efl/src/lib/eo/eo_ptr_indirection.x

786 lines
26 KiB
Plaintext
Raw Normal View History

#include <assert.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#ifdef _WIN32
# include <evil_private.h> /* mmap mprotect */
#else
# include <sys/mman.h>
#endif
#ifdef HAVE_VALGRIND
# include <valgrind.h>
# include <memcheck.h>
#endif
/* Start of pointer indirection:
*
* This feature is responsible of hiding from the developer the real pointer of
* the Eo object to supply a better memory management by preventing bad usage
* of the pointers.
*
* Eo * is no more a pointer but indexes to an entry into an ids table.
* For a better memory usage:
* - a tree structure is used, composed of a top level table pointing at
* mid tables pointing at tables composed of entries.
* - tables are allocated when needed (i.e no more empty entries in allocated tables.
* - empty tables are freed, except one kept as spare table.
*
* An Eo id is contructed by bits manipulation of table indexes and a generation.
*
* id = Mid Table | Table | Entry | Generation
*
* Generation helps finding abuse of ids. When an entry is assigned to an
* object, a generation is inserted into the id. If the developer uses this id
* although the object is freed and another one has replaced it into the same
* entry of the table, the generation will be different and an error will
* occur when accessing with the old id.
*
* Each Table is composed of:
* - an index 'start' indicating which free entry is the next one to use.
* - 2 indexes 'fifo_head' and 'fifo_tail' defining a fifo,
* that will help us to store the entries to be reused. It stores only the
* entries that have been used at least one time. The entries that have
* never been used are "pointed" by the start parameter.
* - entries composed of:
* - a pointer to the object
* - an index 'next_in_fifo' used to chain the free entries in the fifo
* - a flag indicating if the entry is active
* - a generation assigned to the object
*
* When an entry is searched into a table, we first use one of the entries that
* has never been used. If there is none, we try to pop from the fifo.
* If a such entry doesn't exist, we pass to the next table.
* When an entry is found, we reserve it to the object pointer
* then contruct and return the related Eo id.
*
* Assigning all the entries of a table before trying to reuse them from
* the fifo ensures that we are not going to soon recycle a released entry,
* thus minimize the risks of an aggressive del() then use() on a single entry.
*
* The indexes and a reference to the last table which served an entry is kept
* and is reused prior to the others untill it is full.
* When an object is freed, the entry into the table is released by appending
* it to the fifo.
*/
eo - by default on 64bit only use 47 bits because of luajit luajit wants to check the upper 17 bits of pointers and assume they are all 0 if on a 64bit architecture. it will panic and barf if they are used. we have been using them in our eoid's for a long time. my take on this is that this is unportable. assuming how many bits are or are not valid in an opaque pointer (void *) that is simply passed to you and you cannot dereference or use (no size information even for what amount of data it points to etc.), is just wrong. it's not portable and it's trying to be too smart and creating such issues. my take is that luajit needs a fix for this in the longer term. but for now let's have a 47 bit mode and go with that. it does mean i have to drop our generation counter to 10 bits on 64bit (from 27 bits) which increases likelihood of eoid re-use being falsely detected as valid (before on 64bit it was 1 in 130 million or so chance, with this 47 bit change it's 1 in 1000. but to be fair on 32bit it's 7 bits for gen count so 1 in 127 ... so still more than 10x "safer" than on 32bit... but still...). the relevant check in luajit is: (((uint64_t)(p) >> 47) ? (lj_err_msg(L, LJ_ERR_BADLU), NULL) : (p)) it ONLY does this on 64bit. on 32bit pointers are not checked for validity at all. as an aside, armv8.2 seemingly will bring 52bit addresses so luajit is going to fall over flat on a newer rev of armv8. it may be x86 also uses more bits. last i knew it was 48bits so the 47 bit check luajit does i think is even wrong for x86-64. to be detailed i read: amd64 arch == 48 bits (so luajit is wrong). even better In addition, the AMD specification requires that the most significant 16 bits of any virtual address, bits 48 through 63, must be copies of bit 47 (in a manner akin to sign extension). so if the upper bit of 48 is set THEN all the 16 upper bits must be 1... breaking luajit, even if it were 47bit and this rule applied. I read the architecture allows for up to 52bits of actual addresses so