tilelang.intrinsics.mma_layout¶

Attributes¶

`shared_16x8_to_mma_32x4_layout_sr_a`
`shared_16x8_to_mma_32x4_layout_sr_b`
`shared_16x8_to_mma_32x4_layout_rs_a`
`shared_16x8_to_mma_32x4_layout_rs_b`
`shared_16x16_to_mma_32x8_layout_sr_a`
`shared_16x16_to_mma_32x8_layout_sr_b`
`shared_16x16_to_mma_32x8_layout_rs_a`
`shared_16x16_to_mma_32x8_layout_rs_b`
`shared_16x32_to_mma_32x16_layout_sr_a`
`shared_16x32_to_mma_32x16_layout_sr_b`
`shared_16x32_to_mma_32x16_layout_rs_a`
`shared_16x32_to_mma_32x16_layout_rs_b`

Functions¶

`ldmatrix_32x4_to_shared_16x8_layout_a`(thread_id, local_id)
`ldmatrix_32x4_to_shared_16x8_layout_b`(thread_id, local_id)
`ldmatrix_32x8_to_shared_16x16_layout`(thread_id, local_id)
`ldmatrix_trans_32x8_to_shared_16x16_layout`(thread_id, ...)
`ldmatrix_32x16_to_shared_16x32_layout_a`(thread_id, ...)
`ldmatrix_32x16_to_shared_16x32_layout_b`(thread_id, ...)
`mma_store_32x8_to_shared_16x16_layout`(thread_id, local_id)
`mma_store_32x2_to_shared_8x8_layout_fp64`(thread_id, ...)
`shared_16x8_to_mma_a_32x4_layout`(i, j)
`shared_16x8_to_mma_a_32x4_layout_trans`(i, j)
`shared_16x8_to_mma_b_32x4_layout`(i, j)
`shared_16x8_to_mma_b_32x4_layout_trans`(i, j)
`shared_16x16_to_mma_a_32x8_layout`(i, j)
`shared_16x16_to_mma_a_32x8_layout_trans`(i, j)
`shared_16x16_to_mma_b_32x8_layout`(i, j)
`shared_16x16_to_mma_b_32x8_layout_trans`(i, j)
`shared_16x32_to_mma_a_32x16_layout`(i, j)
`shared_32x16_to_mma_a_32x16_layout_trans`(i, j)
`shared_16x32_to_mma_b_32x16_layout`(i, j)
`shared_32x16_to_mma_b_32x16_layout_trans`(i, j)
`mma_32x8_to_shared_16x16_layout`(thread_id, local_id)
`mma_load_a_32x4_to_shared_16x8_layout`(thread_id, local_id)
`mma_load_b_32x4_to_shared_16x8_layout`(thread_id, local_id)
`mma_load_a_32x16_to_shared_16x32_layout`(thread_id, ...)
`mma_load_b_32x16_to_shared_16x32_layout`(thread_id, ...)
`shared_16x16_to_mma_32x8_smoothlayout`(i, j)
`shared_16x32_to_mma_32x16_smoothlayout`(i, j)
`shared_32x16_to_mma_32x16_smoothlayout`(i, j)
`get_swizzle_layout`(row_idx, col_idx, row_size, dtype)
`make_mma_swizzle_layout`(shared_buf[, is_smooth])

Module Contents¶

tilelang.intrinsics.mma_layout.ldmatrix_32x4_to_shared_16x8_layout_a(thread_id, local_id)¶

tilelang.intrinsics.mma_layout.ldmatrix_32x4_to_shared_16x8_layout_b(thread_id, local_id)¶

tilelang.intrinsics.mma_layout.ldmatrix_32x8_to_shared_16x16_layout(thread_id, local_id)¶

tilelang.intrinsics.mma_layout.ldmatrix_trans_32x8_to_shared_16x16_layout(thread_id, local_id)¶

tilelang.intrinsics.mma_layout.ldmatrix_32x16_to_shared_16x32_layout_a(thread_id, local_id)¶

tilelang.intrinsics.mma_layout.ldmatrix_32x16_to_shared_16x32_layout_b(thread_id, local_id)¶

tilelang.intrinsics.mma_layout.mma_store_32x8_to_shared_16x16_layout(thread_id, local_id)¶

