Unify the atomic implementation for x86 architectures

It's almost exactly the same code in both files, so let's have one file
only. That means we need an #ifdef for the special case of 64-bit types
on i386.

Also take the opportunity to add a comment explaining how this works.

Change-Id: I50d274fa026806ae511b1045aa8a5c25daaa0edc
Reviewed-by: Bradley T. Hughes <bradley.hughes@nokia.com>

3 files changed, 96 insertions, 392 deletions

diff --git a/src/corelib/arch/arch.pri b/src/corelib/arch/arch.pri
index a178c24366..51e67abfd4 100644
--- a/src/corelib/arch/arch.pri
+++ b/src/corelib/arch/arch.pri
@@ -9,7 +9,6 @@ HEADERS += \
     arch/qatomic_armv7.h \
     arch/qatomic_bfin.h \
     arch/qatomic_bootstrap.h \
-    arch/qatomic_i386.h \
     arch/qatomic_ia64.h \
     arch/qatomic_mips.h \
     arch/qatomic_power.h \
diff --git a/src/corelib/arch/qatomic_i386.h b/src/corelib/arch/qatomic_i386.h
deleted file mode 100644
index a81376108c..0000000000
--- a/src/corelib/arch/qatomic_i386.h
+++ /dev/null
@@ -1,358 +0,0 @@
-/****************************************************************************
-**
-** Copyright (C) 2012 Nokia Corporation and/or its subsidiary(-ies).
-** Copyright (C) 2011 Thiago Macieira <thiago@kde.org>
-** Contact: http://www.qt-project.org/
-**
-** This file is part of the QtCore module of the Qt Toolkit.
-**
-** $QT_BEGIN_LICENSE:LGPL$
-** GNU Lesser General Public License Usage
-** This file may be used under the terms of the GNU Lesser General Public
-** License version 2.1 as published by the Free Software Foundation and
-** appearing in the file LICENSE.LGPL included in the packaging of this
-** file. Please review the following information to ensure the GNU Lesser
-** General Public License version 2.1 requirements will be met:
-** http://www.gnu.org/licenses/old-licenses/lgpl-2.1.html.
-**
-** In addition, as a special exception, Nokia gives you certain additional
-** rights. These rights are described in the Nokia Qt LGPL Exception
-** version 1.1, included in the file LGPL_EXCEPTION.txt in this package.
-**
-** GNU General Public License Usage
-** Alternatively, this file may be used under the terms of the GNU General
-** Public License version 3.0 as published by the Free Software Foundation
-** and appearing in the file LICENSE.GPL included in the packaging of this
-** file. Please review the following information to ensure the GNU General
-** Public License version 3.0 requirements will be met:
-** http://www.gnu.org/copyleft/gpl.html.
-**
-** Other Usage
-** Alternatively, this file may be used in accordance with the terms and
-** conditions contained in a signed written agreement between you and Nokia.
-**
-**
-**
-**
-**
-**
-** $QT_END_LICENSE$
-**
-****************************************************************************/
-
-#ifndef QATOMIC_I386_H
-#define QATOMIC_I386_H
-
-#include <QtCore/qgenericatomic.h>
-
-QT_BEGIN_HEADER
-QT_BEGIN_NAMESPACE
-
-#if 0
-// silence syncqt warnings
-QT_END_NAMESPACE
-QT_END_HEADER
-
-#pragma qt_sync_stop_processing
-#endif
-
-template<> struct QAtomicIntegerTraits<int> { enum { IsInteger = 1 }; };
-template<> struct QAtomicIntegerTraits<unsigned int> { enum { IsInteger = 1 }; };
-
-#define Q_ATOMIC_INT_REFERENCE_COUNTING_IS_ALWAYS_NATIVE
-#define Q_ATOMIC_INT_REFERENCE_COUNTING_IS_WAIT_FREE
-
