vm: strip out call-counting profiler

This makes the separate "code" and "entry_point" fields in word and quotation redundant, so also remove them to reclaim an additional cell per word and quotation object, which should help with #318.

GNUmakefile

@@ -34,7 +34,6 @@ ifdef CONFIG
 		vm/code_heap.o \
 		vm/compaction.o \
 		vm/contexts.o \
-		vm/counting_profiler.o \
 		vm/data_heap.o \
 		vm/data_heap_checker.o \
 		vm/debug.o \
@@ -77,7 +76,6 @@ ifdef CONFIG
 		vm/contexts.hpp \
 		vm/run.hpp \
 		vm/objects.hpp \
-		vm/counting_profiler.hpp \
 		vm/sampling_profiler.hpp \
 		vm/errors.hpp \
 		vm/bignumint.hpp \

Nmakefile

@@ -37,7 +37,6 @@ DLL_OBJS = $(PLAF_DLL_OBJS) \
 	vm\code_heap.obj \
 	vm\compaction.obj \
 	vm\contexts.obj \
-	vm\counting_profiler.obj \
 	vm\data_heap.obj \
 	vm\data_heap_checker.obj \
 	vm\debug.obj \

basis/bootstrap/image/image.factor

@@ -350,11 +350,8 @@ M: f ' drop \ f type-number ;
                     [ props>> , ]
                     [ pic-def>> , ]
                     [ pic-tail-def>> , ]
-                    [ drop 0 , ] ! count
                     [ word-sub-primitive , ]
-                    [ drop 0 , ] ! xt
-                    [ drop 0 , ] ! code
-                    [ drop 0 , ] ! profiling
+                    [ drop 0 , ] ! entry point
                 } cleave
             ] { } make [ ' ] map
         ] bi
@@ -493,8 +490,7 @@ M: quotation '
             emit ! array
             f ' emit ! cached-effect
             f ' emit ! cache-counter
-            0 emit ! xt
-            0 emit ! code
+            0 emit ! entry point
         ] emit-object
     ] cache-eql-object ;
 

basis/compiler/constants/constants.factor

@@ -20,9 +20,9 @@ CONSTANT: deck-bits 18
 : alien-offset ( -- n ) 4 alien type-number slot-offset ; inline
 : underlying-alien-offset ( -- n ) 1 alien type-number slot-offset ; inline
 : tuple-class-offset ( -- n ) 1 tuple type-number slot-offset ; inline
-: word-entry-point-offset ( -- n ) 10 \ word type-number slot-offset ; inline
+: word-entry-point-offset ( -- n ) 9 \ word type-number slot-offset ; inline
 : quot-entry-point-offset ( -- n ) 4 quotation type-number slot-offset ; inline
-: word-code-offset ( -- n ) 11 \ word type-number slot-offset ; inline
+: code-block-entry-point-offset ( -- n ) 4 bootstrap-cells ; inline
 : array-start-offset ( -- n ) 2 array type-number slot-offset ; inline
 : compiled-header-size ( -- n ) 4 bootstrap-cells ; inline
 : callstack-length-offset ( -- n ) 1 \ callstack type-number slot-offset ; inline

basis/cpu/x86/bootstrap.factor

@@ -76,19 +76,6 @@ big-endian off
     HEX: ffff RET f rc-absolute-2 rel-untagged
 ] callback-stub jit-define
 
-[
-    ! Load word
-    temp0 0 MOV f rc-absolute-cell rel-literal
-    ! Bump profiling counter
-    temp0 profile-count-offset [+] 1 tag-fixnum ADD
-    ! Load word->code
-    temp0 temp0 word-code-offset [+] MOV
-    ! Compute word entry point
-    temp0 compiled-header-size ADD
-    ! Jump to entry point
-    temp0 JMP
-] jit-profiling jit-define
-
 [
     ! load literal
     temp0 0 MOV f rc-absolute-cell rel-literal

