Lower unsigned vsetcc to psubus in certain cases

The current approach to lower a vsetult is to flip the sign bit of the
operands, swap the operands and then use a (signed) pcmpgt.  psubus (unsigned
saturating subtract) can be used to emulate a vsetult more efficiently:

+    case ISD::SETULT: {
+      // If the comparison is against a constant we can turn this into a
+      // setule.  With psubus, setule does not require a swap.  This is
+      // beneficial because the constant in the register is no longer
+      // destructed as the destination so it can be hoisted out of a loop.

I also enable lowering via psubus in a few other cases where it's clearly
beneficial: setule and setuge if minu/maxu cannot be used.
    
rdar://problem/14338765

Patch by Adam Nemet <anemet@apple.com>.

llvm-svn: 202301

1 files changed, 6 insertions, 12 deletions

diff --git a/llvm/test/CodeGen/X86/vec_setcc.ll b/llvm/test/CodeGen/X86/vec_setcc.ll
index fc8a56de791..322dbae0c89 100644
--- a/llvm/test/CodeGen/X86/vec_setcc.ll
+++ b/llvm/test/CodeGen/X86/vec_setcc.ll
@@ -42,12 +42,9 @@ define <8 x i16> @v8i16_icmp_uge(<8 x i16> %a, <8 x i16> %b) nounwind readnone s
   %2 = sext <8 x i1> %1 to <8 x i16>
   ret <8 x i16> %2
 ; SSE2-LABEL: v8i16_icmp_uge:
-; SSE2: movdqa  {{.*}}(%rip), %xmm2
-; SEE2: pxor    %xmm2, %xmm0
-; SSE2: pxor    %xmm1, %xmm2
-; SSE2: pcmpgtw %xmm0, %xmm2
-; SSE2: pcmpeqd %xmm0, %xmm0
-; SSE2: pxor    %xmm2, %xmm0
+; SSE2: psubusw %xmm0, %xmm1
+; SEE2: pxor    %xmm0, %xmm0
+; SSE2: pcmpeqw %xmm1, %xmm0
 
 ; SSE41-LABEL: v8i16_icmp_uge:
 ; SSE41: pmaxuw  %xmm0, %xmm1
@@ -63,12 +60,9 @@ define <8 x i16> @v8i16_icmp_ule(<8 x i16> %a, <8 x i16> %b) nounwind readnone s
   %2 = sext <8 x i1> %1 to <8 x i16>
   ret <8 x i16> %2
 ; SSE2-LABEL: v8i16_icmp_ule:
-; SSE2: movdqa  {{.*}}(%rip), %xmm2
-; SSE2: pxor    %xmm2, %xmm1
-; SSE2: pxor    %xmm2, %xmm0
-; SSE2: pcmpgtw %xmm1, %xmm0
-; SSE2: pcmpeqd %xmm1, %xmm1
-; SSE2: pxor    %xmm0, %xmm1
+; SSE2: psubusw %xmm1, %xmm0
+; SSE2: pxor    %xmm1, %xmm1
+; SSE2: pcmpeqw %xmm0, %xmm1
 ; SSE2: movdqa  %xmm1, %xmm0
 
 ; SSE41-LABEL: v8i16_icmp_ule: