Update to Pytorch 2.3.0

pytorch/src/gen/java/org/bytedeco/pytorch/Half.java

@@ -46,8 +46,9 @@ public class Half extends Pointer {
 // #if defined(__aarch64__) && !defined(C10_MOBILE) && !defined(__CUDACC__)
 // #else
   public Half(float value) { super((Pointer)null); allocate(value); }
-  private native void allocate(float value);
-  public native @Name("operator float") float asFloat();
+  @Namespace @Name("javacpp::allocate_Half") private native void allocate(float value);
+  public float asFloat() { return _asFloat(this); }
+  private static native @Namespace @Name("javacpp::cast_Half_to_float") float _asFloat(Half h);
 // #endif
 
 // #if defined(__CUDACC__) || defined(__HIPCC__)

pytorch/src/gen/java/org/bytedeco/pytorch/global/torch.java

@@ -79474,9 +79474,8 @@ scalar_t sf(scalar_t x, scalar_t y)
     @ByVal(nullValue = "torch::jit::TypeResolver(nullptr)") TypeResolver type_resolver,
     @ByVal(nullValue = "c10::ArrayRef<at::Tensor>{}") TensorArrayRef tensor_table,
     TypeParser type_parser/*=torch::jit::Unpickler::defaultTypeParser*/);
-// Targeting ../VectorReader.java
-
 
+// #ifndef C10_MOBILE
 // #endif
  // namespace jit
  // namespace torch

pytorch/src/main/java/org/bytedeco/pytorch/presets/torch.java

@@ -71,7 +71,8 @@
                 "torch/csrc/jit/serialization/storage_context.h",
 
                 "datasets.h",
-                "pytorch_adapters.h"
+                "pytorch_adapters.h",
+                "platform_unification.h"
             },
             exclude = {"openblas_config.h", "cblas.h", "lapacke_config.h", "lapacke_mangling.h", "lapack.h", "lapacke.h", "lapacke_utils.h"},
             link = {"c10", "torch_cpu", "torch"},
@@ -354,12 +355,12 @@ public void map(InfoMap infoMap) {
             //.put(new Info("c10::DataPtr&&", "at::DataPtr&&").valueTypes("@Cast({\"\", \"c10::DataPtr&&\"}) @StdMove DataPtr").pointerTypes("DataPtr")) // DataPtr::operator= deleted
             .put(new Info("c10::DataPtr", "at::DataPtr").valueTypes("@StdMove DataPtr").pointerTypes("DataPtr"))
             .put(new Info("c10::StorageImpl::UniqueStorageShareExternalPointer(at::DataPtr&&, size_t)",
-                          "c10::Storage::UniqueStorageShareExternalPointer(at::DataPtr&&, size_t)").javaText(
+                "c10::Storage::UniqueStorageShareExternalPointer(at::DataPtr&&, size_t)").javaText(
                 "public native void UniqueStorageShareExternalPointer(@Cast({\"\", \"c10::DataPtr&&\"}) @StdMove DataPtr data_ptr,  @Cast(\"size_t\") long size_bytes);"
-                ))
+            ))
             .put(new Info("c10::GetStorageImplCreate", "c10::SetStorageImplCreate",
                 "c10::intrusive_ptr<c10::StorageImpl> (*)(c10::StorageImpl::use_byte_size_t, c10::SymInt, c10::DataPtr, c10::Allocator*, bool)").skip())
-;
+        ;
         //// Enumerations
         infoMap
             .put(new Info("c10::ScalarType", "at::ScalarType", "torch::Dtype").enumerate().valueTypes("ScalarType").pointerTypes("@Cast(\"c10::ScalarType*\") BytePointer"))
@@ -1113,21 +1114,21 @@ public void map(InfoMap infoMap) {
                 "c10::complex<c10::Half>::operator -=(c10::Half)",
                 "c10::complex<c10::Half>::operator *=(c10::Half)",
                 "c10::complex<c10::Half>::operator /=(c10::Half)"
-                ).skip())
+            ).skip())
             .put(new Info("c10::complex<c10::Half>::complex(const c10::Half&, const c10::Half&)").javaText( // Second argument not optional + add specific functions
-                "public HalfComplex(Half re, Half im) { super((Pointer)null); allocate(re, im); }\n" +
-                "private native void allocate(@Const @ByRef Half re, @Const @ByRef(nullValue = \"c10::Half()\") Half im);\n" +
-                "public HalfComplex(@Const @ByRef FloatComplex value) { super((Pointer)null); allocate(value); }\n" +
-                "private native void allocate(@Const @ByRef FloatComplex value);\n" +
-                "\n" +
-                "// Conversion operator\n" +
-                "public native @ByVal @Name(\"operator c10::complex<float>\") FloatComplex asFloatComplex();\n" +
-                "\n" +
-                "public native @ByRef @Name(\"operator +=\") HalfComplex addPut(@Const @ByRef HalfComplex other);\n" +
-                "\n" +
-                "public native @ByRef @Name(\"operator -=\") HalfComplex subtractPut(@Const @ByRef HalfComplex other);\n" +
-                "\n" +
-                "public native @ByRef @Name(\"operator *=\") HalfComplex multiplyPut(@Const @ByRef HalfComplex other);"
+                    "public HalfComplex(Half re, Half im) { super((Pointer)null); allocate(re, im); }\n" +
+                    "private native void allocate(@Const @ByRef Half re, @Const @ByRef(nullValue = \"c10::Half()\") Half im);\n" +
+                    "public HalfComplex(@Const @ByRef FloatComplex value) { super((Pointer)null); allocate(value); }\n" +
+                    "private native void allocate(@Const @ByRef FloatComplex value);\n" +
+                    "\n" +
+                    "// Conversion operator\n" +
+                    "public native @ByVal @Name(\"operator c10::complex<float>\") FloatComplex asFloatComplex();\n" +
+                    "\n" +
+                    "public native @ByRef @Name(\"operator +=\") HalfComplex addPut(@Const @ByRef HalfComplex other);\n" +
+                    "\n" +
+                    "public native @ByRef @Name(\"operator -=\") HalfComplex subtractPut(@Const @ByRef HalfComplex other);\n" +
+                    "\n" +
+                    "public native @ByRef @Name(\"operator *=\") HalfComplex multiplyPut(@Const @ByRef HalfComplex other);"
                 )
             )
         ;
@@ -1202,7 +1203,7 @@ public void map(InfoMap infoMap) {
             .put(new Info("torch::jit::Wrap<torch::jit::Block>").pointerTypes("BlockWrap"))
             .put(new Info("torch::jit::Wrap<torch::jit::Node>").pointerTypes("JitNodeWrap"))
             .put(new Info("torch::jit::Wrap<torch::jit::Value>").pointerTypes("ValueWrap"))
-            ;
+        ;
 
