search

The-OpenROAD-Project / OpenROAD-flow-scripts

OpenROAD's scripts implementing an RTL-to-GDS Flow. Documentation at https://openroad-flow-scripts.readthedocs.io/en/latest/
https://theopenroadproject.org/
Other
334 stars 288 forks source link

Autotuner Results and Make Results inconsistent (Using parameters chosen by autotuner do make run), Make failes #1393

Closed dalyles closed 1 year ago

dalyles commented 1 year ago

Subject

[Documentation] for documentation errors.

Describe the bug

Hello,

I was running the autotuner on one of my circuits named picorv32 was successfully able to be generated without any violations. These are the parameter results that was determined for a clean run by the autotuner.

This can be found at the autotuner-best-xxxxxxxx.json

{
    "_SDC_CLK_PERIOD": 1,
    "CORE_UTILIZATION": 39,
    "CAP_MARGIN": 39,
    "CORE_ASPECT_RATIO": 1.1435000182737207,
    "CORE_MARGIN": 2,
    "CELL_PAD_IN_SITES_GLOBAL_PLACEMENT": 1,
    "CELL_PAD_IN_SITES_DETAIL_PLACEMENT": 0,
    "_FR_LAYER_ADJUST": 0.15900339639207847,
    "PLACE_DENSITY_LB_ADDON": 0.436940328162555,
    "_PINS_DISTANCE": 1,
    "CTS_CLUSTER_SIZE": 51,
    "CTS_CLUSTER_DIAMETER": 336,
    "_FR_GR_OVERFLOW": 1,
    "best_result": 92.52495
}

What I do then is copy these paramters and put them into the associated config.mk file like so:

export DESIGN_NAME = picorv32
export PLATFORM    = nangate45

export VERILOG_FILES = ./designs/src/$(DESIGN_NAME)/picorv32.v
export SDC_FILE      = ./designs/$(PLATFORM)/$(DESIGN_NAME)/constraint.sdc
export ABC_AREA      = 1

# Adders degrade GCD
export ADDER_MAP_FILE :=

export CORE_UTILIZATION ?= 55
export PLACE_DENSITY_LB_ADDON = 0.20
export TNS_END_PERCENT = 100

export _SDC_CLK_PERIOD = 15
export CORE_UTILIZATION = 39
export CAP_MARGIN = 39
export CORE_ASPECT_RATIO = 1.1435000182737207
export CORE_MARGIN = 2
export CELL_PAD_IN_SITES_GLOBAL_PLACEMENT = 1
export CELL_PAD_IN_SITES_DETAIL_PLACEMENT = 0
export _FR_LAYER_ADJUST = 0.15900339639207847
export PLACE_DENSITY_LB_ADDON = 0.436940328162555
export _PINS_DISTANCE = 1
export CTS_CLUSTER_SIZE = 51
export CTS_CLUSTER_DIAMETER = 336
export _FR_GR_OVERFLOW = 1

My expectation would be that given the parameters determined by the autotuner, this would not result in an error (see "Expected Behavior" below), Is there a possiblity that the way I'm setting up the fixed parameters is not correct?

Thank you

Expected Behavior

The result should be a clean circuit as was done by the autotuner. Here's a snapshot of 6_report.log results from the autotuner

==========================================================================
finish max_slew_violation_count
--------------------------------------------------------------------------
max slew violation count 0

==========================================================================
finish max_fanout_violation_count
--------------------------------------------------------------------------
max fanout violation count 0

==========================================================================
finish max_cap_violation_count
--------------------------------------------------------------------------
max cap violation count 0

==========================================================================
finish setup_violation_count
--------------------------------------------------------------------------
setup violation count 0

==========================================================================
finish hold_violation_count
--------------------------------------------------------------------------
hold violation count 0

==========================================================================
finish critical path delay
--------------------------------------------------------------------------
1.0632

