kotovalexarian-likes-github/ruby--ruby
1
0
Fork 0
mirror of https://github.com/ruby/ruby.git synced 2022-11-09 12:17:21 -05:00
Code Releases Activity
ruby--ruby/vm_callinfo.h

442 lines
12 KiB
C
Raw Normal View History

add #include guard hack According to MSVC manual (*1), cl.exe can skip including a header file when that: - contains #pragma once, or - starts with #ifndef, or - starts with #if ! defined. GCC has a similar trick (*2), but it acts more stricter (e. g. there must be _no tokens_ outside of #ifndef...#endif). Sun C lacked #pragma once for a looong time. Oracle Developer Studio 12.5 finally implemented it, but we cannot assume such recent version. This changeset modifies header files so that each of them include strictly one #ifndef...#endif. I believe this is the most portable way to trigger compiler optimizations. [Bug #16770] *1: https://docs.microsoft.com/en-us/cpp/preprocessor/once *2: https://gcc.gnu.org/onlinedocs/cppinternals/Guard-Macros.html
2020-04-10 01:11:40 -04:00
#ifndef RUBY_VM_CALLINFO_H /*-*-C-*-vi:se ft=c:*/
#define RUBY_VM_CALLINFO_H
/**
* @file
* @author Ruby developers <ruby-core@ruby-lang.org>
* @copyright This file is a part of the programming language Ruby.
* Permission is hereby granted, to either redistribute and/or
* modify this file, provided that the conditions mentioned in the
* file COPYING are met. Consult the file for details.
*/
VALUE size packed callinfo (ci). Now, rb_call_info contains how to call the method with tuple of (mid, orig_argc, flags, kwarg). Most of cases, kwarg == NULL and mid+argc+flags only requires 64bits. So this patch packed rb_call_info to VALUE (1 word) on such cases. If we can not represent it in VALUE, then use imemo_callinfo which contains conventional callinfo (rb_callinfo, renamed from rb_call_info). iseq->body->ci_kw_size is removed because all of callinfo is VALUE size (packed ci or a pointer to imemo_callinfo). To access ci information, we need to use these functions: vm_ci_mid(ci), _flag(ci), _argc(ci), _kwarg(ci). struct rb_call_info_kw_arg is renamed to rb_callinfo_kwarg. rb_funcallv_with_cc() and rb_method_basic_definition_p_with_cc() is temporary removed because cd->ci should be marked.
2020-01-07 18:20:36 -05:00
#include "debug_counter.h"
enum vm_call_flag_bits {
VM_CALL_ARGS_SPLAT_bit, /* m(*args) */
VM_CALL_ARGS_BLOCKARG_bit, /* m(&block) */
VM_CALL_FCALL_bit, /* m(...) */
VM_CALL_VCALL_bit, /* m */
VM_CALL_ARGS_SIMPLE_bit, /* (ci->flag & (SPLAT|BLOCKARG)) && blockiseq == NULL && ci->kw_arg == NULL */
VM_CALL_BLOCKISEQ_bit, /* has blockiseq */
VM_CALL_KWARG_bit, /* has kwarg */
VM_CALL_KW_SPLAT_bit, /* m(**opts) */
VM_CALL_TAILCALL_bit, /* located at tail position */
VM_CALL_SUPER_bit, /* super */
VM_CALL_ZSUPER_bit, /* zsuper */
VM_CALL_OPT_SEND_bit, /* internal flag */
Reduce allocations for keyword argument hashes Previously, passing a keyword splat to a method always allocated a hash on the caller side, and accepting arbitrary keywords in a method allocated a separate hash on the callee side. Passing explicit keywords to a method that accepted a keyword splat did not allocate a hash on the caller side, but resulted in two hashes allocated on the callee side. This commit makes passing a single keyword splat to a method not allocate a hash on the caller side. Passing multiple keyword splats or a mix of explicit keywords and a keyword splat still generates a hash on the caller side. On the callee side, if arbitrary keywords are not accepted, it does not allocate a hash. If arbitrary keywords are accepted, it will allocate a hash, but this commit uses a callinfo flag to indicate whether the caller already allocated a hash, and if so, the callee can use the passed hash without duplicating it. So this commit should make it so that a maximum of a single hash is allocated during method calls. To set the callinfo flag appropriately, method call argument compilation checks if only a single keyword splat is given. If only one keyword splat is given, the VM_CALL_KW_SPLAT_MUT callinfo flag is not set, since in that case the keyword splat is passed directly and not mutable. If more than one splat is used, a new hash needs to be generated on the caller side, and in that case the callinfo flag is set, indicating the keyword splat is mutable by the callee. In compile_hash, used for both hash and keyword argument compilation, if compiling keyword arguments and only a single keyword splat is used, pass the argument directly. On the caller side, in vm_args.c, the callinfo flag needs to be recognized and handled. Because the keyword splat argument may not be a hash, it needs to be converted to a hash first if not. Then, unless the callinfo flag is set, the hash needs to be duplicated. The temporary copy of the callinfo flag, kw_flag, is updated if a hash was duplicated, to prevent the need to duplicate it again. If we are converting to a hash or duplicating a hash, we need to update the argument array, which can including duplicating the positional splat array if one was passed. CALLER_SETUP_ARG and a couple other places needs to be modified to handle similar issues for other types of calls. This includes fairly comprehensive tests for different ways keywords are handled internally, checking that you get equal results but that keyword splats on the caller side result in distinct objects for keyword rest parameters. Included are benchmarks for keyword argument calls. Brief results when compiled without optimization: def kw(a: 1) a end def kws(**kw) kw end h = {a: 1} kw(a: 1) # about same kw(**h) # 2.37x faster kws(a: 1) # 1.30x faster kws(**h) # 2.19x faster kw(a: 1, **h) # 1.03x slower kw(**h, **h) # about same kws(a: 1, **h) # 1.16x faster kws(**h, **h) # 1.14x faster
2020-02-24 15:05:07 -05:00
VM_CALL_KW_SPLAT_MUT_bit, /* kw splat hash can be modified (to avoid allocating a new one) */
VALUE size packed callinfo (ci). Now, rb_call_info contains how to call the method with tuple of (mid, orig_argc, flags, kwarg). Most of cases, kwarg == NULL and mid+argc+flags only requires 64bits. So this patch packed rb_call_info to VALUE (1 word) on such cases. If we can not represent it in VALUE, then use imemo_callinfo which contains conventional callinfo (rb_callinfo, renamed from rb_call_info). iseq->body->ci_kw_size is removed because all of callinfo is VALUE size (packed ci or a pointer to imemo_callinfo). To access ci information, we need to use these functions: vm_ci_mid(ci), _flag(ci), _argc(ci), _kwarg(ci). struct rb_call_info_kw_arg is renamed to rb_callinfo_kwarg. rb_funcallv_with_cc() and rb_method_basic_definition_p_with_cc() is temporary removed because cd->ci should be marked.
