Limited support for Distributed Memory / LUTRAM

[hansemro]

Issue Description

memory_libmap pass in Yosys 0.18 and newer would synthesize LUTRAMs unsupported by nextpnr including:

• RAMS32 (manually instantiated)
• RAMD32 (manually instantiated)
• RAMS64E (manually instantiated)
• RAMD64E (manually instantiated)
• RAM32X1S
• RAM64X1S
• RAM64X1S_1 (same as RAM64X1S with inverted clock)
• RAM128X1S
• RAM256X1S

Part of the issue stems from nextpnr not fully supporting all LUTRAMs in the Distributed RAM packer in xilinx/pack_dram.cc.

Resolving this should also address https://github.com/openXC7/demo-projects/issues/6.

Tasks/Status

TODO: rewrite tasks

Development Branches

• experimental branch with support for RAMS32, RAMS64E, RAM32X1S, RAM64X1S, RAM128X1S, RAM256X1S: https://github.com/hansemro/nextpnr-xilinx/commits/xc7-lutram-dev/
  • use with yosys 0.18 or newer to test
  • rebuild chipdb after building
  • broken as many incorrect assumptions were made in this branch:
    • does not check negative z height for newly supported cells, possibly breaking projects with several newly supported cells.
    • missing RAMS32/RAMD32 to LUT_OR_MEM transformations with DI1 and O6 ports used.

References

See 018-clb-ram minitest. Build and view design checkpoint in Vivado.

https://f4pga.readthedocs.io/projects/prjxray/en/latest/architecture/dram_configuration.html

https://docs.xilinx.com/v/u/en-US/ug474_7Series_CLB

https://docs.xilinx.com/v/u/en-US/ug574-ultrascale-clb

https://docs.xilinx.com/r/en-US/ug953-vivado-7series-libraries

https://docs.xilinx.com/r/en-US/ug974-vivado-ultrascale-libraries

https://www.xilinx.com/content/dam/xilinx/support/documents/sw_manuals/xilinx14_7/7series_hdl.pdf

https://docs.amd.com/v/u/en-US/7series_hdl

https://github.com/Xilinx/XilinxUnisimLibrary/tree/master/verilog/src/unisims

[hansfbaier]

Very good, thanks!

[hansemro]

Added support for RAM128X1S and RAM256X1S though not sufficiently tested. I was able to build litex-ddr-kc705 after latest changes. However, I am now running into DDR memtest issues it seems: https://gist.github.com/hansemro/5f48f4098e59f9db2e34ae25cb0b6ecd

[hansfbaier]

@hansemro Wow, that was quick! We might want to write some basic tests here: https://github.com/openXC7/primitive-tests Debugging the issue inside that complex design is probably too cumbersome.

[hansemro]

@hansfbaier Sounds good. I'll try to write some tests soon. DDR issue could be totally unrelated to this.

[hansfbaier]

@hansemro Yes, very likely your code works. I never got 8 modules working nice with OpenXC7. At some point the timing just falls apart, because of congestion, I suppose. See my comment on your gist.

[hansemro]

Rebased experimental branch on https://github.com/openXC7/nextpnr-xilinx/commit/8120acd7c036cd87af2f1169e9c555546ad7bded (current stable-backports)

[hansfbaier]

@hansemro great, thanks

[hansemro]

Fixed up RAM32X1D not creating RAMD32 instances (was creating RAMD64 instances previously) and made sure RAM32X1S is handled in pack_dram (forgot this one apparently).

[hansemro]

Made initial tests: https://github.com/hansemro/primitive-tests/commits/lutram-tests/

• Targets KC705 with its 200 MHz differential clock.
• Uses basic clock division and reset propagation so that I can visibly check patterns via LED.
• Basic 1,0,1,0,... write pattern.
• Simple FSM: Reset -> Clear -> Write -> Read -> Finish

Notably nextpnr-xilinx hangs with RAM256X1S. Seems #10 made this observation months ago. I'll try to spend some time debugging this.

[hansemro]

RAM256X1S and RAM256X1D were not handled correctly since their address pins are in an array A[N:0] rather than specified individually (A0, A1, A2, ... ). I'll need to validate every transform rule anyway...

[hansfbaier]

@hansemro I pushed an MMCM-fix to stable-backports. The MMCM should work now, if you have time to try.

[hansemro]

@hansfbaier Thanks for the heads up. I will get to it at some point. I decided it is better for me to fix dual-port LUTRAMs before moving forward, because everything is broken (for at least xc7, not sure about ultrascale).

