riscv
/
OpenFPGA
mirror of https://github.com/lnis-uofu/OpenFPGA.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

Adde vtr_benchmarks_template for demo

This commit is contained in:
Ganesh Gore 2022-05-06 22:40:31 -06:00
parent 9473523b6b
commit 522982c9ba
6 changed files with 892 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
openfpga.sh
View File

@ -44,6 +44,7 @@ create-task () {
if [ ${#2} -ge 1 ]; then
if [[ "$2" == "vpr_blif" ]]; then template="template_tasks/${2}_template/";
elif [[ "$2" == "yosys_vpr" ]]; then template="template_tasks/${2}_template/";
elif [[ "$2" == "vtr_benchmarks" ]]; then template="template_tasks/${2}_template/";
else template="$2"
fi
fi

1
openfpga_flow/regression_test_scripts/basic_reg_test.sh
View File

@ -181,6 +181,7 @@ run-task basic_tests/fix_pins --debug --show_thread_logs
echo -e "Testing project templates";
run-task template_tasks/vpr_blif_template --debug --show_thread_logs
run-task template_tasks/yosys_vpr_template --debug --show_thread_logs
run-task template_tasks/vtr_benchmarks_template --debug --show_thread_logs
echo -e "Testing create tsk from template and run task"
create-task _task_copy basic_tests/generate_fabric

54
openfpga_flow/tasks/template_tasks/vtr_benchmarks_template/config/task.conf Normal file
View File

@ -0,0 +1,54 @@
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
# Configuration file for running experiments
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
# timeout_each_job : FPGA Task script splits fpga flow into multiple jobs
# Each job execute fpga_flow script on combination of architecture & benchmark
# timeout_each_job is timeout for each job
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
[GENERAL]
run_engine=openfpga_shell
power_tech_file = ${PATH:OPENFPGA_PATH}/openfpga_flow/tech/PTM_45nm/45nm.xml
power_analysis = false
spice_output=false
verilog_output=true
timeout_each_job = 20*60
fpga_flow=yosys_vpr
[OpenFPGA_SHELL]
openfpga_shell_template=${PATH:TASK_DIR}/vtr_benchmark_template_script.openfpga
openfpga_arch_file=${PATH:OPENFPGA_PATH}/openfpga_flow/openfpga_arch/k6_frac_N10_adder_chain_dpram8K_dsp36_40nm_openfpga.xml
openfpga_sim_setting_file=${PATH:OPENFPGA_PATH}/openfpga_flow/openfpga_simulation_settings/fixed_sim_openfpga.xml
# VPR parameters
# Use a fixed routing channel width to save runtime
vpr_route_chan_width=300
[ARCHITECTURES]
arch0=${PATH:TASK_DIR}/k6_N10_tileable_dpram8K_dsp36_40nm.xml
arch1=${PATH:TASK_DIR}/k6_frac_N10_tileable_adder_chain_dpram8K_dsp36_40nm.xml
[BENCHMARKS]
bench1=${PATH:OPENFPGA_PATH}/openfpga_flow/benchmarks/vtr_benchmark/ch_intrinsics.v
bench2=${PATH:OPENFPGA_PATH}/openfpga_flow/benchmarks/vtr_benchmark/diffeq1.v
bench3=${PATH:OPENFPGA_PATH}/openfpga_flow/benchmarks/vtr_benchmark/diffeq2.v
bench4=${PATH:OPENFPGA_PATH}/openfpga_flow/benchmarks/vtr_benchmark/sha.v
[SYNTHESIS_PARAM]
# Yosys script parameters
bench_yosys_cell_sim_verilog_common=${PATH:OPENFPGA_PATH}/openfpga_flow/openfpga_yosys_techlib/k6_frac_N10_tileable_adder_chain_dpram8K_dsp36_40nm_cell_sim.v
bench_yosys_bram_map_rules_common=${PATH:OPENFPGA_PATH}/openfpga_flow/openfpga_yosys_techlib/k6_frac_N10_tileable_adder_chain_dpram8K_dsp36_40nm_bram.txt
bench_yosys_bram_map_verilog_common=${PATH:OPENFPGA_PATH}/openfpga_flow/openfpga_yosys_techlib/k6_frac_N10_tileable_adder_chain_dpram8K_dsp36_40nm_bram_map.v
bench_yosys_dsp_map_verilog_common=${PATH:OPENFPGA_PATH}/openfpga_flow/openfpga_yosys_techlib/k6_frac_N10_tileable_adder_chain_dpram8K_dsp36_40nm_dsp_map.v
bench_yosys_dsp_map_parameters_common=-D DSP_A_MAXWIDTH=36 -D DSP_B_MAXWIDTH=36 -D DSP_A_MINWIDTH=2 -D DSP_B_MINWIDTH=2 -D DSP_NAME=mult_36x36
bench_read_verilog_options_common = -nolatches
bench_yosys_common=${PATH:OPENFPGA_PATH}/openfpga_flow/misc/ys_tmpl_yosys_vpr_bram_dsp_flow.ys
# Benchmark top_module name
bench1_top = memset
bench2_top = diffeq_paj_convert
bench3_top = diffeq_f_systemC
bench4_top = sha1
[SCRIPT_PARAM_MIN_ROUTE_CHAN_WIDTH]
# end_flow_with_test=
# vpr_fpga_verilog_formal_verification_top_netlist=

2
openfpga_flow/vpr_arch/k6_N10_tileable_adder_chain_dpram8K_dsp36_40nm.xml → openfpga_flow/tasks/template_tasks/vtr_benchmarks_template/k6_N10_tileable_dpram8K_dsp36_40nm.xml
View File

@ -559,7 +559,7 @@
-->
<direct name="clbouts1" input="fle[9:0].out[0:0]" output="clb.O[9:0]"/>
<direct name="clbouts2" input="fle[9:0].out[1:1]" output="clb.O[19:10]"/>
</interconnect>
</pb_type>
<!-- Define general purpose logic block (CLB) ends -->

828
openfpga_flow/tasks/template_tasks/vtr_benchmarks_template/k6_frac_N10_tileable_adder_chain_dpram8K_dsp36_40nm.xml Normal file
View File

@ -0,0 +1,828 @@
<!--
Flagship Heterogeneous Architecture with Carry Chains for VTR 7.0.
- 40 nm technology
- General purpose logic block:
K = 6, N = 10, fracturable 6 LUTs (can operate as one 6-LUT or two 5-LUTs with 8 total FLE inputs (2 inputs of which are shared by the 5-LUTs)
with optionally registered outputs
Each 5-LUT has an arithemtic mode that converts it to a single-bit adder with both inputs driven by 4-LUTs (both 4-LUTs share all 4 inputs)
Carry chain links to vertically adjacent logic blocks
- Memory size 32 Kbits, memory aspect ratios vary from a data width of 1 to data width of 64.
Height = 6, found on every (8n+2)th column
- Multiplier modes: one 36x36, two 18x18, each 18x18 can also operate as two 9x9.
Height = 4, found on every (8n+6)th column
- Routing architecture: L = 4, fc_in = 0.15, Fc_out = 0.1
Details on Modelling:
The electrical design of the architecture described here is NOT from an
optimized, SPICED architecture. Instead, we attempt to create a reasonable
architecture file by using an existing commercial FPGA to approximate the area,
delay, and power of the underlying components. This is combined with a reasonable 40 nm
model of wiring and circuit design for low-level routing components, where available.
The resulting architecture has delays that roughly match a commercial 40 nm FPGA, but also
has wiring electrical parameters that allow the wire lengths and switch patterns to be
modified and you will still get reasonable delay results for the new architecture.
The following describes, in detail, how we obtained the various electrical values for this
architecture.
Rmin for nmos and pmos, routing buffer sizes, and I/O pad delays are from the ifar
architecture created by Ian Kuon: K06 N10 45nm fc 0.15 area-delay optimized architecture.
(n10k06l04.fc15.area1delay1.cmos45nm.bptm.cmos45nm.xml)
This routing architecture was optimized for 45 nm, and we have scaled it linearly to 40 nm to
match the overall target (a 40 nm FPGA).
We obtain delay numbers by measuring delays of routing, soft logic blocks,
memories, and multipliers from test circuits on a Stratix IV GX device
(EP4SGX230DF29C2X, i.e. fastest speed grade). For routing, we took the average delay of H4 and V4
wires. Rmetal and Cmetal values for the routing wires were obtained from work done by Charles
Chiasson. We use a 96 nm half-pitch (corresponding to mid-level metal stack 40 nm routing) and
take the R and C data from the ITRS roadmap.
For the general purpose logic block, we assume that the area and delays of the Stratix IV
crossbar is close enough to the crossbar modelled here.
Stratix IV uses 52 inputs and 20 feedback lines, but only a half-populated crossbar, leading to
36:1 multiplexers. We match these parameters in this architecture.
For LUTs, we include LUT
delays measured from Stratix IV which is dependant on the input used (ie. some
LUT inputs are faster than others). The CAD tools at the time of VTR 7 does
not consider differences in LUT input delays.
Adder delays obtained as approximate values from a Stratix IV EP4SE230F29C3 device.
Delay obtained by compiling a 256 bit adder (registered inputs and outputs,
all pins except clock virtual) then measuring the delays in chip-planner,
sumout delay = 0.271ns to 0.348 ns, intra-block carry delay = 0.011 ns,
inter-block carry delay = 0.327 ns. Given this data, I will approximate
sumout 0.3 ns, intra-block carry-delay = 0.01 ns, and
inter-block carry-delay = 0.16 ns (since Altera inter-block carry delay has
overhead that we don't have, I'll approximate the delay of a simpler chain at
one half what they have. This is very rough, anything from 0.01ns to 0.327ns
can be justified).
Logic block area numbers obtained by scaling overall tile area of a 65nm
Stratix III device, (as given in Wong, Betz and Rose, FPGA 2011) to 40 nm, then subtracting out
routing area at a channel width of 300. We use a channel width of 300 because it can route
all the VTR 6.0 benchmark circuits with an approximately 20% safety margin, and is also close to the
total channel width of Stratix IV. Hence this channel width is close to the commercial practice of
choosing a width that provides high routability. The architecture can be routed at different channel
widths, but we estimate the tile size and hence the physical length of routing wires assuming
a channel width of 300.
Sanity checks employed:
1. We confirmed the routing buffer delay is ~1/3rd of total routing delay at L = 4. This matches
common electrical design.
Authors: Jason Luu, Jeff Goeders, Vaughn Betz
-->
<architecture>
<!--
ODIN II specific config begins
Describes the types of user-specified netlist blocks (in blif, this corresponds to
".model [type_of_block]") that this architecture supports.
Note: Basic LUTs, I/Os, and flip-flops are not included here as there are
already special structures in blif (.names, .input, .output, and .latch)
that describe them.
