// ================================================================ // NVDLA Open Source Project // // Copyright(c) 2016 - 2017 NVIDIA Corporation. Licensed under the // NVDLA Open Hardware License; Check "LICENSE" which comes with // this distribution for more information. // ================================================================ // File Name: NV_NVDLA_RUBIK_fifo.v `define FORCE_CONTENTION_ASSERTION_RESET_ACTIVE 1'b1 `include "simulate_x_tick.vh" module NV_NVDLA_RUBIK_fifo ( nvdla_core_clk , nvdla_core_rstn , idata_prdy , idata_pvld `ifdef FV_RAND_WR_PAUSE , idata_pause `endif , idata_pd , odata_prdy , odata_pvld , odata_pd , pwrbus_ram_pd ); // spyglass disable_block W401 -- clock is not input to module input nvdla_core_clk; input nvdla_core_rstn; output idata_prdy; input idata_pvld; `ifdef FV_RAND_WR_PAUSE input idata_pause; `endif input [255:0] idata_pd; input odata_prdy; output odata_pvld; output [255:0] odata_pd; input [31:0] pwrbus_ram_pd; // Master Clock Gating (SLCG) // // We gate the clock(s) when idle or stalled. // This allows us to turn off numerous miscellaneous flops // that don't get gated during synthesis for one reason or another. // // We gate write side and read side separately. // If the fifo is synchronous, we also gate the ram separately, but if // -master_clk_gated_unified or -status_reg/-status_logic_reg is specified, // then we use one clk gate for write, ram, and read. // wire nvdla_core_clk_mgated_enable; // assigned by code at end of this module wire nvdla_core_clk_mgated; // used only in synchronous fifos NV_CLK_gate_power nvdla_core_clk_mgate( .clk(nvdla_core_clk), .reset_(nvdla_core_rstn), .clk_en(nvdla_core_clk_mgated_enable), .clk_gated(nvdla_core_clk_mgated) ); // // WRITE SIDE // // synopsys translate_off `ifndef SYNTH_LEVEL1_COMPILE `ifndef SYNTHESIS wire wr_pause_rand; // random stalling `endif `endif // synopsys translate_on wire wr_reserving; reg idata_pvld_in; // registered idata_pvld reg [255:0] idata_pd_in; // registered idata_pd reg wr_busy_in; // inputs being held this cycle? assign idata_prdy = !wr_busy_in; wire idata_busy_next; // fwd: fifo busy next? // factor for better timing with distant idata_pvld signal wire wr_busy_in_next_wr_req_eq_1 = idata_busy_next; wire wr_busy_in_next_wr_req_eq_0 = (idata_pvld_in && idata_busy_next) && !wr_reserving; `ifdef FV_RAND_WR_PAUSE wire wr_busy_in_next = (idata_pvld? wr_busy_in_next_wr_req_eq_1 : wr_busy_in_next_wr_req_eq_0) || idata_pause ; `else wire wr_busy_in_next = (idata_pvld? wr_busy_in_next_wr_req_eq_1 : wr_busy_in_next_wr_req_eq_0) // synopsys translate_off `ifndef SYNTH_LEVEL1_COMPILE `ifndef SYNTHESIS || wr_pause_rand `endif `endif // synopsys translate_on ; `endif wire wr_busy_in_int; always @( posedge nvdla_core_clk or negedge nvdla_core_rstn ) begin if ( !nvdla_core_rstn ) begin idata_pvld_in <= 1'b0; wr_busy_in <= 1'b0; end else begin wr_busy_in <= wr_busy_in_next; if ( !wr_busy_in_int ) begin idata_pvld_in <= idata_pvld && !wr_busy_in; end //synopsys translate_off else if ( wr_busy_in_int ) begin end else begin idata_pvld_in <= `x_or_0; end //synopsys translate_on end end always @( posedge nvdla_core_clk ) begin if ( !wr_busy_in && idata_pvld ) begin idata_pd_in <= idata_pd; end //synopsys translate_off else if ( !