| Message ID | 20191003001437.76038-1-luc.vanoostenryck@gmail.com (mailing list archive) |
|---|---|
| State | Mainlined, archived |
| Headers | show |
| Series | expand more builtins like __builtin_ffs() | expand |
On 10/2/19 5:14 PM, Luc Van Oostenryck wrote: > GCC expands at compile time builtins like __builtin_ffs() > when their argument is constant. > > A driver in the kernel uses such a builtin in a case statement, > causing sparse to report: > error: Expected constant expression in case statement > > So, let sparse also expand such builtins, somehow like it was > done for bswap16/32/64 but now for ffs/ffsl/ffsll, clz, ctz, > clrsb, popcount & parity. > > Reported-by: Randy Dunlap <rdunlap@infradead.org> > Signed-off-by: Luc Van Oostenryck <luc.vanoostenryck@gmail.com> Tested-by: Randy Dunlap <rdunlap@infradead.org> Thanks. > --- > builtin.c | 56 +++++++++++++++++++++++++++++++++++++++++++++++++++++++ > 1 file changed, 56 insertions(+) > > diff --git a/builtin.c b/builtin.c > index 221c98991..3e2d77e75 100644 > --- a/builtin.c > +++ b/builtin.c > @@ -230,6 +230,44 @@ static struct symbol_op bswap_op = { > }; > > > +#define EXPAND_FINDBIT(name) \ > +static int expand_##name(struct expression *expr, int cost) \ > +{ \ > + struct expression *arg; \ > + long long val; \ > + \ > + if (cost) \ > + return cost; \ > + \ > + arg = first_expression(expr->args); \ > + val = get_expression_value_silent(arg); \ > + switch (expr->ctype->bit_size) { \ > + case sizeof(int) * 8: \ > + val = __builtin_##name(val); break; \ > + case sizeof(long long) * 8: \ > + val = __builtin_##name##ll(val); break; \ > + default: /* impossible error */ \ > + return SIDE_EFFECTS; \ > + } \ > + \ > + expr->value = val; \ > + expr->type = EXPR_VALUE; \ > + expr->taint = 0; \ > + return 0; \ > +} \ > + \ > +static struct symbol_op name##_op = { \ > + .evaluate = evaluate_pure_unop, \ > + .expand = expand_##name, \ > +} > + > +EXPAND_FINDBIT(clz); > +EXPAND_FINDBIT(ctz); > +EXPAND_FINDBIT(clrsb); > +EXPAND_FINDBIT(ffs); > +EXPAND_FINDBIT(parity); > +EXPAND_FINDBIT(popcount); > + > static int evaluate_fp_unop(struct expression *expr) > { > struct expression *arg; > @@ -334,11 +372,29 @@ static struct sym_init { > { "__builtin_bswap16", &builtin_fn_type, MOD_TOPLEVEL, &bswap_op }, > { "__builtin_bswap32", &builtin_fn_type, MOD_TOPLEVEL, &bswap_op }, > { "__builtin_bswap64", &builtin_fn_type, MOD_TOPLEVEL, &bswap_op }, > + { "__builtin_clrsb", &builtin_fn_type, MOD_TOPLEVEL, &clrsb_op }, > + { "__builtin_clrsbl", &builtin_fn_type, MOD_TOPLEVEL, &clrsb_op }, > + { "__builtin_clrsbll", &builtin_fn_type, MOD_TOPLEVEL, &clrsb_op }, > + { "__builtin_clz", &builtin_fn_type, MOD_TOPLEVEL, &clz_op }, > + { "__builtin_clzl", &builtin_fn_type, MOD_TOPLEVEL, &clz_op }, > + { "__builtin_clzll", &builtin_fn_type, MOD_TOPLEVEL, &clz_op }, > + { "__builtin_ctz", &builtin_fn_type, MOD_TOPLEVEL, &ctz_op }, > + { "__builtin_ctzl", &builtin_fn_type, MOD_TOPLEVEL, &ctz_op }, > + { "__builtin_ctzll", &builtin_fn_type, MOD_TOPLEVEL, &ctz_op }, > + { "__builtin_ffs", &builtin_fn_type, MOD_TOPLEVEL, &ffs_op }, > + { "__builtin_ffsl", &builtin_fn_type, MOD_TOPLEVEL, &ffs_op }, > + { "__builtin_ffsll", &builtin_fn_type, MOD_TOPLEVEL, &ffs_op }, > { "__builtin_isfinite", &builtin_fn_type, MOD_TOPLEVEL, &fp_unop_op }, > { "__builtin_isinf", &builtin_fn_type, MOD_TOPLEVEL, &fp_unop_op }, > { "__builtin_isinf_sign", &builtin_fn_type, MOD_TOPLEVEL, &fp_unop_op }, > { "__builtin_isnan", &builtin_fn_type, MOD_TOPLEVEL, &fp_unop_op }, > { "__builtin_isnormal", &builtin_fn_type, MOD_TOPLEVEL, &fp_unop_op }, > + { "__builtin_parity", &builtin_fn_type, MOD_TOPLEVEL, &parity_op }, > + { "__builtin_parityl", &builtin_fn_type, MOD_TOPLEVEL, &parity_op }, > + { "__builtin_parityll", &builtin_fn_type, MOD_TOPLEVEL, &parity_op }, > + { "__builtin_popcount", &builtin_fn_type, MOD_TOPLEVEL, &popcount_op }, > + { "__builtin_popcountl", &builtin_fn_type, MOD_TOPLEVEL, &popcount_op }, > + { "__builtin_popcountll", &builtin_fn_type, MOD_TOPLEVEL, &popcount_op }, > { "__builtin_signbit", &builtin_fn_type, MOD_TOPLEVEL, &fp_unop_op }, > { "__builtin_add_overflow", &builtin_fn_type, MOD_TOPLEVEL, &overflow_op }, > { "__builtin_sub_overflow", &builtin_fn_type, MOD_TOPLEVEL, &overflow_op }, >
