From 60b860ce3619d5f165bf8eda6ad596d5d458d1e2 Mon Sep 17 00:00:00 2001
From: Florian Fischer <florian.fl.fischer@fau.de>
Date: Tue, 11 Feb 2020 13:23:55 +0100
Subject: move plotting code from src.benchmark to src.plots

---
 src/benchmark.py               | 510 ----------------------------------------
 src/benchmarks/blowup.py       |  12 +-
 src/benchmarks/cfrac.py        |  26 ++-
 src/benchmarks/espresso.py     |  30 +--
 src/benchmarks/falsesharing.py |  39 ++--
 src/benchmarks/fd.py           |  23 +-
 src/benchmarks/httpd.py        |  46 ++--
 src/benchmarks/larson.py       |  15 +-
 src/benchmarks/lld.py          |   9 +-
 src/benchmarks/loop.py         |  48 ++--
 src/benchmarks/raxmlng.py      |  29 +--
 src/plots.py                   | 518 +++++++++++++++++++++++++++++++++++++++++
 12 files changed, 672 insertions(+), 633 deletions(-)
 create mode 100644 src/plots.py

diff --git a/src/benchmark.py b/src/benchmark.py
index fd094d0..6f81ba2 100644
--- a/src/benchmark.py
+++ b/src/benchmark.py
@@ -9,8 +9,6 @@ import os
 import subprocess
 from time import sleep
 
-import matplotlib
-import matplotlib.pyplot as plt
 import numpy as np
 
 import src.globalvars
@@ -18,9 +16,6 @@ import src.util
 from src.util import print_status, print_error, print_warn
 from src.util import print_info0, print_info, print_debug
 
-# This is useful when evaluating strings in the plot functions. str(np.NaN) == "nan"
-nan = np.NaN
-
 
 class Benchmark:
     """Default implementation of most methods allocbench expects from a benchmark"""
@@ -568,508 +563,3 @@ class Benchmark:
                     stats["outliers"][dp] = outliers
 
