int main()
{
// Small constants use immediate field
printf("0x%08x\n", __sync_fetch_and_add(&foo, 1)); // CHECK: 0x5a5a5a5a
printf("0x%08x\n", __sync_add_and_fetch(&foo, 1)); // CHECK: 0x5a5a5a5c
printf("0x%08x\n", __sync_add_and_fetch(&foo, 1)); // CHECK: 0x5a5a5a5d
printf("0x%08x\n", __sync_fetch_and_add(&foo, 1)); // CHECK: 0x5a5a5a5d
// Large constants require a separate load.
printf("0x%08x\n", __sync_add_and_fetch(&foo, 0x10000000)); // CHECK: 0x6a5a5a5e
printf("0x%08x\n", __sync_sub_and_fetch(&foo, 0x20000000)); // CHECK: 0x4a5a5a5e
printf("0x%08x\n", __sync_and_and_fetch(&foo, 0xf0ffffff)); // CHECK: 0x405a5a5e
printf("0x%08x\n", __sync_or_and_fetch(&foo, 0x0f000000)); // CHECK: 0x4f5a5a5e
printf("0x%08x\n", __sync_xor_and_fetch(&foo, 0x05000000)); // CHECK: 0x4a5a5a5e
// Small constants. These will generate immediate instructions. Test for all forms.
printf("0x%08x\n", __sync_sub_and_fetch(&foo, 1)); // CHECK: 0x4a5a5a5d
printf("0x%08x\n", __sync_and_and_fetch(&foo, 1)); // CHECK: 0x00000001
printf("0x%08x\n", __sync_or_and_fetch(&foo, 2)); // CHECK: 0x00000003
printf("0x%08x\n", __sync_xor_and_fetch(&foo, 0xffffffff)); // CHECK: 0xfffffffc
printf("0x%08x\n", __sync_nand_and_fetch(&foo, 0x5fffffff)); // CHECK: 0xa0000003
// Compare and swap
foo = 2;
// successful
printf("0x%08x\n", __sync_val_compare_and_swap(&foo, 2, 3)); // CHECK: 0x00000002
printf("0x%08x\n", foo); // CHECK: 0x00000003
// not successful
printf("0x%08x\n", __sync_val_compare_and_swap(&foo, 2, 4)); // CHECK: 0x00000003
printf("0x%08x\n", foo); // CHECK: 0x00000003
// not successful
printf("0x%08x\n", __sync_bool_compare_and_swap(&foo, 2, 10)); // CHECK: 0x00000000
printf("0x%08x\n", foo); // CHECK: 0x00000003
// successful
printf("0x%08x\n", __sync_bool_compare_and_swap(&foo, 3, 10)); // CHECK: 0x00000001
printf("0x%08x\n", foo); // CHECK: 0x0000000a
// Unlock
foo = 1;
__sync_lock_release(&foo);
printf("foo = %d\n", foo); // CHECK: foo = 0
// Swap
foo = 0x12;
printf("old value 0x%08x\n", __atomic_exchange_n(&foo, 0x17, __ATOMIC_RELEASE)); // CHECK: 0x00000012
printf("new value 0x%08x\n", foo); // CHECK: new value 0x00000017
return 0;
}
abool & abool::operator-=( uint64_t aValue )
{
if ((aValue % 2) > 0) {
__sync_xor_and_fetch(&_value, 0x01);
}
return *this;
}
abool & abool::operator-=( const aint64_t & aValue )
{
if (((int64_t)aValue % 2) > 0) {
__sync_xor_and_fetch(&_value, 0x01);
}
return *this;
}
static void
do_hi (void)
{
if (__sync_fetch_and_add(AL+4, 1) != 0)
abort ();
if (__sync_fetch_and_add(AL+5, 4) != 0)
abort ();
if (__sync_fetch_and_add(AL+6, 22) != 0)
abort ();
if (__sync_fetch_and_sub(AL+7, 12) != 0)
abort ();
if (__sync_fetch_and_and(AL+8, 7) != -1)
abort ();
if (__sync_fetch_and_or(AL+9, 8) != 0)
abort ();
if (__sync_fetch_and_xor(AL+10, 9) != 0)
abort ();
if (__sync_fetch_and_nand(AL+11, 7) != -1)
abort ();
if (__sync_add_and_fetch(AL+12, 1) != 1)
abort ();
