Merging r339536:

------------------------------------------------------------------------
r339536 | ctopper | 2018-08-13 08:53:49 +0200 (Mon, 13 Aug 2018) | 3 lines

[SelectionDAG] In PromoteFloatOp_BITCAST, insert a bitcast after the fp_to_fp16 in case the result type isn't a scalar integer.

This is another variation of PR38533. In this case, the result type of the bitcast is legal and 16-bits wide, but not a scalar integer. So we need to emit the convert to i16 and then bitcast it to the true result type. This new bitcast will be further type legalized if necessary.
------------------------------------------------------------------------

llvm-svn: 339857

2 files changed, 58 insertions, 20 deletions

diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
index feca1c10f52..2e6f6edbce5 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
@@ -1778,15 +1778,16 @@ SDValue DAGTypeLegalizer::PromoteFloatOp_BITCAST(SDNode *N, unsigned OpNo) {
   SDValue Op = N->getOperand(0);
   EVT OpVT = Op->getValueType(0);
 
-  EVT IVT = EVT::getIntegerVT(*DAG.getContext(), OpVT.getSizeInBits());
-  assert (IVT == N->getValueType(0) && "Bitcast to type of different size");
-
   SDValue Promoted = GetPromotedFloat(N->getOperand(0));
   EVT PromotedVT = Promoted->getValueType(0);
 
   // Convert the promoted float value to the desired IVT.
-  return DAG.getNode(GetPromotionOpcode(PromotedVT, OpVT), SDLoc(N), IVT,
-                     Promoted);
+  EVT IVT = EVT::getIntegerVT(*DAG.getContext(), OpVT.getSizeInBits());
+  SDValue Convert = DAG.getNode(GetPromotionOpcode(PromotedVT, OpVT), SDLoc(N),
+                                IVT, Promoted);
+  // The final result type might not be an scalar so we need a bitcast. The
+  // bitcast will be further legalized if needed.
+  return DAG.getBitcast(N->getValueType(0), Convert);
 }
 
 // Promote Operand 1 of FCOPYSIGN.  Operand 0 ought to be handled by
@@ -1941,9 +1942,11 @@ void DAGTypeLegalizer::PromoteFloatResult(SDNode *N, unsigned ResNo) {
 SDValue DAGTypeLegalizer::PromoteFloatRes_BITCAST(SDNode *N) {
   EVT VT = N->getValueType(0);
   EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), VT);
-  // Input type isn't guaranteed to be i16 so bitcast if not. The bitcast
-  // will be legalized further if necessary.
-  SDValue Cast = DAG.getBitcast(MVT::i16, N->getOperand(0));
+  // Input type isn't guaranteed to be a scalar int so bitcast if not. The
+  // bitcast will be legalized further if necessary.
+  EVT IVT = EVT::getIntegerVT(*DAG.getContext(),
+                              N->getOperand(0).getValueType().getSizeInBits());
+  SDValue Cast = DAG.getBitcast(IVT, N->getOperand(0));
   return DAG.getNode(GetPromotionOpcode(VT, NVT), SDLoc(N), NVT, Cast);
 }
 
diff --git a/llvm/test/CodeGen/X86/pr38533.ll b/llvm/test/CodeGen/X86/pr38533.ll
index d6429b7a1fa..96d003ba1a8 100644
--- a/llvm/test/CodeGen/X86/pr38533.ll
+++ b/llvm/test/CodeGen/X86/pr38533.ll
@@ -1,5 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-unknown | FileCheck %s
+; RUN: llc < %s -mtriple=x86_64-unknown | FileCheck %s --check-prefixes=CHECK,SSE
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=avx512f | FileCheck %s --check-prefixes=CHECK,AVX512
 
 ; This test makes sure that a vector that needs to be promoted that is bitcasted to fp16 is legalized correctly without causing a width mismatch.
 define void @constant_fold_vector_to_half() {
@@ -13,18 +14,52 @@ define void @constant_fold_vector_to_half() {
 
 ; Similarly this makes sure that the opposite bitcast of the above is also legalized without crashing.
 define void @pr38533_2(half %x) {
-; CHECK-LABEL: pr38533_2:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    pushq %rax
-; CHECK-NEXT:    .cfi_def_cfa_offset 16
-; CHECK-NEXT:    callq __gnu_f2h_ieee
-; CHECK-NEXT:    movw %ax, {{[0-9]+}}(%rsp)
-; CHECK-NEXT:    movzwl {{[0-9]+}}(%rsp), %eax
-; CHECK-NEXT:    movw %ax, (%rax)
-; CHECK-NEXT:    popq %rax
-; CHECK-NEXT:    .cfi_def_cfa_offset 8
-; CHECK-NEXT:    retq
+; SSE-LABEL: pr38533_2:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pushq %rax
+; SSE-NEXT:    .cfi_def_cfa_offset 16
+; SSE-NEXT:    callq __gnu_f2h_ieee
+; SSE-NEXT:    movw %ax, {{[0-9]+}}(%rsp)
+; SSE-NEXT:    movzwl {{[0-9]+}}(%rsp), %eax
+; SSE-NEXT:    movw %ax, (%rax)
+; SSE-NEXT:    popq %rax
+; SSE-NEXT:    .cfi_def_cfa_offset 8
+; SSE-NEXT:    retq
+;
+; AVX512-LABEL: pr38533_2:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
+; AVX512-NEXT:    vmovd %xmm0, %eax
+; AVX512-NEXT:    movw %ax, -{{[0-9]+}}(%rsp)
+; AVX512-NEXT:    movzwl -{{[0-9]+}}(%rsp), %eax
+; AVX512-NEXT:    movw %ax, (%rax)
+; AVX512-NEXT:    retq
   %a = bitcast half %x to <4 x i4>
   store volatile <4 x i4> %a, <4 x i4>* undef
   ret void
 }
+
+; This case is a bitcast from fp16 to a 16-bit wide legal vector type. In this case the result type is legal when the bitcast gets type legalized.
+define void @pr38533_3(half %x) {
+; SSE-LABEL: pr38533_3:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pushq %rax
+; SSE-NEXT:    .cfi_def_cfa_offset 16
+; SSE-NEXT:    callq __gnu_f2h_ieee
+; SSE-NEXT:    movw %ax, (%rsp)
+; SSE-NEXT:    movzwl (%rsp), %eax
+; SSE-NEXT:    movw %ax, (%rax)
+; SSE-NEXT:    popq %rax
+; SSE-NEXT:    .cfi_def_cfa_offset 8
+; SSE-NEXT:    retq
+;
+; AVX512-LABEL: pr38533_3:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
+; AVX512-NEXT:    vmovd %xmm0, %eax
+; AVX512-NEXT:    movw %ax, (%rax)
+; AVX512-NEXT:    retq
+  %a = bitcast half %x to <16 x i1>
+  store volatile <16 x i1> %a, <16 x i1>* undef
+  ret void
+}