Merge
This commit is contained in:
commit
2a37882c1c
@ -205,11 +205,11 @@ HeapRegion* G1AllocRegion::release() {
|
||||
}
|
||||
|
||||
#if G1_ALLOC_REGION_TRACING
|
||||
void G1AllocRegion::trace(const char* str, size_t word_size, HeapWord* result) {
|
||||
void G1AllocRegion::trace(const char* str, size_t min_word_size, size_t desired_word_size, size_t actual_word_size, HeapWord* result) {
|
||||
// All the calls to trace that set either just the size or the size
|
||||
// and the result are considered part of level 2 tracing and are
|
||||
// skipped during level 1 tracing.
|
||||
if ((word_size == 0 && result == NULL) || (G1_ALLOC_REGION_TRACING > 1)) {
|
||||
if ((actual_word_size == 0 && result == NULL) || (G1_ALLOC_REGION_TRACING > 1)) {
|
||||
const size_t buffer_length = 128;
|
||||
char hr_buffer[buffer_length];
|
||||
char rest_buffer[buffer_length];
|
||||
@ -226,10 +226,10 @@ void G1AllocRegion::trace(const char* str, size_t word_size, HeapWord* result) {
|
||||
|
||||
if (G1_ALLOC_REGION_TRACING > 1) {
|
||||
if (result != NULL) {
|
||||
jio_snprintf(rest_buffer, buffer_length, SIZE_FORMAT " " PTR_FORMAT,
|
||||
word_size, result);
|
||||
} else if (word_size != 0) {
|
||||
jio_snprintf(rest_buffer, buffer_length, SIZE_FORMAT, word_size);
|
||||
jio_snprintf(rest_buffer, buffer_length, "min " SIZE_FORMAT " desired " SIZE_FORMAT " actual " SIZE_FORMAT " " PTR_FORMAT,
|
||||
min_word_size, desired_word_size, actual_word_size, result);
|
||||
} else if (min_word_size != 0) {
|
||||
jio_snprintf(rest_buffer, buffer_length, "min " SIZE_FORMAT " desired " SIZE_FORMAT, min_word_size, desired_word_size);
|
||||
} else {
|
||||
jio_snprintf(rest_buffer, buffer_length, "");
|
||||
}
|
||||
|
@ -104,6 +104,15 @@ private:
|
||||
static inline HeapWord* par_allocate(HeapRegion* alloc_region,
|
||||
size_t word_size,
|
||||
bool bot_updates);
|
||||
// Perform a MT-safe allocation out of the given region, with the given
|
||||
// minimum and desired size. Returns the actual size allocated (between
|
||||
// minimum and desired size) in actual_word_size if the allocation has been
|
||||
// successful.
|
||||
static inline HeapWord* par_allocate(HeapRegion* alloc_region,
|
||||
size_t min_word_size,
|
||||
size_t desired_word_size,
|
||||
size_t* actual_word_size,
|
||||
bool bot_updates);
|
||||
|
||||
// Ensure that the region passed as a parameter has been filled up
|
||||
// so that noone else can allocate out of it any more.
|
||||
@ -159,7 +168,18 @@ public:
|
||||
// First-level allocation: Should be called without holding a
|
||||
// lock. It will try to allocate lock-free out of the active region,
|
||||
// or return NULL if it was unable to.
|
||||
inline HeapWord* attempt_allocation(size_t word_size, bool bot_updates);
|
||||
inline HeapWord* attempt_allocation(size_t word_size,
|
||||
bool bot_updates);
|
||||
// Perform an allocation out of the current allocation region, with the given
|
||||
// minimum and desired size. Returns the actual size allocated (between
|
||||
// minimum and desired size) in actual_word_size if the allocation has been
|
||||
// successful.
|
||||
// Should be called without holding a lock. It will try to allocate lock-free
|
||||
// out of the active region, or return NULL if it was unable to.
|
||||
inline HeapWord* attempt_allocation(size_t min_word_size,
|
||||
size_t desired_word_size,
|
||||
size_t* actual_word_size,
|
||||
bool bot_updates);
|
||||
|
||||
// Second-level allocation: Should be called while holding a
|
||||
// lock. It will try to first allocate lock-free out of the active
|
||||
@ -169,6 +189,14 @@ public:
|
||||
// it conform to its locking protocol.
|
||||
inline HeapWord* attempt_allocation_locked(size_t word_size,
|
||||
bool bot_updates);
|
||||
// Same as attempt_allocation_locked(size_t, bool), but allowing specification
|
||||
// of minimum word size of the block in min_word_size, and the maximum word
|
||||
// size of the allocation in desired_word_size. The actual size of the block is
|
||||
// returned in actual_word_size.
|
||||
inline HeapWord* attempt_allocation_locked(size_t min_word_size,
|
||||
size_t desired_word_size,
|
||||
size_t* actual_word_size,
|
||||
bool bot_updates);
|
||||
|
||||
// Should be called to allocate a new region even if the max of this
|
||||
// type of regions has been reached. Should only be called if other
|
||||
@ -191,9 +219,17 @@ public:
|
||||
virtual HeapRegion* release();
|
||||
|
||||
#if G1_ALLOC_REGION_TRACING
|
||||
void trace(const char* str, size_t word_size = 0, HeapWord* result = NULL);
|
||||
void trace(const char* str,
|
||||
size_t min_word_size = 0,
|
||||
size_t desired_word_size = 0,
|
||||
size_t actual_word_size = 0,
|
||||
HeapWord* result = NULL);
|
||||
#else // G1_ALLOC_REGION_TRACING
|
||||
void trace(const char* str, size_t word_size = 0, HeapWord* result = NULL) { }
|
||||
void trace(const char* str,
|
||||
size_t min_word_size = 0,
|
||||
size_t desired_word_size = 0,
|
||||
size_t actual_word_size = 0,
|
||||
HeapWord* result = NULL) { }
|
||||
#endif // G1_ALLOC_REGION_TRACING
|
||||
};
|
||||
|
||||
|
@ -40,52 +40,74 @@ inline HeapWord* G1AllocRegion::allocate(HeapRegion* alloc_region,
|
||||
}
|
||||
}
|
||||
|
||||
inline HeapWord* G1AllocRegion::par_allocate(HeapRegion* alloc_region, size_t word_size, bool bot_updates) {
|
||||
size_t temp;
|
||||
return par_allocate(alloc_region, word_size, word_size, &temp, bot_updates);
|
||||
}
|
||||
|
||||
inline HeapWord* G1AllocRegion::par_allocate(HeapRegion* alloc_region,
|
||||
size_t word_size,
|
||||
size_t min_word_size,
|
||||
size_t desired_word_size,
|
||||
size_t* actual_word_size,
|
||||
bool bot_updates) {
|
||||
assert(alloc_region != NULL, err_msg("pre-condition"));
|
||||
assert(!alloc_region->is_empty(), err_msg("pre-condition"));
|
||||
|
||||
if (!bot_updates) {
|
||||
return alloc_region->par_allocate_no_bot_updates(word_size);
|
||||
return alloc_region->par_allocate_no_bot_updates(min_word_size, desired_word_size, actual_word_size);
|
||||
} else {
|
||||
return alloc_region->par_allocate(word_size);
|
||||
return alloc_region->par_allocate(min_word_size, desired_word_size, actual_word_size);
|
||||
}
|
||||
}
|
||||
|
||||
inline HeapWord* G1AllocRegion::attempt_allocation(size_t word_size,
|
||||
inline HeapWord* G1AllocRegion::attempt_allocation(size_t word_size, bool bot_updates) {
|
||||
size_t temp;
|
||||
return attempt_allocation(word_size, word_size, &temp, bot_updates);
|
||||
}
|
||||
|
||||
inline HeapWord* G1AllocRegion::attempt_allocation(size_t min_word_size,
|
||||
size_t desired_word_size,
|
||||
size_t* actual_word_size,
|
||||
bool bot_updates) {
|
||||
assert(bot_updates == _bot_updates, ar_ext_msg(this, "pre-condition"));
|
||||
|
||||
HeapRegion* alloc_region = _alloc_region;
|
||||
assert(alloc_region != NULL, ar_ext_msg(this, "not initialized properly"));
|
||||
|
||||
HeapWord* result = par_allocate(alloc_region, word_size, bot_updates);
|
||||
HeapWord* result = par_allocate(alloc_region, min_word_size, desired_word_size, actual_word_size, bot_updates);
|
||||
if (result != NULL) {
|
||||
trace("alloc", word_size, result);
|
||||
trace("alloc", min_word_size, desired_word_size, *actual_word_size, result);
|
||||
return result;
|
||||
}
|
||||
trace("alloc failed", word_size);
|
||||
trace("alloc failed", min_word_size, desired_word_size);
|
||||
return NULL;
|
||||
}
|
||||
|
||||
inline HeapWord* G1AllocRegion::attempt_allocation_locked(size_t word_size,
|
||||
inline HeapWord* G1AllocRegion::attempt_allocation_locked(size_t word_size, bool bot_updates) {
|
||||
size_t temp;
|
||||
return attempt_allocation_locked(word_size, word_size, &temp, bot_updates);
|
||||
}
|
||||
|
||||
inline HeapWord* G1AllocRegion::attempt_allocation_locked(size_t min_word_size,
|
||||
size_t desired_word_size,
|
||||
size_t* actual_word_size,
|
||||
bool bot_updates) {
|
||||
// First we have to redo the allocation, assuming we're holding the
|
||||
// appropriate lock, in case another thread changed the region while
|
||||
// we were waiting to get the lock.