core/bootstrap/primitives.factor

@@ -241,7 +241,6 @@ bi
     "props"
     "pic-def"
     "pic-tail-def"
-    { "counter" { "fixnum" "math" } }
     { "sub-primitive" read-only }
 } define-builtin
 

vm/code_block_visitor.hpp

@@ -59,19 +59,15 @@ void code_block_visitor<Fixup>::visit_object_code_block(object *obj)
 	case WORD_TYPE:
 		{
 			word *w = (word *)obj;
-			if(w->code)
-				w->code = visit_code_block(w->code);
-			if(w->counting_profiler)
-				w->counting_profiler = visit_code_block(w->counting_profiler);
-
-			parent->update_word_entry_point(w);
+			if(w->entry_point)
+				w->entry_point = visit_code_block(w->code())->entry_point();
 			break;
 		}
 	case QUOTATION_TYPE:
 		{
 			quotation *q = (quotation *)obj;
-			if(q->code)
-				parent->set_quot_entry_point(q,visit_code_block(q->code));
+			if(q->entry_point)
+				q->entry_point = visit_code_block(q->code())->entry_point();
 			break;
 		}
 	case CALLSTACK_TYPE:

vm/code_blocks.hpp

@@ -84,4 +84,17 @@ struct code_block
 };
 
 VM_C_API void undefined_symbol(void);
+
+inline code_block *word::code() const {
+	assert(entry_point != NULL);
+	return (code_block*)entry_point - 1;
 }
+
+inline code_block *quotation::code() const {
+	assert(entry_point != NULL);
+	return (code_block*)entry_point - 1;
+}
+
+}
+
+

vm/code_heap.cpp

@@ -178,15 +178,13 @@ void factor_vm::primitive_modify_code_heap()
 					parameters,
 					literals);
 
-				word->code = compiled;
+				word->entry_point = compiled->entry_point();
 			}
 			break;
 		default:
 			critical_error("Expected a quotation or an array",data.value());
 			break;
 		}
-
-		update_word_entry_point(word.untagged());
 	}
 
 	if(update_existing_words)

vm/counting_profiler.cpp

@@ -1,66 +0,0 @@
-#include "master.hpp"
-
-namespace factor
-{
-
-void factor_vm::init_counting_profiler()
-{
-	counting_profiler_p = false;
-}
-
-/* Allocates memory */
-code_block *factor_vm::compile_counting_profiler_stub(cell word_)
-{
-	data_root<word> word(word_,this);
-
-	jit jit(code_block_counting_profiler,word.value(),this);
-	jit.emit_with_literal(special_objects[JIT_PROFILING],word.value());
-
-	return jit.to_code_block();
-}
-
-/* Allocates memory */
-void factor_vm::set_counting_profiler(bool counting_profiler)
-{
-	if(counting_profiler == counting_profiler_p)
-		return;
-
-	/* Push everything to tenured space so that we can heap scan
-	and allocate counting_profiler blocks if necessary */
-	primitive_full_gc();
-
-	data_root<array> words(find_all_words(),this);
-
-	counting_profiler_p = counting_profiler;
-
-	cell length = array_capacity(words.untagged());
-	for(cell i = 0; i < length; i++)
-	{
-		tagged<word> word(array_nth(words.untagged(),i));
-
-		/* Note: can't do w->counting_profiler = ... since LHS evaluates
-		before RHS, and if RHS does a GC, we will have an
-		invalid pointer on the LHS */
-		if(counting_profiler)
-		{
-			if(!word->counting_profiler)
-			{
-				code_block *counting_profiler_block = compile_counting_profiler_stub(word.value());
-				word->counting_profiler = counting_profiler_block;
-			}
-
-			word->counter = tag_fixnum(0);
-		}
-
-		update_word_entry_point(word.untagged());
-	}
-
-	update_code_heap_words(false);
-}
-
-void factor_vm::primitive_counting_profiler()
-{
-	set_counting_profiler(to_boolean(ctx->pop()));
-}
-
-}

vm/counting_profiler.hpp

@@ -1,4 +0,0 @@
-namespace factor
-{
-
-}

vm/cpu-x86.cpp

@@ -63,7 +63,7 @@ void factor_vm::dispatch_signal_handler(cell *sp, cell *pc, cell handler)
 		else
 			assert(false);
 
