Turn on LLVM's "basic regalloc" by default

Can still switch back to the default regalloc using the -b flag,
but it looks like the basic register allocator cuts the LLVM JIT
time significantly, and improves every one of our benchmarks.

README.md

@@ -102,7 +102,7 @@ For a full list, please check out the (Makefile)[https://github.com/dropbox/pyst
   <dd>Use a stripped stdlib.  When running pyston_dbg, the default is to use a stdlib with full debugging symbols enabled.  Passing -r changes this behavior to load a slimmer, stripped stdlib.</dd>
 
 <dt>-b</dt>
-  <dd>Experimental: use the LLVM "basic" register allocator.</dd>
+  <dd>Use the default LLVM register allocator instead of the "basic" register allocator.</dd>
 
 <dt>-x</dt>
   <dd>Experimental: use the pypa parser.</dd>

src/codegen/irgen/hooks.cpp

@@ -122,6 +122,11 @@ static void compileIR(CompiledFunction* cf, EffortLevel::EffortLevel effort) {
         us_jitting.log(us);
         static StatCounter num_jits("num_jits");
         num_jits.log();
+
+        if (VERBOSITY() >= 1&& us > 100000) {
+            printf("Took %.1fs to compile %s\n", us * 0.000001, cf->func->getName().str().c_str());
+            printf("Has %ld basic blocks\n", cf->func->getBasicBlockList().size());
+        }
     }
 
     if (VERBOSITY("irgen") >= 1) {

src/core/options.cpp

@@ -34,7 +34,7 @@ bool TRAP = false;
 bool USE_STRIPPED_STDLIB = true; // always true
 bool ENABLE_INTERPRETER = true;
 bool ENABLE_PYPA_PARSER = false;
-bool USE_REGALLOC_BASIC = false;
+bool USE_REGALLOC_BASIC = true;
 
 static bool _GLOBAL_ENABLE = 1;
 bool ENABLE_ICS = 1 && _GLOBAL_ENABLE;

src/jit.cpp

@@ -83,7 +83,7 @@ int main(int argc, char** argv) {
         } else if (code == 'r') {
             USE_STRIPPED_STDLIB = true;
         } else if (code == 'b') {
-            USE_REGALLOC_BASIC = true;
+            USE_REGALLOC_BASIC = false;
         } else if (code == 'x') {
             ENABLE_PYPA_PARSER = true;
         } else if (code == '?')