Allow the formal engine to perform a same-cycle result in the ALU
[soc.git] / src / soc / simple / issuer.py
index cfad831a3fa76a92c71464fbd95dc0554b8328d0..15bd1760a5ab93f233d8cb7cdff813d7b0833096 100644 (file)
@@ -169,6 +169,21 @@ class TestIssuerBase(Elaboratable):
         self.microwatt_compat = (hasattr(pspec, "microwatt_compat") and
                                  (pspec.microwatt_compat == True))
         self.alt_reset = Signal(reset_less=True) # not connected yet (microwatt)
+        # test if fabric compatibility is to be enabled
+        self.fabric_compat = (hasattr(pspec, "fabric_compat") and
+                                 (pspec.fabric_compat == True))
+
+        if self.microwatt_compat or self.fabric_compat:
+
+            if hasattr(pspec, "microwatt_old"):
+                self.microwatt_old = pspec.microwatt_old
+            else:
+                self.microwatt_old = True # PLEASE DO NOT ALTER THIS
+
+            if hasattr(pspec, "microwatt_debug"):
+                self.microwatt_debug = pspec.microwatt_debug
+            else:
+                self.microwatt_debug = True # set to False when using an FPGA
 
         # test is SVP64 is to be enabled
         self.svp64_en = hasattr(pspec, "svp64") and (pspec.svp64 == True)
@@ -194,7 +209,7 @@ class TestIssuerBase(Elaboratable):
         #self.dbg_domain = "sync"  # sigh "dbgsunc" too problematic
         self.dbg_domain = "dbgsync" # domain for DMI/JTAG clock
         if self.jtag_en:
-            # XXX MUST keep this up-to-date with litex, and
+            # XXX MUST keep this up-to-date with fabric, and
             # soc-cocotb-sim, and err.. all needs sorting out, argh
             subset = ['uart',
                       'mtwi',
@@ -307,7 +322,7 @@ class TestIssuerBase(Elaboratable):
         # hack method of keeping an eye on whether branch/trap set the PC
         self.state_nia = self.core.regs.rf['state'].w_ports['nia']
         self.state_nia.wen.name = 'state_nia_wen'
-        # and whether SPR pipeline sets DEC or TB
+        # and whether SPR pipeline sets DEC or TB (fu/spr/main_stage.py)
         self.state_spr = self.core.regs.rf['state'].w_ports['state1']
 
         # pulse to synchronize the simulator at instruction end
@@ -321,14 +336,15 @@ class TestIssuerBase(Elaboratable):
             self.srcmask = Signal(64)
             self.dstmask = Signal(64)
 
-        # sigh, the wishbone addresses are not wishbone-compliant in microwatt
-        if self.microwatt_compat:
+        # sigh, the wishbone addresses are not wishbone-compliant
+        # in old versions of microwatt, tplaten_3d_game is a new one
+        if self.microwatt_compat or self.fabric_compat:
             self.ibus_adr = Signal(32, name='wishbone_insn_out.adr')
             self.dbus_adr = Signal(32, name='wishbone_data_out.adr')
 
         # add an output of the PC and instruction, and whether it was requested
         # this is for verilator debug purposes
-        if self.microwatt_compat:
+        if self.microwatt_compat or self.fabric_compat:
             self.nia = Signal(64)
             self.msr_o = Signal(64)
             self.nia_req = Signal(1)
@@ -351,7 +367,7 @@ class TestIssuerBase(Elaboratable):
         csd = DomainRenamer(self.core_domain)
         dbd = DomainRenamer(self.dbg_domain)
 
-        if self.microwatt_compat:
+        if self.microwatt_compat or self.fabric_compat:
             m.submodules.core = core = self.core
         else:
             m.submodules.core = core = csd(self.core)
@@ -372,7 +388,7 @@ class TestIssuerBase(Elaboratable):
         # fixup the clocks in microwatt-compat mode (but leave resets alone
         # so that microwatt soc.vhdl can pull a reset on the core or DMI
         # can do it, just like in TestIssuer)
-        if self.microwatt_compat:
+        if self.microwatt_compat or self.fabric_compat:
             intclk = ClockSignal(self.core_domain)
             dbgclk = ClockSignal(self.dbg_domain)
             if self.core_domain != 'sync':
@@ -380,21 +396,26 @@ class TestIssuerBase(Elaboratable):
             if self.dbg_domain != 'sync':
                 comb += dbgclk.eq(ClockSignal())
 
