From: Craig Topper <craig.topper@sifive.com>
Date: Tue, 15 Jun 2021 05:01:45 +0000 (-0700)
Subject: [X86] Use EVT::getVectorVT instead of changeVectorElementType in reduceVMULWidth.
X-Git-Tag: llvmorg-14-init~3967
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=4017d0335a35334835bfae6fc3e258adcd9ed2dc;p=platform%2Fupstream%2Fllvm.git

[X86] Use EVT::getVectorVT instead of changeVectorElementType in reduceVMULWidth.

Changing vector element type doesn't work for v6i32->v6i16 now
that v6i32 is an MVT and v6i16 is not.

I would like to fix this in changeVectorElementType, but you
need a LLVMContext to call getVectorVT which we can't get from
an MVT.

Fixes PR50709.
---

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 90babf3..6b7aea4 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -42902,7 +42902,7 @@ static SDValue reduceVMULWidth(SDNode *N, SelectionDAG &DAG,
   if ((NumElts % 2) != 0)
     return SDValue();
 
-  EVT ReducedVT = VT.changeVectorElementType(MVT::i16);
+  EVT ReducedVT = EVT::getVectorVT(*DAG.getContext(), MVT::i16, NumElts);
 
   // Shrink the operands of mul.
   SDValue NewN0 = DAG.getNode(ISD::TRUNCATE, DL, ReducedVT, N0);
diff --git a/llvm/test/CodeGen/X86/pr50709.ll b/llvm/test/CodeGen/X86/pr50709.ll
new file mode 100644
index 0000000..cb1b527
--- /dev/null
+++ b/llvm/test/CodeGen/X86/pr50709.ll
@@ -0,0 +1,27 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu | FileCheck %s
+
+define <6 x i32> @foo(<6 x i16> %x, <6 x i16> %y) {
+; CHECK-LABEL: foo:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    movq %rdi, %rax
+; CHECK-NEXT:    movdqa %xmm0, %xmm2
+; CHECK-NEXT:    pmulhuw %xmm1, %xmm2
+; CHECK-NEXT:    pmullw %xmm1, %xmm0
+; CHECK-NEXT:    movdqa %xmm0, %xmm1
+; CHECK-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
+; CHECK-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,xmm2[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; CHECK-NEXT:    pslldq {{.*#+}} xmm0 = zero,zero,xmm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; CHECK-NEXT:    punpckhwd {{.*#+}} xmm0 = xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
+; CHECK-NEXT:    movdqa %xmm1, %xmm2
+; CHECK-NEXT:    movsd {{.*#+}} xmm2 = xmm0[0],xmm2[1]
+; CHECK-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,0]
+; CHECK-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,2,2,3]
+; CHECK-NEXT:    movq %xmm0, 16(%rdi)
+; CHECK-NEXT:    movaps %xmm1, (%rdi)
+; CHECK-NEXT:    retq
+  %a = zext <6 x i16> %x to <6 x i32>
+  %b = zext <6 x i16> %y to <6 x i32>
+  %c = mul <6 x i32> %a, %b
+  ret <6 x i32> %c
+}