[RFC][PATCH] gc: rewrite GC stop-the-world algorithm.

signal handlers because pthread_mutex_lock() is not reentrant. Using it from a signal handler may result in a deadlock. This approach uses semaphores to notify GC when a thread enters or leaves a safepoint. Signed-off-by: Tomek Grabiec --- arch/x86/include/arch/memory.h| 13 +++ include/vm/gc.h

[PATCH 3/3] gc: rewrite GC stop-the-world algorithm.

handlers because pthread_mutex_lock() is not reentrant. Using it from a signal handler may result in a deadlock. This approach uses semaphores to notify GC when a thread enters or leaves a safepoint. According to the manual sem_post() is the only async-signal safe pthreads method. Signed-off-by: Tomek

[PATCH 2/3] vm: support for disabling thread creation and deletion

We need this to implement GC during which threads should not be created nor deleted. Signed-off-by: Tomek Grabiec --- include/vm/thread.h |2 ++ vm/thread.c | 37 ++--- 2 files changed, 32 insertions(+), 7 deletions(-) diff --git a/include/vm

[PATCH 1/3] x86: make all safepoint instructions do a safepoint poll before.

Signed-off-by: Tomek Grabiec --- arch/x86/insn-selector.brg | 39 ++- 1 files changed, 18 insertions(+), 21 deletions(-) diff --git a/arch/x86/insn-selector.brg b/arch/x86/insn-selector.brg index 630af2f..5dea125 100644 --- a/arch/x86/insn-selector.brg

[PATCH] vm: dump state of all threads on SIGUSR2

When jato receives SIGUSR2 all threads print their register content and backtrace: pkill -SIGUSR2 jato Signed-off-by: Tomek Grabiec --- arch/x86/backtrace.c | 23 +++- include/vm/backtrace.h |1 + vm/signal.c| 52

[PATCH] vm: fix init_stack_trace_elem_current()

ler's frame. We must make it a macro becasue we can't rely on __builtin_frame_address(1) because of gcc optimizations. Signed-off-by: Tomek Grabiec --- include/vm/stack-trace.h |8 +++- test/vm/stack-trace-stub.c |4 vm/stack-trace.c |7 --- 3 f

[PATCH] x86: fix is_sse_insn()

SSE instructions have two prefixes depending on operand size: 0xf3 for 32-bit XMM and 0xf2 for 64-bit XMM. Signed-off-by: Tomek Grabiec --- arch/x86/emit-code.c |2 +- regression/jvm/PutstaticPatchingTest.java | 36 - 2 files changed, 36

[PATCH 1/5] jit: fix get_reload_before_insn()

LIR position passed to radix_tree_lookup() should be always even. Signed-off-by: Tomek Grabiec --- jit/spill-reload.c |5 +++-- 1 files changed, 3 insertions(+), 2 deletions(-) diff --git a/jit/spill-reload.c b/jit/spill-reload.c index 3dfb0b3..eaa2790 100644 --- a/jit/spill-reload.c +++ b

[PATCH 2/5] jit: fix dead code in __insert_spill_reload_insn()

Reported-by: Pekka Enberg Signed-off-by: Tomek Grabiec --- jit/spill-reload.c |2 +- 1 files changed, 1 insertions(+), 1 deletions(-) diff --git a/jit/spill-reload.c b/jit/spill-reload.c index eaa2790..c7b3b51 100644 --- a/jit/spill-reload.c +++ b/jit/spill-reload.c @@ -225,7 +225,7

[PATCH 4/5] jit: set bytecode offset of spill/reload instructions properly

cases a container of insn list head is returned. Therefore we can't get the bytecode offset by dereferencing the result. A valid bytecode offset value is returned by those function by setting a variable to which pointer is passed as an argument. Signed-off-by: Tomek Grabiec --- jit/spill-rel

[PATCH 5/5] jit: cleanup in spill-reload.c

This changes the API and renames get_spill_after_insn() and get_reload_before_insn() so that they do not return a pointer to struct insn, since it is not always a pointer to struct insn. Signed-off-by: Tomek Grabiec --- jit/spill-reload.c | 57

[PATCH 3/5] jit: make API to insert_reload_insn() the same as for insert_spill_insn()

Signed-off-by: Tomek Grabiec --- jit/spill-reload.c | 20 +--- 1 files changed, 9 insertions(+), 11 deletions(-) diff --git a/jit/spill-reload.c b/jit/spill-reload.c index c7b3b51..ba8ac27 100644 --- a/jit/spill-reload.c +++ b/jit/spill-reload.c @@ -168,20 +168,20

[PATCH 2/2] jit: fix dead code in __insert_spill_reload_insn()

The if statement should be an assertion. Reported-by: Pekka Enberg Signed-off-by: Tomek Grabiec --- jit/spill-reload.c |2 +- 1 files changed, 1 insertions(+), 1 deletions(-) diff --git a/jit/spill-reload.c b/jit/spill-reload.c index 7624bd6..f8fd582 100644 --- a/jit/spill-reload.c +++ b

