[git commit] bc: use more compact parsing data structures
Denys Vlasenko
vda.linux at googlemail.com
Fri Dec 7 11:57:32 UTC 2018
commit: https://git.busybox.net/busybox/commit/?id=18c6b54f820923549135724fee6cf66c26929b07
branch: https://git.busybox.net/busybox/commit/?id=refs/heads/master
function old new delta
dc_lex_token 697 701 +4
bc_parse_next_rel 20 - -20
bc_parse_next_read 20 - -20
bc_parse_next_print 20 - -20
bc_parse_next_param 20 - -20
bc_parse_next_for 20 - -20
bc_parse_next_expr 20 - -20
bc_parse_next_elem 20 - -20
common_parse_expr 62 40 -22
bc_parse_expr 49 24 -25
dc_lex_regs 52 13 -39
bc_parse_name 581 539 -42
bc_parse_expr_empty_ok 2157 2108 -49
dc_parse_insts 332 83 -249
dc_lex_tokens 364 91 -273
bc_parse_stmt 2261 1868 -393
------------------------------------------------------------------------------
(add/remove: 0/7 grow/shrink: 1/8 up/down: 4/-1232) Total: -1228 bytes
text data bss dec hex filename
987037 485 7296 994818 f2e02 busybox_old
985814 485 7296 993595 f293b busybox_unstripped
Signed-off-by: Denys Vlasenko <vda.linux at googlemail.com>
---
miscutils/bc.c | 62 +++++++++++++++++++++++++++++++++-------------------------
1 file changed, 35 insertions(+), 27 deletions(-)
diff --git a/miscutils/bc.c b/miscutils/bc.c
index 0330c43e5..685427e58 100644
--- a/miscutils/bc.c
+++ b/miscutils/bc.c
@@ -632,11 +632,6 @@ typedef struct BcLex {
BC_PARSE_FLAG_LOOP | BC_PARSE_FLAG_LOOP_INNER | BC_PARSE_FLAG_IF | \
BC_PARSE_FLAG_ELSE | BC_PARSE_FLAG_IF_END)))
-typedef struct BcParseNext {
- uint32_t len;
- BcLexType tokens[4];
-} BcParseNext;
-
struct BcParse;
struct BcProgram;
@@ -834,34 +829,40 @@ static const uint8_t bc_parse_ops[] = {
#define bc_parse_op_PREC(i) (bc_parse_ops[i] & 0x0f)
#define bc_parse_op_LEFT(i) (bc_parse_ops[i] & 0x10)
+// Byte array of up to 4 BC_LEX's, packed into 32-bit word
+typedef uint32_t BcParseNext;
+
// These identify what tokens can come after expressions in certain cases.
-#define BC_PARSE_NEXT_TOKENS(...) .tokens = { __VA_ARGS__ }
-#define BC_PARSE_NEXT(a, ...) \
- { \
- .len = (a), BC_PARSE_NEXT_TOKENS(__VA_ARGS__) \
- }
-static const BcParseNext bc_parse_next_expr =
- BC_PARSE_NEXT(4, BC_LEX_NLINE, BC_LEX_SCOLON, BC_LEX_RBRACE, BC_LEX_EOF);
-static const BcParseNext bc_parse_next_param =
- BC_PARSE_NEXT(2, BC_LEX_RPAREN, BC_LEX_COMMA);
-static const BcParseNext bc_parse_next_print =
- BC_PARSE_NEXT(4, BC_LEX_COMMA, BC_LEX_NLINE, BC_LEX_SCOLON, BC_LEX_EOF);
-static const BcParseNext bc_parse_next_rel = BC_PARSE_NEXT(1, BC_LEX_RPAREN);
-static const BcParseNext bc_parse_next_elem = BC_PARSE_NEXT(1, BC_LEX_RBRACKET);
-static const BcParseNext bc_parse_next_for = BC_PARSE_NEXT(1, BC_LEX_SCOLON);
-static const BcParseNext bc_parse_next_read =
- BC_PARSE_NEXT(2, BC_LEX_NLINE, BC_LEX_EOF);
+enum {
