Always_inline codegen rewrite.

Current implementation may end up emitting an undefined reference for
an "inline __attribute__((always_inline))" function by generating an
"available_externally alwaysinline" IR function for it and then failing to
inline all the calls. This happens when a call to such function is in dead
code. As the inliner is an SCC pass, it does not process dead code.

Libc++ relies on the compiler never emitting such undefined reference.

With this patch, we emit a pair of
1. internal alwaysinline definition (called F.alwaysinline)
2a. A stub F() { musttail call F.alwaysinline }
  -- or, depending on the linkage --
2b. A declaration of F.

The frontend ensures that F.inlinefunction is only used for direct
calls, and the stub is used for everything else (taking the address of
the function, really). Declaration (2b) is emitted in the case when
"inline" is meant for inlining only (like __gnu_inline__ and some
other cases).

This approach, among other nice properties, ensures that alwaysinline
functions are always internal, making it impossible for a direct call
to such function to produce an undefined symbol reference.

This patch is based on ideas by Chandler Carruth and Richard Smith.

llvm-svn: 247494

10 files changed, 238 insertions, 17 deletions

diff --git a/clang/test/CodeGen/always-inline.c b/clang/test/CodeGen/always-inline.c
index c9fd1ae2d80..ed8ecc7bfa7 100644
--- a/clang/test/CodeGen/always-inline.c
+++ b/clang/test/CodeGen/always-inline.c
@@ -1,7 +1,9 @@
 // RUN: %clang_cc1 -emit-llvm %s -o - | FileCheck %s
 // RUN: %clang_cc1 -fno-inline -emit-llvm %s -o - | FileCheck %s
 
+// CHECK-LABEL: define void @i_want_bar()
 // CHECK-NOT: foo
+// CHECK: ret void
 