if (__sync_sub_and_fetch(AL+13, 12) != -12)
abort ();
if (__sync_and_and_fetch(AL+14, 7) != 7)
abort ();
if (__sync_or_and_fetch(AL+15, 8) != 8)
abort ();
if (__sync_xor_and_fetch(AL+16, 9) != 9)
abort ();
if (__sync_nand_and_fetch(AL+17, 7) != ~7)
abort ();
}
static void
do_qi (void)
{
if (__sync_fetch_and_add(AI+4, 1) != 0)
abort ();
if (__sync_fetch_and_add(AI+5, 4) != 0)
abort ();
if (__sync_fetch_and_add(AI+6, 22) != 0)
abort ();
if (__sync_fetch_and_sub(AI+7, 12) != 0)
abort ();
if (__sync_fetch_and_and(AI+8, 7) != (char)-1)
abort ();
if (__sync_fetch_and_or(AI+9, 8) != 0)
abort ();
if (__sync_fetch_and_xor(AI+10, 9) != 0)
abort ();
if (__sync_fetch_and_nand(AI+11, 7) != (char)-1)
abort ();
if (__sync_add_and_fetch(AI+12, 1) != 1)
abort ();
if (__sync_sub_and_fetch(AI+13, 12) != (char)-12)
abort ();
if (__sync_and_and_fetch(AI+14, 7) != 7)
abort ();
if (__sync_or_and_fetch(AI+15, 8) != 8)
abort ();
if (__sync_xor_and_fetch(AI+16, 9) != 9)
abort ();
if (__sync_nand_and_fetch(AI+17, 7) != (char)~7)
abort ();
}
/**
* These are the standard RHS operators for this guy, and are all
* handled in a consistent, thread-safe way. The way this works for
* a boolean is that each move "flips" the state, so a simple
* modulo 2 gives us whether or not to flip this value at all.
*/
abool & abool::operator+=( bool aValue )
{
if (aValue) {
__sync_xor_and_fetch(&_value, 0x01);
}
return *this;
}
abool & abool::operator-=( const abool & aValue )
{
if ((bool)aValue) {
__sync_xor_and_fetch(&_value, 0x01);
}
return *this;
}
void test_op_and_fetch (void)
{
sc = __sync_add_and_fetch (&sc, uc);
uc = __sync_add_and_fetch (&uc, uc);
ss = __sync_add_and_fetch (&ss, uc);
us = __sync_add_and_fetch (&us, uc);
si = __sync_add_and_fetch (&si, uc);
ui = __sync_add_and_fetch (&ui, uc);
sll = __sync_add_and_fetch (&sll, uc);
ull = __sync_add_and_fetch (&ull, uc);
sc = __sync_sub_and_fetch (&sc, uc);
uc = __sync_sub_and_fetch (&uc, uc);
ss = __sync_sub_and_fetch (&ss, uc);
us = __sync_sub_and_fetch (&us, uc);
si = __sync_sub_and_fetch (&si, uc);
ui = __sync_sub_and_fetch (&ui, uc);
sll = __sync_sub_and_fetch (&sll, uc);
ull = __sync_sub_and_fetch (&ull, uc);
sc = __sync_or_and_fetch (&sc, uc);
uc = __sync_or_and_fetch (&uc, uc);
ss = __sync_or_and_fetch (&ss, uc);
us = __sync_or_and_fetch (&us, uc);
si = __sync_or_and_fetch (&si, uc);
ui = __sync_or_and_fetch (&ui, uc);
sll = __sync_or_and_fetch (&sll, uc);
ull = __sync_or_and_fetch (&ull, uc);
sc = __sync_xor_and_fetch (&sc, uc);
uc = __sync_xor_and_fetch (&uc, uc);
ss = __sync_xor_and_fetch (&ss, uc);
us = __sync_xor_and_fetch (&us, uc);
si = __sync_xor_and_fetch (&si, uc);
ui = __sync_xor_and_fetch (&ui, uc);
sll = __sync_xor_and_fetch (&sll, uc);
ull = __sync_xor_and_fetch (&ull, uc);
sc = __sync_and_and_fetch (&sc, uc);
uc = __sync_and_and_fetch (&uc, uc);
ss = __sync_and_and_fetch (&ss, uc);
us = __sync_and_and_fetch (&us, uc);
