Bruno Haible <[EMAIL PROTECTED]> writes:
> In m4/round.m4, the test whether floor() is available without libm is useful
> for any uses of floor(). I'm therefore putting it into a separate module
> 'floor' which you can then depend on.
Thanks, done.
> In m4/roundf.m4 you need an AC_REQUIRE([gl_FUNC_FLOORF]), so that the
> determination of FLOORF_LIBM comes before it is used in the assignment to
> ROUNDF_LIBM, not afterwards. Likewise for m4/roundl.m4.
I am not sure about this. gl_FUNC_FLOORF checks for floorf and
provides a substitute if it is not available. But for roundf I
was planning to use the system floorf if it was available and, if
not, use the roundf implementation that does not need floorf.
Thus, gl_FUNC_FLOORF does more than what roundf needs.
> In tests/test-round2.c the include of <assert.h> is not necessary.
Actually, there is one assertion in there, to verify that DOUBLE
and DOUBLE_UINT are the same size. But this can be done at
compile time, so I changed it to use the "verify" macro instead
(and added a dependency on verify to round{,f}-tests).
> tests/test-round2.c seems to have a tab/whitespace indentation problem.
> Apply sed -e 's,^ , ,'
Fixed.
I always have a hard time telling what people expect,
indentation-wise. I now work with a group of anti-tab bigots and
another group of pro-tab bigots. To please both groups, perhaps
I should write my code so that no line is ever indented past
column 6 ;-)
> In tests/test-round2.c, midbits is a constant. I would move it into the
> loop (also since it's confusingly named: lowbits is the low bits of
> janus.i. highbits is the high bits of janus.i. But the middle bits of janus.i
> are not midbits but either midbits or 0.)
OK, done.
>> my m4 code can use some looking over.
>
> I haven't looked over this part; I trust that you have looked at the
> expansion in the configure file.
Yes, I did.
I pushed this to the repository, after making the following
changes relative to the previous patch:
diff --git a/ChangeLog b/ChangeLog
index aa77450..c8f01a4 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,33 @@
+2007-10-20 Ben Pfaff <[EMAIL PROTECTED]>
+
+ * lib/math.in.h: Declare round, roundf, roundl if we are providing
+ implementations.
+ * m4/math_h.m4: New substitutions for round, roundf, roundl modules.
+ * lib/round.c: New file.
+ * lib/roundf.c: New file.
+ * lib/roundl.c: New file.
+ * m4/round.m4: New file.
+ * m4/roundf.m4: New file.
+ * m4/roundl.m4: New file.
+ * m4/check-libm-func-m4: New file.
+ * modules/math: Replace round, roundf, roundl related @VARS@ in
+ math.in.h.
+ * modules/round: New file.
+ * modules/round-tests: New file.
+ * modules/roundf: New file.
+ * modules/roundf-tests: New file.
+ * modules/roundl: New file.
+ * modules/roundl-tests: New file.
+ * tests/test-round1.c: New file.
+ * tests/test-round2.c: New file.
+ * tests/test-roundf1.c: New file.
+ * tests/test-roundf2.c: New file.
+ * tests/test-roundl.c: New file.
+ * doc/functions/round.texi: Mention round module.
+ * doc/functions/roundf.texi: Mention roundf module.
+ * doc/functions/roundl.texi: Mention roundl module.
+ * MODULES.html.sh: Mention new modules.
+
2007-10-20 Jim Meyering <[EMAIL PROTECTED]>
* lib/xprintf.c: Include <config.h> unconditionally.
diff --git a/lib/round.c b/lib/round.c
index a9fd393..5758fb7 100644
--- a/lib/round.c
+++ b/lib/round.c
@@ -20,7 +20,7 @@
#include <config.h>
-/* Specification. */
+#include <float.h>
#include <math.h>
#ifdef USE_LONG_DOUBLE
diff --git a/m4/round.m4 b/m4/round.m4
index 4914aac..edd9e89 100644
--- a/m4/round.m4
+++ b/m4/round.m4
@@ -10,30 +10,6 @@ AC_DEFUN([gl_FUNC_ROUND],
dnl Persuade glibc <math.h> to declare round().
AC_REQUIRE([gl_USE_SYSTEM_EXTENSIONS])
gl_CHECK_LIBM_FUNC([round], [x = round(x);], [], [
- dnl Test whether floor() can be used without libm.
- ROUND_LIBM=?