-#define Q_ATOMIC_INT_TEST_AND_SET_IS_ALWAYS_NATIVE
-#define Q_ATOMIC_INT_TEST_AND_SET_IS_WAIT_FREE
-
-#define Q_ATOMIC_INT_FETCH_AND_STORE_IS_ALWAYS_NATIVE
-#define Q_ATOMIC_INT_FETCH_AND_STORE_IS_WAIT_FREE
-
-#define Q_ATOMIC_INT_FETCH_AND_ADD_IS_ALWAYS_NATIVE
-#define Q_ATOMIC_INT_FETCH_AND_ADD_IS_WAIT_FREE
-
-#define Q_ATOMIC_INT32_IS_SUPPORTED
-
-#define Q_ATOMIC_INT32_REFERENCE_COUNTING_IS_ALWAYS_NATIVE
-#define Q_ATOMIC_INT32_REFERENCE_COUNTING_IS_WAIT_FREE
-
-#define Q_ATOMIC_INT32_TEST_AND_SET_IS_ALWAYS_NATIVE
-#define Q_ATOMIC_INT32_TEST_AND_SET_IS_WAIT_FREE
-
-#define Q_ATOMIC_INT32_FETCH_AND_STORE_IS_ALWAYS_NATIVE
-#define Q_ATOMIC_INT32_FETCH_AND_STORE_IS_WAIT_FREE
-
-#define Q_ATOMIC_INT32_FETCH_AND_ADD_IS_ALWAYS_NATIVE
-#define Q_ATOMIC_INT32_FETCH_AND_ADD_IS_WAIT_FREE
-
-#define Q_ATOMIC_POINTER_TEST_AND_SET_IS_ALWAYS_NATIVE
-#define Q_ATOMIC_POINTER_TEST_AND_SET_IS_WAIT_FREE
-
-#define Q_ATOMIC_POINTER_FETCH_AND_STORE_IS_ALWAYS_NATIVE
-#define Q_ATOMIC_POINTER_FETCH_AND_STORE_IS_WAIT_FREE
-
-#define Q_ATOMIC_POINTER_FETCH_AND_ADD_IS_ALWAYS_NATIVE
-#define Q_ATOMIC_POINTER_FETCH_AND_ADD_IS_WAIT_FREE
-
-template <int size> struct QBasicAtomicOps: QGenericAtomicOps<QBasicAtomicOps<size> >
-{
-    static inline bool isReferenceCountingNative() { return true; }
-    static inline bool isReferenceCountingWaitFree() { return true; }
-    template <typename T> static bool ref(T &_q_value);
-    template <typename T> static bool deref(T &_q_value);
-
-    static inline bool isTestAndSetNative() { return true; }
-    static inline bool isTestAndSetWaitFree() { return true; }
-    template <typename T> static bool testAndSetRelaxed(T &_q_value, T expectedValue, T newValue);
-
-    static inline bool isFetchAndStoreNative() { return true; }
-    static inline bool isFetchAndStoreWaitFree() { return true; }
-    template <typename T> static T fetchAndStoreRelaxed(T &_q_value, T newValue);
-
-    static inline bool isFetchAndAddNative() { return true; }
-    static inline bool isFetchAndAddWaitFree() { return true; }
-    template <typename T> static
-    T fetchAndAddRelaxed(T &_q_value, typename QAtomicAdditiveType<T>::AdditiveT valueToAdd);
-};
-
-template <typename T> struct QAtomicOps : QBasicAtomicOps<sizeof(T)>
-{
-    typedef T Type;
-};
-
-#if defined(Q_CC_GNU) || defined(Q_CC_INTEL)
-
-template<> struct QAtomicIntegerTraits<char> { enum { IsInteger = 1 }; };
-template<> struct QAtomicIntegerTraits<signed char> { enum { IsInteger = 1 }; };
-template<> struct QAtomicIntegerTraits<unsigned char> { enum { IsInteger = 1 }; };
-template<> struct QAtomicIntegerTraits<short> { enum { IsInteger = 1 }; };
-template<> struct QAtomicIntegerTraits<unsigned short> { enum { IsInteger = 1 }; };
-template<> struct QAtomicIntegerTraits<long> { enum { IsInteger = 1 }; };
-template<> struct QAtomicIntegerTraits<unsigned long> { enum { IsInteger = 1 }; };
-template<> struct QAtomicIntegerTraits<long long> { enum { IsInteger = 1 }; };
-template<> struct QAtomicIntegerTraits<unsigned long long> { enum { IsInteger = 1 }; };
-
-template<> template<typename T> inline
-bool QBasicAtomicOps<1>::ref(T &_q_value)
-{
-    unsigned char ret;
-    asm volatile("lock\n"
-                 "addb $1, %0\n"
-                 "setne %1"
-                 : "+m" (_q_value), "=qm" (ret)
-                 :
-                 : "memory");
