[2/2] Remove float compare option from sincosf

Message ID f36da210-930b-2314-53b7-72d74e91ce55@arm.com
State New
Headers show
Series
  • Minor updates to the new math code
Related show

Commit Message

Szabolcs Nagy July 11, 2018, 10:12 a.m.

Patch

From 82d80802dd01b6cbbf361c027e1af02e626bd5bd Mon Sep 17 00:00:00 2001
From: Szabolcs Nagy <szabolcs.nagy@arm.com>
Date: Tue, 10 Jul 2018 17:14:32 +0100
Subject: [PATCH 2/2] Remove float compare option from sincosf

PREFER_FLOAT_COMPARISON setting was not correct as it could raise
spurious exceptions.  Fixing it is easy: just use ISLESS(x, y) instead
of abstop12(x) < abstop12(y) with appropriate non-signaling definition
for ISLESS.  However it seems this setting is not very useful (there is
only minor performance difference on various architectures), so remove
this option for now.
---
 newlib/libm/common/sincosf.h | 11 +----------
 1 file changed, 1 insertion(+), 10 deletions(-)

diff --git a/newlib/libm/common/sincosf.h b/newlib/libm/common/sincosf.h
index 27f24ffe3..3cac8e88f 100644
--- a/newlib/libm/common/sincosf.h
+++ b/newlib/libm/common/sincosf.h
@@ -43,21 +43,12 @@  extern const sincos_t __sincosf_table[2] HIDDEN;
 
 extern const uint32_t __inv_pio4[] HIDDEN;
 
-/* abstop12 assumes floating point reinterpret is fast by default.
-   If floating point comparisons are faster, define PREFER_FLOAT_COMPARISON.  */
-#if PREFER_FLOAT_COMPARISON
-static inline float
-abstop12 (float x)
-{
-  return fabsf (x);
-}
-#else
+/* Top 12 bits of the float representation with the sign bit cleared.  */
 static inline uint32_t
 abstop12 (float x)
 {
   return (asuint (x) >> 20) & 0x7ff;
 }
-#endif
 
 /* Compute the sine and cosine of inputs X and X2 (X squared), using the
    polynomial P and store the results in SINP and COSP.  N is the quadrant,
-- 
2.14.1