si = __sync_and_and_fetch (&si, uc);
ui = __sync_and_and_fetch (&ui, uc);
sll = __sync_and_and_fetch (&sll, uc);
ull = __sync_and_and_fetch (&ull, uc);
sc = __sync_nand_and_fetch (&sc, uc);
uc = __sync_nand_and_fetch (&uc, uc);
ss = __sync_nand_and_fetch (&ss, uc);
us = __sync_nand_and_fetch (&us, uc);
si = __sync_nand_and_fetch (&si, uc);
ui = __sync_nand_and_fetch (&ui, uc);
sll = __sync_nand_and_fetch (&sll, uc);
ull = __sync_nand_and_fetch (&ull, uc);
}
void test_op_and_fetch (void)
{
sc = __sync_add_and_fetch (&sc, uc); // CHECK: atomicrmw add
uc = __sync_add_and_fetch (&uc, uc); // CHECK: atomicrmw add
ss = __sync_add_and_fetch (&ss, uc); // CHECK: atomicrmw add
us = __sync_add_and_fetch (&us, uc); // CHECK: atomicrmw add
si = __sync_add_and_fetch (&si, uc); // CHECK: atomicrmw add
ui = __sync_add_and_fetch (&ui, uc); // CHECK: atomicrmw add
sll = __sync_add_and_fetch (&sll, uc); // CHECK: atomicrmw add
ull = __sync_add_and_fetch (&ull, uc); // CHECK: atomicrmw add
sc = __sync_sub_and_fetch (&sc, uc); // CHECK: atomicrmw sub
uc = __sync_sub_and_fetch (&uc, uc); // CHECK: atomicrmw sub
ss = __sync_sub_and_fetch (&ss, uc); // CHECK: atomicrmw sub
us = __sync_sub_and_fetch (&us, uc); // CHECK: atomicrmw sub
si = __sync_sub_and_fetch (&si, uc); // CHECK: atomicrmw sub
ui = __sync_sub_and_fetch (&ui, uc); // CHECK: atomicrmw sub
sll = __sync_sub_and_fetch (&sll, uc); // CHECK: atomicrmw sub
ull = __sync_sub_and_fetch (&ull, uc); // CHECK: atomicrmw sub
sc = __sync_or_and_fetch (&sc, uc); // CHECK: atomicrmw or
uc = __sync_or_and_fetch (&uc, uc); // CHECK: atomicrmw or
ss = __sync_or_and_fetch (&ss, uc); // CHECK: atomicrmw or
us = __sync_or_and_fetch (&us, uc); // CHECK: atomicrmw or
si = __sync_or_and_fetch (&si, uc); // CHECK: atomicrmw or
ui = __sync_or_and_fetch (&ui, uc); // CHECK: atomicrmw or
sll = __sync_or_and_fetch (&sll, uc); // CHECK: atomicrmw or
ull = __sync_or_and_fetch (&ull, uc); // CHECK: atomicrmw or
sc = __sync_xor_and_fetch (&sc, uc); // CHECK: atomicrmw xor
uc = __sync_xor_and_fetch (&uc, uc); // CHECK: atomicrmw xor
ss = __sync_xor_and_fetch (&ss, uc); // CHECK: atomicrmw xor
us = __sync_xor_and_fetch (&us, uc); // CHECK: atomicrmw xor
si = __sync_xor_and_fetch (&si, uc); // CHECK: atomicrmw xor
ui = __sync_xor_and_fetch (&ui, uc); // CHECK: atomicrmw xor
sll = __sync_xor_and_fetch (&sll, uc); // CHECK: atomicrmw xor
ull = __sync_xor_and_fetch (&ull, uc); // CHECK: atomicrmw xor
sc = __sync_and_and_fetch (&sc, uc); // CHECK: atomicrmw and
uc = __sync_and_and_fetch (&uc, uc); // CHECK: atomicrmw and
ss = __sync_and_and_fetch (&ss, uc); // CHECK: atomicrmw and
us = __sync_and_and_fetch (&us, uc); // CHECK: atomicrmw and
si = __sync_and_and_fetch (&si, uc); // CHECK: atomicrmw and
ui = __sync_and_and_fetch (&ui, uc); // CHECK: atomicrmw and
sll = __sync_and_and_fetch (&sll, uc); // CHECK: atomicrmw and
