Revert "[X86] Elide references to _chkstk for dynamic allocas"
authorReid Kleckner <rnk@google.com>
Wed, 2 Mar 2016 19:20:59 +0000 (19:20 +0000)
committerReid Kleckner <rnk@google.com>
Wed, 2 Mar 2016 19:20:59 +0000 (19:20 +0000)
commit65f9d9cd324fc12f8bf90fa67477e2ddd90803d6
tree59aafeca41a840b088622377df59b63033a5ae15
parentf290912d2248687e688779d1d89999df56c14a09
Revert "[X86] Elide references to _chkstk for dynamic allocas"

This reverts commit r262370.

It turns out there is code out there that does sequences of allocas
greater than 4K: http://crbug.com/591404

The goal of this change was to improve the code size of inalloca call
sequences, but we got tangled up in the mess of dynamic allocas.
Instead, we should come back later with a separate MI pass that uses
dominance to optimize the full sequence. This should also be able to
remove the often unneeded stacksave/stackrestore pairs around the call.

llvm-svn: 262505
llvm/lib/Target/X86/X86ISelLowering.cpp
llvm/test/CodeGen/X86/cleanuppad-inalloca.ll
llvm/test/CodeGen/X86/dynamic-alloca-in-entry.ll
llvm/test/CodeGen/X86/inalloca-ctor.ll
llvm/test/CodeGen/X86/inalloca-invoke.ll
llvm/test/CodeGen/X86/inalloca-stdcall.ll
llvm/test/CodeGen/X86/inalloca.ll
llvm/test/CodeGen/X86/shrink-wrap-chkstk.ll