From 35a59a11932cc550663fe186c314eb626b96fc6d Mon Sep 17 00:00:00 2001
From: Ross Thompson <ross1728@gmail.com>
Date: Wed, 3 May 2023 12:52:32 -0500
Subject: [PATCH] I think ahead pipelining is working for local history.

---
 bin/parseHPMC.py              |   7 ++-
 sim/bpred-sim.py              |  33 +++++++---
 src/ifu/bpred/bpred.sv        |   9 ++-
 src/ifu/bpred/localaheadbp.sv | 111 ++++++++++++++++++++++++++++++++++
 src/ifu/bpred/localbpbasic.sv |   8 +--
 testbench/testbench.sv        |  17 +++++-
 6 files changed, 166 insertions(+), 19 deletions(-)
 create mode 100644 src/ifu/bpred/localaheadbp.sv

diff --git a/bin/parseHPMC.py b/bin/parseHPMC.py
index 5b5e0d98b..7d65f120c 100755
--- a/bin/parseHPMC.py
+++ b/bin/parseHPMC.py
@@ -279,12 +279,13 @@ if(sys.argv[1] == '-b'):
                     dct[PredType] = (currSize, currPercent)
         print(dct)
         fig, axes = plt.subplots()
-        marker={'twobit' : '^', 'gshare' : 'o', 'global' : 's', 'gshareBasic' : '*', 'globalBasic' : 'x', 'btb': 'x', 'twobitCModel' : 'x', 'gshareCModel' : '*'}
-        colors={'twobit' : 'black', 'gshare' : 'blue', 'global' : 'dodgerblue', 'gshareBasic' : 'turquoise', 'globalBasic' : 'lightsteelblue', 'btb' : 'blue', 'twobitCModel' : 'gray', 'gshareCModel' : 'dodgerblue'}
+        marker={'twobit' : '^', 'gshare' : 'o', 'global' : 's', 'gshareBasic' : '*', 'globalBasic' : 'x', 'btb': 'x', 'twobitCModel' : 'x', 'gshareCModel' : '*', 'tenlocal' : '.', 'eightlocal' : ',', 'fourlocal' : 'x'}
+        colors={'twobit' : 'black', 'gshare' : 'blue', 'global' : 'dodgerblue', 'gshareBasic' : 'turquoise', 'globalBasic' : 'lightsteelblue', 'btb' : 'blue', 'twobitCModel' : 'gray', 'gshareCModel' : 'dodgerblue', 'tenlocal' : 'lightblue', 'eightlocal' : 'lightblue', 'fourlocal' : 'lightblue'}
         for cat in dct:
             (x, y) = dct[cat]
             x=[int(2**int(v)) for v in x]
-            print(x, y)
+            #print(x, y)
+            print(cat)
             axes.plot(x,y, color=colors[cat])
             axes.scatter(x,y, label=cat, marker=marker[cat], color=colors[cat])
             #plt.scatter(x, y, label=cat)
diff --git a/sim/bpred-sim.py b/sim/bpred-sim.py
index 60574e371..256658a6c 100755
--- a/sim/bpred-sim.py
+++ b/sim/bpred-sim.py
@@ -46,18 +46,33 @@ configs = [
     )
 ]
 
+# bpdSize = [6, 8, 10, 12, 14, 16]
+# bpdType = ['twobit', 'gshare', 'global', 'gshare_basic', 'global_basic', 'local_basic']
+# for CurrBPType in bpdType:
+#     for CurrBPSize in bpdSize:
+#         name = CurrBPType+str(CurrBPSize)
+#         configOptions = "+define+INSTR_CLASS_PRED=0 +define+BPRED_TYPE=\"BP_" + CurrBPType.upper() + "\" +define+BPRED_SIZE=" + str(CurrBPSize)
+#         tc = TestCase(
+#             name=name,
+#             variant="rv32gc",
+#             cmd="vsim > {} -c <<!\ndo wally-batch.do  rv32gc configOptions " + name + " embench " + configOptions,
+#             grepstr="")
+#         configs.append(tc)
+
 bpdSize = [6, 8, 10, 12, 14, 16]
-bpdType = ['twobit', 'gshare', 'global', 'gshare_basic', 'global_basic']
+LHRSize = [4, 8, 10]
+bpdType = ['local_basic']
 for CurrBPType in bpdType:
     for CurrBPSize in bpdSize:
-        name = CurrBPType+str(CurrBPSize)
-        configOptions = "+define+INSTR_CLASS_PRED=0 +define+BPRED_TYPE=\"BP_" + CurrBPType.upper() + "\" +define+BPRED_SIZE=" + str(CurrBPSize)
-        tc = TestCase(
-            name=name,
-            variant="rv32gc",
-            cmd="vsim > {} -c <<!\ndo wally-batch.do  rv32gc configOptions " + name + " embench " + configOptions,
-            grepstr="")
-        configs.append(tc)
+        for CurrLHRSize in  LHRSize:
+            name = str(CurrLHRSize)+CurrBPType+str(CurrBPSize)
+            configOptions = "+define+INSTR_CLASS_PRED=0 +define+BPRED_TYPE=\"BP_" + CurrBPType.upper() + "\" +define+BPRED_SIZE=" + str(CurrBPSize) + " +define+BPRED_NUM_LHR=" + str(CurrLHRSize) + " "
+            tc = TestCase(
+                name=name,
+                variant="rv32gc",
+                cmd="vsim > {} -c <<!\ndo wally-batch.do  rv32gc configOptions " + name + " embench " + configOptions,
+                grepstr="")
+            configs.append(tc)
 