[PATCH 1/2] jit: fix bug in get_spill_after_insn() and get_reload_after_insn()

) and get_reload_after_insn(). Signed-off-by: Tomek Grabiec --- jit/spill-reload.c | 14 +++--- 1 files changed, 7 insertions(+), 7 deletions(-) diff --git a/jit/spill-reload.c b/jit/spill-reload.c index 3dfb0b3..7624bd6 100644 --- a/jit/spill-reload.c +++ b/jit/spill-reload.c @@ -49,11 +

[PATCH 2/2] x86: Add missing result conversion after native method invocations.

ods do not perform the conversion. Therefore it is now safe to use jbyte, jboolean, jchar and jshort as VM native implementation return type. Signed-off-by: Tomek Grabiec --- arch/x86/insn-selector.brg | 32 1 files changed, 32 insertions(+), 0 deletions(-) di

[PATCH 1/2] mmix: fix compilation error on make check

Signed-off-by: Tomek Grabiec --- arch/mmix/call.c |6 +- 1 files changed, 1 insertions(+), 5 deletions(-) diff --git a/arch/mmix/call.c b/arch/mmix/call.c index 1dc640a..9c3f244 100644 --- a/arch/mmix/call.c +++ b/arch/mmix/call.c @@ -27,10 +27,6 @@ #include "vm/method.h"

[PATCH] jit: fix conversion of dup_x2 when second value is J_LONG or J_DOUBLE

Converter for dup_x2 didn't check the stack size of second stack operand which caused that the forst value was put one level deeper when second value was J_LONG or J_DOUBLE. Signed-off-by: Tomek Grabiec --- jit/ostack-bc.c |4 regression/jvm/DupTest.j |

[PATCH 2/2] x86, vm: implement support for invoking methods with J_LONG return type in reflection.

Signed-off-by: Tomek Grabiec --- arch/x86/call.c | 50 ++ include/vm/types.h |5 +++ regression/java/lang/reflect/MethodTest.java |5 +++ runtime/reflection.c |4 ++- 4 files

[PATCH 1/2] x86: cleanup in call.c

Signed-off-by: Tomek Grabiec --- arch/x86/call.c | 180 ++--- include/vm/call.h |6 +-- vm/call.c |6 -- 3 files changed, 62 insertions(+), 130 deletions(-) diff --git a/arch/x86/call.c b/arch/x86/call.c index 0b96020..389b1a6

[PATCH 3/3] vm: implement VMClassLoader.defineClass()

Signed-off-by: Tomek Grabiec --- include/runtime/classloader.h |1 + include/vm/class.h|1 + include/vm/classloader.h |1 + include/vm/preload.h |1 + runtime/classloader.c | 40 +++- test/vm/preload-stub.c

[PATCH 2/3] vm: make class cache recognize different classloaders.

Two different classloaders may load a class of the same name. We must therefore make a classloader reference a part of class cache key. Signed-off-by: Tomek Grabiec --- include/lib/hash-map.h |1 + include/vm/classloader.h |2 +- lib/hash-map.c |5 +++ runtime

[PATCH 1/3] vm: fix loading of array classes.

different classloaders. The former is always loaded with bootstrap classloader (primitive array) while loading of the latter can be initiated with any classloader. Class [[I must be eventually loaded with bootstrap classloader too. Signed-off-by: Tomek Grabiec --- include/vm/types.h |6 ++ vm

[PATCH] runtime: move VMClassLoader natives to runtime/classloader.c

Signed-off-by: Tomek Grabiec --- Makefile |1 + include/runtime/classloader.h | 10 runtime/classloader.c | 97 + vm/jato.c | 66 +--- 4 files changed, 109 insertions

[PATCH 2/3] jit: handle exception handlers covering subroutines in subroutine inlining.

The code didn't handle the case where exception handler range covers the whole subroutine and more. Signed-off-by: Tomek Grabiec --- jit/subroutine.c | 180 ++ 1 files changed, 127 insertions(+), 53 deletions(-) diff --git a/jit/subrout

[PATCH 3/3] vm: do not return jboolean from VM natives.