==========================================================================
finish critical path slack
--------------------------------------------------------------------------
0.0748

==========================================================================
finish slack div critical path delay
--------------------------------------------------------------------------
7.035365

==========================================================================
finish report_power
--------------------------------------------------------------------------
Group                  Internal  Switching    Leakage      Total
                          Power      Power      Power      Power (Watts)
----------------------------------------------------------------
Sequential             1.25e-02   3.08e-03   1.33e-04   1.57e-02  27.8%
Combinational          1.75e-02   2.29e-02   2.93e-04   4.06e-02  72.2%
Macro                  0.00e+00   0.00e+00   0.00e+00   0.00e+00   0.0%
Pad                    0.00e+00   0.00e+00   0.00e+00   0.00e+00   0.0%
----------------------------------------------------------------
Total                  2.99e-02   2.60e-02   4.25e-04   5.63e-02 100.0%
                          53.2%      46.1%       0.8%

==========================================================================
finish report_design_area
--------------------------------------------------------------------------
Design area 18899 u^2 41% utilization.

libGL error: No matching fbConfigs or visuals found
libGL error: failed to load driver: swrast
Elapsed time: 0:37.10[h:]min:sec. CPU time: user 24.97 sys 0.64 (69%). Peak memory: 727844KB.

Below is the reult after running make on picorv32.v

==========================================================================
finish max_slew_violation_count
--------------------------------------------------------------------------
max slew violation count 0

==========================================================================
finish max_fanout_violation_count
--------------------------------------------------------------------------
max fanout violation count 0

==========================================================================
finish max_cap_violation_count
--------------------------------------------------------------------------
max cap violation count 0

==========================================================================
finish setup_violation_count
--------------------------------------------------------------------------
setup violation count 1

==========================================================================
finish hold_violation_count
--------------------------------------------------------------------------
hold violation count 0

==========================================================================
finish critical path delay
--------------------------------------------------------------------------
0.8677

==========================================================================
finish critical path slack
--------------------------------------------------------------------------
-0.2813

==========================================================================
finish slack div critical path delay
--------------------------------------------------------------------------
-32.419039

==========================================================================
finish report_power
--------------------------------------------------------------------------
Group                  Internal  Switching    Leakage      Total
                          Power      Power      Power      Power (Watts)
----------------------------------------------------------------
Sequential             2.70e-02   6.02e-03   1.33e-04   3.32e-02  20.5%
Combinational          6.13e-02   6.66e-02   4.86e-04   1.28e-01  79.5%
Macro                  0.00e+00   0.00e+00   0.00e+00   0.00e+00   0.0%
Pad                    0.00e+00   0.00e+00   0.00e+00   0.00e+00   0.0%
----------------------------------------------------------------
Total                  8.84e-02   7.26e-02   6.18e-04   1.62e-01 100.0%
                          54.7%      44.9%       0.4%

==========================================================================
finish report_design_area
--------------------------------------------------------------------------
Design area 23904 u^2 52% utilization.

libGL error: No matching fbConfigs or visuals found
libGL error: failed to load driver: swrast
Elapsed time: 0:24.83[h:]min:sec. CPU time: user 13.99 sys 0.37 (57%). Peak memory: 729004KB.

Environment

.

To Reproduce

1) Run auotuner with the following autotuner.json file:

{
    "_SDC_FILE_PATH": "constraint.sdc",
    "_SDC_CLK_PERIOD": {
        "type": "float",
        "minmax": [
            0.67,
            0.67
        ],
        "step": 0
    },
    "CORE_UTILIZATION": {
        "type": "int",
        "minmax": [
            5,
            100
        ],
        "step": 1
    },
    "CAP_MARGIN": {
            "type":"int",
            "minmax": [
                    5,
                    100
            ],
            "step": 1
    },
    "CORE_ASPECT_RATIO": {
        "type": "float",
        "minmax": [
            0.5,
            2.0
        ],
        "step": 0
    },
    "CORE_MARGIN": {
        "type": "int",
        "minmax": [
            2,
            2
        ],
        "step": 0
    },
    "CELL_PAD_IN_SITES_GLOBAL_PLACEMENT": {
        "type": "int",
        "minmax": [
            0,
            5
        ],
        "step": 1
    },
    "CELL_PAD_IN_SITES_DETAIL_PLACEMENT": {
        "type": "int",
        "minmax": [
            0,
            5
        ],
        "step": 1
    },
    "_FR_LAYER_ADJUST": {
        "type": "float",
        "minmax": [
            0.1,
            0.7
        ],
        "step": 0
    },
    "PLACE_DENSITY_LB_ADDON": {
        "type": "float",
        "minmax": [
            0.0,
            0.99
        ],
        "step": 0
    },
    "_PINS_DISTANCE": {
        "type": "int",
        "minmax": [
            1,
            4
        ],
        "step": 1
    },
    "CTS_CLUSTER_SIZE": {
        "type": "int",
        "minmax": [
            10,
            200
        ],
        "step": 1
    },
    "CTS_CLUSTER_DIAMETER": {
        "type": "int",
        "minmax": [
            20,
            400
        ],
        "step": 1
    },
    "_FR_FILE_PATH": "../../../platforms/nangate45/fastroute.tcl",
    "_FR_GR_OVERFLOW": {
        "type": "int",
        "minmax": [
            1,
            1
        ],
        "step": 0
    }
}

2) Get parameters determined by the autotuner and inset into config.mk 3) Run make with the given parameters, if the autotuner run is clean, the make run should be clean as well.

Relevant log output

No response

Screenshots

No response

Additional Context

No response

maliberty commented 1 year ago

The _ variables do not correspond to flow variables and are handled specially by the autotuner script. For example _SDC_CLOCK_PERIOD updates the set_clock_period value in the sdc.

dalyles commented 1 year ago

Is there some kind of correspondence or mapping with with variables such as _SDC_CLOCK_PERIOD, _FR_LAYER_ADJUST, _PINS_DISTANCE, and _FR_GR_OVERFLOW? I'd like to set those values directly in config.mk so I can be able to have a successful run.

Thank you

maliberty commented 1 year ago

The point is that they don't correspond to anything in config.mk and are handled specially. See https://github.com/The-OpenROAD-Project/OpenROAD-flow-scripts/blob/753d91cddbd990fcb82d0592d59645053f3b2d3a/flow/util/distributed.py#L328

dalyles commented 1 year ago

Hi @maliberty I see what you're saying here. According to the code, would it be as straightforward as removing the underscore and prefix for these variables? I've given this a try and it seems to have not yielded to a clean result. Here is what I have for my config.mk:

export DESIGN_NAME = picorv32
export PLATFORM    = nangate45

export VERILOG_FILES = ./designs/src/$(DESIGN_NAME)/picorv32.v
export SDC_FILE      = ./designs/$(PLATFORM)/$(DESIGN_NAME)/constraint.sdc
export ABC_AREA      = 1

# Adders degrade GCD
export ADDER_MAP_FILE :=

export CORE_UTILIZATION ?= 55
export PLACE_DENSITY_LB_ADDON = 0.20
export TNS_END_PERCENT = 100

export CLK_PERIOD = 1
export CORE_UTILIZATION = 39
export CAP_MARGIN = 39
export CORE_ASPECT_RATIO = 1.1435000182737207
export CORE_MARGIN = 2
export CELL_PAD_IN_SITES_GLOBAL_PLACEMENT = 1
export CELL_PAD_IN_SITES_DETAIL_PLACEMENT = 0
export LAYER_ADJUST = 0.15900339639207847
export PLACE_DENSITY_LB_ADDON = 0.436940328162555
export DISTANCE = 1
export CTS_CLUSTER_SIZE = 51
export CTS_CLUSTER_DIAMETER = 336
export GR_OVERFLOW = 1

