/* * Distributed under the Boost Software License, Version 1.0. * (See accompanying file LICENSE_1_0.txt or copy at * http://www.boost.org/LICENSE_1_0.txt) * * Copyright (c) 2009 Helge Bahmann * Copyright (c) 2012 Tim Blechmann * Copyright (c) 2014 - 2018 Andrey Semashev */ /*! * \file atomic/detail/ops_gcc_x86_dcas.hpp * * This header contains implementation of the double-width CAS primitive for x86. */ #ifndef BOOST_ATOMIC_DETAIL_OPS_GCC_X86_DCAS_HPP_INCLUDED_ #define BOOST_ATOMIC_DETAIL_OPS_GCC_X86_DCAS_HPP_INCLUDED_ #include #include #include #include #include #include #ifdef BOOST_HAS_PRAGMA_ONCE #pragma once #endif namespace boost { namespace atomics { namespace detail { // Note: In the 32-bit PIC code guarded with BOOST_ATOMIC_DETAIL_X86_ASM_PRESERVE_EBX below we have to avoid using memory // operand constraints because the compiler may choose to use ebx as the base register for that operand. At least, clang // is known to do that. For this reason we have to pre-compute a pointer to storage and pass it in edi. For the same reason // we cannot save ebx to the stack with a mov instruction, so we use esi as a scratch register and restore it afterwards. // Alternatively, we could push/pop the register to the stack, but exchanging the registers is faster. // The need to pass a pointer in edi is a bit wasteful because normally the memory operand would use a base pointer // with an offset (e.g. `this` + offset). But unfortunately, there seems to be no way around it. #if defined(BOOST_ATOMIC_DETAIL_X86_HAS_CMPXCHG8B) template< bool Signed > struct gcc_dcas_x86 { typedef typename make_storage_type< 8u >::type storage_type; typedef typename make_storage_type< 8u >::aligned aligned_storage_type; typedef uint32_t BOOST_ATOMIC_DETAIL_MAY_ALIAS aliasing_uint32_t; static BOOST_CONSTEXPR_OR_CONST bool full_cas_based = true; static BOOST_CONSTEXPR_OR_CONST bool is_always_lock_free = true; static BOOST_FORCEINLINE void store(storage_type volatile& storage, storage_type v, memory_order) BOOST_NOEXCEPT { if (BOOST_LIKELY((((uint32_t)&storage) & 0x00000007) == 0u)) { #if defined(__SSE__) typedef float xmm_t __attribute__((__vector_size__(16))); xmm_t xmm_scratch; __asm__ __volatile__ ( #if defined(__AVX__) "vmovq %[value], %[xmm_scratch]\n\t" "vmovq %[xmm_scratch], %[storage]\n\t" #elif defined(__SSE2__) "movq %[value], %[xmm_scratch]\n\t" "movq %[xmm_scratch], %[storage]\n\t" #else "xorps %[xmm_scratch], %[xmm_scratch]\n\t" "movlps %[value], %[xmm_scratch]\n\t" "movlps %[xmm_scratch], %[storage]\n\t" #endif : [storage] "=m" (storage), [xmm_scratch] "=x" (xmm_scratch) : [value] "m" (v) : "memory" ); #else __asm__ __volatile__ ( "fildll %[value]\n\t" "fistpll %[storage]\n\t" : [storage] "=m" (storage) : [value] "m" (v) : "memory" ); #endif } else { #if defined(BOOST_ATOMIC_DETAIL_X86_ASM_PRESERVE_EBX) __asm__ __volatile__ ( "xchgl %%ebx, %%esi\n\t" "movl %%eax, %%ebx\n\t" "movl (%[dest]), %%eax\n\t" "movl 4(%[dest]), %%edx\n\t" ".align 16\n\t" "1: lock; cmpxchg8b (%[dest])\n\t" "jne 1b\n\t" "xchgl %%ebx, %%esi\n\t" : : "a" ((uint32_t)v), "c" ((uint32_t)(v >> 32)), [dest] "D" (&storage) : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "edx", "memory" ); #else // defined(BOOST_ATOMIC_DETAIL_X86_ASM_PRESERVE_EBX) __asm__ __volatile__ ( "movl %[dest_lo], %%eax\n\t" "movl %[dest_hi], %%edx\n\t" ".align 16\n\t" "1: lock; cmpxchg8b %[dest_lo]\n\t" "jne 1b\n\t" : [dest_lo] "=m" (storage), [dest_hi] "=m" (reinterpret_cast< volatile aliasing_uint32_t* >(&storage)[1]) : [value_lo] "b" ((uint32_t)v), "c" ((uint32_t)(v >> 32)) : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "eax", "edx", "memory" ); #endif // defined(BOOST_ATOMIC_DETAIL_X86_ASM_PRESERVE_EBX) } } static BOOST_FORCEINLINE storage_type load(storage_type const volatile& storage, memory_order) BOOST_NOEXCEPT { storage_type value; if (BOOST_LIKELY((((uint32_t)&storage) & 0x00000007) == 0u)) { #if defined(__SSE__) typedef float xmm_t __attribute__((__vector_size__(16))); xmm_t xmm_scratch; __asm__ __volatile__ ( #if defined(__AVX__) "vmovq %[storage], %[xmm_scratch]\n\t" "vmovq %[xmm_scratch], %[value]\n\t" #elif defined(__SSE2__) "movq %[storage], %[xmm_scratch]\n\t" "movq %[xmm_scratch], %[value]\n\t" #else "xorps %[xmm_scratch], %[xmm_scratch]\n\t" "movlps %[storage], %[xmm_scratch]\n\t" "movlps %[xmm_scratch], %[value]\n\t" #endif : [value] "=m" (value), [xmm_scratch] "=x" (xmm_scratch) : [storage] "m" (storage) : "memory" ); #else __asm__ __volatile__ ( "fildll %[storage]\n\t" "fistpll %[value]\n\t" : [value] "=m" (value) : [storage] "m" (storage) : "memory" ); #endif } else { // Note that despite const qualification cmpxchg8b below may issue a store to the storage. The storage value // will not change, but this prevents the storage to reside in read-only memory. #if defined(BOOST_ATOMIC_DETAIL_X86_NO_ASM_AX_DX_PAIRS) uint32_t value_bits[2]; // We don't care for comparison result here; the previous value will be stored into value anyway. // Also we don't care for ebx and ecx values, they just have to be equal to eax and edx before cmpxchg8b. __asm__ __volatile__ ( "movl %%ebx, %%eax\n\t" "movl %%ecx, %%edx\n\t" "lock; cmpxchg8b %[storage]\n\t" : "=&a" (value_bits[0]), "=&d" (value_bits[1]) : [storage] "m" (storage) : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory" ); BOOST_ATOMIC_DETAIL_MEMCPY(&value, value_bits, sizeof(value)); #else // defined(BOOST_ATOMIC_DETAIL_X86_NO_ASM_AX_DX_PAIRS) // We don't care for comparison result here; the previous value will be stored into value anyway. // Also we don't care for ebx and ecx values, they just have to be equal to eax and edx before cmpxchg8b. __asm__ __volatile__ ( "movl %%ebx, %%eax\n\t" "movl %%ecx, %%edx\n\t" "lock; cmpxchg8b %[storage]\n\t" : "=&A" (value) : [storage] "m" (storage) : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory" ); #endif // defined(BOOST_ATOMIC_DETAIL_X86_NO_ASM_AX_DX_PAIRS) } return value; } static BOOST_FORCEINLINE bool compare_exchange_strong( storage_type