Github_Repos/cvw
1
0
Fork 1
mirror of https://github.com/openhwgroup/cvw synced 2025-02-03 10:15:19 +00:00
Code Issues Packages Projects Releases Wiki Activity

Cleanup branch predictor files.

Browse Source
This commit is contained in:
Ross Thompson 2023-02-20 18:45:45 -06:00
parent 7df3a84060
commit 2c2c1b5221
7 changed files with 13 additions and 729 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

146
src/ifu/bpred/foldedgshare.sv
View File

@ -1,146 +0,0 @@
///////////////////////////////////////////
// gsharePredictor.sv
//
// Written: Shreya Sanghai
// Email: ssanghai@hmc.edu
// Created: March 16, 2021
// Modified:
//
// Purpose: Global History Branch predictor with parameterized global history register
//
// A component of the CORE-V-WALLY configurable RISC-V project.
//
// Copyright (C) 2021-23 Harvey Mudd College & Oklahoma State University
//
// SPDX-License-Identifier: Apache-2.0 WITH SHL-2.1
//
// Licensed under the Solderpad Hardware License v 2.1 (the “License”); you may not use this file
// except in compliance with the License, or, at your option, the Apache License version 2.0. You
// may obtain a copy of the License at
//
// https://solderpad.org/licenses/SHL-2.1/
//
// Unless required by applicable law or agreed to in writing, any work distributed under the
// License is distributed on an “AS IS” BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND,
// either express or implied. See the License for the specific language governing permissions
// and limitations under the License.
////////////////////////////////////////////////////////////////////////////////////////////////
`include "wally-config.vh"
module foldedgshare #(parameter k = 16, depth = 10) (
input logic clk,
input logic reset,
input logic StallF, StallD, StallE, StallM, StallW,
input logic FlushD, FlushE, FlushM, FlushW,
// input logic [`XLEN-1:0] LookUpPC,
output logic [1:0] DirPredictionF,
output logic DirPredictionWrongE,
// update
input logic [`XLEN-1:0] PCNextF, PCF, PCD, PCE, PCM,
input logic BranchInstrF, BranchInstrD, BranchInstrE, BranchInstrM, BranchInstrW,
input logic PCSrcE
);
logic MatchF, MatchD, MatchE, MatchM, MatchW;
logic MatchNextX, MatchXF;
logic [1:0] TableDirPredictionF, DirPredictionD, DirPredictionE;
logic [1:0] NewDirPredictionF, NewDirPredictionD, NewDirPredictionE, NewDirPredictionM, NewDirPredictionW;
logic [k-1:0] GHRF;
logic [k:0] GHRD, OldGHRE, GHRE, GHRM, GHRW;
logic [k-1:0] GHRNextF;
logic [k:0] GHRNextD, GHRNextE, GHRNextM, GHRNextW;
logic [k-1:0] IndexNextF, IndexF;
logic [k-1:0] IndexD, IndexE, IndexM, IndexW;
logic [depth-1:0] FinalIndexNextF, FinalIndexW;
logic PCSrcM, PCSrcW;
logic [`XLEN-1:0] PCW;
logic [1:0] ForwardNewDirPrediction, ForwardDirPredictionF;
localparam delta = 2 * depth - k;
assign IndexNextF = GHRNextF ^ {PCNextF[k+1] ^ PCNextF[1], PCNextF[k:2]};
assign IndexF = GHRF ^ {PCF[k+1] ^ PCF[1], PCF[k:2]};
assign IndexD = GHRD[k-1:0] ^ {PCD[k+1] ^ PCD[1], PCD[k:2]};
assign IndexE = GHRE[k-1:0] ^ {PCE[k+1] ^ PCE[1], PCE[k:2]};
assign IndexM = GHRM[k-1:0] ^ {PCM[k+1] ^ PCM[1], PCM[k:2]};
assign IndexW = GHRW[k-1:0] ^ {PCW[k+1] ^ PCW[1], PCW[k:2]};
// just be dumb for now.
//localparam int kToDepthRatio = (k+depth) / depth;
assign FinalIndexNextF = IndexNextF[depth-1:0] ^ {{delta{1'b0}} , IndexNextF[k-1:depth]};
assign FinalIndexW = IndexW[depth-1:0] ^ {{delta{1'b0}} , IndexW[k-1:depth]};
ram2p1r1wbe #(2**depth, 2) PHT(.clk(clk),
.ce1(~StallF | reset), .ce2(~StallW & ~FlushW),
.ra1(FinalIndexNextF),
.rd1(TableDirPredictionF),
.wa2(FinalIndexW),
.wd2(NewDirPredictionW),
.we2(BranchInstrW & ~StallW & ~FlushW),
.bwe2(1'b1));
// if there are non-flushed branches in the pipeline we need to forward the prediction from that stage to the NextF demi stage
// and then register for use in the Fetch stage.
assign MatchF = BranchInstrF & ~FlushD & (IndexNextF == IndexF);
assign MatchD = BranchInstrD & ~FlushE & (IndexNextF == IndexD);
assign MatchE = BranchInstrE & ~FlushM & (IndexNextF == IndexE);
assign MatchM = BranchInstrM & ~FlushW & (IndexNextF == IndexM);
assign MatchW = BranchInstrW & (IndexNextF == IndexW);
assign MatchNextX = MatchF | MatchD | MatchE | MatchM | MatchW;
flopenr #(1) MatchReg(clk, reset, ~StallF, MatchNextX, MatchXF);
assign ForwardNewDirPrediction = MatchF ? NewDirPredictionF :
MatchD ? NewDirPredictionD :
MatchE ? NewDirPredictionE :
MatchM ? NewDirPredictionM :
NewDirPredictionW;
flopenr #(2) ForwardDirPredicitonReg(clk, reset, ~StallF, ForwardNewDirPrediction, ForwardDirPredictionF);
assign DirPredictionF = MatchXF ? ForwardDirPredictionF : TableDirPredictionF;
// DirPrediction pipeline
flopenr #(2) PredictionRegD(clk, reset, ~StallD, DirPredictionF, DirPredictionD);
flopenr #(2) PredictionRegE(clk, reset, ~StallE, DirPredictionD, DirPredictionE);
// New prediction pipeline
satCounter2 BPDirUpdateF(.BrDir(DirPredictionF[1]), .OldState(DirPredictionF), .NewState(NewDirPredictionF));
flopenr #(2) NewPredDReg(clk, reset, ~StallD, NewDirPredictionF, NewDirPredictionD);
satCounter2 BPDirUpdateE(.BrDir(PCSrcE), .OldState(DirPredictionE), .NewState(NewDirPredictionE));
flopenr #(2) NewPredMReg(clk, reset, ~StallM, NewDirPredictionE, NewDirPredictionM);
flopenr #(2) NewPredWReg(clk, reset, ~StallW, NewDirPredictionM, NewDirPredictionW);
// PCSrc pipeline
flopenrc #(1) PCSrcMReg(clk, reset, FlushM, ~StallM, PCSrcE, PCSrcM);
flopenrc #(1) PCSrcWReg(clk, reset, FlushW, ~StallW, PCSrcM, PCSrcW);
// GHR pipeline
assign GHRNextF = FlushD ? GHRNextD[k:1] :
BranchInstrF ? {DirPredictionF[1], GHRF[k-1:1]} :
GHRF;
flopenr #(k) GHRFReg(clk, reset, (~StallF) | FlushD, GHRNextF, GHRF);
assign GHRNextD = FlushD ? GHRNextE : {DirPredictionF[1], GHRF};
flopenr #(k+1) GHRDReg(clk, reset, (~StallD) | FlushD, GHRNextD, GHRD);
assign GHRNextE = FlushE ? GHRNextM : GHRD;
flopenr #(k+1) GHREReg(clk, reset, (~StallE) | FlushE, GHRNextE, OldGHRE);
assign GHRE = BranchInstrE ? {PCSrcE, OldGHRE[k-1:0]} : OldGHRE;
assign GHRNextM = FlushM ? GHRNextW : GHRE;
flopenr #(k+1) GHRMReg(clk, reset, (~StallM) | FlushM, GHRNextM, GHRM);
assign GHRNextW = FlushW ? GHRW : GHRM;
flopenr #(k+1) GHRWReg(clk, reset, (BranchInstrM & ~StallW) | FlushW, GHRNextW, GHRW);
assign DirPredictionWrongE = PCSrcE != DirPredictionE[1] & BranchInstrE;
flopenr #(`XLEN) PCWReg(clk, reset, ~StallW, PCM, PCW);
endmodule

