Implement the DSP primitive. (#239)

* Implement the DSP primitive. For chips that have these capabilities, a DSP implementation has been added in the form of all the primitives described in the Gowin documentation (UG287-1.3.3E_Gowin Digital Signal Processing (DSP) User Guide), namely: - PADD9 - PADD18 - MULT9X9 - MULT18X18 - MULT36X36 - MULTALU18X18 - MULTALU36X18 - MULTADDALU18X18 - ALU54D The most complex but also the most useful is the MULTADDALU18X18 primitive - it allows you to easily make a typical FIR filter, while all connections between these primitives in the chain will be implemented by direct fixed wires with minimal delay. MULT36X36 are not combined into chains, but they have a different task - this primitive can be found in Linux SOCs. Added examples (in the examples/himbaechel directory) that are based on the tiny Riscv demonstrating UART calculations. Only the TXD pin is used (can be found in the specific .CST file for each board), so on the large computer side, only GND and RXD are enough. Port speed 115200, no parity, 8 data bits, 1 stop bit, linefeed only. Picocom launch example: ``` shell picocom -l --imap lfcrlf -b 115200 /dev/ttyU0 ``` The source code for the riscv test programs is provided along with the assembly instructions, but they are not built during the compilation of the examples due to additional compilers. Implemented the combination of primitives into chains using wires CASO-CASI, SO(A, B)-SI(A, B), as well as SBO-SBI for PADD. Signed-off-by: YRabbit <[email protected]> * I forgot to remove the debugging part. Fixed. Signed-off-by: YRabbit <[email protected]> --------- Signed-off-by: YRabbit <[email protected]>
YosysHQ · Mar 31, 2024 · 91807b0 · 91807b0
1 parent 1ff94b7
commit 91807b0
Show file tree

Hide file tree

Showing 37 changed files with 8,885 additions and 72 deletions.
diff --git a/apycula/attrids.py b/apycula/attrids.py
diff --git a/apycula/chipdb.py b/apycula/chipdb.py
diff --git a/apycula/gowin_pack.py b/apycula/gowin_pack.py
diff --git a/apycula/gowin_unpack.py b/apycula/gowin_unpack.py
@@ -22,6 +22,12 @@
         'GW2A-18C' : 'PBGA256S'
 }
 
+def print_sorted_dict(start, d):
+    print(start, end='{')
+    for i in sorted(d):
+        print(f'{i}:{d[i]}, ', end='')
+    print('}')
+
 # bank iostandards
 # XXX default io standard may be board-dependent!
 _banks = {'0': "LVCMOS18", '1': "LVCMOS18", '2': "LVCMOS18", '3': "LVCMOS18"}
@@ -59,6 +65,7 @@ def get_attr_name(attrname_table, code):
     for name, cod in attrname_table.items():
         if cod == code:
             return name
+    print(f'Unknown attr name for {code}/0x{code:x}.')
     return ''
 
 # fix names and types of the PLL attributes
@@ -293,7 +300,7 @@ def get_pll_A(db, row, col, typ):
         'IDDRX8': 'IDES16',
         }
 
-# BSRAM have 3 cells: BSRAM, BSRAM0 and BSRAM1
+# BSRAM has 3 cells: BSRAM, BSRAM0 and BSRAM1
 # { (row, col) : idx }
 _bsram_cells = {}
 def get_bsram_main_cell(db, row, col, typ):
@@ -303,6 +310,12 @@ def get_bsram_main_cell(db, row, col, typ):
             col -= 2
     return row, col
 
