tilelang.engine.phase¶

Functions¶

allow_warp_specialized([pass_ctx, target])

allow_tma_and_warp_specialized([pass_ctx, target])

allow_fence_proxy([target])

allow_vectorize([pass_ctx])

allow_global_thread_synchronization([pass_ctx])

should_enable_aggressive_merge([pass_ctx, target])

LowerAndLegalize(mod, target)

Bind target information and progressively legalize and lower frontend Tile IR into a form suitable for downstream optimization and codegen.

OptimizeForTarget(mod, target)

Module Contents¶

tilelang.engine.phase.allow_warp_specialized(pass_ctx=None, target=None)¶
Parameters:

  • pass_ctx (Optional[tilelang.transform.PassContext])

  • target (Optional[tvm.target.Target])

Return type:

bool

tilelang.engine.phase.allow_tma_and_warp_specialized(pass_ctx=None, target=None)¶
Parameters:

  • pass_ctx (Optional[tilelang.transform.PassContext])

  • target (Optional[tvm.target.Target])

Return type:

bool

tilelang.engine.phase.allow_fence_proxy(target=None)¶
Parameters:

target (Optional[tvm.target.Target])

Return type:

bool

tilelang.engine.phase.allow_vectorize(pass_ctx=None)¶
Parameters:

pass_ctx (Optional[tilelang.transform.PassContext])

Return type:

bool

tilelang.engine.phase.allow_global_thread_synchronization(pass_ctx=None)¶
Parameters:

pass_ctx (Optional[tilelang.transform.PassContext])

Return type:

bool

tilelang.engine.phase.should_enable_aggressive_merge(pass_ctx=None, target=None)¶
Parameters:

  • pass_ctx (Optional[tilelang.transform.PassContext])

  • target (Optional[tvm.target.Target])

Return type:

bool

tilelang.engine.phase.LowerAndLegalize(mod, target)¶

Bind target information and progressively legalize and lower frontend Tile IR into a form suitable for downstream optimization and codegen.

This pass pipeline: - Binds the provided target to the module. - Legalizes frontend Tile IR into TVM-compatible constructs. - Simplifies expressions. - Configures reducer layouts and performs layout inference for fragments and shared memory. - Lowers high-level tile operations and L2 persistent maps. - Legalizes vectorized loops and inserts safety checks for memory accesses. - Re-simplifies to remove redundancies introduced by safety checks. - Attempts loop vectorization for dynamic-shaped loops.

Parameters:

  • mod (IRModule) – The input IR module containing frontend Tile IR.

  • target (Target) – Target device information to bind into the module.

Returns:

The transformed module, ready for target-specific optimization passes.

Return type:

IRModule

tilelang.engine.phase.OptimizeForTarget(mod, target)¶
Parameters:

  • mod (tvm.IRModule)

  • target (tvm.target.Target)

Return type:

tvm.IRModule