tilelang.language.tir.ir¶

Attributes¶

`abs`
`acos`
`acosh`
`address_of`
`asin`
`asinh`
`atan`
`atan2`
`atanh`
`bitwise_and`
`bitwise_not`
`bitwise_or`
`bitwise_xor`
`ceil`
`clz`
`copysign`
`cos`
`cosh`
`erf`
`exp`
`exp2`
`exp10`
`floor`
`ceildiv`
`cdiv`
`floordiv`
`floormod`
`fmod`
`hypot`
`if_then_else`
`infinity`
`isfinite`
`isinf`
`isnan`
`isnullptr`
`ldexp`
`likely`
`log`
`log1p`
`log2`
`log10`
`lookup_param`
`max_value`
`min_value`
`nearbyint`
`nextafter`
`popcount`
`pow`
`q_multiply_shift`
`q_multiply_shift_per_axis`
`ret`
`round`
`rsqrt`
`shift_left`
`shift_right`
`sigmoid`
`sin`
`sinh`
`sqrt`
`tan`
`tanh`
`trunc`
`truncdiv`
`truncmod`
`tvm_access_ptr`
`tvm_throw_last_error`
`tvm_stack_alloca`
`tvm_stack_make_shape`
`tvm_stack_make_array`
`tvm_check_return`
`call_packed`
`call_cpacked`
`call_packed_lowered`
`call_cpacked_lowered`
`tvm_tuple`
`tvm_struct_set`
`tvm_struct_get`
`tvm_thread_invariant`
`tvm_thread_allreduce`
`tvm_load_matrix_sync`
`tvm_mma_sync`
`tvm_bmma_sync`
`tvm_fill_fragment`
`tvm_store_matrix_sync`
`tvm_storage_sync`
`tvm_warp_shuffle`
`tvm_warp_shuffle_up`
`tvm_warp_shuffle_down`
`tvm_warp_activemask`
`ptx_wait_group`
`ptx_commit_group`
`ptx_cp_async_barrier`
`ptx_init_barrier_thread_count`
`ptx_fence_barrier_init`
`ptx_arrive_barrier`
`ptx_arrive_barrier_expect_tx`
`ptx_wait_barrier`
`create_barriers`
`assume`
`undef`
`TVMBackendAllocWorkspace`
`TVMBackendFreeWorkspace`
`start_profile_intrinsic`
`end_profile_intrinsic`
`anylist_getitem`
`anylist_resetitem`
`anylist_setitem_call_packed`
`anylist_setitem_call_cpacked`
`vscale`
`reinterpret`
`call_extern`
`call_intrin`
`call_llvm_intrin`
`call_llvm_pure_intrin`
`call_pure_extern`
`ptx_mma`
`ptx_mma_sp`
`ptx_wgmma_ss`
`ptx_wgmma_rs`
`ptx_wgmma_sp_ss`
`ptx_wgmma_sp_rs`
`ptx_tcgen05_mma_ss`
`ptx_tcgen05_mma_ts`
`ptx_tcgen05_mma_blockscaled_ss`
`ptx_ldmatrix`
`ptx_cp_async`
`ptx_cp_async_bulk`
`mma_store`
`mma_fill`
`vectorlow`
`vectorhigh`
`vectorcombine`
`tvm_mfma`
`tvm_mfma_store`
`tvm_rdna_wmma`
`tvm_rdna_wmma_store`

Functions¶

`serial`(start[, stop, annotations])	The serial For statement.
`parallel`(start[, stop, annotations])	The parallel For statement.
`vectorized`(start[, stop, annotations])	The vectorized For statement.
`unroll`(start[, stop, annotations])	The unrolled For statement.
`thread_binding`(start[, stop, thread, annotations])	The thread-binding For statement.
`grid`(*extents)	The grid For statement.
`cast`(value, dtype[, round, sat, rbits, span])	Cast `value` to `dtype` with optional PTX-style rounding hints.

Module Contents¶

tilelang.language.tir.ir.serial(start, stop=None, *, annotations=None)¶

The serial For statement.

Parameters:

start (PrimExpr) – The minimum value of iteration.
stop (PrimExpr) – The maximum value of iteration.
annotations (Dict[str, Any]) – The optional annotations of the For statement.

Returns:

res – The ForFrame.

Return type:

frame.ForFrame

tilelang.language.tir.ir.parallel(start, stop=None, *, annotations=None)¶

The parallel For statement.

Parameters:

start (PrimExpr) – The minimum value of iteration.
stop (PrimExpr) – The maximum value of iteration.
annotations (Dict[str, Any]) – The optional annotations of the For statement.

Returns:

res – The ForFrame.

Return type:

frame.ForFrame

tilelang.language.tir.ir.vectorized(start, stop=None, *, annotations=None)¶

The vectorized For statement.