(!wr_busy_in && idata_pvld) ) begin end else begin idata_pd_in <= {256{`x_or_0}}; end //synopsys translate_on end reg idata_busy_int; // copy for internal use assign wr_reserving = idata_pvld_in && !idata_busy_int; // reserving write space? reg wr_popping; // fwd: write side sees pop? reg [6:0] idata_count; // write-side count wire [6:0] wr_count_next_wr_popping = wr_reserving ? idata_count : (idata_count - 1'd1); // spyglass disable W164a W484 wire [6:0] wr_count_next_no_wr_popping = wr_reserving ? (idata_count + 1'd1) : idata_count; // spyglass disable W164a W484 wire [6:0] wr_count_next = wr_popping ? wr_count_next_wr_popping : wr_count_next_no_wr_popping; wire wr_count_next_no_wr_popping_is_80 = ( wr_count_next_no_wr_popping == 7'd80 ); wire wr_count_next_is_80 = wr_popping ? 1'b0 : wr_count_next_no_wr_popping_is_80; wire [6:0] wr_limit_muxed; // muxed with simulation/emulation overrides wire [6:0] wr_limit_reg = wr_limit_muxed; // VCS coverage off assign idata_busy_next = wr_count_next_is_80 || // busy next cycle? (wr_limit_reg != 7'd0 && // check idata_limit if != 0 wr_count_next >= wr_limit_reg) ; // VCS coverage on assign wr_busy_in_int = idata_pvld_in && idata_busy_int; always @( posedge nvdla_core_clk_mgated or negedge nvdla_core_rstn ) begin if ( !nvdla_core_rstn ) begin idata_busy_int <= 1'b0; idata_count <= 7'd0; end else begin idata_busy_int <= idata_busy_next; if ( wr_reserving ^ wr_popping ) begin idata_count <= wr_count_next; end //synopsys translate_off else if ( !(wr_reserving ^ wr_popping) ) begin end else begin idata_count <= {7{`x_or_0}}; end //synopsys translate_on end end wire wr_pushing = wr_reserving; // data pushed same cycle as idata_pvld_in // // RAM // reg [6:0] idata_adr; // current write address wire [6:0] odata_adr_p; // read address to use for ram wire [255:0] odata_pd_p; // read data directly out of ram wire rd_enable; wire ore; wire [31 : 0] pwrbus_ram_pd; // Adding parameter for fifogen to disable wr/rd contention assertion in ramgen. // Fifogen handles this by ignoring the data on the ram data out for that cycle. nv_ram_rwsp_80x256 #(`FORCE_CONTENTION_ASSERTION_RESET_ACTIVE) ram ( .clk ( nvdla_core_clk ) , .pwrbus_ram_pd ( pwrbus_ram_pd ) , .wa ( idata_adr ) , .we ( wr_pushing ) , .di ( idata_pd_in ) , .ra ( odata_adr_p ) , .re ( rd_enable ) , .dout ( odata_pd_p ) , .ore ( ore ) ); // next idata_adr if wr_pushing=1 wire [6:0] wr_adr_next = (idata_adr == 7'd79) ? 7'd0 : (idata_adr + 1'd1); // spyglass disable W484 // spyglass disable_block W484 always @( posedge nvdla_core_clk_mgated or negedge nvdla_core_rstn ) begin if ( !nvdla_core_rstn ) begin idata_adr <= 7'd0; end else begin if ( wr_pushing ) begin idata_adr <= wr_adr_next; end //synopsys translate_off else if ( !(wr_pushing) ) begin end else begin idata_adr <= {7{`x_or_0}}; end //synopsys translate_on end end // spyglass enable_block W484 wire rd_popping; // read side doing pop this cycle? reg [6:0] odata_adr; // current read address // next read address wire [6:0] rd_adr_next = (odata_adr == 7'd79) ? 