Native implementation of coroutines to improve performance of fibers

git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@65834 b2dd03c8-39d4-4d8f-98ff-823fe69b080e

.gitignore

@@ -18,7 +18,6 @@
 *.pch
 *.pdb
 *.rej
-*.s
 *.sav
 *.swp
 *.yarb

common.mk

@@ -895,6 +895,7 @@ strstr.$(OBJEXT): {$(VPATH)}strstr.c
 nt.$(OBJEXT): {$(VPATH)}nt.c
 ia64.$(OBJEXT): {$(VPATH)}ia64.s
 	$(CC) $(CFLAGS) -c $<
+coroutine/amd64/Context.$(OBJECT): {$(VPATH)}coroutine/amd64/Context.s
 
 ###
 

configure.ac

@@ -2274,6 +2274,12 @@ AS_IF([test x"$target_cpu" = xia64], [
     ])
 ])
 
+AS_CASE(["$target_cpu-$target_os"],
+  [*-darwin*], [
+    AC_LIBOBJ([coroutine/amd64/Context])
+  ],
+)
+
 AC_CACHE_CHECK(whether right shift preserve sign bit, rb_cv_rshift_sign,
     [AC_COMPILE_IFELSE([AC_LANG_BOOL_COMPILE_TRY([], [(-1==(-1>>1))])],
 	rb_cv_rshift_sign=yes,

cont.c

@@ -26,6 +26,27 @@
  * in Proc. of 51th Programming Symposium, pp.21--28 (2010) (in Japanese).
  */
 
+/*
+  Enable this include to make fiber yield/resume about twice as fast.
+  
+  # Without libcoro
+  koyoko% ./build/bin/ruby ./fiber_benchmark.rb 10000 1000
+  setup time for 10000 fibers:   0.099961
+  execution time for 1000 messages:  19.505909
+
+  # With libcoro
+  koyoko% ./build/bin/ruby ./fiber_benchmark.rb 10000 1000
+  setup time for 10000 fibers:   0.099268
+  execution time for 1000 messages:   8.491746
+*/
+
+#define FIBER_USE_COROUTINE
+
+#ifdef FIBER_USE_COROUTINE
+#include "coroutine/amd64/Context.h"
+#define FIBER_USE_NATIVE 1
+#endif
+
 #if !defined(FIBER_USE_NATIVE)
 # if defined(HAVE_GETCONTEXT) && defined(HAVE_SETCONTEXT)
 #   if 0
@@ -139,7 +160,7 @@ enum fiber_status {
 #define FIBER_TERMINATED_P(fib) ((fib)->status == FIBER_TERMINATED)
 #define FIBER_RUNNABLE_P(fib)   (FIBER_CREATED_P(fib) || FIBER_SUSPENDED_P(fib))
 
-#if FIBER_USE_NATIVE && !defined(_WIN32)
+#if FIBER_USE_NATIVE && !defined(FIBER_USE_COROUTINE) && !defined(_WIN32)
 static inline int
 fiber_context_create(ucontext_t *context, void (*func)(), void *arg, void *ptr, size_t size)
 {
@@ -181,7 +202,11 @@ struct rb_fiber_struct {
     unsigned int transferred : 1;
 
 #if FIBER_USE_NATIVE
-#ifdef _WIN32
+#if defined(FIBER_USE_COROUTINE)
+    coroutine_context context;
+    void *ss_sp;
+    size_t ss_size;
+#elif defined(_WIN32)
     void *fib_handle;
 #else
     ucontext_t context;
@@ -382,8 +407,16 @@ cont_free(void *ptr)
     else {
 	/* fiber */
 	const rb_fiber_t *fib = (rb_fiber_t*)cont;
-#ifdef _WIN32
-        if (!fiber_is_root_p(fib)) {
+#if defined(FIBER_USE_COROUTINE)
+	coroutine_destroy(&fib->context);
+	if (fib->ss_sp != NULL) {
+	    if (fiber_is_root_p(fib)) {
+		rb_bug("Illegal root fiber parameter");
+	    }
+	    munmap((void*)fib->ss_sp, fib->ss_size);
+	}
+#elif defined(_WIN32)
+	if (!fiber_is_root_p(fib)) {
 	    /* don't delete root fiber handle */
 	    if (fib->fib_handle) {
 		DeleteFiber(fib->fib_handle);
@@ -799,9 +832,7 @@ fiber_entry(void *arg)
 }
 #else /* _WIN32 */
 