-		*pc = (cell)handler_word->code->entry_point();
+		*pc = (cell)handler_word->entry_point;
 	}
 }
 

vm/entry_points.cpp

@@ -22,18 +22,14 @@ void factor_vm::c_to_factor(cell quot)
 
 template<typename Func> Func factor_vm::get_entry_point(cell n)
 {
-	/* We return word->code->entry_point() and not word->entry_point,
-	because if the counting profiler is enabled, we don't want to go through the
-	entry point's counting profiler stub. This clobbers registers, since entry
-	points use the C ABI and not the Factor ABI. */
 	tagged<word> entry_point_word(special_objects[n]);
-	return (Func)entry_point_word->code->entry_point();
+	return (Func)entry_point_word->entry_point;
 }
 
 void factor_vm::unwind_native_frames(cell quot, stack_frame *to)
 {
 	tagged<word> entry_point_word(special_objects[UNWIND_NATIVE_FRAMES_WORD]);
-	void *func = entry_point_word->code->entry_point();
+	void *func = entry_point_word->entry_point;
 	CODE_TO_FUNCTION_POINTER(func);
 	((unwind_native_frames_func_type)func)(quot,to);
 }
@@ -41,7 +37,7 @@ void factor_vm::unwind_native_frames(cell quot, stack_frame *to)
 cell factor_vm::get_fpu_state()
 {
 	tagged<word> entry_point_word(special_objects[GET_FPU_STATE_WORD]);
-	void *func = entry_point_word->code->entry_point();
+	void *func = entry_point_word->entry_point;
 	CODE_TO_FUNCTION_POINTER(func);
 	return ((get_fpu_state_func_type)func)();
 }
@@ -49,7 +45,7 @@ cell factor_vm::get_fpu_state()
 void factor_vm::set_fpu_state(cell state)
 {
 	tagged<word> entry_point_word(special_objects[SET_FPU_STATE_WORD]);
-	void *func = entry_point_word->code->entry_point();
+	void *func = entry_point_word->entry_point;
 	CODE_TO_FUNCTION_POINTER(func);
 	((set_fpu_state_func_type)func)(state);
 }

vm/jit.cpp

@@ -4,7 +4,6 @@ namespace factor
 {
 
 /* Simple code generator used by:
-- counting_profiler (counting_profiler.cpp),
 - quotation compiler (quotations.cpp),
 - megamorphic caches (dispatch.cpp),
 - polymorphic inline caches (inline_cache.cpp) */

vm/layouts.hpp

@@ -98,7 +98,6 @@ enum code_block_type
 {
 	code_block_unoptimized,
 	code_block_optimized,
-	code_block_counting_profiler,
 	code_block_pic
 };
 
@@ -253,7 +252,11 @@ struct string : public object {
 
 struct code_block;
 
-/* Assembly code makes assumptions about the layout of this struct */
+/* Assembly code makes assumptions about the layout of this struct:
+	basis/bootstrap/images/images.factor
+	basis/compiler/constants/constants.factor
+	core/bootstrap/primitives.factor
+*/
 struct word : public object {
 	static const cell type_number = WORD_TYPE;
 	/* TAGGED hashcode */
@@ -270,16 +273,14 @@ struct word : public object {
 	cell pic_def;
 	/* TAGGED alternative entry point for direct tail calls. Used for inline caching */
 	cell pic_tail_def;
-	/* TAGGED call count for counting_profiler */
-	cell counter;
 	/* TAGGED machine code for sub-primitive */
 	cell subprimitive;
 	/* UNTAGGED entry point: jump here to execute word */
 	void *entry_point;
 	/* UNTAGGED compiled code block */
-	code_block *code;
-	/* UNTAGGED counting_profiler stub */
-	code_block *counting_profiler;
+
+	/* defined in code_blocks.hpp */
+	code_block *code() const;
 };
 