7'd0 : (odata_adr + 1'd1); // spyglass disable W484 assign odata_adr_p = rd_popping ? rd_adr_next : odata_adr; // for ram // spyglass disable_block W484 always @( posedge nvdla_core_clk_mgated or negedge nvdla_core_rstn ) begin if ( !nvdla_core_rstn ) begin odata_adr <= 7'd0; end else begin if ( rd_popping ) begin odata_adr <= rd_adr_next; end //synopsys translate_off else if ( !rd_popping ) begin end else begin odata_adr <= {7{`x_or_0}}; end //synopsys translate_on end end // spyglass enable_block W484 // // SYNCHRONOUS BOUNDARY // always @( posedge nvdla_core_clk_mgated or negedge nvdla_core_rstn ) begin if ( !nvdla_core_rstn ) begin wr_popping <= 1'b0; end else begin wr_popping <= rd_popping; end end reg rd_pushing; always @( posedge nvdla_core_clk_mgated or negedge nvdla_core_rstn ) begin if ( !nvdla_core_rstn ) begin rd_pushing <= 1'b0; end else begin rd_pushing <= wr_pushing; // let data go into ram first end end // // READ SIDE // reg odata_pvld_p; // data out of fifo is valid reg odata_pvld_int; // internal copy of odata_pvld assign odata_pvld = odata_pvld_int; assign rd_popping = odata_pvld_p && !(odata_pvld_int && !odata_prdy); reg [6:0] odata_count_p; // read-side fifo count // spyglass disable_block W164a W484 wire [6:0] rd_count_p_next_rd_popping = rd_pushing ? odata_count_p : (odata_count_p - 1'd1); wire [6:0] rd_count_p_next_no_rd_popping = rd_pushing ? (odata_count_p + 1'd1) : odata_count_p; // spyglass enable_block W164a W484 wire [6:0] rd_count_p_next = rd_popping ? rd_count_p_next_rd_popping : rd_count_p_next_no_rd_popping; wire rd_count_p_next_rd_popping_not_0 = rd_count_p_next_rd_popping != 0; wire rd_count_p_next_no_rd_popping_not_0 = rd_count_p_next_no_rd_popping != 0; wire rd_count_p_next_not_0 = rd_popping ? rd_count_p_next_rd_popping_not_0 : rd_count_p_next_no_rd_popping_not_0; assign rd_enable = ((rd_count_p_next_not_0) && ((~odata_pvld_p) || rd_popping)); // anytime data's there and not stalled always @( posedge nvdla_core_clk_mgated or negedge nvdla_core_rstn ) begin if ( !nvdla_core_rstn ) begin odata_count_p <= 7'd0; odata_pvld_p <= 1'b0; end else begin if ( rd_pushing || rd_popping ) begin odata_count_p <= rd_count_p_next; end //synopsys translate_off else if ( !(rd_pushing || rd_popping ) ) begin end else begin odata_count_p <= {7{`x_or_0}}; end //synopsys translate_on if ( rd_pushing || rd_popping ) begin odata_pvld_p <= (rd_count_p_next_not_0); end //synopsys translate_off else if ( !(rd_pushing || rd_popping ) ) begin end else begin odata_pvld_p <= `x_or_0; end //synopsys translate_on end end wire rd_req_next = (odata_pvld_p || (odata_pvld_int && !odata_prdy)) ; always @( posedge nvdla_core_clk_mgated or negedge nvdla_core_rstn ) begin if ( !nvdla_core_rstn ) begin odata_pvld_int <= 1'b0; end else begin odata_pvld_int <= rd_req_next; end end assign odata_pd = odata_pd_p; assign ore = rd_popping; // Master Clock Gating (SLCG) Enables // // plusarg for disabling this stuff: // synopsys translate_off `ifndef SYNTH_LEVEL1_COMPILE `ifndef SYNTHESIS reg master_clk_gating_disabled; initial master_clk_gating_disabled = $test$plusargs( "fifogen_disable_master_clk_gating" ) != 0; `endif `endif // synopsys