|
||||
HeapWord* result = attempt_allocation(word_size, bot_updates);
|
||||
HeapWord* result = attempt_allocation(min_word_size, desired_word_size, actual_word_size, bot_updates);
|
||||
if (result != NULL) {
|
||||
return result;
|
||||
}
|
||||
|
||||
retire(true /* fill_up */);
|
||||
result = new_alloc_region_and_allocate(word_size, false /* force */);
|
||||
result = new_alloc_region_and_allocate(desired_word_size, false /* force */);
|
||||
if (result != NULL) {
|
||||
trace("alloc locked (second attempt)", word_size, result);
|
||||
*actual_word_size = desired_word_size;
|
||||
trace("alloc locked (second attempt)", min_word_size, desired_word_size, *actual_word_size, result);
|
||||
return result;
|
||||
}
|
||||
trace("alloc locked failed", word_size);
|
||||
trace("alloc locked failed", min_word_size, desired_word_size);
|
||||
return NULL;
|
||||
}
|
||||
|
||||
@ -94,13 +116,13 @@ inline HeapWord* G1AllocRegion::attempt_allocation_force(size_t word_size,
|
||||
assert(bot_updates == _bot_updates, ar_ext_msg(this, "pre-condition"));
|
||||
assert(_alloc_region != NULL, ar_ext_msg(this, "not initialized properly"));
|
||||
|
||||
trace("forcing alloc");
|
||||
trace("forcing alloc", word_size, word_size);
|
||||
HeapWord* result = new_alloc_region_and_allocate(word_size, true /* force */);
|
||||
if (result != NULL) {
|
||||
trace("alloc forced", word_size, result);
|
||||
trace("alloc forced", word_size, word_size, word_size, result);
|
||||
return result;
|
||||
}
|
||||
trace("alloc forced failed", word_size);
|
||||
trace("alloc forced failed", word_size, word_size);
|
||||
return NULL;
|
||||
}
|
||||
|
||||
|
@ -24,6 +24,7 @@
|
||||
|
||||
#include "precompiled.hpp"
|
||||
#include "gc/g1/g1Allocator.inline.hpp"
|
||||
#include "gc/g1/g1AllocRegion.inline.hpp"
|
||||
#include "gc/g1/g1CollectedHeap.inline.hpp"
|
||||
#include "gc/g1/g1CollectorPolicy.hpp"
|
||||
#include "gc/g1/g1MarkSweep.hpp"
|
||||
@ -143,11 +144,24 @@ size_t G1Allocator::unsafe_max_tlab_alloc(AllocationContext_t context) {
|
||||
HeapWord* G1Allocator::par_allocate_during_gc(InCSetState dest,
|
||||
size_t word_size,
|
||||
AllocationContext_t context) {
|
||||
size_t temp = 0;
|
||||
HeapWord* result = par_allocate_during_gc(dest, word_size, word_size, &temp, context);
|
||||
assert(result == NULL || temp == word_size,
|
||||
err_msg("Requested " SIZE_FORMAT " words, but got " SIZE_FORMAT " at " PTR_FORMAT,
|
||||
word_size, temp, p2i(result)));
|
||||
return result;
|
||||
}
|
||||
|
||||
HeapWord* G1Allocator::par_allocate_during_gc(InCSetState dest,
|
||||
size_t min_word_size,
|
||||
size_t desired_word_size,
|
||||
size_t* actual_word_size,
|
||||
AllocationContext_t context) {
|
||||
switch (dest.value()) {
|
||||
case InCSetState::Young:
|
||||
return survivor_attempt_allocation(word_size, context);
|
||||
return survivor_attempt_allocation(min_word_size, desired_word_size, actual_word_size, context);
|
||||
case InCSetState::Old:
|
||||
return old_attempt_allocation(word_size, context);
|
||||
return old_attempt_allocation(min_word_size, desired_word_size, actual_word_size, context);
|
||||
default:
|
||||
ShouldNotReachHere();
|
||||
return NULL; // Keep some compilers happy
|
||||
@ -170,37 +184,49 @@ void G1Allocator::set_old_full(AllocationContext_t context) {
|
||||
_old_is_full = true;
|
||||
}
|
||||
|
||||
HeapWord* G1Allocator::survivor_attempt_allocation(size_t word_size,
|
||||
HeapWord* G1Allocator::survivor_attempt_allocation(size_t min_word_size,
|
||||
size_t desired_word_size,
|
||||
size_t* actual_word_size,
|
||||
AllocationContext_t context) {
|
||||
assert(!_g1h->is_humongous(word_size),
|
||||
assert(!_g1h->is_humongous(desired_word_size),
|
||||
"we should not be seeing humongous-size allocations in this path");
|
||||
|
||||
HeapWord* result = survivor_gc_alloc_region(context)->attempt_allocation(word_size,
|
||||
HeapWord* result = survivor_gc_alloc_region(context)->attempt_allocation(min_word_size,
|
||||
desired_word_size,
|
||||
actual_word_size,
|
||||
false /* bot_updates */);
|
||||
if (result == NULL && !survivor_is_full(context)) {
|
||||
MutexLockerEx x(FreeList_lock, Mutex::_no_safepoint_check_flag);
|
||||
result = survivor_gc_alloc_region(context)->attempt_allocation_locked(word_size,
|
||||
result = survivor_gc_alloc_region(context)->attempt_allocation_locked(min_word_size,
|
||||
desired_word_size,
|
||||
actual_word_size,
|
||||
false /* bot_updates */);
|
||||
if (result == NULL) {
|
||||
set_survivor_full(context);
|
||||
}
|
||||
}
|
||||
if (result != NULL) {
|
||||
_g1h->dirty_young_block(result, word_size);
|
||||
_g1h->dirty_young_block(result, *actual_word_size);
|
||||
}
|
||||
return result;
|
||||
}
|
||||
|
||||
HeapWord* G1Allocator::old_attempt_allocation(size_t word_size,
|
||||
HeapWord* G1Allocator::old_attempt_allocation(size_t min_word_size,
|
||||
size_t desired_word_size,
|
||||
size_t* actual_word_size,
|
||||
AllocationContext_t context) {
|
||||
assert(!_g1h->is_humongous(word_size),
|
||||
assert(!_g1h->is_humongous(desired_word_size),
|
||||
"we should not be seeing humongous-size allocations in this path");
|
||||
|
||||
HeapWord* result = old_gc_alloc_region(context)->attempt_allocation(word_size,
|
||||
HeapWord* result = old_gc_alloc_region(context)->attempt_allocation(min_word_size,
|
||||
desired_word_size,
|
||||
actual_word_size,
|
||||
true /* bot_updates */);
|
||||
if (result == NULL && !old_is_full(context)) {
|
||||
MutexLockerEx x(FreeList_lock, Mutex::_no_safepoint_check_flag);
|
||||
result = old_gc_alloc_region(context)->attempt_allocation_locked(word_size,
|
||||
result = old_gc_alloc_region(context)->attempt_allocation_locked(min_word_size,
|
||||
desired_word_size,
|
||||
actual_word_size,
|
||||
true /* bot_updates */);
|
||||
if (result == NULL) {
|
||||
set_old_full(context);
|
||||
@ -242,10 +268,19 @@ HeapWord* G1PLABAllocator::allocate_direct_or_new_plab(InCSetState dest,
|
||||
G1PLAB* alloc_buf = alloc_buffer(dest, context);
|
||||
alloc_buf->retire();
|
||||
|
||||
HeapWord* buf = _allocator->par_allocate_during_gc(dest, plab_word_size, context);
|
||||
size_t actual_plab_size = 0;
|
||||
HeapWord* buf = _allocator->par_allocate_during_gc(dest,
|
||||
required_in_plab,
|
||||
plab_word_size,
|
||||
&actual_plab_size,
|
||||
context);
|
||||
|
||||
assert(buf == NULL || ((actual_plab_size >= required_in_plab) && (actual_plab_size <= plab_word_size)),
|
||||
err_msg("Requested at minimum " SIZE_FORMAT ", desired " SIZE_FORMAT " words, but got " SIZE_FORMAT " at " PTR_FORMAT,
|
||||
required_in_plab, plab_word_size, actual_plab_size, p2i(buf)));
|
||||
|
||||
if (buf != NULL) {
|
||||
// Otherwise.
|
||||
alloc_buf->set_buf(buf, plab_word_size);
|
||||
alloc_buf->set_buf(buf, actual_plab_size);
|
||||
|
||||
HeapWord* const obj = alloc_buf->allocate(word_sz);
|
||||
assert(obj != NULL, err_msg("PLAB should have been big enough, tried to allocate "
|
||||
|
@ -57,10 +57,14 @@ protected:
|
||||
virtual OldGCAllocRegion* old_gc_alloc_region(AllocationContext_t context) = 0;
|
||||
|
||||
// Allocation attempt during GC for a survivor object / PLAB.
|
||||
inline HeapWord* survivor_attempt_allocation(size_t word_size,
|
||||
inline HeapWord* survivor_attempt_allocation(size_t min_word_size,
|
||||
size_t desired_word_size,
|
||||
size_t* actual_word_size,
|
||||
AllocationContext_t context);
|
||||
// Allocation attempt during GC for an old object / PLAB.
|
||||
inline HeapWord* old_attempt_allocation(size_t word_size,
|
||||
inline HeapWord* old_attempt_allocation(size_t min_word_size,
|
||||
size_t desired_word_size,
|
||||
size_t* actual_word_size,
|
||||
AllocationContext_t context);
|
||||
public:
|
||||
G1Allocator(G1CollectedHeap* heap) : _g1h(heap), _survivor_is_full(false), _old_is_full(false) { }
|
||||
@ -102,6 +106,12 @@ public:
|
||||
size_t word_size,
|
||||
AllocationContext_t context);
|
||||
|
||||
HeapWord* par_allocate_during_gc(InCSetState dest,
|
||||
size_t min_word_size,
|
||||
size_t desired_word_size,
|
||||
size_t* actual_word_size,
|
||||
AllocationContext_t context);
|
||||
|
||||
virtual size_t used_in_alloc_regions() = 0;
|
||||
};
|
||||
|
||||
|
@ -26,7 +26,7 @@
|
||||
#define SHARE_VM_GC_G1_G1BLOCKOFFSETTABLE_INLINE_HPP
|
||||
|
||||
#include "gc/g1/g1BlockOffsetTable.hpp"
|
||||
#include "gc/g1/heapRegion.inline.hpp"
|
||||
#include "gc/g1/heapRegion.hpp"
|
||||
#include "gc/shared/space.hpp"
|
||||
|
||||
inline HeapWord* G1BlockOffsetTable::block_start(const void* addr) {
|
||||
|
@ -109,7 +109,7 @@ public:
|
||||
// evacuation pauses between two cleanups, which is _highly_ unlikely.
|
||||
class G1OffsetTableContigSpace: public CompactibleSpace {
|
||||
friend class VMStructs;
|
||||
HeapWord* _top;
|
||||
HeapWord* volatile _top;
|
||||
HeapWord* volatile _scan_top;
|
||||
protected:
|
||||
G1BlockOffsetArrayContigSpace _offsets;
|
||||
@ -134,10 +134,18 @@ class G1OffsetTableContigSpace: public CompactibleSpace {
|
||||
// Reset the G1OffsetTableContigSpace.
|
||||
virtual void initialize(MemRegion mr, bool clear_space, bool mangle_space);
|
||||
|
||||
HeapWord** top_addr() { return &_top; }
|
||||
// Allocation helpers (return NULL if full).
|
||||
inline HeapWord* allocate_impl(size_t word_size, HeapWord* end_value);
|
||||
inline HeapWord* par_allocate_impl(size_t word_size, HeapWord* end_value);
|
||||
HeapWord* volatile* top_addr() { return &_top; }
|
||||
// Try to allocate at least min_word_size and up to desired_size from this Space.
|
||||
// Returns NULL if not possible, otherwise sets actual_word_size to the amount of
|
||||
// space allocated.
|
||||
// This version assumes that all allocation requests to this Space are properly
|
||||
// synchronized.
|
||||
inline HeapWord* allocate_impl(size_t min_word_size, size_t desired_word_size, size_t* actual_word_size);
|
||||
// Try to allocate at least min_word_size and up to desired_size from this Space.
|
||||
// Returns NULL if not possible, otherwise sets actual_word_size to the amount of
|
||||
// space allocated.
|
||||
// This version synchronizes with other calls to par_allocate_impl().
|
||||
inline HeapWord* par_allocate_impl(size_t min_word_size, size_t desired_word_size, size_t* actual_word_size);
|
||||
|
||||
public:
|
||||
void reset_after_compaction() { set_top(compaction_top()); }
|
||||
@ -179,9 +187,14 @@ class G1OffsetTableContigSpace: public CompactibleSpace {
|
||||
HeapWord* block_start(const void* p);
|
||||
HeapWord* block_start_const(const void* p) const;
|
||||
|
||||
// Add offset table update.
|
||||
// Allocation (return NULL if full). Assumes the caller has established
|
||||
// mutually exclusive access to the space.
|
||||
HeapWord* allocate(size_t min_word_size, size_t desired_word_size, size_t* actual_word_size);
|
||||
// Allocation (return NULL if full). Enforces mutual exclusion internally.
|
||||
HeapWord* par_allocate(size_t min_word_size, size_t desired_word_size, size_t* actual_word_size);
|
||||
|
||||
virtual HeapWord* allocate(size_t word_size);
|
||||
HeapWord* par_allocate(size_t word_size);
|
||||
virtual HeapWord* par_allocate(size_t word_size);
|
||||
|
||||
HeapWord* saved_mark_word() const { ShouldNotReachHere(); return NULL; }
|
||||
|
||||
@ -351,8 +364,9 @@ class HeapRegion: public G1OffsetTableContigSpace {
|
||||
// Override for scan_and_forward support.
|
||||
void prepare_for_compaction(CompactPoint* cp);
|
||||
|
||||
inline HeapWord* par_allocate_no_bot_updates(size_t word_size);
|
||||
inline HeapWord* par_allocate_no_bot_updates(size_t min_word_size, size_t desired_word_size, size_t* word_size);
|
||||
inline HeapWord* allocate_no_bot_updates(size_t word_size);
|
||||
inline HeapWord* allocate_no_bot_updates(size_t min_word_size, size_t desired_word_size, size_t* actual_size);
|
||||
|
||||
// If this region is a member of a HeapRegionManager, the index in that
|
||||
// sequence, otherwise -1.
|
||||
|
@ -32,33 +32,39 @@
|
||||
#include "oops/oop.inline.hpp"
|
||||
#include "runtime/atomic.inline.hpp"
|
||||
|
||||
// This version requires locking.
