[PATCH] [X86, AVX] adjust tablegen patterns to generate better code for scalar insertion into zero vector (PR23073)

[Sanjay Patel]

================
Comment at: test/CodeGen/X86/vector-shuffle-256-v4.ll:846
@@ -845,3 +845,3 @@
 ; ALL:       # BB#0:
-; ALL-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
+; ALL:         vxorpd %xmm1, %xmm1, %xmm1
 ; ALL-NEXT:    vmovsd {{.*#+}} xmm0 = xmm0[0],xmm1[1]
----------------
craig.topper wrote:
> Why is this not ALL-NEXT anymore?
Hi Craig -

We now print a 'kill' ahead of the xor. I didn't see any other tests matching on that in this file, so I loosened the check. But I see that other regression test files do have that kind of check, so I'll add another check line:

   _insert_reg_and_zero_v4f64:             ## @insert_reg_and_zero_v4f64
   ## BB#0:
                                        ## kill: XMM0<def> XMM0<kill> YMM0<def>
   vxorpd	%xmm1, %xmm1, %xmm1
   vmovsd	%xmm0, %xmm1, %xmm0     ## xmm0 = xmm0[0],xmm1[1]
   retq

http://reviews.llvm.org/D8794

EMAIL PREFERENCES
  http://reviews.llvm.org/settings/panel/emailpreferences/