Revised [PATCH] ARM half-precision floating point, 7/8 (library support)

[Sandra Loosemore <sandra at codesourcery dot com>]

Here's another slightly revised patch.  Compared to the version I posted last 
month, the only change here is that I updated the copyright notice and license 
on the new fp16.c file.

OK to commit now?

-Sandra

2009-05-15  Paul Brook  <paul@codesourcery.com>
	    Sandra Loosemore  <sandra@codesourcery.com>

	gcc/
	* config/arm/sfp-machine.h (_FP_NANFRAC_H, _FP_NANSIGN_H): Define.
	(__extendhfsf2, __truncsfhf2): Define.
	* config/arm/fp16.c: New file.
	* config/arm/t-bpabi (LIB2FUNCS_STATIC_EXTRA): Add fp16.c.
	* config/arm/t-symbian (LIB2FUNCS_STATIC_EXTRA):  Add fp16.c.

Index: gcc/config/arm/sfp-machine.h
===================================================================
--- gcc/config/arm/sfp-machine.h	(revision 146010)
+++ gcc/config/arm/sfp-machine.h	(working copy)
@@ -14,9 +14,11 @@
 #define _FP_DIV_MEAT_D(R,X,Y)	_FP_DIV_MEAT_2_udiv(D,R,X,Y)
 #define _FP_DIV_MEAT_Q(R,X,Y)	_FP_DIV_MEAT_4_udiv(Q,R,X,Y)
 
+#define _FP_NANFRAC_H		((_FP_QNANBIT_H << 1) - 1)
 #define _FP_NANFRAC_S		((_FP_QNANBIT_S << 1) - 1)
 #define _FP_NANFRAC_D		((_FP_QNANBIT_D << 1) - 1), -1
 #define _FP_NANFRAC_Q		((_FP_QNANBIT_Q << 1) - 1), -1, -1, -1
+#define _FP_NANSIGN_H		0
 #define _FP_NANSIGN_S		0
 #define _FP_NANSIGN_D		0
 #define _FP_NANSIGN_Q		0
@@ -92,5 +94,7 @@
 #define __fixdfdi	__aeabi_d2lz
 #define __fixunsdfdi	__aeabi_d2ulz
 #define __floatdidf	__aeabi_l2d
+#define __extendhfsf2	__gnu_h2f_ieee
+#define __truncsfhf2	__gnu_f2h_ieee
 