|
||||
inline HeapWord* G1OffsetTableContigSpace::allocate_impl(size_t size,
|
||||
HeapWord* const end_value) {
|
||||
inline HeapWord* G1OffsetTableContigSpace::allocate_impl(size_t min_word_size,
|
||||
size_t desired_word_size,
|
||||
size_t* actual_size) {
|
||||
HeapWord* obj = top();
|
||||
if (pointer_delta(end_value, obj) >= size) {
|
||||
HeapWord* new_top = obj + size;
|
||||
size_t available = pointer_delta(end(), obj);
|
||||
size_t want_to_allocate = MIN2(available, desired_word_size);
|
||||
if (want_to_allocate >= min_word_size) {
|
||||
HeapWord* new_top = obj + want_to_allocate;
|
||||
set_top(new_top);
|
||||
assert(is_aligned(obj) && is_aligned(new_top), "checking alignment");
|
||||
*actual_size = want_to_allocate;
|
||||
return obj;
|
||||
} else {
|
||||
return NULL;
|
||||
}
|
||||
}
|
||||
|
||||
// This version is lock-free.
|
||||
inline HeapWord* G1OffsetTableContigSpace::par_allocate_impl(size_t size,
|
||||
HeapWord* const end_value) {
|
||||
inline HeapWord* G1OffsetTableContigSpace::par_allocate_impl(size_t min_word_size,
|
||||
size_t desired_word_size,
|
||||
size_t* actual_size) {
|
||||
do {
|
||||
HeapWord* obj = top();
|
||||
if (pointer_delta(end_value, obj) >= size) {
|
||||
HeapWord* new_top = obj + size;
|
||||
size_t available = pointer_delta(end(), obj);
|
||||
size_t want_to_allocate = MIN2(available, desired_word_size);
|
||||
if (want_to_allocate >= min_word_size) {
|
||||
HeapWord* new_top = obj + want_to_allocate;
|
||||
HeapWord* result = (HeapWord*)Atomic::cmpxchg_ptr(new_top, top_addr(), obj);
|
||||
// result can be one of two:
|
||||
// the old top value: the exchange succeeded
|
||||
// otherwise: the new value of the top is returned.
|
||||
if (result == obj) {
|
||||
assert(is_aligned(obj) && is_aligned(new_top), "checking alignment");
|
||||
*actual_size = want_to_allocate;
|
||||
return obj;
|
||||
}
|
||||
} else {
|
||||
@ -67,20 +73,34 @@ inline HeapWord* G1OffsetTableContigSpace::par_allocate_impl(size_t size,
|
||||
} while (true);
|
||||
}
|
||||
|
||||
inline HeapWord* G1OffsetTableContigSpace::allocate(size_t size) {
|
||||
HeapWord* res = allocate_impl(size, end());
|
||||
inline HeapWord* G1OffsetTableContigSpace::allocate(size_t min_word_size,
|
||||
size_t desired_word_size,
|
||||
size_t* actual_size) {
|
||||
HeapWord* res = allocate_impl(min_word_size, desired_word_size, actual_size);
|
||||
if (res != NULL) {
|
||||
_offsets.alloc_block(res, size);
|
||||
_offsets.alloc_block(res, *actual_size);
|
||||
}
|
||||
return res;
|
||||
}
|
||||
|
||||
inline HeapWord* G1OffsetTableContigSpace::allocate(size_t word_size) {
|
||||
size_t temp;
|
||||
return allocate(word_size, word_size, &temp);
|
||||
}
|
||||
|
||||
inline HeapWord* G1OffsetTableContigSpace::par_allocate(size_t word_size) {
|
||||
size_t temp;
|
||||
return par_allocate(word_size, word_size, &temp);
|
||||
}
|
||||
|
||||
// Because of the requirement of keeping "_offsets" up to date with the
|
||||
// allocations, we sequentialize these with a lock. Therefore, best if
|
||||
// this is used for larger LAB allocations only.
|
||||
inline HeapWord* G1OffsetTableContigSpace::par_allocate(size_t size) {
|
||||
inline HeapWord* G1OffsetTableContigSpace::par_allocate(size_t min_word_size,
|
||||
size_t desired_word_size,
|
||||
size_t* actual_size) {
|
||||
MutexLocker x(&_par_alloc_lock);
|
||||
return allocate(size);
|
||||
return allocate(min_word_size, desired_word_size, actual_size);
|
||||
}
|
||||
|
||||
inline HeapWord* G1OffsetTableContigSpace::block_start(const void* p) {
|
||||
@ -128,14 +148,23 @@ HeapRegion::block_size(const HeapWord *addr) const {
|
||||
return pointer_delta(next, addr);
|
||||
}
|
||||
|
||||
inline HeapWord* HeapRegion::par_allocate_no_bot_updates(size_t word_size) {
|
||||
inline HeapWord* HeapRegion::par_allocate_no_bot_updates(size_t min_word_size,
|
||||
size_t desired_word_size,
|
||||
size_t* actual_word_size) {
|
||||
assert(is_young(), "we can only skip BOT updates on young regions");
|
||||
return par_allocate_impl(word_size, end());
|
||||
return par_allocate_impl(min_word_size, desired_word_size, actual_word_size);
|
||||
}
|
||||
|
||||
inline HeapWord* HeapRegion::allocate_no_bot_updates(size_t word_size) {
|
||||
size_t temp;
|
||||
return allocate_no_bot_updates(word_size, word_size, &temp);
|
||||
}
|
||||
|
||||
inline HeapWord* HeapRegion::allocate_no_bot_updates(size_t min_word_size,
|
||||
size_t desired_word_size,
|
||||
size_t* actual_word_size) {
|
||||
assert(is_young(), "we can only skip BOT updates on young regions");
|
||||
return allocate_impl(word_size, end());
|
||||
return allocate_impl(min_word_size, desired_word_size, actual_word_size);
|
||||
}
|
||||
|
||||
inline void HeapRegion::note_start_of_marking() {
|
||||
|
@ -428,7 +428,7 @@ uint HeapRegionManager::shrink_by(uint num_regions_to_remove) {
|
||||
|
||||
uncommit_regions(idx_last_found + num_last_found - to_remove, to_remove);
|
||||
|
||||
cur -= num_last_found;
|
||||
cur = idx_last_found;
|
||||
removed += to_remove;
|
||||
}
|
||||
|
||||
|
@ -34,7 +34,7 @@
|
||||
static_field(HeapRegion, GrainBytes, size_t) \
|
||||
static_field(HeapRegion, LogOfHRGrainBytes, int) \
|
||||
\
|
||||
nonstatic_field(G1OffsetTableContigSpace, _top, HeapWord*) \
|
||||
nonstatic_field(G1OffsetTableContigSpace, _top, HeapWord* volatile) \
|
||||
\
|
||||
nonstatic_field(G1HeapRegionTable, _base, address) \
|
||||
nonstatic_field(G1HeapRegionTable, _length, size_t) \
|
||||
|
@ -626,3 +626,75 @@ bool ArrayCopyNode::may_modify(const TypeOopPtr *t_oop, PhaseTransform *phase) {
|
||||
|
||||
return CallNode::may_modify_arraycopy_helper(dest_t, t_oop, phase);
|
||||
}
|
||||
|
||||
bool ArrayCopyNode::may_modify_helper(const TypeOopPtr *t_oop, Node* n, PhaseTransform *phase) {
|
||||
if (n->is_Proj()) {
|
||||
n = n->in(0);
|
||||
if (n->is_Call() && n->as_Call()->may_modify(t_oop, phase)) {
|
||||
return true;
|
||||
}
|
||||
}
|
||||
return false;
|
||||
}
|
||||
|
||||
bool ArrayCopyNode::may_modify(const TypeOopPtr *t_oop, MemBarNode* mb, PhaseTransform *phase) {
|
||||
Node* mem = mb->in(TypeFunc::Memory);
|
||||
|
||||
if (mem->is_MergeMem()) {
|
||||
Node* n = mem->as_MergeMem()->memory_at(Compile::AliasIdxRaw);
|
||||
if (may_modify_helper(t_oop, n, phase)) {
|
||||
return true;
|
||||
} else if (n->is_Phi()) {
|
||||
for (uint i = 1; i < n->req(); i++) {
|
||||
if (n->in(i) != NULL) {
|
||||
if (may_modify_helper(t_oop, n->in(i), phase)) {
|
||||
return true;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return false;
|
||||
}
|
||||
|
||||
// Does this array copy modify offsets between offset_lo and offset_hi
|
||||
// in the destination array
|
||||
// if must_modify is false, return true if the copy could write
|
||||
// between offset_lo and offset_hi
|
||||
// if must_modify is true, return true if the copy is guaranteed to
|
||||
// write between offset_lo and offset_hi
|
||||
bool ArrayCopyNode::modifies(intptr_t offset_lo, intptr_t offset_hi, PhaseTransform* phase, bool must_modify) {
|
||||
assert(_kind == ArrayCopy || _kind == CopyOf || _kind == CopyOfRange, "only for real array copies");
|
||||
|
||||
Node* dest = in(ArrayCopyNode::Dest);
|
||||
Node* src_pos = in(ArrayCopyNode::SrcPos);
|
||||
Node* dest_pos = in(ArrayCopyNode::DestPos);
|
||||
Node* len = in(ArrayCopyNode::Length);
|
||||
|
||||
const TypeInt *dest_pos_t = phase->type(dest_pos)->isa_int();
|
||||
const TypeInt *len_t = phase->type(len)->isa_int();
|
||||
const TypeAryPtr* ary_t = phase->type(dest)->isa_aryptr();
|
||||
|
||||
if (dest_pos_t != NULL && len_t != NULL && ary_t != NULL) {
|
||||
BasicType ary_elem = ary_t->klass()->as_array_klass()->element_type()->basic_type();
|
||||
uint header = arrayOopDesc::base_offset_in_bytes(ary_elem);
|
||||
uint elemsize = type2aelembytes(ary_elem);
|
||||
|
||||
intptr_t dest_pos_plus_len_lo = (((intptr_t)dest_pos_t->_lo) + len_t->_lo) * elemsize + header;
|
||||
intptr_t dest_pos_plus_len_hi = (((intptr_t)dest_pos_t->_hi) + len_t->_hi) * elemsize + header;
|
||||
intptr_t dest_pos_lo = ((intptr_t)dest_pos_t->_lo) * elemsize + header;
|
||||
intptr_t dest_pos_hi = ((intptr_t)dest_pos_t->_hi) * elemsize + header;
|
||||
|
||||
if (must_modify) {
|
||||
if (offset_lo >= dest_pos_hi && offset_hi < dest_pos_plus_len_lo) {
|
||||
return true;
|
||||
}
|
||||
} else {
|
||||
if (offset_hi >= dest_pos_lo && offset_lo < dest_pos_plus_len_hi) {
|
||||
return true;
|
||||
}
|
||||
}
|
||||
}
|
||||
return false;
|
||||
}
|
||||
|
@ -108,6 +108,7 @@ private:
|
||||
BasicType copy_type, const Type* value_type, int count);
|
||||
bool finish_transform(PhaseGVN *phase, bool can_reshape,
|
||||
Node* ctl, Node *mem);
|
||||
static bool may_modify_helper(const TypeOopPtr *t_oop, Node* n, PhaseTransform *phase);
|
||||
|
||||
public:
|
||||
|
||||
@ -162,6 +163,9 @@ public:
|
||||
|
||||
bool is_alloc_tightly_coupled() const { return _alloc_tightly_coupled; }
|
||||
|
||||
static bool may_modify(const TypeOopPtr *t_oop, MemBarNode* mb, PhaseTransform *phase);
|
||||
bool modifies(intptr_t offset_lo, intptr_t offset_hi, PhaseTransform* phase, bool must_modify);
|
||||
|
||||
#ifndef PRODUCT
|
||||
virtual void dump_spec(outputStream *st) const;
|
||||
virtual void dump_compact_spec(outputStream* st) const;
|
||||
|
@ -161,7 +161,7 @@ bool C2Compiler::is_intrinsic_supported(methodHandle method, bool is_virtual) {
|
||||
vmIntrinsics::ID id = method->intrinsic_id();
|
||||
assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
|
||||
|
||||
if (id < vmIntrinsics::FIRST_ID || id >= vmIntrinsics::LAST_COMPILER_INLINE) {
|
||||
if (id < vmIntrinsics::FIRST_ID || id > vmIntrinsics::LAST_COMPILER_INLINE) {
|
||||
return false;
|
||||
}
|
||||
|
||||
|
@ -742,7 +742,7 @@ uint CallNode::match_edge(uint idx) const {
|
||||
//
|
||||
bool CallNode::may_modify(const TypeOopPtr *t_oop, PhaseTransform *phase) {
|
||||
assert((t_oop != NULL), "sanity");
|
||||
if (is_call_to_arraycopystub()) {
|
||||
if (is_call_to_arraycopystub() && strcmp(_name, "unsafe_arraycopy") != 0) {
|
||||
const TypeTuple* args = _tf->domain();
|
||||
Node* dest = NULL;
|
||||
// Stubs that can be called once an ArrayCopyNode is expanded have
|
||||
|
@ -2730,7 +2730,22 @@ bool LibraryCallKit::inline_unsafe_load_store(BasicType type, LoadStoreKind kind
|
||||
load_store = _gvn.transform(new CompareAndSwapPNode(control(), mem, adr, newval, oldval));
|
||||
}
|
||||
}
|
||||
post_barrier(control(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
|
||||
if (kind == LS_cmpxchg) {
|
||||
// Emit the post barrier only when the actual store happened.
