[Python-checkins] r70071 - in python/trunk: Doc/library/dis.rst Include/opcode.h Lib/compiler/pyassem.py Lib/compiler/pycodegen.py Lib/opcode.py Lib/test/test_dis.py Lib/test/test_peepholer.py Python/ceval.c Python/compile.c Python/import.c Python/peephole.c
jeffrey.yasskin
python-checkins at python.org
Sat Feb 28 20:03:21 CET 2009
Author: jeffrey.yasskin
Date: Sat Feb 28 20:03:21 2009
New Revision: 70071
Log:
Backport r69961 to trunk, replacing JUMP_IF_{TRUE,FALSE} with
POP_JUMP_IF_{TRUE,FALSE} and JUMP_IF_{TRUE,FALSE}_OR_POP. This avoids executing
a POP_TOP on each conditional and sometimes allows the peephole optimizer to
skip a JUMP_ABSOLUTE entirely. It speeds up list comprehensions significantly.
Modified:
python/trunk/Doc/library/dis.rst
python/trunk/Include/opcode.h
python/trunk/Lib/compiler/pyassem.py
python/trunk/Lib/compiler/pycodegen.py
python/trunk/Lib/opcode.py
python/trunk/Lib/test/test_dis.py
python/trunk/Lib/test/test_peepholer.py
python/trunk/Python/ceval.c
python/trunk/Python/compile.c
python/trunk/Python/import.c
python/trunk/Python/peephole.c
Modified: python/trunk/Doc/library/dis.rst
==============================================================================
--- python/trunk/Doc/library/dis.rst (original)
+++ python/trunk/Doc/library/dis.rst Sat Feb 28 20:03:21 2009
@@ -664,16 +664,26 @@
Increments bytecode counter by *delta*.
-.. opcode:: JUMP_IF_TRUE (delta)
+.. opcode:: POP_JUMP_IF_TRUE (target)
- If TOS is true, increment the bytecode counter by *delta*. TOS is left on the
- stack.
+ If TOS is true, sets the bytecode counter to *target*. TOS is popped.
-.. opcode:: JUMP_IF_FALSE (delta)
+.. opcode:: POP_JUMP_IF_FALSE (target)
- If TOS is false, increment the bytecode counter by *delta*. TOS is not
- changed.
+ If TOS is false, sets the bytecode counter to *target*. TOS is popped.
+
+
+.. opcode:: JUMP_IF_TRUE_OR_POP (target)
+
+ If TOS is true, sets the bytecode counter to *target* and leaves TOS
+ on the stack. Otherwise (TOS is false), TOS is popped.
+
+
+.. opcode:: JUMP_IF_FALSE_OR_POP (target)
+
+ If TOS is false, sets the bytecode counter to *target* and leaves
+ TOS on the stack. Otherwise (TOS is true), TOS is popped.
.. opcode:: JUMP_ABSOLUTE (target)
Modified: python/trunk/Include/opcode.h
==============================================================================
--- python/trunk/Include/opcode.h (original)
+++ python/trunk/Include/opcode.h Sat Feb 28 20:03:21 2009
@@ -106,9 +106,12 @@
#define IMPORT_FROM 108 /* Index in name list */
#define JUMP_FORWARD 110 /* Number of bytes to skip */
-#define JUMP_IF_FALSE 111 /* "" */
-#define JUMP_IF_TRUE 112 /* "" */
-#define JUMP_ABSOLUTE 113 /* Target byte offset from beginning of code */
+#define JUMP_IF_FALSE_OR_POP 111 /* Target byte offset from beginning
+ of code */
+#define JUMP_IF_TRUE_OR_POP 112 /* "" */
+#define JUMP_ABSOLUTE 113 /* "" */
+#define POP_JUMP_IF_FALSE 114 /* "" */
+#define POP_JUMP_IF_TRUE 115 /* "" */
#define LOAD_GLOBAL 116 /* Index in name list */
Modified: python/trunk/Lib/compiler/pyassem.py
==============================================================================
--- python/trunk/Lib/compiler/pyassem.py (original)
+++ python/trunk/Lib/compiler/pyassem.py Sat Feb 28 20:03:21 2009
@@ -122,6 +122,9 @@
for b in remaining:
if __debug__ and b.next:
assert b is b.next[0].prev[0], (b, b.next)
+ # Make sure every block appears in dominators, even if no
+ # other block must precede it.
