cpython: Enable LTO on all builds as it doesn't break reproducibility

LTO allows us to optimise the binaries we are shipping a bit further
than just with the regular -Ox and other compiler flags. It also is
deterministic and doesn't harm our reproducibility efforts while
providing us with up to 10% performance gain (and sometimes slightly
slower). See the table below for a comparsion of this version with the
Python 3.9 build that sets -fno-semantic-interposition.

+-------------------------+--------------------+------------------------+--------------+------------------------+
| Benchmark               | py39-nsip.nix.json | py39-nsip-lto.nix.json | Change       | Significance           |
+=========================+====================+========================+==============+========================+
| 2to3                    | 642 ms             | 620 ms                 | 1.03x faster | Significant (t=12.04)  |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| chameleon               | 14.6 ms            | 14.4 ms                | 1.02x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| chaos                   | 182 ms             | 182 ms                 | 1.00x slower | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| crypto_pyaes            | 175 ms             | 172 ms                 | 1.02x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| deltablue               | 11.2 ms            | 11.2 ms                | 1.00x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| django_template         | 82.0 ms            | 81.4 ms                | 1.01x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| dulwich_log             | 101 ms             | 99.8 ms                | 1.01x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| fannkuch                | 634 ms             | 638 ms                 | 1.01x slower | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| float                   | 176 ms             | 189 ms                 | 1.08x slower | Significant (t=-4.30)  |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| go                      | 366 ms             | 365 ms                 | 1.00x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| hexiom                  | 14.9 ms            | 15.1 ms                | 1.01x slower | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| json_dumps              | 18.7 ms            | 18.5 ms                | 1.01x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| json_loads              | 37.9 us            | 37.3 us                | 1.02x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| logging_format          | 14.2 us            | 14.3 us                | 1.01x slower | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| logging_silent          | 305 ns             | 313 ns                 | 1.02x slower | Significant (t=-3.91)  |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| logging_simple          | 13.0 us            | 13.2 us                | 1.01x slower | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| mako                    | 24.7 ms            | 23.3 ms                | 1.06x faster | Significant (t=10.74)  |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| meteor_contest          | 130 ms             | 128 ms                 | 1.01x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| nbody                   | 201 ms             | 201 ms                 | 1.00x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| nqueens                 | 152 ms             | 154 ms                 | 1.01x slower | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| pathlib                 | 26.3 ms            | 26.2 ms                | 1.01x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| pickle                  | 13.1 us            | 13.7 us                | 1.05x slower | Significant (t=-10.36) |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| pickle_dict             | 26.6 us            | 27.1 us                | 1.02x slower | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| pickle_list             | 4.34 us            | 4.31 us                | 1.01x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| pickle_pure_python      | 738 us             | 759 us                 | 1.03x slower | Significant (t=-5.26)  |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| pidigits                | 181 ms             | 181 ms                 | 1.00x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| pyflate                 | 959 ms             | 974 ms                 | 1.02x slower | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| python_startup          | 34.0 ms            | 31.4 ms                | 1.08x faster | Significant (t=29.75)  |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| python_startup_no_site  | 15.3 ms            | 14.5 ms                | 1.06x faster | Significant (t=17.58)  |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| raytrace                | 849 ms             | 846 ms                 | 1.00x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| regex_compile           | 261 ms             | 261 ms                 | 1.00x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| regex_dna               | 187 ms             | 221 ms                 | 1.18x slower | Significant (t=-46.94) |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| regex_effbot            | 3.66 ms            | 3.98 ms                | 1.09x slower | Significant (t=-18.55) |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| regex_v8                | 29.7 ms            | 29.3 ms                | 1.01x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| richards                | 113 ms             | 110 ms                 | 1.03x faster | Significant (t=3.47)   |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| scimark_fft             | 592 ms             | 590 ms                 | 1.00x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| scimark_lu              | 253 ms             | 265 ms                 | 1.05x slower | Significant (t=-8.97)  |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| scimark_monte_carlo     | 170 ms             | 177 ms                 | 1.04x slower | Significant (t=-4.96)  |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| scimark_sor             | 310 ms             | 315 ms                 | 1.02x slower | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| scimark_sparse_mat_mult | 8.36 ms            | 8.33 ms                | 1.00x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| spectral_norm           | 232 ms             | 229 ms                 | 1.02x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| sqlalchemy_declarative  | 185 ms             | 183 ms                 | 1.01x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| sqlalchemy_imperative   | 27.4 ms            | 27.6 ms                | 1.01x slower | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| sqlite_synth            | 4.73 us            | 4.73 us                | 1.00x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| sympy_expand            | 813 ms             | 819 ms                 | 1.01x slower | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| sympy_integrate         | 31.6 ms            | 31.8 ms                | 1.01x slower | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| sympy_str               | 477 ms             | 479 ms                 | 1.00x slower | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| sympy_sum               | 245 ms             | 247 ms                 | 1.01x slower | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| telco                   | 11.3 ms            | 11.4 ms                | 1.00x slower | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| tornado_http            | 172 ms             | 172 ms                 | 1.00x slower | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| unpack_sequence         | 56.2 ns            | 51.2 ns                | 1.10x faster | Significant (t=2.50)   |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| unpickle                | 19.8 us            | 19.5 us                | 1.02x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| unpickle_list           | 5.75 us            | 5.75 us                | 1.00x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| unpickle_pure_python    | 524 us             | 522 us                 | 1.01x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| xml_etree_generate      | 148 ms             | 148 ms                 | 1.00x slower | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| xml_etree_iterparse     | 129 ms             | 131 ms                 | 1.02x slower | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| xml_etree_parse         | 179 ms             | 177 ms                 | 1.01x faster | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+
| xml_etree_process       | 118 ms             | 119 ms                 | 1.00x slower | Not significant        |
+-------------------------+--------------------+------------------------+--------------+------------------------+

