tilelang.intrinsics.wgmma_macro_generator

Attributes

lift

Classes

SwizzleMode

Enum where members are also (and must be) ints

TensorCoreIntrinEmitter

To eliminate Python syntax within TIR Macro.

Module Contents

tilelang.intrinsics.wgmma_macro_generator.lift
class tilelang.intrinsics.wgmma_macro_generator.SwizzleMode

Bases: enum.IntEnum

Enum where members are also (and must be) ints

NONE = 0
SWIZZLE_128B = 1
SWIZZLE_64B = 2
SWIZZLE_32B = 3
is_none()
返回类型:

bool

is_swizzle_32b()
返回类型:

bool

is_swizzle_64b()
返回类型:

bool

is_swizzle_128b()
返回类型:

bool

swizzle_byte_size()
返回类型:

int

swizzle_atom_size()
返回类型:

int

class tilelang.intrinsics.wgmma_macro_generator.TensorCoreIntrinEmitter(a_dtype=T.float16, b_dtype=T.float16, accum_dtype=T.float16, a_transposed=False, b_transposed=False, block_row_warps=2, block_col_warps=2, warp_row_tiles=8, warp_col_tiles=8, chunk=16, reduce_k=1, num_elems_per_byte=1, is_m_first=False, thread_var=None)

Bases: tilelang.intrinsics.mma_macro_generator.TensorCoreIntrinEmitter

To eliminate Python syntax within TIR Macro.

参数:

  • a_dtype (str)

  • b_dtype (str)

  • accum_dtype (str)

  • a_transposed (bool)

  • b_transposed (bool)

  • block_row_warps (int)

  • block_col_warps (int)

  • warp_row_tiles (int)

  • warp_col_tiles (int)

  • chunk (int)

  • reduce_k (int)

  • num_elems_per_byte (int)

  • is_m_first (bool | None)

  • thread_var (tvm.tir.Var | None)

wgmma_prefix: str
wgmma_inst_m: int
wgmma_inst_n: int
a_shared_layout: tilelang.layout.Layout = None
b_shared_layout: tilelang.layout.Layout = None
wgmma(A_region, B_region, C_region, clear_accum=False, wg_wait=0)
参数:

  • A_region (tvm.tir.BufferRegion)

  • B_region (tvm.tir.BufferRegion)

  • C_region (tvm.tir.BufferRegion)

  • clear_accum (tvm.tir.PrimExpr)

  • wg_wait (int)

wgmma_rs(A_region, B_region, C_region, clear_accum=False, wg_wait=0)
参数:

  • A_region (tvm.tir.BufferRegion)

  • B_region (tvm.tir.BufferRegion)

  • C_region (tvm.tir.BufferRegion)

  • clear_accum (tvm.tir.PrimExpr)

  • wg_wait (int)

make_mma_load_layout(local_buf, matrix='A')

Create a layout function for storing MMA results into a fragment buffer. This layout is used in conjunction with inverse_mma_store_layout to map fragment indices to threads and local indices.

参数:

  • local_buf (tir.Buffer) -- The local buffer representing a fragment of a matrix.

  • matrix (str)

返回:

A fragment object that describes how threads and indices in local_buf are laid out.

返回类型:

T.Fragment

抛出:

AssertionError -- If local_buf is not detected to be a fragment buffer.

make_mma_store_layout(local_buf)

Create a layout function for storing MMA results into a fragment buffer. This layout is used in conjunction with inverse_mma_store_layout to map fragment indices to threads and local indices.

参数:

local_buf (tir.Buffer) -- The local buffer representing a fragment of a matrix.

返回:

A fragment object that describes how threads and indices in local_buf are laid out.

返回类型:

T.Fragment

抛出:

AssertionError -- If local_buf is not detected to be a fragment buffer.