[SelectionDAG] Improve support for promotion of <1 x fX> floating point argument types (PR31088)

PR31088 demonstrated that we were assuming that only integers require promotion from <1 x iX> types, when in fact float types may require it as well - in this case half floats.

This patch adds support for extension/truncation for both integer and float types.

Differential Revision: https://reviews.llvm.org/D32391

llvm-svn: 301910

include/llvm/CodeGen/SelectionDAG.h

@@ -688,6 +688,10 @@ public:
   /// Example: shuffle A, B, <0,5,2,7> -> shuffle B, A, <4,1,6,3>
   SDValue getCommutedVectorShuffle(const ShuffleVectorSDNode &SV);
 
+  /// Convert Op, which must be of float type, to the
+  /// float type VT, by either extending or rounding (by truncation).
+  SDValue getFPExtendOrRound(SDValue Op, const SDLoc &DL, EVT VT);
+
   /// Convert Op, which must be of integer type, to the
   /// integer type VT, by either any-extending or truncating it.
   SDValue getAnyExtOrTrunc(SDValue Op, const SDLoc &DL, EVT VT);

lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp

@@ -523,16 +523,17 @@ SDValue DAGTypeLegalizer::ScalarizeVecOp_CONCAT_VECTORS(SDNode *N) {
   return DAG.getBuildVector(N->getValueType(0), SDLoc(N), Ops);
 }
 
-/// If the input is a vector that needs to be scalarized, it must be <1 x ty>,
+/// If the input is a vector that needs to be scalarized, it must be <1 x ty>,
-/// so just return the element, ignoring the index.
+/// so just return the element, ignoring the index.
-SDValue DAGTypeLegalizer::ScalarizeVecOp_EXTRACT_VECTOR_ELT(SDNode *N) {
+SDValue DAGTypeLegalizer::ScalarizeVecOp_EXTRACT_VECTOR_ELT(SDNode *N) {
-  EVT VT = N->getValueType(0);
+  EVT VT = N->getValueType(0);
-  SDValue Res = GetScalarizedVector(N->getOperand(0));
+  SDValue Res = GetScalarizedVector(N->getOperand(0));
-  if (Res.getValueType() != VT)
+  if (Res.getValueType() != VT)
-    Res = DAG.getNode(ISD::ANY_EXTEND, SDLoc(N), VT, Res);
+    Res = VT.isFloatingPoint()
-  return Res;
+              ? DAG.getNode(ISD::FP_EXTEND, SDLoc(N), VT, Res)
-}
+              : DAG.getNode(ISD::ANY_EXTEND, SDLoc(N), VT, Res);
-
+  return Res;
+}
 
 /// If the input condition is a vector that needs to be scalarized, it must be
 /// <1 x i1>, so just convert to a normal ISD::SELECT

lib/CodeGen/SelectionDAG/SelectionDAG.cpp

@@ -959,6 +959,12 @@ void SelectionDAG::clear() {
   DbgInfo->clear();
 }
 
+SDValue SelectionDAG::getFPExtendOrRound(SDValue Op, const SDLoc &DL, EVT VT) {
+  return VT.bitsGT(Op.getValueType())
+             ? getNode(ISD::FP_EXTEND, DL, VT, Op)
+             : getNode(ISD::FP_ROUND, DL, VT, Op, getIntPtrConstant(0, DL));
+}
+
 SDValue SelectionDAG::getAnyExtOrTrunc(SDValue Op, const SDLoc &DL, EVT VT) {
   return VT.bitsGT(Op.getValueType()) ?
     getNode(ISD::ANY_EXTEND, DL, VT, Op) :

lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

@@ -350,7 +350,8 @@ static SDValue getCopyFromPartsVector(SelectionDAG &DAG, const SDLoc &DL,
 
   EVT ValueSVT = ValueVT.getVectorElementType();
   if (ValueVT.getVectorNumElements() == 1 && ValueSVT != PartEVT)
-    Val = DAG.getAnyExtOrTrunc(Val, DL, ValueSVT);
+    Val = ValueVT.isFloatingPoint() ? DAG.getFPExtendOrRound(Val, DL, ValueSVT)
+                                    : DAG.getAnyExtOrTrunc(Val, DL, ValueSVT);
 
   return DAG.getBuildVector(ValueVT, DL, Val);
 }
@@ -543,10 +544,9 @@ static void getCopyToPartsVector(SelectionDAG &DAG, const SDLoc &DL,
       Val = DAG.getNode(
           ISD::EXTRACT_VECTOR_ELT, DL, PartVT, Val,
           DAG.getConstant(0, DL, TLI.getVectorIdxTy(DAG.getDataLayout())));
-
-      Val = DAG.getAnyExtOrTrunc(Val, DL, PartVT);
     }
 
+    assert(Val.getValueType() == PartVT && "Unexpected vector part value type");
     Parts[0] = Val;
     return;
   }

test/CodeGen/NVPTX/f16-instructions.ll

@@ -36,6 +36,21 @@ define half @test_fadd(half %a, half %b) #0 {
   ret half %r
 }
 
