ENH: Speed up trim_zeros (#16911)

* Added a benchmark for `trim_zeros()`

* Improve the performance of `np.trim_zeros()`

* Increase the variety of the tests

Fall back to the old `np.trim_zeros()` implementation if an exception is encountered.
Emit a `DeprecationWarning` in such case.


* DEP,REL: Added a deprecation release note

1 files changed, 27 insertions, 0 deletions

diff --git a/benchmarks/benchmarks/bench_trim_zeros.py b/benchmarks/benchmarks/bench_trim_zeros.py
new file mode 100644
index 000000000..4e25a8b02
--- /dev/null
+++ b/benchmarks/benchmarks/bench_trim_zeros.py
@@ -0,0 +1,27 @@
+from .common import Benchmark
+
+import numpy as np
+
+_FLOAT = np.dtype('float64')
+_COMPLEX = np.dtype('complex128')
+_INT = np.dtype('int64')
+_BOOL = np.dtype('bool')
+
+
+class TrimZeros(Benchmark):
+    param_names = ["dtype", "size"]
+    params = [
+        [_INT, _FLOAT, _COMPLEX, _BOOL],
+        [3000, 30_000, 300_000]
+    ]
+
+    def setup(self, dtype, size):
+        n = size // 3
+        self.array = np.hstack([
+            np.zeros(n),
+            np.random.uniform(size=n),
+            np.zeros(n),
+        ]).astype(dtype)
+
+    def time_trim_zeros(self, dtype, size):
+        np.trim_zeros(self.array)