architecture-wise this is even wrong... So I smell a core bug here in luajit. Certainly in the number of bits it insists must be 0 (the upper 17 bits where on amd64/x86-64 it should be the upper 16 bits... and even then these may NOT be 0 if bit 47 (the upper bit of the lower 48 is 1).... so the whole check is invalid... :( at least the above is at a theoretical level. i believe that the addresses divide the 48 bits into 2 chunks (thus 47)... but at the PHYSICAL level with no mmu and virtual memory. arm64 has this: https://www.kernel.org/doc/Documentation/arm64/memory.txt note in all cases the 2nd chunk of memory has at leats some upper bits of physical addresses beign 1 ... which makes luajit invalid tyo use without virtual memory remapping these away from high bits. @fix
2017-11-22 20:27:52 -08:00
// enable this to test and use all 64bits of a pointer, otherwise limit to
// 47 bits because of luajit. it wants to check if any bits in the upper 17 are
// set for a sanity check for lightuserdata ... basically it does this:
// #define checklightudptr(L, p) (((uint64_t)(p) >> 47) ? (lj_err_msg(L, LJ_ERR_BADLU), NULL) : (p))
//#define EO_FULL64BIT 1
#if SIZEOF_UINTPTR_T == 4
/* 32 bits */
# define BITS_MID_TABLE_ID 5
# define BITS_TABLE_ID 5
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
# define BITS_ENTRY_ID 11
# define BITS_GENERATION_COUNTER 7
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
# define BITS_DOMAIN 2
# define BITS_CLASS 1
# define REF_TAG_SHIFT 31
# define DROPPED_TABLES 0
# define DROPPED_ENTRIES 4
typedef int16_t Table_Index;
typedef uint16_t Generation_Counter;
#else
eo - by default on 64bit only use 47 bits because of luajit luajit wants to check the upper 17 bits of pointers and assume they are all 0 if on a 64bit architecture. it will panic and barf if they are used. we have been using them in our eoid's for a long time. my take on this is that this is unportable. assuming how many bits are or are not valid in an opaque pointer (void *) that is simply passed to you and you cannot dereference or use (no size information even for what amount of data it points to etc.), is just wrong. it's not portable and it's trying to be too smart and creating such issues. my take is that luajit needs a fix for this in the longer term. but for now let's have a 47 bit mode and go with that. it does mean i have to drop our generation counter to 10 bits on 64bit (from 27 bits) which increases likelihood of eoid re-use being falsely detected as valid (before on 64bit it was 1 in 130 million or so chance, with this 47 bit change it's 1 in 1000. but to be fair on 32bit it's 7 bits for gen count so 1 in 127 ... so still more than 10x "safer" than on 32bit... but still...). the relevant check in luajit is: (((uint64_t)(p) >> 47) ? (lj_err_msg(L, LJ_ERR_BADLU), NULL) : (p)) it ONLY does this on 64bit. on 32bit pointers are not checked for validity at all. as an aside, armv8.2 seemingly will bring 52bit addresses so luajit is going to fall over flat on a newer rev of armv8. it may be x86 also uses more bits. last i knew it was 48bits so the 47 bit check luajit does i think is even wrong for x86-64. to be detailed i read: amd64 arch == 48 bits (so luajit is wrong). even better In addition, the AMD specification requires that the most significant 16 bits of any virtual address, bits 48 through 63, must be copies of bit 47 (in a manner akin to sign extension). so if the upper bit of 48 is set THEN all the 16 upper bits must be 1... breaking luajit, even if it were 47bit and this rule applied. I read the architecture allows for up to 52bits of actual addresses so architecture-wise this is even wrong... So I smell a core bug here in luajit. Certainly in the number of bits it insists must be 0 (the upper 17 bits where on amd64/x86-64 it should be the upper 16 bits... and even then these may NOT be 0 if bit 47 (the upper bit of the lower 48 is 1).... so the whole check is invalid... :( at least the above is at a theoretical level. i believe that the addresses divide the 48 bits into 2 chunks (thus 47)... but at the PHYSICAL level with no mmu and virtual memory. arm64 has this: https://www.kernel.org/doc/Documentation/arm64/memory.txt note in all cases the 2nd chunk of memory has at leats some upper bits of physical addresses beign 1 ... which makes luajit invalid tyo use without virtual memory remapping these away from high bits. @fix
2017-11-22 20:27:52 -08:00
# ifndef EO_FULL64BIT
/* 47 bits */
# define BITS_MID_TABLE_ID 11
# define BITS_TABLE_ID 11
# define BITS_ENTRY_ID 11
# define BITS_GENERATION_COUNTER 10
# define BITS_DOMAIN 2
# define BITS_CLASS 1
# define REF_TAG_SHIFT 46
# define DROPPED_TABLES 2
# define DROPPED_ENTRIES 3
typedef int16_t Table_Index;