diff --git a/builtin.c b/builtin.c index 221c98991..3e2d77e75 100644 --- a/builtin.c +++ b/builtin.c @@ -230,6 +230,44 @@ static struct symbol_op bswap_op = { }; +#define EXPAND_FINDBIT(name) \ +static int expand_##name(struct expression *expr, int cost) \ +{ \ + struct expression *arg; \ + long long val; \ + \ + if (cost) \ + return cost; \ + \ + arg = first_expression(expr->args); \ + val = get_expression_value_silent(arg); \ + switch (expr->ctype->bit_size) { \ + case sizeof(int) * 8: \ + val = __builtin_##name(val); break; \ + case sizeof(long long) * 8: \ + val = __builtin_##name##ll(val); break; \ + default: /* impossible error */ \ + return SIDE_EFFECTS; \ + } \ + \ + expr->value = val; \ + expr->type = EXPR_VALUE; \ + expr->taint = 0; \ + return 0; \ +} \ + \ +static struct symbol_op name##_op = { \ + .evaluate = evaluate_pure_unop, \ + .expand = expand_##name, \ +} + +EXPAND_FINDBIT(clz); +EXPAND_FINDBIT(ctz); +EXPAND_FINDBIT(clrsb); +EXPAND_FINDBIT(ffs); +EXPAND_FINDBIT(parity); +EXPAND_FINDBIT(popcount); + static int evaluate_fp_unop(struct expression *expr) { struct expression *arg; @@ -334,11 +372,29 @@ static struct sym_init { { "__builtin_bswap16", &builtin_fn_type, MOD_TOPLEVEL, &bswap_op }, { "__builtin_bswap32", &builtin_fn_type, MOD_TOPLEVEL, &bswap_op }, { "__builtin_bswap64", &builtin_fn_type, MOD_TOPLEVEL, &bswap_op }, + { "__builtin_clrsb", &builtin_fn_type, MOD_TOPLEVEL, &clrsb_op }, + { "__builtin_clrsbl", &builtin_fn_type, MOD_TOPLEVEL, &clrsb_op }, + { "__builtin_clrsbll", &builtin_fn_type, MOD_TOPLEVEL, &clrsb_op }, + { "__builtin_clz", &builtin_fn_type, MOD_TOPLEVEL, &clz_op }, + { "__builtin_clzl", &builtin_fn_type, MOD_TOPLEVEL, &clz_op }, + { "__builtin_clzll", &builtin_fn_type, MOD_TOPLEVEL, &clz_op }, + { "__builtin_ctz", &builtin_fn_type, MOD_TOPLEVEL, &ctz_op }, + { "__builtin_ctzl", &builtin_fn_type, MOD_TOPLEVEL, &ctz_op }, + { "__builtin_ctzll", &builtin_fn_type, MOD_TOPLEVEL, &ctz_op }, + { "__builtin_ffs", &builtin_fn_type, MOD_TOPLEVEL, &ffs_op }, + { "__builtin_ffsl", &builtin_fn_type, MOD_TOPLEVEL, &ffs_op }, + { "__builtin_ffsll", &builtin_fn_type, MOD_TOPLEVEL, &ffs_op }, { "__builtin_isfinite", &builtin_fn_type, MOD_TOPLEVEL, &fp_unop_op }, { "__builtin_isinf", &builtin_fn_type, MOD_TOPLEVEL, &fp_unop_op }, { "__builtin_isinf_sign", &builtin_fn_type, MOD_TOPLEVEL, &fp_unop_op }, { "__builtin_isnan", &builtin_fn_type, MOD_TOPLEVEL, &fp_unop_op }, { "__builtin_isnormal", &builtin_fn_type, MOD_TOPLEVEL, &fp_unop_op }, + { "__builtin_parity", &builtin_fn_type, MOD_TOPLEVEL, &parity_op }, + { "__builtin_parityl", &builtin_fn_type, MOD_TOPLEVEL, &parity_op }, + { "__builtin_parityll", &builtin_fn_type, MOD_TOPLEVEL, &parity_op }, + { "__builtin_popcount", &builtin_fn_type, MOD_TOPLEVEL, &popcount_op }, + { "__builtin_popcountl", &builtin_fn_type, MOD_TOPLEVEL, &popcount_op }, + { "__builtin_popcountll", &builtin_fn_type, MOD_TOPLEVEL, &popcount_op }, { "__builtin_signbit", &builtin_fn_type, MOD_TOPLEVEL, &fp_unop_op }, { "__builtin_add_overflow", &builtin_fn_type, MOD_TOPLEVEL, &overflow_op }, { "__builtin_sub_overflow", &builtin_fn_type, MOD_TOPLEVEL, &overflow_op },
GCC expands at compile time builtins like __builtin_ffs() when their argument is constant. A driver in the kernel uses such a builtin in a case statement, causing sparse to report: error: Expected constant expression in case statement So, let sparse also expand such builtins, somehow like it was done for bswap16/32/64 but now for ffs/ffsl/ffsll, clz, ctz, clrsb, popcount & parity. Reported-by: Randy Dunlap <rdunlap@infradead.org> Signed-off-by: Luc Van Oostenryck <luc.vanoostenryck@gmail.com> --- builtin.c | 56 +++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 56 insertions(+)