2020-01-07 18:20:36 -05:00
VM_CALL__END
};
#define VM_CALL_ARGS_SPLAT (0x01 << VM_CALL_ARGS_SPLAT_bit)
#define VM_CALL_ARGS_BLOCKARG (0x01 << VM_CALL_ARGS_BLOCKARG_bit)
#define VM_CALL_FCALL (0x01 << VM_CALL_FCALL_bit)
#define VM_CALL_VCALL (0x01 << VM_CALL_VCALL_bit)
#define VM_CALL_ARGS_SIMPLE (0x01 << VM_CALL_ARGS_SIMPLE_bit)
#define VM_CALL_BLOCKISEQ (0x01 << VM_CALL_BLOCKISEQ_bit)
#define VM_CALL_KWARG (0x01 << VM_CALL_KWARG_bit)
#define VM_CALL_KW_SPLAT (0x01 << VM_CALL_KW_SPLAT_bit)
#define VM_CALL_TAILCALL (0x01 << VM_CALL_TAILCALL_bit)
#define VM_CALL_SUPER (0x01 << VM_CALL_SUPER_bit)
#define VM_CALL_ZSUPER (0x01 << VM_CALL_ZSUPER_bit)
#define VM_CALL_OPT_SEND (0x01 << VM_CALL_OPT_SEND_bit)
Reduce allocations for keyword argument hashes Previously, passing a keyword splat to a method always allocated a hash on the caller side, and accepting arbitrary keywords in a method allocated a separate hash on the callee side. Passing explicit keywords to a method that accepted a keyword splat did not allocate a hash on the caller side, but resulted in two hashes allocated on the callee side. This commit makes passing a single keyword splat to a method not allocate a hash on the caller side. Passing multiple keyword splats or a mix of explicit keywords and a keyword splat still generates a hash on the caller side. On the callee side, if arbitrary keywords are not accepted, it does not allocate a hash. If arbitrary keywords are accepted, it will allocate a hash, but this commit uses a callinfo flag to indicate whether the caller already allocated a hash, and if so, the callee can use the passed hash without duplicating it. So this commit should make it so that a maximum of a single hash is allocated during method calls. To set the callinfo flag appropriately, method call argument compilation checks if only a single keyword splat is given. If only one keyword splat is given, the VM_CALL_KW_SPLAT_MUT callinfo flag is not set, since in that case the keyword splat is passed directly and not mutable. If more than one splat is used, a new hash needs to be generated on the caller side, and in that case the callinfo flag is set, indicating the keyword splat is mutable by the callee. In compile_hash, used for both hash and keyword argument compilation, if compiling keyword arguments and only a single keyword splat is used, pass the argument directly. On the caller side, in vm_args.c, the callinfo flag needs to be recognized and handled. Because the keyword splat argument may not be a hash, it needs to be converted to a hash first if not. Then, unless the callinfo flag is set, the hash needs to be duplicated. The temporary copy of the callinfo flag, kw_flag, is updated if a hash was duplicated, to prevent the need to duplicate it again. If we are converting to a hash or duplicating a hash, we need to update the argument array, which can including duplicating the positional splat array if one was passed. CALLER_SETUP_ARG and a couple other places needs to be modified to handle similar issues for other types of calls. This includes fairly comprehensive tests for different ways keywords are handled internally, checking that you get equal results but that keyword splats on the caller side result in distinct objects for keyword rest parameters. Included are benchmarks for keyword argument calls. Brief results when compiled without optimization: def kw(a: 1) a end def kws(**kw) kw end h = {a: 1} kw(a: 1) # about same kw(**h) # 2.37x faster kws(a: 1) # 1.30x faster kws(**h) # 2.19x faster kw(a: 1, **h) # 1.03x slower kw(**h, **h) # about same kws(a: 1, **h) # 1.16x faster kws(**h, **h) # 1.14x faster
2020-02-24 15:05:07 -05:00
#define VM_CALL_KW_SPLAT_MUT (0x01 << VM_CALL_KW_SPLAT_MUT_bit)
VALUE size packed callinfo (ci). Now, rb_call_info contains how to call the method with tuple of (mid, orig_argc, flags, kwarg). Most of cases, kwarg == NULL and mid+argc+flags only requires 64bits. So this patch packed rb_call_info to VALUE (1 word) on such cases. If we can not represent it in VALUE, then use imemo_callinfo which contains conventional callinfo (rb_callinfo, renamed from rb_call_info). iseq->body->ci_kw_size is removed because all of callinfo is VALUE size (packed ci or a pointer to imemo_callinfo). To access ci information, we need to use these functions: vm_ci_mid(ci), _flag(ci), _argc(ci), _kwarg(ci). struct rb_call_info_kw_arg is renamed to rb_callinfo_kwarg. rb_funcallv_with_cc() and rb_method_basic_definition_p_with_cc() is temporary removed because cd->ci should be marked.
2020-01-07 18:20:36 -05:00
struct rb_callinfo_kwarg {
int keyword_len;
Avoid UB with flexible array member Accessing past the end of an array is technically UB. Use C99 flexible array member instead to avoid the UB and simplify allocation size calculation. See also: DCL38-C in the SEI CERT C Coding Standard
2020-04-12 15:19:06 -04:00
VALUE keywords[];
VALUE size packed callinfo (ci). Now, rb_call_info contains how to call the method with tuple of (mid, orig_argc, flags, kwarg). Most of cases, kwarg == NULL and mid+argc+flags only requires 64bits. So this patch packed rb_call_info to VALUE (1 word) on such cases. If we can not represent it in VALUE, then use imemo_callinfo which contains conventional callinfo (rb_callinfo, renamed from rb_call_info). iseq->body->ci_kw_size is removed because all of callinfo is VALUE size (packed ci or a pointer to imemo_callinfo). To access ci information, we need to use these functions: vm_ci_mid(ci), _flag(ci), _argc(ci), _kwarg(ci). struct rb_call_info_kw_arg is renamed to rb_callinfo_kwarg. rb_funcallv_with_cc() and rb_method_basic_definition_p_with_cc() is temporary removed because cd->ci should be marked.
2020-01-07 18:20:36 -05:00
};
static inline size_t
rb_callinfo_kwarg_bytes(int keyword_len)
{
return rb_size_mul_add_or_raise(
Avoid UB with flexible array member Accessing past the end of an array is technically UB. Use C99 flexible array member instead to avoid the UB and simplify allocation size calculation. See also: DCL38-C in the SEI CERT C Coding Standard
2020-04-12 15:19:06 -04:00
keyword_len,
VALUE size packed callinfo (ci). Now, rb_call_info contains how to call the method with tuple of (mid, orig_argc, flags, kwarg). Most of cases, kwarg == NULL and mid+argc+flags only requires 64bits. So this patch packed rb_call_info to VALUE (1 word) on such cases. If we can not represent it in VALUE, then use imemo_callinfo which contains conventional callinfo (rb_callinfo, renamed from rb_call_info). iseq->body->ci_kw_size is removed because all of callinfo is VALUE size (packed ci or a pointer to imemo_callinfo). To access ci information, we need to use these functions: vm_ci_mid(ci), _flag(ci), _argc(ci), _kwarg(ci). struct rb_call_info_kw_arg is renamed to rb_callinfo_kwarg. rb_funcallv_with_cc() and rb_method_basic_definition_p_with_cc() is temporary removed because cd->ci should be marked.
2020-01-07 18:20:36 -05:00
sizeof(VALUE),
sizeof(struct rb_callinfo_kwarg),
rb_eRuntimeError);
}
// imemo_callinfo
struct rb_callinfo {
VALUE flags;
const struct rb_callinfo_kwarg *kwarg;
VALUE mid;
VALUE flag;
VALUE argc;
};
#ifndef USE_EMBED_CI
#define USE_EMBED_CI 1
#endif
#if SIZEOF_VALUE == 8
#define CI_EMBED_TAG_bits 1
#define CI_EMBED_ARGC_bits 15
#define CI_EMBED_FLAG_bits 16
#define CI_EMBED_ID_bits 32
#elif SIZEOF_VALUE == 4
#define CI_EMBED_TAG_bits 1
Reduce allocations for keyword argument hashes Previously, passing a keyword splat to a method always allocated a hash on the caller side, and accepting arbitrary keywords in a method allocated a separate hash on the callee side. Passing explicit keywords to a method that accepted a keyword splat did not allocate a hash on the caller side, but resulted in two hashes allocated on the callee side. This commit makes passing a single keyword splat to a method not allocate a hash on the caller side. Passing multiple keyword splats or a mix of explicit keywords and a keyword splat still generates a hash on the caller side. On the callee side, if arbitrary keywords are not accepted, it does not allocate a hash. If arbitrary keywords are accepted, it will allocate a hash, but this commit uses a callinfo flag to indicate whether the caller already allocated a hash, and if so, the callee can use the passed hash without duplicating it. So this commit should make it so that a maximum of a single hash is allocated during method calls. To set the callinfo flag appropriately, method call argument compilation checks if only a single keyword splat is given. If only one keyword splat is given, the VM_CALL_KW_SPLAT_MUT callinfo flag is not set, since in that case the keyword splat is passed directly and not mutable. If more than one splat is used, a new hash needs to be generated on the caller side, and in that case the callinfo flag is set, indicating the keyword splat is mutable by the callee. In compile_hash, used for both hash and keyword argument compilation, if compiling keyword arguments and only a single keyword splat is used, pass the argument directly. On the caller side, in vm_args.c, the callinfo flag needs to be recognized and handled. Because the keyword splat argument may not be a hash, it needs to be converted to a hash first if not. Then, unless the callinfo flag is set, the hash needs to be duplicated. The temporary copy of the callinfo flag, kw_flag, is updated if a hash was duplicated, to prevent the need to duplicate it again. If we are converting to a hash or duplicating a hash, we need to update the argument array, which can including duplicating the positional splat array if one was passed. CALLER_SETUP_ARG and a couple other places needs to be modified to handle similar issues for other types of calls. This includes fairly comprehensive tests for different ways keywords are handled internally, checking that you get equal results but that keyword splats on the caller side result in distinct objects for keyword rest parameters. Included are benchmarks for keyword argument calls. Brief results when compiled without optimization: def kw(a: 1) a end def kws(**kw) kw end h = {a: 1} kw(a: 1) # about same kw(**h) # 2.37x faster kws(a: 1) # 1.30x faster kws(**h) # 2.19x faster kw(a: 1, **h) # 1.03x slower kw(**h, **h) # about same kws(a: 1, **h) # 1.16x faster kws(**h, **h) # 1.14x faster
2020-02-24 15:05:07 -05:00
#define CI_EMBED_ARGC_bits 3
#define CI_EMBED_FLAG_bits 13
VALUE size packed callinfo (ci). Now, rb_call_info contains how to call the method with tuple of (mid, orig_argc, flags, kwarg). Most of cases, kwarg == NULL and mid+argc+flags only requires 64bits. So this patch packed rb_call_info to VALUE (1 word) on such cases. If we can not represent it in VALUE, then use imemo_callinfo which contains conventional callinfo (rb_callinfo, renamed from rb_call_info). iseq->body->ci_kw_size is removed because all of callinfo is VALUE size (packed ci or a pointer to imemo_callinfo). To access ci information, we need to use these functions: vm_ci_mid(ci), _flag(ci), _argc(ci), _kwarg(ci). struct rb_call_info_kw_arg is renamed to rb_callinfo_kwarg. rb_funcallv_with_cc() and rb_method_basic_definition_p_with_cc() is temporary removed because cd->ci should be marked.
2020-01-07 18:20:36 -05:00
#define CI_EMBED_ID_bits 15
#endif
#if (CI_EMBED_TAG_bits + CI_EMBED_ARGC_bits + CI_EMBED_FLAG_bits + CI_EMBED_ID_bits) != (SIZEOF_VALUE * 8)
#error
#endif
#define CI_EMBED_FLAG 0x01
#define CI_EMBED_ARGC_SHFT (CI_EMBED_TAG_bits)
Introduce disposable call-cache. This patch contains several ideas: (1) Disposable inline method cache (IMC) for race-free inline method cache * Making call-cache (CC) as a RVALUE (GC target object) and allocate new CC on cache miss. * This technique allows race-free access from parallel processing elements like RCU. (2) Introduce per-Class method cache (pCMC) * Instead of fixed-size global method cache (GMC), pCMC allows flexible cache size. * Caching CCs reduces CC allocation and allow sharing CC's fast-path between same call-info (CI) call-sites. (3) Invalidate an inline method cache by invalidating corresponding method entries (MEs) * Instead of using class serials, we set "invalidated" flag for method entry itself to represent cache invalidation. * Compare with using class serials, the impact of method modification (add/overwrite/delete) is small. * Updating class serials invalidate all method caches of the class and sub-classes. * Proposed approach only invalidate the method cache of only one ME. See [Feature #16614] for more details.
2020-01-08 02:14:01 -05:00
#define CI_EMBED_ARGC_MASK ((((VALUE)1)<<CI_EMBED_ARGC_bits) - 1)
VALUE size packed callinfo (ci). Now, rb_call_info contains how to call the method with tuple of (mid, orig_argc, flags, kwarg). Most of cases, kwarg == NULL and mid+argc+flags only requires 64bits. So this patch packed rb_call_info to VALUE (1 word) on such cases. If we can not represent it in VALUE, then use imemo_callinfo which contains conventional callinfo (rb_callinfo, renamed from rb_call_info). iseq->body->ci_kw_size is removed because all of callinfo is VALUE size (packed ci or a pointer to imemo_callinfo). To access ci information, we need to use these functions: vm_ci_mid(ci), _flag(ci), _argc(ci), _kwarg(ci). struct rb_call_info_kw_arg is renamed to rb_callinfo_kwarg. rb_funcallv_with_cc() and rb_method_basic_definition_p_with_cc() is temporary removed because cd->ci should be marked.
2020-01-07 18:20:36 -05:00
#define CI_EMBED_FLAG_SHFT (CI_EMBED_TAG_bits + CI_EMBED_ARGC_bits)
Introduce disposable call-cache. This patch contains several ideas: (1) Disposable inline method cache (IMC) for race-free inline method cache * Making call-cache (CC) as a RVALUE (GC target object) and allocate new CC on cache miss. * This technique allows race-free access from parallel processing elements like RCU. (2) Introduce per-Class method cache (pCMC) * Instead of fixed-size global method cache (GMC), pCMC allows flexible cache size. * Caching CCs reduces CC allocation and allow sharing CC's fast-path between same call-info (CI) call-sites. (3) Invalidate an inline method cache by invalidating corresponding method entries (MEs) * Instead of using class serials, we set "invalidated" flag for method entry itself to represent cache invalidation. * Compare with using class serials, the impact of method modification (add/overwrite/delete) is small. * Updating class serials invalidate all method caches of the class and sub-classes. * Proposed approach only invalidate the method cache of only one ME. See [Feature #16614] for more details.
