This is the mail archive of the libc-alpha@sourceware.org mailing list for the glibc project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [Patch soft-fp] Add support for various half-precision conversion routines


On Wed, Sep 07, 2016 at 04:20:43PM +0000, Joseph Myers wrote:
> On Wed, 7 Sep 2016, James Greenhalgh wrote:
>
> > What I'm not clear about is whether this patch is sufficient on the glibc
> > side, or if I also need to wire something up in a Makefile over here?
>
> No, it's not expected to be wired up in a Makefile, and testing in GCC is
> appropriate in this case.
>
> > In the patch below, soft-fp/half.h is derived from soft-fp/single.h .
>
> You are, correctly, not including the various macros for arithmetic
> operations not using this code, but I think that means a comment to that
> effect as in <https://gcc.gnu.org/ml/gcc-patches/2009-04/msg01132.html> is
> appropriate.

Thanks. The pointer to this patch was very helpful for resolving your
other comments. I've applied the suggested changes, and re-run the GCC
testing, which showed no issues in a native bootstrap and test with the
_Float16 type enabled, and using these library functions.

> But is DF to HF is handled *correctly* by GCC proper (avoiding double
> rounding), or is such a libgcc call actually needed to get correct
> conversions rather than going via SFmode (or is that double rounding issue
> only an issue for AArch32 or for older instruction set variants you're not
> dealing with in this patch)?

Yes, the instruction emitted is:

  fcvt    d0, h0

Which is a one-step conversion from DFmode to HFmode.

As you correctly identify, the older AArch32 instruction sets do not
provide this one-step conversion. I expect I'll need to add truncdfhf2 when
I start work on the ARM support for _Float16.

> It's reasonable to include just the functions you need, as long as the
> write-up, the ChangeLog and the patch itself are consistent.

Hopefully I've squashed such bugs for this submission.

---

This patch adds conversion routines required for _Float16 support in
AArch64.

These are one-step conversions to and from TImode and TFmode. We need
these on AArch64 regardless of presence of the ARMv8.2-A 16-bit floating-point
extensions.

In the patch below, soft-fp/half.h is derived from soft-fp/single.h .
The conversion routines are derivatives of their respective SFmode
variants.

OK? If so, I'll need help applying the patch to glibc, as I have no commit
rights over here.

Thanks,
James

---
2016-09-19  James Greenhalgh  <james.greenhalgh@arm.com>

	* soft-fp/extendhftf2.c: New.
	* soft-fp/fixhfti.c: Likewise.
	* soft-fp/fixunshfti.c: Likewise.
	* soft-fp/floattihf.c: Likewise.
	* soft-fp/floatuntihf.c: Likewise.
	* soft-fp/half.h: Likewise.
	* soft-fp/trunctfhf2.c: Likewise.