+# The DSP has 9 cells: the main one and a group of auxiliary ones.
+def get_dsp_main_cell(db, row, col, typ):
+    if type[-6:-2] == '_AUX':
+        col = 1 + (col - 1) // 9
+    return row, col
+
 # noiostd --- this is the case when the function is called
 # with iostd by default, e.g. from the clock fuzzer
 # With normal gowin_unpack io standard is determined first and it is known.
@@ -361,6 +374,23 @@ def parse_tile_(db, row, col, tile, default=True, noalias=False, noiostd = True)
             #print(row, col, name, idx, tiledata.ttyp, attrvals)
             bels[f'{name}'] = {}
             continue
+        if name.startswith("ALU54D"):
+            continue
+        if name.startswith("DSP") or name.startswith("DSP_AUX"):
+            modes = set()
+            idx = name[-1]
+            #print(row, col, name, idx, tiledata.ttyp)
+            if name.startswith("DSP_AUX"):
+                row, col = get_dsp_main_cell(db, row, col, name)
+
+            if f'DSP{idx}' in db.shortval[tiledata.ttyp]:
+                attrvals = parse_attrvals(tile, db.logicinfo['DSP'], db.shortval[tiledata.ttyp][f'DSP{idx}'], attrids.dsp_attrids)
+                #print_sorted_dict(f'{row}, {col}, {name}, {idx}, {tiledata.ttyp} - ', attrvals)
+                for attrval in attrvals:
+                    modes.add(attrval)
+            if modes and not name.startswith("DSP_AUX"):
+                bels[f'{name}{idx}'] = modes
+            continue
         if name.startswith("IOLOGIC"):
             idx = name[-1]
             attrvals = parse_attrvals(tile, db.logicinfo['IOLOGIC'], db.shortval[tiledata.ttyp][f'IOLOGIC{idx}'], attrids.iologic_attrids)
@@ -802,7 +832,7 @@ def tile2verilog(dbrow, dbcol, bels, pips, clock_pips, mod, cst, db):
         mod.wires.update({srcg, destg})
         mod.assigns.append((destg, srcg))
 
-    belre = re.compile(r"(IOB|LUT|DFF|BANK|CFG|ALU|RAM16|ODDR|OSC[ZFHWO]?|BUFS|RPLL[AB]|PLLVR|IOLOGIC|BSRAM)(\w*)")
+    belre = re.compile(r"(IOB|LUT|DFF|BANK|CFG|ALU|RAM16|ODDR|OSC[ZFHWO]?|BUFS|RPLL[AB]|PLLVR|IOLOGIC|BSRAM|DSP)(\w*)")
     bels_items = move_iologic(bels)
 
     iologic_detected = set()