I guess I have a follow-up question then: Does a clean run on the autotuner also imply that a run using make (with the autotuner parameters chosen) will also be clean?

maliberty commented 1 year ago

"would it be as straightforward as removing the underscore and prefix for these variables?" No - there is no variable you can set as I said earlier.

dalyles commented 1 year ago

I have a follow up question: If there are no such variables that can be set, how can you compare autotuner vs make runs? Is this even possible to replicate one or the other?

maliberty commented 1 year ago

You have to modify the corresponding files the same way that the autotuner does. For example you have to edit the sdc to set the clock period.

dalyles commented 1 year ago

Ok, I think I better understand what you're saying.

The autotuner already has the files generated for constraint.sdc and fastroute.tcl which includes the parameters that have the appropriate prefixes, so I can use that in my make run. As for the other parameters, I can manually place them in the config.mk file since they belong in the "options" section in distributed.py. Is this understanding correct?

vijayank88 commented 1 year ago

yes @dalyles

dalyles commented 1 year ago

I just want to confirm with you if this setup is correct (it worked for me, just need a sanity check) So for my parameter list which I will show below from the autotuner:

{
  "CAP_MARGIN": 39,
  "CELL_PAD_IN_SITES_DETAIL_PLACEMENT": 0,
  "CELL_PAD_IN_SITES_GLOBAL_PLACEMENT": 1,
  "CORE_ASPECT_RATIO": 1.1435000182737207,
  "CORE_MARGIN": 2,
  "CORE_UTILIZATION": 39,
  "CTS_CLUSTER_DIAMETER": 336,
  "CTS_CLUSTER_SIZE": 51,
  "PLACE_DENSITY_LB_ADDON": 0.436940328162555,
  "_FR_GR_OVERFLOW": 1,
  "_FR_LAYER_ADJUST": 0.15900339639207847,
  "_PINS_DISTANCE": 1,
  "_SDC_CLK_PERIOD": 1
}

I have the parameters distributed as follows:

config.mk

current_design picorv32

set clk_name  core_clock
set clk_port_name clk
set clk_period 1
set clk_io_pct 0.2

set clk_port [get_ports $clk_port_name]

create_clock -name $clk_name -period $clk_period $clk_port

set non_clock_inputs [lsearch -inline -all -not -exact [all_inputs] $clk_port]

set_input_delay  [expr $clk_period * $clk_io_pct] -clock $clk_name $non_clock_inputs
set_output_delay [expr $clk_period * $clk_io_pct] -clock $clk_name [all_outputs]

fastroute.tcl

set_global_routing_layer_adjustment metal2-metal3 0.5
set_global_routing_layer_adjustment metal4-$::env(MAX_ROUTING_LAYER) 0.25

set_routing_layers -signal $::env(MIN_ROUTING_LAYER)-$::env(MAX_ROUTING_LAYER)

set_global_routing_layer_adjustment $::env(MIN_ROUTING_LAYER)-$::env(MAX_ROUTING_LAYER) 0.15900339639207847

And finally, the config.mk (note that I changed the fastroute reference to the updated fastroute.tcl)

export DESIGN_NAME = picorv32
export PLATFORM    = nangate45

export VERILOG_FILES = ./designs/src/$(DESIGN_NAME)/picorv32.v
export SDC_FILE      = ./designs/$(PLATFORM)/$(DESIGN_NAME)/constraint.sdc
export FASTROUTE_TCL = ./fastroute.tcl
export ABC_AREA      = 1

