tilelang.contrib.cutedsl.threadblock_swizzle¶
Classes¶
Three-dimensional CUDA index tuple. |
Functions¶
Return the current CUDA thread index. |
|
|
Return the current CUDA block index. |
|
Return the CUDA grid dimensions. |
|
Map block indices to row-major swizzled rasterization coordinates. |
|
Map block indices to column-major swizzled rasterization coordinates. |
|
Row-major swizzle at cluster granularity while preserving intra-cluster x rank. |
|
Column-major swizzle at cluster granularity while preserving intra-cluster x rank. |
Module Contents¶
- class tilelang.contrib.cutedsl.threadblock_swizzle.dim3¶
Three-dimensional CUDA index tuple.
- x: int¶
- y: int¶
- z: int¶
- tilelang.contrib.cutedsl.threadblock_swizzle.ThreadIdx()¶
Return the current CUDA thread index.
- Return type:
- tilelang.contrib.cutedsl.threadblock_swizzle.BlockIdx()¶
Return the current CUDA block index.
- Return type:
- tilelang.contrib.cutedsl.threadblock_swizzle.GridDim()¶
Return the CUDA grid dimensions.
- Return type:
- tilelang.contrib.cutedsl.threadblock_swizzle.rasterization2DRow(panel_width)¶
Map block indices to row-major swizzled rasterization coordinates.
- Parameters:
panel_width (cutlass.Constexpr[int])
- Return type:
- tilelang.contrib.cutedsl.threadblock_swizzle.rasterization2DColumn(panel_width)¶
Map block indices to column-major swizzled rasterization coordinates.
- Parameters:
panel_width (cutlass.Constexpr[int])
- Return type:
- tilelang.contrib.cutedsl.threadblock_swizzle.rasterization2DRowWithCluster(panel_width, cluster_dim_x)¶
Row-major swizzle at cluster granularity while preserving intra-cluster x rank.
- Parameters:
panel_width (cutlass.Constexpr[int])
cluster_dim_x (cutlass.Constexpr[int])
- Return type: