[PowerPC] Add missing handling for half precision

tstellar · tstellar · commit 1abba52044dd · 2020-06-22T15:03:26.000-07:00
The fix for PR39865 took care of some of the handling for half precision but it missed a number of issues that still exist. This patch fixes the remaining issues that cause crashes in the PPC back end. Fixes: https://bugs.llvm.org/show_bug.cgi?id=45776 Differential revision: https://reviews.llvm.org/D79283 (cherry picked from commit 1a493b0)
diff --git a/llvm/include/llvm/Target/TargetSelectionDAG.td b/llvm/include/llvm/Target/TargetSelectionDAG.td
@@ -959,6 +959,10 @@ def extloadi32 : PatFrag<(ops node:$ptr), (extload node:$ptr)> {
   let IsLoad = 1;
   let MemoryVT = i32;
 }
+def extloadf16 : PatFrag<(ops node:$ptr), (extload node:$ptr)> {
+  let IsLoad = 1;
+  let MemoryVT = f16;
+}
 def extloadf32 : PatFrag<(ops node:$ptr), (extload node:$ptr)> {
   let IsLoad = 1;
   let MemoryVT = f32;
@@ -1094,6 +1098,11 @@ def truncstorei32 : PatFrag<(ops node:$val, node:$ptr),
   let IsStore = 1;
   let MemoryVT = i32;
 }
+def truncstoref16 : PatFrag<(ops node:$val, node:$ptr),
+                            (truncstore node:$val, node:$ptr)> {
+  let IsStore = 1;
+  let MemoryVT = f16;
+}
 def truncstoref32 : PatFrag<(ops node:$val, node:$ptr),
                             (truncstore node:$val, node:$ptr)> {
   let IsStore = 1;
diff --git a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -167,6 +167,23 @@ PPCTargetLowering::PPCTargetLowering(const PPCTargetMachine &TM,
     setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i8, Expand);
   }
 
+  if (Subtarget.isISA3_0()) {
+    setLoadExtAction(ISD::EXTLOAD, MVT::f64, MVT::f16, Legal);
+    setLoadExtAction(ISD::EXTLOAD, MVT::f32, MVT::f16, Legal);
+    setTruncStoreAction(MVT::f64, MVT::f16, Legal);
+    setTruncStoreAction(MVT::f32, MVT::f16, Legal);
+  } else {
+    // No extending loads from f16 or HW conversions back and forth.
+    setLoadExtAction(ISD::EXTLOAD, MVT::f64, MVT::f16, Expand);
+    setOperationAction(ISD::FP16_TO_FP, MVT::f64, Expand);
+    setOperationAction(ISD::FP_TO_FP16, MVT::f64, Expand);
+    setLoadExtAction(ISD::EXTLOAD, MVT::f32, MVT::f16, Expand);
+    setOperationAction(ISD::FP16_TO_FP, MVT::f32, Expand);
+    setOperationAction(ISD::FP_TO_FP16, MVT::f32, Expand);
+    setTruncStoreAction(MVT::f64, MVT::f16, Expand);
+    setTruncStoreAction(MVT::f32, MVT::f16, Expand);
+  }
+
   setTruncStoreAction(MVT::f64, MVT::f32, Expand);
 
   // PowerPC has pre-inc load and store's.
@@ -10361,6 +10378,7 @@ SDValue PPCTargetLowering::LowerFP_EXTEND(SDValue Op, SelectionDAG &DAG) const {
   assert(Op.getOpcode() == ISD::FP_EXTEND &&
          "Should only be called for ISD::FP_EXTEND");
 