 void bar() {
 }
diff --git a/clang/test/CodeGen/always_inline-unused.c b/clang/test/CodeGen/always_inline-unused.c
new file mode 100644
index 00000000000..eb733f72c7e
--- /dev/null
+++ b/clang/test/CodeGen/always_inline-unused.c
@@ -0,0 +1,31 @@
+// Test alwaysinline definitions w/o any non-direct-call uses.
+// None of the declarations are emitted. Stub are only emitted when the original
+// function can not be discarded.
+
+// RUN: %clang_cc1 -disable-llvm-optzns -emit-llvm %s -o - | FileCheck %s
+
+void __attribute__((__always_inline__)) f1() {}
+inline void __attribute__((__always_inline__)) f2() {}
+static inline void __attribute__((__always_inline__)) f3() {}
+inline void __attribute__((gnu_inline, __always_inline__)) f4() {}
+static inline void __attribute__((gnu_inline, __always_inline__)) f5() {}
+inline void __attribute__((visibility("hidden"), __always_inline__)) f6() {}
+inline void __attribute__((visibility("hidden"), gnu_inline, __always_inline__)) f7() {}
+
+void g() {
+  f1();
+  f2();
+  f3();
+  f4();
+  f5();
+  f6();
+  f7();
+}
+
+// CHECK: define void @f1()
+// CHECK-NOT: void @f2()
+// CHECK-NOT: void @f3()
+// CHECK: define void @f4()
+// CHECK-NOT: void @f5()
+// CHECK-NOT: void @f6()
+// CHECK: define hidden void @f7()
diff --git a/clang/test/CodeGen/always_inline-wrappers.c b/clang/test/CodeGen/always_inline-wrappers.c
new file mode 100644
index 00000000000..ba0b148a7c5
--- /dev/null
+++ b/clang/test/CodeGen/always_inline-wrappers.c
@@ -0,0 +1,108 @@
+// Test different kinds of alwaysinline definitions.
+
+// RUN: %clang_cc1 -disable-llvm-optzns -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-INLINE
+// RUN: %clang_cc1 -disable-llvm-optzns -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK-USE
+// RUN: %clang_cc1 -disable-llvm-optzns -fno-inline -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK
+// RUN: %clang_cc1 -disable-llvm-optzns -fno-inline -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK-USE
+
+void __attribute__((__always_inline__)) f1() {}
+inline void __attribute__((__always_inline__)) f2() {}
+static inline void __attribute__((__always_inline__)) f3() {}
+inline void __attribute__((gnu_inline, __always_inline__)) f4() {}
+static inline void __attribute__((gnu_inline, __always_inline__)) f5() {}
+inline void __attribute__((visibility("hidden"), __always_inline__)) f6() {}
+inline void __attribute__((visibility("hidden"), gnu_inline, __always_inline__)) f7() {}
+
+void g() {
+  f1();
+  f2();
+  f3();
+  f4();
+  f5();
+  f6();
+  f7();
+}
+
+void (*p)(void);
+void h() {
+  p = f1;
+  p = f2;
+  p = f3;
+  p = f4;
+  p = f5;
+  p = f6;
+  p = f7;
+}
+
+void (*const cp1)(void) = f1;
+void (*p1)(void) = f1;
+void (*p2)(int) = (void (*)(int))f1;
+
+void __attribute__((__always_inline__)) f8(void(*f)(void)) {}
+
+void call() {
+  f8(f1);
+}
+
+// CHECK-DAG: define internal void @f1.alwaysinline() #[[AI:[0-9]+]]
+// CHECK-DAG: define internal void @f2.alwaysinline() #[[AI_IH:[0-9]+]]
+// CHECK-DAG: define internal void @f3.alwaysinline() #[[AI_IH]]
+// CHECK-DAG: define internal void @f4.alwaysinline() #[[AI_IH]]
+// CHECK-DAG: define internal void @f5.alwaysinline() #[[AI_IH]]
+// CHECK-DAG: define internal void @f6.alwaysinline() #[[AI_IH]]
+// CHECK-DAG: define internal void @f7.alwaysinline() #[[AI_IH]]
+
+
+// CHECK-DAG: define void @f1() #[[NOAI:[01-9]+]]
+// CHECK-DAG: musttail call void @f1.alwaysinline()
+
+// CHECK-DAG: declare void @f2() #[[NOAI:[01-9]+]]
+
+// CHECK-DAG: define internal void @f3() #[[NOAI:[01-9]+]]
+// CHECK-DAG: musttail call void @f3.alwaysinline()
+
+// CHECK-DAG: define void @f4() #[[NOAI:[01-9]+]]
+// CHECK-DAG: musttail call void @f4.alwaysinline()
+
+// CHECK-DAG: define internal void @f5() #[[NOAI:[01-9]+]]
+// CHECK-DAG: musttail call void @f5.alwaysinline()
+
+// CHECK-DAG: declare hidden void @f6() #[[NOAI:[01-9]+]]
+
+// CHECK-DAG: define hidden void @f7() #[[NOAI:[01-9]+]]
+// CHECK-DAG: musttail call void @f7.alwaysinline()
+
+
+// CHECK-DAG: @cp1 = constant void ()* @f1, align
+// CHECK-DAG: @p1 = global void ()* @f1, align
+// CHECK-DAG: @p2 = global void (i32)* bitcast (void ()* @f1 to void (i32)*), align
+
+// CHECK: attributes #[[AI]] = {{.*alwaysinline.*}}
+// CHECK-INLINE: attributes #[[AI_IH]] = {{.*alwaysinline.*inlinehint.*}}
+// CHECK-NOT: attributes #[[NOAI]] = {{.*alwaysinline.*}}
+
+// CHECK-USE-LABEL: define void @g()
+// CHECK-USE-NOT:      ret void
+// CHECK-USE:          call void @f1.alwaysinline()
+// CHECK-USE-NEXT:     call void @f2.alwaysinline()
+// CHECK-USE-NEXT:     call void @f3.alwaysinline()
+// CHECK-USE-NEXT:     call void @f4.alwaysinline()
+// CHECK-USE-NEXT:     call void @f5.alwaysinline()
+// CHECK-USE-NEXT:     call void @f6.alwaysinline()
+// CHECK-USE-NEXT:     call void @f7.alwaysinline()
+// CHECK-USE-NEXT:     ret void
+
+// CHECK-USE-LABEL: define void @h()
+// CHECK-USE-NOT:      ret void
+// CHECK-USE:          store void ()* @f1,
+// CHECK-USE-NEXT:     store void ()* @f2,
+// CHECK-USE-NEXT:     store void ()* @f3,
+// CHECK-USE-NEXT:     store void ()* @f4,
+// CHECK-USE-NEXT:     store void ()* @f5,
+// CHECK-USE-NEXT:     store void ()* @f6,
+// CHECK-USE-NEXT:     store void ()* @f7,
+// CHECK-USE-NEXT:     ret void
+
+// CHECK-USE-LABEL:  define void @call()
+// CHECK-USE:           call void @f8.alwaysinline(void ()* @f1)
+// CHECK-USE:           ret void
diff --git a/clang/test/CodeGen/always_inline.c b/clang/test/CodeGen/always_inline.c
index c91fd43f276..00b4234dc71 100644
--- a/clang/test/CodeGen/always_inline.c
+++ b/clang/test/CodeGen/always_inline.c
@@ -1,8 +1,5 @@
-// RUN: %clang -emit-llvm -S -o %t %s
-// RUN: not grep '@f0' %t
-// RUN: not grep 'call ' %t
-// RUN: %clang -mllvm -disable-llvm-optzns -emit-llvm -S -o %t %s
-// RUN: grep '@f0' %t | count 2
+// RUN: %clang -target x86_64-pc-linux-gnu -emit-llvm -S -o - %s | FileCheck %s
+// RUN: %clang -target x86_64-pc-linux-gnu -mllvm -disable-llvm-optzns -emit-llvm -S -o - %s | FileCheck %s --check-prefix=CHECK-NO-OPTZNS
 