80
src/ifu/bpred/globalhistory.sv
View File

@ -1,80 +0,0 @@
///////////////////////////////////////////
// globalHistoryPredictor.sv
//
// Written: Shreya Sanghai
// Email: ssanghai@hmc.edu
// Created: March 16, 2021
// Modified:
//
// Purpose: Global History Branch predictor with parameterized global history register
//
// A component of the CORE-V-WALLY configurable RISC-V project.
//
// Copyright (C) 2021-23 Harvey Mudd College & Oklahoma State University
//
// SPDX-License-Identifier: Apache-2.0 WITH SHL-2.1
//
// Licensed under the Solderpad Hardware License v 2.1 (the “License”); you may not use this file
// except in compliance with the License, or, at your option, the Apache License version 2.0. You
// may obtain a copy of the License at
//
// https://solderpad.org/licenses/SHL-2.1/
//
// Unless required by applicable law or agreed to in writing, any work distributed under the
// License is distributed on an “AS IS” BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND,
// either express or implied. See the License for the specific language governing permissions
// and limitations under the License.
////////////////////////////////////////////////////////////////////////////////////////////////
`include "wally-config.vh"
module globalhistory #(parameter k = 10) (
input logic clk,
input logic reset,
input logic StallF, StallD, StallE, StallM, StallW,
input logic FlushD, FlushE, FlushM, FlushW,
output logic [1:0] DirPredictionF,
output logic DirPredictionWrongE,
// update
input logic BranchInstrE, BranchInstrM, PCSrcE
);
logic [k-1:0] IndexNextF, IndexE;
logic [1:0] DirPredictionD, DirPredictionE;
logic [1:0] NewDirPredictionE, NewDirPredictionM;
logic [k-1:0] GHRF, GHRD, GHRE, GHRM, GHR;
logic [k-1:0] GHRNext;
logic PCSrcM;
assign IndexNextF = GHR;
assign IndexE = GHRM;
ram2p1r1wbe #(2**k, 2) PHT(.clk(clk),
.ce1(~StallF), .ce2(~StallM & ~FlushM),
.ra1(IndexNextF),
.rd1(DirPredictionF),
.wa2(IndexE),
.wd2(NewDirPredictionM),
.we2(BranchInstrM & ~StallW & ~FlushW),
.bwe2(1'b1));
flopenrc #(2) PredictionRegD(clk, reset, FlushD, ~StallD, DirPredictionF, DirPredictionD);
flopenrc #(2) PredictionRegE(clk, reset, FlushE, ~StallE, DirPredictionD, DirPredictionE);
satCounter2 BPDirUpdateE(.BrDir(PCSrcE), .OldState(DirPredictionE), .NewState(NewDirPredictionE));
flopenrc #(2) NewPredictionRegM(clk, reset, FlushM, ~StallM, NewDirPredictionE, NewDirPredictionM);
assign DirPredictionWrongE = PCSrcE != DirPredictionE[1] & BranchInstrE;
assign GHRNext = BranchInstrM ? {PCSrcM, GHR[k-1:1]} : GHR;
flopenr #(k) GHRReg(clk, reset, ~StallM & ~FlushM & BranchInstrM, GHRNext, GHR);
flopenrc #(1) PCSrcMReg(clk, reset, FlushM, ~StallM, PCSrcE, PCSrcM);
flopenrc #(k) GHRFReg(clk, reset, FlushD, ~StallF, GHR, GHRF);
flopenrc #(k) GHRDReg(clk, reset, FlushD, ~StallD, GHRF, GHRD);
flopenrc #(k) GHREReg(clk, reset, FlushE, ~StallE, GHRD, GHRE);
flopenrc #(k) GHRMReg(clk, reset, FlushM, ~StallM, GHRE, GHRM);
endmodule

