lloda 推送了 main 分支的代碼到 lloda/ra-ra
- a72390fce9 Merge big.hh and small.hh into arrays.hh
* src/arrays.hh: As stated.
* src/ply.hh: Move wlen/from here, from small.hh.
Elsewhere fix references.
- df2e410cff Trim size of ViewBig, CellBig
This makes these the size of P when rank is 0.
* ra/small.hh (SmallArray): Use special type for size 0. Remove prevent_zero_size.
See also aad94465a94bcdd41b9d1e14c95dac30de06bc10.
- 查看 2 次提交的內容比對 »
2 周之前
lloda 推送了 main 分支的代碼到 lloda/ra-ra
- fe96b07727 Looking for an angle to from(iota, ...)
* ra/expr.hh (clen): Renamed from maybe_len.
(cadd, csub): New functions.
(Ptr::constant): Unused, remove.
(reverse(Ptr)): Reuse csub.
* ra/ra.hh (opt): Renamed from optimize. Reuse cadd/csub. Fix users.
* ra/small.hh (is_iota_any): Do not exclude UNB.
(beatable): Move the iota/UNB exclusion here.
(fromb): Fix iota predicate; must also be beatable.
2 周之前
lloda 推送了 main 分支的代碼到 lloda/ra-ra
- a07a94727f Patch for bug10
* ra/small.hh (ViewSmall(nested braces)): Add barrier after init.
* test/bug10.cc: Don't downgrade warning, check values.
* test/ra-7.cc: Remove, merge into ra-9.cc.
* ra/expr.hh (Scalar): Simplify operator*().
3 周之前
lloda 推送了 main 分支的代碼到 lloda/ra-ra
- a4f3347665 Remove unnecessary forwarding
* ra/ply.hh (ply_fixed, for_each, early): Expressions get consumed here, so
there's no reason to forward.
- c6d2a1b1c8 Clear spurious (?) warning in filldimv()
* ra/expr.hh (filldimv): Resize, take Iterator, manually fill dimv[..].len.
(default_dims): Adapt user.
(Ptr::mov, Cell::mov): Remove aggressive-optimizations diagnostic override,
caused by Seq<int>.
* ra/big.hh: Adapt users.
* test/wrank.c, test/small-1.cc: Don't use Seq<int>.
* ra/ra.hh: Add binary op %.
- 查看 2 次提交的內容比對 »
3 周之前
lloda 推送了 main 分支的代碼到 lloda/ra-ra
- 8a08f22bbd Provide separate at(), at_view()
This fixes an inconsistency where at(view, i) could return any prefix slice, but
at(iter, i) only returned elements; moreover, at(view, i) could be a lot slower
than at(start(view), i) because the rank of the result depended on the length of
i, which can be dynamic.
* ra/small.hh (ViewSmall::at): Do not support slicing; require size(i)==rank(a).
* ra/big.hh (ViewBig::at): Idem.
* ra/ra.hh (at_view): Redefine, including rank extension.
(at): Define rank extension of expr.at().
* docs/ra-ra.texi (at): Update.
(at_view): New defun.
* test/at.cc: Examples from doc. Move misc at() tests here from fromu.cc.
* test/at.cc, test/ra-0.cc, test/ra-11.cc, test/small-1.cc, test/operators.cc:
Fix tests that assumed the old behavior of at() to use at_view() instead.
- 0c6e24a47b Check check for non-constant ptr's length
- 查看 2 次提交的內容比對 »
3 周之前
lloda 推送了 main 分支的代碼到 lloda/ra-ra
- 16b5e6e948 No need to forward in Ptr::at(i)
Forwarding hurt performance a lot in bench-at, but I'm not sure why that is --
if anything it should be the opposite, since start(RA_FW(i)) is a nop when i is
a temp.
- c9cad45df3 Fix scalar conversions for views
* ra/big.hh (ViewBig): Scalar conversion returns *cp not T &. This fixes scalar
conversion for sequence view, which returns rvalue.
* ra/small.hh (ViewSmall): Likewise.
* bench/bench-at.cc: Exercise the conversions here.
- 查看 2 次提交的內容比對 »
3 周之前
lloda 推送了 main 分支的代碼到 lloda/ra-ra
- 51626cb767 Look again at len/len_s vs global len_s
In the end I didn't change much.
* ra/expr.hh: Move Cell here, from small.hh.
* ra/expr.hh (Match): New static member len(is_constant).
* test/constexpr.cc: Test Match::len_s(k, check true or false).
1 月之前
lloda 推送了 main 分支的代碼到 lloda/ra-ra
- f76f5cfd3b Reorder declarations, looking to replace Ptr()
* ra/small.hh (Wlen): Move after all Iterators have been defined.
