This is useful for certain cross-compilation workloads, and to cache
rarely changing build products.
To use this functionality, build e.g. as follows:
cmake . -B build-export -DEXPORT_BBA_FILES=../bba-files -DARCH=all
cmake --build build-export -t nextpnr-all-bba
cmake . -B build-import -DIMPORT_BBA_FILES=../bba-files -DARCH=all
cmake --build build-import
The impetus for this commit is the fact that it causes rare but
build-breaking race conditions when used with `make -jN` with `N > 1`.
These race conditions are difficult to track down or fix because of
the very rudmentary debugging tools provided by `make` and opaque
semantics of CMake's Makefile generator. They break the build by
running two `.bba` generation processes, then one of them renaming
the `.bba.new` file once it's done, leaving the other one to fail.
After reflection (as the author of this code path) and discussion with
community members who use it, I've concluded that this isn't the right
approach.
1. In practice, on targets where `-DSERIALIZE_CHIPDBS=` matters, you
also care about other build steps, like linking nextpnr, which
are not serializable this way. So you use a workaround anyway, like
`make`ing individual targets instead.
2. The way to serialize the build with Make is the `-j1` option. Trying
to work around `-jN` to make it work like `-j1` is inherently error
prone. While there is some utility in not serializing C++ compilation
this utility could be more easily achieved by providing a single
target that builds all chipdbs, running `make <chipdb-target> -j1`,
then running `make -jN` for the rest of the build.
Two user-visible changes were made:
* `-DUSE_RUST` is replaced with `-DBUILD_RUST`, by analogy with
`-DBUILD_PYTHON`
* `-DCOVERAGE` was removed as it doesn't work with either modern GCC
or Clang
This removes the atomic rename for bbasm outputs because it embeds
the resulting paths into the `.cc` files in embed mode. In any case
the write should be fast enough to not be a big risk for interrupted
builds.
This was tested with Clang 19 only (gcc hasn't had a release that
supports `#embed` yet).
While it served a purpose (granting the ability to build `.bba` files
separately from the rest of nextpnr), it made things excessively
convoluted, especially around paths.
This commit removes the ability to pre-generate chip databases. As far
as I know, I was the primary user of that feature. It can be added back
if there is demand for it.
In exchange the per-family `CMakeLists.txt` files are now much easier
to understand.
The two main changes, done together in this commit, are:
* Eliminating most instances of `aux_source_directory()`, replacing
them with explicit file listings; and
* Moving these file listings into respective subdirectories by
representing respective nextpnr components as interface libraries.
In addition, the GUI CMake script tree was simplified since it had
a lot of unused/redundant code.
The `aux_source_directory()` command is not recommended for use by
CMake itself because it misses dependency changes when adding/removing
files, and consequently causes build failures requiring a clean rebuild.
This commit does not touch anything related to architectures/families,
which are very complex and redundant all on their own.
* ng-ultra: new architecture
* Implementation as in D2 deliverable
* Support for nxdesignsuite-24.0.0.0-20240429T102300
* Save memory by directly outputing json
* Add support for bidirectional IOs
* cleanup
* Create BFRs properly
* Add IOM insertion
* Cleanup
* Block certain pips depending of DDFR mode
* Add LUT bypass to improve routability
* Add bypass for CSC mode of GCK
* Fix IOM case
* Initial memory support
* Better RF/XRF handling
* fix
* RF placement and legalization
* Disconnect non available ports for NX_RAM
* cleanup
* Add RFB/RAM context support for latest release
* Remove ports that must not be used
* Proper port used only on RFB
* Add structure for clock sinks
* Use cell type where applicable
* Add clock sinks for other cell types
* Validation check fixes
* Commented too restrictive placement
* Added more crossbar wire type
* Hande IO termination input
* Fail early due to NX tools limitation for now
* Validations and fixes for RAM I/Os
* Fix for latest version of tools
* Use ctx->idf where applicable
* warn if RAM ports are not actually used
* Fix IOM packing
* Fix CY packing
* Change how constants are handled on CY
* Post placement optimization for CY