Signed-off-by: Tomek Grabiec --- include/runtime/class.h |6 +++--- runtime/class.c |6 +++--- vm/jato.c |6 +++--- 3 files changed, 9 insertions(+), 9 deletions(-) diff --git a/include/runtime/class.h b/include/runtime/class.h index 50d1ec0..7691ac9 100644 --- a/

[PATCH 1/3] jit: fix print_exception_table()

Signed-off-by: Tomek Grabiec --- jit/exception.c |2 +- 1 files changed, 1 insertions(+), 1 deletions(-) diff --git a/jit/exception.c b/jit/exception.c index b412e25..8274eaa 100644 --- a/jit/exception.c +++ b/jit/exception.c @@ -352,7 +352,7 @@ print_exception_table(const struct vm_method

[PATCH] vm: fix bug in native_field_get()

Signed-off-by: Tomek Grabiec --- vm/reflection.c |2 +- 1 files changed, 1 insertions(+), 1 deletions(-) diff --git a/vm/reflection.c b/vm/reflection.c index d97fec9..faf27bd 100644 --- a/vm/reflection.c +++ b/vm/reflection.c @@ -621,7 +621,7 @@ struct vm_object *native_field_get(struct

[PATCH 3/4] x86: make native_call() hard crash for not implemented types.

Signed-off-by: Tomek Grabiec --- arch/x86/call.c |8 1 files changed, 4 insertions(+), 4 deletions(-) diff --git a/arch/x86/call.c b/arch/x86/call.c index 717d53d..0b96020 100644 --- a/arch/x86/call.c +++ b/arch/x86/call.c @@ -137,7 +137,7 @@ void native_call(struct vm_method

[PATCH 2/4] regression: fix TestCase.assertObjectEquals()

Signed-off-by: Tomek Grabiec --- regression/jvm/TestCase.java |2 +- 1 files changed, 1 insertions(+), 1 deletions(-) diff --git a/regression/jvm/TestCase.java b/regression/jvm/TestCase.java index fd00063..2a53a8b 100644 --- a/regression/jvm/TestCase.java +++ b/regression/jvm/TestCase.java

[PATCH 1/4] vm: fix native_vmclass_get_declared_methods()

We should perform the vm_method_is_special() check also when public_only == false; Signed-off-by: Tomek Grabiec --- vm/reflection.c | 23 ++- 1 files changed, 10 insertions(+), 13 deletions(-) diff --git a/vm/reflection.c b/vm/reflection.c index 6d03a57..d97fec9 100644

[PATCH 4/4] regression: add regression test for method invokation via reflection

Signed-off-by: Tomek Grabiec --- regression/jvm/MethodTest.java | 20 +++- 1 files changed, 19 insertions(+), 1 deletions(-) diff --git a/regression/jvm/MethodTest.java b/regression/jvm/MethodTest.java index c633f8c..b805da4 100644 --- a/regression/jvm/MethodTest.java +++ b

[PATCH 3/4] vm: rename encapsulate_value() to wrap()

While at it, make it more robust by using valueOf() functions. Signed-off-by: Tomek Grabiec --- include/vm/preload.h |8 +++ vm/preload.c | 56 ++ vm/reflection.c | 153 -- 3 files changed, 124 insertions(+), 93

[PATCH 4/4] vm: fix call_static_method() and call_virtual_method()

Those functions should return values wrapped in approptiate objects, not the value itself. This caused a SIGSEGV when running clojure because an integer value was accessed like an object reference. Signed-off-by: Tomek Grabiec --- vm/reflection.c |4 ++-- 1 files changed, 2 insertions(+), 2

[PATCH 1/4] vm: change call result return method for vm_call_method_*() so that it can handle any type

saved. This way we can use vm_call_method_*() functions in a machine and type independed way. While at it, implement JNI funciton families: CallStatic*Method() CallStatic*MethodV() Call*Method() Signed-off-by: Tomek Grabiec --- arch/x86/call.c| 222

[PATCH 2/4] vm: implement unwrap() for J_LONG and J_DOUBLE

Signed-off-by: Tomek Grabiec --- include/vm/preload.h |2 ++ vm/preload.c | 14 ++ vm/reflection.c |7 ++- 3 files changed, 22 insertions(+), 1 deletions(-) diff --git a/include/vm/preload.h b/include/vm/preload.h index c71d8f9..a99fa89 100644 --- a

[PATCH] jit: skip conversion of unreachable basic blocks

[main][ 359 ] 0xb0 areturn Signed-off-by: Tomek Grabiec --- jit/bytecode-to-ir.c |3 +++ 1 files changed, 3 insertions(+), 0 deletions(-) diff --git a/jit/bytecode-to-ir.c b/jit/bytecode-to-ir.c index a93a2ca..5f66587 100644 --- a/jit/bytecode-to-ir.c +++ b/jit/bytecode-to-ir.c @@ -434,6

[PATCH 2/3] vm: do not return and methods in native_vmclass_get_declared_methods()

The API spec says we shouldn't do so. Signed-off-by: Tomek Grabiec --- include/vm/method.h |5 + vm/reflection.c |6 +- 2 files changed, 10 insertions(+), 1 deletions(-) diff --git a/include/vm/method.h b/include/vm/method.h index ccf1db8..7c9e0d7 100644 --- a/inclu

[PATCH 1/3] vm: rename unwrap_and_set_field() to unwrap()

Signed-off-by: Tomek Grabiec --- vm/reflection.c |8 1 files changed, 4 insertions(+), 4 deletions(-) diff --git a/vm/reflection.c b/vm/reflection.c index c7fc9e5..d0bff58 100644 --- a/vm/reflection.c +++ b/vm/reflection.c @@ -624,7 +624,7 @@ jint

[PATCH 3/3] jit: fix bug in spill_interval()

due to register corruption, which prevented java.awt.image.Raster.createWritableRaster() from correct working. Signed-off-by: Tomek Grabiec --- jit/linear-scan.c | 14 ++ 1 files changed, 14 insertions(+), 0 deletions(-) diff --git a/jit/linear-scan.c b/jit/linear-scan.c index

[PATCH] vm: fix unwrap_and_set_field()

'This' pointer was not passed to vm_call_method_this_a() in args which caused a SIGSEGV inside that function. Signed-off-by: Tomek Grabiec --- vm/call.c |1 + vm/reflection.c |6 -- 2 files changed, 5 insertions(+), 2 deletions(-) diff --git a/vm/call.c b/vm/ca

[PATCH] vm: fix call_virtual_method()

'This' pointer must be added to args manually. vm_call_method_this_a() expects that it is already in args to avoid unnecessary copying. Signed-off-by: Tomek Grabiec --- vm/call.c |6 +- vm/reflection.c |3 ++- 2 files changed, 7 insertions(+), 2 deletions(-) diff

[PATCH] cafebabe: fix bug in cafebabe_stream_close()

The bracket was misplaced causing close() to use wrong file descriptor. This led to the following error message for tetris: .: Fatal IO error 9 (Bad file descriptor) on X server :0.0. CC: Vegard Nossum Signed-off-by: Tomek Grabiec --- cafebabe/src/cafebabe/stream.c |2 +- 1 files changed

[PATCH 3/5] vm: implement all GetStatic*Field() JNI functions

Signed-off-by: Tomek Grabiec --- vm/jni-interface.c | 52 ++-- 1 files changed, 26 insertions(+), 26 deletions(-) diff --git a/vm/jni-interface.c b/vm/jni-interface.c index b3b205e..ac8188f 100644 --- a/vm/jni-interface.c +++ b/vm/jni

[PATCH 5/5] vm: fix pack_args()

The condition should be negated. Signed-off-by: Tomek Grabiec --- vm/jni-interface.c |6 +++--- 1 files changed, 3 insertions(+), 3 deletions(-) diff --git a/vm/jni-interface.c b/vm/jni-interface.c index cc1e7ca..dca174a 100644 --- a/vm/jni-interface.c +++ b/vm/jni-interface.c @@ -912,12

[PATCH 2/5] vm: fix vm_jni_set_static_*()

Those functions were using non-static field setters on a class object, which is incorrect. Signed-off-by: Tomek Grabiec --- vm/jni-interface.c | 20 ++-- 1 files changed, 10 insertions(+), 10 deletions(-) diff --git a/vm/jni-interface.c b/vm/jni-interface.c index 7e502b7

[PATCH 4/5] vm: fix vm_jni_new_object_a()

We should call the constructor virtually not statically. Signed-off-by: Tomek Grabiec --- vm/jni-interface.c |2 +- 1 files changed, 1 insertions(+), 1 deletions(-) diff --git a/vm/jni-interface.c b/vm/jni-interface.c index ac8188f..cc1e7ca 100644 --- a/vm/jni-interface.c +++ b/vm/jni

[PATCH 1/5] vm: fix vm_jni_get_static_double_field()

0x1EEF47A7: ??? ==6229==by 0x1EEF47A7: ??? ==6229== Address 0x1f0d51f8 is not stack'd, malloc'd or (recently) free'd Signed-off-by: Tomek Grabiec --- vm/jni-interface.c |2 +- 1 files changed, 1 insertions(+), 1 deletions(-) diff --git a/vm/jni-interface.c b/vm/jni-in

[PATCH 4/5] vm: implement thread interruption operations.

The following natives are implemented: java/lang/VMThread.isInterrupted() java/lang/VMThread.interrupted() java/lang/VMThread.interrupt() Signed-off-by: Tomek Grabiec --- Makefile|1 + include/vm/preload.h|1 + include/vm/thread.h |7

[PATCH 5/5] x86: introduce valgrind workaround for exception guards.

ntrol to the exception handler or unwind. The workaround is to inline exception check and don't use signal handlers when run on valgrind. Signed-off-by: Tomek Grabiec --- arch/x86/insn-selector.brg |9 +++-- arch/x86/unwind_32.S | 27 +++ include/jit/e

[PATCH 1/5] vm: fix parse_method_type()

Call arguments were incorrectly inserted to args list causing the list to represent the reverse order of arguments. Signed-off-by: Tomek Grabiec --- vm/types.c |2 +- 1 files changed, 1 insertions(+), 1 deletions(-) diff --git a/vm/types.c b/vm/types.c index e20aecf..f4fe84d 100644 --- a