Parameters:

start (PrimExpr) – The minimum value of iteration.
stop (PrimExpr) – The maximum value of iteration.
annotations (Dict[str, Any]) – The optional annotations of the For statement.

Returns:

res – The ForFrame.

Return type:

frame.ForFrame

tilelang.language.tir.ir.unroll(start, stop=None, *, annotations=None)¶

The unrolled For statement.

Parameters:

start (PrimExpr) – The minimum value of iteration.
stop (PrimExpr) – The maximum value of iteration.
annotations (Dict[str, Any]) – The optional annotations of the For statement.

Returns:

res – The ForFrame.

Return type:

frame.ForFrame

tilelang.language.tir.ir.thread_binding(start, stop=None, thread=None, *, annotations=None)¶

The thread-binding For statement.

Parameters:

start (PrimExpr) – The minimum value of iteration.
stop (PrimExpr) – The maximum value of iteration.
thread (str) – The thread for loop variable to bind.
annotations (Dict[str, Any]) – The optional annotations of the For statement.

Returns:

res – The ForFrame.

Return type:

frame.ForFrame

tilelang.language.tir.ir.grid(*extents)¶

The grid For statement.

Parameters:: extents (PrimExpr) – The extents of the iteration.
Returns:: res – The ForFrame.
Return type:: frame.ForFrame