+#define BC_PARSE_NEXT4(a,b,c,d) ( (a) | ((b)<<8) | ((c)<<16) | ((((d)|0x80)<<24)) )
+#define BC_PARSE_NEXT2(a,b) BC_PARSE_NEXT4(a,b,0xff,0xff)
+#define BC_PARSE_NEXT1(a) BC_PARSE_NEXT4(a,0xff,0xff,0xff)
+ bc_parse_next_expr = BC_PARSE_NEXT4(BC_LEX_NLINE, BC_LEX_SCOLON, BC_LEX_RBRACE, BC_LEX_EOF),
+ bc_parse_next_param = BC_PARSE_NEXT2(BC_LEX_RPAREN, BC_LEX_COMMA),
+ bc_parse_next_print = BC_PARSE_NEXT4(BC_LEX_COMMA, BC_LEX_NLINE, BC_LEX_SCOLON, BC_LEX_EOF),
+ bc_parse_next_rel = BC_PARSE_NEXT1(BC_LEX_RPAREN),
+ bc_parse_next_elem = BC_PARSE_NEXT1(BC_LEX_RBRACKET),
+ bc_parse_next_for = BC_PARSE_NEXT1(BC_LEX_SCOLON),
+ bc_parse_next_read = BC_PARSE_NEXT2(BC_LEX_NLINE, BC_LEX_EOF),
+#undef BC_PARSE_NEXT4
+#undef BC_PARSE_NEXT2
+#undef BC_PARSE_NEXT1
+};
#endif // ENABLE_BC
#if ENABLE_DC
-static const BcLexType dc_lex_regs[] = {
+static const //BcLexType - should be this type, but narrower type saves size:
+uint8_t
+dc_lex_regs[] = {
BC_LEX_OP_REL_EQ, BC_LEX_OP_REL_LE, BC_LEX_OP_REL_GE, BC_LEX_OP_REL_NE,
BC_LEX_OP_REL_LT, BC_LEX_OP_REL_GT, BC_LEX_SCOLON, BC_LEX_COLON,
BC_LEX_ELSE, BC_LEX_LOAD, BC_LEX_LOAD_POP, BC_LEX_OP_ASSIGN,
BC_LEX_STORE_PUSH,
};
-static const BcLexType dc_lex_tokens[] = {
+static const //BcLexType - should be this type
+uint8_t
+dc_lex_tokens[] = {
BC_LEX_OP_MODULUS, BC_LEX_INVALID, BC_LEX_INVALID, BC_LEX_LPAREN,
BC_LEX_INVALID, BC_LEX_OP_MULTIPLY, BC_LEX_OP_PLUS, BC_LEX_INVALID,
BC_LEX_OP_MINUS, BC_LEX_INVALID, BC_LEX_OP_DIVIDE,
@@ -889,7 +890,9 @@ static const BcLexType dc_lex_tokens[] = {
BC_LEX_INVALID
};
-static const BcInst dc_parse_insts[] = {
+static const //BcInst - should be this type. Using signed narrow type since BC_INST_INVALID is -1
+int8_t
+dc_parse_insts[] = {
BC_INST_INVALID, BC_INST_INVALID, BC_INST_INVALID, BC_INST_REL_GE,
BC_INST_INVALID, BC_INST_POWER, BC_INST_MULTIPLY, BC_INST_DIVIDE,
BC_INST_MODULUS, BC_INST_PLUS, BC_INST_MINUS,
@@ -4768,7 +4771,7 @@ static BcStatus bc_parse_expr_empty_ok(BcParse *p, uint8_t flags, BcParseNext ne
BcInst prev = BC_INST_PRINT;
BcLexType top, t = p->l.t.t;
size_t nexprs = 0, ops_bgn = p->ops.len;
- uint32_t i, nparens, nrelops;
+ unsigned nparens, nrelops;
bool paren_first, paren_expr, rprn, done, get_token, assign, bin_last;
paren_first = p->l.t.t == BC_LEX_LPAREN;
@@ -4993,9 +4996,14 @@ static BcStatus bc_parse_expr_empty_ok(BcParse *p, uint8_t flags, BcParseNext ne
if (prev == BC_INST_BOOL_NOT || nexprs != 1)
return bc_error_bad_expression();
- for (i = 0; i < next.len; ++i)
- if (t == next.tokens[i])
+ // next is BcParseNext, byte array of up to 4 BC_LEX's, packed into 32-bit word
+ for (;;) {
+ if (t == (next & 0x7f))
goto ok;
+ if (next & 0x80) // last element?
+ break;
+ next >>= 8;
+ }
return bc_error_bad_expression();
ok:
More information about the busybox-cvs
mailing list