 import os
 from multiprocessing import Pool, TimeoutError
diff --git a/src/ifu/bpred/bpred.sv b/src/ifu/bpred/bpred.sv
index c3ac87232..2034851bc 100644
--- a/src/ifu/bpred/bpred.sv
+++ b/src/ifu/bpred/bpred.sv
@@ -128,8 +128,13 @@ module bpred (
       .BranchE, .BranchM, .PCSrcE);
   
   end else if (`BPRED_TYPE == "BP_LOCAL_BASIC") begin:Predictor
-    localbpbasic #(`BPRED_NUM_LHR, `BPRED_SIZE)
-DirPredictor(.clk, .reset, .StallF, .StallD, .StallE, .StallM, .StallW, .FlushD, .FlushE, .FlushM, .FlushW,
+    localbpbasic #(`BPRED_NUM_LHR, `BPRED_SIZE) DirPredictor(.clk, .reset, 
+      .StallF, .StallD, .StallE, .StallM, .StallW, .FlushD, .FlushE, .FlushM, .FlushW,
+      .PCNextF, .PCM, .BPDirPredF, .BPDirPredWrongE,
+      .BranchE, .BranchM, .PCSrcE);
+  end else if (`BPRED_TYPE == "BP_LOCAL_AHEAD") begin:Predictor
+    localaheadbp #(`BPRED_NUM_LHR, `BPRED_SIZE) DirPredictor(.clk, .reset, 
+      .StallF, .StallD, .StallE, .StallM, .StallW, .FlushD, .FlushE, .FlushM, .FlushW,
       .PCNextF, .PCM, .BPDirPredF, .BPDirPredWrongE,
       .BranchE, .BranchM, .PCSrcE);
   end 
diff --git a/src/ifu/bpred/localaheadbp.sv b/src/ifu/bpred/localaheadbp.sv
new file mode 100644
index 000000000..a0bc12fe3
--- /dev/null
+++ b/src/ifu/bpred/localaheadbp.sv
@@ -0,0 +1,111 @@
+///////////////////////////////////////////
+// gsharebasic.sv
+//
+// Written: Ross Thompson
+// Email: ross1728@gmail.com
+// Created: 16 March 2021
+// 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
+// 
+// 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 localaheadbp #(parameter m = 6, // 2^m = number of local history branches 
+                      parameter k = 10) ( // number of past branches stored
+  input logic             clk,
+  input logic             reset,
+  input logic             StallF, StallD, StallE, StallM, StallW,
+  input logic             FlushD, FlushE, FlushM, FlushW,
+  output logic [1:0]      BPDirPredF, 
+  output logic            BPDirPredWrongE,
+  // update
+  input logic [`XLEN-1:0] PCNextF, PCM,
+  input logic             BranchE, BranchM, PCSrcE
+);
+
+  logic [k-1:0]           IndexNextF, IndexM;
+  logic [1:0]             BPDirPredD, BPDirPredE;
+  logic [1:0]             NewBPDirPredE, NewBPDirPredM;
+
+  logic [k-1:0]           LHRF, LHRD, LHRE, LHRM, LHRNextF;
+  logic [k-1:0]           LHRNextW;
+  logic                   PCSrcM;
+  logic [2**m-1:0][k-1:0] LHRArray;
+  logic [m-1:0]           IndexLHRNextF, IndexLHRM;
+  logic [`XLEN-1:0]       PCW;
+  
+  
+  logic                    UpdateM;
+
+  //assign IndexNextF = LHR;
+  assign IndexM = LHRM;
+  
+  ram2p1r1wbe #(2**k, 2) PHT(.clk(clk),
+    .ce1(~StallF), .ce2(~StallW & ~FlushW),
+    .ra1(LHRNextF),
+    .rd1(BPDirPredF),
+    .wa2(IndexM),
+    .wd2(NewBPDirPredM),
+    .we2(BranchM),
+    .bwe2(1'b1));
+
+  flopenrc #(2) PredictionRegD(clk, reset,  FlushD, ~StallD, BPDirPredF, BPDirPredD);
+  flopenrc #(2) PredictionRegE(clk, reset,  FlushE, ~StallE, BPDirPredD, BPDirPredE);
+
+  satCounter2 BPDirUpdateE(.BrDir(PCSrcE), .OldState(BPDirPredE), .NewState(NewBPDirPredE));
+  flopenrc #(2) NewPredictionRegM(clk, reset,  FlushM, ~StallM, NewBPDirPredE, NewBPDirPredM);
+
+  assign BPDirPredWrongE = PCSrcE != BPDirPredE[1] & BranchE;
+
+  // This is the main difference between global and local history basic implementations. In global, 
+  // the ghr wraps back into itself directly without
+  // being pipelined.  I.E. GHR is not read in F and then pipelined to M where it is updated.  Instead
+  // GHR is both read and update in M.  GHR is still pipelined so that the PHT is updated with the correct
+  // GHR.  Local history in contrast must pipeline the specific history register read during F and then update
+  // that same one in M.  This implementation does not forward if a branch matches in the D, E, or M stages.
+  assign LHRNextW = BranchM ? {PCSrcM, LHRM[k-1:1]} : LHRM;
+
+  // this is local history
+  //genvar      index;
+  //assign UpdateM = BranchM & ~StallW & ~FlushW;
+  assign IndexLHRM = {PCW[m+1] ^ PCW[1], PCW[m:2]};
+  assign IndexLHRNextF = {PCNextF[m+1] ^ PCNextF[1], PCNextF[m:2]};
+
+  ram2p1r1wbe #(2**m, k) BHT(.clk(clk),
+    .ce1(~StallF), .ce2(~StallW & ~FlushW),
+    .ra1(IndexLHRNextF),
+    .rd1(LHRNextF),
+    .wa2(IndexLHRM),
+    .wd2(LHRNextW),
+    .we2(BranchM),
+    .bwe2('1));  
+
+  flopenrc #(1) PCSrcMReg(clk, reset, FlushM, ~StallM, PCSrcE, PCSrcM);
+    
+  flopenrc #(k) LHRFReg(clk, reset, FlushD, ~StallF, LHRNextF, LHRF);
+  flopenrc #(k) LHRDReg(clk, reset, FlushD, ~StallD, LHRF, LHRD);
+  flopenrc #(k) LHREReg(clk, reset, FlushE, ~StallE, LHRD, LHRE);
+  flopenrc #(k) LHRMReg(clk, reset, FlushM, ~StallM, LHRE, LHRM);
+
+  flopenr #(`XLEN) PCWReg(clk, reset, ~StallW, PCM, PCW);
+
+endmodule
diff --git a/src/ifu/bpred/localbpbasic.sv b/src/ifu/bpred/localbpbasic.sv
index de0d94414..a874dfce2 100644
--- a/src/ifu/bpred/localbpbasic.sv
+++ b/src/ifu/bpred/localbpbasic.sv
@@ -74,10 +74,10 @@ module localbpbasic #(parameter m = 6, // 2^m = number of local history branches
 
