| Message ID | 87wp09ok1f.fsf@linaro.org |
|---|---|
| State | New |
| Headers | show |
| Series | Fix vect_float markup for a couple of tests (PR 83888) | expand |
On Mon, Jan 22, 2018 at 6:55 PM, Richard Sandiford <richard.sandiford@linaro.org> wrote: > vect_float is true for arm*-*-* targets, but the support is only > available when -funsafe-math-optimizations is on. This caused > failures in two tests that disable fast-math. > > The easiest fix seemed to be to add a new target selector for > "vect_float without special options". > > Tested on aarch64-linux-gnu, x86_64-linux-gnu and powerpc64le-linux-gnu. > OK to install? Ok - so arm float vectors are not IEEE compliant? Richard. > Richard > > > 2018-01-22 Richard Sandiford <richard.sandiford@linaro.org> > > gcc/ > PR testsuite/83888 > * doc/sourcebuild.texi (vect_float): Say that the selector > only describes the situation when -funsafe-math-optimizations is on. > (vect_float_strict): Document. > > gcc/testsuite/ > PR testsuite/83888 > * lib/target-supports.exp (check_effective_target_vect_float): Say > that the result only holds when -funsafe-math-optimizations is on. > (check_effective_target_vect_float_strict): New procedure. > * gcc.dg/vect/no-fast-math-vect16.c: Use vect_float_strict instead > of vect_float. > * gcc.dg/vect/vect-reduc-6.c: Likewise. > > Index: gcc/doc/sourcebuild.texi > =================================================================== > --- gcc/doc/sourcebuild.texi 2018-01-22 17:51:03.860579049 +0000 > +++ gcc/doc/sourcebuild.texi 2018-01-22 17:54:02.172848564 +0000 > @@ -1403,7 +1403,13 @@ The target's preferred vector alignment > alignment. > > @item vect_float > -Target supports hardware vectors of @code{float}. > +Target supports hardware vectors of @code{float} when > +@option{-funsafe-math-optimizations} is in effect. > + > +@item vect_float_strict > +Target supports hardware vectors of @code{float} when > +@option{-funsafe-math-optimizations} is not in effect. > +This implies @code{vect_float}. > > @item vect_int > Target supports hardware vectors of @code{int}. > Index: gcc/testsuite/lib/target-supports.exp > =================================================================== > --- gcc/testsuite/lib/target-supports.exp 2018-01-22 17:51:03.817580787 +0000 > +++ gcc/testsuite/lib/target-supports.exp 2018-01-22 17:54:02.173848531 +0000 > @@ -5492,7 +5492,8 @@ proc check_effective_target_vect_long { > return $answer > } > > -# Return 1 if the target supports hardware vectors of float, 0 otherwise. > +# Return 1 if the target supports hardware vectors of float when > +# -funsafe-math-optimizations is enabled, 0 otherwise. > # > # This won't change for different subtargets so cache the result. > > @@ -5525,6 +5526,14 @@ proc check_effective_target_vect_float { > return $et_vect_float_saved($et_index) > } > > +# Return 1 if the target supports hardware vectors of float without > +# -funsafe-math-optimizations being enabled, 0 otherwise. > + > +proc check_effective_target_vect_float_strict { } { > + return [expr { [check_effective_target_vect_float] > + && ![istarget arm*-*-*] }] > +} > + > # Return 1 if the target supports hardware vectors of double, 0 otherwise. > # > # This won't change for different subtargets so cache the result. > Index: gcc/testsuite/gcc.dg/vect/no-fast-math-vect16.c > =================================================================== > --- gcc/testsuite/gcc.dg/vect/no-fast-math-vect16.c 2018-01-13 18:01:15.293116922 +0000 > +++ gcc/testsuite/gcc.dg/vect/no-fast-math-vect16.c 2018-01-22 17:54:02.172848564 +0000 > @@ -1,4 +1,4 @@ > -/* { dg-require-effective-target vect_float } */ > +/* { dg-require-effective-target vect_float_strict } */ > > #include <stdarg.h> > #include "tree-vect.h" > Index: gcc/testsuite/gcc.dg/vect/vect-reduc-6.c > =================================================================== > --- gcc/testsuite/gcc.dg/vect/vect-reduc-6.c 2018-01-13 18:01:15.294116882 +0000 > +++ gcc/testsuite/gcc.dg/vect/vect-reduc-6.c 2018-01-22 17:54:02.172848564 +0000 > @@ -1,4 +1,4 @@ > -/* { dg-require-effective-target vect_float } */ > +/* { dg-require-effective-target vect_float_strict } */ > /* { dg-additional-options "-fno-fast-math" } */ > > #include <stdarg.h>