-->
<models>
<model name="adder">
<input_ports>
<port name="a" combinational_sink_ports="sumout cout"/>
<port name="b" combinational_sink_ports="sumout cout"/>
<port name="cin" combinational_sink_ports="sumout cout"/>
</input_ports>
<output_ports>
<port name="cout"/>
<port name="sumout"/>
</output_ports>
</model>
<!-- A virtual model for I/O to be used in the physical mode of io block -->
<model name="io">
<input_ports>
<port name="outpad"/>
</input_ports>
<output_ports>
<port name="inpad"/>
</output_ports>
</model>
<!-- A virtual model for I/O to be used in the physical mode of io block -->
<model name="frac_lut6">
<input_ports>
<port name="in"/>
</input_ports>
<output_ports>
<port name="lut4_out"/>
<port name="lut5_out"/>
<port name="lut6_out"/>
</output_ports>
</model>
<model name="dpram_1024x8">
<input_ports>
<!-- write address lines -->
<port name="waddr" clock="clk"/>
<!-- read address lines -->
<port name="raddr" clock="clk"/>
<!-- data lines can be broken down into smaller bit widths minimum size 1 -->
<port name="data_in" clock="clk"/>
<!-- write enable -->
<port name="wen" clock="clk"/>
<!-- read enable -->
<port name="ren" clock="clk"/>
<!-- memories are often clocked -->
<port name="clk" is_clock="1"/>
</input_ports>
<output_ports>
<!-- output can be broken down into smaller bit widths minimum size 1 -->
<port name="data_out" clock="clk"/>
</output_ports>
</model>
<model name="mult_36">
<input_ports>
<port name="A" combinational_sink_ports="Y"/>
<port name="B" combinational_sink_ports="Y"/>
</input_ports>
<output_ports>
<port name="Y"/>
</output_ports>
</model>
</models>
<tiles>
<tile name="io" capacity="8" area="0">
<equivalent_sites>
<site pb_type="io"/>
</equivalent_sites>
<input name="outpad" num_pins="1"/>
<output name="inpad" num_pins="1"/>
<fc in_type="frac" in_val="0.15" out_type="frac" out_val="0.10"/>
<pinlocations pattern="custom">
<loc side="left">io.outpad io.inpad</loc>
<loc side="top">io.outpad io.inpad</loc>
<loc side="right">io.outpad io.inpad</loc>
<loc side="bottom">io.outpad io.inpad</loc>
</pinlocations>
</tile>
<tile name="clb" area="53894">
<equivalent_sites>
<site pb_type="clb"/>
</equivalent_sites>
<input name="I" num_pins="40" equivalent="full"/>
<input name="cin" num_pins="1"/>
<output name="O" num_pins="20" equivalent="none"/>
<output name="cout" num_pins="1"/>
<clock name="clk" num_pins="1"/>
<fc in_type="frac" in_val="0.15" out_type="frac" out_val="0.10">
<fc_override port_name="cin" fc_type="frac" fc_val="0"/>
<fc_override port_name="cout" fc_type="frac" fc_val="0"/>
</fc>
<!-- Highly recommand to customize pin location when direct connection is used!!! -->
<!--pinlocations pattern="spread"/-->
<pinlocations pattern="custom">
<loc side="left">clb.clk</loc>
<loc side="top">clb.cin</loc>
<loc side="right">clb.O[9:0] clb.I[19:0]</loc>
<loc side="bottom">clb.cout clb.O[19:10] clb.I[39:20]</loc>
</pinlocations>
</tile>
<tile name="memory" height="2" area="548000">
<equivalent_sites>
<site pb_type="memory"/>
</equivalent_sites>
<input name="waddr" num_pins="10"/>
<input name="raddr" num_pins="10"/>
<input name="data_in" num_pins="8"/>
<input name="wen" num_pins="1"/>
<input name="ren" num_pins="1"/>
<output name="data_out" num_pins="8"/>
<clock name="clk" num_pins="1"/>
<fc in_type="frac" in_val="0.15" out_type="frac" out_val="0.10"/>
<!--pinlocations pattern="spread"/-->
<pinlocations pattern="custom">
<loc side="left">memory.clk</loc>
<loc side="top"></loc>
<loc side="right">memory.waddr[4:0] memory.raddr[4:0] memory.data_in[3:0] memory.wen memory.data_out[3:0]</loc>
<loc side="bottom">memory.waddr[9:5] memory.raddr[9:5] memory.data_in[7:4] memory.ren memory.data_out[7:4]</loc>
</pinlocations>
</tile>
<tile name="mult_36" height="6" area="396000">
<equivalent_sites>
<site pb_type="mult_36" pin_mapping="direct"/>
</equivalent_sites>
<input name="a" num_pins="36"/>
<input name="b" num_pins="36"/>
<output name="out" num_pins="72"/>
<fc in_type="frac" in_val="0.15" out_type="frac" out_val="0.10"/>
<!-- Highly recommand to customize pin location when direct connection is used!!! -->
<!-- pinlocations are designed to spread pin on 4 sides evenly -->
<pinlocations pattern="custom">
<loc side="left">mult_36.b[0:9] mult_36.b[10:35] mult_36.out[36:71]</loc>
<loc side="top"></loc>
<loc side="right">mult_36.a[0:9] mult_36.a[10:35] mult_36.out[0:35]</loc>
<loc side="bottom"></loc>
</pinlocations>
</tile>
</tiles>
<!-- ODIN II specific config ends -->
<!-- Physical descriptions begin -->
<layout tileable="true" through_channel="false">
<auto_layout aspect_ratio="1.0">
<!--Perimeter of 'io' blocks with 'EMPTY' blocks at corners-->
<perimeter type="io" priority="100"/>
<corners type="EMPTY" priority="101"/>
<!--Fill with 'clb'-->
<fill type="clb" priority="10"/>
<!--Column of 'memory' with 'EMPTY' blocks wherever a 'memory' does not fit. Vertical offset by 1 for perimeter.-->
<col type="memory" startx="2" starty="1" repeatx="8" priority="20"/>
<!--Column of 'mult_36' with 'EMPTY' blocks wherever a 'mult_36' does not fit. Vertical offset by 1 for perimeter.-->
<col type="mult_36" startx="6" starty="1" repeatx="8" priority="20"/>
<col type="EMPTY" startx="2" repeatx="8" starty="1" priority="19"/>
</auto_layout>
<fixed_layout name="3x2" width="5" height="4">
<!--Perimeter of 'io' blocks with 'EMPTY' blocks at corners-->
<perimeter type="io" priority="100"/>
<corners type="EMPTY" priority="101"/>
<!--Fill with 'clb'-->
<fill type="clb" priority="10"/>
<!--Column of 'memory' with 'EMPTY' blocks wherever a 'memory' does not fit. Vertical offset by 1 for perimeter.-->
<col type="memory" startx="2" starty="1" repeatx="8" priority="20"/>
<col type="EMPTY" startx="2" repeatx="8" starty="1" priority="19"/>
</fixed_layout>
</layout>
<device>
<!-- VB & JL: Using Ian Kuon's transistor sizing and drive strength data for routing, at 40 nm. Ian used BPTM
models. We are modifying the delay values however, to include metal C and R, which allows more architecture
experimentation. We are also modifying the relative resistance of PMOS to be 1.8x that of NMOS
(vs. Ian's 3x) as 1.8x lines up with Jeff G's data from a 45 nm process (and is more typical of
45 nm in general). I'm upping the Rmin_nmos from Ian's just over 6k to nearly 9k, and dropping
RminW_pmos from 18k to 16k to hit this 1.8x ratio, while keeping the delays of buffers approximately
lined up with Stratix IV.
We are using Jeff G.'s capacitance data for 45 nm (in tech/ptm_45nm).
Jeff's tables list C in for transistors with widths in multiples of the minimum feature size (45 nm).
The minimum contactable transistor is 2.5 * 45 nm, so I need to multiply drive strength sizes in this file
by 2.5x when looking up in Jeff's tables.
The delay values are lined up with Stratix IV, which has an architecture similar to this
proposed FPGA, and which is also 40 nm
C_ipin_cblock: input capacitance of a track buffer, which VPR assumes is a single-stage
4x minimum drive strength buffer. -->
<sizing R_minW_nmos="8926" R_minW_pmos="16067"/>
<!-- The grid_logic_tile_area below will be used for all blocks that do not explicitly set their own (non-routing)
area; set to 0 since we explicitly set the area of all blocks currently in this architecture file.
-->
<area grid_logic_tile_area="0"/>
<chan_width_distr>
<x distr="uniform" peak="1.000000"/>
<y distr="uniform" peak="1.000000"/>
</chan_width_distr>
<switch_block type="wilton" fs="3"/>
<connection_block input_switch_name="ipin_cblock"/>
</device>
<switchlist>
<!-- VB: the mux_trans_size and buf_size data below is in minimum width transistor *areas*, assuming the purple
book area formula. This means the mux transistors are about 5x minimum drive strength.
We assume the first stage of the buffer is 3x min drive strength to be reasonable given the large
mux transistors, and this gives a reasonable stage ratio of a bit over 5x to the second stage. We assume
the n and p transistors in the first stage are equal-sized to lower the buffer trip point, since it's fed
by a pass transistor mux. We can then reverse engineer the buffer second stage to hit the specified
buf_size (really buffer area) - 16.2x minimum drive nmos and 1.8*16.2 = 29.2x minimum drive.
I then took the data from Jeff G.'s PTM modeling of 45 nm to get the Cin (gate of first stage) and Cout
(diff of second stage) listed below. Jeff's models are in tech/ptm_45nm, and are in min feature multiples.
The minimum contactable transistor is 2.5 * 45 nm, so I need to multiply the drive strength sizes above by
2.5x when looking up in Jeff's tables.
Finally, we choose a switch delay (58 ps) that leads to length 4 wires having a delay equal to that of SIV of 126 ps.
This also leads to the switch being 46% of the total wire delay, which is reasonable. -->
<switch type="mux" name="0" R="551" Cin=".77e-15" Cout="4e-15" Tdel="58e-12" mux_trans_size="2.630740" buf_size="27.645901"/>
<!--switch ipin_cblock resistance set to yeild for 4x minimum drive strength buffer-->
<switch type="mux" name="ipin_cblock" R="2231.5" Cout="0." Cin="1.47e-15" Tdel="7.247000e-11" mux_trans_size="1.222260" buf_size="auto"/>
</switchlist>
<segmentlist>
<!--- VB & JL: using ITRS metal stack data, 96 nm half pitch wires, which are intermediate metal width/space.
With the 96 nm half pitch, such wires would take 60 um of height, vs. a 90 nm high (approximated as square) Stratix IV tile so this seems
reasonable. Using a tile length of 90 nm, corresponding to the length of a Stratix IV tile if it were square. -->
<!-- GIVE a specific name for the segment! OpenFPGA appreciate that! -->
<segment name="L4" freq="1.000000" length="4" type="unidir" Rmetal="101" Cmetal="22.5e-15">
<mux name="0"/>
<sb type="pattern">1 1 1 1 1</sb>
<cb type="pattern">1 1 1 1</cb>
</segment>
</segmentlist>
<directlist>
<direct name="adder_carry" from_pin="clb.cout" to_pin="clb.cin" x_offset="0" y_offset="-1" z_offset="0"/>
</directlist>
<complexblocklist>
<!-- Define I/O pads begin -->
<!-- Capacity is a unique property of I/Os, it is the maximum number of I/Os that can be placed at the same (X,Y) location on the FPGA -->
<!-- Not sure of the area of an I/O (varies widely), and it's not relevant to the design of the FPGA core, so we're setting it to 0. -->
<pb_type name="io">
<input name="outpad" num_pins="1"/>
<output name="inpad" num_pins="1"/>
<!-- Do NOT add clock pins to I/O here!!! VPR does not build clock network in the way that OpenFPGA can support
If you need to register the I/O, define clocks in the circuit models
These clocks can be handled in back-end
-->
<!-- A mode denotes the physical implementation of an I/O
This mode will be not packable but is mainly used for fabric verilog generation
-->
<mode name="physical" disable_packing="true">
<pb_type name="iopad" blif_model=".subckt io" num_pb="1">
<input name="outpad" num_pins="1"/>
<output name="inpad" num_pins="1"/>
</pb_type>
<interconnect>
<direct name="outpad" input="io.outpad" output="iopad.outpad">
<delay_constant max="1.394e-11" in_port="io.outpad" out_port="iopad.outpad"/>
</direct>
<direct name="inpad" input="iopad.inpad" output="io.inpad">
<delay_constant max="4.243e-11" in_port="iopad.inpad" out_port="io.inpad"/>
</direct>
</interconnect>
</mode>
<!-- IOs can operate as either inputs or outputs.
Delays below come from Ian Kuon. They are small, so they should be interpreted as
the delays to and from registers in the I/O (and generally I/Os are registered
today and that is when you timing analyze them.
-->
<mode name="inpad">
<pb_type name="inpad" blif_model=".input" num_pb="1">
<output name="inpad" num_pins="1"/>
</pb_type>
<interconnect>
<direct name="inpad" input="inpad.inpad" output="io.inpad">
<delay_constant max="4.243e-11" in_port="inpad.inpad" out_port="io.inpad"/>
</direct>
</interconnect>
</mode>
<mode name="outpad">
<pb_type name="outpad" blif_model=".output" num_pb="1">
<input name="outpad" num_pins="1"/>
</pb_type>
<interconnect>
<direct name="outpad" input="io.outpad" output="outpad.outpad">
<delay_constant max="1.394e-11" in_port="io.outpad" out_port="outpad.outpad"/>
</direct>
</interconnect>
</mode>
<!-- Every input pin is driven by 15% of the tracks in a channel, every output pin is driven by 10% of the tracks in a channel -->
<!-- IOs go on the periphery of the FPGA, for consistency,
make it physically equivalent on all sides so that only one definition of I/Os is needed.