+ dominators.setdefault(b, set())
# preceeding blocks dominate following blocks
for c in b.get_followers():
while 1:
Modified: python/trunk/Lib/compiler/pycodegen.py
==============================================================================
--- python/trunk/Lib/compiler/pycodegen.py (original)
+++ python/trunk/Lib/compiler/pycodegen.py Sat Feb 28 20:03:21 2009
@@ -421,13 +421,11 @@
self.set_lineno(test)
self.visit(test)
nextTest = self.newBlock()
- self.emit('JUMP_IF_FALSE', nextTest)
+ self.emit('POP_JUMP_IF_FALSE', nextTest)
self.nextBlock()
- self.emit('POP_TOP')
self.visit(suite)
self.emit('JUMP_FORWARD', end)
self.startBlock(nextTest)
- self.emit('POP_TOP')
if node.else_:
self.visit(node.else_)
self.nextBlock(end)
@@ -446,15 +444,13 @@
self.set_lineno(node, force=True)
self.visit(node.test)
- self.emit('JUMP_IF_FALSE', else_ or after)
+ self.emit('POP_JUMP_IF_FALSE', else_ or after)
self.nextBlock()
- self.emit('POP_TOP')
self.visit(node.body)
self.emit('JUMP_ABSOLUTE', loop)
self.startBlock(else_) # or just the POPs if not else clause
- self.emit('POP_TOP')
self.emit('POP_BLOCK')
self.setups.pop()
if node.else_:
@@ -525,26 +521,23 @@
self.visit(child)
self.emit(jump, end)
self.nextBlock()
- self.emit('POP_TOP')
self.visit(node.nodes[-1])
self.nextBlock(end)
def visitAnd(self, node):
- self.visitTest(node, 'JUMP_IF_FALSE')
+ self.visitTest(node, 'JUMP_IF_FALSE_OR_POP')
def visitOr(self, node):
- self.visitTest(node, 'JUMP_IF_TRUE')
+ self.visitTest(node, 'JUMP_IF_TRUE_OR_POP')
def visitIfExp(self, node):
endblock = self.newBlock()
elseblock = self.newBlock()
self.visit(node.test)
- self.emit('JUMP_IF_FALSE', elseblock)
- self.emit('POP_TOP')
+ self.emit('POP_JUMP_IF_FALSE', elseblock)
self.visit(node.then)
self.emit('JUMP_FORWARD', endblock)
self.nextBlock(elseblock)
- self.emit('POP_TOP')
self.visit(node.else_)
self.nextBlock(endblock)
@@ -556,9 +549,8 @@
self.emit('DUP_TOP')
self.emit('ROT_THREE')
self.emit('COMPARE_OP', op)
- self.emit('JUMP_IF_FALSE', cleanup)
+ self.emit('JUMP_IF_FALSE_OR_POP', cleanup)
self.nextBlock()
- self.emit('POP_TOP')
# now do the last comparison
if node.ops:
op, code = node.ops[-1]
@@ -593,11 +585,7 @@
for start, cont, anchor in stack:
if cont:
- skip_one = self.newBlock()
- self.emit('JUMP_FORWARD', skip_one)
- self.startBlock(cont)
- self.emit('POP_TOP')
- self.nextBlock(skip_one)
+ self.nextBlock(cont)
self.emit('JUMP_ABSOLUTE', start)
self.startBlock(anchor)
@@ -617,9 +605,8 @@
def visitListCompIf(self, node, branch):
self.set_lineno(node, force=True)
self.visit(node.test)
- self.emit('JUMP_IF_FALSE', branch)
+ self.emit('POP_JUMP_IF_FALSE', branch)
self.newBlock()
- self.emit('POP_TOP')
def _makeClosure(self, gen, args):
frees = gen.scope.get_free_vars()
@@ -665,11 +652,7 @@
for start, cont, anchor, end in stack:
if cont:
- skip_one = self.newBlock()
- self.emit('JUMP_FORWARD', skip_one)
- self.startBlock(cont)
- self.emit('POP_TOP')
- self.nextBlock(skip_one)
+ self.nextBlock(cont)
self.emit('JUMP_ABSOLUTE', start)
self.startBlock(anchor)
self.emit('POP_BLOCK')
@@ -702,9 +685,8 @@
def visitGenExprIf(self, node, branch):
self.set_lineno(node, force=True)
self.visit(node.test)
- self.emit('JUMP_IF_FALSE', branch)
+ self.emit('POP_JUMP_IF_FALSE', branch)
self.newBlock()
- self.emit('POP_TOP')