                 self.results["stats"][alloc][perm] = stats
-
-    ###### Summary helpers ######
-    def _eval_with_stat(self, evaluation, alloc, perm, stat):
-        try:
-            s = evaluation.format(**self.results["stats"][alloc][perm][stat])
-        except KeyError as e:
-            import traceback
-            print_warn(traceback.format_exc())
-            print_warn(f"For {alloc} in {perm}")
-            return nan
-        return eval(s)
-
-    def plot_single_arg(self, yval, ylabel="'y-label'", xlabel="'x-label'",
-                        autoticks=True, title="'default title'", filepostfix="",
-                        sumdir="", arg="", scale=None, file_ext=src.globalvars.summary_file_ext):
-
-        args = self.results["args"]
-        allocators = self.results["allocators"]
-
-        arg = arg or list(args.keys())[0]
-
-        if not autoticks:
-            x_vals = list(range(1, len(args[arg]) + 1))
-        else:
-            x_vals = args[arg]
-
-        for allocator in allocators:
-            y_vals = []
-            for perm in self.iterate_args(args=args):
-                if scale:
-                    if scale == allocator:
-                        y_vals = [1] * len(x_vals)
-                    else:
-                        mean = self._eval_with_stat(yval, allocator, perm, "mean")
-                        norm_mean = self._eval_with_stat(yval, scale, perm, "mean")
-                        y_vals.append(mean / norm_mean)
-                else:
-                    y_vals.append(self._eval_with_stat(yval, allocator, perm, "mean"))
-
-            plt.plot(x_vals, y_vals, marker='.', linestyle='-',
-                     label=allocator, color=allocators[allocator]["color"])
-
-        plt.legend(loc="best")
-        if not autoticks:
-            plt.xticks(x_vals, args[arg])
-        plt.xlabel(eval(xlabel))
-        plt.ylabel(eval(ylabel))
-        plt.title(eval(title))
-        figname = os.path.join(sumdir, f"{self.name}.{filepostfix}.{file_ext}")
-        if figname.endswith(".tex"):
-            import tikzplotlib
-            tikzplotlib.save(figname)
-        else:
-            plt.savefig(figname)
-        plt.clf()
-
-    def barplot_single_arg(self, yval, ylabel="'y-label'", xlabel="'x-label'",
-                           title="'default title'", filepostfix="", sumdir="",
-                           arg="", scale=None, file_ext=src.globalvars.summary_file_ext, yerr=True):
-
-        args = self.results["args"]
-        allocators = self.results["allocators"]
-        nallocators = len(allocators)
-
-        if arg:
-            arg = args[arg]
-        elif args.keys():
-            arg = args[list(args.keys())[0]]
-        else:
-            arg = [""]
-
-        narg = len(arg)
-
-        for i, allocator in enumerate(allocators):
-            x_vals = list(range(i, narg * (nallocators+1), nallocators+1))
-            y_vals = []
-            y_errs = None
-            if yerr:
-                y_errs = []
-
-            for perm in self.iterate_args(args=args):
-                if scale:
-                    if scale == allocator:
-                        y_vals = [1] * len(x_vals)
-                    else:
-                        mean = self._eval_with_stat(yval, allocator, perm, "mean")
-                        norm_mean = self._eval_with_stat(yval, scale, perm, "mean")
-                        y_vals.append(mean / norm_mean)
-                else:
-                    y_vals.append(self._eval_with_stat(yval, allocator, perm, "mean"))
-
-                if yerr:
-                    y_errs.append(self._eval_with_stat(yval, allocator, perm, "std"))
-
-            plt.bar(x_vals, y_vals, width=1, label=allocator, yerr=y_errs,
-                    color=allocators[allocator]["color"])
-
-        plt.legend(loc="best")
-        plt.xticks(list(range(int(np.floor(nallocators/2)), narg*(nallocators+1), nallocators+1)), arg)
-        plt.xlabel(eval(xlabel))
-        plt.ylabel(eval(ylabel))
-        plt.title(eval(title))
-        figname = os.path.join(sumdir, f"{self.name}.{filepostfix}.{file_ext}")
-        if figname.endswith(".tex"):
-            import tikzplotlib
-            tikzplotlib.save(figname)
-        else:
-            plt.savefig(figname)
-        plt.clf()
-
-    def plot_fixed_arg(self, yval, ylabel="'y-label'", xlabel="loose_arg",
-                       autoticks=True, title="'default title'", filepostfix="",
-                       sumdir="", fixed=[], file_ext=src.globalvars.summary_file_ext, scale=None):
-
-        args = self.results["args"]
-        allocators = self.results["allocators"]
-
-        for arg in fixed or args:
-            loose_arg = [a for a in args if a != arg][0]
-
-            if not autoticks:
-                x_vals = list(range(1, len(args[loose_arg]) + 1))
-            else:
-                x_vals = args[loose_arg]
-
-            for arg_value in args[arg]:
-                for allocator in allocators:
-                    y_vals = []
-                    for perm in self.iterate_args_fixed({arg: arg_value}, args=args):
-                        if scale:
-                            if scale == allocator:
-                                y_vals = [1] * len(x_vals)
-                            else:
-                                mean = self._eval_with_stat(yval, allocator, perm, "mean")
-                                norm_mean = self._eval_with_stat(yval, scale, perm, "mean")
-                                y_vals.append(mean / norm_mean)
-                        else:
-                            y_vals.append(self._eval_with_stat(yval, allocator, perm, "mean"))
-
-                    plt.plot(x_vals, y_vals, marker='.', linestyle='-',
-                             label=allocator, color=allocators[allocator]["color"])
-
-                plt.legend(loc="best")
-                if not autoticks:
-                    plt.xticks(x_vals, args[loose_arg])
-                plt.xlabel(eval(xlabel))
-                plt.ylabel(eval(ylabel))
-                plt.title(eval(title))
-                figname = os.path.join(sumdir,
-                                       f"{self.name}.{arg}.{arg_value}.{filepostfix}.{file_ext}")
-                if figname.endswith(".tex"):
-                    import tikzplotlib
-                    tikzplotlib.save(figname)
-                else:
-                    plt.savefig(figname)
-                plt.clf()
-
-    def export_facts_to_file(self, comment_symbol, f):
-        """Write collected facts about used system and benchmark to file"""
-        print(comment_symbol, self.name, file=f)
-        print(file=f)
-        print(comment_symbol, "Common facts:", file=f)
-        for k, v in src.facter.FACTS.items():
-            print(comment_symbol, k + ":", v, file=f)
-        print(file=f)
-        print(comment_symbol, "Benchmark facts:", file=f)
-        for k, v in self.results["facts"].items():
-            print(comment_symbol, k + ":", v, file=f)
-        print(file=f)
-
-    def export_stats_to_csv(self, datapoint, path=None):
-        """Write descriptive statistics about datapoint to csv file"""
-        allocators = self.results["allocators"]
-        args = self.results["args"]
-        stats = self.results["stats"]
-
-        if path is None:
-            path = datapoint
-
-        path = path + ".csv"
-
-        stats_fields = list(stats[list(allocators)[0]][list(self.iterate_args(args=args))[0]])
-        fieldnames = ["allocator", *args, *stats_fields]
-        widths = []
-        for fieldname in fieldnames:
-            widths.append(len(fieldname) + 2)
-
-        # collect rows
-        rows = {}
-        for alloc in allocators:
-            rows[alloc] = {}
-            for perm in self.iterate_args(args=args):
-                d = []
-                d.append(alloc)
-                d += list(perm._asdict().values())
-                d += [stats[alloc][perm][s][datapoint] for s in stats[alloc][perm]]
-                d[-1] = (",".join([str(x) for x in d[-1]]))
-                rows[alloc][perm] = d
-
-        # calc widths
-        for i in range(0, len(fieldnames)):
-            for alloc in allocators:
-                for perm in self.iterate_args(args=args):
-                    field_len = len(str(rows[alloc][perm][i])) + 2
-                    if field_len > widths[i]:
-                        widths[i] = field_len
-
-        with open(path, "w") as f:
-            headerline = ""
-            for i, h in enumerate(fieldnames):
-                headerline += h.capitalize().ljust(widths[i]).replace("_", "-")
-            print(headerline, file=f)
-
-            for alloc in allocators:
-                for perm in self.iterate_args(args=args):
-                    line = ""
-                    for i, x in enumerate(rows[alloc][perm]):
-                        line += str(x).ljust(widths[i])
-                    print(line.replace("_", "-"), file=f)
-
-    def export_stats_to_dataref(self, datapoint, path=None):
-        """Write descriptive statistics about datapoint to dataref file"""
-        stats = self.results["stats"]
-
-        if path is None:
-            path = datapoint
-
-        path = path + ".dataref"
-
-        # Example: \drefset{/mysql/glibc/40/Lower-whisker}{71552.0}
-        line = "\\drefset{{/{}/{}/{}/{}}}{{{}}}"
-
-        with open(path, "w") as f:
-            # Write facts to file
-            self.export_facts_to_file("%", f)
-
-            for alloc in self.results["allocators"]:
-                for perm in self.iterate_args(args=self.results["args"]):
-                    for statistic, values in stats[alloc][perm].items():
-                        cur_line = line.format(self.name, alloc,
-                                               "/".join([str(p) for p in list(perm)]),
-                                               statistic, values[datapoint])
-                        # Replace empty outliers
-                        cur_line.replace("[]", "")
-                        # Replace underscores
-                        cur_line.replace("_", "-")
-                        print(cur_line, file=f)
-
-    def write_best_doublearg_tex_table(self, evaluation, sort=">",
-                                       filepostfix="", sumdir="", std=False):
-        args = self.results["args"]
-        keys = list(args.keys())
-        allocators = self.results["allocators"]
-
-        header_arg = keys[0] if len(args[keys[0]]) < len(args[keys[1]]) else keys[1]
-        row_arg = [arg for arg in args if arg != header_arg][0]
-
-        headers = args[header_arg]
-        rows = args[row_arg]
-
-        cell_text = []
-        for av in rows:
-            row = []
-            for perm in self.iterate_args_fixed({row_arg: av}, args=args):
-                best = []
-                best_val = None
-                for allocator in allocators:
-                    d = []
-                    for m in self.results[allocator][perm]:
-                        d.append(eval(evaluation.format(**m)))
-                    mean = np.mean(d)
-                    if not best_val:
-                        best = [allocator]
-                        best_val = mean
-                    elif ((sort == ">" and mean > best_val)
-                          or (sort == "<" and mean < best_val)):
-                        best = [allocator]
-                        best_val = mean
-                    elif mean == best_val:
-                        best.append(allocator)
-
-                row.append("{}: {:.3f}".format(best[0], best_val))
-            cell_text.append(row)
-
-        fname = os.path.join(sumdir, ".".join([self.name, filepostfix, "tex"]))
-        with open(fname, "w") as f:
-            print("\\documentclass{standalone}", file=f)
-            print("\\begin{document}", file=f)
-            print("\\begin{tabular}{|", end="", file=f)
-            print(" l |" * len(headers), "}", file=f)
-
-            print(header_arg+"/"+row_arg, end=" & ", file=f)
-            for header in headers[:-1]:
-                print(header, end="& ", file=f)
-            print(headers[-1], "\\\\", file=f)
-
-            for i, row in enumerate(cell_text):
-                print(rows[i], end=" & ", file=f)
-                for e in row[:-1]:
-                    print(e, end=" & ", file=f)
-                print(row[-1], "\\\\", file=f)
-            print("\\end{tabular}", file=f)
-            print("\\end{document}", file=f)
-
-    def write_tex_table(self, entries, sort=">",
-                        filepostfix="", sumdir="", std=False):
-        """generate a latex standalone table from an list of entries dictionaries
-
-        Entries must have at least the two keys: "label" and "expression".
-        The optional "sort" key specifies the direction of the order:
-            ">" : bigger is better.
-            "<" : smaller is better.
-
-        Table layout:
-
-        |    alloc1     |    alloc2    | ....
-        ---------------------------------------
-        | name1  name2  | ...
-        ---------------------------------------
-        perm1 | eavl1  eval2  | ...
-        perm2 | eval1  eval2  | ...
-        """
-        args = self.results["args"]
-        allocators = self.results["allocators"]
-        nallocators = len(allocators)
-        nentries = len(entries)
-        perm_fields = self.Perm._fields
-        nperm_fields = len(perm_fields)
-
-        alloc_header_line = f"\\multicolumn{{{nperm_fields}}}{{c|}}{{}} &"
-        for alloc in allocators:
-            alloc_header_line += f"\\multicolumn{{{nentries}}}{{c|}}{{{alloc}}} &"
-        alloc_header_line = alloc_header_line[:-1] + "\\\\"
-
-        perm_fields_header = ""
-        for field in self.Perm._fields:
-            perm_fields_header += f'{field} &'
-        entry_header_line = ""
-        for entry in entries:
-            entry_header_line += f'{entry["label"]} &'
-        entry_header_line = perm_fields_header + entry_header_line * nallocators
-        entry_header_line = entry_header_line[:-1] + "\\\\"
-
-        fname = os.path.join(sumdir, ".".join([self.name, filepostfix, "tex"]))
-        with open(fname, "w") as f:
-            print("\\documentclass{standalone}", file=f)
-            print("\\usepackage{booktabs}", file=f)
-            print("\\usepackage{xcolor}", file=f)
-            print("\\begin{document}", file=f)
-            print("\\begin{tabular}{|", f"{'c|'*nperm_fields}", f"{'c'*nentries}|"*nallocators, "}", file=f)
-            print("\\toprule", file=f)
-
-            print(alloc_header_line, file=f)
-            print("\\hline", file=f)
-            print(entry_header_line, file=f)
-            print("\\hline", file=f)
-
-            for perm in self.iterate_args(args=args):
-                values = [[] for _ in entries]
-                maxs = [None for _ in entries]
-                mins = [None for _ in entries]
-                for allocator in allocators:
-                    for i, entry in enumerate(entries):
-                        expr = entry["expression"]
-                        values[i].append(eval(expr.format(**self.results["stats"][allocator][perm]["mean"])))
-
-                # get max and min for each entry
-                for i, entry in enumerate(entries):
-                    if not "sort" in entry:
-                        continue
-                    # bigger is better
-                    elif entry["sort"] == ">":
-                        maxs[i] = max(values[i])
-                        mins[i] = min(values[i])
-                    # smaller is better
-                    elif entry["sort"] == "<":
-                        mins[i] = max(values[i])
-                        maxs[i] = min(values[i])
-
-                # build row
-                row = ""
-                perm_dict = perm._asdict()
-                for field in perm_fields:
-                    row += str(perm_dict[field]) + "&"
-
-                for i, _ in enumerate(allocators):
-                    for y, entry_vals in enumerate(values):
-                        val = entry_vals[i]
-
-                        # format
-                        val_str = str(val)
-                        if type(val) == float:
-                            val_str = f"{val:.2f}"
-
-                        # colorize
-                        if val == maxs[y]:
-                            val_str = f"\\textcolor{{green}}{{{val_str}}}"
-                        elif val == mins[y]:
-                            val_str = f"\\textcolor{{red}}{{{val_str}}}"
-                        row += f"{val_str} &"
-                #escape _ for latex
-                row = row.replace("_", "\\_")
-                print(row[:-1], "\\\\", file=f)
-
-            print("\\end{tabular}", file=f)
-            print("\\end{document}", file=f)
-
-    def pgfplot_legend(self, sumdir=""):
-
-        allocators = self.results["allocators"]
-        s =\
-"""
-\\documentclass{standalone}
-\\usepackage{pgfplots}
-
-\\usepackage{pgfkeys}
-
-\\newenvironment{customlegend}[1][]{%
-\t\\begingroup
-\t\\csname pgfplots@init@cleared@structures\\endcsname
-\t\\pgfplotsset{#1}%
-}{%
-\t\\csname pgfplots@createlegend\\endcsname
-\t\\endgroup
-}%
-\\def\\addlegendimage{\\csname pgfplots@addlegendimage\\endcsname}
-
-\\usepackage{xcolor}
-"""
-
-        for alloc_name, alloc_dict in allocators.items():
-            # define color
-            rgb = matplotlib.colors.to_rgb(alloc_dict["color"])
-            s += f"\\providecolor{{{alloc_name}-color}}{{rgb}}{{{rgb[0]},{rgb[1]},{rgb[2]}}}\n"
-
-        s +=\
-"""
-\\begin{document}
-\\begin{tikzpicture}
-\\begin{customlegend}[
-\tlegend entries={"""
-
-        alloc_list = ""
-        addlegendimage_list = ""
-        for alloc_name in allocators:
-            alloc_list += f"{alloc_name}, "
-            addlegendimage_list += "\t\\addlegendimage{}\n"
-
-        s += alloc_list[:-2] + "},\n]"
-        s += addlegendimage_list
-        s +=\
-"""
-\\end{customlegend}
-\\end{tikzpicture}
-\\end{document}"""
-
-        with open(os.path.join(sumdir, "legend.tex"), "w") as legend_file:
-            print(s, file=legend_file)
-
-    def pgfplot_linear(self, perms, xval, yval, ylabel="'y-label'", xlabel="'x-label'",
-                       title="'default title'", postfix="", sumdir="", scale=None):
-
-        allocators = self.results["allocators"]
-        perms = list(perms)
-        title = eval(title)
-        s =\
-"""\\documentclass{standalone}
-\\usepackage{pgfplots}
-\\usepackage{xcolor}
-"""
-
-        for alloc_name, alloc_dict in allocators.items():
-            s += f"\\begin{{filecontents*}}{{{alloc_name}.dat}}\n"
-            for i, perm in enumerate(perms):
-                x = self._eval_with_stat(xval, alloc_name, perm, "mean")
-                y = self._eval_with_stat(yval, alloc_name, perm, "mean")
-                s += f"{x} {y}\n"
-            s += "\\end{filecontents*}\n"
-
-            # define color
-            rgb = matplotlib.colors.to_rgb(alloc_dict["color"])
-            s += f"\\providecolor{{{alloc_name}-color}}{{rgb}}{{{rgb[0]},{rgb[1]},{rgb[2]}}}\n"
-
-        s +=\
-f"""
-\\begin{{document}}
-\\begin{{tikzpicture}}
-\\begin{{axis}}[
-\ttitle={{{title}}},
-\txlabel={{{eval(xlabel)}}},
-\tylabel={{{eval(ylabel)}}},
-]
-"""
-
-        for alloc_name in allocators:
-            s += f"\\addplot [{alloc_name}-color] table {{{alloc_name}.dat}};\n"
-            # s += f"\t\\addplot table {{{alloc_name}.dat}};\n"
-
-        s +=\
-"""\\end{axis}
-\\end{tikzpicture}
-\\end{document}"""
-
-        with open(os.path.join(sumdir, f"{self.name}.{postfix}.tex"), "w") as plot_file:
-            print(s, file=plot_file)
diff --git a/src/benchmarks/blowup.py b/src/benchmarks/blowup.py
index aca9293..8f8e550 100644
--- a/src/benchmarks/blowup.py
+++ b/src/benchmarks/blowup.py
@@ -17,6 +17,7 @@
 """Definition of the blowup micro benchmark"""
 