~For example, while tracing how RAM128X1D is handled, twice as many RAMD64Es are created with z/height decremented for each one. We end up with negative z?! Not sure what the intention was but this does not seem right to me.~

[hansfbaier]

Yes definitely, that is more important. Thanks for working on this.

[hansemro]

I misspoke since I was testing RAM256X1D (thought I was testing RAM256X1S) which does not fit in xc7 anyhow. Still, yosys should not allow unsupported LUTRAMs to be synthesized!

[hansemro]

Resolved two issues:

• Fixed address port detection for single port LUTRAM with c073116536800022adbdeb0ba4b785026a6116b4
• Fixed MUXF tree z offset for RAM256X1S with 4792453ba7bde361d090e91a18456a1601e7b5e2
  • This fixes placer stalling when handling designs with RAM256X1S
  • Will need to revisit for Ultrascale since they have different offsets (more LUTs in a slice).

[hansfbaier]

@hansemro How are things going? Have you been able to test the changes?

[hansemro]

@hansfbaier

Have you been able to test the changes?

MMCM is confirmed working with multiple clock outputs on KC705 though with a BUFG on all clock outputs. fasm2frames would throw segment DB errors if I didn't have them. Test branch: https://github.com/hansemro/primitive-tests/commits/mmcm-blinky-kc705-db-error/

Interestingly, the first clock output didn't require me to place BUFG buffer, though it should probably have one.

How are things going?

Things were going well until I had to handle each LUTRAM as an edge case. Initially, I was less bothered to write things down, but now I feel it is appropriate to actually spend time documenting the port/parameter transformations for all cells (including ultrascale-only cells). I intend to resume work on validation and get xc7 cells covered.

Anyhow, it turns out I made some incorrect assumptions about some things. Here are some TODOs:

1. create_dram32_lut should be able to map LUT5 or LUT6 BEL site, but currently doesn't
2. check whether address ports are connected to both A{6:1} and WA{8:1} (WA{9:1} for ultrascale) ports in LUT6/LUT5 BEL for single-port LUTRAM

Will elaborate further with a follow-up post on LUTRAM transformations to RAMS/RAMD primitive cells to LUT5/LUT6 BELs in more detail.

[hansfbaier]

Yes, I had similar observations. CLKOUT 1-3 had missing pips. CLKOUT 0,5,6 worked fine out of the box: https://github.com/openXC7/primitive-tests/blob/main/mmcm-blinky-kintex/blinky.v But only on Kintex. On other series all CLKOUT ports were fine.

[hansfbaier]

Thanks for the update!

[hansemro]

Naming Notation:

I'll define the following name notation to fold port/parameter names:

• A comma-separated list or range enclosed by curly brackets {} maps to individual names with an entry from the list/range.
  • Curly brackets may be inserted anywhere in the name.
  • List example: ADR{3, 2, 1, 0} maps to ADR3, ADR2, ADR1, ADR0
  • Range example: {D:A}_O maps to D_O, C_O, B_O, A_O
• Verilog-style array with a range enclosed by square brackets [].
  • The range indicates the bit width of the signal/parameter. If there are no square brackets, assume 1-bit width.
  • Square brackets can only be placed at the end of a name.
  • Example: ADR[3:0] maps to ADR[3], ADR[2], ADR[1], ADR[0]
• Combined example:
  • ADDR{D:A}[4:0] maps to ADDRD[4:0], ADDRC[4:0], ADDRB[4:0], ADDRA[4:0]

LUTRAM Cell Table:

Additional notes:

• LUT_OR_MEM5/LUT_OR_MEM6 BEL cannot be instantiated manually. Port and parameter details of these BELs are purely behavioral.
• LUT_OR_MEM6 BEL has SIN and MC31 pins which are used for shift register functionality and can be ignored for LUTRAM.

