2025-08-14 23:02.30: New job: test ahrefs/ocannl https://github.com/ahrefs/ocannl.git#refs/heads/master (8923c9bfb01efddeefcde961b2cca56db40e5a8a) (linux-x86_64:(lint-fmt)) Base: ocaml/opam:debian-12-ocaml-4.08@sha256:474656ea1593a299054f8966c700443fa0944c9534de3da94ca6dfab4a44c47a ocamlformat version: version 0.27.0 (from opam) To reproduce locally: git clone --recursive "https://github.com/ahrefs/ocannl.git" -b "master" && cd "ocannl" && git reset --hard 8923c9bf cat > Dockerfile <<'END-OF-DOCKERFILE' FROM ocaml/opam:debian-12-ocaml-4.08@sha256:474656ea1593a299054f8966c700443fa0944c9534de3da94ca6dfab4a44c47a USER 1000:1000 RUN cd ~/opam-repository && (git cat-file -e b8021439f8c57ba6435bc2263f6596671f4f4466 || git fetch origin master) && git reset -q --hard b8021439f8c57ba6435bc2263f6596671f4f4466 && git log --no-decorate -n1 --oneline && opam update -u RUN opam depext -i dune WORKDIR /src RUN opam depext -i ocamlformat=0.27.0 COPY --chown=1000:1000 . /src/ RUN opam exec -- dune build @fmt --ignore-promoted-rules || (echo "dune build @fmt failed"; exit 2) END-OF-DOCKERFILE docker build . END-REPRO-BLOCK 2025-08-14 23:02.30: Using cache hint "ahrefs/ocannl-ocaml/opam:debian-12-ocaml-4.08@sha256:474656ea1593a299054f8966c700443fa0944c9534de3da94ca6dfab4a44c47a-debian-12-4.08_opam-2.4-ocamlformat-b8021439f8c57ba6435bc2263f6596671f4f4466" 2025-08-14 23:02.30: Using OBuilder spec: ((from ocaml/opam:debian-12-ocaml-4.08@sha256:474656ea1593a299054f8966c700443fa0944c9534de3da94ca6dfab4a44c47a) (user (uid 1000) (gid 1000)) (run (cache (opam-archives (target /home/opam/.opam/download-cache))) (network host) (shell "cd ~/opam-repository && (git cat-file -e b8021439f8c57ba6435bc2263f6596671f4f4466 || git fetch origin master) && git reset -q --hard b8021439f8c57ba6435bc2263f6596671f4f4466 && git log --no-decorate -n1 --oneline && opam update -u")) (run (cache (opam-archives (target /home/opam/.opam/download-cache))) (network host) (shell "opam depext -i dune")) (workdir /src) (run (cache (opam-archives (target /home/opam/.opam/download-cache))) (network host) (shell "opam depext -i ocamlformat=0.27.0")) (copy (src .) (dst /src/)) (run (shell "opam exec -- dune build @fmt --ignore-promoted-rules || (echo \"dune build @fmt failed\"; exit 2)")) ) 2025-08-14 23:02.30: Waiting for resource in pool OCluster 2025-08-14 23:02.30: Waiting for worker… 2025-08-14 23:02.30: Got resource from pool OCluster Building on eumache HEAD is now at 075bc6a5 The previous change is_bijective -> is_surjective was a mistake, but we can do better altogether, collab with Claude HEAD is now at 8923c9bf Fix the most recent changes to lowering accumulations (from ocaml/opam:debian-12-ocaml-4.08@sha256:474656ea1593a299054f8966c700443fa0944c9534de3da94ca6dfab4a44c47a) 2025-08-14 23:02.31 ---> using "1d0024db739bd078f91b2384c47919652a4b72a425e3e24ce24cfd1f6debdfbc" from cache /: (user (uid 1000) (gid 1000)) /: (run (cache (opam-archives (target /home/opam/.opam/download-cache))) (network host) (shell "cd ~/opam-repository && (git cat-file -e b8021439f8c57ba6435bc2263f6596671f4f4466 || git fetch origin master) && git reset -q --hard b8021439f8c57ba6435bc2263f6596671f4f4466 && git log --no-decorate -n1 --oneline && opam update -u")) b8021439f8 Merge pull request #28261 from kit-ty-kate/deploy-fix-25819 <><> Updating package repositories ><><><><><><><><><><><><><><><><><><><><><><> [default] Initialised default (at git+file:///home/opam/opam-repository): [INFO] opam 2.1 and 2.2 include many performance and security improvements over 2.0; please consider upgrading (https://opam.ocaml.org/doc/Install.html) Everything as up-to-date as possible (run with --verbose to show unavailable upgrades). However, you may "opam upgrade" these packages