Author: Wim Lavrijsen <wlavrij...@lbl.gov>
Branch: reflex-support
Changeset: r63535:608cb355ea66
Date: 2013-04-19 10:48 -0700
http://bitbucket.org/pypy/pypy/changeset/608cb355ea66/
Log: merge default into branch; hopefully solves gc crash ...
diff --git a/rpython/jit/backend/arm/assembler.py
b/rpython/jit/backend/arm/assembler.py
--- a/rpython/jit/backend/arm/assembler.py
+++ b/rpython/jit/backend/arm/assembler.py
@@ -56,7 +56,7 @@
if we_are_translated():
self.debug = False
self.current_clt = looptoken.compiled_loop_token
- self.mc = InstrBuilder(self.cpu.arch_version)
+ self.mc = InstrBuilder(self.cpu.cpuinfo.arch_version)
self.pending_guards = []
assert self.datablockwrapper is None
allblocks = self.get_asmmemmgr_blocks(looptoken)
@@ -80,7 +80,7 @@
if not self.cpu.propagate_exception_descr:
return # not supported (for tests, or non-translated)
#
- mc = InstrBuilder(self.cpu.arch_version)
+ mc = InstrBuilder(self.cpu.cpuinfo.arch_version)
self._store_and_reset_exception(mc, r.r0)
ofs = self.cpu.get_ofs_of_frame_field('jf_guard_exc')
# make sure ofs fits into a register
@@ -165,7 +165,7 @@
# | my own retaddr | <-- sp
# +-----------------------+
#
- mc = InstrBuilder(self.cpu.arch_version)
+ mc = InstrBuilder(self.cpu.cpuinfo.arch_version)
# save argument registers and return address
mc.PUSH([reg.value for reg in r.argument_regs] + [r.ip.value,
r.lr.value])
# stack is aligned here
@@ -206,7 +206,7 @@
# write barriers. It must save all registers, and optionally
# all vfp registers. It takes a single argument which is in r0.
# It must keep stack alignment accordingly.
- mc = InstrBuilder(self.cpu.arch_version)
+ mc = InstrBuilder(self.cpu.cpuinfo.arch_version)
#
exc0 = exc1 = None
mc.PUSH([r.ip.value, r.lr.value]) # push two words to keep alignment
@@ -249,8 +249,10 @@
else:
self.wb_slowpath[withcards + 2 * withfloats] = rawstart
- def _build_malloc_slowpath(self):
- mc = InstrBuilder(self.cpu.arch_version)
+ def _build_malloc_slowpath(self, kind):
+ if kind != 'fixed':
+ return 0
+ mc = InstrBuilder(self.cpu.cpuinfo.arch_version)
self._push_all_regs_to_jitframe(mc, [r.r0, r.r1],
self.cpu.supports_floats)
ofs = self.cpu.get_ofs_of_frame_field('jf_gcmap')
# store the gc pattern
@@ -287,9 +289,7 @@
self.store_reg(mc, r.ip, r.fp, ofs)
# return
mc.POP([r.ip.value, r.pc.value])
-
- rawstart = mc.materialize(self.cpu.asmmemmgr, [])
- self.malloc_slowpath = rawstart
+ return mc.materialize(self.cpu.asmmemmgr, [])
def _reload_frame_if_necessary(self, mc):
gcrootmap = self.cpu.gc_ll_descr.gcrootmap
@@ -362,7 +362,7 @@
self.load_reg(mc, vfpr, r.fp, ofs)
def _build_failure_recovery(self, exc, withfloats=False):
- mc = InstrBuilder(self.cpu.arch_version)
+ mc = InstrBuilder(self.cpu.cpuinfo.arch_version)
self._push_all_regs_to_jitframe(mc, [], withfloats)
if exc:
@@ -645,7 +645,7 @@
expected_size=expected_size)
