weekly_report_notes

#14.09. - 20.09.

• finally got bank account and submitted it + summer job hours to hr -> can finally get employee account + get paid

• firesim

  • got firesim repo to build and software to run on lobo
    • qemu can't be built
      • requires newer gthread and glib versions than available
      • only necessary for FireMarshal -> test linux images elsewhere -> needs to be commented out in firesim/target-design/chipyard/scripts/build-toolchains.sh
    • make dromajo compile
      • Makfile should be generated by cmake -> no use for pull request -> add -D__STDC_FORMAT_MACROS to CXXFLAGS in firesim/target-design/chipyard/tools/dromajo/dromajo-src/src/Makefile -> lobo now usable
  • adjusted firesim makefiles to provide targets fpga and convey-app for PLATFORM=convey
    • fpga generates files for convey toolchain and builds fpga image
    • convey-app compiles host application for firesim (new host application for every image because bridge addresses can change)
  • firesim hangs in initialisation phase
    • communication with convey to axi shell works from firesim binary -> builds correctly
    • communication with bridge registers over AXI fails -> get verilator simulation of wolverine to work for firesim
      • WolverineShim as top verilog
      • simif_convey_emul
      • found out that I was using word addresses instead of byte addresses... -> simulation is now at least starting
      • implement + check memory next -> turns out that the maximum transaction size what set too large -> dhrystone works in simulation
    • firesim works for ibda.elf
      • hangs every second time - some reset not working?
        • quickfix: don't let dispatch complete -> fpga bitstream is reloaded for every execution -> slow startup -> reset csr in convey harness?
    • doesn't work for dhrystone
      • run individual instruction tests
        • something broken for amo instructions -> difference between real and simulation memory interface?
      • create simple tracer to find out which insn fails
        • TracerS

###Objectives for last week

• get firesim to work ###Accomplishments last week
• finally got bank account and submitted it + summer job hours to hr -> can finally get employee account + get paid
• got firesim to build on lobo
• integrated convey bitstream build process into firesim
• implemented verilator simulation of convey interface for firesim ###Problems and solutions
• fixed some memory interface issues with the simulator
• firesim currently fails for programs with amo instructions
  • works in simulator
  • simple tracer to find out exact failing point (TracerV without DMA) ###Next week's objectives
• get amo instructions to work on firesim
  • finish implementing tracer
  • improve simulation model once source is found
• improve firesim makefile to make usage easier
• run simulations on firesim ###Long-term plan
• get thesis work running on new firesim
• improve critical path
• investigate OpenRoad ASIC toolflow
• decide what to do next

#21.09. - 27.09.

• firesim
  • got TracerS to work
  • failing applications either stop right after bootloader or at 0x8000003E - last before 0x80000040 -> elf loading does not work
    • disabled loadmem interface because it does not work propperly
      • replacing with wolverine
        • memcopy does not work while accelerator is running -> doesn't work
      • use TSI instead
        • TSI is too slow
          • adjust +fesvr-step-size and let serial act without blocking
          • rework serial bridge to have bigger buffers + better interface
  • applications work
  • linux hangs during boot
    • update FireMarshall + generate new image -> linux boots and works

###Objectives for last week

• get amo instructions to work on firesim
  • finish implementing tracer
  • improve simulation model once source is found
• improve firesim makefile to make usage easier
• run simulations on firesim ###Accomplishments last week
• got firesim to work
  • working TracerS (currently 1-wide only)
  • working linux
  • decent speeds
  • experimented a bit with FASED memory model ###Problems and solutions
• failing applications either stop right after bootloader or at 0x8000003E - last before 0x80000040 -> elf loading does not work
  • disabled loadmem interface because it does not work propperly
    • replacing with wolverine
      • memcopy does not work while accelerator is running -> doesn't work
    • use TSI instead
      • TSI is too slow
        • adjust +fesvr-step-size and let serial act without blocking
        • rework serial bridge to have bigger buffers + better interface ###Next week's objectives
• improve firesim makefile to make usage easier
• run simulations on firesim
• find out how to decouple FASED memory freq. from system freq.
  • old method deprecated and not replaced yet? ###Long-term plan
• get thesis work running on new firesim
• improve critical path
• investigate OpenRoad ASIC toolflow
• decide what to do next