libstdc++: Correct NTTP and simd_mask ctor call
Commit Message
OK for master and all backports (after 11.4 is done)?
tested on powerpc64le-linux-gnu and x86_64-pc-linux-gnu
----------- 8< ------------
Signed-off-by: Matthias Kretz <m.kretz@gsi.de>
libstdc++-v3/ChangeLog:
PR libstdc++/109822
* include/experimental/bits/simd.h (to_native): Use int NTTP
as specified in PTS2.
(to_compatible): Likewise. Add missing tag to call mask
generator ctor.
* testsuite/experimental/simd/pr109822_cast_functions.cc: New
test.
---
libstdc++-v3/include/experimental/bits/simd.h | 7 ++-
.../simd/pr109822_cast_functions.cc | 63 +++++++++++++++++++
2 files changed, 67 insertions(+), 3 deletions(-)
create mode 100644 libstdc++-v3/testsuite/experimental/simd/
pr109822_cast_functions.cc
--
──────────────────────────────────────────────────────────────────────────
Dr. Matthias Kretz https://mattkretz.github.io
GSI Helmholtz Centre for Heavy Ion Research https://gsi.de
stdₓ::simd
──────────────────────────────────────────────────────────────────────────
Comments
On Fri, 26 May 2023 at 12:11, Matthias Kretz via Libstdc++ <
libstdc++@gcc.gnu.org> wrote:
> OK for master and all backports (after 11.4 is done)?
>
OK
>
> tested on powerpc64le-linux-gnu and x86_64-pc-linux-gnu
>
> ----------- 8< ------------
>
> Signed-off-by: Matthias Kretz <m.kretz@gsi.de>
>
> libstdc++-v3/ChangeLog:
>
> PR libstdc++/109822
> * include/experimental/bits/simd.h (to_native): Use int NTTP
> as specified in PTS2.
> (to_compatible): Likewise. Add missing tag to call mask
> generator ctor.
> * testsuite/experimental/simd/pr109822_cast_functions.cc: New
> test.
> ---
> libstdc++-v3/include/experimental/bits/simd.h | 7 ++-
> .../simd/pr109822_cast_functions.cc | 63 +++++++++++++++++++
> 2 files changed, 67 insertions(+), 3 deletions(-)
> create mode 100644 libstdc++-v3/testsuite/experimental/simd/
> pr109822_cast_functions.cc
>
>
> --
> ──────────────────────────────────────────────────────────────────────────
> Dr. Matthias Kretz https://mattkretz.github.io
> GSI Helmholtz Centre for Heavy Ion Research https://gsi.de
> stdₓ::simd
> ──────────────────────────────────────────────────────────────────────────
Hello, Matthias,
On May 26, 2023, Matthias Kretz via Libstdc++ <libstdc++@gcc.gnu.org> wrote:
> OK for master and all backports (after 11.4 is done)?
> tested on powerpc64le-linux-gnu and x86_64-pc-linux-gnu
> * testsuite/experimental/simd/pr109822_cast_functions.cc: New
> test.
This testcase fails to compile on PowerPC targets without VSX: 64-bit
integer and floating-point types cannot be vectorized.
I wonder if the test is malformed (and should be amended to test for
available simd types), or whether a patch like this would be desirable
to make simd constructs more portable. I'm not sure about the
requirements.
[libstdc++] [simd] [ppc] use nonvector intrinsic fallback types
From: Alexandre Oliva <oliva@adacore.com>
Compiling such tests as pr109822_cast_functions.cc on powerpc targets
that don't support VSX fails because some intrinsic types that are
expected to be vectorizable are not defined without VSX.
Introduce fallback non-vector types to enable the code to compile.
for libstdc++-v3/ChangeLog
* include/experimental/bits/simd.h: Introduce fallback
non-vector intrinsic_type_impl specializations for PowerPC
without VSX.