volatile& storage, storage_type& expected, storage_type desired, memory_order, memory_order) BOOST_NOEXCEPT { #if defined(__clang__) // Clang cannot allocate eax:edx register pairs but it has sync intrinsics storage_type old_expected = expected; expected = __sync_val_compare_and_swap(&storage, old_expected, desired); return expected == old_expected; #elif defined(BOOST_ATOMIC_DETAIL_X86_ASM_PRESERVE_EBX) bool success; #if defined(BOOST_ATOMIC_DETAIL_ASM_HAS_FLAG_OUTPUTS) __asm__ __volatile__ ( "xchgl %%ebx, %%esi\n\t" "lock; cmpxchg8b (%[dest])\n\t" "xchgl %%ebx, %%esi\n\t" : "+A" (expected), [success] "=@ccz" (success) : "S" ((uint32_t)desired), "c" ((uint32_t)(desired >> 32)), [dest] "D" (&storage) : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory" ); #else // defined(BOOST_ATOMIC_DETAIL_ASM_HAS_FLAG_OUTPUTS) __asm__ __volatile__ ( "xchgl %%ebx, %%esi\n\t" "lock; cmpxchg8b (%[dest])\n\t" "xchgl %%ebx, %%esi\n\t" "sete %[success]\n\t" : "+A" (expected), [success] "=qm" (success) : "S" ((uint32_t)desired), "c" ((uint32_t)(desired >> 32)), [dest] "D" (&storage) : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory" ); #endif // defined(BOOST_ATOMIC_DETAIL_ASM_HAS_FLAG_OUTPUTS) return success; #else // defined(BOOST_ATOMIC_DETAIL_X86_ASM_PRESERVE_EBX) bool success; #if defined(BOOST_ATOMIC_DETAIL_ASM_HAS_FLAG_OUTPUTS) __asm__ __volatile__ ( "lock; cmpxchg8b %[dest]\n\t" : "+A" (expected), [dest] "+m" (storage), [success] "=@ccz" (success) : "b" ((uint32_t)desired), "c" ((uint32_t)(desired >> 32)) : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory" ); #else // defined(BOOST_ATOMIC_DETAIL_ASM_HAS_FLAG_OUTPUTS) __asm__ __volatile__ ( "lock; cmpxchg8b %[dest]\n\t" "sete %[success]\n\t" : "+A" (expected), [dest] "+m" (storage), [success] "=qm" (success) : "b" ((uint32_t)desired), "c" ((uint32_t)(desired >> 32)) : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory" ); #endif // defined(BOOST_ATOMIC_DETAIL_ASM_HAS_FLAG_OUTPUTS) return success; #endif // defined(BOOST_ATOMIC_DETAIL_X86_ASM_PRESERVE_EBX) } static BOOST_FORCEINLINE bool compare_exchange_weak( storage_type volatile& storage, storage_type& expected, storage_type desired, memory_order success_order, memory_order failure_order) BOOST_NOEXCEPT { return compare_exchange_strong(storage, expected, desired, success_order, failure_order); } static BOOST_FORCEINLINE storage_type exchange(storage_type volatile& storage, storage_type v, memory_order) BOOST_NOEXCEPT { #if defined(BOOST_ATOMIC_DETAIL_X86_ASM_PRESERVE_EBX) #if defined(BOOST_ATOMIC_DETAIL_X86_NO_ASM_AX_DX_PAIRS) uint32_t old_bits[2]; __asm__ __volatile__ ( "xchgl %%ebx, %%esi\n\t" "movl (%[dest]), %%eax\n\t" "movl 4(%[dest]), %%edx\n\t" ".align 16\n\t" "1: lock; cmpxchg8b (%[dest])\n\t" "jne 1b\n\t" "xchgl %%ebx, %%esi\n\t" : "=a" (old_bits[0]), "=d" (old_bits[1]) : "S" ((uint32_t)v), "c" ((uint32_t)(v >> 32)), [dest] "D" (&storage) : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory" ); storage_type