[UM] Added templated umalloc

examples/c/1_add_vectors/main.c

@@ -86,7 +86,7 @@ int main(int argc, const char **argv) {
   occaCopyMemToPtr(ab, o_ab, occaAllBytes, 0, occaDefault);
 
   // Assert values
-  for (i = 0; i < 5; ++i) {
+  for (i = 0; i < entries; ++i) {
     printf("%d = %f\n", i, ab[i]);
   }
   for (i = 0; i < entries; ++i) {

examples/c/2_background_device/main.c

@@ -57,9 +57,9 @@ int main(int argc, const char **argv) {
   //   making it safe to use them again
   occaFinish();
 
-  for (i = 0; i < 5; ++i)
+  for (i = 0; i < entries; ++i) {
     printf("%d = %f\n", i, ab[i]);
-
+  }
   for (i = 0; i < entries; ++i) {
     if (ab[i] != (a[i] + b[i])) {
       exit(1);

examples/cpp/01_add_vectors/main.cpp

@@ -64,7 +64,7 @@ int main(int argc, const char **argv) {
   o_ab.copyTo(ab);
 
   // Assert values
-  for (int i = 0; i < 5; ++i) {
+  for (int i = 0; i < entries; ++i) {
     std::cout << i << ": " << ab[i] << '\n';
   }
   for (int i = 0; i < entries; ++i) {

examples/cpp/02_background_device/main.cpp

@@ -21,9 +21,9 @@ int main(int argc, const char **argv) {
 
   int entries = 5;
 
-  float *a  = (float*) occa::umalloc(entries, occa::dtype::float_);
-  float *b  = (float*) occa::umalloc(entries, occa::dtype::float_);
-  float *ab = (float*) occa::umalloc(entries, occa::dtype::float_);
+  float *a  = occa::umalloc<float>(entries);
+  float *b  = occa::umalloc<float>(entries);
+  float *ab = occa::umalloc<float>(entries);
 
   for (int i = 0; i < entries; ++i) {
     a[i]  = i;
@@ -47,9 +47,9 @@ int main(int argc, const char **argv) {
   //   making it safe to use them again
   occa::finish();
 
-  for (int i = 0; i < 5; ++i)
+  for (int i = 0; i < entries; ++i) {
     std::cout << i << ": " << ab[i] << '\n';
-
+  }
   for (int i = 0; i < entries; ++i) {
     if (ab[i] != (a[i] + b[i]))
       throw 1;

examples/cpp/03_inline_kernels/main.cpp

@@ -11,20 +11,16 @@ int main(int argc, const char **argv) {
 
   int entries = 5;
 
-  float *a  = new float[entries];
-  float *b  = new float[entries];
-  float *ab = new float[entries];
+  float *a  = occa::umalloc<float>(entries);
+  float *b  = occa::umalloc<float>(entries);
+  float *ab = occa::umalloc<float>(entries);
 
   for (int i = 0; i < entries; ++i) {
     a[i]  = i;
     b[i]  = 1 - i;
     ab[i] = 0;
   }
 
-  occa::memory o_a  = occa::malloc(entries, occa::dtype::float_, a);
-  occa::memory o_b  = occa::malloc(entries, occa::dtype::float_, b);
-  occa::memory o_ab = occa::malloc(entries, occa::dtype::float_);
-
   // Props are
   occa::properties props;
   props["defines/TILE_SIZE"] = 16;
@@ -49,35 +45,32 @@ int main(int argc, const char **argv) {
   //       - Resolved once 'auto' is supported. Function arguments of
   //         type 'auto' will act as templated typed variables
   //
-  //   - Cannot use unified memory from occa::umalloc
-  //       - dtype::get<> needs to check a pointer type
-  //
   //   ~ Cannot use external functions
   //       - Potentially can happen with another macro OCCA_INLINED_FUNCTION
   OCCA_INLINED_KERNEL(
-    (entries, o_a, o_b, o_ab),
+    (entries, a, b, ab),
     props,
     (
       for (int i = 0; i < entries; ++i; @tile(TILE_SIZE, @outer, @inner)) {
-        o_ab[i] = o_a[i] + o_b[i];
+        ab[i] = a[i] + b[i];
       }
     )
   );
 
   // Copy result to the host
-  o_ab.copyTo(ab);
+  occa::finish();
 
-  for (int i = 0; i < 5; ++i)
+  for (int i = 0; i < entries; ++i) {
     std::cout << i << ": " << ab[i] << '\n';
-
+  }
   for (int i = 0; i < entries; ++i) {
     if (ab[i] != (a[i] + b[i]))
       throw 1;
   }
 
-  delete [] a;
-  delete [] b;
-  delete [] ab;
+  occa::free(a);
+  occa::free(b);
+  occa::free(ab);
 
   return 0;
 }

examples/cpp/05_building_kernels/main.cpp

@@ -88,9 +88,9 @@ int main(int argc, const char **argv) {
 
   o_vec2.copyTo(vec2);
 
-  for (int i = 0; i < 5; ++i)
+  for (int i = 0; i < entries; ++i) {
     std::cout << vec[i] << "^2 = " << vec2[i] << '\n';
-
+  }
   for (int i = 0; i < entries; ++i) {
     if (vec2[i] != (vec[i] * vec[i]))
       throw 1;

examples/cpp/06_unified_memory/main.cpp

@@ -41,7 +41,7 @@ int main(int argc, const char **argv) {
   //   making it safe to use them again
   occa::finish();
 
-  for (int i = 0; i < 5; ++i) {
+  for (int i = 0; i < entries; ++i) {
     std::cout << i << ": " << ab[i] << '\n';
   }
   for (int i = 0; i < entries; ++i) {

examples/cpp/interop_examples/cuda/main.cpp

@@ -58,7 +58,7 @@ int main(int argc, char **argv) {
 
   o_ab.copyTo(ab);
 
-  for (int i = 0; i < 5; ++i) {
+  for (int i = 0; i < entries; ++i) {
     std::cout << i << ": " << ab[i] << '\n';
   }
 

examples/cpp/interop_examples/native/addVectors/main.cpp

@@ -53,7 +53,7 @@ int main(int argc, char **argv) {
 
   o_ab.copyTo(ab);
 
-  for (int i = 0; i < 5; ++i) {
+  for (int i = 0; i < entries; ++i) {
     std::cout << i << ": " << ab[i] << '\n';
   }
   for (int i = 0; i < entries; ++i) {

examples/cpp/interop_examples/opencl/main.cpp

@@ -60,7 +60,7 @@ int main(int argc, char **argv) {
 
   o_ab.copyTo(ab);
 
-  for (int i = 0; i < 5; ++i) {
+  for (int i = 0; i < entries; ++i) {
     std::cout << i << ": " << ab[i] << '\n';
   }
 

examples/cpp/interop_examples/openmp/main.cpp

@@ -35,7 +35,7 @@ int main(int argc, char **argv) {
 
   o_ab.copyTo(ab);
 
-  for (int i = 0; i < 5; ++i) {
+  for (int i = 0; i < entries; ++i) {
     std::cout << i << ": " << ab[i] << '\n';
   }
 

include/occa/core/base.hpp

@@ -74,9 +74,12 @@ namespace occa {
                 const void *src = NULL,
                 const occa::properties &props = occa::properties());
 
-  void* umalloc(const dim_t bytes,
-                const void *src = NULL,
-                const occa::properties &props = occa::properties());
+  template <class TM>
+  TM* umalloc(const dim_t bytes,
+              const void *src = NULL,
+              const occa::properties &props = occa::properties()) {
+    return (TM*) umalloc(bytes, dtype::get<TM>(), src, props);
+  }
 
   void memcpy(void *dest, const void *src,
               const dim_t bytes,

include/occa/core/device.hpp

@@ -4,17 +4,17 @@
 #include <iostream>
 #include <sstream>
 
-#include <occa/defines.hpp>
 #include <occa/core/kernel.hpp>
+#include <occa/core/memory.hpp>
 #include <occa/core/stream.hpp>
+#include <occa/defines.hpp>
 #include <occa/dtype.hpp>
 #include <occa/io/output.hpp>
 #include <occa/tools/gc.hpp>
 #include <occa/tools/uva.hpp>
 
 namespace occa {
   class modeKernel_t; class kernel;
-  class modeMemory_t; class memory;
   class modeDevice_t; class device;
   class modeStreamTag_t; class streamTag;
   class deviceInfo;
@@ -270,16 +270,25 @@ namespace occa {
                   const dtype_t &dtype,
                   const occa::properties &props);
 
