feat(qualcomm): Qnn AOT Lowering pass #580

oreomaker · 2026-01-05T12:59:26Z

Please check Guidelines for Contributing.

Summary by CodeRabbit

Release Notes

New Features
- Added support for CastType, Embedding, Index, and View operations in QNN backend.
- Introduced Mul operation support alongside existing Add operation.
- Added Sigmoid neural network layer.
Refactor
- Simplified tensor creation logic in QNN backend by removing caching and registration overhead.
- Restructured pattern registration for improved modularity and extensibility.
- Unified operation pattern handling interface.

_{✏️ Tip: You can customize this high-level summary in your review settings.}

…iew operations - Add new QnnAOTCastTypePattern to handle CastType operations with quantization support for float/int conversions and QNN Quantize/Dequantize/Cast operations - Add new QnnAOTEmbeddingPattern to convert embedding operations to QNN Gather - Add new QnnAOTIndexPattern to handle index operations using QNN Gather with proper axis detection - Add new QnnAOTViewPattern to convert view operations to QNN Reshape - Refactor LLM2QnnLoweringPass to use template-based pattern registration instead of manual insertion - Update Elewise visitor to add QnnAOTMulPattern for multiplication operations - Modify base pattern to use rewrite method instead of compile method

coderabbitai · 2026-01-05T12:59:37Z

📝 Walkthrough

Walkthrough

The PR adds five new QNN AOT visitor patterns (CastType, Embedding, Index, View, Mul), introduces a Sigmoid neural network layer, and refactors the pattern registration system to support bulk initialization. It also removes tensor caching and runtime registration side effects from QNN tensor creation and changes the base pattern class from compile-based to rewrite-based API.

Changes

Cohort / File(s)	Summary
QNN AOT Tensor Creation `mllm/backends/qnn/aot/QnnWrappersAPI.cpp`	Simplified tensor creation by removing caching (static_tensor_, all_tensors_) and QNN runtime registration calls (tensorCreateContextTensor, tensorCreateGraphTensor). Tensors are now only constructed and returned without side effects.
QNN AOT Pattern Base Refactoring `mllm/backends/qnn/aot/visitor/Base.hpp`	Removed pure virtual `compile` method from QnnAOTBasePattern and QnnAOTQuantRecipeBasePattern. Replaced with no-op `rewrite` methods returning false, decoupling pattern rewriting from compilation logic.
QNN AOT Lowering Infrastructure `mllm/backends/qnn/aot/passes/LLM2QnnLoweringPass.cpp`, `mllm/backends/qnn/aot/passes/LLM2QnnLoweringPass.hpp`	Introduced variadic template helper `registerPatterns()` to bulk-register multiple pattern types. Migrated from single QnnAOTAddPattern to consolidated registration of Embedding, CastType, Add, View, and Index patterns.
QNN AOT CastType Pattern `mllm/backends/qnn/aot/visitor/CastType.hpp`, `mllm/backends/qnn/aot/visitor/CastType.cpp`	New pattern to convert linalg::CastTypeOp into QNN operations (Quantize, Dequantize, Cast). Validates qnn_graph_name/qnn_context_name attributes and registers the QNN node.
QNN AOT Embedding Pattern `mllm/backends/qnn/aot/visitor/Embedding.hpp`, `mllm/backends/qnn/aot/visitor/Embedding.cpp`	New pattern to convert linalg::EmbeddingOp into QNN Gather operations. Extracts weight and indices tensors, creates Gather node with axis parameter, and registers in AOT context.
QNN AOT Index Pattern `mllm/backends/qnn/aot/visitor/Index.hpp`, `mllm/backends/qnn/aot/visitor/Index.cpp`	New pattern to convert linalg::IndexOp into QNN Gather operations. Infers axis from indices, constructs indices tensor, and registers operation with axis parameter.
QNN AOT View Pattern `mllm/backends/qnn/aot/visitor/View.hpp`, `mllm/backends/qnn/aot/visitor/View.cpp`	New pattern to convert linalg::ViewOp into QNN Reshape operations. Computes output shape, builds shape tensor, and registers Reshape node.
QNN AOT Elewise Patterns `mllm/backends/qnn/aot/visitor/Elewise.hpp`, `mllm/backends/qnn/aot/visitor/Elewise.cpp`	Renamed QnnAOTAddPattern method from `compile` to `rewrite`. Added new QnnAOTMulPattern with identical structure for ElementWiseMultiply operations.
Sigmoid Layer `mllm/nn/layers/Sigmoid.hpp`, `mllm/nn/layers/Sigmoid.cpp`, `mllm/nn/Nn.hpp`	New Sigmoid layer implementation with default and options-based constructors. Exported via Nn.hpp public interface. Includes in-place operation support via macros.
Sigmoid Op Registration `mllm/compile/ir/linalg/Op.hpp`, `mllm/compile/ir/linalg/Op.cpp`	Moved SigmoidOp registration from CustomizedOps block to main LINALG operator definitions block for consistency.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Possibly related PRs