Cell	Cell Type	US-only	*INIT* Parameter	*CLK_INVERTED Parameter	Clock Input	Write Enable Input	Write Data Input	Write Select Input	Write Address Input	Read Address Input	Read Data Output
LUT_OR_MEM5	BEL	false	INIT[31:0]	N/A	CLK	WE	DI1	N/A	WA{5:1}	A{5:1}	O5
LUT_OR_MEM6	BEL	false	INIT[63:0]	N/A	CLK	WE	DI2	N/A	US ? WA{9:1} : WA{8:1}	A{6:1}	O6
RAMS32	LUTRAM Primitive	false	INIT[31:0]	IS_CLK_INVERTED	CLK	WE	I	N/A	ADR{4:0}	ADR{4:0}	O
RAMD32	LUTRAM Primitive	false	INIT[31:0]	IS_CLK_INVERTED	CLK	WE	I	N/A	WADR{4:0}	RADR{4:0}	O
RAMD32M64	LUTRAM Primitive	true	INIT[63:0]	IS_CLK_INVERTED	CLK	WE	I	N/A	WADR{4:0}	RADR{5:0}	O
RAM32X1S	LUTRAM	false	INIT[31:0]	IS_WCLK_INVERTED	WCLK	WE	D	N/A	A{4:0}	A{4:0}	O
RAM32X1D	LUTRAM	false	INIT[31:0]	IS_WCLK_INVERTED	WCLK	WE	D	N/A	A{4:0}	A{4:0}; DPRA{4:0}	SPO; DPO
RAM32X16DR8	Asymmetric LUTRAM	true	N/A?	IS_WCLK_INVERTED	WCLK	WE	DI{H:A}[1:0]	N/A	ADDRH[4:0];ADDR{G:A}[5:0]	ADDRH[4:0];ADDR{G:A}[5:0]	DOH[1:0]; DO{G:A}
RAM32M	SelectRAM	false	INIT_{D:A}[63:0]	IS_WCLK_INVERTED	WCLK	WE	DI{D:A}[1:0]	N/A	ADDR{D:A}[4:0]	ADDR{D:A}[4:0]	DO{D:A}[1:0]
RAM32M16	SelectRAM	true	INIT_{H:A}[63:0]	IS_WCLK_INVERTED	WCLK	WE	DI{H:A}[1:0]	N/A	ADDR{H:A}[4:0]	ADDR{H:A}[4:0]	DO{H:A}[1:0]
RAMS64E	LUTRAM Primitive	false	INIT[63:0]	IS_CLK_INVERTED	CLK	WE	I	N/A	(WADR{7:6}), ADR{5:0}	ADR{5:0}	O
RAMS64E1	LUTRAM Primitive	true?	INIT[63:0]	IS_CLK_INVERTED	CLK	WE	I	N/A	(WADR{8:6}), ADR{5:0}	ADR{5:0}	O
RAMD64E	LUTRAM Primitive	false	INIT[63:0]	IS_CLK_INVERTED	CLK	WE	I	N/A	WADR{7:0}	RADR{5:0}	O
RAM64X1S	LUTRAM	false	INIT[63:0]	IS_WCLK_INVERTED	WCLK	WE	D	N/A	A{5:0}	A{5:0}	O
RAM64X1D	LUTRAM	false	INIT[63:0]	IS_WCLK_INVERTED	WCLK	WE	D	N/A	A{5:0}	A{5:0}; DPRA{5:0}	SPO; DPO
RAM64X8SW	SelectRAM	true	INIT_{H:A}[63:0]	IS_WCLK_INVERTED	WCLK	WE	D	WSEL[2:0]	A[5:0]	A[5:0]	O[7:0]
RAM64M	SelectRAM	false	INIT_{D:A}[63:0]	IS_WCLK_INVERTED	WCLK	WE	DI{D:A}	N/A	ADDR{D:A}[5:0]	ADDR{D:A}[5:0]	DO{D:A}
RAM64M8	SelectRAM	true	INIT_{H:A}[63:0]	IS_WCLK_INVERTED	WCLK	WE	DI{H:A}	N/A	ADDR{H:A}[5:0]	ADDR{H:A}[5:0]	DO{H:A}
RAM128X1S	LUTRAM	false	INIT[127:0]	IS_WCLK_INVERTED	WCLK	WE	D	N/A	A{6:0}	A{6:0}	O
RAM128X1D	LUTRAM	false	INIT[127:0]	IS_WCLK_INVERTED	WCLK	WE	D	N/A	A[6:0]	A[6:0]; DPRA[6:0]	SPO; DPO
RAM256X1S	LUTRAM	false	INIT[255:0]	IS_WCLK_INVERTED	WCLK	WE	D	N/A	A[7:0]	A[7:0]	O
RAM256X1D	LUTRAM	true	INIT[255:0]	IS_WCLK_INVERTED	WCLK	WE	D	N/A	A[7:0]	A[7:0]; DPRA[7:0]	SPO; DPO
RAM512X1S	LUTRAM	true	INIT[511:0]	IS_WCLK_INVERTED	WCLK	WE	D	N/A	A[8:0]	A[8:0]	O

[hansemro]

XC7 LUTRAM to LUTRAM Primitive Transformations:

LUTRAMs are broken down to primitive cell(s) that will eventually map to SLICEM LUT_OR_MEM6/LUT_OR_MEM5 BEL site(s) once placed.