2020-01-08 02:14:01 -05:00
#define CI_EMBED_FLAG_MASK ((((VALUE)1)<<CI_EMBED_FLAG_bits) - 1)
VALUE size packed callinfo (ci). Now, rb_call_info contains how to call the method with tuple of (mid, orig_argc, flags, kwarg). Most of cases, kwarg == NULL and mid+argc+flags only requires 64bits. So this patch packed rb_call_info to VALUE (1 word) on such cases. If we can not represent it in VALUE, then use imemo_callinfo which contains conventional callinfo (rb_callinfo, renamed from rb_call_info). iseq->body->ci_kw_size is removed because all of callinfo is VALUE size (packed ci or a pointer to imemo_callinfo). To access ci information, we need to use these functions: vm_ci_mid(ci), _flag(ci), _argc(ci), _kwarg(ci). struct rb_call_info_kw_arg is renamed to rb_callinfo_kwarg. rb_funcallv_with_cc() and rb_method_basic_definition_p_with_cc() is temporary removed because cd->ci should be marked.
2020-01-07 18:20:36 -05:00
#define CI_EMBED_ID_SHFT (CI_EMBED_TAG_bits + CI_EMBED_ARGC_bits + CI_EMBED_FLAG_bits)
Introduce disposable call-cache. This patch contains several ideas: (1) Disposable inline method cache (IMC) for race-free inline method cache * Making call-cache (CC) as a RVALUE (GC target object) and allocate new CC on cache miss. * This technique allows race-free access from parallel processing elements like RCU. (2) Introduce per-Class method cache (pCMC) * Instead of fixed-size global method cache (GMC), pCMC allows flexible cache size. * Caching CCs reduces CC allocation and allow sharing CC's fast-path between same call-info (CI) call-sites. (3) Invalidate an inline method cache by invalidating corresponding method entries (MEs) * Instead of using class serials, we set "invalidated" flag for method entry itself to represent cache invalidation. * Compare with using class serials, the impact of method modification (add/overwrite/delete) is small. * Updating class serials invalidate all method caches of the class and sub-classes. * Proposed approach only invalidate the method cache of only one ME. See [Feature #16614] for more details.
2020-01-08 02:14:01 -05:00
#define CI_EMBED_ID_MASK ((((VALUE)1)<<CI_EMBED_ID_bits) - 1)
VALUE size packed callinfo (ci). Now, rb_call_info contains how to call the method with tuple of (mid, orig_argc, flags, kwarg). Most of cases, kwarg == NULL and mid+argc+flags only requires 64bits. So this patch packed rb_call_info to VALUE (1 word) on such cases. If we can not represent it in VALUE, then use imemo_callinfo which contains conventional callinfo (rb_callinfo, renamed from rb_call_info). iseq->body->ci_kw_size is removed because all of callinfo is VALUE size (packed ci or a pointer to imemo_callinfo). To access ci information, we need to use these functions: vm_ci_mid(ci), _flag(ci), _argc(ci), _kwarg(ci). struct rb_call_info_kw_arg is renamed to rb_callinfo_kwarg. rb_funcallv_with_cc() and rb_method_basic_definition_p_with_cc() is temporary removed because cd->ci should be marked.
2020-01-07 18:20:36 -05:00
Introduce disposable call-cache. This patch contains several ideas: (1) Disposable inline method cache (IMC) for race-free inline method cache * Making call-cache (CC) as a RVALUE (GC target object) and allocate new CC on cache miss. * This technique allows race-free access from parallel processing elements like RCU. (2) Introduce per-Class method cache (pCMC) * Instead of fixed-size global method cache (GMC), pCMC allows flexible cache size. * Caching CCs reduces CC allocation and allow sharing CC's fast-path between same call-info (CI) call-sites. (3) Invalidate an inline method cache by invalidating corresponding method entries (MEs) * Instead of using class serials, we set "invalidated" flag for method entry itself to represent cache invalidation. * Compare with using class serials, the impact of method modification (add/overwrite/delete) is small. * Updating class serials invalidate all method caches of the class and sub-classes. * Proposed approach only invalidate the method cache of only one ME. See [Feature #16614] for more details.
2020-01-08 02:14:01 -05:00
static inline bool
VALUE size packed callinfo (ci). Now, rb_call_info contains how to call the method with tuple of (mid, orig_argc, flags, kwarg). Most of cases, kwarg == NULL and mid+argc+flags only requires 64bits. So this patch packed rb_call_info to VALUE (1 word) on such cases. If we can not represent it in VALUE, then use imemo_callinfo which contains conventional callinfo (rb_callinfo, renamed from rb_call_info). iseq->body->ci_kw_size is removed because all of callinfo is VALUE size (packed ci or a pointer to imemo_callinfo). To access ci information, we need to use these functions: vm_ci_mid(ci), _flag(ci), _argc(ci), _kwarg(ci). struct rb_call_info_kw_arg is renamed to rb_callinfo_kwarg. rb_funcallv_with_cc() and rb_method_basic_definition_p_with_cc() is temporary removed because cd->ci should be marked.
2020-01-07 18:20:36 -05:00
vm_ci_packed_p(const struct rb_callinfo *ci)
{
#if USE_EMBED_CI
if (LIKELY(((VALUE)ci) & 0x01)) {
return 1;
}
else {
Introduce disposable call-cache. This patch contains several ideas: (1) Disposable inline method cache (IMC) for race-free inline method cache * Making call-cache (CC) as a RVALUE (GC target object) and allocate new CC on cache miss. * This technique allows race-free access from parallel processing elements like RCU. (2) Introduce per-Class method cache (pCMC) * Instead of fixed-size global method cache (GMC), pCMC allows flexible cache size. * Caching CCs reduces CC allocation and allow sharing CC's fast-path between same call-info (CI) call-sites. (3) Invalidate an inline method cache by invalidating corresponding method entries (MEs) * Instead of using class serials, we set "invalidated" flag for method entry itself to represent cache invalidation. * Compare with using class serials, the impact of method modification (add/overwrite/delete) is small. * Updating class serials invalidate all method caches of the class and sub-classes. * Proposed approach only invalidate the method cache of only one ME. See [Feature #16614] for more details.
2020-01-08 02:14:01 -05:00
VM_ASSERT(IMEMO_TYPE_P(ci, imemo_callinfo));
VALUE size packed callinfo (ci). Now, rb_call_info contains how to call the method with tuple of (mid, orig_argc, flags, kwarg). Most of cases, kwarg == NULL and mid+argc+flags only requires 64bits. So this patch packed rb_call_info to VALUE (1 word) on such cases. If we can not represent it in VALUE, then use imemo_callinfo which contains conventional callinfo (rb_callinfo, renamed from rb_call_info). iseq->body->ci_kw_size is removed because all of callinfo is VALUE size (packed ci or a pointer to imemo_callinfo). To access ci information, we need to use these functions: vm_ci_mid(ci), _flag(ci), _argc(ci), _kwarg(ci). struct rb_call_info_kw_arg is renamed to rb_callinfo_kwarg. rb_funcallv_with_cc() and rb_method_basic_definition_p_with_cc() is temporary removed because cd->ci should be marked.
2020-01-07 18:20:36 -05:00
return 0;
}
#else
return 0;
#endif
}
Introduce disposable call-cache. This patch contains several ideas: (1) Disposable inline method cache (IMC) for race-free inline method cache * Making call-cache (CC) as a RVALUE (GC target object) and allocate new CC on cache miss. * This technique allows race-free access from parallel processing elements like RCU. (2) Introduce per-Class method cache (pCMC) * Instead of fixed-size global method cache (GMC), pCMC allows flexible cache size. * Caching CCs reduces CC allocation and allow sharing CC's fast-path between same call-info (CI) call-sites. (3) Invalidate an inline method cache by invalidating corresponding method entries (MEs) * Instead of using class serials, we set "invalidated" flag for method entry itself to represent cache invalidation. * Compare with using class serials, the impact of method modification (add/overwrite/delete) is small. * Updating class serials invalidate all method caches of the class and sub-classes. * Proposed approach only invalidate the method cache of only one ME. See [Feature #16614] for more details.