explicitly, which will ask permission to downgrade or uninstall the conflicting packages. Nothing to do. # Run eval $(opam env) to update the current shell environment 2025-08-14 23:02.31 ---> using "76d9d96bb26da3c78200d383fd35f876d80571baf05962331a1fff5f47db0e2e" from cache /: (run (cache (opam-archives (target /home/opam/.opam/download-cache))) (network host) (shell "opam depext -i dune")) # Detecting depexts using vars: arch=x86_64, os=linux, os-distribution=debian, os-family=debian # No extra OS packages requirements found. # All required OS packages found. # Now letting opam install the packages The following actions will be performed: - install dune 3.19.1 <><> Gathering sources ><><><><><><><><><><><><><><><><><><><><><><><><><><><><> [dune.3.19.1] found in cache <><> Processing actions <><><><><><><><><><><><><><><><><><><><><><><><><><><><> -> installed dune.3.19.1 Done. # Run eval $(opam env) to update the current shell environment 2025-08-14 23:02.31 ---> using "da0888a20a067de19f6183f0b497dcc2d1ea7f7036861cc50f633c662efdce4f" from cache /: (workdir /src) /src: (run (cache (opam-archives (target /home/opam/.opam/download-cache))) (network host) (shell "opam depext -i ocamlformat=0.27.0")) # Detecting depexts using vars: arch=x86_64, os=linux, os-distribution=debian, os-family=debian # No extra OS packages requirements found. # All required OS packages found. # Now letting opam install the packages The following actions will be performed: - install sexplib0 v0.14.0 [required by base] - install menhirLib 20240715 [required by ocamlformat-lib] - install menhirCST 20240715 [required by menhir] - install ocamlbuild 0.16.1 [required by fpath, astring, uuseg] - install cmdliner 1.3.0 [required by ocamlformat] - install menhirSdk 20240715 [required by ocamlformat-lib] - install either 1.0.0 [required by ocamlformat-lib] - install csexp 1.5.2 [required by ocamlformat] - install dune-build-info 3.19.1 [required by ocamlformat-lib] - install camlp-streams 5.0.1 [required by ocamlformat-lib] - install seq base [required by re] - install fix 20250428 [required by ocamlformat-lib] - install ocaml-version 4.0.1 [required by ocamlformat-lib] - install ocamlfind 1.9.8 [required by ocp-indent, astring, fpath, uuseg] - install menhir 20240715 [required by ocamlformat-lib] - install dune-configurator 3.19.1 [required by base] - install re 1.11.0 [required by ocamlformat] - install topkg 1.1.0 [required by fpath, astring, uuseg] - install base-bytes base [required by ocp-indent] - install base v0.14.3 [required by ocamlformat-lib] - install uutf 1.0.4 [required by ocamlformat-lib] - install astring 0.8.5 [required by ocamlformat-lib] - install ocp-indent 1.8.1 [required by ocamlformat-lib] - install stdio v0.14.0 [required by ocamlformat-lib] - install uucp 15.0.0 [required by uuseg] - install fpath 0.7.3 [required by ocamlformat-lib] - install uuseg 15.0.0 [required by ocamlformat-lib] - install ocamlformat-lib 0.27.0 [required by ocamlformat] - install ocamlformat 0.27.0 ===== 29 to install ===== <><> Gathering sources ><><><><><><><><><><><><><><><><><><><><><><><><><><><><> [astring.0.8.5] found in cache [base.v0.14.3] found in cache [camlp-streams.5.0.1] found in cache [cmdliner.1.3.0] found in cache [csexp.1.5.2] found in cache [dune-build-info.3.19.1] found in cache [dune-configurator.3.19.1] found in cache [either.1.0.0] found in cache [fix.20250428] found in cache [fpath.0.7.3] found in cache [menhir.20240715] found in cache [menhirCST.20240715] found in cache [menhirLib.20240715] found in cache [menhirSdk.20240715] found in cache [ocaml-version.4.0.1] found in cache [ocamlbuild.0.16.1] found in cache [ocamlfind.1.9.8] found in cache [ocamlformat.0.27.0] found in cache [ocamlformat-lib.0.27.0] found