diff --git a/examples/himbaechel/Makefile.himbaechel b/examples/himbaechel/Makefile.himbaechel
@@ -12,7 +12,10 @@ all: \
 	bsram-pROM-tangnano20k.fs bsram-SDPB-tangnano20k.fs bsram-SP-tangnano20k.fs \
 	bsram-DPB-tangnano20k.fs bsram-pROMX9-tangnano20k.fs bsram-SDPX9B-tangnano20k.fs \
 	bsram-SPX9-tangnano20k.fs bsram-DPX9B-tangnano20k.fs \
-	femto-riscv-15-tangnano20k.fs femto-riscv-16-tangnano20k.fs femto-riscv-16-tangnano20k.fs \
+	femto-riscv-15-tangnano20k.fs femto-riscv-16-tangnano20k.fs femto-riscv-18-tangnano20k.fs \
+	dsp-mult36x36-tangnano20k.fs dsp-padd9-tangnano20k.fs dsp-padd18-tangnano20k.fs \
+	dsp-mult9x9-tangnano20k.fs dsp-alu54d-tangnano20k.fs dsp-multalu18x18-tangnano20k.fs \
+	dsp-multalu36x18-tangnano20k.fs dsp-multaddalu18x18-tangnano20k.fs \
 	\
 	blinky-primer20k.fs shift-primer20k.fs blinky-tbuf-primer20k.fs blinky-oddr-primer20k.fs \
 	blinky-osc-primer20k.fs tlvds-primer20k.fs elvds-primer20k.fs oddr-tlvds-primer20k.fs \
@@ -22,6 +25,10 @@ all: \
 	bsram-pROM-primer20k.fs bsram-SDPB-primer20k.fs bsram-SP-primer20k.fs \
 	bsram-DPB-primer20k.fs bsram-pROMX9-primer20k.fs bsram-SDPX9B-primer20k.fs \
 	bsram-SPX9-primer20k.fs bsram-DPX9B-primer20k.fs \
+	femto-riscv-15-primer20k.fs femto-riscv-16-primer20k.fs femto-riscv-18-primer20k.fs \
+	dsp-mult36x36-primer20k.fs dsp-padd9-primer20k.fs dsp-padd18-primer20k.fs \
+	dsp-mult9x9-primer20k.fs dsp-alu54d-primer20k.fs dsp-multalu18x18-primer20k.fs \
+	dsp-multalu36x18-primer20k.fs dsp-multaddalu18x18-primer20k.fs \
 	\
 	blinky-tangnano.fs shift-tangnano.fs blinky-tbuf-tangnano.fs blinky-oddr-tangnano.fs \
 	blinky-osc-tangnano.fs elvds-tangnano.fs oddr-elvds-tangnano.fs pll-nanolcd-tangnano.fs \
@@ -46,7 +53,10 @@ all: \
 	ides16-tangnano4k.fs \
 	ides4-tangnano4k.fs ivideo-tangnano4k.fs ides8-tangnano4k.fs ides10-tangnano4k.fs \
 	oser10-tlvds-tangnano4k.fs \
-	femto-riscv-15-tangnano4k.fs femto-riscv-16-tangnano4k.fs femto-riscv-16-tangnano4k.fs \
+	femto-riscv-15-tangnano4k.fs femto-riscv-16-tangnano4k.fs femto-riscv-18-tangnano4k.fs \
+	dsp-mult36x36-tangnano4k.fs dsp-padd9-tangnano4k.fs dsp-padd18-tangnano4k.fs \
+	dsp-mult9x9-tangnano4k.fs dsp-alu54d-tangnano4k.fs dsp-multalu18x18-tangnano4k.fs \
+	dsp-multalu36x18-tangnano4k.fs dsp-multaddalu18x18-tangnano4k.fs \
 	\
 	blinky-tangnano9k.fs shift-tangnano9k.fs blinky-tbuf-tangnano9k.fs blinky-oddr-tangnano9k.fs \