# exception related
@@ -719,9 +701,8 @@
# is a sort of renaming op.
self.nextBlock()
self.visit(node.test)
- self.emit('JUMP_IF_TRUE', end)
+ self.emit('POP_JUMP_IF_TRUE', end)
self.nextBlock()
- self.emit('POP_TOP')
self.emit('LOAD_GLOBAL', 'AssertionError')
if node.fail:
self.visit(node.fail)
@@ -729,7 +710,6 @@
else:
self.emit('RAISE_VARARGS', 1)
self.nextBlock(end)
- self.emit('POP_TOP')
def visitRaise(self, node):
self.set_lineno(node)
@@ -772,9 +752,8 @@
self.visit(expr)
self.emit('COMPARE_OP', 'exception match')
next = self.newBlock()
- self.emit('JUMP_IF_FALSE', next)
+ self.emit('POP_JUMP_IF_FALSE', next)
self.nextBlock()
- self.emit('POP_TOP')
self.emit('POP_TOP')
if target:
self.visit(target)
@@ -787,8 +766,6 @@
self.nextBlock(next)
else:
self.nextBlock()
- if expr: # XXX
- self.emit('POP_TOP')
self.emit('END_FINALLY')
if node.else_:
self.nextBlock(lElse)
Modified: python/trunk/Lib/opcode.py
==============================================================================
--- python/trunk/Lib/opcode.py (original)
+++ python/trunk/Lib/opcode.py Sat Feb 28 20:03:21 2009
@@ -146,9 +146,11 @@
name_op('IMPORT_FROM', 108) # Index in name list
jrel_op('JUMP_FORWARD', 110) # Number of bytes to skip
-jrel_op('JUMP_IF_FALSE', 111) # ""
-jrel_op('JUMP_IF_TRUE', 112) # ""
-jabs_op('JUMP_ABSOLUTE', 113) # Target byte offset from beginning of code
+jabs_op('JUMP_IF_FALSE_OR_POP', 111) # Target byte offset from beginning of code
+jabs_op('JUMP_IF_TRUE_OR_POP', 112) # ""
+jabs_op('JUMP_ABSOLUTE', 113) # ""
+jabs_op('POP_JUMP_IF_FALSE', 114) # ""
+jabs_op('POP_JUMP_IF_TRUE', 115) # ""
name_op('LOAD_GLOBAL', 116) # Index in name list
Modified: python/trunk/Lib/test/test_dis.py
==============================================================================
--- python/trunk/Lib/test/test_dis.py (original)
+++ python/trunk/Lib/test/test_dis.py Sat Feb 28 20:03:21 2009
@@ -54,25 +54,23 @@
dis_bug1333982 = """\
%-4d 0 LOAD_CONST 1 (0)
- 3 JUMP_IF_TRUE 33 (to 39)
- 6 POP_TOP
- 7 LOAD_GLOBAL 0 (AssertionError)
- 10 BUILD_LIST 0
- 13 LOAD_FAST 0 (x)
- 16 GET_ITER
- >> 17 FOR_ITER 12 (to 32)
- 20 STORE_FAST 1 (s)
- 23 LOAD_FAST 1 (s)
- 26 LIST_APPEND 2
- 29 JUMP_ABSOLUTE 17
-
- %-4d >> 32 LOAD_CONST 2 (1)
- 35 BINARY_ADD
- 36 RAISE_VARARGS 2
- >> 39 POP_TOP
+ 3 POP_JUMP_IF_TRUE 38
+ 6 LOAD_GLOBAL 0 (AssertionError)
+ 9 BUILD_LIST 0
+ 12 LOAD_FAST 0 (x)
+ 15 GET_ITER
+ >> 16 FOR_ITER 12 (to 31)
+ 19 STORE_FAST 1 (s)
+ 22 LOAD_FAST 1 (s)
+ 25 LIST_APPEND 2
+ 28 JUMP_ABSOLUTE 16
+
+ %-4d >> 31 LOAD_CONST 2 (1)
+ 34 BINARY_ADD
+ 35 RAISE_VARARGS 2
- %-4d 40 LOAD_CONST 0 (None)
- 43 RETURN_VALUE
+ %-4d >> 38 LOAD_CONST 0 (None)
+ 41 RETURN_VALUE
"""%(bug1333982.func_code.co_firstlineno + 1,
bug1333982.func_code.co_firstlineno + 2,
bug1333982.func_code.co_firstlineno + 3)
Modified: python/trunk/Lib/test/test_peepholer.py
==============================================================================
--- python/trunk/Lib/test/test_peepholer.py (original)
+++ python/trunk/Lib/test/test_peepholer.py Sat Feb 28 20:03:21 2009
@@ -19,15 +19,14 @@
class TestTranforms(unittest.TestCase):
def test_unot(self):
- # UNARY_NOT JUMP_IF_FALSE POP_TOP --> JUMP_IF_TRUE POP_TOP'
+ # UNARY_NOT POP_JUMP_IF_FALSE --> POP_JUMP_IF_TRUE
def unot(x):
if not x == 2:
del x
asm = disassemble(unot)
- for elem in ('UNARY_NOT', 'JUMP_IF_FALSE'):
+ for elem in ('UNARY_NOT', 'POP_JUMP_IF_FALSE'):
self.assert_(elem not in asm)
- for elem in ('JUMP_IF_TRUE', 'POP_TOP'):
- self.assert_(elem in asm)
+ self.assert_('POP_JUMP_IF_TRUE' in asm)
def test_elim_inversion_of_is_or_in(self):
for line, elem in (
@@ -56,13 +55,13 @@
self.assert_('LOAD_GLOBAL' not in disassemble(f))
def test_while_one(self):
- # Skip over: LOAD_CONST trueconst JUMP_IF_FALSE xx POP_TOP
+ # Skip over: LOAD_CONST trueconst POP_JUMP_IF_FALSE xx
def f():
while 1:
pass
return list
asm = disassemble(f)
- for elem in ('LOAD_CONST', 'JUMP_IF_FALSE'):
+ for elem in ('LOAD_CONST', 'POP_JUMP_IF_FALSE'):
self.assert_(elem not in asm)
for elem in ('JUMP_ABSOLUTE',):
self.assert_(elem in asm)
Modified: python/trunk/Python/ceval.c
==============================================================================
--- python/trunk/Python/ceval.c (original)
+++ python/trunk/Python/ceval.c Sat Feb 28 20:03:21 2009
@@ -764,8 +764,8 @@
/* OpCode prediction macros
Some opcodes tend to come in pairs thus making it possible to
predict the second code when the first is run. For example,
- COMPARE_OP is often followed by JUMP_IF_FALSE or JUMP_IF_TRUE. And,
- those opcodes are often followed by a POP_TOP.
