generic: Cell timing support

Signed-off-by: David Shah <dave@ds0.me>

common/nextpnr.h

@@ -181,9 +181,9 @@ struct GraphicElement
 
     float x1 = 0, y1 = 0, x2 = 0, y2 = 0, z = 0;
     std::string text;
-    GraphicElement() {};
-    GraphicElement(type_t type, style_t style, float x1, float y1, float x2, float y2, float z) : type(type),
-        style(style), x1(x1), y1(y1), x2(x2), y2(y2), z(z) {};
+    GraphicElement(){};
+    GraphicElement(type_t type, style_t style, float x1, float y1, float x2, float y2, float z)
+            : type(type), style(style), x1(x1), y1(y1), x2(x2), y2(y2), z(z){};
 };
 
 struct Loc

docs/generic.md

@@ -74,6 +74,26 @@ Sets the number of input pins a LUT in the architecture has. Only affects the ge
 
 Set the linear scaling vs distance and fixed offset (both values in nanoseconds) for routing delay estimates.
 
+### void addCellTimingClock(IdString cell, IdString port);
+
+Set the timing class of a port on a particular cell to a clock input.
+
+_NOTE: All cell timing functions apply to an individual named cell and not a cell type. This is because
+cell-specific configuration might affect timing, e.g. whether or not the register is used for a slice._
+
+### void addCellTimingDelay(IdString cell, IdString fromPort, IdString toPort, DelayInfo delay);
+
+Specify the combinational delay between two ports of a cell, and set the timing class of
+ those ports as combinational input/output.
+
+### void addCellTimingSetupHold(IdString cell, IdString port, IdString clock, DelayInfo setup, DelayInfo hold);
+
+Specify setup and hold timings for a port of a cell, and set the timing class of that port as register input.
+
+### void addCellTimingClockToOut(IdString cell, IdString port, IdString clock, DelayInfo clktoq);
+
+Specify clock-to-out time for a port of a cell, and set the timing class of that port as register output.
+
 ## Generic Packer
 
 The generic packer combines K-input LUTs (`LUT` cells) and simple D-type flip flops (`DFF` cells) (posedge clock only, no set/reset or enable) into a `GENERIC_SLICE` cell. It also inserts `GENERIC_IOB`s onto any top level IO pins without an IO buffer.

generic/arch.cc

@@ -17,8 +17,8 @@
  *
  */
 
-#include <math.h>
 #include <iostream>
+#include <math.h>
 #include "nextpnr.h"
 #include "placer1.h"
 #include "router1.h"
@@ -200,9 +200,42 @@ void Arch::setDelayScaling(double scale, double offset)
     args.delayOffset = offset;
 }
 
+void Arch::addCellTimingClock(IdString cell, IdString port) { cellTiming[cell].portClasses[port] = TMG_CLOCK_INPUT; }
+
+void Arch::addCellTimingDelay(IdString cell, IdString fromPort, IdString toPort, DelayInfo delay)
+{
+    if (get_or_default(cellTiming[cell].portClasses, fromPort, TMG_IGNORE) == TMG_IGNORE)
+        cellTiming[cell].portClasses[fromPort] = TMG_COMB_INPUT;
+    if (get_or_default(cellTiming[cell].portClasses, toPort, TMG_IGNORE) == TMG_IGNORE)
+        cellTiming[cell].portClasses[toPort] = TMG_COMB_OUTPUT;
+    cellTiming[cell].combDelays[CellDelayKey{fromPort, toPort}] = delay;
+}
+
+void Arch::addCellTimingSetupHold(IdString cell, IdString port, IdString clock, DelayInfo setup, DelayInfo hold)
+{
+    TimingClockingInfo ci;
+    ci.clock_port = clock;
+    ci.edge = RISING_EDGE;
+    ci.setup = setup;
+    ci.hold = hold;
+    cellTiming[cell].clockingInfo[port].push_back(ci);
+    cellTiming[cell].portClasses[port] = TMG_REGISTER_INPUT;
+}
+
+void Arch::addCellTimingClockToOut(IdString cell, IdString port, IdString clock, DelayInfo clktoq)
+{
+    TimingClockingInfo ci;
+    ci.clock_port = clock;
+    ci.edge = RISING_EDGE;
+    ci.clockToQ = clktoq;
+    cellTiming[cell].clockingInfo[port].push_back(ci);
+    cellTiming[cell].portClasses[port] = TMG_REGISTER_OUTPUT;
+}
+
 // ---------------------------------------------------------------
 
-Arch::Arch(ArchArgs args) : chipName("generic"), args(args) {
+Arch::Arch(ArchArgs args) : chipName("generic"), args(args)
+{
     // Dummy for empty decals
     decal_graphics[IdString()];
 }
@@ -473,7 +506,8 @@ bool Arch::route() { return router1(getCtx(), Router1Cfg(getCtx())); }
 
