Merge branch 'release/debian/4.27-1'HEADdebian/4.27-1master

1 files changed, 107 insertions, 0 deletions

diff --git a/include/libHX/endian_float.h b/include/libHX/endian_float.h
new file mode 100644
index 0000000..f36e110
--- /dev/null
+++ b/include/libHX/endian_float.h
@@ -0,0 +1,107 @@
+#ifndef _LIBHX_ENDIAN_FLOAT_H
+#define _LIBHX_ENDIAN_FLOAT_H 1
+#include <string.h>
+#include <libHX/endian.h>
+
+/*
+ * While construction of integers from bytes was easy, it would be more work
+ * for floats — and compilers probably won't be able to optimize it.
+ *
+ * So then, we make some shortcuts/assumptions here in endian_float.h:
+ * - that the host platform uses the same byte order for integers as for floats
+ * - that the host platform is using IEEE754/IEC559
+ *
+ * This holds for the typical Linux on {arm gnueabi LE, arm gnueabi
+ * BE, aarch64 LE, aarch64 BE, i386, amd64, hppa, loongarch64, m68k,
+ * mips, ppc64, ppc64le, sparc, sparc64, riscv64, s390x}.
+ */
+
+/*
+ * Unlike cpu_to_be32, we will offer no float_cpu_to_be32. Values comprised of
+ * inverted bytes should probably not be passed around in memory.
+ */
+LIBHX_DBG_INLINE float LIBHX_OPT_O2 float_be32p_to_cpu(const void *p)
+{
+	uint32_t v = be32p_to_cpu(p);
+	float w;
+#ifdef __cplusplus
+	static_assert(sizeof(v) == sizeof(w));
+#endif
+	memcpy(&w, &v, sizeof(w));
+	return w;
+}
+
+LIBHX_DBG_INLINE double LIBHX_OPT_O2 float_le32p_to_cpu(const void *p)
+{
+	uint32_t v = le32p_to_cpu(p);
+	float w;
+#ifdef __cplusplus
+	static_assert(sizeof(v) == sizeof(w));
+#endif
+	memcpy(&w, &v, sizeof(w));
+	return w;
+}
+
+LIBHX_DBG_INLINE float LIBHX_OPT_O2 float_be64p_to_cpu(const void *p)
+{
+	uint64_t v = be64p_to_cpu(p);
+	double w;
+#ifdef __cplusplus
+	static_assert(sizeof(v) == sizeof(w));
+#endif
+	memcpy(&w, &v, sizeof(w));
+	return w;
+}
+
+LIBHX_DBG_INLINE double LIBHX_OPT_O2 float_le64p_to_cpu(const void *p)
+{
+	uint64_t v = le64p_to_cpu(p);
+	double w;
+#ifdef __cplusplus
+	static_assert(sizeof(v) == sizeof(w));
+#endif
+	memcpy(&w, &v, sizeof(w));
+	return w;
+}
+
+LIBHX_DBG_INLINE void LIBHX_OPT_O2 float_cpu_to_be32p(void *p, float v)
+{
+	uint32_t w;
+#ifdef __cplusplus
+	static_assert(sizeof(v) == sizeof(w));
+#endif
+	memcpy(&w, &v, sizeof(w));
+	cpu_to_be32p(p, w);
+}
+
+LIBHX_DBG_INLINE void LIBHX_OPT_O2 float_cpu_to_le32p(void *p, float v)
+{
+	uint32_t w;
+#ifdef __cplusplus
+	static_assert(sizeof(v) == sizeof(w));
+#endif
+	memcpy(&w, &v, sizeof(w));
+	cpu_to_le32p(p, w);
+}
+
+LIBHX_DBG_INLINE void LIBHX_OPT_O2 float_cpu_to_be64p(void *p, double v)
+{
+	uint64_t w;
+#ifdef __cplusplus
+	static_assert(sizeof(v) == sizeof(w));
+#endif
+	memcpy(&w, &v, sizeof(w));
+	cpu_to_be64p(p, w);
+}
+
+LIBHX_DBG_INLINE void LIBHX_OPT_O2 float_cpu_to_le64p(void *p, double v)
+{
+	uint64_t w;
+#ifdef __cplusplus
+	static_assert(sizeof(v) == sizeof(w));
+#endif
+	memcpy(&w, &v, sizeof(w));
+	cpu_to_le64p(p, w);
+}
+
+#endif /* _LIBHX_ENDIAN_FLOAT_H */