-    return ret != 0;
-}
-
-template<> template<typename T> inline
-bool QBasicAtomicOps<2>::ref(T &_q_value)
-{
-    unsigned char ret;
-    asm volatile("lock\n"
-                 "incw %0\n"
-                 "setne %1"
-                 : "+m" (_q_value), "=qm" (ret)
-                 :
-                 : "memory");
-    return ret != 0;
-}
-
-template<> template<typename T> inline
-bool QBasicAtomicOps<4>::ref(T &_q_value)
-{
-    unsigned char ret;
-    asm volatile("lock\n"
-                 "addl $1, %0\n"
-                 "setne %1"
-                 : "+m" (_q_value), "=qm" (ret)
-                 :
-                 : "memory");
-    return ret != 0;
-}
-
-template<> template <typename T> inline
-bool QBasicAtomicOps<1>::deref(T &_q_value)
-{
-    unsigned char ret;
-    asm volatile("lock\n"
-                 "subb $1, %0\n"
-                 "setne %1"
-                 : "+m" (_q_value), "=qm" (ret)
-                 :
-                 : "memory");
-    return ret != 0;
-}
-
-template<> template <typename T> inline
-bool QBasicAtomicOps<2>::deref(T &_q_value)
-{
-    unsigned char ret;
-    asm volatile("lock\n"
-                 "decw %0\n"
-                 "setne %1"
-                 : "+m" (_q_value), "=qm" (ret)
-                 :
-                 : "memory");
-    return ret != 0;
-}
-
-template<> template <typename T> inline
-bool QBasicAtomicOps<4>::deref(T &_q_value)
-{
-    unsigned char ret;
-    asm volatile("lock\n"
-                 "subl $1, %0\n"
-                 "setne %1"
-                 : "+m" (_q_value), "=qm" (ret)
-                 :
-                 : "memory");
-    return ret != 0;
-}
-
-template<int size> template <typename T> inline
-bool QBasicAtomicOps<size>::testAndSetRelaxed(T &_q_value, T expectedValue, T newValue)
-{
-    unsigned char ret;
-    asm volatile("lock\n"
-                 "cmpxchg %3,%2\n"
-                 "sete %1\n"
-                 : "=a" (newValue), "=qm" (ret), "+m" (_q_value)
-                 : "r" (newValue), "0" (expectedValue)
-                 : "memory");
-    return ret != 0;
-}
-
-template<> template <typename T> inline
-bool QBasicAtomicOps<1>::testAndSetRelaxed(T &_q_value, T expectedValue, T newValue)
-{
-    unsigned char ret;
-    asm volatile("lock\n"
-                 "cmpxchg %3,%2\n"
-                 "sete %1\n"
-                 : "=a" (newValue), "=qm" (ret), "+m" (_q_value)
-                 : "q" (newValue), "0" (expectedValue)
-                 : "memory");
-    return ret != 0;
-}
-
-template<int size> template <typename T> inline
-T QBasicAtomicOps<size>::fetchAndStoreRelaxed(T &_q_value, T newValue)
-{
-    asm volatile("xchg %0,%1"
-                 : "=r" (newValue), "+m" (_q_value)
-                 : "0" (newValue)
-                 : "memory");
-    return newValue;
-}
-
-template<> template <typename T> inline
-T QBasicAtomicOps<1>::fetchAndStoreRelaxed(T &_q_value, T newValue)
-{
-    asm volatile("xchg %0,%1"
-                 : "=q" (newValue), "+m" (_q_value)
-                 : "0" (newValue)
-                 : "memory");
-    return newValue;
-}
-
-template<int size> template <typename T> inline
-T QBasicAtomicOps<size>::fetchAndAddRelaxed(T &_q_value, typename QAtomicAdditiveType<T>::AdditiveT valueToAdd)
-{
-    T result;
-    asm volatile("lock\n"
-                 "xadd %0,%1"
