[PATCH] D113661: [X86][FP16] Avoid to generate VZEXT_MOVL with i16

[Phoebe Wang via Phabricator via llvm-commits]

pengfei updated this revision to Diff 386472.
pengfei added a comment.

Swapped the condition order.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D113661/new/

https://reviews.llvm.org/D113661

Files:
  llvm/lib/Target/X86/X86ISelLowering.cpp
  llvm/test/CodeGen/X86/avx512fp16-mov.ll


Index: llvm/test/CodeGen/X86/avx512fp16-mov.ll
===================================================================
--- llvm/test/CodeGen/X86/avx512fp16-mov.ll
+++ llvm/test/CodeGen/X86/avx512fp16-mov.ll
@@ -1926,3 +1926,32 @@
   store <4 x half> %c, <4 x half>* %z
   ret void
 }
+
+define <8 x half> @test21(half %a, half %b, half %c) nounwind {
+; X64-LABEL: test21:
+; X64:       # %bb.0:
+; X64-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X64-NEXT:    vmovsh %xmm2, %xmm3, %xmm2
+; X64-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; X64-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
+; X64-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; X64-NEXT:    vpbroadcastw %xmm1, %xmm1
+; X64-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,0]
+; X64-NEXT:    retq
+;
+; X86-LABEL: test21:
+; X86:       # %bb.0:
+; X86-NEXT:    vmovsh {{[0-9]+}}(%esp), %xmm0
+; X86-NEXT:    vmovsh {{[0-9]+}}(%esp), %xmm1
+; X86-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; X86-NEXT:    vmovsh {{[0-9]+}}(%esp), %xmm1
+; X86-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; X86-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; X86-NEXT:    vpbroadcastw %xmm1, %xmm1
+; X86-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,0]
+; X86-NEXT:    retl
+  %1 = insertelement <8 x half> <half poison, half poison, half poison, half 0xH0000, half 0xH0000, half 0xH0000, half 0xH0000, half 0xH0000>, half %a, i32 0
+  %2 = insertelement <8 x half> %1, half %b, i32 1
+  %3 = insertelement <8 x half> %2, half %c, i32 2
+  ret <8 x half> %3
+}
Index: llvm/lib/Target/X86/X86ISelLowering.cpp
===================================================================
--- llvm/lib/Target/X86/X86ISelLowering.cpp
+++ llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -36129,8 +36129,9 @@
         (V1.getOpcode() == ISD::SCALAR_TO_VECTOR &&
          isUndefOrZeroInRange(Mask, 1, NumMaskElts - 1))) {
       Shuffle = X86ISD::VZEXT_MOVL;
-      SrcVT = DstVT =
-          !Subtarget.hasSSE2() && MaskEltSize == 32 ? MVT::v4f32 : MaskVT;
+      SrcVT = DstVT = MaskEltSize == 16      ? MVT::v8f16
+                      : !Subtarget.hasSSE2() ? MVT::v4f32
+                                             : MaskVT;
       return true;
     }
   }


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D113661.386472.patch
Type: text/x-patch
Size: 2331 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20211111/e5a64583/attachment.bin>