+        # if using old version of microwatt
         # drop the first 3 bits of the incoming wishbone addresses
-        # this can go if using later versions of microwatt (not now)
-        if self.microwatt_compat:
+        if self.microwatt_compat or self.fabric_compat:
             ibus = self.imem.ibus
             dbus = self.core.l0.cmpi.wb_bus()
-            comb += self.ibus_adr.eq(Cat(Const(0, 3), ibus.adr))
-            comb += self.dbus_adr.eq(Cat(Const(0, 3), dbus.adr))
-            # microwatt verilator debug purposes
-            pi = self.core.l0.cmpi.pi.pi
-            comb += self.ldst_req.eq(pi.addr_ok_o)
-            comb += self.ldst_addr.eq(pi.addr)
+            if self.microwatt_old:
+                comb += self.ibus_adr.eq(Cat(Const(0, 3), ibus.adr))
+                comb += self.dbus_adr.eq(Cat(Const(0, 3), dbus.adr))
+            else:
+                comb += self.ibus_adr.eq(ibus.adr)
+                comb += self.dbus_adr.eq(dbus.adr)
+            if self.microwatt_debug:
+                # microwatt verilator debug purposes
+                pi = self.core.l0.cmpi.pi.pi
+                comb += self.ldst_req.eq(pi.addr_ok_o)
+                comb += self.ldst_addr.eq(pi.addr)
 
         cur_state = self.cur_state
 
-        # 4x 4k SRAM blocks.  these simply "exist", they get routed in litex
+        # 4x 4k SRAM blocks.  these simply "exist", they get routed in fabric
         if self.sram4x4k:
             for i, sram in enumerate(self.sram4k):
                 m.submodules["sram4k_%d" % i] = csd(sram)
@@ -414,7 +435,7 @@ class TestIssuerBase(Elaboratable):
             m.submodules.simple_gpio = simple_gpio = csd(self.simple_gpio)
 
         # connect one GPIO output to ICS bit 15 (like in microwatt soc.vhdl)
-        # XXX causes litex ECP5 test to get wrong idea about input and output
+        # XXX causes fabric ECP5 test to get wrong idea about input and output
         # (but works with verilator sim *sigh*)
         # if self.gpio and self.xics:
         #   comb += self.int_level_i[15].eq(simple_gpio.gpio_o[0])
@@ -555,6 +576,7 @@ class TestIssuerBase(Elaboratable):
         state_r_dectb = state_rf.r_ports['issue']  # DEC/TB
         state_w_dectb = state_rf.w_ports['issue']  # DEC/TB
 
+
         with m.FSM() as fsm:
 
             # initiates read of current DEC
@@ -615,6 +637,10 @@ class TestIssuerBase(Elaboratable):
         # reset current state if core reset requested
         with m.If(core_rst):
             m.d.sync += self.cur_state.eq(0)
+            # and, sigh, set configured values, which are also done in regfile
+            # XXX ??? what the hell is the shift for??
+            m.d.sync += self.cur_state.pc.eq(self.core.pc_at_reset)
+            m.d.sync += self.cur_state.msr.eq(self.core.msr_at_reset)
 
         # check halted condition: requested PC to execute matches DMI stop addr
         # and immediately stop. address of 0xffff_ffff_ffff_ffff can never
@@ -679,7 +705,7 @@ class TestIssuerBase(Elaboratable):
                 sync += self.sv_changed.eq(1)
 
