Actually fixed non-power of 2 issue with RAS.

Added RAS swapping to branch predictor scripts and configurations.

bin/parseTest.py

@@ -32,6 +32,10 @@ import math
 import numpy as np
 import argparse
 
+RefData = [('twobitCModel6', 'twobitCModel', 64, 9.65280765420711), ('twobitCModel8', 'twobitCModel', 256, 8.75120245829945), ('twobitCModel10', 'twobitCModel', 1024, 8.1318382397263),
+           ('twobitCModel12', 'twobitCModel', 4096, 7.53026646633342), ('twobitCModel14', 'twobitCModel', 16384, 6.07679338544009), ('twobitCModel16', 'twobitCModel', 65536, 6.07679338544009),
+           ('gshareCModel6', 'gshareCModel', 64, 10.6602835418646), ('gshareCModel8', 'gshareCModel', 256, 8.38384710559667), ('gshareCModel10', 'gshareCModel', 1024, 6.36847432155534),
+           ('gshareCModel12', 'gshareCModel', 4096, 3.91108491151983), ('gshareCModel14', 'gshareCModel', 16384, 2.83926519215395), ('gshareCModel16', 'gshareCModel', 65536, .60213659066941)]
 
 def ParseBranchListFile(path):
     '''Take the path to the list of Questa Sim log files containing the performance counters outputs.  File
@@ -65,7 +69,7 @@ def ProcessFile(fileName):
             HPMClist = { }
         elif(len(lineToken) > 4 and lineToken[1][0:3] == 'Cnt'):
             countToken = line.split('=')[1].split()
-            value = int(countToken[0])
+            value = int(countToken[0]) if countToken[0] != 'x' else 0
             name = ' '.join(countToken[1:])
             HPMClist[name] = value
         elif ('is done' in line):
@@ -111,7 +115,7 @@ def ComputeGeometricAverage(benchmarks):
     benchmarks.append(('Mean', '', AllAve))
 
 def GenerateName(predictorType, predictorParams):
-    if(predictorType == 'gshare' or  predictorType == 'twobit'):
+    if(predictorType == 'gshare' or  predictorType == 'twobit' or predictorType == 'btb' or predictorType == 'class'):
         return predictorType + predictorParams[0]
     elif(predictorParams == 'local'):
         return predictorType + predictorParams[0] + '_' + predictorParams[1]
@@ -120,7 +124,7 @@ def GenerateName(predictorType, predictorParams):
         sys.exit(-1)
 
 def ComputePredNumEntries(predictorType, predictorParams):
-    if(predictorType == 'gshare' or  predictorType == 'twobit'):
+    if(predictorType == 'gshare' or  predictorType == 'twobit' or predictorType == 'btb' or predictorType == 'class'):
         return 2**int(predictorParams[0])
     elif(predictorParams == 'local'):
         return 2**int(predictorParams[0]) * int(predictorParams[1]) + 2**int(predictorParams[1])
@@ -286,7 +290,7 @@ def ReportAsGraph(benchmarkDict, bar):
               'ClassMPR': 'Class Misprediction'}
     if(args.summary):
         markers = ['x', '.', '+', '*', '^', 'o', ',', 's']
-        colors = ['black', 'blue', 'dodgerblue', 'turquoise', 'lightsteelblue', 'gray', 'black', 'blue']
+        colors = ['blue', 'black', 'dodgerblue', 'gray', 'lightsteelblue', 'turquoise', 'black', 'blue']
         temp = benchmarkDict['Mean']
 
         # the benchmarkDict['Mean'] contains sequencies of results for multiple
@@ -429,6 +433,7 @@ performanceCounterList = BuildDataBase(predictorLogs)         # builds a databas
 benchmarkFirstList = ReorderDataBase(performanceCounterList)  # reorder first by benchmark then trace
 benchmarkDict = ExtractSelectedData(benchmarkFirstList)       # filters to just the desired performance counter metric
 
+if(args.reference): benchmarkDict['Mean'].extend(RefData)
 #print(benchmarkDict['Mean'])
 #print(benchmarkDict['aha-mont64Speed'])
 #print(benchmarkDict)

config/buildroot/config.vh

@@ -142,6 +142,7 @@ localparam BPRED_TYPE = `BP_GSHARE; // BP_GSHARE_BASIC, BP_GLOBAL, BP_GLOBAL_BAS
 localparam BPRED_SIZE = 32'd10;
 localparam BPRED_NUM_LHR = 32'd6;
 localparam BTB_SIZE = 32'd10;
+localparam RAS_SIZE = 32'd16;
 
 
 localparam SVADU_SUPPORTED = 1;