Convention:

• Title: Source Cell to Destination Cell
• Transformations: Source Name to Destination Name (if owner is not specified)

RAM32X1S -> 1x RAMS32

Cell Rules:

• RAMS32 (6LUT) with /SP appended to name
• No MUXF cells
• 1 output

Parameter Rules:

• IS_WCLK_INVERTED to IS_CLK_INVERTED

Port Rules:

• A{4:0} to /SP's ADR{4:0}
• D to I
• O to O
• WCLK to CLK
• WE to WE

RAM32X1D -> 2x RAMD32

Cell Rules:

• RAMD32 (D6LUT/B6LUT) with /SP appended to name
• RAMD32 (C6LUT/A6LUT) with /DP appended to name
• No MUXF cells
• 2 outputs

Parameter Rules:

• IS_WCLK_INVERTED to IS_CLK_INVERTED

Port Rules:

• A{4:0} to /SP's RADR{4:0}
• A{4:0} to /SP's WADR{4:0}
• DPRA{4:0} to /DP's RADR{4:0}
• A{4:0} to /DP's WADR{4:0}
• D to all I
• SPO to /SP's O
• DPO to /DP's O
• WCLK to all CLK
• WE to all WE

RAM32M -> 2x RAMS32 + 6x RAMD32

Cell Rules:

• RAMS32 (D5LUT) with /RAMD appended to name
• RAMS32 (D6LUT) with /RAMD_D1 appended to name
• RAMD32 (C5LUT) with /RAMC appended to name
• RAMD32 (C6LUT) with /RAMC_D1 appended to name
• RAMD32 (B5LUT) with /RAMB appended to name
• RAMD32 (B6LUT) with /RAMB_D1 appended to name
• RAMD32 (A5LUT) with /RAMA appended to name
• RAMD32 (A6LUT) with /RAMA_D1 appended to name
• No MUXF cells
• 8 outputs

Parameter Rules:

• IS_WCLK_INVERTED to IS_CLK_INVERTED

Port Rules:

• WCLK to all CLK
• WE to all WE
• DIA[0] to /RAMA's I
• DIA[1] to /RAMA_D1's I
• DIB[0] to /RAMB's I
• DIB[1] to /RAMB_D1's I
• DIC[0] to /RAMC's I
• DIC[1] to /RAMC_D1's I
• DID[0] to /RAMD's I
• DID[1] to /RAMD_D1's I
• DOA[0] to /RAMA's O
• DOA[1] to /RAMA_D1's O
• DOB[0] to /RAMB's O
• DOB[1] to /RAMB_D1's O
• DOC[0] to /RAMC's O
• DOC[1] to /RAMC_D1's O
• DOD[0] to /RAMD's O
• DOD[1] to /RAMD_D1's O
• ADDRA[4:0] to /RAMA and /RAMA_D1's RADR{4:0}
• ADDRB[4:0] to /RAMB and /RAMB_D1's RADR{4:0}
• ADDRC[4:0] to /RAMC and /RAMC_D1's RADR{4:0}
• ADDRD[4:0] to all WADR{4:0}
• ADDRD[4:0] to /RAMD and /RAMD_D1's `ADR{4:0}

RAM64X1S -> RAMS64E

Cell Rules:

• RAMS64E (6LUT)
• No MUXF cells
• 1 output

Parameter Rules:

• IS_WCLK_INVERTED to IS_CLK_INVERTED

Port Rules:

• WCLK to CLK
• WE to WE
• D to I
• O to O
• A{5:0} to ADR{5:0}
• RAMS64E's WADR6 and WADR7 ports not connected

RAM64X1D -> 2x RAMD64E

Cell Rules:

• RAMD64E (D6LUT/B6LUT) with /SP appended to name
• RAMD64E (C6LUT/A6LUT) with /DP appended to name
• No MUXF cells
• 2 outputs

Parameter Rules:

• IS_WCLK_INVERTED to IS_CLK_INVERTED

Port Rules:

• WCLK to all CLK
• WE to all WE
• A{5:0} to all RADR{5:0}
• A{5:0} to all WADR{5:0}
• DPRA{5:0} to /DP's RADR{5:0}
• D to all I
• SPO to /SP's O
• DPO to /DP's O

RAM64M -> 4x RAMD64E

Cell Rules:

• RAMD64E (A6LUT) with /RAMA appended to name
• RAMD64E (B6LUT) with /RAMB appended to name
• RAMD64E (C6LUT) with /RAMC appended to name
• RAMD64E (D6LUT) with /RAMD appended to name
• No MUXF cells
• 4 outputs

Parameter Rules:

• IS_WCLK_INVERTED to IS_CLK_INVERTED

Port Rules:

• WCLK to all CLK
• WE to all WE
• DI{A:D} to /RAM{A:D}'s I
• DO{A:D} to /RAM{A:D}'s O
• DO{A:D} to /RAM{A:D}'s O
• ADDR{A:D}[5:0] to /RAM{A:D}'s RADR{5:0}
• ADDRD[5:0] to all WADR{5:0}
• all WADR6 and WADR7 RAMD64E ports are not connected

RAM128X1S -> 2x RAMS64E

Cell Rules:

• RAMS64E (D6LUT/B6LUT) with /LOW appended to name
• RAMS64E (C6LUT/A6LUT) with /HIGH appended to name
• MUXF7 cell with /F7 for single output

Parameter Rules:

• IS_WCLK_INVERTED to IS_CLK_INVERTED

Port Rules:

• WCLK to all CLK
• D to all I
• A{5:0} to all ADR{5:0}
• A6 to all WADR6
• A6 to /F7's S
• all WADR7 ports are not connected
• /LOW's O to /F7's I0
• /HIGH's O to /F7's I1
• O to /F7's O

  RAM128X1D -> 4x RAMD64E

Cell Rules:

• RAMD64E (D6LUT) with /SP.LOW appended to name
• RAMD64E (C6LUT) with /SP.HIGH appended to name
• RAMD64E (B6LUT) with /DP.LOW appended to name
• RAMD64E (A6LUT) with /DP.HIGH appended to name
• MUXF7 with /F7.SP appended to name for single port output
• MUXF7 with /F7.DP appended to name for dual port output
• 2 outputs

Parameter Rules:

• IS_WCLK_INVERTED to IS_CLK_INVERTED

Port Rules:

• WCLK to all CLK
• WE to all WE
• D to all I
• A[6:0] to all WADR{6:0}
• A[6:0] to /SP.LOW's RADR{6:0}
• A[6:0] to /SP.HIGH's RADR{6:0}
• A[6] to /F7.SP's S
• DPRA[6:0] to /DP.LOW's RADR{6:0}
• DPRA[6:0] to /DP.HIGH's RADR{6:0}
• DPRA[6] to /F7.DP's S
• /SP.LOW's O to /F7.SP's I0
• /SP.HIGH's O to /F7.SP's I1
• /DP.LOW's O to /F7.DP's I0
• /DP.HIGH's O to /F7.DP's I1
• SPO to /F7.SP's O
• DPO to /F7.DP's O

RAM256X1S -> 4x RAMS64E

Cell Rules:

• RAMS64E (D6LUT) with /RAMS64E_D appended to name
• RAMS64E (C6LUT) with /RAMS64E_C appended to name
• RAMS64E (B6LUT) with /RAMS64E_B appended to name
• RAMS64E (A6LUT) with /RAMS64E_A appended to name
• MUXF7 with /F7.A appended to name
• MUXF7 with /F7.B appended to name
• MUXF8 with /F8 appended to name for single output

Parameter Rules:

• IS_WCLK_INVERTED to IS_CLK_INVERTED

Port Rules:

• WCLK to all CLK
• WE to all WE
• D to all I
• A[5:0] to all ADR{5:0}
• A[7:6] to all WADR{7:6}
• A[6] to /F7.A's S
• A[6] to /F7.B's S
• A[7] to /F8's S
• /RAMS64E_D's O to /F7.B's I0
• /RAMS64E_C's O to /F7.B's I1
• /RAMS64E_B's O to /F7.A's I0
• /RAMS64E_A's O to /F7.A's I1
• /F7.B's O to /F8's I0
• /F7.A's O to /F8's I1
• O to /F8's O

[hansemro]

LUTRAM Primitive to BEL Transformations:

Additional notes:

• Unclear to me how CLKINV bit is set for SLICE_LUTX in nextpnr/fasm.
  • Also unclear to me how CLKINV status for all BELs in the SLICE site are checked to match.
• Some of the transformation details have yet been implemented or verified in my development branch.