- AC_TRY_LINK([
- #ifndef __NO_MATH_INLINES
- # define __NO_MATH_INLINES 1 /* for glibc */
- #endif
- #include <math.h>
- double x;],
- [x = floor(x);],
- [ROUND_LIBM=])
- if test "$ROUND_LIBM" = "?"; then
- save_LIBS="$LIBS"
- LIBS="$LIBS -lm"
- AC_TRY_LINK([
- #ifndef __NO_MATH_INLINES
- # define __NO_MATH_INLINES 1 /* for glibc */
- #endif
- #include <math.h>
- double x;],
- [x = floor(x);],
- [ROUND_LIBM="-lm"])
- LIBS="$save_LIBS"
- fi
- if test "$ROUND_LIBM" = "?"; then
- ROUND_LIBM=
- fi
+ AC_REQUIRE([gl_FUNC_FLOOR])
+ ROUND_LIBM=$FLOOR_LIBM
AC_LIBOBJ([round])])])
diff --git a/modules/round b/modules/round
index 4bb01a9..0dd2526 100644
--- a/modules/round
+++ b/modules/round
@@ -8,6 +8,7 @@ m4/round.m4
Depends-on:
float
+floor
math
extensions
diff --git a/modules/round-tests b/modules/round-tests
index 2ba583e..d061431 100644
--- a/modules/round-tests
+++ b/modules/round-tests
@@ -7,6 +7,7 @@ isnan-nolibm
stdbool
stdint
fprintf-posix
+verify
configure.ac:
diff --git a/modules/roundf-tests b/modules/roundf-tests
index 8f9ef53..9fc69eb 100644
--- a/modules/roundf-tests
+++ b/modules/roundf-tests
@@ -8,6 +8,7 @@ isnanf-nolibm
stdbool
stdint
fprintf-posix
+verify
configure.ac:
diff --git a/tests/test-round2.c b/tests/test-round2.c
index 9c07a5f..1da0d9f 100644
--- a/tests/test-round2.c
+++ b/tests/test-round2.c
@@ -25,15 +25,15 @@
#define FLOOR_FREE_ROUND round_reference2
#include "round.c"
-#include <assert.h>
#include <math.h>
-
#include <float.h>
#include <stdbool.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
+#include "verify.h"
+
#ifdef USE_LONG_DOUBLE
# error Long double not supported.
#elif ! defined USE_FLOAT
@@ -63,7 +63,7 @@ equal (const char *message, DOUBLE x, DOUBLE y0, DOUBLE y1)
else
{
fprintf (stderr, "%s: "FUNCTION"(%g(%a)) = %g(%a) or %g(%a)?\n",
- message, x, x, y0, y0, y1, y1);
+ message, x, x, y0, y0, y1, y1);
return false;
}
}
@@ -87,21 +87,22 @@ check (DOUBLE x)
int
main (void)
{
- DOUBLE_UINT highbits, lowbits, midbits;
+ DOUBLE_UINT highbits, lowbits;
int error = 0;
- midbits = ((DOUBLE_UINT) -1) >> (NUM_LOWBITS + NUM_HIGHBITS) << NUM_LOWBITS;
for (highbits = 0; highbits < (1 << NUM_HIGHBITS); highbits++)
for (lowbits = 0; lowbits < (1 << NUM_LOWBITS); lowbits++)
{
/* Combine highbits and lowbits into a floating-point number,
- sign-extending the lowbits to DOUBLE_BITS-NUM_HIGHBITS bits. */
+ sign-extending the lowbits to DOUBLE_BITS-NUM_HIGHBITS bits. */
union { DOUBLE f; DOUBLE_UINT i; } janus;
- assert (sizeof janus.f == sizeof janus.i);
- janus.i = lowbits | (highbits << (DOUBLE_BITS - NUM_HIGHBITS));
- if (lowbits >> (NUM_LOWBITS - 1))
- janus.i |= midbits;
+ verify (sizeof janus.f == sizeof janus.i);
+ janus.i = lowbits | (highbits << (DOUBLE_BITS - NUM_HIGHBITS));
+ if (lowbits >> (NUM_LOWBITS - 1))
+ janus.i |= ((DOUBLE_UINT) -1
+ >> (NUM_LOWBITS + NUM_HIGHBITS)
+ << NUM_LOWBITS);
if (!check (janus.f))
- error = true;
+ error = true;
}
return (error ? 1 : 0);
}
--
Ben Pfaff
http://benpfaff.org
