Author: Armin Rigo <ar...@tunes.org>
Branch: stackroot-speedup-2
Changeset: r75710:1c9c0cc23c80
Date: 2015-02-04 19:01 +0100
http://bitbucket.org/pypy/pypy/changeset/1c9c0cc23c80/
Log: Untested code
diff --git a/rpython/jit/backend/x86/arch.py b/rpython/jit/backend/x86/arch.py
--- a/rpython/jit/backend/x86/arch.py
+++ b/rpython/jit/backend/x86/arch.py
@@ -35,7 +35,9 @@
PASS_ON_MY_FRAME = 15
JITFRAME_FIXED_SIZE = 6 + 8 * 2 # 6 GPR + 8 XMM * 2 WORDS/float
# 'threadlocal_addr' is passed as 2nd argument on the stack,
- # and it can be left here for when it is needed
+ # and it can be left here for when it is needed. As an additional hack,
+ # with asmgcc, it is made odd-valued to mean "already seen this frame
+ # during the previous minor collection".
THREADLOCAL_OFS = (FRAME_FIXED_SIZE + 2) * WORD
else:
# rbp + rbx + r12 + r13 + r14 + r15 + threadlocal + 12 extra words = 19
@@ -43,7 +45,9 @@
PASS_ON_MY_FRAME = 12
JITFRAME_FIXED_SIZE = 28 # 13 GPR + 15 XMM
# 'threadlocal_addr' is passed as 2nd argument in %esi,
- # and is moved into this frame location
+ # and is moved into this frame location. As an additional hack,
+ # with asmgcc, it is made odd-valued to mean "already seen this frame
+ # during the previous minor collection".
THREADLOCAL_OFS = (FRAME_FIXED_SIZE - 1) * WORD
assert PASS_ON_MY_FRAME >= 12 # asmgcc needs at least JIT_USE_WORDS + 3
diff --git a/rpython/jit/backend/x86/assembler.py
b/rpython/jit/backend/x86/assembler.py
--- a/rpython/jit/backend/x86/assembler.py
+++ b/rpython/jit/backend/x86/assembler.py
@@ -1980,6 +1980,23 @@
def _call_assembler_emit_call(self, addr, argloc, _):
threadlocal_loc = RawEspLoc(THREADLOCAL_OFS, INT)
+ if self._is_asmgcc():
+ # We need to remove the bit "already seen during the
+ # previous minor collection" instead of passing this
+ # value directly.
+ if IS_X86_64:
+ tmploc = esi # already the correct place
+ if argloc is tmploc:
+ self.mc.MOV_rr(esi.value, edi.value)
+ argloc = edi
+ else:
+ tmploc = eax
+ if tmploc is argloc:
+ tmploc = edx
+ self.mc.MOV(tmploc, threadlocal_ofs)
+ self.mc.AND_ri(tmploc.value, ~1)
+ threadlocal_ofs = tmploc
+ #
self.simple_call(addr, [argloc, threadlocal_loc])
def _call_assembler_emit_helper_call(self, addr, arglocs, result_loc):
@@ -2355,6 +2372,8 @@
assert self.cpu.translate_support_code
assert isinstance(resloc, RegLoc)
self.mc.MOV_rs(resloc.value, THREADLOCAL_OFS)
+ if self._is_asmgcc():
+ self.mc.AND_ri(resloc.value, ~1)
self.load_from_mem(resloc, addr_add_const(resloc, offset),
imm(size), imm(sign))
diff --git a/rpython/jit/backend/x86/callbuilder.py
b/rpython/jit/backend/x86/callbuilder.py
--- a/rpython/jit/backend/x86/callbuilder.py
+++ b/rpython/jit/backend/x86/callbuilder.py
@@ -167,6 +167,8 @@
self.tlofs_reg = r12
self.mc.MOV_rs(self.tlofs_reg.value,
THREADLOCAL_OFS - self.current_esp)
+ if self.asm._is_asmgcc():
+ self.mc.AND_ri(self.tlofs_reg.value, ~1)
return self.tlofs_reg
def save_stack_position(self):
diff --git a/rpython/memory/gctransform/asmgcroot.py
b/rpython/memory/gctransform/asmgcroot.py
--- a/rpython/memory/gctransform/asmgcroot.py
+++ b/rpython/memory/gctransform/asmgcroot.py
@@ -343,6 +343,7 @@
def walk_stack_roots(self, collect_stack_root, is_minor=False):
gcdata = self.gcdata
gcdata._gc_collect_stack_root = collect_stack_root
+ gcdata._gc_collect_is_minor = is_minor
pypy_asm_stackwalk(llhelper(ASM_CALLBACK_PTR, self._asm_callback),
gcrootanchor)
@@ -477,6 +478,14 @@
addr = self.getlocation(callee, ebp_in_caller, location)
caller.regs_stored_at[reg] = addr
reg -= 1
+ #
+ # small hack: the JIT reserves THREADLOCAL_OFS's last bit for
+ # us. We use it to store an "already traced past this frame"
+ # flag.
+ if self._with_jit:
+ is_minor = self.gcdata._gc_collect_is_minor
+ if self.mark_jit_frame_can_stop(callee, is_minor):
+ return False
location = self._shape_decompressor.next()
caller.frame_address = self.getlocation(callee, ebp_in_caller,
@@ -548,6 +557,23 @@
else: # kind == LOC_EBP_MINUS: at -N(%ebp)
return ebp_in_caller - offset
+ def mark_jit_frame_can_stop(self, callee, is_minor):
+ location = self._shape_decompressor.get_threadlocal_loc()
+ if location == LOC_NOWHERE:
+ return False
+ addr = self.getlocation(callee, llmemory.NULL, location)
+ #
+ x = addr.signed[0]
+ if is_minor:
+ if x & 1:
+ return True # this JIT stack frame is already
marked!
+ else:
+ addr.signed[0] = x | 1 # otherwise, mark it but don't stop
+ return False
+ else:
+ addr.signed[0] = x & ~1 # 'is_minor' is False, remove the marks
+ return False
+
LOC_REG = 0
LOC_ESP_PLUS = 1
@@ -729,6 +755,19 @@
llop.debug_fatalerror(lltype.Void, "asmgcroot: invalid index")
return 0 # annotator fix
+ def get_threadlocal_loc(self):
+ index = self.jit_index
+ if index < 0:
+ return LOC_NOWHERE # case "outside the jit"
+ else:
+ # case "in the jit"
+ from rpython.jit.backend.x86.arch import THREADLOCAL_OFS
+ from rpython.jit.backend.x86.arch import PASS_ON_MY_FRAME
+ stack_depth = PASS_ON_MY_FRAME + self.extra_stack_depth
+ return (LOC_ESP_PLUS |
+ ((THREADLOCAL_OFS // WORD + self.extra_stack_depth) << 2))
+
+
# ____________________________________________________________
#
_______________________________________________
pypy-commit mailing list
pypy-commit@python.org
https://mail.python.org/mailman/listinfo/pypy-commit