13
src/ifu/bpred/gshare.sv
View File

@ -1,12 +1,13 @@
/////////////////////////////////////////// ///////////////////////////////////////////
// globalHistoryPredictor.sv // gshare.sv
// //
// Written: Shreya Sanghai // Written: Ross Thompson
// Email: ssanghai@hmc.edu // Email: ross1728@gmail.com
// Created: March 16, 2021 // Created: 16 March 2021
// Modified: // Adapted from ssanghai@hmc.edu (Shreya Sanghai)
// Modified: 20 February 2023
// //
// Purpose: Global History Branch predictor with parameterized global history register // Purpose: gshare and Global History Branch predictors
// //
// A component of the CORE-V-WALLY configurable RISC-V project. // A component of the CORE-V-WALLY configurable RISC-V project.
// //

11
src/ifu/bpred/gsharebasic.sv
View File

@ -1,10 +1,11 @@
/////////////////////////////////////////// ///////////////////////////////////////////
// globalHistoryPredictor.sv // gsharebasic.sv
// //
// Written: Shreya Sanghai // Written: Ross Thompson
// Email: ssanghai@hmc.edu // Email: ross1728@gmail.com
// Created: March 16, 2021 // Created: 16 March 2021
// Modified: // Adapted from ssanghai@hmc.edu (Shreya Sanghai) global history predictor implementation.
// Modified: 20 February 2023
// //
// Purpose: Global History Branch predictor with parameterized global history register // Purpose: Global History Branch predictor with parameterized global history register
// //