RAMD64E -> LUT_OR_MEM6 BEL

• nextpnr-specific rules:
  • set type to id_SLICE_LUTX
  • transform parameter IS_CLK_INVERTED to IS_WCLK_INVERTED
  • set attribute X_LUT_AS_DRAM to 1
• RADR{5:0} to A{6:1}
• WADR{7:0} to WA{8:1}
• I to DI1
• O to O6
• WE to WE
• CLK to CLK

RAMS64E -> LUT_OR_MEM6 BEL

• nextpnr-specific rules:
  • set type to id_SLICE_LUTX
  • transform parameter IS_CLK_INVERTED to IS_WCLK_INVERTED
  • set attribute X_LUT_AS_DRAM to 1
• ADR{5:0} to A{6:1}
• ADR{5:0} to WA{6:1}
• WADR{7:6} to WA{8:7}
• I to DI1
• O to O6
• WE to WE
• CLK to CLK

RAMD32 -> LUT_OR_MEM6 BEL

• nextpnr-specific rules:
  • set type to id_SLICE_LUTX
  • transform parameter IS_CLK_INVERTED to IS_WCLK_INVERTED
  • set attribute X_LUT_AS_DRAM to 1
• RADR{4:0} to A{5:1}
• WADR{4:0} to WA{5:1}
• I to DI1 or DI2 if DI1 already used by LUT_OR_MEM5
• O to O6
• WE to WE
• CLK to CLK

RAMD32 -> LUT_OR_MEM5 BEL

• nextpnr-specific rules:
  • set type to id_SLICE_LUTX
  • transform parameter IS_CLK_INVERTED to IS_WCLK_INVERTED
  • set attribute X_LUT_AS_DRAM to 1
• RADR{4:0} to A{5:1}
• WADR{4:0} to WA{5:1}
• I to DI1
• O to O5
• WE to WE
• CLK to CLK

RAMS32 -> LUT_OR_MEM6 BEL

• nextpnr-specific rules:
  • set type to id_SLICE_LUTX
  • transform parameter IS_CLK_INVERTED to IS_WCLK_INVERTED
  • set attribute X_LUT_AS_DRAM to 1
• ADR{4:0} to A{5:1}
• ADR{4:0} to WA{5:1}
• I to DI1 or DI2 if DI1 already used by LUT_OR_MEM5
• O to O6
• WE to WE
• CLK to CLK

RAMS32 -> LUT_OR_MEM5 BEL

• nextpnr-specific rules:
  • set type to id_SLICE_LUTX
  • transform parameter IS_CLK_INVERTED to IS_WCLK_INVERTED
  • set attribute X_LUT_AS_DRAM to 1
• ADR{4:0} to A{5:1}
• ADR{4:0} to WA{5:1}
• I to DI1
• O to O5
• WE to WE
• CLK to CLK

[hansfbaier]

Thanks for the effort! I am looking forward to what you will come up with!

[hansemro]

Issue: nextpnr checks INIT{A:D} instead of INIT_{A:D} parameters for RAM32M/RAM64M

While working on an INIT parameter test, I noticed that the INIT parameters for RAM32M/RAM64M were not being set with correct values in the FASM result.

https://github.com/openXC7/nextpnr-xilinx/blob/1c57f511f80945a709d1d43478d39ad0b6cd51d2/xilinx/pack_dram.cc#L455-L469

Merely adding underscores does not immediately solve the issue, so I will need to look more into this later.

WIP INIT Parameter Test: https://github.com/hansemro/primitive-tests/tree/xc7-lutram-tests/lutram-tests/init-test

[hansemro]

Issue: nextpnr checks INIT{A:D} instead of INIT_{A:D} parameters for RAM32M/RAM64M

This should be fixed in this branch: https://github.com/hansemro/nextpnr-xilinx/tree/fix-ram32m-ram64m-init

However, I am noticing some discrepancies compared to Vivado that I will need to verify. Notice how the upper and lower 32-bits of the ?LUT.INIT pattern are swapped.

