[llvm-commits] [llvm] r105228 - in /llvm/trunk: lib/CodeGen/SelectionDAG/SelectionDAG.cpp test/CodeGen/X86/memcpy.ll

[Chris Lattner]

Author: lattner
Date: Mon May 31 12:30:14 2010
New Revision: 105228

URL: http://llvm.org/viewvc/llvm-project?rev=105228&view=rev
Log:
fix PR6623: when optimizing for size, don't inline memcpy/memsets
that are too large.  This causes the freebsd bootloader to be too
large apparently.

It's unclear if this should be an -Os or -Oz thing.  Thoughts welcome.

Modified:
    llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
    llvm/trunk/test/CodeGen/X86/memcpy.ll

Modified: llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp?rev=105228&r1=105227&r2=105228&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp (original)
+++ llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp Mon May 31 12:30:14 2010
@@ -3266,6 +3266,15 @@
     if (VT.bitsGT(LVT))
       VT = LVT;
   }
+  
+  // If we're optimizing for size, and there is a limit, bump the maximum number
+  // of operations inserted down to 4.  This is a wild guess that approximates
+  // the size of a call to memcpy or memset (3 arguments + call).
+  if (Limit != ~0U) {
+    const Function *F = DAG.getMachineFunction().getFunction();
+    if (F->hasFnAttr(Attribute::OptimizeForSize))
+      Limit = 4;
+  }
 
   unsigned NumMemOps = 0;
   while (Size != 0) {

Modified: llvm/trunk/test/CodeGen/X86/memcpy.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/memcpy.ll?rev=105228&r1=105227&r2=105228&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/memcpy.ll (original)
+++ llvm/trunk/test/CodeGen/X86/memcpy.ll Mon May 31 12:30:14 2010
@@ -25,3 +25,33 @@
 ; CHECK: memcpy
 }
 
+; Large constant memcpy's should lower to a call when optimizing for size.
+; PR6623
+define void @test3(i8* nocapture %A, i8* nocapture %B) nounwind optsize noredzone {
+entry:
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %A, i8* %B, i64 64, i32 1, i1 false)
+  ret void
+; CHECK: test3:
+; CHECK: memcpy
+}
+
+; Large constant memcpy's should be inlined when not optimizing for size.
+define void @test4(i8* nocapture %A, i8* nocapture %B) nounwind noredzone {
+entry:
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %A, i8* %B, i64 64, i32 1, i1 false)
+  ret void
+; CHECK: test4:
+; CHECK: movq
+; CHECK: movq
+; CHECK: movq
+; CHECK: movq
+; CHECK: movq
+; CHECK: movq
+; CHECK: movq
+; CHECK: movq
+; CHECK: movq
+; CHECK: movq
+; CHECK: movq
+; CHECK: movq
+}
+