On Mon, Jun 20, 2011 at 09:38:22PM +0200, Bernd Schmidt wrote: > On 06/16/2011 06:25 PM, Richard Henderson wrote: > > On 06/16/2011 05:44 AM, Bernd Schmidt wrote: > >> +@deftypefn {Built-in Function} int __builtin_clrsb (unsigned int x) > >> +Returns the number of leading redundant sign bits in @var{x}, starting > >> +at the most significant bit position. > >> +@end deftypefn > > > > Do we want a signed argument, since we're talking about signs? > > Err, yes. It's signed everywhere else (builtins.def etc.).
But unfortunately not during expansion. gcc.c-torture/execute/builtin-bitops-1.c fails on x86_64-linux, because there is no SImode libcall, only DImode, and it was zero-extending instead of sign-extending the argument, and not subtracting the mode difference afterwards. So int foo (int x) { return __builtin_clrsb (x); } expanded the same as int foo (int x) { return __builtin_clrsbl ((unsigned) x); } instead of the correct int foo (int x) { return __builtin_clrsbl (x) - 32; } Fixed thusly, I've also added folding of constant arguments during tree optimizations for it. Ok for trunk? 2011-06-21 Jakub Jelinek <ja...@redhat.com> PR middle-end/49489 * builtins.c (expand_builtin_unop): Call expand_unop with 0 as unsignedp argument instead of 1 for clrsb_optab. (fold_builtin_bitop): Fix masking for width > HOST_BITS_PER_WIDE_INT and < 2 * HOST_BITS_PER_WIDE_INT. Optimize BUILT_IN_CLRSB*. (fold_builtin_1): Call fold_builtin_binop for BUILT_IN_CLRSB*. * optabs.c (widen_leading): Call widen_operand and expand_unop with 0 as unsignedp argument instead of 1 for clrsb_optab. (expand_unop): Subtract difference of mode sizes also for clrsb_optab. --- gcc/builtins.c.jj 2011-06-21 16:46:01.000000000 +0200 +++ gcc/builtins.c 2011-06-21 17:36:38.000000000 +0200 @@ -4578,7 +4578,7 @@ expand_builtin_unop (enum machine_mode t /* Compute op, into TARGET if possible. Set TARGET to wherever the result comes back. */ target = expand_unop (TYPE_MODE (TREE_TYPE (CALL_EXPR_ARG (exp, 0))), - op_optab, op0, target, 1); + op_optab, op0, target, op_optab != clrsb_optab); gcc_assert (target); return convert_to_mode (target_mode, target, 0); @@ -7265,7 +7265,8 @@ fold_builtin_bitop (tree fndecl, tree ar { hi = TREE_INT_CST_HIGH (arg); if (width < 2 * HOST_BITS_PER_WIDE_INT) - hi &= ~((HOST_WIDE_INT) (-1) >> (width - HOST_BITS_PER_WIDE_INT)); + hi &= ~((unsigned HOST_WIDE_INT) (-1) + << (width - HOST_BITS_PER_WIDE_INT)); } else { @@ -7303,6 +7304,26 @@ fold_builtin_bitop (tree fndecl, tree ar result = width; break; + CASE_INT_FN (BUILT_IN_CLRSB): + if (width > HOST_BITS_PER_WIDE_INT + && (hi & ((unsigned HOST_WIDE_INT) 1 + << (width - HOST_BITS_PER_WIDE_INT - 1))) != 0) + { + hi = ~hi & ~((unsigned HOST_WIDE_INT) (-1) + << (width - HOST_BITS_PER_WIDE_INT - 1)); + lo = ~lo; + } + else if (width <= HOST_BITS_PER_WIDE_INT + && (lo & ((unsigned HOST_WIDE_INT) 1 << (width - 1))) != 0) + lo = ~lo & ~((unsigned HOST_WIDE_INT) (-1) << (width - 1)); + if (hi != 0) + result = width - floor_log2 (hi) - 2 - HOST_BITS_PER_WIDE_INT; + else if (lo != 0) + result = width - floor_log2 (lo) - 2; + else + result = width - 1; + break; + CASE_INT_FN (BUILT_IN_POPCOUNT): result = 0; while (lo) @@ -9737,6 +9758,7 @@ fold_builtin_1 (location_t loc, tree fnd CASE_INT_FN (BUILT_IN_FFS): CASE_INT_FN (BUILT_IN_CLZ): CASE_INT_FN (BUILT_IN_CTZ): + CASE_INT_FN (BUILT_IN_CLRSB): CASE_INT_FN (BUILT_IN_POPCOUNT): CASE_INT_FN (BUILT_IN_PARITY): return fold_builtin_bitop (fndecl, arg0); --- gcc/optabs.c.jj 2011-06-21 16:46:01.000000000 +0200 +++ gcc/optabs.c 2011-06-21 17:50:35.000000000 +0200 @@ -2350,8 +2350,10 @@ widen_leading (enum machine_mode mode, r if (target == 0) target = gen_reg_rtx (mode); - xop0 = widen_operand (op0, wider_mode, mode, true, false); - temp = expand_unop (wider_mode, unoptab, xop0, NULL_RTX, true); + xop0 = widen_operand (op0, wider_mode, mode, + unoptab != clrsb_optab, false); + temp = expand_unop (wider_mode, unoptab, xop0, NULL_RTX, + unoptab != clrsb_optab); if (temp != 0) temp = expand_binop (wider_mode, sub_optab, temp, GEN_INT (GET_MODE_BITSIZE (wider_mode) @@ -3075,8 +3077,9 @@ expand_unop (enum machine_mode mode, opt unsignedp); /* If we are generating clz using wider mode, adjust the - result. */ - if (unoptab == clz_optab && temp != 0) + result. Similarly for clrsb. */ + if ((unoptab == clz_optab || unoptab == clrsb_optab) + && temp != 0) temp = expand_binop (wider_mode, sub_optab, temp, GEN_INT (GET_MODE_BITSIZE (wider_mode) - GET_MODE_BITSIZE (mode)), Jakub