 // ---------------------------------------------------------------
 
-const std::vector<GraphicElement> &Arch::getDecalGraphics(DecalId decal) const {
+const std::vector<GraphicElement> &Arch::getDecalGraphics(DecalId decal) const
+{
     if (!decal_graphics.count(decal)) {
         std::cerr << "No decal named " << decal.str(this) << std::endl;
         log_error("No decal named %s!\n", decal.c_str(this));
@@ -493,18 +527,37 @@ DecalXY Arch::getGroupDecal(GroupId group) const { return groups.at(group).decal
 
 bool Arch::getCellDelay(const CellInfo *cell, IdString fromPort, IdString toPort, DelayInfo &delay) const
 {
-    return false;
+    if (!cellTiming.count(cell->name))
+        return false;
+    const auto &tmg = cellTiming.at(cell->name);
+    auto fnd = tmg.combDelays.find(CellDelayKey{fromPort, toPort});
+    if (fnd != tmg.combDelays.end()) {
+        delay = fnd->second;
+        return true;
+    } else {
+        return false;
+    }
 }
 
 // Get the port class, also setting clockPort if applicable
 TimingPortClass Arch::getPortTimingClass(const CellInfo *cell, IdString port, int &clockInfoCount) const
 {
-    return TMG_IGNORE;
+    if (!cellTiming.count(cell->name))
+        return TMG_IGNORE;
+    const auto &tmg = cellTiming.at(cell->name);
+    if (tmg.clockingInfo.count(port))
+        clockInfoCount = int(tmg.clockingInfo.at(port).size());
+    else
+        clockInfoCount = 0;
+    return get_or_default(tmg.portClasses, port, TMG_IGNORE);
 }
 
 TimingClockingInfo Arch::getPortClockingInfo(const CellInfo *cell, IdString port, int index) const
 {
-    NPNR_ASSERT_FALSE("no clocking info for generic");
+    NPNR_ASSERT(cellTiming.count(cell->name));
+    const auto &tmg = cellTiming.at(cell->name);
+    NPNR_ASSERT(tmg.clockingInfo.count(port));
+    return tmg.clockingInfo.at(port).at(index);
 }
 
 bool Arch::isValidBelForCell(CellInfo *cell, BelId bel) const

generic/arch.h

@@ -86,6 +86,33 @@ struct GroupInfo
     DecalXY decalxy;
 };
 
+struct CellDelayKey
+{
+    IdString from, to;
+    inline bool operator==(const CellDelayKey &other) const { return from == other.from && to == other.to; }
+};
+
+NEXTPNR_NAMESPACE_END
+namespace std {
+template <> struct hash<NEXTPNR_NAMESPACE_PREFIX CellDelayKey>
+{
+    std::size_t operator()(const NEXTPNR_NAMESPACE_PREFIX CellDelayKey &dk) const noexcept
+    {
+        std::size_t seed = std::hash<NEXTPNR_NAMESPACE_PREFIX IdString>()(dk.from);
+        seed ^= std::hash<NEXTPNR_NAMESPACE_PREFIX IdString>()(dk.to) + 0x9e3779b9 + (seed << 6) + (seed >> 2);
+        return seed;
+    }
+};
+} // namespace std
+NEXTPNR_NAMESPACE_BEGIN
+
+struct CellTiming
+{
+    std::unordered_map<IdString, TimingPortClass> portClasses;
+    std::unordered_map<CellDelayKey, DelayInfo> combDelays;
+    std::unordered_map<IdString, std::vector<TimingClockingInfo>> clockingInfo;
+};
+
 struct Arch : BaseCtx
 {
     std::string chipName;
@@ -106,6 +133,8 @@ struct Arch : BaseCtx
     std::vector<std::vector<int>> tileBelDimZ;
     std::vector<std::vector<int>> tilePipDimZ;
 
+    std::unordered_map<IdString, CellTiming> cellTiming;
+
     void addWire(IdString name, IdString type, int x, int y);
     void addPip(IdString name, IdString type, IdString srcWire, IdString dstWire, DelayInfo delay, Loc loc);
     void addAlias(IdString name, IdString type, IdString srcWire, IdString dstWire, DelayInfo delay);
@@ -133,6 +162,11 @@ struct Arch : BaseCtx
     void setLutK(int K);
     void setDelayScaling(double scale, double offset);
 
+    void addCellTimingClock(IdString cell, IdString port);
+    void addCellTimingDelay(IdString cell, IdString fromPort, IdString toPort, DelayInfo delay);
+    void addCellTimingSetupHold(IdString cell, IdString port, IdString clock, DelayInfo setup, DelayInfo hold);
+    void addCellTimingClockToOut(IdString cell, IdString port, IdString clock, DelayInfo clktoq);
+
     // ---------------------------------------------------------------
     // Common Arch API. Every arch must provide the following methods.
 