 /* Assembly code makes assumptions about the layout of this struct */
@@ -299,7 +300,11 @@ struct boxed_float : object {
 	double n;
 };
 
-/* Assembly code makes assumptions about the layout of this struct */
+/* Assembly code makes assumptions about the layout of this struct:
+	basis/bootstrap/images/images.factor
+	basis/compiler/constants/constants.factor
+	core/bootstrap/primitives.factor
+*/
 struct quotation : public object {
 	static const cell type_number = QUOTATION_TYPE;
 	/* tagged */
@@ -310,8 +315,9 @@ struct quotation : public object {
 	cell cache_counter;
 	/* UNTAGGED entry point; jump here to call quotation */
 	void *entry_point;
-	/* UNTAGGED compiled code block */
-	code_block *code;
+
+	/* defined in code_blocks.hpp */
+	code_block *code() const;
 };
 
 /* Assembly code makes assumptions about the layout of this struct */

vm/master.hpp

@@ -96,7 +96,6 @@ namespace factor
 #include "contexts.hpp"
 #include "run.hpp"
 #include "objects.hpp"
-#include "counting_profiler.hpp"
 #include "sampling_profiler.hpp"
 #include "errors.hpp"
 #include "bignumint.hpp"

vm/primitives.hpp

@@ -46,7 +46,6 @@ namespace factor
 	_(context_object) \
 	_(context_object_for) \
 	_(current_callback) \
-	_(counting_profiler) \
 	_(data_room) \
 	_(datastack) \
 	_(datastack_for) \

vm/quotations.cpp

@@ -284,12 +284,6 @@ void quotation_jit::iterate_quotation()
 	}
 }
 
-void factor_vm::set_quot_entry_point(quotation *quot, code_block *code)
-{
-	quot->code = code;
-	quot->entry_point = code->entry_point();
-}
-
 /* Allocates memory */
 code_block *factor_vm::jit_compile_quot(cell owner_, cell quot_, bool relocating)
 {
@@ -313,7 +307,7 @@ void factor_vm::jit_compile_quot(cell quot_, bool relocating)
 	if(!quot_compiled_p(quot.untagged()))
 	{
 		code_block *compiled = jit_compile_quot(quot.value(),quot.value(),relocating);
-		set_quot_entry_point(quot.untagged(),compiled);
+		quot.untagged()->entry_point = compiled->entry_point();
 	}
 }
 
@@ -322,9 +316,9 @@ void factor_vm::primitive_jit_compile()
 	jit_compile_quot(ctx->pop(),true);
 }
 
-code_block *factor_vm::lazy_jit_compile_block()
+void *factor_vm::lazy_jit_compile_entry_point()
 {
-	return untag<word>(special_objects[LAZY_JIT_COMPILE_WORD])->code;
+	return untag<word>(special_objects[LAZY_JIT_COMPILE_WORD])->entry_point;
 }
 
 /* push a new quotation on the stack */
@@ -335,7 +329,7 @@ void factor_vm::primitive_array_to_quotation()
 	quot->array = ctx->peek();
 	quot->cached_effect = false_object;
 	quot->cache_counter = false_object;
-	set_quot_entry_point(quot,lazy_jit_compile_block());
+	quot->entry_point = lazy_jit_compile_entry_point();
 
 	ctx->replace(tag<quotation>(quot));
 }
@@ -344,8 +338,8 @@ void factor_vm::primitive_quotation_code()
 {
 	quotation *quot = untag_check<quotation>(ctx->pop());
 
-	ctx->push(from_unsigned_cell((cell)quot->code->entry_point()));
-	ctx->push(from_unsigned_cell((cell)quot->code + quot->code->size()));
+	ctx->push(from_unsigned_cell((cell)quot->entry_point));
+	ctx->push(from_unsigned_cell((cell)quot->code() + quot->code()->size()));
 }
 