-                 : "=r" (result), "+m" (_q_value)
-                 : "0" (T(valueToAdd * QAtomicAdditiveType<T>::AddScale))
-                 : "memory");
-    return result;
-}
-
-template<> template <typename T> inline
-T QBasicAtomicOps<1>::fetchAndAddRelaxed(T &_q_value, typename QAtomicAdditiveType<T>::AdditiveT valueToAdd)
-{
-    T result;
-    asm volatile("lock\n"
-                 "xadd %0,%1"
-                 : "=q" (result), "+m" (_q_value)
-                 : "0" (T(valueToAdd * QAtomicAdditiveType<T>::AddScale))
-                 : "memory");
-    return result;
-}
-
-#define Q_ATOMIC_INT8_IS_SUPPORTED
-
-#define Q_ATOMIC_INT8_REFERENCE_COUNTING_IS_ALWAYS_NATIVE
-#define Q_ATOMIC_INT8_REFERENCE_COUNTING_IS_WAIT_FREE
-
-#define Q_ATOMIC_INT8_TEST_AND_SET_IS_ALWAYS_NATIVE
-#define Q_ATOMIC_INT8_TEST_AND_SET_IS_WAIT_FREE
-
-#define Q_ATOMIC_INT8_FETCH_AND_STORE_IS_ALWAYS_NATIVE
-#define Q_ATOMIC_INT8_FETCH_AND_STORE_IS_WAIT_FREE
-
-#define Q_ATOMIC_INT8_FETCH_AND_ADD_IS_ALWAYS_NATIVE
-#define Q_ATOMIC_INT8_FETCH_AND_ADD_IS_WAIT_FREE
-
-#define Q_ATOMIC_INT16_IS_SUPPORTED
-
-#define Q_ATOMIC_INT16_REFERENCE_COUNTING_IS_ALWAYS_NATIVE
-#define Q_ATOMIC_INT16_REFERENCE_COUNTING_IS_WAIT_FREE
-
-#define Q_ATOMIC_INT16_TEST_AND_SET_IS_ALWAYS_NATIVE
-#define Q_ATOMIC_INT16_TEST_AND_SET_IS_WAIT_FREE
-
-#define Q_ATOMIC_INT16_FETCH_AND_STORE_IS_ALWAYS_NATIVE
-#define Q_ATOMIC_INT16_FETCH_AND_STORE_IS_WAIT_FREE
-
-#define Q_ATOMIC_INT16_FETCH_AND_ADD_IS_ALWAYS_NATIVE
-#define Q_ATOMIC_INT16_FETCH_AND_ADD_IS_WAIT_FREE
-
-template <> struct QBasicAtomicOps<8>: QGenericAtomicOps<QBasicAtomicOps<8> >
-{
-    static inline bool isTestAndSetNative() { return true; }
-    static inline bool isTestAndSetWaitFree() { return true; }
-    template <typename T> static inline
-    bool testAndSetRelaxed(T &_q_value, T expectedValue, T newValue)
-    {
-#ifdef __PIC__
-# define EBX_reg "r"
-# define EBX_load(reg) "xchg " reg ", %%ebx\n"
-#else
-# define EBX_reg "b"
-# define EBX_load(reg)
-#endif
-        quint32 highExpectedValue = quint32(newValue >> 32); // ECX
-        asm volatile(EBX_load("%3")
-                     "lock\n"
-                     "cmpxchg8b %0\n"
-                     EBX_load("%3")
-                     "sete %%cl\n"
-                     : "+m" (_q_value), "+c" (highExpectedValue), "+&A" (expectedValue)
-                     : EBX_reg (quint32(newValue & 0xffffffff))
-                     : "memory");
-        // if the comparison failed, expectedValue here contains the current value
-        return quint8(highExpectedValue) != 0;
-#undef EBX_reg
-#undef EBX_load
-    }
-};
-#define Q_ATOMIC_INT64_IS_SUPPORTED
-
-#define Q_ATOMIC_INT64_REFERENCE_COUNTING_IS_NOT_NATIVE
-
-#define Q_ATOMIC_INT64_TEST_AND_SET_IS_NOT_NATIVE
-#define Q_ATOMIC_INT64_TEST_AND_SET_IS_WAIT_FREE
-
-#define Q_ATOMIC_INT64_FETCH_AND_STORE_IS_NATIVE
-
-#define Q_ATOMIC_INT64_FETCH_AND_ADD_IS_NOT_NATIVE
-
-#else
-#  error "This compiler for i386 is not supported"
-#endif
-
-QT_END_NAMESPACE
-QT_END_HEADER
-
-#endif // QATOMIC_I386_H
diff --git a/src/corelib/arch/qatomic_x86.h b/src/corelib/arch/qatomic_x86.h
index 58505e2aa9..5212e8014b 100644
--- a/src/corelib/arch/qatomic_x86.h
+++ b/src/corelib/arch/qatomic_x86.h
@@ -40,8 +40,8 @@
 **
 ****************************************************************************/
 
