<tpb>
Title: Eblif write support by acomodi · Pull Request #518 · verilog-to-routing/vtr-verilog-to-routing · GitHub (at github.com)
<sf-slack2>
<acomodi> mithro: it correctly writes both .params and .attrs in the output blif
anuejn_ has joined #symbiflow
plaes_ has joined #symbiflow
noopwafe1 has joined #symbiflow
tmichalak1 has joined #symbiflow
Bertl_ has joined #symbiflow
Vonter_ has joined #symbiflow
Vonter has quit [*.net *.split]
tmichalak has quit [*.net *.split]
Bertl has quit [*.net *.split]
anuejn has quit [*.net *.split]
plaes has quit [*.net *.split]
noopwafel has quit [*.net *.split]
jevinskie has quit [Ping timeout: 246 seconds]
Bertl_ is now known as Bertl
jevinskie has joined #symbiflow
alexhw has joined #symbiflow
futarisIRCcloud has quit [Quit: Connection closed for inactivity]
<mithro>
kgugala: Your idea of using yosys on the .sim.v files is a pretty good idea
<sf-slack2>
<kgugala> @mithro I did some initial tests on that and implemented a simple python tool for updating blif generated from those sim.v files so they correspond with what the test architecture expects
<tpb>
Title: Use Yosys on .sim.v files to generate a .eblif needed for architecture testing · Issue #746 · SymbiFlow/symbiflow-arch-defs · GitHub (at github.com)
<sf-slack2>
<kgugala> @mithro we have a WIP implementation of points 1 to 3 of the above issue
<tpb>
Title: Add v2x to use specify section rather than the extra timing attributes · Issue #748 · SymbiFlow/symbiflow-arch-defs · GitHub (at github.com)
<tpb>
Title: Flip Flops - Google Sheets (at docs.google.com)
citypw has joined #symbiflow
jevinskie has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
Bertl is now known as bertl_oO
bertl_oO is now known as Bertl_oO
<hackerfoo>
Apparently Vivado infers the NO_CHANGE write mode for block RAM even if you specify READ_FIRST.
<litghost>
hackerfoo: There are cases where NO_CHANGE is equiv, but presumably that is an optimization?
<litghost>
hackerfoo: Read the memory guide, I believe there is a paragraph describing this behavior?
<hackerfoo>
Yeah, I know, I just didn't expect it to do that for me.
<hackerfoo>
Probably only when you have a single write port, which is a pretty simple test.
<litghost>
hackerfoo: If I had to guess, maybe Vivado proved that write and reading over the same address was impossible, so NO_CHANGE is equiv to READ_FIRST?
<hackerfoo>
Okay - It's also ignoring WRITE_FIRST? That doesn't seem right.
<litghost>
hackerfoo: If there is never a read/write address conflict, the write mode doesn't matter
<litghost>
hackerfoo: And NO_CHANGE is the least power/fastest mode
<litghost>
hackerfoo: WRITE_FIRST/READ_FIRST/NO_CHANGE have an impact if a read and write of the same address occurs in one cycle
jevinskie has joined #symbiflow
<hackerfoo>
litghost: Ah, DO of the write port isn't connected so it really doesn't matter.
<litghost>
hackerfoo: Right, so the mode is unobservable, in which case the cheapest/fastest mode should be used
<litghost>
hackerfoo: FYI this simple optimization (DO/DOP disconnected -> NO_CHANGE) is fairly easy to write in the techmap
<litghost>
hackerfoo: We don't have to do it at the first cut, but it isn't hard to do
<hackerfoo>
I think I'm just going to keep it simple for now, and just do widths of 8/9, 16/18, and 32/36 for now, and no optimizations.
<litghost>
hackerfoo: I agree
<litghost>
hackerfoo: It seems reasonable to write a short issue with the known limitations of the current approach and what work could be done to make it complete