feat(Qnn AOT): AOT and AOT Runtime. Qwen3 AOT Mode. #567: Modifies QnnWrappersAPI.cpp tensor/context creation and AOT scaffolding, overlapping tensor registration logic changes.
feat(Qnn AOT): Implement LLMQuantRecipePass and associated patterns for quantization #572: Touches QNN AOT backend infrastructure (QnnWrappersAPI.cpp, visitor/Base.hpp, pattern implementations) with similar registration/caching refactoring.
feat: add LLM2QnnLoweringPass and update graph splitting logic #577: Modifies QNN AOT lowering pipeline and QnnWrappersAPI tensor/graph creation paths, affecting overlapping code areas.

Suggested reviewers

liang1232018
chenghuaWang
yirongjie

Poem

🐰 Five patterns hop into the QNN,
Embedding, Index, View—quite the keen bunch!
Sigmoid learns to curve with graceful flair,
While tensors dance free, unregistered in air. ✨

Pre-merge checks and finishing touches

❌ Failed checks (2 warnings, 1 inconclusive)

Check name	Status	Explanation	Resolution
Description check	⚠️ Warning	The PR description only contains the template boilerplate with no actual content describing the changes, objectives, or rationale for the modifications.	Replace the template text with a substantive description of the changes, including what patterns/features were added, why they were added, and any relevant implementation details or testing information.
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	You can run `@coderabbitai generate docstrings` to improve docstring coverage.
Title check	❓ Inconclusive	The title 'Qnn aot' is too vague and generic, not clearly describing the primary changes or scope of the pull request.	Provide a more descriptive title that captures the main changes, such as 'Add QNN AOT patterns and SigmoidOp support' or 'Refactor QNN AOT lowering with multiple new patterns'.

✨ Finishing touches

📝 Generate docstrings

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

coderabbitai

Actionable comments posted: 3

Fix all issues with AI Agents 🤖

In @mllm/backends/qnn/aot/passes/LLM2QnnLoweringPass.cpp:
- Line 24: registerPatterns is missing QnnAOTMulPattern registration even though
QnnAOTMulPattern is implemented (in Elewise.hpp) like QnnAOTAddPattern; update
the registerPatterns template instantiation to include QnnAOTMulPattern
alongside QnnAOTAddPattern (so the list becomes QnnAOTEmbeddingPattern,
QnnAOTCastTypePattern, QnnAOTAddPattern, QnnAOTMulPattern, QnnAOTViewPattern,
QnnAOTIndexPattern) to ensure multiplication ops are lowered by
LLM2QnnLoweringPass.

In @mllm/backends/qnn/aot/visitor/Embedding.cpp:
- Around line 34-35: The lookup of the weight symbol via
writer.getContext()->lookupSymbolTable(a_op->getName() + ".weight") can return
nullptr; validate the returned pointer before calling outputs().front() (the
code that assigns to variable weight). If lookup returns nullptr, handle it
(e.g., log or raise an informative error including a_op->getName(), or
return/throw) to avoid null dereference; only call
outputs().front()->cast_<ir::tensor::TensorValue>() when the symbol table
pointer is non-null. Ensure the error path provides clear context about the
missing "<opName>.weight" symbol.