-#ifndef QATOMIC_X86_64_H
-#define QATOMIC_X86_64_H
+#ifndef QATOMIC_X86_H
+#define QATOMIC_X86_H
 
 #include <QtCore/qgenericatomic.h>
 
@@ -121,7 +121,7 @@ template <typename T> struct QAtomicOps : QBasicAtomicOps<sizeof(T)>
     typedef T Type;
 };
 
-#if defined(Q_CC_GNU) || defined(Q_CC_INTEL)
+#if defined(Q_CC_GNU)
 
 template<> struct QAtomicIntegerTraits<char> { enum { IsInteger = 1 }; };
 template<> struct QAtomicIntegerTraits<signed char> { enum { IsInteger = 1 }; };
@@ -133,6 +133,34 @@ template<> struct QAtomicIntegerTraits<unsigned long> { enum { IsInteger = 1 };
 template<> struct QAtomicIntegerTraits<long long> { enum { IsInteger = 1 }; };
 template<> struct QAtomicIntegerTraits<unsigned long long> { enum { IsInteger = 1 }; };
 
+/*
+ * Guide for the inline assembly below:
+ *
+ * x86 instructions are in the form "{opcode}{length} {source}, {destination}",
+ * where the length is one of the letters "b" (byte), "w" (word, 16-bit), "l"
+ * (dword, 32-bit), "q" (qword, 64-bit).
+ *
+ * In most cases, we can omit the length because it's inferred from one of the
+ * registers. For example, "xchg %0,%1" doesn't need the length suffix because
+ * we can only exchange data of the same size and one of the operands must be a
+ * register.
+ *
+ * The exception is the increment and decrement functions, where we add and
+ * subtract an immediate value (1). For those, we need to specify the length.
+ * GCC and ICC support the syntax "add%z0 $1, %0", where "%z0" expands to the
+ * length of the operand. Unfortunately, clang as of 3.0 doesn't support that.
+ * For that reason, the ref() and deref() functions are rolled out for all
+ * sizes.
+ *
+ * The functions are also rolled out for the 1-byte operations since those
+ * require a special register constraint "q" to force the compiler to schedule
+ * one of the 8-bit registers. It's probably a compiler bug that it tries to
+ * use a register that doesn't exist.
+ *
+ * Finally, 64-bit operations are supported via the cmpxchg8b instruction on
+ * 32-bit processors, via specialisation below.
+ */
+
 template<> template<typename T> inline
 bool QBasicAtomicOps<1>::ref(T &_q_value)
 {
@@ -172,19 +200,6 @@ bool QBasicAtomicOps<4>::ref(T &_q_value)
     return ret != 0;
 }
 