# Adders degrade GCD
export ADDER_MAP_FILE :=

export CORE_UTILIZATION ?= 55
export PLACE_DENSITY_LB_ADDON = 0.20
export TNS_END_PERCENT = 100

export CAP_MARGIN = 39
export CELL_PAD_IN_SITES_DETAIL_PLACEMENT = 0
export CELL_PAD_IN_SITES_GLOBAL_PLACEMENT = 1
export CORE_ASPECT_RATIO = 1.1435000182737207
export CORE_MARGIN = 2
export CORE_UTILIZATION = 39
export CTS_CLUSTER_DIAMETER = 336
export CTS_CLUSTER_SIZE = 51
export PLACE_DENSITY_LB_ADDON = 0.436940328162555
export _PINS_DISTANCE = 1
vijayank88 commented 1 year ago

constraint.sdc looks ok.

fastroute.tcl should be updated as:

set_global_routing_layer_adjustment $::env(MIN_ROUTING_LAYER)-$::env(MAX_ROUTING_LAYER) 0.15900339639207847
set_routing_layers -signal $::env(MIN_ROUTING_LAYER)-$::env(MAX_ROUTING_LAYER)

config.mk as follows


export DESIGN_NAME = picorv32
export PLATFORM    = nangate45

export VERILOG_FILES = ./designs/src/$(DESIGN_NAME)/picorv32.v
export SDC_FILE      = ./designs/$(PLATFORM)/$(DESIGN_NAME)/constraint.sdc
export FASTROUTE_TCL = ./fastroute.tcl
export ABC_AREA      = 1

export TNS_END_PERCENT = 100

export CAP_MARGIN = 39
export CELL_PAD_IN_SITES_DETAIL_PLACEMENT = 0
export CELL_PAD_IN_SITES_GLOBAL_PLACEMENT = 1
export CORE_ASPECT_RATIO = 1.1435000182737207
export CORE_MARGIN = 2
export CORE_UTILIZATION = 39
export CTS_CLUSTER_DIAMETER = 336
export CTS_CLUSTER_SIZE = 51
export PLACE_DENSITY_LB_ADDON = 0.436940328162555
export PLACE_PINS_ARGS="-min_distance 1"
dalyles commented 1 year ago

Hi @vijayank88 ,

Thanks so much for your input, as for the fastroute.tcl, I directly copied and pasted from the autotuner's generated fastroute.tcl file so I didn't touch the first two lines of code, just to add context.

I'll update the last line in config.mk and give a run.

Thank you!

dalyles commented 1 year ago

For some reason, the last line you gave me in config.mk export PLACE_PINS_ARGS="-min_distance 1" gives an error when I run:

Elapsed time: 0:01.75[h:]min:sec. CPU time: user 1.70 sys 0.05 (100%). Peak memory: 138192KB.
(/usr/bin/time -f 'Elapsed time: %E[h:]min:sec. CPU time: user %U sys %S (%P). Peak memory: %MKB.' /home/dlyles/OpenROAD-flow-scripts/tools/install/OpenROAD/bin/openroad -exit -no_init  ./scripts/io_placement_random.tcl -metrics ./logs/nangate45/picorv32/base/2_2_floorplan_io.json) 2>&1 | tee ./logs/nangate45/picorv32/base/2_2_floorplan_io.log
OpenROAD v2.0-9070-gbb341ffb9 
This program is licensed under the BSD-3 license. See the LICENSE file for details.
Components of this program may be licensed under more restrictive licenses which must be honored.
[ERROR STA-0402] place_pins -min_distance 1 is not a known keyword or flag.
Error: io_placement_random.tcl, 14 STA-0402
Command exited with non-zero status 1
Elapsed time: 0:00.28[h:]min:sec. CPU time: user 0.25 sys 0.03 (100%). Peak memory: 102284KB.
make: *** [Makefile:473: results/nangate45/picorv32/base/2_2_floorplan_io.odb] Error 1
vijayank88 commented 1 year ago

Just replace with export PLACE_PINS_ARGS=-min_distance 1

dalyles commented 1 year ago

It works!

vijayank88 commented 1 year ago

@dalyles is anything else pending here to resolve?

dalyles commented 1 year ago

This issue has been resolved.

Thanks!