[PATCH 3/5] vm: unify code in vm_monitor_wait() and vm_monitor_timedwait()

Signed-off-by: Tomek Grabiec --- vm/monitor.c | 78 + 1 files changed, 34 insertions(+), 44 deletions(-) diff --git a/vm/monitor.c b/vm/monitor.c index 4a22584..5705954 100644 --- a/vm/monitor.c +++ b/vm/monitor.c @@ -141,10 +141,9

[PATCH 2/5] vm: move monitor operations to vm/monitor.c

Signed-off-by: Tomek Grabiec --- Makefile |1 + include/vm/object.h|1 + test/arch-x86/Makefile |1 + vm/jato.c |1 + vm/monitor.c | 242 vm/object.c| 191

[PATCH 1/2] vm: parse method and field type on initialization

sulated in struct vm_method_arg and linked in a list pointed by struct vm_method.args. This change removes plenty of calls to type string parsing which should not be done at run-time but rather on method or field initialization. Signed-off-by: Tomek Grabiec --- include/jit/expression.h

[PATCH 2/2] vm: implement java/lang/reflect/Method.getReturnType()

Signed-off-by: Tomek Grabiec --- include/vm/reflection.h |1 + vm/jato.c |1 + vm/reflection.c | 18 ++ 3 files changed, 20 insertions(+), 0 deletions(-) diff --git a/include/vm/reflection.h b/include/vm/reflection.h index f9c709a..c1e3aa4 100644

[PATCH] README: fix package dependencies for Ubuntu

We need valgrind headers to compile jato, since valgrind workarounds were introduced. Signed-off-by: Tomek Grabiec --- README |2 +- 1 files changed, 1 insertions(+), 1 deletions(-) diff --git a/README b/README index 03985d6..97a9467 100644 --- a/README +++ b/README @@ -13,7 +13,7 @@ How

[PATCH] jit: fix race condition in fixup_direct_calls()

0x88D73CC: ??? Signed-off-by: Tomek Grabiec --- arch/x86/emit-code.c | 12 +++- arch/x86/insn-selector.brg |6 +++--- include/jit/compilation-unit.h |1 + include/jit/compiler.h | 11 +-- jit/compilation-unit.c | 18 ++

Re: [PATCH 15/19] x86-64: move received parameters to non-fixed registers

CE because it could cause that integers would be interpreted as pointers. I think the proper solution would be to iterate over method's call argument types, and for argument type T we should allocate register of type mimic_stack_type(T). I also think that we only have to emit movs for registers

[PATCH] jit: fix conversion of invoke* instructions

. Invocations are now handled as statements and their result is pushed onto mimic stack. Signed-off-by: Tomek Grabiec --- Makefile |5 +- arch/x86/insn-selector.brg | 236 ++ include/jit/expression.h | 48 +- include/jit

[PATCH 4/4] jit: fix mimic stack content propagation

oblems with GC. Signed-off-by: Tomek Grabiec --- include/jit/basic-block.h |9 include/lib/stack.h | 11 +- jit/basic-block.c |1 + jit/bytecode-to-ir.c | 50 lib/stack.c | 17 +++

[PATCH 1/4] jit: make clear_mimic_stack() work on stack instead of basic block

Signed-off-by: Tomek Grabiec --- include/jit/basic-block.h |2 +- jit/basic-block.c |6 +++--- jit/exception-bc.c|2 +- jit/invoke-bc.c |4 ++-- 4 files changed, 7 insertions(+), 7 deletions(-) diff --git a/include/jit/basic-block.h b/include/jit/basic

[PATCH 2/4] jit: split tableswitch/lookupswitch bsaic block at the end.

This way, the order of basic blocks matches the values of .start and .end It also makes created basic blocks have empty bytecode range (start == end). Signed-off-by: Tomek Grabiec --- jit/basic-block.c |2 +- jit/switch-bc.c |6 +++--- test/jit/basic-block-test.c

[PATCH 3/4] jit: set .has_branch flag for created basic blocks for tableswitch/lookupswitch.

We must set that flag so that mimic stack resolution will put moves before branch instructions. Signed-off-by: Tomek Grabiec --- jit/switch-bc.c |7 --- 1 files changed, 4 insertions(+), 3 deletions(-) diff --git a/jit/switch-bc.c b/jit/switch-bc.c index b1ece1a..fb3fef2 100644 --- a

[PATCH 2/2][SQUASH TO 4/4] lib: fix compilation error

Signed-off-by: Tomek Grabiec --- include/lib/stack.h |2 ++ 1 files changed, 2 insertions(+), 0 deletions(-) diff --git a/include/lib/stack.h b/include/lib/stack.h index 7ec599d..aa60109 100644 --- a/include/lib/stack.h +++ b/include/lib/stack.h @@ -7,6 +7,8 @@ #include #include

[PATCH 1/2][SQUASH to 2/4] test: fix 'make check' breakage

Signed-off-by: Tomek Grabiec --- test/jit/basic-block-test.c |2 +- 1 files changed, 1 insertions(+), 1 deletions(-) diff --git a/test/jit/basic-block-test.c b/test/jit/basic-block-test.c index 7195487..a22d3bf 100644 --- a/test/jit/basic-block-test.c +++ b/test/jit/basic-block-test.c

[PATCH 3/4] jit: set .has_branch flag for created basic blocks for tableswitch/lookupswitch.

