Gowin. Fix pipeline mode in BSRAM.

It seems that the internal registers on the BSRAM output pins in
READ_MODE=1'b1 (pipeline) mode do not function properly because in the
images generated by Gowin IDE an external register is added to each pin,
and the BSRAM itself switches to READ_MODE=1'b0 (bypass) mode .

This is observed on Tangnano9k and Tangnano20k boards.

Here we repeat this fix.

Signed-off-by: YRabbit <rabbit@yrabbit.cyou>

himbaechel/uarch/gowin/constids.inc

@@ -892,6 +892,7 @@ X(WRE)
 X(BSRAM_SUBTYPE)
 X(WRITE_MODE)
 X(READ_MODE)
+X(RESET_MODE)
 X(BIT_WIDTH)
 X(BIT_WIDTH_0)
 X(BIT_WIDTH_1)

himbaechel/uarch/gowin/gowin.h

@@ -97,6 +97,7 @@ NPNR_PACKED_STRUCT(struct Extra_chip_data_POD {
     // chip flags
     static constexpr int32_t HAS_SP32 = 1;
     static constexpr int32_t NEED_SP_FIX = 2;
+    static constexpr int32_t NEED_BSRAM_OUTREG_FIX = 4;
 });
 
 } // namespace

himbaechel/uarch/gowin/gowin_arch_gen.py

@@ -15,8 +15,9 @@ from apycula import chipdb
 BEL_FLAG_SIMPLE_IO = 0x100
 
 # Chip flags
-CHIP_HAS_SP32    = 0x1
-CHIP_NEED_SP_FIX = 0x2
+CHIP_HAS_SP32             = 0x1
+CHIP_NEED_SP_FIX          = 0x2
+CHIP_NEED_BSRAM_OUTREG_FIX = 0x4
 
 # Z of the bels
 # sync with C++ part!
@@ -1018,6 +1019,8 @@ def main():
             chip_flags |= CHIP_HAS_SP32;
         if "NEED_SP_FIX" in db.chip_flags:
             chip_flags |= CHIP_NEED_SP_FIX;
+        if "NEED_BSRAM_OUTREG_FIX" in db.chip_flags:
+            chip_flags |= CHIP_NEED_BSRAM_OUTREG_FIX;
 
     X = db.cols;
     Y = db.rows;

himbaechel/uarch/gowin/gowin_utils.cc

@@ -124,6 +124,12 @@ bool GowinUtils::need_SP_fix(void)
     return extra->chip_flags & Extra_chip_data_POD::NEED_SP_FIX;
 }
 
+bool GowinUtils::need_BSRAM_OUTREG_fix(void)
+{
+    const Extra_chip_data_POD *extra = reinterpret_cast<const Extra_chip_data_POD *>(ctx->chip_info->extra_data.get());
+    return extra->chip_flags & Extra_chip_data_POD::NEED_BSRAM_OUTREG_FIX;
+}
+
 std::unique_ptr<CellInfo> GowinUtils::create_cell(IdString name, IdString type)
 {
     NPNR_ASSERT(!ctx->cells.count(name));

himbaechel/uarch/gowin/gowin_utils.h

@@ -37,6 +37,7 @@ struct GowinUtils
     // BSRAM
     bool have_SP32(void);
     bool need_SP_fix(void);
+    bool need_BSRAM_OUTREG_fix(void);
 
