For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /max/get-started.md).
Mojo function
elementwise_epilogue_c_tile
def elementwise_epilogue_c_tile[simd_width: Int, c_type: DType, func: def[dtype: DType, width: SIMDLength, *, alignment: Int = Int(1)](IndexList[Int(2)], SIMD[dtype, width]) capturing thin -> None](offset: GemmShape, tile_len: GemmShape, c: TileTensor[c_type, Storage=c.Storage, linear_idx_type=c.linear_idx_type])
Applies a vectorized epilogue function over a 2D C output tile.
Iterates column chunks of tile_len.N using SIMD width simd_width,
calling func with the global (m, n) coordinates and the loaded values.
Parameters:
- βsimd_width (
Int): SIMD vector width for epilogue loads. - βc_type (
DType): Data type of the C matrix. - βfunc (
def[dtype: DType, width: SIMDLength, *, alignment: Int = Int(1)](IndexList[Int(2)], SIMD[dtype, width]) capturing thin -> None): Epilogue function called with coordinates and a SIMD value chunk.
Args:
- βoffset (
GemmShape): Starting (M, N, K) offset within the global matmul space. - βtile_len (
GemmShape): Number of rows and columns to process. - βc (
TileTensor[c_type, Storage=c.Storage, linear_idx_type=c.linear_idx_type]): Read-only view of the C output tile.
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!