 	blinky-osc-tangnano9k.fs tlvds-tangnano9k.fs elvds-tangnano9k.fs oddr-tlvds-tangnano9k.fs \
@@ -58,6 +68,9 @@ all: \
 	bsram-SPX9-tangnano9k.fs bsram-DPX9B-tangnano9k.fs \
 	oser10-elvds-tangnano9k.fs \
 	femto-riscv-15-tangnano9k.fs femto-riscv-16-tangnano9k.fs femto-riscv-18-tangnano9k.fs \
+	dsp-mult36x36-tangnano9k.fs dsp-padd9-tangnano9k.fs dsp-padd18-tangnano9k.fs \
+	dsp-mult9x9-tangnano9k.fs dsp-alu54d-tangnano9k.fs dsp-multalu18x18-tangnano9k.fs \
+	dsp-multalu36x18-tangnano9k.fs dsp-multaddalu18x18-tangnano9k.fs \
 	\
 	blinky-szfpga.fs shift-szfpga.fs blinky-tbuf-szfpga.fs blinky-oddr-szfpga.fs \
 	blinky-osc-szfpga.fs tlvds-szfpga.fs elvds-szfpga.fs oddr-tlvds-szfpga.fs \
@@ -68,19 +81,29 @@ all: \
 	bsram-pROM-szfpga.fs bsram-SDPB-szfpga.fs bsram-SP-szfpga.fs \
 	bsram-pROMX9-szfpga.fs bsram-SDPX9B-szfpga.fs \
 	bsram-SPX9-szfpga.fs \
+	femto-riscv-15-szfpga.fs femto-riscv-16-szfpga.fs femto-riscv-18-szfpga.fs \
+	dsp-mult36x36-szfpga.fs dsp-padd9-szfpga.fs dsp-padd18-szfpga.fs \
+	dsp-mult9x9-szfpga.fs dsp-alu54d-szfpga.fs dsp-multalu18x18-szfpga.fs \
+	dsp-multalu36x18-szfpga.fs dsp-multaddalu18x18-szfpga.fs \
 	\
 	blinky-tec0117.fs shift-tec0117.fs blinky-tbuf-tec0117.fs blinky-oddr-tec0117.fs \
 	blinky-osc-tec0117.fs tlvds-tec0117.fs elvds-tec0117.fs oddr-tlvds-tec0117.fs \
 	oddr-elvds-tec0117.fs blinky-pll-tec0117.fs oser16-tec0117.fs attosoc-tec0117.fs \
 	oser4-tec0117.fs ovideo-tec0117.fs oser8-tec0117.fs oser10-tec0117.fs \
 	ides16-tec0117.fs \
 	ides4-tec0117.fs ivideo-tec0117.fs ides8-tec0117.fs ides10-tec0117.fs \
+	dsp-mult36x36-tec0117.fs dsp-padd9-tec0117.fs dsp-padd18-tec0117.fs \
+	dsp-mult9x9-tec0117.fs dsp-alu54d-tec0117.fs dsp-multalu18x18-tec0117.fs \
+	dsp-multalu36x18-tec0117.fs dsp-multaddalu18x18-tec0117.fs \
 	\
 	blinky-runber.fs shift-runber.fs blinky-tbuf-runber.fs blinky-oddr-runber.fs \
 	blinky-osc-runber.fs tlvds-runber.fs elvds-runber.fs oddr-tlvds-runber.fs \
 	oddr-elvds-runber.fs blinky-pll-runber.fs \
 	oser4-runber.fs ovideo-runber.fs oser8-runber.fs oser10-runber.fs \
-	ides4-runber.fs ivideo-runber.fs ides8-runber.fs ides10-runber.fs 
+	ides4-runber.fs ivideo-runber.fs ides8-runber.fs ides10-runber.fs \
+	dsp-mult36x36-runber.fs dsp-padd9-runber.fs dsp-padd18-runber.fs \
+	dsp-mult9x9-runber.fs dsp-alu54d-runber.fs dsp-multalu18x18-runber.fs \
+	dsp-multalu36x18-runber.fs dsp-multaddalu18x18-runber.fs
 
