Commit Graph

  • 89016fc88d Update Reorder Axes implementation comments Marshall Lochbaum 2023-03-29 20:23:16 -0400
  • 597e25af4d Full simplification pass for Reorder Axes Marshall Lochbaum 2023-03-29 20:08:36 -0400
  • d492cd0865 And strided height, so all orders are fast if axes are long Marshall Lochbaum 2023-03-29 11:31:52 -0400
  • e923a71881 Use strided width in transpose-based Reorder Axes Marshall Lochbaum 2023-03-29 08:22:15 -0400
  • 814a677676 Make transpose movement functions strided Marshall Lochbaum 2023-03-28 21:52:13 -0400
  • de18fb996b Save a shape object in Reorder Axes Marshall Lochbaum 2023-03-28 20:17:04 -0400
  • 7fea4ca2ad Combine transp_c2 temp allocations into one TALLOC Marshall Lochbaum 2023-03-28 16:32:41 -0400
  • 5b17ee44d6 SIMD kernel for Reorder Axes transposing last two axes Marshall Lochbaum 2023-03-27 22:12:12 -0400
  • 8413461074 Fast bottom-level loop for Reorder Axes loop structure Marshall Lochbaum 2023-03-27 20:44:31 -0400
  • bbedbbc585 Transpose large cells with mut_copyG Marshall Lochbaum 2023-03-26 22:28:53 -0400
  • b829076271 Put CPU-size transpose functions in a table Marshall Lochbaum 2023-03-25 15:48:53 -0400
  • 410d51501d Check for AVX2 in Singeli transpose so it can build on non-x86 architectures Marshall Lochbaum 2023-03-25 15:27:25 -0400
  • bf04b3aab0 Move w128i and similar from sse3 and avx.singeli to base Marshall Lochbaum 2023-03-26 20:33:24 -0400
  • ebbddf9dcf Move height/width 2 transposes to Singeli Marshall Lochbaum 2023-03-25 15:18:50 -0400
  • 37f481ea0d Separate out transpose data movement function and share with monadic ⍉⁼ Marshall Lochbaum 2023-03-25 10:43:35 -0400
  • 3c32844fe0 Separate out transpose_move function and use for el_B Marshall Lochbaum 2023-03-25 10:18:19 -0400
  • 968bfd2fc6 Separate transpose el_B case and fix dropped fill for shape 2‿n Marshall Lochbaum 2023-03-24 16:17:33 -0400
  • dea063cb7b Easy optimizations: (1-˜=𝕩)⍉𝕩 to ⍉𝕩 and detect trailing unchanged axes Marshall Lochbaum 2023-03-23 22:02:00 -0400
  • b555e3c035 Reorder Axes translated from runtime version Marshall Lochbaum 2023-03-23 20:56:25 -0400
  • 432b4eaaa6 Transpose implementation comments Marshall Lochbaum 2023-03-23 13:41:06 -0400
  • 21b9ad5c3c don't include debug.singeli dzaima 2023-03-27 17:53:01 +0300
  • 9780cacacd commented-out heap alignment dzaima 2023-03-27 17:38:26 +0300
  • ebd6d9d889 only check src directory for dirty flag for commit info dzaima 2023-03-27 15:33:49 +0300
  • f175bc9e19 fix printf of sample count dzaima 2023-03-26 15:57:09 +0300
  • 1f2456042c clean up calls.h includes dzaima 2023-03-26 15:53:04 +0300
  • 1f70c08aa2 fillarr NOGC dzaima 2023-03-26 15:36:29 +0300
  • 709c1896d3 better selfElType dzaima 2023-03-26 14:30:49 +0300
  • b02be62486 build/build-using makefile targets dzaima 2023-03-25 00:02:47 +0200
  • 37a32eb15a fix native non-Singeli builds dzaima 2023-03-24 23:51:18 +0200
  • 830c752e87 instruction pointer storing profiling dzaima 2023-03-24 18:07:48 +0200
  • e4c95dce22 fix profiler buffer size usage dzaima 2023-03-24 13:47:33 +0200
  • 98f4df88d0 whitespace dzaima 2023-03-23 18:35:55 +0200
