compress copy (scalar) (#1663)

* compress copy tests
* compress copy scalar

include/eve/detail/function/simd/common/to_logical.hpp

@@ -64,10 +64,9 @@ to_logical(C c, eve::as<T>) noexcept
   // represented. We then use an unsigned version of the index type.
   // We don't just use unsigned indexes all the time cause on most cases,
   // signed comparisons are faster and this will lead to pessimisation.
-  using i_t = std::conditional_t< (T::size()>=128 && sizeof(element_type_t<T>) == 1)
-                                , as_integer_t<typename as_logical_t<T>::mask_type,unsigned>
-                                , as_integer_t<typename as_logical_t<T>::mask_type>
-                                >;
+  using i_t = std::conditional_t<(T::size() >= 128 && sizeof(element_type_t<l_t>) == 1),
+                                 typename l_t::bits_type,
+                                 as_integer_t<typename l_t::bits_type, signed>>;
 
   if constexpr( std::same_as<C, ignore_all_> ) return l_t {false};
   else if constexpr( std::same_as<C, ignore_none_> ) return l_t {true};

include/eve/module/core/compress/compress_copy.hpp

@@ -58,10 +58,12 @@ namespace eve
 //!     * the ignored elements can are not loaded (same as load[ignore])
 //!     * they are treated as not selected, regardless of the mask value
 //!   2nd is the output side ignore:
-//!     * elements that are ignored, will not be written. Example:
-//!       if the ignore_first(1) is passed, the first selected element
-//!       will not appear anywhere in the output and the 2nd selected
-//!       element will be written in (out + 1)
+//!     * elements that are ignored, will not be written.
+//!       having initial offset is equvialent to offeting the `o + offset`.
+//!       followed by keep_first(count)
+//!     Example:
+//!       if the `eve::ignore_extrema(1, L::size() - 2)` is passed,
+//`       the first selected element will be written at o + 1.
 //!     Defaults to 1st ignore.
 //!
 //!   If the mask == true this and it's `unsafe` variation, this is the same behaviour as
@@ -95,7 +97,7 @@ namespace eve
 //!                typename I,
 //!                logical_simd_value L,
 //!                typename O>
-//!      autoo compress_copy
+//!      auto compress_copy
 //!        [safe/unsafe][sparse/dense]
 //!        [C1 ignore_in][C2 ignore_out](
 //!           I in,
@@ -107,7 +109,7 @@ namespace eve
 //!                typename I,
 //!                logical_simd_value L,
 //!                typename O>
-//!      autoo compress_copy
+//!      auto compress_copy
 //!        [safe/unsafe][sparse/dense]
 //!        [C1 ignore_in][C2 ignore_out](
 //!           I in,

include/eve/module/core/compress/compress_copy_scalar.hpp

@@ -0,0 +1,94 @@
+//==================================================================================================
+/*
+  EVE - Expressive Vector Engine
+  Copyright : EVE Project Contributors
+  SPDX-License-Identifier: BSL-1.0
+*/
+//==================================================================================================
+#pragma once
+
+#include <eve/arch.hpp>
+#include <eve/detail/overload.hpp>
+
+namespace eve
+{
+EVE_REGISTER_CALLABLE(compress_copy_scalar_impl_)
+EVE_DECLARE_CALLABLE(compress_copy_scalar_impl_, compress_copy_scalar_impl)
+
+namespace detail
+{
+  EVE_ALIAS_CALLABLE(compress_copy_scalar_impl_, compress_copy_scalar_impl);
+}
+
+EVE_CALLABLE_API(compress_copy_scalar_impl_, compress_copy_scalar_impl)
+}
+
+//================================================================================================
+//! @addtogroup core_compress
+//! @{
+//!    @var compress_copy_scalar
+//!    @brief One of the implementations for eve::compress_copy that relies on
+//!    only scalar reads/writes
+//!
+//!   **Defined in Header**
+//!
+//!   @code
+//!   #include <eve/module/core.hpp>
+//!   @endcode
+//!
+//!   @warning you should probably use eve::compress_copy.
+//!
+//!   eve::compress_copy will either call this or eve::compress_copy_simd.
+//!   We believe eve::compress_copy makes correct selection of which one to use
+//!   but this is exposed in case we didn't.
+//!
+//!   Api completly matches eve::compress_copy.
+//!
+//!   @groupheader{Callable Signatures}
+//!
+//!   @code
+//!   namespace eve
+//!   {
+//!
+//!      template <relative_conditional_expr C1,
+//!                relative_conditional_expr C2,
+//!                typename I,
+//!                logical_simd_value L,
+//!                typename O>
+//!      auto compress_copy_scalar
+//!        [safe/unsafe][sparse/dense]
+//!        [C1 ignore_in][C2 ignore_out](
+//!           I in,
+//!           L mask,
+//!           O out) -> unaligned_t<O>;   // (1)
+//!
+//!      template <relative_conditional_expr C1,
+//!                relative_conditional_expr C2,
+//!                typename I,
+//!                logical_simd_value L,
+//!                typename O>
+//!      auto compress_copy_scalar
+//!        [safe/unsafe][sparse/dense]
+//!        [C1 ignore_in][C2 ignore_out](
+//!           I in,
+//!           wide<value_type_t<I>, fixed<L::size()>> preloaded,
+//!           L mask,
+//!           O out) -> unaligned_t<O>;   // (2)
+//!   }
+//!   @endcode
+//!
+//!   **Parameters**
+//!
+//!   @see eve::compress_copy
+//!
+//!   **Return value**
+//!
+//!   @see eve::compress_copy
+//! @}
+//================================================================================================
+
+#include <eve/module/core/compress/simd/common/compress_copy_scalar.hpp>
+
+#if defined(EVE_INCLUDE_SVE_HEADER)
+#  include <eve/module/core/compress/simd/arm/sve/compress_copy_scalar.hpp>
+#endif