---
libstdc++-v3/include/experimental/bits/simd.h | 12 ++++++++++++
1 file changed, 12 insertions(+)
diff --git a/libstdc++-v3/include/experimental/bits/simd.h b/libstdc++-v3/include/experimental/bits/simd.h
index 834fe923065bd..2691823e869e8 100644
--- a/libstdc++-v3/include/experimental/bits/simd.h
+++ b/libstdc++-v3/include/experimental/bits/simd.h
@@ -2431,9 +2431,14 @@ template <typename _Tp>
#define _GLIBCXX_SIMD_PPC_INTRIN(_Tp) \
template <> \
struct __intrinsic_type_impl<_Tp> { using type = __vector _Tp; }
+#define _GLIBCXX_SIMD_PPC_INTRIN_NOVEC(_Tp) \
+ template <> \
+ struct __intrinsic_type_impl<_Tp> { using type = _Tp; }
_GLIBCXX_SIMD_PPC_INTRIN(float);
#ifdef __VSX__
_GLIBCXX_SIMD_PPC_INTRIN(double);
+#else
+_GLIBCXX_SIMD_PPC_INTRIN_NOVEC(double);
#endif
_GLIBCXX_SIMD_PPC_INTRIN(signed char);
_GLIBCXX_SIMD_PPC_INTRIN(unsigned char);
@@ -2444,12 +2449,19 @@ _GLIBCXX_SIMD_PPC_INTRIN(unsigned int);
#if defined __VSX__ || __SIZEOF_LONG__ == 4
_GLIBCXX_SIMD_PPC_INTRIN(signed long);
_GLIBCXX_SIMD_PPC_INTRIN(unsigned long);
+#else
+_GLIBCXX_SIMD_PPC_INTRIN_NOVEC(signed long);
+_GLIBCXX_SIMD_PPC_INTRIN_NOVEC(unsigned long);
#endif
#ifdef __VSX__
_GLIBCXX_SIMD_PPC_INTRIN(signed long long);
_GLIBCXX_SIMD_PPC_INTRIN(unsigned long long);
+#else
+_GLIBCXX_SIMD_PPC_INTRIN_NOVEC(signed long long);
+_GLIBCXX_SIMD_PPC_INTRIN_NOVEC(unsigned long long);
#endif
#undef _GLIBCXX_SIMD_PPC_INTRIN
+#undef _GLIBCXX_SIMD_PPC_INTRIN_NOVEC
template <typename _Tp, size_t _Bytes>
struct __intrinsic_type<_Tp, _Bytes, enable_if_t<__is_vectorizable_v<_Tp> && _Bytes <= 16>>
Hello Alexandre,
On Friday, 2 June 2023 10:32:40 CEST Alexandre Oliva wrote:
> On May 26, 2023, Matthias Kretz via Libstdc++ <libstdc++@gcc.gnu.org> wrote:
> > OK for master and all backports (after 11.4 is done)?
> > tested on powerpc64le-linux-gnu and x86_64-pc-linux-gnu
> >
> > * testsuite/experimental/simd/pr109822_cast_functions.cc: New
> > test.
>
> This testcase fails to compile on PowerPC targets without VSX: 64-bit
> integer and floating-point types cannot be vectorized.
Yes, and the simd implementation already encodes that both in
__vectorized_sizeof<T>() and __intrinsic_type<T>.
> I wonder if the test is malformed (and should be amended to test for
> available simd types), or whether a patch like this would be desirable
> to make simd constructs more portable. I'm not sure about the
> requirements.
The test is correct. The stdx::simd implementation has a latent bug (my
dejagnu boards included only POWER7-POWER9; I'm at POWER5-POWER10 by now). The
_S_store function is trying to work around bad code-gen but fails to notice
that long long vectors can't be used.
I'm looking at that function again, also in light of recent improvements wrt.
code-gen, and will remove that assumption, that long long is vectorizable.