In @mllm/backends/qnn/aot/visitor/View.cpp:
- Around line 38-55: The computed but unused shape variables should be removed:
delete the local variables `shape_data` and `shape_tensor_shape` and the loop
that fills `shape_data` (the for-loop over `output_shape`) in View.cpp so only
the output shape query remains; keep the QNN Reshape op creation
(`QnnAOTNodeOperation::create("Reshape")`), package name set, inputs/outputs via
`env->captureQnnAOTNodeTensor(...)`, `setName(view_op->getAOp()->getName())`,
and the final `env->captureAOTNodeOp(...)` call unchanged.

🧹 Nitpick comments (4)

mllm/backends/qnn/aot/visitor/CastType.cpp (1)
15-25: Consider using a switch statement or lookup for maintainability.

The isInt function has a long chain of OR conditions covering all integer and quantized types. While correct, a more maintainable approach could use a switch statement or a set-based lookup.
🔎 Alternative implementation
 static bool isFloat(DataTypes dtype) { return dtype == kFloat32 || dtype == kFloat16; }
 
 static bool isInt(DataTypes dtype) {
-  return dtype == kInt8 || dtype == kInt16 || dtype == kInt32 || dtype == kUInt8 || dtype == kUInt16 || dtype == kUInt32
-         || dtype == kInt8PerTensorSym || dtype == kInt8PerChannelSym || dtype == kUInt8PerTensorSym
-         || dtype == kUInt8PerChannelSym || dtype == kInt16PerTensorSym || dtype == kInt16PerChannelSym
-         || dtype == kUInt16PerTensorSym || dtype == kUInt16PerChannelSym || dtype == kInt8PerTensorAsy
-         || dtype == kInt8PerChannelAsy || dtype == kUInt8PerTensorAsy || dtype == kUInt8PerChannelAsy
-         || dtype == kInt16PerTensorAsy || dtype == kInt16PerChannelAsy || dtype == kUInt16PerTensorAsy
-         || dtype == kUInt16PerChannelAsy;
+  switch (dtype) {
+    case kInt8: case kInt16: case kInt32:
+    case kUInt8: case kUInt16: case kUInt32:
+    case kInt8PerTensorSym: case kInt8PerChannelSym:
+    case kUInt8PerTensorSym: case kUInt8PerChannelSym:
+    case kInt16PerTensorSym: case kInt16PerChannelSym:
+    case kUInt16PerTensorSym: case kUInt16PerChannelSym:
+    case kInt8PerTensorAsy: case kInt8PerChannelAsy:
+    case kUInt8PerTensorAsy: case kUInt8PerChannelAsy:
+    case kInt16PerTensorAsy: case kInt16PerChannelAsy:
+    case kUInt16PerTensorAsy: case kUInt16PerChannelAsy:
+      return true;
+    default:
+      return false;
+  }
 }
mllm/compile/ir/linalg/Op.hpp (1)
31-31: Duplicate forward declaration of SigmoidOp.

SigmoidOp is forward-declared both at line 31 and line 83. The duplicate at line 83 appears to be a remnant. Consider removing one to avoid redundancy.
🔎 Proposed fix

Remove line 83:
 class WhereOp;
-class SigmoidOp;
 }  // namespace mllm
Also applies to: 83-83
mllm/backends/qnn/aot/visitor/Index.cpp (1)

47-76: Axis detection logic is correct but complex—consider a brief inline comment.

The loop correctly identifies the gather axis by skipping "select all" slices while incrementing the axis counter, then breaks upon finding vector or tensor indices. The early exit and error handling for multiple axes are well implemented.

A brief comment explaining the axis-counting strategy could improve readability for future maintainers.

mllm/backends/qnn/aot/visitor/Elewise.cpp (1)

48-81: LGTM with optional refactor opportunity.

QnnAOTMulPattern::rewrite correctly implements element-wise multiplication support using ElementWiseMultiply. The implementation mirrors AddPattern which is acceptable for clarity.

Optional: Consider extracting a shared helper for the common boilerplate (env retrieval, attribute validation, input/output extraction, node registration) if more element-wise patterns are added in the future.

📜 Review details

Configuration used: defaults

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 6ada683 and 18717f8.

