tvm_tirx.backend.metal.op

Metal-owned TIR intrinsic builders.

Functions

make_filled_simdgroup_matrix(d, index, value)	Create a filled SIMDGroup matrix.
simdgroup_load(d, index, ptr, stride[, col, ...])	Load data from device or threadgroup memory to simdgroup.
simdgroup_store(d, index, ptr, stride[, ...])	Store data from simdgroup to device or threadgroup memory.
simdgroup_multiply_accumulate(d, index_d, a, ...)	Multiply and accumulate two matrices in simdgroup.
simd_shuffle(var, lane, *[, ty, loc])	Shuffle a value from the selected SIMD lane.
simd_shuffle_up(var, delta, *[, ty, loc])	Shuffle a value from a lower SIMD lane.
simd_shuffle_down(var, delta, *[, ty, loc])	Shuffle a value from a higher SIMD lane.
cooperative_tensor_fill(d, index, value, ...)
cooperative_tensor_load(d, index, ptr, ...)
cooperative_tensor_store(d, index, ptr, ...)
cooperative_tensor_multiply_accumulate(d, ...)