URL
https://opencores.org/ocsvn/openrisc/openrisc/trunk
Subversion Repositories openrisc
[/] [openrisc/] [trunk/] [orpsocv2/] [rtl/] [verilog/] [or1200/] [or1200_fpu.v] - Rev 522
Go to most recent revision | Compare with Previous | Blame | View Log
////////////////////////////////////////////////////////////////////// //// //// //// OR1200's FPU Wrapper //// //// //// //// This file is part of the OpenRISC 1200 project //// //// http://opencores.org/project,or1k //// //// //// //// Description //// //// Wrapper for floating point unit. //// //// Interface based on MULT/MAC unit. //// //// //// //// To Do: //// //// - lf.rem.s and lf.madd.s instruction support //// //// - implement FP SPRs as needed //// //// //// //// Author(s): //// //// - Julius Baxter, julius@opencores.org //// //// //// ////////////////////////////////////////////////////////////////////// //// //// //// Copyright (C) 2009,2010 Authors and OPENCORES.ORG //// //// //// //// This source file may be used and distributed without //// //// restriction provided that this copyright statement is not //// //// removed from the file and that any derivative work contains //// //// the original copyright notice and the associated disclaimer. //// //// //// //// This source file is free software; you can redistribute it //// //// and/or modify it under the terms of the GNU Lesser General //// //// Public License as published by the Free Software Foundation; //// //// either version 2.1 of the License, or (at your option) any //// //// later version. //// //// //// //// This source is distributed in the hope that it will be //// //// useful, but WITHOUT ANY WARRANTY; without even the implied //// //// warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR //// //// PURPOSE. See the GNU Lesser General Public License for more //// //// details. //// //// //// //// You should have received a copy of the GNU Lesser General //// //// Public License along with this source; if not, download it //// //// from http://www.opencores.org/lgpl.shtml //// //// //// ////////////////////////////////////////////////////////////////////// // synopsys translate_off `include "timescale.v" // synopsys translate_on `include "or1200_defines.v" module or1200_fpu( // Clock and reset clk, rst, // FPU interface ex_freeze, a, b, fpu_op, result, done, // Flag controls flagforw, flag_we, // Exception signal sig_fp, except_started, // FPCSR system register fpcsr_we, fpcsr, // SPR interface -- currently unused spr_cs, spr_write, spr_addr, spr_dat_i, spr_dat_o ); parameter width = `OR1200_OPERAND_WIDTH; // // I/O // // // Clock and reset // input clk; input rst; // // FPU interface // input ex_freeze; input [width-1:0] a; input [width-1:0] b; input [`OR1200_FPUOP_WIDTH-1:0] fpu_op; output [width-1:0] result; output done; // // Flag signals // output flagforw; output flag_we; // // FPCSR interface // input fpcsr_we; output [`OR1200_FPCSR_WIDTH-1:0] fpcsr; // // Exception signal // output sig_fp; input except_started; // // SPR interface // input spr_cs; input spr_write; input [31:0] spr_addr; input [31:0] spr_dat_i; output [31:0] spr_dat_o; `ifndef OR1200_FPU_IMPLEMENTED // No FPU needed assign result = 0; assign flagforw = 0; assign flag_we = 0; assign sig_fp = 0; assign spr_dat_o = 0; assign fpcsr = 0; assign done = 1; `else // // Internals // wire fpu_op_is_arith, fpu_op_is_conv, fpu_op_is_comp; wire fpu_op_r_is_arith, fpu_op_r_is_conv, fpu_op_r_is_comp; wire fpu_arith_done, fpu_conv_done, fpu_comp_done; wire [width-1:0] result_arith, result_conv; reg [`OR1200_FPUOP_WIDTH-1:0] fpu_op_r; reg [`OR1200_FPCSR_WIDTH-1:0] fpcsr_r; wire fpu_op_valid; reg fpu_op_valid_re; wire fpu_check_op; wire inf, inv_inf_op_in,snan, snan_in,qnan, ine, overflow, underflow, zero, dbz, dbz_in, mul_z_inf, nan_in; wire altb, blta, aeqb, inf_cmp, zero_cmp, unordered ; wire snan_conv, ine_conv, inv_conv, zero_conv, underflow_conv, overflow_conv; wire inv_comp; reg flag; assign spr_dat_o = 0; assign fpcsr = fpcsr_r; assign sig_fp = fpcsr_r[`OR1200_FPCSR_FPEE] & (|fpcsr_r[`OR1200_FPCSR_WIDTH-1:`OR1200_FPCSR_OVF]); // Top bit indicates FPU instruction assign fpu_op_valid = fpu_op[`OR1200_FPUOP_WIDTH-1]; assign fpu_check_op = !ex_freeze & fpu_op_valid; // Generate signals to latch fpu_op from decode instruction, then latch // operands when they appear during execute stage assign fpu_op_is_arith = !(|fpu_op[3:2]); assign fpu_op_is_conv = fpu_op[2] & !fpu_op[3]; assign fpu_op_is_comp = fpu_op[3]; assign fpu_op_r_is_arith = !(|fpu_op_r[3:2]); assign fpu_op_r_is_conv = fpu_op_r[2] & !fpu_op_r[3]; assign fpu_op_r_is_comp = fpu_op_r[3]; assign done = (fpu_op_r_is_arith & fpu_arith_done) | (fpu_op_r_is_conv & fpu_conv_done) | (fpu_op_r_is_comp & fpu_comp_done) ; // Register fpu_op (remove FPU op valid bit [7], replace with 0) always @(posedge clk) if (fpu_check_op) fpu_op_r <= {1'b0,fpu_op[`OR1200_FPUOP_WIDTH-2:0]}; // Indicate new FPU op always @(posedge clk or `OR1200_RST_EVENT rst) if (rst == `OR1200_RST_VALUE) fpu_op_valid_re <= 0; else if (fpu_op_valid_re) fpu_op_valid_re <= 0; else if (fpu_check_op) fpu_op_valid_re <= 1; // // FPCSR system group register implementation // always @(posedge clk or `OR1200_RST_EVENT rst) begin if (rst == `OR1200_RST_VALUE) fpcsr_r <= 0; else begin if (fpcsr_we) fpcsr_r <= b[`OR1200_FPCSR_WIDTH-1:0]; else if (done) begin fpcsr_r[`OR1200_FPCSR_OVF] <= (overflow & fpu_op_r_is_arith); fpcsr_r[`OR1200_FPCSR_UNF] <= (underflow & fpu_op_r_is_arith) | (underflow_conv & fpu_op_r_is_conv); fpcsr_r[`OR1200_FPCSR_SNF] <= (snan & fpu_op_r_is_arith)| (snan_conv & fpu_op_r_is_conv); fpcsr_r[`OR1200_FPCSR_QNF] <= (qnan & fpu_op_r_is_arith); fpcsr_r[`OR1200_FPCSR_ZF] <= (zero & fpu_op_r_is_arith) | (zero_cmp & fpu_op_r_is_comp) | (zero_conv & fpu_op_r_is_conv); fpcsr_r[`OR1200_FPCSR_IXF] <= (ine & fpu_op_r_is_arith) | (ine_conv & fpu_op_r_is_conv); fpcsr_r[`OR1200_FPCSR_IVF] <= ((snan_in | dbz_in | inv_inf_op_in | mul_z_inf) & fpu_op_r_is_arith) | ((inv_conv | snan_conv) & fpu_op_r_is_conv) | (inv_comp & fpu_op_r_is_comp); fpcsr_r[`OR1200_FPCSR_INF] <= (inf & fpu_op_r_is_arith) | (inf_cmp & fpu_op_r_is_comp); fpcsr_r[`OR1200_FPCSR_DZF] <= (dbz & fpu_op_r_is_arith); end // if (fpu_arith_done | fpu_conv_done) if (except_started) fpcsr_r[`OR1200_FPCSR_FPEE] <= 0; end // else: !if(rst) end // always @ (posedge clk or `OR1200_RST_EVENT rst) // // Comparison flag generation // always @* begin // Get rid of top bit - is FPU op valid bit case({1'b0,fpu_op_r[`OR1200_FPUOP_WIDTH-2:0]}) `OR1200_FPCOP_SFEQ: begin flag = aeqb; end `OR1200_FPCOP_SFNE: begin flag = !aeqb; end `OR1200_FPCOP_SFGT: begin flag = blta & !aeqb; end `OR1200_FPCOP_SFGE: begin flag = blta | aeqb; end `OR1200_FPCOP_SFLT: begin flag = altb & !aeqb; end `OR1200_FPCOP_SFLE: begin flag = altb | aeqb; end default: begin flag = 0; end endcase // case (fpu_op_r) end // always@ (posedge clk) assign flagforw = flag; // Determine here where we do the write, ie how much we pipeline the // comparison assign flag_we = fpu_op_r_is_comp & fpu_comp_done; // MUX for outputs from arith and conversion