RAM32M NextPNR fasm result:

CLBLM_L_X84Y126.SLICEM_X0.ALUT.INIT[63:0] = 64'b0000000000000000111011100100010000000000000000000101000001010000
CLBLM_L_X84Y126.SLICEM_X0.ALUT.DI1MUX.AI
CLBLM_L_X84Y126.SLICEM_X0.ALUT.SMALL
CLBLM_L_X84Y126.SLICEM_X0.ALUT.RAM
CLBLM_L_X84Y126.SLICEM_X0.AOUTMUX.O5
CLBLM_L_X84Y126.SLICEM_X0.BLUT.INIT[63:0] = 64'b0000000000000000111011100100010000000000000000001111101011111010
CLBLM_L_X84Y126.SLICEM_X0.BLUT.DI1MUX.BI
CLBLM_L_X84Y126.SLICEM_X0.BLUT.SMALL
CLBLM_L_X84Y126.SLICEM_X0.BLUT.RAM
CLBLM_L_X84Y126.SLICEM_X0.BOUTMUX.O5
CLBLM_L_X84Y126.SLICEM_X0.CLUT.INIT[63:0] = 64'b0000000000000000000100011011101100000000000000001010111110101111
CLBLM_L_X84Y126.SLICEM_X0.CLUT.DI1MUX.CI
CLBLM_L_X84Y126.SLICEM_X0.CLUT.SMALL
CLBLM_L_X84Y126.SLICEM_X0.CLUT.RAM
CLBLM_L_X84Y126.SLICEM_X0.COUTMUX.O5
CLBLM_L_X84Y126.SLICEM_X0.DLUT.INIT[63:0] = 64'b0000000000000000000100011011101100000000000000000000010100000101
CLBLM_L_X84Y126.SLICEM_X0.DLUT.SMALL
CLBLM_L_X84Y126.SLICEM_X0.DLUT.RAM
CLBLM_L_X84Y126.SLICEM_X0.DOUTMUX.O5
CLBLM_L_X84Y126.SLICEM_X0.WEMUX.CE
CLBLM_L_X84Y126.SLICEM_X0.NOCLKINV
CLBLM_L_X84Y126.SLICEL_X1.NOCLKINV

RAM32M Vivado bit2fasm result:

CLBLM_L_X48Y95.SLICEL_X1.NOCLKINV
CLBLM_L_X48Y95.SLICEL_X1.PRECYINIT.C0
CLBLM_L_X48Y95.SLICEM_X0.ALUT.DI1MUX.BDI1_BMC31
CLBLM_L_X48Y95.SLICEM_X0.ALUT.INIT[46:0] = 47'b10100000101000000000000000000001110111001000100
CLBLM_L_X48Y95.SLICEM_X0.ALUT.RAM
CLBLM_L_X48Y95.SLICEM_X0.ALUT.SMALL
CLBLM_L_X48Y95.SLICEM_X0.AOUTMUX.O5
CLBLM_L_X48Y95.SLICEM_X0.BLUT.DI1MUX.DI_CMC31
CLBLM_L_X48Y95.SLICEM_X0.BLUT.INIT[47:0] = 48'b111110101111101000000000000000001110111001000100
CLBLM_L_X48Y95.SLICEM_X0.BLUT.RAM
CLBLM_L_X48Y95.SLICEM_X0.BLUT.SMALL
CLBLM_L_X48Y95.SLICEM_X0.BOUTMUX.O5
CLBLM_L_X48Y95.SLICEM_X0.CLUT.DI1MUX.DI_DMC31
CLBLM_L_X48Y95.SLICEM_X0.CLUT.INIT[47:0] = 48'b101011111010111100000000000000000001000110111011
CLBLM_L_X48Y95.SLICEM_X0.CLUT.RAM
CLBLM_L_X48Y95.SLICEM_X0.CLUT.SMALL
CLBLM_L_X48Y95.SLICEM_X0.COUTMUX.O5
CLBLM_L_X48Y95.SLICEM_X0.DLUT.INIT[42:0] = 43'b1010000010100000000000000000001000110111011
CLBLM_L_X48Y95.SLICEM_X0.DLUT.RAM
CLBLM_L_X48Y95.SLICEM_X0.DLUT.SMALL
CLBLM_L_X48Y95.SLICEM_X0.DOUTMUX.O5
CLBLM_L_X48Y95.SLICEM_X0.NOCLKINV
CLBLM_L_X48Y95.SLICEM_X0.PRECYINIT.C0
CLBLM_L_X48Y95.SLICEM_X0.WEMUX.CE

[hansemro]

Issue: nextpnr does nothing with IS_*CLK_INVERTED property for LUTRAM cells

While working on CLKINV property test, I noticed that nextpnr does not set the CLKINV bit when IS_*CLK_INVERTED property for a LUTRAM is set. Instead, nextpnr fasm writer ignores the property and sets NOCLKINV for the SLICEM site.

Also note that, on XC7, LUTRAMs and FFs share the same CLKINV routing BEL. However, on Ultrascale(+), LUTRAMs have their own dedicated clock inverter provided by the LCLKINV routing BEL.