 //static int f0() { 
 static int __attribute__((always_inline)) f0() { 
@@ -18,3 +15,14 @@ inline int f2() __attribute__((always_inline));
 int f2() { return 7; }
 int f3(void) { return f2(); }
 
+// CHECK-LABEL: define i32 @f1()
+// CHECK: ret i32 1
+// CHECK-LABEL: define i32 @f2()
+// CHECK: ret i32 7
+// CHECK-LABEL: define i32 @f3()
+// CHECK: ret i32 7
+
+// CHECK-NO-OPTZNS: define i32 @f3()
+// CHECK-NO-OPTZNS-NOT: ret i32
+// CHECK-NO-OPTZNS:   call i32 @f2.alwaysinline()
+// CHECK-NO-OPTZNS-NEXT:   ret i32
diff --git a/clang/test/CodeGen/function-attributes.c b/clang/test/CodeGen/function-attributes.c
index 177ad848f74..4a7d4fc0edb 100644
--- a/clang/test/CodeGen/function-attributes.c
+++ b/clang/test/CodeGen/function-attributes.c
@@ -25,8 +25,8 @@ void f6(signed short x) { }
 
 void f7(unsigned short x) { }
 
-// CHECK-LABEL: define void @f8()
-// CHECK: [[AI:#[0-9]+]]
+// CHECK: define void @f8()
+// CHECK: [[NUW:#[0-9]+]]
 // CHECK: {
 void __attribute__((always_inline)) f8(void) { }
 
@@ -129,7 +129,6 @@ void f20(void) {
 }
 
 // CHECK: attributes [[NUW]] = { nounwind optsize readnone{{.*}} }
-// CHECK: attributes [[AI]] = { alwaysinline nounwind optsize readnone{{.*}} }
 // CHECK: attributes [[ALIGN]] = { nounwind optsize readnone alignstack=16{{.*}} }
 // CHECK: attributes [[RT]] = { nounwind optsize returns_twice{{.*}} }
 // CHECK: attributes [[NR]] = { noreturn nounwind optsize }
diff --git a/clang/test/CodeGen/pr9614.c b/clang/test/CodeGen/pr9614.c
index 63cb5af1868..d264ef3a9ee 100644
--- a/clang/test/CodeGen/pr9614.c
+++ b/clang/test/CodeGen/pr9614.c
@@ -1,4 +1,5 @@
-// RUN: %clang_cc1 -triple x86_64-pc-linux -emit-llvm %s -o - | FileCheck %s
+// RUN: %clang_cc1 -triple x86_64-pc-linux -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK0
+// RUN: %clang_cc1 -triple x86_64-pc-linux -O1 -disable-llvm-optzns -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK1
 
 extern void foo_alias (void) __asm ("foo");
 inline void foo (void) {
@@ -24,7 +25,7 @@ extern inline __attribute__((__always_inline__, __gnu_inline__)) void *memchr(vo
 
 void f(void) {
   foo();
-  abs(0);
+  int x = abs(0);
   strrchr_foo("", '.');
   prefetch();
   memchr("", '.', 0);
@@ -32,9 +33,10 @@ void f(void) {
 