     // DSP
     inline int get_dsp_18_z(int z) const { return z & (~3); }

himbaechel/uarch/gowin/pack.cc

@@ -1331,6 +1331,73 @@ struct GowinPacker
         }
     }
 
+    // Some chips cannot, for some reason, use internal BSRAM registers to
+    // implement READ_MODE=1'b1 (pipeline) with a word width other than 32 or
+    // 36 bits.
+    // We work around this by adding an external DFF and using BSRAM
+    // as READ_MODE=1'b0 (bypass).
+    void bsram_fix_outreg(CellInfo *ci, std::vector<std::unique_ptr<CellInfo>> &new_cells)
+    {
+        int bit_width = ci->params.at(id_BIT_WIDTH).as_int64();
+        if (bit_width == 32 || bit_width == 36) {
+            return;
+        }
+        int read_mode = ci->params.at(id_READ_MODE).as_int64();
+        if (read_mode == 0) {
+            return;
+        }
+        NetInfo *ce_net = ci->getPort(id_CE);
+        NetInfo *oce_net = ci->getPort(id_OCE);
+        if (ce_net == nullptr || oce_net == nullptr) {
+            return;
+        }
+        if (ce_net->name == ctx->id("$PACKER_GND") || oce_net->name == ctx->id("$PACKER_GND")) {
+            return;
+        }
+
+        if (ctx->verbose) {
+            log_info("  apply the BSRAM OUTREG fix\n");
+        }
+        ci->setParam(id_READ_MODE, 0);
+        ci->disconnectPort(id_OCE);
+        ci->connectPort(id_OCE, ce_net);
+
+        NetInfo *reset_net = ci->getPort(id_RESET);
+        bool sync_reset = ci->params.at(id_RESET_MODE).as_string() == std::string("SYNC");
+        IdString dff_type = sync_reset ? id_DFFRE : id_DFFCE;
+        IdString reset_port = sync_reset ? id_RESET : id_CLEAR;
+
+        for (int i = 0; i < bit_width; ++i) {
+            IdString do_name = ctx->idf("DO[%d]", i);
+            const NetInfo *net = ci->getPort(do_name);
+            if (net != nullptr) {
+                if (net->users.empty()) {
+                    ci->disconnectPort(do_name);
+                    continue;
+                }
+
+                // create DFF
+                auto cache_dff_cell = gwu.create_cell(create_aux_name(ci->name, i, "_cache_dff$"), dff_type);
+                CellInfo *cache_dff = cache_dff_cell.get();
+                cache_dff->addInput(id_CE);
+                cache_dff->connectPort(id_CE, oce_net);
+
+                cache_dff->addInput(reset_port);
+                cache_dff->connectPort(reset_port, reset_net);
+
+                ci->copyPortTo(id_CLK, cache_dff, id_CLK);
+
+                cache_dff->addOutput(id_Q);
+                ci->movePortTo(do_name, cache_dff, id_Q);
+
+                cache_dff->addInput(id_D);
+                ci->connectPorts(do_name, cache_dff, id_D);
+
+                new_cells.push_back(std::move(cache_dff_cell));
+            }
+        }
+    }
+
     // Analysis of the images generated by the IDE showed that some components
     // are being added at the input and output of the BSRAM.  Two LUTs are
     // added on the WRE and CE inputs (strangely, OCE is not affected), a pair
@@ -1342,6 +1409,9 @@ struct GowinPacker
     {
         int bit_width = ci->params.at(id_BIT_WIDTH).as_int64();
 
+        if (ctx->verbose) {
+            log_info("  apply the SP fix\n");
+        }
         // create WRE LUT
         auto wre_lut_cell = gwu.create_cell(create_aux_name(ci->name, 0, "_wre_lut$"), id_LUT4);
         CellInfo *wre_lut = wre_lut_cell.get();
@@ -1654,6 +1724,11 @@ struct GowinPacker
             bsram_fix_sp(ci, new_cells);
         }
 
+        // Some chips have faulty output registers
+        if (gwu.need_BSRAM_OUTREG_fix()) {
+            bsram_fix_outreg(ci, new_cells);
+        }
+
         // XXX UG285-1.3.6_E Gowin BSRAM & SSRAM User Guide:
         // For GW1N-9/GW1NR-9/GW1NS-4 series, 32/36-bit SP/SPX9 is divided into two
         // SP/SPX9s, which occupy two BSRAMs.