This patch substantially simplifies and cleans up handling of basic blocks
authorChris Lattner <sabre@nondot.org>
Wed, 8 Oct 2003 22:52:54 +0000 (22:52 +0000)
committerChris Lattner <sabre@nondot.org>
Wed, 8 Oct 2003 22:52:54 +0000 (22:52 +0000)
commit4ee8ef2a5d5a28cf5255b986d57dc658fa060479
treed08ab107eb58849ab3ed824df815aab381e4aa00
parent6e44802ec03647549c19a7d0313fae009fac7a49
This patch substantially simplifies and cleans up handling of basic blocks
in the bytecode parser.  Before we tried to shoehorn basic blocks into the
"getValue" code path with other types of values.  For a variety of reasons
this was a bad idea, so this patch separates it out into its own data structure.

This simplifies the code, makes it fit in 80 columns, and is also much faster.
In a testcase provided by Bill, which has lots of PHI nodes, this patch speeds
up bytecode parsing from taking 6.9s to taking 2.32s.  More speedups to
follow later.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@8977 91177308-0d34-0410-b5e6-96231b3b80d8
lib/Bytecode/Reader/InstructionReader.cpp
lib/Bytecode/Reader/Reader.cpp
lib/Bytecode/Reader/ReaderInternals.h