 from src.benchmark import Benchmark
+import src.plots as plt
 
 
 class BenchmarkBlowup(Benchmark):
@@ -53,15 +54,16 @@ class BenchmarkBlowup(Benchmark):
                 }
             }
 
-        self.barplot_single_arg("{VmHWM}/1000",
-                                ylabel='"VmHWM in MB"',
-                                title='"blowup test"',
-                                filepostfix="vmhwm")
+        plt.barplot_single_arg(self,
+                               "{VmHWM}/1000",
+                               ylabel='"VmHWM in MB"',
+                               title='"blowup test"',
+                               filepostfix="vmhwm")
 
         del allocators["Ideal-RSS"]
         del self.results["stats"]["Ideal-RSS"]
 
-        self.export_stats_to_dataref("VmHWM")
+        plt.export_stats_to_dataref(self, "VmHWM")
 
 
 blowup = BenchmarkBlowup()
diff --git a/src/benchmarks/cfrac.py b/src/benchmarks/cfrac.py
index dfd87d6..34c0894 100644
--- a/src/benchmarks/cfrac.py
+++ b/src/benchmarks/cfrac.py
@@ -59,6 +59,7 @@ API function as well as memory placement strategies with good data locality.
 """
 
 from src.benchmark import Benchmark
+import src.plots as plt
 
 
 class BenchmarkCfrac(Benchmark):
@@ -75,13 +76,15 @@ class BenchmarkCfrac(Benchmark):
 
     def summary(self):
         # Speed
-        self.barplot_single_arg("{task-clock}/1000",
-                                ylabel='"cpu-second"',
-                                title='"Cfrac: runtime"',
-                                filepostfix="time")
+        plt.barplot_single_arg(self,
+                               "{task-clock}/1000",
+                               ylabel='"cpu-second"',
+                               title='"Cfrac: runtime"',
+                               filepostfix="time")
 