+  // FIXME: handle extends from half precision float vectors on P9.
   // We only want to custom lower an extend from v2f32 to v2f64.
   if (Op.getValueType() != MVT::v2f64 ||
       Op.getOperand(0).getValueType() != MVT::v2f32)
@@ -10574,6 +10592,11 @@ void PPCTargetLowering::ReplaceNodeResults(SDNode *N,
   case ISD::BITCAST:
     // Don't handle bitcast here.
     return;
+  case ISD::FP_EXTEND:
+    SDValue Lowered = LowerFP_EXTEND(SDValue(N, 0), DAG);
+    if (Lowered)
+      Results.push_back(Lowered);
+    return;
   }
 }
 
diff --git a/llvm/lib/Target/PowerPC/PPCISelLowering.h b/llvm/lib/Target/PowerPC/PPCISelLowering.h
@@ -637,7 +637,7 @@ namespace llvm {
     /// then the VPERM for the shuffle. All in all a very slow sequence.
     TargetLoweringBase::LegalizeTypeAction getPreferredVectorAction(MVT VT)
       const override {
-      if (VT.getScalarSizeInBits() % 8 == 0)
+      if (VT.getVectorNumElements() != 1 && VT.getScalarSizeInBits() % 8 == 0)
         return TypeWidenVector;
       return TargetLoweringBase::getPreferredVectorAction(VT);
     }
diff --git a/llvm/lib/Target/PowerPC/PPCInstrVSX.td b/llvm/lib/Target/PowerPC/PPCInstrVSX.td
@@ -3343,6 +3343,23 @@ let AddedComplexity = 400, Predicates = [HasP9Vector] in {
   def : Pat<(v2i64 (scalar_to_vector ScalarLoads.SELi16i64)),
             (v2i64 (XXPERMDIs (VEXTSH2Ds (LXSIHZX xoaddr:$src)), 0))>;
 
