TST: Fix test_numeric on riscv64

markdryan · markdryan · commit 6ac005152acd · 2023-12-19T12:22:10.000Z
Modify test_numeric so that it passes on riscv64.  The subtest
TestBoolCmp::test_float currently fails on riscv64 as it assumes that
the sign of -np.nan is retained when stored to and read back from an
array.  This is not always the case on riscv64.  Many RISC-V
instructions that produce NaNs return a canonical NaN, as defined by
the RISC-V specification. The canonical NaNs are always positive.  In
this particular test the negative sign of the -np.nan is lost when it
is converted from a double to a float before being stored in
self.signf.  It's best not to make any assumptions about the retention
of signs for negative NaNs on RISC-V so we disable both sign tests for
-np.nan on riscv64.
diff --git a/numpy/_core/tests/test_numeric.py b/numpy/_core/tests/test_numeric.py
@@ -798,8 +798,15 @@ def setup_method(self):
         self.signd[self.ed] *= -1.
         self.signf[1::6][self.ef[1::6]] = -np.inf
         self.signd[1::6][self.ed[1::6]] = -np.inf
-        self.signf[3::6][self.ef[3::6]] = -np.nan
-        self.signd[3::6][self.ed[3::6]] = -np.nan
+        # On RISC-V, many operations that produce NaNs, such as converting
+        # a -NaN from f64 to f32, return a canonical NaN.  The canonical
+        # NaNs are always positive.  See section 11.3 NaN Generation and
+        # Propagation of the RISC-V Unprivileged ISA for more details.
+        # We disable the sign test on riscv64 for -np.nan as we
+        # cannot assume that its sign will be honoured in these tests.
+        if platform.processor() != 'riscv64':
+            self.signf[3::6][self.ef[3::6]] = -np.nan
+            self.signd[3::6][self.ed[3::6]] = -np.nan
         self.signf[4::6][self.ef[4::6]] = -0.
         self.signd[4::6][self.ed[4::6]] = -0.
 