in cache [ocp-indent.1.8.1] found in cache [re.1.11.0] found in cache [sexplib0.v0.14.0] found in cache [stdio.v0.14.0] found in cache [topkg.1.1.0] found in cache [uucp.15.0.0] found in cache [uuseg.15.0.0] found in cache [uutf.1.0.4] found in cache <><> Processing actions <><><><><><><><><><><><><><><><><><><><><><><><><><><><> -> installed seq.base -> installed camlp-streams.5.0.1 -> installed csexp.1.5.2 -> installed either.1.0.0 -> installed fix.20250428 -> installed cmdliner.1.3.0 -> installed menhirCST.20240715 -> installed menhirLib.20240715 -> installed menhirSdk.20240715 -> installed ocaml-version.4.0.1 -> installed re.1.11.0 -> installed sexplib0.v0.14.0 -> installed dune-build-info.3.19.1 -> installed dune-configurator.3.19.1 -> installed ocamlfind.1.9.8 -> installed base-bytes.base -> installed ocp-indent.1.8.1 -> installed ocamlbuild.0.16.1 -> installed base.v0.14.3 -> installed stdio.v0.14.0 -> installed topkg.1.1.0 -> installed uutf.1.0.4 -> installed astring.0.8.5 -> installed fpath.0.7.3 -> installed menhir.20240715 -> installed uucp.15.0.0 -> installed uuseg.15.0.0 -> installed ocamlformat-lib.0.27.0 -> installed ocamlformat.0.27.0 Done. <><> ocp-indent.1.8.1 installed successfully ><><><><><><><><><><><><><><><><><> => This package requires additional configuration for use in editors. Install package 'user-setup', or manually: * for Emacs, add these lines to ~/.emacs: (add-to-list 'load-path "/home/opam/.opam/4.08/share/emacs/site-lisp") (require 'ocp-indent) * for Vim, add this line to ~/.vimrc: set rtp^="/home/opam/.opam/4.08/share/ocp-indent/vim" # Run eval $(opam env) to update the current shell environment 2025-08-14 23:02.31 ---> using "86ec8dcb8046a1e5dacfb1841e8c026d30cfead67649bcb6d6a8f9ddd6fb153d" from cache /src: (copy (src .) (dst /src/)) 2025-08-14 23:02.32 ---> saved as "33448d827dedee7c96d6086b752ca02dcdd51619f7698250f644cf45c686cbfe" /src: (run (shell "opam exec -- dune build @fmt --ignore-promoted-rules || (echo \"dune build @fmt failed\"; exit 2)")) File "arrayjit/bin/dune", line 6, characters 30-43: 6 | (pps ppx_here ppx_minidebug ppx_sexp_conv)) ^^^^^^^^^^^^^ Error: Library "ppx_sexp_conv" not found. -> required by _build/default/arrayjit/bin/read_config.exe -> required by %{dep:../../../arrayjit/bin/read_config.exe} at test/operations/dune:17 -> required by _build/default/test/operations/config/ocannl_backend.txt -> required by %{read:config/ocannl_backend.txt} at test/operations/dune:33 -> required by Computing directory contents of _build/default/test/operations File "arrayjit/bin/dune", line 6, characters 7-15: 6 | (pps ppx_here ppx_minidebug ppx_sexp_conv)) ^^^^^^^^ Error: Library "ppx_here" not found. -> required by _build/default/arrayjit/bin/.merlin-conf/exe-read_config -> required by _build/default/arrayjit/bin/read_config.exe -> required by %{dep:../../../arrayjit/bin/read_config.exe} at test/operations/dune:17 -> required by _build/default/test/operations/config/ocannl_backend.txt -> required by %{read:config/ocannl_backend.txt} at test/operations/dune:33 -> required by Computing directory contents of _build/default/test/operations File "test/einsum/surjectivity.ml", line 1, characters 0-0: diff --git a/_build/default/test/einsum/surjectivity.ml b/_build/default/test/einsum/.formatted/surjectivity.ml index e1cdc46..d5a8079 100644 --- a/_build/default/test/einsum/surjectivity.ml +++ b/_build/default/test/einsum/.formatted/surjectivity.ml @@ -22,14 +22,15 @@ let%expect_test "diagonal_tensor_initialization" = (* Create a diagonal tensor using einsum: i->ii *) let input = TDSL.range 5 in let%op diagonal = input ++ "i=>ii" in - + (* Ensure the diagonal tensor is hosted *) Train.set_hosted diagonal.value; ignore (Train.forward_once