tech.compute.tvm.tensor-math — tvm-clj 4.5

*fn-map* clj

as-cpu-tensor clj

(as-cpu-tensor data & {:keys [shape datatype]})

assign! clj

(assign! stream lhs rhs)

Broadcasting, marshalling assignment of rhs to lhs lhs = rhs

assign-constant! clj

(assign-constant! stream tensor value)

build-bind-map clj

(build-bind-map tensor-var-map)

The special cases in the bind map must relate to the name mangling of the function

cpu-fallback-impl clj macro

(cpu-fallback-impl tm-fn stream & args)

cpu-stream-fallback clj macro

(cpu-stream-fallback fn-name stream-item & args)

explode-read-tensor clj

(explode-read-tensor tensor n-dims)

get-fn-name clj

(get-fn-name stem dest & args)

get-or-create-fn clj

(get-or-create-fn stream fn-name fn-create-fn)

left-pad-ones clj

(left-pad-ones shape-vec n-dims)

n-dim-compute-op clj

(n-dim-compute-op n-dims compute-fn & {:keys [name] :or {name "compute_op"}})

n-dims->shape-stride-tuples clj

(n-dims->shape-stride-tuples n-dims arg-name)

tensor-read clj

(tensor-read placeholder index-vars shape-stride-tuples)

Perform the math required to get the absolute element offset from the shape/stride combined with the max shape variables

tensor-read-arg->mangle-str clj

(tensor-read-arg->mangle-str tensor)

For arguments where we ourselves are interpreting the argument.

tensor-read-dims->vars clj

(tensor-read-dims->vars n-dims tensor arg-name)

tensor-read-placeholder clj

(tensor-read-placeholder tensor arg-name)

tensor-result-arg->mangle-str clj

(tensor-result-arg->mangle-str tensor)

y-dim-tvm-fn clj

(y-dim-tvm-fn n-dims detailed-fn)