         # L1 cache misses
-        self.barplot_single_arg(
+        plt.barplot_single_arg(
+            self,
             "({L1-dcache-load-misses}/{L1-dcache-loads})*100",
             ylabel='"L1 misses in %"',
             title='"Cfrac l1 cache misses"',
@@ -89,10 +92,11 @@ class BenchmarkCfrac(Benchmark):
             yerr=False)
 
         # Memusage
-        self.barplot_single_arg("{VmHWM}",
-                                ylabel='"VmHWM in KB"',
-                                title='"Cfrac VmHWM"',
-                                filepostfix="vmhwm")
+        plt.barplot_single_arg(self,
+                               "{VmHWM}",
+                               ylabel='"VmHWM in KB"',
+                               title='"Cfrac VmHWM"',
+                               filepostfix="vmhwm")
 
         self.write_tex_table([{
             "label": "Runtime [ms]",
@@ -105,9 +109,9 @@ class BenchmarkCfrac(Benchmark):
         }],
                              filepostfix="table")
 
-        self.export_stats_to_dataref("task-clock")
+        plt.export_stats_to_dataref(self, "task-clock")
 
-        self.export_stats_to_dataref("VmHWM")
+        plt.export_stats_to_dataref(self, "VmHWM")
 
 
 cfrac = BenchmarkCfrac()
diff --git a/src/benchmarks/espresso.py b/src/benchmarks/espresso.py
index 2f8b8bf..1c9d4d2 100644
--- a/src/benchmarks/espresso.py
+++ b/src/benchmarks/espresso.py
@@ -59,6 +59,7 @@ import os
 
 from src.benchmark import Benchmark
 import src.globalvars
+import src.plots as plt
 
 
 class BenchmarkEspresso(Benchmark):
@@ -79,13 +80,15 @@ class BenchmarkEspresso(Benchmark):
 
     def summary(self):
         # Speed
-        self.barplot_single_arg("{task-clock}/1000",
-                                ylabel='"cpu-second"',
-                                title='"Espresso: runtime"',
-                                filepostfix="time")
+        plt.barplot_single_arg(self,
+                               "{task-clock}/1000",
+                               ylabel='"cpu-second"',
+                               title='"Espresso: runtime"',
+                               filepostfix="time")
 
         # L1 cache misses
-        self.barplot_single_arg(
+        plt.barplot_single_arg(
+            self,
             "({L1-dcache-load-misses}/{L1-dcache-loads})*100",
             ylabel='"L1 misses in %"',
             title='"Espresso l1 cache misses"',
@@ -93,12 +96,13 @@ class BenchmarkEspresso(Benchmark):
             yerr=False)
 
         # Memusage
-        self.barplot_single_arg("{VmHWM}",
-                                ylabel='"VmHWM in KB"',
-                                title='"Espresso VmHWM"',
-                                filepostfix="vmhwm")
+        plt.barplot_single_arg(self,
+                               "{VmHWM}",
+                               ylabel='"VmHWM in KB"',
+                               title='"Espresso VmHWM"',
+                               filepostfix="vmhwm")
 
-        self.write_tex_table([{
+        plt.write_tex_table(self, [{
             "label": "Runtime [ms]",
             "expression": "{task-clock}",
             "sort": "<"
@@ -107,11 +111,11 @@ class BenchmarkEspresso(Benchmark):
             "expression": "{VmHWM}",
             "sort": "<"
         }],
-                             filepostfix="table")
+                            filepostfix="table")
 
-        self.export_stats_to_dataref("task-clock")
+        plt.export_stats_to_dataref(self, "task-clock")
 
-        self.export_stats_to_dataref("VmHWM")
+        plt.export_stats_to_dataref(self, "VmHWM")
 
 
 espresso = BenchmarkEspresso()
diff --git a/src/benchmarks/falsesharing.py b/src/benchmarks/falsesharing.py
index 530ca99..626104a 100644
--- a/src/benchmarks/falsesharing.py
+++ b/src/benchmarks/falsesharing.py
@@ -23,6 +23,7 @@ import numpy as np
 
 from src.benchmark import Benchmark
 from src.globalvars import summary_file_ext
+import src.plots as plt
 
 TIME_RE = re.compile("^Time elapsed = (?P<time>\\d*\\.\\d*) seconds.$")
 
@@ -78,14 +79,16 @@ class BenchmarkFalsesharing(Benchmark):
         del self.results["stats"]
         self.calc_desc_statistics()
 
-        self.plot_fixed_arg("{speedup}",
-                            ylabel="'Speedup'",
-                            title="'Speedup: ' + arg + ' ' + str(arg_value)",
-                            filepostfix="speedup",
-                            autoticks=False,
-                            fixed=["bench"])
+        plt.plot_fixed_arg(self,
+                           "{speedup}",
+                           ylabel="'Speedup'",
+                           title="'Speedup: ' + arg + ' ' + str(arg_value)",
+                           filepostfix="speedup",
+                           autoticks=False,
+                           fixed=["bench"])
 
-        self.plot_fixed_arg(
+        plt.plot_fixed_arg(
+            self,
             "{l1chache_misses}",
             ylabel="'l1 cache misses in %'",
             title="'cache misses: ' + arg + ' ' + str(arg_value)",
@@ -93,23 +96,23 @@ class BenchmarkFalsesharing(Benchmark):
             autoticks=False,
             fixed=["bench"])
 
-        self.plot_fixed_arg(
-            "({LLC-load-misses}/{LLC-loads})*100",
-            ylabel="'llc cache misses in %'",
-            title="'LLC misses: ' + arg + ' ' + str(arg_value)",
-            filepostfix="llc-misses",
-            autoticks=False,
-            fixed=["bench"])
+        plt.plot_fixed_arg(self,
+                           "({LLC-load-misses}/{LLC-loads})*100",
+                           ylabel="'llc cache misses in %'",
+                           title="'LLC misses: ' + arg + ' ' + str(arg_value)",
+                           filepostfix="llc-misses",
+                           autoticks=False,
+                           fixed=["bench"])
 
-        self.write_tex_table([{
+        plt.write_tex_table(self, [{
             "label": "Speedup",
             "expression": "{speedup}",
             "sort": ">"
         }],
-                             filepostfix="speedup.table")
+                            filepostfix="speedup.table")
 
-        self.export_stats_to_csv("speedup", "time")
-        self.export_stats_to_csv("l1chache_misses", "l1-misses")
+        plt.export_stats_to_csv(self, "speedup", "time")
+        plt.export_stats_to_csv(self, "l1chache_misses", "l1-misses")
 
 
 falsesharing = BenchmarkFalsesharing()
diff --git a/src/benchmarks/fd.py b/src/benchmarks/fd.py
index 4fab958..bede21a 100644
--- a/src/benchmarks/fd.py
+++ b/src/benchmarks/fd.py
@@ -22,6 +22,7 @@ from urllib.request import urlretrieve
 
 from src.artifact import ArchiveArtifact, GitArtifact
 from src.benchmark import Benchmark
+import src.plots as plt
 from src.util import print_info
 
 
@@ -66,19 +67,21 @@ class BenchmarkFd(Benchmark):
             os.link(src, dest)
 
     def summary(self):
-        self.barplot_single_arg("{task-clock}",
-                                ylabel='"runtime in ms"',
-                                title='"fd runtime"',
-                                filepostfix="runtime")
+        plt.barplot_single_arg(self,
+                               "{task-clock}",
+                               ylabel='"runtime in ms"',
+                               title='"fd runtime"',
+                               filepostfix="runtime")
 
-        self.export_stats_to_dataref("task-clock")
+        plt.export_stats_to_dataref(self, "task-clock")
 