def _patch_frame_depth(self, adr, allocated_depth):
- mc = InstrBuilder(self.cpu.arch_version)
+ mc = InstrBuilder(self.cpu.cpuinfo.arch_version)
mc.gen_load_int(r.lr.value, allocated_depth)
mc.copy_to_raw_memory(adr)
@@ -721,7 +721,7 @@
# f) store the address of the new jitframe in the shadowstack
# c) set the gcmap field to 0 in the new jitframe
# g) restore registers and return
- mc = InstrBuilder(self.cpu.arch_version)
+ mc = InstrBuilder(self.cpu.cpuinfo.arch_version)
self._push_all_regs_to_jitframe(mc, [], self.cpu.supports_floats)
# this is the gcmap stored by push_gcmap(mov=True) in
_check_stack_frame
# and the expected_size pushed in _check_stack_frame
@@ -781,7 +781,7 @@
self.target_tokens_currently_compiling = None
def _patch_stackadjust(self, adr, allocated_depth):
- mc = InstrBuilder(self.cpu.arch_version)
+ mc = InstrBuilder(self.cpu.cpuinfo.arch_version)
mc.gen_load_int(r.lr.value, allocated_depth)
mc.copy_to_raw_memory(adr)
@@ -821,7 +821,7 @@
# patch the guard jumpt to the stub
# overwrite the generate NOP with a B_offs to the pos of the
# stub
- mc = InstrBuilder(self.cpu.arch_version)
+ mc = InstrBuilder(self.cpu.cpuinfo.arch_version)
mc.B_offs(relative_offset, c.get_opposite_of(tok.fcond))
mc.copy_to_raw_memory(guard_pos)
else:
@@ -900,7 +900,7 @@
self.mc.ASR_ri(resloc.value, resloc.value, 16)
def patch_trace(self, faildescr, looptoken, bridge_addr, regalloc):
- b = InstrBuilder(self.cpu.arch_version)
+ b = InstrBuilder(self.cpu.cpuinfo.arch_version)
patch_addr = faildescr._arm_failure_recovery_block
assert patch_addr != 0
b.B(bridge_addr)
diff --git a/rpython/jit/backend/arm/detect.py
b/rpython/jit/backend/arm/detect.py
--- a/rpython/jit/backend/arm/detect.py
+++ b/rpython/jit/backend/arm/detect.py
@@ -1,7 +1,6 @@
import os
from rpython.translator.tool.cbuild import ExternalCompilationInfo
-from rpython.rlib.clibffi import FFI_DEFAULT_ABI, FFI_SYSV, FFI_VFP
from rpython.rtyper.tool import rffi_platform
from rpython.translator.platform import CompilationError
from rpython.rlib.debug import debug_print, debug_start, debug_stop
@@ -46,9 +45,11 @@
# "Processor : ARMv%d-compatible processor rev 7 (v6l)"
i = buf.find('ARMv')
if i == -1:
- raise ValueError("Unknown Processor entry")
-
- n = int(buf[i + 4])
+ n = 6
+ debug_print("Could not detect architecture version, "
+ "falling back to", "ARMv%d" % n)
+ else:
+ n = int(buf[i + 4])
if n < 6:
raise ValueError("Unsupported ARM architecture version")
diff --git a/rpython/jit/backend/arm/opassembler.py
b/rpython/jit/backend/arm/opassembler.py
--- a/rpython/jit/backend/arm/opassembler.py
+++ b/rpython/jit/backend/arm/opassembler.py
@@ -354,7 +354,7 @@
# whether to worry about a CALL that can collect; this
# is always true except in call_release_gil
can_collect=True):
- if self.cpu.hf_abi:
+ if self.cpu.cpuinfo.hf_abi:
stack_args, adr = self._setup_call_hf(adr, arglocs, fcond,
resloc, result_info)
else:
@@ -382,7 +382,7 @@
# ensure the result is wellformed and stored in the correct location
if resloc is not None:
- if resloc.is_vfp_reg() and not self.cpu.hf_abi:
+ if resloc.is_vfp_reg() and not self.cpu.cpuinfo.hf_abi:
# move result to the allocated register
self.mov_to_vfp_loc(r.r0, r.r1, resloc)
elif resloc.is_reg() and result_info != (-1, -1):
@@ -1230,7 +1230,7 @@
baseofs = self.cpu.get_baseofs_of_frame_field()
newlooptoken.compiled_loop_token.update_frame_info(
oldlooptoken.compiled_loop_token, baseofs)
- mc = InstrBuilder(self.cpu.arch_version)
+ mc = InstrBuilder(self.cpu.cpuinfo.arch_version)
mc.B(target)
mc.copy_to_raw_memory(oldadr)
diff --git a/rpython/jit/backend/arm/runner.py
b/rpython/jit/backend/arm/runner.py
--- a/rpython/jit/backend/arm/runner.py
+++ b/rpython/jit/backend/arm/runner.py
@@ -11,6 +11,10 @@
jitframe.STATICSIZE = JITFRAME_FIXED_SIZE
+class CPUInfo(object):
+ hf_abi = False
+ arch_version = 6
+
class AbstractARMCPU(AbstractLLCPU):
IS_64_BIT = False
@@ -26,13 +30,11 @@
float_regs = VFPRegisterManager.all_regs
frame_reg = fp
- hf_abi = False # use hard float abi flag
- arch_version = 6 # assume ARMv6 as base case
-
def __init__(self, rtyper, stats, opts=None, translate_support_code=False,
gcdescr=None):
AbstractLLCPU.__init__(self, rtyper, stats, opts,
translate_support_code, gcdescr)
+ self.cpuinfo = CPUInfo()
def set_debug(self, flag):
return self.assembler.set_debug(flag)
@@ -47,8 +49,8 @@
self.assembler = AssemblerARM(self, self.translate_support_code)
def setup_once(self):
- self.arch_version = detect_arch_version()
- self.hf_abi = detect_hardfloat()
+ self.cpuinfo.arch_version = detect_arch_version()
+ self.cpuinfo.hf_abi = detect_hardfloat()
self.assembler.setup_once()
def finish_once(self):
@@ -92,7 +94,7 @@
from rpython.jit.backend.arm.codebuilder import InstrBuilder
for jmp, tgt in looptoken.compiled_loop_token.invalidate_positions:
- mc = InstrBuilder(self.arch_version)
+ mc = InstrBuilder(self.cpuinfo.arch_version)
mc.B_offs(tgt)
mc.copy_to_raw_memory(jmp)
# positions invalidated
diff --git a/rpython/jit/backend/arm/test/test_detect.py
b/rpython/jit/backend/arm/test/test_detect.py
--- a/rpython/jit/backend/arm/test/test_detect.py
+++ b/rpython/jit/backend/arm/test/test_detect.py
@@ -3,7 +3,32 @@
from rpython.jit.backend.arm.detect import detect_arch_version
cpuinfo = "Processor : ARMv%d-compatible processor rev 7 (v6l)"""
-
+cpuinfo2 = """processor : 0
+vendor_id : GenuineIntel
+cpu family : 6
+model : 23
+model name : Intel(R) Core(TM)2 Duo CPU E8400 @ 3.00GHz
+stepping : 10
+microcode : 0xa07
+cpu MHz : 2997.000
+cache size : 6144 KB
+physical id : 0
+siblings : 2
+core id : 0
+cpu cores : 2
+apicid : 0
+initial apicid : 0
+fpu : yes
+fpu_exception : yes
+cpuid level : 13
+wp : yes
+flags : fpu vme ...