+  // Load/convert and convert/store patterns for f16.
+  def : Pat<(f64 (extloadf16 xoaddr:$src)),
+            (f64 (XSCVHPDP (LXSIHZX xoaddr:$src)))>;
+  def : Pat<(truncstoref16 f64:$src, xoaddr:$dst),
+            (STXSIHX (XSCVDPHP $src), xoaddr:$dst)>;
+  def : Pat<(f32 (extloadf16 xoaddr:$src)),
+            (f32 (COPY_TO_REGCLASS (XSCVHPDP (LXSIHZX xoaddr:$src)), VSSRC))>;
+  def : Pat<(truncstoref16 f32:$src, xoaddr:$dst),
+            (STXSIHX (XSCVDPHP (COPY_TO_REGCLASS $src, VSFRC)), xoaddr:$dst)>;
+  def : Pat<(f64 (f16_to_fp i32:$A)),
+            (f64 (XSCVHPDP (MTVSRWZ $A)))>;
+  def : Pat<(f32 (f16_to_fp i32:$A)),
+            (f32 (COPY_TO_REGCLASS (XSCVHPDP (MTVSRWZ $A)), VSSRC))>;
+  def : Pat<(i32 (fp_to_f16 f32:$A)),
+            (i32 (MFVSRWZ (XSCVDPHP (COPY_TO_REGCLASS $A, VSFRC))))>;
+  def : Pat<(i32 (fp_to_f16 f64:$A)), (i32 (MFVSRWZ (XSCVDPHP $A)))>;
+
   let Predicates = [IsBigEndian, HasP9Vector] in {
   // Scalar stores of i8
   def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 0)), xoaddr:$dst),
diff --git a/llvm/test/CodeGen/PowerPC/scalar_vector_test_2.ll b/llvm/test/CodeGen/PowerPC/scalar_vector_test_2.ll
@@ -11,46 +11,34 @@
 define void @test_liwzx1(<1 x float>* %A, <1 x float>* %B, <1 x float>* %C) {
 ; P9LE-LABEL: test_liwzx1:
 ; P9LE:       # %bb.0:
-; P9LE-NEXT:    lfiwzx f0, 0, r3
-; P9LE-NEXT:    lfiwzx f1, 0, r4
-; P9LE-NEXT:    xxpermdi vs0, f0, f0, 2
-; P9LE-NEXT:    xxpermdi vs1, f1, f1, 2
-; P9LE-NEXT:    xvaddsp vs0, vs0, vs1
-; P9LE-NEXT:    xxsldwi vs0, vs0, vs0, 2
-; P9LE-NEXT:    stfiwx f0, 0, r5
+; P9LE-NEXT:    lfs f0, 0(r3)
+; P9LE-NEXT:    lfs f1, 0(r4)
+; P9LE-NEXT:    xsaddsp f0, f0, f1
+; P9LE-NEXT:    stfs f0, 0(r5)
 ; P9LE-NEXT:    blr
 ;
 ; P9BE-LABEL: test_liwzx1:
 ; P9BE:       # %bb.0:
-; P9BE-NEXT:    lfiwzx f0, 0, r3
-; P9BE-NEXT:    lfiwzx f1, 0, r4
-; P9BE-NEXT:    xxsldwi vs0, f0, f0, 1
-; P9BE-NEXT:    xxsldwi vs1, f1, f1, 1
-; P9BE-NEXT:    xvaddsp vs0, vs0, vs1
-; P9BE-NEXT:    xxsldwi vs0, vs0, vs0, 3
-; P9BE-NEXT:    stfiwx f0, 0, r5
+; P9BE-NEXT:    lfs f0, 0(r3)
+; P9BE-NEXT:    lfs f1, 0(r4)
+; P9BE-NEXT:    xsaddsp f0, f0, f1
+; P9BE-NEXT:    stfs f0, 0(r5)
 ; P9BE-NEXT:    blr
 ;
 ; P8LE-LABEL: test_liwzx1:
 ; P8LE:       # %bb.0:
-; P8LE-NEXT:    lfiwzx f0, 0, r3
-; P8LE-NEXT:    lfiwzx f1, 0, r4
-; P8LE-NEXT:    xxpermdi vs0, f0, f0, 2
-; P8LE-NEXT:    xxpermdi vs1, f1, f1, 2
-; P8LE-NEXT:    xvaddsp vs0, vs0, vs1
-; P8LE-NEXT:    xxsldwi vs0, vs0, vs0, 2
-; P8LE-NEXT:    stfiwx f0, 0, r5
+; P8LE-NEXT:    lfsx f0, 0, r3
+; P8LE-NEXT:    lfsx f1, 0, r4
+; P8LE-NEXT:    xsaddsp f0, f0, f1
+; P8LE-NEXT:    stfsx f0, 0, r5
 ; P8LE-NEXT:    blr
 ;
 ; P8BE-LABEL: test_liwzx1:
 ; P8BE:       # %bb.0:
-; P8BE-NEXT:    lfiwzx f0, 0, r3
-; P8BE-NEXT:    lfiwzx f1, 0, r4
-; P8BE-NEXT:    xxsldwi vs0, f0, f0, 1
-; P8BE-NEXT:    xxsldwi vs1, f1, f1, 1
-; P8BE-NEXT:    xvaddsp vs0, vs0, vs1
-; P8BE-NEXT:    xxsldwi vs0, vs0, vs0, 3
-; P8BE-NEXT:    stfiwx f0, 0, r5
+; P8BE-NEXT:    lfsx f0, 0, r3
+; P8BE-NEXT:    lfsx f1, 0, r4
+; P8BE-NEXT:    xsaddsp f0, f0, f1
+; P8BE-NEXT:    stfsx f0, 0, r5
 ; P8BE-NEXT:    blr
 
