tests: math: interpolation: disable floating point contraction

This test cannot tolerate any loss of precision, including the one
caused by the compiler contracting floating points operations together,
like in fused multiply-add (FMA). Some toolchains enable FP contraction
by default, so disable it for the specific test and let the user decide
themselves whether precision or performance is needed for their
specific application.

Co-authored-by: Jordan Yates <[email protected]>
Co-authored-by: Ilya Tagunov <[email protected]>
Signed-off-by: Ilya Tagunov <[email protected]>

Author: 2 people

tests/lib/math/interpolation/src/main.c

@@ -6,6 +6,18 @@
 #include <math.h>
 
 #include <zephyr/ztest.h>
+
+#ifdef __clang__
+/*
+ * Floating-point contraction merges an operation of the form (a*b + c) from
+ * two operations (fmul, fadd) into a single operation (fmadd). This can change
+ * the value of the resulting LSB, causing problems when the expected value is
+ * some multiple of 0.5f and the rounding functions are used. Disable
+ * contraction for the tests to ensure this doesn't occur.
+ */
+#pragma STDC FP_CONTRACT OFF
+#endif
+
 #include <zephyr/math/interpolation.h>
 
 ZTEST(interpolation, test_interpolation_oob)