[llvm] 3f94a38 - [SVE] Fix invalid INSERT_SUBVECTOR creation when lowering fixed length fp-int conversions.

[Paul Walker via llvm-commits]

Author: Paul Walker
Date: 2023-01-24T12:29:25Z
New Revision: 3f94a38388fd07f4436da97ddef6abb48ebd0554

URL: https://github.com/llvm/llvm-project/commit/3f94a38388fd07f4436da97ddef6abb48ebd0554
DIFF: https://github.com/llvm/llvm-project/commit/3f94a38388fd07f4436da97ddef6abb48ebd0554.diff

LOG: [SVE] Fix invalid INSERT_SUBVECTOR creation when lowering fixed length fp-int conversions.

The original logic resulted in inserting an integer vector into
a floating point one and vice versa. Patch also adds the missing
assert that would have caught the issue.

Differential Revision: https://reviews.llvm.org/D142303

Added: 
    

Modified: 
    llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
    llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
    llvm/test/CodeGen/AArch64/sve-streaming-mode-fixed-length-int-to-fp.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index 7a390ba7ce904..953e940039cce 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -6711,6 +6711,8 @@ SDValue SelectionDAG::getNode(unsigned Opcode, const SDLoc &DL, EVT VT,
            "Dest and insert subvector source types must match!");
     assert(VT.isVector() && N2VT.isVector() &&
            "Insert subvector VTs must be vectors!");
+    assert(VT.getVectorElementType() == N2VT.getVectorElementType() &&
+           "Insert subvector VTs must have the same element type!");
     assert((VT.isScalableVector() || N2VT.isFixedLengthVector()) &&
            "Cannot insert a scalable vector into a fixed length vector!");
     assert((VT.isScalableVector() != N2VT.isScalableVector() ||

diff  --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index abe4ef7447142..6f2058c721577 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -23619,10 +23619,10 @@ AArch64TargetLowering::LowerFixedLengthIntToFPToSVE(SDValue Op,
     Val = DAG.getNode(IsSigned ? ISD::SIGN_EXTEND : ISD::ZERO_EXTEND, DL,
                       VT.changeTypeToInteger(), Val);
 
-    Val = convertToScalableVector(DAG, ContainerSrcVT, Val);
-    Val = getSVESafeBitCast(ContainerDstVT.changeTypeToInteger(), Val, DAG);
-    // Safe to use a larger than specified operand since we just unpacked the
-    // data, hence the upper bits are zero.
+    // Safe to use a larger than specified operand because by promoting the
+    // value nothing has changed from an arithmetic point of view.
+    Val =
+        convertToScalableVector(DAG, ContainerDstVT.changeTypeToInteger(), Val);
     Val = DAG.getNode(Opcode, DL, ContainerDstVT, Pg, Val,
                       DAG.getUNDEF(ContainerDstVT));
     return convertFromScalableVector(DAG, VT, Val);
@@ -23665,7 +23665,7 @@ AArch64TargetLowering::LowerFixedLengthFPToIntToSVE(SDValue Op,
     Val = DAG.getNode(ISD::BITCAST, DL, SrcVT.changeTypeToInteger(), Val);
     Val = DAG.getNode(ISD::ANY_EXTEND, DL, VT, Val);
 
-    Val = convertToScalableVector(DAG, ContainerSrcVT, Val);
+    Val = convertToScalableVector(DAG, ContainerDstVT, Val);
     Val = getSVESafeBitCast(CvtVT, Val, DAG);
     Val = DAG.getNode(Opcode, DL, ContainerDstVT, Pg, Val,
                       DAG.getUNDEF(ContainerDstVT));

diff  --git a/llvm/test/CodeGen/AArch64/sve-streaming-mode-fixed-length-int-to-fp.ll b/llvm/test/CodeGen/AArch64/sve-streaming-mode-fixed-length-int-to-fp.ll
index 493778e17f7cd..ab0c348af18c8 100644
--- a/llvm/test/CodeGen/AArch64/sve-streaming-mode-fixed-length-int-to-fp.ll
+++ b/llvm/test/CodeGen/AArch64/sve-streaming-mode-fixed-length-int-to-fp.ll
@@ -182,8 +182,8 @@ define void @ucvtf_v8i16_v8f64(ptr %a, ptr %b) #0 {
 ; CHECK-NEXT:    uunpklo z3.d, z0.s
 ; CHECK-NEXT:    ext z0.b, z0.b, z0.b, #8
 ; CHECK-NEXT:    uunpklo z0.d, z0.s
-; CHECK-NEXT:    ucvtf z3.d, p0/m, z3.d
 ; CHECK-NEXT:    ext z1.b, z1.b, z1.b, #8
+; CHECK-NEXT:    ucvtf z3.d, p0/m, z3.d
 ; CHECK-NEXT:    ucvtf z0.d, p0/m, z0.d
 ; CHECK-NEXT:    ucvtf z2.d, p0/m, z2.d
 ; CHECK-NEXT:    uunpklo z1.d, z1.s
@@ -758,8 +758,8 @@ define void @scvtf_v8i16_v8f64(ptr %a, ptr %b) #0 {
 ; CHECK-NEXT:    sunpklo z3.d, z0.s
 ; CHECK-NEXT:    ext z0.b, z0.b, z0.b, #8
 ; CHECK-NEXT:    sunpklo z0.d, z0.s
-; CHECK-NEXT:    scvtf z3.d, p0/m, z3.d
 ; CHECK-NEXT:    ext z1.b, z1.b, z1.b, #8
+; CHECK-NEXT:    scvtf z3.d, p0/m, z3.d
 ; CHECK-NEXT:    scvtf z0.d, p0/m, z0.d
 ; CHECK-NEXT:    scvtf z2.d, p0/m, z2.d
 ; CHECK-NEXT:    sunpklo z1.d, z1.s