219
src/ifu/bpred/optgshare.sv
View File

@ -1,219 +0,0 @@
///////////////////////////////////////////
// gsharePredictor.sv
//
// Written: Shreya Sanghai
// Email: ssanghai@hmc.edu
// Created: March 16, 2021
// Modified:
//
// Purpose: Global History Branch predictor with parameterized global history register
//
// A component of the Wally configurable RISC-V project.
//
// Copyright (C) 2021 Harvey Mudd College & Oklahoma State University
//
// SPDX-License-Identifier: Apache-2.0 WITH SHL-2.1
//
// Licensed under the Solderpad Hardware License v 2.1 (the “License”); you may not use this file
// except in compliance with the License, or, at your option, the Apache License version 2.0. You
// may obtain a copy of the License at
//
// https://solderpad.org/licenses/SHL-2.1/
//
// Unless required by applicable law or agreed to in writing, any work distributed under the
// License is distributed on an “AS IS” BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND,
// either express or implied. See the License for the specific language governing permissions
// and limitations under the License.
////////////////////////////////////////////////////////////////////////////////////////////////
`include "wally-config.vh"
module optgshare #(parameter k = 10) (
input logic clk,
input logic reset,
input logic StallF, StallD, StallE, StallM, StallW,
input logic FlushD, FlushE, FlushM, FlushW,
// input logic [`XLEN-1:0] LookUpPC,
output logic [1:0] DirPredictionF,
output logic DirPredictionWrongE,
// update
input logic [`XLEN-1:0] PCNextF, PCF, PCD, PCE, PCM,
input logic BranchInstrF, BranchInstrD, BranchInstrE, BranchInstrM, BranchInstrW,
input logic PCSrcE
);
logic MatchF, MatchD, MatchE, MatchM, MatchW;
logic MatchNextX, MatchXF;
logic [1:0] TableDirPredictionF, DirPredictionD, DirPredictionE;
logic [1:0] NewDirPredictionF, NewDirPredictionD, NewDirPredictionE, NewDirPredictionM, NewDirPredictionW;
logic [k-1:0] GHRF;
logic [k:0] GHRD, OldGHRE, GHRE, GHRM, GHRW;
logic [k-1:0] GHRNextF;
logic [k:0] GHRNextD, GHRNextE, GHRNextM, GHRNextW;
logic [k-1:0] IndexNextF, IndexF;
logic [k-1:0] IndexD, IndexE, IndexM, IndexW;
logic PCSrcM, PCSrcW;
logic [`XLEN-1:0] PCW;
logic [1:0] ForwardNewDirPrediction, ForwardDirPredictionF;
logic [k+4:0] GHRNext, GHR;
logic GHRUpdateEn;
assign GHRUpdateEn = BranchInstrF | (DirPredictionWrongE & BranchInstrE) |
FlushD | FlushE | FlushM | FlushW;
// it doesn't work this way. Instead we need to see how many branch instructions are flushed.
// then shift over by that amount.
logic RemoveBrW, RemoveBrM, RemoveBrE, RemoveBrD, RemoveBrF, RemoveBrNextF;
assign RemoveBrW = '0;
assign RemoveBrM = BranchInstrM & FlushW;
assign RemoveBrE = BranchInstrE & FlushM;
assign RemoveBrD = BranchInstrD & FlushE;
assign RemoveBrF = BranchInstrF & FlushD;
assign RemoveBrNextF = BranchInstrF & FlushD;
always_comb begin
casez ({BranchInstrF, DirPredictionWrongE, RemoveBrF, RemoveBrD, RemoveBrE, RemoveBrM})
6'b00_0000: GHRNext = GHR; // no change
6'b00_0001: GHRNext = {GHR[k+4:k+1], GHR[k-1:0], 1'b0}; // RemoveBrM
6'b0?_0010: GHRNext = {GHR[k+4:k+2], GHR[k:0], 1'b0}; // RemoveBrE
6'b0?_0011: GHRNext = {GHR[k+4:k+2], GHR[k-1:0], 2'b0}; // RemoveBrE, RemoveBrM
