[llvm-commits] [llvm] r50309 - in /llvm/branches/Apple/Tak: lib/Transforms/IPO/GlobalOpt.cpp test/Transforms/GlobalOpt/2008-04-26-SROA-Global-Align.ll
Bill Wendling
isanbard at gmail.com
Sat Apr 26 02:30:01 PDT 2008
Author: void
Date: Sat Apr 26 04:30:01 2008
New Revision: 50309
URL: http://llvm.org/viewvc/llvm-project?rev=50309&view=rev
Log:
Pulled r50308 into Tak.
Added:
llvm/branches/Apple/Tak/test/Transforms/GlobalOpt/2008-04-26-SROA-Global-Align.ll
Modified:
llvm/branches/Apple/Tak/lib/Transforms/IPO/GlobalOpt.cpp
Modified: llvm/branches/Apple/Tak/lib/Transforms/IPO/GlobalOpt.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/Apple/Tak/lib/Transforms/IPO/GlobalOpt.cpp?rev=50309&r1=50308&r2=50309&view=diff
==============================================================================
--- llvm/branches/Apple/Tak/lib/Transforms/IPO/GlobalOpt.cpp (original)
+++ llvm/branches/Apple/Tak/lib/Transforms/IPO/GlobalOpt.cpp Sat Apr 26 04:30:01 2008
@@ -28,6 +28,7 @@
#include "llvm/Support/Compiler.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/GetElementPtrTypeIterator.h"
+#include "llvm/Support/MathExtras.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Statistic.h"
@@ -455,7 +456,7 @@
/// behavior of the program in a more fine-grained way. We have determined that
/// this transformation is safe already. We return the first global variable we
/// insert so that the caller can reprocess it.
-static GlobalVariable *SRAGlobal(GlobalVariable *GV) {
+static GlobalVariable *SRAGlobal(GlobalVariable *GV, const TargetData &TD) {
// Make sure this global only has simple uses that we can SRA.
if (!GlobalUsersSafeToSRA(GV))
return 0;
@@ -467,8 +468,14 @@
std::vector<GlobalVariable*> NewGlobals;
Module::GlobalListType &Globals = GV->getParent()->getGlobalList();
+ // Get the alignment of the global, either explicit or target-specific.
+ unsigned StartAlignment = GV->getAlignment();
+ if (StartAlignment == 0)
+ StartAlignment = TD.getABITypeAlignment(GV->getType());
+
if (const StructType *STy = dyn_cast<StructType>(Ty)) {
NewGlobals.reserve(STy->getNumElements());
+ const StructLayout &Layout = *TD.getStructLayout(STy);
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
Constant *In = getAggregateConstantElement(Init,
ConstantInt::get(Type::Int32Ty, i));
@@ -480,19 +487,28 @@
GV->isThreadLocal());
Globals.insert(GV, NGV);
NewGlobals.push_back(NGV);
+
+ // Calculate the known alignment of the field. If the original aggregate
+ // had 256 byte alignment for example, something might depend on that:
+ // propagate info to each field.
+ uint64_t FieldOffset = Layout.getElementOffset(i);
+ unsigned NewAlign = (unsigned)MinAlign(StartAlignment, FieldOffset);
+ if (NewAlign > TD.getABITypeAlignment(STy->getElementType(i)))
+ NGV->setAlignment(NewAlign);
}
} else if (const SequentialType *STy = dyn_cast<SequentialType>(Ty)) {
unsigned NumElements = 0;
if (const ArrayType *ATy = dyn_cast<ArrayType>(STy))
NumElements = ATy->getNumElements();
- else if (const VectorType *PTy = dyn_cast<VectorType>(STy))
- NumElements = PTy->getNumElements();
else
- assert(0 && "Unknown aggregate sequential type!");
+ NumElements = cast<VectorType>(STy)->getNumElements();
if (NumElements > 16 && GV->hasNUsesOrMore(16))
return 0; // It's not worth it.