modules assign result = fpu_op_r_is_conv ? result_conv : result_arith; // // Instantiate FPU modules // // FPU 100 VHDL core from OpenCores.org: http://opencores.org/project,fpu100 // Used only for add,sub,mul,div or1200_fpu_arith fpu_arith ( .clk_i(clk), .opa_i(a), .opb_i(b), .fpu_op_i({1'b0,fpu_op_r[1:0]}), // Only bottom 2 bits .rmode_i(fpcsr_r[`OR1200_FPCSR_RM]), .output_o(result_arith), .start_i(fpu_op_valid_re & fpu_op_r_is_arith), .ready_o(fpu_arith_done), .ine_o(ine), .overflow_o(overflow), .underflow_o(underflow), .div_zero_o(dbz), .inf_o(inf), .zero_o(zero), .qnan_o(qnan), .snan_o(snan) ); // Logic for detection of signaling NaN on input // signaling NaN: exponent is 8hff, [22] is zero, rest of fract is non-zero // quiet NaN: exponent is 8hff, [22] is 1 reg a_is_snan, b_is_snan; reg a_is_qnan, b_is_qnan; always @(posedge clk) begin a_is_snan <= (a[30:23]==8'hff) & !a[22] & (|a[21:0]); b_is_snan <= (b[30:23]==8'hff) & !b[22] & (|b[21:0]); a_is_qnan <= (a[30:23]==8'hff) & a[22]; b_is_qnan <= (b[30:23]==8'hff) & b[22]; end // Signal to indicate there was a signaling NaN on input assign snan_in = a_is_snan | b_is_snan; // Check for, add with opposite signed infinities, or subtract with // same signed infinities. reg a_is_inf, b_is_inf, a_b_sign_xor; always @(posedge clk) begin a_is_inf <= (a[30:23]==8'hff) & !(|a[22:0]); b_is_inf <= (b[30:23]==8'hff) & !(|a[22:0]); a_b_sign_xor <= a[31] ^ b[31]; end assign inv_inf_op_in = (a_is_inf & b_is_inf) & ((a_b_sign_xor & ({1'b0,fpu_op_r[`OR1200_FPUOP_WIDTH-2:0]} == `OR1200_FPUOP_ADD)) | (!a_b_sign_xor & ({1'b0,fpu_op_r[`OR1200_FPUOP_WIDTH-2:0]} == `OR1200_FPUOP_SUB))) ; // Check if it's 0.0/0.0 to generate invalid signal (ignore sign bit) reg a_is_zero, b_is_zero; always @(posedge clk) begin a_is_zero <= !(|a[30:0]); b_is_zero <= !(|b[30:0]); end assign dbz_in = ({1'b0,fpu_op_r[`OR1200_FPUOP_WIDTH-2:0]} == `OR1200_FPUOP_DIV) & (a_is_zero & b_is_zero); assign mul_z_inf = ({1'b0,fpu_op_r[`OR1200_FPUOP_WIDTH-2:0]} == `OR1200_FPUOP_MUL) & ((a_is_zero & b_is_inf) | (b_is_zero & a_is_inf)); assign nan_in = (a_is_snan | b_is_snan | a_is_qnan | b_is_qnan); // 32-bit integer <-> single precision floating point conversion unit or1200_fpu_intfloat_conv fpu_intfloat_conv ( .clk(clk), .rmode(fpcsr_r[`OR1200_FPCSR_RM]), .fpu_op(fpu_op_r[2:0]), .opa(a), .out(result_conv), .snan(snan_conv), .ine(ine_conv), .inv(inv_conv), .overflow(overflow_conv), .underflow(underflow_conv), .zero(zero_conv) ); // 5-long shift reg for conversion ready counter reg [6:0] fpu_conv_shr; always @(posedge clk) fpu_conv_shr <= {fpu_conv_shr[5:0],fpu_check_op & fpu_op_is_conv}; assign fpu_conv_done = fpu_conv_shr[6]; // Single precision floating point number comparison module or1200_fpu_fcmp fpu_fcmp ( .opa(a), .opb(b), .unordered(unordered), // I am convinced the comparison logic is wrong way around in this // module, simplest to swap them on output -- julius .altb(blta), .blta(altb), .aeqb(aeqb), .inf(inf_cmp), .zero(zero_cmp)); reg fpu_op_valid_re_r; always @(posedge clk) fpu_op_valid_re_r <= fpu_op_valid_re; assign fpu_comp_done = fpu_op_valid_re_r & fpu_op_r_is_comp; // Comparison invalid when sNaN in on an equal comparison, or any NaN // for any other comparison. assign inv_comp = (snan_in & ({1'b0,fpu_op_r[`OR1200_FPUOP_WIDTH-2:0]} == `OR1200_FPCOP_SFEQ)) | (nan_in & ({1'b0,fpu_op_r[`OR1200_FPUOP_WIDTH-2:0]} != `OR1200_FPCOP_SFEQ)); `endif // !`ifndef OR1200_FPU_IMPLEMENTED endmodule // or1200_fpu
Go to most recent revision | Compare with Previous | Blame | View Log