-template<> template<typename T> inline
-bool QBasicAtomicOps<8>::ref(T &_q_value)
-{
-    unsigned char ret;
-    asm volatile("lock\n"
-                 "addq $1, %0\n"
-                 "setne %1"
-                 : "=m" (_q_value), "=qm" (ret)
-                 : "m" (_q_value)
-                 : "memory");
-    return ret != 0;
-}
-
 template<> template <typename T> inline
 bool QBasicAtomicOps<1>::deref(T &_q_value)
 {
@@ -223,19 +238,6 @@ bool QBasicAtomicOps<4>::deref(T &_q_value)
     return ret != 0;
 }
 
-template<> template <typename T> inline
-bool QBasicAtomicOps<8>::deref(T &_q_value)
-{
-    unsigned char ret;
-    asm volatile("lock\n"
-                 "subq $1, %0\n"
-                 "setne %1"
-                 : "=m" (_q_value), "=qm" (ret)
-                 : "m" (_q_value)
-                 : "memory");
-    return ret != 0;
-}
-
 template<int size> template <typename T> inline
 bool QBasicAtomicOps<size>::testAndSetRelaxed(T &_q_value, T expectedValue, T newValue)
 {
@@ -348,12 +350,73 @@ T QBasicAtomicOps<1>::fetchAndAddRelaxed(T &_q_value, typename QAtomicAdditiveTy
 #define Q_ATOMIC_INT64_FETCH_AND_ADD_IS_ALWAYS_NATIVE
 #define Q_ATOMIC_INT64_FETCH_AND_ADD_IS_WAIT_FREE
 
-#else // !Q_CC_INTEL && !Q_CC_GNU
-#  error "This compiler for x86_64 is not supported"
-#endif // Q_CC_GNU || Q_CC_INTEL
+#ifdef Q_PROCESSOR_X86_64
+// native support for 64-bit types
+template<> template<typename T> inline
+bool QBasicAtomicOps<8>::ref(T &_q_value)
+{
+    unsigned char ret;
+    asm volatile("lock\n"
+                 "addq $1, %0\n"
+                 "setne %1"
+                 : "=m" (_q_value), "=qm" (ret)
+                 : "m" (_q_value)
+                 : "memory");
+    return ret != 0;
+}
+
+template<> template <typename T> inline
+bool QBasicAtomicOps<8>::deref(T &_q_value)
+{
+    unsigned char ret;
+    asm volatile("lock\n"
+                 "subq $1, %0\n"
+                 "setne %1"
+                 : "=m" (_q_value), "=qm" (ret)
+                 : "m" (_q_value)
+                 : "memory");
+    return ret != 0;
+}
+#else
+// i386 architecture, emulate 64-bit support via cmpxchg8b
+template <> struct QBasicAtomicOps<8>: QGenericAtomicOps<QBasicAtomicOps<8> >
+{
+    static inline bool isTestAndSetNative() { return true; }
+    static inline bool isTestAndSetWaitFree() { return true; }
+    template <typename T> static inline
+    bool testAndSetRelaxed(T &_q_value, T expectedValue, T newValue)
+    {
+#ifdef __PIC__
+# define EBX_reg "r"
+# define EBX_load(reg) "xchg " reg ", %%ebx\n"
+#else
+# define EBX_reg "b"
+# define EBX_load(reg)
+#endif
+        quint32 highExpectedValue = quint32(newValue >> 32); // ECX
+        asm volatile(EBX_load("%3")
+                     "lock\n"
+                     "cmpxchg8b %0\n"
+                     EBX_load("%3")
+                     "sete %%cl\n"
+                     : "+m" (_q_value), "+c" (highExpectedValue), "+&A" (expectedValue)
+                     : EBX_reg (quint32(newValue & 0xffffffff))
+                     : "memory");
+        // if the comparison failed, expectedValue here contains the current value
+        return quint8(highExpectedValue) != 0;
+#undef EBX_reg
+#undef EBX_load
+    }
+};
+#endif
+
+#else
+#  error "This compiler for x86 is not supported"
+#endif
+
 
 QT_END_NAMESPACE
 
 QT_END_HEADER
 
-#endif // QATOMIC_X86_64_H
+#endif // QATOMIC_X86_H