+ GET_ITER is often followed by FOR_ITER. And FOR_ITER is often
+ followed by STORE_FAST or UNPACK_SEQUENCE.
Verifying the prediction costs a single high-speed test of a register
variable against a constant. If the pairing was good, then the
@@ -1110,7 +1110,6 @@
SETLOCAL(oparg, v);
goto fast_next_opcode;
- PREDICTED(POP_TOP);
case POP_TOP:
v = POP();
Py_DECREF(v);
@@ -2220,8 +2219,8 @@
Py_DECREF(w);
SET_TOP(x);
if (x == NULL) break;
- PREDICT(JUMP_IF_FALSE);
- PREDICT(JUMP_IF_TRUE);
+ PREDICT(POP_JUMP_IF_FALSE);
+ PREDICT(POP_JUMP_IF_TRUE);
continue;
case IMPORT_NAME:
@@ -2298,41 +2297,45 @@
JUMPBY(oparg);
goto fast_next_opcode;
- PREDICTED_WITH_ARG(JUMP_IF_FALSE);
- case JUMP_IF_FALSE:
- w = TOP();
+ PREDICTED_WITH_ARG(POP_JUMP_IF_FALSE);
+ case POP_JUMP_IF_FALSE:
+ w = POP();
if (w == Py_True) {
- PREDICT(POP_TOP);
+ Py_DECREF(w);
goto fast_next_opcode;
}
if (w == Py_False) {
- JUMPBY(oparg);
+ Py_DECREF(w);
+ JUMPTO(oparg);
goto fast_next_opcode;
}
err = PyObject_IsTrue(w);
+ Py_DECREF(w);
if (err > 0)
err = 0;
else if (err == 0)
- JUMPBY(oparg);
+ JUMPTO(oparg);
else
break;
continue;
- PREDICTED_WITH_ARG(JUMP_IF_TRUE);
- case JUMP_IF_TRUE:
- w = TOP();
+ PREDICTED_WITH_ARG(POP_JUMP_IF_TRUE);
+ case POP_JUMP_IF_TRUE:
+ w = POP();
if (w == Py_False) {
- PREDICT(POP_TOP);
+ Py_DECREF(w);
goto fast_next_opcode;
}
if (w == Py_True) {
- JUMPBY(oparg);
+ Py_DECREF(w);
+ JUMPTO(oparg);
goto fast_next_opcode;
}
err = PyObject_IsTrue(w);
+ Py_DECREF(w);
if (err > 0) {
err = 0;
- JUMPBY(oparg);
+ JUMPTO(oparg);
}
else if (err == 0)
;
@@ -2340,6 +2343,53 @@
break;
continue;
+ case JUMP_IF_FALSE_OR_POP:
+ w = TOP();
+ if (w == Py_True) {
+ STACKADJ(-1);
+ Py_DECREF(w);
+ goto fast_next_opcode;
+ }
+ if (w == Py_False) {
+ JUMPTO(oparg);
+ goto fast_next_opcode;
+ }
+ err = PyObject_IsTrue(w);
+ if (err > 0) {
+ STACKADJ(-1);
+ Py_DECREF(w);
+ err = 0;
+ }
+ else if (err == 0)
+ JUMPTO(oparg);
+ else
+ break;
+ continue;
+
+ case JUMP_IF_TRUE_OR_POP:
+ w = TOP();
+ if (w == Py_False) {
+ STACKADJ(-1);
+ Py_DECREF(w);
+ goto fast_next_opcode;
+ }
+ if (w == Py_True) {
+ JUMPTO(oparg);
+ goto fast_next_opcode;
+ }
+ err = PyObject_IsTrue(w);
+ if (err > 0) {
+ err = 0;
+ JUMPTO(oparg);
+ }
+ else if (err == 0) {
+ STACKADJ(-1);
+ Py_DECREF(w);
+ }
+ else
+ break;
+ continue;
+
PREDICTED_WITH_ARG(JUMP_ABSOLUTE);
case JUMP_ABSOLUTE:
JUMPTO(oparg);
Modified: python/trunk/Python/compile.c
==============================================================================
--- python/trunk/Python/compile.c (original)
