[PATCH] D125159: [CodeGen] Fix ConvertNodeToLibcall for STRICT_FPOWI

[Xiang Zhang via Phabricator via llvm-commits]

xiangzhangllvm created this revision.
xiangzhangllvm added reviewers: craig.topper, pengfei, LuoYuanke, RKSimon.
Herald added subscribers: StephenFan, hiraditya.
Herald added a project: All.
xiangzhangllvm requested review of this revision.
Herald added a project: LLVM.
Herald added a subscriber: llvm-commits.

Correct the Operand index when convert STRICT_FPOWI to FPOW.
(the old code miss considering the chain operand of STRICT_FPOWI, will cause crash.)


https://reviews.llvm.org/D125159

Files:
  llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
  llvm/test/CodeGen/X86/float-strict-powi-convert.ll


Index: llvm/test/CodeGen/X86/float-strict-powi-convert.ll
===================================================================
--- /dev/null
+++ llvm/test/CodeGen/X86/float-strict-powi-convert.ll
@@ -0,0 +1,21 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=x86_64-pc-windows-msvc -disable-strictnode-mutation %s -o - | FileCheck %s -check-prefix=WIN
+; RUN: llc -mtriple=x86_64-pc-linux -disable-strictnode-mutation %s -o -| FileCheck %s -check-prefix=UNIX
+
+declare float @llvm.experimental.constrained.powi.f32.i32(float, i32, metadata, metadata)
+
+define float @powi_f64(float %a, i32 %b) nounwind strictfp {
+; WIN-LABEL: powi_f64:
+; WIN:       # %bb.0:
+; WIN-NEXT:    cvtsi2ss %edx, %xmm1
+; WIN-NEXT:    jmp powf # TAILCALL
+;
+; UNIX-LABEL: powi_f64:
+; UNIX:       # %bb.0:
+; UNIX-NEXT:    pushq %rax
+; UNIX-NEXT:    callq __powisf2 at PLT
+; UNIX-NEXT:    popq %rax
+; UNIX-NEXT:    retq
+  %1 = call float @llvm.experimental.constrained.powi.f32.i32(float %a, i32 %b, metadata !"round.tonearest", metadata !"fpexcept.ignore") strictfp
+  ret float %1
+}
Index: llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
===================================================================
--- llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
+++ llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
@@ -4096,17 +4096,18 @@
   case ISD::STRICT_FPOWI: {
     RTLIB::Libcall LC = RTLIB::getPOWI(Node->getSimpleValueType(0));
     assert(LC != RTLIB::UNKNOWN_LIBCALL && "Unexpected fpowi.");
+    unsigned Offset = Node->isStrictFPOpcode() ? 1 : 0;
     if (!TLI.getLibcallName(LC)) {
       // Some targets don't have a powi libcall; use pow instead.
       SDValue Exponent = DAG.getNode(ISD::SINT_TO_FP, SDLoc(Node),
                                      Node->getValueType(0),
-                                     Node->getOperand(1));
+                                     Node->getOperand(1 + Offset));
       Results.push_back(DAG.getNode(ISD::FPOW, SDLoc(Node),
-                                    Node->getValueType(0), Node->getOperand(0),
+                                    Node->getValueType(0),
+                                    Node->getOperand(Offset),
                                     Exponent));
       break;
     }
-    unsigned Offset = Node->isStrictFPOpcode() ? 1 : 0;
     bool ExponentHasSizeOfInt =
         DAG.getLibInfo().getIntSize() ==
         Node->getOperand(1 + Offset).getValueType().getSizeInBits();


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D125159.427826.patch
Type: text/x-patch
Size: 2487 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20220507/a9805329/attachment-0001.bin>