translate_on // synopsys translate_off `ifndef SYNTH_LEVEL1_COMPILE `ifndef SYNTHESIS reg wr_pause_rand_dly; always @( posedge nvdla_core_clk or negedge nvdla_core_rstn ) begin if ( !nvdla_core_rstn ) begin wr_pause_rand_dly <= 1'b0; end else begin wr_pause_rand_dly <= wr_pause_rand; end end `endif `endif // synopsys translate_on assign nvdla_core_clk_mgated_enable = ((wr_reserving || wr_pushing || rd_popping || wr_popping || (idata_pvld_in && !idata_busy_int) || (idata_busy_int != idata_busy_next)) || (rd_pushing || rd_popping || (odata_pvld_int && odata_prdy) || wr_pushing)) `ifdef FIFOGEN_MASTER_CLK_GATING_DISABLED || 1'b1 `endif // synopsys translate_off `ifndef SYNTH_LEVEL1_COMPILE `ifndef SYNTHESIS || master_clk_gating_disabled || (wr_pause_rand != wr_pause_rand_dly) `endif `endif // synopsys translate_on ; // Simulation and Emulation Overrides of wr_limit(s) // `ifdef EMU `ifdef EMU_FIFO_CFG // Emulation Global Config Override // assign wr_limit_muxed = `EMU_FIFO_CFG.NV_NVDLA_RUBIK_fifo_wr_limit_override ? `EMU_FIFO_CFG.NV_NVDLA_RUBIK_fifo_wr_limit : 7'd0; `else // No Global Override for Emulation // assign wr_limit_muxed = 7'd0; `endif // EMU_FIFO_CFG `else // !EMU `ifdef SYNTH_LEVEL1_COMPILE // No Override for GCS Compiles // assign wr_limit_muxed = 7'd0; `else `ifdef SYNTHESIS // No Override for RTL Synthesis // assign wr_limit_muxed = 7'd0; `else // RTL Simulation Plusarg Override // VCS coverage off reg wr_limit_override; reg [6:0] wr_limit_override_value; assign wr_limit_muxed = wr_limit_override ? wr_limit_override_value : 7'd0; `ifdef NV_ARCHPRO event reinit; initial begin $display("fifogen reinit initial block %m"); -> reinit; end `endif `ifdef NV_ARCHPRO always @( reinit ) begin `else initial begin `endif wr_limit_override = 0; wr_limit_override_value = 0; // to keep viva happy with dangles if ( $test$plusargs( "NV_NVDLA_RUBIK_fifo_wr_limit" ) ) begin wr_limit_override = 1; $value$plusargs( "NV_NVDLA_RUBIK_fifo_wr_limit=%d", wr_limit_override_value); end end // VCS coverage on `endif `endif `endif // Random Write-Side Stalling // synopsys translate_off `ifndef SYNTH_LEVEL1_COMPILE `ifndef SYNTHESIS // VCS coverage off // leda W339 OFF -- Non synthesizable operator // leda W372 OFF -- Undefined PLI task // leda W373 OFF -- Undefined PLI function // leda W599 OFF -- This construct is not supported by Synopsys // leda W430 OFF -- Initial statement is not synthesizable // leda W182 OFF -- Illegal statement for synthesis // leda W639 OFF -- For synthesis, operands of a division or modulo operation need to be constants // leda DCVER_274_NV OFF -- This system task is not supported by DC integer stall_probability; // prob of stalling integer stall_cycles_min; // min cycles to stall integer stall_cycles_max; // max cycles to stall integer stall_cycles_left; // stall cycles left `ifdef NV_ARCHPRO always @( reinit ) begin `else initial begin `endif stall_probability = 0; // no stalling by default stall_cycles_min = 1; stall_cycles_max = 10; `ifdef NO_PLI `else if ( $test$plusargs( "NV_NVDLA_RUBIK_fifo_fifo_stall_probability" ) ) begin $value$plusargs( "NV_NVDLA_RUBIK_fifo_fifo_stall_probability=%d", stall_probability); end