+++ python/trunk/Python/compile.c Sat Feb 28 20:03:21 2009
@@ -836,11 +836,15 @@
return 1;
case JUMP_FORWARD:
- case JUMP_IF_FALSE:
- case JUMP_IF_TRUE:
+ case JUMP_IF_TRUE_OR_POP: /* -1 if jump not taken */
+ case JUMP_IF_FALSE_OR_POP: /* "" */
case JUMP_ABSOLUTE:
return 0;
+ case POP_JUMP_IF_FALSE:
+ case POP_JUMP_IF_TRUE:
+ return -1;
+
case LOAD_GLOBAL:
return 1;
@@ -1499,12 +1503,10 @@
if (next == NULL)
return 0;
VISIT(c, expr, e->v.IfExp.test);
- ADDOP_JREL(c, JUMP_IF_FALSE, next);
- ADDOP(c, POP_TOP);
+ ADDOP_JABS(c, POP_JUMP_IF_FALSE, next);
VISIT(c, expr, e->v.IfExp.body);
ADDOP_JREL(c, JUMP_FORWARD, end);
compiler_use_next_block(c, next);
- ADDOP(c, POP_TOP);
VISIT(c, expr, e->v.IfExp.orelse);
compiler_use_next_block(c, end);
return 1;
@@ -1597,9 +1599,6 @@
end = compiler_new_block(c);
if (end == NULL)
return 0;
- next = compiler_new_block(c);
- if (next == NULL)
- return 0;
constant = expr_constant(s->v.If.test);
/* constant = 0: "if 0"
@@ -1611,15 +1610,21 @@
} else if (constant == 1) {
VISIT_SEQ(c, stmt, s->v.If.body);
} else {
+ if (s->v.If.orelse) {
+ next = compiler_new_block(c);
+ if (next == NULL)
+ return 0;
+ }
+ else
+ next = end;
VISIT(c, expr, s->v.If.test);
- ADDOP_JREL(c, JUMP_IF_FALSE, next);
- ADDOP(c, POP_TOP);
+ ADDOP_JABS(c, POP_JUMP_IF_FALSE, next);
VISIT_SEQ(c, stmt, s->v.If.body);
ADDOP_JREL(c, JUMP_FORWARD, end);
- compiler_use_next_block(c, next);
- ADDOP(c, POP_TOP);
- if (s->v.If.orelse)
+ if (s->v.If.orelse) {
+ compiler_use_next_block(c, next);
VISIT_SEQ(c, stmt, s->v.If.orelse);
+ }
}
compiler_use_next_block(c, end);
return 1;
@@ -1693,8 +1698,7 @@
so we need to set an extra line number. */
c->u->u_lineno_set = false;
VISIT(c, expr, s->v.While.test);
- ADDOP_JREL(c, JUMP_IF_FALSE, anchor);
- ADDOP(c, POP_TOP);
+ ADDOP_JABS(c, POP_JUMP_IF_FALSE, anchor);
}
VISIT_SEQ(c, stmt, s->v.While.body);
ADDOP_JABS(c, JUMP_ABSOLUTE, loop);
@@ -1705,7 +1709,6 @@
if (constant == -1) {
compiler_use_next_block(c, anchor);
- ADDOP(c, POP_TOP);
ADDOP(c, POP_BLOCK);
}
compiler_pop_fblock(c, LOOP, loop);
@@ -1826,20 +1829,17 @@
[tb, val, exc] L1: DUP )
[tb, val, exc, exc] <evaluate E1> )
[tb, val, exc, exc, E1] COMPARE_OP EXC_MATCH ) only if E1
- [tb, val, exc, 1-or-0] JUMP_IF_FALSE L2 )
- [tb, val, exc, 1] POP )
+ [tb, val, exc, 1-or-0] POP_JUMP_IF_FALSE L2 )
[tb, val, exc] POP
[tb, val] <assign to V1> (or POP if no V1)
[tb] POP
[] <code for S1>
JUMP_FORWARD L0
- [tb, val, exc, 0] L2: POP
- [tb, val, exc] DUP
+ [tb, val, exc] L2: DUP
.............................etc.......................