6'b00_0100: GHRNext = {GHR[k+4:k+2], GHR[k-1:0], 2'b0}; // RemoveBrD
6'b00_0101: GHRNext = {GHR[k+4:k+3], GHR[k+1:0], 1'b0}; // RemoveBrD, RemoveBrM
6'b0?_0110: GHRNext = {GHR[k+4:k+3], GHR[k+1], GHR[k-1:0], 2'b0}; // RemoveBrD, RemoveBrE
6'b0?_0111: GHRNext = {GHR[k+4:k+3], GHR[k-1:0], 3'b0}; // RemoveBrD, RemoveBrE, RemoveBrM
6'b?0_1000: GHRNext = {GHR[k+2:0], 2'b0}; // RemoveBrF,
6'b?0_1001: GHRNext = {GHR[k+2:k+1], GHR[k-1:0], 3'b0}; // RemoveBrF, RemoveBrM
6'b??_1010: GHRNext = {GHR[k+2], GHR[k:0], 3'b0}; // RemoveBrF, RemoveBrE
6'b??_1011: GHRNext = {GHR[k+2], GHR[k-1:0], 4'b0}; // RemoveBrF, RemoveBrE, RemoveBrM
6'b?0_1100: GHRNext = {GHR[k+1:0], 3'b0}; // RemoveBrF, RemoveBrD
6'b?0_1101: GHRNext = {GHR[k+1], GHR[k-1:0], 4'b0}; // RemoveBrF, RemoveBrD, RemoveBrM
6'b??_1110: GHRNext = {GHR[k:0], 4'b0}; // RemoveBrF, RemoveBrD, RemoveBrE
6'b??_1111: GHRNext = {GHR[k-1:0], 5'b0}; // RemoveBrF, RemoveBrD, RemoveBrE, RemoveBrM
6'b?1_0000: GHRNext = {PCSrcE, GHR[k+3:0]}; // Miss prediction, no branches to flushes
6'b?1_0001: GHRNext = {PCSrcE, GHR[k+3:k], GHR[k-1:1], 1'b0}; // Miss prediction, branch in Memory stage dropped
6'b?1_1100: GHRNext = {PCSrcE, GHR[k+1:0], 2'b00}; // Miss prediction, cannot have RemoveBrE
6'b?1_1101: GHRNext = {PCSrcE, GHR[k+1], GHR[k-1:0], 3'b0}; // Miss prediction, cannot have RemoveBrE
6'b10_0000: GHRNext = {DirPredictionF[1], GHR[k+4:1]};
6'b10_0001: GHRNext = {DirPredictionF[1], GHR[k+4:k+1], GHR[k-1:1], 1'b0};
6'b10_0010: GHRNext = {DirPredictionF[1], GHR[k+4:k+2], GHR[k:1], 1'b0};
6'b10_0011: GHRNext = {DirPredictionF[1], GHR[k+4:k+2], GHR[k-1:1], 2'b0};
6'b10_0100: GHRNext = {DirPredictionF[1], GHR[k+4:k+3], GHR[k+1:1], 1'b0};
6'b10_0101: GHRNext = {DirPredictionF[1], GHR[k+4:k+3], GHR[k+1], GHR[k-1:1], 2'b0};
6'b10_0110: GHRNext = {DirPredictionF[1], GHR[k+4:k+3], GHR[k], GHR[k-1:1], 2'b0};
6'b10_0111: GHRNext = {DirPredictionF[1], GHR[k+4:k+3], GHR[k-1:1], 3'b0};
default: GHRNext = GHR;
endcase
end
flopenr #(k+5) GHRReg(clk, reset, GHRUpdateEn, GHRNext, GHR);
logic [k-1:0] GHRNextF_temp, GHRF_temp;
logic [k:0] GHRD_temp, GHRE_temp, GHRM_temp, GHRW_temp;
logic GHRFExtra_temp;
// these are also in the ieu controller. should create inputs.
logic InstrValidF, InstrValidD, InstrValidE, InstrValidM, InstrValidW;
flopenrc #(1) InstrValidFReg(clk, reset, FlushD, ~StallF, 1'b1, InstrValidF);
flopenrc #(1) InstrValidDReg(clk, reset, FlushD, ~StallD, InstrValidF, InstrValidD);
flopenrc #(1) InstrValidEReg(clk, reset, FlushE, ~StallE, InstrValidD, InstrValidE);
flopenrc #(1) InstrValidMReg(clk, reset, FlushM, ~StallM, InstrValidE, InstrValidM);
flopenrc #(1) InstrValidWReg(clk, reset, FlushW, ~StallW, InstrValidM, InstrValidW);
assign GHRNextF_temp = GHRNext[k+4:5];
assign GHRF_temp = InstrValidF ? GHR[k+3:4] : GHRNextF_temp;
assign GHRFExtra_temp = InstrValidF ? 1'b0 : GHR[k+4];
assign GHRD_temp = InstrValidD ? GHR[k+3:3] : {GHRFExtra_temp, GHRF_temp};
assign GHRE_temp = InstrValidE ? GHR[k+2:2] : GHRD_temp;
assign GHRM_temp = InstrValidM ? GHR[k+1:1] : GHRE_temp;
assign GHRW_temp = InstrValidW ? GHR[k:0] : GHRM_temp;
assign IndexNextF = GHRNextF ^ {PCNextF[k+1] ^ PCNextF[1], PCNextF[k:2]};