backend diagonal); - + (* Print the diagonal tensor *) Train.printf ~here:[%here] ~with_code:false ~with_grad:false diagonal; - [%expect {| + [%expect + {| HERE: test/einsum/surjectivity.ml:31:21 ┌──────────────────────────────────────┐ │[1]: =>_diagonal shape 0:6,1:6 │ @@ -60,12 +61,13 @@ let%expect_test "sparse_assignment_with_fixed_indices" = (* Create a sparse tensor using fixed indices: i->i0j *) let input = TDSL.range 4 in let%op sparse = input ++ "i=>i0j" in - + Train.set_hosted sparse.value; ignore (Train.forward_once backend sparse); - + Train.printf ~here:[%here] ~with_code:false ~with_grad:false sparse; - [%expect {| + [%expect + {| HERE: test/einsum/surjectivity.ml:67:21 ┌─────────────────────────────────┐ │[1]: =>_sparse shape 0:5,1:1,2:1 │ @@ -103,14 +105,15 @@ let%expect_test "multiple_sparse_axes" = in (* Test with multiple fixed indices: ij->i1j2 *) - let input = TDSL.range_of_shape ~output_dims:[3; 4] () in + let input = TDSL.range_of_shape ~output_dims:[ 3; 4 ] () in let%op sparse_multi = input ++ "ij=>i1j2" in - + Train.set_hosted sparse_multi.value; ignore (Train.forward_once backend sparse_multi); - + Train.printf ~here:[%here] ~with_code:false ~with_grad:false sparse_multi; - [%expect {| + [%expect + {| HERE: test/einsum/surjectivity.ml:112:21 ┌───────────────────────────────────────────┐ │[1]: =>_sparse_multi shape 0:3,1:2,2:4,3:3 │ @@ -156,4 +159,4 @@ let%expect_test "multiple_sparse_axes" = ││ │ 0.00 0.00 1.10e+1 │ │ │└──────┴─────────────────────┘ │ └───────────────────────────────────────────┘ - |}] \ No newline at end of file + |}] File "arrayjit/lib/backend_impl.ml", line 1, characters 0-0: diff --git a/_build/default/arrayjit/lib/backend_impl.ml b/_build/default/arrayjit/lib/.formatted/backend_impl.ml index 2b853fb..c867f49 100644 --- a/_build/default/arrayjit/lib/backend_impl.ml +++ b/_build/default/arrayjit/lib/.formatted/backend_impl.ml @@ -54,11 +54,11 @@ module No_device_buffer_and_copying () : (* FIXME(#344, #355): This is not efficient, but it won't be used for long. *) let ptr = alloc_impl ~size_in_bytes in (* Zero-initialize the allocated memory *) - if size_in_bytes > 0 then ( - let arr = Ctypes.from_voidp Ctypes.uint8_t ptr in - for i = 0 to size_in_bytes - 1 do - Ctypes.(arr +@ i <-@ Unsigned.UInt8.zero) - done); + (if size_in_bytes > 0 then + let arr = Ctypes.from_voidp Ctypes.uint8_t ptr in + for i = 0 to size_in_bytes - 1 do + Ctypes.(arr +@ i <-@ Unsigned.UInt8.zero) + done); ptr let%track7_sexp alloc_buffer ?(old_buffer : buffer_ptr Backend_intf.buffer option) File "test/einsum/test_surjectivity.ml", line 1, characters 0-0: diff --git a/_build/default/test/einsum/test_surjectivity.ml b/_build/default/test/einsum/.formatted/test_surjectivity.ml index 0c225ed..1bb021e 100644 --- a/_build/default/test/einsum/test_surjectivity.ml +++ b/_build/default/test/einsum/.formatted/test_surjectivity.ml @@ -11,15 +11,14 @@ let test_diagonal_tensor () = Stdio.printf "\nTesting diagonal tensor initialization:\n"; Tensor.unsafe_reinitialize (); let module Backend = (val Backends.fresh_backend ()) in - (* Create a diagonal tensor using einsum: i->ii *) let input = TDSL.range 5 in let%op diagonal = input ++ "i=>ii" in - + (* Ensure the diagonal tensor is hosted *) Train.set_hosted diagonal.value; ignore (Train.forward_once (module Backend) diagonal); - + (* Print the diagonal tensor *) Train.printf ~here:[%here] ~with_code:false ~with_grad:false diagonal; Stdio.printf "\n" @@ -28,14 +27,13 @@ let test_sparse_fixed_index () = Stdio.printf "\nTesting sparse assignment with fixed index:\n"; Tensor.unsafe_reinitialize (); let module Backend = (val Backends.fresh_backend ()) in - (* Create a sparse tensor using fixed indices: i->i0j *) let input = TDSL.range 4 in