 #endif /* __ARM_EABI__ */
Index: gcc/config/arm/fp16.c
===================================================================
--- gcc/config/arm/fp16.c	(revision 0)
+++ gcc/config/arm/fp16.c	(revision 0)
@@ -0,0 +1,145 @@
+/* Half-float conversion routines.
+
+   Copyright (C) 2008, 2009 Free Software Foundation, Inc.
+   Contributed by CodeSourcery.
+
+   This file is free software; you can redistribute it and/or modify it
+   under the terms of the GNU General Public License as published by the
+   Free Software Foundation; either version 3, or (at your option) any
+   later version.
+
+   This file is distributed in the hope that it will be useful, but
+   WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   General Public License for more details.
+
+   Under Section 7 of GPL version 3, you are granted additional
+   permissions described in the GCC Runtime Library Exception, version
+   3.1, as published by the Free Software Foundation.
+
+   You should have received a copy of the GNU General Public License and
+   a copy of the GCC Runtime Library Exception along with this program;
+   see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
+   <http://www.gnu.org/licenses/>.  */
+
+static inline unsigned short
+__gnu_f2h_internal(unsigned int a, int ieee)
+{
+  unsigned short sign = (a >> 16) & 0x8000;
+  int aexp = (a >> 23) & 0xff;
+  unsigned int mantissa = a & 0x007fffff;
+  unsigned int mask;
+  unsigned int increment;
+
+  if (aexp == 0xff)
+    {
+      if (!ieee)
+	return sign;
+      return sign | 0x7e00 | (mantissa >> 13);
+    }
+  
+  if (aexp == 0 && mantissa == 0)
+    return sign;
+
+  aexp -= 127;
+
+  /* Decimal point between bits 22 and 23.  */
+  mantissa |= 0x00800000;
+  if (aexp < -14)
+    {
+      mask = 0x007fffff;
+      if (aexp < -25)
+	aexp = -26;
+      else if (aexp != -25)
+	mask >>= 24 + aexp;
+    }
+  else
+    mask = 0x00001fff;
+
+  /* Round.  */
+  if (mantissa & mask)
+    {
+      increment = (mask + 1) >> 1;
+      if ((mantissa & mask) == increment)
+	increment = mantissa & (increment << 1);
+      mantissa += increment;
+      if (mantissa >= 0x01000000)
+       	{
+	  mantissa >>= 1;
+	  aexp++;
+	}
+    }
+
+  if (ieee)
+    {
+      if (aexp > 15)
+	return sign | 0x7c00;
+    }
+  else
+    {
+      if (aexp > 16)
+	return sign | 0x7fff;
+    }
+
+  if (aexp < -24)
+    return sign;
+
+  if (aexp < -14)
+    {
+      mantissa >>= -14 - aexp;
+      aexp = -14;
+    }
+
+  /* We leave the leading 1 in the mantissa, and subtract one
+     from the exponent bias to compensate.  */
+  return sign | (((aexp + 14) << 10) + (mantissa >> 13));
+}
+
+unsigned int
+__gnu_h2f_internal(unsigned short a, int ieee)
+{
+  unsigned int sign = (unsigned int)(a & 0x8000) << 16;
+  int aexp = (a >> 10) & 0x1f;
+  unsigned int mantissa = a & 0x3ff;
+
+  if (aexp == 0x1f && ieee)
+    return sign | 0x7f800000 | (mantissa << 13);
+
+  if (aexp == 0)
+    {
+      int shift;
+
+      if (mantissa == 0)
+	return sign;
+
+      shift = __builtin_clz(mantissa) - 21;
+      mantissa <<= shift;
+      aexp = -shift;
+    }
+
+  return sign | (((aexp + 0x70) << 23) + (mantissa << 13));
+}
+
+unsigned short
+__gnu_f2h_ieee(unsigned int a)
+{
+  return __gnu_f2h_internal(a, 1);
+}
+
+unsigned int
+__gnu_h2f_ieee(unsigned short a)
+{
+  return __gnu_h2f_internal(a, 1);
+}
+
+unsigned short
+__gnu_f2h_alternative(unsigned int x)
+{
+  return __gnu_f2h_internal(x, 0);
+}
+
+unsigned int
+__gnu_h2f_alternative(unsigned short a)
+{
+  return __gnu_h2f_internal(a, 0);
+}
Index: gcc/config/arm/t-bpabi
===================================================================
--- gcc/config/arm/t-bpabi	(revision 146010)
+++ gcc/config/arm/t-bpabi	(working copy)
@@ -5,6 +5,8 @@ LIB1ASMFUNCS += _aeabi_lcmp _aeabi_ulcmp
 LIB2FUNCS_EXTRA = $(srcdir)/config/arm/bpabi.c \
 		  $(srcdir)/config/arm/unaligned-funcs.c
 
+LIB2FUNCS_STATIC_EXTRA = $(srcdir)/config/arm/fp16.c
+
 UNWIND_H = $(srcdir)/config/arm/unwind-arm.h
 LIB2ADDEH = $(srcdir)/config/arm/unwind-arm.c \
   $(srcdir)/config/arm/libunwind.S \
Index: gcc/config/arm/t-symbian
===================================================================
--- gcc/config/arm/t-symbian	(revision 146010)
+++ gcc/config/arm/t-symbian	(working copy)
@@ -17,6 +17,9 @@ UNWIND_H = $(srcdir)/config/arm/unwind-a
 LIB2ADDEH = $(srcdir)/unwind-c.c $(srcdir)/config/arm/pr-support.c
 LIB2ADDEHDEP = $(UNWIND_H)
 
+# Include half-float helpers.
+LIB2FUNCS_STATIC_EXTRA = $(srcdir)/config/arm/fp16.c
+
 # Create a multilib for processors with VFP floating-point, and a
 # multilib for those without -- using the soft-float ABI in both
 # cases.  Symbian OS object should be compiled with interworking