Kernels for max min mean median #102
Conversation
Codecov Report
@@ Coverage Diff @@
## master #102 +/- ##
=======================================
Coverage 51.45% 51.45%
=======================================
Files 19 19
Lines 1201 1201
Branches 312 312
=======================================
Hits 618 618
Misses 333 333
Partials 250 250 Continue to review full report at Codecov.
|
dpnp/backend_statistics.pyx
Outdated
|
|
||
|
|
||
| # C function pointer to the C library template functions | ||
| ctypedef void(*custom_statistic_1in_1out_func_ptr_t)(void * , void * , size_t) |
There was a problem hiding this comment.
You could use existing fptr_1in_1out_t from backend instead of custom_statistic_1in_1out_func_ptr_t.
dpnp/backend_statistics.pyx
Outdated
| @@ -87,7 +91,26 @@ cpdef dparray dpnp_cov(dparray array1): | |||
| return result | |||
|
|
|||
|
|
|||
| cpdef dparray _dpnp_max(dparray input): | |||
| cdef DPNPFuncType param1_type = dpnp_dtype_to_DPNPFuncType(input.dtype) | |||
| cdef DPNPFuncType output_type = dpnp_dtype_to_DPNPFuncType(input.dtype) | |||
There was a problem hiding this comment.
output_type is equal to param1_type, so you could just use param1_type instead of output_type. I'm not sure about cython but maybe it will work:
cdef DPNPFuncType param1_type = output_type = dpnp_dtype_to_DPNPFuncType(input.dtype)
or you could use DPNP_FT_NONE as the third argument of get_dpnp_function_ptr.
There was a problem hiding this comment.
I thought that the third parameter of get_dpnp_function_ptr isn't output type, but it's type of the second input parameter. Please correct me if I'm wrong.
dpnp/backend_statistics.pyx
Outdated
| @@ -166,7 +189,26 @@ cpdef dparray dpnp_max(dparray input, axis): | |||
| return dpnp_result_array | |||
|
|
|||
|
|
|||
| cpdef dparray _dpnp_mean(dparray input): | |||
| cdef DPNPFuncType param1_type = dpnp_dtype_to_DPNPFuncType(input.dtype) | |||
| cdef DPNPFuncType output_type = dpnp_dtype_to_DPNPFuncType(numpy.float64 if input.dtype != numpy.float32 else numpy.float32) | |||
There was a problem hiding this comment.
Isn't it resolved by fmap[DPNPFuncName::DPNP_FN_MEAN][eft_FLT][eft_FLT] = {eft_FLT, (void*)custom_mean_c<float, float>};?
| _DataType* sum = reinterpret_cast<_DataType*>(dpnp_memory_alloc_c(1 * sizeof(_DataType))); | ||
|
|
||
| custom_sum_c<_DataType>(array1_in, sum, size); |
There was a problem hiding this comment.
| _DataType* sum = reinterpret_cast<_DataType*>(dpnp_memory_alloc_c(1 * sizeof(_DataType))); | |
| custom_sum_c<_DataType>(array1_in, sum, size); | |
| _DataType sum; | |
| custom_sum_c<_DataType>(array1_in, &sum, size); |
There was a problem hiding this comment.
oh... no. we need USM memory for kernel. sorry - my proposal is wrong.
perhaps it is better to use *result memory for this
No description provided.