-        self.barplot_single_arg("{VmHWM}",
-                                ylabel='"VmHWM in KB"',
-                                title='"fd memusage"',
-                                filepostfix="memusage")
+        plt.barplot_single_arg(self,
+                               "{VmHWM}",
+                               ylabel='"VmHWM in KB"',
+                               title='"fd memusage"',
+                               filepostfix="memusage")
 
-        self.export_stats_to_dataref("VmHWM")
+        plt.export_stats_to_dataref(self, "VmHWM")
 
 
 fd = BenchmarkFd()
diff --git a/src/benchmarks/httpd.py b/src/benchmarks/httpd.py
index 64f8e11..2a848c5 100644
--- a/src/benchmarks/httpd.py
+++ b/src/benchmarks/httpd.py
@@ -20,6 +20,7 @@ import re
 
 from src.benchmark import Benchmark
 import src.facter
+import src.plots as plt
 
 
 class BenchmarkHTTPD(Benchmark):
@@ -59,28 +60,29 @@ class BenchmarkHTTPD(Benchmark):
             "Requests per second:\\s*(\\d*\\.\\d*) .*", stdout).group(1)
 
     def summary(self):
-        allocators = self.results["allocators"]
-
-        self.plot_fixed_arg("{requests}",
-                            xlabel='"threads"',
-                            ylabel='"requests/s"',
-                            autoticks=False,
-                            filepostfix="requests",
-                            title='perm.site + ": requests/s"')
-
-        self.plot_fixed_arg("{nginx_vmhwm}",
-                            xlabel='"threads"',
-                            ylabel='"VmHWM in KB"',
-                            title='perm.site + ": nginx memory usage"',
-                            filepostfix="httpd_vmhwm",
-                            autoticks=False)
-
-        self.plot_fixed_arg("{php-fpm_vmhwm}",
-                            xlabel='"threads"',
-                            ylabel='"VmHWM in KB"',
-                            title='perm.site + ": php-fpm memory usage"',
-                            filepostfix="php-fpm_vmhwm",
-                            autoticks=False)
+        plt.plot_fixed_arg(self,
+                           "{requests}",
+                           xlabel='"threads"',
+                           ylabel='"requests/s"',
+                           autoticks=False,
+                           filepostfix="requests",
+                           title='perm.site + ": requests/s"')
+
+        plt.plot_fixed_arg(self,
+                           "{nginx_vmhwm}",
+                           xlabel='"threads"',
+                           ylabel='"VmHWM in KB"',
+                           title='perm.site + ": nginx memory usage"',
+                           filepostfix="httpd_vmhwm",
+                           autoticks=False)
+
+        plt.plot_fixed_arg(self,
+                           "{php-fpm_vmhwm}",
+                           xlabel='"threads"',
+                           ylabel='"VmHWM in KB"',
+                           title='perm.site + ": php-fpm memory usage"',
+                           filepostfix="php-fpm_vmhwm",
+                           autoticks=False)
 
 
 httpd = BenchmarkHTTPD()
diff --git a/src/benchmarks/larson.py b/src/benchmarks/larson.py
index 6e55fbe..8fce9ed 100644
--- a/src/benchmarks/larson.py
+++ b/src/benchmarks/larson.py
@@ -48,6 +48,7 @@ false sharing because it uses multiple threads, which pass memory around.
 import re
 
 from src.benchmark import Benchmark
+import src.plots as plt
 
 THROUGHPUT_RE = re.compile(
     "^Throughput =\\s*(?P<throughput>\\d+) operations per second.$")
@@ -80,12 +81,14 @@ class BenchmarkLarson(Benchmark):
 
     def summary(self):
         # Plot threads->throughput and maxsize->throughput
-        self.plot_fixed_arg("{throughput}/1000000",
-                            ylabel="'MOPS/s'",
-                            title="'Larson: ' + arg + ' ' + str(arg_value)",
-                            filepostfix="throughput")
-
-        self.plot_fixed_arg(
+        plt.plot_fixed_arg(self,
+                           "{throughput}/1000000",
+                           ylabel="'MOPS/s'",
+                           title="'Larson: ' + arg + ' ' + str(arg_value)",
+                           filepostfix="throughput")
+
+        plt.plot_fixed_arg(
+            self,
             "({L1-dcache-load-misses}/{L1-dcache-loads})*100",
             ylabel="'l1 cache misses in %'",
             title="'Larson cache misses: ' + arg + ' ' + str(arg_value)",
diff --git a/src/benchmarks/lld.py b/src/benchmarks/lld.py
index 3657896..b2da35b 100644
--- a/src/benchmarks/lld.py
+++ b/src/benchmarks/lld.py
@@ -204,6 +204,7 @@ import matplotlib.pyplot as plt
 from src.artifact import ArchiveArtifact
 from src.benchmark import Benchmark
 import src.facter
+import src.plots
 from src.globalvars import summary_file_ext
 
 
@@ -283,17 +284,17 @@ class BenchmarkLld(Benchmark):
             plt.clf()
 
         # self.export_stats_to_csv("VmHWM")
-        self.export_stats_to_csv("task-clock")
+        src.plots.export_stats_to_csv("task-clock")
 
         # self.export_stats_to_dataref("VmHWM")
-        self.export_stats_to_dataref("task-clock")
+        src.plots.export_stats_to_dataref("task-clock")
 
-        self.write_tex_table([{
+        src.plots.write_tex_table([{
             "label": "Runtime [ms]",
             "expression": "{task-clock}",
             "sort": "<"
         }],
-                             filepostfix="table")
+                                  filepostfix="table")
 
 
 lld = BenchmarkLld()
diff --git a/src/benchmarks/loop.py b/src/benchmarks/loop.py
index a969f4a..75c56ee 100644
--- a/src/benchmarks/loop.py
+++ b/src/benchmarks/loop.py
@@ -35,6 +35,7 @@ Interpretation:
 """
 
 from src.benchmark import Benchmark
+import src.plots as plt
 
 
 class BenchmarkLoop(Benchmark):
@@ -57,14 +58,16 @@ class BenchmarkLoop(Benchmark):
 
     def summary(self):
         # Speed
-        self.plot_fixed_arg("perm.nthreads / ({task-clock}/1000)",
-                            ylabel='"MOPS/cpu-second"',
-                            title='"Loop: " + arg + " " + str(arg_value)',
-                            filepostfix="time",
-                            autoticks=False)
+        plt.plot_fixed_arg(self,
+                           "perm.nthreads / ({task-clock}/1000)",
+                           ylabel='"MOPS/cpu-second"',
+                           title='"Loop: " + arg + " " + str(arg_value)',
+                           filepostfix="time",
+                           autoticks=False)
 
         # L1 cache misses
-        self.plot_fixed_arg(
+        plt.plot_fixed_arg(
+            self,
             "({L1-dcache-load-misses}/{L1-dcache-loads})*100",
             ylabel='"L1 misses in %"',
             title='"Loop l1 cache misses: " + arg + " " + str(arg_value)',
@@ -72,29 +75,34 @@ class BenchmarkLoop(Benchmark):
             autoticks=False)
 
         # Speed Matrix
-        self.write_best_doublearg_tex_table(
-            "perm.nthreads / ({task-clock}/1000)", filepostfix="time.matrix")
+        plt.write_best_doublearg_tex_table(
+            self,
+            "perm.nthreads / ({task-clock}/1000)",
+            filepostfix="time.matrix")
 
-        self.write_tex_table([{
+        plt.write_tex_table(self, [{
             "label": "MOPS/s",
             "expression": "perm.nthreads / ({task-clock}/1000)",
             "sort": ">"
         }],
-                             filepostfix="mops.table")
+                            filepostfix="mops.table")
 
-        self.export_stats_to_csv("task-clock")
-        self.export_stats_to_dataref("task-clock")
+        plt.export_stats_to_csv(self, "task-clock")
+        plt.export_stats_to_dataref(self, "task-clock")
 
