react-native-executorch 0.4.9 → 0.5.0-nightly-6f4dd53-20251211

package/third-party/include/executorch/runtime/core/portable_type/c10/c10/util/overflows.h

@@ -0,0 +1,95 @@
+#pragma once
+
+#include <c10/macros/Macros.h>
+#include <c10/util/TypeSafeSignMath.h>
+#include <c10/util/complex.h>
+
+#include <cmath>
+#include <limits>
+#include <type_traits>
+
+namespace c10 {
+// In some versions of MSVC, there will be a compiler error when building.
+// C4146: unary minus operator applied to unsigned type, result still unsigned
+// C4804: unsafe use of type 'bool' in operation
+// It can be addressed by disabling the following warning.
+#ifdef _MSC_VER
+#pragma warning(push)
+#pragma warning(disable : 4146)
+#pragma warning(disable : 4804)
+#pragma warning(disable : 4018)
+#endif
+
+// The overflow checks may involve float to int conversion which may
+// trigger precision loss warning. Re-enable the warning once the code
+// is fixed. See T58053069.
+C10_CLANG_DIAGNOSTIC_PUSH()
+#if C10_CLANG_HAS_WARNING("-Wimplicit-float-conversion")
+C10_CLANG_DIAGNOSTIC_IGNORE("-Wimplicit-float-conversion")
+#endif
+
+// bool can be converted to any type.
+// Without specializing on bool, in pytorch_linux_trusty_py2_7_9_build:
+// `error: comparison of constant '255' with boolean expression is always false`
+// for `f > limit::max()` below
+template <typename To, typename From>
+std::enable_if_t<std::is_same_v<From, bool>, bool>
+overflows(From /*f*/, bool strict_unsigned [[maybe_unused]] = false) {
+  return false;
+}
+
+// skip isnan and isinf check for integral types
+template <typename To, typename From>
+std::enable_if_t<std::is_integral_v<From> && !std::is_same_v<From, bool>, bool>
+overflows(From f, bool strict_unsigned = false) {
+  using limit = std::numeric_limits<typename scalar_value_type<To>::type>;
+  if constexpr (!limit::is_signed && std::numeric_limits<From>::is_signed) {
+    // allow for negative numbers to wrap using two's complement arithmetic.
+    // For example, with uint8, this allows for `a - b` to be treated as
+    // `a + 255 * b`.
+    if (!strict_unsigned) {
+      return greater_than_max<To>(f) ||
+             (c10::is_negative(f) &&
+              -static_cast<uint64_t>(f) > static_cast<uint64_t>(limit::max()));
+    }
+  }
+  return c10::less_than_lowest<To>(f) || greater_than_max<To>(f);
+}
+
+template <typename To, typename From>
+std::enable_if_t<std::is_floating_point_v<From>, bool>
+overflows(From f, bool strict_unsigned [[maybe_unused]] = false) {
+  using limit = std::numeric_limits<typename scalar_value_type<To>::type>;
+  if (limit::has_infinity && std::isinf(static_cast<double>(f))) {
+    return false;
+  }
+  if (!limit::has_quiet_NaN && (f != f)) {
+    return true;
+  }
+  return f < limit::lowest() || f > limit::max();
+}
+
+C10_CLANG_DIAGNOSTIC_POP()
+
+#ifdef _MSC_VER
+#pragma warning(pop)
+#endif
+
+template <typename To, typename From>
+std::enable_if_t<is_complex<From>::value, bool>
+overflows(From f, bool strict_unsigned = false) {
+  // casts from complex to real are considered to overflow if the
+  // imaginary component is non-zero
+  if (!is_complex<To>::value && f.imag() != 0) {
+    return true;
+  }
+  // Check for overflow componentwise
+  // (Technically, the imag overflow check is guaranteed to be false
+  // when !is_complex<To>, but any optimizer worth its salt will be
+  // able to figure it out.)
+  return overflows<typename scalar_value_type<To>::type,
+                   typename From::value_type>(f.real(), strict_unsigned) ||
+         overflows<typename scalar_value_type<To>::type,
+                   typename From::value_type>(f.imag(), strict_unsigned);
+}
+} // namespace c10

package/third-party/include/executorch/runtime/core/portable_type/c10/c10/util/safe_numerics.h