ull = __sync_and_and_fetch (&ull, uc); // CHECK: atomicrmw and
}
/* Now check return values. */
static void
do_ret_di (void)
{
if (__sync_val_compare_and_swap (AL+0, 0x100000002ll, 0x1234567890ll) !=
0x100000002ll) abort ();
if (__sync_bool_compare_and_swap (AL+1, 0x200000003ll, 0x1234567890ll) !=
1) abort ();
if (__sync_lock_test_and_set (AL+2, 1) != 0) abort ();
__sync_lock_release (AL+3); /* no return value, but keep to match results. */
/* The following tests should not change the value since the
original does NOT match. */
if (__sync_val_compare_and_swap (AL+4, 0x000000002ll, 0x1234567890ll) !=
0x100000002ll) abort ();
if (__sync_val_compare_and_swap (AL+5, 0x100000000ll, 0x1234567890ll) !=
0x100000002ll) abort ();
if (__sync_bool_compare_and_swap (AL+6, 0x000000002ll, 0x1234567890ll) !=
0) abort ();
if (__sync_bool_compare_and_swap (AL+7, 0x100000000ll, 0x1234567890ll) !=
0) abort ();
if (__sync_fetch_and_add (AL+8, 1) != 0) abort ();
if (__sync_fetch_and_add (AL+9, 0xb000e0000000ll) != 0x1000e0de0000ll) abort ();
if (__sync_fetch_and_sub (AL+10, 22) != 42) abort ();
if (__sync_fetch_and_sub (AL+11, 0xb000e0000000ll) != 0xc001c0de0000ll)
abort ();
if (__sync_fetch_and_and (AL+12, 0x300000007ll) != -1ll) abort ();
if (__sync_fetch_and_or (AL+13, 0x500000009ll) != 0) abort ();
if (__sync_fetch_and_xor (AL+14, 0xe00000001ll) != 0xff00ff0000ll) abort ();
if (__sync_fetch_and_nand (AL+15, 0xa00000007ll) != -1ll) abort ();
/* These should be the same as the fetch_and_* cases except for
return value. */
if (__sync_add_and_fetch (AL+16, 1) != 1) abort ();
if (__sync_add_and_fetch (AL+17, 0xb000e0000000ll) != 0xc001c0de0000ll)
abort ();
if (__sync_sub_and_fetch (AL+18, 22) != 20) abort ();
if (__sync_sub_and_fetch (AL+19, 0xb000e0000000ll) != 0x1000e0de0000ll)
abort ();
if (__sync_and_and_fetch (AL+20, 0x300000007ll) != 0x300000007ll) abort ();
if (__sync_or_and_fetch (AL+21, 0x500000009ll) != 0x500000009ll) abort ();
if (__sync_xor_and_fetch (AL+22, 0xe00000001ll) != 0xf100ff0001ll) abort ();
if (__sync_nand_and_fetch (AL+23, 0xa00000007ll) != ~0xa00000007ll) abort ();
}
static void
do_di (void)
{
if (__sync_val_compare_and_swap(AL+0, 0, 1) != 0)
abort ();
if (__sync_val_compare_and_swap(AL+0, 0, 1) != 1)
abort ();
if (__sync_bool_compare_and_swap(AL+1, 0, 1) != 1)
abort ();
if (__sync_bool_compare_and_swap(AL+1, 0, 1) != 0)
abort ();
if (__sync_lock_test_and_set(AL+2, 1) != 0)
abort ();
if (__sync_fetch_and_add(AL+4, 1) != 0)
abort ();
if (__sync_fetch_and_add(AL+5, 4) != 0)
abort ();
if (__sync_fetch_and_add(AL+6, 22) != 0)
abort ();
if (__sync_fetch_and_sub(AL+7, 12) != 0)
abort ();
if (__sync_fetch_and_and(AL+8, 7) != -1)
abort ();
if (__sync_fetch_and_or(AL+9, 8) != 0)
abort ();
if (__sync_fetch_and_xor(AL+10, 9) != 0)
abort ();
if (__sync_fetch_and_nand(AL+11, 7) != 0)
abort ();
if (__sync_add_and_fetch(AL+12, 1) != 1)
abort ();
if (__sync_sub_and_fetch(AL+13, 12) != -12)