config/fpga/config.vh

@@ -156,6 +156,7 @@ localparam BPRED_TYPE = `BP_GSHARE; // BP_GSHARE_BASIC, BP_GLOBAL, BP_GLOBAL_BAS
 localparam BPRED_NUM_LHR = 32'd6;
 localparam BPRED_SIZE = 32'd12;
 localparam BTB_SIZE = 32'd10;
+localparam RAS_SIZE = 32'd16;
 
 localparam SVADU_SUPPORTED = 1;
 localparam ZMMUL_SUPPORTED = 0;

config/rv32e/config.vh

@@ -144,6 +144,7 @@ localparam BPRED_TYPE = `BP_GSHARE; // BP_GSHARE_BASIC, BP_GLOBAL, BP_GLOBAL_BAS
 localparam BPRED_SIZE = 32'd10;
 localparam BPRED_NUM_LHR = 32'd6;
 localparam BTB_SIZE = 32'd10;
+localparam RAS_SIZE = 32'd16;
 
 localparam SVADU_SUPPORTED = 0;
 localparam ZMMUL_SUPPORTED = 0;

config/rv32gc/config.vh

@@ -150,7 +150,13 @@ localparam BPRED_TYPE = `BP_GSHARE; // BP_GSHARE_BASIC, BP_GLOBAL, BP_GLOBAL_BAS
 localparam BPRED_SIZE = 32'd10;
 `endif
 localparam BPRED_NUM_LHR = 32'd6;
+`ifdef BTB_OVERRIDE
+localparam BTB_SIZE = `BTB_SIZE;
+localparam RAS_SIZE = `RAS_SIZE;
+`else
 localparam BTB_SIZE = 32'd10;
+localparam RAS_SIZE = 32'd16;
+`endif
 
 localparam SVADU_SUPPORTED = 1;
 localparam ZMMUL_SUPPORTED = 0;

config/rv32i/config.vh

@@ -144,6 +144,7 @@ localparam BPRED_TYPE = `BP_GSHARE; // BP_GSHARE_BASIC, BP_GLOBAL, BP_GLOBAL_BAS
 localparam BPRED_SIZE = 32'd10;
 localparam BPRED_NUM_LHR = 32'd6;
 localparam BTB_SIZE = 32'd10;
+localparam RAS_SIZE = 32'd16;
 
 localparam SVADU_SUPPORTED = 0;
 localparam ZMMUL_SUPPORTED = 0;

config/rv32imc/config.vh

@@ -143,6 +143,7 @@ localparam BPRED_TYPE = `BP_GSHARE; // BP_GSHARE_BASIC, BP_GLOBAL, BP_GLOBAL_BAS
 localparam BPRED_SIZE = 32'd10;
 localparam BPRED_NUM_LHR = 32'd6;
 localparam BTB_SIZE = 32'd10;
+localparam RAS_SIZE = 32'd16;
 
 localparam SVADU_SUPPORTED = 0;
 localparam ZMMUL_SUPPORTED = 0;

config/rv64fpquad/config.vh

@@ -146,6 +146,7 @@ localparam BPRED_TYPE = `BP_GSHARE; // BP_GSHARE_BASIC, BP_GLOBAL, BP_GLOBAL_BAS
 localparam BPRED_SIZE = 32'd10;
 localparam BPRED_NUM_LHR = 32'd6;
 localparam BTB_SIZE = 32'd10;
+localparam RAS_SIZE = 32'd16;
 
 localparam SVADU_SUPPORTED = 0;
 localparam ZMMUL_SUPPORTED = 0;

config/rv64gc/config.vh

@@ -147,8 +147,9 @@ localparam PLIC_SDC_ID = 32'd9;
 localparam BPRED_SUPPORTED = 1;
 localparam BPRED_TYPE = `BP_GSHARE; // BP_GSHARE_BASIC, BP_GLOBAL, BP_GLOBAL_BASIC, BP_TWOBIT
 localparam BPRED_NUM_LHR = 32'd6;
-localparam BPRED_SIZE = 32'd10;
+localparam BPRED_SIZE = 32'd6;
 localparam BTB_SIZE = 32'd10;
+localparam RAS_SIZE = 32'd16;
 
 localparam SVADU_SUPPORTED = 1;
 localparam ZMMUL_SUPPORTED = 0;

config/rv64i/config.vh

@@ -146,6 +146,7 @@ localparam BPRED_TYPE = `BP_GSHARE; // BP_GSHARE_BASIC, BP_GLOBAL, BP_GLOBAL_BAS
 localparam BPRED_SIZE = 32'd10;
 localparam BPRED_NUM_LHR = 32'd6;
 localparam BTB_SIZE = 32'd10;
+localparam RAS_SIZE = 32'd16;
 
 localparam SVADU_SUPPORTED = 0;
 localparam ZMMUL_SUPPORTED = 0;

config/shared/parameter-defs.vh

@@ -89,6 +89,7 @@ localparam cvw_t P = '{
   BPRED_SIZE :        BPRED_SIZE,
   BPRED_NUM_LHR : BPRED_NUM_LHR,                       
   BTB_SIZE :        BTB_SIZE,
+  RAS_SIZE :        RAS_SIZE,
   RADIX :        RADIX,
   DIVCOPIES :        DIVCOPIES,
   ZBA_SUPPORTED :        ZBA_SUPPORTED,

sim/bpred-sim.py

@@ -46,19 +46,41 @@ configs = [
     )
 ]
 