old_value; BOOST_ATOMIC_DETAIL_MEMCPY(&old_value, old_bits, sizeof(old_value)); return old_value; #else // defined(BOOST_ATOMIC_DETAIL_X86_NO_ASM_AX_DX_PAIRS) storage_type old_value; __asm__ __volatile__ ( "xchgl %%ebx, %%esi\n\t" "movl (%[dest]), %%eax\n\t" "movl 4(%[dest]), %%edx\n\t" ".align 16\n\t" "1: lock; cmpxchg8b (%[dest])\n\t" "jne 1b\n\t" "xchgl %%ebx, %%esi\n\t" : "=A" (old_value) : "S" ((uint32_t)v), "c" ((uint32_t)(v >> 32)), [dest] "D" (&storage) : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory" ); return old_value; #endif // defined(BOOST_ATOMIC_DETAIL_X86_NO_ASM_AX_DX_PAIRS) #else // defined(BOOST_ATOMIC_DETAIL_X86_ASM_PRESERVE_EBX) #if defined(__MINGW32__) && ((__GNUC__+0) * 100 + (__GNUC_MINOR__+0)) < 407 // MinGW gcc up to 4.6 has problems with allocating registers in the asm blocks below uint32_t old_bits[2]; __asm__ __volatile__ ( "movl (%[dest]), %%eax\n\t" "movl 4(%[dest]), %%edx\n\t" ".align 16\n\t" "1: lock; cmpxchg8b (%[dest])\n\t" "jne 1b\n\t" : "=&a" (old_bits[0]), "=&d" (old_bits[1]) : "b" ((uint32_t)v), "c" ((uint32_t)(v >> 32)), [dest] "DS" (&storage) : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory" ); storage_type old_value; BOOST_ATOMIC_DETAIL_MEMCPY(&old_value, old_bits, sizeof(old_value)); return old_value; #elif defined(BOOST_ATOMIC_DETAIL_X86_NO_ASM_AX_DX_PAIRS) uint32_t old_bits[2]; __asm__ __volatile__ ( "movl %[dest_lo], %%eax\n\t" "movl %[dest_hi], %%edx\n\t" ".align 16\n\t" "1: lock; cmpxchg8b %[dest_lo]\n\t" "jne 1b\n\t" : "=&a" (old_bits[0]), "=&d" (old_bits[1]), [dest_lo] "+m" (storage), [dest_hi] "+m" (reinterpret_cast< volatile aliasing_uint32_t* >(&storage)[1]) : "b" ((uint32_t)v), "c" ((uint32_t)(v >> 32)) : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory" ); storage_type old_value; BOOST_ATOMIC_DETAIL_MEMCPY(&old_value, old_bits, sizeof(old_value)); return old_value; #else // defined(BOOST_ATOMIC_DETAIL_X86_NO_ASM_AX_DX_PAIRS) storage_type old_value; __asm__ __volatile__ ( "movl %[dest_lo], %%eax\n\t" "movl %[dest_hi], %%edx\n\t" ".align 16\n\t" "1: lock; cmpxchg8b %[dest_lo]\n\t" "jne 1b\n\t" : "=&A" (old_value), [dest_lo] "+m" (storage), [dest_hi] "+m" (reinterpret_cast< volatile aliasing_uint32_t* >(&storage)[1]) : "b" ((uint32_t)v), "c" ((uint32_t)(v >> 32)) : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory" ); return old_value; #endif // defined(BOOST_ATOMIC_DETAIL_X86_NO_ASM_AX_DX_PAIRS) #endif // defined(BOOST_ATOMIC_DETAIL_X86_ASM_PRESERVE_EBX) } }; #endif // defined(BOOST_ATOMIC_DETAIL_X86_HAS_CMPXCHG8B) #if defined(BOOST_ATOMIC_DETAIL_X86_HAS_CMPXCHG16B) template< bool Signed > struct gcc_dcas_x86_64 { typedef typename make_storage_type< 16u >::type storage_type; typedef typename make_storage_type< 16u >::aligned aligned_storage_type; typedef uint64_t BOOST_ATOMIC_DETAIL_MAY_ALIAS aliasing_uint64_t; static BOOST_CONSTEXPR_OR_CONST bool full_cas_based = true; static BOOST_CONSTEXPR_OR_CONST bool is_always_lock_free = true; static