Cherrypick upstream volatile _Complex alignment patches

Specifically:
  r186564 - Fix volatile _Complex alignment test on platforms where 64-bit floating point isn't 64-bit aligned
  r186490 - Propagate alignment for _Complex

These should fix GCC torture test failures, as well as the all-important uses of volatile _Complex numbers in C, and their alignment being incorrect.

BUG= PNaCl FYI bots red on torture tests
TEST= ./tools/toolchain_tester/torture_test.py pnacl x86-64 --concurrency=32 >& torture-x86-64.log ; ./tools/toolchain_tester/torture_test.py pnacl x86-32 --concurrency=32 >& torture-x86-32.log ; ./tools/toolchain_tester/torture_test.py pnacl arm --concurrency=32 >& torture-arm.log
R=stichnot@chromium.org

Review URL: https://codereview.chromium.org/19915003

4 files changed, 120 insertions, 34 deletions

diff --git a/lib/CodeGen/CGExprComplex.cpp b/lib/CodeGen/CGExprComplex.cpp
index 36f974a313..3a3971479d 100644
--- a/lib/CodeGen/CGExprComplex.cpp
+++ b/lib/CodeGen/CGExprComplex.cpp
@@ -18,6 +18,7 @@
 #include "llvm/ADT/SmallString.h"
 #include "llvm/IR/Constants.h"
 #include "llvm/IR/Function.h"
+#include <algorithm>
 using namespace clang;
 using namespace CodeGen;
 
@@ -294,19 +295,26 @@ ComplexPairTy ComplexExprEmitter::EmitLoadOfLValue(LValue lvalue) {
 
   llvm::Value *SrcPtr = lvalue.getAddress();
   bool isVolatile = lvalue.isVolatileQualified();
+  unsigned AlignR = lvalue.getAlignment().getQuantity();
+  ASTContext &C = CGF.getContext();
+  QualType ComplexTy = lvalue.getType();
+  unsigned ComplexAlign = C.getTypeAlignInChars(ComplexTy).getQuantity();
+  unsigned AlignI = std::min(AlignR, ComplexAlign);
 
   llvm::Value *Real=0, *Imag=0;
 
   if (!IgnoreReal || isVolatile) {
     llvm::Value *RealP = Builder.CreateStructGEP(SrcPtr, 0,
                                                  SrcPtr->getName() + ".realp");
-    Real = Builder.CreateLoad(RealP, isVolatile, SrcPtr->getName() + ".real");
+    Real = Builder.CreateAlignedLoad(RealP, AlignR, isVolatile,
+                                     SrcPtr->getName() + ".real");
   }
 
   if (!IgnoreImag || isVolatile) {
     llvm::Value *ImagP = Builder.CreateStructGEP(SrcPtr, 1,
                                                  SrcPtr->getName() + ".imagp");
-    Imag = Builder.CreateLoad(ImagP, isVolatile, SrcPtr->getName() + ".imag");
+    Imag = Builder.CreateAlignedLoad(ImagP, AlignI, isVolatile,
+                                     SrcPtr->getName() + ".imag");
   }
   return ComplexPairTy(Real, Imag);
 }
@@ -322,10 +330,16 @@ void ComplexExprEmitter::EmitStoreOfComplex(ComplexPairTy Val,
   llvm::Value *Ptr = lvalue.getAddress();
   llvm::Value *RealPtr = Builder.CreateStructGEP(Ptr, 0, "real");
   llvm::Value *ImagPtr = Builder.CreateStructGEP(Ptr, 1, "imag");
-
-  // TODO: alignment
-  Builder.CreateStore(Val.first, RealPtr, lvalue.isVolatileQualified());
-  Builder.CreateStore(Val.second, ImagPtr, lvalue.isVolatileQualified());
+  unsigned AlignR = lvalue.getAlignment().getQuantity();
+  ASTContext &C = CGF.getContext();
+  QualType ComplexTy = lvalue.getType();
+  unsigned ComplexAlign = C.getTypeAlignInChars(ComplexTy).getQuantity();
+  unsigned AlignI = std::min(AlignR, ComplexAlign);
+
+  Builder.CreateAlignedStore(Val.first, RealPtr, AlignR,
+                             lvalue.isVolatileQualified());
+  Builder.CreateAlignedStore(Val.second, ImagPtr, AlignI,
+                             lvalue.isVolatileQualified());
 }
 
