WIP shadowram

6 files changed, 175 insertions, 0 deletions

diff --git a/gowin/arch.cc b/gowin/arch.cc
index 4fc2cd43..d554bd64 100644
--- a/gowin/arch.cc
+++ b/gowin/arch.cc
@@ -791,6 +791,39 @@ void Arch::read_cst(std::istream &in)
     settings[id_cst] = 1;
 }
 
+void Arch::addShadowRamBels(const DatabasePOD *db, int row, int col)
+{
+    IdString belname, bel_id;
+    char buf[32];
+    snprintf(buf, 32, "R%dC%d_RAMW", row + 1, col + 1);
+    belname = id(buf);
+    addBel(belname, id_RAMW, Loc(col, row, BelZ::lutram_0_z), false);
+
+    snprintf(buf, 32, "R%dC%d_A%d", row + 1, col + 1, 4);
+    addBelInput(belname, id_A4, id(buf));
+    snprintf(buf, 32, "R%dC%d_B%d", row + 1, col + 1, 4);
+    addBelInput(belname, id_B4, id(buf));
+    snprintf(buf, 32, "R%dC%d_C%d", row + 1, col + 1, 4);
+    addBelInput(belname, id_C4, id(buf));
+    snprintf(buf, 32, "R%dC%d_D%d", row + 1, col + 1, 4);
+    addBelInput(belname, id_D4, id(buf));
+
+    snprintf(buf, 32, "R%dC%d_A%d", row + 1, col + 1, 5);
+    addBelInput(belname, id_A5, id(buf));
+    snprintf(buf, 32, "R%dC%d_B%d", row + 1, col + 1, 5);
+    addBelInput(belname, id_B5, id(buf));
+    snprintf(buf, 32, "R%dC%d_C%d", row + 1, col + 1, 5);
+    addBelInput(belname, id_C5, id(buf));
+    snprintf(buf, 32, "R%dC%d_D%d", row + 1, col + 1, 5);
+    addBelInput(belname, id_D5, id(buf));
+
+    snprintf(buf, 32, "R%dC%d_CLK%d", row + 1, col + 1, 2);
+    addBelInput(belname, id_CLK, id(buf));
+    snprintf(buf, 32, "R%dC%d_LSR%d", row + 1, col + 1, 2);
+    addBelInput(belname, id_LSR, id(buf));
+}
+
+
 // Add all MUXes for the cell
 void Arch::addMuxBels(const DatabasePOD *db, int row, int col)
 {
@@ -1112,6 +1145,9 @@ Arch::Arch(ArchArgs args) : args(args)
                 if (z == 0) {
                     addMuxBels(db, row, col);
                 }
+                if (z == 4) {
+                    addShadowRamBels(db, row, col);
+                }
                 if (z % 2 == 0) {
                     snprintf(buf, 32, "R%dC%d_LUT_GRP%d", row + 1, col + 1, z);
                     grpname = id(buf);
diff --git a/gowin/arch.h b/gowin/arch.h
index 14181d79..80f887bb 100644
--- a/gowin/arch.h
+++ b/gowin/arch.h
@@ -344,6 +344,7 @@ struct Arch : BaseArch<ArchRanges>
     DelayQuad getWireTypeDelay(IdString wire);
     void read_cst(std::istream &in);
     void addMuxBels(const DatabasePOD *db, int row, int col);
+    void addShadowRamBels(const DatabasePOD *db, int row, int col);
 
     // ---------------------------------------------------------------
     // Common Arch API. Every arch must provide the following methods.
@@ -485,6 +486,7 @@ enum
 {
     mux_0_z = 10,     // start Z for the MUX2LUT5 bels
     iologic_0_z = 20, // start Z for the IOLOGIC bels
+    lutram_0_z = 30, // start Z for the IOLOGIC bels
     vcc_0_z = 277,    // virtual VCC bel Z
     gnd_0_z = 278,    // virtual VSS bel Z
     osc_z = 280       // Z for the oscillator bels
diff --git a/gowin/cells.cc b/gowin/cells.cc
index c3b21782..8285dc4e 100644
--- a/gowin/cells.cc
+++ b/gowin/cells.cc
@@ -166,4 +166,37 @@ void gwio_to_iob(Context *ctx, CellInfo *nxio, CellInfo *iob, pool<IdString> &to
     }
 }
 