 unpacked:\
 	blinky-tangnano20k-unpacked.v shift-tangnano20k-unpacked.v \
@@ -173,7 +196,8 @@ unpacked:\
 	elvds-runber-unpacked.v oddr-tlvds-runber-unpacked.v oddr-elvds-runber-unpacked.v \
 	blinky-pll-runber-unpacked.v oser4-runber-unpacked.v ovideo-runber-unpacked.v \
 	oser8-runber-unpacked.v oser10-runber-unpacked.v ides4-runber-unpacked.v \
-	ivideo-runber-unpacked.v ides8-runber-unpacked.v ides10-runber-unpacked.v 
+	ivideo-runber-unpacked.v ides8-runber-unpacked.v ides10-runber-unpacked.v
+
 clean: 
 	rm -f *.json *.fs *-unpacked.v
 
@@ -208,7 +232,7 @@ attosoc-tangnano20k-synth.json: attosoc/attosoc.v attosoc/picorv32.v
 	$(NEXTPNR) --json $< --write $@ --device GW2A-LV18PG256C8/I7 --vopt family=GW2A-18 --vopt cst=primer20k.cst
 
 %-primer20k-synth.json: %.v
-	$(YOSYS) -D LEDS_NR=6 -D OSC_TYPE_OSC -D INV_BTN=0 -p "read_verilog $^; synth_gowin -json $@"
+	$(YOSYS) -D LEDS_NR=6 -D OSC_TYPE_OSC -D INV_BTN=0 -D CPU_FREQ=27 -D BAUD_RATE=115200 -p "read_verilog $^; synth_gowin -json $@"
 
 pll-nanolcd-primer20k-synth.json: pll/GW2A-18-dyn.vh pll-nanolcd/TOP.v pll-nanolcd/VGAMod.v
 	$(YOSYS) -D INV_BTN=0 -p "read_verilog $^; synth_gowin -json $@"
@@ -299,7 +323,7 @@ bsram-%-tangnano9k-synth.json: pll/GW1N-9C-dyn.vh %-image-rom.v %-video-ram.v %.
 	$(NEXTPNR) --json $< --write $@ --device GW1NR-LV9LQ144PC6/I5 --vopt family=GW1N-9 --vopt cst=szfpga.cst
 
 %-szfpga-synth.json: %.v
-	$(YOSYS) -D LEDS_NR=4 -D OSC_TYPE_OSC -D INV_BTN=0 -p "read_verilog $^; synth_gowin -json $@"
+	$(YOSYS) -D LEDS_NR=4 -D OSC_TYPE_OSC -D INV_BTN=0 -D CPU_FREQ=50 -D BAUD_RATE=115200 -p "read_verilog $^; synth_gowin -json $@"
 
 blinky-pll-szfpga-synth.json: pll/GW1N-9-dyn.vh blinky-pll.v
 	$(YOSYS) -D INV_BTN=0 -D LEDS_NR=4 -p "read_verilog $^; synth_gowin -json $@"
@@ -316,7 +340,7 @@ bsram-%-szfpga-synth.json: pll/GW1N-9-dyn.vh %-image-rom.v %-video-ram.v %.v
 	$(NEXTPNR) --json $< --write $@ --device GW1NR-LV9QN88C6/I5 --vopt family=GW1N-9 --vopt cst=tec0117.cst
 
 %-tec0117-synth.json: %.v
-	$(YOSYS) -D LEDS_NR=8 -D OSC_TYPE_OSC -D INV_BTN=0 -p "read_verilog $^; synth_gowin -json $@"
+	$(YOSYS) -D LEDS_NR=8 -D OSC_TYPE_OSC -D INV_BTN=0 -D CPU_FREQ=12 -D BAUD_RATE=115200 -p "read_verilog $^; synth_gowin -json $@"
 
 blinky-pll-tec0117-synth.json: pll/GW1N-9-dyn.vh blinky-pll.v
 	$(YOSYS) -D INV_BTN=0 -D LEDS_NR=8 -p "read_verilog $^; synth_gowin -json $@"
@@ -330,7 +354,7 @@ blinky-pll-tec0117-synth.json: pll/GW1N-9-dyn.vh blinky-pll.v
 	$(NEXTPNR) --json $< --write $@ --device GW1N-UV4LQ144C6/I5 --vopt cst=runber.cst
 
 %-runber-synth.json: %.v
-	$(YOSYS) -D LEDS_NR=8 -D OSC_TYPE_OSC -D INV_BTN=0 -p "read_verilog $^; synth_gowin -json $@"
+	$(YOSYS) -D LEDS_NR=8 -D OSC_TYPE_OSC -D INV_BTN=0 -D FORCE_BRAM -D CPU_FREQ=12 -D BAUD_RATE=115200 -p "read_verilog $^; synth_gowin -json $@"
 
 blinky-pll-runber-synth.json: pll/GW1N-4-dyn.vh blinky-pll.v
 	$(YOSYS) -D INV_BTN=0 -D LEDS_NR=8 -p "read_verilog $^; synth_gowin -json $@"

