[X86][SSE] Match bitcasted v4i32 BUILD_VECTORS for v2i64 shifts on 64-bit targets...

We were already doing this for 32-bit targets, but we can generate these on 64-bits as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@315155 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index e2f5807aa55b7542cd3fd21c15f87d600dea353e..6e45621bd960939452e3a979528a6be924c24c28 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -22279,9 +22279,8 @@ static SDValue LowerScalarVariableShift(SDValue Op, SelectionDAG &DAG,
     }
   }
 
-  // Special case in 32-bit mode, where i64 is expanded into high and low parts.
-  if (!Subtarget.is64Bit() && VT == MVT::v2i64  &&
-      Amt.getOpcode() == ISD::BITCAST &&
+  // Check cases (mainly 32-bit) where i64 is expanded into high and low parts.
+  if (VT == MVT::v2i64  && Amt.getOpcode() == ISD::BITCAST &&
       Amt.getOperand(0).getOpcode() == ISD::BUILD_VECTOR) {
     Amt = Amt.getOperand(0);
     unsigned Ratio = Amt.getSimpleValueType().getVectorNumElements() /

diff --git a/test/CodeGen/X86/pr34855.ll b/test/CodeGen/X86/pr34855.ll
new file mode 100644 (file)
index 0000000..286d7d2
--- /dev/null
+++ b/test/CodeGen/X86/pr34855.ll
@@ -0,0 +1,38 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i686-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X86
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X64
+
+define void @PR34855(<2 x i32> *%p0, <2 x i32> *%p1, <2 x i32> *%p2) {
+; X86-LABEL: PR34855:
+; X86:       # BB#0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
+; X86-NEXT:    movlps %xmm0, (%eax)
+; X86-NEXT:    retl
+;
+; X64-LABEL: PR34855:
+; X64:       # BB#0:
+; X64-NEXT:    movslq 4(%rdi), %rax
+; X64-NEXT:    movq %rax, %xmm0
+; X64-NEXT:    movslq (%rdi), %rax
+; X64-NEXT:    movq %rax, %xmm1
+; X64-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
+; X64-NEXT:    pxor %xmm0, %xmm0
+; X64-NEXT:    movdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
+; X64-NEXT:    psrlq %xmm0, %xmm2
+; X64-NEXT:    psrlq %xmm0, %xmm1
+; X64-NEXT:    pxor %xmm2, %xmm1
+; X64-NEXT:    psubq %xmm2, %xmm1
+; X64-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
+; X64-NEXT:    movq %xmm0, (%rdx)
+; X64-NEXT:    retq
+  %tmp = load <2 x i32>, <2 x i32>* %p0, align 8
+  %tmp1 = load <2 x i32>, <2 x i32>* %p1, align 8
+  %mul = mul <2 x i32> zeroinitializer, %tmp1
+  %mul1 = mul <2 x i32> <i32 -8190, i32 -8190>, %mul
+  %mul2 = mul <2 x i32> <i32 3, i32 3>, %mul1
+  %shr = ashr <2 x i32> %tmp, %mul2
+  store <2 x i32> %shr, <2 x i32>* %p2, align 8
+  ret void
+}