- [tb, val, exc, 0] Ln+1: POP
- [tb, val, exc] END_FINALLY # re-raise exception
+ [tb, val, exc] Ln+1: END_FINALLY # re-raise exception
[] L0: <next statement>
@@ -1881,8 +1881,7 @@
ADDOP(c, DUP_TOP);
VISIT(c, expr, handler->v.ExceptHandler.type);
ADDOP_I(c, COMPARE_OP, PyCmp_EXC_MATCH);
- ADDOP_JREL(c, JUMP_IF_FALSE, except);
- ADDOP(c, POP_TOP);
+ ADDOP_JABS(c, POP_JUMP_IF_FALSE, except);
}
ADDOP(c, POP_TOP);
if (handler->v.ExceptHandler.name) {
@@ -1895,8 +1894,6 @@
VISIT_SEQ(c, stmt, handler->v.ExceptHandler.body);
ADDOP_JREL(c, JUMP_FORWARD, end);
compiler_use_next_block(c, except);
- if (handler->v.ExceptHandler.type)
- ADDOP(c, POP_TOP);
}
ADDOP(c, END_FINALLY);
compiler_use_next_block(c, orelse);
@@ -2084,8 +2081,7 @@
end = compiler_new_block(c);
if (end == NULL)
return 0;
- ADDOP_JREL(c, JUMP_IF_TRUE, end);
- ADDOP(c, POP_TOP);
+ ADDOP_JABS(c, POP_JUMP_IF_TRUE, end);
ADDOP_O(c, LOAD_GLOBAL, assertion_error, names);
if (s->v.Assert.msg) {
VISIT(c, expr, s->v.Assert.msg);
@@ -2095,7 +2091,6 @@
ADDOP_I(c, RAISE_VARARGS, 1);
}
compiler_use_next_block(c, end);
- ADDOP(c, POP_TOP);
return 1;
}
@@ -2473,9 +2468,9 @@
assert(e->kind == BoolOp_kind);
if (e->v.BoolOp.op == And)
- jumpi = JUMP_IF_FALSE;
+ jumpi = JUMP_IF_FALSE_OR_POP;
else
- jumpi = JUMP_IF_TRUE;
+ jumpi = JUMP_IF_TRUE_OR_POP;
end = compiler_new_block(c);
if (end == NULL)
return 0;
@@ -2484,8 +2479,7 @@
assert(n >= 0);
for (i = 0; i < n; ++i) {
VISIT(c, expr, (expr_ty)asdl_seq_GET(s, i));
- ADDOP_JREL(c, jumpi, end);
- ADDOP(c, POP_TOP)
+ ADDOP_JABS(c, jumpi, end);
}
VISIT(c, expr, (expr_ty)asdl_seq_GET(s, n));
compiler_use_next_block(c, end);
@@ -2543,9 +2537,8 @@
ADDOP_I(c, COMPARE_OP,
cmpop((cmpop_ty)(asdl_seq_GET(
e->v.Compare.ops, i - 1))));
- ADDOP_JREL(c, JUMP_IF_FALSE, cleanup);
+ ADDOP_JABS(c, JUMP_IF_FALSE_OR_POP, cleanup);
NEXT_BLOCK(c);
- ADDOP(c, POP_TOP);
if (i < (n - 1))
VISIT(c, expr,
(expr_ty)asdl_seq_GET(e->v.Compare.comparators, i));
@@ -2636,9 +2629,8 @@
for (i = 0; i < n; i++) {
expr_ty e = (expr_ty)asdl_seq_GET(l->ifs, i);
VISIT(c, expr, e);
- ADDOP_JREL(c, JUMP_IF_FALSE, if_cleanup);
+ ADDOP_JABS(c, POP_JUMP_IF_FALSE, if_cleanup);
NEXT_BLOCK(c);
- ADDOP(c, POP_TOP);
}
if (++gen_index < asdl_seq_LEN(generators))
@@ -2652,12 +2644,7 @@
compiler_use_next_block(c, skip);
}
- for (i = 0; i < n; i++) {
- ADDOP_I(c, JUMP_FORWARD, 1);
- if (i == 0)
- compiler_use_next_block(c, if_cleanup);
- ADDOP(c, POP_TOP);
- }
+ compiler_use_next_block(c, if_cleanup);
ADDOP_JABS(c, JUMP_ABSOLUTE, start);
compiler_use_next_block(c, anchor);
@@ -2720,10 +2707,9 @@
for (i = 0; i < n; i++) {
expr_ty e = (expr_ty)asdl_seq_GET(ge->ifs, i);
VISIT(c, expr, e);
- ADDOP_JREL(c, JUMP_IF_FALSE, if_cleanup);
+ ADDOP_JABS(c, POP_JUMP_IF_FALSE, if_cleanup);
NEXT_BLOCK(c);
- ADDOP(c, POP_TOP);
- }
+ }
if (++gen_index < asdl_seq_LEN(generators))
if (!compiler_genexp_generator(c, generators, gen_index, elt))
@@ -2737,13 +2723,7 @@
compiler_use_next_block(c, skip);
}
- for (i = 0; i < n; i++) {
- ADDOP_I(c, JUMP_FORWARD, 1);
- if (i == 0)
- compiler_use_next_block(c, if_cleanup);
-
- ADDOP(c, POP_TOP);
- }
+ compiler_use_next_block(c, if_cleanup);
ADDOP_JABS(c, JUMP_ABSOLUTE, start);
compiler_use_next_block(c, anchor);
ADDOP(c, POP_BLOCK);
Modified: python/trunk/Python/import.c
==============================================================================
--- python/trunk/Python/import.c (original)
+++ python/trunk/Python/import.c Sat Feb 28 20:03:21 2009
@@ -72,9 +72,11 @@
Python 2.6a0: 62151 (peephole optimizations and STORE_MAP opcode)
Python 2.6a1: 62161 (WITH_CLEANUP optimization)
Python 2.7a0: 62171 (optimize list comprehensions/change LIST_APPEND)
+ Python 2.7a0: 62181 (optimize conditional branches:
+ introduce POP_JUMP_IF_FALSE and POP_JUMP_IF_TRUE)
.