__intrinsic_type_t<T> should never be T, but always the type that can be
passed to corresponding platform intrinsics. There are traits for the
implementation to detect whether the intrinsics types are available.
- Matthias
>
>
> [libstdc++] [simd] [ppc] use nonvector intrinsic fallback types
>
> From: Alexandre Oliva <oliva@adacore.com>
>
> Compiling such tests as pr109822_cast_functions.cc on powerpc targets
> that don't support VSX fails because some intrinsic types that are
> expected to be vectorizable are not defined without VSX.
>
> Introduce fallback non-vector types to enable the code to compile.
>
>
> for libstdc++-v3/ChangeLog
>
> * include/experimental/bits/simd.h: Introduce fallback
> non-vector intrinsic_type_impl specializations for PowerPC
> without VSX.
> ---
> libstdc++-v3/include/experimental/bits/simd.h | 12 ++++++++++++
> 1 file changed, 12 insertions(+)
>
> diff --git a/libstdc++-v3/include/experimental/bits/simd.h
> b/libstdc++-v3/include/experimental/bits/simd.h index
> 834fe923065bd..2691823e869e8 100644
> --- a/libstdc++-v3/include/experimental/bits/simd.h
> +++ b/libstdc++-v3/include/experimental/bits/simd.h
> @@ -2431,9 +2431,14 @@ template <typename _Tp>
> #define _GLIBCXX_SIMD_PPC_INTRIN(_Tp)
> \ template <>
> \ struct __intrinsic_type_impl<_Tp> { using type = __vector _Tp; }
> +#define _GLIBCXX_SIMD_PPC_INTRIN_NOVEC(_Tp)
\
> + template <>
> \ + struct __intrinsic_type_impl<_Tp> { using type = _Tp; }
> _GLIBCXX_SIMD_PPC_INTRIN(float);
> #ifdef __VSX__
> _GLIBCXX_SIMD_PPC_INTRIN(double);
> +#else
> +_GLIBCXX_SIMD_PPC_INTRIN_NOVEC(double);
> #endif
> _GLIBCXX_SIMD_PPC_INTRIN(signed char);
> _GLIBCXX_SIMD_PPC_INTRIN(unsigned char);
> @@ -2444,12 +2449,19 @@ _GLIBCXX_SIMD_PPC_INTRIN(unsigned int);
> #if defined __VSX__ || __SIZEOF_LONG__ == 4
> _GLIBCXX_SIMD_PPC_INTRIN(signed long);
> _GLIBCXX_SIMD_PPC_INTRIN(unsigned long);
> +#else
> +_GLIBCXX_SIMD_PPC_INTRIN_NOVEC(signed long);
> +_GLIBCXX_SIMD_PPC_INTRIN_NOVEC(unsigned long);
> #endif
> #ifdef __VSX__
> _GLIBCXX_SIMD_PPC_INTRIN(signed long long);
> _GLIBCXX_SIMD_PPC_INTRIN(unsigned long long);
> +#else
> +_GLIBCXX_SIMD_PPC_INTRIN_NOVEC(signed long long);
> +_GLIBCXX_SIMD_PPC_INTRIN_NOVEC(unsigned long long);
> #endif
> #undef _GLIBCXX_SIMD_PPC_INTRIN
> +#undef _GLIBCXX_SIMD_PPC_INTRIN_NOVEC
>
> template <typename _Tp, size_t _Bytes>
> struct __intrinsic_type<_Tp, _Bytes, enable_if_t<__is_vectorizable_v<_Tp>
> && _Bytes <= 16>>
On Jun 2, 2023, Matthias Kretz <m.kretz@gsi.de> wrote:
> I'm looking at that function again, also in light of recent improvements wrt.
> code-gen, and will remove that assumption, that long long is vectorizable.
Thanks, I'll leave that to you, then.