|
||||
// This makes sense to check only for compareAndSet that can fail to set the value.
|
||||
// CAS success path is marked more likely since we anticipate this is a performance
|
||||
// critical path, while CAS failure path can use the penalty for going through unlikely
|
||||
// path as backoff. Which is still better than doing a store barrier there.
|
||||
IdealKit ideal(this);
|
||||
ideal.if_then(load_store, BoolTest::ne, ideal.ConI(0), PROB_STATIC_FREQUENT); {
|
||||
sync_kit(ideal);
|
||||
post_barrier(ideal.ctrl(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
|
||||
ideal.sync_kit(this);
|
||||
} ideal.end_if();
|
||||
final_sync(ideal);
|
||||
} else {
|
||||
post_barrier(control(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
|
||||
}
|
||||
break;
|
||||
default:
|
||||
fatal(err_msg_res("unexpected type %d: %s", type, type2name(type)));
|
||||
|
@ -1175,7 +1175,7 @@ void CountedLoopEndNode::dump_spec(outputStream *st) const {
|
||||
//=============================================================================
|
||||
//------------------------------is_member--------------------------------------
|
||||
// Is 'l' a member of 'this'?
|
||||
int IdealLoopTree::is_member( const IdealLoopTree *l ) const {
|
||||
bool IdealLoopTree::is_member(const IdealLoopTree *l) const {
|
||||
while( l->_nest > _nest ) l = l->_parent;
|
||||
return l == this;
|
||||
}
|
||||
|
@ -384,7 +384,7 @@ public:
|
||||
{ }
|
||||
|
||||
// Is 'l' a member of 'this'?
|
||||
int is_member( const IdealLoopTree *l ) const; // Test for nested membership
|
||||
bool is_member(const IdealLoopTree *l) const; // Test for nested membership
|
||||
|
||||
// Set loop nesting depth. Accumulate has_call bits.
|
||||
int set_nest( uint depth );
|
||||
@ -1086,6 +1086,8 @@ private:
|
||||
bool split_up( Node *n, Node *blk1, Node *blk2 );
|
||||
void sink_use( Node *use, Node *post_loop );
|
||||
Node *place_near_use( Node *useblock ) const;
|
||||
Node* try_move_store_before_loop(Node* n, Node *n_ctrl);
|
||||
void try_move_store_after_loop(Node* n);
|
||||
|
||||
bool _created_loop_node;
|
||||
public:
|
||||
|
@ -653,6 +653,209 @@ Node *PhaseIdealLoop::conditional_move( Node *region ) {
|
||||
return iff->in(1);
|
||||
}
|
||||
|
||||
#ifdef ASSERT
|
||||
static void enqueue_cfg_uses(Node* m, Unique_Node_List& wq) {
|
||||
for (DUIterator_Fast imax, i = m->fast_outs(imax); i < imax; i++) {
|
||||
Node* u = m->fast_out(i);
|
||||
if (u->is_CFG()) {
|
||||
if (u->Opcode() == Op_NeverBranch) {
|
||||
u = ((NeverBranchNode*)u)->proj_out(0);
|
||||
enqueue_cfg_uses(u, wq);
|
||||
} else {
|
||||
wq.push(u);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
#endif
|
||||
|
||||
// Try moving a store out of a loop, right before the loop
|
||||
Node* PhaseIdealLoop::try_move_store_before_loop(Node* n, Node *n_ctrl) {
|
||||
// Store has to be first in the loop body
|
||||
IdealLoopTree *n_loop = get_loop(n_ctrl);
|
||||
if (n->is_Store() && n_loop != _ltree_root && n_loop->is_loop()) {
|
||||
assert(n->in(0), "store should have control set");
|
||||
Node* address = n->in(MemNode::Address);
|
||||
Node* value = n->in(MemNode::ValueIn);
|
||||
Node* mem = n->in(MemNode::Memory);
|
||||
IdealLoopTree* address_loop = get_loop(get_ctrl(address));
|
||||
IdealLoopTree* value_loop = get_loop(get_ctrl(value));
|
||||
|
||||
// - address and value must be loop invariant
|
||||
// - memory must be a memory Phi for the loop
|
||||
// - Store must be the only store on this memory slice in the
|
||||
// loop: if there's another store following this one then value
|
||||
// written at iteration i by the second store could be overwritten
|
||||
// at iteration i+n by the first store: it's not safe to move the
|
||||
// first store out of the loop
|
||||
// - nothing must observe the Phi memory: it guarantees no read
|
||||
// before the store and no early exit out of the loop
|
||||
// With those conditions, we are also guaranteed the store post
|
||||
// dominates the loop head. Otherwise there would be extra Phi
|
||||
// involved between the loop's Phi and the store.
|
||||
|
||||
if (!n_loop->is_member(address_loop) &&
|
||||
!n_loop->is_member(value_loop) &&
|
||||
mem->is_Phi() && mem->in(0) == n_loop->_head &&
|
||||
mem->outcnt() == 1 &&
|
||||
mem->in(LoopNode::LoopBackControl) == n) {
|
||||
|
||||
#ifdef ASSERT
|
||||
// Verify that store's control does post dominate loop entry and
|
||||
// that there's no early exit of the loop before the store.
|
||||
bool ctrl_ok = false;
|
||||
{
|
||||
// Follow control from loop head until n, we exit the loop or
|
||||
// we reach the tail
|
||||
ResourceMark rm;
|
||||
Unique_Node_List wq;
|
||||
wq.push(n_loop->_head);
|
||||
assert(n_loop->_tail != NULL, "need a tail");
|
||||
for (uint next = 0; next < wq.size(); ++next) {
|
||||
Node *m = wq.at(next);
|
||||
if (m == n->in(0)) {
|
||||
ctrl_ok = true;
|
||||
continue;
|
||||
}
|
||||
assert(!has_ctrl(m), "should be CFG");
|
||||
if (!n_loop->is_member(get_loop(m)) || m == n_loop->_tail) {
|
||||
ctrl_ok = false;
|
||||
break;
|
||||
}
|
||||
enqueue_cfg_uses(m, wq);
|
||||
}
|
||||
}
|
||||
assert(ctrl_ok, "bad control");
|
||||
#endif
|
||||
|
||||
// move the Store
|
||||
_igvn.replace_input_of(mem, LoopNode::LoopBackControl, mem);
|
||||
_igvn.replace_input_of(n, 0, n_loop->_head->in(LoopNode::EntryControl));
|
||||
_igvn.replace_input_of(n, MemNode::Memory, mem->in(LoopNode::EntryControl));
|
||||
// Disconnect the phi now. An empty phi can confuse other
|
||||
// optimizations in this pass of loop opts.
|
||||
_igvn.replace_node(mem, mem->in(LoopNode::EntryControl));
|
||||
n_loop->_body.yank(mem);
|
||||
|
||||
IdealLoopTree* new_loop = get_loop(n->in(0));
|
||||
set_ctrl_and_loop(n, n->in(0));
|
||||
|
||||
return n;
|
||||
}
|
||||
}
|
||||
return NULL;
|
||||
}
|
||||
|
||||
// Try moving a store out of a loop, right after the loop
|
||||
void PhaseIdealLoop::try_move_store_after_loop(Node* n) {
|
||||
if (n->is_Store()) {
|
||||
assert(n->in(0), "store should have control set");
|
||||
Node *n_ctrl = get_ctrl(n);
|
||||
IdealLoopTree *n_loop = get_loop(n_ctrl);
|
||||
// Store must be in a loop
|
||||
if (n_loop != _ltree_root && !n_loop->_irreducible) {
|
||||
Node* address = n->in(MemNode::Address);
|
||||
Node* value = n->in(MemNode::ValueIn);
|
||||
IdealLoopTree* address_loop = get_loop(get_ctrl(address));
|
||||
// address must be loop invariant
|
||||
if (!n_loop->is_member(address_loop)) {
|
||||
// Store must be last on this memory slice in the loop and
|
||||
// nothing in the loop must observe it
|
||||
Node* phi = NULL;
|
||||
for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
|
||||
Node* u = n->fast_out(i);
|
||||
if (has_ctrl(u)) { // control use?
|
||||
IdealLoopTree *u_loop = get_loop(get_ctrl(u));
|
||||
if (!n_loop->is_member(u_loop)) {
|
||||
continue;
|
||||
}
|
||||
if (u->is_Phi() && u->in(0) == n_loop->_head) {
|
||||
assert(_igvn.type(u) == Type::MEMORY, "bad phi");
|
||||
assert(phi == NULL, "already found");
|
||||
phi = u;
|
||||
continue;
|
||||
}
|
||||
}
|
||||
phi = NULL;
|
||||
break;
|
||||
}
|
||||
if (phi != NULL) {
|
||||
// Nothing in the loop before the store (next iteration)
|
||||
// must observe the stored value
|
||||
bool mem_ok = true;
|
||||
{
|
||||
ResourceMark rm;
|
||||
Unique_Node_List wq;
|
||||
wq.push(phi);
|
||||
for (uint next = 0; next < wq.size() && mem_ok; ++next) {
|
||||
Node *m = wq.at(next);
|
||||
for (DUIterator_Fast imax, i = m->fast_outs(imax); i < imax && mem_ok; i++) {
|
||||
Node* u = m->fast_out(i);
|
||||
if (u->is_Store() || u->is_Phi()) {
|
||||
if (u != n) {
|
||||
wq.push(u);
|
||||
mem_ok = (wq.size() <= 10);
|
||||
}
|
||||
} else {
|
||||
mem_ok = false;
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
if (mem_ok) {
|
||||
// Move the Store out of the loop creating clones along
|
||||
// all paths out of the loop that observe the stored value
|
||||
_igvn.rehash_node_delayed(phi);
|
||||
int count = phi->replace_edge(n, n->in(MemNode::Memory));
|
||||
assert(count > 0, "inconsistent phi");
|
||||
for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
|
||||
Node* u = n->fast_out(i);
|
||||
Node* c = get_ctrl(u);
|
||||
|
||||
if (u->is_Phi()) {
|
||||
c = u->in(0)->in(u->find_edge(n));
|
||||
}
|
||||
IdealLoopTree *u_loop = get_loop(c);
|
||||
assert (!n_loop->is_member(u_loop), "only the phi should have been a use in the loop");
|
||||
while(true) {
|
||||
Node* next_c = find_non_split_ctrl(idom(c));
|
||||
if (n_loop->is_member(get_loop(next_c))) {
|
||||
break;
|
||||
}
|
||||
c = next_c;
|
||||
}
|
||||
|
||||
Node* st = n->clone();
|
||||
st->set_req(0, c);
|
||||
_igvn.register_new_node_with_optimizer(st);
|
||||
|
||||
set_ctrl(st, c);
|
||||
IdealLoopTree* new_loop = get_loop(c);
|
||||
assert(new_loop != n_loop, "should be moved out of loop");
|
||||
if (new_loop->_child == NULL) new_loop->_body.push(st);
|
||||
|
||||
_igvn.replace_input_of(u, u->find_edge(n), st);
|
||||
--imax;
|
||||
--i;
|
||||
}
|
||||
|
||||
|
||||
assert(n->outcnt() == 0, "all uses should be gone");
|
||||
_igvn.replace_input_of(n, MemNode::Memory, C->top());
|
||||
// Disconnect the phi now. An empty phi can confuse other
|
||||
// optimizations in this pass of loop opts..