📒 Files selected for processing (19)

mllm/backends/qnn/aot/QnnWrappersAPI.cpp
mllm/backends/qnn/aot/passes/LLM2QnnLoweringPass.cpp
mllm/backends/qnn/aot/passes/LLM2QnnLoweringPass.hpp
mllm/backends/qnn/aot/visitor/Base.hpp
mllm/backends/qnn/aot/visitor/CastType.cpp
mllm/backends/qnn/aot/visitor/CastType.hpp
mllm/backends/qnn/aot/visitor/Elewise.cpp
mllm/backends/qnn/aot/visitor/Elewise.hpp
mllm/backends/qnn/aot/visitor/Embedding.cpp
mllm/backends/qnn/aot/visitor/Embedding.hpp
mllm/backends/qnn/aot/visitor/Index.cpp
mllm/backends/qnn/aot/visitor/Index.hpp
mllm/backends/qnn/aot/visitor/View.cpp
mllm/backends/qnn/aot/visitor/View.hpp
mllm/compile/ir/linalg/Op.cpp
mllm/compile/ir/linalg/Op.hpp
mllm/nn/Nn.hpp
mllm/nn/layers/Sigmoid.cpp
mllm/nn/layers/Sigmoid.hpp

💤 Files with no reviewable changes (1)

mllm/backends/qnn/aot/QnnWrappersAPI.cpp

🧰 Additional context used

📓 Path-based instructions (4)

{mllm,mllm-cli,pymllm}/**/*

📄 CodeRabbit inference engine (.github/copilot-instructions.md)

{mllm,mllm-cli,pymllm}/**/*: Files must not contain C0 control codes 0x00–0x08, 0x0B–0x0C, 0x0E–0x1F, C1 control codes 0x7F–0x9F, or DEL 0x7F. Horizontal tab (0x09) and line feed (0x0A) are explicitly allowed.
All files must be encoded in UTF-8 without BOM.
Any violation of character set (Rule 1) or encoding (Rule 2) requirements must cause the review to fail.
No line may end with trailing whitespace.
Use Unix line endings (LF).
File and directory names must consist only of printable Unicode characters, excluding C0 control codes 0x00–0x08, 0x0B–0x0C, 0x0E–0x1F, C1 control codes 0x7F–0x9F, and DEL 0x7F.
Only use acceptable file extensions: .c, .cc, .cpp, .cxx, .h, .hh, .hpp, .py, .pyi, .sh, .txt, .md, .yml, .yaml, .json, .toml.
Optional license headers, if present, must comply with character set rules (no C0/C1 control codes except tab and line feed).

Files:

mllm/backends/qnn/aot/visitor/Embedding.hpp
mllm/backends/qnn/aot/visitor/CastType.hpp
mllm/backends/qnn/aot/visitor/View.cpp
mllm/backends/qnn/aot/visitor/Elewise.hpp
mllm/backends/qnn/aot/visitor/Index.hpp
mllm/backends/qnn/aot/visitor/Elewise.cpp
mllm/backends/qnn/aot/visitor/Base.hpp
mllm/backends/qnn/aot/visitor/Embedding.cpp
mllm/backends/qnn/aot/visitor/CastType.cpp
mllm/backends/qnn/aot/visitor/View.hpp
mllm/backends/qnn/aot/passes/LLM2QnnLoweringPass.hpp
mllm/backends/qnn/aot/passes/LLM2QnnLoweringPass.cpp
mllm/backends/qnn/aot/visitor/Index.cpp
mllm/nn/layers/Sigmoid.cpp
mllm/compile/ir/linalg/Op.hpp
mllm/nn/Nn.hpp
mllm/compile/ir/linalg/Op.cpp
mllm/nn/layers/Sigmoid.hpp

{mllm,mllm-cli,pymllm}/**/*.{c,cc,cpp,cxx,h,hh,hpp,py,pyi,sh}

📄 CodeRabbit inference engine (.github/copilot-instructions.md)

{mllm,mllm-cli,pymllm}/**/*.{c,cc,cpp,cxx,h,hh,hpp,py,pyi,sh}: TODO and FIXME comments must be written as 'TODO:' or 'FIXME:' followed by UTF-8 text that adheres to character set rules.
Encourage consistent coding style and patterns with the existing codebase.
Ensure code is portable across supported platforms (e.g., Linux, Windows) unless explicitly platform-specific.