 
diff --git a/test/CodeGen/volatile-1.c b/test/CodeGen/volatile-1.c
index 65511593d3..09c75ee83b 100644
--- a/test/CodeGen/volatile-1.c
+++ b/test/CodeGen/volatile-1.c
@@ -26,45 +26,45 @@ int printf(const char *, ...);
 void test() {
   // CHECK: load volatile [[INT]]* @i
   i;
-  // CHECK-NEXT: load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 0)
-  // CHECK-NEXT: load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 1)
+  // CHECK-NEXT: load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 0), align 4
+  // CHECK-NEXT: load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 1), align 4
   // CHECK-NEXT: sitofp [[INT]]
   (float)(ci);
-  // CHECK-NEXT: load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 0)
-  // CHECK-NEXT: load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 1)
+  // CHECK-NEXT: load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 0), align 4
+  // CHECK-NEXT: load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 1), align 4
   (void)ci;
   // CHECK-NEXT: bitcast
   // CHECK-NEXT: memcpy
   (void)a;
-  // CHECK-NEXT: [[R:%.*]] = load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 0)
-  // CHECK-NEXT: [[I:%.*]] = load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 1)
-  // CHECK-NEXT: store volatile [[INT]] [[R]], [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 0)
-  // CHECK-NEXT: store volatile [[INT]] [[I]], [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 1)
+  // CHECK-NEXT: [[R:%.*]] = load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 0), align 4
+  // CHECK-NEXT: [[I:%.*]] = load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 1), align 4
+  // CHECK-NEXT: store volatile [[INT]] [[R]], [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 0), align 4
+  // CHECK-NEXT: store volatile [[INT]] [[I]], [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 1), align 4
   (void)(ci=ci);
   // CHECK-NEXT: [[T:%.*]] = load volatile [[INT]]* @j
   // CHECK-NEXT: store volatile [[INT]] [[T]], [[INT]]* @i
   (void)(i=j);
-  // CHECK-NEXT: [[R1:%.*]] = load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 0)
-  // CHECK-NEXT: [[I1:%.*]] = load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 1)
-  // CHECK-NEXT: [[R2:%.*]] = load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 0)
-  // CHECK-NEXT: [[I2:%.*]] = load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 1)
+  // CHECK-NEXT: [[R1:%.*]] = load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 0), align 4
+  // CHECK-NEXT: [[I1:%.*]] = load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 1), align 4
+  // CHECK-NEXT: [[R2:%.*]] = load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 0), align 4
+  // CHECK-NEXT: [[I2:%.*]] = load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 1), align 4
   // Not sure why they're ordered this way.
   // CHECK-NEXT: [[R:%.*]] = add [[INT]] [[R2]], [[R1]]
   // CHECK-NEXT: [[I:%.*]] = add [[INT]] [[I2]], [[I1]]
-  // CHECK-NEXT: store volatile [[INT]] [[R]], [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 0)
-  // CHECK-NEXT: store volatile [[INT]] [[I]], [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 1)
+  // CHECK-NEXT: store volatile [[INT]] [[R]], [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 0), align 4
+  // CHECK-NEXT: store volatile [[INT]] [[I]], [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 1), align 4
   ci+=ci;
 