+bogomips : 5993.08
+clflush size : 64
+cache_alignment : 64
+address sizes : 36 bits physical, 48 bits virtual
+power management:
+"""
def write_cpuinfo(info):
filepath = udir.join('get_arch_version')
@@ -20,5 +45,4 @@
assert detect_arch_version(write_cpuinfo(cpuinfo % 8)) == 7
py.test.raises(ValueError,
'detect_arch_version(write_cpuinfo(cpuinfo % 5))')
- py.test.raises(ValueError,
- 'detect_arch_version(write_cpuinfo("Lorem ipsum dolor sit amet,
consectetur"))')
+ assert detect_arch_version(write_cpuinfo(cpuinfo2)) == 6
diff --git a/rpython/jit/backend/arm/test/test_runner.py
b/rpython/jit/backend/arm/test/test_runner.py
--- a/rpython/jit/backend/arm/test/test_runner.py
+++ b/rpython/jit/backend/arm/test/test_runner.py
@@ -10,6 +10,7 @@
from rpython.rtyper.annlowlevel import llhelper
from rpython.jit.codewriter.effectinfo import EffectInfo
from rpython.jit.metainterp.history import JitCellToken, TargetToken
+from rpython.jit.backend.arm.detect import detect_arch_version
CPU = getcpuclass()
@@ -27,7 +28,8 @@
bridge_loop_instructions = ['ldr', 'mov', 'nop', 'cmp', 'bge',
'push', 'mov', 'mov', 'push', 'mov', 'mov',
'blx', 'mov', 'mov', 'bx']
- if CPU.arch_version == 7:
+ arch_version = detect_arch_version()
+ if arch_version == 7:
bridge_loop_instructions = ['ldr', 'mov', 'nop', 'cmp', 'bge',
'push', 'mov', 'mov', 'push', 'mov', 'mov',
'blx', 'mov', 'mov', 'bx']
diff --git a/rpython/jit/backend/test/runner_test.py
b/rpython/jit/backend/test/runner_test.py
--- a/rpython/jit/backend/test/runner_test.py
+++ b/rpython/jit/backend/test/runner_test.py
@@ -3935,3 +3935,19 @@
descr = self.cpu.get_latest_descr(frame)
assert descr.identifier == 42
assert not self.cpu.grab_exc_value(frame)
+
+ def test_setarrayitem_raw_short(self):
+ # setarrayitem_raw(140737353744432, 0, 30583, descr=<ArrayS 2>)
+ A = rffi.CArray(rffi.SHORT)
+ arraydescr = self.cpu.arraydescrof(A)
+ a = lltype.malloc(A, 2, flavor='raw')
+ a[0] = rffi.cast(rffi.SHORT, 666)
+ a[1] = rffi.cast(rffi.SHORT, 777)
+ a_int = rffi.cast(lltype.Signed, a)
+ print 'a_int:', a_int
+ self.execute_operation(rop.SETARRAYITEM_RAW,
+ [ConstInt(a_int), ConstInt(0), ConstInt(-7654)],
+ 'void', descr=arraydescr)
+ assert rffi.cast(lltype.Signed, a[0]) == -7654
+ assert rffi.cast(lltype.Signed, a[1]) == 777
+ lltype.free(a, flavor='raw')
diff --git a/rpython/jit/backend/x86/regloc.py
b/rpython/jit/backend/x86/regloc.py
--- a/rpython/jit/backend/x86/regloc.py
+++ b/rpython/jit/backend/x86/regloc.py
@@ -521,17 +521,6 @@
return func_with_new_name(INSN, "INSN_" + name)
- def _16_bit_binaryop(name):
- def INSN(self, loc1, loc2):
- # Select 16-bit operand mode
- self.writechar('\x66')
- # XXX: Hack to let immediate() in rx86 know to do a 16-bit encoding
- self._use_16_bit_immediate = True
- getattr(self, name)(loc1, loc2)
- self._use_16_bit_immediate = False
-
- return INSN
-
def _addr_as_reg_offset(self, addr):
# Encodes a (64-bit) address as an offset from the scratch register.
# If we are within a "reuse_scratch_register" block, we remember the
@@ -616,10 +605,10 @@
NEG = _unaryop('NEG')
CMP = _binaryop('CMP')
- CMP16 = _16_bit_binaryop('CMP')
+ CMP16 = _binaryop('CMP16')
MOV = _binaryop('MOV')
MOV8 = _binaryop('MOV8')
- MOV16 = _16_bit_binaryop('MOV')
+ MOV16 = _binaryop('MOV16')
MOVZX8 = _binaryop('MOVZX8')
MOVSX8 = _binaryop('MOVSX8')
MOVZX16 = _binaryop('MOVZX16')
diff --git a/rpython/jit/backend/x86/rx86.py b/rpython/jit/backend/x86/rx86.py
--- a/rpython/jit/backend/x86/rx86.py
+++ b/rpython/jit/backend/x86/rx86.py
@@ -125,10 +125,7 @@
elif width == 'q' and mc.WORD == 8:
mc.writeimm64(immediate)
else:
- if mc._use_16_bit_immediate:
- mc.writeimm16(immediate)
- else:
- mc.writeimm32(immediate)
+ mc.writeimm32(immediate)
return 0
def immediate(argnum, width='i'):
@@ -282,16 +279,20 @@
# (the immediate address itself must be explicitely encoded as well,
# with immediate(argnum)).
-def encode_abs(mc, _1, _2, orbyte):
+@specialize.arg(2)
+def encode_abs(mc, immediate, _, orbyte):
# expands to either '\x05' on 32-bit, or '\x04\x25' on 64-bit
if mc.WORD == 8:
mc.writechar(chr(0x04 | orbyte))
mc.writechar(chr(0x25))
else:
mc.writechar(chr(0x05 | orbyte))
+ # followed by an immediate, always 32 bits
+ mc.writeimm32(immediate)
return 0
-abs_ = encode_abs, 0, None, None
+def abs_(argnum):
+ return encode_abs, argnum, None, None
# ____________________________________________________________
# For 64-bits mode: the REX.W, REX.R, REX.X, REG.B prefixes
@@ -305,9 +306,6 @@
def encode_rex(mc, rexbyte, basevalue, orbyte):
if mc.WORD == 8:
assert 0 <= rexbyte < 8
- # XXX: Hack. Ignore REX.W if we are using 16-bit operands
- if mc._use_16_bit_immediate:
- basevalue &= ~REX_W
if basevalue != 0 or rexbyte != 0:
if basevalue == 0:
basevalue = 0x40
@@ -374,9 +372,8 @@
INSN_br = insn(rex_w, chr(base+1), register(2,8), stack_bp(1))
INSN_rb = insn(rex_w, chr(base+3), register(1,8), stack_bp(2))
INSN_rm = insn(rex_w, chr(base+3), register(1,8), mem_reg_plus_const(2))
- INSN_rj = insn(rex_w, chr(base+3), register(1,8), abs_, immediate(2))
- INSN_ji8 = insn(rex_w, '\x83', orbyte(base), abs_, immediate(1),
- immediate(2,'b'))
+ INSN_rj = insn(rex_w, chr(base+3), register(1,8), abs_(2))
+ INSN_ji8 = insn(rex_w, '\x83', orbyte(base), abs_(1), immediate(2,'b'))
INSN_mi8 = insn(rex_w, '\x83', orbyte(base), mem_reg_plus_const(1),
immediate(2,'b'))
INSN_bi8 = insn(rex_w, '\x83', orbyte(base), stack_bp(1), immediate(2,'b'))
@@ -449,9 +446,6 @@
class AbstractX86CodeBuilder(object):
"""Abstract base class."""