diff --git a/examples/himbaechel/dsp-alu54d.v b/examples/himbaechel/dsp-alu54d.v
@@ -0,0 +1,101 @@
+`default_nettype none
+// alu0 mode 0 - simple substraction with accumulator in C
+// alu1 mode 1 - addition with CASI and accumulator in A
+// alu2 mode 2 - addition with CASI
+module idsp(input wire clk, input wire reset,
+	output wire [63:0] product, 
+	output wire [63:0] product1, 
+	output wire [63:0] product2, 
+	output wire [63:0] product3, 
+	output wire [63:0] product4
+	);
+
+wire [17:0] soa;
+wire [17:0] sob;
+wire [17:0] soa0;
+wire [17:0] sob0;
+wire [17:0] soa1;
+wire [17:0] sob1;
+wire [17:0] soa2;
+wire [17:0] sob2;
+wire gnd = 1'b0;
+
+wire [54:0]caso;
+wire [54:0]caso0;
+
+ALU54D alu0(
+	.A(54'hde1ec7ab1e),
+	.B(54'hcad),
+	.DOUT(product[53:0]),
+	.CASI(gnd),
+	.CASO(caso),
+	.ASIGN(gnd),
+	.BSIGN(gnd),
+	.ACCLOAD(1'b1),
+	.CLK(clk),
+	.CE(1'b1),
+	.RESET(reset)
+);
+	defparam alu0.AREG=1'b0;
+	defparam alu0.BREG=1'b0;
+	defparam alu0.ASIGN_REG=1'b0;
+	defparam alu0.BSIGN_REG=1'b0;
+	defparam alu0.ACCLOAD_REG=1'b0;
+	defparam alu0.OUT_REG=1'b1;
+	defparam alu0.B_ADD_SUB=1'b1;
+	defparam alu0.C_ADD_SUB=1'b0;
+	defparam alu0.ALUD_MODE=2'b0;
+	defparam alu0.ALU_RESET_MODE="SYNC";
+
+ALU54D alu1(
+	.A(54'h1111),
+	.B(54'h2),
+	.DOUT(product1[53:0]),
+	.CASI(caso),
+	.CASO(caso0),
+	.ASIGN(gnd),
+	.BSIGN(gnd),
+	.ACCLOAD(gnd),
+	.CLK(clk),
+	.CE(1'b1),
+	.RESET(reset)
+);
+	defparam alu1.AREG=1'b1;
+	defparam alu1.BREG=1'b0;
+	defparam alu1.ASIGN_REG=1'b0;
+	defparam alu1.BSIGN_REG=1'b0;
+	defparam alu1.ACCLOAD_REG=1'b0;
+	defparam alu1.OUT_REG=1'b0;
+	defparam alu1.B_ADD_SUB=1'b0;
+	defparam alu1.C_ADD_SUB=1'b0;
+	defparam alu1.ALUD_MODE=1;
+	defparam alu1.ALU_RESET_MODE="SYNC";
+
+ALU54D alu2(
+	.A(54'h100000000),
+	.B(54'h00000f000),
+	.DOUT(product2[53:0]),
+	.CASI(caso0),
+	.CASO(),
+	.ASIGN(gnd),
+	.BSIGN(gnd),
+	.ACCLOAD(gnd),
+	.CLK(clk),
+	.CE(1'b1),
+	.RESET(reset)
+);
+	defparam alu2.AREG=1'b1;
+	defparam alu2.BREG=1'b1;
+	defparam alu2.ASIGN_REG=1'b1;
+	defparam alu2.BSIGN_REG=1'b1;
+	defparam alu2.ACCLOAD_REG=1'b1;
+	defparam alu2.OUT_REG=1'b0;
+	defparam alu2.B_ADD_SUB=1'b0;
+	defparam alu2.C_ADD_SUB=1'b0;
+	defparam alu2.ALUD_MODE=2;
+	defparam alu2.ALU_RESET_MODE="SYNC";
+endmodule
+
+`define FIRMWARE "riscv-dsp-firmware/alu54d.hex"
+`include "dsp-riscv.v"
+