These two functions should produce the same code: int f(int y, int x) { return x & ((~x) | y); } int g(int y, int x) { return y & x; } x & ((~x) | y) is the same as (x & ~x) | (x & y) using distribution and that becomes (x & y) because (x & ~x) is always 0 and 0 | x is x. This should be done at both the tree level and the RTL level.
Mine for RTL level: Index: simplify-rtx.c =================================================================== --- simplify-rtx.c (revision 2035) +++ simplify-rtx.c (revision 2036) @@ -1885,6 +1885,18 @@ simplify_binary_operation_1 (enum rtx_co return simplify_gen_unary (NOT, mode, tem, mode); } + /* (and X (ior (not X) Y) -> (and X Y) */ + if (GET_CODE (op1) == IOR + && GET_CODE (XEXP (op1, 0)) == NOT + && op0 == XEXP (XEXP (op1, 0), 0)) + return simplify_gen_binary (AND, mode, op0, XEXP (op1, 1)); + + /* (and (ior (not X) Y) X) -> (and X Y) */ + if (GET_CODE (op0) == IOR + && GET_CODE (XEXP (op0, 0)) == NOT + && op1 == XEXP (XEXP (op0, 0), 0)) + return simplify_gen_binary (AND, mode, op1, XEXP (op0, 1)); + tem = simplify_associative_operation (code, mode, op0, op1); if (tem) return tem;
Subject: Bug 33512 Author: pinskia Date: Sat Feb 23 17:58:48 2008 New Revision: 132575 URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=132575 Log: 2008-02-23 Andrew Pinski <andrew_pinski@playstation.sony.com> PR rtl-opt/33512 * simplify-rtx.c (simplify_binary_operation_1): Add simplification of (and X (ior (not X) Y) and (and (ior (not X) Y) X). 2008-02-23 Andrew Pinski <andrew_pinski@playstation.sony.com> PR rtl-opt/33512 * gcc.dg/and-1.c: New test. Added: trunk/gcc/testsuite/gcc.dg/and-1.c Modified: trunk/gcc/ChangeLog trunk/gcc/simplify-rtx.c trunk/gcc/testsuite/ChangeLog
The RTL level has been fixed. The tree level needs fixing still but I am not working on that.
Test gcc.dg/and-1.c, added as a fix for this PR, fails on powerpc64-linux. The code generated for "-m64 -O2" is: f: mr 0,3 neg 3,3 nand 3,3,0 and 3,3,0 blr
(In reply to comment #4) > Test gcc.dg/and-1.c, added as a fix for this PR, fails on powerpc64-linux. The > code generated for "-m64 -O2" is: The issue is we see (and X (ior Y (not X) ) ), I will add this extra check later this weekend. In GCC 4.1.1, we saw what I had expected but for some reasons GCC 4.3.0 has a different order of arguments.
Mine still.
*** Bug 38218 has been marked as a duplicate of this bug. ***
So what asm do we expect that we should get form the and-1.c testcase?
I will be submitting a patch for 4.8.0 to fix this on the tree level in fold-const.c.
Author: pinskia Date: Tue Apr 24 07:05:09 2012 New Revision: 186749 URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=186749 Log: 2012-04-24 Andrew Pinski <apinski@cavium.com> PR tree-opt/33512 * tree-ssa-forwprop.c (defcodefor_name): New function. (simplify_bitwise_binary): Use defcodefor_name instead of manually Simplify "( X | Y) & X" to X and "( X & Y) | X" to X. Simplify "(~X | Y) & X" to "X & Y" and "(~X & Y) | X" to "X | Y". 2012-04-24 Andrew Pinski <apinski@cavium.com> PR tree-opt/33512 * gcc.dg/tree-ssa/andor-3.c: New testcase. * gcc.dg/tree-ssa/andor-4.c: New testcase. * gcc.dg/tree-ssa/andor-5.c: New testcase. Added: trunk/gcc/testsuite/gcc.dg/tree-ssa/andor-3.c trunk/gcc/testsuite/gcc.dg/tree-ssa/andor-4.c trunk/gcc/testsuite/gcc.dg/tree-ssa/andor-5.c Modified: trunk/gcc/ChangeLog trunk/gcc/testsuite/ChangeLog trunk/gcc/tree-ssa-forwprop.c
Fixed.
