megaboom: branch predictors as macros

Signed-off-by: Øyvind Harboe <[email protected]>

BUILD.bazel

@@ -377,6 +377,16 @@ SWEEP = {
         },
         "stage_sources": {"floorplan": ["write_macro_placement"]},
     },
+    "1": {
+        "description": "Hierarchical, branch predictors as macros",
+        "variables": {
+            "HOLD_SLACK_MARGIN": "-900",
+            "PDN_TCL": "$(location :pdn.tcl)",
+            "MAX_ROUTING_LAYER": "M9",
+        },
+        "macros": ["ComposedBranchPredictorBank_generate_abstract"],
+        "stage_sources": {"floorplan": ["pdn.tcl"]},
+    },
 }
 
 BOOMTILE_VARIABLES = SKIP_REPORT_METRICS | FAST_BUILD_SETTINGS | {
@@ -428,14 +438,44 @@ boomtile_all_rams = boom_regfile_rams.keys() + all_srams.keys()
 
 ALL_VARIANTS = SWEEP | OTHER_VARIANTS
 
+BRANCH_PREDICTOR_VARIABLES = SKIP_REPORT_METRICS | FAST_BUILD_SETTINGS | {
+    "PDN_TCL": "$(PLATFORM_DIR)/openRoad/pdn/BLOCKS_grid_strategy.tcl",
+    "IO_CONSTRAINTS": "$(location :io-boomtile)",
+    "PLACE_DENSITY": "0.24",
+    # We only need hierarchical synthesis when we are running through floorplan
+    # write_macro_placement macros.tcl
+    "SYNTH_HIERARCHICAL": "1",
+    "SDC_FILE": "$(location :constraints-boomtile)",
+    "MACRO_PLACE_HALO": "4 4",
+    "PLACE_PINS_ARGS": "-annealing",
+    "DIE_AREA": "0 0 300 400",
+    "CORE_AREA": "0 0 298 398",
+    # Saves hours of build time, specific to BoomTile
+    "SKIP_LAST_GASP": "1",
+    "SETUP_SLACK_MARGIN": "-1300",
+    "HOLD_SLACK_MARGIN": "-200",
+}
+
+orfs_flow(
+    name = "ComposedBranchPredictorBank",
+    abstract_stage = "cts",
+    arguments = BRANCH_PREDICTOR_VARIABLES,
+    macros = [
+        ":" + m + "_generate_abstract"
+        for m in boomtile_all_rams
+    ],
+    stage_sources = BOOMTILE_SOURCES,
+    verilog_files = all_source_files,
+)
+
 [orfs_flow(
     name = "BoomTile",
     arguments = BOOMTILE_VARIABLES | ALL_VARIANTS[variant].get("variables", {}),
     macros = [
         ":" + m + "_generate_abstract"
         for m in boomtile_all_rams
         if m not in ALL_VARIANTS[variant].get("dissolve", [])
-    ],
+    ] + ALL_VARIANTS[variant].get("macros", []),
     previous_stage = ALL_VARIANTS[variant].get("previous_stage", {}),
     renamed_inputs = ALL_VARIANTS[variant].get("renamed_inputs", {}),
     # Explictily set arguments for a stage when stages are not listed in

constraints-boomtile.sdc

@@ -11,4 +11,20 @@ set clk_port_name clock
 # Ignore synchronous reset for now.
 set_false_path -from [get_ports reset]
 
-source $env(PLATFORM_DIR)/constraints.sdc
+set clk_port [get_ports $clk_port_name]
+create_clock -period $clk_period -waveform [list 0 [expr $clk_period / 2]] -name $clk_name $clk_port
+
+set non_clk_inputs  [lsearch -inline -all -not -exact [all_inputs] $clk_port]
+
+if {[llength [all_registers]] > 0} {
+    set all_register_outputs [get_pins -of_objects [all_registers] -filter {direction == output}]
+    set_max_delay [expr {[info exists in2reg_max] ? $in2reg_max : 80}] -from $non_clk_inputs -to [all_registers]
+    set_max_delay [expr {[info exists reg2out_max] ? $reg2out_max : 80}] -from $all_register_outputs -to [all_outputs]
+
+    group_path -name in2reg -from $non_clk_inputs -to [all_registers]
+    group_path -name reg2out -from [all_registers] -to [all_outputs]
+    group_path -name reg2reg -from [all_registers] -to [all_registers]
+}
+
+set_max_delay [expr {[info exists in2out_max] ? $in2out_max : 80}] -from $non_clk_inputs -to [all_outputs]
+group_path -name in2out -from $non_clk_inputs -to [all_outputs]

pdn.tcl

@@ -0,0 +1,35 @@
+# based on ORFS/platforms/asap7/openRoad/pdn/BLOCK_grid_strategy.tcl,
+# but added power higher up
+
+####################################
+# global connections
+####################################
+add_global_connection -net {VDD} -inst_pattern {.*} -pin_pattern {^VDD$} -power
+add_global_connection -net {VSS} -inst_pattern {.*} -pin_pattern {^VSS$} -ground
+
+####################################
+# voltage domains
+####################################
+set_voltage_domain -name {CORE} -power {VDD} -ground {VSS}
+
+####################################
+# standard cell grid
+# M1 M2 are for follow pin, width derived from PG rail in standard cell
+# M5 stripe width rerived from one of width allowed in LEF, offset and pitch
+#  put stripe on M5 track
+# M4 M5 ring follow stripe width
+####################################
+define_pdn_grid -name {top} -voltage_domains {CORE}
+
+add_pdn_ring  -grid {top} -layers {M5 M4} -widths {0.12 0.12} -spacings {0.072} -core_offset {0.084}
+
+add_pdn_stripe -grid {top} -layer {M1} -width {0.018} -pitch {0.54} -offset {0} -followpins
+add_pdn_stripe -grid {top} -layer {M2} -width {0.018} -pitch {0.54} -offset {0} -followpins
+
+add_pdn_stripe -grid {top} -layer {M5} -width {0.12} -spacing {0.072} -pitch {2.976} -offset {1.5} -extend_to_core_ring
+
+add_pdn_connect -grid {top} -layers {M1 M2}
+add_pdn_connect -grid {top} -layers {M2 M5}
+add_pdn_connect -grid {top} -layers {M4 M5}
+# allow routing at all metal levels
+add_pdn_connect -grid {top} -layers {M5 M9}