Fix/public internal header (apache#12374)

* move random-gen header to internal code

* uncomment header

* remove public method from cc

* move cuda randgen logic to cc

* remove private files

* include correct header in cu random_gen

* fix include in cu random-gen

* fix template

* remove static kw

* add missing template expression

* change dtype to float

* remove old header

* fix lint

* fix imports

* fix space

* fix template

* fix template

* add todo

* fix lint

src/common/random_generator.h → include/mxnet/random_generator.h

@@ -22,16 +22,15 @@
  * \file random_generator.h
  * \brief Parallel random number generator.
  */
-#ifndef MXNET_COMMON_RANDOM_GENERATOR_H_
-#define MXNET_COMMON_RANDOM_GENERATOR_H_
+#ifndef MXNET_RANDOM_GENERATOR_H_
+#define MXNET_RANDOM_GENERATOR_H_
 
-#include <mxnet/base.h>
 #include <random>
 #include <new>
+#include "./base.h"
 
 #if MXNET_USE_CUDA
 #include <curand_kernel.h>
-#include "../common/cuda_utils.h"
 #endif  // MXNET_USE_CUDA
 
 namespace mxnet {
@@ -50,6 +49,7 @@ class RandGenerator<cpu, DType> {
   static const int kNumRandomStates;
 
   // implementation class for random number generator
+  // TODO(alexzai): move impl class to separate file - tracked in MXNET-948
   class Impl {
    public:
     typedef typename std::conditional<std::is_floating_point<DType>::value,
@@ -116,6 +116,7 @@ class RandGenerator<gpu, DType> {
   // by using 1.0-curand_uniform().
   // Needed as some samplers in sampler.h won't be able to deal with
   // one of the boundary cases.
+  // TODO(alexzai): move impl class to separate file - tracked in MXNET-948
   class Impl {
    public:
     Impl &operator=(const Impl &) = delete;
@@ -150,14 +151,9 @@ class RandGenerator<gpu, DType> {
     curandStatePhilox4_32_10_t state_;
   };  // class RandGenerator<gpu, DType>::Impl
 
-  static void AllocState(RandGenerator<gpu, DType> *inst) {
-    CUDA_CALL(cudaMalloc(&inst->states_,
-                         kNumRandomStates * sizeof(curandStatePhilox4_32_10_t)));
-  }
+  static void AllocState(RandGenerator<gpu, DType> *inst);
 
-  static void FreeState(RandGenerator<gpu, DType> *inst) {
-    CUDA_CALL(cudaFree(inst->states_));
-  }
+  static void FreeState(RandGenerator<gpu, DType> *inst);
 
   void Seed(mshadow::Stream<gpu> *s, uint32_t seed);
 
@@ -172,6 +168,7 @@ class RandGenerator<gpu, double> {
   // by using 1.0-curand_uniform().
   // Needed as some samplers in sampler.h won't be able to deal with
   // one of the boundary cases.
+  // TODO(alexzai): move impl class to separate file - tracked in MXNET-948
   class Impl {
    public:
     Impl &operator=(const Impl &) = delete;
@@ -215,4 +212,4 @@ class RandGenerator<gpu, double> {
 }  // namespace random
 }  // namespace common
 }  // namespace mxnet
-#endif  // MXNET_COMMON_RANDOM_GENERATOR_H_
+#endif  // MXNET_RANDOM_GENERATOR_H_

include/mxnet/resource.h

@@ -28,7 +28,7 @@
 #include <dmlc/logging.h>
 #include "./base.h"
 #include "./engine.h"
-#include "../../src/common/random_generator.h"
+#include "./random_generator.h"
 
 namespace mxnet {
 

src/common/random_generator.cu

@@ -23,8 +23,8 @@
  * \brief gpu implements for parallel random number generator.
  */
 
+#include <mxnet/random_generator.h>
 #include <algorithm>
-#include "./random_generator.h"
 #include "../operator/mxnet_op.h"
 
 namespace mxnet {
@@ -59,6 +59,17 @@ void RandGenerator<gpu, float>::Seed(mshadow::Stream<gpu> *s, uint32_t seed) {
   s->Wait();
 }
 
+template<>
+void RandGenerator<gpu, float>::AllocState(RandGenerator<gpu> *inst) {
+  CUDA_CALL(cudaMalloc(&inst->states_,
+                       kNumRandomStates * sizeof(curandStatePhilox4_32_10_t)));
+}
+
+template<>
+void RandGenerator<gpu, float>::FreeState(RandGenerator<gpu> *inst) {
+  CUDA_CALL(cudaFree(inst->states_));
+}
+
 }  // namespace random
 }  // namespace common
 }  // namespace mxnet

src/operator/leaky_relu-inl.h

@@ -28,13 +28,13 @@
 
 #include <dmlc/logging.h>
 #include <dmlc/parameter.h>
+#include <mxnet/random_generator.h>
 #include <mxnet/operator.h>
 #include <cstring>
 #include <map>
 #include <string>
 #include <vector>
 #include <utility>
-#include "../common/random_generator.h"
 #include "./operator_common.h"
 #include "./mshadow_op.h"
 #include "./random/sampler.h"

src/resource.cc

@@ -27,12 +27,12 @@
 #include <dmlc/thread_local.h>
 #include <mxnet/base.h>
 #include <mxnet/engine.h>
+#include <mxnet/random_generator.h>
 #include <mxnet/resource.h>
 #include <mxnet/storage.h>
 #include <limits>
 #include <atomic>
 #include "./common/lazy_alloc_array.h"
-#include "./common/random_generator.h"
 #include "./common/utils.h"
 
 namespace mxnet {