We must set that flag so that mimic stack resolution will put moves before branch instructions. Signed-off-by: Tomek Grabiec --- jit/switch-bc.c |7 --- 1 files changed, 4 insertions(+), 3 deletions(-) diff --git a/jit/switch-bc.c b/jit/switch-bc.c index b1ece1a..fb3fef2 100644 --- a

[PATCH 4/4] jit: fix mimic stack content propagation

is defined. The register appears to be always live before it's use position, which is wrong, and could cause problems with GC. Signed-off-by: Tomek Grabiec --- include/jit/basic-block.h |3 ++ include/lib/stack.h | 24 +- jit/bytecode

[PATCH 2/4] jit: split tableswitch/lookupswitch bsaic block at the end.

This way, the order of basic blocks matches the values of .start and .end It also makes created basic blocks have empty bytecode range (start == end). Signed-off-by: Tomek Grabiec --- jit/basic-block.c |2 +- jit/switch-bc.c |6 +++--- 2 files changed, 4 insertions(+), 4 deletions

[PATCH 1/4] jit: make clear_mimic_stack() work on stack instead of basic block

Signed-off-by: Tomek Grabiec --- include/jit/basic-block.h |2 +- jit/basic-block.c |6 +++--- jit/exception-bc.c|2 +- jit/invoke-bc.c |4 ++-- 4 files changed, 7 insertions(+), 7 deletions(-) diff --git a/include/jit/basic-block.h b/include/jit/basic

Re: [PATCH 1/6] vm: introduce -Xnobuf

2009/9/1 Pekka Enberg : > Hi Tomek, > > On Tue, 2009-09-01 at 10:06 +0200, Tomek Grabiec wrote: >> It disables buffering of strings passed to trace_printf(). >> >> Signed-off-by: Tomek Grabiec > > Does something like this work for your scenario? > >

[PATCH 6/6] x86: fix writes below (%esp) in insn-selector.brg

It is incorrect to write or read from memory below %esp. This patch fixes valgrind complaints about access of uninitilized data. Signed-off-by: Tomek Grabiec --- arch/x86/insn-selector.brg | 152 +++- 1 files changed, 80 insertions(+), 72 deletions

[PATCH 3/6] x86: cleanup emit_mov_reg_memlocal()

Signed-off-by: Tomek Grabiec --- arch/x86/emit-code.c | 16 +--- 1 files changed, 1 insertions(+), 15 deletions(-) diff --git a/arch/x86/emit-code.c b/arch/x86/emit-code.c index a99296c..e1eb5ba 100644 --- a/arch/x86/emit-code.c +++ b/arch/x86/emit-code.c @@ -861,21 +861,7

[PATCH 4/6] x86: implement INSN_(FSTP|FLD)_{64}_MEMLOCAL

Signed-off-by: Tomek Grabiec --- arch/x86/emit-code.c| 24 arch/x86/include/arch/instruction.h |4 arch/x86/lir-printer.c | 28 arch/x86/use-def.c | 12 4 files

[PATCH 5/6] jit: introduce lazy-allocated scratch stack slots

They are needed to hold temporary results when register can not be used. Signed-off-by: Tomek Grabiec --- include/jit/compilation-unit.h | 20 +++- jit/compilation-unit.c | 16 2 files changed, 35 insertions(+), 1 deletions(-) diff --git a/include

[PATCH 1/6] vm: introduce -Xnobuf

It disables buffering of strings passed to trace_printf(). Signed-off-by: Tomek Grabiec --- include/vm/trace.h |2 ++ vm/jato.c |7 +++ vm/trace.c |9 - 3 files changed, 17 insertions(+), 1 deletions(-) diff --git a/include/vm/trace.h b/include/vm

[PATCH 2/6] x86: implement INSN_MOV_IMM_MEMLOCAL

Signed-off-by: Tomek Grabiec --- arch/x86/emit-code.c|9 - arch/x86/include/arch/instruction.h |2 ++ arch/x86/instruction.c | 12 arch/x86/lir-printer.c | 14 ++ arch/x86/use-def.c |1