-NORETURN(static void fiber_entry(void *arg));
-static void
-fiber_entry(void *arg)
+COROUTINE fiber_entry(coroutine_context * from, coroutine_context * to)
 {
     rb_fiber_start();
 }
@@ -862,7 +893,17 @@ fiber_initialize_machine_stack_context(rb_fiber_t *fib, size_t size)
 {
     rb_execution_context_t *sec = &fib->cont.saved_ec;
 
-#ifdef _WIN32
+#if defined(FIBER_USE_COROUTINE)
+    char *ptr;
+    STACK_GROW_DIR_DETECTION;
+
+    ptr = fiber_machine_stack_alloc(size);
+    fib->ss_sp = ptr;
+    fib->ss_size = size;
+    coroutine_initialize(&fib->context, fiber_entry, ptr+size, size);
+    sec->machine.stack_start = (VALUE*)(ptr + STACK_DIR_UPPER(0, size));
+    sec->machine.stack_maxsize = size - RB_PAGE_SIZE;
+#elif defined(_WIN32)
 # if defined(_MSC_VER) && _MSC_VER <= 1200
 #   define CreateFiberEx(cs, stacksize, flags, entry, param) \
     CreateFiber((stacksize), (entry), (param))
@@ -925,15 +966,15 @@ fiber_setcontext(rb_fiber_t *newfib, rb_fiber_t *oldfib)
     /* restore thread context */
     fiber_restore_thread(th, newfib);
 
-#ifndef _WIN32
+    /* swap machine context */
+#if defined(FIBER_USE_COROUTINE)
+    coroutine_transfer(&oldfib->context, &newfib->context);
+#elif defined(_WIN32)
+    SwitchToFiber(newfib->fib_handle);
+#else
     if (!newfib->context.uc_stack.ss_sp && th->root_fiber != newfib) {
 	rb_bug("non_root_fiber->context.uc_stac.ss_sp should not be NULL");
     }
-#endif
-    /* swap machine context */
-#ifdef _WIN32
-    SwitchToFiber(newfib->fib_handle);
-#else
     swapcontext(&oldfib->context, &newfib->context);
 #endif
 }
@@ -1532,7 +1573,9 @@ root_fiber_alloc(rb_thread_t *th)
     fib->cont.self = fibval;
 
 #if FIBER_USE_NATIVE
-#ifdef _WIN32
+#if defined(FIBER_USE_COROUTINE)
+    coroutine_initialize(&fib->context, NULL, NULL, 0);
+#elif defined(_WIN32)
     /* setup fib_handle for root Fiber */
     if (fib->fib_handle == 0) {
         if ((fib->fib_handle = ConvertThreadToFiber(0)) == 0) {
@@ -1794,9 +1837,13 @@ rb_fiber_terminate(rb_fiber_t *fib, int need_interrupt)
     VM_ASSERT(FIBER_RESUMED_P(fib));
     rb_fiber_close(fib);
 
-#if FIBER_USE_NATIVE && !defined(_WIN32)
+#if FIBER_USE_NATIVE
+#if defined(FIBER_USE_COROUTINE)
+    coroutine_destroy(&fib->context);
+#elif !defined(_WIN32)
     fib->context.uc_stack.ss_sp = NULL;
 #endif
+#endif
 #ifdef MAX_MACHINE_STACK_CACHE
     /* Ruby must not switch to other thread until storing terminated_machine_stack */
     terminated_machine_stack.ptr = fib->ss_sp;