assign IndexF = GHRF ^ {PCF[k+1] ^ PCF[1], PCF[k:2]};
assign IndexD = GHRD[k-1:0] ^ {PCD[k+1] ^ PCD[1], PCD[k:2]};
assign IndexE = GHRE[k-1:0] ^ {PCE[k+1] ^ PCE[1], PCE[k:2]};
assign IndexM = GHRM[k-1:0] ^ {PCM[k+1] ^ PCM[1], PCM[k:2]};
assign IndexW = GHRW[k-1:0] ^ {PCW[k+1] ^ PCW[1], PCW[k:2]};
ram2p1r1wbe #(2**k, 2) PHT(.clk(clk),
.ce1(~StallF | reset), .ce2(~StallW & ~FlushW),
.ra1(IndexNextF),
.rd1(TableDirPredictionF),
.wa2(IndexW),
.wd2(NewDirPredictionW),
.we2(BranchInstrW & ~StallW & ~FlushW),
.bwe2(1'b1));
// if there are non-flushed branches in the pipeline we need to forward the prediction from that stage to the NextF demi stage
// and then register for use in the Fetch stage.
assign MatchF = BranchInstrF & ~FlushD & (IndexNextF == IndexF);
assign MatchD = BranchInstrD & ~FlushE & (IndexNextF == IndexD);
assign MatchE = BranchInstrE & ~FlushM & (IndexNextF == IndexE);
assign MatchM = BranchInstrM & ~FlushW & (IndexNextF == IndexM);
assign MatchW = BranchInstrW & (IndexNextF == IndexW);
assign MatchNextX = MatchF | MatchD | MatchE | MatchM | MatchW;
flopenr #(1) MatchReg(clk, reset, ~StallF, MatchNextX, MatchXF);
assign ForwardNewDirPrediction = MatchF ? NewDirPredictionF :
MatchD ? NewDirPredictionD :
MatchE ? NewDirPredictionE :
MatchM ? NewDirPredictionM :
NewDirPredictionW;
flopenr #(2) ForwardDirPredicitonReg(clk, reset, ~StallF, ForwardNewDirPrediction, ForwardDirPredictionF);
assign DirPredictionF = MatchXF ? ForwardDirPredictionF : TableDirPredictionF;
// DirPrediction pipeline
flopenr #(2) PredictionRegD(clk, reset, ~StallD, DirPredictionF, DirPredictionD);
flopenr #(2) PredictionRegE(clk, reset, ~StallE, DirPredictionD, DirPredictionE);
// New prediction pipeline
satCounter2 BPDirUpdateF(.BrDir(DirPredictionF[1]), .OldState(DirPredictionF), .NewState(NewDirPredictionF));
flopenr #(2) NewPredDReg(clk, reset, ~StallD, NewDirPredictionF, NewDirPredictionD);
satCounter2 BPDirUpdateE(.BrDir(PCSrcE), .OldState(DirPredictionE), .NewState(NewDirPredictionE));
flopenr #(2) NewPredMReg(clk, reset, ~StallM, NewDirPredictionE, NewDirPredictionM);
flopenr #(2) NewPredWReg(clk, reset, ~StallW, NewDirPredictionM, NewDirPredictionW);
// PCSrc pipeline
flopenrc #(1) PCSrcMReg(clk, reset, FlushM, ~StallM, PCSrcE, PCSrcM);
flopenrc #(1) PCSrcWReg(clk, reset, FlushW, ~StallW, PCSrcM, PCSrcW);
// GHR pipeline
assign GHRNextF = FlushD ? GHRNextD[k:1] :
BranchInstrF ? {DirPredictionF[1], GHRF[k-1:1]} :
GHRF;
flopenr #(k) GHRFReg(clk, reset, (~StallF) | FlushD, GHRNextF, GHRF);
assign GHRNextD = FlushD ? GHRNextE : {DirPredictionF[1], GHRF};
flopenr #(k+1) GHRDReg(clk, reset, (~StallD) | FlushD, GHRNextD, GHRD);
assign GHRNextE = FlushE ? GHRNextM : GHRD;
flopenr #(k+1) GHREReg(clk, reset, (~StallE) | FlushE, GHRNextE, OldGHRE);
assign GHRE = BranchInstrE ? {PCSrcE, OldGHRE[k-1:0]} : OldGHRE;
assign GHRNextM = FlushM ? GHRNextW : GHRE;
flopenr #(k+1) GHRMReg(clk, reset, (~StallM) | FlushM, GHRNextM, GHRM);
assign GHRNextW = FlushW ? GHRW : GHRM;
flopenr #(k+1) GHRWReg(clk, reset, (BranchInstrM & ~StallW) | FlushW, GHRNextW, GHRW);
assign DirPredictionWrongE = PCSrcE != DirPredictionE[1] & BranchInstrE;
flopenr #(`XLEN) PCWReg(clk, reset, ~StallW, PCM, PCW);
endmodule