 // CHECK-LABEL: define void @f()
 // CHECK: call void @foo()
-// CHECK: call i32 @abs(i32 0)
+// CHECK: call i32 @abs(
 // CHECK: call i8* @strrchr(
-// CHECK: call void @llvm.prefetch(
+// CHECK0: call void @llvm.prefetch(
+// CHECK1: call void @prefetch.alwaysinline(
 // CHECK: call i8* @memchr(
 // CHECK: ret void
 
diff --git a/clang/test/CodeGenCXX/alwaysinline.cpp b/clang/test/CodeGenCXX/alwaysinline.cpp
new file mode 100644
index 00000000000..c694ae41d22
--- /dev/null
+++ b/clang/test/CodeGenCXX/alwaysinline.cpp
@@ -0,0 +1,68 @@
+// Test different kinds of alwaysinline *structor definitions.
+
+// RUN: %clang_cc1 -triple x86_64-pc-linux-gnu -disable-llvm-optzns -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK
+// RUN: %clang_cc1 -triple x86_64-pc-linux-gnu -disable-llvm-optzns -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK-CALL
+
+// RUN: %clang_cc1 -triple x86_64-pc-linux-gnu -mconstructor-aliases -disable-llvm-optzns -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK
+// RUN: %clang_cc1 -triple x86_64-pc-linux-gnu -mconstructor-aliases -disable-llvm-optzns -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK-CALL
+
+struct A1 {
+  __attribute__((__always_inline__)) A1() {}
+  __attribute__((__always_inline__)) ~A1() {}
+};
+
+void g1() {
+  A1 a1;
+}
+
+struct A2 {
+  inline __attribute__((__always_inline__)) A2() {}
+  inline __attribute__((__always_inline__)) ~A2() {}
+};
+
+void g2() {
+  A2 a2;
+}
+
+struct A3 {
+  inline __attribute__((gnu_inline, __always_inline__)) A3() {}
+  inline __attribute__((gnu_inline, __always_inline__)) ~A3() {}
+};
+
+void g3() {
+  A3 a3;
+}
+
+// CHECK-DAG: define internal void @_ZN2A1C1Ev.alwaysinline(%struct.A1* %this) unnamed_addr #[[AI:[01-9]+]]
+// CHECK-DAG: define internal void @_ZN2A1C2Ev.alwaysinline(%struct.A1* %this) unnamed_addr #[[AI]]
+// CHECK-DAG: define internal void @_ZN2A1D1Ev.alwaysinline(%struct.A1* %this) unnamed_addr #[[AI]]
+// CHECK-DAG: define internal void @_ZN2A1D2Ev.alwaysinline(%struct.A1* %this) unnamed_addr #[[AI]]
+
+// CHECK-DAG: define internal void @_ZN2A2C1Ev.alwaysinline(%struct.A2* %this) unnamed_addr #[[AIIH:[01-9]+]]
+// CHECK-DAG: define internal void @_ZN2A2C2Ev.alwaysinline(%struct.A2* %this) unnamed_addr #[[AIIH]]
+// CHECK-DAG: define internal void @_ZN2A2D1Ev.alwaysinline(%struct.A2* %this) unnamed_addr #[[AIIH]]
+// CHECK-DAG: define internal void @_ZN2A2D2Ev.alwaysinline(%struct.A2* %this) unnamed_addr #[[AIIH]]
+
+// CHECK-DAG: define internal void @_ZN2A3C1Ev.alwaysinline(%struct.A3* %this) unnamed_addr #[[AIIH]]
+// CHECK-DAG: define internal void @_ZN2A3C2Ev.alwaysinline(%struct.A3* %this) unnamed_addr #[[AIIH]]
+// CHECK-DAG: define internal void @_ZN2A3D1Ev.alwaysinline(%struct.A3* %this) unnamed_addr #[[AIIH]]
+// CHECK-DAG: define internal void @_ZN2A3D2Ev.alwaysinline(%struct.A3* %this) unnamed_addr #[[AIIH]]
+
+// CHECK-DAG: attributes #[[AI]] = {{.*alwaysinline.*}}
+// CHECK-DAG: attributes #[[AIIH]] = {{.*alwaysinline.*inlinehint.*}}
+// CHECK-NOT: attributes #[[NOAI]] = {{.*alwaysinline.*}}
+
+// CHECK-CALL-LABEL: define void @_Z2g1v()
+// CHECK-CALL:       call void @_ZN2A1C1Ev.alwaysinline
+// CHECK-CALL:       call void @_ZN2A1D1Ev.alwaysinline
+// CHECK-CALL:       ret void
+
+// CHECK-CALL-LABEL: define void @_Z2g2v()
+// CHECK-CALL:       call void @_ZN2A2C1Ev.alwaysinline
+// CHECK-CALL:       call void @_ZN2A2D1Ev.alwaysinline
+// CHECK-CALL:       ret void
+
+// CHECK-CALL-LABEL: define void @_Z2g3v()
+// CHECK-CALL:       call void @_ZN2A3C1Ev.alwaysinline
+// CHECK-CALL:       call void @_ZN2A3D1Ev.alwaysinline
+// CHECK-CALL:       ret void
diff --git a/clang/test/Frontend/optimization-remark-line-directive.c b/clang/test/Frontend/optimization-remark-line-directive.c
index f4c0011fb4f..23d63c9aa1e 100644
--- a/clang/test/Frontend/optimization-remark-line-directive.c
+++ b/clang/test/Frontend/optimization-remark-line-directive.c
@@ -5,8 +5,9 @@
 // RUN: %clang_cc1 %s -Rpass=inline -gline-tables-only -dwarf-column-info -emit-llvm-only -verify
 
