Here is a new version. gcc/ChangeLog 2012-05-08 Marc Glisse <marc.gli...@inria.fr>
* simplify-rtx.c (simplify_binary_operation_1): Optimize shuffle of concatenations. gcc/testsuite/ChangeLog 2012-05-08 Marc Glisse <marc.gli...@inria.fr> * gcc.target/i386/shuf-concat.c: New test. On Tue, 8 May 2012, Richard Sandiford wrote:
Very minor, but this code probably belongs in the else part of the if (!VECTOR_MODE_P (mode)) block.
I moved it in the block. Note that the piece of code right below, that starts with:
if (XVECLEN (trueop1, 0) == 1 && CONST_INT_P (XVECEXP (trueop1, 0, 0)) && GET_CODE (trueop0) == VEC_CONCAT)could probably move too. I had put mine right below because they do similar things. By the way, reusing that piece of code and applying it to each of the 2 selected parts of the vector would be one way to generalize my patch so it also applies to the vpermilpd case.
Note to self: if you want to grep for "shuf" in the asm, don't put "shuf" in the name of the file...
-- Marc Glisse
Index: testsuite/gcc.target/i386/shuf-concat.c =================================================================== --- testsuite/gcc.target/i386/shuf-concat.c (revision 0) +++ testsuite/gcc.target/i386/shuf-concat.c (revision 0) @@ -0,0 +1,13 @@ +/* { dg-do compile } */ +/* { dg-options "-O" } */ + +typedef double v2df __attribute__ ((__vector_size__ (16))); + +v2df f(double d,double e){ + v2df x={-d,d}; + v2df y={-e,e}; + return __builtin_ia32_shufpd(x,y,1); +} + +/* { dg-final { scan-assembler-not "shufpd" } } */ +/* { dg-final { scan-assembler-times "unpck" 1 } } */ Property changes on: testsuite/gcc.target/i386/shuf-concat.c ___________________________________________________________________ Added: svn:eol-style + native Added: svn:keywords + Author Date Id Revision URL Index: simplify-rtx.c =================================================================== --- simplify-rtx.c (revision 187276) +++ simplify-rtx.c (working copy) @@ -1,10 +1,10 @@ /* RTL simplification functions for GNU compiler. Copyright (C) 1987, 1988, 1989, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, - 2011 Free Software Foundation, Inc. + 2011, 2012 Free Software Foundation, Inc. This file is part of GCC. GCC is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 3, or (at your option) any later @@ -3239,12 +3239,33 @@ simplify_binary_operation_1 (enum rtx_co RTVEC_ELT (v, i) = CONST_VECTOR_ELT (trueop0, INTVAL (x)); } return gen_rtx_CONST_VECTOR (mode, v); } + + /* If we build {a,b} then permute it, build the result directly. */ + if (XVECLEN (trueop1, 0) == 2 + && CONST_INT_P (XVECEXP (trueop1, 0, 0)) + && CONST_INT_P (XVECEXP (trueop1, 0, 1)) + && GET_CODE (trueop0) == VEC_CONCAT + && GET_CODE (XEXP (trueop0, 0)) == VEC_CONCAT + && GET_MODE (XEXP (trueop0, 0)) == mode + && GET_CODE (XEXP (trueop0, 1)) == VEC_CONCAT + && GET_MODE (XEXP (trueop0, 1)) == mode) + { + unsigned int i0 = INTVAL (XVECEXP (trueop1, 0, 0)); + unsigned int i1 = INTVAL (XVECEXP (trueop1, 0, 1)); + rtx subop0, subop1; + + gcc_assert (i0 < 4 && i1 < 4); + subop0 = XEXP (XEXP (trueop0, i0 / 2), i0 % 2); + subop1 = XEXP (XEXP (trueop0, i1 / 2), i1 % 2); + + return simplify_gen_binary (VEC_CONCAT, mode, subop0, subop1); + } } if (XVECLEN (trueop1, 0) == 1 && CONST_INT_P (XVECEXP (trueop1, 0, 0)) && GET_CODE (trueop0) == VEC_CONCAT) {