提交 d79d38c1 authored 作者: Arnaud Bergeron's avatar Arnaud Bergeron

Fix clients of ceil_intdiv.

上级 33d29806
...@@ -613,6 +613,15 @@ class GpuCAReduceCuda(GpuKernelBase, HideC, CAReduceDtype): ...@@ -613,6 +613,15 @@ class GpuCAReduceCuda(GpuKernelBase, HideC, CAReduceDtype):
def c_headers(self): def c_headers(self):
return ['<numpy_compat.h>', '<gpuarray/types.h>'] return ['<numpy_compat.h>', '<gpuarray/types.h>']
def c_support_code(self):
return """
template <typename T>
static T ceil_intdiv(T a, T b)
{
return (a/b) + ((a % b) ? 1: 0);
}
"""
def c_code(self, node, name, inp, out, sub): def c_code(self, node, name, inp, out, sub):
x, = inp x, = inp
z, = out z, = out
......
...@@ -242,6 +242,15 @@ class GpuImages2Neibs(GpuKernelBase, Images2Neibs, Op): ...@@ -242,6 +242,15 @@ class GpuImages2Neibs(GpuKernelBase, Images2Neibs, Op):
flags=flags, objvar=k_var)) flags=flags, objvar=k_var))
return kernels return kernels
def c_support_code(self):
return """
template <typename T>
static T ceil_intdiv(T a, T b)
{
return (a/b) + ((a % b) ? 1: 0);
}
"""
def c_code(self, node, name, inp, out, sub): def c_code(self, node, name, inp, out, sub):
dtype_ten4 = node.inputs[0].dtype dtype_ten4 = node.inputs[0].dtype
dtype_neib_shape = node.inputs[1].dtype dtype_neib_shape = node.inputs[1].dtype
......
Markdown 格式
0%
您添加了 0 到此讨论。请谨慎行事。
请先完成此评论的编辑!
注册 或者 后发表评论