let%op sparse = input ++ "i=>i0" in - + Train.set_hosted sparse.value; ignore (Train.forward_once (module Backend) sparse); - + Train.printf ~here:[%here] ~with_code:false ~with_grad:false sparse; Stdio.printf "\n" @@ -43,31 +41,29 @@ let test_multi_sparse () = Stdio.printf "\nTesting multiple sparse axes:\n"; Tensor.unsafe_reinitialize (); let module Backend = (val Backends.fresh_backend ()) in - (* Create tensor with multiple dims and test sparse assignment *) - let input = TDSL.range_of_shape ~output_dims:[3; 4] () in + let input = TDSL.range_of_shape ~output_dims:[ 3; 4 ] () in let%op result = input ++ "ij=>i1j" in - + Train.set_hosted result.value; ignore (Train.forward_once (module Backend) result); - + Train.printf ~here:[%here] ~with_code:false ~with_grad:false result; Stdio.printf "\n" - let _test_stride_gap () = - Stdio.printf "\nTesting stride gap:\n"; - Tensor.unsafe_reinitialize (); - let module Backend = (val Backends.fresh_backend ()) in - - (* Create tensor with multiple dims and test sparse assignment *) - let input = TDSL.range_of_shape ~output_dims:[2; 5] () in - let%op result = input ++ "ij=>i+3*j" in - - Train.set_hosted result.value; - ignore (Train.forward_once (module Backend) result); - - Train.printf ~here:[%here] ~with_code:false ~with_grad:false result; - Stdio.printf "\n" +let _test_stride_gap () = + Stdio.printf "\nTesting stride gap:\n"; + Tensor.unsafe_reinitialize (); + let module Backend = (val Backends.fresh_backend ()) in + (* Create tensor with multiple dims and test sparse assignment *) + let input = TDSL.range_of_shape ~output_dims:[ 2; 5 ] () in + let%op result = input ++ "ij=>i+3*j" in + + Train.set_hosted result.value; + ignore (Train.forward_once (module Backend) result); + + Train.printf ~here:[%here] ~with_code:false ~with_grad:false result; + Stdio.printf "\n" let () = test_diagonal_tensor (); @@ -75,4 +71,4 @@ let () = test_multi_sparse (); (* FIXME(#354): Projections inference for convolution-style expressions not implemented yet. *) (* test_stride_gap (); *) - Stdio.printf "All surjectivity tests completed.\n" \ No newline at end of file + Stdio.printf "All surjectivity tests completed.\n" File "test/einsum/test_accumulation_semantics.ml", line 1, characters 0-0: diff --git a/_build/default/test/einsum/test_accumulation_semantics.ml b/_build/default/test/einsum/.formatted/test_accumulation_semantics.ml index 54307b7..6925d41 100644 --- a/_build/default/test/einsum/test_accumulation_semantics.ml +++ b/_build/default/test/einsum/.formatted/test_accumulation_semantics.ml @@ -12,28 +12,26 @@ let test_einsum_reduction () = Stdio.printf "\n=== Testing einsum reduction (surjective but not injective) ===\n"; Tensor.unsafe_reinitialize (); let module Backend = (val Backends.fresh_backend ()) in - (* Create input tensor with shape [batch=2, input=3, output=4] *) let input = TDSL.range_of_shape ~batch_dims:[ 2 ] ~input_dims:[ 3 ] ~output_dims:[ 4 ] () in - - (* The einsum "b|i->o => b|i" reduces over the output dimension - This projection is: - - Surjective: all positions in result tensor get written - - NOT injective: multiple source positions map to same target (needs accumulation) *) + + (* The einsum "b|i->o => b|i" reduces over the output dimension This projection is: - Surjective: + all positions in result tensor get written - NOT injective: multiple source positions map to + same target (needs accumulation) *) let%op result = input ++ "b|i->o => b|i" in - + Train.set_hosted input.value; Train.set_hosted result.value; Train.every_non_literal_on_host result; - + ignore (Train.forward_once (module Backend) result); - + Stdio.printf "Input tensor (shape: batch=2, input=3, output=4):\n"; Train.printf ~here:[%here] ~with_code:false ~with_grad:false