vpblend intrinsics combines as shifts intrinsics due to absence return stmt between them

[rob.khasanov]

Summary

Fix PR20088

[andreadb]

Hi Robert,

Please address the two minor comments below related to the test.
Otherwise, the patch looks good to me.

Thanks!

test/CodeGen/X86/pr20088.ll

3–4 ↗

(On Diff #10796)

Remove the comment at line 3.
Remove the reference to attribute group #0.

6–7 ↗

(On Diff #10796)

Remove the comment at line 6.
Remove the reference to attribute group #1 since it is not defined.

[rob.khasanov]

Hi Andrea,

Thanks for review, patch is updated

[andreadb]

LGTM. Thanks :)

[rob.khasanov]

Closed by commit rL211617 (authored by rkhasanov).

llvm/trunk/lib/Target/X86/X86ISelLowering.cpp

case Intrinsic::x86_sse41_pblendvb: {
     
     // Simplify the case where the mask is a constant i32 value.
     if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Mask)) {
       if (C->isNullValue())
         return Op0;
       if (C->isAllOnesValue())
         return Op1;
     }
+
+    return SDValue();
   }
 
   // Packed SSE2/AVX2 arithmetic shift immediate intrinsics.
   case Intrinsic::x86_sse2_psrai_w:
   case Intrinsic::x86_sse2_psrai_d:
   case Intrinsic::x86_avx2_psrai_w:
   case Intrinsic::x86_avx2_psrai_d:
   case Intrinsic::x86_sse2_psra_w:

llvm/trunk/test/CodeGen/X86/pr20088.ll

+; RUN: llc < %s -mattr=+avx | FileCheck %s
+
+declare <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8>, <16 x i8>, <16 x i8>)
+
+define <16 x i8> @foo(<16 x i8> %x) {
+; CHECK: vpblendvb
+  %res = call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> zeroinitializer, <16 x i8> <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>, <16 x i8> %x)
+  ret <16 x i8> %res;
+}