-    void* umalloc(const dim_t bytes,
-                  const void *src = NULL,
-                  const occa::properties &props = occa::properties());
+    template <class TM>
+    TM* umalloc(const dim_t entries,
+                const void *src = NULL,
+                const occa::properties &props = occa::properties()) {
+      return (TM*) umalloc(entries, dtype::get<TM>(), src, props);
+    }
 
-    void* umalloc(const dim_t bytes,
-                  const occa::memory src,
-                  const occa::properties &props = occa::properties());
+    template <class TM>
+    TM* umalloc(const dim_t entries,
+                const occa::memory src,
+                const occa::properties &props = occa::properties()) {
+      return (TM*) umalloc(entries, dtype::get<TM>(), src, props);
+    }
 
-    void* umalloc(const dim_t bytes,
-                  const occa::properties &props);
+    template <class TM>
+    TM* umalloc(const dim_t entries,
+                const occa::properties &props) {
+      return (TM*) umalloc(entries, dtype::get<TM>(), props);
+    }
     //  |===============================
   };
 

src/c/base.cpp

@@ -182,11 +182,11 @@ void* OCCA_RFUNC occaUMalloc(const occaUDim_t bytes,
                              occaProperties props) {
 
   if (occa::c::isDefault(props)) {
-    return occa::umalloc(bytes, src);
+    return occa::umalloc<void>(bytes, src);
   }
-  return occa::umalloc(bytes,
-                       src,
-                       occa::c::properties(props));
+  return occa::umalloc<void>(bytes,
+                             src,
+                             occa::c::properties(props));
 }
 
 void* OCCA_RFUNC occaTypedUMalloc(const occaUDim_t entries,

src/c/device.cpp

@@ -226,14 +226,13 @@ void* OCCA_RFUNC occaDeviceUMalloc(occaDevice device,
   occa::device device_ = occa::c::device(device);
 
   if (occa::c::isDefault(props)) {
-    return device_.umalloc(bytes, src);
+    return device_.umalloc<void>(bytes, src);
   }
-  return device_.umalloc(bytes,
-                         src,
-                         occa::c::properties(props));
+  return device_.umalloc<void>(bytes,
+                               src,
+                               occa::c::properties(props));
 }
 
-
 void* OCCA_RFUNC occaDeviceTypedUMalloc(occaDevice device,
                                         const occaUDim_t entries,
                                         const occaDtype dtype,

src/core/base.cpp

@@ -115,12 +115,6 @@ namespace occa {
     return getDevice().umalloc(entries, dtype, src, props);
   }
 
-  void* umalloc(const dim_t bytes,
-                const void *src,
-                const occa::properties &props) {
-    return getDevice().umalloc(bytes, src, props);
-  }
-
   void memcpy(void *dest, const void *src,
               const dim_t bytes,
               const occa::properties &props) {

src/core/device.cpp

@@ -627,23 +627,6 @@ namespace occa {
                         const occa::properties &props) {
     return umalloc(entries, dtype, NULL, props);
   }
-
-  void* device::umalloc(const dim_t bytes,
-                        const void *src,
-                        const occa::properties &props) {
-    return umalloc(bytes, dtype::byte, src, props);
-  }
-
-  void* device::umalloc(const dim_t bytes,
-                        const occa::memory src,
-                        const occa::properties &props) {
-    return umalloc(bytes, dtype::byte, src, props);
-  }
-
-  void* device::umalloc(const dim_t bytes,
-                        const occa::properties &props) {
-    return umalloc(bytes, dtype::byte, NULL, props);
-  }
   //  |=================================
 
   template <>