+; CHECK-LABEL: test_fadd_v1f16(
+; CHECK-DAG:  ld.param.b16    [[A:%h[0-9]+]], [test_fadd_v1f16_param_0];
+; CHECK-DAG:  ld.param.b16    [[B:%h[0-9]+]], [test_fadd_v1f16_param_1];
+; CHECK-F16-NEXT:   add.rn.f16     [[R:%h[0-9]+]], [[A]], [[B]];
+; CHECK-NOF16-DAG:  cvt.f32.f16    [[A32:%f[0-9]+]], [[A]]
+; CHECK-NOF16-DAG:  cvt.f32.f16    [[B32:%f[0-9]+]], [[B]]
+; CHECK-NOF16-NEXT: add.rn.f32     [[R32:%f[0-9]+]], [[A32]], [[B32]];
+; CHECK-NOF16-NEXT: cvt.rn.f16.f32 [[R:%h[0-9]+]], [[R32]]
+; CHECK-NEXT: st.param.b16    [func_retval0+0], [[R]];
+; CHECK-NEXT: ret;
+define <1 x half> @test_fadd_v1f16(<1 x half> %a, <1 x half> %b) #0 {
+  %r = fadd <1 x half> %a, %b
+  ret <1 x half> %r
+}
+
 ; Check that we can lower fadd with immediate arguments.
 ; CHECK-LABEL: test_fadd_imm_0(
 ; CHECK-DAG:  ld.param.b16    [[B:%h[0-9]+]], [test_fadd_imm_0_param_0];

test/CodeGen/X86/pr31088.ll

@@ -3,6 +3,63 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X64
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+f16c | FileCheck %s --check-prefix=F16C
 
+define <1 x half> @ir_fadd_v1f16(<1 x half> %arg0, <1 x half> %arg1) nounwind {
+; X86-LABEL: ir_fadd_v1f16:
+; X86:       # BB#0:
+; X86-NEXT:    subl $28, %esp
+; X86-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X86-NEXT:    movss %xmm0, (%esp)
+; X86-NEXT:    calll __gnu_f2h_ieee
+; X86-NEXT:    movzwl %ax, %eax
+; X86-NEXT:    movl %eax, (%esp)
+; X86-NEXT:    calll __gnu_h2f_ieee
+; X86-NEXT:    fstpt {{[0-9]+}}(%esp) # 10-byte Folded Spill
+; X86-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X86-NEXT:    movss %xmm0, (%esp)
+; X86-NEXT:    calll __gnu_f2h_ieee
+; X86-NEXT:    movzwl %ax, %eax
+; X86-NEXT:    movl %eax, (%esp)
+; X86-NEXT:    fldt {{[0-9]+}}(%esp) # 10-byte Folded Reload
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    calll __gnu_h2f_ieee
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X86-NEXT:    addss {{[0-9]+}}(%esp), %xmm0
+; X86-NEXT:    movss %xmm0, {{[0-9]+}}(%esp)
+; X86-NEXT:    flds {{[0-9]+}}(%esp)
+; X86-NEXT:    addl $28, %esp
+; X86-NEXT:    retl
+;
+; X64-LABEL: ir_fadd_v1f16:
+; X64:       # BB#0:
+; X64-NEXT:    pushq %rax
+; X64-NEXT:    movss %xmm0, {{[0-9]+}}(%rsp) # 4-byte Spill
+; X64-NEXT:    movaps %xmm1, %xmm0
+; X64-NEXT:    callq __gnu_f2h_ieee
+; X64-NEXT:    movzwl %ax, %edi
+; X64-NEXT:    callq __gnu_h2f_ieee
+; X64-NEXT:    movss %xmm0, (%rsp) # 4-byte Spill
+; X64-NEXT:    movss {{[0-9]+}}(%rsp), %xmm0 # 4-byte Reload
+; X64-NEXT:    # xmm0 = mem[0],zero,zero,zero
+; X64-NEXT:    callq __gnu_f2h_ieee
+; X64-NEXT:    movzwl %ax, %edi
+; X64-NEXT:    callq __gnu_h2f_ieee
+; X64-NEXT:    addss (%rsp), %xmm0 # 4-byte Folded Reload
+; X64-NEXT:    popq %rax
+; X64-NEXT:    retq
+;
+; F16C-LABEL: ir_fadd_v1f16:
+; F16C:       # BB#0:
+; F16C-NEXT:    vcvtps2ph $4, %xmm1, %xmm1
+; F16C-NEXT:    vcvtph2ps %xmm1, %xmm1
+; F16C-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
+; F16C-NEXT:    vcvtph2ps %xmm0, %xmm0
+; F16C-NEXT:    vaddss %xmm1, %xmm0, %xmm0
+; F16C-NEXT:    retq
+  %retval = fadd <1 x half> %arg0, %arg1
+  ret <1 x half> %retval
+}
+
 define <2 x half> @ir_fadd_v2f16(<2 x half> %arg0, <2 x half> %arg1) nounwind {
 ; X86-LABEL: ir_fadd_v2f16:
 ; X86:       # BB#0: