Mojo module
tcgen05
This module includes utilities for working with the tensorcore 5th generation (tcgen05) instructions.
Aliases
check_blackwell_constraint
alias check_blackwell_constraint = constrained[::Bool,::StringSlice[::Bool[_has_blackwell_tcgen05(), __init__[__mlir_type.!kgen.string]("The tcgen05 instructions are only applicable on nVidia Blackwell (sm_100a, sm_101a) hardware."), ?]
Structs
-
TensorMemory
: A wrapper around tensor memory allocated for tcgen05 instructions.
Functions
-
tcgen05_alloc
: Allocates tensor memory for use with tcgen05 instructions. -
tcgen05_cp
: Copies data from shared memory described by the matrix descriptors_desc
to tensor memorytmem_addr
. -
tcgen05_dealloc
: Deallocates tensor memory allocated by tcgen05_alloc(). -
tcgen05_fence_after
: Orders all the subsequent asynchronoustcgen05
operations. -
tcgen05_fence_before
: Orders all the prior asynchronoustcgen05
operations. -
tcgen05_ld
: Loads data from tensor memory into registers. -
tcgen05_load_wait
: Waits for tensor memory loads to complete. -
tcgen05_release_allocation_lock
: Releases the allocation lock for the current CTA group. -
tcgen05_st
: Stores data from registers into tensor memory. -
tcgen05_store_wait
: Waits for tensor memory stores to complete.
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!