 /* Allocates memory */
@@ -369,7 +363,7 @@ cell factor_vm::lazy_jit_compile(cell quot_)
 	assert(!quot_compiled_p(quot.untagged()));
 
 	code_block *compiled = jit_compile_quot(quot.value(),quot.value(),true);
-	set_quot_entry_point(quot.untagged(),compiled);
+	quot.untagged()->entry_point = compiled->entry_point();
 
 	return quot.value();
 }
@@ -381,7 +375,7 @@ VM_C_API cell lazy_jit_compile(cell quot, factor_vm *parent)
 
 bool factor_vm::quot_compiled_p(quotation *quot)
 {
-	return quot->code != NULL && quot->code != lazy_jit_compile_block();
+	return quot->entry_point != NULL && quot->entry_point != lazy_jit_compile_entry_point();
 }
 
 void factor_vm::primitive_quot_compiled_p()
@@ -404,8 +398,8 @@ void factor_vm::initialize_all_quotations()
 	for(cell i = 0; i < length; i++)
 	{
 		data_root<quotation> quot(array_nth(quotations.untagged(),i),this);
-		if(!quot->code)
-			set_quot_entry_point(quot.untagged(),lazy_jit_compile_block());
+		if(!quot->entry_point)
+			quot.untagged()->entry_point = lazy_jit_compile_entry_point();
 	}
 }
 

vm/slot_visitor.hpp

@@ -65,13 +65,13 @@ cell object::binary_payload_start(Fixup fixup) const
 		return 0;
 	/* these objects have some binary data at the end */
 	case WORD_TYPE:
-		return sizeof(word) - sizeof(cell) * 3;
+		return sizeof(word) - sizeof(cell);
 	case ALIEN_TYPE:
 		return sizeof(cell) * 3;
 	case DLL_TYPE:
 		return sizeof(cell) * 2;
 	case QUOTATION_TYPE:
-		return sizeof(quotation) - sizeof(cell) * 2;
+		return sizeof(quotation) - sizeof(cell);
 	case STRING_TYPE:
 		return sizeof(string);
 	/* everything else consists entirely of pointers */

vm/vm.cpp

@@ -8,7 +8,6 @@ factor_vm::factor_vm(THREADHANDLE thread) :
         thread(thread),
 	callback_id(0),
 	c_to_factor_func(NULL),
-	counting_profiler_p(false),
 	sampling_profiler_p(false),
 	signal_pipe_input(0),
 	signal_pipe_output(0),

vm/vm.hpp

@@ -64,7 +64,6 @@ struct factor_vm
 	c_to_factor_func_type c_to_factor_func;
 
 	/* Is profiling enabled? */
-	bool counting_profiler_p;
 	volatile cell sampling_profiler_p;
 	fixnum samples_per_second;
 
@@ -199,13 +198,7 @@ struct factor_vm
 	void primitive_clone();
 	void primitive_become();
 
-	// counting_profiler
-	void init_counting_profiler();
-	code_block *compile_counting_profiler_stub(cell word_);
-	void set_counting_profiler(bool counting_profiler);
-	void primitive_counting_profiler();
-
-	/* Sampling profiler */
+	// sampling_profiler
 	void clear_samples();
 	void record_sample();
 	void record_callstack_sample(cell *begin, cell *end);
@@ -458,7 +451,6 @@ struct factor_vm
 	word *allot_word(cell name_, cell vocab_, cell hashcode_);
 	void primitive_word();
 	void primitive_word_code();
-	void update_word_entry_point(word *w_);
 	void primitive_optimized_p();
 	void primitive_wrapper();
 	void jit_compile_word(cell word_, cell def_, bool relocating);
@@ -658,10 +650,9 @@ struct factor_vm
 
