Add a natural stack alignment field to TargetData, and prevent InstCombine from
promoting allocas to preferred alignments that exceed the natural
alignment. This avoids some potentially expensive dynamic stack realignments.
The natural stack alignment is set in target data strings via the "S<size>"
option. Size is in bits and must be a multiple of 8. The natural stack alignment
defaults to "unspecified" (represented by a zero value), and the "unspecified"
value does not prevent any alignment promotions. Target maintainers that care
about avoiding promotions should explicitly add the "S<size>" option to their
target data strings.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141599 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/Target/ARM/ARMTargetMachine.cpp b/lib/Target/ARM/ARMTargetMachine.cpp
index f779ede..96b1e89 100644
--- a/lib/Target/ARM/ARMTargetMachine.cpp
+++ b/lib/Target/ARM/ARMTargetMachine.cpp
@@ -53,9 +53,12 @@
: ARMBaseTargetMachine(T, TT, CPU, FS, RM, CM), InstrInfo(Subtarget),
DataLayout(Subtarget.isAPCS_ABI() ?
std::string("e-p:32:32-f64:32:64-i64:32:64-"
- "v128:32:128-v64:32:64-n32") :
+ "v128:32:128-v64:32:64-n32-S32") :
+ Subtarget.isAAPCS_ABI() ?
std::string("e-p:32:32-f64:64:64-i64:64:64-"
- "v128:64:128-v64:64:64-n32")),
+ "v128:64:128-v64:64:64-n32-S64") :
+ std::string("e-p:32:32-f64:64:64-i64:64:64-"
+ "v128:64:128-v64:64:64-n32-S32")),
ELFWriterInfo(*this),
TLInfo(*this),
TSInfo(*this),
@@ -75,10 +78,14 @@
DataLayout(Subtarget.isAPCS_ABI() ?
std::string("e-p:32:32-f64:32:64-i64:32:64-"
"i16:16:32-i8:8:32-i1:8:32-"
- "v128:32:128-v64:32:64-a:0:32-n32") :
+ "v128:32:128-v64:32:64-a:0:32-n32-S32") :
+ Subtarget.isAAPCS_ABI() ?
std::string("e-p:32:32-f64:64:64-i64:64:64-"
"i16:16:32-i8:8:32-i1:8:32-"
- "v128:64:128-v64:64:64-a:0:32-n32")),
+ "v128:64:128-v64:64:64-a:0:32-n32-S64") :
+ std::string("e-p:32:32-f64:64:64-i64:64:64-"
+ "i16:16:32-i8:8:32-i1:8:32-"
+ "v128:64:128-v64:64:64-a:0:32-n32-S32")),
ELFWriterInfo(*this),
TLInfo(*this),
TSInfo(*this),
diff --git a/lib/Target/TargetData.cpp b/lib/Target/TargetData.cpp
index 1dfd9a8..bd6a6b6 100644
--- a/lib/Target/TargetData.cpp
+++ b/lib/Target/TargetData.cpp
@@ -139,6 +139,7 @@
PointerMemSize = 8;
PointerABIAlign = 8;
PointerPrefAlign = PointerABIAlign;
+ StackNaturalAlign = 0;
// Default alignments
setAlignment(INTEGER_ALIGN, 1, 1, 1); // i1
@@ -218,7 +219,12 @@
Token = Split.second;
} while (!Specifier.empty() || !Token.empty());
break;
-
+ case 'S': // Stack natural alignment.
+ StackNaturalAlign = getInt(Specifier.substr(1));
+ StackNaturalAlign /= 8;
+ // FIXME: Should we really be truncating these alingments and
+ // sizes silently?
