diff options
| author | Cedric Nugteren <web@cedricnugteren.nl> | 2018-10-23 20:52:25 +0200 |
|---|---|---|
| committer | Cedric Nugteren <web@cedricnugteren.nl> | 2018-10-23 20:52:25 +0200 |
| commit | d45911b61dedafcbd74f65df263b4197697d6a81 (patch) | |
| tree | 1f5d30db2f1b77a8258c5d0b331581938574d17e /scripts/generator/generator.py | |
| parent | 44b630fc222c6e22446c20995411994b51bc2f21 (diff) | |
Added groundwork for col2im algorithm plus first non-working version of kernel and test
Diffstat (limited to 'scripts/generator/generator.py')
| -rwxr-xr-x | scripts/generator/generator.py | 1 |
1 files changed, 1 insertions, 0 deletions
diff --git a/scripts/generator/generator.py b/scripts/generator/generator.py index c2637037..27107739 100755 --- a/scripts/generator/generator.py +++ b/scripts/generator/generator.py @@ -181,6 +181,7 @@ ROUTINES = [ Routine(True, True, 0, False, "x", "had", T, [S,D,C,Z,H], ["n"], [], ["x","y"], ["z"], [xn,yn,zn], ["alpha","beta"], "", "Element-wise vector product (Hadamard)", "Performs the Hadamard element-wise product _z = alpha * x * y + beta * z_, in which _x_, _y_, and _z_ are vectors and _alpha_ and _beta_ are scalar constants.", []), Routine(True, True, 0, False, "x", "omatcopy", T, [S,D,C,Z,H], ["m","n"], ["layout","a_transpose"], ["a"], ["b"], [amn,bnma], ["alpha"], "", "Scaling and out-place transpose/copy (non-BLAS function)", "Performs scaling and out-of-place transposition/copying of matrices according to _B = alpha*op(A)_, in which _A_ is an input matrix (_m_ rows by _n_ columns), _B_ an output matrix, and _alpha_ a scalar value. The operation _op_ can be a normal matrix copy, a transposition or a conjugate transposition.", [ald_m, bld_n]), Routine(True, True, 0, False, "x", "im2col", T, [S,D,C,Z,H], im2col_constants, [], ["im"], ["col"], [im,col], [""], "", "Im2col function (non-BLAS function)", "Performs the im2col algorithm, in which _im_ is the input matrix and _col_ is the output matrix.", []), + Routine(True, True, 0, False, "x", "col2im", T, [S,D,C,Z,H], im2col_constants, [], ["col"], ["im"], [col,im], [""], "", "Col2im function (non-BLAS function)", "Performs the col2im algorithm, in which _col_ is the input matrix and _im_ is the output matrix.", []), Routine(True, True, 0, False, "x", "convgemm", T, [S,D,H], convgemm_constants, [], ["im","kernel"], ["result"], [imb,kernel,result],[""], "", "Batched convolution as GEMM (non-BLAS function)", "Integrates im2col and GEMM for batched 3D convolution, in which _im_ is the 4D input tensor (NCHW - batch-channelin-height-width), _kernel_ the 4D kernel weights tensor (KCHW - channelout-channelin-height-width), and _result_ the 4D output tensor (NCHW - batch-channelout-height-width).", []), # Batched routines: Routine(True, True, 1, False, "x", "axpy", T, [S,D,C,Z,H], ["n"], [], ["x"], ["y"], [xn,yn], ["alpha"], "", "Batched version of AXPY", "As AXPY, but multiple operations are batched together for better performance.", []), |