diff --git a/soft-fp/extendhftf2.c b/soft-fp/extendhftf2.c
new file mode 100644
index 0000000..6ff6438
--- /dev/null
+++ b/soft-fp/extendhftf2.c
@@ -0,0 +1,53 @@
+/* Software floating-point emulation.
+   Return an IEEE half converted to IEEE quad
+   Copyright (C) 1997-2016 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   In addition to the permissions in the GNU Lesser General Public
+   License, the Free Software Foundation gives you unlimited
+   permission to link the compiled version of this file into
+   combinations with other programs, and to distribute those
+   combinations without any restriction coming from the use of this
+   file.  (The Lesser General Public License restrictions do apply in
+   other respects; for example, they cover modification of the file,
+   and distribution when not linked into a combine executable.)
+
+   The GNU C Library 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
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#define FP_NO_EXACT_UNDERFLOW
+#include "soft-fp.h"
+#include "half.h"
+#include "quad.h"
+
+TFtype
+__extendhftf2 (HFtype a)
+{
+  FP_DECL_EX;
+  FP_DECL_H (A);
+  FP_DECL_Q (R);
+  TFtype r;
+
+  FP_INIT_EXCEPTIONS;
+  FP_UNPACK_RAW_H (A, a);
+#if (2 * _FP_W_TYPE_SIZE) < _FP_FRACBITS_Q
+  FP_EXTEND (Q, H, 4, 1, R, A);
+#else
+  FP_EXTEND (Q, H, 2, 1, R, A);
+#endif
+  FP_PACK_RAW_Q (r, R);
+  FP_HANDLE_EXCEPTIONS;
+
+  return r;
+}
diff --git a/soft-fp/fixhfti.c b/soft-fp/fixhfti.c
new file mode 100644
index 0000000..3610f4c
--- /dev/null
+++ b/soft-fp/fixhfti.c
@@ -0,0 +1,45 @@
+/* Software floating-point emulation.
+   Convert IEEE half to 128bit signed integer
+   Copyright (C) 2007-2016 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   In addition to the permissions in the GNU Lesser General Public
+   License, the Free Software Foundation gives you unlimited
+   permission to link the compiled version of this file into
+   combinations with other programs, and to distribute those
+   combinations without any restriction coming from the use of this
+   file.  (The Lesser General Public License restrictions do apply in
+   other respects; for example, they cover modification of the file,
+   and distribution when not linked into a combine executable.)
+
+   The GNU C Library 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
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#include "soft-fp.h"
+#include "half.h"
+
+TItype
+__fixhfti (HFtype a)
+{
+  FP_DECL_EX;
+  FP_DECL_H (A);
+  UTItype r;
+
+  FP_INIT_EXCEPTIONS;
+  FP_UNPACK_RAW_H (A, a);
+  FP_TO_INT_H (r, A, TI_BITS, 1);
+  FP_HANDLE_EXCEPTIONS;
+
+  return r;
+}
diff --git a/soft-fp/fixunshfti.c b/soft-fp/fixunshfti.c
new file mode 100644
index 0000000..30edbfe
--- /dev/null
+++ b/soft-fp/fixunshfti.c
@@ -0,0 +1,45 @@
+/* Software floating-point emulation.
+   Convert IEEE half to 128bit unsigned integer
+   Copyright (C) 2007-2016 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   In addition to the permissions in the GNU Lesser General Public
+   License, the Free Software Foundation gives you unlimited
+   permission to link the compiled version of this file into
+   combinations with other programs, and to distribute those
+   combinations without any restriction coming from the use of this
+   file.  (The Lesser General Public License restrictions do apply in
+   other respects; for example, they cover modification of the file,
+   and distribution when not linked into a combine executable.)
+
+   The GNU C Library 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
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#include "soft-fp.h"
+#include "half.h"
+
+UTItype
+__fixunshfti (HFtype a)
+{
+  FP_DECL_EX;
+  FP_DECL_H (A);
+  UTItype r;
+
+  FP_INIT_EXCEPTIONS;
+  FP_UNPACK_RAW_H (A, a);
+  FP_TO_INT_H (r, A, TI_BITS, 0);
+  FP_HANDLE_EXCEPTIONS;
+
+  return r;
+}
diff --git a/soft-fp/floattihf.c b/soft-fp/floattihf.c
new file mode 100644
index 0000000..74ac83a
--- /dev/null
+++ b/soft-fp/floattihf.c
@@ -0,0 +1,45 @@
+/* Software floating-point emulation.
+   Convert a 128bit signed integer to IEEE half
+   Copyright (C) 2007-2016 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   In addition to the permissions in the GNU Lesser General Public
