mirror of https://github.com/openXC7/prjxray.git
a8de0afdb3
The open-flow (Yosys → nextpnr-xilinx → FASM → bitstream) now produces
silicon-functional bits on VC707 xc7vx485tffg1761-2 for:
- rst_to_led (IBUF↔OBUF passthrough)
- counter_skewfree (button-clocked 8b counter, general routing)
- counter_sw_bufr (button → BUFR → 8b counter)
- counter_bufr (200 MHz LVDS sysclk → IBUFDS → BUFR → 8b counter)
- counter_2bufg (2× BUFGCTRL on the same source)
- vc707_telegraph (125 MHz crystal → IBUFDS_GTE2 → BUFG → UART smoke test)
- vc707_picosoc (picorv32 + simpleuart + BRAM @ 125 MHz; UART prints
'PicoSoC alive on VC707 @ 125 MHz' on /dev/ttyUSB0)
Highlights of this drop:
utils/fasm2frames.py (+223 net):
- Bank-glue auto-injection for HP-bank IOB18 — IBUF/OBUF (Y0+Y1) +
IBUFDS differential pair. Fires off the FASM-level direction
heuristic (.IN/.IN_ONLY/IBUFDISABLE for IBUF, .DRIVE. for OBUF,
.IN_DIFF for IBUFDS; .SLEW. is unreliable as a marker — gets emitted
on default-state IOBs too).
- INT_L_X32Y49 DCI cascade / bank-active markers when any LIOB18_X81
Y1 OBUF is present.
- PUDC_B emission rewritten for HP-bank IOSTANDARDs (10 features
cover Y0 + Y1 default-state; all 9 historic 'PUDC_B glue' bits
flow naturally from the existing IOSTANDARD segbits).
- HCLK_L per-BUFRCLK-channel 'active' marker — currently codified
for BUFRCLK3 (the channel exercised by counter_sw_bufr).
- GFAN T-tie root glue — INT_L_X62Y(N+10).GFAN_TIE_ROOT_GLUE when
INT_L_X62Y(N).GFAN0.GND_WIRE appears (OBUF.T → GND routing).
- PUDC_B tile excluded from the bank-glue walk (its IN features are
virtual; injecting OBUF_HP_BANK_GLUE on it produces spurious bits).
utils/utils.tcl (+47):
- write_pip_txtdata bulk-fetch — replaces per-net foreach pip with
bulk get_pips + bulk get_property IS_DIRECTIONAL + cached
dst_wire_to_num_pips. ~4× speed-up on xc7vx485t (per-spec time on
041-clk-hrow-pips / 045-hclk-cmt-pips drops from ~1.5 h to ~25 min).
utils/mergedb.sh (+15):
- LIOI / LIOI_TBYTESRC / LIOI_TBYTETERM / LIOB18 / mask_liob18 sed
rewrites for the L-side IOI/IOB18 tiles on HP-only parts (xc7vx485t
uses left-side IOB18 too; upstream kintex7 mergedb only knew the
right side).
11 fuzzers patched for virtex7 readiness:
- 030-iob18 Makefile: split DB target for virtex7 (HP-only); the BUFR
HP-bank results come from the actual fuzzer rather than HR-side sed.
- 037-iob18-pips: L-side mirror tiles (LIOI / LIOI_TBYTESRC /
LIOI_TBYTETERM) added to segdata glob; *_SING tiles excluded;
EXCLUDE_RE updated for L-side prefixes.
- 039-hclk-config: split virtex7 vs kintex7 (HCLK_IOI vs HCLK_IOI3);
XRAY_IOSTANDARD env var; IOB18M/IOB33M alternation.
- 047a-hclk-idelayctrl-pips: accepts both HCLK_IOI and HCLK_IOI3.
- 041, 045, 034, 034b, 043, 044, 046: removed local
write_pip_txtdata override that shadowed the patched utils.tcl
bulk-fetch (was re-introducing the slow per-net Tcl path).
README.md (+86):
- 'Virtex-7 Port Status (virtex7-support branch)' section —
achievements, goals, work-in-progress, constraints.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
|
||
|---|---|---|
| .. | ||
| Makefile | ||
| README.md | ||
| cmt_top_upper_t.tcl | ||
| fixup_and_group.py | ||
| generate.py | ||
| generate.tcl | ||
| tag_groups.txt | ||
| top.py | ||
README.md
Fuzzer for the PIPs of CMT_TOP_[LR]_UPPER_T tiles.
The fuzzer instantiates a PLL in each available site with 2/3 probability of using it. Once used it is connected randomly to various clock and logic resources.
For some nets a randomized "manual" route is chosen to cover as many routing scenarios as possible.
The information whether a PLL is used or not is stored in a file ("design.txt") along with the randomized route (route.txt)
After the design synthesis the generate.py sets fixed routes on some nets which is read from the route.txt file. The rest of the design is routed in the regular way. The script also dumps all used PIPs (as reported by Vivado) to the design_pips.txt.
The tag generation is done in the following way:
- If a PLL site is occupied then tags for all active PIPs are emitted as 1s. No tags are emitted for inactive PIPs.
- When a PLL site is not occupied (IN_USE=0) then tags for all PIPs for the CMT tile are emitted as 0s.
- The IN_USE tag is emitted directly.
The raw solution of tag bits is postprocessed via the custom script fixup_and_group.py. The script does two things:
- Clears all bits found for the IN_USE tag in all other tags. Those bits are common to all of them.
- Groups tags according to the group definitions read from the
tag_groups.txtfile. Bits that are common to the group are set as 0 in each tag that belongs to it (tags within a group are exclusive).