abort ();
if (__sync_and_and_fetch(AL+14, 7) != 7)
abort ();
if (__sync_or_and_fetch(AL+15, 8) != 8)
abort ();
if (__sync_xor_and_fetch(AL+16, 9) != 9)
abort ();
if (__sync_nand_and_fetch(AL+17, 7) != 7)
abort ();
}
/* First check they work in terms of what they do to memory. */
static void
do_noret_di (void)
{
__sync_val_compare_and_swap (AL+0, 0x100000002ll, 0x1234567890ll);
__sync_bool_compare_and_swap (AL+1, 0x200000003ll, 0x1234567890ll);
__sync_lock_test_and_set (AL+2, 1);
__sync_lock_release (AL+3);
/* The following tests should not change the value since the
original does NOT match. */
__sync_val_compare_and_swap (AL+4, 0x000000002ll, 0x1234567890ll);
__sync_val_compare_and_swap (AL+5, 0x100000000ll, 0x1234567890ll);
__sync_bool_compare_and_swap (AL+6, 0x000000002ll, 0x1234567890ll);
__sync_bool_compare_and_swap (AL+7, 0x100000000ll, 0x1234567890ll);
__sync_fetch_and_add (AL+8, 1);
__sync_fetch_and_add (AL+9, 0xb000e0000000ll); /* + to both halves & carry. */
__sync_fetch_and_sub (AL+10, 22);
__sync_fetch_and_sub (AL+11, 0xb000e0000000ll);
__sync_fetch_and_and (AL+12, 0x300000007ll);
__sync_fetch_and_or (AL+13, 0x500000009ll);
__sync_fetch_and_xor (AL+14, 0xe00000001ll);
__sync_fetch_and_nand (AL+15, 0xa00000007ll);
/* These should be the same as the fetch_and_* cases except for
return value. */
__sync_add_and_fetch (AL+16, 1);
/* add to both halves & carry. */
__sync_add_and_fetch (AL+17, 0xb000e0000000ll);
__sync_sub_and_fetch (AL+18, 22);
__sync_sub_and_fetch (AL+19, 0xb000e0000000ll);
__sync_and_and_fetch (AL+20, 0x300000007ll);
__sync_or_and_fetch (AL+21, 0x500000009ll);
__sync_xor_and_fetch (AL+22, 0xe00000001ll);
__sync_nand_and_fetch (AL+23, 0xa00000007ll);
}
int main()
{
{
T x = HILO(5, 3);
T y = __sync_add_and_fetch(&x, DUP(1));
assert(y == HILO(6, 4));
assert(x == HILO(6, 4));
volatile T n = HILO(2, 1);
if (emscripten_has_threading_support())
{
for(int i = 0; i < NUM_THREADS; ++i) pthread_create(&thread[i], NULL, thread_add_and_fetch, (void*)&n);
for(int i = 0; i < NUM_THREADS; ++i) pthread_join(thread[i], NULL);
printf("n: %llx\n", n);
assert(n == HILO(NUM_THREADS*10000ULL+2ULL, NUM_THREADS*10000ULL+1ULL));
}
}
{
T x = HILO(15, 13);
T y = __sync_sub_and_fetch(&x, HILO(10, 10));
assert(y == HILO(5, 3));
assert(x == HILO(5, 3));
volatile T n = HILO(NUM_THREADS*10000ULL+5ULL, NUM_THREADS*10000ULL+3ULL);
if (emscripten_has_threading_support())
{
for(int i = 0; i < NUM_THREADS; ++i) pthread_create(&thread[i], NULL, thread_sub_and_fetch, (void*)&n);
for(int i = 0; i < NUM_THREADS; ++i) pthread_join(thread[i], NULL);
printf("n: %llx\n", n);
assert(n == HILO(5,3));
}
}
{
T x = HILO(32768 + 5, 5);
T y = __sync_or_and_fetch(&x, HILO(65536 + 9, 9));
assert(y == HILO(32768 + 65536 + 13, 13));
assert(x == HILO(32768 + 65536 + 13, 13));
for(int x = 0; x < 100; ++x) // Test a few times for robustness, since this test is so short-lived.
