Commit Graph

675 Commits

Author SHA1 Message Date
Eddie Hung aa6d06c1b5 Revert "Revert "synth_* with -retime option now calls abc with -D 1 as well""
This reverts commit 6008bb7002.
2019-12-30 13:28:29 -08:00
Marcin Kościelnicki 13a3041030
Merge pull request #1593 from YosysHQ/mwk/dsp48a1-pmgen
xilinx_dsp: Initial DSP48A/DSP48A1 support.
2019-12-25 16:18:44 +01:00
Marcin Kościelnicki dadaf7ed78 xilinx: Test our DSP48A/DSP48A1 simulation models. 2019-12-23 20:36:43 +01:00
Marcin Kościelnicki 666c6128a9 xilinx_dsp: Initial DSP48A/DSP48A1 support. 2019-12-22 20:51:14 +01:00
Eddie Hung 5986a4df40 Add abc9_arrival times for RAM{32,64}M 2019-12-20 14:06:59 -08:00
Eddie Hung 7928eb113c Add RAM{32,64}M to abc9_map.v 2019-12-20 13:41:23 -08:00
Eddie Hung 10e82e103f
Revert "Optimise write_xaiger" 2019-12-20 12:05:45 -08:00
Eddie Hung df626ee7ab
Merge pull request #1558 from YosysHQ/eddie/xaiger_cleanup
Optimise write_xaiger
2019-12-19 12:24:03 -05:00
Marcin Kościelnicki 8b2c9f4518 xilinx: Add simulation models for remaining CLB primitives. 2019-12-19 18:04:04 +01:00
Marcin Kościelnicki 561ae1c5c4 xilinx_dffopt: Keep order of LUT inputs.
See rationale at https://github.com/YosysHQ/yosys/pull/1557#discussion_r359196549
2019-12-19 18:01:43 +01:00
Marcin Kościelnicki a235250403 xilinx: Add xilinx_dffopt pass (#1557) 2019-12-18 13:43:43 +01:00
Marcin Kościelnicki aff6ad1ce0 xilinx: Improve flip-flop handling.
This adds support for infering more kinds of flip-flops:

- FFs with async set/reset and clock enable
- FFs with sync set/reset
- FFs with sync set/reset and clock enable

Some passes have been moved (and some added) in order for dff2dffs to
work correctly.

This gives us complete coverage of Virtex 6+ and Spartan 6 flip-flop
capabilities (though not latch capabilities).  Older FPGAs also support
having both a set and a reset input, which will be handled at a later
data.
2019-12-18 13:43:43 +01:00
Eddie Hung a73f96594f
Merge pull request #1574 from YosysHQ/eddie/xilinx_lutram
xilinx: add LUTRAM rules for RAM32M, RAM64M
2019-12-16 21:48:21 -08:00
Eddie Hung 5a00d5578c Add unconditional match blocks for force RAM 2019-12-16 13:31:15 -08:00
Eddie Hung d910bec8e0 Update xc7/xcu bram rules 2019-12-16 13:00:58 -08:00
Eddie Hung 5d00996426 Merge branch 'eddie/xilinx_lutram' of github.com:YosysHQ/yosys into eddie/xilinx_lutram 2019-12-16 12:06:47 -08:00
Eddie Hung 7545ab3814 Populate DID/DOD even if unused 2019-12-16 11:57:04 -08:00
Eddie Hung c4d37813cb Rename *RAM{32,64}M rules to RAM{32X2,64X1}Q 2019-12-16 10:41:13 -08:00
Diego H f3f59910eb Removing fixed attribute value to !ramstyle rules 2019-12-15 23:51:58 -06:00
Diego H b35559fc33 Merging attribute rules into a single match block; Adding tests 2019-12-15 23:33:09 -06:00
Diego H 266993408a Refactoring memory attribute matching based on IEEE 1364.1 and Tool specific 2019-12-13 15:43:24 -06:00
Eddie Hung 52875b0d61
Merge pull request #1533 from dh73/bram_xilinx
Adjust Xilinx xc7/xcu BRAM min bits threshold for RAMB18E1
2019-12-13 12:01:03 -08:00
Eddie Hung c3262d6075 Disable RAM16X1D match rule; carry-over from LUT4 arches 2019-12-13 08:59:17 -08:00
Eddie Hung d6514fc2e1 RAM64M8 to also have [5:0] for address 2019-12-13 08:54:19 -08:00
Eddie Hung 8925bf4b96 Add RAM32X6SDP and RAM64X3SDP modes 2019-12-12 18:52:28 -08:00
Eddie Hung 50e0c83560 Fix RAM64M model to have 6 bit address bus 2019-12-12 18:52:03 -08:00
Eddie Hung 7a9d1be97d Add memory rules for RAM16X1D, RAM32M, RAM64M 2019-12-12 17:44:59 -08:00
Diego H 751a18d7e9 Fixing citation in xc7_xcu_brams.txt file. Fixing RAMB36E1 test. 2019-12-12 17:32:58 -06:00
Eddie Hung 9ab1feeaf1 abc9_map.v: fix Xilinx LUTRAM 2019-12-12 14:56:52 -08:00
Diego H 937ec1ee78 Updating RAMB36E1 thresholds. Adding test for both RAMB18E1/RAMB36E1 2019-12-12 13:50:36 -06:00
Diego H ab6ac8327f Merge https://github.com/YosysHQ/yosys into bram_xilinx 2019-12-12 13:40:05 -06:00
Eddie Hung 98c9ea605b techmap/aigmap of whiteboxes to occur before abc9 instead of in write_xaiger 2019-12-06 17:05:02 -08:00
Marcin Kościelnicki fcce94010f
xilinx: Add tristate buffer mapping. (#1528)
Fixes #1225.
2019-12-04 09:44:00 +01:00
Marcin Kościelnicki 10014e2643
xilinx: Add models for LUTRAM cells. (#1537) 2019-12-04 06:31:09 +01:00
Marcin Kościelnicki 2badaa9adb xilinx: Add missing blackbox cell for BUFPLL. 2019-11-29 16:56:27 +01:00
Diego H 3a5a65829c Adjusting Vivado's BRAM min bits threshold for RAMB18E1 2019-11-27 12:05:04 -06:00
Marcin Kościelnicki 0466c48533 xilinx: Add simulation models for IOBUF and OBUFT. 2019-11-26 08:15:20 +01:00
Marcin Kościelnicki 6cdea425b8 clkbufmap: Add support for inverters in clock path. 2019-11-25 20:40:39 +01:00
Marcin Kościelnicki 7562e7304e xilinx: Use INV instead of LUT1 when applicable 2019-11-25 20:40:39 +01:00
Marcin Kościelnicki 7a9081440c xilinx: Add simulation models for MULT18X18* and DSP48A*.
This adds simulation models for the following primitives:

- MULT18X18 and MULT18X18S (Virtex 2*, Spartan 3)
- MULT18X18SIO (Spartan 3E, Spartan 3A)
- DSP48A (Spartan 3A DSP) — implemented in terms of DSP48A1
- DSP48A1 (Spartan 6)
2019-11-19 01:00:58 +01:00
Marcin Kościelnicki c4bd318e76 synth_xilinx: Merge blackbox primitive libraries.
First, there are no longer separate cell libraries for xc6s/xc7/xcu.
Manually instantiating a primitive for a "wrong" family will result
in yosys passing it straight through to the output, and it will be
either upgraded or rejected by the P&R tool.

Second, the blackbox library is expanded to cover many more families:
everything from Spartan 3 up is included.  Primitives for Virtex and
Virtex 2 are listed in the Python file as well if we ever want to
include them, but that would require having two different ISE versions
(10.1 and 14.7) available when running cells_xtra.py, and so is probably
more trouble than it's worth.

Third, the blockram blackboxes are no longer in separate files — there
is no practical reason to do so (from synthesis PoV, they are no
different from any other cells_xtra blackbox), and they needlessly
complicated the flow (among other things, merging them allows the user
to use eg. Series 7 primitives and have them auto-upgraded to
Ultrascale).

Last, since xc5v logic synthesis appears to work reasonably well
(the only major problem is lack of blockram inference support), xc5v is
now an accepted setting for the -family option.
2019-11-06 15:11:27 +01:00
David Shah 3506eaf290 xilinx: Add URAM288 mapping for xcup
Signed-off-by: David Shah <dave@ds0.me>
2019-10-23 11:47:44 +01:00
David Shah 6769d31ddb xilinx: Add support for UltraScale[+] BRAM mapping
Signed-off-by: David Shah <dave@ds0.me>
2019-10-23 11:47:37 +01:00
Marcin Kościelnicki 7b350cacd4 xilinx: Support multiplier mapping for all families.
This supports several older families that are not yet supported for
actual logic synthesis — the intention is to add them soon.
2019-10-22 18:06:57 +02:00
Clifford Wolf a3a7bb9bf7
Merge pull request #1452 from nakengelhardt/fix_dsp_mem_reg
Call memory_dff before DSP mapping to reserve registers (fixes #1447)
2019-10-22 17:36:54 +02:00
Sean Cross 82f60ba938 Makefile: don't assume python is called `python3`
On some architectures, notably on Windows, the official name for the
Python binary from python.org is `python`.  The build system assumes
that python is called `python3`, which breaks under this architecture.

There is already infrastructure in place to determine the name of the
Python binary when building PYOSYS.  Since Python is now always required
to build Yosys, enable this check universally which sets the
`PYTHON_EXECUTABLE` variable.

Then, reuse this variable in other Makefiles as necessary, rather than
hardcoding `python3` everywhere.

Signed-off-by: Sean Cross <sean@xobs.io>
2019-10-19 14:04:52 +08:00
N. Engelhardt 3b405d985e Call memory_dff before DSP mapping to reserve registers (fixes #1447) 2019-10-17 21:33:54 +02:00
Marcin Kościelnicki 526fe4cb89 xilinx: Add simulation model for IBUFG. 2019-10-10 13:16:03 +02:00
Eddie Hung 9fd2ddb14c
Merge pull request #1437 from YosysHQ/eddie/abc_to_abc9
Rename abc_* names/attributes to more precisely be abc9_*
2019-10-08 10:53:38 -07:00
Eddie Hung 6c5e1234e1 Add comment on why partial multipliers are 18x18 2019-10-04 22:31:04 -07:00