else if ( $test$plusargs( "default_fifo_stall_probability" ) ) begin $value$plusargs( "default_fifo_stall_probability=%d", stall_probability); end if ( $test$plusargs( "NV_NVDLA_RUBIK_fifo_fifo_stall_cycles_min" ) ) begin $value$plusargs( "NV_NVDLA_RUBIK_fifo_fifo_stall_cycles_min=%d", stall_cycles_min); end else if ( $test$plusargs( "default_fifo_stall_cycles_min" ) ) begin $value$plusargs( "default_fifo_stall_cycles_min=%d", stall_cycles_min); end if ( $test$plusargs( "NV_NVDLA_RUBIK_fifo_fifo_stall_cycles_max" ) ) begin $value$plusargs( "NV_NVDLA_RUBIK_fifo_fifo_stall_cycles_max=%d", stall_cycles_max); end else if ( $test$plusargs( "default_fifo_stall_cycles_max" ) ) begin $value$plusargs( "default_fifo_stall_cycles_max=%d", stall_cycles_max); end `endif if ( stall_cycles_min < 1 ) begin stall_cycles_min = 1; end if ( stall_cycles_min > stall_cycles_max ) begin stall_cycles_max = stall_cycles_min; end end `ifdef NO_PLI `else // randomization globals `ifdef SIMTOP_RANDOMIZE_STALLS always @( `SIMTOP_RANDOMIZE_STALLS.global_stall_event ) begin if ( ! $test$plusargs( "NV_NVDLA_RUBIK_fifo_fifo_stall_probability" ) ) stall_probability = `SIMTOP_RANDOMIZE_STALLS.global_stall_fifo_probability; if ( ! $test$plusargs( "NV_NVDLA_RUBIK_fifo_fifo_stall_cycles_min" ) ) stall_cycles_min = `SIMTOP_RANDOMIZE_STALLS.global_stall_fifo_cycles_min; if ( ! $test$plusargs( "NV_NVDLA_RUBIK_fifo_fifo_stall_cycles_max" ) ) stall_cycles_max = `SIMTOP_RANDOMIZE_STALLS.global_stall_fifo_cycles_max; end `endif `endif always @( negedge nvdla_core_clk or negedge nvdla_core_rstn ) begin if ( !nvdla_core_rstn ) begin stall_cycles_left <= 0; end else begin `ifdef NO_PLI stall_cycles_left <= 0; `else if ( idata_pvld && !(!idata_prdy) && stall_probability != 0 ) begin if ( prand_inst0(1, 100) <= stall_probability ) begin stall_cycles_left <= prand_inst1(stall_cycles_min, stall_cycles_max); end else if ( stall_cycles_left !== 0 ) begin stall_cycles_left <= stall_cycles_left - 1; end end else if ( stall_cycles_left !== 0 ) begin stall_cycles_left <= stall_cycles_left - 1; end `endif end end assign wr_pause_rand = (stall_cycles_left !== 0) ; // VCS coverage on `endif `endif // synopsys translate_on // VCS coverage on // leda W339 ON // leda W372 ON // leda W373 ON // leda W599 ON // leda W430 ON // leda W182 ON // leda W639 ON // leda DCVER_274_NV ON // // Histogram of fifo depth (from write side's perspective) // // NOTE: it will reference `SIMTOP.perfmon_enabled, so that // has to at least be defined, though not initialized. // tbgen testbenches have it already and various // ways to turn it on and off. // `ifdef PERFMON_HISTOGRAM // synopsys translate_off `ifndef SYNTH_LEVEL1_COMPILE `ifndef SYNTHESIS perfmon_histogram perfmon ( .clk ( nvdla_core_clk ) , .max ( {25'd0, (wr_limit_reg == 7'd0) ? 