Re: [Patch, fortran] Use BUILT_IN_IROUND
On Fri, Mar 16, 2012 at 9:17 AM, Janne Blomqvist
wrote:
> On Thu, Mar 15, 2012 at 22:28, Janne Blomqvist
> wrote:
>> On Thu, Mar 15, 2012 at 22:14, Tobias Burnus wrote:
>>> Janne Blomqvist wrote:
since some time GCC has BUILT_IN_IROUND{F,,L}, similar to lround() and
llround() but the result is returned as an integer.
Regtested on x86_64-unknown-linux-gnu, Ok for trunk?
>>>
>>>
>>> OK. Thanks for the patch! Nit: Could you check mathbuiltins.def - at least
>>> in the diff, "iround" seems to be misaligned (one " " missing).
>>
>> Ah, a tab had sneaked in, fixed. Committed the fixed patch as r185442.
>> Thanks for the quick review!
>>
>> --
>> Janne Blomqvist
>
> I realized there was a bug in the patch, and testing revealed my
> suspicion as true. Namely as there is no library fallback for iround,
> and the middle-end machinery takes care of converting it to lround
> only for float, double, and long double, it doesn't work for
> __float128. In the testcase in my original patch mail, changing x to
> real(16) generated a call to iroundq, which doesn't exist. Committed
> the patch below as obvious.
This seems to have broken SPEC 2000 FPU quite a lot as we now get
/gcc/spec/sb-vangelis-head-64/x86_64/install-201203152354/lib/gcc/x86_64-unknown-linux-gnu/4.8.0/../../../../lib64/libgfortran.so:
undefined reference to `iroundq'
collect2: error: ld returned 1 exit status
specmake: *** [apsi] Error 1
though maybe it was Jakubs patch as libgfortran has the reference?
Jakub?
Richard.
> Index: ChangeLog
> ===
> --- ChangeLog (revision 185452)
> +++ ChangeLog (working copy)
> @@ -1,3 +1,8 @@
> +2012-03-16 Janne Blomqvist
> +
> + * trans-intrinsic.c (build_round_expr): Don't use BUILT_IN_IROUND
> + for __float128.
> +
> 2012-03-15 Janne Blomqvist
>
> * f95-lang.c (gfc_init_builtin_functions): Initialize
> Index: trans-intrinsic.c
> ===
> --- trans-intrinsic.c (revision 185452)
> +++ trans-intrinsic.c (working copy)
> @@ -383,10 +383,11 @@ build_round_expr (tree arg, tree restype
> resprec = TYPE_PRECISION (restype);
>
> /* Depending on the type of the result, choose the int intrinsic
> - (iround, available only as a builtin), long int intrinsic
> (lround
> - family) or long long intrinsic (llround). We might also need to
> - convert the result afterwards. */
> - if (resprec <= INT_TYPE_SIZE)
> + (iround, available only as a builtin, therefore cannot use it
> for
> + __float128), long int intrinsic (lround family) or long long
> + intrinsic (llround). We might also need to convert the result
> + afterwards. */
> + if (resprec <= INT_TYPE_SIZE && argprec <= LONG_DOUBLE_TYPE_SIZE)
> fn = builtin_decl_for_precision (BUILT_IN_IROUND, argprec);
> else if (resprec <= LONG_TYPE_SIZE)
> fn = builtin_decl_for_precision (BUILT_IN_LROUND, argprec);
>
>
>
> --
> Janne Blomqvist
Re: [Patch, fortran] Use BUILT_IN_IROUND
On Thu, Mar 15, 2012 at 22:28, Janne Blomqvist
wrote:
> On Thu, Mar 15, 2012 at 22:14, Tobias Burnus wrote:
>> Janne Blomqvist wrote:
>>>
>>> since some time GCC has BUILT_IN_IROUND{F,,L}, similar to lround() and
>>> llround() but the result is returned as an integer.
>>>
>>> Regtested on x86_64-unknown-linux-gnu, Ok for trunk?
>>
>>
>> OK. Thanks for the patch! Nit: Could you check mathbuiltins.def - at least
>> in the diff, "iround" seems to be misaligned (one " " missing).
>
> Ah, a tab had sneaked in, fixed. Committed the fixed patch as r185442.
> Thanks for the quick review!
>
> --
> Janne Blomqvist
I realized there was a bug in the patch, and testing revealed my
suspicion as true. Namely as there is no library fallback for iround,
and the middle-end machinery takes care of converting it to lround
only for float, double, and long double, it doesn't work for
__float128. In the testcase in my original patch mail, changing x to
real(16) generated a call to iroundq, which doesn't exist. Committed
the patch below as obvious.