coroutine/amd64/Context.h

@@ -0,0 +1,62 @@
+//
+//  amd64.h
+//  File file is part of the "Coroutine" project and released under the MIT License.
+//
+//  Created by Samuel Williams on 10/5/2018.
+//  Copyright, 2018, by Samuel Williams. All rights reserved.
+//
+
+#pragma once
+
+#include <assert.h>
+#include <string.h>
+
+#if __cplusplus
+extern "C" {
+#endif
+
+#define COROUTINE __attribute__((noreturn)) void
+
+const size_t COROUTINE_REGISTERS = 6;
+
+// The fiber context (stack pointer).
+typedef struct
+{
+	void **stack_pointer;
+} coroutine_context;
+
+// The initialization function.
+typedef void(* coroutine_start)(coroutine_context *from, coroutine_context *self);
+
+inline void coroutine_initialize(
+	coroutine_context *context,
+	coroutine_start start,
+	void *stack_pointer,
+	size_t stack_size
+) {
+	/* Force 16-byte alignment */
+	context->stack_pointer = (void**)((uintptr_t)stack_pointer & ~0xF);
+
+	if (!start) {
+		assert(!context->stack_pointer);
+		/* We are main coroutine for this thread */
+		return;
+	}
+
+	*--context->stack_pointer = NULL;
+	*--context->stack_pointer = (void*)start;
+
+	context->stack_pointer -= COROUTINE_REGISTERS;
+	memset(context->stack_pointer, 0, sizeof(void*) * COROUTINE_REGISTERS);
+}
+
+coroutine_context * coroutine_transfer(coroutine_context * current, coroutine_context * target);
+
+inline void coroutine_destroy(coroutine_context * context)
+{
+	context->stack_pointer = NULL;
+}
+
+#if __cplusplus
+}
+#endif

coroutine/amd64/Context.s

@@ -0,0 +1,44 @@
+##
+##  amd64.c
+##  File file is part of the "Coroutine" project and released under the MIT License.
+##
+##  Created by Samuel Williams on 10/5/2018.
+##  Copyright, 2018, by Samuel Williams. All rights reserved.
+##
+
+.text
+
+.globl coroutine_transfer
+coroutine_transfer:
+
+# For older linkers
+.globl _coroutine_transfer
+_coroutine_transfer:
+
+	# Save caller state
+	pushq %rbp
+	pushq %rbx
+	pushq %r12
+	pushq %r13
+	pushq %r14
+	pushq %r15
+	
+	# Save caller stack pointer
+	movq %rsp, (%rdi)
+	
+	# Restore callee stack pointer
+	movq (%rsi), %rsp
+	
+	# Restore callee stack
+	popq %r15
+	popq %r14
+	popq %r13
+	popq %r12
+	popq %rbx
+	popq %rbp
+	
+	# Put the first argument into the return value
+	# movq %rdi, %rax
+	
+	# We pop the return address and jump to it
+	ret

coroutine/arm32/Context.h

@@ -0,0 +1,59 @@
+//
+//  amd64.h
+//  File file is part of the "Coroutine" project and released under the MIT License.
+//
+//  Created by Samuel Williams on 10/5/2018.
+//  Copyright, 2018, by Samuel Williams. All rights reserved.
+//
+
+#pragma once
+
+#include <assert.h>
+#include <string.h>
+
+#if __cplusplus
+extern "C" {
+#endif
+
+#define COROUTINE void
+
+const size_t COROUTINE_REGISTERS = 9;
+
+// The fiber context (stack pointer).
+typedef struct
+{
+	void **stack_pointer;
+} coroutine_context;
+
+// The initialization function.
+typedef void(* coroutine_start)(coroutine_context *from, coroutine_context *self);
+
+inline void coroutine_initialize(
+	coroutine_context *context,
+	coroutine_start start,
+	void *stack_pointer,
+	size_t stack_size
+) {
+	context->stack_pointer = (void**)stack_pointer;
+
+	if (!start) {
+		assert(!context->stack_pointer);
+		/* We are main coroutine for this thread */
+		return;
+	}
+
+	*--context->stack_pointer = (void*)start;
+
+	context->stack_pointer -= COROUTINE_REGISTERS;
+	memset(context->stack_pointer, 0, sizeof(void*) * COROUTINE_REGISTERS);
+}
+
+coroutine_context * coroutine_transfer(coroutine_context * current, coroutine_context * target);
+
+inline void coroutine_destroy(coroutine_context * context)
+{
+}
+
+#if __cplusplus
+}
+#endif