136
src/ifu/bpred/speculativeglobalhistory.sv
View File

@ -1,136 +0,0 @@
///////////////////////////////////////////
// speculativeglobalhistory.sv
//
// Written: Shreya Sanghai
// Email: ssanghai@hmc.edu
// Created: March 16, 2021
// Modified:
//
// Purpose: Global History Branch predictor with parameterized global history register
//
// A component of the CORE-V-WALLY configurable RISC-V project.
//
// Copyright (C) 2021-23 Harvey Mudd College & Oklahoma State University
//
// SPDX-License-Identifier: Apache-2.0 WITH SHL-2.1
//
// Licensed under the Solderpad Hardware License v 2.1 (the “License”); you may not use this file
// except in compliance with the License, or, at your option, the Apache License version 2.0. You
// may obtain a copy of the License at
//
// https://solderpad.org/licenses/SHL-2.1/
//
// Unless required by applicable law or agreed to in writing, any work distributed under the
// License is distributed on an “AS IS” BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND,
// either express or implied. See the License for the specific language governing permissions
// and limitations under the License.
////////////////////////////////////////////////////////////////////////////////////////////////
`include "wally-config.vh"
module speculativeglobalhistory #(parameter int k = 10 ) (
input logic clk,
input logic reset,
input logic StallF, StallD, StallE, StallM, StallW,
input logic FlushD, FlushE, FlushM, FlushW,
output logic [1:0] DirPredictionF,
output logic DirPredictionWrongE,
// update
input logic [3:0] PredInstrClassF,
input logic [3:0] InstrClassD, InstrClassE, InstrClassM,
input logic [3:0] WrongPredInstrClassD,
input logic PCSrcE
);
logic MatchF, MatchD, MatchE;
logic MatchNextX, MatchXF;
logic [1:0] TableDirPredictionF, DirPredictionD, DirPredictionE;
logic [1:0] NewDirPredictionE;
logic [k-1:0] GHRF, GHRD, GHRE, GHRM;
logic GHRLastF;
logic [k-1:0] GHRNextF, GHRNextD, GHRNextE, GHRNextM;
logic [k-1:0] IndexNextF, IndexF, IndexD, IndexE;
logic [1:0] ForwardNewDirPrediction, ForwardDirPredictionF;
logic FlushDOrDirWrong;
assign IndexNextF = GHRNextF;
assign IndexF = GHRF ;
assign IndexD = GHRD[k-1:0] ;
assign IndexE = GHRE[k-1:0] ;
ram2p1r1wbe #(2**k, 2) PHT(.clk(clk),
.ce1(~StallF | reset), .ce2(~StallM & ~FlushM),
.ra1(IndexNextF),
.rd1(TableDirPredictionF),
.wa2(IndexE),
.wd2(NewDirPredictionE),
.we2(InstrClassE[0]),
.bwe2(1'b1));
// if there are non-flushed branches in the pipeline we need to forward the prediction from that stage to the NextF demi stage
// and then register for use in the Fetch stage.
assign MatchF = PredInstrClassF[0] & ~FlushD & (IndexNextF == IndexF);
assign MatchD = InstrClassD[0] & ~FlushE & (IndexNextF == IndexD);
assign MatchE = InstrClassE[0] & ~FlushM & (IndexNextF == IndexE);
assign MatchNextX = MatchF | MatchD | MatchE;
flopenr #(1) MatchReg(clk, reset, ~StallF, MatchNextX, MatchXF);
assign ForwardNewDirPrediction = MatchF ? {2{DirPredictionF[1]}} :
MatchD ? {2{DirPredictionD[1]}} :
NewDirPredictionE ;
flopenr #(2) ForwardDirPredicitonReg(clk, reset, ~StallF, ForwardNewDirPrediction, ForwardDirPredictionF);
assign DirPredictionF = MatchXF ? ForwardDirPredictionF : TableDirPredictionF;
// DirPrediction pipeline
flopenr #(2) PredictionRegD(clk, reset, ~StallD, DirPredictionF, DirPredictionD);
flopenr #(2) PredictionRegE(clk, reset, ~StallE, DirPredictionD, DirPredictionE);
satCounter2 BPDirUpdateE(.BrDir(PCSrcE), .OldState(DirPredictionE), .NewState(NewDirPredictionE));
// GHR pipeline
// If Fetch has a branch, speculatively insert prediction into the GHR
// If the front end is flushed or the direction prediction is wrong, reset to
// most recent valid GHR. For a BP wrong this is GHRD with the correct prediction shifted in.
// For FlushE this is GHRE. GHRNextE is both.
logic [k-1:0] GHRClassWrongAndPredF;
logic [k-1:0] GHRClassWrong;
mux2 #(k) GHRClassWrongAndPredFMux({DirPredictionF[1], DirPredictionD[1], GHRF[k-1:2]}, {DirPredictionF[1], GHRF[k-2:0]}, ~InstrClassD[0], GHRClassWrongAndPredF);
assign FlushDOrDirWrong = FlushD | DirPredictionWrongE;
// mux3 #(k) GHRFMux(GHRF, {DirPredictionF[1], GHRF[k-1:1]}, GHRNextE[k-1:0],
// {FlushDOrDirWrong, PredInstrClassF[0]}, GHRNextF);
mux5 #(k) GHRFMux(GHRF, {DirPredictionF[1], GHRF[k-1:1]}, GHRClassWrong, GHRClassWrongAndPredF, GHRNextE[k-1:0],
{FlushDOrDirWrong, WrongPredInstrClassD[0], PredInstrClassF[0]}, GHRNextF);
// Need 1 extra bit to store the shifted out GHRF if repair needs to back shift.
flopenr #(k) GHRFReg(clk, reset, ~StallF | FlushDOrDirWrong, GHRNextF, GHRF);
flopenr #(1) GHRFLastReg(clk, reset, ~StallF | FlushDOrDirWrong, GHRF[0], GHRLastF);
// With instruction class prediction, the class could be wrong and is checked in Decode.
// If it is wrong and branch does exist then shift right and insert the prediction.
// If the branch does not exist then shift left and use GHRLastF to restore the LSB.
mux2 #(k) GHRClassWrongMux({DirPredictionD[1], GHRF[k-1:1]}, {GHRF[k-2:0], GHRLastF}, ~InstrClassD[0], GHRClassWrong);
// As with GHRF FlushD and wrong direction prediction flushes the pipeline and restores to GHRNextE.
mux3 #(k) GHRDMux(GHRF, GHRClassWrong, GHRNextE, {FlushDOrDirWrong, WrongPredInstrClassD[0]}, GHRNextD);
flopenr #(k) GHRDReg(clk, reset, ~StallD | FlushDOrDirWrong, GHRNextD, GHRD);
mux3 #(k) GHREMux(GHRD, GHRNextM, {PCSrcE, GHRD[k-2:0]}, {InstrClassE[0] & ~FlushM, FlushE}, GHRNextE);
flopenr #(k) GHREReg(clk, reset, (~StallE) | FlushE, GHRNextE, GHRE);
assign GHRNextM = FlushM ? GHRM : GHRE;
flopenr #(k) GHRMReg(clk, reset, (InstrClassM[0] & ~StallM) | FlushM, GHRNextM, GHRM);
assign DirPredictionWrongE = PCSrcE != DirPredictionE[1] & InstrClassE[0];
endmodule

