Resolve gh-2882

CHANGELOG.md

@@ -30,6 +30,7 @@ This release is compatible with NumPy 2.4.5.
 * Fixed incorrect `dpnp.tensor.acosh` result for `complex(±0, NaN)` special case to match the Python Array API specification [#2914](https://github.com/IntelPython/dpnp/pull/2914)
 * Fixed fork PR documentation workflow failures by implementing conditional publishing strategy: upstream PRs publish to GitHub Pages with comment, fork PRs upload artifacts [#2910](https://github.com/IntelPython/dpnp/pull/2910)
 * Fixed missing `libtensor` headers in the installed `dpnp` package [#2915](https://github.com/IntelPython/dpnp/pull/2915)
+* Fixed a bug in `astype` where casting floating point types to unsigned integral types could cause an intermediate signed integral type to overflow, leading to incorrect results [#2930](https://github.com/IntelPython/dpnp/pull/2930)
 
 ### Security
 

dpnp/tensor/libtensor/include/utils/type_utils.hpp

@@ -99,9 +99,14 @@ dstTy convert_impl(const srcTy &v)
     else if constexpr (!std::is_integral_v<srcTy> &&
                        !std::is_same_v<dstTy, bool> &&
                        std::is_integral_v<dstTy> && std::is_unsigned_v<dstTy>) {
-        // first cast to signed variant, the cast to unsigned one
-        using signedT = typename std::make_signed_t<dstTy>;
-        return static_cast<dstTy>(convert_impl<signedT, srcTy>(v));
+        // for negative values, cast through signed integer to get two's
+        // complement wrapping
+        using intermediateT =
+            std::conditional_t<sizeof(dstTy) < sizeof(std::int32_t),
+                               std::int32_t, std::int64_t>;
+        return (v < srcTy{0})
+                   ? static_cast<dstTy>(static_cast<intermediateT>(v))
+                   : static_cast<dstTy>(v);
     }
     else {
         return static_cast<dstTy>(v);

dpnp/tests/tensor/test_usm_ndarray_ctor.py

@@ -1100,6 +1100,15 @@ def test_astype_gh_2121():
     assert dpt.all(res == expected)
 
 
+def test_astype_gh_2882():
+    get_queue_or_skip()
+
+    x = dpt.asarray([160.0, 120.0], dtype="f4")
+    r = dpt.astype(x, dpt.uint8)
+    expected = dpt.asarray([160, 120], dtype="u1")
+    assert dpt.all(r == expected)
+
+
 def test_copy():
     try:
         X = dpt.usm_ndarray((5, 5), "i4")[2:4, 1:4]