Author: Maciej Fijalkowski <fij...@gmail.com>
Branch: fast-slowpath
Changeset: r65481:23817367279e
Date: 2013-07-19 16:43 +0200
http://bitbucket.org/pypy/pypy/changeset/23817367279e/
Log: try to improve on slowpath + fixes
diff --git a/rpython/jit/backend/llsupport/assembler.py
b/rpython/jit/backend/llsupport/assembler.py
--- a/rpython/jit/backend/llsupport/assembler.py
+++ b/rpython/jit/backend/llsupport/assembler.py
@@ -106,8 +106,10 @@
kind='unicode')
else:
self.malloc_slowpath_unicode = None
- self.cond_call_slowpath = [self._build_cond_call_slowpath(False),
- self._build_cond_call_slowpath(True)]
+ self.cond_call_slowpath = [self._build_cond_call_slowpath(False,
False),
+ self._build_cond_call_slowpath(False, True),
+ self._build_cond_call_slowpath(True, False),
+ self._build_cond_call_slowpath(True, True)]
self._build_stack_check_slowpath()
self._build_release_gil(gc_ll_descr.gcrootmap)
diff --git a/rpython/jit/backend/llsupport/regalloc.py
b/rpython/jit/backend/llsupport/regalloc.py
--- a/rpython/jit/backend/llsupport/regalloc.py
+++ b/rpython/jit/backend/llsupport/regalloc.py
@@ -382,7 +382,7 @@
loc = self.reg_bindings.get(v, None)
if loc is not None and loc not in self.no_lower_byte_regs:
return loc
- for i in range(len(self.free_regs)):
+ for i in range(len(self.free_regs) - 1, -1, -1):
reg = self.free_regs[i]
if reg not in self.no_lower_byte_regs:
if loc is not None:
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
@@ -141,7 +141,7 @@
gcrootmap = self.cpu.gc_ll_descr.gcrootmap
if gcrootmap and gcrootmap.is_shadow_stack:
- self._load_shadowstack_top_in_ebx(mc, gcrootmap)
+ self._load_shadowstack_top_in_reg(mc, gcrootmap)
mc.MOV_mr((ebx.value, -WORD), eax.value)
mc.MOV_bi(gcmap_ofs, 0)
@@ -149,15 +149,15 @@
mc.RET()
self._frame_realloc_slowpath = mc.materialize(self.cpu.asmmemmgr, [])
- def _build_cond_call_slowpath(self, supports_floats):
+ def _build_cond_call_slowpath(self, supports_floats, callee_only):
""" This builds a general call slowpath, for whatever call happens to
come.
"""
mc = codebuf.MachineCodeBlockWrapper()
- self._push_all_regs_to_frame(mc, [], supports_floats,
callee_only=False)
+ self._push_all_regs_to_frame(mc, [], supports_floats, callee_only)
gcrootmap = self.cpu.gc_ll_descr.gcrootmap
if gcrootmap and gcrootmap.is_shadow_stack:
- self._call_header_shadowstack(mc, gcrootmap)
+ self._call_header_shadowstack(mc, gcrootmap, selected_reg=r8)
mc.SUB(esp, imm(WORD))
self.set_extra_stack_depth(mc, 2 * WORD)
# args are in their respective positions
@@ -166,9 +166,9 @@
self.set_extra_stack_depth(mc, 0)
self._reload_frame_if_necessary(mc, align_stack=True)
if gcrootmap and gcrootmap.is_shadow_stack:
- self._call_footer_shadowstack(mc, gcrootmap)
- self._pop_all_regs_from_frame(mc, [], self.cpu.supports_floats,
- callee_only=False)
+ self._call_footer_shadowstack(mc, gcrootmap, selected_reg=r8)
