Example sram_32x32_1rw configuration fails to generate

[tajayi]

Hi, The example freepdk45.cfg fails to generate for me. Specifically sram_32x32_1rw fails with the following error

top 3 best memory configurations are:
Memory    cap: 80 GB num_bobs: 1 bw: 533 (MHz) cost: $731.2 energy: 32.6101 (nJ) 
 {
 (0)  BoB       cap: 80 GB num_channels: 1 bw: 533 (MHz) cost: $731.2 energy: 32.6101 (nJ) 
    ==============
    (0) cap: 80 GB bw: 533 (MHz) cost: $731.2 dpc: 3 energy: 32.6101 (nJ)  DIMM:  RDIMM  low power: F [ 0(4GB) 0(8GB) 1(16GB) 2(32GB) 0(64GB) ]
    ==============

 }

=============================================

Error: y = 5.38 < minimum (14.82)
Aborting.
make: *** [run] Error 1

Also failing with similar issues are free45_1rw_d64_w7, free45_1rw_d64_w96 and free45_1rw_d512_w64 provided as examples for the bp design.

Note that sram_8x512_1rw succeeds.

Thanks

[stdavids]

Hey Tutu,

Thanks for the issue. I'm not able to replicate this on my end, can you attach your cfg file as well as cacti.cfg and cacti.cfg.out file that should reside in the failing SRAM's results/ directory. The error is after cacti so those files should exist.

[tajayi]

Thanks for the quick response. Find all the results folders attached results.tar.gz

[stdavids]

Thanks Tutu,

So it looks like the issue is with the "aspect ratio" output of cacti. Your SRAMs have huge aspect ratios like 22x while mine are ~0.3. 22x wider than it is tall is a bit outside the normal scope so I'm not surprised the tools died :)

What OS are you running on? And what version of gcc is installed? I'd like to try and bootstrap a vm to see if I can replicate this output. I'll also give cacti a look and see if anything pops out at me but I'll likely need your help to verify if any fixes work. In the meantime, I'll send you the generated files for you to work with.

[tajayi]

Thanks. Here's the info you requested

Linux  2.6.32-642.6.1.el6.x86_64 #1 SMP Thu Aug 25 12:42:19 EDT 2016 x86_64 x86_64 x86_64 GNU/Linux
gcc version 5.4.0 (GCC) 
Red Hat Enterprise Linux Server release 6.8 (Santiago)

[tajayi]

This issue appears to be related to the gcc compiler (and possibly configuration of our servers). On 5.4.0, i get a really weird aspect ratio in the cacti output. I though switching to 4.4.7 fixed the issue, but it only resulted in the aspect ration from cacti being "NaN", which lead to a lot of coordinates in the lef file being set to NaN. Another version of gcc had "Inf" as the aspect ratio.

My gut is that some variable or operator needs to be set explicitly in cacti. I was able to generate what appears to be usable using gcc 4.8.5

[taylor-bsg]

For posterity what is our version of GCC? If we release this to people we need to give some sane range of GCC versions.

See cacti issues, e.g.

https://github.com/HewlettPackard/cacti/issues/8

Maybe we can do a pull request to cacti with some fixes.

M

On Tue, Oct 22, 2019 at 2:12 PM Tutu Ajayi notifications@github.com wrote:

This issue appears to be related to the gcc compiler (and possibly configuration of our servers). On 5.4.0, i get a really weird aspect ratio in the cacti output. I though switching to 4.4.7 fixed the issue, but it only resulted in the aspect ration from cacti being "NaN", which lead to a lot of coordinates in the lef file being set to NaN. Another version of gcc had "Inf" as the aspect ratio.

My gut is that some variable or operator needs to be set explicitly in cacti. I was able to generate what appears to be usable using gcc 4.8.5

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/bespoke-silicon-group/bsg_fakeram/issues/2?email_source=notifications&email_token=AEFG5ACTOGFSADT6727XFG3QP47ALA5CNFSM4I2OKV6KYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEB6V4GI#issuecomment-545087001, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEFG5AG5ITU4N43ENVLIWBDQP47ALANCNFSM4I2OKV6A .

[taylor-bsg]

Also https://github.com/HewlettPackard/cacti/issues/6

On Tue, Oct 22, 2019 at 2:44 PM Michael Nguyen Taylor < michael.b.taylor@gmail.com> wrote:

For posterity what is our version of GCC? If we release this to people we need to give some sane range of GCC versions.

See cacti issues, e.g.

https://github.com/HewlettPackard/cacti/issues/8

Maybe we can do a pull request to cacti with some fixes.

M

On Tue, Oct 22, 2019 at 2:12 PM Tutu Ajayi notifications@github.com wrote:

This issue appears to be related to the gcc compiler (and possibly configuration of our servers). On 5.4.0, i get a really weird aspect ratio in the cacti output. I though switching to 4.4.7 fixed the issue, but it only resulted in the aspect ration from cacti being "NaN", which lead to a lot of coordinates in the lef file being set to NaN. Another version of gcc had "Inf" as the aspect ratio.

My gut is that some variable or operator needs to be set explicitly in cacti. I was able to generate what appears to be usable using gcc 4.8.5

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/bespoke-silicon-group/bsg_fakeram/issues/2?email_source=notifications&email_token=AEFG5ACTOGFSADT6727XFG3QP47ALA5CNFSM4I2OKV6KYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEB6V4GI#issuecomment-545087001, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEFG5AG5ITU4N43ENVLIWBDQP47ALANCNFSM4I2OKV6A .

[stdavids]

We use GCC 4.8.5

[taylor-bsg]

Okay I see. I suggest we run valgrind on cacti to find the issue.

M

On Tue, Oct 22, 2019 at 3:18 PM Scott Davidson notifications@github.com wrote:

We use GCC 4.8.5

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/bespoke-silicon-group/bsg_fakeram/issues/2?email_source=notifications&email_token=AEFG5AGF4YTA52HO5JY2KLDQP5GZVA5CNFSM4I2OKV6KYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEB64JFQ#issuecomment-545113238, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEFG5AAGYRR4XQW7Y5WS3RLQP5GZVANCNFSM4I2OKV6A .