(start(is_builtin_array)): Use Cell<> directly.
* ra/expr.hh (wlen): Move together with Len, Wlen forward decl.
1 月之前
lloda 推送了 main 分支的代碼到 lloda/ra-ra
- 563da6ab25 Bug in assignment to nested type
This bug is also in 5cccf4a963aae54690b0fe257cc3de19e237611b (Before slicer
rewrite), so it's not related to the new slicer patches.
* test/bug10.cc: As stated.
* ra/ply.hh (format): Avoid every() + ops, which come in a later header.
1 月之前
lloda 推送了 main 分支的代碼到 lloda/ra-ra
- d024383856 Update version and fix manual error
* docs/ra-ra.texi: As stated.
(Slicing): Correct condition for from(...) to be a view.
1 月之前
lloda 推送了 main 分支的代碼到 lloda/ra-ra
- 734d37552d Adopt new slicer
Some loose ends remain, but nothing that didn't exist before.
* ra/small.hh (from etc.): As stated.
(ViewSmall::operator()): Reuse from().
* ra/big.hh (ViewBig::operator()): Reuse from().
* test/genfrom.cc: New tests.
* test/bench-at.cc: Bench newly enabled from(small, var rank).
* test/frame-old.cc, test/from.cc, test/ra-5.cc: Fix tests for
(rank 2 A)(rank 1 i) which is now rank 2 and not nested rank 1 of rank 1.
* test/fromb.cc, test/view-ops.cc: Test mixed u/b cases that weren't supported
before.
- 5cccf4a963 Before slicer rewrite
- 查看 2 次提交的內容比對 »
1 月之前
lloda 推送了 main 分支的代碼到 lloda/ra-ra
- de8728184a Fix fmt(..., scalar) and -DRA_CHECK=2
* ra/ply.hh (std::formatter<ra::Fmt>): start() arg to be formatted.
* test/io.cc: New test.
- a635c72d65 Minor fixes
* ra/ra.hh (at): No need to forward a.
(lexical_compare): Renamed.
* ra/base.hh (mp::append, mp::cons): Simplify.
* ra/small.hh (cat): Simplify.
- 查看 2 次提交的內容比對 »
2 月之前
lloda 推送了標籤 last-gcc-11 到 lloda/ra-ra
2 月之前
lloda 推送了標籤 last-c++20 到 lloda/ra-ra
2 月之前
lloda
已經刪除在 lloda/ra-ra 上的標籤 c++17-last
3 月之前
lloda 推送了標籤 last-c++17 到 lloda/ra-ra
3 月之前
lloda 推送了 main 分支的代碼到 lloda/ra-ra
- 0258eadcb9 Fix explode for zero-sized sub
* ra/small.hh (explode_dims): As stated.
* test/explode-0.cc: New test.
- 32422a3e97 Reuse to_scalar for all scalar conversions
* ra/expr.hh (to_scalar): Fix ambiguous overload.
* ra/small.hh, ra/big.hh: As stated.
- 查看 2 次提交的內容比對 »
3 月之前
lloda 推送了 main 分支的代碼到 lloda/ra-ra
- d532d3171f Fix static conditions in Match
* ra/expr.hh (Match): As stated.
- 04174d97f2 Fix type bug in bench/bench-gemm
Was missing update in BLAS section, which isn't exercised consistently.
* bench/bench-gemm.cc: As stated.
* .github/workflows/gcc-14-no-sanitize.yml: Use cmake sanitize flag.
- 查看 2 次提交的內容比對 »
4 月之前
lloda 推送了 main 分支的代碼到 lloda/ra-ra
- 85653fd786 Restrict forward in array constructors
Generic (auto && x) grabbed (Container & x) or (T & x), which required extra
overloads to patch it all up.
* ra/big.hh (ViewBig): Forward only for Iterator argument.
(Container): Remove (Container &) constructor and operator=.
* ra/small.hh (ViewSmall): Forward only for Iterator argument.
(SmallArray): Remove (X && x) constructor for non-T.
- 8a6f6f9716 Do not forward in array constructors
This caused caused conflicts by (auto && x) grabbing (Container & x), which
required extra overloads, etc.
In principle, temp iterators get copied if start()ed after going through (auto
const &). Might bring (auto && x) back if I find that matters.
* ra/big.hh (ViewBig): Remove special operator=(X && x), just reuse macro.
(Container): Remove (Container &) constructor and operator=.
* ra/small.hh (ViewSmall): Remove special operator=(X && x), just reuse macro.
(SmallArray): Remove (X && x) constructor for non-T.
- 查看 2 次提交的內容比對 »
4 月之前