|
||||
if (phi->in(LoopNode::LoopBackControl) == phi) {
|
||||
_igvn.replace_node(phi, phi->in(LoopNode::EntryControl));
|
||||
n_loop->_body.yank(phi);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
//------------------------------split_if_with_blocks_pre-----------------------
|
||||
// Do the real work in a non-recursive function. Data nodes want to be
|
||||
// cloned in the pre-order so they can feed each other nicely.
|
||||
@ -683,6 +886,11 @@ Node *PhaseIdealLoop::split_if_with_blocks_pre( Node *n ) {
|
||||
Node *n_ctrl = get_ctrl(n);
|
||||
if( !n_ctrl ) return n; // Dead node
|
||||
|
||||
Node* res = try_move_store_before_loop(n, n_ctrl);
|
||||
if (res != NULL) {
|
||||
return n;
|
||||
}
|
||||
|
||||
// Attempt to remix address expressions for loop invariants
|
||||
Node *m = remix_address_expressions( n );
|
||||
if( m ) return m;
|
||||
@ -691,16 +899,18 @@ Node *PhaseIdealLoop::split_if_with_blocks_pre( Node *n ) {
|
||||
// Returns the block to clone thru.
|
||||
Node *n_blk = has_local_phi_input( n );
|
||||
if( !n_blk ) return n;
|
||||
|
||||
// Do not clone the trip counter through on a CountedLoop
|
||||
// (messes up the canonical shape).
|
||||
if( n_blk->is_CountedLoop() && n->Opcode() == Op_AddI ) return n;
|
||||
|
||||
// Check for having no control input; not pinned. Allow
|
||||
// dominating control.
|
||||
if( n->in(0) ) {
|
||||
if (n->in(0)) {
|
||||
Node *dom = idom(n_blk);
|
||||
if( dom_lca( n->in(0), dom ) != n->in(0) )
|
||||
if (dom_lca(n->in(0), dom) != n->in(0)) {
|
||||
return n;
|
||||
}
|
||||
}
|
||||
// Policy: when is it profitable. You must get more wins than
|
||||
// policy before it is considered profitable. Policy is usually 0,
|
||||
@ -1029,6 +1239,8 @@ void PhaseIdealLoop::split_if_with_blocks_post( Node *n ) {
|
||||
}
|
||||
}
|
||||
|
||||
try_move_store_after_loop(n);
|
||||
|
||||
// Check for Opaque2's who's loop has disappeared - who's input is in the
|
||||
// same loop nest as their output. Remove 'em, they are no longer useful.
|
||||
if( n_op == Op_Opaque2 &&
|
||||
|
@ -324,18 +324,28 @@ static Node *scan_mem_chain(Node *mem, int alias_idx, int offset, Node *start_me
|
||||
return in;
|
||||
} else if (in->is_Call()) {
|
||||
CallNode *call = in->as_Call();
|
||||
if (!call->may_modify(tinst, phase)) {
|
||||
mem = call->in(TypeFunc::Memory);
|
||||
if (call->may_modify(tinst, phase)) {
|
||||
assert(call->is_ArrayCopy(), "ArrayCopy is the only call node that doesn't make allocation escape");
|
||||
|
||||
if (call->as_ArrayCopy()->modifies(offset, offset, phase, false)) {
|
||||
return in;
|
||||
}
|
||||
}
|
||||
mem = in->in(TypeFunc::Memory);
|
||||
} else if (in->is_MemBar()) {
|
||||
if (ArrayCopyNode::may_modify(tinst, in->as_MemBar(), phase)) {
|
||||
assert(in->in(0)->is_Proj() && in->in(0)->in(0)->is_ArrayCopy(), "should be arraycopy");
|
||||
ArrayCopyNode* ac = in->in(0)->in(0)->as_ArrayCopy();
|
||||
assert(ac->is_clonebasic(), "Only basic clone is a non escaping clone");
|
||||
return ac;
|
||||
}
|
||||
mem = in->in(TypeFunc::Memory);
|
||||
} else {
|
||||
assert(false, "unexpected projection");
|
||||
}
|
||||
} else if (mem->is_Store()) {
|
||||
const TypePtr* atype = mem->as_Store()->adr_type();
|
||||
int adr_idx = Compile::current()->get_alias_index(atype);
|
||||
int adr_idx = phase->C->get_alias_index(atype);
|
||||
if (adr_idx == alias_idx) {
|
||||
assert(atype->isa_oopptr(), "address type must be oopptr");
|
||||
int adr_offset = atype->offset();
|
||||
@ -373,7 +383,7 @@ static Node *scan_mem_chain(Node *mem, int alias_idx, int offset, Node *start_me
|
||||
adr = mem->in(3); // Destination array
|
||||
}
|
||||
const TypePtr* atype = adr->bottom_type()->is_ptr();
|
||||
int adr_idx = Compile::current()->get_alias_index(atype);
|
||||
int adr_idx = phase->C->get_alias_index(atype);
|
||||
if (adr_idx == alias_idx) {
|
||||
assert(false, "Object is not scalar replaceable if a LoadStore node access its field");
|
||||
return NULL;
|
||||
@ -386,12 +396,63 @@ static Node *scan_mem_chain(Node *mem, int alias_idx, int offset, Node *start_me
|
||||
}
|
||||
}
|
||||
|
||||
// Generate loads from source of the arraycopy for fields of
|
||||
// destination needed at a deoptimization point
|
||||
Node* PhaseMacroExpand::make_arraycopy_load(ArrayCopyNode* ac, intptr_t offset, Node* ctl, BasicType ft, const Type *ftype, AllocateNode *alloc) {
|
||||
BasicType bt = ft;
|
||||
const Type *type = ftype;
|
||||
if (ft == T_NARROWOOP) {
|
||||
bt = T_OBJECT;
|
||||
type = ftype->make_oopptr();
|
||||
}
|
||||
Node* res = NULL;
|
||||
if (ac->is_clonebasic()) {
|
||||
Node* base = ac->in(ArrayCopyNode::Src)->in(AddPNode::Base);
|
||||
Node* adr = _igvn.transform(new AddPNode(base, base, MakeConX(offset)));
|
||||
const TypePtr* adr_type = _igvn.type(base)->is_ptr()->add_offset(offset);
|
||||
Node* m = ac->in(TypeFunc::Memory);
|
||||
while (m->is_MergeMem()) {
|
||||
m = m->as_MergeMem()->memory_at(C->get_alias_index(adr_type));
|
||||
if (m->is_Proj() && m->in(0)->is_MemBar()) {
|
||||
m = m->in(0)->in(TypeFunc::Memory);
|
||||
}
|
||||
}
|
||||
res = LoadNode::make(_igvn, ctl, m, adr, adr_type, type, bt, MemNode::unordered, LoadNode::Pinned);
|
||||
} else {
|
||||
if (ac->modifies(offset, offset, &_igvn, true)) {
|
||||
assert(ac->in(ArrayCopyNode::Dest) == alloc->result_cast(), "arraycopy destination should be allocation's result");
|
||||
uint shift = exact_log2(type2aelembytes(bt));
|
||||
Node* diff = _igvn.transform(new SubINode(ac->in(ArrayCopyNode::SrcPos), ac->in(ArrayCopyNode::DestPos)));
|
||||
#ifdef _LP64
|
||||
diff = _igvn.transform(new ConvI2LNode(diff));
|
||||
#endif
|
||||
diff = _igvn.transform(new LShiftXNode(diff, intcon(shift)));
|
||||
|
||||
Node* off = _igvn.transform(new AddXNode(MakeConX(offset), diff));
|
||||
Node* base = ac->in(ArrayCopyNode::Src);
|
||||
Node* adr = _igvn.transform(new AddPNode(base, base, off));
|
||||
const TypePtr* adr_type = _igvn.type(base)->is_ptr()->add_offset(offset);
|
||||
Node* m = ac->in(TypeFunc::Memory);
|
||||
res = LoadNode::make(_igvn, ctl, m, adr, adr_type, type, bt, MemNode::unordered, LoadNode::Pinned);
|
||||
}
|
||||
}
|
||||
if (res != NULL) {
|
||||
res = _igvn.transform(res);
|
||||
if (ftype->isa_narrowoop()) {
|
||||
// PhaseMacroExpand::scalar_replacement adds DecodeN nodes
|
||||
res = _igvn.transform(new EncodePNode(res, ftype));
|
||||
}
|
||||
return res;
|
||||
}
|
||||
return NULL;
|
||||
}
|
||||
|
||||
//
|
||||
// Given a Memory Phi, compute a value Phi containing the values from stores
|
||||
// on the input paths.
|
||||
// Note: this function is recursive, its depth is limied by the "level" argument
|
||||
// Note: this function is recursive, its depth is limited by the "level" argument
|
||||
// Returns the computed Phi, or NULL if it cannot compute it.
|
||||
Node *PhaseMacroExpand::value_from_mem_phi(Node *mem, BasicType ft, const Type *phi_type, const TypeOopPtr *adr_t, Node *alloc, Node_Stack *value_phis, int level) {
|
||||
Node *PhaseMacroExpand::value_from_mem_phi(Node *mem, BasicType ft, const Type *phi_type, const TypeOopPtr *adr_t, AllocateNode *alloc, Node_Stack *value_phis, int level) {
|
||||
assert(mem->is_Phi(), "sanity");
|
||||
int alias_idx = C->get_alias_index(adr_t);
|
||||
int offset = adr_t->offset();
|
||||
@ -458,6 +519,9 @@ Node *PhaseMacroExpand::value_from_mem_phi(Node *mem, BasicType ft, const Type *
|
||||
assert(val->in(0)->is_LoadStore() || val->in(0)->Opcode() == Op_EncodeISOArray, "sanity");
|
||||
assert(false, "Object is not scalar replaceable if a LoadStore node access its field");
|
||||
return NULL;
|
||||
} else if (val->is_ArrayCopy()) {
|
||||
Node* res = make_arraycopy_load(val->as_ArrayCopy(), offset, val->in(0), ft, phi_type, alloc);
|
||||
values.at_put(j, res);
|
||||
} else {
|
||||
#ifdef ASSERT
|
||||
val->dump();
|
||||
@ -479,7 +543,7 @@ Node *PhaseMacroExpand::value_from_mem_phi(Node *mem, BasicType ft, const Type *
|
||||
}
|
||||
|
||||
// Search the last value stored into the object's field.
|
||||
Node *PhaseMacroExpand::value_from_mem(Node *sfpt_mem, BasicType ft, const Type *ftype, const TypeOopPtr *adr_t, Node *alloc) {
|
||||
Node *PhaseMacroExpand::value_from_mem(Node *sfpt_mem, Node *sfpt_ctl, BasicType ft, const Type *ftype, const TypeOopPtr *adr_t, AllocateNode *alloc) {
|
||||
assert(adr_t->is_known_instance_field(), "instance required");
|
||||
int instance_id = adr_t->instance_id();
|
||||
assert((uint)instance_id == alloc->_idx, "wrong allocation");
|
||||
@ -538,6 +602,8 @@ Node *PhaseMacroExpand::value_from_mem(Node *sfpt_mem, BasicType ft, const Type
|
||||
} else {
|
||||
done = true;
|
||||
}
|
||||
} else if (mem->is_ArrayCopy()) {
|
||||
done = true;
|
||||
} else {
|
||||
assert(false, "unexpected node");
|
||||
}
|
||||
@ -562,6 +628,13 @@ Node *PhaseMacroExpand::value_from_mem(Node *sfpt_mem, BasicType ft, const Type
|
||||
value_phis.pop();
|
||||
}
|
||||
}
|
||||
} else if (mem->is_ArrayCopy()) {
|
||||
Node* ctl = mem->in(0);
|
||||
if (sfpt_ctl->is_Proj() && sfpt_ctl->as_Proj()->is_uncommon_trap_proj(Deoptimization::Reason_none)) {
|
||||
// pin the loads in the uncommon trap path
|
||||
ctl = sfpt_ctl;
|
||||
}
|
||||
return make_arraycopy_load(mem->as_ArrayCopy(), offset, ctl, ft, ftype, alloc);
|
||||
}
|
||||
}
|
||||
// Something go wrong.