2020-01-08 02:14:01 -05:00
static inline bool
vm_ci_p(const struct rb_callinfo *ci)
{
if (vm_ci_packed_p(ci) || IMEMO_TYPE_P(ci, imemo_callinfo)) {
return 1;
}
else {
return 0;
}
}
VALUE size packed callinfo (ci). Now, rb_call_info contains how to call the method with tuple of (mid, orig_argc, flags, kwarg). Most of cases, kwarg == NULL and mid+argc+flags only requires 64bits. So this patch packed rb_call_info to VALUE (1 word) on such cases. If we can not represent it in VALUE, then use imemo_callinfo which contains conventional callinfo (rb_callinfo, renamed from rb_call_info). iseq->body->ci_kw_size is removed because all of callinfo is VALUE size (packed ci or a pointer to imemo_callinfo). To access ci information, we need to use these functions: vm_ci_mid(ci), _flag(ci), _argc(ci), _kwarg(ci). struct rb_call_info_kw_arg is renamed to rb_callinfo_kwarg. rb_funcallv_with_cc() and rb_method_basic_definition_p_with_cc() is temporary removed because cd->ci should be marked.
2020-01-07 18:20:36 -05:00
static inline ID
vm_ci_mid(const struct rb_callinfo *ci)
{
if (vm_ci_packed_p(ci)) {
return (((VALUE)ci) >> CI_EMBED_ID_SHFT) & CI_EMBED_ID_MASK;
}
else {
return (ID)ci->mid;
}
}
static inline unsigned int
vm_ci_flag(const struct rb_callinfo *ci)
{
if (vm_ci_packed_p(ci)) {
return (unsigned int)((((VALUE)ci) >> CI_EMBED_FLAG_SHFT) & CI_EMBED_FLAG_MASK);
}
else {
return (unsigned int)ci->flag;
}
}
static inline unsigned int
vm_ci_argc(const struct rb_callinfo *ci)
{
if (vm_ci_packed_p(ci)) {
return (unsigned int)((((VALUE)ci) >> CI_EMBED_ARGC_SHFT) & CI_EMBED_ARGC_MASK);
}
else {
return (unsigned int)ci->argc;
}
}
static inline const struct rb_callinfo_kwarg *
vm_ci_kwarg(const struct rb_callinfo *ci)
{
if (vm_ci_packed_p(ci)) {
return NULL;
}
else {
return ci->kwarg;
}
}
static inline void
vm_ci_dump(const struct rb_callinfo *ci)
{
if (vm_ci_packed_p(ci)) {
fprintf(stderr, "packed_ci ID:%s flag:%x argc:%u\n",
rb_id2name(vm_ci_mid(ci)), vm_ci_flag(ci), vm_ci_argc(ci));
}
else {
rp(ci);
}
}
#define vm_ci_new(mid, flag, argc, kwarg) vm_ci_new_(mid, flag, argc, kwarg, __FILE__, __LINE__)
#define vm_ci_new_runtime(mid, flag, argc, kwarg) vm_ci_new_runtime_(mid, flag, argc, kwarg, __FILE__, __LINE__)
static inline const struct rb_callinfo *
vm_ci_new_(ID mid, unsigned int flag, unsigned int argc, const struct rb_callinfo_kwarg *kwarg, const char *file, int line)
{
#if USE_EMBED_CI
if ((mid & ~CI_EMBED_ID_MASK) == 0 &&
(argc & ~CI_EMBED_ARGC_MASK) == 0 &&
kwarg == NULL) {
VALUE embed_ci =
Introduce disposable call-cache. This patch contains several ideas: (1) Disposable inline method cache (IMC) for race-free inline method cache * Making call-cache (CC) as a RVALUE (GC target object) and allocate new CC on cache miss. * This technique allows race-free access from parallel processing elements like RCU. (2) Introduce per-Class method cache (pCMC) * Instead of fixed-size global method cache (GMC), pCMC allows flexible cache size. * Caching CCs reduces CC allocation and allow sharing CC's fast-path between same call-info (CI) call-sites. (3) Invalidate an inline method cache by invalidating corresponding method entries (MEs) * Instead of using class serials, we set "invalidated" flag for method entry itself to represent cache invalidation. * Compare with using class serials, the impact of method modification (add/overwrite/delete) is small. * Updating class serials invalidate all method caches of the class and sub-classes. * Proposed approach only invalidate the method cache of only one ME. See [Feature #16614] for more details.
2020-01-08 02:14:01 -05:00
1L |
VALUE size packed callinfo (ci). Now, rb_call_info contains how to call the method with tuple of (mid, orig_argc, flags, kwarg). Most of cases, kwarg == NULL and mid+argc+flags only requires 64bits. So this patch packed rb_call_info to VALUE (1 word) on such cases. If we can not represent it in VALUE, then use imemo_callinfo which contains conventional callinfo (rb_callinfo, renamed from rb_call_info). iseq->body->ci_kw_size is removed because all of callinfo is VALUE size (packed ci or a pointer to imemo_callinfo). To access ci information, we need to use these functions: vm_ci_mid(ci), _flag(ci), _argc(ci), _kwarg(ci). struct rb_call_info_kw_arg is renamed to rb_callinfo_kwarg. rb_funcallv_with_cc() and rb_method_basic_definition_p_with_cc() is temporary removed because cd->ci should be marked.
2020-01-07 18:20:36 -05:00
((VALUE)argc << CI_EMBED_ARGC_SHFT) |
((VALUE)flag << CI_EMBED_FLAG_SHFT) |
((VALUE)mid << CI_EMBED_ID_SHFT);
RB_DEBUG_COUNTER_INC(ci_packed);
return (const struct rb_callinfo *)embed_ci;
}
#endif
Introduce disposable call-cache. This patch contains several ideas: (1) Disposable inline method cache (IMC) for race-free inline method cache * Making call-cache (CC) as a RVALUE (GC target object) and allocate new CC on cache miss. * This technique allows race-free access from parallel processing elements like RCU. (2) Introduce per-Class method cache (pCMC) * Instead of fixed-size global method cache (GMC), pCMC allows flexible cache size. * Caching CCs reduces CC allocation and allow sharing CC's fast-path between same call-info (CI) call-sites. (3) Invalidate an inline method cache by invalidating corresponding method entries (MEs) * Instead of using class serials, we set "invalidated" flag for method entry itself to represent cache invalidation. * Compare with using class serials, the impact of method modification (add/overwrite/delete) is small. * Updating class serials invalidate all method caches of the class and sub-classes. * Proposed approach only invalidate the method cache of only one ME. See [Feature #16614] for more details.
2020-01-08 02:14:01 -05:00
VALUE size packed callinfo (ci). Now, rb_call_info contains how to call the method with tuple of (mid, orig_argc, flags, kwarg). Most of cases, kwarg == NULL and mid+argc+flags only requires 64bits. So this patch packed rb_call_info to VALUE (1 word) on such cases. If we can not represent it in VALUE, then use imemo_callinfo which contains conventional callinfo (rb_callinfo, renamed from rb_call_info). iseq->body->ci_kw_size is removed because all of callinfo is VALUE size (packed ci or a pointer to imemo_callinfo). To access ci information, we need to use these functions: vm_ci_mid(ci), _flag(ci), _argc(ci), _kwarg(ci). struct rb_call_info_kw_arg is renamed to rb_callinfo_kwarg. rb_funcallv_with_cc() and rb_method_basic_definition_p_with_cc() is temporary removed because cd->ci should be marked.