 	// quotations
 	void primitive_jit_compile();
-	code_block *lazy_jit_compile_block();
+	void *lazy_jit_compile_entry_point();
 	void primitive_array_to_quotation();
 	void primitive_quotation_code();
-	void set_quot_entry_point(quotation *quot, code_block *code);
 	code_block *jit_compile_quot(cell owner_, cell quot_, bool relocating);
 	void jit_compile_quot(cell quot_, bool relocating);
 	fixnum quot_code_offset_to_scan(cell quot_, cell offset);

vm/words.cpp

@@ -11,11 +11,11 @@ void factor_vm::jit_compile_word(cell word_, cell def_, bool relocating)
 
 	/* Refuse to compile this word more than once, because quot_compiled_p()
 	depends on the identity of its code block */
-	if(word->code && word.value() == special_objects[LAZY_JIT_COMPILE_WORD])
+	if(word->entry_point && word.value() == special_objects[LAZY_JIT_COMPILE_WORD])
 		return;
 
 	code_block *compiled = jit_compile_quot(word.value(),def.value(),relocating);
-	word->code = compiled;
+	word->entry_point = compiled->entry_point();
 
 	if(to_boolean(word->pic_def)) jit_compile_quot(word->pic_def,relocating);
 	if(to_boolean(word->pic_tail_def)) jit_compile_quot(word->pic_tail_def,relocating);
@@ -35,10 +35,8 @@ void factor_vm::compile_all_words()
 	{
 		data_root<word> word(array_nth(words.untagged(),i),this);
 
-		if(!word->code || !word->code->optimized_p())
+		if(!word->entry_point || !word->code()->optimized_p())
 			jit_compile_word(word.value(),word->def,false);
-
-		update_word_entry_point(word.untagged());
 	}
 }
 
@@ -54,22 +52,12 @@ word *factor_vm::allot_word(cell name_, cell vocab_, cell hashcode_)
 	new_word->name = name.value();
 	new_word->def = special_objects[OBJ_UNDEFINED];
 	new_word->props = false_object;
-	new_word->counter = tag_fixnum(0);
 	new_word->pic_def = false_object;
 	new_word->pic_tail_def = false_object;
 	new_word->subprimitive = false_object;
-	new_word->counting_profiler = NULL;
-	new_word->code = NULL;
+	new_word->entry_point = NULL;
 
 	jit_compile_word(new_word.value(),new_word->def,true);
-	if(counting_profiler_p)
-	{
-		code_block *counting_profiler_block = compile_counting_profiler_stub(new_word.value());
-		new_word->counting_profiler = counting_profiler_block;
-		initialize_code_block(new_word->counting_profiler);
-	}
-
-	update_word_entry_point(new_word.untagged());
 
 	return new_word.untagged();
 }
@@ -89,30 +77,14 @@ void factor_vm::primitive_word_code()
 	data_root<word> w(ctx->pop(),this);
 	w.untag_check(this);
 
-	if(counting_profiler_p)
-	{
-		ctx->push(from_unsigned_cell((cell)w->counting_profiler->entry_point()));
-		ctx->push(from_unsigned_cell((cell)w->counting_profiler + w->counting_profiler->size()));
-	}
-	else
-	{
-		ctx->push(from_unsigned_cell((cell)w->code->entry_point()));
-		ctx->push(from_unsigned_cell((cell)w->code + w->code->size()));
-	}
-}
-
-void factor_vm::update_word_entry_point(word *w)
-{
-	if(counting_profiler_p && w->counting_profiler)
-		w->entry_point = w->counting_profiler->entry_point();
-	else
-		w->entry_point = w->code->entry_point();
+	ctx->push(from_unsigned_cell((cell)w->entry_point));
+	ctx->push(from_unsigned_cell((cell)w->code() + w->code()->size()));
 }
 
 void factor_vm::primitive_optimized_p()
 {
 	word *w = untag_check<word>(ctx->peek());
-	ctx->replace(tag_boolean(w->code->optimized_p()));
+	ctx->replace(tag_boolean(w->code()->optimized_p()));
 }
 
 void factor_vm::primitive_wrapper()