x86_64: Use peephole2 to eliminate redundant moves.

The recent fix for mul_widen_cost revealed an interesting
quirk of ira/reload register allocation on x86_64.  As shown in
https://gcc.gnu.org/pipermail/gcc-patches/2020-August/551648.html
for gcc.target/i386/pr71321.c we generate the following code that
performs unnecessary register shuffling.

        movl    $-51, %edx
        movl    %edx, %eax
        mulb    %dil

Various discussions in bugzilla seem to point to reload preferring
not to load constants directly into CLASS_LIKELY_SPILLED_P registers.
Whatever the cause, one solution (workaround), that doesn't involve
rewriting a register allocator, is to use peephole2 to spot this
wierdness and eliminate it.  With this peephole2 the above three
instructions (from pr71321.c) are replaced with:

        movl    $-51, %eax
        mulb    %dil

2020-08-12  Roger Sayle  <roger@nextmovesoftware.com>
	    Uroš Bizjak  <ubizjak@gmail.com>

gcc/ChangeLog
	* config/i386/i386.md (peephole2): Reduce unnecessary
	register shuffling produced by register allocation.

diff --git a/gcc/config/i386/i386.md b/gcc/config/i386/i386.md
index 4e916bf3c32990b5645e6dad7fb03a18dab740b8..f3799ac3e6f91db3c1ff8776761616fca06d05a7 100644 (file)
--- a/gcc/config/i386/i386.md
+++ b/gcc/config/i386/i386.md
@@ -18946,6 +18946,16 @@
   operands[2] = gen_rtx_REG (GET_MODE (operands[0]), FLAGS_REG);
   ix86_expand_clear (operands[1]);
 })
+
+;; Reload dislikes loading constants directly into class_likely_spilled
+;; hard registers.  Try to tidy things up here.
+(define_peephole2
+  [(set (match_operand:SWI 0 "general_reg_operand")
+       (match_operand:SWI 1 "x86_64_general_operand"))
+   (set (match_operand:SWI 2 "general_reg_operand")
+       (match_dup 0))]
+  "peep2_reg_dead_p (2, operands[0])"
+  [(set (match_dup 2) (match_dup 1))])
 \f
 ;; Misc patterns (?)