RF using async memory (#759)

kuznia-rdzeni · Nov 26, 2024 · 240734d · 240734d
1 parent 40c82ee
commit 240734d
Show file tree

Hide file tree

Showing 2 changed files with 22 additions and 16 deletions.
diff --git a/coreblocks/core_structs/rf.py b/coreblocks/core_structs/rf.py
@@ -1,10 +1,10 @@
 from amaranth import *
+import amaranth.lib.memory as memory
 from transactron import Method, Transaction, def_method, TModule
 from coreblocks.interface.layouts import RFLayouts
 from coreblocks.params import GenParams
 from transactron.lib.metrics import HwExpHistogram, TaggedLatencyMeasurer
 from transactron.utils.amaranth_ext.functions import popcount
-from transactron.utils.transactron_helpers import make_layout
 
 __all__ = ["RegisterFile"]
 
@@ -13,12 +13,9 @@ class RegisterFile(Elaboratable):
     def __init__(self, *, gen_params: GenParams):
         self.gen_params = gen_params
         layouts = gen_params.get(RFLayouts)
-        self.internal_layout = make_layout(("reg_val", gen_params.isa.xlen), ("valid", 1))
         self.read_layout = layouts.rf_read_out
-        self.entries = Array(
-            Signal(self.internal_layout, init={"reg_val": 0, "valid": k == 0})
-            for k in range(2**gen_params.phys_regs_bits)
-        )
+        self.entries = memory.Memory(shape=gen_params.isa.xlen, depth=2**gen_params.phys_regs_bits, init=[])
+        self.valids = Array(Signal(init=k == 0) for k in range(2**gen_params.phys_regs_bits))
 
         self.read1 = Method(i=layouts.rf_read_in, o=layouts.rf_read_out)
         self.read2 = Method(i=layouts.rf_read_in, o=layouts.rf_read_out)
@@ -41,49 +38,57 @@ def __init__(self, *, gen_params: GenParams):
     def elaborate(self, platform):
         m = TModule()
 
-        m.submodules += [self.perf_rf_valid_time, self.perf_num_valid]
+        m.submodules += [self.entries, self.perf_rf_valid_time, self.perf_num_valid]
 
         being_written = Signal(self.gen_params.phys_regs_bits)
         written_value = Signal(self.gen_params.isa.xlen)
 
+        write_port = self.entries.write_port()
+        read_port_1 = self.entries.read_port(domain="comb")
+        read_port_2 = self.entries.read_port(domain="comb")
+
         @def_method(m, self.read1)
         def _(reg_id: Value):
             forward = Signal()
             m.d.av_comb += forward.eq((being_written == reg_id) & (reg_id != 0))
+            m.d.av_comb += read_port_1.addr.eq(reg_id)
             return {
-                "reg_val": Mux(forward, written_value, self.entries[reg_id].reg_val),
-                "valid": Mux(forward, 1, self.entries[reg_id].valid),
+                "reg_val": Mux(forward, written_value, read_port_1.data),
+                "valid": Mux(forward, 1, self.valids[reg_id]),
             }
 
         @def_method(m, self.read2)
         def _(reg_id: Value):
             forward = Signal()
             m.d.av_comb += forward.eq((being_written == reg_id) & (reg_id != 0))
+            m.d.av_comb += read_port_2.addr.eq(reg_id)
             return {
-                "reg_val": Mux(forward, written_value, self.entries[reg_id].reg_val),
-                "valid": Mux(forward, 1, self.entries[reg_id].valid),
+                "reg_val": Mux(forward, written_value, read_port_2.data),
+                "valid": Mux(forward, 1, self.valids[reg_id]),
             }
 
         @def_method(m, self.write)
         def _(reg_id: Value, reg_val: Value):
             zero_reg = reg_id == 0
             m.d.comb += being_written.eq(reg_id)
             m.d.av_comb += written_value.eq(reg_val)
+            m.d.av_comb += write_port.addr.eq(reg_id)
+            m.d.av_comb += write_port.data.eq(reg_val)
             with m.If(~(zero_reg)):
-                m.d.sync += self.entries[reg_id].reg_val.eq(reg_val)
-                m.d.sync += self.entries[reg_id].valid.eq(1)
+                m.d.comb += write_port.en.eq(1)
+                m.d.sync += self.valids[reg_id].eq(1)
                 self.perf_rf_valid_time.start(m, slot=reg_id)
 
         @def_method(m, self.free)
         def _(reg_id: Value):
             with m.If(reg_id != 0):
-                m.d.sync += self.entries[reg_id].valid.eq(0)
+                m.d.sync += self.valids[reg_id].eq(0)
                 self.perf_rf_valid_time.stop(m, slot=reg_id)
 
         if self.perf_num_valid.metrics_enabled():
             num_valid = Signal(self.gen_params.phys_regs_bits + 1)
             m.d.comb += num_valid.eq(
-                popcount(Cat(self.entries[reg_id].valid for reg_id in range(2**self.gen_params.phys_regs_bits)))
+                popcount(Cat(self.valids[reg_id] for reg_id in range(2**self.gen_params.phys_regs_bits)))
             )
             with Transaction(name="perf").body(m):
                 self.perf_num_valid.add(m, num_valid)

diff --git a/test/test_core.py b/test/test_core.py
@@ -70,7 +70,8 @@ def get_phys_reg_rrat(self, sim: TestbenchContext, reg_id):
         return sim.get(self.m.core.RRAT.entries[reg_id])
 
     def get_arch_reg_val(self, sim: TestbenchContext, reg_id):
-        return sim.get(self.m.core.RF.entries[(self.get_phys_reg_rrat(sim, reg_id))].reg_val)
+        # TODO: better stubs for memory, remove ignore
+        return sim.get(self.m.core.RF.entries.data[(self.get_phys_reg_rrat(sim, reg_id))])  # type: ignore
 
 
 class TestCoreAsmSourceBase(TestCoreBase):