If I do not make a physically equivalent definition, then I need to define 4 different I/Os, one for each side of the FPGA
-->
<!-- Place I/Os on the sides of the FPGA -->
<power method="ignore"/>
</pb_type>
<!-- Define I/O pads ends -->
<!-- Define general purpose logic block (CLB) begin -->
<!--- Area calculation: Total Stratix IV tile area is about 8100 um^2, and a minimum width transistor
area is 60 L^2 yields a tile area of 84375 MWTAs.
Routing at W=300 is 30481 MWTAs, leaving us with a total of 53000 MWTAs for logic block area
This means that only 37% of our area is in the general routing, and 63% is inside the logic
block. Note that the crossbar / local interconnect is considered part of the logic block
area in this analysis. That is a lower proportion of of routing area than most academics
assume, but note that the total routing area really includes the crossbar, which would push
routing area up significantly, we estimate into the ~70% range.
-->
<pb_type name="clb">
<input name="I" num_pins="40" equivalent="full"/>
<input name="cin" num_pins="1"/>
<output name="O" num_pins="20" equivalent="none"/>
<output name="cout" num_pins="1"/>
<clock name="clk" num_pins="1"/>
<!-- Describe fracturable logic element.
Each fracturable logic element has a 6-LUT that can alternatively operate as two 5-LUTs with shared inputs.
The outputs of the fracturable logic element can be optionally registered
-->
<pb_type name="fle" num_pb="10">
<input name="in" num_pins="6"/>
<input name="cin" num_pins="1"/>
<output name="out" num_pins="2"/>
<output name="cout" num_pins="1"/>
<clock name="clk" num_pins="1"/>
<!-- Physical mode definition begin (physical implementation of the fle) -->
<mode name="physical" disable_packing="true">
<pb_type name="fabric" num_pb="1">
<input name="in" num_pins="6"/>
<input name="cin" num_pins="1"/>
<output name="out" num_pins="2"/>
<output name="cout" num_pins="1"/>
<clock name="clk" num_pins="1"/>
<pb_type name="frac_logic" num_pb="1">
<input name="in" num_pins="6"/>
<output name="lut4_out" num_pins="4"/>
<output name="out" num_pins="2"/>
<!-- Define LUT -->
<pb_type name="frac_lut6" blif_model=".subckt frac_lut6" num_pb="1">
<input name="in" num_pins="6"/>
<output name="lut4_out" num_pins="4"/>
<output name="lut5_out" num_pins="2"/>
<output name="lut6_out" num_pins="1"/>
</pb_type>
<interconnect>
<direct name="direct1" input="frac_logic.in" output="frac_lut6.in"/>
<direct name="direct2" input="frac_lut6.lut4_out" output="frac_logic.lut4_out"/>
<direct name="direct3" input="frac_lut6.lut5_out[1]" output="frac_logic.out[1]"/>
<!-- Xifan Tang: I use out[0] because the output of lut6 in lut6 mode is wired to the out[0] -->
<mux name="mux1" input="frac_lut6.lut6_out frac_lut6.lut5_out[0]" output="frac_logic.out[0]"/>
</interconnect>
</pb_type>
<!-- Define flip-flop -->
<pb_type name="ff" blif_model=".latch" num_pb="2" class="flipflop">
<input name="D" num_pins="1" port_class="D"/>
<output name="Q" num_pins="1" port_class="Q"/>
<clock name="clk" num_pins="1" port_class="clock"/>
<T_setup value="66e-12" port="ff.D" clock="clk"/>
<T_clock_to_Q max="124e-12" port="ff.Q" clock="clk"/>
</pb_type>
<!-- Define adders -->
<pb_type name="adder" blif_model=".subckt adder" num_pb="2">
<input name="a" num_pins="1"/>
<input name="b" num_pins="1"/>
<input name="cin" num_pins="1"/>
<output name="cout" num_pins="1"/>
<output name="sumout" num_pins="1"/>
<delay_constant max="0.3e-9" in_port="adder.a" out_port="adder.sumout"/>
<delay_constant max="0.3e-9" in_port="adder.b" out_port="adder.sumout"/>
<delay_constant max="0.3e-9" in_port="adder.cin" out_port="adder.sumout"/>
<delay_constant max="0.3e-9" in_port="adder.a" out_port="adder.cout"/>
<delay_constant max="0.3e-9" in_port="adder.b" out_port="adder.cout"/>
<delay_constant max="0.01e-9" in_port="adder.cin" out_port="adder.cout"/>
</pb_type>
<interconnect>
<direct name="direct1" input="fabric.in" output="frac_logic.in"/>
<direct name="direct2" input="frac_logic.out[1:0]" output="ff[1:0].D"/>
<direct name="direct3" input="fabric.cin" output="adder[0:0].cin"/>
<direct name="direct4" input="adder[0:0].cout" output="adder[1:1].cin"/>
<direct name="direct5" input="adder[1:1].cout" output="fabric.cout"/>
<direct name="direct6" input="frac_logic.lut4_out[0:0]" output="adder[0:0].a"/>
<direct name="direct7" input="frac_logic.lut4_out[1:1]" output="adder[0:0].b"/>
<direct name="direct8" input="frac_logic.lut4_out[2:2]" output="adder[1:1].a"/>
<direct name="direct9" input="frac_logic.lut4_out[3:3]" output="adder[1:1].b"/>
<complete name="direct10" input="fabric.clk" output="ff[1:0].clk"/>
<mux name="mux1" input="adder[0].sumout ff[0].Q frac_logic.out[0]" output="fabric.out[0]">
<!-- LUT to output is faster than FF to output on a Stratix IV -->
<delay_constant max="25e-12" in_port="frac_logic.out[0]" out_port="fabric.out[0]"/>
<delay_constant max="45e-12" in_port="ff[0].Q" out_port="fabric.out[0]"/>
</mux>
<mux name="mux2" input="adder[1].sumout ff[1].Q frac_logic.out[1]" output="fabric.out[1]">
<!-- LUT to output is faster than FF to output on a Stratix IV -->
<delay_constant max="25e-12" in_port="frac_logic.out[1]" out_port="fabric.out[1]"/>
<delay_constant max="45e-12" in_port="ff[1].Q" out_port="fabric.out[1]"/>
</mux>
</interconnect>
</pb_type>
<interconnect>
<direct name="direct1" input="fle.in" output="fabric.in"/>
<direct name="direct2" input="fle.cin" output="fabric.cin"/>
<direct name="direct3" input="fabric.out" output="fle.out"/>
<direct name="direct4" input="fabric.cout" output="fle.cout"/>
<direct name="direct5" input="fle.clk" output="fabric.clk"/>
</interconnect>