* Address comments for PR
* pack and export GCK, WFG and PLL
* Cover more global routing cases
* Constraing to location if provided
* Place at LOC
* Pack and export DSP
* wip
* wip
* notes
* wip
* wip
* Validate DSPs
* DSP cascading
* Check mandatory parameters for DSP
* existing gck
* wip
* export all the rest for bitstream
* CDC packing
* add more sinks
* place FIFO
* map rest of FIFO ports
* enable pll by default
* cleanup
* Initial XLUT support
* Fix statistics
* Properly duplicate GCKs
* RRSTO and WRSTO are not used on XFIFO
* Fix for latest version of JSON format
* Implement GCK limitations
* cleanup
* cleanup
* Add more signals and use lowskew name
* cleanup code a bit
* Fix wfb
* detect cascaded GCKs
* Handle DFR
* Route dfr clock properly
* Cleanup
* Cleanup bitstream code
* Review issues addressed
* Move helper routines
* Expose private members for unit tests
* cleanup
* remove scale factor
* make all location helper arrays static
* Addressed review comments
* Support post-routing CSC and SCC
* Support NX_BFF
* Place CSS and SCC only on allowed locations
* Support latest Impulse
* ng_ultra: Expand bounding box further for left-edge IO
Signed-off-by: gatecat <gatecat@ds0.me>
* Export all IO parameters in bitstream
* Handle new CSV order or parameters and additional validation
* Add some more undocumented values for CSV
* Support for old and new CSV formats
* Initial DDFR support
* Display warning message once per file
* Address review issues
* Fix crash on memory access
* Make boundbox fit NG-Ultra internal design
* Update attributes after dff rewrite
* Implement basic NG-Ultra LUT-DFF unit tests
* Always use first seen xbar input
Signed-off-by: gatecat <gatecat@ds0.me>
* Simplified crossbar pip detection
* Change order to prevent issues with some unconnected constants
* Pack LUT and multiple DFF in stripe
* Place DFF chains
* Improve large DFF chains
* Rename to pack_dff_chains
* Better use XLUTs when possible
* pack output DFF together with XLUT
* option to disable XLUT optimiziations
* Make more optimizations optional
* fix to use pre-increment
* GCK for lowskew signals
* Bugfix for nets that are not part of lowskew network
* Fix bitstream export for PLL cell
* Remove separate route lowskew
* Allow WFG mode 2
* Merge inverter into GCK
* Add CSC per TILE when needed
* Improve reusage of existing cell for CSC
* Take preferred CSC
* Cleanup
* When in place CSC size not important
* Cleanup
* Reset and Load restriction
* make csc optimisation optional
* Proper count for IO resources
* Detect when there is no next cell for DSP chain
* Do not incorporate loops in XLUT
* Check if output exists
* Update copyright for delivery
* Make building NG-Ultra chip database optional, follow filename convention
* Ported drawing code to new API
* Update expandBoundingBox for NG-Ultra
* Copyright and license update
* Add README information
* cleanup and constids
* Using ctx->idf where applicable
* remove if_using_basecluster
* refactor extra data usage
* refactor to use create_cell_ptr only
* optimized getCSC
* optimize critical path a bit
* clangformat
* disable clangformat where applicable
---------
Signed-off-by: gatecat <gatecat@ds0.me>
Co-authored-by: Lofty <dan.ravensloft@gmail.com>
Co-authored-by: gatecat <gatecat@ds0.me>
Eigen considers the EIGEN3_INCLUDE_DIRS and EIGEN3_DEFINITIONS variables
to be deprecated and they will no longer be exported in the next release
after 3.4.0:
f2984cd077
Use the IMPORTED target instead, which seems to be the preferred way of
consuming third-party CMake libraries.
Currently the lookahead is disabled by default because of the time to
compute and RAM usage. However it does appear to work reasonably well
in testing. Further effort is required to lower RAM usage after initial
computation, and explore trade-off for cheaper time to compute.
Signed-off-by: Keith Rothman <537074+litghost@users.noreply.github.com>
Abseil has a hard dependency on threads (not just in the build system
but in many places in the base libraries), so there is no way to use
it on WASI at the moment.
This lowers the CPU cost of using the flat wire map in router2, and should
use less memory as well.
Signed-off-by: Keith Rothman <537074+litghost@users.noreply.github.com>
-Werror is not enabled by default, except on CI and for a few specific common traps, to avoid the inevitable breakages when new compiler versions add new diagnostics.
Signed-off-by: gatecat <gatecat@ds0.me>
