futhark

Commits

• 1c921c3: Mention #2009. (Troels Henriksen)
• 2666a6e: Fix #1993. (Troels Henriksen)
• 1dc25af: Avoid duplication of existential sizes. (Troels Henriksen)

Commits

• c665832: Clarify. (Troels Henriksen)

Commits

• 8fc7e8f: Start on another useful section. (Troels Henriksen)

Commits

• a5e1700: Restore ISPC backend. (Troels Henriksen)

Commits

• 960968e: Define this centrally. (Troels Henriksen)

Commits

• cb567eb: Switch away from deprecated names. (Troels Henriksen)

Commits

• ef2387b: mapOnName is actually always on a QualName. (Troels Henriksen)

Commits

• 01a4f23: Do not astMap names in binding position. (Troels Henriksen)

Commits

• e17e260: Another #1997-related fix. (Troels Henriksen)

Commits

• 55a17e3: Add another small simplification rule. (Troels Henriksen)

Commits

• 3542b4a: Better prettyprinting of ModExp. (Troels Henriksen)
• 79d4c74: Fix #1992. (Troels Henriksen)

Added

• Flattening/unflattening as the final operation in an entry point no
  only forces a copy.

• The opencl backend no longer always fails on platforms that do
  not support 64-bit integer atomics, although it will still fail if
  the program needs them.

• Various performance improvements to the compiler itself;
  particularly the frontend. It should be moderately faster.

Fixed

• Code generation for f16 literals in CUDA backend (#1979).

• Branches that return arrays differing in sign of their stride
  (#1984).

Added

• Arbitrary expressions of type i64 are now allowed as sizes. Work
  by Lubin Bailly.

• New prelude function resize.

Removed

• The prelude functions concat_to and flatten_to. They are often
  not necessary now, and otherwise resize is available.

Changed

• The prelude functions flatten and unflatten (and their
  multidimensional variants), as well as split, now have more
  precise types.

• Local and anonymous (lambda) functions that must return unique
  results (because they are passed to a higher order function that
  requires this) must now have an explicit return type ascription that
  declares this, using *. This is very rare (in practice
  unobserved) in real programs.

Fixed

• futhark doc produced some invalid links.

• flatten did not properly check for claimed negative array sizes.

• Type checker crash on some ill-typed programs (#1926).

• Some soundness bugs in memory short circuiting (#1927, #1930).

• Another compiler crash in block tiling (#1933, #1940).

• Global arrays with size parameters no longer have aliases.

• futhark eval no longer crashes on ambiguously typed expressions (#1946).

• A code motion pass was ignorant of consumption constraints, leading
  to compiler crash (#1947).

• Type checker could get confused and think unknown sizes were
  available when they really weren't (#1950).

• Some index optimisations removed certificates (#1952).

• GPU backends can now transpose arrays whose size does not fit in a
  32-bit integer (#1953).

• Bug in alias checking for the core language type checker (#1949).
  Actually (finally) a proper fix of #803.

• Defunctionalisation duplicates less code (#1968).

Fixed

• Certain cases of noninlined functions in multicore backend.

• Defunctionalisation of match where the constructors carry
  functions (#1917).

• Shape coercions involving sum types (#1918). This required
  tightening the rules a little bit, so some coercions involving
  anonymous sizes may now be rejected. Add expected sizes as needed.

• Defunctionalisation somtimes forgot about sizes bound at top level
  (#1920).

Added

• futhark literate (and FutharkScript in general) is now able to do
  a bit of type-coercion of constants.

Fixed

• Accumulators (produced by AD) had defective code generation for
  intra-group GPU kernel versions. (#1895)

• The header file generated by --library contained a prototype for
  an undefined function. (#1896)

• Crashing bug in LSP caused by stray undefined (#1907).

• Missing check for anonymous sizes in type abbreviations (#1903).

• Defunctionalisation crashed on projection of holes.

• Loop optimisation would sometimes remove holes.

• A potential barrier divergence for certain GPU kernels that fail
  bounds checking.

• A potential infinite loop when looking up aliases (#1915).

• futhark literate: less extraneous whitespace.

Added

• The manifest file now lists which tuning parameters are relevant for
  each entry point. (#1884)

• A command tuning_params has been added to the server protocol.

Changed

• If part of a function parameter is marked as consuming ("unique"),
  the entire parameter is now marked as consuming.

Fixed

• A somewhat obscure simplification rule could mess up use of memory.

• Corner case optimisation for mapping over iota (#1874).

• AD for certain combinations of map and indexing (#1878).

• Barrier divergence in generated code in some exotic cases (#1883).