2020-01-07 18:20:36 -05:00
const bool debug = 0;
if (debug) fprintf(stderr, "%s:%d ", file, line);
Introduce disposable call-cache. This patch contains several ideas: (1) Disposable inline method cache (IMC) for race-free inline method cache * Making call-cache (CC) as a RVALUE (GC target object) and allocate new CC on cache miss. * This technique allows race-free access from parallel processing elements like RCU. (2) Introduce per-Class method cache (pCMC) * Instead of fixed-size global method cache (GMC), pCMC allows flexible cache size. * Caching CCs reduces CC allocation and allow sharing CC's fast-path between same call-info (CI) call-sites. (3) Invalidate an inline method cache by invalidating corresponding method entries (MEs) * Instead of using class serials, we set "invalidated" flag for method entry itself to represent cache invalidation. * Compare with using class serials, the impact of method modification (add/overwrite/delete) is small. * Updating class serials invalidate all method caches of the class and sub-classes. * Proposed approach only invalidate the method cache of only one ME. See [Feature #16614] for more details.
2020-01-08 02:14:01 -05:00
// TODO: dedup
VALUE size packed callinfo (ci). Now, rb_call_info contains how to call the method with tuple of (mid, orig_argc, flags, kwarg). Most of cases, kwarg == NULL and mid+argc+flags only requires 64bits. So this patch packed rb_call_info to VALUE (1 word) on such cases. If we can not represent it in VALUE, then use imemo_callinfo which contains conventional callinfo (rb_callinfo, renamed from rb_call_info). iseq->body->ci_kw_size is removed because all of callinfo is VALUE size (packed ci or a pointer to imemo_callinfo). To access ci information, we need to use these functions: vm_ci_mid(ci), _flag(ci), _argc(ci), _kwarg(ci). struct rb_call_info_kw_arg is renamed to rb_callinfo_kwarg. rb_funcallv_with_cc() and rb_method_basic_definition_p_with_cc() is temporary removed because cd->ci should be marked.
2020-01-07 18:20:36 -05:00
const struct rb_callinfo *ci = (const struct rb_callinfo *)
rb_imemo_new(imemo_callinfo,
(VALUE)mid,
(VALUE)flag,
(VALUE)argc,
(VALUE)kwarg);
if (debug) rp(ci);
if (kwarg) {
RB_DEBUG_COUNTER_INC(ci_kw);
}
else {
RB_DEBUG_COUNTER_INC(ci_nokw);
}
VM_ASSERT(vm_ci_flag(ci) == flag);
VM_ASSERT(vm_ci_argc(ci) == argc);
return ci;
}
static inline const struct rb_callinfo *
vm_ci_new_runtime_(ID mid, unsigned int flag, unsigned int argc, const struct rb_callinfo_kwarg *kwarg, const char *file, int line)
{
RB_DEBUG_COUNTER_INC(ci_runtime);
return vm_ci_new_(mid, flag, argc, kwarg, file, line);
}
Introduce disposable call-cache. This patch contains several ideas: (1) Disposable inline method cache (IMC) for race-free inline method cache * Making call-cache (CC) as a RVALUE (GC target object) and allocate new CC on cache miss. * This technique allows race-free access from parallel processing elements like RCU. (2) Introduce per-Class method cache (pCMC) * Instead of fixed-size global method cache (GMC), pCMC allows flexible cache size. * Caching CCs reduces CC allocation and allow sharing CC's fast-path between same call-info (CI) call-sites. (3) Invalidate an inline method cache by invalidating corresponding method entries (MEs) * Instead of using class serials, we set "invalidated" flag for method entry itself to represent cache invalidation. * Compare with using class serials, the impact of method modification (add/overwrite/delete) is small. * Updating class serials invalidate all method caches of the class and sub-classes. * Proposed approach only invalidate the method cache of only one ME. See [Feature #16614] for more details.
2020-01-08 02:14:01 -05:00
typedef VALUE (*vm_call_handler)(
struct rb_execution_context_struct *ec,
struct rb_control_frame_struct *cfp,
struct rb_calling_info *calling,
struct rb_call_data *cd);
// imemo_callcache
struct rb_callcache {
* remove trailing spaces. [ci skip]
2020-02-21 19:59:23 -05:00
const VALUE flags;
Introduce disposable call-cache. This patch contains several ideas: (1) Disposable inline method cache (IMC) for race-free inline method cache * Making call-cache (CC) as a RVALUE (GC target object) and allocate new CC on cache miss. * This technique allows race-free access from parallel processing elements like RCU. (2) Introduce per-Class method cache (pCMC) * Instead of fixed-size global method cache (GMC), pCMC allows flexible cache size. * Caching CCs reduces CC allocation and allow sharing CC's fast-path between same call-info (CI) call-sites. (3) Invalidate an inline method cache by invalidating corresponding method entries (MEs) * Instead of using class serials, we set "invalidated" flag for method entry itself to represent cache invalidation. * Compare with using class serials, the impact of method modification (add/overwrite/delete) is small. * Updating class serials invalidate all method caches of the class and sub-classes. * Proposed approach only invalidate the method cache of only one ME. See [Feature #16614] for more details.
2020-01-08 02:14:01 -05:00
/* inline cache: key */
const VALUE klass; // should not mark it because klass can not be free'd
// because of this marking. When klass is collected,
// cc will be cleared (cc->klass = 0) at vm_ccs_free().
/* inline cache: values */
const struct rb_callable_method_entry_struct * const cme_;
const vm_call_handler call_;
union {
const unsigned int attr_index;
const enum method_missing_reason method_missing_reason; /* used by method_missing */
vm_cc_fill() need to clear aux. vm_cc_fill() fills CC information into stack allocated memory so it is not cleared. So we need to clear CC->aux.
2020-03-02 00:25:35 -05:00
VALUE v;
Introduce disposable call-cache. This patch contains several ideas: (1) Disposable inline method cache (IMC) for race-free inline method cache * Making call-cache (CC) as a RVALUE (GC target object) and allocate new CC on cache miss. * This technique allows race-free access from parallel processing elements like RCU. (2) Introduce per-Class method cache (pCMC) * Instead of fixed-size global method cache (GMC), pCMC allows flexible cache size. * Caching CCs reduces CC allocation and allow sharing CC's fast-path between same call-info (CI) call-sites. (3) Invalidate an inline method cache by invalidating corresponding method entries (MEs) * Instead of using class serials, we set "invalidated" flag for method entry itself to represent cache invalidation. * Compare with using class serials, the impact of method modification (add/overwrite/delete) is small. * Updating class serials invalidate all method caches of the class and sub-classes. * Proposed approach only invalidate the method cache of only one ME. See [Feature #16614] for more details.