{
or_and_fetch_data = HILO(65536 + (1<<NUM_THREADS), 1<<NUM_THREADS);
if (emscripten_has_threading_support())
{
for(int i = 0; i < NUM_THREADS; ++i)
{
threadArg[i] = DUP(1 << i);
pthread_create(&thread[i], NULL, thread_or_and_fetch, (void*)&threadArg[i]);
}
for(int i = 0; i < NUM_THREADS; ++i) pthread_join(thread[i], NULL);
assert(or_and_fetch_data == HILO(65536 + (1<<(NUM_THREADS+1))-1, (1<<(NUM_THREADS+1))-1));
}
}
}
{
T x = HILO(32768 + 5, 5);
T y = __sync_and_and_fetch(&x, HILO(32768 + 9, 9));
assert(y == HILO(32768 + 1, 1));
assert(x == HILO(32768 + 1, 1));
if (emscripten_has_threading_support())
{
for(int x = 0; x < 100; ++x) // Test a few times for robustness, since this test is so short-lived.
{
and_and_fetch_data = HILO(65536 + (1<<(NUM_THREADS+1))-1, (1<<(NUM_THREADS+1))-1);
for(int i = 0; i < NUM_THREADS; ++i)
{
threadArg[i] = DUP(~(1UL<<i));
pthread_create(&thread[i], NULL, thread_and_and_fetch, (void*)&threadArg[i]);
}
for(int i = 0; i < NUM_THREADS; ++i) pthread_join(thread[i], NULL);
assert(and_and_fetch_data == HILO(65536 + (1<<NUM_THREADS), 1<<NUM_THREADS));
}
}
}
{
T x = HILO(32768 + 5, 5);
T y = __sync_xor_and_fetch(&x, HILO(16384 + 9, 9));
assert(y == HILO(32768 + 16384 + 12, 12));
assert(x == HILO(32768 + 16384 + 12, 12));
if (emscripten_has_threading_support())
{
for(int x = 0; x < 100; ++x) // Test a few times for robustness, since this test is so short-lived.
{
xor_and_fetch_data = HILO(32768 + (1<<NUM_THREADS), 1<<NUM_THREADS);
for(int i = 0; i < NUM_THREADS; ++i)
{
threadArg[i] = DUP(~(1UL<<i));
pthread_create(&thread[i], NULL, thread_xor_and_fetch, (void*)&threadArg[i]);
}
for(int i = 0; i < NUM_THREADS; ++i) pthread_join(thread[i], NULL);
assert(xor_and_fetch_data == HILO(32768 + ((1<<(NUM_THREADS+1))-1), (1<<(NUM_THREADS+1))-1));
}
}
}
// XXX NAND support does not exist in Atomics API.
#if 0
{
T x = 5;
T y = __sync_nand_and_fetch(&x, 9);
assert(y == 5);
assert(x == -2);
const int oddNThreads = NUM_THREADS-1;
for(int x = 0; x < 100; ++x) // Test a few times for robustness, since this test is so short-lived.