7'd80 : wr_limit_reg} ) , .curr ( {25'd0, idata_count} ) ); `endif `endif // synopsys translate_on `endif // spyglass disable_block W164a W164b W116 W484 W504 `ifdef SPYGLASS `else `ifdef FV_ASSERT_ON `else // synopsys translate_off `endif `ifdef ASSERT_ON `ifdef SPYGLASS wire disable_assert_plusarg = 1'b0; `else `ifdef FV_ASSERT_ON wire disable_assert_plusarg = 1'b0; `else wire disable_assert_plusarg = $test$plusargs("DISABLE_NESS_FLOW_ASSERTIONS"); `endif `endif wire assert_enabled = 1'b1 && !disable_assert_plusarg; `endif `ifdef FV_ASSERT_ON `else // synopsys translate_on `endif `ifdef ASSERT_ON //synopsys translate_off `ifndef SYNTH_LEVEL1_COMPILE `ifndef SYNTHESIS always @(assert_enabled) begin if ( assert_enabled === 1'b0 ) begin $display("Asserts are disabled for %m"); end end `endif `endif //synopsys translate_on `endif `endif // spyglass enable_block W164a W164b W116 W484 W504 //The NV_BLKBOX_SRC0 module is only present when the FIFOGEN_MODULE_SEARCH // define is set. This is to aid fifogen team search for fifogen fifo // instance and module names in a given design. `ifdef FIFOGEN_MODULE_SEARCH NV_BLKBOX_SRC0 dummy_breadcrumb_fifogen_blkbox (.Y()); `endif // spyglass enable_block W401 -- clock is not input to module // synopsys dc_script_begin // set_boundary_optimization find(design, "NV_NVDLA_RUBIK_fifo") true // synopsys dc_script_end `ifdef SYNTH_LEVEL1_COMPILE `else `ifdef SYNTHESIS `else `ifdef PRAND_VERILOG // Only verilog needs any local variables reg [47:0] prand_local_seed0; reg prand_initialized0; reg prand_no_rollpli0; `endif `endif `endif function [31:0] prand_inst0; //VCS coverage off input [31:0] min; input [31:0] max; reg [32:0] diff; begin `ifdef SYNTH_LEVEL1_COMPILE prand_inst0 = min; `else `ifdef SYNTHESIS prand_inst0 = min; `else `ifdef PRAND_VERILOG if (prand_initialized0 !== 1'b1) begin prand_no_rollpli0 = $test$plusargs("NO_ROLLPLI"); if (!prand_no_rollpli0) prand_local_seed0 = {$prand_get_seed(0), 16'b0}; prand_initialized0 = 1'b1; end if (prand_no_rollpli0) begin prand_inst0 = min; end else begin diff = max - min + 1; prand_inst0 = min + prand_local_seed0[47:16] % diff; // magic numbers taken from Java's random class (same as lrand48) prand_local_seed0 = prand_local_seed0 * 48'h5deece66d + 48'd11; end `else `ifdef PRAND_OFF prand_inst0 = min; `else prand_inst0 = $RollPLI(min, max, "auto"); `endif `endif `endif `endif end //VCS coverage on endfunction `ifdef SYNTH_LEVEL1_COMPILE `else `ifdef SYNTHESIS `else `ifdef PRAND_VERILOG // Only verilog needs any local variables reg [47:0] prand_local_seed1; reg prand_initialized1; reg prand_no_rollpli1; `endif `endif `endif function [31:0] prand_inst1; //VCS coverage off input [31:0] min; input [31:0] max; reg [32:0] diff; begin `ifdef SYNTH_LEVEL1_COMPILE prand_inst1 = min; `else `ifdef SYNTHESIS prand_inst1 = min; `else `ifdef PRAND_VERILOG if (prand_initialized1 !== 1'b1) begin prand_no_rollpli1 = $test$plusargs("NO_ROLLPLI"); if (!prand_no_rollpli1) prand_local_seed1 = {$prand_get_seed(1), 16'b0}; prand_initialized1 = 1'b1; end if (prand_no_rollpli1) begin prand_inst1 = min; end else begin diff = max - min + 1; prand_inst1 = min + prand_local_seed1[47:16] % diff; // magic numbers taken from Java's random class (same as lrand48) prand_local_seed1 = prand_local_seed1 * 48'h5deece66d + 48'd11; end `else `ifdef PRAND_OFF prand_inst1 = min; `else prand_inst1 = $RollPLI(min, max, "auto"); `endif `endif `endif `endif end //VCS coverage on endfunction endmodule // NV_NVDLA_RUBIK_fifo