[PATCH] x86: fix writes below (%esp) which can be fixed at no cost

Signed-off-by: Tomek Grabiec --- arch/x86/insn-selector.brg | 30 -- 1 files changed, 16 insertions(+), 14 deletions(-) diff --git a/arch/x86/insn-selector.brg b/arch/x86/insn-selector.brg index 2e7367f..85f2fa6 100644 --- a/arch/x86/insn-selector.brg +++ b/arch

[PATCH] [APPEND-TO-LAST-SERIES] jit: cleanup handling of expired ranges of intervals

liveness(). Signed-off-by: Tomek Grabiec --- include/jit/vars.h | 18 -- jit/interval.c | 41 +++-- jit/linear-scan.c | 15 +++ 3 files changed, 42 insertions(+), 32 deletions(-) diff --git a/include/jit/vars.h b/include/jit/vars

[PATCH] x86: Introduce workarounds for valgrind to work with jato.

Jato can be compiled with workarounds which make valgrind work with jato. To do so, define VALGRIND variable for make: make jato VALGRIND=y Currently workarounds eliminate class initialization from signal handler by unconditionally selecting calls to vm_class_ensure_init(). Signed-off-by: Tomek

[PATCH 13/14] jit: optimize register use and block position calculation in regalloc

those registers. For example, we do not have to check at which position XMM0 register is available if we are allocating a general purpose register. Signed-off-by: Tomek Grabiec --- jit/linear-scan.c |9 + 1 files changed, 9 insertions(+), 0 deletions(-) diff --git a/jit/linear-scan.c

[PATCH 12/14] jit: do not put to inactive list fixed reg intervals for ESP and EBP

Those register are not considered for allocation and their numbers are > NR_REGISTERS. Letting those fixed intervals into register allocator can cause memory corruption becuase use position arrays are of size NR_REGISTERS. Signed-off-by: Tomek Grabiec --- jit/linear-scan.c |9 ++---

[PATCH 10/14] jit: print variable types in regalloc trace

Signed-off-by: Tomek Grabiec --- jit/trace-jit.c |1 + 1 files changed, 1 insertions(+), 0 deletions(-) diff --git a/jit/trace-jit.c b/jit/trace-jit.c index 58ecaa4..915571d 100644 --- a/jit/trace-jit.c +++ b/jit/trace-jit.c @@ -360,6 +360,7 @@ void trace_regalloc(struct compilation_unit

[PATCH 11/14] jit: fix spilling of 64-bit registers.

This cleans handling of 64-bit stack slots up and fixes the buggy spilling code. We always allocated 32-bit spill slot regardless of register type which caused memory corruption. Signed-off-by: Tomek Grabiec --- arch/mmix/include/arch/instruction.h | 15 +- arch/mmix/instruction.c

[PATCH 14/14] x86: remove unconditional saving and restoring of XMM registers

We do not longer need to do this because this bug has been solved: http://jato.lighthouseapp.com/projects/29055/tickets/5-sse-registers-are-saved-and-registered-unconditionally Signed-off-by: Tomek Grabiec --- arch/x86/emit-code.c| 83 --- arch

[PATCH 06/14] jit: implement precise live range calculation

; this can be optimized out in the future add ebx, ebx Signed-off-by: Tomek Grabiec --- arch/x86/include/arch/instruction.h |5 -- arch/x86/instruction.c |5 ++ arch/x86/use-def.c | 28 +++ include/jit/instruction.h |6 +++ i

[PATCH 09/14] jit: ensure that spill variable has the same vm_type as original variable.

This is a bug fix. The bug caused that floating point variables were spilled as if they were a general purpose registers which led to corruption of general purpose registers. Signed-off-by: Tomek Grabiec --- jit/compilation-unit.c |4 ++-- jit/interval.c |1 + jit/spill-reload.c

[PATCH 08/14] jit: force spill of intervals containing caller saved registers before calls.

This fixes the follwing bug: http://jato.lighthouseapp.com/projects/29055/tickets/7-ebcdx-are-unavailable-for-allocation-after-some-call-instructions Signed-off-by: Tomek Grabiec --- arch/mmix/include/arch/instruction.h |5 + arch/mmix/include/arch/registers.h |3 +++ arch/mmix

[PATCH 04/14] jit: introduce multiple live ranges per interval.

This is needed for precise modeling of live ranges. Signed-off-by: Tomek Grabiec --- include/jit/vars.h | 83 +--- jit/interval.c | 228 +- jit/linear-scan.c | 51 ++ jit/liveness.c

[PATCH 05/14] jit: move arch independent stuff from arch/instruction.h to jit/instruction.h

Signed-off-by: Tomek Grabiec --- arch/mmix/include/arch/instruction.h |8 arch/x86/include/arch/instruction.h |8 arch/x86/instruction.c |3 +-- arch/x86/use-def.c |2 +- include/jit/instruction.h| 10

[PATCH 07/14] x86: ensure fixed-reg variables are not returned as rule results

t be used in place of regular virtual registers. Signed-off-by: Tomek Grabiec --- arch/x86/insn-selector.brg | 134 ++-- 1 files changed, 79 insertions(+), 55 deletions(-) diff --git a/arch/x86/insn-selector.brg b/arch/x86/insn-selector.brg index 8522667..