2 files changed, 7 insertions, 0 deletions

diff --git a/pkgs/development/interpreters/python/cpython/default.nix b/pkgs/development/interpreters/python/cpython/default.nix
index e0b941e009d..4ee67a452fe 100644
--- a/pkgs/development/interpreters/python/cpython/default.nix
+++ b/pkgs/development/interpreters/python/cpython/default.nix
@@ -41,6 +41,8 @@
 , enableOptimizations ? false
 # enableNoSemanticInterposition is a subset of the enableOptimizations flag that doesn't harm reproducibility.
 , enableNoSemanticInterposition ? true
+# enableLTO is a subset of the enableOptimizations flag that doesn't harm reproducibility.
+, enableLTO ? true
 , reproducibleBuild ? true
 , pythonAttr ? "python${sourceVersion.major}${sourceVersion.minor}"
 }:
@@ -102,6 +104,8 @@ let
   ] ++ optionals (stdenv.hostPlatform != stdenv.buildPlatform) [
     buildPackages.stdenv.cc
     pythonForBuild
+  ] ++ optionals (stdenv.cc.isClang && enableLTO) [
+    stdenv.cc.cc.libllvm.out
   ];
 
   buildInputs = filter (p: p != null) ([
@@ -280,6 +284,8 @@ in with passthru; stdenv.mkDerivation {
     "--with-system-ffi"
   ] ++ optionals enableOptimizations [
     "--enable-optimizations"
+  ] ++ optionals enableLTO [
+    "--with-lto"
   ] ++ optionals (pythonOlder "3.7") [
     # This is unconditionally true starting in CPython 3.7.
     "--with-threads"
diff --git a/pkgs/development/interpreters/python/default.nix b/pkgs/development/interpreters/python/default.nix
index 176f8db4bff..be0d6ce1e43 100644
--- a/pkgs/development/interpreters/python/default.nix
+++ b/pkgs/development/interpreters/python/default.nix
@@ -225,6 +225,7 @@ in {
     stripBytecode = true;
     includeSiteCustomize = false;
     enableOptimizations = false;
+    enableLTO = false;
     mimetypesSupport = false;
   } // sources.python39)).overrideAttrs(old: {
     pname = "python3-minimal";