         # start renaming some of the ports to match microwatt
-        if self.microwatt_compat:
+        if self.microwatt_compat or self.fabric_compat:
             self.core.o.core_terminate_o.name = "terminated_out"
             # names of DMI interface
             self.dbg.dmi.addr_i.name = 'dmi_addr'
@@ -690,26 +716,28 @@ class TestIssuerBase(Elaboratable):
             self.dbg.dmi.ack_o.name  = 'dmi_ack'
             # wishbone instruction bus
             ibus = self.imem.ibus
-            ibus.adr.name = 'wishbone_insn_out.adr'
-            ibus.dat_w.name = 'wishbone_insn_out.dat'
-            ibus.sel.name = 'wishbone_insn_out.sel'
-            ibus.cyc.name = 'wishbone_insn_out.cyc'
-            ibus.stb.name = 'wishbone_insn_out.stb'
-            ibus.we.name = 'wishbone_insn_out.we'
-            ibus.dat_r.name = 'wishbone_insn_in.dat'
-            ibus.ack.name = 'wishbone_insn_in.ack'
-            ibus.stall.name = 'wishbone_insn_in.stall'
+            if self.microwatt_compat:
+                ibus.adr.name = 'wishbone_insn_out.adr'
+                ibus.dat_w.name = 'wishbone_insn_out.dat'
+                ibus.sel.name = 'wishbone_insn_out.sel'
+                ibus.cyc.name = 'wishbone_insn_out.cyc'
+                ibus.stb.name = 'wishbone_insn_out.stb'
+                ibus.we.name = 'wishbone_insn_out.we'
+                ibus.dat_r.name = 'wishbone_insn_in.dat'
+                ibus.ack.name = 'wishbone_insn_in.ack'
+                ibus.stall.name = 'wishbone_insn_in.stall'
             # wishbone data bus
             dbus = self.core.l0.cmpi.wb_bus()
-            dbus.adr.name = 'wishbone_data_out.adr'
-            dbus.dat_w.name = 'wishbone_data_out.dat'
-            dbus.sel.name = 'wishbone_data_out.sel'
-            dbus.cyc.name = 'wishbone_data_out.cyc'
-            dbus.stb.name = 'wishbone_data_out.stb'
-            dbus.we.name = 'wishbone_data_out.we'
-            dbus.dat_r.name = 'wishbone_data_in.dat'
-            dbus.ack.name = 'wishbone_data_in.ack'
-            dbus.stall.name = 'wishbone_data_in.stall'
+            if self.microwatt_compat:
+                dbus.adr.name = 'wishbone_data_out.adr'
+                dbus.dat_w.name = 'wishbone_data_out.dat'
+                dbus.sel.name = 'wishbone_data_out.sel'
+                dbus.cyc.name = 'wishbone_data_out.cyc'
+                dbus.stb.name = 'wishbone_data_out.stb'
+                dbus.we.name = 'wishbone_data_out.we'
+                dbus.dat_r.name = 'wishbone_data_in.dat'
+                dbus.ack.name = 'wishbone_data_in.ack'
+                dbus.stall.name = 'wishbone_data_in.stall'
 
         return m
 
@@ -727,15 +755,24 @@ class TestIssuerBase(Elaboratable):
         return list(self)
 
     def external_ports(self):
-        if self.microwatt_compat:
-            ports = [self.core.o.core_terminate_o,
-                     self.ext_irq,
-                     self.alt_reset, # not connected yet
-                     self.nia, self.insn, self.nia_req, self.msr_o,
-                     self.ldst_req, self.ldst_addr,
-                     ClockSignal(),
-                     ResetSignal(),
-                    ]
+        if self.microwatt_compat or self.fabric_compat:
+            if self.fabric_compat:
+                ports = [self.core.o.core_terminate_o,
+                         self.alt_reset, # not connected yet
+                         self.nia, self.insn, self.nia_req, self.msr_o,
+                         self.ldst_req, self.ldst_addr,
+                         ClockSignal(),
+                         ResetSignal(),
+                        ]
+            else:
+                ports = [self.core.o.core_terminate_o,
+                         self.ext_irq,
+                         self.alt_reset, # not connected yet
+                         self.nia, self.insn, self.nia_req, self.msr_o,
+                         self.ldst_req, self.ldst_addr,
+                         ClockSignal(),
+                         ResetSignal(),
+                        ]
             ports += list(self.dbg.dmi.ports())
             # for dbus/ibus microwatt, exclude err btw and cti
             for name, sig in self.imem.ibus.fields.items():
@@ -747,7 +784,10 @@ class TestIssuerBase(Elaboratable):
             # microwatt non-compliant with wishbone
             ports.append(self.ibus_adr)
             ports.append(self.dbus_adr)
-            return ports
+
+            if self.microwatt_compat:
+                # Ignore the remaining ports in microwatt compat mode
+                return ports
 