  • c1c50a97b0 move moreCfgs.sh to build/build dzaima 2023-03-23 18:31:23 +0200
  • 0fc700b4f4
    Merge pull request #70 from mlochbaum/transpose dzaima 2023-03-23 18:24:21 +0200
  • fa8437ca25 don't use _mm256_loadu2_m128i dzaima 2023-03-23 18:21:34 +0200
  • 3178241d9a update Singeli, move "make forbuild" to "make for-build" dzaima 2023-03-23 18:08:28 +0200
  • 662fd83c40 fix typo dzaima 2023-03-23 17:58:50 +0200
  • 699faea69c Dedicated width-overflow loop for SIMD transpose Marshall Lochbaum 2023-03-22 17:30:02 -0400
  • 5a2bc15f2a Overlap SIMD transpose on width too Marshall Lochbaum 2023-03-22 15:13:18 -0400
  • 61eefe0776 Fix writing past end in i16 transpose with overlapped halves Marshall Lochbaum 2023-03-22 13:58:54 -0400
  • 8ff40cb2e5 Overlap SIMD transpose to handle uneven heights (not widths yet) Marshall Lochbaum 2023-03-22 11:35:57 -0400
  • fad7f3aa8b Simplify Marshall Lochbaum 2023-03-21 22:04:51 -0400
  • dd141add3f Merge i16 SIMD transpose with the rest of them Marshall Lochbaum 2023-03-21 22:01:21 -0400
  • 9673771187 Combine code for the different transpose kernel geometries Marshall Lochbaum 2023-03-21 21:52:31 -0400
  • e40feaa81c Make transpose kernel code type-generic Marshall Lochbaum 2023-03-21 19:17:00 -0400
  • 923c485cc2 Use packed-halves transpose kernel for 8x8 i16 Marshall Lochbaum 2023-03-21 18:54:46 -0400
  • 2a8e6e9c8a Separate most transpose kernel logic from loops and move i8 case into main Marshall Lochbaum 2023-03-21 18:42:43 -0400
  • 7d800bf349 SIMD transpose on 1-byte elements Marshall Lochbaum 2023-03-21 15:17:17 -0400
  • 5fccf4cda3 Refactoring; check for small transpose width/height in Singeli, not C Marshall Lochbaum 2023-03-21 11:26:44 -0400
  • ff6361e363 Move tranpose to its own file Marshall Lochbaum 2023-03-21 11:01:09 -0400
  • 0e2e315e51 Allow an 8x8 row for 2-byte transpose Marshall Lochbaum 2023-03-20 21:43:34 -0400
  • 239c765998 SIMD transpose on 2-byte elements Marshall Lochbaum 2023-03-20 20:57:52 -0400
  • f6c6e72661 More transpose kernel refactoring Marshall Lochbaum 2023-03-20 20:01:39 -0400
  • c0aaa6f615 SIMD transpose on 8-byte elements Marshall Lochbaum 2023-03-20 19:47:02 -0400
  • a0e85db702 Refactor vtranspose with more general code Marshall Lochbaum 2023-03-20 18:14:10 -0400
  • ca44d41daa Faster i32 transpose when height is a multiple of 16 by writing cache lines Marshall Lochbaum 2023-03-19 17:23:18 -0400
  • bfb2519aed Remove transpose_rec (was unused): the whole cache-oblivious thing seems to be useless Marshall Lochbaum 2023-03-19 17:18:09 -0400
  • 5c6075c29d support namespace a.b in native compiler dzaima 2023-03-22 19:50:01 +0200
  • dc583d754c separate FLDG for gid-based field lookup dzaima 2023-03-22 19:49:32 +0200
  • 66e82f078c include bootstrap utilities dzaima 2023-03-22 18:51:16 +0200
  • 39a0cb63c6 proper example mlochbaum/BQN path in readme dzaima 2023-03-22 18:00:47 +0200
  • 9337de5a0e fix LST_/ARM_ comments in vm.h dzaima 2023-03-21 23:04:39 +0200
  • d9ebdf060f )t:1s dzaima 2023-03-21 04:01:37 +0200
  • fe36c74080 support exponential for )t dzaima 2023-03-21 02:56:39 +0200