+ break;
default:
break;
}
@@ -372,7 +378,9 @@
OS << (LittleEndian ? "e" : "E")
<< "-p:" << PointerMemSize*8 << ':' << PointerABIAlign*8
- << ':' << PointerPrefAlign*8;
+ << ':' << PointerPrefAlign*8
+ << "-S" << StackNaturalAlign*8;
+
for (unsigned i = 0, e = Alignments.size(); i != e; ++i) {
const TargetAlignElem &AI = Alignments[i];
OS << '-' << (char)AI.AlignType << AI.TypeBitWidth << ':'
diff --git a/lib/Target/X86/README-SSE.txt b/lib/Target/X86/README-SSE.txt
index f16ec02..7d901af 100644
--- a/lib/Target/X86/README-SSE.txt
+++ b/lib/Target/X86/README-SSE.txt
@@ -862,7 +862,7 @@
This IR (from PR6194):
-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-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
+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-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
target triple = "x86_64-apple-darwin10.0.0"
%0 = type { double, double }
diff --git a/lib/Target/X86/README.txt b/lib/Target/X86/README.txt
index b98f5fb..b407955 100644
--- a/lib/Target/X86/README.txt
+++ b/lib/Target/X86/README.txt
@@ -1217,7 +1217,7 @@
Take the following:
-target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128"
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128-S128"
target triple = "i386-apple-darwin8"
@in_exit.4870.b = internal global i1 false ; <i1*> [#uses=2]
define fastcc void @abort_gzip() noreturn nounwind {
diff --git a/lib/Target/X86/X86TargetMachine.cpp b/lib/Target/X86/X86TargetMachine.cpp
index 683d6aa..1dcad35 100644
--- a/lib/Target/X86/X86TargetMachine.cpp
+++ b/lib/Target/X86/X86TargetMachine.cpp
@@ -34,11 +34,11 @@
Reloc::Model RM, CodeModel::Model CM)
: X86TargetMachine(T, TT, CPU, FS, RM, CM, false),
DataLayout(getSubtargetImpl()->isTargetDarwin() ?
- "e-p:32:32-f64:32:64-i64:32:64-f80:128:128-f128:128:128-n8:16:32" :
+ "e-p:32:32-f64:32:64-i64:32:64-f80:128:128-f128:128:128-n8:16:32-S128" :
(getSubtargetImpl()->isTargetCygMing() ||
getSubtargetImpl()->isTargetWindows()) ?
- "e-p:32:32-f64:64:64-i64:64:64-f80:32:32-f128:128:128-n8:16:32" :
- "e-p:32:32-f64:32:64-i64:32:64-f80:32:32-f128:128:128-n8:16:32"),
+ "e-p:32:32-f64:64:64-i64:64:64-f80:32:32-f128:128:128-n8:16:32-S32" :
+ "e-p:32:32-f64:32:64-i64:32:64-f80:32:32-f128:128:128-n8:16:32-S32"),
InstrInfo(*this),
TSInfo(*this),
TLInfo(*this),
@@ -50,7 +50,7 @@
StringRef CPU, StringRef FS,
Reloc::Model RM, CodeModel::Model CM)
: X86TargetMachine(T, TT, CPU, FS, RM, CM, true),
- DataLayout("e-p:64:64-s:64-f64:64:64-i64:64:64-f80:128:128-f128:128:128-n8:16:32:64"),
+ DataLayout("e-p:64:64-s:64-f64:64:64-i64:64:64-f80:128:128-f128:128:128-n8:16:32:64-S128"),
InstrInfo(*this),
TSInfo(*this),
TLInfo(*this),
diff --git a/lib/Transforms/Utils/Local.cpp b/lib/Transforms/Utils/Local.cpp
index be0aa82..7034feb 100644
--- a/lib/Transforms/Utils/Local.cpp
+++ b/lib/Transforms/Utils/Local.cpp
@@ -721,10 +721,14 @@
/// their preferred alignment from the beginning.
///
static unsigned enforceKnownAlignment(Value *V, unsigned Align,
- unsigned PrefAlign) {
+ unsigned PrefAlign, const TargetData *TD) {
V = V->stripPointerCasts();
if (AllocaInst *AI = dyn_cast<AllocaInst>(V)) {
+ // If the preferred alignment is greater than the natural stack alignment
+ // then don't round up. This avoids dynamic stack realignment.
+ if (TD && TD->exceedsNaturalStackAlignment(PrefAlign))
+ return Align;
// If there is a requested alignment and if this is an alloca, round up.
if (AI->getAlignment() >= PrefAlign)
return AI->getAlignment();
@@ -775,7 +779,7 @@
Align = std::min(Align, +Value::MaximumAlignment);
if (PrefAlign > Align)
- Align = enforceKnownAlignment(V, Align, PrefAlign);
+ Align = enforceKnownAlignment(V, Align, PrefAlign, TD);
// We don't need to make any adjustment.
return Align;