typedef uint16_t Generation_Counter;
# else
/* 64 bits */
eo - by default on 64bit only use 47 bits because of luajit luajit wants to check the upper 17 bits of pointers and assume they are all 0 if on a 64bit architecture. it will panic and barf if they are used. we have been using them in our eoid's for a long time. my take on this is that this is unportable. assuming how many bits are or are not valid in an opaque pointer (void *) that is simply passed to you and you cannot dereference or use (no size information even for what amount of data it points to etc.), is just wrong. it's not portable and it's trying to be too smart and creating such issues. my take is that luajit needs a fix for this in the longer term. but for now let's have a 47 bit mode and go with that. it does mean i have to drop our generation counter to 10 bits on 64bit (from 27 bits) which increases likelihood of eoid re-use being falsely detected as valid (before on 64bit it was 1 in 130 million or so chance, with this 47 bit change it's 1 in 1000. but to be fair on 32bit it's 7 bits for gen count so 1 in 127 ... so still more than 10x "safer" than on 32bit... but still...). the relevant check in luajit is: (((uint64_t)(p) >> 47) ? (lj_err_msg(L, LJ_ERR_BADLU), NULL) : (p)) it ONLY does this on 64bit. on 32bit pointers are not checked for validity at all. as an aside, armv8.2 seemingly will bring 52bit addresses so luajit is going to fall over flat on a newer rev of armv8. it may be x86 also uses more bits. last i knew it was 48bits so the 47 bit check luajit does i think is even wrong for x86-64. to be detailed i read: amd64 arch == 48 bits (so luajit is wrong). even better In addition, the AMD specification requires that the most significant 16 bits of any virtual address, bits 48 through 63, must be copies of bit 47 (in a manner akin to sign extension). so if the upper bit of 48 is set THEN all the 16 upper bits must be 1... breaking luajit, even if it were 47bit and this rule applied. I read the architecture allows for up to 52bits of actual addresses so architecture-wise this is even wrong... So I smell a core bug here in luajit. Certainly in the number of bits it insists must be 0 (the upper 17 bits where on amd64/x86-64 it should be the upper 16 bits... and even then these may NOT be 0 if bit 47 (the upper bit of the lower 48 is 1).... so the whole check is invalid... :( at least the above is at a theoretical level. i believe that the addresses divide the 48 bits into 2 chunks (thus 47)... but at the PHYSICAL level with no mmu and virtual memory. arm64 has this: https://www.kernel.org/doc/Documentation/arm64/memory.txt note in all cases the 2nd chunk of memory has at leats some upper bits of physical addresses beign 1 ... which makes luajit invalid tyo use without virtual memory remapping these away from high bits. @fix
2017-11-22 20:27:52 -08:00
# define BITS_MID_TABLE_ID 11
# define BITS_TABLE_ID 11
# define BITS_ENTRY_ID 11
# define BITS_GENERATION_COUNTER 27
# define BITS_DOMAIN 2
# define BITS_CLASS 1
# define REF_TAG_SHIFT 63
# define DROPPED_TABLES 2
# define DROPPED_ENTRIES 3
typedef int16_t Table_Index;
typedef uint32_t Generation_Counter;
eo - by default on 64bit only use 47 bits because of luajit luajit wants to check the upper 17 bits of pointers and assume they are all 0 if on a 64bit architecture. it will panic and barf if they are used. we have been using them in our eoid's for a long time. my take on this is that this is unportable. assuming how many bits are or are not valid in an opaque pointer (void *) that is simply passed to you and you cannot dereference or use (no size information even for what amount of data it points to etc.), is just wrong. it's not portable and it's trying to be too smart and creating such issues. my take is that luajit needs a fix for this in the longer term. but for now let's have a 47 bit mode and go with that. it does mean i have to drop our generation counter to 10 bits on 64bit (from 27 bits) which increases likelihood of eoid re-use being falsely detected as valid (before on 64bit it was 1 in 130 million or so chance, with this 47 bit change it's 1 in 1000. but to be fair on 32bit it's 7 bits for gen count so 1 in 127 ... so still more than 10x "safer" than on 32bit... but still...). the relevant check in luajit is: (((uint64_t)(p) >> 47) ? (lj_err_msg(L, LJ_ERR_BADLU), NULL) : (p)) it ONLY does this on 64bit. on 32bit pointers are not checked for validity at all. as an aside, armv8.2 seemingly will bring 52bit addresses so luajit is going to fall over flat on a newer rev of armv8. it may be x86 also uses more bits. last i knew it was 48bits so the 47 bit check luajit does i think is even wrong for x86-64. to be detailed i