coroutine/arm32/Context.s

@@ -0,0 +1,15 @@
+##
+##  arm.c
+##  File file is part of the "Coroutine" project and released under the MIT License.
+##
+##  Created by Samuel Williams on 10/5/2018.
+##  Copyright, 2018, by Samuel Williams. All rights reserved.
+##
+
+.text
+
+.globl coroutine_transfer
+coroutine_transfer:
+	stmia r1!, {r4-r11,sp,lr}
+	ldmia r0!, {r4-r11,sp,pc}
+	bx lr

coroutine/arm64/Context.h

@@ -0,0 +1,60 @@
+//
+//  amd64.h
+//  File file is part of the "Coroutine" project and released under the MIT License.
+//
+//  Created by Samuel Williams on 10/5/2018.
+//  Copyright, 2018, by Samuel Williams. All rights reserved.
+//
+
+#pragma once
+
+#include <assert.h>
+#include <string.h>
+
+#if __cplusplus
+extern "C" {
+#endif
+
+#define COROUTINE void
+
+const size_t COROUTINE_REGISTERS = 0xb0 / 8;
+
+// The fiber context (stack pointer).
+typedef struct
+{
+	void **stack_pointer;
+} coroutine_context;
+
+// The initialization function.
+typedef void(* coroutine_start)(coroutine_context *from, coroutine_context *self);
+
+inline void coroutine_initialize(
+	coroutine_context *context,
+	coroutine_start start,
+	void *stack_pointer,
+	size_t stack_size
+) {
+	/* Force 16-byte alignment */
+	context->stack_pointer = (void**)((uintptr_t)stack_pointer & ~0xF);
+
+	if (!start) {
+		assert(!context->stack_pointer);
+		/* We are main coroutine for this thread */
+		return;
+	}
+
+	context->stack_pointer -= COROUTINE_REGISTERS;
+	memset(context->stack_pointer, 0, sizeof(void*) * COROUTINE_REGISTERS);
+
+	context->stack_pointer[0xa0 / 8] = (void*)start;
+}
+
+coroutine_context * coroutine_transfer(coroutine_context * current, coroutine_context * target);
+
+inline void coroutine_destroy(coroutine_context * context)
+{
+}
+
+#if __cplusplus
+}
+#endif

coroutine/arm64/Context.s

@@ -0,0 +1,58 @@
+##
+##  arm64.s
+##  File file is part of the "Coroutine" project and released under the MIT License.
+##
+##  Created by Samuel Williams on 10/5/2018.
+##  Copyright, 2018, by Samuel Williams. All rights reserved.
+##
+
+.text
+.align 2
+.global coroutine_transfer
+coroutine_transfer:
+	# Make space on the stack for caller registers
+	sub sp, sp, 0xb0
+	
+	# Save caller registers
+	stp d8, d9, [sp, 0x00]
+	stp d10, d11, [sp, 0x10]
+	stp d12, d13, [sp, 0x20]
+	stp d14, d15, [sp, 0x30]
+	stp x19, x20, [sp, 0x40]
+	stp x21, x22, [sp, 0x50]
+	stp x23, x24, [sp, 0x60]
+	stp x25, x26, [sp, 0x70]
+	stp x27, x28, [sp, 0x80]
+	stp x29, x30, [sp, 0x90]
+
+	# Save return address
+	str x30, [sp, 0xa0]
+
+	# Save stack pointer to x0 (first argument)
+	mov x2, sp
+	str x2, [x0, 0]
+
+	# Load stack pointer from x1 (second argument)
+	ldr x3, [x1, 0]
+	mov sp, x3
+
+	# Restore caller registers
+	ldp d8, d9, [sp, 0x00]
+	ldp d10, d11, [sp, 0x10]
+	ldp d12, d13, [sp, 0x20]
+	ldp d14, d15, [sp, 0x30]
+	ldp x19, x20, [sp, 0x40]
+	ldp x21, x22, [sp, 0x50]
+	ldp x23, x24, [sp, 0x60]
+	ldp x25, x26, [sp, 0x70]
+	ldp x27, x28, [sp, 0x80]
+	ldp x29, x30, [sp, 0x90]
+
+	# Load return address into x4
+	ldr x4, [sp, 0xa0]
+
+	# Pop stack frame
+	add sp, sp, 0xb0
+
+	# Jump to return address (in x4)
+	ret x4