2020-01-08 02:14:01 -05:00
} aux_;
};
#define VM_CALLCACHE_UNMARKABLE IMEMO_FL_USER0
static inline const struct rb_callcache *
vm_cc_new(VALUE klass,
const struct rb_callable_method_entry_struct *cme,
vm_call_handler call)
{
const struct rb_callcache *cc = (const struct rb_callcache *)rb_imemo_new(imemo_callcache, (VALUE)cme, (VALUE)call, 0, klass);
RB_DEBUG_COUNTER_INC(cc_new);
return cc;
}
static inline const struct rb_callcache *
vm_cc_fill(struct rb_callcache *cc,
VALUE klass,
const struct rb_callable_method_entry_struct *cme,
vm_call_handler call)
{
struct rb_callcache cc_body = {
.flags = T_IMEMO | (imemo_callcache << FL_USHIFT) | VM_CALLCACHE_UNMARKABLE,
.klass = klass,
.cme_ = cme,
.call_ = call,
vm_cc_fill() need to clear aux. vm_cc_fill() fills CC information into stack allocated memory so it is not cleared. So we need to clear CC->aux.
2020-03-02 00:25:35 -05:00
.aux_.v = 0,
Introduce disposable call-cache. This patch contains several ideas: (1) Disposable inline method cache (IMC) for race-free inline method cache * Making call-cache (CC) as a RVALUE (GC target object) and allocate new CC on cache miss. * This technique allows race-free access from parallel processing elements like RCU. (2) Introduce per-Class method cache (pCMC) * Instead of fixed-size global method cache (GMC), pCMC allows flexible cache size. * Caching CCs reduces CC allocation and allow sharing CC's fast-path between same call-info (CI) call-sites. (3) Invalidate an inline method cache by invalidating corresponding method entries (MEs) * Instead of using class serials, we set "invalidated" flag for method entry itself to represent cache invalidation. * Compare with using class serials, the impact of method modification (add/overwrite/delete) is small. * Updating class serials invalidate all method caches of the class and sub-classes. * Proposed approach only invalidate the method cache of only one ME. See [Feature #16614] for more details.
2020-01-08 02:14:01 -05:00
};
MEMCPY(cc, &cc_body, struct rb_callcache, 1);
return cc;
}
static inline bool
vm_cc_class_check(const struct rb_callcache *cc, VALUE klass)
{
VM_ASSERT(IMEMO_TYPE_P(cc, imemo_callcache));
VM_ASSERT(cc->klass == 0 ||
RB_TYPE_P(cc->klass, T_CLASS) || RB_TYPE_P(cc->klass, T_ICLASS));
return cc->klass == klass;
}
static inline const struct rb_callable_method_entry_struct *
vm_cc_cme(const struct rb_callcache *cc)
{
VM_ASSERT(IMEMO_TYPE_P(cc, imemo_callcache));
return cc->cme_;
}
static inline vm_call_handler
vm_cc_call(const struct rb_callcache *cc)
{
VM_ASSERT(IMEMO_TYPE_P(cc, imemo_callcache));
return cc->call_;
}
static inline unsigned int
vm_cc_attr_index(const struct rb_callcache *cc)
{
VM_ASSERT(IMEMO_TYPE_P(cc, imemo_callcache));
return cc->aux_.attr_index;
}
static inline unsigned int
vm_cc_cmethod_missing_reason(const struct rb_callcache *cc)
{
VM_ASSERT(IMEMO_TYPE_P(cc, imemo_callcache));
return cc->aux_.method_missing_reason;
}
static inline int
vm_cc_markable(const struct rb_callcache *cc)
{
VM_ASSERT(IMEMO_TYPE_P(cc, imemo_callcache));
Merge pull request #2991 from shyouhei/ruby.h Split ruby.h
2020-04-08 00:28:13 -04:00
return FL_TEST_RAW((VALUE)cc, VM_CALLCACHE_UNMARKABLE) == 0;
Introduce disposable call-cache. This patch contains several ideas: (1) Disposable inline method cache (IMC) for race-free inline method cache * Making call-cache (CC) as a RVALUE (GC target object) and allocate new CC on cache miss. * This technique allows race-free access from parallel processing elements like RCU. (2) Introduce per-Class method cache (pCMC) * Instead of fixed-size global method cache (GMC), pCMC allows flexible cache size. * Caching CCs reduces CC allocation and allow sharing CC's fast-path between same call-info (CI) call-sites. (3) Invalidate an inline method cache by invalidating corresponding method entries (MEs) * Instead of using class serials, we set "invalidated" flag for method entry itself to represent cache invalidation. * Compare with using class serials, the impact of method modification (add/overwrite/delete) is small. * Updating class serials invalidate all method caches of the class and sub-classes. * Proposed approach only invalidate the method cache of only one ME. See [Feature #16614] for more details.
2020-01-08 02:14:01 -05:00
}
Pin and inline cme in JIT-ed method calls ``` $ benchmark-driver benchmark.yml -v --rbenv 'before --jit;after --jit' --repeat-count=12 --output=all before --jit: ruby 2.8.0dev (2020-03-11T07:43:12Z master e89ebdcb87) +JIT [x86_64-linux] after --jit: ruby 2.8.0dev (2020-03-11T07:54:18Z master 143776a0da) +JIT [x86_64-linux] Calculating ------------------------------------- before --jit after --jit Optcarrot Lan_Master.nes 73.86976729561439 77.20184819316513 fps 74.46997176460742 78.43493030231805 77.59686308754307 78.55714131655935 78.53693921126656 79.08984255596820 80.10158944910573 79.17751731838183 80.12254974411167 79.60853122429181 80.28678655204945 79.74674066871896 80.38690681095379 79.90624544440300 80.79223498756919 80.57881084206193 80.82857188422419 80.70677614429169 81.06447745878245 81.03868541295149 81.21620802278490 82.16354660940607 ```
2020-03-11 03:54:18 -04:00
// For MJIT. cc_cme is supposed to have inlined `vm_cc_cme(cc)`.
Introduce disposable call-cache. This patch contains several ideas: (1) Disposable inline method cache (IMC) for race-free inline method cache * Making call-cache (CC) as a RVALUE (GC target object) and allocate new CC on cache miss. * This technique allows race-free access from parallel processing elements like RCU. (2) Introduce per-Class method cache (pCMC) * Instead of fixed-size global method cache (GMC), pCMC allows flexible cache size. * Caching CCs reduces CC allocation and allow sharing CC's fast-path between same call-info (CI) call-sites. (3) Invalidate an inline method cache by invalidating corresponding method entries (MEs) * Instead of using class serials, we set "invalidated" flag for method entry itself to represent cache invalidation. * Compare with using class serials, the impact of method modification (add/overwrite/delete) is small. * Updating class serials invalidate all method caches of the class and sub-classes. * Proposed approach only invalidate the method cache of only one ME. See [Feature #16614] for more details.