Cause bootstrap failure ./../gcc/gcc/tree-ssa-ifcombine.c -o tree-ssa-ifcombine.o ../../gcc/gcc/tree-ssa-forwprop.c: In function `simplify_bitwise_binary': ../../gcc/gcc/tree-ssa-forwprop.c:1916: error: `def1' undeclared (first use in this function) ../../gcc/gcc/tree-ssa-forwprop.c:1916: error: (Each undeclared identifier is reported only once ../../gcc/gcc/tree-ssa-forwprop.c:1916: error: for each function it appears in.) ../../gcc/gcc/tree-ssa-forwprop.c:1917: error: `def2' undeclared (first use in this function)
Patch which I am testing: Index: tree-ssa-forwprop.c =================================================================== --- tree-ssa-forwprop.c (revision 186755) +++ tree-ssa-forwprop.c (working copy) @@ -1913,10 +1913,10 @@ simplify_bitwise_binary (gimple_stmt_ite /* Simplify (A & B) OP0 (C & B) to (A OP0 C) & B. */ if (def1_code == def2_code && def1_code == BIT_AND_EXPR - && operand_equal_for_phi_arg_p (gimple_assign_rhs2 (def1), - gimple_assign_rhs2 (def2))) + && operand_equal_for_phi_arg_p (def1_arg2, + def2_arg2)) { - tree b = gimple_assign_rhs2 (def1); + tree b = def1_arg2; tree a = def1_arg1; tree c = def2_arg1; tree inner = fold_build2 (code, TREE_TYPE (arg2), a, c); --- CUT --- This was definitely my fault. I tested both patches independently and forgot to test them together. Sorry about that.
Hmm, I have right now in my tree Index: tree-ssa-forwprop.c =================================================================== --- tree-ssa-forwprop.c (revision 186753) +++ tree-ssa-forwprop.c (working copy) @@ -1912,17 +1912,25 @@ /* Simplify (A & B) OP0 (C & B) to (A OP0 C) & B. */ if (def1_code == def2_code - && def1_code == BIT_AND_EXPR - && operand_equal_for_phi_arg_p (gimple_assign_rhs2 (def1), - gimple_assign_rhs2 (def2))) + && def1_code == BIT_AND_EXPR) { - tree b = gimple_assign_rhs2 (def1); - tree a = def1_arg1; - tree c = def2_arg1; - tree inner = fold_build2 (code, TREE_TYPE (arg2), a, c); + tree inner, a = NULL_TREE, b = NULL_TREE, c = NULL_TREE; + + if (operand_equal_for_phi_arg_p (def1_arg2, def2_arg2)) + { b = def1_arg2; a = def1_arg1; c = def2_arg1; } + else if (operand_equal_for_phi_arg_p (def1_arg1, def2_arg1)) + { b = def1_arg1; a = def1_arg2; c = def2_arg2; } + else if (operand_equal_for_phi_arg_p (def1_arg1, def2_arg2)) + { b = def1_arg1; a = def1_arg2; c = def2_arg1; } + else if (operand_equal_for_phi_arg_p (def1_arg2, def2_arg1)) + { b = def1_arg2; a = def1_arg1; c = def2_arg2; } + if (a && c) + inner = fold_build2 (code, TREE_TYPE (arg2), a, c); + if (!a || !b || !c) + ; /* If A OP0 C (this usually means C is the same as A) is 0 then fold it down correctly. */ - if (integer_zerop (inner)) + else if (integer_zerop (inner)) { gimple_assign_set_rhs_from_tree (gsi, inner); update_stmt (stmt); It has the advantage of handling also cases for (A & B) OP0 (C & B) to (A OP0 C) & B, (B & A) OP0 (C & B) to (A OP0 C) & B, (A & B) OP0 (B & C) to (A OP0 C) & B, and (B & A) OP0 (B & C) to (A OP0 C) & B. It bootstraps fine.
>Hmm, I have right now in my tree The most common case is with b being a constant which is why I only looked into that case.
Fixed the bootstrap issue now so closing as fixed.