         # pgfplot test
-        self.pgfplot_linear(self.iterate_args_fixed({"maxsize": 1024}, args=self.results["args"]),
-                            "int(perm.nthreads)",
-                            "perm.nthreads / ({task-clock}/1000)",
-                            xlabel='"Threads"',
-                            ylabel='"MOPS/cpu-second"',
-                            title='"Loop: 1024B"',
-                            postfix='mops_1024B')
+        plt.pgfplot_linear(self,
+                           self.iterate_args_fixed({"maxsize": 1024},
+                                                   args=self.results["args"]),
+                           "int(perm.nthreads)",
+                           "perm.nthreads / ({task-clock}/1000)",
+                           xlabel='"Threads"',
+                           ylabel='"MOPS/cpu-second"',
+                           title='"Loop: 1024B"',
+                           postfix='mops_1024B')
 
         # create pgfplot legend
-        self.pgfplot_legend()
+        plt.pgfplot_legend(self)
+
 
 loop = BenchmarkLoop()
diff --git a/src/benchmarks/raxmlng.py b/src/benchmarks/raxmlng.py
index 05a6ca7..228c220 100644
--- a/src/benchmarks/raxmlng.py
+++ b/src/benchmarks/raxmlng.py
@@ -18,11 +18,10 @@
 
 import os
 import re
-import sys
-from urllib.request import urlretrieve
 
 from src.artifact import GitArtifact
 from src.benchmark import Benchmark
+import src.plots as plt
 from src.util import print_info, run_cmd
 
 RUNTIME_RE = re.compile("Elapsed time: (?P<runtime>(\\d*.\\d*)) seconds")
@@ -58,8 +57,8 @@ class BenchmarkRaxmlng(Benchmark):
         os.makedirs(raxmlng_builddir, exist_ok=True)
 
         # building raxml-ng
-        run_cmd( ["cmake", ".."], cwd=raxmlng_builddir)
-        run_cmd( ["make"], cwd=raxmlng_builddir)
+        run_cmd(["cmake", ".."], cwd=raxmlng_builddir)
+        run_cmd(["make"], cwd=raxmlng_builddir)
 
         # create symlinks
         for exe in ["raxml-ng"]:
@@ -83,19 +82,21 @@ class BenchmarkRaxmlng(Benchmark):
         result["runtime"] = RUNTIME_RE.search(stdout).group("runtime")
 
     def summary(self):
-        self.barplot_single_arg("{runtime}",
-                                ylabel='"runtime in s"',
-                                title='"raxml-ng tree inference benchmark"',
-                                filepostfix="runtime")
+        plt.barplot_single_arg(self,
+                               "{runtime}",
+                               ylabel='"runtime in s"',
+                               title='"raxml-ng tree inference benchmark"',
+                               filepostfix="runtime")
 
-        self.export_stats_to_dataref("runtime")
+        plt.export_stats_to_dataref(self, "runtime")
 
-        self.barplot_single_arg("{VmHWM}",
-                                ylabel='"VmHWM in KB"',
-                                title='"raxml-ng memusage"',
-                                filepostfix="memusage")
+        plt.barplot_single_arg(self,
+                               "{VmHWM}",
+                               ylabel='"VmHWM in KB"',
+                               title='"raxml-ng memusage"',
+                               filepostfix="memusage")
 
-        self.export_stats_to_dataref("VmHWM")
+        plt.export_stats_to_dataref(self, "VmHWM")
 
