batchnorm rejects unsupported layouts

AI summary

This issue proposes adding integration tests for the MIOpen plugin to ensure it correctly rejects unsupported tensor layouts for Batchnorm operations. The tests will cover scenarios like 3D tensors, mixed layouts, and non-standard stride orders, verifying that the plugin returns an error without crashing and provides informative error messages.

good first issueTestsproject: hipdnnproject: miopen-provider
381
Difficulty
2/5

[miopen-provider]: Add integration tests that verify Fused BN+Activation rejects unsupported modes

AI summary

This issue proposes adding integration tests for the MIOpen provider to ensure it correctly rejects fused batchnorm and activation operations when unsupported activation modes are used. The tests should verify that `graph.build()` returns an error and that specific unsupported modes like SIGMOID, TANH, and LEAKY_RELU are gracefully handled without crashes, providing informative error messages.

good first issueTestsproject: hipdnnproject: miopen-provider
381
Difficulty
2/5

[miopen-provider]: Add Integration tests to verify Batchnorm training rejects insufficient spatial

AI summary

This issue proposes adding integration tests for the MIOpen plugin to ensure it correctly rejects batchnorm training operations when the product of batch and spatial dimensions is less than or equal to 1. The tests should verify that `graph.build()` returns an error and that the error message indicates the spatial dimension requirement, without causing crashes.

good first issueTestsproject: hipdnnproject: miopen-provider
381
Difficulty
2/5

Also browsing: Python , TypeScript , JavaScript