</mode>
<!-- Physical mode definition end (physical implementation of the fle) -->
<!-- BEGIN fle mode of dual lut5 -->
<mode name="n2_lut5">
<pb_type name="ble5" num_pb="2">
<input name="in" num_pins="5"/>
<output name="out" num_pins="1"/>
<clock name="clk" num_pins="1"/>
<!-- Regular LUT mode -->
<pb_type name="lut5" blif_model=".names" num_pb="1" class="lut">
<input name="in" num_pins="5" port_class="lut_in"/>
<output name="out" num_pins="1" port_class="lut_out"/>
<!-- LUT timing using delay matrix -->
<!-- These are the physical delay inputs on a Stratix IV LUT but because VPR cannot do LUT rebalancing,
we instead take the average of these numbers to get more stable results
82e-12
173e-12
261e-12
263e-12
398e-12
-->
<delay_matrix type="max" in_port="lut5.in" out_port="lut5.out">
235e-12
235e-12
235e-12
235e-12
235e-12
</delay_matrix>
</pb_type>
<pb_type name="ff" blif_model=".latch" num_pb="1" class="flipflop">
<input name="D" num_pins="1" port_class="D"/>
<output name="Q" num_pins="1" port_class="Q"/>
<clock name="clk" num_pins="1" port_class="clock"/>
<T_setup value="66e-12" port="ff.D" clock="clk"/>
<T_clock_to_Q max="124e-12" port="ff.Q" clock="clk"/>
</pb_type>
<interconnect>
<direct name="direct1" input="ble5.in" output="lut5.in"/>
<direct name="direct2" input="lut5.out" output="ff.D">
<pack_pattern name="ble5" in_port="lut5.out" out_port="ff.D"/>
</direct>
<direct name="direct3" input="ble5.clk" output="ff.clk"/>
<mux name="mux1" input="ff.Q lut5.out" output="ble5.out">
<delay_constant max="25e-12" in_port="lut5.out" out_port="ble5.out"/>
<delay_constant max="45e-12" in_port="ff.Q" out_port="ble5.out"/>
</mux>
</interconnect>
</pb_type>
<interconnect>
<direct name="direct1" input="fle.in[4:0]" output="ble5[0:0].in"/>
<direct name="direct2" input="fle.in[4:0]" output="ble5[1:1].in"/>
<complete name="direct3" input="fle.clk" output="ble5.clk"/>
<direct name="direct4" input="ble5.out" output="fle.out"/>
</interconnect>
</mode>
<!-- END fle mode of dual lut5 -->
<!-- BEGIN arithmetic mode of dual lut4 + adders -->
<mode name="arithmetic">
<pb_type name="arithmetic" num_pb="2">
<input name="in" num_pins="4"/>
<input name="cin" num_pins="1"/>
<output name="out" num_pins="1"/>
<output name="cout" num_pins="1"/>
<clock name="clk" num_pins="1"/>
<!-- Special dual-LUT mode that drives adder only -->
<pb_type name="lut4" blif_model=".names" num_pb="2" class="lut">
<input name="in" num_pins="4" port_class="lut_in"/>
<output name="out" num_pins="1" port_class="lut_out"/>
<!-- LUT timing using delay matrix -->
<!-- These are the physical delay inputs on a Stratix IV LUT but because VPR cannot do LUT rebalancing,
we instead take the average of these numbers to get more stable results
82e-12
173e-12
261e-12
263e-12
-->
<delay_matrix type="max" in_port="lut4.in" out_port="lut4.out">
195e-12
195e-12
195e-12
195e-12
</delay_matrix>
</pb_type>
<pb_type name="adder" blif_model=".subckt adder" num_pb="1">
<input name="a" num_pins="1"/>
<input name="b" num_pins="1"/>
<input name="cin" num_pins="1"/>
<output name="cout" num_pins="1"/>
<output name="sumout" num_pins="1"/>
<delay_constant max="0.3e-9" in_port="adder.a" out_port="adder.sumout"/>
<delay_constant max="0.3e-9" in_port="adder.b" out_port="adder.sumout"/>
<delay_constant max="0.3e-9" in_port="adder.cin" out_port="adder.sumout"/>
<delay_constant max="0.3e-9" in_port="adder.a" out_port="adder.cout"/>
<delay_constant max="0.3e-9" in_port="adder.b" out_port="adder.cout"/>
<delay_constant max="0.01e-9" in_port="adder.cin" out_port="adder.cout"/>
</pb_type>
<pb_type name="ff" blif_model=".latch" num_pb="1" class="flipflop">
<input name="D" num_pins="1" port_class="D"/>
<output name="Q" num_pins="1" port_class="Q"/>
<clock name="clk" num_pins="1" port_class="clock"/>
<T_setup value="66e-12" port="ff.D" clock="clk"/>
<T_clock_to_Q max="124e-12" port="ff.Q" clock="clk"/>
</pb_type>
<interconnect>
<direct name="clock" input="arithmetic.clk" output="ff.clk"/>
<direct name="lut_in1" input="arithmetic.in[3:0]" output="lut4[0:0].in[3:0]"/>
<direct name="lut_in2" input="arithmetic.in[3:0]" output="lut4[1:1].in[3:0]"/>
<direct name="lut_to_add1" input="lut4[0:0].out" output="adder.a">
</direct>
<direct name="lut_to_add2" input="lut4[1:1].out" output="adder.b">
</direct>
<direct name="add_to_ff" input="adder.sumout" output="ff.D">
<pack_pattern name="chain" in_port="adder.sumout" out_port="ff.D"/>
</direct>
<direct name="carry_in" input="arithmetic.cin" output="adder.cin">
<pack_pattern name="chain" in_port="arithmetic.cin" out_port="adder.cin"/>
</direct>
<direct name="carry_out" input="adder.cout" output="arithmetic.cout">
<pack_pattern name="chain" in_port="adder.cout" out_port="arithmetic.cout"/>
</direct>
<mux name="sumout" input="ff.Q adder.sumout" output="arithmetic.out">
<delay_constant max="25e-12" in_port="adder.sumout" out_port="arithmetic.out"/>
<delay_constant max="45e-12" in_port="ff.Q" out_port="arithmetic.out"/>
</mux>
</interconnect>
</pb_type>
<interconnect>
<direct name="direct1" input="fle.in[3:0]" output="arithmetic[0:0].in"/>
<direct name="direct2" input="fle.in[3:0]" output="arithmetic[1:1].in"/>
<direct name="carry_in" input="fle.cin" output="arithmetic[0:0].cin">
<pack_pattern name="chain" in_port="fle.cin" out_port="arithmetic[0:0].cin"/>
</direct>
<direct name="carry_inter" input="arithmetic[0:0].cout" output="arithmetic[1:1].cin">
<pack_pattern name="chain" in_port="arithmetic[0:0].cout" out_port="arithmetic[1:1].cin"/>
</direct>
<direct name="carry_out" input="arithmetic[1:1].cout" output="fle.cout">
<pack_pattern name="chain" in_port="arithmetic.cout" out_port="fle.cout"/>
