Added groundwork for col2im algorithm plus first non-working version of kernel and test

2 files changed, 6 insertions, 3 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.", []),
diff --git a/scripts/generator/generator/routine.py b/scripts/generator/generator/routine.py
index 7321349d..3b5a6b76 100644
--- a/scripts/generator/generator/routine.py
+++ b/scripts/generator/generator/routine.py
@@ -205,7 +205,7 @@ class Routine:
 
     def no_scalars(self):
         """Determines whether or not this routine has scalar arguments (alpha/beta)"""
-        return self.scalars == [] or self.name in ["im2col", "convgemm"]
+        return self.scalars == [] or self.name in ["im2col", "col2im", "convgemm"]
 
     def has_layout(self):
         """Determines whether the layout is an argument"""
@@ -226,12 +226,14 @@ class Routine:
         """Determines which buffers go first (between alpha and beta) and which ones go after"""
         if self.level == "2b" or self.name == "had":
             return ["x", "y"]
-        return ["ap", "a", "b", "x", "im", "kernel"]
+        extra_buffer = "col" if self.name == "col2im" else "im"
+        return ["ap", "a", "b", "x", extra_buffer, "kernel"]
 
     def buffers_second(self):
         if self.level == "2b" or self.name == "had":
             return ["z", "ap", "a", "b", "c"]
-        return ["y", "c", "col", "result"]
+        extra_buffer = "im" if self.name == "col2im" else "col"
+        return ["y", "c", extra_buffer, "result"]
 
     def buffer(self, name):
         """Retrieves a variable name for a specific input/output vector/matrix (e.g. 'x')"""