I also noticed the same test is failing on rtems6 (at least with gcc
11). AFAICT the problem is that _GLIBCXX_SIMD_MATH_CALL* macros in
simd_math.h expect the named functions to be in std::, but I get such
errors as:
.../aarch64-rtems6/include/c++/11.4.1/experimental/bits/simd_math.h:1299: error: 'remainder' is not a member of 'std'
.../aarch64-rtems6/include/c++/11.4.1/experimental/bits/simd_math.h:1299: note: suggested alternatives:
[...]
.../aarch64-rtems6/include/math.h:346: note: 'remainder'
[...]
.../aarch64-rtems6/include/c++/11.4.1/experimental/bits/simd_math.h:1299: note: 'std::experimental::parallelism_v2::remainder'
.../aarch64-rtems6/include/c++/11.4.1/experimental/bits/simd_math.h:1299: error: template argument 1 is invalid
[...]
.../aarch64-rtems6/include/c++/11.4.1/experimental/bits/simd_math.h:1328: error: 'fmin' is not a member of 'std'; did you mean 'min'?
.../aarch64-rtems6/include/c++/11.4.1/experimental/bits/simd_math.h:1328: error: 'fmin' is not a member of 'std'; did you mean 'min'?
.../aarch64-rtems6/include/c++/11.4.1/experimental/bits/simd_math.h:1328: error: template argument 1 is invalid
ISTM that rtems is missing some of the math.h functions expected by
libstdc++, but also that even those that are present are not visible in
namespace ::std::, where the macros reasonably expect to find them. Is
this known? Should I file a PR about it?
On Friday, 2 June 2023 11:30:17 CEST Alexandre Oliva wrote:
> I also noticed the same test is failing on rtems6 (at least with gcc
> 11). AFAICT the problem is that _GLIBCXX_SIMD_MATH_CALL* macros in
> simd_math.h expect the named functions to be in std::, but I get such
> errors as:
>
> .../aarch64-rtems6/include/c++/11.4.1/experimental/bits/simd_math.h:1299:
> error: 'remainder' is not a member of 'std'
> .../aarch64-rtems6/include/c++/11.4.1/experimental/bits/simd_math.h:1299:
> note: suggested alternatives: [...]
> .../aarch64-rtems6/include/math.h:346: note: 'remainder'
> [...]
> .../aarch64-rtems6/include/c++/11.4.1/experimental/bits/simd_math.h:1299:
> note: 'std::experimental::parallelism_v2::remainder'
> .../aarch64-rtems6/include/c++/11.4.1/experimental/bits/simd_math.h:1299:
> error: template argument 1 is invalid [...]
> .../aarch64-rtems6/include/c++/11.4.1/experimental/bits/simd_math.h:1328:
> error: 'fmin' is not a member of 'std'; did you mean 'min'?
> .../aarch64-rtems6/include/c++/11.4.1/experimental/bits/simd_math.h:1328:
> error: 'fmin' is not a member of 'std'; did you mean 'min'?
> .../aarch64-rtems6/include/c++/11.4.1/experimental/bits/simd_math.h:1328:
> error: template argument 1 is invalid
>
> ISTM that rtems is missing some of the math.h functions expected by
> libstdc++, but also that even those that are present are not visible in
> namespace ::std::, where the macros reasonably expect to find them. Is
> this known? Should I file a PR about it?
I had/have no idea. Is rtems6 using the "freestanding" subset of C++? In which
case simd shouldn't be there at all. Otherwise <cmath> should work, no?
- Matthias
On Fri, 2 Jun 2023 at 10:30, Alexandre Oliva via Libstdc++ <
libstdc++@gcc.gnu.org> wrote:
>
> ISTM that rtems is missing some of the math.h functions expected by
> libstdc++, but also that even those that are present are not visible in
> namespace ::std::, where the macros reasonably expect to find them. Is
> this known? Should I file a PR about it?
>
That looks like https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109818
We only import the C99 <math.h> functions into namespace std when the
target libc supports all of them.