   assign BPDirPredWrongE = PCSrcE != BPDirPredE[1] & BranchE;
 
-  // this is the main difference between global and local history basic implementations rather than
-  // having multiple history registers.  In global, the ghr wraps back into itself directly without
-  // being pipelined.  IE. GHR is not read in F and then pipelined to M where it is updated.  Instead
-  // GHR is just read in M and updated.  GHR is still pipelined so that the PHT is updated with the correct
+  // This is the main difference between global and local history basic implementations. In global, 
+  // the ghr wraps back into itself directly without
+  // being pipelined.  I.E. GHR is not read in F and then pipelined to M where it is updated.  Instead
+  // GHR is both read and update in M.  GHR is still pipelined so that the PHT is updated with the correct
   // GHR.  Local history in contrast must pipeline the specific history register read during F and then update
   // that same one in M.  This implementation does not forward if a branch matches in the D, E, or M stages.
   assign LHRNextW = BranchM ? {PCSrcM, LHRM[k-1:1]} : LHRM;
diff --git a/testbench/testbench.sv b/testbench/testbench.sv
index 9021b6448..0112c24bd 100644
--- a/testbench/testbench.sv
+++ b/testbench/testbench.sv
@@ -28,7 +28,7 @@
 `include "wally-config.vh"
 `include "tests.vh"
 
-`define PrintHPMCounters 0
+`define PrintHPMCounters 1
 `define BPRED_LOGGER 0
 `define I_CACHE_ADDR_LOGGER 0
 `define D_CACHE_ADDR_LOGGER 0
@@ -536,6 +536,21 @@ module testbench;
   if (`BPRED_SUPPORTED) begin
     integer adrindex;
 
+    // local history only
+    if (`BPRED_TYPE == "BP_LOCAL_AHEAD") begin
+      always @(*) begin
+        if(reset) begin
+          for(adrindex = 0; adrindex < 2**`BPRED_NUM_LHR; adrindex++) begin
+            force dut.core.ifu.bpred.bpred.Predictor.DirPredictor.BHT.mem[adrindex] = 0;
+          end
+            #1;
+          for(adrindex = 0; adrindex < 2**`BPRED_NUM_LHR; adrindex++) begin
+            release dut.core.ifu.bpred.bpred.Predictor.DirPredictor.BHT.mem[adrindex];
+          end
+        end
+      end
+    end
+
     always @(*) begin
       if(reset) begin
         for(adrindex = 0; adrindex < 2**`BTB_SIZE; adrindex++) begin