 
@@ -65,50 +53,38 @@ define void @test_liwzx1(<1 x float>* %A, <1 x float>* %B, <1 x float>* %C) {
 define <1 x float>* @test_liwzx2(<1 x float>* %A, <1 x float>* %B, <1 x float>* %C) {
 ; P9LE-LABEL: test_liwzx2:
 ; P9LE:       # %bb.0:
-; P9LE-NEXT:    lfiwzx f0, 0, r3
-; P9LE-NEXT:    lfiwzx f1, 0, r4
-; P9LE-NEXT:    xxpermdi vs0, f0, f0, 2
-; P9LE-NEXT:    xxpermdi vs1, f1, f1, 2
-; P9LE-NEXT:    xvsubsp vs0, vs0, vs1
-; P9LE-NEXT:    xxsldwi vs0, vs0, vs0, 2
+; P9LE-NEXT:    lfs f0, 0(r3)
 ; P9LE-NEXT:    mr r3, r5
-; P9LE-NEXT:    stfiwx f0, 0, r5
+; P9LE-NEXT:    lfs f1, 0(r4)
+; P9LE-NEXT:    xssubsp f0, f0, f1
+; P9LE-NEXT:    stfs f0, 0(r5)
 ; P9LE-NEXT:    blr
 ;
 ; P9BE-LABEL: test_liwzx2:
 ; P9BE:       # %bb.0:
-; P9BE-NEXT:    lfiwzx f0, 0, r3
-; P9BE-NEXT:    lfiwzx f1, 0, r4
-; P9BE-NEXT:    xxsldwi vs0, f0, f0, 1
-; P9BE-NEXT:    xxsldwi vs1, f1, f1, 1
-; P9BE-NEXT:    xvsubsp vs0, vs0, vs1
-; P9BE-NEXT:    xxsldwi vs0, vs0, vs0, 3
+; P9BE-NEXT:    lfs f0, 0(r3)
 ; P9BE-NEXT:    mr r3, r5
-; P9BE-NEXT:    stfiwx f0, 0, r5
+; P9BE-NEXT:    lfs f1, 0(r4)
+; P9BE-NEXT:    xssubsp f0, f0, f1
+; P9BE-NEXT:    stfs f0, 0(r5)
 ; P9BE-NEXT:    blr
 ;
 ; P8LE-LABEL: test_liwzx2:
 ; P8LE:       # %bb.0:
-; P8LE-NEXT:    lfiwzx f0, 0, r3
-; P8LE-NEXT:    lfiwzx f1, 0, r4
+; P8LE-NEXT:    lfsx f0, 0, r3
+; P8LE-NEXT:    lfsx f1, 0, r4
 ; P8LE-NEXT:    mr r3, r5
-; P8LE-NEXT:    xxpermdi vs0, f0, f0, 2
-; P8LE-NEXT:    xxpermdi vs1, f1, f1, 2
-; P8LE-NEXT:    xvsubsp vs0, vs0, vs1
-; P8LE-NEXT:    xxsldwi vs0, vs0, vs0, 2
-; P8LE-NEXT:    stfiwx f0, 0, r5
+; P8LE-NEXT:    xssubsp f0, f0, f1
+; P8LE-NEXT:    stfsx f0, 0, r5
 ; P8LE-NEXT:    blr
 ;
 ; P8BE-LABEL: test_liwzx2:
 ; P8BE:       # %bb.0:
-; P8BE-NEXT:    lfiwzx f0, 0, r3
-; P8BE-NEXT:    lfiwzx f1, 0, r4
+; P8BE-NEXT:    lfsx f0, 0, r3
+; P8BE-NEXT:    lfsx f1, 0, r4
 ; P8BE-NEXT:    mr r3, r5
-; P8BE-NEXT:    xxsldwi vs0, f0, f0, 1
-; P8BE-NEXT:    xxsldwi vs1, f1, f1, 1
-; P8BE-NEXT:    xvsubsp vs0, vs0, vs1
-; P8BE-NEXT:    xxsldwi vs0, vs0, vs0, 3
-; P8BE-NEXT:    stfiwx f0, 0, r5
+; P8BE-NEXT:    xssubsp f0, f0, f1
+; P8BE-NEXT:    stfsx f0, 0, r5
 ; P8BE-NEXT:    blr
 
 
diff --git a/llvm/test/CodeGen/PowerPC/vector-constrained-fp-intrinsics.ll b/llvm/test/CodeGen/PowerPC/vector-constrained-fp-intrinsics.ll

Original file line number	Diff line number	Diff line change
`@@ -637,7 +637,7 @@ namespace llvm {`
`637`	`637`	`/// then the VPERM for the shuffle. All in all a very slow sequence.`
`638`	`638`	`TargetLoweringBase::LegalizeTypeAction getPreferredVectorAction(MVT VT)`
`639`	`639`	`const override {`
`640`		`- if (VT.getScalarSizeInBits() % 8 == 0)`
	`640`	`+ if (VT.getVectorNumElements() != 1 && VT.getScalarSizeInBits() % 8 == 0)`
`641`	`641`	`return TypeWidenVector;`
`642`	`642`	`return TargetLoweringBase::getPreferredVectorAction(VT);`
`643`	`643`	`}`