Add a note.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77584 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Evan Cheng 2009-07-30 08:56:19 +00:00
parent fe0bfe4fa4
commit 7216920791

View File

@ -1932,3 +1932,23 @@ Replacing an icmp+select with a shift should always be considered profitable in
instcombine.
//===---------------------------------------------------------------------===//
Re-implement atomic builtins __sync_add_and_fetch() and __sync_sub_and_fetch
properly.
When the return value is not used (i.e. only care about the value in the
memory), x86 does not have to use add to implement these. Instead, it can use
add, sub, inc, dec instructions with the "lock" prefix.
This is currently implemented using a bit of instruction selection trick. The
issue is the target independent pattern produces one output and a chain and we
want to map it into one that just output a chain. The current trick is to select
it into a MERGE_VALUES with the first definition being an implicit_def. The
proper solution is to add new ISD opcodes for the no-output variant. DAG
combiner can then transform the node before it gets to target node selection.
Problem #2 is we are adding a whole bunch of x86 atomic instructions when in
fact these instructions are identical to the non-lock versions. We need a way to
add target specific information to target nodes and have this information
carried over to machine instructions. Asm printer (or JIT) can use this
information to add the "lock" prefix.