Change from long double to double for storing floating point values in modcc.

WSL, and Windows generally, treats long double differently than Linux, which leads to inconsistencies in literal numeric values in code generated by modcc on the two platforms.

Fixes #1245

* Update CMake CXX version.
* Make CUDA 10 a minimum requirement and remove CUDA 9 support code.
* Set up g++ Travis tests to use g++ 8.
* Explicitly set C++14 version for nvcc.
* Properly split cuda compilation, including in unit tests.
* Remove unnecessary modcc `SOLVE` warning.
* Update pybind11 module to tag v2.5.0
* Replace `util::size` and `util::data` with `std::` equivalents.

The `vec_t_` time vector is already available to the mechanisms, but not exposed. It is indexed by `cell_index` instead of directly by the `node_index` (`vec_t[vec_ci_[node_index_[i]]]`). This kind of indexing was previously unavailble. It has not been added to the printers. 
This PR also includes some cleanup in the vectorized code printer. 

Address #1109

* Use assign when reading memory in simd mode.
* Expand assign to work for numbers and other simd_impls.
* Don't expose detail namespace in simd.hpp.

* Add zero assignments to local variables in modcc output.

* Add new API for the SIMD library that is compatible with the ARM 'sizeless' SVE vectors. Language restrictions prevent the use of the operator overload interface used up to this point for SIMD.
* Add `indirect_expressions` and `indirect_indexed_expressions` for describing memory reads/writes. `where_expressions` control masked access to simd vectors.
* Implement the SVE SIMD back-end in accordance with the alternate SIMD API.
* Retrieve vector width information from compiled mechanisms.
* Use alternate SIMD API in modcc-generated mechanisms.
* Add assertion in generated mechanism code that checks runtime vector width compatibility.

Fixess #1021. 

Add a new catalogue comprising optimised version of some of the mechanisms commonly used in models of the Allen Cell Types Database, eg http://celltypes.brain-map.org/mouse/experiment/electrophysiology/323535089.
This made required extending modcc to some new NMODL features, as well as to expose some optimisations.

In addition clean up the default catalogue by moving test mechanisms to their proper place in the unit tests.

* Add ranges in STATE blocks of the form FROM ... TO ...
* Unit test for the above
* Add type check for <..., ...> ranges
* Add units in procedure argument lists
* Unit test for the above.

Add a simple re-write to eliminate division by a known constant.

* [WIP] Investigate the effect of __restrict__ specifier on GPU code.

* Fix range parsing and add unit test

- Add option to compile Arbor for HIP/CUDA backend using Clang. 
- Add new CMake option `ARB_GPU_COMPILE_TYPE` to distinguish between three possible gpu platforms/builds: `cuda-nvcc`; `cuda-clang`; `hip-clang`
- Add gpu wrapper functions in `arbor/backends/gpu/gpu_api.hpp` and `arborenv/gpu_api.hpp` which call HIP/CUDA functions depending on the platform
- Rename functions and files: cuda -> gpu
- Add downgraded warp primitives for HIP
- Implement `uuid` workaround for HIP
- Set correct alignment and warp size for AMD gpus
- Update installation guide and docs.
- Update pip/setuptools to support new gpu targets

Fixes #833

* Replace TCLAP with tinyopt, remove TCLAP from `ext/`.
* Replace `sup/tinyopt.hpp` with copy of external tinyopt, update use of `tinyopt` in unit tests and examples to suit.
* Use tinyopt's 'smallopt' interface in `modcc`, `brunel` example.
* Update tinyopt interface in `single` example.
 Fix our attributions and add HBP acknowledgement

Addresses #557 #695 #803 

- Reset the error of the semantic pass of an expression at the start of the function call. 
- Check the errors of sub-expressions after applying to them their semantic passes, and report errors if found. If the error could disrupt the rest of the original semantic pass, return.
- Check errors in functions/procedures before and after inlining, but not during. 

Fixes #931 

* Add support for conditionals with the SIMD printer by using `where_expresssions` to mask assignments to range variables that happen inside if/else blocks. Assignments to non-range variables are executed non-masked when correctness is ensured. 

Fixes #836 

* Choose the pivot on the diagonal if possible when performing symbolic Gaussian elimination in modcc-generated mechanisms.

