Merge pull request #1521 from dh73/diego/memattr

Adding support for Xilinx memory attribute 'block' in single port mode.

passes/memory/memory_bram.cc

@@ -134,6 +134,7 @@ struct rules_t
 		dict<string, int> min_limits, max_limits;
 		bool or_next_if_better, make_transp, make_outreg;
 		char shuffle_enable;
+		vector<vector<std::tuple<bool,IdString,Const>>> attributes;
 	};
 
 	dict<IdString, vector<bram_t>> brams;
@@ -327,6 +328,20 @@ struct rules_t
 				continue;
 			}
 
+			if (GetSize(tokens) >= 2 && tokens[0] == "attribute") {
+				data.attributes.emplace_back();
+				for (int idx = 1; idx <= GetSize(tokens)-1; idx++) {
+					size_t c1 = tokens[idx][0] == '!' ? 1 : 0;
+					size_t c2 = tokens[idx].find("=");
+					bool exists = (c1 == 0);
+					IdString key = RTLIL::escape_id(tokens[idx].substr(c1, c2));
+					Const val = c2 != std::string::npos ? tokens[idx].substr(c2+1) : RTLIL::Const(1);
+
+					data.attributes.back().emplace_back(exists, key, val);
+				}
+				continue;
+			}
+
 			syntax_error();
 		}
 	}
@@ -724,7 +739,7 @@ grow_read_ports:;
 					if (match.make_transp && wr_ports <= 1) {
 						pi.make_transp = true;
 						if (pi.clocks != 0) {
-							if (wr_ports == 1 && wr_clkdom != clkdom) {								
+							if (wr_ports == 1 && wr_clkdom != clkdom) {
 								log("        Bram port %c%d.%d cannot have soft transparency logic added as read and write clock domains differ.\n", pi.group + 'A', pi.index + 1, pi.dupidx + 1);
 								goto skip_bram_rport;
 							}
@@ -813,6 +828,45 @@ grow_read_ports:;
 			return false;
 		}
 
+		for (const auto &sums : match.attributes) {
+			bool found = false;
+			for (const auto &term : sums) {
+				bool exists = std::get<0>(term);
+				IdString key = std::get<1>(term);
+				const Const &value = std::get<2>(term);
+				auto it = cell->attributes.find(key);
+				if (it == cell->attributes.end()) {
+					if (exists)
+						continue;
+					found = true;
+					break;
+				}
+				else if (!exists)
+					continue;
+				if (it->second != value)
+					continue;
+				found = true;
+				break;
+			}
+			if (!found) {
+				std::stringstream ss;
+				bool exists = std::get<0>(sums.front());
+				if (!exists)
+					ss << "!";
+				IdString key = std::get<1>(sums.front());
+				ss << key.str();
+				const Const &value = std::get<2>(sums.front());
+				if (exists)
+					ss << "=";
+				if (value != Const(1))
+					ss << "\"" << value.decode_string() << "\"";
+
+				log("    Rule for bram type %s rejected: requirement 'attribute %s ...' not met.\n",
+						log_id(match.name), ss.str().c_str());
+				return false;
+			}
+		}
+
 		if (mode == 1)
 			return true;
 	}
@@ -1100,6 +1154,45 @@ void handle_cell(Cell *cell, const rules_t &rules)
 				goto next_match_rule;
 			}
 
+			for (const auto &sums : match.attributes) {
+				bool found = false;
+				for (const auto &term : sums) {
+					bool exists = std::get<0>(term);
+					IdString key = std::get<1>(term);
+					const Const &value = std::get<2>(term);
+					auto it = cell->attributes.find(key);
+					if (it == cell->attributes.end()) {
+						if (exists)
+							continue;
+						found = true;
+						break;
+					}
+					else if (!exists)
+						continue;
+					if (it->second != value)
+						continue;
+					found = true;
+					break;
+				}
+				if (!found) {
+					std::stringstream ss;
+					bool exists = std::get<0>(sums.front());
+					if (!exists)
+						ss << "!";
+					IdString key = std::get<1>(sums.front());
+					ss << key.str();
+					const Const &value = std::get<2>(sums.front());
+					if (exists)
+						ss << "=";
+					if (value != Const(1))
+						ss << "\"" << value.decode_string() << "\"";
+
+					log("    Rule for bram type %s (variant %d) rejected: requirement 'attribute %s ...' not met.\n",
+							log_id(bram.name), bram.variant, ss.str().c_str());
+					goto next_match_rule;
+				}
+			}
+
 			log("    Rule #%d for bram type %s (variant %d) accepted.\n", i+1, log_id(bram.name), bram.variant);
 