Index: ChangeLog
===
--- ChangeLog (revision 185452)
+++ ChangeLog (working copy)
@@ -1,3 +1,8 @@
+2012-03-16 Janne Blomqvist
+
+ * trans-intrinsic.c (build_round_expr): Don't use BUILT_IN_IROUND
+ for __float128.
+
2012-03-15 Janne Blomqvist
* f95-lang.c (gfc_init_builtin_functions): Initialize
Index: trans-intrinsic.c
===
--- trans-intrinsic.c (revision 185452)
+++ trans-intrinsic.c (working copy)
@@ -383,10 +383,11 @@ build_round_expr (tree arg, tree restype
resprec = TYPE_PRECISION (restype);
/* Depending on the type of the result, choose the int intrinsic
- (iround, available only as a builtin), long int intrinsic
(lround
- family) or long long intrinsic (llround). We might also need to
- convert the result afterwards. */
- if (resprec <= INT_TYPE_SIZE)
+ (iround, available only as a builtin, therefore cannot use it
for
+ __float128), long int intrinsic (lround family) or long long
+ intrinsic (llround). We might also need to convert the result
+ afterwards. */
+ if (resprec <= INT_TYPE_SIZE && argprec <= LONG_DOUBLE_TYPE_SIZE)
fn = builtin_decl_for_precision (BUILT_IN_IROUND, argprec);
else if (resprec <= LONG_TYPE_SIZE)
fn = builtin_decl_for_precision (BUILT_IN_LROUND, argprec);
--
Janne Blomqvist
Re: [Patch, fortran] Use BUILT_IN_IROUND
On Thu, Mar 15, 2012 at 22:14, Tobias Burnus wrote:
> Janne Blomqvist wrote:
>>
>> since some time GCC has BUILT_IN_IROUND{F,,L}, similar to lround() and
>> llround() but the result is returned as an integer.
>>
>> Regtested on x86_64-unknown-linux-gnu, Ok for trunk?
>
>
> OK. Thanks for the patch! Nit: Could you check mathbuiltins.def - at least
> in the diff, "iround" seems to be misaligned (one " " missing).
Ah, a tab had sneaked in, fixed. Committed the fixed patch as r185442.
Thanks for the quick review!
--
Janne Blomqvist
Re: [Patch, fortran] Use BUILT_IN_IROUND
Janne Blomqvist wrote:
since some time GCC has BUILT_IN_IROUND{F,,L}, similar to lround() and
llround() but the result is returned as an integer.
Regtested on x86_64-unknown-linux-gnu, Ok for trunk?
OK. Thanks for the patch! Nit: Could you check mathbuiltins.def - at
least in the diff, "iround" seems to be misaligned (one " " missing).
Tobias
2012-03-15 Janne Blomqvist
* f95-lang.c (gfc_init_builtin_functions): Initialize
BUILT_IN_IROUND.
* mathbuiltins.def: Add IROUND.
* trans-intrinsic.c (build_round_expr): Use BUILT_IN_IROUND if
type size matches.
(gfc_build_intrinsic_lib_fndecls): Build iround functions.
[Patch, fortran] Use BUILT_IN_IROUND
Hi,
since some time GCC has BUILT_IN_IROUND{F,,L}, similar to lround() and
llround() but the result is returned as an integer. As there is no
corresponding libc function, this builtin is expanded to lround{f,l}
except when -ffast-math is used, in which case it enables slightly
shorter and faster code to be generated inline. Attached patch enables
this builtin in the gfortran frontend. For the testcase
function my_nint(x)
implicit none
real :: x
integer :: my_nint
my_nint = nint(x)
end function my_nint
compiled with "-O2 -ffast-math" on x86-64, the difference between the
assembler output of trunk and trunk+patch:
--- iround.trunk.O2fastmath.s 2012-03-15 20:12:40.045069324 +0200
+++ iround.s2012-03-15 20:24:19.501320278 +0200
@@ -12,7 +12,7 @@ my_nint_:
movss .LC0(%rip), %xmm0
orps%xmm1, %xmm0
addss %xmm2, %xmm0
- cvttss2siq %xmm0, %rax
+ cvttss2si %xmm0, %eax
ret
.cfi_endproc
.LFE0:
Comparing the size of the object files, in the patched version the
text section is one byte shorter.
Regtested on x86_64-unknown-linux-gnu, Ok for trunk?
(As an aside, there were some recent problems with __builtin_iround(),
see PR 52592, but they seem to be fixed now so I think this should be
safe)
2012-03-15 Janne Blomqvist
* f95-lang.c (gfc_init_builtin_functions): Initialize
BUILT_IN_IROUND.