tilelang.intrinsics.mma_layout.mma_store_32x2_to_shared_8x8_layout_fp64(thread_id, local_id)¶

tilelang.intrinsics.mma_layout.shared_16x8_to_mma_a_32x4_layout(i, j)¶

tilelang.intrinsics.mma_layout.shared_16x8_to_mma_a_32x4_layout_trans(i, j)¶

tilelang.intrinsics.mma_layout.shared_16x8_to_mma_b_32x4_layout(i, j)¶

tilelang.intrinsics.mma_layout.shared_16x8_to_mma_b_32x4_layout_trans(i, j)¶

tilelang.intrinsics.mma_layout.shared_16x8_to_mma_32x4_layout_sr_a¶

tilelang.intrinsics.mma_layout.shared_16x8_to_mma_32x4_layout_sr_b¶

tilelang.intrinsics.mma_layout.shared_16x8_to_mma_32x4_layout_rs_a¶

tilelang.intrinsics.mma_layout.shared_16x8_to_mma_32x4_layout_rs_b¶

tilelang.intrinsics.mma_layout.shared_16x16_to_mma_a_32x8_layout(i, j)¶

tilelang.intrinsics.mma_layout.shared_16x16_to_mma_a_32x8_layout_trans(i, j)¶

tilelang.intrinsics.mma_layout.shared_16x16_to_mma_b_32x8_layout(i, j)¶

tilelang.intrinsics.mma_layout.shared_16x16_to_mma_b_32x8_layout_trans(i, j)¶

tilelang.intrinsics.mma_layout.shared_16x16_to_mma_32x8_layout_sr_a¶

tilelang.intrinsics.mma_layout.shared_16x16_to_mma_32x8_layout_sr_b¶

tilelang.intrinsics.mma_layout.shared_16x16_to_mma_32x8_layout_rs_a¶

tilelang.intrinsics.mma_layout.shared_16x16_to_mma_32x8_layout_rs_b¶

tilelang.intrinsics.mma_layout.shared_16x32_to_mma_a_32x16_layout(i, j)¶

tilelang.intrinsics.mma_layout.shared_32x16_to_mma_a_32x16_layout_trans(i, j)¶

tilelang.intrinsics.mma_layout.shared_16x32_to_mma_b_32x16_layout(i, j)¶

tilelang.intrinsics.mma_layout.shared_32x16_to_mma_b_32x16_layout_trans(i, j)¶

tilelang.intrinsics.mma_layout.shared_16x32_to_mma_32x16_layout_sr_a¶

tilelang.intrinsics.mma_layout.shared_16x32_to_mma_32x16_layout_sr_b¶

tilelang.intrinsics.mma_layout.shared_16x32_to_mma_32x16_layout_rs_a¶

tilelang.intrinsics.mma_layout.shared_16x32_to_mma_32x16_layout_rs_b¶

tilelang.intrinsics.mma_layout.mma_32x8_to_shared_16x16_layout(thread_id, local_id)¶

tilelang.intrinsics.mma_layout.mma_load_a_32x4_to_shared_16x8_layout(thread_id, local_id)¶

tilelang.intrinsics.mma_layout.mma_load_b_32x4_to_shared_16x8_layout(thread_id, local_id)¶

tilelang.intrinsics.mma_layout.mma_load_a_32x16_to_shared_16x32_layout(thread_id, local_id)¶

tilelang.intrinsics.mma_layout.mma_load_b_32x16_to_shared_16x32_layout(thread_id, local_id)¶

tilelang.intrinsics.mma_layout.shared_16x16_to_mma_32x8_smoothlayout(i, j)¶

tilelang.intrinsics.mma_layout.shared_16x32_to_mma_32x16_smoothlayout(i, j)¶

tilelang.intrinsics.mma_layout.shared_32x16_to_mma_32x16_smoothlayout(i, j)¶

tilelang.intrinsics.mma_layout.get_swizzle_layout(row_idx, col_idx, row_size, dtype, swizzle_bytes=None)¶

Parameters:: dtype (tvm.DataType | str)

tilelang.intrinsics.mma_layout.make_mma_swizzle_layout(shared_buf, is_smooth=False)¶

Parameters:: is_smooth (bool)