[llvm] r371343 - [X86] Teach materializeVectorConstant to not call getZeroVector/getOnesVector on the types we already have isel patterns for.

[Craig Topper via llvm-commits]

Author: ctopper
Date: Sun Sep  8 12:24:29 2019
New Revision: 371343

URL: http://llvm.org/viewvc/llvm-project?rev=371343&view=rev
Log:
[X86] Teach materializeVectorConstant to not call getZeroVector/getOnesVector on the types we already have isel patterns for.

Modified:
    llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
    llvm/trunk/test/CodeGen/X86/fold-load-vec.ll

Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=371343&r1=371342&r2=371343&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Sun Sep  8 12:24:29 2019
@@ -9137,7 +9137,8 @@ static SDValue materializeVectorConstant
   if (ISD::isBuildVectorAllZeros(Op.getNode())) {
     // Canonicalize this to <4 x i32> to 1) ensure the zero vectors are CSE'd
     // and 2) ensure that i64 scalars are eliminated on x86-32 hosts.
-    if (VT == MVT::v4i32 || VT == MVT::v8i32 || VT == MVT::v16i32)
+    if (VT.isFloatingPoint() ||
+        VT == MVT::v4i32 || VT == MVT::v8i32 || VT == MVT::v16i32)
       return Op;
 
     return getZeroVector(VT, Subtarget, DAG, DL);
@@ -9147,8 +9148,7 @@ static SDValue materializeVectorConstant
   // vectors or broken into v4i32 operations on 256-bit vectors. AVX2 can use
   // vpcmpeqd on 256-bit vectors.
   if (Subtarget.hasSSE2() && ISD::isBuildVectorAllOnes(Op.getNode())) {
-    if (VT == MVT::v4i32 || VT == MVT::v16i32 ||
-        (VT == MVT::v8i32 && Subtarget.hasInt256()))
+    if (VT == MVT::v4i32 || VT == MVT::v8i32 || VT == MVT::v16i32)
       return Op;
 
     return getOnesVector(VT, DAG, DL);

Modified: llvm/trunk/test/CodeGen/X86/fold-load-vec.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/fold-load-vec.ll?rev=371343&r1=371342&r2=371343&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/fold-load-vec.ll (original)
+++ llvm/trunk/test/CodeGen/X86/fold-load-vec.ll Sun Sep  8 12:24:29 2019
@@ -12,7 +12,7 @@ define void @sample_test(<4 x float>* %s
 ; CHECK-NEXT:    movq %rsi, {{[0-9]+}}(%rsp)
 ; CHECK-NEXT:    xorps %xmm0, %xmm0
 ; CHECK-NEXT:    movlps %xmm0, (%rsp)
-; CHECK-NEXT:    insertps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[2,3]
+; CHECK-NEXT:    unpcklps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
 ; CHECK-NEXT:    movlps %xmm0, (%rsp)
 ; CHECK-NEXT:    movlps %xmm0, (%rsi)
 ; CHECK-NEXT:    movq {{[0-9]+}}(%rsp), %rax