Eric, this patch is the second half of the fix for PR61605.
It make sure in pass_cprop_hardreg that, if available we use the call-specific information provided by fuse-caller-save, rather than the generic regs_invalidated_by_call info.
The 2 patches together allow an insn to removed from the gcc.target/i386/fuse-caller-save.c testcase, which is updated accordingly in this patch.
Bootstrapped and reg-tested on x86_64. OK for trunk? Thanks, - Tom
2014-10-13 Tom de Vries <t...@codesourcery.com> PR rtl-optimization/61605 * regcprop.c (copyprop_hardreg_forward_1): Use regs_invalidated_by_this_call instead of regs_invalidated_by_call. * gcc.target/i386/fuse-caller-save.c: Update addition check. Add movl absence check. diff --git a/gcc/regcprop.c b/gcc/regcprop.c index c71de98..304820f 100644 --- a/gcc/regcprop.c +++ b/gcc/regcprop.c @@ -1000,6 +1000,7 @@ copyprop_hardreg_forward_1 (basic_block bb, struct value_data *vd) unsigned int set_nregs = 0; unsigned int regno; rtx exp; + HARD_REG_SET regs_invalidated_by_this_call; for (exp = CALL_INSN_FUNCTION_USAGE (insn); exp; exp = XEXP (exp, 1)) { @@ -1018,8 +1019,11 @@ copyprop_hardreg_forward_1 (basic_block bb, struct value_data *vd) } } + get_call_reg_set_usage (insn, + ®s_invalidated_by_this_call, + regs_invalidated_by_call); for (regno = 0; regno < FIRST_PSEUDO_REGISTER; regno++) - if ((TEST_HARD_REG_BIT (regs_invalidated_by_call, regno) + if ((TEST_HARD_REG_BIT (regs_invalidated_by_this_call, regno) || HARD_REGNO_CALL_PART_CLOBBERED (regno, vd->e[regno].mode)) && (regno < set_regno || regno >= set_regno + set_nregs)) kill_value_regno (regno, 1, vd); diff --git a/gcc/testsuite/gcc.target/i386/fuse-caller-save.c b/gcc/testsuite/gcc.target/i386/fuse-caller-save.c index 7e2b11d..a6e8f1c 100644 --- a/gcc/testsuite/gcc.target/i386/fuse-caller-save.c +++ b/gcc/testsuite/gcc.target/i386/fuse-caller-save.c @@ -20,5 +20,13 @@ foo (int y) /* { dg-final { scan-assembler-not "push" } } */ /* { dg-final { scan-assembler-not "pop" } } */ -/* Check that addition uses dx. */ -/* { dg-final { scan-assembler-times "addl\t%\[re\]?dx, %\[re\]?ax" 1 } } */ +/* PR61605. If the first argument register and the return register differ, then + bar leaves the first argument register intact. That means in foo that the + first argument register still contains y after bar has been called, and + there's no need to copy y to a different register before the call, to be able + to use it after the call. + Check that the copy is absent. */ +/* { dg-final { scan-assembler-not "movl" { target { ! ia32 } } } } */ + +/* Check that addition uses di (in case of no copy) or dx (in case of copy). */ +/* { dg-final { scan-assembler-times "addl\t%\[re\]?d\[ix\], %\[re\]?ax" 1 } } */ -- 1.9.1