-  // CHECK-NEXT: [[R1:%.*]] = load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 0)
-  // CHECK-NEXT: [[I1:%.*]] = load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 1)
-  // CHECK-NEXT: [[R2:%.*]] = load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 0)
-  // CHECK-NEXT: [[I2:%.*]] = load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 1)
+  // CHECK-NEXT: [[R1:%.*]] = load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 0), align 4
+  // CHECK-NEXT: [[I1:%.*]] = load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 1), align 4
+  // CHECK-NEXT: [[R2:%.*]] = load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 0), align 4
+  // CHECK-NEXT: [[I2:%.*]] = load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 1), align 4
   // CHECK-NEXT: [[R:%.*]] = add [[INT]] [[R2]], [[R1]]
   // CHECK-NEXT: [[I:%.*]] = add [[INT]] [[I2]], [[I1]]
-  // CHECK-NEXT: store volatile [[INT]] [[R]], [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 0)
-  // CHECK-NEXT: store volatile [[INT]] [[I]], [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 1)
-  // CHECK-NEXT: [[R2:%.*]] = load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 0)
-  // CHECK-NEXT: [[I2:%.*]] = load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 1)
+  // CHECK-NEXT: store volatile [[INT]] [[R]], [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 0), align 4
+  // CHECK-NEXT: store volatile [[INT]] [[I]], [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 1), align 4
+  // CHECK-NEXT: [[R2:%.*]] = load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 0), align 4
+  // CHECK-NEXT: [[I2:%.*]] = load volatile [[INT]]* getelementptr inbounds ([[CINT]]* @ci, i32 0, i32 1), align 4
   // These additions can be elided
   // CHECK-NEXT: add [[INT]] [[R]], [[R2]]
   // CHECK-NEXT: add [[INT]] [[I]], [[I2]]
diff --git a/test/CodeGen/volatile-2.c b/test/CodeGen/volatile-2.c
index 3d342de690..9233b2a167 100644
--- a/test/CodeGen/volatile-2.c
+++ b/test/CodeGen/volatile-2.c
@@ -3,8 +3,8 @@
 void test0() {
   // CHECK: define void @test0()
   // CHECK:      [[F:%.*]] = alloca float
-  // CHECK-NEXT: [[REAL:%.*]] = load volatile float* getelementptr inbounds ({ float, float }* @test0_v, i32 0, i32 0)
-  // CHECK-NEXT: load volatile float* getelementptr inbounds ({{.*}} @test0_v, i32 0, i32 1)
+  // CHECK-NEXT: [[REAL:%.*]] = load volatile float* getelementptr inbounds ({ float, float }* @test0_v, i32 0, i32 0), align 4
+  // CHECK-NEXT: load volatile float* getelementptr inbounds ({{.*}} @test0_v, i32 0, i32 1), align 4
   // CHECK-NEXT: store float [[REAL]], float* [[F]], align 4
   // CHECK-NEXT: ret void
   extern volatile _Complex float test0_v;
@@ -13,10 +13,10 @@ void test0() {
 