read: amd64 arch == 48 bits (so luajit is wrong). even better In addition, the AMD specification requires that the most significant 16 bits of any virtual address, bits 48 through 63, must be copies of bit 47 (in a manner akin to sign extension). so if the upper bit of 48 is set THEN all the 16 upper bits must be 1... breaking luajit, even if it were 47bit and this rule applied. I read the architecture allows for up to 52bits of actual addresses so architecture-wise this is even wrong... So I smell a core bug here in luajit. Certainly in the number of bits it insists must be 0 (the upper 17 bits where on amd64/x86-64 it should be the upper 16 bits... and even then these may NOT be 0 if bit 47 (the upper bit of the lower 48 is 1).... so the whole check is invalid... :( at least the above is at a theoretical level. i believe that the addresses divide the 48 bits into 2 chunks (thus 47)... but at the PHYSICAL level with no mmu and virtual memory. arm64 has this: https://www.kernel.org/doc/Documentation/arm64/memory.txt note in all cases the 2nd chunk of memory has at leats some upper bits of physical addresses beign 1 ... which makes luajit invalid tyo use without virtual memory remapping these away from high bits. @fix
2017-11-22 20:27:52 -08:00
# endif
#endif
/* Shifts macros to manipulate the Eo id */
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
#define SHIFT_DOMAIN (BITS_MID_TABLE_ID + BITS_TABLE_ID + \
BITS_ENTRY_ID + BITS_GENERATION_COUNTER)
#define SHIFT_MID_TABLE_ID (BITS_TABLE_ID + \
BITS_ENTRY_ID + BITS_GENERATION_COUNTER)
#define SHIFT_TABLE_ID (BITS_ENTRY_ID + BITS_GENERATION_COUNTER)
#define SHIFT_ENTRY_ID (BITS_GENERATION_COUNTER)
/* Maximum ranges - a few tables and entries are dropped to minimize the amount
* of wasted bytes, see _eo_id_mem_alloc */
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
#define MAX_DOMAIN (1 << BITS_DOMAIN)
#define MAX_MID_TABLE_ID (1 << BITS_MID_TABLE_ID)
#define MAX_TABLE_ID ((1 << BITS_TABLE_ID) - DROPPED_TABLES )
#define MAX_ENTRY_ID ((1 << BITS_ENTRY_ID) - DROPPED_ENTRIES)
#define MAX_GENERATIONS (1 << BITS_GENERATION_COUNTER)
/* Masks */
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
#define MASK_DOMAIN (MAX_DOMAIN - 1)
#define MASK_MID_TABLE_ID (MAX_MID_TABLE_ID - 1)
#define MASK_TABLE_ID ((1 << BITS_TABLE_ID) - 1)
#define MASK_ENTRY_ID ((1 << BITS_ENTRY_ID) - 1)
#define MASK_GENERATIONS (MAX_GENERATIONS - 1)
#define MASK_OBJ_TAG (((Eo_Id) 1) << (REF_TAG_SHIFT))
/* This only applies to classes. Used to artificially enlarge the class ids
* to reduce the likelihood of a clash with normal integers. */
#define CLASS_TAG_SHIFT (REF_TAG_SHIFT - 1)
#define MASK_CLASS_TAG (((Eo_Id) 1) << (CLASS_TAG_SHIFT))
#define MEM_HEADER_SIZE 16
#define MEM_PAGE_SIZE 4096
#define MEM_MAGIC 0x3f61ec8a
typedef struct _Mem_Header
{
size_t size;
size_t magic;
} Mem_Header;
static void *
_eo_id_mem_alloc(size_t size)
{
#ifdef HAVE_MMAP
# ifdef HAVE_VALGRIND
if (RUNNING_ON_VALGRIND) return malloc(size);
else
# endif
{
void *ptr;
Mem_Header *hdr;
size_t newsize;
newsize = MEM_PAGE_SIZE * ((size + MEM_HEADER_SIZE + MEM_PAGE_SIZE - 1) /
MEM_PAGE_SIZE);
ptr = mmap(NULL, newsize, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANON, -1, 0);
if (ptr == MAP_FAILED)
{
ERR("mmap of eo id table region failed!");
return NULL;
}
hdr = ptr;
hdr->size = newsize;
hdr->magic = MEM_MAGIC;
/* DBG("asked:%lu allocated:%lu wasted:%lu bytes", size, newsize, (newsize - size)); */
return (void *)(((unsigned char *)ptr) + MEM_HEADER_SIZE);
}
#else
return malloc(size);
#endif
}
static void *
_eo_id_mem_calloc(size_t num, size_t size)
{
void *ptr = _eo_id_mem_alloc(num * size);
if (!ptr) return NULL;
memset(ptr, 0, num * size);
return ptr;
}
static void
_eo_id_mem_free(void *ptr)
{
#ifdef HAVE_MMAP
# ifdef HAVE_VALGRIND
if (RUNNING_ON_VALGRIND) free(ptr);
else
# endif
{
Mem_Header *hdr;
if (!ptr) return;
hdr = (Mem_Header *)(((unsigned char *)ptr) - MEM_HEADER_SIZE);
if (hdr->magic != MEM_MAGIC)
{
ERR("unmap of eo table region has bad magic!");
return;
}
munmap(hdr, hdr->size);
}
#else
free(ptr);
#endif
}
#ifdef EINA_DEBUG_MALLOC
static void
_eo_id_mem_protect(void *ptr, Eina_Bool may_not_write)
{
# ifdef HAVE_MMAP
# ifdef HAVE_VALGRIND
if (RUNNING_ON_VALGRIND) { return; }
else
# endif
{
Mem_Header *hdr;
if (!ptr) return;
hdr = (Mem_Header *)(((unsigned char *)ptr) - MEM_HEADER_SIZE);
if (hdr->magic != MEM_MAGIC)
{
ERR("mprotect of eo table region has bad magic!");
return;
}
mprotect(hdr, hdr->size, PROT_READ | ( may_not_write ? 0 : PROT_WRITE) );
}
# endif
}
# define PROTECT(_ptr_) _eo_id_mem_protect((_ptr_), EINA_TRUE)
# define UNPROTECT(_ptr_) _eo_id_mem_protect((_ptr_), EINA_FALSE)
#else
# define PROTECT(_ptr_)
# define UNPROTECT(_ptr_)
#endif