Files:

mllm/backends/qnn/aot/visitor/Embedding.hpp
mllm/backends/qnn/aot/visitor/CastType.hpp
mllm/backends/qnn/aot/visitor/View.cpp
mllm/backends/qnn/aot/visitor/Elewise.hpp
mllm/backends/qnn/aot/visitor/Index.hpp
mllm/backends/qnn/aot/visitor/Elewise.cpp
mllm/backends/qnn/aot/visitor/Base.hpp
mllm/backends/qnn/aot/visitor/Embedding.cpp
mllm/backends/qnn/aot/visitor/CastType.cpp
mllm/backends/qnn/aot/visitor/View.hpp
mllm/backends/qnn/aot/passes/LLM2QnnLoweringPass.hpp
mllm/backends/qnn/aot/passes/LLM2QnnLoweringPass.cpp
mllm/backends/qnn/aot/visitor/Index.cpp
mllm/nn/layers/Sigmoid.cpp
mllm/compile/ir/linalg/Op.hpp
mllm/nn/Nn.hpp
mllm/compile/ir/linalg/Op.cpp
mllm/nn/layers/Sigmoid.hpp

{mllm,mllm-cli,pymllm}/**/*.{c,cc,cpp,cxx,h,hh,hpp,py,pyi}

📄 CodeRabbit inference engine (.github/copilot-instructions.md)

{mllm,mllm-cli,pymllm}/**/*.{c,cc,cpp,cxx,h,hh,hpp,py,pyi}: Ensure public APIs, classes, and functions have clear docstrings or comments explaining purpose, parameters, returns, and errors.
Adhere to language-specific best practices and idioms (e.g., PEP 8 for Python, Google C++ Style Guide for C++).

Files:

mllm/backends/qnn/aot/visitor/Embedding.hpp
mllm/backends/qnn/aot/visitor/CastType.hpp
mllm/backends/qnn/aot/visitor/View.cpp
mllm/backends/qnn/aot/visitor/Elewise.hpp
mllm/backends/qnn/aot/visitor/Index.hpp
mllm/backends/qnn/aot/visitor/Elewise.cpp
mllm/backends/qnn/aot/visitor/Base.hpp
mllm/backends/qnn/aot/visitor/Embedding.cpp
mllm/backends/qnn/aot/visitor/CastType.cpp
mllm/backends/qnn/aot/visitor/View.hpp
mllm/backends/qnn/aot/passes/LLM2QnnLoweringPass.hpp
mllm/backends/qnn/aot/passes/LLM2QnnLoweringPass.cpp
mllm/backends/qnn/aot/visitor/Index.cpp
mllm/nn/layers/Sigmoid.cpp
mllm/compile/ir/linalg/Op.hpp
mllm/nn/Nn.hpp
mllm/compile/ir/linalg/Op.cpp
mllm/nn/layers/Sigmoid.hpp

{mllm,mllm-cli,pymllm}/**/*.{c,cc,cpp,cxx,py,pyi}

📄 CodeRabbit inference engine (.github/copilot-instructions.md)

{mllm,mllm-cli,pymllm}/**/*.{c,cc,cpp,cxx,py,pyi}: Prioritize production-ready code quality by evaluating time and space complexity of algorithms and data structures, and suggest more efficient alternatives for operations with high complexity (e.g., O(n^2) or worse) when feasible.
Avoid unnecessary object creation in loops or hot paths.
Check for proper error handling and resource cleanup (e.g., using try-finally, context managers, or RAII).
Ensure functions that can fail return appropriate error codes or raise exceptions.
Validate inputs for public APIs and critical internal functions.
Add comments for complex algorithms or non-obvious logic.
Identify potential security issues (e.g., buffer overflows, injection risks, insecure temporary files) and recommend using secure alternatives (e.g., parameterized queries, secure random generators).
Suggest adding unit tests for untested complex logic or edge cases.
Ensure code is testable by avoiding global state and using dependency injection.
Flag overly complex functions (e.g., high cyclomatic complexity) and suggest breaking them down.
Use named constants instead of magic numbers.
Add appropriate logging (e.g., debug, info, warning, error) for significant events and errors, avoiding sensitive data exposure.

