tvm_tirx.tirx.tile.views#
Typed views of tensor instruction operands during tile lowering.
Classes#
|
Base class for unary operators: unary(output, input). |
Extended unary operator with bias and scale parameters: unary_with_scale_bias(output, input, scale, bias). |
|
|
Base class for binary operators: binary(output, input0, input1). |
|
Base class for reduction operators: reduce(output, input, reduce_axes, accum). |
|
Zero out all elements in src and store to dst. |
|
Compute sqrt(src) and store to dst. |
Compute sqrt(src * scale + bias) and store to dst. |
|
|
Fill dst with a scalar value. |
|
Add src1 and src2 element-wise and store to dst. |
|
Subtract src2 from src1 element-wise and store to dst. |
|
Multiply src1 and src2 element-wise and store to dst. |
|
Divide src1 by src2 element-wise using floating point division and store to dst. |
|
Fused multiply-add: output = input * scale + bias. |
|
Cast src to dst. |
|
Copy all elements from src to dst. |
Copy all elements from src to dst asynchronously. |
|
|
General matrix multiplication: D = A * B * alpha + C * beta. |
General matrix multiplication asynchronously. |
|
|
Sum elements in src along specified axes and store in dst. |
|
Compute maximum value in src along specified axes and store in dst. |
|
Compute minimum value in src along specified axes and store in dst. |
Compute reciprocal (1/x) for all elements in src and store to dst. |
|
|
Compute SiLU (x * sigmoid(x)) for all elements in src and store to dst. |
|
Set all elements in dst to a specified value. |
|
Compute element-wise maximum of src1 and src2 and store to dst. |
|
Compute element-wise minimum of src1 and src2 and store to dst. |
|
Compute exp(src) and store to dst. |
Compute exp(src * scale + bias) and store to dst. |
|
|
Compute exp2(src) and store to dst. |
Compute exp2(src * scale + bias) and store to dst. |
|
|
Compute log2(src) and store to dst. |
Compute log2(src * scale + bias) and store to dst. |
|
|
Select elements from src1 or src2 based on the predicate. |
Combine a binary operation with a reduction operation. |
|
Combine a unary operation with a reduction operation. |
|
Write unary(input * scale + bias), then reduce the unary output. |
|
Chain multiple binary operations together. |
|
Negate the result of a reduction operation. |