+void sram_to_ramw_split(Context *ctx, CellInfo *ram, CellInfo *ramw)
+{
+    if (ramw->hierpath == IdString())
+        ramw->hierpath = ramw->hierpath;
+    ram->movePortTo(ctx->id("WAD[0]"), ramw, id_A4);
+    ram->movePortTo(ctx->id("WAD[1]"), ramw, id_B4);
+    ram->movePortTo(ctx->id("WAD[2]"), ramw, id_C4);
+    ram->movePortTo(ctx->id("WAD[3]"), ramw, id_D4);
+
+    ram->movePortTo(ctx->id("DI[0]"), ramw, id_A5);
+    ram->movePortTo(ctx->id("DI[1]"), ramw, id_B5);
+    ram->movePortTo(ctx->id("DI[2]"), ramw, id_C5);
+    ram->movePortTo(ctx->id("DI[3]"), ramw, id_D5);
+    
+    ram->movePortTo(ctx->id("CLK"), ramw, id_CLK);
+    ram->movePortTo(ctx->id("WRE"), ramw, id_LSR);
+}
+
+void sram_to_slice(Context *ctx, CellInfo *ram, CellInfo *slice, int index)
+{
+    char buf1[32];
+    if (slice->hierpath == IdString())
+        slice->hierpath = slice->hierpath;
+
+    snprintf(buf1, 32, "DO[%d]", index);
+    ram->movePortTo(ctx->id(buf1), slice, id_F);
+
+    ram->copyPortTo(ctx->id("RAD[0]"), slice, id_A);
+    ram->copyPortTo(ctx->id("RAD[1]"), slice, id_B);
+    ram->copyPortTo(ctx->id("RAD[2]"), slice, id_C);
+    ram->copyPortTo(ctx->id("RAD[3]"), slice, id_D);
+}
+
 NEXTPNR_NAMESPACE_END
diff --git a/gowin/cells.h b/gowin/cells.h
index 8f0636b8..b6d86497 100644
--- a/gowin/cells.h
+++ b/gowin/cells.h
@@ -111,6 +111,8 @@ inline bool is_ff(const BaseCtx *ctx, const CellInfo *cell)
 
 inline bool is_lc(const BaseCtx *ctx, const CellInfo *cell) { return cell->type == id_SLICE; }
 
+inline bool is_sram(const BaseCtx *ctx, const CellInfo *cell) { return cell->type == id_RAM16SDP4; }
+
 // Convert a LUT primitive to (part of) an GENERIC_SLICE, swapping ports
 // as needed. Set no_dff if a DFF is not being used, so that the output
 // can be reconnected
@@ -125,6 +127,12 @@ void dff_to_lc(const Context *ctx, CellInfo *dff, CellInfo *lc, bool pass_thru_l
 // Convert a Gowin IO buffer to a IOB bel
 void gwio_to_iob(Context *ctx, CellInfo *nxio, CellInfo *sbio, pool<IdString> &todelete_cells);
 
+// Convert RAM16 to write port
+void sram_to_ramw_split(Context *ctx, CellInfo *ram, CellInfo *ramw);
+
+// Convert RAM16 to slice
+void sram_to_slice(Context *ctx, CellInfo *ram, CellInfo *slice, int index);
+
 NEXTPNR_NAMESPACE_END
 
 #endif
diff --git a/gowin/constids.inc b/gowin/constids.inc
index 3691c506..d9ca1f0c 100644
--- a/gowin/constids.inc
+++ b/gowin/constids.inc
@@ -748,6 +748,19 @@ X(DFFNPE)
 X(DFFNC)
 X(DFFNCE)
 
+// Shadow RAM
+X(RAMW)
+X(RAM16SDP4)
+X(WADA)
+X(WADB)
+X(WADC)
+X(WADD)
+X(DIA)
+X(DIB)
+X(DIC)
+X(DID)
+X(WRE)
+
 // IOB types
 X(IBUF)
 X(OBUF)
diff --git a/gowin/pack.cc b/gowin/pack.cc
index 4adfec1a..fa88772a 100644
--- a/gowin/pack.cc
+++ b/gowin/pack.cc
@@ -692,6 +692,89 @@ static void pack_gsr(Context *ctx)
     }
 }
 