-bpdSize = [6, 8, 10, 12, 14, 16]
+# bpdSize = [6, 8, 10, 12, 14, 16]
-bpdType = ['twobit', 'gshare', 'global', 'gshare_basic', 'global_basic', 'local_basic']
+# bpdType = ['twobit', 'gshare', 'global', 'gshare_basic', 'global_basic', 'local_basic']
-for CurrBPType in bpdType:
+# for CurrBPType in bpdType:
-    for CurrBPSize in bpdSize:
+#     for CurrBPSize in bpdSize:
-        name = CurrBPType+str(CurrBPSize)
+#         name = CurrBPType+str(CurrBPSize)
-        configOptions = "+define+INSTR_CLASS_PRED=0 +define+BPRED_OVERRIDE +define+BPRED_TYPE=" + str(bpdType.index(CurrBPType)) + "+define+BPRED_SIZE=" + str(CurrBPSize)
+#         configOptions = "+define+INSTR_CLASS_PRED=0 +define+BPRED_OVERRIDE +define+BPRED_TYPE=" + str(bpdType.index(CurrBPType)) + "+define+BPRED_SIZE=" + str(CurrBPSize)
-        tc = TestCase(
+#         tc = TestCase(
-            name=name,
+#             name=name,
-            variant="rv32gc",
+#             variant="rv32gc",
-            cmd="vsim > {} -c <<!\ndo wally-batch.do  rv32gc configOptions " + name + " embench " + configOptions,
+#             cmd="vsim > {} -c <<!\ndo wally-batch.do  rv32gc configOptions " + name + " embench " + configOptions,
-            grepstr="")
+#             grepstr="")
-        configs.append(tc)
+#         configs.append(tc)
 
+# bpdSize = [6, 8, 10, 12, 14, 16]
+# for CurrBPSize in bpdSize:
+#     name = 'BTB'+str(CurrBPSize)
+#     configOptions = "+define+INSTR_CLASS_PRED=1 +define+BPRED_OVERRIDE +define+BPRED_TYPE=\`BP_GSHARE" + "+define+BPRED_SIZE=16" + "+define+BTB_SIZE=" + str(CurrBPSize) + "+define+BTB_OVERRIDE"
+#     tc = TestCase(
+#         name=name,
+#         variant="rv32gc",
+#         cmd="vsim > {} -c <<!\ndo wally-batch.do  rv32gc configOptions " + name + " embench " + configOptions,
+#         grepstr="")
+#     configs.append(tc)
+
+bpdSize = [2, 3, 4, 6, 10, 16]
+for CurrBPSize in bpdSize:
+    name = 'RAS'+str(CurrBPSize)
+    configOptions = "+define+INSTR_CLASS_PRED=0 +define+BPRED_OVERRIDE +define+BPRED_TYPE=\`BP_GSHARE" + "+define+BPRED_SIZE=16" + "+define+BTB_SIZE=16" + "+define+RAS_SIZE=" + str(CurrBPSize) + "+define+BTB_OVERRIDE+define+RAS_OVERRIDE"
+    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]
 # LHRSize = [4, 8, 10]
 # bpdType = ['local_repair']

src/cvw.sv

@@ -149,6 +149,7 @@ typedef struct packed {
   int                  BPRED_NUM_LHR;
   int                  BPRED_SIZE;
   int                  BTB_SIZE;
+  int                  RAS_SIZE;
 
 // FPU division architecture
   int           RADIX;

src/ifu/bpred/RASPredictor.sv

@@ -77,7 +77,10 @@ module RASPredictor import cvw::*;  #(parameter cvw_t P)(
   mux2 #(Depth) PtrMux(P1, M1, DecrementPtr, IncDecPtr);
   logic [Depth-1:0] Sum;
   assign Sum = Ptr + IncDecPtr;
-  assign NextPtr = Sum == P.RAS_SIZE[Depth-1:0] ? 0 : Sum; // wrap back around if our stack is not a power of 2
+  if(|P.RAS_SIZE[Depth-1:0])
+    assign NextPtr = Sum >= P.RAS_SIZE[Depth-1:0] ? 0 : Sum; // wrap back around if our stack is not a power of 2
+  else
+    assign NextPtr = Sum;
   //assign NextPtr = Ptr + IncDecPtr;
 
   flopenr #(Depth) PTR(clk, reset, CounterEn, NextPtr, Ptr);