qalgora-Q Docs Hub量子文档 ✦ Ask AI✦ 问问文档
/

Writing an MLIR Pass for qalgora-Q (规划中 · 暂不可运行)

○ Planned · Not yet implemented
The capabilities described on this page are planned and not yet implemented or released. They explain future design directions and should not be interpreted as delivered features.

Extend the compiler with a custom transformation over qalgora IR — for a new optimization, an analysis, or hardware-specific lowering.

规划中 · 暂不可运行 — planned, not runnable yet
The MLIR pass framework and the qalgora-opt tool shown here are a planned interface; they are not published in the open reference build, so nothing on this page compiles or runs today. The code below is illustrative pseudocode: it omits the implementation of nextOnSameQubit and the MLIR pattern-rewrite details.

A minimal pass

#include "qalgora/Optimizer/Transforms/Passes.h"
#include "mlir/Pass/Pass.h"

// Pseudocode sketch: nextOnSameQubit() and the rewrite plumbing are elided.
// The top-level op (mlir::ModuleOp here) depends on how qalgora IR nests ops.
struct CancelAdjacentH
    : public mlir::PassWrapper<CancelAdjacentH, mlir::OperationPass<mlir::ModuleOp>> {
  void runOnOperation() override {
    getOperation()->walk([](qalgora::HOp op) {
      // if the next op on this qubit is another H, erase both
      if (auto next = nextOnSameQubit(op))
        // NOTE: erasing op/next while walk() iterates can invalidate the
        // iterator — collect matches first, or use the pattern rewriter.
        if (mlir::isa<qalgora::HOp>(next)) { op.erase(); next.erase(); }
    });
  }
};

QALGORA_REGISTER_PASS(CancelAdjacentH, "cancel-adjacent-h")
C++ API is planned — won’t run today
The runnable reference build is Python-only; the C++ library, its headers and the build tooling shown here are a planned interface and are not published yet, so this snippet will not compile or run as-is. Use the Python API against the reference build to actually run these examples today.

Running your pass

qalgora-opt input.qke --cancel-adjacent-h -o output.qke
Pass pipelines
Register your pass so it can be slotted into the optimization pipeline alongside built-ins like canonicalization and gate fusion.

为 qalgora-Q 编写 MLIR Pass （规划中 · 暂不可运行）

○ 规划中
本页所述能力属于规划功能，当前尚未发布或尚未实现。相关内容仅用于说明未来设计方向，不应理解为已交付能力。

写一个对 qalgora IR 做自定义变换的 pass，就能扩展编译器——无论是新增优化、分析，还是面向特定硬件的降级。

规划中 · 暂不可运行
此处展示的 MLIR pass 框架和 qalgora-opt 工具属于规划中的接口，开放参考实现尚未发布，因此本页代码当前无法编译或运行。以下代码为伪代码示意，省略了 nextOnSameQubit 的实现和 MLIR pattern rewrite 细节。

最小化 pass 示例

#include "qalgora/Optimizer/Transforms/Passes.h"
#include "mlir/Pass/Pass.h"

// 伪代码示意：nextOnSameQubit() 与 rewrite 细节均已省略。
// 顶层算子（此处为 mlir::ModuleOp）取决于 qalgora IR 如何嵌套算子。
struct CancelAdjacentH
    : public mlir::PassWrapper<CancelAdjacentH, mlir::OperationPass<mlir::ModuleOp>> {
  void runOnOperation() override {
    getOperation()->walk([](qalgora::HOp op) {
      // if the next op on this qubit is another H, erase both
      if (auto next = nextOnSameQubit(op))
        // 注意：在 walk() 迭代期间 erase op/next 可能使迭代器失效——
        // 应先收集匹配项，或改用 pattern rewriter。
        if (mlir::isa<qalgora::HOp>(next)) { op.erase(); next.erase(); }
    });
  }
};

QALGORA_REGISTER_PASS(CancelAdjacentH, "cancel-adjacent-h")
C++ 接口为规划中 · 暂不可运行
可运行的参考实现仅提供 Python；此处展示的 C++ 库 头文件与构建工具属于规划中的接口 尚未发布 因此该片段当前无法直接编译或运行。若要真正运行这些示例 请使用 Python API 对接参考实现。

运行自定义 pass

qalgora-opt input.qke --cancel-adjacent-h -o output.qke
Pass 流水线
把你的 pass 注册进来，它就能和规范化、门融合等内置 pass 一起编入优化流水线。