+   License, the Free Software Foundation gives you unlimited
+   permission to link the compiled version of this file into
+   combinations with other programs, and to distribute those
+   combinations without any restriction coming from the use of this
+   file.  (The Lesser General Public License restrictions do apply in
+   other respects; for example, they cover modification of the file,
+   and distribution when not linked into a combine executable.)
+
+   The GNU C Library 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
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#include "soft-fp.h"
+#include "half.h"
+
+HFtype
+__floattihf (TItype i)
+{
+  FP_DECL_EX;
+  FP_DECL_H (A);
+  HFtype a;
+
+  FP_INIT_ROUNDMODE;
+  FP_FROM_INT_H (A, i, TI_BITS, UTItype);
+  FP_PACK_RAW_H (a, A);
+  FP_HANDLE_EXCEPTIONS;
+
+  return a;
+}
diff --git a/soft-fp/floatuntihf.c b/soft-fp/floatuntihf.c
new file mode 100644
index 0000000..90e77c2
--- /dev/null
+++ b/soft-fp/floatuntihf.c
@@ -0,0 +1,45 @@
+/* Software floating-point emulation.
+   Convert a 128bit unsigned integer to IEEE half.
+   Copyright (C) 2007-2016 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   In addition to the permissions in the GNU Lesser General Public
+   License, the Free Software Foundation gives you unlimited
+   permission to link the compiled version of this file into
+   combinations with other programs, and to distribute those
+   combinations without any restriction coming from the use of this
+   file.  (The Lesser General Public License restrictions do apply in
+   other respects; for example, they cover modification of the file,
+   and distribution when not linked into a combine executable.)
+
+   The GNU C Library 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
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#include "soft-fp.h"
+#include "half.h"
+
+HFtype
+__floatuntihf (UTItype i)
+{
+  FP_DECL_EX;
+  FP_DECL_H (A);
+  HFtype a;
+
+  FP_INIT_ROUNDMODE;
+  FP_FROM_INT_H (A, i, TI_BITS, UTItype);
+  FP_PACK_RAW_H (a, A);
+  FP_HANDLE_EXCEPTIONS;
+
+  return a;
+}
diff --git a/soft-fp/half.h b/soft-fp/half.h
new file mode 100644
index 0000000..ba75297
--- /dev/null
+++ b/soft-fp/half.h
@@ -0,0 +1,170 @@
+/* Software floating-point emulation.
+   Definitions for IEEE Half Precision.
+   Copyright (C) 1997-2016 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   In addition to the permissions in the GNU Lesser General Public
+   License, the Free Software Foundation gives you unlimited
+   permission to link the compiled version of this file into
+   combinations with other programs, and to distribute those
+   combinations without any restriction coming from the use of this
+   file.  (The Lesser General Public License restrictions do apply in
+   other respects; for example, they cover modification of the file,
+   and distribution when not linked into a combine executable.)
+
+   The GNU C Library 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
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#ifndef SOFT_FP_HALF_H
+#define SOFT_FP_HALF_H	1
+
+#if _FP_W_TYPE_SIZE < 32
+# error "Here's a nickel kid.  Go buy yourself a real computer."
+#endif
+
+#define _FP_FRACTBITS_H		(_FP_W_TYPE_SIZE)
+
+#define _FP_FRACTBITS_DW_H	(_FP_W_TYPE_SIZE)
+
+#define _FP_FRACBITS_H		11
+#define _FP_FRACXBITS_H		(_FP_FRACTBITS_H - _FP_FRACBITS_H)
+#define _FP_WFRACBITS_H		(_FP_WORKBITS + _FP_FRACBITS_H)
+#define _FP_WFRACXBITS_H	(_FP_FRACTBITS_H - _FP_WFRACBITS_H)
+#define _FP_EXPBITS_H		5
+#define _FP_EXPBIAS_H		15
+#define _FP_EXPMAX_H		31
+
+#define _FP_QNANBIT_H		((_FP_W_TYPE) 1 << (_FP_FRACBITS_H-2))
+#define _FP_QNANBIT_SH_H	((_FP_W_TYPE) 1 << (_FP_FRACBITS_H-2+_FP_WORKBITS))
+#define _FP_IMPLBIT_H		((_FP_W_TYPE) 1 << (_FP_FRACBITS_H-1))
+#define _FP_IMPLBIT_SH_H	((_FP_W_TYPE) 1 << (_FP_FRACBITS_H-1+_FP_WORKBITS))
+#define _FP_OVERFLOW_H		((_FP_W_TYPE) 1 << (_FP_WFRACBITS_H))
+
+#define _FP_WFRACBITS_DW_H	(2 * _FP_WFRACBITS_H)
+#define _FP_WFRACXBITS_DW_H	(_FP_FRACTBITS_DW_H - _FP_WFRACBITS_DW_H)
+#define _FP_HIGHBIT_DW_H	\
+  ((_FP_W_TYPE) 1 << (_FP_WFRACBITS_DW_H - 1) % _FP_W_TYPE_SIZE)
+
+/* The implementation of _FP_MUL_MEAT_H and _FP_DIV_MEAT_H should be
