From f4d1aca56808ba143da704fb51bdf0df52ca6fef Mon Sep 17 00:00:00 2001
From: Michael Kuperstein <michael.m.kuperstein@intel.com>
Date: Thu, 12 Feb 2015 14:17:35 +0000
Subject: [PATCH] [X86] Call frame optimization - allow stack-relative movs to
 be folded into a push

Since we track esp precisely, there's no reason not to allow this.

llvm-svn: 228924
---
 llvm/lib/Target/X86/X86CallFrameOptimization.cpp |  6 ------
 llvm/test/CodeGen/X86/movtopush.ll               | 22 +++++++++++-----------
 2 files changed, 11 insertions(+), 17 deletions(-)

diff --git a/llvm/lib/Target/X86/X86CallFrameOptimization.cpp b/llvm/lib/Target/X86/X86CallFrameOptimization.cpp
index c0c009b..1ede5c7 100644
--- a/llvm/lib/Target/X86/X86CallFrameOptimization.cpp
+++ b/llvm/lib/Target/X86/X86CallFrameOptimization.cpp
@@ -469,12 +469,6 @@ MachineInstr *X86CallFrameOptimization::canFoldIntoRegPush(
       DefMI->getParent() != FrameSetup->getParent())
     return nullptr;
 
-  // Be careful with movs that load from a stack slot, since it may get
-  // resolved incorrectly.
-  // TODO: Again, we already have the infrastructure, so this should work.
-  if (!DefMI->getOperand(1).isReg())
-    return nullptr;
-
   // Now, make sure everything else up until the ADJCALLSTACK is a sequence
   // of MOVs. To be less conservative would require duplicating a lot of the
   // logic from PeepholeOptimizer.
diff --git a/llvm/test/CodeGen/X86/movtopush.ll b/llvm/test/CodeGen/X86/movtopush.ll
index b0cb91a..4a5d903 100644
--- a/llvm/test/CodeGen/X86/movtopush.ll
+++ b/llvm/test/CodeGen/X86/movtopush.ll
@@ -103,7 +103,8 @@ entry:
 ; NORMAL-NEXT: addl $16, %esp
 define void @test3(i32 %k) optsize {
 entry:
-  call void @good(i32 %k, i32 2, i32 3, i32 4)
+  %f = add i32 %k, 1
+  call void @good(i32 %f, i32 2, i32 3, i32 4)
   ret void
 }
 
@@ -200,21 +201,20 @@ entry:
   ret void
 }
 
-; But we don't want to fold stack-relative loads into the push,
-; because the offset will be wrong
+; Fold stack-relative loads into the push, with correct offset
+; In particular, at the second push, %b was at 12(%esp) and
+; %a wast at 8(%esp), but the second push bumped %esp, so %a
+; is now it at 12(%esp)
 ; NORMAL-LABEL: test8:
-; NORMAL-NOT: subl {{.*}} %esp
-; NORMAL: movl 4(%esp), [[EAX:%e..]]
-; NORMAL-NEXT: pushl   $4
-; NORMAL-NEXT: pushl   [[EAX]]
-; NORMAL-NEXT: pushl   $2
+; NORMAL: pushl   $4
+; NORMAL-NEXT: pushl   12(%esp)
+; NORMAL-NEXT: pushl   12(%esp)
 ; NORMAL-NEXT: pushl   $1
 ; NORMAL-NEXT: call
 ; NORMAL-NEXT: addl $16, %esp
-define void @test8(i32* %ptr) optsize {
+define void @test8(i32 %a, i32 %b) optsize {
 entry:
-  %val = ptrtoint i32* %ptr to i32
-  call void @good(i32 1, i32 2, i32 %val, i32 4)
+  call void @good(i32 1, i32 %a, i32 %b, i32 4)
   ret void
 }
 
-- 
2.7.4