- # Used by the 16-bit version of instructions
- _use_16_bit_immediate = False
-
def writechar(self, char):
raise NotImplementedError
@@ -489,15 +483,13 @@
CMP_mi = select_8_or_32_bit_immed(CMP_mi8, CMP_mi32)
CMP_mr = insn(rex_w, '\x39', register(2, 8), mem_reg_plus_const(1))
- CMP_ji8 = insn(rex_w, '\x83', orbyte(7<<3), abs_,
- immediate(1), immediate(2, 'b'))
- CMP_ji32 = insn(rex_w, '\x81', orbyte(7<<3), abs_,
- immediate(1), immediate(2))
+ CMP_ji8 = insn(rex_w, '\x83', orbyte(7<<3), abs_(1), immediate(2, 'b'))
+ CMP_ji32 = insn(rex_w, '\x81', orbyte(7<<3), abs_(1), immediate(2))
CMP_ji = select_8_or_32_bit_immed(CMP_ji8, CMP_ji32)
- CMP_jr = insn(rex_w, '\x39', register(2, 8), abs_, immediate(1))
+ CMP_jr = insn(rex_w, '\x39', register(2, 8), abs_(1))
CMP32_mi = insn(rex_nw, '\x81', orbyte(7<<3), mem_reg_plus_const(1),
immediate(2))
-
+ CMP16_mi = insn('\x66', rex_nw, '\x81', orbyte(7<<3),
mem_reg_plus_const(1), immediate(2, 'h'))
CMP8_ri = insn(rex_fw, '\x80', byte_register(1), '\xF8', immediate(2, 'b'))
AND8_rr = insn(rex_fw, '\x20', byte_register(1), byte_register(2,8),
'\xC0')
@@ -505,7 +497,7 @@
OR8_rr = insn(rex_fw, '\x08', byte_register(1), byte_register(2,8), '\xC0')
OR8_mi = insn(rex_nw, '\x80', orbyte(1<<3), mem_reg_plus_const(1),
immediate(2, 'b'))
- OR8_ji = insn(rex_nw, '\x80', orbyte(1<<3), abs_, immediate(1),
+ OR8_ji = insn(rex_nw, '\x80', orbyte(1<<3), abs_(1),
immediate(2, 'b'))
NEG_r = insn(rex_w, '\xF7', register(1), '\xD8')
@@ -556,7 +548,7 @@
LEA32_rb = insn(rex_w, '\x8D', register(1,8),stack_bp(2,force_32bits=True))
LEA_ra = insn(rex_w, '\x8D', register(1, 8),
mem_reg_plus_scaled_reg_plus_const(2))
LEA_rm = insn(rex_w, '\x8D', register(1, 8), mem_reg_plus_const(2))
- LEA_rj = insn(rex_w, '\x8D', register(1, 8), abs_, immediate(2))
+ LEA_rj = insn(rex_w, '\x8D', register(1, 8), abs_(2))
CALL_l = insn('\xE8', relative(1))
CALL_r = insn(rex_nw, '\xFF', register(1), chr(0xC0 | (2<<3)))
@@ -583,11 +575,11 @@
TEST8_mi = insn(rex_nw, '\xF6', orbyte(0<<3), mem_reg_plus_const(1),
immediate(2, 'b'))
TEST8_bi = insn(rex_nw, '\xF6', orbyte(0<<3), stack_bp(1), immediate(2,
'b'))
- TEST8_ji = insn(rex_nw, '\xF6', orbyte(0<<3), abs_, immediate(1),
immediate(2, 'b'))
+ TEST8_ji = insn(rex_nw, '\xF6', orbyte(0<<3), abs_(1), immediate(2, 'b'))
TEST_rr = insn(rex_w, '\x85', register(2,8), register(1), '\xC0')
BTS_mr = insn(rex_w, '\x0F\xAB', register(2,8), mem_reg_plus_const(1))
- BTS_jr = insn(rex_w, '\x0F\xAB', register(2,8), abs_, immediate(1))
+ BTS_jr = insn(rex_w, '\x0F\xAB', register(2,8), abs_(1))
# x87 instructions
FSTPL_b = insn('\xDD', orbyte(3<<3), stack_bp(1)) # rffi.DOUBLE ('as'
wants L??)