Files:

mllm/backends/qnn/aot/visitor/View.cpp
mllm/backends/qnn/aot/visitor/Elewise.cpp
mllm/backends/qnn/aot/visitor/Embedding.cpp
mllm/backends/qnn/aot/visitor/CastType.cpp
mllm/backends/qnn/aot/passes/LLM2QnnLoweringPass.cpp
mllm/backends/qnn/aot/visitor/Index.cpp
mllm/nn/layers/Sigmoid.cpp
mllm/compile/ir/linalg/Op.cpp

🧬 Code graph analysis (12)

mllm/backends/qnn/aot/visitor/Embedding.hpp (1)

mllm/backends/qnn/aot/visitor/Base.hpp (8)

op (17-17)

op (17-17)

op (26-26)

op (26-26)

writer (19-19)

writer (19-19)

writer (28-28)

writer (28-28)

mllm/backends/qnn/aot/visitor/CastType.hpp (5)

mllm/backends/qnn/aot/visitor/Base.hpp (8)

op (17-17)

op (17-17)

op (26-26)

op (26-26)

writer (19-19)

writer (19-19)

writer (28-28)

writer (28-28)

mllm/backends/qnn/aot/visitor/Elewise.hpp (4)

op (14-14)

op (25-25)

writer (16-16)

writer (27-27)

mllm/backends/qnn/aot/visitor/Embedding.hpp (2)

op (14-14)

writer (16-16)

mllm/backends/qnn/aot/visitor/Index.hpp (2)

op (14-14)

writer (16-16)

mllm/backends/qnn/aot/visitor/View.hpp (2)

op (14-14)

writer (16-16)

mllm/backends/qnn/aot/visitor/View.cpp (9)

mllm/backends/qnn/aot/visitor/CastType.cpp (4)

isMatch (27-29)

isMatch (27-27)

rewrite (31-81)

rewrite (31-31)

mllm/backends/qnn/aot/visitor/Elewise.cpp (8)

isMatch (13-15)

isMatch (13-13)

isMatch (48-50)

isMatch (48-48)

rewrite (17-46)

rewrite (17-17)

rewrite (52-81)

rewrite (52-52)

mllm/backends/qnn/aot/visitor/Embedding.cpp (4)

isMatch (13-15)

isMatch (13-13)

rewrite (17-59)

rewrite (17-17)

mllm/backends/qnn/aot/visitor/Index.cpp (4)

isMatch (14-16)

isMatch (14-14)

rewrite (18-99)

rewrite (18-18)

mllm/backends/qnn/aot/visitor/Base.hpp (8)

op (17-17)

op (17-17)

op (26-26)

op (26-26)

writer (19-19)

writer (19-19)

writer (28-28)

writer (28-28)

mllm/backends/qnn/aot/visitor/Elewise.hpp (4)

op (14-14)

op (25-25)

writer (16-16)

writer (27-27)

mllm/backends/qnn/aot/visitor/Embedding.hpp (2)

op (14-14)

writer (16-16)

mllm/backends/qnn/aot/visitor/Index.hpp (2)

op (14-14)

writer (16-16)

mllm/backends/qnn/aot/visitor/View.hpp (2)

op (14-14)

writer (16-16)

mllm/backends/qnn/aot/visitor/Elewise.hpp (9)

mllm/backends/qnn/aot/visitor/Base.hpp (8)

writer (19-19)

writer (19-19)

writer (28-28)

writer (28-28)

op (17-17)

op (17-17)

op (26-26)

op (26-26)

mllm/backends/qnn/aot/visitor/CastType.hpp (2)

writer (16-16)

op (14-14)

mllm/backends/qnn/aot/visitor/Embedding.hpp (2)

writer (16-16)

op (14-14)

mllm/backends/qnn/aot/visitor/Index.hpp (2)

writer (16-16)

op (14-14)

mllm/backends/qnn/aot/visitor/View.hpp (2)

writer (16-16)