 			if (or_next_if_better || !best_rule_cache.empty())
@@ -1225,6 +1318,13 @@ struct MemoryBramPass : public Pass {
 		log("    dcells  .......  number of cells in 'data-direction'\n");
 		log("    cells  ........  total number of cells (acells*dcells*dups)\n");
 		log("\n");
+		log("A match containing the command 'attribute' followed by a list of space\n");
+		log("separated 'name[=string_value]' values requires that the memory contains any\n");
+		log("one of the given attribute name and string values (where specified), or name\n");
+		log("and integer 1 value (if no string_value given, since Verilog will interpret\n");
+		log("'(* attr *)' as '(* attr=1 *)').\n");
+		log("A name prefixed with '!' indicates that the attribute must not exist.\n");
+		log("\n");
 		log("The interface for the created bram instances is derived from the bram\n");
 		log("description. Use 'techmap' to convert the created bram instances into\n");
 		log("instances of the actual bram cells of your target architecture.\n");

techlibs/xilinx/xc7_xcu_brams.txt

@@ -77,6 +77,26 @@ endbram
 # https://www.xilinx.com/support/documentation/user_guides/ug473_7Series_Memory_Resources.pdf
 
 match $__XILINX_RAMB36_SDP
+  attribute !ram_style
+  attribute !logic_block
+  min bits 1024
+  min efficiency 5
+  shuffle_enable B
+  make_transp
+  or_next_if_better
+endmatch
+
+match $__XILINX_RAMB36_SDP
+  attribute ram_style=block ram_block
+  attribute !logic_block
+  shuffle_enable B
+  make_transp
+  or_next_if_better
+endmatch
+
+match $__XILINX_RAMB18_SDP
+  attribute !ram_style
+  attribute !logic_block
   min bits 1024
   min efficiency 5
   shuffle_enable B
@@ -85,6 +105,16 @@ match $__XILINX_RAMB36_SDP
 endmatch
 
 match $__XILINX_RAMB18_SDP
+  attribute ram_style=block ram_block
+  attribute !logic_block
+  shuffle_enable B
+  make_transp
+  or_next_if_better
+endmatch
+
+match $__XILINX_RAMB36_TDP
+  attribute !ram_style
+  attribute !logic_block
   min bits 1024
   min efficiency 5
   shuffle_enable B
@@ -93,6 +123,16 @@ match $__XILINX_RAMB18_SDP
 endmatch
 
 match $__XILINX_RAMB36_TDP
+  attribute ram_style=block ram_block
+  attribute !logic_block
+  shuffle_enable B
+  make_transp
+  or_next_if_better
+endmatch
+
+match $__XILINX_RAMB18_TDP
+  attribute !ram_style
+  attribute !logic_block
   min bits 1024
   min efficiency 5
   shuffle_enable B
@@ -101,8 +141,9 @@ match $__XILINX_RAMB36_TDP
 endmatch
 
 match $__XILINX_RAMB18_TDP
-  min bits 1024
+  attribute ram_style=block ram_block
-  min efficiency 5
+  attribute !logic_block
   shuffle_enable B
   make_transp
 endmatch
+

