fix a test that has no checks

[spatel]

Summary

This test has no checks; it doesn't even pipe output to grep.

In an amazing display of architectural restraint, x86 doesn't have fneg instructions. :)
So check for the xor (and possibly sub) that are required to make fneg happen.

[chandlerc]

LGTM. Minor comment below.

test/CodeGen/X86/vec_fneg.ll

8

I suspect you're going to have trouble getting this to portably match. I would use a triple such as the bare metal 'x86_64-unknown-unknown' one to get a more stable pattern.

What we really should do is add comprehensive verbose asm comments that expand constant pool operands in the comment so that we can file check them in-line.

You can see some of how to do this with my changes to X86MCInstLower.cpp for comments on PSHUFB instructions if your interested... But it'll be somewhat complex. Clearly no need to do that for this patch.

[spatel]

Closed by commit rL214992 (authored by @spatel).

test/CodeGen/X86/vec_fneg.ll

8

Thanks for the suggestion. I'll just use a triple for now rather than -march/-mattr.

I'm all for improving the asm comments - a few weeks ago, I filed:
http://llvm.org/bugs/show_bug.cgi?id=20347

I'll see if I can make any progress there.

test/CodeGen/X86/vec_fneg.ll

-; RUN: llc < %s -march=x86 -mattr=+sse2
+; RUN: llc < %s -march=x86 -mattr=+sse2 | FileCheck %s
 
+; FNEG is defined as subtraction from -0.0.
+
+; This test verifies that we use an xor with a constant to flip the sign bits; no subtraction needed.
 define <4 x float> @t1(<4 x float> %Q) {
-        %tmp15 = fsub <4 x float> < float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00 >, %Q
-  ret <4 x float> %tmp15
+; CHECK-LABEL: t1:
+; CHECK: xorps  {{.*}}LCPI0_0{{.*}}, %xmm0
+; CHECK-NEXT: retl
+        %tmp = fsub <4 x float> < float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00 >, %Q
+  ret <4 x float> %tmp
 }
 
+; This test verifies that we generate an FP subtraction because "0.0 - x" is not an fneg.
 define <4 x float> @t2(<4 x float> %Q) {
-        %tmp15 = fsub <4 x float> zeroinitializer, %Q
-  ret <4 x float> %tmp15
+; CHECK-LABEL: t2:
+; CHECK: xorps  %[[X:xmm[0-9]+]], %[[X]]
+; CHECK-NEXT: subps %xmm0, %[[X]]
+; CHECK-NEXT: movaps  %[[X]], %xmm0
+; CHECK-NEXT: retl
+        %tmp = fsub <4 x float> zeroinitializer, %Q
+  ret <4 x float> %tmp
 }