coroutine/win32/Context.asm

@@ -0,0 +1,43 @@
+;;
+;;  win32.asm
+;;  File file is part of the "Coroutine" project and released under the MIT License.
+;;
+;;  Created by Samuel Williams on 10/5/2018.
+;;  Copyright, 2018, by Samuel Williams. All rights reserved.
+;;
+
+.model flat
+
+.code
+
+; Using fastcall is a big win (and it's the same has how x64 works).
+; In coroutine transfer, the arguments are passed in ecx and edx. We don't need
+; to touch these in order to pass them to the destination coroutine.
+
+@coroutine_transfer@8 proc
+	; Save caller registers
+	push ebp
+	push ebx
+	push edi
+	push esi
+	
+	; Save caller stack pointer
+	mov dword ptr [ecx], esp
+	
+	; Restore callee stack pointer
+	mov esp, dword ptr [edx]
+
+	; Restore callee stack
+	pop esi
+	pop edi
+	pop ebx
+	pop ebp
+
+	; Save the first argument as the return value
+	mov eax, dword ptr ecx
+
+	; Jump to the address on the stack
+	ret
+@coroutine_transfer@8 endp
+
+end

coroutine/win32/Context.h

@@ -0,0 +1,64 @@
+//
+//  win32.h
+//  File file is part of the "Coroutine" project and released under the MIT License.
+//
+//  Created by Samuel Williams on 10/5/2018.
+//  Copyright, 2018, by Samuel Williams. All rights reserved.
+//
+
+#pragma once
+
+#include <assert.h>
+#include <string.h>
+
+#if __cplusplus
+extern "C" {
+#endif
+
+#define COROUTINE void __fastcall
+
+const size_t COROUTINE_REGISTERS = 4;
+
+// The fiber context (stack pointer).
+struct coroutine_context
+{
+	void **stack_pointer;
+};
+
+// The initialization function.
+typedef void(__fastcall * coroutine_start)(coroutine_context *from, coroutine_context *self);
+
+inline void coroutine_initialize(
+	coroutine_context *context,
+	coroutine_start start,
+	void *stack_pointer,
+	size_t stack_size
+) {
+	context->stack_pointer = (void**)stack_pointer;
+
+	if (!start) {
+		assert(!context->stack_pointer);
+		/* We are main coroutine for this thread */
+		return;
+	}
+
+	/* Windows Thread Information Block */
+	*--context->stack_pointer = 0;
+	*--context->stack_pointer = stack_pointer;
+	*--context->stack_pointer = (void*)stack_size;
+
+	*--context->stack_pointer = (void*)start;
+
+	context->stack_pointer -= COROUTINE_REGISTERS;
+	memset(context->stack_pointer, 0, sizeof(void*) * COROUTINE_REGISTERS);
+}
+
+coroutine_context * __fastcall coroutine_transfer(coroutine_context * current, coroutine_context * target);
+
+inline void coroutine_destroy(coroutine_context * context)
+{
+}
+
+#if __cplusplus
+}
+#endif

coroutine/win64/Context.asm

@@ -0,0 +1,44 @@
+;;
+;;  win64.asm
+;;  File file is part of the "Coroutine" project and released under the MIT License.
+;;
+;;  Created by Samuel Williams on 10/5/2018.
+;;  Copyright, 2018, by Samuel Williams. All rights reserved.
+;;
+
+.code
+
+coroutine_transfer proc
+	push rbp
+	push rbx
+	push rdi
+	push rsi
+	push r12
+	push r13
+	push r14
+	push r15
+	
+	; Save caller stack pointer
+	mov [rcx], rsp 
+	
+	; Restore callee stack pointer
+	mov rsp, [rdx]
+
+	; Restore callee stack
+	pop r15
+	pop r14
+	pop r13
+	pop r12
+	pop rsi
+	pop rdi
+	pop rbx
+	pop rbp
+
+	; Put the first argument into the return value
+	mov rax, rcx
+
+	; We pop the return address and jump to it
+	ret
+coroutine_transfer endp
+
+end