• Handling of higher-order holes (#1885).

Added

• f16.log1p/f32.log1p/f64.log1p by nbos (#1820).

• Better syntax errors for invalid use of !.

• futhark literate now supports a $loadaudio builtin function for loading
  audio to Futhark programs (#1829).

• You can now mix consumption and higher-order terms in slightly more
  cases (#1836).

• futhark pkg now invokes Git directly rather than scraping
  GitHub/GitLab. This means package paths can now refer to any Git
  repository, as long as git clone works. In particular, you can
  use private and self-hosted repositories.

• Significant reduction in compilation time by doing internal sanity
  checks in separate thread.

• New command: futhark eval. Evaluates Futhark expressions
  provided as command line arguments, optionally allowing a file
  import (#1408).

• script input now allows the use of $loaddata.

• Datasets used in futhark test and futhark bench can now be named
  (#1859).

• New command futhark benchcmp by William Due.

Changed

• The C API function futhark_context_new_with_command_queue() for
  the OpenCL backend has been replaced with a configuration setting
  futhark_context_config_set_command_queue().

Fixed

• Minor parser regression that mostly affects the REPL (#1822).

• Parser did not recognise custom infix operators that did not have a
  builtin operator as prefix (#1824).

• GPU backends: expansion of irregular nested allocations involving
  consumption (#1837, #1838).

• CLI executables now handle entry points with names that are not
  valid C identifiers (#1841).

• Various oversights in the type checking of uniqueness annotations
  for higher-order functions (#1842).

• Invalid short-circuiting could cause compiler crashes (#1843).

• Defunctionalisation could mess up sum types, leading to invalid code
  generation by internalisation, leading to a compiler crash (#1847).

• The #[break] attribute now provides the right environment to
  futhark repl, allowing local variables to be inspected.

• Simplification of concatenations (#1851).

• Array payloads in sum types no longer need parens (#1853).

• When a file is loaded with futhark repl, local declarations are
  now available.

• Missing alias propagation when pattern matching incompletely known
  sum types (#1855).

• reduce_by_index and hist were in some cases unable to handle
  input sizes that do not fit in a 32-bit integer.

• A fusion bug related to fusing across transpositions could result in
  a compiler crash (#1858).

Added

• futhark literate now supports an :audio directive for producing audio
  files from arrays of i8 (#1810).

• futhark multicore now parallelises copies (#1799).

• futhark multicore now uses an allocator that better handles large
  allocations (#1768).

Fixed

• Some record field names could cause generation of invalid C API names (#1806).

• Memory block merging was extremely and unnecessarily slow for
  programs with many entry points.

• Simplification mistake could lead to out-of-bounds reads (#1808).

• futhark lsp now handles some bookkeeping messages sent by Eglot.

• Parser can now handle arbitrarily complex chaining of indexing and
  projection.

• Detect and complain about source files without .fut extension (#1813).

• Overly conservative checking of whether a function parameter is
  allowed by be consumed - it was disallowed if it contained any
  scalars (#1816).

Added

• Slightly better type errors for sum types (#1792).

• Better tracing output in interpreter (#1795).

• Improved optimisation of code that uses zero-element arrays
  (sometimes used for type witnesses).

Fixed

• Mishandling of bounds checks in parallel backends could cause
  compiler crashes (#1791).

• Mis-simplification of certain sequentialised scatters into
  single-element arrays (#1793).

• Invalid scatter fusion would cause an internal compiler error
  (#1794).

• The code generator flipped the order of match cases.

• Simpification of concatenations (#1796).

• Defunctionalisation error for fancy size programming (#1798).

• Code generation for load-unbalanced reductions in multicore backend
  (#1800).

• Futhark now works on CUDA 12 (#1801).

• mul_hi and mad_hi for signed integer types now actually do
  signed multiplication (previously it was always unsigned).

Added

• Memory short circuiting now also applied to the multicore backend.

• Fixes for AD of scan with nonscalar operators. Work by Lotte
  Bruun and Ulrik Larsen.

• Generalised histograms now supported in AD. Work by Lotte Bruun and
  Ulrik Larsen.

• OpenCL kernels now embed group size information, which can
  potentially be used for better register allocation by the OpenCL
  implementation.

Fixed

• A hole in the type checker that allowed one to sneak functions out
  if conditionals (#1787).

• futhark repl: unsigned integers are now printed correctly.

• A bug in the type-checking of match (#1783).

• Missing sequencing in type-checking of in-place let expressions (#1786).

• Crash in defunctionaliser caused by duplicate parameter names
  (#1780).

• Infelicities in multicore codegen for scans (#1777).