Hello! This patch fixes asymmetry for TARGET_SSE4_1, where pinsr was used to create vector copy, but not pextr when creating scalar copy with -mstv.
2017-02-03 Uros Bizjak <ubiz...@gmail.com> * config/i386/i386.c (dimode_scalar_chain::convert_reg): Use pextrd for TARGET_SSE4_1 when creating scalar copy. Bootstrapped and regression tested on x86_64-linux-gnu {,-m32}, configured with --with-arch=core-avx-i. Committed to mainline SVN. Uros.
diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c index 01a05c6..3a65945 100644 --- a/gcc/config/i386/i386.c +++ b/gcc/config/i386/i386.c @@ -3622,11 +3622,26 @@ dimode_scalar_chain::convert_reg (unsigned regno) if (scalar_copy) { - rtx vcopy = gen_reg_rtx (V2DImode); - start_sequence (); - if (TARGET_INTER_UNIT_MOVES_FROM_VEC) + if (TARGET_SSE4_1) + { + rtx tmp = gen_rtx_PARALLEL (VOIDmode, gen_rtvec (1, const0_rtx)); + emit_insn + (gen_rtx_SET + (gen_rtx_SUBREG (SImode, scopy, 0), + gen_rtx_VEC_SELECT (SImode, + gen_rtx_SUBREG (V4SImode, reg, 0), tmp))); + + tmp = gen_rtx_PARALLEL (VOIDmode, gen_rtvec (1, const1_rtx)); + emit_insn + (gen_rtx_SET + (gen_rtx_SUBREG (SImode, scopy, 4), + gen_rtx_VEC_SELECT (SImode, + gen_rtx_SUBREG (V4SImode, reg, 0), tmp))); + } + else if (TARGET_INTER_UNIT_MOVES_FROM_VEC) { + rtx vcopy = gen_reg_rtx (V2DImode); emit_move_insn (vcopy, gen_rtx_SUBREG (V2DImode, reg, 0)); emit_move_insn (gen_rtx_SUBREG (SImode, scopy, 0), gen_rtx_SUBREG (SImode, vcopy, 0));