 
         //// Data loader
@@ -1786,7 +1787,7 @@ public void map(InfoMap infoMap) {
                 .put(new Info("torch::optim::" + opt + "Options", "torch::optim::" + opt + "ParamState")) // Help qualification
                 .put(new Info("torch::optim::OptimizerCloneableOptions<torch::optim::" + opt + "Options>").pointerTypes("OptimizerCloneable" + opt + "Options"))
                 .put(new Info("torch::optim::OptimizerCloneableParamState<torch::optim::" + opt + "ParamState>").pointerTypes("OptimizerCloneable" + opt + "ParamState"))
-                ;
+            ;
             new PointerInfo("torch::optim::" + opt + "Options").makeUnique(infoMap);
             new PointerInfo("torch::optim::OptimizerCloneableParamState<torch::optim::" + opt + "ParamState>").javaBaseName("OptimizerCloneable" + opt + "AdagradParamState").makeUnique(infoMap);
             new PointerInfo("torch::optim::OptimizerCloneableOptions<torch::optim::" + opt + "Options>").javaBaseName("OptimizerCloneable" + opt + "Options").makeUnique(infoMap);
@@ -2342,6 +2343,7 @@ We need either to put an annotation info on each member, or javaName("@NoOffset
             "torch::jit::Suspend",
             "torch::jit::TokenTrie",
             "torch::jit::TaggedRange",
+            "torch::jit::VectorReader",
             "torch::jit::WithCurrentScope",
             "torch::jit::WithInsertPoint",
             "torch::jit::variable_tensor_list",
@@ -2432,8 +2434,7 @@ We need either to put an annotation info on each member, or javaName("@NoOffset
                    "std::enable_shared_from_this<torch::jit::tracer::TracingState>", "std::enable_shared_from_this<TracingState>",
                    "std::enable_shared_from_this<torch::nn::Module>", "std::enable_shared_from_this<Module>"
                ).pointerTypes("Pointer").cast())
-            .put(new Info("MTLCommandBuffer_t", "DispatchQueue_t").valueTypes("Pointer").pointerTypes("PointerPointer").skip());
-
+               .put(new Info("MTLCommandBuffer_t", "DispatchQueue_t").valueTypes("Pointer").pointerTypes("PointerPointer").skip());
 
 
         ///// Special cases needing javaText
@@ -2539,6 +2540,14 @@ We need either to put an annotation info on each member, or javaName("@NoOffset
 
         infoMap.put(new Info("c10::VaryingShape<c10::Stride>::merge").skip()); // https://github.com/pytorch/pytorch/issues/123248, waiting for the fix in 2.3.1 or 2.4
 
+        //// Different C++ API between platforms
+        infoMap
+            .put(new Info("c10::Half::Half(float)").annotations("@Namespace @Name(\"javacpp::allocate_Half\")"))
+            .put(new Info("c10::Half::operator float()").javaText(
+                "public float asFloat() { return _asFloat(this); }\n" +
+                "private static native @Namespace @Name(\"javacpp::cast_Half_to_float\") float _asFloat(Half h);"
+            ))
+        ;
     }
 
     private static String template(String t, String... args) {

pytorch/src/main/resources/org/bytedeco/pytorch/include/platform_unification.h

@@ -0,0 +1,21 @@
+/* Add the necessary C++ calls to present a unified C++ API whatever the platform */
+
+namespace javacpp {
+
+   inline c10::Half *allocate_Half(jfloat value) {
+#if defined(__aarch64__) && !defined(C10_MOBILE) && !defined(__CUDACC__)
+       return new c10::Half((float16_t) value);
+#else
+       return new c10::Half((float) value);
+#endif
+   }
+
+   inline float cast_Half_to_float(c10::Half *h) {
+#if defined(__aarch64__) && !defined(C10_MOBILE) && !defined(__CUDACC__)
+       return (float) (float16_t) *h;
+#else
+       return (float) *h;
+#endif
+   }
+
+}