* Rewrite the function expander which lowers function calls and function arguments, such that both steps are done in the same visitor, and both visitors are applied to the entire procedure. 
* Rewrite the function inliner such that it is also applied to an entire procedure, after all functions have been expanded. The function inliner iterates over the body of the procedure, inlining one function call at a time, until all function calls have been inlined. 
* Conductivity and current accumulations are also modified to be done in one step at the end of `nrn_current`

* Group common linear system solver code across `SparseSolverVisitor`, `SparseNonlinearSolverVisitor`, and `LinearSolverVisitor`.
* Normalize intermediate values in Gaussian elimination when the matrix has more than 5 rows.

Fixes #890

Nora Abi Akar authored 5 years ago and Benjamin Cumming committed 5 years ago

Also change the way we handle unitson/unitsoff.
UNITSON, UNITSOFF, COMMENT, ENDCOMMENT are now all handled in Lexer::parse

Fixes #14 
Fixes #885

Nora Abi Akar authored 5 years ago and Benjamin Cumming committed 5 years ago

Also renumber operator precedence.

Fixes #25

Nora Abi Akar authored 5 years ago and Benjamin Cumming committed 5 years ago

Also emit "abs" in cexpr_emit.cpp because the simd backend implements abs

Fixes #887

Nora Abi Akar authored 5 years ago and Benjamin Cumming committed 5 years ago

Fixes #892

* Fix modcc segmentation fault caused by improper check for the presence of a symbol, triggered when a symbol defined in a RANGE is absent from the ASSIGNED and PARAMETER blocks.

* Accept `tok::ne` for conditional expressions in modcc, and emit correctly as '!=' in `cexpr_emit`.

* Add support for non-linear kinetic schemes in modcc, using a non-linear backwards Euler method with a fixed number of iterations of Newton's method.

Nora Abi Akar authored 5 years ago and Benjamin Cumming committed 5 years ago

Sets the diameter of a CV as the diameter at the middle of a CV
Exposes the diameter in `shared_state` and `mechanism`
Adds support in modcc to treat `diam` as a special parameter 
Implements unit tests

Fixes #850 

* Modify `SparseSolverVisitor` to allow solving kinetic equations at steady state. 

Addresses #837 

Nora Abi Akar authored 5 years ago and Benjamin Cumming committed 5 years ago

- Implement inlining functions with multiple statements
- Implement inlining nested function calls
- Implement inlining if/else statements. Compilation fails if a function fails to set the return value.

Fixes #741, #742, #862 

WIP on implementation of new cable cell API.

* Hide implementation and state using PIMPL
* Add consistent `place` methods for adding synapse, stimuls, gap junction site and spike detectors to `cable_cell`.
* Add default constructor for `region` and `locset` that create empty `nil` instances.

Sam Yates authored 5 years ago and Benjamin Cumming committed 5 years ago

* Remove or implement move ctor/assign operations for ostream-derived classes.
* Add explicit test for a zero row in modcc's sparse and linear solver finalize routines.
* Forward declare `morphology_impl` as a struct.
* Remove redundant capture in a lambda expression in fvm_layout.cpp.
* Add extra parens in an assign-and-test condition in mechcat.cpp.

* Adds modcc support for COMPARTMENT statements of the form: `COMPARTMENT v {state_0, state_1, ..., state_n}`.
* Use `COMPARTMENT` values `v` to multiply the derivative of state variables `state_0, state_1, ..., state_n` in associated kinetic scheme blocks.

Fixes #838.

Nora Abi Akar authored 5 years ago and Benjamin Cumming committed 5 years ago

Add support for parsing and processing `LINEAR` blocks: 

Changes: 
* `SOLVE` expressions can be called from inside an `INITIAL` block, but only if they are solving a linear system
* Tilde expressions can now be either linear expressions or reaction expressions
* Linear expressions need to be rewritten before being sent to the solver, this is done using `LinearRewriter`
* The linear system is setup in `LinearSolverVisitor` fills the lhs and rhs of the symbolic matrix   
* The matrix is recued using `gj_reduce`, which now works on non-diagonal matrices. 