tilelang.language.tir.ir.abs¶

tilelang.language.tir.ir.acos¶

tilelang.language.tir.ir.acosh¶

tilelang.language.tir.ir.address_of¶

tilelang.language.tir.ir.asin¶

tilelang.language.tir.ir.asinh¶

tilelang.language.tir.ir.atan¶

tilelang.language.tir.ir.atan2¶

tilelang.language.tir.ir.atanh¶

tilelang.language.tir.ir.bitwise_and¶

tilelang.language.tir.ir.bitwise_not¶

tilelang.language.tir.ir.bitwise_or¶

tilelang.language.tir.ir.bitwise_xor¶

tilelang.language.tir.ir.ceil¶

tilelang.language.tir.ir.clz¶

tilelang.language.tir.ir.copysign¶

tilelang.language.tir.ir.cos¶

tilelang.language.tir.ir.cosh¶

tilelang.language.tir.ir.erf¶

tilelang.language.tir.ir.exp¶

tilelang.language.tir.ir.exp2¶

tilelang.language.tir.ir.exp10¶

tilelang.language.tir.ir.floor¶

tilelang.language.tir.ir.ceildiv¶

tilelang.language.tir.ir.cdiv¶

tilelang.language.tir.ir.floordiv¶

tilelang.language.tir.ir.floormod¶

tilelang.language.tir.ir.fmod¶

tilelang.language.tir.ir.hypot¶

tilelang.language.tir.ir.if_then_else¶

tilelang.language.tir.ir.infinity¶

tilelang.language.tir.ir.isfinite¶

tilelang.language.tir.ir.isinf¶

tilelang.language.tir.ir.isnan¶

tilelang.language.tir.ir.isnullptr¶

tilelang.language.tir.ir.ldexp¶

tilelang.language.tir.ir.likely¶

tilelang.language.tir.ir.log¶

tilelang.language.tir.ir.log1p¶

tilelang.language.tir.ir.log2¶

tilelang.language.tir.ir.log10¶

tilelang.language.tir.ir.lookup_param¶

tilelang.language.tir.ir.max_value¶

tilelang.language.tir.ir.min_value¶

tilelang.language.tir.ir.nearbyint¶

tilelang.language.tir.ir.nextafter¶

tilelang.language.tir.ir.popcount¶

tilelang.language.tir.ir.pow¶

tilelang.language.tir.ir.q_multiply_shift¶

tilelang.language.tir.ir.q_multiply_shift_per_axis¶

tilelang.language.tir.ir.ret¶

tilelang.language.tir.ir.round¶

tilelang.language.tir.ir.rsqrt¶

tilelang.language.tir.ir.shift_left¶

tilelang.language.tir.ir.shift_right¶

tilelang.language.tir.ir.sigmoid¶

tilelang.language.tir.ir.sin¶

tilelang.language.tir.ir.sinh¶

tilelang.language.tir.ir.sqrt¶

tilelang.language.tir.ir.tan¶

tilelang.language.tir.ir.tanh¶

tilelang.language.tir.ir.trunc¶

tilelang.language.tir.ir.truncdiv¶

tilelang.language.tir.ir.truncmod¶

tilelang.language.tir.ir.tvm_access_ptr¶

tilelang.language.tir.ir.tvm_throw_last_error¶

tilelang.language.tir.ir.tvm_stack_alloca¶

tilelang.language.tir.ir.tvm_stack_make_shape¶

tilelang.language.tir.ir.tvm_stack_make_array¶

tilelang.language.tir.ir.tvm_check_return¶

tilelang.language.tir.ir.call_packed¶

tilelang.language.tir.ir.call_cpacked¶

tilelang.language.tir.ir.call_packed_lowered¶

tilelang.language.tir.ir.call_cpacked_lowered¶

tilelang.language.tir.ir.tvm_tuple¶

tilelang.language.tir.ir.tvm_struct_set¶

tilelang.language.tir.ir.tvm_struct_get¶

tilelang.language.tir.ir.tvm_thread_invariant¶

tilelang.language.tir.ir.tvm_thread_allreduce¶

tilelang.language.tir.ir.tvm_load_matrix_sync¶

tilelang.language.tir.ir.tvm_mma_sync¶

tilelang.language.tir.ir.tvm_bmma_sync¶

tilelang.language.tir.ir.tvm_fill_fragment¶

tilelang.language.tir.ir.tvm_store_matrix_sync¶

tilelang.language.tir.ir.tvm_storage_sync¶

tilelang.language.tir.ir.tvm_warp_shuffle¶

tilelang.language.tir.ir.tvm_warp_shuffle_up¶

tilelang.language.tir.ir.tvm_warp_shuffle_down¶

tilelang.language.tir.ir.tvm_warp_activemask¶

tilelang.language.tir.ir.ptx_wait_group¶

tilelang.language.tir.ir.ptx_commit_group¶

tilelang.language.tir.ir.ptx_cp_async_barrier¶

tilelang.language.tir.ir.ptx_init_barrier_thread_count¶

tilelang.language.tir.ir.ptx_fence_barrier_init¶

tilelang.language.tir.ir.ptx_arrive_barrier¶

tilelang.language.tir.ir.ptx_arrive_barrier_expect_tx¶

tilelang.language.tir.ir.ptx_wait_barrier¶

tilelang.language.tir.ir.create_barriers¶

tilelang.language.tir.ir.assume¶

tilelang.language.tir.ir.undef¶

tilelang.language.tir.ir.TVMBackendAllocWorkspace¶

tilelang.language.tir.ir.TVMBackendFreeWorkspace¶

tilelang.language.tir.ir.start_profile_intrinsic¶

tilelang.language.tir.ir.end_profile_intrinsic¶

tilelang.language.tir.ir.anylist_getitem¶

tilelang.language.tir.ir.anylist_resetitem¶

tilelang.language.tir.ir.anylist_setitem_call_packed¶

tilelang.language.tir.ir.anylist_setitem_call_cpacked¶

tilelang.language.tir.ir.vscale¶

tilelang.language.tir.ir.reinterpret¶

tilelang.language.tir.ir.call_extern¶

tilelang.language.tir.ir.call_intrin¶

tilelang.language.tir.ir.call_llvm_intrin¶

tilelang.language.tir.ir.call_llvm_pure_intrin¶

tilelang.language.tir.ir.call_pure_extern¶

tilelang.language.tir.ir.ptx_mma¶

tilelang.language.tir.ir.ptx_mma_sp¶

tilelang.language.tir.ir.ptx_wgmma_ss¶

tilelang.language.tir.ir.ptx_wgmma_rs¶

tilelang.language.tir.ir.ptx_wgmma_sp_ss¶

tilelang.language.tir.ir.ptx_wgmma_sp_rs¶

tilelang.language.tir.ir.ptx_tcgen05_mma_ss¶

tilelang.language.tir.ir.ptx_tcgen05_mma_ts¶

tilelang.language.tir.ir.ptx_tcgen05_mma_blockscaled_ss¶

tilelang.language.tir.ir.ptx_ldmatrix¶

tilelang.language.tir.ir.ptx_cp_async¶

tilelang.language.tir.ir.ptx_cp_async_bulk¶

tilelang.language.tir.ir.mma_store¶

tilelang.language.tir.ir.mma_fill¶

tilelang.language.tir.ir.vectorlow¶

tilelang.language.tir.ir.vectorhigh¶

tilelang.language.tir.ir.vectorcombine¶

tilelang.language.tir.ir.tvm_mfma¶

tilelang.language.tir.ir.tvm_mfma_store¶

tilelang.language.tir.ir.tvm_rdna_wmma¶

tilelang.language.tir.ir.tvm_rdna_wmma_store¶

tilelang.language.tir.ir.cast(value, dtype, round='', sat=True, rbits=None, span=None)¶

Cast value to dtype with optional PTX-style rounding hints.

Parameters:

value (object) – The source operand.
dtype (str) – The target data type.
round (str, optional) – PTX rounding modifier (e.g. "rn", "rz", "rp", "rm", "rs"). Empty string means use the backend default. Currently only "" and "rs" are lowered by the CUDA backend; the other modifiers are reserved.
sat (bool, optional) – Saturate to finite (True = PTX .satfinite, default).
rbits (PrimExpr, optional) – Random bits operand for stochastic rounding (round="rs").
span (Optional[Span]) – The location of this operator in the source.

Returns:

op – The cast expression.

Return type:

tvm.tir.PrimExpr