@@ -3304,7 +3304,7 @@ to_native(const fixed_size_simd<_Tp, _Np>& __x)
return {__mem, vector_aligned};
}
-template <typename _Tp, size_t _Np>
+template <typename _Tp, int _Np>
_GLIBCXX_SIMD_INTRINSIC
enable_if_t<(_Np == native_simd_mask<_Tp>::size()), native_simd_mask<_Tp>>
to_native(const fixed_size_simd_mask<_Tp, _Np>& __x)
@@ -3315,7 +3315,7 @@ to_native(const fixed_size_simd_mask<_Tp, _Np>& __x)
}
// to_compatible {{{2
-template <typename _Tp, size_t _Np>
+template <typename _Tp, int _Np>
_GLIBCXX_SIMD_INTRINSIC enable_if_t<(_Np == simd<_Tp>::size()), simd<_Tp>>
to_compatible(const simd<_Tp, simd_abi::fixed_size<_Np>>& __x)
{
@@ -3324,12 +3324,13 @@ to_compatible(const simd<_Tp, simd_abi::fixed_size<_Np>>& __x)
return {__mem, vector_aligned};
}
-template <typename _Tp, size_t _Np>
+template <typename _Tp, int _Np>
_GLIBCXX_SIMD_INTRINSIC
enable_if_t<(_Np == simd_mask<_Tp>::size()), simd_mask<_Tp>>
to_compatible(const simd_mask<_Tp, simd_abi::fixed_size<_Np>>& __x)
{
return simd_mask<_Tp>(
+ __private_init,
[&](auto __i) constexpr _GLIBCXX_SIMD_ALWAYS_INLINE_LAMBDA { return __x[__i]; });
}
new file mode 100644
@@ -0,0 +1,63 @@
+// { dg-options "-std=gnu++17" }
+// { dg-do compile { target c++17 } }
+
+#include <experimental/simd>
+
+namespace stdx = std::experimental;
+
+template <typename T, typename V>
+ void
+ test01()
+ {
+ using M = typename V::mask_type;
+ [[maybe_unused]] auto x = to_fixed_size(V());
+ [[maybe_unused]] auto k = to_fixed_size(M());
+ if constexpr (stdx::simd<T>::size() == V::size())
+ {
+ [[maybe_unused]] auto xx = to_compatible(x);
+ [[maybe_unused]] auto kk = to_compatible(k);
+ x = to_fixed_size(xx);
+ k = to_fixed_size(kk);
+ }
+ if constexpr (stdx::native_simd<T>::size() == V::size())
+ {
+ [[maybe_unused]] auto xx = to_native(x);
+ [[maybe_unused]] auto kk = to_native(k);
+ x = to_fixed_size(xx);
+ k = to_fixed_size(kk);
+ }
+ }
+
+template <typename T>
+ void
+ iterate_abis()
+ {
+ test01<T, stdx::simd<T, stdx::simd_abi::scalar>>();
+ test01<T, stdx::simd<T>>();
+ test01<T, stdx::native_simd<T>>();
+ test01<T, stdx::fixed_size_simd<T, 3>>();
+ test01<T, stdx::fixed_size_simd<T, stdx::simd_abi::max_fixed_size<T> - 4>>();
+ }
+
+int
+main()
+{
+ iterate_abis<char>();
+ iterate_abis<wchar_t>();
+ iterate_abis<char16_t>();
+ iterate_abis<char32_t>();
+
+ iterate_abis<signed char>();
+ iterate_abis<unsigned char>();
+ iterate_abis<short>();
+ iterate_abis<unsigned short>();
+ iterate_abis<int>();
+ iterate_abis<unsigned int>();
+ iterate_abis<long>();
+ iterate_abis<unsigned long>();
+ iterate_abis<long long>();
+ iterate_abis<unsigned long long>();
+ iterate_abis<float>();
+ iterate_abis<double>();
+ iterate_abis<long double>();
+}