 void test1() {
   // CHECK: define void @test1()
-  // CHECK:      [[REAL:%.*]] = load volatile float* getelementptr inbounds ({{.*}} @test1_v, i32 0, i32 0)
-  // CHECK-NEXT: [[IMAG:%.*]] = load volatile float* getelementptr inbounds ({{.*}} @test1_v, i32 0, i32 1)
-  // CHECK-NEXT: store volatile float [[REAL]], float* getelementptr inbounds ({{.*}} @test1_v, i32 0, i32 0)
-  // CHECK-NEXT: store volatile float [[IMAG]], float* getelementptr inbounds ({{.*}} @test1_v, i32 0, i32 1)
+  // CHECK:      [[REAL:%.*]] = load volatile float* getelementptr inbounds ({{.*}} @test1_v, i32 0, i32 0), align 4
+  // CHECK-NEXT: [[IMAG:%.*]] = load volatile float* getelementptr inbounds ({{.*}} @test1_v, i32 0, i32 1), align 4
+  // CHECK-NEXT: store volatile float [[REAL]], float* getelementptr inbounds ({{.*}} @test1_v, i32 0, i32 0), align 4
+  // CHECK-NEXT: store volatile float [[IMAG]], float* getelementptr inbounds ({{.*}} @test1_v, i32 0, i32 1), align 4
   // CHECK-NEXT: ret void
   extern volatile _Complex float test1_v;
   test1_v = test1_v;
diff --git a/test/CodeGen/volatile-complex.c b/test/CodeGen/volatile-complex.c
new file mode 100644
index 0000000000..92ffd9de45
--- /dev/null
+++ b/test/CodeGen/volatile-complex.c
@@ -0,0 +1,72 @@
+// RUN: %clang_cc1 -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck %s
+
+// Validate that volatile _Complex loads and stores are generated
+// properly, including their alignment (even when overaligned).
+//
+// This test assumes that floats are 32-bit aligned and doubles are
+// 64-bit aligned, and uses x86-64 as a target that should have this
+// datalayout.
+
+// CHECK: target datalayout = "{{.*}}f32:32:32-f64:64:64{{.*}}"
+
+volatile _Complex float cf;
+volatile _Complex double cd;
+volatile _Complex float cf32 __attribute__((aligned(32)));
+volatile _Complex double cd32 __attribute__((aligned(32)));
+
+
+// CHECK: define void @test_cf()
+// CHECK-NEXT: entry:
+void test_cf() {
+  // CHECK-NEXT: load volatile float* getelementptr inbounds ({ float, float }* @cf, i32 0, i32 0), align 4
+  // CHECK-NEXT: load volatile float* getelementptr inbounds ({ float, float }* @cf, i32 0, i32 1), align 4
+  (void)(cf);
+  // CHECK-NEXT: [[R:%.*]] = load volatile float* getelementptr inbounds ({ float, float }* @cf, i32 0, i32 0), align 4
+  // CHECK-NEXT: [[I:%.*]] = load volatile float* getelementptr inbounds ({ float, float }* @cf, i32 0, i32 1), align 4
+  // CHECK-NEXT: store volatile float [[R]], float* getelementptr inbounds ({ float, float }* @cf, i32 0, i32 0), align 4
+  // CHECK-NEXT: store volatile float [[I]], float* getelementptr inbounds ({ float, float }* @cf, i32 0, i32 1), align 4
+  (void)(cf=cf);
+  // CHECK-NEXT: ret void
+}
+
+// CHECK: define void @test_cd()
+// CHECK-NEXT: entry:
+void test_cd() {
+  // CHECK-NEXT: load volatile double* getelementptr inbounds ({ double, double }* @cd, i32 0, i32 0), align 8
+  // CHECK-NEXT: load volatile double* getelementptr inbounds ({ double, double }* @cd, i32 0, i32 1), align 8
+  (void)(cd);
+  // CHECK-NEXT: [[R:%.*]] = load volatile double* getelementptr inbounds ({ double, double }* @cd, i32 0, i32 0), align 8
+  // CHECK-NEXT: [[I:%.*]] = load volatile double* getelementptr inbounds ({ double, double }* @cd, i32 0, i32 1), align 8
+  // CHECK-NEXT: store volatile double [[R]], double* getelementptr inbounds ({ double, double }* @cd, i32 0, i32 0), align 8
+  // CHECK-NEXT: store volatile double [[I]], double* getelementptr inbounds ({ double, double }* @cd, i32 0, i32 1), align 8
+  (void)(cd=cd);
+  // CHECK-NEXT: ret void
+}
+
+// CHECK: define void @test_cf32()
+// CHECK-NEXT: entry:
+void test_cf32() {
+  // CHECK-NEXT: load volatile float* getelementptr inbounds ({ float, float }* @cf32, i32 0, i32 0), align 32
+  // CHECK-NEXT: load volatile float* getelementptr inbounds ({ float, float }* @cf32, i32 0, i32 1), align 4
+  (void)(cf32);
+  // CHECK-NEXT: [[R:%.*]] = load volatile float* getelementptr inbounds ({ float, float }* @cf32, i32 0, i32 0), align 32
+  // CHECK-NEXT: [[I:%.*]] = load volatile float* getelementptr inbounds ({ float, float }* @cf32, i32 0, i32 1), align 4
+  // CHECK-NEXT: store volatile float [[R]], float* getelementptr inbounds ({ float, float }* @cf32, i32 0, i32 0), align 32
+  // CHECK-NEXT: store volatile float [[I]], float* getelementptr inbounds ({ float, float }* @cf32, i32 0, i32 1), align 4
+  (void)(cf32=cf32);
+  // CHECK-NEXT: ret void
+}
+
+// CHECK: define void @test_cd32()
+// CHECK-NEXT: entry:
+void test_cd32() {
+  // CHECK-NEXT: load volatile double* getelementptr inbounds ({ double, double }* @cd32, i32 0, i32 0), align 32
+  // CHECK-NEXT: load volatile double* getelementptr inbounds ({ double, double }* @cd32, i32 0, i32 1), align 8
+  (void)(cd32);
+  // CHECK-NEXT: [[R:%.*]] = load volatile double* getelementptr inbounds ({ double, double }* @cd32, i32 0, i32 0), align 32
+  // CHECK-NEXT: [[I:%.*]] = load volatile double* getelementptr inbounds ({ double, double }* @cd32, i32 0, i32 1), align 8
+  // CHECK-NEXT: store volatile double [[R]], double* getelementptr inbounds ({ double, double }* @cd32, i32 0, i32 0), align 32
+  // CHECK-NEXT: store volatile double [[I]], double* getelementptr inbounds ({ double, double }* @cd32, i32 0, i32 1), align 8
+  (void)(cd32=cd32);
+  // CHECK-NEXT: ret void
+}