Fixes #839 

Nora Abi Akar authored 5 years ago and Benjamin Cumming committed 5 years ago

Add modcc support for processing `CONSERVE` statements in `KINETIC` blocks.

The `KineticRewriter` remains unchanged. The `SparseSolverVisitor` is modified. 
If one or more `CONSERVE` statements are present, corresponding rows in the symbolic matrix, which would otherwise represent differential equations, are replaced by the content of the conserve statements. 

Addresses #828 and #830 

Addresses #824.

* Add modcc support for `CONSTANT` blocks in NMODL, subject to the following constraints:
    - Any identifier declared in the `CONSTANT` block may only be used after that declaration (including within the `CONSTANT` block itself).
    - Units in the `CONSTANT` block are parsed but not processed.
    - Values assigned to identifiers inside the `CONSTANT` block may only be signed numeric literals, or an already declared constant, possibly preceded by unary minus.

* Collect cable cell parameter setting structures into `include/cable_cell_param.hpp`.
* Restructure electrical specifications (axial resistance, membrane capacitance) and ionic concentrations and reversal potentials on cable cells, so that these can be specified with a global default, per-cell defaults, and per-segment values.
* Allow reversal potentials to be set by a mechanism of a new kind 'revpot', which are prohibited from maintaining state or writing to any shared state other than ionic reversal potentials.
* Specify reversal potential mechanisms as global defaults or per-cell. Reversal potential mechanisms may not be specified at the level of a segment in order to avoid non-linearities arising from the discretization.
* Supply default cable cell parameter data that matches NEURON values (this is _not_ used by default).
* Replace the d_lambda calculation with one that approximates more faithfully the effect of tapered segments, and which will ...

Nora Abi Akar authored 5 years ago and Benjamin Cumming committed 5 years ago

Add missing check of whether to use direct or indirect indexing for simd procedure calls.

fixes #804

Sam Yates authored 5 years ago and Benjamin Cumming committed 5 years ago

Address inconsistencies in unit conversions between current updates, initialization, ionic currents etc.

Fixes issues #793 and #798.

* Correct units in comments for ion state iX_ to a current density.
* Remove conversion factor for NMODL current densities [mA/cm²] to Arbor current densities [A/m²] from layout weight: from the point of view of Arbor, mechanisms now write in units that Arbor uses natively.
* Remove weight_ as a magic modcc variable; applying weight_ is now performed directly by code generated by printers.
* Change IndexedVariable interface: remove redundant index_name; remove 'op'. Whether external data is accumulated or assigned is now a property of the corresponding sourceKind.
* Conversion factors for NMODL <-> Arbor are collated in printerutils.cpp, determined by the external data sourceKind.
* Permit direct assignment of external data, as opposed to a weighted sum. This behaviour will be required for reversal potential handling.
* `current_` and `conducitivity_` get a different sourceKind in point mechanisms, allowing us to distinguish current contributions from current density contributions (and applying correct unit scaling accordingly.)
* Add unit test for correct initialization and scaling of ionic currents, together with two corresponding unit test mechanisms.

* Add `::arb::` decoration to `fvm_value_type` and `fvm_index_type` in modcc simd cprinter output.
* Return stream from unwind stack trace printer.

Fixes #801 and also a long standing (but low impact) bug in unwind code.

* Correctly decorate required names from arb:: namespace.
* Put the unit test mechanisms in the 'testing::' namespace, to exercise this functionality.

Sam Yates authored 5 years ago and Benjamin Cumming committed 5 years ago

Extended USEION VALENCE support.

* Augment modcc parser to accept identifier or signed integer after VALENCE.
* Treat identifer after VALENCE as global scalar, tied to ionic charge.
* Add ionic charge field to mechanism ion parameter pointers, replacing scalar in shared state with a length-1 array.
* Verify presence of ion name and check ion charge against numeric valence in mechanism info when discretizing in `fvm_build_mechanism_data`.
* Pass ion info, coalesce_synapse global state to `fvm_build_mechanism_data` just via cable cell global properties.
* Add more unit-test only mechanisms (test_ca_read_valence, test_cl_valence).
* Add unit tests for valence verification and reading into mechanism state (multicore only).
* Update existing unit tests for new interfaces.

Implements #780.