*/
-#define MAGIC (62171 | ((long)'\r'<<16) | ((long)'\n'<<24))
+#define MAGIC (62181 | ((long)'\r'<<16) | ((long)'\n'<<24))
/* Magic word as global; note that _PyImport_Init() can change the
value of this global to accommodate for alterations of how the
Modified: python/trunk/Python/peephole.c
==============================================================================
--- python/trunk/Python/peephole.c (original)
+++ python/trunk/Python/peephole.c Sat Feb 28 20:03:21 2009
@@ -13,7 +13,12 @@
#define GETARG(arr, i) ((int)((arr[i+2]<<8) + arr[i+1]))
#define UNCONDITIONAL_JUMP(op) (op==JUMP_ABSOLUTE || op==JUMP_FORWARD)
-#define ABSOLUTE_JUMP(op) (op==JUMP_ABSOLUTE || op==CONTINUE_LOOP)
+#define CONDITIONAL_JUMP(op) (op==POP_JUMP_IF_FALSE || op==POP_JUMP_IF_TRUE \
+ || op==JUMP_IF_FALSE_OR_POP || op==JUMP_IF_TRUE_OR_POP)
+#define ABSOLUTE_JUMP(op) (op==JUMP_ABSOLUTE || op==CONTINUE_LOOP \
+ || op==POP_JUMP_IF_FALSE || op==POP_JUMP_IF_TRUE \
+ || op==JUMP_IF_FALSE_OR_POP || op==JUMP_IF_TRUE_OR_POP)
+#define JUMPS_ON_TRUE(op) (op==POP_JUMP_IF_TRUE || op==JUMP_IF_TRUE_OR_POP)
#define GETJUMPTGT(arr, i) (GETARG(arr,i) + (ABSOLUTE_JUMP(arr[i]) ? 0 : i+3))
#define SETARG(arr, i, val) arr[i+2] = val>>8; arr[i+1] = val & 255
#define CODESIZE(op) (HAS_ARG(op) ? 3 : 1)
@@ -245,8 +250,10 @@
switch (opcode) {
case FOR_ITER:
case JUMP_FORWARD:
- case JUMP_IF_FALSE:
- case JUMP_IF_TRUE:
+ case JUMP_IF_FALSE_OR_POP:
+ case JUMP_IF_TRUE_OR_POP:
+ case POP_JUMP_IF_FALSE:
+ case POP_JUMP_IF_TRUE:
case JUMP_ABSOLUTE:
case CONTINUE_LOOP:
case SETUP_LOOP:
@@ -338,29 +345,24 @@
assert(PyList_Check(consts));
for (i=0 ; i<codelen ; i += CODESIZE(codestr[i])) {
+ reoptimize_current:
opcode = codestr[i];
lastlc = cumlc;
cumlc = 0;
switch (opcode) {
-
- /* Replace UNARY_NOT JUMP_IF_FALSE POP_TOP with
- with JUMP_IF_TRUE POP_TOP */
+ /* Replace UNARY_NOT POP_JUMP_IF_FALSE
+ with POP_JUMP_IF_TRUE */
case UNARY_NOT:
- if (codestr[i+1] != JUMP_IF_FALSE ||
- codestr[i+4] != POP_TOP ||
- !ISBASICBLOCK(blocks,i,5))
- continue;
- tgt = GETJUMPTGT(codestr, (i+1));
- if (codestr[tgt] != POP_TOP)
+ if (codestr[i+1] != POP_JUMP_IF_FALSE
+ || !ISBASICBLOCK(blocks,i,4))
continue;
- j = GETARG(codestr, i+1) + 1;
- codestr[i] = JUMP_IF_TRUE;
+ j = GETARG(codestr, i+1);
+ codestr[i] = POP_JUMP_IF_TRUE;
SETARG(codestr, i, j);
- codestr[i+3] = POP_TOP;
- codestr[i+4] = NOP;
- break;
+ codestr[i+3] = NOP;
+ goto reoptimize_current;
/* not a is b --> a is not b
not a in b --> a not in b
@@ -400,16 +402,16 @@
break;
/* Skip over LOAD_CONST trueconst
- JUMP_IF_FALSE xx POP_TOP */