include/eve/module/core/compress/simd/arm/sve/compress_copy_scalar.hpp

@@ -0,0 +1,95 @@
+//==================================================================================================
+/*
+  EVE - Expressive Vector Engine
+  Copyright : EVE Project Contributors
+  SPDX-License-Identifier: BSL-1.0
+*/
+//==================================================================================================
+#pragma once
+
+#include <eve/module/core/regular/logical_andnot.hpp>
+
+namespace eve::detail
+{
+
+template <typename I, typename L, typename O>
+EVE_FORCEINLINE O
+compress_copy_sparse_sve_no_limits(I f, L m, O o)
+{
+  if constexpr( has_aggregated_abi_v<L> )
+  {
+    auto [lo, hi] = m.slice();
+    o             = compress_copy_sparse_sve_no_limits(f, lo, o);
+    return compress_copy_sparse_sve_no_limits(f + (L::size() / 2), hi, o);
+  }
+  else
+  {
+    while( any(m) )
+    {
+      L ignored_before = svbrka_z(sve_true<element_type_t<L>>(), m);
+      m = svbic_z(sve_true<element_type_t<L>>(), m, ignored_before);
+      auto cur = f + count_true(ignored_before) - 1;
+      eve::write(eve::read(cur), o++);
+    }
+    return o;
+  }
+}
+
+template <typename I, typename L, typename O>
+EVE_FORCEINLINE O
+compress_copy_sparse_sve_limited(I f, L m, O o, O limit)
+{
+  if constexpr( has_aggregated_abi_v<L> )
+  {
+    auto [lo, hi] = m.slice();
+    o             = compress_copy_sparse_sve_limited(f, lo, o, limit);
+    return compress_copy_sparse_sve_limited(f + (L::size() / 2), hi, o, limit);
+  }
+  else
+  {
+    while( any(m) )
+    {
+      if (o == limit) return o;
+      L ignored_before = svbrka_z(sve_true<element_type_t<L>>(), m);
+      m = svbic_z(sve_true<element_type_t<L>>(), m, ignored_before);
+      auto cur = f + count_true(ignored_before) - 1;
+      eve::write(eve::read(cur), o++);
+    }
+    return o;
+  }
+}
+
+template<typename Settings, typename I, typename L, typename O>
+EVE_FORCEINLINE auto
+compress_copy_scalar_impl_(EVE_SUPPORTS(sve_), Settings settings, I f, L m, O o) -> O
+    requires(Settings::is_sparse) //
+{
+  using IC = typename Settings::cond_in_t;
+  using OC = typename Settings::cond_out_t;
+
+  if constexpr( L::size() < 4 )
+  {
+    return compress_copy_scalar_impl(
+        compress_callable_settings(settings.safety, dense, settings.c_in, settings.c_out), f, m, o);
+  }
+  else if constexpr( !IC::is_complete )
+  {
+    return compress_copy_scalar_impl(
+        compress_callable_settings(settings.safety, settings.density, ignore_none, settings.c_out),
+        f,
+        m && to_logical(settings.c_in, as(m)),
+        o);
+  }
+  else if constexpr ( !OC::is_complete )
+  {
+    o += settings.c_out.offset(eve::as(m));
+    O limit = o + settings.c_out.count(eve::as(m));
+    return compress_copy_sparse_sve_limited(f, m, o, limit);
+  }
+  else
+  {
+    return compress_copy_sparse_sve_no_limits(f, m, o);
+  }
+}
+
+}