NewGlobals.reserve(NumElements);
+
+ uint64_t EltSize = TD.getABITypeSize(STy->getElementType());
+ unsigned EltAlign = TD.getABITypeAlignment(STy->getElementType());
for (unsigned i = 0, e = NumElements; i != e; ++i) {
Constant *In = getAggregateConstantElement(Init,
ConstantInt::get(Type::Int32Ty, i));
@@ -505,6 +521,13 @@
GV->isThreadLocal());
Globals.insert(GV, NGV);
NewGlobals.push_back(NGV);
+
+ // Calculate the known alignment of the field. If the original aggregate
+ // had 256 byte alignment for example, something might depend on that:
+ // propagate info to each field.
+ unsigned NewAlign = (unsigned)MinAlign(StartAlignment, EltSize*i);
+ if (NewAlign > EltAlign)
+ NGV->setAlignment(NewAlign);
}
}
@@ -804,6 +827,9 @@
GV->getName()+".body",
(Module *)NULL,
GV->isThreadLocal());
+ // FIXME: This new global should have the alignment returned by malloc. Code
+ // could depend on malloc returning large alignment (on the mac, 16 bytes) but
+ // this would only guarantee some lower alignment.
GV->getParent()->getGlobalList().insert(GV, NewGV);
// Anything that used the malloc now uses the global directly.
@@ -1520,7 +1546,8 @@
++NumMarked;
return true;
} else if (!GV->getInitializer()->getType()->isFirstClassType()) {
- if (GlobalVariable *FirstNewGV = SRAGlobal(GV)) {
+ if (GlobalVariable *FirstNewGV = SRAGlobal(GV,
+ getAnalysis<TargetData>())) {
GVI = FirstNewGV; // Don't skip the newly produced globals!
return true;
}
Added: llvm/branches/Apple/Tak/test/Transforms/GlobalOpt/2008-04-26-SROA-Global-Align.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/Apple/Tak/test/Transforms/GlobalOpt/2008-04-26-SROA-Global-Align.ll?rev=50309&view=auto
==============================================================================
--- llvm/branches/Apple/Tak/test/Transforms/GlobalOpt/2008-04-26-SROA-Global-Align.ll (added)
+++ llvm/branches/Apple/Tak/test/Transforms/GlobalOpt/2008-04-26-SROA-Global-Align.ll Sat Apr 26 04:30:01 2008
@@ -0,0 +1,32 @@
+; Verify that when @G is SROA'd that the new globals have correct
+; alignments. Elements 0 and 2 must be 16-byte aligned, and element
+; 1 must be at least 8 byte aligned (but could be more).
+
+; RUN: llvm-as < %s | opt -globalopt | llvm-dis | grep {@G.0 = internal global .*align 16}
+; RUN: llvm-as < %s | opt -globalopt | llvm-dis | grep {@G.1 = internal global .*align 8}
+; RUN: llvm-as < %s | opt -globalopt | llvm-dis | grep {@G.2 = internal global .*align 16}
+; rdar://5891920
+
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:32:32-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128"
+target triple = "x86_64-apple-darwin8"
+
+%T = type { double, double, double }
+
+ at G = internal global %T zeroinitializer, align 16
+
+
+define void @test() {
+ store double 1.0, double* getelementptr (%T* @G, i32 0, i32 0), align 16
+ store double 2.0, double* getelementptr (%T* @G, i32 0, i32 1), align 8
+ store double 3.0, double* getelementptr (%T* @G, i32 0, i32 2), align 16
+ ret void
+}
+
+define double @test2() {
+ %V1 = load double* getelementptr (%T* @G, i32 0, i32 0), align 16
+ %V2 = load double* getelementptr (%T* @G, i32 0, i32 1), align 8
+ %V3 = load double* getelementptr (%T* @G, i32 0, i32 2), align 16
+ %R = add double %V1, %V2
+ %R2 = add double %R, %V3
+ ret double %R2
+}
More information about the llvm-commits
mailing list