[EarlyCSE] Force hash collision with EXPENSIVE_CHECKS

[bryanpkc]

Summary

This patch turns on EarlyCSEDebugHash when EXPENSIVE_CHECKS is defined, to
help detect inconsistencies between hashing and equivalence tests.

Running CTMark at O2 and O3 shows that this change does not significantly
increase the compile time, both with a baseline release build with assertions
enabled, and a release build with assertions and expensive checks enabled.

Program                                        O2     O2-debughash diff
 test-suite...ark/tramp3d-v4/tramp3d-v4.test    31.10  32.64        4.9%
 test-sutie...:: CTMark/ClamAV/clamscan.test    21.25  21.65        1.8%
 test-suite...TMark/7zip/7zip-benchmark.test    50.85  51.46        1.2%
 test-suite :: CTMark/kimwitu++/kc.test         19.14  19.23        0.4%
 test-suite...-typeset/consumer-typeset.test    16.01  16.07        0.4%
 test-suite :: CTMark/lencode/lencod.test       22.79  22.86        0.3%
 test-suite...Mark/mafft/pairlocalalign.test    13.16  13.14       -0.2%
 test-suite :: CTMark/SPASS/SPASS.test          19.50  19.13       -1.9%
 test-suite :: CTMark/Bullet/bullet.test        37.42  36.66       -2.0%
 test-suite...:: CTMark/sqlite3/sqlite3.test    21.44  20.78       -3.1%
 Geomean difference                                                 0.2%

Program                                        ExpnsvO3 ExpnsvO3-dbghsh diff
 test-suite...ark/tramp3d-v4/tramp3d-v4.test   1829.05  1976.60          8.1%
 test-suite...Mark/mafft/pairlocalalign.test     28.06    28.39          1.2%
 test-suite...-typeset/consumer-typeset.test     33.73    33.88          0.4%
 test-suite...TMark/7zip/7zip-benchmark.test     99.23    99.19         -0.0%
 test-suite :: CTMark/lencode/lencod.test        47.97    47.90         -0.1%
 test-suite :: CTMark/SPASS/SPASS.test           39.19    39.09         -0.3%
 test-sutie...:: CTMark/ClamAV/clamscan.test     40.95    40.70         -0.6%
 test-suite :: CTMark/Bullet/bullet.test         73.79    73.32         -0.6%
 test-suite :: CTMark/kimwitu++/kc.test         18.32    136.15         -1.6%
 Geomean difference                                                      0.7%

[lebedev.ri]

That doesn't sound all that bad to me :)
I'd like to hear from those who commented in D86863, but this doesn't seem too bad, given what it protects agains.

[efriedma]

I'm primarily concerned about running into some basic block with a few thousand instructions and compile-time spiking essentially to infinity. Maybe CTMark doesn't have any functions like that, but I see them regularly.

[lebedev.ri]

This review seems to be stuck/dead, consider abandoning if no longer relevant.

llvm/lib/Transforms/Scalar/EarlyCSE.cpp

DEBUG_COUNTER(CSECounter, "early-cse",
               "Controls which instructions are removed");
 
 static cl::opt<unsigned> EarlyCSEMssaOptCap(
     "earlycse-mssa-optimization-cap", cl::init(500), cl::Hidden,
     cl::desc("Enable imprecision in EarlyCSE in pathological cases, in exchange "
              "for faster compile. Caps the MemorySSA clobbering calls."));
 
 static cl::opt<bool> EarlyCSEDebugHash(
-    "earlycse-debug-hash", cl::init(false), cl::Hidden,
-    cl::desc("Perform extra assertion checking to verify that SimpleValue's hash "
-             "function is well-behaved w.r.t. its isEqual predicate"));
+    "earlycse-debug-hash",
+#ifdef EXPENSIVE_CHECKS
+    cl::init(true),
+#else
+    cl::init(false),
+#endif
+    cl::Hidden,
+    cl::desc(
+        "Perform extra assertion checking to verify that SimpleValue's hash "
 
 //===----------------------------------------------------------------------===//
 // SimpleValue
 //===----------------------------------------------------------------------===//
 
 namespace {
 
 /// Struct representing the available values in the scoped hash table.