tests/arch/common/memory_attributes/attributes_test.v

@@ -0,0 +1,88 @@
+`default_nettype none
+module block_ram #(parameter DATA_WIDTH=4, ADDRESS_WIDTH=10)
+   (input  wire                      write_enable, clk,
+    input  wire  [DATA_WIDTH-1:0]    data_in,
+    input  wire  [ADDRESS_WIDTH-1:0] address_in,
+    output wire  [DATA_WIDTH-1:0]    data_out);
+
+   localparam WORD  = (DATA_WIDTH-1);
+   localparam DEPTH = (2**ADDRESS_WIDTH-1);
+
+   reg [WORD:0] data_out_r;
+   reg [WORD:0] memory [0:DEPTH];
+
+   always @(posedge clk) begin
+      if (write_enable)
+        memory[address_in] <= data_in;
+      data_out_r <= memory[address_in];
+   end
+
+   assign data_out = data_out_r;
+endmodule // block_ram
+
+`default_nettype none
+module distributed_ram #(parameter DATA_WIDTH=8, ADDRESS_WIDTH=4)
+   (input  wire                      write_enable, clk,
+    input  wire  [DATA_WIDTH-1:0]    data_in,
+    input  wire  [ADDRESS_WIDTH-1:0] address_in,
+    output wire  [DATA_WIDTH-1:0]    data_out);
+
+   localparam WORD  = (DATA_WIDTH-1);
+   localparam DEPTH = (2**ADDRESS_WIDTH-1);
+
+   reg [WORD:0] data_out_r;
+   reg [WORD:0] memory [0:DEPTH];
+
+   always @(posedge clk) begin
+      if (write_enable)
+        memory[address_in] <= data_in;
+      data_out_r <= memory[address_in];
+   end
+
+   assign data_out = data_out_r;
+endmodule // distributed_ram
+
+`default_nettype none
+module distributed_ram_manual #(parameter DATA_WIDTH=8, ADDRESS_WIDTH=4)
+   (input  wire                      write_enable, clk,
+    input  wire  [DATA_WIDTH-1:0]    data_in,
+    input  wire  [ADDRESS_WIDTH-1:0] address_in,
+    output wire  [DATA_WIDTH-1:0]    data_out);
+
+   localparam WORD  = (DATA_WIDTH-1);
+   localparam DEPTH = (2**ADDRESS_WIDTH-1);
+
+   reg [WORD:0] data_out_r;
+   (* ram_style = "block" *) reg [WORD:0] memory [0:DEPTH];
+
+   always @(posedge clk) begin
+      if (write_enable)
+        memory[address_in] <= data_in;
+      data_out_r <= memory[address_in];
+   end
+
+   assign data_out = data_out_r;
+endmodule // distributed_ram
+
+`default_nettype none
+module distributed_ram_manual_syn #(parameter DATA_WIDTH=8, ADDRESS_WIDTH=4)
+   (input  wire                      write_enable, clk,
+    input  wire  [DATA_WIDTH-1:0]    data_in,
+    input  wire  [ADDRESS_WIDTH-1:0] address_in,
+    output wire  [DATA_WIDTH-1:0]    data_out);
+
+   localparam WORD  = (DATA_WIDTH-1);
+   localparam DEPTH = (2**ADDRESS_WIDTH-1);
+
+   reg [WORD:0] data_out_r;
+   (* synthesis, ram_block *) reg [WORD:0] memory [0:DEPTH];
+
+   always @(posedge clk) begin
+      if (write_enable)
+        memory[address_in] <= data_in;
+      data_out_r <= memory[address_in];
+   end
+
+   assign data_out = data_out_r;
+endmodule // distributed_ram
+

tests/arch/xilinx/attributes_test.ys

@@ -0,0 +1,47 @@
+# Check that blockram memory without parameters is not modified
+read_verilog ../common/memory_attributes/attributes_test.v
+hierarchy -top block_ram
+synth_xilinx -top block_ram
+cd block_ram # Constrain all select calls below inside the top module
+select -assert-count 1 t:RAMB18E1
+ 
+# Check that distributed memory without parameters is not modified
+design -reset
+read_verilog ../common/memory_attributes/attributes_test.v
+hierarchy -top distributed_ram
+synth_xilinx -top distributed_ram
+cd distributed_ram # Constrain all select calls below inside the top module
+select -assert-count 8 t:RAM32X1D
+ 
+# Set ram_style distributed to blockram memory; will be implemented as distributed
+design -reset
+read_verilog ../common/memory_attributes/attributes_test.v
+prep
+setattr -mod -set ram_style "distributed" block_ram
+synth_xilinx -top block_ram
+cd block_ram # Constrain all select calls below inside the top module
+select -assert-count 32 t:RAM128X1D
+ 
+# Set synthesis, logic_block to blockram memory; will be implemented as distributed
+design -reset
+read_verilog ../common/memory_attributes/attributes_test.v
+prep
+setattr -mod -set logic_block 1 block_ram
+synth_xilinx -top block_ram
+cd block_ram # Constrain all select calls below inside the top module
+select -assert-count 0 t:RAMB18E1
+select -assert-count 32 t:RAM128X1D
+ 
+# Set ram_style block to a distributed memory; will be implemented as blockram
+design -reset
+read_verilog ../common/memory_attributes/attributes_test.v
+synth_xilinx -top distributed_ram_manual
+cd distributed_ram_manual # Constrain all select calls below inside the top module
+select -assert-count 1 t:RAMB18E1
+ 
+# Set synthesis, ram_block block to a distributed memory; will be implemented as blockram
+design -reset
+read_verilog ../common/memory_attributes/attributes_test.v
+synth_xilinx -top distributed_ram_manual_syn
+cd distributed_ram_manual_syn # Constrain all select calls below inside the top module
+select -assert-count 1 t:RAMB18E1

tests/arch/xilinx/blockram.ys

@@ -0,0 +1,97 @@
+### TODO: Not running equivalence checking because BRAM models does not exists
+###       currently. Checking instance counts instead.
+# Memory bits <= 18K; Data width <= 36; Address width <= 14: -> RAMB18E1
+read_verilog ../common/blockram.v
+chparam -set ADDRESS_WIDTH 10 -set DATA_WIDTH 1 sync_ram_sdp
+synth_xilinx -top sync_ram_sdp
+cd sync_ram_sdp
+select -assert-count 1 t:RAMB18E1
+
+design -reset
+read_verilog ../common/blockram.v
+chparam -set ADDRESS_WIDTH 8 -set DATA_WIDTH 18 sync_ram_sdp
+synth_xilinx -top sync_ram_sdp
+cd sync_ram_sdp
+select -assert-count 1 t:RAMB18E1
+
+design -reset
+read_verilog ../common/blockram.v
+chparam -set ADDRESS_WIDTH 14 -set DATA_WIDTH 1 sync_ram_sdp
+synth_xilinx -top sync_ram_sdp
+cd sync_ram_sdp
+select -assert-count 1 t:RAMB18E1
+
+design -reset
+read_verilog ../common/blockram.v
+chparam -set ADDRESS_WIDTH 9 -set DATA_WIDTH 36 sync_ram_sdp
+synth_xilinx -top sync_ram_sdp
+cd sync_ram_sdp
+select -assert-count 1 t:RAMB18E1
+
+# Anything memory bits < 1024 -> LUTRAM
+design -reset
+read_verilog ../common/blockram.v
+chparam -set ADDRESS_WIDTH 8 -set DATA_WIDTH 2 sync_ram_sdp
+synth_xilinx -top sync_ram_sdp
+cd sync_ram_sdp
+select -assert-count 0 t:RAMB18E1
+select -assert-count 4 t:RAM128X1D
+
+# More than 18K bits, data width <= 36 (TDP), and address width from 10 to 15b (non-cascaded) -> RAMB36E1
+design -reset
+read_verilog ../common/blockram.v
+chparam -set ADDRESS_WIDTH 10 -set DATA_WIDTH 36 sync_ram_sdp
+synth_xilinx -top sync_ram_sdp
+cd sync_ram_sdp
+select -assert-count 1 t:RAMB36E1
+
+
+### With parameters
+
+design -reset
+read_verilog ../common/blockram.v
+hierarchy -top sync_ram_sdp -chparam ADDRESS_WIDTH 10 -chparam DATA_WIDTH 1 
+setattr -set ram_style "block" m:memory
+synth_xilinx -top sync_ram_sdp
+cd sync_ram_sdp
+select -assert-count 1 t:RAMB18E1
+
+design -reset
+read_verilog ../common/blockram.v
+hierarchy -top sync_ram_sdp -chparam ADDRESS_WIDTH 10 -chparam DATA_WIDTH 1 
+setattr -set ram_block 1 m:memory
+synth_xilinx -top sync_ram_sdp
+cd sync_ram_sdp
+select -assert-count 1 t:RAMB18E1
+
+design -reset
+read_verilog ../common/blockram.v
+hierarchy -top sync_ram_sdp -chparam ADDRESS_WIDTH 10 -chparam DATA_WIDTH 1 
+setattr -set ram_style "dont_infer_a_ram_pretty_please" m:memory
+synth_xilinx -top sync_ram_sdp
+cd sync_ram_sdp
+select -assert-count 0 t:RAMB18E1
+
+design -reset
+read_verilog ../common/blockram.v
+hierarchy -top sync_ram_sdp -chparam ADDRESS_WIDTH 10 -chparam DATA_WIDTH 1 
+setattr -set logic_block 1 m:memory
+synth_xilinx -top sync_ram_sdp
+cd sync_ram_sdp
+select -assert-count 0 t:RAMB18E1
+
+design -reset
+read_verilog ../common/blockram.v
+hierarchy -top sync_ram_sdp -chparam ADDRESS_WIDTH 8 -chparam DATA_WIDTH 1
+setattr -set ram_style "block" m:memory
+synth_xilinx -top sync_ram_sdp
+cd sync_ram_sdp
+select -assert-count 1 t:RAMB18E1
+
+design -reset
+read_verilog ../common/blockram.v
+hierarchy -top sync_ram_sdp -chparam ADDRESS_WIDTH 8 -chparam DATA_WIDTH 1
+setattr -set ram_block 1 m:memory
+synth_xilinx -top sync_ram_sdp
+cd sync_ram_sdp
+select -assert-count 1 t:RAMB18E1