+ POP_JUMP_IF_FALSE xx. This improves
+ "while 1" performance. */
case LOAD_CONST:
cumlc = lastlc + 1;
j = GETARG(codestr, i);
- if (codestr[i+3] != JUMP_IF_FALSE ||
- codestr[i+6] != POP_TOP ||
- !ISBASICBLOCK(blocks,i,7) ||
+ if (codestr[i+3] != POP_JUMP_IF_FALSE ||
+ !ISBASICBLOCK(blocks,i,6) ||
!PyObject_IsTrue(PyList_GET_ITEM(consts, j)))
continue;
- memset(codestr+i, NOP, 7);
+ memset(codestr+i, NOP, 6);
cumlc = 0;
break;
@@ -498,27 +500,49 @@
"if a or b:"
"a and b or c"
"(a and b) and c"
- x:JUMP_IF_FALSE y y:JUMP_IF_FALSE z --> x:JUMP_IF_FALSE z
- x:JUMP_IF_FALSE y y:JUMP_IF_TRUE z --> x:JUMP_IF_FALSE y+3
+ x:JUMP_IF_FALSE_OR_POP y y:JUMP_IF_FALSE_OR_POP z
+ --> x:JUMP_IF_FALSE_OR_POP z
+ x:JUMP_IF_FALSE_OR_POP y y:JUMP_IF_TRUE_OR_POP z
+ --> x:POP_JUMP_IF_FALSE y+3
where y+3 is the instruction following the second test.
*/
- case JUMP_IF_FALSE:
- case JUMP_IF_TRUE:
+ case JUMP_IF_FALSE_OR_POP:
+ case JUMP_IF_TRUE_OR_POP:
tgt = GETJUMPTGT(codestr, i);
j = codestr[tgt];
- if (j == JUMP_IF_FALSE || j == JUMP_IF_TRUE) {
- if (j == opcode) {
- tgttgt = GETJUMPTGT(codestr, tgt) - i - 3;
+ if (CONDITIONAL_JUMP(j)) {
+ /* NOTE: all possible jumps here are
+ absolute! */
+ if (JUMPS_ON_TRUE(j) == JUMPS_ON_TRUE(opcode)) {
+ /* The second jump will be
+ taken iff the first is. */
+ tgttgt = GETJUMPTGT(codestr, tgt);
+ /* The current opcode inherits
+ its target's stack behaviour */
+ codestr[i] = j;
SETARG(codestr, i, tgttgt);
+ goto reoptimize_current;
} else {
- tgt -= i;
- SETARG(codestr, i, tgt);
+ /* The second jump is not taken
+ if the first is (so jump past
+ it), and all conditional
+ jumps pop their argument when
+ they're not taken (so change
+ the first jump to pop its
+ argument when it's taken). */
+ if (JUMPS_ON_TRUE(opcode))
+ codestr[i] = POP_JUMP_IF_TRUE;
+ else
+ codestr[i] = POP_JUMP_IF_FALSE;
+ SETARG(codestr, i, (tgt + 3));
+ goto reoptimize_current;
}
- break;
}
- /* Intentional fallthrough */
+ /* Intentional fallthrough */
/* Replace jumps to unconditional jumps */
+ case POP_JUMP_IF_FALSE:
+ case POP_JUMP_IF_TRUE:
case FOR_ITER:
case JUMP_FORWARD:
case JUMP_ABSOLUTE:
@@ -591,14 +615,16 @@
case JUMP_ABSOLUTE:
case CONTINUE_LOOP:
+ case POP_JUMP_IF_FALSE:
+ case POP_JUMP_IF_TRUE:
+ case JUMP_IF_FALSE_OR_POP:
+ case JUMP_IF_TRUE_OR_POP:
j = addrmap[GETARG(codestr, i)];
SETARG(codestr, i, j);
break;
case FOR_ITER:
case JUMP_FORWARD:
- case JUMP_IF_FALSE:
- case JUMP_IF_TRUE:
case SETUP_LOOP:
case SETUP_EXCEPT:
case SETUP_FINALLY:
More information about the Python-checkins
mailing list