+   chosen by the target machine.  */
+
+typedef float HFtype __attribute__ ((mode (HF)));
+
+union _FP_UNION_H
+{
+  HFtype flt;
+  struct _FP_STRUCT_LAYOUT
+  {
+#if __BYTE_ORDER == __BIG_ENDIAN
+    unsigned sign : 1;
+    unsigned exp  : _FP_EXPBITS_H;
+    unsigned frac : _FP_FRACBITS_H - (_FP_IMPLBIT_H != 0);
+#else
+    unsigned frac : _FP_FRACBITS_H - (_FP_IMPLBIT_H != 0);
+    unsigned exp  : _FP_EXPBITS_H;
+    unsigned sign : 1;
+#endif
+  } bits __attribute__ ((packed));
+};
+
+#define FP_DECL_H(X)		_FP_DECL (1, X)
+#define FP_UNPACK_RAW_H(X, val)	_FP_UNPACK_RAW_1 (H, X, (val))
+#define FP_UNPACK_RAW_HP(X, val)	_FP_UNPACK_RAW_1_P (H, X, (val))
+#define FP_PACK_RAW_H(val, X)	_FP_PACK_RAW_1 (H, (val), X)
+#define FP_PACK_RAW_HP(val, X)			\
+  do						\
+    {						\
+      if (!FP_INHIBIT_RESULTS)			\
+	_FP_PACK_RAW_1_P (H, (val), X);		\
+    }						\
+  while (0)
+
+#define FP_UNPACK_H(X, val)			\
+  do						\
+    {						\
+      _FP_UNPACK_RAW_1 (H, X, (val));		\
+      _FP_UNPACK_CANONICAL (H, 1, X);		\
+    }						\
+  while (0)
+
+#define FP_UNPACK_HP(X, val)			\
+  do						\
+    {						\
+      _FP_UNPACK_RAW_1_P (H, X, (val));		\
+      _FP_UNPACK_CANONICAL (H, 1, X);		\
+    }						\
+  while (0)
+
+#define FP_UNPACK_SEMIRAW_H(X, val)		\
+  do						\
+    {						\
+      _FP_UNPACK_RAW_1 (H, X, (val));		\
+      _FP_UNPACK_SEMIRAW (H, 1, X);		\
+    }						\
+  while (0)
+
+#define FP_UNPACK_SEMIRAW_HP(X, val)		\
+  do						\
+    {						\
+      _FP_UNPACK_RAW_1_P (H, X, (val));		\
+      _FP_UNPACK_SEMIRAW (H, 1, X);		\
+    }						\
+  while (0)
+
+#define FP_PACK_H(val, X)			\
+  do						\
+    {						\
+      _FP_PACK_CANONICAL (H, 1, X);		\
+      _FP_PACK_RAW_1 (H, (val), X);		\
+    }						\
+  while (0)
+
+#define FP_PACK_HP(val, X)			\
+  do						\
+    {						\
+      _FP_PACK_CANONICAL (H, 1, X);		\
+      if (!FP_INHIBIT_RESULTS)			\
+	_FP_PACK_RAW_1_P (H, (val), X);		\
+    }						\
+  while (0)
+
+#define FP_PACK_SEMIRAW_H(val, X)		\
+  do						\
+    {						\
+      _FP_PACK_SEMIRAW (H, 1, X);		\
+      _FP_PACK_RAW_1 (H, (val), X);		\
+    }						\
+  while (0)
+
+#define FP_PACK_SEMIRAW_HP(val, X)		\
+  do						\
+    {						\
+      _FP_PACK_SEMIRAW (H, 1, X);		\
+      if (!FP_INHIBIT_RESULTS)			\
+	_FP_PACK_RAW_1_P (H, (val), X);		\
+    }						\
+  while (0)
+
+#define FP_TO_INT_H(r, X, rsz, rsg)	_FP_TO_INT (H, 1, (r), X, (rsz), (rsg))
+#define FP_TO_INT_ROUND_H(r, X, rsz, rsg)	\
+  _FP_TO_INT_ROUND (H, 1, (r), X, (rsz), (rsg))
+#define FP_FROM_INT_H(X, r, rs, rt)	_FP_FROM_INT (H, 1, X, (r), (rs), rt)
+
+/* HFmode arithmetic is not implemented.  */
+
+#define _FP_FRAC_HIGH_H(X)	_FP_FRAC_HIGH_1 (X)
+#define _FP_FRAC_HIGH_RAW_H(X)	_FP_FRAC_HIGH_1 (X)
+#define _FP_FRAC_HIGH_DW_H(X)	_FP_FRAC_HIGH_1 (X)
+
+#endif /* !SOFT_FP_HALF_H */
diff --git a/soft-fp/trunctfhf2.c b/soft-fp/trunctfhf2.c
new file mode 100644
index 0000000..0500b14
--- /dev/null
+++ b/soft-fp/trunctfhf2.c
@@ -0,0 +1,52 @@
+/* Software floating-point emulation.
+   Truncate IEEE quad into IEEE half.
+   Copyright (C) 1997-2015 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   In addition to the permissions in the GNU Lesser General Public
+   License, the Free Software Foundation gives you unlimited
+   permission to link the compiled version of this file into
+   combinations with other programs, and to distribute those
+   combinations without any restriction coming from the use of this
+   file.  (The Lesser General Public License restrictions do apply in
+   other respects; for example, they cover modification of the file,
+   and distribution when not linked into a combine executable.)
+
+   The GNU C Library 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
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#include "soft-fp.h"
+#include "half.h"
+#include "quad.h"
+
+HFtype
+__trunctfhf2 (TFtype a)
+{
+  FP_DECL_EX;
+  FP_DECL_Q (A);
+  FP_DECL_H (R);
+  HFtype r;
+
+  FP_INIT_ROUNDMODE;
+  FP_UNPACK_SEMIRAW_Q (A, a);
+#if (2 * _FP_W_TYPE_SIZE) < _FP_FRACBITS_Q
+  FP_TRUNC (H, Q, 1, 4, R, A);
+#else
+  FP_TRUNC (H, Q, 1, 2, R, A);
+#endif
+  FP_PACK_SEMIRAW_H (r, R);
+  FP_HANDLE_EXCEPTIONS;
+
+  return r;
+}

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]