@@ -0,0 +1,97 @@
+#pragma once
+#include <c10/macros/Macros.h>
+
+#include <cstddef>
+#include <cstdint>
+
+// GCC has __builtin_mul_overflow from before it supported __has_builtin
+#ifdef _MSC_VER
+#define C10_HAS_BUILTIN_OVERFLOW() (0)
+#include <c10/util/llvmMathExtras.h>
+#include <intrin.h>
+#else
+#define C10_HAS_BUILTIN_OVERFLOW() (1)
+#endif
+
+namespace c10 {
+
+C10_ALWAYS_INLINE bool add_overflows(uint64_t a, uint64_t b, uint64_t *out) {
+#if C10_HAS_BUILTIN_OVERFLOW()
+  return __builtin_add_overflow(a, b, out);
+#else
+  unsigned long long tmp;
+#if defined(_M_IX86) || defined(_M_X64)
+  auto carry = _addcarry_u64(0, a, b, &tmp);
+#else
+  tmp = a + b;
+  unsigned long long vector = (a & b) ^ ((a ^ b) & ~tmp);
+  auto carry = vector >> 63;
+#endif
+  *out = tmp;
+  return carry;
+#endif
+}
+
+template <typename T> C10_ALWAYS_INLINE bool mul_overflows(T a, T b, T *out) {
+#if C10_HAS_BUILTIN_OVERFLOW()
+  return __builtin_mul_overflow(a, b, out);
+#else
+  static_assert(std::is_integral_v<T>,
+                "mul_overflows only supports integral types");
+
+  if constexpr (std::is_signed_v<T>) {
+    // For signed types, use the division-based check
+    volatile T tmp = a * b;
+    *out = tmp;
+    if (a == 0 || b == 0) {
+      return false;
+    }
+    return !(a == tmp / b);
+  } else {
+    // For unsigned types, use leading zeros approach
+    // This test isn't exact, but avoids doing integer division
+    *out = a * b;
+    constexpr int bits = sizeof(T) * 8;
+    return ((c10::llvm::countLeadingZeros(a) +
+             c10::llvm::countLeadingZeros(b)) < bits);
+  }
+#endif
+}
+
+C10_ALWAYS_INLINE bool mul_overflows(uint64_t a, uint64_t b, uint64_t *out) {
+  return mul_overflows<uint64_t>(a, b, out);
+}
+
+template <typename It>
+bool safe_multiplies_u64(It first, It last, uint64_t *out) {
+#if C10_HAS_BUILTIN_OVERFLOW()
+  uint64_t prod = 1;
+  bool overflow = false;
+  for (; first != last; ++first) {
+    overflow |= c10::mul_overflows(prod, *first, &prod);
+  }
+  *out = prod;
+  return overflow;
+#else
+  uint64_t prod = 1;
+  uint64_t prod_log2 = 0;
+  bool is_zero = false;
+  for (; first != last; ++first) {
+    auto x = static_cast<uint64_t>(*first);
+    prod *= x;
+    // log2(0) isn't valid, so need to track it specially
+    is_zero |= (x == 0);
+    prod_log2 += c10::llvm::Log2_64_Ceil(x);
+  }
+  *out = prod;
+  // This test isn't exact, but avoids doing integer division
+  return !is_zero && (prod_log2 >= 64);
+#endif
+}
+
+template <typename Container>
+bool safe_multiplies_u64(const Container &c, uint64_t *out) {
+  return safe_multiplies_u64(c.begin(), c.end(), out);
+}
+
+} // namespace c10

package/third-party/include/executorch/runtime/core/portable_type/c10/torch/headeronly/macros/Export.h

