Added new tests

Author: David Schall

.github/workflows/build-and-run.yml

@@ -23,7 +23,7 @@ jobs:
       #
       # To add more build types (Release, Debug, RelWithDebInfo, etc.) customize the build_type list.
       matrix:
-        os: [ubuntu-latest]
+        os: [ubuntu-latest, ubuntu-20.04, ubuntu-22.04]
         build_type: [Release]
         c_compiler: [gcc, clang]
         include:
@@ -37,6 +37,7 @@ jobs:
 
     - name: Cache CMake build
       uses: actions/cache@v4
+      if: ${{ matrix.os == 'ubuntu-latest' }}
       with:
         path: build
         key: ${{ matrix.os }}-${{ matrix.c_compiler }}-build

README.md

@@ -29,12 +29,15 @@ The framework supports a timing approximation by clocking the predictor for ever
 
 ## Prerequisites
 
-The infrastructure has been tested with the following system configuration:
+The infrastructure and following commands have been tested with the following system configuration:
 
 * Ubuntu 22.04.2 LTS
 * gcc 11.4.0
 * cmake 3.22.1
 
+> See the [CI pipeline](https://github.com/dhschall/LLBP/actions/workflows/build-and-run.yml) for other tested system configurations.
+
+
 
 ## Install Dependencies
 
@@ -62,7 +65,7 @@ cmake --build ./build -j 8
 
 ## Server traces
 
-The traces use to evaluate LLBP collected by running the server applications on gem5 in full-system mode. The OS of the disk image is Ubuntu 20.04 and the kernel version is 5.4.84. The traces are in the [ChampSim](https://github.com/ChampSim/ChampSim) format and contains both user and kernel space instructions. The traces are available on Zenodo at [10.5281/zenodo.13133243](https://doi.org/10.5281/zenodo.13133243).
+The traces use to evaluate LLBP collected by running the server applications on gem5 in full-system mode. The OS of the disk image is Ubuntu 20.04 and the kernel version is 5.4.84. The traces are in the [ChampSim](https://github.com/ChampSim/ChampSim) format and contains both user and kernel space instructions. The traces are available on Zenodo at [10.5281/zenodo.13133242](https://doi.org/10.5281/zenodo.13133242).
 
 The `download_traces.sh` script in the utils folder will download all traces from Zenodo and stores them into the `traces` directory.:
 

analysis/mpki.ipynb

@@ -84,13 +84,12 @@ LLBP::LLBP(LLBPConfig cfg)
     : TageSCL(cfg.tsclConfig),
     llbpStorage(cfg.numContexts, cfg.numPatterns,
                  cfg.ctxAssoc, cfg.ptrnAssoc),
-    rcr(HASHVALS,14),
+    rcr(HASHVALS,cfg.CTWidth),
     patternBuffer(cfg.pbSize, cfg.pbAssoc),
 
     numContexts(cfg.numContexts),
     numPatterns(cfg.numPatterns),
     TTWidth(cfg.TTWidth),
-    CTWidth(cfg.CTWidth),
     CtrWidth(cfg.CtrWidth),
     ReplCtrWidth(cfg.ReplCtrWidth),
     CtxReplCtrWidth(cfg.CtxReplCtrWidth),

bpmodels/llbp/llbp.cc

@@ -151,8 +151,8 @@ class LLBP : public TageSCL {
             patterns(n, assoc)
         {}
 
-        // Before a pattern in the pattern set is replaced, the patterns are 
-        // sorted from the highest to the lowest confidence. This is done to 
+        // Before a pattern in the pattern set is replaced, the patterns are
+        // sorted from the highest to the lowest confidence. This is done to
         // determine which pattern should be evicted.
         void sortPatters(const uint64_t key) {
             auto& set = patterns.getSet(key);
@@ -171,7 +171,7 @@ class LLBP : public TageSCL {
      * LLBPs high-capacity structure to store all pattern sets.
      * It's implemented as a set associative cache.
      * The Context directory (CD) can be thought of as the tag array while the
-     * LLBPStorage is the data array. In this simulation model, both LLBP 
+     * LLBPStorage is the data array. In this simulation model, both LLBP
      * and CD are represented with a single data structure.
      ********************************************************************/
     class LLBPStorage : public BaseCache<uint64_t, Context>{
@@ -399,8 +399,6 @@ class LLBP : public TageSCL {
     const int numPatterns;
     // Bit width for pattern tag.
     const int TTWidth;
-    // Bit width for context tag.
-    const int CTWidth;
     // Constants for the patterns counter widths
     const int CtrWidth;
     const int ReplCtrWidth;

bpmodels/llbp/llbp.h

@@ -59,9 +59,7 @@ cmake --build ./build --target predictor -j $(nproc)
 