op (14-14)

mllm/backends/qnn/aot/passes/LLM2QnnLoweringPass.hpp (1)

op (22-22)

mllm/backends/qnn/aot/passes/OpNamingPass.hpp (1)

op (17-17)

mllm/backends/qnn/aot/passes/MarkTensorIO.hpp (1)

op (17-17)

mllm/backends/qnn/aot/passes/MarkQnnGraphPass.hpp (1)

op (17-17)

mllm/backends/qnn/aot/visitor/Elewise.cpp (2)

mllm/backends/qnn/aot/visitor/Base.hpp (8)

writer (19-19)

writer (19-19)

writer (28-28)

writer (28-28)

op (17-17)

op (17-17)

op (26-26)

op (26-26)

mllm/backends/qnn/aot/QnnWrappersAPI.hpp (3)

qnn_context_name (216-216)

qnn_context_name (218-219)

qnn_context_name (221-222)

mllm/backends/qnn/aot/visitor/Base.hpp (6)

mllm/backends/qnn/aot/visitor/CastType.hpp (1)

writer (16-16)

mllm/backends/qnn/aot/visitor/Elewise.hpp (2)

writer (16-16)

writer (27-27)

mllm/backends/qnn/aot/visitor/Embedding.hpp (1)

writer (16-16)

mllm/backends/qnn/aot/visitor/Index.hpp (1)

writer (16-16)

mllm/backends/qnn/aot/visitor/View.hpp (1)

writer (16-16)

mllm/compile/ir/linalg/Attribute.hpp (6)

node (27-27)

node (27-27)

node (327-327)

node (327-327)

node (348-348)

node (348-348)

mllm/backends/qnn/aot/visitor/CastType.cpp (1)

mllm/backends/qnn/aot/visitor/CastType.hpp (2)

op (14-14)

writer (16-16)

mllm/backends/qnn/aot/visitor/View.hpp (6)

mllm/backends/qnn/aot/passes/LLM2QnnLoweringPass.hpp (1)

op (22-22)

mllm/backends/qnn/aot/visitor/Base.hpp (8)

op (17-17)

op (17-17)

op (26-26)

op (26-26)

writer (19-19)

writer (19-19)

writer (28-28)

writer (28-28)

mllm/backends/qnn/aot/visitor/CastType.hpp (2)

op (14-14)

writer (16-16)

mllm/backends/qnn/aot/visitor/Elewise.hpp (4)

op (14-14)

op (25-25)

writer (16-16)

writer (27-27)

mllm/backends/qnn/aot/visitor/Embedding.hpp (2)

op (14-14)

writer (16-16)

mllm/backends/qnn/aot/visitor/Index.hpp (2)

op (14-14)

writer (16-16)

mllm/backends/qnn/aot/visitor/Index.cpp (3)

mllm/backends/qnn/aot/visitor/CastType.cpp (4)

isMatch (27-29)

isMatch (27-27)

rewrite (31-81)

rewrite (31-31)

mllm/backends/qnn/aot/visitor/Embedding.cpp (4)

isMatch (13-15)

isMatch (13-13)

rewrite (17-59)

rewrite (17-17)

mllm/backends/qnn/aot/visitor/View.cpp (4)

isMatch (14-16)

isMatch (14-14)

rewrite (18-58)

rewrite (18-18)

mllm/nn/layers/Sigmoid.cpp (1)

mllm/nn/layers/Sigmoid.hpp (2)

Sigmoid (13-13)

Sigmoid (15-15)

mllm/compile/ir/linalg/Op.hpp (2)

mllm/core/aops/SigmoidOp.cpp (1)

SigmoidOp (12-12)

mllm/core/aops/SigmoidOp.hpp (1)

SigmoidOp (15-15)

mllm/compile/ir/linalg/Op.cpp (2)

mllm/core/aops/SigmoidOp.cpp (1)

SigmoidOp (12-12)

mllm/core/aops/SigmoidOp.hpp (1)

SigmoidOp (15-15)

🪛 Clang (14.0.6)

mllm/backends/qnn/aot/visitor/Embedding.hpp

[error] 14-14: parameter name 'op' is too short, expected at least 3 characters