+ self._pop_all_regs_from_frame(mc, [], supports_floats,
+ callee_only)
mc.RET()
return mc.materialize(self.cpu.asmmemmgr, [])
@@ -757,34 +757,34 @@
self.mc.ADD_ri(esp.value, FRAME_FIXED_SIZE * WORD)
self.mc.RET()
- def _load_shadowstack_top_in_ebx(self, mc, gcrootmap):
+ def _load_shadowstack_top_in_reg(self, mc, gcrootmap, selected_reg=ebx):
rst = gcrootmap.get_root_stack_top_addr()
if rx86.fits_in_32bits(rst):
- mc.MOV_rj(ebx.value, rst) # MOV ebx, [rootstacktop]
+ mc.MOV_rj(selected_reg.value, rst) # MOV ebx, [rootstacktop]
else:
mc.MOV_ri(X86_64_SCRATCH_REG.value, rst) # MOV r11, rootstacktop
- mc.MOV_rm(ebx.value, (X86_64_SCRATCH_REG.value, 0))
+ mc.MOV_rm(selected_reg.value, (X86_64_SCRATCH_REG.value, 0))
# MOV ebx, [r11]
#
return rst
- def _call_header_shadowstack(self, mc, gcrootmap):
- rst = self._load_shadowstack_top_in_ebx(mc, gcrootmap)
- mc.MOV_mr((ebx.value, 0), ebp.value) # MOV [ebx], ebp
- mc.ADD_ri(ebx.value, WORD)
+ def _call_header_shadowstack(self, mc, gcrootmap, selected_reg=ebx):
+ rst = self._load_shadowstack_top_in_reg(mc, gcrootmap, selected_reg)
+ mc.MOV_mr((selected_reg.value, 0), ebp.value) # MOV [ebx], ebp
+ mc.ADD_ri(selected_reg.value, WORD)
if rx86.fits_in_32bits(rst):
- mc.MOV_jr(rst, ebx.value) # MOV [rootstacktop], ebx
+ mc.MOV_jr(rst, selected_reg.value) # MOV [rootstacktop], ebx
else:
mc.MOV_mr((X86_64_SCRATCH_REG.value, 0),
- ebx.value) # MOV [r11], ebx
+ selected_reg.value) # MOV [r11], ebx
- def _call_footer_shadowstack(self, mc, gcrootmap):
+ def _call_footer_shadowstack(self, mc, gcrootmap, selected_reg=ebx):
rst = gcrootmap.get_root_stack_top_addr()
if rx86.fits_in_32bits(rst):
mc.SUB_ji8(rst, WORD) # SUB [rootstacktop], WORD
else:
- mc.MOV_ri(ebx.value, rst) # MOV ebx, rootstacktop
- mc.SUB_mi8((ebx.value, 0), WORD) # SUB [ebx], WORD
+ mc.MOV_ri(selected_reg.value, rst) # MOV ebx,
rootstacktop
+ mc.SUB_mi8((selected_reg.value, 0), WORD) # SUB [ebx], WORD
def redirect_call_assembler(self, oldlooptoken, newlooptoken):
# some minimal sanity checking
@@ -2152,10 +2152,17 @@
self.mc.J_il8(rx86.Conditions['Z'], 0) # patched later
jmp_adr = self.mc.get_relative_pos()
self.push_gcmap(self.mc, gcmap, store=True)
- if self._regalloc is not None and self._regalloc.xrm.reg_bindings:
- cond_call_adr = self.cond_call_slowpath[1]
- else:
- cond_call_adr = self.cond_call_slowpath[0]
+ callee_only = False
+ floats = False
+ if self._regalloc is not None:
+ for reg in self._regalloc.rm.reg_bindings.values():
+ if reg not in self._regalloc.rm.save_around_call_regs:
+ break
+ else:
+ callee_only = True
+ if self._regalloc.xrm.reg_bindings:
+ floats = True
+ cond_call_adr = self.cond_call_slowpath[floats * 2 + callee_only]
self.mc.CALL(imm(cond_call_adr))
self.pop_gcmap(self.mc)
# never any result value
_______________________________________________
pypy-commit mailing list
pypy-commit@python.org
http://mail.python.org/mailman/listinfo/pypy-commit
