commit 0c9497da9485ba688c592e5f452b7b1305a519c0
author Mark Mendell <mark.p.mendell@intel.com> Fri Aug 21 09:30:05 2015 -0400
committer Mark Mendell <mark.p.mendell@intel.com> Wed Aug 26 11:18:59 2015 -0400
tree 7ebcf9c4ebea923131bf6064fd5924d54306e2b7
parent 73f455ecb76d063846a82735eb80596ceee8cee3

X86: Use short forward jumps if possible

The optimizing compiler uses 32 bit relative jumps for all forward
jumps, just in case the offset is too large to fit in one byte.  Some of
the generated code knows that the jumps will in fact fit.

Use the 'NearLabel' class for the code generator and intrinsics.

Use the jecxz/jrcxz instruction for string intrinsics.

Unfortunately, conditional jumps to basic blocks don't know enough to
use this, as we don't know how much code will be generated.

This saves a whopping 0.24% for core.oat and boot.oat sizes, but
every little bit helps, and it reduces icache footprint slightly.

Change-Id: I633fe3b2e0e810b4ce12fdad8c02135644b63506
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>

compiler/optimizing/intrinsics_x86.cc

diff --git a/compiler/optimizing/intrinsics_x86.cc b/compiler/optimizing/intrinsics_x86.cc
index b7126b2..cc4e705 100644
--- a/compiler/optimizing/intrinsics_x86.cc
+++ b/compiler/optimizing/intrinsics_x86.cc
@@ -508,7 +508,7 @@
 
   XmmRegister op2 = op2_loc.AsFpuRegister<XmmRegister>();
 
-  Label nan, done, op2_label;
+  NearLabel nan, done, op2_label;
   if (is_double) {
     __ ucomisd(out, op2);
   } else {
@@ -842,7 +842,7 @@
   Register out = locations->Out().AsRegister<Register>();
   XmmRegister maxInt = locations->GetTemp(0).AsFpuRegister<XmmRegister>();
   XmmRegister inPlusPointFive = locations->GetTemp(1).AsFpuRegister<XmmRegister>();
-  Label done, nan;
+  NearLabel done, nan;
   X86Assembler* assembler = GetAssembler();
 
   // Generate 0.5 into inPlusPointFive.
@@ -971,9 +971,7 @@
   Register edi = locations->GetTemp(1).AsRegister<Register>();
   Register esi = locations->Out().AsRegister<Register>();
 
-  Label end;
-  Label return_true;
-  Label return_false;
+  NearLabel end, return_true, return_false;
 
   // Get offsets of count, value, and class fields within a string object.
   const uint32_t count_offset = mirror::String::CountOffset().Uint32Value();
@@ -1005,8 +1003,7 @@
   __ cmpl(ecx, Address(arg, count_offset));
   __ j(kNotEqual, &return_false);
   // Return true if both strings are empty.
-  __ testl(ecx, ecx);
-  __ j(kEqual, &return_true);
+  __ jecxz(&return_true);
 
   // Load starting addresses of string values into ESI/EDI as required for repe_cmpsl instruction.
   __ leal(esi, Address(str, value_offset));
@@ -1116,7 +1113,7 @@
 
   // Do a zero-length check.
   // TODO: Support jecxz.
-  Label not_found_label;
+  NearLabel not_found_label;
   __ testl(string_length, string_length);
   __ j(kEqual, &not_found_label);
 
@@ -1159,7 +1156,7 @@
   __ subl(string_length, counter);
   __ leal(out, Address(string_length, -1));
 
-  Label done;
+  NearLabel done;
   __ jmp(&done);
 
   // Failed to match; return -1.
@@ -1879,7 +1876,7 @@
     }
 
     // BSR sets ZF if the input was zero, and the output is undefined.
-    Label all_zeroes, done;
+    NearLabel all_zeroes, done;
     __ j(kEqual, &all_zeroes);
 
     // Correct the result from BSR to get the final CLZ result.
@@ -1898,7 +1895,7 @@
   DCHECK(src.IsRegisterPair());
   Register src_lo = src.AsRegisterPairLow<Register>();
   Register src_hi = src.AsRegisterPairHigh<Register>();
-  Label handle_low, done, all_zeroes;
+  NearLabel handle_low, done, all_zeroes;
 
   // Is the high word zero?
   __ testl(src_hi, src_hi);