Richard Biener <richard.guenther@gmail.com> writes: > On Mon, Jan 22, 2018 at 6:55 PM, Richard Sandiford > <richard.sandiford@linaro.org> wrote: >> vect_float is true for arm*-*-* targets, but the support is only >> available when -funsafe-math-optimizations is on. This caused >> failures in two tests that disable fast-math. >> >> The easiest fix seemed to be to add a new target selector for >> "vect_float without special options". >> >> Tested on aarch64-linux-gnu, x86_64-linux-gnu and powerpc64le-linux-gnu. >> OK to install? > > Ok - so arm float vectors are not IEEE compliant? Not for subnormals apparently: ; Note that NEON operations don't support the full IEEE 754 standard: in ; particular, denormal values are flushed to zero. This means that GCC cannot ; use those instructions for autovectorization, etc. unless ; -funsafe-math-optimizations is in effect (in which case flush-to-zero ; behavior is permissible). Thanks, Richard
Index: gcc/doc/sourcebuild.texi =================================================================== --- gcc/doc/sourcebuild.texi 2018-01-22 17:51:03.860579049 +0000 +++ gcc/doc/sourcebuild.texi 2018-01-22 17:54:02.172848564 +0000 @@ -1403,7 +1403,13 @@ The target's preferred vector alignment alignment. @item vect_float -Target supports hardware vectors of @code{float}. +Target supports hardware vectors of @code{float} when +@option{-funsafe-math-optimizations} is in effect. + +@item vect_float_strict +Target supports hardware vectors of @code{float} when +@option{-funsafe-math-optimizations} is not in effect. +This implies @code{vect_float}. @item vect_int Target supports hardware vectors of @code{int}. Index: gcc/testsuite/lib/target-supports.exp =================================================================== --- gcc/testsuite/lib/target-supports.exp 2018-01-22 17:51:03.817580787 +0000 +++ gcc/testsuite/lib/target-supports.exp 2018-01-22 17:54:02.173848531 +0000 @@ -5492,7 +5492,8 @@ proc check_effective_target_vect_long { return $answer } -# Return 1 if the target supports hardware vectors of float, 0 otherwise. +# Return 1 if the target supports hardware vectors of float when +# -funsafe-math-optimizations is enabled, 0 otherwise. # # This won't change for different subtargets so cache the result. @@ -5525,6 +5526,14 @@ proc check_effective_target_vect_float { return $et_vect_float_saved($et_index) } +# Return 1 if the target supports hardware vectors of float without +# -funsafe-math-optimizations being enabled, 0 otherwise. + +proc check_effective_target_vect_float_strict { } { + return [expr { [check_effective_target_vect_float] + && ![istarget arm*-*-*] }] +} + # Return 1 if the target supports hardware vectors of double, 0 otherwise. # # This won't change for different subtargets so cache the result. Index: gcc/testsuite/gcc.dg/vect/no-fast-math-vect16.c =================================================================== --- gcc/testsuite/gcc.dg/vect/no-fast-math-vect16.c 2018-01-13 18:01:15.293116922 +0000 +++ gcc/testsuite/gcc.dg/vect/no-fast-math-vect16.c 2018-01-22 17:54:02.172848564 +0000 @@ -1,4 +1,4 @@ -/* { dg-require-effective-target vect_float } */ +/* { dg-require-effective-target vect_float_strict } */ #include <stdarg.h> #include "tree-vect.h" Index: gcc/testsuite/gcc.dg/vect/vect-reduc-6.c =================================================================== --- gcc/testsuite/gcc.dg/vect/vect-reduc-6.c 2018-01-13 18:01:15.294116882 +0000 +++ gcc/testsuite/gcc.dg/vect/vect-reduc-6.c 2018-01-22 17:54:02.172848564 +0000 @@ -1,4 +1,4 @@ -/* { dg-require-effective-target vect_float } */ +/* { dg-require-effective-target vect_float_strict } */ /* { dg-additional-options "-fno-fast-math" } */ #include <stdarg.h>