coroutine/win64/Context.h

@@ -0,0 +1,64 @@
+//
+//  win64.h
+//  File file is part of the "Coroutine" project and released under the MIT License.
+//
+//  Created by Samuel Williams on 10/5/2018.
+//  Copyright, 2018, by Samuel Williams. All rights reserved.
+//
+
+#pragma once
+
+#include <assert.h>
+#include <string.h>
+
+#if __cplusplus
+extern "C" {
+#endif
+
+#define COROUTINE void
+
+const size_t COROUTINE_REGISTERS = 8;
+
+// The fiber context (stack pointer).
+struct coroutine_context
+{
+	void **stack_pointer;
+};
+
+// The initialization function.
+typedef void(* coroutine_start)(coroutine_context *from, coroutine_context *self);
+
+inline void coroutine_initialize(
+	coroutine_context *context,
+	coroutine_start start,
+	void *stack_pointer,
+	size_t stack_size
+) {
+	context->stack_pointer = (void**)stack_pointer;
+
+	if (!start) {
+		assert(!context->stack_pointer);
+		/* We are main coroutine for this thread */
+		return;
+	}
+
+	/* Windows Thread Information Block */
+	*--context->stack_pointer = 0;
+	*--context->stack_pointer = stack_pointer;
+	*--context->stack_pointer = (void*)stack_size;
+
+	*--context->stack_pointer = (void*)start;
+
+	context->stack_pointer -= COROUTINE_REGISTERS;
+	memset(context->stack_pointer, 0, sizeof(void*) * COROUTINE_REGISTERS);
+}
+
+coroutine_context * coroutine_transfer(coroutine_context * current, coroutine_context * target);
+
+inline void coroutine_destroy(coroutine_context * context)
+{
+}
+
+#if __cplusplus
+}
+#endif

fiber_benchmark.rb

@@ -0,0 +1,103 @@
+#!/usr/bin/env ruby
+
+require 'fiber'
+require 'benchmark'
+
+class Ring
+   attr_reader :id
+   attr_accessor :attach 
+
+   def initialize(id)
+      @id = id
+      #puts "Creating ring ... #{id}"
+      @fiber = Fiber.new do
+         pass_message
+      end
+   end
+
+   def |(other)
+      other.attach = self if !other.nil?
+      #puts "attaching #{@id} to #{other.id}" if !other.nil?
+      other
+   end
+
+   def resume
+      @fiber.resume
+    end
+
+   def pass_message
+      #puts "I'm fiber #{@id}"
+      while message = message_in
+         #puts "... #{@id} I received message #{message}"
+         # do something with message
+         message_out(message)      
+      end
+   end
+
+   def message_in
+      #puts "Resuming #{@attach.id}" if !@attach.nil?
+      @attach.resume if !@attach.nil?
+   end
+
+   def message_out(message)
+      Fiber.yield(message)
+   end
+
+end
+
+class RingStart < Ring
+   attr_accessor :message
+   def initialize(n, m, message)
+      @m = m
+      @message = message
+      super(n)
+   end
+   
+   def pass_message 
+      loop { message_out(@message) }
+   end
+
+end
+
+
+def create_chain_r(i, chain)
+   # recursive version
+   return chain if i<=0
+   r = chain.nil? ? Ring.new(i) :  chain | Ring.new(i)
+   create_chain(i-1, r)
+end
+
+def create_chain(n, chain)
+   # loop version
+   # needed to avoid stack overflow for high n
+   n.downto(0) {
+      chain = chain | Ring.new(n)
+   }
+   chain
+end
+
+def run_benchmark(n, m)
+  mess = :hello
+  ringu = nil
+  chain = nil
+
+  tm = Benchmark.measure {
+     ringu = RingStart.new(0, m, mess)
+     chain = create_chain(n, ringu)
+  }.format("%10.6r\n").gsub!(/\(|\)/, "")
+
+  puts "setup time for #{n} fibers: #{tm}"
+
+  tm  = Benchmark.measure {
+     m.times { ringu.message = chain.resume }
+  }.format("%10.6r\n").gsub!(/\(|\)/, "")
+
+  puts "execution time for #{m} messages: #{tm}"
+end
+
+n = (ARGV[0] || 1000).to_i
+m = (ARGV[1] || 10000).to_i
+
+5.times do
+  run_benchmark(n, m)
+end