#define EO_ALIGN_SIZE(size) eina_mempool_alignof(size)
/* Entry */
typedef struct
{
/* Pointer to the object */
_Eo_Object *ptr;
/* Indicates where to find the next entry to recycle */
Table_Index next_in_fifo;
/* Active flag */
unsigned int active : 1;
/* Generation */
unsigned int generation : BITS_GENERATION_COUNTER;
} _Eo_Id_Entry;
/* Table */
typedef struct
{
/* Indicates where start the "never used" entries */
Table_Index start;
/* Indicates where to find the next entry to recycle */
Table_Index fifo_head;
/* Indicates where to add an entry to recycle */
Table_Index fifo_tail;
/* Packed mid table and table indexes */
Eo_Id partial_id;
/* Counter of free entries */
unsigned int free_entries;
/* Entries of the table holding real pointers and generations */
_Eo_Id_Entry entries[MAX_ENTRY_ID];
} _Eo_Ids_Table;
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
//////////////////////////////////////////////////////////////////////////
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
typedef struct _Eo_Id_Data Eo_Id_Data;
typedef struct _Eo_Id_Table_Data Eo_Id_Table_Data;
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
struct _Eo_Id_Table_Data
{
/* Cached eoid lookups */
struct
{
Eo_Id id;
_Eo_Object *object;
const Eo *isa_id;
const Efl_Class *klass;
Eina_Bool isa;
}
cache;
/* Tables handling pointers indirection */
_Eo_Ids_Table **eo_ids_tables[MAX_MID_TABLE_ID];
/* Current table used for following allocations */
_Eo_Ids_Table *current_table;
/* Spare empty table */
_Eo_Ids_Table *empty_table;
/* Optional lock around all objects in eoid table - only used if shared */
Eina_Lock obj_lock;
/* Next generation to use when assigning a new entry to a Eo pointer */
Generation_Counter generation;
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
/* are we shared so we need lock/unlock? */
Eina_Bool shared;
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
};
struct _Eo_Id_Data
{
Eo_Id_Table_Data *tables[4];
unsigned char local_domain;
unsigned char stack_top;
unsigned char domain_stack[255 - (sizeof(void *) * 4) - 2];
};
extern Eina_TLS _eo_table_data;
extern Eo_Id_Data *_eo_table_data_shared;
extern Eo_Id_Table_Data *_eo_table_data_shared_data;
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
static inline Eo_Id_Table_Data *
_eo_table_data_table_new(Efl_Id_Domain domain)
{
Eo_Id_Table_Data *tdata;
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
tdata = calloc(1, sizeof(Eo_Id_Table_Data));
if (!tdata) return NULL;
if (domain == EFL_ID_DOMAIN_SHARED)
{
if (!eina_lock_recursive_new(&(tdata->obj_lock)))
{
free(tdata);
return NULL;
}
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
tdata->shared = EINA_TRUE;
}
tdata->generation = rand() % MAX_GENERATIONS;
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
return tdata;
}
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
static inline Eo_Id_Data *
_eo_table_data_new(Efl_Id_Domain domain)
{
Eo_Id_Data *data;
data = calloc(1, sizeof(Eo_Id_Data));
if (!data) return NULL;
data->local_domain = domain;
data->domain_stack[data->stack_top] = data->local_domain;
data->tables[data->local_domain] =
_eo_table_data_table_new(data->local_domain);
if (domain != EFL_ID_DOMAIN_SHARED)
data->tables[EFL_ID_DOMAIN_SHARED] = _eo_table_data_shared_data;
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
return data;
}
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
static void
_eo_table_data_table_free(Eo_Id_Table_Data *tdata)
{
if (tdata->shared) eina_lock_free(&(tdata->obj_lock));
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
free(tdata);
}
static inline Eo_Id_Data *
_eo_table_data_get(void)
{
Eo_Id_Data *data = eina_tls_get(_eo_table_data);
if (EINA_LIKELY(data != NULL)) return data;
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
data = _eo_table_data_new(EFL_ID_DOMAIN_THREAD);
if (!data) return NULL;
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
eina_tls_set(_eo_table_data, data);
return data;
}
static inline Eo_Id_Table_Data *
_eo_table_data_current_table_get(Eo_Id_Data *data)
{
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
return data->tables[data->domain_stack[data->stack_top]];
}
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
static inline Eo_Id_Table_Data *
_eo_table_data_table_get(Eo_Id_Data *data, Efl_Id_Domain domain)
{
return data->tables[domain];
}
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
static inline Eina_Bool
_eo_id_domain_compatible(const Eo *o1, const Eo *o2)
{
Efl_Id_Domain domain1 = ((Eo_Id)o1 >> SHIFT_DOMAIN) & MASK_DOMAIN;
Efl_Id_Domain domain2 = ((Eo_Id)o2 >> SHIFT_DOMAIN) & MASK_DOMAIN;
if (domain1 == domain2) return EINA_TRUE;
ERR("Object %p and %p are not compatible. Domain %i and %i do not match",