|
||||
@ -738,6 +811,7 @@ bool PhaseMacroExpand::scalar_replacement(AllocateNode *alloc, GrowableArray <Sa
|
||||
while (safepoints.length() > 0) {
|
||||
SafePointNode* sfpt = safepoints.pop();
|
||||
Node* mem = sfpt->memory();
|
||||
Node* ctl = sfpt->control();
|
||||
assert(sfpt->jvms() != NULL, "missed JVMS");
|
||||
// Fields of scalar objs are referenced only at the end
|
||||
// of regular debuginfo at the last (youngest) JVMS.
|
||||
@ -789,7 +863,7 @@ bool PhaseMacroExpand::scalar_replacement(AllocateNode *alloc, GrowableArray <Sa
|
||||
|
||||
const TypeOopPtr *field_addr_type = res_type->add_offset(offset)->isa_oopptr();
|
||||
|
||||
Node *field_val = value_from_mem(mem, basic_elem_type, field_type, field_addr_type, alloc);
|
||||
Node *field_val = value_from_mem(mem, ctl, basic_elem_type, field_type, field_addr_type, alloc);
|
||||
if (field_val == NULL) {
|
||||
// We weren't able to find a value for this field,
|
||||
// give up on eliminating this allocation.
|
||||
|
@ -85,8 +85,8 @@ private:
|
||||
Node* length,
|
||||
const TypeFunc* slow_call_type,
|
||||
address slow_call_address);
|
||||
Node *value_from_mem(Node *mem, BasicType ft, const Type *ftype, const TypeOopPtr *adr_t, Node *alloc);
|
||||
Node *value_from_mem_phi(Node *mem, BasicType ft, const Type *ftype, const TypeOopPtr *adr_t, Node *alloc, Node_Stack *value_phis, int level);
|
||||
Node *value_from_mem(Node *mem, Node *ctl, BasicType ft, const Type *ftype, const TypeOopPtr *adr_t, AllocateNode *alloc);
|
||||
Node *value_from_mem_phi(Node *mem, BasicType ft, const Type *ftype, const TypeOopPtr *adr_t, AllocateNode *alloc, Node_Stack *value_phis, int level);
|
||||
|
||||
bool eliminate_boxing_node(CallStaticJavaNode *boxing);
|
||||
bool eliminate_allocate_node(AllocateNode *alloc);
|
||||
@ -200,6 +200,8 @@ private:
|
||||
Node* old_eden_top, Node* new_eden_top,
|
||||
Node* length);
|
||||
|
||||
Node* make_arraycopy_load(ArrayCopyNode* ac, intptr_t offset, Node* ctl, BasicType ft, const Type *ftype, AllocateNode *alloc);
|
||||
|
||||
public:
|
||||
PhaseMacroExpand(PhaseIterGVN &igvn) : Phase(Macro_Expand), _igvn(igvn), _has_locks(false) {
|
||||
_igvn.set_delay_transform(true);
|
||||
|
@ -108,37 +108,6 @@ extern void print_alias_types();
|
||||
|
||||
#endif
|
||||
|
||||
static bool membar_for_arraycopy_helper(const TypeOopPtr *t_oop, Node* n, PhaseTransform *phase) {
|
||||
if (n->is_Proj()) {
|
||||
n = n->in(0);
|
||||
if (n->is_Call() && n->as_Call()->may_modify(t_oop, phase)) {
|
||||
return true;
|
||||
}
|
||||
}
|
||||
return false;
|
||||
}
|
||||
|
||||
static bool membar_for_arraycopy(const TypeOopPtr *t_oop, MemBarNode* mb, PhaseTransform *phase) {
|
||||
Node* mem = mb->in(TypeFunc::Memory);
|
||||
|
||||
if (mem->is_MergeMem()) {
|
||||
Node* n = mem->as_MergeMem()->memory_at(Compile::AliasIdxRaw);
|
||||
if (membar_for_arraycopy_helper(t_oop, n, phase)) {
|
||||
return true;
|
||||
} else if (n->is_Phi()) {
|
||||
for (uint i = 1; i < n->req(); i++) {
|
||||
if (n->in(i) != NULL) {
|
||||
if (membar_for_arraycopy_helper(t_oop, n->in(i), phase)) {
|
||||
return true;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return false;
|
||||
}
|
||||
|
||||
Node *MemNode::optimize_simple_memory_chain(Node *mchain, const TypeOopPtr *t_oop, Node *load, PhaseGVN *phase) {
|
||||
assert((t_oop != NULL), "sanity");
|
||||
bool is_instance = t_oop->is_known_instance_field();
|
||||
@ -183,7 +152,7 @@ Node *MemNode::optimize_simple_memory_chain(Node *mchain, const TypeOopPtr *t_oo
|
||||
}
|
||||
}
|
||||
} else if (proj_in->is_MemBar()) {
|
||||
if (membar_for_arraycopy(t_oop, proj_in->as_MemBar(), phase)) {
|
||||
if (ArrayCopyNode::may_modify(t_oop, proj_in->as_MemBar(), phase)) {
|
||||
break;
|
||||
}
|
||||
result = proj_in->in(TypeFunc::Memory);
|
||||
@ -545,35 +514,12 @@ Node* LoadNode::find_previous_arraycopy(PhaseTransform* phase, Node* ld_alloc, N
|
||||
Node* dest = ac->in(ArrayCopyNode::Dest);
|
||||
|
||||
if (dest == ld_base) {
|
||||
Node* src_pos = ac->in(ArrayCopyNode::SrcPos);
|
||||
Node* dest_pos = ac->in(ArrayCopyNode::DestPos);
|
||||
Node* len = ac->in(ArrayCopyNode::Length);
|
||||
|
||||
const TypeInt *dest_pos_t = phase->type(dest_pos)->isa_int();
|
||||
const TypeX *ld_offs_t = phase->type(ld_offs)->isa_intptr_t();
|
||||
const TypeInt *len_t = phase->type(len)->isa_int();
|
||||
const TypeAryPtr* ary_t = phase->type(dest)->isa_aryptr();
|
||||
|
||||
if (dest_pos_t != NULL && ld_offs_t != NULL && len_t != NULL && ary_t != NULL) {
|
||||
BasicType ary_elem = ary_t->klass()->as_array_klass()->element_type()->basic_type();
|
||||
uint header = arrayOopDesc::base_offset_in_bytes(ary_elem);
|
||||
uint elemsize = type2aelembytes(ary_elem);
|
||||
|
||||
intptr_t dest_pos_plus_len_lo = (((intptr_t)dest_pos_t->_lo) + len_t->_lo) * elemsize + header;
|
||||
intptr_t dest_pos_plus_len_hi = (((intptr_t)dest_pos_t->_hi) + len_t->_hi) * elemsize + header;
|
||||
intptr_t dest_pos_lo = ((intptr_t)dest_pos_t->_lo) * elemsize + header;
|
||||
intptr_t dest_pos_hi = ((intptr_t)dest_pos_t->_hi) * elemsize + header;
|
||||
|
||||
if (can_see_stored_value) {
|
||||
if (ld_offs_t->_lo >= dest_pos_hi && ld_offs_t->_hi < dest_pos_plus_len_lo) {
|
||||
return ac;
|
||||
}
|
||||
} else {
|
||||
if (ld_offs_t->_hi < dest_pos_lo || ld_offs_t->_lo >= dest_pos_plus_len_hi) {
|
||||
mem = ac->in(TypeFunc::Memory);
|
||||
}
|
||||
return ac;
|
||||
}
|
||||
if (ac->modifies(ld_offs_t->_lo, ld_offs_t->_hi, phase, can_see_stored_value)) {
|
||||
return ac;
|
||||
}
|
||||
if (!can_see_stored_value) {
|
||||
mem = ac->in(TypeFunc::Memory);
|
||||
}
|
||||
}
|
||||
}
|
||||
@ -703,7 +649,7 @@ Node* MemNode::find_previous_store(PhaseTransform* phase) {
|
||||
continue; // (a) advance through independent call memory
|
||||
}
|
||||
} else if (mem->is_Proj() && mem->in(0)->is_MemBar()) {
|
||||
if (membar_for_arraycopy(addr_t, mem->in(0)->as_MemBar(), phase)) {
|
||||
if (ArrayCopyNode::may_modify(addr_t, mem->in(0)->as_MemBar(), phase)) {
|
||||
break;
|
||||
}
|
||||
mem = mem->in(0)->in(TypeFunc::Memory);
|
||||
@ -883,18 +829,17 @@ static bool skip_through_membars(Compile::AliasType* atp, const TypeInstPtr* tp,
|
||||
// Is the value loaded previously stored by an arraycopy? If so return
|
||||
// a load node that reads from the source array so we may be able to
|
||||
// optimize out the ArrayCopy node later.
|
||||
Node* MemNode::can_see_arraycopy_value(Node* st, PhaseTransform* phase) const {
|
||||
Node* LoadNode::can_see_arraycopy_value(Node* st, PhaseTransform* phase) const {
|
||||
Node* ld_adr = in(MemNode::Address);
|
||||
intptr_t ld_off = 0;
|
||||
AllocateNode* ld_alloc = AllocateNode::Ideal_allocation(ld_adr, phase, ld_off);
|
||||
Node* ac = find_previous_arraycopy(phase, ld_alloc, st, true);
|
||||
if (ac != NULL) {
|
||||
assert(ac->is_ArrayCopy(), "what kind of node can this be?");
|
||||
assert(is_Load(), "only for loads");
|
||||
|
||||
Node* ld = clone();
|
||||
if (ac->as_ArrayCopy()->is_clonebasic()) {
|
||||
assert(ld_alloc != NULL, "need an alloc");
|
||||
Node* ld = clone();
|
||||
Node* addp = in(MemNode::Address)->clone();
|
||||
assert(addp->is_AddP(), "address must be addp");
|
||||
assert(addp->in(AddPNode::Base) == ac->in(ArrayCopyNode::Dest)->in(AddPNode::Base), "strange pattern");
|
||||
@ -906,9 +851,7 @@ Node* MemNode::can_see_arraycopy_value(Node* st, PhaseTransform* phase) const {
|
||||
assert(ld_alloc->in(0) != NULL, "alloc must have control");
|
||||
ld->set_req(0, ld_alloc->in(0));
|
||||
}
|
||||
return ld;
|
||||
} else {
|
||||
Node* ld = clone();
|
||||
Node* addp = in(MemNode::Address)->clone();
|
||||
assert(addp->in(AddPNode::Base) == addp->in(AddPNode::Address), "should be");
|
||||
addp->set_req(AddPNode::Base, ac->in(ArrayCopyNode::Src));
|
||||
@ -933,8 +876,10 @@ Node* MemNode::can_see_arraycopy_value(Node* st, PhaseTransform* phase) const {
|
||||
assert(ac->in(0) != NULL, "alloc must have control");
|
||||
ld->set_req(0, ac->in(0));
|
||||
}
|
||||
return ld;
|
||||
}
|
||||
// load depends on the tests that validate the arraycopy
|
||||
ld->as_Load()->_depends_only_on_test = Pinned;
|
||||
return ld;
|
||||
}
|
||||
return NULL;
|
||||
}
|
||||
@ -2426,40 +2371,47 @@ Node *StoreNode::Ideal(PhaseGVN *phase, bool can_reshape) {
|
||||
|
||||
Node* mem = in(MemNode::Memory);
|
||||
Node* address = in(MemNode::Address);
|
||||
|
||||
// Back-to-back stores to same address? Fold em up. Generally
|
||||
// unsafe if I have intervening uses... Also disallowed for StoreCM
|
||||
// since they must follow each StoreP operation. Redundant StoreCMs
|
||||
// are eliminated just before matching in final_graph_reshape.
|
||||
if (mem->is_Store() && mem->in(MemNode::Address)->eqv_uncast(address) &&
|
||||
mem->Opcode() != Op_StoreCM) {
|
||||
// Looking at a dead closed cycle of memory?
|
||||
assert(mem != mem->in(MemNode::Memory), "dead loop in StoreNode::Ideal");
|
||||
{
|
||||
Node* st = mem;
|
||||
// If Store 'st' has more than one use, we cannot fold 'st' away.