  • 1ec37410c6 faster constint|intarr dzaima 2023-03-21 02:45:42 +0200
  • d333902f74 fast integer 𝕨|𝕩 dzaima 2023-03-21 02:02:23 +0200
  • fb645e467a build.bqn: don't include options without docs in width calculation dzaima 2023-03-20 19:17:44 +0200
  • e14a3629f6 don't try lowercasing × dzaima 2023-03-20 18:18:16 +0200
  • b0201131e2 array type support for •FFI dzaima 2023-03-20 16:33:22 +0200
  • ca0c100b73 name normalization for •ns.Get & •ns.Has dzaima 2023-03-19 23:25:53 +0200
  • 0b32ef96ce inline → inplace in existing-scope-mutating function names dzaima 2023-03-19 22:47:00 +0200
  • d157fac17f use •ParseFloat instead of •BQN to parse numbers in build.bqn dzaima 2023-03-19 21:31:57 +0200
  • 4a6877a87a Rank checks dzaima 2023-03-19 15:42:11 +0200
  • afd4288cf2 check for hash success in makefile dzaima 2023-03-18 21:56:54 +0200
  • a3d74c6af2 MUT_APPEND dzaima 2023-03-16 22:27:44 +0200
  • 390ee503fb extract function for a+↕b dzaima 2023-03-16 22:05:44 +0200
  • a9b83ca70a V_POS, V_DEPTH dzaima 2023-03-16 22:03:08 +0200
  • db8a4d3af0 don't use def as a variable name dzaima 2023-03-12 00:11:30 +0200
  • 1178e5df56 new Singeli syntax dzaima 2023-03-11 23:17:30 +0200
  • 8194f51b30 MAKE_MUT_INIT, move mut_init to mut.c dzaima 2023-03-06 22:22:33 +0200
  • 3f0bc57503 move diagnostics coloring option outside of hash dzaima 2023-03-11 16:43:32 +0200
  • 01c457b77b make all cpy.*Arr function results Arr* dzaima 2023-03-06 22:21:22 +0200
  • 92763fa8b2 proper hash values for derived functions dzaima 2023-03-06 20:29:35 +0200
  • ab5b8f92b4 fast F˜˝˘ and F˜˝ and F˜⌜ for pervasive F dzaima 2023-03-05 16:48:10 +0200
  • 680d0f8601 use mut_copy for building fold_rows result dzaima 2023-03-05 15:20:12 +0200
  • 1c0b797456
    Merge pull request #69 from mlochbaum/transfold dzaima 2023-03-05 15:10:26 +0200
  • 98c963453b •internal.GC dzaima 2023-03-05 14:15:05 +0200
  • 4b6dd1bf81 immediately mark&sweep unknown reference objects instead of storing dzaima 2023-03-05 14:04:36 +0200
  • c5aedaa414 switch to GCv2, various fixes/cleanup dzaima 2023-03-04 20:40:56 +0200
  • f4d2e26e58 Don't do transposing fold unless height is greater than width Marshall Lochbaum 2023-03-04 14:42:00 -0500
  • 176152b445 Handle fold-cells on row length 1 with ⊏˘ Marshall Lochbaum 2023-03-04 14:28:22 -0500
  • 4b35acb82b Transpose-based fold-cells implementation Marshall Lochbaum 2023-03-04 14:24:56 -0500
  • 16274e5952 bucket coalescing on GC dzaima 2023-03-04 20:09:12 +0200
  • ead637b135 don't leak c1fn/c2fn arguments from array function dzaima 2023-03-04 18:58:01 +0200
  • 717074a2f8 fast transpose on shape n‿2 𝕩 dzaima 2023-03-03 18:53:18 +0200
  • 76d26db4c2 faster F˘´n‿2⥊𝕩 for pervasive F dzaima 2023-03-03 17:45:16 +0200
  • fd72d416db manual prefix character typing detecting dzaima 2023-03-03 15:54:03 +0200
  • 093beda394 faster path for 𝕨/𝕩 with non-boolean 𝕨 with sum 0 dzaima 2023-03-03 13:27:42 +0200
  • 397f8745ce go back to regular completions if no system commands match dzaima 2023-02-27 23:51:35 +0200
  • dfd78115c3 remove repl initialization before it's needed dzaima 2023-02-26 18:58:08 +0200