 
 raxmlng = BenchmarkRaxmlng()
diff --git a/src/plots.py b/src/plots.py
new file mode 100644
index 0000000..b6ef4f8
--- /dev/null
+++ b/src/plots.py
@@ -0,0 +1,518 @@
+import os
+
+import matplotlib
+import matplotlib.pyplot as plt
+import numpy as np
+import tikzplotlib
+
+from src.benchmark import Benchmark
+import src.globalvars
+from src.util import print_warn
+
+# This is useful when evaluating strings in the plot functions. str(np.NaN) == "nan"
+nan = np.NaN
+
+
+###### Summary helpers ######
+def _eval_with_stat(bench, evaluation, alloc, perm, stat):
+    try:
+        s = evaluation.format(**bench.results["stats"][alloc][perm][stat])
+    except KeyError as e:
+        import traceback
+        print_warn(traceback.format_exc())
+        print_warn(f"For {alloc} in {perm}")
+        return nan
+    return eval(s)
+
+def plot_single_arg(bench, yval, ylabel="'y-label'", xlabel="'x-label'",
+                    autoticks=True, title="'default title'", filepostfix="",
+                    sumdir="", arg="", scale=None, file_ext=src.globalvars.summary_file_ext):
+
+    args = bench.results["args"]
+    allocators = bench.results["allocators"]
+
+    arg = arg or list(args.keys())[0]
+
+    if not autoticks:
+        x_vals = list(range(1, len(args[arg]) + 1))
+    else:
+        x_vals = args[arg]
+
+    for allocator in allocators:
+        y_vals = []
+        for perm in bench.iterate_args(args=args):
+            if scale:
+                if scale == allocator:
+                    y_vals = [1] * len(x_vals)
+                else:
+                    mean = _eval_with_stat(bench, yval, allocator, perm, "mean")
+                    norm_mean = _eval_with_stat(bench, yval, scale, perm, "mean")
+                    y_vals.append(mean / norm_mean)
+            else:
+                y_vals.append(_eval_with_stat(bench, yval, allocator, perm, "mean"))
+
+        plt.plot(x_vals, y_vals, marker='.', linestyle='-',
+                 label=allocator, color=allocators[allocator]["color"])
+
+    plt.legend(loc="best")
+    if not autoticks:
+        plt.xticks(x_vals, args[arg])
+    plt.xlabel(eval(xlabel))
+    plt.ylabel(eval(ylabel))
+    plt.title(eval(title))
+    figname = os.path.join(sumdir, f"{bench.name}.{filepostfix}.{file_ext}")
+    if figname.endswith(".tex"):
+        tikzplotlib.save(figname)
+    else:
+        plt.savefig(figname)
+    plt.clf()
+
+def barplot_single_arg(bench, yval, ylabel="'y-label'", xlabel="'x-label'",
+                       title="'default title'", filepostfix="", sumdir="",
+                       arg="", scale=None, file_ext=src.globalvars.summary_file_ext, yerr=True):
+
+    args = bench.results["args"]
+    allocators = bench.results["allocators"]
+    nallocators = len(allocators)
+
+    if arg:
+        arg = args[arg]
+    elif args.keys():
+        arg = args[list(args.keys())[0]]
+    else:
+        arg = [""]
+
+    narg = len(arg)
+
+    for i, allocator in enumerate(allocators):
+        x_vals = list(range(i, narg * (nallocators+1), nallocators+1))
+        y_vals = []
+        y_errs = None
+        if yerr:
+            y_errs = []
+
+        for perm in bench.iterate_args(args=args):
+            if scale:
+                if scale == allocator:
+                    y_vals = [1] * len(x_vals)
+                else:
+                    mean = _eval_with_stat(bench, yval, allocator, perm, "mean")
+                    norm_mean = _eval_with_stat(bench, yval, scale, perm, "mean")
+                    y_vals.append(mean / norm_mean)
+            else:
+                y_vals.append(_eval_with_stat(bench, yval, allocator, perm, "mean"))
+
+            if yerr:
+                y_errs.append(_eval_with_stat(bench, yval, allocator, perm, "std"))
+
+        plt.bar(x_vals, y_vals, width=1, label=allocator, yerr=y_errs,
+                color=allocators[allocator]["color"])
+
+    plt.legend(loc="best")
+    plt.xticks(list(range(int(np.floor(nallocators/2)), narg*(nallocators+1), nallocators+1)), arg)
+    plt.xlabel(eval(xlabel))
+    plt.ylabel(eval(ylabel))
+    plt.title(eval(title))
+    figname = os.path.join(sumdir, f"{bench.name}.{filepostfix}.{file_ext}")
+    if figname.endswith(".tex"):
+        import tikzplotlib
+        tikzplotlib.save(figname)
+    else:
+        plt.savefig(figname)
+    plt.clf()
+
+def plot_fixed_arg(bench, yval, ylabel="'y-label'", xlabel="loose_arg",
+                   autoticks=True, title="'default title'", filepostfix="",
+                   sumdir="", fixed=None, file_ext=src.globalvars.summary_file_ext, scale=None):
+
+    args = bench.results["args"]
+    allocators = bench.results["allocators"]
+
+    for arg in fixed or args:
+        loose_arg = [a for a in args if a != arg][0]
+
+        if not autoticks:
+            x_vals = list(range(1, len(args[loose_arg]) + 1))
+        else:
+            x_vals = args[loose_arg]
+
+        for arg_value in args[arg]:
+            for allocator in allocators:
+                y_vals = []
+                for perm in bench.iterate_args_fixed({arg: arg_value}, args=args):
+                    if scale:
+                        if scale == allocator:
+                            y_vals = [1] * len(x_vals)
+                        else:
+                            mean = _eval_with_stat(bench, yval, allocator, perm, "mean")
+                            norm_mean = _eval_with_stat(bench, yval, scale, perm, "mean")
+                            y_vals.append(mean / norm_mean)
+                    else:
+                        y_vals.append(_eval_with_stat(bench, yval, allocator, perm, "mean"))
+
+                plt.plot(x_vals, y_vals, marker='.', linestyle='-',
+                         label=allocator, color=allocators[allocator]["color"])
+
+            plt.legend(loc="best")
+            if not autoticks:
+                plt.xticks(x_vals, args[loose_arg])
+            plt.xlabel(eval(xlabel))
+            plt.ylabel(eval(ylabel))
+            plt.title(eval(title))
+            figname = os.path.join(sumdir,
+                                   f"{bench.name}.{arg}.{arg_value}.{filepostfix}.{file_ext}")
+            if figname.endswith(".tex"):
+                import tikzplotlib
+                tikzplotlib.save(figname)
+            else:
+                plt.savefig(figname)
+            plt.clf()
+
+def export_facts_to_file(bench, comment_symbol, f):
+    """Write collected facts about used system and benchmark to file"""
+    print(comment_symbol, bench.name, file=f)
+    print(file=f)
+    print(comment_symbol, "Common facts:", file=f)
+    for k, v in src.facter.FACTS.items():
+        print(comment_symbol, k + ":", v, file=f)
+    print(file=f)
+    print(comment_symbol, "Benchmark facts:", file=f)
+    for k, v in bench.results["facts"].items():
+        print(comment_symbol, k + ":", v, file=f)
+    print(file=f)
+
+def export_stats_to_csv(bench, datapoint, path=None):
+    """Write descriptive statistics about datapoint to csv file"""
+    allocators = bench.results["allocators"]
+    args = bench.results["args"]
+    stats = bench.results["stats"]
+
+    if path is None:
+        path = datapoint
+
+    path = path + ".csv"
+
+    stats_fields = list(stats[list(allocators)[0]][list(bench.iterate_args(args=args))[0]])
+    fieldnames = ["allocator", *args, *stats_fields]
+    widths = []
+    for fieldname in fieldnames:
+        widths.append(len(fieldname) + 2)
+
+    # collect rows
+    rows = {}
+    for alloc in allocators:
+        rows[alloc] = {}
+        for perm in bench.iterate_args(args=args):
+            d = []
+            d.append(alloc)
+            d += list(perm._asdict().values())
+            d += [stats[alloc][perm][s][datapoint] for s in stats[alloc][perm]]
+            d[-1] = (",".join([str(x) for x in d[-1]]))
+            rows[alloc][perm] = d
+
+    # calc widths
+    for i in range(0, len(fieldnames)):
+        for alloc in allocators:
+            for perm in bench.iterate_args(args=args):
+                field_len = len(str(rows[alloc][perm][i])) + 2
+                if field_len > widths[i]:
+                    widths[i] = field_len
+
+    with open(path, "w") as f:
+        headerline = ""
+        for i, h in enumerate(fieldnames):
+            headerline += h.capitalize().ljust(widths[i]).replace("_", "-")
+        print(headerline, file=f)
+
+        for alloc in allocators:
+            for perm in bench.iterate_args(args=args):
+                line = ""
+                for i, x in enumerate(rows[alloc][perm]):
+                    line += str(x).ljust(widths[i])
+                print(line.replace("_", "-"), file=f)
+
+def export_stats_to_dataref(bench, datapoint, path=None):
+    """Write descriptive statistics about datapoint to dataref file"""
+    stats = bench.results["stats"]
+
+    if path is None:
+        path = datapoint
+
+    path = path + ".dataref"
+
+    # Example: \drefset{/mysql/glibc/40/Lower-whisker}{71552.0}
+    line = "\\drefset{{/{}/{}/{}/{}}}{{{}}}"
+
+    with open(path, "w") as f:
+        # Write facts to file
+        export_facts_to_file(bench, "%", f)
+
+        for alloc in bench.results["allocators"]:
+            for perm in bench.iterate_args(args=bench.results["args"]):
+                for statistic, values in stats[alloc][perm].items():