{
nand_and_fetch_data = 0;
for(int i = 0; i < oddNThreads; ++i) pthread_create(&thread[i], NULL, thread_nand_and_fetch, (void*)-1);
for(int i = 0; i < oddNThreads; ++i) pthread_join(thread[i], NULL);
assert(nand_and_fetch_data == -1);
}
}
#endif
#ifdef REPORT_RESULT
REPORT_RESULT(0);
#endif
}
void *thread_xor_and_fetch(void *arg)
{
for(int i = 0; i < 9999; ++i) // Odd number of times so that the operation doesn't cancel itself out.
__sync_xor_and_fetch((T*)&xor_and_fetch_data, *(T*)arg);
pthread_exit(0);
}
UInt8 BitXorAtomic8(UInt32 val, UInt8* ptr)
{
return __sync_xor_and_fetch(ptr, val);
}
T xorAndFetch(T value) { return __sync_xor_and_fetch(&_value, value); }
UInt16 BitXorAtomic16(UInt32 val, UInt16* ptr)
{
return __sync_xor_and_fetch(ptr, val);
}
void NANOS_atomic ( int op, int type, void * variable, void * operand )
{
if ( ( type == 0 ) && ( op == 1 || op == 2 || op == 5 || op == 6 || op == 7) )
{ // variable has integer type and the operation is some kind of the following compound assignments:
// plus, minus, and, ior, xor
printf("info: 'atomic' construct implemented using atomic builtins.\n");
int tmp = *((int *) operand);
switch (op)
{
case 1: __sync_add_and_fetch((int *) variable, tmp);
break;
case 2: __sync_sub_and_fetch((int *) variable, tmp);
break;
case 5: __sync_and_and_fetch((int *) variable, tmp);
break;
case 6: __sync_or_and_fetch((int *) variable, tmp);
break;
case 7: __sync_xor_and_fetch((int *) variable, tmp);
break;
};
}
else if ( ( type == 0 ) && ( op == 10 || op == 11) )
{ // variable has integer type and the operation is a pre-/post- incr-/decr- ement
printf("info: 'atomic' construct implemented using atomic builtins.\n");
if (op == 10)
__sync_add_and_fetch((int *) variable, 1);
else
__sync_sub_and_fetch((int *) variable, 1);
}
else
{ // any other case
printf("info: 'atomic' construct implemented using compare and exchange.\n");
if (type == 1)
{ // Float type
// float tmp = *((float *) operand);
printf("Nanos support for Atomic access to floats is not yet implemented\n");
abort();
}
else if (type == 2)
{ // Double type
double tmp = *((double *) operand);
double oldval, newval;
unsigned int sizeof_var = sizeof(variable);
do {
oldval = *((double *) variable);
switch (op)
{
case 1: newval = oldval + tmp;
break;
case 2: newval = oldval - tmp;
break;
case 3: newval = oldval * tmp;
break;
case 4: newval = oldval / tmp;
break;
case 10: newval = oldval + 1;
break;
case 11: newval = oldval - 1;
break;
default: printf("Unhandled operation type while generating Nanos code for OpenMP atomic contruct.");
abort();
}
__sync_synchronize();
} while ( (sizeof_var == 4) ? !__sync_bool_compare_and_swap_4((double *) variable,
*(unsigned int *) &oldval,
*(unsigned int *) &newval) :
(sizeof_var == 8) ? !__sync_bool_compare_and_swap_8((double *) variable,
*(unsigned long *) &oldval,
*(unsigned long *) &newval) :
0 );
}
else
{
printf("Unhandled variable type while generating Nanos code for OpenMP atomic contruct.");
abort( );
}
}
}
/**
* These are the prefix and postfix decrement operators, in that
* order, and the first returns a reference to the same instance
* while the latter has to receive a copy prior to the decrement.
* In the case of the boolean, these simply "flip" the state of the
* bool from true->false->true, etc.
*/
abool & abool::operator--()
{
__sync_xor_and_fetch(&_value, 0x01);
return *this;
}
abool abool::operator--(int)
{
abool pre(*this);
__sync_xor_and_fetch(&_value, 0x01);
return pre;
}