@@ -718,7 +710,7 @@
add_insn('s', stack_sp(modrm_argnum))
add_insn('m', mem_reg_plus_const(modrm_argnum))
add_insn('a', mem_reg_plus_scaled_reg_plus_const(modrm_argnum))
- add_insn('j', abs_, immediate(modrm_argnum))
+ add_insn('j', abs_(modrm_argnum))
# Define a regular MOV, and a variant MOV32 that only uses the low 4 bytes of a
# register
@@ -729,6 +721,8 @@
define_modrm_modes('MOV8_*r', [rex_fw, '\x88', byte_register(2, 8)],
regtype='BYTE')
define_modrm_modes('MOV8_*i', [rex_fw, '\xC6', orbyte(0<<3)], [immediate(2,
'b')], regtype='BYTE')
+define_modrm_modes('MOV16_*r', ['\x66', rex_nw, '\x89', register(2, 8)])
+define_modrm_modes('MOV16_*i', ['\x66', rex_nw, '\xC7', orbyte(0<<3)],
[immediate(2, 'h')])
define_modrm_modes('MOVZX8_r*', [rex_w, '\x0F\xB6', register(1, 8)],
regtype='BYTE')
define_modrm_modes('MOVSX8_r*', [rex_w, '\x0F\xBE', register(1, 8)],
regtype='BYTE')
@@ -766,7 +760,7 @@
#
assert insnname_template.count('*') == 1
add_insn('x', register(2), '\xC0')
- add_insn('j', abs_, immediate(2))
+ add_insn('j', abs_(2))
add_insn('m', mem_reg_plus_const(2))
define_pxmm_insn('PADDQ_x*', '\xD4')
diff --git a/rpython/memory/gc/env.py b/rpython/memory/gc/env.py
--- a/rpython/memory/gc/env.py
+++ b/rpython/memory/gc/env.py
@@ -279,7 +279,7 @@
def best_nursery_size_for_L2cache(L2cache):
# Heuristically, the best nursery size to choose is about half
# of the L2 cache.
- if L2cache > 1024 * 1024: # we don't want to have nursery estimated
+ if L2cache > 2 * 1024 * 1024: # we don't want to have nursery estimated
# on L2 when L3 is present
return L2cache // 2
else:
diff --git a/rpython/memory/gc/minimark.py b/rpython/memory/gc/minimark.py
--- a/rpython/memory/gc/minimark.py
+++ b/rpython/memory/gc/minimark.py
@@ -130,6 +130,7 @@
FORWARDSTUB = lltype.GcStruct('forwarding_stub',
('forw', llmemory.Address))
FORWARDSTUBPTR = lltype.Ptr(FORWARDSTUB)
+NURSARRAY = lltype.Array(llmemory.Address)
# ____________________________________________________________
@@ -263,7 +264,7 @@
self.nursery_top = NULL
self.nursery_real_top = NULL
self.debug_tiny_nursery = -1
- self.debug_rotating_nurseries = None
+ self.debug_rotating_nurseries = lltype.nullptr(NURSARRAY)
self.extra_threshold = 0
#
# The ArenaCollection() handles the nonmovable objects allocation.
@@ -350,8 +351,6 @@
# hacking at the current nursery position in collect_and_reserve().
if newsize <= 0:
newsize = env.estimate_best_nursery_size()
- # 4*1024*1024 # fixed to 4MB by default
- # (it was env.estimate_best_nursery_size())
if newsize <= 0:
newsize = defaultsize
if newsize < minsize:
@@ -471,23 +470,32 @@
# and use them alternatively, while mprotect()ing the unused
# ones to detect invalid access.
debug_start("gc-debug")
- self.debug_rotating_nurseries = []
- for i in range(22):
+ self.debug_rotating_nurseries = lltype.malloc(
+ NURSARRAY, 22, flavor='raw', track_allocation=False)
+ i = 0
+ while i < 22:
nurs = self._alloc_nursery()
llarena.arena_protect(nurs, self._nursery_memory_size(), True)
- self.debug_rotating_nurseries.append(nurs)
+ self.debug_rotating_nurseries[i] = nurs
+ i += 1
debug_print("allocated", len(self.debug_rotating_nurseries),
"extra nurseries")
debug_stop("gc-debug")
def debug_rotate_nursery(self):
- if self.debug_rotating_nurseries is not None:
+ if self.debug_rotating_nurseries:
debug_start("gc-debug")
oldnurs = self.nursery
llarena.arena_protect(oldnurs, self._nursery_memory_size(), True)
- self.debug_rotating_nurseries.append(oldnurs)
#
- newnurs = self.debug_rotating_nurseries.pop(0)
+ newnurs = self.debug_rotating_nurseries[0]
+ i = 0
+ while i < len(self.debug_rotating_nurseries) - 1:
+ self.debug_rotating_nurseries[i] = (
+ self.debug_rotating_nurseries[i + 1])
+ i += 1
+ self.debug_rotating_nurseries[i] = oldnurs
+ #
llarena.arena_protect(newnurs, self._nursery_memory_size(), False)
self.nursery = newnurs
self.nursery_top = self.nursery + self.initial_cleanup
diff --git a/rpython/memory/gctransform/framework.py
b/rpython/memory/gctransform/framework.py
--- a/rpython/memory/gctransform/framework.py
+++ b/rpython/memory/gctransform/framework.py
@@ -619,6 +619,12 @@
func = getattr(graph, 'func', None)
if func and getattr(func, '_gc_no_collect_', False):
if self.collect_analyzer.analyze_direct_call(graph):
+ print '!'*79
+ ca = CollectAnalyzer(self.translator)
+ ca.verbose = True
+ ca.analyze_direct_call(graph)
+ # ^^^ for the dump of which operation in which graph actually
+ # causes it to return True
raise Exception("'no_collect' function can trigger collection:"
" %s" % func)
diff --git a/rpython/translator/backendopt/graphanalyze.py
b/rpython/translator/backendopt/graphanalyze.py
--- a/rpython/translator/backendopt/graphanalyze.py
+++ b/rpython/translator/backendopt/graphanalyze.py
@@ -80,21 +80,21 @@
if graph is None:
x = self.analyze_external_call(op, seen)
if self.verbose and x:
- print '\tanalyze_external_call %s: %r' % (op, x)
+ self.dump_info('analyze_external_call %s: %r' % (op, x))
return x
x = self.analyze_direct_call(graph, seen)
if self.verbose and x:
- print '\tanalyze_direct_call(%s): %r' % (graph, x)
+ self.dump_info('analyze_direct_call(%s): %r' % (graph, x))
return x
elif op.opname == "indirect_call":
graphs = op.args[-1].value
if graphs is None:
if self.verbose:
- print '\t%s to unknown' % (op,)
+ self.dump_info('%s to unknown' % (op,))
return self.top_result()
x = self.analyze_indirect_call(graphs, seen)
if self.verbose and x:
- print '\tanalyze_indirect_call(%s): %r' % (graphs, x)
+ self.dump_info('analyze_indirect_call(%s): %r' % (graphs, x))
return x
elif op.opname == "oosend":
name = op.args[0].value
@@ -106,9 +106,12 @@
return self.analyze_oosend(TYPE, name, seen)
x = self.analyze_simple_operation(op, graphinfo)
if self.verbose and x:
- print '\t%s: %r' % (op, x)
+ self.dump_info('%s: %r' % (op, x))
return x
+ def dump_info(self, info):
+ print '[%s] %s' % (self.__class__.__name__, info)
+
def analyze_direct_call(self, graph, seen=None):
if seen is None:
seen = DependencyTracker(self)
_______________________________________________
pypy-commit mailing list
pypy-commit@python.org
http://mail.python.org/mailman/listinfo/pypy-commit