@@ -0,0 +1,154 @@
+#define C10_USING_CUSTOM_GENERATED_MACROS
+#pragma once
+
+#ifndef C10_MACROS_EXPORT_H_
+#define C10_MACROS_EXPORT_H_
+
+#ifndef C10_USING_CUSTOM_GENERATED_MACROS
+#include <torch/headeronly/macros/cmake_macros.h>
+#endif // C10_USING_CUSTOM_GENERATED_MACROS
+
+/* Header file to define the common scaffolding for exported symbols.
+ *
+ * Export is by itself a quite tricky situation to deal with, and if you are
+ * hitting this file, make sure you start with the background here:
+ * - Linux: https://gcc.gnu.org/wiki/Visibility
+ * - Windows:
+ * https://docs.microsoft.com/en-us/cpp/cpp/dllexport-dllimport?view=vs-2017
+ *
+ * Do NOT include this file directly. Instead, use c10/macros/Macros.h
+ */
+
+// You do not need to edit this part of file unless you are changing the core
+// pytorch export abstractions.
+//
+// This part defines the C10 core export and import macros. This is controlled
+// by whether we are building shared libraries or not, which is determined
+// during build time and codified in c10/core/cmake_macros.h.
+// When the library is built as a shared lib, EXPORT and IMPORT will contain
+// visibility attributes. If it is being built as a static lib, then EXPORT
+// and IMPORT basically have no effect.
+
+// As a rule of thumb, you should almost NEVER mix static and shared builds for
+// libraries that depend on c10. AKA, if c10 is built as a static library, we
+// recommend everything dependent on c10 to be built statically. If c10 is built
+// as a shared library, everything dependent on it should be built as shared. In
+// the PyTorch project, all native libraries shall use the macro
+// C10_BUILD_SHARED_LIB to check whether pytorch is building shared or static
+// libraries.
+
+// For build systems that do not directly depend on CMake and directly build
+// from the source directory (such as Buck), one may not have a cmake_macros.h
+// file at all. In this case, the build system is responsible for providing
+// correct macro definitions corresponding to the cmake_macros.h.in file.
+//
+// In such scenarios, one should define the macro
+//     C10_USING_CUSTOM_GENERATED_MACROS
+// to inform this header that it does not need to include the cmake_macros.h
+// file.
+
+#ifdef _WIN32
+#define C10_HIDDEN
+#if defined(C10_BUILD_SHARED_LIBS)
+#define C10_EXPORT __declspec(dllexport)
+#define C10_IMPORT __declspec(dllimport)
+#else
+#define C10_EXPORT
+#define C10_IMPORT
+#endif
+#else // _WIN32
+#if defined(__GNUC__)
+#define C10_EXPORT __attribute__((__visibility__("default")))
+#define C10_HIDDEN __attribute__((__visibility__("hidden")))
+#else // defined(__GNUC__)
+#define C10_EXPORT
+#define C10_HIDDEN
+#endif // defined(__GNUC__)
+#define C10_IMPORT C10_EXPORT
+#endif // _WIN32
+
+#ifdef NO_EXPORT
+#undef C10_EXPORT
+#define C10_EXPORT
+#endif
+
+// Definition of an adaptive XX_API macro, that depends on whether you are
+// building the library itself or not, routes to XX_EXPORT and XX_IMPORT.
+// Basically, you will need to do this for each shared library that you are
+// building, and the instruction is as follows: assuming that you are building
+// a library called libawesome.so. You should:
+// (1) for your cmake target (usually done by "add_library(awesome, ...)"),
+//     define a macro called AWESOME_BUILD_MAIN_LIB using
+//     target_compile_options.
+// (2) define the AWESOME_API macro similar to the one below.
+// And in the source file of your awesome library, use AWESOME_API to
+// annotate public symbols.
+
+// Here, for the C10 library, we will define the macro C10_API for both import
+// and export.
+
+// This one is being used by libc10.so
+#ifdef C10_BUILD_MAIN_LIB
+#define C10_API C10_EXPORT
+#else
+#define C10_API C10_IMPORT
+#endif
+
+// This one is being used by libtorch.so
+#ifdef CAFFE2_BUILD_MAIN_LIB
+#define TORCH_API C10_EXPORT
+#else
+#define TORCH_API C10_IMPORT
+#endif
+
+// You may be wondering why we have TORCH_CUDA_CPP_API and TORCH_CUDA_CU_API
+// belonging to the same library instead of just one TORCH_CUDA_API. Well, it
+// can indeed just be one TORCH_CUDA_API (and used to be)! TORCH_CUDA_CPP_API
+// and TORCH_CUDA_CU_API are artifacts of when we needed a split build to
+// avoid relocation marker linking errors. The context is as follows:
+//
+// Once upon a time, there _was_ only TORCH_CUDA_API. All was happy until we
+// tried to compile PyTorch for CUDA 11.1, which ran into relocation marker
+// issues when linking big binaries.
+// (https://github.com/pytorch/pytorch/issues/39968) We had two choices:
+//    (1) Stop supporting so many GPU architectures
+//    (2) Do something else
+// We chose #2 and decided to split the behemoth that was torch_cuda into two
+// smaller libraries, one with most of the core kernel functions (torch_cuda_cu)
+// and the other that had..well..everything else (torch_cuda_cpp). The idea was
+// this: instead of linking our static libraries (like the hefty
+// libcudnn_static.a) with another huge library, torch_cuda, and run into pesky
+// relocation marker issues, we could link our static libraries to a smaller
+// part of torch_cuda (torch_cuda_cpp) and avoid the issues.
+
+// libtorch_cuda.so (where torch_cuda_cu and torch_cuda_cpp are a part of the
+// same api)
+#ifdef TORCH_CUDA_BUILD_MAIN_LIB
+#define TORCH_CUDA_CPP_API C10_EXPORT
+#define TORCH_CUDA_CU_API C10_EXPORT
+#else
+#define TORCH_CUDA_CPP_API C10_IMPORT
+#define TORCH_CUDA_CU_API C10_IMPORT
+#endif
+
+#if defined(TORCH_HIP_BUILD_MAIN_LIB)
+#define TORCH_HIP_CPP_API C10_EXPORT
+#define TORCH_HIP_API C10_EXPORT
+#else
+#define TORCH_HIP_CPP_API C10_IMPORT
+#define TORCH_HIP_API C10_IMPORT
+#endif
+
+#if defined(TORCH_XPU_BUILD_MAIN_LIB)
+#define TORCH_XPU_API C10_EXPORT
+#else
+#define TORCH_XPU_API C10_IMPORT
+#endif
+
+// Enums only need to be exported on windows for non-CUDA files
+#if defined(_WIN32) && defined(__CUDACC__)
+#define C10_API_ENUM C10_API
+#else
+#define C10_API_ENUM
+#endif
+#endif // C10_MACROS_EXPORT_H_