[PATCH 03/14] jit: cleanup interval spilling

Signed-off-by: Tomek Grabiec --- jit/linear-scan.c | 86 +++-- 1 files changed, 37 insertions(+), 49 deletions(-) diff --git a/jit/linear-scan.c b/jit/linear-scan.c index 5538bc7..8baa914 100644 --- a/jit/linear-scan.c +++ b/jit/linear-scan.c

[PATCH 01/14] jit: add missing trace_flush() to trace_return_value()

Signed-off-by: Tomek Grabiec --- jit/trace-jit.c |4 +++- 1 files changed, 3 insertions(+), 1 deletions(-) diff --git a/jit/trace-jit.c b/jit/trace-jit.c index 0246c3b..1bde5e1 100644 --- a/jit/trace-jit.c +++ b/jit/trace-jit.c @@ -811,8 +811,10 @@ void trace_return_value(struct vm_method

[PATCH 02/14] jit: assign two LIR positions for each instruction.

positions are even yet. Signed-off-by: Tomek Grabiec --- include/jit/compilation-unit.h |2 + include/jit/vars.h | 10 + jit/compilation-unit.c |4 ++- jit/liveness.c |4 +- jit/spill-reload.c | 16 +++-- jit/t

[PATCH 3/3] x86: fix wrong argument cleanup count for EXPR_ANEWARRAY

Selected code was adding 4 bytes too much to ESP, which led to a memory corruption. Signed-off-by: Tomek Grabiec --- arch/x86/insn-selector.brg |2 +- 1 files changed, 1 insertions(+), 1 deletions(-) diff --git a/arch/x86/insn-selector.brg b/arch/x86/insn-selector.brg index d7f506b

[PATCH 2/3] jit: remove redundant compute_boundaries()

The values of .start_insn and .end_insn can be computed in compute_insn_positions(). Signed-off-by: Tomek Grabiec --- jit/compilation-unit.c |4 jit/liveness.c | 17 - 2 files changed, 4 insertions(+), 17 deletions(-) diff --git a/jit/compilation-unit.c b

[PATCH 1/3] jit: remove unused basic block sorting code

Signed-off-by: Tomek Grabiec --- include/jit/compilation-unit.h |1 - jit/compilation-unit.c | 15 --- 2 files changed, 0 insertions(+), 16 deletions(-) diff --git a/include/jit/compilation-unit.h b/include/jit/compilation-unit.h index 66aef32..f6fb0e9 100644 --- a

[PATCH] jit: remove trace_flush() from the middle of compile()

Signed-off-by: Tomek Grabiec --- jit/compiler.c |2 +- 1 files changed, 1 insertions(+), 1 deletions(-) diff --git a/jit/compiler.c b/jit/compiler.c index 645e9b2..1fa321a 100644 --- a/jit/compiler.c +++ b/jit/compiler.c @@ -66,7 +66,7 @@ int compile(struct compilation_unit *cu

[RFC][PATCH 3/3] jit: fix data flow resolution code in presence of empty basic blocks.

before the slot was assigned a correct value. Empty basic blocks can be a result of the following bytecode: iconst_0 iconst_1 ifeq lab pop // This and the next instruction generate no iconst_1 // LIR instructions. lab: Cc: Arthur HUILLET Signed-off-by: Tomek Grabiec --- i

[PATCH 2/3] x86: make insn_is_branch() return true for INSN_RET

INSN_RET should be considered a branch because it jumps to different basic block. It is used in LIR for athrow. Signed-off-by: Tomek Grabiec --- arch/x86/include/arch/instruction.h |1 + 1 files changed, 1 insertions(+), 0 deletions(-) diff --git a/arch/x86/include/arch/instruction.h b

[PATCH 1/3] jit: introduce include/jit/instruction.h for arch-independent stuff

Signed-off-by: Tomek Grabiec --- include/jit/instruction.h | 12 1 files changed, 12 insertions(+), 0 deletions(-) create mode 100644 include/jit/instruction.h diff --git a/include/jit/instruction.h b/include/jit/instruction.h new file mode 100644 index 000..376e278

Re: [PATCH 3/3] spill-reload: Use radix_tree_lookup() in insert_mov_insns()

2009/8/19 Pekka Enberg : > Hi Tomek, > > On Tue, 2009-08-18 at 21:21 +0200, Tomek Grabiec wrote: >> Like I said on IRC, this will not work for empty basic blocks, because >> spill_at_insn will belong to the preceding >> basic block. This causes that instructions

  1   2   3   4   5   6   7   8   9   >