         ports = self.pc_i.ports()
         ports = self.msr_i.ports()
@@ -792,7 +832,7 @@ class TestIssuerInternal(TestIssuerBase):
     easy understanding) come later.
     """
 
-    def fetch_fsm(self, m, dbg, core, pc, msr, svstate, nia, is_svp64_mode,
+    def fetch_fsm(self, m, dbg, core, core_rst, nia, is_svp64_mode,
                         fetch_pc_o_ready, fetch_pc_i_valid,
                         fetch_insn_o_valid, fetch_insn_i_ready):
         """fetch FSM
@@ -806,6 +846,7 @@ class TestIssuerInternal(TestIssuerBase):
         pdecode2 = self.pdecode2
         cur_state = self.cur_state
         dec_opcode_i = pdecode2.dec.raw_opcode_in # raw opcode
+        pc, msr, svstate = cur_state.pc, cur_state.msr, cur_state.svstate
 
         # also note instruction fetch failed
         if hasattr(core, "icache"):
@@ -825,7 +866,7 @@ class TestIssuerInternal(TestIssuerBase):
             # allow fetch to not run at startup due to I-Cache reset not
             # having time to settle.  power-on-reset holds dbg.core_stopped_i
             with m.State("PRE_IDLE"):
-                with m.If(~dbg.core_stopped_i & ~dbg.core_stop_o):
+                with m.If(~dbg.core_stopped_i & ~dbg.core_stop_o & ~core_rst):
                     m.next = "IDLE"
 
             # waiting (zzz)
@@ -843,11 +884,6 @@ class TestIssuerInternal(TestIssuerBase):
                     comb += self.imem.a_pc_i.eq(pc)
                     comb += self.imem.a_i_valid.eq(1)
                     comb += self.imem.f_i_valid.eq(1)
-                    # transfer state to output
-                    sync += cur_state.pc.eq(pc)
-                    sync += cur_state.svstate.eq(svstate)  # and svstate
-                    sync += cur_state.msr.eq(msr)  # and msr
-
                     m.next = "INSN_READ"  # move to "wait for bus" phase
 
             # dummy pause to find out why simulation is not keeping up
@@ -906,7 +942,7 @@ class TestIssuerInternal(TestIssuerBase):
                             # not SVP64 - 32-bit only
                             sync += nia.eq(cur_state.pc + 4)
                             sync += dec_opcode_i.eq(insn)
-                            if self.microwatt_compat:
+                            if self.microwatt_compat or self.fabric_compat:
                                 # for verilator debug purposes
                                 comb += self.insn.eq(insn)
                                 comb += self.nia.eq(cur_state.pc)
@@ -1175,8 +1211,14 @@ class TestIssuerInternal(TestIssuerBase):
             fetch_failed = Const(0, 1)
             flush_needed = False
 
+        sync += fetch_pc_i_valid.eq(0)
+
         with m.FSM(name="issue_fsm"):
 