 int foo(int x, int y) __attribute__((always_inline));
+// expected-remark@+1 {{foo.alwaysinline inlined into foo}}
 int foo(int x, int y) { return x + y; }
 
-// expected-remark@+2 {{foo inlined into bar}} expected-note@+2 {{could not determine the original source location for /bad/path/to/original.c:1230:25}}
+// expected-remark@+2 {{foo.alwaysinline inlined into bar}} expected-note@+2 {{could not determine the original source location for /bad/path/to/original.c:1230:25}}
 #line 1230 "/bad/path/to/original.c"
 int bar(int j) { return foo(j, j - 2); }
diff --git a/clang/test/Frontend/optimization-remark.c b/clang/test/Frontend/optimization-remark.c
index 6ada0030a70..604fec00204 100644
--- a/clang/test/Frontend/optimization-remark.c
+++ b/clang/test/Frontend/optimization-remark.c
@@ -32,6 +32,8 @@
 // CHECK-NOT: !llvm.dbg.cu = !{
 
 int foo(int x, int y) __attribute__((always_inline));
+// expected-remark@+2 {{foo.alwaysinline should always be inlined}}
+// expected-remark@+1 {{foo.alwaysinline inlined into foo}}
 int foo(int x, int y) { return x + y; }
 
 float foz(int x, int y) __attribute__((noinline));
@@ -45,7 +47,7 @@ int bar(int j) {
 // expected-remark@+5 {{foz will not be inlined into bar}}
 // expected-remark@+4 {{foz should never be inlined}}
 // expected-remark@+3 {{foz will not be inlined into bar}}
-// expected-remark@+2 {{foo should always be inlined}}
-// expected-remark@+1 {{foo inlined into bar}}
+// expected-remark@+2 {{foo.alwaysinline should always be inlined}}
+// expected-remark@+1 {{foo.alwaysinline inlined into bar}}
   return foo(j, j - 2) * foz(j - 2, j);
 }
diff --git a/clang/test/Modules/cxx-irgen.cpp b/clang/test/Modules/cxx-irgen.cpp
index 7cdb0d6eb59..73d17467892 100644
--- a/clang/test/Modules/cxx-irgen.cpp
+++ b/clang/test/Modules/cxx-irgen.cpp
@@ -26,14 +26,14 @@ namespace EmitInlineMethods {
   };
 }
 
-// CHECK-DAG: define available_externally hidden {{.*}}{{signext i32|i32}} @_ZN1SIiE1gEv({{.*}} #[[ALWAYS_INLINE:.*]] align
+// CHECK-DAG: define internal i32 @_ZN1SIiE1gEv.alwaysinline() #[[ALWAYS_INLINE:.*]] align
 int a = S<int>::g();
 
 int b = h();
 
 // CHECK-DAG: define linkonce_odr {{.*}}{{signext i32|i32}} @_Z3minIiET_S0_S0_(i32
 int c = min(1, 2);
-// CHECK: define available_externally {{.*}}{{signext i32|i32}} @_ZN1SIiE1fEv({{.*}} #[[ALWAYS_INLINE]] align
+// CHECK-DAG: define internal {{.*}}{{signext i32|i32}} @_ZN1SIiE1fEv.alwaysinline() #[[ALWAYS_INLINE]] align
 
 namespace ImplicitSpecialMembers {
   // CHECK-LABEL: define {{.*}} @_ZN22ImplicitSpecialMembers1BC2ERKS0_(