Author: Armin Rigo <[email protected]>
Branch: stringbuilder2-perf
Changeset: r72056:070ac1adb8f6
Date: 2014-06-15 12:44 +0200
http://bitbucket.org/pypy/pypy/changeset/070ac1adb8f6/
Log: clean-ups
diff --git a/rpython/rtyper/lltypesystem/rbuilder.py
b/rpython/rtyper/lltypesystem/rbuilder.py
--- a/rpython/rtyper/lltypesystem/rbuilder.py
+++ b/rpython/rtyper/lltypesystem/rbuilder.py
@@ -1,16 +1,14 @@
from rpython.rlib import rgc, jit
-from rpython.rlib.objectmodel import enforceargs, specialize
+from rpython.rlib.objectmodel import enforceargs
from rpython.rlib.rarithmetic import ovfcheck
from rpython.rlib.debug import ll_assert
from rpython.rtyper.rptr import PtrRepr
-from rpython.rtyper.lltypesystem import lltype, llmemory, rffi, rstr
+from rpython.rtyper.lltypesystem import lltype, rffi, rstr
from rpython.rtyper.lltypesystem.lltype import staticAdtMethod, nullptr
from rpython.rtyper.lltypesystem.rstr import (STR, UNICODE, char_repr,
string_repr, unichar_repr, unicode_repr)
from rpython.rtyper.rbuilder import AbstractStringBuilderRepr
from rpython.tool.sourcetools import func_with_new_name
-from rpython.rtyper.llannotation import SomePtr
-from rpython.rtyper.annlowlevel import llstr, llunicode
# ------------------------------------------------------------
@@ -19,25 +17,19 @@
# - A StringBuilder has a rstr.STR of the specified initial size
# (100 by default), which is filled gradually.
#
-# - When it is full, we allocate extra buffers as *raw* memory
-# held by STRINGPIECE objects. The STRINGPIECE has a destructor
-# that frees the memory, but usually the memory is freed explicitly
-# at build() time.
+# - When it is full, we allocate extra buffers as an extra rstr.STR,
+# and the already-filled one is added to a chained list of STRINGPIECE
+# objects.
+#
+# - At build() time, we consolidate all these pieces into a single
+# rstr.STR, which is both returned and re-attached to the StringBuilder,
+# replacing the STRINGPIECEs.
#
# - The data is copied at most twice, and only once in case it fits
# into the initial size (and the GC supports shrinking the STR).
#
-# XXX too much a mess to handle the case where the JIT sees this code.
-# Think about an easier alternative, like using raw_store(current_buf, ..)
-# uniformly, where current_buf is a GC pointer that can be NULL. We'd
-# need support in the JIT to map that to virtual string index. We'd also
-# need a way to express c_memcpy() below --- similar to copystrcontent,
-# but without the assumption that it's about a string (or unicode).
-#
-# XXX alternatively, a simpler solution might be to allocate all pieces
-# as GC-managed rstr.STR. To avoid filling the old generation with
-# garbage we could add a weakref holding the most recently built chain
-# of STRs, and reuse it the next time if it's still there.
+# XXX in build(), we could try keeping around a global weakref to the
+# chain of STRINGPIECEs and reuse them the next time.
#
# ------------------------------------------------------------
@@ -140,16 +132,6 @@
)
-@always_inline
-def ll_str2raw(ll_str, charoffset):
- STRTYPE = lltype.typeOf(ll_str).TO
- ofs = (rffi.offsetof(STRTYPE, 'chars') +
- rffi.itemoffsetof(STRTYPE.chars, 0))
- ofs = llmemory.raw_malloc_usage(ofs) # for direct run
- ofs += rffi.sizeof(STRTYPE.chars.OF) * charoffset
- return rffi.ptradd(rffi.cast(rffi.CCHARP, ll_str), ofs)
-
-
class BaseStringBuilderRepr(AbstractStringBuilderRepr):
def empty(self):
return nullptr(self.lowleveltype.TO)
@@ -181,6 +163,9 @@
@staticmethod
def ll_append_char_2(ll_builder, char0, char1):
+ # this is only used by the JIT, when appending a small, known-length
+ # string. Unlike two consecutive ll_append_char(), it can do that
+ # with only one conditional_call.
ll_builder.skip = 2
jit.conditional_call(
ll_builder.current_end - ll_builder.current_pos < 2,
_______________________________________________
pypy-commit mailing list
[email protected]
https://mail.python.org/mailman/listinfo/pypy-commit