input; - + Stdio.printf "\nResult after reduction 'b|i->o => b|i' (should sum over output dimension):\n"; Train.printf ~here:[%here] ~with_code:false ~with_grad:false result; - + (* Verify the accumulation is correct *) Stdio.printf "\nExpected values (summing over output dimension):\n"; Stdio.printf " Batch 0: [0+3+6+9=18, 1+4+7+10=22, 2+5+8+11=26]\n"; @@ -44,23 +42,21 @@ let test_diagonal_tensor () = Stdio.printf "\n=== Testing diagonal tensor (not surjective) ===\n"; Tensor.unsafe_reinitialize (); let module Backend = (val Backends.fresh_backend ()) in - - (* Create a diagonal tensor using einsum: i=>ii - This projection is: - - NOT surjective: off-diagonal positions never get written (need Zero_out) - - Injective: each source position maps to exactly one target *) + (* Create a diagonal tensor using einsum: i=>ii This projection is: - NOT surjective: off-diagonal + positions never get written (need Zero_out) - Injective: each source position maps to exactly + one target *) let input = TDSL.range 5 in let%op diagonal = input ++ "i=>ii" in - + Train.set_hosted diagonal.value; ignore (Train.forward_once (module Backend) diagonal); - + Stdio.printf "Input (1D tensor of size 5):\n"; Train.printf ~here:[%here] ~with_code:false ~with_grad:false input; - + Stdio.printf "\nDiagonal tensor 'i=>ii' (5x5 with zeros off-diagonal):\n"; Train.printf ~here:[%here] ~with_code:false ~with_grad:false diagonal; - + Stdio.printf "\nNote: Off-diagonal elements should be zero (initialized by Zero_out)\n" (** Test fixed index projection (not surjective) *) @@ -68,24 +64,22 @@ let test_fixed_index_projection () = Stdio.printf "\n=== Testing fixed index projection (not surjective) ===\n"; Tensor.unsafe_reinitialize (); let module Backend = (val Backends.fresh_backend ()) in - - (* Create a sparse tensor using fixed index: i=>i0 - This projection is: - - NOT surjective: only column 0 gets written (need Zero_out for other columns) - - Injective: each source position maps to exactly one target *) + (* Create a sparse tensor using fixed index: i=>i0 This projection is: - NOT surjective: only + column 0 gets written (need Zero_out for other columns) - Injective: each source position maps + to exactly one target *) let input = TDSL.range 4 in let%op sparse = input ++ "i=>i0" in let%op _ = sparse ++ "i2=>i" in - + Train.set_hosted sparse.value; ignore (Train.forward_once (module Backend) sparse); - + Stdio.printf "Input (1D tensor of size 4):\n"; Train.printf ~here:[%here] ~with_code:false ~with_grad:false input; - + Stdio.printf "\nSparse tensor 'i=>i0' (only first column populated):\n"; Train.printf ~here:[%here] ~with_code:false ~with_grad:false sparse; - + Stdio.printf "\nNote: Only column 0 should have values, others should be zero\n" (** Test bijective mapping (no initialization or accumulation needed) *) @@ -93,24 +87,21 @@ let test_bijective_transpose () = Stdio.printf "\n=== Testing bijective transpose (optimization case) ===\n"; Tensor.unsafe_reinitialize (); let module Backend = (val Backends.fresh_backend ()) in - - (* Simple transpose: ij=>ji - This projection is: - - Surjective: all target positions get written - - Injective: each source maps to exactly one target - - Therefore BIJECTIVE: can skip both Zero_out and accumulation *) - let input = TDSL.range_of_shape ~output_dims:[3; 4] () in + (* Simple transpose: ij=>ji This projection is: - Surjective: all target positions get written - + Injective: each source maps to exactly one target - Therefore BIJECTIVE: can skip both Zero_out + and accumulation *) + let input = TDSL.range_of_shape ~output_dims:[ 3; 4 ] () in let%op transposed = input ++ "ij=>ji" in - + Train.set_hosted transposed.value; ignore (Train.forward_once (module Backend) transposed); - + Stdio.printf "Input (3x4 matrix):\n"; Train.printf ~here:[%here] ~with_code:false ~with_grad:false input; - + Stdio.printf "\nTransposed 'ij=>ji' (4x3 matrix):\n"; Train.printf ~here:[%here] ~with_code:false ~with_grad:false transposed; - + Stdio.printf "\nNote: Simple bijective mapping - no initialization or accumulation needed\n" let () = @@ -118,4 +109,4 @@ let () = test_diagonal_tensor (); test_fixed_index_projection (); test_bijective_transpose (); - Stdio.printf "\n=== All accumulation semantics tests completed ===\n" \ No newline at end of file + Stdio.printf "\n=== All accumulation semantics tests completed ===\n" File "arrayjit/lib/indexing.ml", line 1, characters 0-0: diff --git a/_build/default/arrayjit/lib/indexing.ml b/_build/default/arrayjit/lib/.formatted/indexing.ml index 38c40f2..9917ec5 100644 --- a/_build/default/arrayjit/lib/indexing.ml +++ b/_build/default/arrayjit/lib/.formatted/indexing.ml @@ -218,7 +218,7 @@ let is_surjective proj = let is_injective proj = let product_iterator_set = Set.of_array (module Symbol) proj.product_iterators in - + (* Check each LHS index for injectivity *) let lhs_symbols, is_injective_mapping = Array.fold proj.project_lhs ~init:([], true) ~f:(fun (syms, still_injective) idx -> @@ -229,19 +229,17 @@ let is_injective proj = | Fixed_idx _ -> (syms, true) | Affine { symbols; _ } -> (* Filter for symbols that are product iterators *) - let product_symbols = - List.filter symbols ~f:(fun (_coeff, s) -> - Set.mem product_iterator_set s) + let product_symbols = + List.filter symbols ~f:(fun (_coeff, s) -> Set.mem product_iterator_set s) in (* If more than one product iterator in this Affine index, not injective *) - if List.length product_symbols > 1 then - (syms, false) + if List.length product_symbols > 1 then (syms, false) else (* (coefficients don't matter for injectivity) *) (List.map product_symbols ~f:snd @ syms, true) | Sub_axis -> (syms, true)) in - + if not is_injective_mapping then false else let lhs_symbol_set = Set.of_list (module Symbol) lhs_symbols in File "arrayjit/lib/assignments.ml", line 1, characters 0-0: diff --git a/_build/default/arrayjit/lib/assignments.ml b/_build/default/arrayjit/lib/.formatted/assignments.ml index 77ab907..85d166f 100644 --- a/_build/default/arrayjit/lib/assignments.ml +++ b/_build/default/arrayjit/lib/.formatted/assignments.ml @@ -247,14 +247,12 @@ let%track4_sexp to_low_level code = } in let for_loops = for_loop [] (Array.to_list projections.product_space) in - (* Need initialization if: - - initialize_neutral is true AND - - (not surjective OR not injective) - Not surjective: some positions never written (need init to avoid garbage) - Not injective: accumulation needed (need init for first += operation) *) - let needs_init = - initialize_neutral && - not (Indexing.is_surjective projections && Indexing.is_injective projections) + (* Need initialization if: - initialize_neutral is true AND - (not surjective OR not injective) + Not surjective: some positions never written (need init to avoid garbage) Not injective: + accumulation needed (need init for first += operation) *) + let needs_init = + initialize_neutral + && not (Indexing.is_surjective projections && Indexing.is_injective projections) in if needs_init then let dims = lazy projections.lhs_dims in @@ -341,7 +339,8 @@ let%track4_sexp to_low_level code = | Fetch { array; fetch_op = Constant c; dims } -> Low_level.loop_over_dims (Lazy.force dims) ~body:(fun idcs -> set array idcs @@ Constant c) | Fetch { array; fetch_op = Constant_bits i; dims } -> - Low_level.loop_over_dims (Lazy.force dims) ~body:(fun idcs -> set array idcs @@ Constant_bits i) + Low_level.loop_over_dims (Lazy.force dims) ~body:(fun idcs -> + set array idcs @@ Constant_bits i) | Fetch { array; fetch_op = Slice { batch_idx = { static_symbol = idx; _ }; sliced }; dims } -> (* TODO: doublecheck this always gets optimized away. *) Low_level.loop_over_dims (Lazy.force dims) ~body:(fun idcs -> File "lib/ppx_op.ml", line 1, characters 0-0: diff --git a/_build/default/lib/ppx_op.ml b/_build/default/lib/.formatted/ppx_op.ml index 59f0690..32726dc 100644 --- a/_build/default/lib/ppx_op.ml +++ b/_build/default/lib/.formatted/ppx_op.ml @@ -89,10 +89,7 @@ let rec translate ~num_configs ~is_toplevel ~has_config ?label expr = let axis = Ast_helper.Exp.constant ~loc:pexp_loc (Pconst_string (String.of_char ch, pexp_loc, None)) in - ( no_vbs, - [%expr - TDSL.bits ?label:[%e opt_expr ~loc label] ~axis_label:[%e axis] [%e i]] - ) + (no_vbs, [%expr TDSL.bits ?label:[%e opt_expr ~loc label] ~axis_label:[%e axis] [%e i]]) | [%expr [%e? { pexp_desc = Pexp_constant (Pconst_char ch); pexp_loc; _ }] [%e? { pexp_desc = Pexp_constant (Pconst_integer _); _ } as i]] -> File "lib/tensor.ml", line 1, characters 0-0: diff --git a/_build/default/lib/tensor.ml b/_build/default/lib/.formatted/tensor.ml index b4df5d4..79ee104 100644 --- a/_build/default/lib/tensor.ml +++ b/_build/default/lib/.formatted/tensor.ml @@ -477,8 +477,8 @@ let%track7_sexp term ?init_data ?fetch_op ?grad_spec ?(label = []) ?(top_down_pr match fetch_op with | None -> Asgns.empty_comp | Some - (( Constant _ | Constant_bits _ | Slice _ | Embed_symbol _ | Embed_self_id | Range_over_offsets - | Constant_fill _ ) as fetch_op) -> + (( Constant _ | Constant_bits _ | Slice _ | Embed_symbol _ | Embed_self_id + | Range_over_offsets | Constant_fill _ ) as fetch_op) -> Asgns.to_comp @@ Fetch { array = v; fetch_op; dims } in let grad_asn ~t:_ ~g:_ ~projections:_ = Asgns.empty_comp in File "arrayjit/lib/cuda_backend.ml", line 1, characters 0-0: diff --git a/_build/default/arrayjit/lib/cuda_backend.ml b/_build/default/arrayjit/lib/.formatted/cuda_backend.ml index a626193..72d9c69 100644 --- a/_build/default/arrayjit/lib/cuda_backend.ml +++ b/_build/default/arrayjit/lib/.formatted/cuda_backend.ml @@ -351,7 +351,9 @@ end) : Ir.Backend_impl.Lowered_backend = struct (string "hexp2(hlog2(" ^^ v1 ^^ string ")," ^^ ifflat (space ^^ v2) (nest 2 (break 1 ^^ v2)) ^^ string ")") - | ToPowOf, (Byte_prec _ | Uint16_prec _ | Int32_prec _ | Int64_prec _ | Fp8_prec _ | Uint4x32_prec _) -> + | ( ToPowOf, + (Byte_prec _ | Uint16_prec _ | Int32_prec _ | Int64_prec _ | Fp8_prec _ | Uint4x32_prec _) + ) -> invalid_arg "Cuda_backend.binop_syntax: ToPowOf not supported for integer precisions" | ToPowOf, Bfloat16_prec _ -> fun v1 v2 -> @@ -666,7 +668,7 @@ end) : Ir.Backend_impl.Lowered_backend = struct let vec_unop_syntax prec op v = let open PPrint in match (op, prec) with - | Ops.Uint4x32_to_prec_uniform, _ -> + | Ops.Uint4x32_to_prec_uniform, _ -> group (string ("uint4x32_to_" ^ Ops.prec_string prec ^ "_uniform_vec(") ^^ v ^^ rparen) let ternop_syntax prec v = @@ -698,7 +700,7 @@ end) : Ir.Backend_impl.Lowered_backend = struct | Single_prec _, Half_prec _ -> ("__float2half(", ")") | Byte_prec _, Half_prec _ -> ("__ushort2half_rn((unsigned short int)", ")") | Double_prec _, Uint4x32_prec _ -> ("{(unsigned int)(", "), 0, 0, 0}") - | Single_prec _, Uint4x32_prec _ -> ("{(unsigned int)(", "), 0, 0, 0}") + | Single_prec _, Uint4x32_prec _ -> ("{(unsigned int)(", "), 0, 0, 0}") | Int32_prec _, Uint4x32_prec _ -> ("{(unsigned int)(", "), 0, 0, 0}") | Int64_prec _, Uint4x32_prec _ -> ("int64_to_uint4x32(", ")") | Uint4x32_prec _, _ -> ("", ".v[0]") dune build @fmt failed "/usr/bin/env" "bash" "-c" "opam exec -- dune build @fmt --ignore-promoted-rules || (echo "dune build @fmt failed"; exit 2)" failed with exit status 2 2025-08-14 23:02.33: Job failed: Failed: Build failed