BOOST_FORCEINLINE void store(storage_type volatile& storage, storage_type v, memory_order) BOOST_NOEXCEPT { __asm__ __volatile__ ( "movq %[dest_lo], %%rax\n\t" "movq %[dest_hi], %%rdx\n\t" ".align 16\n\t" "1: lock; cmpxchg16b %[dest_lo]\n\t" "jne 1b\n\t" : [dest_lo] "=m" (storage), [dest_hi] "=m" (reinterpret_cast< volatile aliasing_uint64_t* >(&storage)[1]) : "b" (reinterpret_cast< const aliasing_uint64_t* >(&v)[0]), "c" (reinterpret_cast< const aliasing_uint64_t* >(&v)[1]) : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "rax", "rdx", "memory" ); } static BOOST_FORCEINLINE storage_type load(storage_type const volatile& storage, memory_order) BOOST_NOEXCEPT { // Note that despite const qualification cmpxchg16b below may issue a store to the storage. The storage value // will not change, but this prevents the storage to reside in read-only memory. #if defined(BOOST_ATOMIC_DETAIL_X86_NO_ASM_AX_DX_PAIRS) uint64_t value_bits[2]; // We don't care for comparison result here; the previous value will be stored into value anyway. // Also we don't care for rbx and rcx values, they just have to be equal to rax and rdx before cmpxchg16b. __asm__ __volatile__ ( "movq %%rbx, %%rax\n\t" "movq %%rcx, %%rdx\n\t" "lock; cmpxchg16b %[storage]\n\t" : "=&a" (value_bits[0]), "=&d" (value_bits[1]) : [storage] "m" (storage) : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory" ); storage_type value; BOOST_ATOMIC_DETAIL_MEMCPY(&value, value_bits, sizeof(value)); return value; #else // defined(BOOST_ATOMIC_DETAIL_X86_NO_ASM_AX_DX_PAIRS) storage_type value; // We don't care for comparison result here; the previous value will be stored into value anyway. // Also we don't care for rbx and rcx values, they just have to be equal to rax and rdx before cmpxchg16b. __asm__ __volatile__ ( "movq %%rbx, %%rax\n\t" "movq %%rcx, %%rdx\n\t" "lock; cmpxchg16b %[storage]\n\t" : "=&A" (value) : [storage] "m" (storage) : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory" ); return value; #endif // defined(BOOST_ATOMIC_DETAIL_X86_NO_ASM_AX_DX_PAIRS) } static BOOST_FORCEINLINE bool compare_exchange_strong( storage_type volatile& storage, storage_type& expected, storage_type desired, memory_order, memory_order) BOOST_NOEXCEPT { #if defined(__clang__) // Clang cannot allocate rax:rdx register pairs but it has sync intrinsics storage_type old_expected = expected; expected = __sync_val_compare_and_swap(&storage, old_expected, desired); return expected == old_expected; #elif defined(BOOST_ATOMIC_DETAIL_X86_NO_ASM_AX_DX_PAIRS) // Some compilers can't allocate rax:rdx register pair either but also don't support 128-bit __sync_val_compare_and_swap bool success; __asm__ __volatile__ ( "lock; cmpxchg16b %[dest]\n\t" "sete %[success]\n\t" : [dest] "+m" (storage), "+a" (reinterpret_cast< aliasing_uint64_t* >(&expected)[0]), "+d" (reinterpret_cast< aliasing_uint64_t* >(&expected)[1]), [success] "=q" (success) : "b" (reinterpret_cast< const aliasing_uint64_t* >(&desired)[0]), "c" (reinterpret_cast< const