Different machine params supported?

[dpetrisko]

On bigblade branch, trying to make a smaller manycore (for verilator compilation on small machine, etc.)

This is my patch:

diff --git a/machines/pod_X1Y1_ruche_X16Y8_hbm/Makefile.machine.include b/machines/pod_X1Y1_ruche_X16Y8_hbm/Makefile.machine.include
index 6cb711f..b8ff129 100644
--- a/machines/pod_X1Y1_ruche_X16Y8_hbm/Makefile.machine.include
+++ b/machines/pod_X1Y1_ruche_X16Y8_hbm/Makefile.machine.include
@@ -56,7 +56,7 @@ BSG_MACHINE_MEM_DRAMSIM3_PKG          = bsg_dramsim3_hbm2_8gb_x128_pkg
 BSG_MACHINE_MEM_CFG                   = e_vcache_hbm2

 # Victim Cache Parameters
-BSG_MACHINE_VCACHE_PER_DRAM_CHANNEL   = 16
+BSG_MACHINE_VCACHE_PER_DRAM_CHANNEL   = 4
 BSG_MACHINE_VCACHE_SET                = 64
 BSG_MACHINE_VCACHE_WAY                = 8
 BSG_MACHINE_VCACHE_LINE_WORDS         = 32
@@ -72,8 +72,8 @@ BSG_MACHINE_NOC_COORD_X_WIDTH         = 7
 BSG_MACHINE_NOC_COORD_Y_WIDTH         = 7

 # Pod Tile Dimensions (Rarely Changed)
-BSG_MACHINE_POD_TILES_X               = 16
-BSG_MACHINE_POD_TILES_Y               = 8
+BSG_MACHINE_POD_TILES_X               = 4
+BSG_MACHINE_POD_TILES_Y               = 4

 # Pod Tile subarray dimensions (Rarely changed, for physical design only)
 BSG_MACHINE_POD_TILES_SUBARRAY_X      = 1

Getting hundreds of these:

[BSG_ERROR] Errant fwd packet detected: src_x=x, src_y=x, dest_x=x, dest_y=x.
Error: "/mnt/users/ssd0/no_backup/petrisko/bsg_bladerunner/bsg_manycore/v/bsg_manycore_link_sif_tieoff.v", 42: replicant_tb_top.testbench.io_e_tieoff: at time 23815000 ps
[BSG_ERROR] Errant fwd packet detected: src_x=x, src_y=x, dest_x=x, dest_y=x.
Error: "/mnt/users/ssd0/no_backup/petrisko/bsg_bladerunner/bsg_manycore/v/bsg_manycore_link_sif_tieoff.v", 42: replicant_tb_top.testbench.io_e_tieoff: at time 23816000 ps

Any other parameters I should be aware of? I can look into if there's a hardware bug, but hopefully it's "just software" ;)

[drichmond]

Dan, should we update this issue for 4x4?

[dpetrisko]

Latest update is that 4x4 manycore fails a subset of regression tests. It seems to be software, as 4x2 is able to pass manycore spmd tests