o1, o2, domain1, domain2);
return EINA_FALSE;
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
}
static inline void
_eo_obj_pointer_done(const Eo_Id obj_id)
{
Efl_Id_Domain domain = (obj_id >> SHIFT_DOMAIN) & MASK_DOMAIN;
if (EINA_LIKELY(domain != EFL_ID_DOMAIN_SHARED)) return;
eina_lock_release(&(_eo_table_data_shared_data->obj_lock));
}
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
//////////////////////////////////////////////////////////////////////////
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
/* Macro used to compose an Eo id */
#define EO_COMPOSE_PARTIAL_ID(MID_TABLE, TABLE) \
(((Eo_Id) 0x1 << REF_TAG_SHIFT) | \
((Eo_Id)(MID_TABLE & MASK_MID_TABLE_ID) << SHIFT_MID_TABLE_ID) | \
((Eo_Id)(TABLE & MASK_TABLE_ID) << SHIFT_TABLE_ID))
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
#define EO_COMPOSE_FINAL_ID(PARTIAL_ID, ENTRY, DOMAIN, GENERATION) \
(PARTIAL_ID | \
(((Eo_Id)DOMAIN & MASK_DOMAIN) << SHIFT_DOMAIN) | \
((ENTRY & MASK_ENTRY_ID) << SHIFT_ENTRY_ID) | \
(GENERATION & MASK_GENERATIONS))
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
/* Macro to extract from an Eo id the indexes of the tables */
#define EO_DECOMPOSE_ID(ID, MID_TABLE, TABLE, ENTRY, GENERATION) \
MID_TABLE = (ID >> SHIFT_MID_TABLE_ID) & MASK_MID_TABLE_ID; \
TABLE = (ID >> SHIFT_TABLE_ID) & MASK_TABLE_ID; \
ENTRY = (ID >> SHIFT_ENTRY_ID) & MASK_ENTRY_ID; \
GENERATION = ID & MASK_GENERATIONS;
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
/* Macro used for readability */
#define TABLE_FROM_IDS tdata->eo_ids_tables[mid_table_id][table_id]
static inline _Eo_Id_Entry *
_get_available_entry(_Eo_Ids_Table *table)
{
_Eo_Id_Entry *entry = NULL;
if (table->start != MAX_ENTRY_ID)
{
/* Serve never used entries first */
entry = &(table->entries[table->start]);
UNPROTECT(table);
table->start++;
table->free_entries--;
}
else if (table->fifo_head != -1)
{
/* Pop a free entry from the fifo */
entry = &(table->entries[table->fifo_head]);
UNPROTECT(table);
if (entry->next_in_fifo == -1)
table->fifo_head = table->fifo_tail = -1;
else
table->fifo_head = entry->next_in_fifo;
table->free_entries--;
}
return entry;
}
static inline _Eo_Id_Entry *
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
_search_tables(Eo_Id_Table_Data *tdata)
{
_Eo_Ids_Table *table;
_Eo_Id_Entry *entry;
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
if (!tdata) return NULL;
for (Table_Index mid_table_id = 0; mid_table_id < MAX_MID_TABLE_ID; mid_table_id++)
{
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
if (!tdata->eo_ids_tables[mid_table_id])
{
/* Allocate a new intermediate table */
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
tdata->eo_ids_tables[mid_table_id] = _eo_id_mem_calloc(MAX_TABLE_ID, sizeof(_Eo_Ids_Table*));
}
for (Table_Index table_id = 0; table_id < MAX_TABLE_ID; table_id++)
{
table = TABLE_FROM_IDS;
if (!table)
{
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
if (tdata->empty_table)
{
/* Recycle the available empty table */
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
table = tdata->empty_table;
tdata->empty_table = NULL;
UNPROTECT(table);
}
else
{
/* Allocate a new table */
table = _eo_id_mem_calloc(1, sizeof(_Eo_Ids_Table));
}
/* Initialize the table and reserve the first entry */
table->start = 1;
table->free_entries = MAX_ENTRY_ID - 1;
table->fifo_head = table->fifo_tail = -1;
table->partial_id = EO_COMPOSE_PARTIAL_ID(mid_table_id, table_id);
entry = &(table->entries[0]);
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
UNPROTECT(tdata->eo_ids_tables[mid_table_id]);
TABLE_FROM_IDS = table;
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
PROTECT(tdata->eo_ids_tables[mid_table_id]);
}
else
entry = _get_available_entry(table);
if (entry)
{
/* Store table info into current table */
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
tdata->current_table = table;
return entry;
}
}
}
ERR("no more available entries to store eo objects");
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
tdata->current_table = NULL;
return NULL;
}
/* Gives a fake id that serves as a marker if eo id is off. */
static inline Eo_Id
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
_eo_id_allocate(const _Eo_Object *obj, const Eo *parent_id)
{
_Eo_Id_Entry *entry = NULL;
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
Eo_Id_Data *data;
Eo_Id_Table_Data *tdata;
Eo_Id id;
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
data = _eo_table_data_get();
if (parent_id)
{
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
Efl_Id_Domain domain = ((Eo_Id)parent_id >> SHIFT_DOMAIN) & MASK_DOMAIN;
tdata = _eo_table_data_table_get(data, domain);
}
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
else tdata = _eo_table_data_current_table_get(data);