</direct>
<complete name="direct3" input="fle.clk" output="arithmetic.clk"/>
<direct name="direct4" input="arithmetic.out" output="fle.out"/>
</interconnect>
</mode>
<!-- n2_lut5 -->
<mode name="n1_lut6">
<pb_type name="ble6" num_pb="1">
<input name="in" num_pins="6"/>
<output name="out" num_pins="1"/>
<clock name="clk" num_pins="1"/>
<pb_type name="lut6" blif_model=".names" num_pb="1" class="lut">
<input name="in" num_pins="6" port_class="lut_in"/>
<output name="out" num_pins="1" port_class="lut_out"/>
<!-- LUT timing using delay matrix -->
<!-- These are the physical delay inputs on a Stratix IV LUT but because VPR cannot do LUT rebalancing,
we instead take the average of these numbers to get more stable results
82e-12
173e-12
261e-12
263e-12
398e-12
397e-12
-->
<delay_matrix type="max" in_port="lut6.in" out_port="lut6.out">
261e-12
261e-12
261e-12
261e-12
261e-12
261e-12
</delay_matrix>
</pb_type>
<pb_type name="ff" blif_model=".latch" num_pb="1" class="flipflop">
<input name="D" num_pins="1" port_class="D"/>
<output name="Q" num_pins="1" port_class="Q"/>
<clock name="clk" num_pins="1" port_class="clock"/>
<T_setup value="66e-12" port="ff.D" clock="clk"/>
<T_clock_to_Q max="124e-12" port="ff.Q" clock="clk"/>
</pb_type>
<interconnect>
<direct name="direct1" input="ble6.in" output="lut6[0:0].in"/>
<direct name="direct2" input="lut6.out" output="ff.D">
<pack_pattern name="ble6" in_port="lut6.out" out_port="ff.D"/>
</direct>
<direct name="direct3" input="ble6.clk" output="ff.clk"/>
<mux name="mux1" input="ff.Q lut6.out" output="ble6.out">
<delay_constant max="25e-12" in_port="lut6.out" out_port="ble6.out"/>
<delay_constant max="45e-12" in_port="ff.Q" out_port="ble6.out"/>
</mux>
</interconnect>
</pb_type>
<interconnect>
<direct name="direct1" input="fle.in[5:0]" output="ble6.in"/>
<direct name="direct2" input="ble6.out" output="fle.out[0:0]"/>
<direct name="direct3" input="fle.clk" output="ble6.clk"/>
</interconnect>
</mode>
<!-- n1_lut6 -->
</pb_type>
<interconnect>
<!-- We use a 50% depop crossbar built using small full xbars to get sets of logically equivalent pins at inputs of CLB
The delays below come from Stratix IV. the delay through a connection block
input mux + the crossbar in Stratix IV is 167 ps. We already have a 72 ps
delay on the connection block input mux (modeled by Ian Kuon), so the remaining
delay within the crossbar is 95 ps.
The delays of cluster feedbacks in Stratix IV is 100 ps, when driven by a LUT.
Since all our outputs LUT outputs go to a BLE output, and have a delay of
25 ps to do so, we subtract 25 ps from the 100 ps delay of a feedback
to get the part that should be marked on the crossbar. -->
<complete name="crossbar" input="clb.I fle[9:0].out" output="fle[9:0].in">
<delay_constant max="95e-12" in_port="clb.I" out_port="fle[9:0].in"/>
<delay_constant max="75e-12" in_port="fle[9:0].out" out_port="fle[9:0].in"/>
</complete>
<complete name="clks" input="clb.clk" output="fle[9:0].clk">
</complete>
<!-- This way of specifying direct connection to clb outputs is important because this architecture uses automatic spreading of opins.
By grouping to output pins in this fashion, if a logic block is completely filled by 6-LUTs,
then the outputs those 6-LUTs take get evenly distributed across all four sides of the CLB instead of clumped on two sides (which is what happens with a more
naive specification).
-->
<direct name="clbouts1" input="fle[9:0].out[0:0]" output="clb.O[9:0]"/>
<direct name="clbouts2" input="fle[9:0].out[1:1]" output="clb.O[19:10]"/>
<!-- Carry chain links -->
<direct name="carry_in" input="clb.cin" output="fle[0:0].cin">
<!-- Put all inter-block carry chain delay on this one edge -->
<delay_constant max="0.16e-9" in_port="clb.cin" out_port="fle[0:0].cin"/>
<pack_pattern name="chain" in_port="clb.cin" out_port="fle[0:0].cin"/>
</direct>
<direct name="carry_out" input="fle[9:9].cout" output="clb.cout">
<pack_pattern name="chain" in_port="fle[9:9].cout" out_port="clb.cout"/>
</direct>
<direct name="carry_link" input="fle[8:0].cout" output="fle[9:1].cin">
<pack_pattern name="chain" in_port="fle[8:0].cout" out_port="fle[9:1].cin"/>
</direct>
</interconnect>
</pb_type>
<!-- Define general purpose logic block (CLB) ends -->
<!-- Define 36-bit multiplier begin -->
<pb_type name="mult_36">
<input name="a" num_pins="36"/>
<input name="b" num_pins="36"/>
<output name="out" num_pins="72"/>
<mode name="mult_36x36">
<pb_type name="mult_36x36_slice" num_pb="1">
<input name="A_cfg" num_pins="36"/>
<input name="B_cfg" num_pins="36"/>
<output name="OUT_cfg" num_pins="72"/>
<pb_type name="mult_36x36" blif_model=".subckt mult_36" num_pb="1">
<input name="A" num_pins="36"/>
<input name="B" num_pins="36"/>
<output name="Y" num_pins="72"/>
<delay_constant max="1.523e-9" min="0.776e-9" in_port="mult_36x36.A" out_port="mult_36x36.Y"/>
<delay_constant max="1.523e-9" min="0.776e-9" in_port="mult_36x36.B" out_port="mult_36x36.Y"/>
</pb_type>
<interconnect>
<direct name="a2a" input="mult_36x36_slice.A_cfg" output="mult_36x36.A">
</direct>
<direct name="b2b" input="mult_36x36_slice.B_cfg" output="mult_36x36.B">
</direct>
<direct name="out2out" input="mult_36x36.Y" output="mult_36x36_slice.OUT_cfg">
</direct>
</interconnect>
<power method="pin-toggle">
<port name="A_cfg" energy_per_toggle="2.13e-12"/>
<port name="B_cfg" energy_per_toggle="2.13e-12"/>
<static_power power_per_instance="0.0"/>
</power>
</pb_type>
<interconnect>
<!-- Stratix IV input delay of 207ps is conservative for this architecture because this architecture does not have an input crossbar in the multiplier.