generic/arch_pybindings.cc

@@ -42,8 +42,8 @@ void arch_wrap_python()
     auto arch_cls = class_<Arch, Arch *, bases<BaseCtx>, boost::noncopyable>("Arch", init<ArchArgs>());
 
     auto dxy_cls = class_<ContextualWrapper<DecalXY>>("DecalXY_", no_init);
-            readwrite_wrapper<DecalXY, decltype(&DecalXY::decal), &DecalXY::decal, conv_to_str<DecalId>,
-                              conv_from_str<DecalId>>::def_wrap(dxy_cls, "decal");
+    readwrite_wrapper<DecalXY, decltype(&DecalXY::decal), &DecalXY::decal, conv_to_str<DecalId>,
+                      conv_from_str<DecalId>>::def_wrap(dxy_cls, "decal");
     readwrite_wrapper<DecalXY, decltype(&DecalXY::x), &DecalXY::x, pass_through<float>, pass_through<float>>::def_wrap(
             dxy_cls, "x");
     readwrite_wrapper<DecalXY, decltype(&DecalXY::y), &DecalXY::y, pass_through<float>, pass_through<float>>::def_wrap(
@@ -213,6 +213,22 @@ void arch_wrap_python()
     fn_wrapper_2a_v<Context, decltype(&Context::setDelayScaling), &Context::setDelayScaling, pass_through<double>,
                     pass_through<double>>::def_wrap(ctx_cls, "setDelayScaling", (arg("scale"), "offset"));
 
+    fn_wrapper_2a_v<Context, decltype(&Context::addCellTimingClock), &Context::addCellTimingClock,
+                    conv_from_str<IdString>, conv_from_str<IdString>>::def_wrap(ctx_cls, "addCellTimingClock",
+                                                                                (arg("cell"), "port"));
+    fn_wrapper_4a_v<Context, decltype(&Context::addCellTimingDelay), &Context::addCellTimingDelay,
+                    conv_from_str<IdString>, conv_from_str<IdString>, conv_from_str<IdString>,
+                    pass_through<DelayInfo>>::def_wrap(ctx_cls, "addCellTimingDelay",
+                                                       (arg("cell"), "fromPort", "toPort", "delay"));
+    fn_wrapper_5a_v<Context, decltype(&Context::addCellTimingSetupHold), &Context::addCellTimingSetupHold,
+                    conv_from_str<IdString>, conv_from_str<IdString>, conv_from_str<IdString>, pass_through<DelayInfo>,
+                    pass_through<DelayInfo>>::def_wrap(ctx_cls, "addCellTimingSetupHold",
+                                                       (arg("cell"), "port", "clock", "setup", "hold"));
+    fn_wrapper_4a_v<Context, decltype(&Context::addCellTimingClockToOut), &Context::addCellTimingClockToOut,
+                    conv_from_str<IdString>, conv_from_str<IdString>, conv_from_str<IdString>,
+                    pass_through<DelayInfo>>::def_wrap(ctx_cls, "addCellTimingClockToOut",
+                                                       (arg("cell"), "port", "clock", "clktoq"));
+
     WRAP_MAP_UPTR(CellMap, "IdCellMap");
     WRAP_MAP_UPTR(NetMap, "IdNetMap");
     WRAP_VECTOR(const std::vector<IdString>, conv_to_str<IdString>);

generic/examples/README.md

@@ -4,6 +4,8 @@ This contains a simple, artificial, example of the nextpnr generic API.
 
  - simple.py procedurally generates a simple FPGA architecture with IO at the edges,
    logic slices in all other tiles, and interconnect only between adjacent tiles
+ 
+ - simple_timing.py annotates cells with timing data (this is a separate script that must be run after packing)
 
  - report.py stores design information after place-and-route to blinky.txt in place
    of real bitstream generation

generic/examples/simple.sh

@@ -1,4 +1,4 @@
 #!/usr/bin/bash
 set -ex
 yosys -p "tcl ../synth/synth_generic.tcl 4 blinky.json" blinky.v
-../../nextpnr-generic --pre-pack simple.py --json blinky.json --post-route report.py
+../../nextpnr-generic --pre-pack simple.py --pre-place simple_timing.py --json blinky.json --post-route report.py

generic/examples/simple_timing.py

@@ -0,0 +1,15 @@
+for cname, cell in ctx.cells:
+	if cell.type != "GENERIC_SLICE":
+		continue
+	if cname in ("$PACKER_GND", "$PACKER_VCC"):
+		continue
+	K = int(cell.params["K"])
+	if cell.params["FF_USED"] == "1":
+		ctx.addCellTimingClock(cell=cname, port="CLK")
+		for i in range(K):
+			ctx.addCellTimingSetupHold(cell=cname, port="I[%d]" % i, clock="CLK",
+				setup=ctx.getDelayFromNS(0.2), hold=ctx.getDelayFromNS(0))
+		ctx.addCellTimingClockToOut(cell=cname, port="Q", clock="CLK", clktoq=ctx.getDelayFromNS(0.2))
+	else:
+		for i in range(K):
+			ctx.addCellTimingDelay(cell=cname, fromPort="I[%d]" % i, toPort="Q", delay=ctx.getDelayFromNS(0.2))