* mathbuiltins.def: Add IROUND.
* trans-intrinsic.c (build_round_expr): Use BUILT_IN_IROUND if
type size matches.
(gfc_build_intrinsic_lib_fndecls): Build iround functions.
--
Janne Blomqvist
diff --git a/gcc/fortran/f95-lang.c b/gcc/fortran/f95-lang.c
index 05b598f..3f28e67 100644
--- a/gcc/fortran/f95-lang.c
+++ b/gcc/fortran/f95-lang.c
@@ -773,7 +773,11 @@ gfc_init_builtin_functions (void)
gfc_define_builtin ("__builtin_fmodf", mfunc_float[1],
BUILT_IN_FMODF, "fmodf", ATTR_CONST_NOTHROW_LEAF_LIST);
- /* lround{f,,l} and llround{f,,l} */
+ /* iround{f,,l}, lround{f,,l} and llround{f,,l} */
+ ftype = build_function_type_list (integer_type_node,
+float_type_node, NULL_TREE);
+ gfc_define_builtin("__builtin_iroundf", ftype, BUILT_IN_IROUNDF,
+ "iroundf", ATTR_CONST_NOTHROW_LEAF_LIST);
ftype = build_function_type_list (long_integer_type_node,
float_type_node, NULL_TREE);
gfc_define_builtin ("__builtin_lroundf", ftype, BUILT_IN_LROUNDF,
@@ -783,6 +787,10 @@ gfc_init_builtin_functions (void)
gfc_define_builtin ("__builtin_llroundf", ftype, BUILT_IN_LLROUNDF,
"llroundf", ATTR_CONST_NOTHROW_LEAF_LIST);
+ ftype = build_function_type_list (integer_type_node,
+double_type_node, NULL_TREE);
+ gfc_define_builtin("__builtin_iround", ftype, BUILT_IN_IROUND,
+ "iround", ATTR_CONST_NOTHROW_LEAF_LIST);
ftype = build_function_type_list (long_integer_type_node,
double_type_node, NULL_TREE);
gfc_define_builtin ("__builtin_lround", ftype, BUILT_IN_LROUND,
@@ -792,6 +800,10 @@ gfc_init_builtin_functions (void)
gfc_define_builtin ("__builtin_llround", ftype, BUILT_IN_LLROUND,
"llround", ATTR_CONST_NOTHROW_LEAF_LIST);
+ ftype = build_function_type_list (integer_type_node,
+long_double_type_node, NULL_TREE);
+ gfc_define_builtin("__builtin_iroundl", ftype, BUILT_IN_IROUNDL,
+ "iroundl", ATTR_CONST_NOTHROW_LEAF_LIST);
ftype = build_function_type_list (long_integer_type_node,
long_double_type_node, NULL_TREE);
gfc_define_builtin ("__builtin_lroundl", ftype, BUILT_IN_LROUNDL,
diff --git a/gcc/fortran/mathbuiltins.def b/gcc/fortran/mathbuiltins.def
index b0bcc1f..f6d9586 100644
--- a/gcc/fortran/mathbuiltins.def
+++ b/gcc/fortran/mathbuiltins.def
@@ -64,6 +64,7 @@ OTHER_BUILTIN (FMOD, "fmod", 2, true)
OTHER_BUILTIN (FREXP, "frexp", frexp, false)
OTHER_BUILTIN (LLROUND, "llround", llround, true)
OTHER_BUILTIN (LROUND,"lround",lround, true)
+OTHER_BUILTIN (IROUND, "iround",iround, true)
OTHER_BUILTIN (NEXTAFTER, "nextafter", 2, true)
OTHER_BUILTIN (POW, "pow", 1, true)
OTHER_BUILTIN (ROUND, "round", 1, true)
diff --git a/gcc/fortran/trans-intrinsic.c b/gcc/fortran/trans-intrinsic.c
index ac9f507..5e54d8e 100644
--- a/gcc/fortran/trans-intrinsic.c
+++ b/gcc/fortran/trans-intrinsic.c
@@ -376,28 +376,24 @@ build_round_expr (tree arg, tree restype)
{
tree argtype;
tree fn;
- bool longlong;
int argprec, resprec;
argtype = TREE_TYPE (arg);
argprec = TYPE_PRECISION (argtype);
resprec = TYPE_PRECISION (restype);
- /* Depending on the type of the result, choose the long int intrinsic
- (lround family) or long long intrinsic (llround). We might also
- need to convert the result afterwards. */
- if (res