if (!tdata) return 0;
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
if (EINA_LIKELY(!tdata->shared))
{
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
if (tdata->current_table)
entry = _get_available_entry(tdata->current_table);
if (!entry) entry = _search_tables(tdata);
if (!tdata->current_table || !entry)
{
return 0;
}
UNPROTECT(tdata->current_table);
/* [1;max-1] thus we never generate an Eo_Id equal to 0 */
tdata->generation++;
if (tdata->generation >= MAX_GENERATIONS) tdata->generation = 1;
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
/* Fill the entry and return it's Eo Id */
entry->ptr = (_Eo_Object *)obj;
entry->active = 1;
entry->generation = tdata->generation;
PROTECT(tdata->current_table);
id = EO_COMPOSE_FINAL_ID(tdata->current_table->partial_id,
(entry - tdata->current_table->entries),
data->domain_stack[data->stack_top],
entry->generation);
}
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
else
{
eina_lock_take(&(_eo_table_data_shared_data->obj_lock));
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
if (tdata->current_table)
entry = _get_available_entry(tdata->current_table);
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
if (!entry) entry = _search_tables(tdata);
if (!tdata->current_table || !entry)
{
id = 0;
goto shared_err;
}
UNPROTECT(tdata->current_table);
/* [1;max-1] thus we never generate an Eo_Id equal to 0 */
tdata->generation++;
if (tdata->generation == MAX_GENERATIONS) tdata->generation = 1;
/* Fill the entry and return it's Eo Id */
entry->ptr = (_Eo_Object *)obj;
entry->active = 1;
entry->generation = tdata->generation;
PROTECT(tdata->current_table);
id = EO_COMPOSE_FINAL_ID(tdata->current_table->partial_id,
(entry - tdata->current_table->entries),
EFL_ID_DOMAIN_SHARED,
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
entry->generation);
shared_err:
eina_lock_release(&(_eo_table_data_shared_data->obj_lock));
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
}
return id;
}
static inline void
_eo_id_release(const Eo_Id obj_id)
{
_Eo_Ids_Table *table;
_Eo_Id_Entry *entry;
Generation_Counter generation;
Table_Index mid_table_id, table_id, entry_id;
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
Efl_Id_Domain domain;
Eo_Id_Data *data;
Eo_Id_Table_Data *tdata;
domain = (obj_id >> SHIFT_DOMAIN) & MASK_DOMAIN;
data = _eo_table_data_get();
tdata = _eo_table_data_table_get(data, domain);
if (!tdata) return;
EO_DECOMPOSE_ID(obj_id, mid_table_id, table_id, entry_id, generation);
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
if (EINA_LIKELY(domain != EFL_ID_DOMAIN_SHARED))
{
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
// Check the validity of the entry
if (tdata->eo_ids_tables[mid_table_id] && (table = TABLE_FROM_IDS))
{
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
entry = &(table->entries[entry_id]);
if (entry && entry->active && (entry->generation == generation))
{
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
UNPROTECT(table);
table->free_entries++;
// Disable the entry
entry->active = 0;
entry->next_in_fifo = -1;
// Push the entry into the fifo
if (table->fifo_tail == -1)
table->fifo_head = table->fifo_tail = entry_id;
else
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
{
table->entries[table->fifo_tail].next_in_fifo = entry_id;
table->fifo_tail = entry_id;
}
PROTECT(table);
if (table->free_entries == MAX_ENTRY_ID)
{
UNPROTECT(tdata->eo_ids_tables[mid_table_id]);
TABLE_FROM_IDS = NULL;
PROTECT(tdata->eo_ids_tables[mid_table_id]);
// Recycle or free the empty table
if (!tdata->empty_table) tdata->empty_table = table;
else _eo_id_mem_free(table);
if (tdata->current_table == table)
tdata->current_table = NULL;
}
// In case an object is destroyed, wipe out the cache
if (tdata->cache.id == obj_id)
{
tdata->cache.id = 0;
tdata->cache.object = NULL;
}
if ((Eo_Id)tdata->cache.isa_id == obj_id)
{
tdata->cache.isa_id = NULL;
tdata->cache.klass = NULL;;
tdata->cache.isa = EINA_FALSE;
}
return;
}
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
}
}
else
{
eina_lock_take(&(_eo_table_data_shared_data->obj_lock));
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
// Check the validity of the entry
if (tdata->eo_ids_tables[mid_table_id] && (table = TABLE_FROM_IDS))
{
entry = &(table->entries[entry_id]);
if (entry && entry->active && (entry->generation == generation))
{
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
UNPROTECT(table);
table->free_entries++;
// Disable the entry
entry->active = 0;
entry->next_in_fifo = -1;
// Push the entry into the fifo
if (table->fifo_tail == -1)
table->fifo_head = table->fifo_tail = entry_id;
else
{
table->entries[table->fifo_tail].next_in_fifo = entry_id;
table->fifo_tail = entry_id;
}