|
||||
// For example, 'st' might be the final state at a conditional
|
||||
// return. Or, 'st' might be used by some node which is live at
|
||||
// the same time 'st' is live, which might be unschedulable. So,
|
||||
// require exactly ONE user until such time as we clone 'mem' for
|
||||
// each of 'mem's uses (thus making the exactly-1-user-rule hold
|
||||
// true).
|
||||
while (st->is_Store() && st->outcnt() == 1 && st->Opcode() != Op_StoreCM) {
|
||||
// Looking at a dead closed cycle of memory?
|
||||
assert(st != st->in(MemNode::Memory), "dead loop in StoreNode::Ideal");
|
||||
assert(Opcode() == st->Opcode() ||
|
||||
st->Opcode() == Op_StoreVector ||
|
||||
Opcode() == Op_StoreVector ||
|
||||
phase->C->get_alias_index(adr_type()) == Compile::AliasIdxRaw ||
|
||||
(Opcode() == Op_StoreL && st->Opcode() == Op_StoreI), // expanded ClearArrayNode
|
||||
err_msg_res("no mismatched stores, except on raw memory: %s %s", NodeClassNames[Opcode()], NodeClassNames[st->Opcode()]));
|
||||
|
||||
assert(Opcode() == mem->Opcode() ||
|
||||
phase->C->get_alias_index(adr_type()) == Compile::AliasIdxRaw,
|
||||
"no mismatched stores, except on raw memory");
|
||||
|
||||
if (mem->outcnt() == 1 && // check for intervening uses
|
||||
mem->as_Store()->memory_size() <= this->memory_size()) {
|
||||
// If anybody other than 'this' uses 'mem', we cannot fold 'mem' away.
|
||||
// For example, 'mem' might be the final state at a conditional return.
|
||||
// Or, 'mem' might be used by some node which is live at the same time
|
||||
// 'this' is live, which might be unschedulable. So, require exactly
|
||||
// ONE user, the 'this' store, until such time as we clone 'mem' for
|
||||
// each of 'mem's uses (thus making the exactly-1-user-rule hold true).
|
||||
if (can_reshape) { // (%%% is this an anachronism?)
|
||||
set_req_X(MemNode::Memory, mem->in(MemNode::Memory),
|
||||
phase->is_IterGVN());
|
||||
} else {
|
||||
// It's OK to do this in the parser, since DU info is always accurate,
|
||||
// and the parser always refers to nodes via SafePointNode maps.
|
||||
set_req(MemNode::Memory, mem->in(MemNode::Memory));
|
||||
if (st->in(MemNode::Address)->eqv_uncast(address) &&
|
||||
st->as_Store()->memory_size() <= this->memory_size()) {
|
||||
Node* use = st->raw_out(0);
|
||||
phase->igvn_rehash_node_delayed(use);
|
||||
if (can_reshape) {
|
||||
use->set_req_X(MemNode::Memory, st->in(MemNode::Memory), phase->is_IterGVN());
|
||||
} else {
|
||||
// It's OK to do this in the parser, since DU info is always accurate,
|
||||
// and the parser always refers to nodes via SafePointNode maps.
|
||||
use->set_req(MemNode::Memory, st->in(MemNode::Memory));
|
||||
}
|
||||
return this;
|
||||
}
|
||||
return this;
|
||||
st = st->in(MemNode::Memory);
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
// Capture an unaliased, unconditional, simple store into an initializer.
|
||||
// Or, if it is independent of the allocation, hoist it above the allocation.
|
||||
if (ReduceFieldZeroing && /*can_reshape &&*/
|
||||
|
@ -126,7 +126,6 @@ public:
|
||||
// Can this node (load or store) accurately see a stored value in
|
||||
// the given memory state? (The state may or may not be in(Memory).)
|
||||
Node* can_see_stored_value(Node* st, PhaseTransform* phase) const;
|
||||
Node* can_see_arraycopy_value(Node* st, PhaseTransform* phase) const;
|
||||
|
||||
#ifndef PRODUCT
|
||||
static void dump_adr_type(const Node* mem, const TypePtr* adr_type, outputStream *st);
|
||||
@ -252,6 +251,9 @@ public:
|
||||
protected:
|
||||
const Type* load_array_final_field(const TypeKlassPtr *tkls,
|
||||
ciKlass* klass) const;
|
||||
|
||||
Node* can_see_arraycopy_value(Node* st, PhaseTransform* phase) const;
|
||||
|
||||
// depends_only_on_test is almost always true, and needs to be almost always
|
||||
// true to enable key hoisting & commoning optimizations. However, for the
|
||||
// special case of RawPtr loads from TLS top & end, and other loads performed by
|
||||
|
@ -1565,6 +1565,7 @@ typedef CompactHashtable<Symbol*, char> SymbolCompactHashTable;
|
||||
declare_toplevel_type(Generation*) \
|
||||
declare_toplevel_type(GenerationSpec**) \
|
||||
declare_toplevel_type(HeapWord*) \
|
||||
declare_toplevel_type(HeapWord* volatile) \
|
||||
declare_toplevel_type(MemRegion*) \
|
||||
declare_toplevel_type(OffsetTableContigSpace*) \
|
||||
declare_toplevel_type(Space*) \
|
||||
|
@ -0,0 +1,204 @@
|
||||
/*
|
||||
* Copyright (c) 2015, Oracle and/or its affiliates. All rights reserved.
|
||||
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
|
||||
*
|
||||
* This code is free software; you can redistribute it and/or modify it
|
||||
* under the terms of the GNU General Public License version 2 only, as
|
||||
* published by the Free Software Foundation.
|
||||
*
|
||||
* This code is distributed in the hope that it will be useful, but WITHOUT
|
||||
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
|
||||
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
||||
* version 2 for more details (a copy is included in the LICENSE file that
|
||||
* accompanied this code).
|
||||
*
|
||||
* You should have received a copy of the GNU General Public License version
|
||||
* 2 along with this work; if not, write to the Free Software Foundation,
|
||||
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
|
||||
*
|
||||
* Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
|
||||
* or visit www.oracle.com if you need additional information or have any
|
||||
* questions.
|
||||
*/
|
||||
|
||||
/*
|
||||
* @test
|
||||
* @bug 8130847
|
||||
* @summary Eliminated instance/array written to by an array copy variant must be correctly initialized when reallocated at a deopt
|
||||
* @run main/othervm -XX:-BackgroundCompilation -XX:-UseOnStackReplacement TestEliminatedArrayCopyDeopt
|
||||
*
|
||||
*/
|
||||
|
||||
// Test that if an ArrayCopy node is eliminated because it doesn't
|
||||
// escape, then the correct field/array element values are captured so
|
||||
// on a deoptimization, when the object/array is reallocated, it is
|
||||
// correctly initialized
|
||||
|
||||
public class TestEliminatedArrayCopyDeopt {
|
||||
|
||||
static class A implements Cloneable {
|
||||
int f0;
|
||||
int f1;
|
||||
int f2;
|
||||
int f3;
|
||||
int f4;
|
||||
int f5;
|
||||
int f6;
|
||||
int f7;
|
||||
int f8;
|
||||
int f9;
|
||||
int f10;
|
||||
int f11;
|
||||
int f12;
|
||||
int f13;
|
||||
int f14;
|
||||
int f15;
|
||||
|
||||
public Object clone() throws CloneNotSupportedException {
|
||||
return super.clone();
|
||||
}
|
||||
}
|
||||
|
||||
// Clone
|
||||
static boolean m1(A a, boolean flag) throws CloneNotSupportedException {
|
||||
A c = (A)a.clone();
|
||||
if (flag) {
|
||||
// never taken branch that causes the deoptimization
|
||||
if (c.f0 != 0x42) {
|
||||
return false;
|
||||
}
|
||||
}
|
||||
return true;
|
||||
}
|
||||
|
||||
// Array clone
|
||||
static int[] m2_src = null;
|
||||
static boolean m2(boolean flag) throws CloneNotSupportedException {
|
||||
int[] src = new int[10];
|
||||
m2_src = src;
|
||||
for (int i = 0; i < src.length; i++) {
|
||||
src[i] = 0x42+i;
|
||||
}
|
||||
int[] c = (int[])src.clone();
|
||||
if (flag) {
|
||||
for (int i = 0; i < c.length; i++) {
|
||||
if (c[i] != src[i]) {
|
||||
return false;
|
||||
}
|
||||
}
|
||||
}
|
||||
return true;
|
||||
}
|
||||
|
||||
// Array copy
|
||||
static boolean m3(int[] src, boolean flag) {
|
||||
int[] dst = new int[10];
|
||||
System.arraycopy(src, 0, dst, 0, 10);
|
||||
if (flag) {
|
||||
for (int i = 0; i < dst.length; i++) {
|
||||
if (dst[i] != src[i]) {
|
||||
return false;
|
||||
}
|
||||
}
|
||||
}
|
||||
return true;
|
||||
}
|
||||
|
||||
// Array copy of subrange
|
||||
static boolean m4(int[] src, boolean flag) {
|
||||
int[] dst = new int[10];
|
||||
dst[0] = 0x42;
|
||||
dst[1] = 0x42 - 1;
|
||||
dst[2] = 0x42 - 2;
|
||||
dst[8] = 0x42 - 8;
|
||||
dst[9] = 0x42 - 9;
|
||||
int src_off = 2;
|
||||
int dst_off = 3;
|
||||
int len = 5;
|
||||
System.arraycopy(src, src_off, dst, dst_off, len);
|
||||
if (flag) {
|
||||
for (int i = 0; i < dst.length; i++) {
|
||||
if (i >= dst_off && i < dst_off + len) {
|
||||
if (dst[i] != src[i - dst_off + src_off]) {
|
||||
return false;
|
||||
}
|
||||
} else {
|
||||
if (dst[i] != 0x42-i) {
|
||||
return false;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
return true;
|
||||
}
|
||||
|
||||
// Array copy with Phi
|
||||
static boolean m5(int[] src, boolean flag1, boolean flag2) {
|
||||
int[] dst = new int[10];
|
||||
if (flag1) {
|
||||
System.arraycopy(src, 0, dst, 0, 10);
|
||||
}
|
||||
if (flag2) {
|
||||
for (int i = 0; i < dst.length; i++) {
|
||||
if (dst[i] != src[i]) {
|
||||
return false;
|
||||
}
|
||||
}
|
||||
}
|
||||
return true;
|
||||
}
|
||||
|
||||
static public void main(String[] args) throws Exception {
|
||||
boolean success = true;
|
||||
A a = new A();
|
||||
a.f0 = 0x42;
|
||||
for (int i = 0; i < 20000; i++) {
|
||||
m1(a, false);
|
||||
}
|
||||
if (!m1(a, true)) {
|
||||
System.out.println("m1 failed");
|
||||
success = false;
|
||||
}
|
||||
|
||||
for (int i = 0; i < 20000; i++) {
|
||||
m2(false);
|
||||
}
|
||||
if (!m2(true)) {
|
||||
System.out.println("m2 failed");
|
||||
success = false;
|
||||
}
|
||||
|
||||
int[] src = new int[10];
|
||||
for (int i = 0; i < src.length; i++) {
|
||||
src[i] = 0x42+i;
|
||||
}
|
||||
|
||||
for (int i = 0; i < 20000; i++) {
|
||||
m3(src, false);
|
||||
}
|
||||
if (!m3(src, true)) {
|
||||
System.out.println("m3 failed");
|
||||
success = false;
|
||||
}
|
||||
|
||||
for (int i = 0; i < 20000; i++) {
|
||||
m4(src, false);
|
||||
}
|
||||
if (!m4(src, true)) {
|
||||
System.out.println("m4 failed");
|
||||
success = false;
|
||||
}
|
||||
|
||||
for (int i = 0; i < 20000; i++) {
|
||||
m5(src, i%2 == 0, false);
|
||||
}
|
||||
if (!m5(src, true, true)) {
|
||||
System.out.println("m4 failed");
|
||||
success = false;
|
||||
}
|
||||
|
||||
if (!success) {
|
||||
throw new RuntimeException("Test failed");
|
||||
}
|
||||
}
|
||||
}
|
310
hotspot/test/compiler/loopopts/TestMoveStoresOutOfLoops.java
Normal file
310
hotspot/test/compiler/loopopts/TestMoveStoresOutOfLoops.java
Normal file
@ -0,0 +1,310 @@
|
||||
/*
|
||||
* Copyright (c) 2015, Oracle and/or its affiliates. All rights reserved.