aliasing_uint64_t* >(&desired)[1]) : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory" ); return success; #else // defined(BOOST_ATOMIC_DETAIL_X86_NO_ASM_AX_DX_PAIRS) bool success; #if defined(BOOST_ATOMIC_DETAIL_ASM_HAS_FLAG_OUTPUTS) __asm__ __volatile__ ( "lock; cmpxchg16b %[dest]\n\t" : "+A" (expected), [dest] "+m" (storage), "=@ccz" (success) : "b" (reinterpret_cast< const aliasing_uint64_t* >(&desired)[0]), "c" (reinterpret_cast< const aliasing_uint64_t* >(&desired)[1]) : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory" ); #else // defined(BOOST_ATOMIC_DETAIL_ASM_HAS_FLAG_OUTPUTS) __asm__ __volatile__ ( "lock; cmpxchg16b %[dest]\n\t" "sete %[success]\n\t" : "+A" (expected), [dest] "+m" (storage), [success] "=qm" (success) : "b" (reinterpret_cast< const aliasing_uint64_t* >(&desired)[0]), "c" (reinterpret_cast< const aliasing_uint64_t* >(&desired)[1]) : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory" ); #endif // defined(BOOST_ATOMIC_DETAIL_ASM_HAS_FLAG_OUTPUTS) return success; #endif // defined(BOOST_ATOMIC_DETAIL_X86_NO_ASM_AX_DX_PAIRS) } static BOOST_FORCEINLINE bool compare_exchange_weak( storage_type volatile& storage, storage_type& expected, storage_type desired, memory_order success_order, memory_order failure_order) BOOST_NOEXCEPT { return compare_exchange_strong(storage, expected, desired, success_order, failure_order); } static BOOST_FORCEINLINE storage_type exchange(storage_type volatile& storage, storage_type v, memory_order) BOOST_NOEXCEPT { #if defined(BOOST_ATOMIC_DETAIL_X86_NO_ASM_AX_DX_PAIRS) uint64_t old_bits[2]; __asm__ __volatile__ ( "movq %[dest_lo], %%rax\n\t" "movq %[dest_hi], %%rdx\n\t" ".align 16\n\t" "1: lock; cmpxchg16b %[dest_lo]\n\t" "jne 1b\n\t" : [dest_lo] "+m" (storage), [dest_hi] "+m" (reinterpret_cast< volatile aliasing_uint64_t* >(&storage)[1]), "=&a" (old_bits[0]), "=&d" (old_bits[1]) : "b" (reinterpret_cast< const aliasing_uint64_t* >(&v)[0]), "c" (reinterpret_cast< const aliasing_uint64_t* >(&v)[1]) : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory" ); storage_type old_value; BOOST_ATOMIC_DETAIL_MEMCPY(&old_value, old_bits, sizeof(old_value)); return old_value; #else // defined(BOOST_ATOMIC_DETAIL_X86_NO_ASM_AX_DX_PAIRS) storage_type old_value; __asm__ __volatile__ ( "movq %[dest_lo], %%rax\n\t" "movq %[dest_hi], %%rdx\n\t" ".align 16\n\t" "1: lock; cmpxchg16b %[dest_lo]\n\t" "jne 1b\n\t" : "=&A" (old_value), [dest_lo] "+m" (storage), [dest_hi] "+m" (reinterpret_cast< volatile aliasing_uint64_t* >(&storage)[1]) : "b" (reinterpret_cast< const aliasing_uint64_t* >(&v)[0]), "c" (reinterpret_cast< const aliasing_uint64_t* >(&v)[1]) : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory" ); return old_value; #endif // defined(BOOST_ATOMIC_DETAIL_X86_NO_ASM_AX_DX_PAIRS) } }; #endif // defined(BOOST_ATOMIC_DETAIL_X86_HAS_CMPXCHG16B) } // namespace detail } // namespace atomics } // namespace boost #endif // BOOST_ATOMIC_DETAIL_OPS_GCC_X86_DCAS_HPP_INCLUDED_