+// Pack shadow RAM
+void pack_sram(Context *ctx)
+{
+    pool<IdString> packed_cells;
+    std::vector<std::unique_ptr<CellInfo>> new_cells;
+
+    for (auto &cell : ctx->cells) {
+        CellInfo *ci = cell.second.get();
+        if (is_sram(ctx, ci)) {
+
+            // Create RAMW slice
+            std::unique_ptr<CellInfo> ramw_slice =
+                    create_generic_cell(ctx, id_RAMW, ci->name.str(ctx) + "$RAMW_SLICE");
+            sram_to_ramw_split(ctx, ci, ramw_slice.get());
+
+            // Create actual RAM slices
+            std::unique_ptr<CellInfo> ram_comb[4];
+            for (int i = 0; i < 4; i++) {
+                ram_comb[i] = create_generic_cell(ctx, id_SLICE,
+                                                ci->name.str(ctx) + "$SRAM_SLICE" + std::to_string(i));
+                sram_to_slice(ctx, ci, ram_comb[i].get(), i);
+            }
+            // Create 'block' SLICEs as a placement hint that these cells are mutually exclusive with the RAMW
+            std::unique_ptr<CellInfo> ramw_block[2];
+            for (int i = 0; i < 2; i++) {
+                ramw_block[i] = create_generic_cell(ctx, id_SLICE,
+                                                    ci->name.str(ctx) + "$RAMW_BLOCK" + std::to_string(i));
+                ramw_block[i]->params[id_FF_TYPE] = std::string("RAMW_BLOCK");
+            }
+
+            // Disconnect ports of original cell after packing
+            // ci->disconnectPort(id_WCK);
+            // ci->disconnectPort(id_WRE);
+
+            for (int i = 0; i < 4; i++)
+                ci->disconnectPort(ctx->id(stringf("RAD[%d]", i)));
+
+            // Setup placement constraints
+            // Use the 0th bit as an anchor
+            ram_comb[0]->constr_abs_z = true;
+            ram_comb[0]->constr_z = 0;
+            ram_comb[0]->cluster = ram_comb[0]->name;
+            for (int i = 1; i < 4; i++) {
+                ram_comb[i]->cluster = ram_comb[0]->name;
+                ram_comb[i]->constr_abs_z = true;
+                ram_comb[i]->constr_x = 0;
+                ram_comb[i]->constr_y = 0;
+                ram_comb[i]->constr_z = i;
+                ram_comb[0]->constr_children.push_back(ram_comb[i].get());
+            }
+            for (int i = 0; i < 2; i++) {
+                ramw_block[i]->cluster = ram_comb[0]->name;
+                ramw_block[i]->constr_abs_z = true;
+                ramw_block[i]->constr_x = 0;
+                ramw_block[i]->constr_y = 0;
+                ramw_block[i]->constr_z = i + 4;
+                ram_comb[0]->constr_children.push_back(ramw_block[i].get());
+            }
+
+            ramw_slice->cluster = ram_comb[0]->name;
+            ramw_slice->constr_abs_z = true;
+            ramw_slice->constr_x = 0;
+            ramw_slice->constr_y = 0;
+            ramw_slice->constr_z = 4;
+            ram_comb[0]->constr_children.push_back(ramw_slice.get());
+
+            for (int i = 0; i < 4; i++)
+                new_cells.push_back(std::move(ram_comb[i]));
+            for (int i = 0; i < 2; i++)
+                new_cells.push_back(std::move(ramw_block[i]));
+            new_cells.push_back(std::move(ramw_slice));
+            packed_cells.insert(ci->name);
+        }
+    }
+    for (auto pcell : packed_cells) {
+        ctx->cells.erase(pcell);
+    }
+    for (auto &ncell : new_cells) {
+        ctx->cells[ncell->name] = std::move(ncell);
+    }
+}
+
+
 static bool is_nextpnr_iob(const Context *ctx, CellInfo *cell)
 {
     return cell->type == ctx->id("$nextpnr_ibuf") || cell->type == ctx->id("$nextpnr_obuf") ||