This pass is a proper subset of opt_rmdff, which is called by opt, which
is called by every synth flow in the coarse part. Thus, it never
actually does anything and can be safely removed.
This commit:
* renames all remaining instances of "DRAM" (which is ambiguous)
to "LUTRAM" (which is not), finishing the work started in
the commit 698ab9be;
* renames memory rule files to brams.txt/lutrams.txt;
* adds/renames script labels map_bram/map_lutram;
* extracts where necessary script labels map_ffram and map_gates;
* adds where necessary options -nobram/-nolutram.
The end result is that BRAM/LUTRAM/FFRAM aspects of every target
are now consistent with each other.
Per architecture:
* anlogic: rename drams.txt→lutrams.txt, add -nolutram, add
:map_lutram, :map_ffram, :map_gates
* ecp5: rename bram.txt→brams.txt, lutram.txt→lutrams.txt
* efinix: rename bram.txt→brams.txt, add -nobram, add :map_ffram,
:map_gates
* gowin: rename bram.txt→brams.txt, dram.txt→lutrams.txt,
rename -nodram→-nolutram (-nodram still recognized), rename
:bram→:map_bram, :dram→:map_lutram, add :map_ffram, :map_gates
The initial list of hits was generated with the codespell command
below, and each hit was evaluated and fixed manually while taking
context into consideration.
DIRS="kernel/ frontends/ backends/ passes/ techlibs/"
DIRS="${DIRS} libs/ezsat/ libs/subcircuit"
codespell $DIRS -S *.o -L upto,iff,thru,synopsys,uint
More hits were found by looking through comments and strings manually.
As the TD tool doesn't accept the DRAM cell to contain unknown values in
the initial value, the initialzation support of DRAM is previously
skipped.
Now add the support by add a new pass to determine unknown values in the
initial value.
Signed-off-by: Icenowy Zheng <icenowy@aosc.io>
During the addition of DRAM inferring support, the installation of
eagle_bb.v is accidentally removed.
Fix this issue.
Signed-off-by: Icenowy Zheng <icenowy@aosc.io>
According to the datasheet of Anlogic Eagle FPGAs, The LUTs natively
in an Anlogic FPGA is LUT4 (in MSLICEs) and "Enhanced LUT5" (in
LSLICEs). An "Enhanced LUT5" can be divided into two LUT4s.
So a LUT5 will cost around 2x resource of a LUT4, and a LUT6 will cost
2x resource of a LUT5.
Change the -lut parameter passed to the abc command to pass this cost
info to the ABC process.
Signed-off-by: Icenowy Zheng <icenowy@aosc.io>
As dffinit has already supported for different initialization strings
for DFFs and check for re-initialization, initialization of Anlogic
DFFs are now ready to go.
Support for set the init values of Anlogic DFFs.
Signed-off-by: Icenowy Zheng <icenowy@aosc.io>
The dbits of DRAM16X4 is wrong set to 2, which leads to waste of DRAM
bits.
Fix the dbits number in the RAM configuration.
Signed-off-by: Icenowy Zheng <icenowy@aosc.io>
The MSLICEs on the Eagle series of FPGA can be configured as Distributed
RAM.
Enable to synthesis to DRAM.
As the Anlogic software suite doesn't support any 'bx to exist in the
initializtion data of DRAM, do not enable the initialization support of
the inferred DRAM.
Signed-off-by: Icenowy Zheng <icenowy@aosc.io>
This reverts commit 43030db5ff.
For a synthesis tool, generating EG_LOGIC cells are a good choice, as
they can be furtherly optimized when PnR, although sometimes EG_LOGIC is
not as blackbox as EG_PHY cells (because the latter is more close to the
hardware implementation).
Signed-off-by: Icenowy Zheng <icenowy@aosc.io>
Claire Xenia Wolf
gatecat
Marcelina Kościelnicka
whitequark
Eddie Hung
Miodrag Milanovic
Clifford Wolf
Larry Doolittle
Icenowy Zheng