PROTECT(table);
if (table->free_entries == MAX_ENTRY_ID)
{
UNPROTECT(tdata->eo_ids_tables[mid_table_id]);
TABLE_FROM_IDS = NULL;
PROTECT(tdata->eo_ids_tables[mid_table_id]);
// Recycle or free the empty table
if (!tdata->empty_table) tdata->empty_table = table;
else _eo_id_mem_free(table);
if (tdata->current_table == table)
tdata->current_table = NULL;
}
// In case an object is destroyed, wipe out the cache
if (tdata->cache.id == obj_id)
{
tdata->cache.id = 0;
tdata->cache.object = NULL;
}
if ((Eo_Id)tdata->cache.isa_id == obj_id)
{
tdata->cache.isa_id = NULL;
tdata->cache.klass = NULL;;
tdata->cache.isa = EINA_FALSE;
}
eina_lock_release(&(_eo_table_data_shared_data->obj_lock));
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
return;
}
}
eina_lock_release(&(_eo_table_data_shared_data->obj_lock));
}
ERR("obj_id %p is not pointing to a valid object. Maybe it has already been freed.", (void *)obj_id);
}
static inline void
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
_eo_free_ids_tables(Eo_Id_Data *data)
{
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
Eo_Id_Table_Data *tdata;
if (!data) return;
tdata = data->tables[data->local_domain];
for (Table_Index mid_table_id = 0; mid_table_id < MAX_MID_TABLE_ID; mid_table_id++)
{
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
if (tdata->eo_ids_tables[mid_table_id])
{
for (Table_Index table_id = 0; table_id < MAX_TABLE_ID; table_id++)
{
if (TABLE_FROM_IDS)
{
_eo_id_mem_free(TABLE_FROM_IDS);
}
}
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
_eo_id_mem_free(tdata->eo_ids_tables[mid_table_id]);
}
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
tdata->eo_ids_tables[mid_table_id] = NULL;
}
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
if (tdata->empty_table) _eo_id_mem_free(tdata->empty_table);
tdata->empty_table = tdata->current_table = NULL;
_eo_table_data_table_free(tdata);
data->tables[data->local_domain] = NULL;
free(data);
}
#ifdef EFL_DEBUG
static inline void
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
_eo_print(Eo_Id_Table_Data *tdata)
{
_Eo_Id_Entry *entry;
unsigned long obj_number = 0;
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
for (Table_Index mid_table_id = 0; mid_table_id < MAX_MID_TABLE_ID; mid_table_id++)
{
eo - make eo id table TLS private data for thread safety and speed This moved all the eoid tables, eoid lookup caches, generation count information ad eo_isa cache into a TLS segment of memory that is thread private. There is also a shared domain for EO objects that all threads can access, but it has an added cost of a lock. This means objects accessed outside the thread they were created in cannot be accessed by another thread unless they are adopted in temporarily, or create4d with the shared domain active at the time of creation. child objects will use their parent object domain if created with a parent object passed in. If you were accessing EO (EFL) objects across threads before then this will actually now cause your code to fail as it was invalid before to do this as no actual objects were threadsafe in EFL, so this will force things to "fail early". ecore_thread_main_loop_begin() and end() still work as this uses the eo domain adoption features to temporarily adopt a domain during this section and then return it when done. This returns speed back to eo brining the overhead in my tests of lookup for the elm genlist autobounce test in elementary from about 5-7% down to 2.5-2.6%. A steep drop. This does not mean everything is perfect. Still to do are: 1. Tests in the test suite 2. Some API's to help for sending objects from thread to thread 3. Make the eo call cache TLS data to make it also safe 4. Look at other locks in eo and probably move them to TLS data 5. Make eo resolve and call wrappers that call the real method func do recursive mutex wrapping of the given object IF it is a shared object to provide threadsafety transparently for shared objects (but adding some overhead as a result) 6. Test test est, and that is why this commit is going in now for wider testing 7. Decide how to make this work with sending IPC (between threads) 8. Deciding what makes an object sendable (a sendable property in base?) 9. Deciding what makes an object shareable (a sharable property in base?)
2016-09-07 01:53:33 -07:00
if (tdata->eo_ids_tables[mid_table_id])
{
for (Table_Index table_id = 0; table_id < MAX_TABLE_ID; table_id++)
{
if (TABLE_FROM_IDS)
{
for (Table_Index entry_id = 0; entry_id < MAX_ENTRY_ID; entry_id++)
{
entry = &(TABLE_FROM_IDS->entries[entry_id]);
if (entry->active)
{
printf("%ld: %p -> (%p, %p, %p, %p)\n", obj_number++,
entry->ptr,
(void *)mid_table_id, (void *)table_id, (void *)entry_id,
(void *)entry->generation);
}
}
}
}
}
}
}
#endif