 
 OUT=results/
-POSTFIX="ae-AlwaysUD"
-POSTFIX="ae-CorrectedNT"
-POSTFIX="ae-New2-noBucket-D5clk"
+POSTFIX="ae"
 
 
 
@@ -115,4 +113,4 @@ echo "Running ${#commands[@]} simulations"
 parallel --jobs $(nproc) ::: "${commands[@]}"
 
 wait
-echo "Simulation complete. Go reap the results, you deserve it!!!!!!"
+echo "Simulation complete."

eval_all.sh

@@ -59,30 +59,29 @@ class Histogram {
         panic_if(bs <= 0, "Bucket size must be positive");
 
         buckets.resize(bins + 2);
-        buckets[0] = Bucket{
-            boost::str(boost::format("< %.1f") % lower), 0, 0};
-        buckets[bins + 1] = Bucket{
-            boost::str(boost::format("> %.1f") % upper), 0, 0};
+        {
+            std::ostringstream oss;
+            oss << "< " << std::fixed << std::setprecision(1) << lower;
+            buckets[0] = Bucket{oss.str(), 0, 0};
+        }
+        {
+            std::ostringstream oss;
+            oss << "> " << std::fixed << std::setprecision(1) << upper;
+            buckets[bins + 1] = Bucket{oss.str(), 0, 0};
+        }
 
         for (int i = 1; i <= bins; ++i) {
+            std::ostringstream oss;
             if (bs == 1) {
-                buckets[i] = Bucket{
-                    boost::str(boost::format("%d") %
-                        (lower + bs * static_cast<T>(i-1))),
-                    0,
-                    0
-                };
+                oss << (lower + bs * static_cast<T>(i-1));
+                buckets[i] = Bucket{oss.str(), 0, 0};
             } else {
-                buckets[i] = Bucket{
-                    boost::str(boost::format("%.1f-%.1f") %
-                        (lower + bs * static_cast<T>(i-1)) %
-                        (lower + bs * static_cast<T>(i))),
-                    0,
-                    0
-                };
+                oss << std::fixed << std::setprecision(1)
+                    << (lower + bs * static_cast<T>(i-1)) << "-"
+                    << (lower + bs * static_cast<T>(i));
+                buckets[i] = Bucket{oss.str(), 0, 0};
             }
         }
-
     }
 
     void insert(T value, int count=1) {
@@ -133,27 +132,32 @@ class Histogram {
         }
 
         std::ostringstream res;
-        res << boost::format("N:%d Min:%d, Max:%d, Sum:%d, Avg:%.2f\n")
-            % samples % min % max % sum
-            % getAvg();
+
+        res << "N:" << samples
+            << " Min:" << min
+            << " Max:" << max
+            << " Sum:" << sum
+            << " Avg:" << std::fixed << std::setprecision(2) << getAvg()
+            << "\n";
 
 
         int cum = 0;
         for (const auto& b : buckets) {
             int barLen = _max > 0 ? static_cast<int>((b.count * width + _max / 2) / _max) : 0;
 
-            res << boost::format("%-10s [%-4d]\t") % b.mark % b.count;
+            res << std::left << std::setw(10) << b.mark
+                << " [" << std::setw(4) << b.count << "]\t";
 
             if (perc) {
-                res << boost::format("%.1f%%\t") % (100.0 * b.count / samples);
+                res << std::fixed << std::setprecision(1) << (100.0 * b.count / samples) << "%\t";
             }
 
             if (cdf) {
                 cum += b.count;
-                res << boost::format("%.1f%%\t") % (100.0 * cum / samples);
+                res << std::fixed << std::setprecision(1) << (100.0 * cum / samples) << "%\t";
             }
 
-            res << boost::format("|%s\n") % std::string(barLen, '*');
+            res << "|" << std::string(barLen, '*') << "\n";
 
         }
         return res.str();