|
||||
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
|
||||
*
|
||||
* This code is free software; you can redistribute it and/or modify it
|
||||
* under the terms of the GNU General Public License version 2 only, as
|
||||
* published by the Free Software Foundation.
|
||||
*
|
||||
* This code is distributed in the hope that it will be useful, but WITHOUT
|
||||
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
|
||||
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
||||
* version 2 for more details (a copy is included in the LICENSE file that
|
||||
* accompanied this code).
|
||||
*
|
||||
* You should have received a copy of the GNU General Public License version
|
||||
* 2 along with this work; if not, write to the Free Software Foundation,
|
||||
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
|
||||
*
|
||||
* Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
|
||||
* or visit www.oracle.com if you need additional information or have any
|
||||
* questions.
|
||||
*
|
||||
*/
|
||||
|
||||
/**
|
||||
* @test
|
||||
* @bug 8080289
|
||||
* @summary Sink stores out of loops if possible
|
||||
* @run main/othervm -XX:-UseOnStackReplacement -XX:-BackgroundCompilation -XX:+PrintCompilation -XX:CompileCommand=dontinline,TestMoveStoresOutOfLoops::test* TestMoveStoresOutOfLoops
|
||||
*
|
||||
*/
|
||||
|
||||
import java.lang.reflect.*;
|
||||
import java.util.*;
|
||||
import java.util.function.*;
|
||||
|
||||
public class TestMoveStoresOutOfLoops {
|
||||
|
||||
private static long[] array = new long[10];
|
||||
private static long[] array2 = new long[10];
|
||||
private static boolean[] array3 = new boolean[1000];
|
||||
private static byte[] byte_array = new byte[10];
|
||||
|
||||
// Array store should be moved out of the loop, value stored
|
||||
// should be 999, the loop should be eliminated
|
||||
static void test_after_1(int idx) {
|
||||
for (int i = 0; i < 1000; i++) {
|
||||
array[idx] = i;
|
||||
}
|
||||
}
|
||||
|
||||
// Array store can't be moved out of loop because of following
|
||||
// non loop invariant array access
|
||||
static void test_after_2(int idx) {
|
||||
for (int i = 0; i < 1000; i++) {
|
||||
array[idx] = i;
|
||||
array2[i%10] = i;
|
||||
}
|
||||
}
|
||||
|
||||
// Array store can't be moved out of loop because of following
|
||||
// use
|
||||
static void test_after_3(int idx) {
|
||||
for (int i = 0; i < 1000; i++) {
|
||||
array[idx] = i;
|
||||
if (array[0] == -1) {
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Array store can't be moved out of loop because of preceding
|
||||
// use
|
||||
static void test_after_4(int idx) {
|
||||
for (int i = 0; i < 1000; i++) {
|
||||
if (array[0] == -2) {
|
||||
break;
|
||||
}
|
||||
array[idx] = i;
|
||||
}
|
||||
}
|
||||
|
||||
// All array stores should be moved out of the loop, one after
|
||||
// the other
|
||||
static void test_after_5(int idx) {
|
||||
for (int i = 0; i < 1000; i++) {
|
||||
array[idx] = i;
|
||||
array[idx+1] = i;
|
||||
array[idx+2] = i;
|
||||
array[idx+3] = i;
|
||||
array[idx+4] = i;
|
||||
array[idx+5] = i;
|
||||
}
|
||||
}
|
||||
|
||||
// Array store can be moved after the loop but needs to be
|
||||
// cloned on both exit paths
|
||||
static void test_after_6(int idx) {
|
||||
for (int i = 0; i < 1000; i++) {
|
||||
array[idx] = i;
|
||||
if (array3[i]) {
|
||||
return;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Optimize out redundant stores
|
||||
static void test_stores_1(int ignored) {
|
||||
array[0] = 0;
|
||||
array[1] = 1;
|
||||
array[2] = 2;
|
||||
array[0] = 0;
|
||||
array[1] = 1;
|
||||
array[2] = 2;
|
||||
}
|
||||
|
||||
static void test_stores_2(int idx) {
|
||||
array[idx+0] = 0;
|
||||
array[idx+1] = 1;
|
||||
array[idx+2] = 2;
|
||||
array[idx+0] = 0;
|
||||
array[idx+1] = 1;
|
||||
array[idx+2] = 2;
|
||||
}
|
||||
|
||||
static void test_stores_3(int idx) {
|
||||
byte_array[idx+0] = 0;
|
||||
byte_array[idx+1] = 1;
|
||||
byte_array[idx+2] = 2;
|
||||
byte_array[idx+0] = 0;
|
||||
byte_array[idx+1] = 1;
|
||||
byte_array[idx+2] = 2;
|
||||
}
|
||||
|
||||
// Array store can be moved out of the loop before the loop header
|
||||
static void test_before_1(int idx) {
|
||||
for (int i = 0; i < 1000; i++) {
|
||||
array[idx] = 999;
|
||||
}
|
||||
}
|
||||
|
||||
// Array store can't be moved out of the loop before the loop
|
||||
// header because there's more than one store on this slice
|
||||
static void test_before_2(int idx) {
|
||||
for (int i = 0; i < 1000; i++) {
|
||||
array[idx] = 999;
|
||||
array[i%2] = 0;
|
||||
}
|
||||
}
|
||||
|
||||
// Array store can't be moved out of the loop before the loop
|
||||
// header because of use before store
|
||||
static int test_before_3(int idx) {
|
||||
int res = 0;
|
||||
for (int i = 0; i < 1000; i++) {
|
||||
res += array[i%10];
|
||||
array[idx] = 999;
|
||||
}
|
||||
return res;
|
||||
}
|
||||
|
||||
// Array store can't be moved out of the loop before the loop
|
||||
// header because of possible early exit
|
||||
static void test_before_4(int idx) {
|
||||
for (int i = 0; i < 1000; i++) {
|
||||
if (idx / (i+1) > 0) {
|
||||
return;
|
||||
}
|
||||
array[idx] = 999;
|
||||
}
|
||||
}
|
||||
|
||||
// Array store can't be moved out of the loop before the loop
|
||||
// header because it doesn't postdominate the loop head
|
||||
static void test_before_5(int idx) {
|
||||
for (int i = 0; i < 1000; i++) {
|
||||
if (i % 2 == 0) {
|
||||
array[idx] = 999;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Array store can be moved out of the loop before the loop header
|
||||
static int test_before_6(int idx) {
|
||||
int res = 0;
|
||||
for (int i = 0; i < 1000; i++) {
|
||||
if (i%2 == 1) {
|
||||
res *= 2;
|
||||
} else {
|
||||
res++;
|
||||
}
|
||||
array[idx] = 999;
|
||||
}
|
||||
return res;
|
||||
}
|
||||
|
||||
final HashMap<String,Method> tests = new HashMap<>();
|
||||
{
|
||||
for (Method m : this.getClass().getDeclaredMethods()) {
|
||||
if (m.getName().matches("test_(before|after|stores)_[0-9]+")) {
|
||||
assert(Modifier.isStatic(m.getModifiers())) : m;
|
||||
tests.put(m.getName(), m);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
boolean success = true;
|
||||
void doTest(String name, Runnable init, Function<String, Boolean> check) throws Exception {
|
||||
Method m = tests.get(name);
|
||||
for (int i = 0; i < 20000; i++) {
|
||||
init.run();
|
||||
m.invoke(null, 0);
|
||||
success = success && check.apply(name);
|
||||
if (!success) {
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
static void array_init() {
|
||||
array[0] = -1;
|
||||
}
|
||||
|
||||
static boolean array_check(String name) {
|
||||
boolean success = true;
|
||||
if (array[0] != 999) {
|
||||
success = false;
|
||||
System.out.println(name + " failed: array[0] = " + array[0]);
|
||||
}
|
||||
return success;
|
||||
}
|
||||
|
||||
static void array_init2() {
|
||||
for (int i = 0; i < 6; i++) {
|
||||
array[i] = -1;
|
||||
}
|
||||
}
|
||||
|
||||
static boolean array_check2(String name) {
|
||||
boolean success = true;
|
||||
for (int i = 0; i < 6; i++) {
|
||||
if (array[i] != 999) {
|
||||
success = false;
|
||||
System.out.println(name + " failed: array[" + i + "] = " + array[i]);
|
||||
}
|
||||
}
|
||||
return success;
|
||||
}
|
||||
|
||||
static void array_init3() {
|
||||
for (int i = 0; i < 3; i++) {
|
||||
array[i] = -1;
|
||||
}
|
||||
}
|
||||
|
||||
static boolean array_check3(String name) {
|
||||
boolean success = true;
|
||||
for (int i = 0; i < 3; i++) {
|
||||
if (array[i] != i) {
|
||||
success = false;
|
||||
System.out.println(name + " failed: array[" + i + "] = " + array[i]);
|
||||
}
|
||||
}
|
||||
return success;
|
||||
}
|
||||
|
||||
static void array_init4() {
|
||||
for (int i = 0; i < 3; i++) {
|
||||
byte_array[i] = -1;
|
||||
}
|
||||
}
|
||||
|
||||
static boolean array_check4(String name) {
|
||||
boolean success = true;
|
||||
for (int i = 0; i < 3; i++) {
|
||||
if (byte_array[i] != i) {
|
||||
success = false;
|
||||
System.out.println(name + " failed: byte_array[" + i + "] = " + byte_array[i]);
|
||||
}
|
||||
}
|
||||
return success;
|
||||
}
|
||||
|
||||
static public void main(String[] args) throws Exception {
|
||||
TestMoveStoresOutOfLoops test = new TestMoveStoresOutOfLoops();
|
||||
test.doTest("test_after_1", TestMoveStoresOutOfLoops::array_init, TestMoveStoresOutOfLoops::array_check);
|
||||
test.doTest("test_after_2", TestMoveStoresOutOfLoops::array_init, TestMoveStoresOutOfLoops::array_check);
|
||||
test.doTest("test_after_3", TestMoveStoresOutOfLoops::array_init, TestMoveStoresOutOfLoops::array_check);
|
||||
test.doTest("test_after_4", TestMoveStoresOutOfLoops::array_init, TestMoveStoresOutOfLoops::array_check);
|
||||
test.doTest("test_after_5", TestMoveStoresOutOfLoops::array_init2, TestMoveStoresOutOfLoops::array_check2);
|
||||
test.doTest("test_after_6", TestMoveStoresOutOfLoops::array_init, TestMoveStoresOutOfLoops::array_check);
|
||||
array3[999] = true;
|
||||
test.doTest("test_after_6", TestMoveStoresOutOfLoops::array_init, TestMoveStoresOutOfLoops::array_check);
|
||||
|
||||
test.doTest("test_stores_1", TestMoveStoresOutOfLoops::array_init3, TestMoveStoresOutOfLoops::array_check3);
|
||||
test.doTest("test_stores_2", TestMoveStoresOutOfLoops::array_init3, TestMoveStoresOutOfLoops::array_check3);
|
||||
test.doTest("test_stores_3", TestMoveStoresOutOfLoops::array_init4, TestMoveStoresOutOfLoops::array_check4);
|
||||
|
||||
test.doTest("test_before_1", TestMoveStoresOutOfLoops::array_init, TestMoveStoresOutOfLoops::array_check);
|
||||
test.doTest("test_before_2", TestMoveStoresOutOfLoops::array_init, TestMoveStoresOutOfLoops::array_check);
|
||||
test.doTest("test_before_3", TestMoveStoresOutOfLoops::array_init, TestMoveStoresOutOfLoops::array_check);
|
||||
test.doTest("test_before_4", TestMoveStoresOutOfLoops::array_init, TestMoveStoresOutOfLoops::array_check);
|
||||
test.doTest("test_before_5", TestMoveStoresOutOfLoops::array_init, TestMoveStoresOutOfLoops::array_check);
|
||||
test.doTest("test_before_6", TestMoveStoresOutOfLoops::array_init, TestMoveStoresOutOfLoops::array_check);
|
||||
|
||||
if (!test.success) {
|
||||
throw new RuntimeException("Some tests failed");
|
||||
}
|
||||
}
|
||||
}
|
Loading…
x
Reference in New Issue
Block a user