tests/arch/xilinx/blockram_params.ys

@@ -1,47 +0,0 @@
-## TODO: Not running equivalence checking because BRAM models does not exists
-##       currently. Checking instance counts instead.
-# Memory bits <= 18K; Data width <= 36; Address width <= 14: -> RAMB18E1
-read_verilog ../common/blockram_params.v
-chparam -set ADDRESS_WIDTH 10 -set DATA_WIDTH 1 sync_ram_sdp
-synth_xilinx -top sync_ram_sdp
-cd sync_ram_sdp
-select -assert-count 1 t:RAMB18E1
-
-design -reset
-read_verilog ../common/blockram_params.v
-chparam -set ADDRESS_WIDTH 8 -set DATA_WIDTH 18 sync_ram_sdp
-synth_xilinx -top sync_ram_sdp
-cd sync_ram_sdp
-select -assert-count 1 t:RAMB18E1
-
-design -reset
-read_verilog ../common/blockram_params.v
-chparam -set ADDRESS_WIDTH 14 -set DATA_WIDTH 1 sync_ram_sdp
-synth_xilinx -top sync_ram_sdp
-cd sync_ram_sdp
-select -assert-count 1 t:RAMB18E1
-
-design -reset
-read_verilog ../common/blockram_params.v
-chparam -set ADDRESS_WIDTH 9 -set DATA_WIDTH 36 sync_ram_sdp
-synth_xilinx -top sync_ram_sdp
-cd sync_ram_sdp
-select -assert-count 1 t:RAMB18E1
-
-# Anything memory bits < 1024 -> LUTRAM
-design -reset
-read_verilog ../common/blockram_params.v
-chparam -set ADDRESS_WIDTH 8 -set DATA_WIDTH 2 sync_ram_sdp
-synth_xilinx -top sync_ram_sdp
-cd sync_ram_sdp
-select -assert-count 0 t:RAMB18E1
-select -assert-count 4 t:RAM128X1D
-
-# More than 18K bits, data width <= 36 (TDP), and address width from 10 to 15b (non-cascaded) -> RAMB36E1 
-design -reset
-read_verilog ../common/blockram_params.v
-chparam -set ADDRESS_WIDTH 10 -set DATA_WIDTH 36 sync_ram_sdp
-synth_xilinx -top sync_ram_sdp
-cd sync_ram_sdp
-select -assert-count 1 t:RAMB36E1
-