+                    cur_line = line.format(bench.name, alloc,
+                                           "/".join([str(p) for p in list(perm)]),
+                                           statistic, values[datapoint])
+                    # Replace empty outliers
+                    cur_line.replace("[]", "")
+                    # Replace underscores
+                    cur_line.replace("_", "-")
+                    print(cur_line, file=f)
+
+def write_best_doublearg_tex_table(bench, evaluation, sort=">",
+                                   filepostfix="", sumdir="", std=False):
+    args = bench.results["args"]
+    keys = list(args.keys())
+    allocators = bench.results["allocators"]
+
+    header_arg = keys[0] if len(args[keys[0]]) < len(args[keys[1]]) else keys[1]
+    row_arg = [arg for arg in args if arg != header_arg][0]
+
+    headers = args[header_arg]
+    rows = args[row_arg]
+
+    cell_text = []
+    for av in rows:
+        row = []
+        for perm in bench.iterate_args_fixed({row_arg: av}, args=args):
+            best = []
+            best_val = None
+            for allocator in allocators:
+                d = []
+                for m in bench.results[allocator][perm]:
+                    d.append(eval(evaluation.format(**m)))
+                mean = np.mean(d)
+                if not best_val:
+                    best = [allocator]
+                    best_val = mean
+                elif ((sort == ">" and mean > best_val)
+                      or (sort == "<" and mean < best_val)):
+                    best = [allocator]
+                    best_val = mean
+                elif mean == best_val:
+                    best.append(allocator)
+
+            row.append("{}: {:.3f}".format(best[0], best_val))
+        cell_text.append(row)
+
+    fname = os.path.join(sumdir, ".".join([bench.name, filepostfix, "tex"]))
+    with open(fname, "w") as f:
+        print("\\documentclass{standalone}", file=f)
+        print("\\begin{document}", file=f)
+        print("\\begin{tabular}{|", end="", file=f)
+        print(" l |" * len(headers), "}", file=f)
+
+        print(header_arg+"/"+row_arg, end=" & ", file=f)
+        for header in headers[:-1]:
+            print(header, end="& ", file=f)
+        print(headers[-1], "\\\\", file=f)
+
+        for i, row in enumerate(cell_text):
+            print(rows[i], end=" & ", file=f)
+            for e in row[:-1]:
+                print(e, end=" & ", file=f)
+            print(row[-1], "\\\\", file=f)
+        print("\\end{tabular}", file=f)
+        print("\\end{document}", file=f)
+
+def write_tex_table(bench, entries, sort=">",
+                    filepostfix="", sumdir="", std=False):
+    """generate a latex standalone table from an list of entries dictionaries
+
+    Entries must have at least the two keys: "label" and "expression".
+    The optional "sort" key specifies the direction of the order:
+        ">" : bigger is better.
+        "<" : smaller is better.
+
+    Table layout:
+
+    |    alloc1     |    alloc2    | ....
+    ---------------------------------------
+    | name1  name2  | ...
+    ---------------------------------------
+    perm1 | eavl1  eval2  | ...
+    perm2 | eval1  eval2  | ...
+    """
+    args = bench.results["args"]
+    allocators = bench.results["allocators"]
+    nallocators = len(allocators)
+    nentries = len(entries)
+    perm_fields = bench.Perm._fields
+    nperm_fields = len(perm_fields)
+
+    alloc_header_line = f"\\multicolumn{{{nperm_fields}}}{{c|}}{{}} &"
+    for alloc in allocators:
+        alloc_header_line += f"\\multicolumn{{{nentries}}}{{c|}}{{{alloc}}} &"
+    alloc_header_line = alloc_header_line[:-1] + "\\\\"
+
+    perm_fields_header = ""
+    for field in bench.Perm._fields:
+        perm_fields_header += f'{field} &'
+    entry_header_line = ""
+    for entry in entries:
+        entry_header_line += f'{entry["label"]} &'
+    entry_header_line = perm_fields_header + entry_header_line * nallocators
+    entry_header_line = entry_header_line[:-1] + "\\\\"
+
+    fname = os.path.join(sumdir, ".".join([bench.name, filepostfix, "tex"]))
+    with open(fname, "w") as f:
+        print("\\documentclass{standalone}", file=f)
+        print("\\usepackage{booktabs}", file=f)
+        print("\\usepackage{xcolor}", file=f)
+        print("\\begin{document}", file=f)
+        print("\\begin{tabular}{|", f"{'c|'*nperm_fields}", f"{'c'*nentries}|"*nallocators, "}", file=f)
+        print("\\toprule", file=f)
+
+        print(alloc_header_line, file=f)
+        print("\\hline", file=f)
+        print(entry_header_line, file=f)
+        print("\\hline", file=f)
+
+        for perm in bench.iterate_args(args=args):
+            values = [[] for _ in entries]
+            maxs = [None for _ in entries]
+            mins = [None for _ in entries]
+            for allocator in allocators:
+                for i, entry in enumerate(entries):
+                    expr = entry["expression"]
+                    values[i].append(eval(expr.format(**bench.results["stats"][allocator][perm]["mean"])))
+
+            # get max and min for each entry
+            for i, entry in enumerate(entries):
+                if not "sort" in entry:
+                    continue
+                # bigger is better
+                elif entry["sort"] == ">":
+                    maxs[i] = max(values[i])
+                    mins[i] = min(values[i])
+                # smaller is better
+                elif entry["sort"] == "<":
+                    mins[i] = max(values[i])
+                    maxs[i] = min(values[i])
+
+            # build row
+            row = ""
+            perm_dict = perm._asdict()
+            for field in perm_fields:
+                row += str(perm_dict[field]) + "&"
+
+            for i, _ in enumerate(allocators):
+                for y, entry_vals in enumerate(values):
+                    val = entry_vals[i]
+
+                    # format
+                    val_str = str(val)
+                    if type(val) == float:
+                        val_str = f"{val:.2f}"
+
+                    # colorize
+                    if val == maxs[y]:
+                        val_str = f"\\textcolor{{green}}{{{val_str}}}"
+                    elif val == mins[y]:
+                        val_str = f"\\textcolor{{red}}{{{val_str}}}"
+                    row += f"{val_str} &"
+            #escape _ for latex
+            row = row.replace("_", "\\_")
+            print(row[:-1], "\\\\", file=f)
+
+        print("\\end{tabular}", file=f)
+        print("\\end{document}", file=f)
+
+def pgfplot_legend(bench, sumdir=""):
+
+    allocators = bench.results["allocators"]
+    s =\
+"""
+\\documentclass{standalone}
+\\usepackage{pgfplots}
+
+\\usepackage{pgfkeys}
+
+\\newenvironment{customlegend}[1][]{%
+\t\\begingroup
+\t\\csname pgfplots@init@cleared@structures\\endcsname
+\t\\pgfplotsset{#1}%
+}{%
+\t\\csname pgfplots@createlegend\\endcsname
+\t\\endgroup
+}%
+\\def\\addlegendimage{\\csname pgfplots@addlegendimage\\endcsname}
+
+\\usepackage{xcolor}
+"""
+
+    for alloc_name, alloc_dict in allocators.items():
+        # define color
+        rgb = matplotlib.colors.to_rgb(alloc_dict["color"])
+        s += f"\\providecolor{{{alloc_name}-color}}{{rgb}}{{{rgb[0]},{rgb[1]},{rgb[2]}}}\n"
+
+    s +=\
+"""
+\\begin{document}
+\\begin{tikzpicture}
+\\begin{customlegend}[
+\tlegend entries={"""
+
+    alloc_list = ""
+    addlegendimage_list = ""
+    for alloc_name in allocators:
+        alloc_list += f"{alloc_name}, "
+        addlegendimage_list += "\t\\addlegendimage{}\n"
+
+    s += alloc_list[:-2] + "},\n]"
+    s += addlegendimage_list
+    s +=\
+"""
+\\end{customlegend}
+\\end{tikzpicture}
+\\end{document}"""
+
+    with open(os.path.join(sumdir, "legend.tex"), "w") as legend_file:
+        print(s, file=legend_file)
+
+def pgfplot_linear(bench, perms, xval, yval, ylabel="'y-label'", xlabel="'x-label'",
+                   title="'default title'", postfix="", sumdir="", scale=None):
+
+    allocators = bench.results["allocators"]
+    perms = list(perms)
+    title = eval(title)
+    s =\
+"""\\documentclass{standalone}
+\\usepackage{pgfplots}
+\\usepackage{xcolor}
+"""
+
+    for alloc_name, alloc_dict in allocators.items():
+        s += f"\\begin{{filecontents*}}{{{alloc_name}.dat}}\n"
+        for i, perm in enumerate(perms):
+            x = _eval_with_stat(bench, xval, alloc_name, perm, "mean")
+            y = _eval_with_stat(bench, yval, alloc_name, perm, "mean")
+            s += f"{x} {y}\n"
+        s += "\\end{filecontents*}\n"
+
+        # define color
+        rgb = matplotlib.colors.to_rgb(alloc_dict["color"])
+        s += f"\\providecolor{{{alloc_name}-color}}{{rgb}}{{{rgb[0]},{rgb[1]},{rgb[2]}}}\n"
+
+    s +=\
+f"""
+\\begin{{document}}
+\\begin{{tikzpicture}}
+\\begin{{axis}}[
+\ttitle={{{title}}},
+\txlabel={{{eval(xlabel)}}},
+\tylabel={{{eval(ylabel)}}},
+]
+"""
+
+    for alloc_name in allocators:
+        s += f"\\addplot [{alloc_name}-color] table {{{alloc_name}.dat}};\n"
+        # s += f"\t\\addplot table {{{alloc_name}.dat}};\n"
+
+    s +=\
+"""\\end{axis}
+\\end{tikzpicture}
+\\end{document}"""
+
+    with open(os.path.join(sumdir, f"{bench.name}.{postfix}.tex"), "w") as plot_file:
+        print(s, file=plot_file)
-- 
cgit v1.2.3