|
1 |
| -This tarfile contains: |
| 1 | +This archive contains VPR, an FPGA placement and routing tool, and VPACK, |
| 2 | +a program to pack LUTs and flip flops into coarser grained logic blocks and |
| 3 | +convert a netlist from blif format to VPR's .net format. |
2 | 4 |
|
3 |
| -manual.ps: Postscript version of the VPR user's manual. |
4 |
| -vpr: Directory containing source and sample input files for VPR. |
5 |
| -blifmap: Directory containing source and a sample input file for blifmap. |
| 5 | +To see VPR in action on the sample files included here, type: |
| 6 | + |
| 7 | +vpr e64.net sample.arch out.p out.r |
| 8 | + |
| 9 | + |
| 10 | +If you want only global routing to be performed, use: |
| 11 | + |
| 12 | +vpr e64.net sample.arch out.p out.r -route_type global |
| 13 | + |
| 14 | +============================================================================= |
| 15 | + |
| 16 | +Major changes since the last version (Version 3.22) of VPR include: |
| 17 | + |
| 18 | +- The blifmap netlist translation program has been replaced by the more |
| 19 | + powerful VPACK program. |
| 20 | + |
| 21 | +- VPR can now perform either combined global and detailed routing, or global |
| 22 | + routing alone. |
| 23 | + |
| 24 | +To convert a technology-mapped file in .blif format to a logic block containing |
| 25 | +one 4-input LUT and a FF use: |
| 26 | + |
| 27 | +vpack input.blif output.net -lut_size 4 -no_clustering |
| 28 | + |
| 29 | + |
| 30 | +To convert the same blif file into a netlist in which every logic block |
| 31 | +contains 4 LUTs and 4 FFs, and there are only 10 inputs to the LUTs, use: |
| 32 | + |
| 33 | +vpack input.blif output.net -lut_size 4 -cluster_size 4 -inputs_per_cluster 10 |
| 34 | + |
| 35 | + |
| 36 | + |
| 37 | +To place and route this netlist in VPR, type: |
| 38 | + |
| 39 | +vpr input.net input.arch output.placed output.routed |
| 40 | + |
| 41 | +Note that the logic block used in the netlist and the logic block described |
| 42 | +in the architecture file must be the same. |
| 43 | + |
| 44 | +The speed quality/tradeoff of the VPR placer is best controlled with the |
| 45 | +-inner_num option. Specifying -inner_num 1 on the command line will speed |
| 46 | +up the placer by a factor of 10, at the cost of about a 10% reduction in |
| 47 | +placement quality. |
| 48 | + |
| 49 | +The speed quality/tradeoff in the VPR router is best controlled with the |
| 50 | +-max_router_iterations and -initial_pres_fac options. Setting |
| 51 | +-max_router_iterations to 15 will roughly double the speed of the router, at |
| 52 | +the cost of some additional track usage, by making the router give up |
| 53 | +and decide a circuit is unroutable at a given channel width sooner. That is, |
| 54 | +the router will not take as long on unsuccessful routing attempts. Setting |
| 55 | +-initial_pres_fac to a large number (say 1000) will greatly speed up successful |
| 56 | +routings; it won't speed up unsuccessful routing attempts. |
| 57 | + |
| 58 | + |
| 59 | +============================================================================== |
| 60 | +Contents of the archive: |
| 61 | + |
| 62 | +manual.ps: Postscript manual of VPR. |
| 63 | +README: This file. |
| 64 | + |
| 65 | +./vpr |
| 66 | + |
| 67 | + *.c, *.h: Source code for VPR. |
| 68 | + |
| 69 | + makefile: Makefile for VPR. Currently set for Solaris and gcc; you may |
| 70 | + have to modify the library paths and compiler options on your |
| 71 | + machine. |
| 72 | + |
| 73 | + descript.txt: A revision history of VPR. |
| 74 | + |
| 75 | + e64.net: A sample netlist file from the MCNC bencmark set. The logic |
| 76 | + block contains one LUT + one FF. |
| 77 | + |
| 78 | + sample.arch: A sample FPGA architecture file. The logic block consists |
| 79 | + of one LUT + one FF in the configuration shown in manual.ps. |
| 80 | + |
| 81 | + e64.p: A placement of e64 generated by VPR. |
| 82 | + |
| 83 | + |
| 84 | +./vpack |
| 85 | + |
| 86 | + *.c, *.h: Source code for VPACK. |
| 87 | + |
| 88 | + makefile: Makefile for VPACK. May have to be modified for non Solaris |
| 89 | + machines or for compilers other than gcc. |
| 90 | + |
| 91 | + descript.txt: Revision history of VPACK. |
| 92 | + |
| 93 | + e64.blif: MCNC benchmark circuit e64 technology-mapped by Flowmap to |
| 94 | + 4-input LUTs. This is a combinational circuit. |
| 95 | + |
| 96 | + s1423.blif: A sequential MCNC benchmark circuit. It has been technology- |
| 97 | + mapped to 4-input LUTs and flip flops by Flowmap. |
| 98 | + |
| 99 | + |
| 100 | +============================================================================== |
| 101 | + |
| 102 | +Changes from VPR Version 3.22 to this Version (3.99): |
| 103 | + |
| 104 | +1) The blifmap program has been replaced by the VPACK program. The command: |
| 105 | + |
| 106 | + vpack input.blif output.net -lut_size 4 -no_clustering |
| 107 | + |
| 108 | + is equivalent to |
| 109 | + |
| 110 | + blifmap input.blif output.net -lut_size 4 -global |
| 111 | + |
| 112 | + |
| 113 | +2) The .net netlist format used as input to VPR has had a new line, the |
| 114 | + subblock line, added to it. VPACK will automatically add this new line |
| 115 | + to every clb, so if you used blifmap to convert .blif files to .net |
| 116 | + format, just use VPACK to reconvert the .blif files to this new format. |
| 117 | + Old .net files will not work with this version of VPR. |
| 118 | + See the manual for details of the new format. |
| 119 | + |
| 120 | + |
| 121 | +3) Several new lines have been added to the architecture file. See the |
| 122 | + manual for details. |
| 123 | + |
| 124 | + |
| 125 | +4) VPR now defaults to combined global and detailed routing, and tries |
| 126 | + 30 routing iterations before deciding a circuit is unroutable at a |
| 127 | + certain track width. To do global routing with a maximum of 15 routing |
| 128 | + iterations (the default in VPR version 3.22), use the command: |
| 129 | + |
| 130 | + vpr input.net input.arch output.p output.r -route_type global |
| 131 | + -max_router_iterations 15 |
| 132 | + |
| 133 | + |
| 134 | +5) The placement file format has been slightly changed. Block numbers are |
| 135 | + no longer required in the placement file, and blocks can now be listed |
| 136 | + in any order in a placement file used as input to VPR. This should make |
| 137 | + it much easier to read placements generated by other tools into VPR. |
| 138 | + See the manual for details. |
| 139 | + |
| 140 | + |
| 141 | +6) One can now read in a fixed I/O configuration and force VPR to use these |
| 142 | + I/O locations. Use the -fix_pins <file.pads> option. |
0 commit comments