WIP CLKINV property test: https://github.com/hansemro/primitive-tests/tree/xc7-lutram-tests/lutram-tests/clkinv-test

[hansfbaier]

Great work!

[hansemro]

Something that bothered me about how RAM{S,D}64E maps to LUT_OR_MEM6 BEL is that only DI1 data input is used to write to both internal LUT_OR_MEM5 BELs. Somehow one of the LUT_OR_MEM5 BEL can select between DI1 and DI2 data inputs, but this is not really well documented.

Recently, I stumbled on the following physical design rules (in $VIVADO_2017.2_ROOT/ids_lite/ISE/msg/usenglish/PhysDesignRules.msg) that seem to correspond to RAM.SMALL configuration bit being what controls data input selection:

1383
Issue with pin connections and/or configuration on block:<%s>:<%s>.  For RAMMODE programming set with DPRAM64 or SPRAM64 or SRL32 the DI1 input pin must be connected.\n
Issue with pin connections and/or configuration on block:<!%1!>:<!%2!>.  For RAMMODE programming set with DPRAM64 or SPRAM64 or SRL32 the DI1 input pin must be connected.\n
;;
1384
Issue with pin connections and/or configuration on block:<%s>:<%s>.  For RAMMODE programming set with DPRAM64 or SPRAM64 or SRL32 the DI2 input pin cannot be connected.\n
Issue with pin connections and/or configuration on block:<!%1!>:<!%2!>.  For RAMMODE programming set with DPRAM64 or SPRAM64 or SRL32 the DI2 input pin cannot be connected.\n
;;
1385
Issue with pin connections and/or configuration on block:<%s>:<%s>.  For RAMMODE programming set with DPRAM32 or SPRAM32 or SRL16 the DI2 input pin must be connected.\n
Issue with pin connections and/or configuration on block:<!%1!>:<!%2!>.  For RAMMODE programming set with DPRAM32 or SPRAM32 or SRL16 the DI2 input pin must be connected.\n
;;
1386
Issue with pin connections and/or configuration on block:<%s>:<%s>.  For RAMMODE programming set with DPRAM32 or SPRAM32 or SRL16 the DI1 input pin cannot be connected.\n
Issue with pin connections and/or configuration on block:<!%1!>:<!%2!>.  For RAMMODE programming set with DPRAM32 or SPRAM32 or SRL16 the DI1 input pin cannot be connected.\n

Coincidentally, 018-clb-ram prjxray fuzzer found that RAM.SMALL configuration bit is set for RAM32M/RAM32X1{S,D} and SRL16E, but not set for RAM64M/RAM{64,128}X{S,D}/RAM256X1S and SRLC32E. This all seems to indicate RAM.SMALL bit is used for data input selection for upper LUT_OR_MEM5 BEL (one with DI2 input and initialized with INIT[63:32]).

Here is a block diagram to help visualize what I see of LUT_OR_MEM6 BEL:

[hansfbaier]

Good to see you making progress!

[lehaifeng000]

@hansemro the DI2 can't be used, it would make the Infinite loop in place. there is some check in the place, if the lutram used the DI2 port, and the wa7-8 ports are not being config, It cannot complete the place

[hansemro]

@lehaifeng000 Yes, nextpnr does not currently have the capacity to pack/place RAM32X1S/RAMS32 which would occupy and utilize both LUT_OR_MEM5 BELs. As you say, placer will get stuck because LUT_OR_MEM5/RAM32* with DI2 is not yet accepted as legal: https://github.com/openXC7/nextpnr-xilinx/blob/9debb871624163d4043150e576793c78cba503f8/xilinx/arch_place.cc#L114-L117.

Additionally, how {C,B}X pins are connected to {C,B}LUTs and WA7USED/WA8USED BEL should also be considered in legalization. I believe, Vivado avoids this by making it illegal to place mismatched LUTRAM types in the same CLB SLICEM site. However, I will need to look more into this.

If we update the packer to utilize DI2, we will need to and should update the placer and legalization accordingly.

[lehaifeng000]

I tend to understand and try to modify place, inserting rules into the placement process.

[lehaifeng000]

@hansfbaier @hansemro I'm wondering if you could provide an email or other contact information. Sometimes, materials that aren't suitable for public sharing can be sent to you privately.

[hansfbaier]

@lehaifeng000 my email address foss@hans-baier.de should be visible in every git commit. Same for @hansemro

[lehaifeng000]

@lehaifeng000 my email address foss@hans-baier.de should be visible in every git commit. Same for @hansemro OK, I got it