Subtract 72.5 ps delay, which is already in the connection block input mux, leading
to a 134 ps delay.
The interconnect difference for DSP blocks is 0.5523, which leads to a minimum delay of 74 ps
-->
<direct name="a2a" input="mult_36.a" output="mult_36x36_slice.A_cfg">
<delay_constant max="134e-12" min="74e-12" in_port="mult_36.a" out_port="mult_36x36_slice.A_cfg"/>
</direct>
<direct name="b2b" input="mult_36.b" output="mult_36x36_slice.B_cfg">
<delay_constant max="134e-12" min="74e-12" in_port="mult_36.b" out_port="mult_36x36_slice.B_cfg"/>
</direct>
<direct name="out2out" input="mult_36x36_slice.OUT_cfg" output="mult_36.out">
<delay_constant max="1.93e-9" min="74e-12" in_port="mult_36x36_slice.OUT_cfg" out_port="mult_36.out"/>
</direct>
</interconnect>
</mode>
<!-- Place this multiplier block every 8 columns from (and including) the sixth column -->
<power method="sum-of-children"/>
</pb_type>
<!-- Define fracturable multiplier end -->
<!-- Define single-mode dual-port memory begin -->
<pb_type name="memory">
<input name="waddr" num_pins="10"/>
<input name="raddr" num_pins="10"/>
<input name="data_in" num_pins="8"/>
<input name="wen" num_pins="1"/>
<input name="ren" num_pins="1"/>
<output name="data_out" num_pins="8"/>
<clock name="clk" num_pins="1"/>
<!-- Specify the 1024x8=8Kbit memory block
Note: the delay numbers are extracted from VPR flagship XML without modification
Should align to the process technology we using to create the 16K dual-port RAM
-->
<mode name="mem_1024x8_dp">
<pb_type name="mem_1024x8_dp" blif_model=".subckt dpram_1024x8" num_pb="1">
<input name="waddr" num_pins="10" port_class="address1"/>
<input name="raddr" num_pins="10" port_class="address2"/>
<input name="data_in" num_pins="8" port_class="data_in1"/>
<input name="wen" num_pins="1" port_class="write_en1"/>
<input name="ren" num_pins="1" port_class="write_en2"/>
<output name="data_out" num_pins="8" port_class="data_out1"/>
<clock name="clk" num_pins="1" port_class="clock"/>
<T_setup value="509e-12" port="mem_1024x8_dp.waddr" clock="clk"/>
<T_setup value="509e-12" port="mem_1024x8_dp.raddr" clock="clk"/>
<T_setup value="509e-12" port="mem_1024x8_dp.data_in" clock="clk"/>
<T_setup value="509e-12" port="mem_1024x8_dp.wen" clock="clk"/>
<T_setup value="509e-12" port="mem_1024x8_dp.ren" clock="clk"/>
<T_clock_to_Q max="1.234e-9" port="mem_1024x8_dp.data_out" clock="clk"/>
<power method="pin-toggle">
<port name="clk" energy_per_toggle="17.9e-12"/>
<static_power power_per_instance="0.0"/>
</power>
</pb_type>
<interconnect>
<direct name="waddress" input="memory.waddr" output="mem_1024x8_dp.waddr">
<delay_constant max="132e-12" in_port="memory.waddr" out_port="mem_1024x8_dp.waddr"/>
</direct>
<direct name="raddress" input="memory.raddr" output="mem_1024x8_dp.raddr">
<delay_constant max="132e-12" in_port="memory.raddr" out_port="mem_1024x8_dp.raddr"/>
</direct>
<direct name="data_input" input="memory.data_in" output="mem_1024x8_dp.data_in">
<delay_constant max="132e-12" in_port="memory.data_in" out_port="mem_1024x8_dp.data_in"/>
</direct>
<direct name="writeen" input="memory.wen" output="mem_1024x8_dp.wen">
<delay_constant max="132e-12" in_port="memory.wen" out_port="mem_1024x8_dp.wen"/>
</direct>
<direct name="readen" input="memory.ren" output="mem_1024x8_dp.ren">
<delay_constant max="132e-12" in_port="memory.ren" out_port="mem_1024x8_dp.ren"/>
</direct>
<direct name="dataout" input="mem_1024x8_dp.data_out" output="memory.data_out">
<delay_constant max="40e-12" in_port="mem_1024x8_dp.data_out" out_port="memory.data_out"/>
</direct>
<direct name="clk" input="memory.clk" output="mem_1024x8_dp.clk">
</direct>
</interconnect>
</mode>
</pb_type>
<!-- Define single-mode dual-port memory end -->
</complexblocklist>
</architecture>

7
openfpga_flow/tasks/template_tasks/vtr_benchmarks_template/vtr_benchmark_template_script.openfpga Normal file
View File

@ -0,0 +1,7 @@
# Execute VPR for architecture exploration
vpr ${VPR_ARCH_FILE} ${VPR_TESTBENCH_BLIF} \
--route_chan_width ${VPR_ROUTE_CHAN_WIDTH} \
--constant_net_method route
exit