+            with m.State("PRE_IDLE"):
+                with m.If(~dbg.core_stop_o & ~core_rst):
+                    m.next = "ISSUE_START"
+
             # sync with the "fetch" phase which is reading the instruction
             # at this point, there is no instruction running, that
             # could inadvertently update the PC.
@@ -1186,7 +1228,10 @@ class TestIssuerInternal(TestIssuerBase):
                 # wait on "core stop" release, before next fetch
                 # need to do this here, in case we are in a VL==0 loop
                 with m.If(~dbg.core_stop_o & ~core_rst):
-                    comb += fetch_pc_i_valid.eq(1)  # tell fetch to start
+                    sync += fetch_pc_i_valid.eq(1)  # tell fetch to start
+                    sync += cur_state.pc.eq(dbg.state.pc)
+                    sync += cur_state.svstate.eq(dbg.state.svstate)
+                    sync += cur_state.msr.eq(dbg.state.msr)
                     with m.If(fetch_pc_o_ready):   # fetch acknowledged us
                         m.next = "INSN_WAIT"
                 with m.Else():
@@ -1460,6 +1505,7 @@ class TestIssuerInternal(TestIssuerBase):
         sync = m.d.sync
         dbg = self.dbg
         pdecode2 = self.pdecode2
+        cur_state = self.cur_state
 
         # temporaries
         core_busy_o = core.n.o_data.busy_o  # core is busy
@@ -1485,18 +1531,26 @@ class TestIssuerInternal(TestIssuerBase):
 
             # instruction started: must wait till it finishes
             with m.State("INSN_ACTIVE"):
-                # note changes to MSR, PC and SVSTATE, and DEC/TB
-                # these last two are done together, and passed to the
-                # DEC/TB FSM
+                # note changes to MSR, PC and SVSTATE
                 with m.If(self.state_nia.wen & (1 << StateRegs.SVSTATE)):
                     sync += self.sv_changed.eq(1)
                 with m.If(self.state_nia.wen & (1 << StateRegs.MSR)):
                     sync += self.msr_changed.eq(1)
                 with m.If(self.state_nia.wen & (1 << StateRegs.PC)):
                     sync += self.pc_changed.eq(1)
-                with m.If((self.state_spr.wen &
-                          ((1 << StateRegs.DEC) | (1 << StateRegs.TB))).bool()):
+                # and note changes to DEC/TB, to be passed to DEC/TB FSM
+                with m.If(self.state_spr.wen & (1 << StateRegs.TB)):
+                    comb += self.pause_dec_tb.eq(1)
+                # but also zero-out the cur_state DEC so that, on
+                # the next instruction, if it is "enable interrupt"
+                # the delay between the DEC/TB FSM reading and updating
+                # cur_state.dec doesn't trigger a spurious interrupt.
+                # the DEC/TB FSM will read the regfile and update to
+                # the correct value, so having cur_state.dec set to zero
+                # for a while is no big deal.
+                with m.If(self.state_spr.wen & (1 << StateRegs.DEC)):
                     comb += self.pause_dec_tb.eq(1)
+                    sync += cur_state.dec.eq(0) # only needs top bit clear
                 with m.If(~core_busy_o):  # instruction done!
                     comb += exec_pc_o_valid.eq(1)
                     with m.If(exec_pc_i_ready):
@@ -1583,8 +1637,7 @@ class TestIssuerInternal(TestIssuerBase):
         # Issue is where the VL for-loop # lives.  the ready/valid
         # signalling is used to communicate between the four.
 
-        self.fetch_fsm(m, dbg, core, dbg.state.pc, dbg.state.msr,
-                       dbg.state.svstate, nia, is_svp64_mode,
+        self.fetch_fsm(m, dbg, core, core_rst, nia, is_svp64_mode,
                        fetch_pc_o_ready, fetch_pc_i_valid,
                        fetch_insn_o_valid, fetch_insn_i_ready)
 
@@ -1606,9 +1659,10 @@ class TestIssuerInternal(TestIssuerBase):
                          exec_insn_i_valid, exec_insn_o_ready,
                          exec_pc_o_valid, exec_pc_i_ready)
 
-        # whatever was done above, over-ride it if core reset is held
+        # whatever was done above, over-ride it if core reset is held.
+        # set NIA to pc_at_reset
         with m.If(core_rst):
-            sync += nia.eq(0)
+            sync += nia.eq(self.core.pc_at_reset)
 
         return m