2020-01-08 02:14:01 -05:00
static inline bool
Pin and inline cme in JIT-ed method calls ``` $ benchmark-driver benchmark.yml -v --rbenv 'before --jit;after --jit' --repeat-count=12 --output=all before --jit: ruby 2.8.0dev (2020-03-11T07:43:12Z master e89ebdcb87) +JIT [x86_64-linux] after --jit: ruby 2.8.0dev (2020-03-11T07:54:18Z master 143776a0da) +JIT [x86_64-linux] Calculating ------------------------------------- before --jit after --jit Optcarrot Lan_Master.nes 73.86976729561439 77.20184819316513 fps 74.46997176460742 78.43493030231805 77.59686308754307 78.55714131655935 78.53693921126656 79.08984255596820 80.10158944910573 79.17751731838183 80.12254974411167 79.60853122429181 80.28678655204945 79.74674066871896 80.38690681095379 79.90624544440300 80.79223498756919 80.57881084206193 80.82857188422419 80.70677614429169 81.06447745878245 81.03868541295149 81.21620802278490 82.16354660940607 ```
2020-03-11 03:54:18 -04:00
vm_cc_valid_p(const struct rb_callcache *cc, const rb_callable_method_entry_t *cc_cme, VALUE klass)
Introduce disposable call-cache. This patch contains several ideas: (1) Disposable inline method cache (IMC) for race-free inline method cache * Making call-cache (CC) as a RVALUE (GC target object) and allocate new CC on cache miss. * This technique allows race-free access from parallel processing elements like RCU. (2) Introduce per-Class method cache (pCMC) * Instead of fixed-size global method cache (GMC), pCMC allows flexible cache size. * Caching CCs reduces CC allocation and allow sharing CC's fast-path between same call-info (CI) call-sites. (3) Invalidate an inline method cache by invalidating corresponding method entries (MEs) * Instead of using class serials, we set "invalidated" flag for method entry itself to represent cache invalidation. * Compare with using class serials, the impact of method modification (add/overwrite/delete) is small. * Updating class serials invalidate all method caches of the class and sub-classes. * Proposed approach only invalidate the method cache of only one ME. See [Feature #16614] for more details.
2020-01-08 02:14:01 -05:00
{
VM_ASSERT(IMEMO_TYPE_P(cc, imemo_callcache));
Pin and inline cme in JIT-ed method calls ``` $ benchmark-driver benchmark.yml -v --rbenv 'before --jit;after --jit' --repeat-count=12 --output=all before --jit: ruby 2.8.0dev (2020-03-11T07:43:12Z master e89ebdcb87) +JIT [x86_64-linux] after --jit: ruby 2.8.0dev (2020-03-11T07:54:18Z master 143776a0da) +JIT [x86_64-linux] Calculating ------------------------------------- before --jit after --jit Optcarrot Lan_Master.nes 73.86976729561439 77.20184819316513 fps 74.46997176460742 78.43493030231805 77.59686308754307 78.55714131655935 78.53693921126656 79.08984255596820 80.10158944910573 79.17751731838183 80.12254974411167 79.60853122429181 80.28678655204945 79.74674066871896 80.38690681095379 79.90624544440300 80.79223498756919 80.57881084206193 80.82857188422419 80.70677614429169 81.06447745878245 81.03868541295149 81.21620802278490 82.16354660940607 ```
2020-03-11 03:54:18 -04:00
if (cc->klass == klass && !METHOD_ENTRY_INVALIDATED(cc_cme)) {
Introduce disposable call-cache. This patch contains several ideas: (1) Disposable inline method cache (IMC) for race-free inline method cache * Making call-cache (CC) as a RVALUE (GC target object) and allocate new CC on cache miss. * This technique allows race-free access from parallel processing elements like RCU. (2) Introduce per-Class method cache (pCMC) * Instead of fixed-size global method cache (GMC), pCMC allows flexible cache size. * Caching CCs reduces CC allocation and allow sharing CC's fast-path between same call-info (CI) call-sites. (3) Invalidate an inline method cache by invalidating corresponding method entries (MEs) * Instead of using class serials, we set "invalidated" flag for method entry itself to represent cache invalidation. * Compare with using class serials, the impact of method modification (add/overwrite/delete) is small. * Updating class serials invalidate all method caches of the class and sub-classes. * Proposed approach only invalidate the method cache of only one ME. See [Feature #16614] for more details.
2020-01-08 02:14:01 -05:00
return 1;
}
else {
return 0;
}
}
#ifndef MJIT_HEADER
extern const struct rb_callcache *vm_empty_cc;
#else
extern const struct rb_callcache *rb_vm_empty_cc(void);
#endif
static inline const struct rb_callcache *
vm_cc_empty(void)
{
#ifndef MJIT_HEADER
return vm_empty_cc;
#else
return rb_vm_empty_cc();
#endif
}
/* callcache: mutete */
static inline void
vm_cc_cme_set(const struct rb_callcache *cc, const struct rb_callable_method_entry_struct *cme)
{
VM_ASSERT(IMEMO_TYPE_P(cc, imemo_callcache));
VM_ASSERT(cc != vm_cc_empty());
VM_ASSERT(vm_cc_cme(cc) != NULL);
VM_ASSERT(vm_cc_cme(cc)->called_id == cme->called_id);
VM_ASSERT(!vm_cc_markable(cc)); // only used for vm_eval.c
*((const struct rb_callable_method_entry_struct **)&cc->cme_) = cme;
}
static inline void
vm_cc_call_set(const struct rb_callcache *cc, vm_call_handler call)
{
VM_ASSERT(IMEMO_TYPE_P(cc, imemo_callcache));
VM_ASSERT(cc != vm_cc_empty());
*(vm_call_handler *)&cc->call_ = call;
}
static inline void
vm_cc_attr_index_set(const struct rb_callcache *cc, int index)
{
VM_ASSERT(IMEMO_TYPE_P(cc, imemo_callcache));
VM_ASSERT(cc != vm_cc_empty());
*(int *)&cc->aux_.attr_index = index;
}
static inline void
vm_cc_method_missing_reason_set(const struct rb_callcache *cc, enum method_missing_reason reason)
{
VM_ASSERT(IMEMO_TYPE_P(cc, imemo_callcache));
VM_ASSERT(cc != vm_cc_empty());
*(enum method_missing_reason *)&cc->aux_.method_missing_reason = reason;
}
static inline void
vm_cc_invalidate(const struct rb_callcache *cc)
{
VM_ASSERT(IMEMO_TYPE_P(cc, imemo_callcache));
VM_ASSERT(cc != vm_cc_empty());
VM_ASSERT(cc->klass != 0); // should be enable
*(VALUE *)&cc->klass = 0;
RB_DEBUG_COUNTER_INC(cc_ent_invalidate);
}
/* calldata */
struct rb_call_data {
const struct rb_callinfo *ci;
const struct rb_callcache *cc;
};
struct rb_class_cc_entries {
#if VM_CHECK_MODE > 0
VALUE debug_sig;
#endif
int capa;
int len;
const struct rb_callable_method_entry_struct *cme;
struct rb_class_cc_entries_entry {
const struct rb_callinfo *ci;
const struct rb_callcache *cc;
} *entries;
};
#if VM_CHECK_MODE > 0
static inline bool
vm_ccs_p(const struct rb_class_cc_entries *ccs)
{
return ccs->debug_sig == ~(VALUE)ccs;
}
#endif
// gc.c
void rb_vm_ccs_free(struct rb_class_cc_entries *ccs);
add #include guard hack According to MSVC manual (*1), cl.exe can skip including a header file when that: - contains #pragma once, or - starts with #ifndef, or - starts with #if ! defined. GCC has a similar trick (*2), but it acts more stricter (e. g. there must be _no tokens_ outside of #ifndef...#endif). Sun C lacked #pragma once for a looong time. Oracle Developer Studio 12.5 finally implemented it, but we cannot assume such recent version. This changeset modifies header files so that each of them include strictly one #ifndef...#endif. I believe this is the most portable way to trigger compiler optimizations. [Bug #16770] *1: https://docs.microsoft.com/en-us/cpp/preprocessor/once *2: https://gcc.gnu.org/onlinedocs/cppinternals/Guard-Macros.html
2020-04-10 01:11:40 -04:00
#endif /* RUBY_VM_CALLINFO_H */
Copy permalink