137
src/ifu/bpred/speculativegshare.sv
View File

@ -1,137 +0,0 @@
///////////////////////////////////////////
// gsharePredictor.sv
//
// Written: Shreya Sanghai
// Email: ssanghai@hmc.edu
// Created: March 16, 2021
// Modified:
//
// Purpose: Global History Branch predictor with parameterized global history register
//
// A component of the CORE-V-WALLY configurable RISC-V project.
//
// Copyright (C) 2021-23 Harvey Mudd College & Oklahoma State University
//
// SPDX-License-Identifier: Apache-2.0 WITH SHL-2.1
//
// Licensed under the Solderpad Hardware License v 2.1 (the “License”); you may not use this file
// except in compliance with the License, or, at your option, the Apache License version 2.0. You
// may obtain a copy of the License at
//
// https://solderpad.org/licenses/SHL-2.1/
//
// Unless required by applicable law or agreed to in writing, any work distributed under the
// License is distributed on an “AS IS” BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND,
// either express or implied. See the License for the specific language governing permissions
// and limitations under the License.
////////////////////////////////////////////////////////////////////////////////////////////////
`include "wally-config.vh"
module speculativegshare #(parameter int k = 10 ) (
input logic clk,
input logic reset,
input logic StallF, StallD, StallE, StallM, StallW,
input logic FlushD, FlushE, FlushM, FlushW,
output logic [1:0] DirPredictionF,
output logic DirPredictionWrongE,
// update
input logic [`XLEN-1:0] PCNextF, PCF, PCD, PCE,
input logic [3:0] PredInstrClassF,
input logic [3:0] InstrClassD, InstrClassE, InstrClassM,
input logic [3:0] WrongPredInstrClassD,
input logic PCSrcE
);
logic MatchF, MatchD, MatchE;
logic MatchNextX, MatchXF;
logic [1:0] TableDirPredictionF, DirPredictionD, DirPredictionE;
logic [1:0] NewDirPredictionE;
logic [k-1:0] GHRF, GHRD, GHRE, GHRM;
logic GHRLastF;
logic [k-1:0] GHRNextF, GHRNextD, GHRNextE, GHRNextM;
logic [k-1:0] IndexNextF, IndexF, IndexD, IndexE;
logic [1:0] ForwardNewDirPrediction, ForwardDirPredictionF;
logic FlushDOrDirWrong;
assign IndexNextF = GHRNextF ^ {PCNextF[k+1] ^ PCNextF[1], PCNextF[k:2]};
assign IndexF = GHRF ^ {PCF[k+1] ^ PCF[1], PCF[k:2]};
assign IndexD = GHRD[k-1:0] ^ {PCD[k+1] ^ PCD[1], PCD[k:2]};
assign IndexE = GHRE[k-1:0] ^ {PCE[k+1] ^ PCE[1], PCE[k:2]};
ram2p1r1wbe #(2**k, 2) PHT(.clk(clk),
.ce1(~StallF | reset), .ce2(~StallM & ~FlushM),
.ra1(IndexNextF),
.rd1(TableDirPredictionF),
.wa2(IndexE),
.wd2(NewDirPredictionE),
.we2(InstrClassE[0]),
.bwe2(1'b1));
// if there are non-flushed branches in the pipeline we need to forward the prediction from that stage to the NextF demi stage
// and then register for use in the Fetch stage.
assign MatchF = PredInstrClassF[0] & ~FlushD & (IndexNextF == IndexF);
assign MatchD = InstrClassD[0] & ~FlushE & (IndexNextF == IndexD);
assign MatchE = InstrClassE[0] & ~FlushM & (IndexNextF == IndexE);
assign MatchNextX = MatchF | MatchD | MatchE;
flopenr #(1) MatchReg(clk, reset, ~StallF, MatchNextX, MatchXF);
assign ForwardNewDirPrediction = MatchF ? {2{DirPredictionF[1]}} :
MatchD ? {2{DirPredictionD[1]}} :
NewDirPredictionE ;
flopenr #(2) ForwardDirPredicitonReg(clk, reset, ~StallF, ForwardNewDirPrediction, ForwardDirPredictionF);
assign DirPredictionF = MatchXF ? ForwardDirPredictionF : TableDirPredictionF;
// DirPrediction pipeline
flopenr #(2) PredictionRegD(clk, reset, ~StallD, DirPredictionF, DirPredictionD);
flopenr #(2) PredictionRegE(clk, reset, ~StallE, DirPredictionD, DirPredictionE);
satCounter2 BPDirUpdateE(.BrDir(PCSrcE), .OldState(DirPredictionE), .NewState(NewDirPredictionE));
// GHR pipeline
// If Fetch has a branch, speculatively insert prediction into the GHR
// If the front end is flushed or the direction prediction is wrong, reset to
// most recent valid GHR. For a BP wrong this is GHRD with the correct prediction shifted in.
// For FlushE this is GHRE. GHRNextE is both.
logic [k-1:0] GHRClassWrongAndPredF;
logic [k-1:0] GHRClassWrong;
mux2 #(k) GHRClassWrongAndPredFMux({DirPredictionF[1], DirPredictionD[1], GHRF[k-1:2]}, {DirPredictionF[1], GHRF[k-2:0]}, ~InstrClassD[0], GHRClassWrongAndPredF);
assign FlushDOrDirWrong = FlushD | DirPredictionWrongE;
// mux3 #(k) GHRFMux(GHRF, {DirPredictionF[1], GHRF[k-1:1]}, GHRNextE[k-1:0],
// {FlushDOrDirWrong, PredInstrClassF[0]}, GHRNextF);
mux5 #(k) GHRFMux(GHRF, {DirPredictionF[1], GHRF[k-1:1]}, GHRClassWrong, GHRClassWrongAndPredF, GHRNextE[k-1:0],
{FlushDOrDirWrong, WrongPredInstrClassD[0], PredInstrClassF[0]}, GHRNextF);
// Need 1 extra bit to store the shifted out GHRF if repair needs to back shift.
flopenr #(k) GHRFReg(clk, reset, ~StallF | FlushDOrDirWrong, GHRNextF, GHRF);
flopenr #(1) GHRFLastReg(clk, reset, ~StallF | FlushDOrDirWrong, GHRF[0], GHRLastF);
// With instruction class prediction, the class could be wrong and is checked in Decode.
// If it is wrong and branch does exist then shift right and insert the prediction.
// If the branch does not exist then shift left and use GHRLastF to restore the LSB.
mux2 #(k) GHRClassWrongMux({DirPredictionD[1], GHRF[k-1:1]}, {GHRF[k-2:0], GHRLastF}, ~InstrClassD[0], GHRClassWrong);
// As with GHRF FlushD and wrong direction prediction flushes the pipeline and restores to GHRNextE.
mux3 #(k) GHRDMux(GHRF, GHRClassWrong, GHRNextE, {FlushDOrDirWrong, WrongPredInstrClassD[0]}, GHRNextD);
flopenr #(k) GHRDReg(clk, reset, ~StallD | FlushDOrDirWrong, GHRNextD, GHRD);
mux3 #(k) GHREMux(GHRD, GHRNextM, {PCSrcE, GHRD[k-2:0]}, {InstrClassE[0] & ~FlushM, FlushE}, GHRNextE);
flopenr #(k) GHREReg(clk, reset, (~StallE) | FlushE, GHRNextE, GHRE);
assign GHRNextM = FlushM ? GHRM : GHRE;
flopenr #(k) GHRMReg(clk, reset, (InstrClassM[0] & ~StallM) | FlushM, GHRNextM, GHRM);
assign DirPredictionWrongE = PCSrcE != DirPredictionE[1] & InstrClassE[0];
endmodule