[llvm] de681d5 - [PowerPC] Do not attempt to combine fptoui without FPCVT

[Nemanja Ivanovic via llvm-commits]

Author: Nemanja Ivanovic
Date: 2023-05-24T11:14:26-05:00
New Revision: de681d53ba8278ebb14b056bd7a2385bbe48ad82

URL: https://github.com/llvm/llvm-project/commit/de681d53ba8278ebb14b056bd7a2385bbe48ad82
DIFF: https://github.com/llvm/llvm-project/commit/de681d53ba8278ebb14b056bd7a2385bbe48ad82.diff

LOG: [PowerPC] Do not attempt to combine fptoui without FPCVT

Commit 8064caf83fb166b709bfe0e7641c5181341cb064 added a call
to a function that performs this combine without checking whether
the target supports FPCVT. This caused asserts to trip on BE bots
as the default target does not have this feature.

Added: 
    llvm/test/CodeGen/PowerPC/fptoui-be-crash.ll

Modified: 
    llvm/lib/Target/PowerPC/PPCISelLowering.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
index 358be0a69a0cc..b51abdb17d31f 100644
--- a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -15515,8 +15515,10 @@ SDValue PPCTargetLowering::PerformDAGCombine(SDNode *N,
 
     EVT Op1VT = N->getOperand(1).getValueType();
     unsigned Opcode = N->getOperand(1).getOpcode();
+    bool NeedsFPCVT = Opcode == ISD::FP_TO_UINT && Op1VT == MVT::i64;
 
-    if (Opcode == ISD::FP_TO_SINT || Opcode == ISD::FP_TO_UINT) {
+    if ((Opcode == ISD::FP_TO_SINT || Opcode == ISD::FP_TO_UINT) &&
+        (!NeedsFPCVT || Subtarget.hasFPCVT())) {
       SDValue Val= combineStoreFPToInt(N, DCI);
       if (Val)
         return Val;

diff  --git a/llvm/test/CodeGen/PowerPC/fptoui-be-crash.ll b/llvm/test/CodeGen/PowerPC/fptoui-be-crash.ll
new file mode 100644
index 0000000000000..fc93f893b1d5d
--- /dev/null
+++ b/llvm/test/CodeGen/PowerPC/fptoui-be-crash.ll
@@ -0,0 +1,76 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -verify-machineinstrs -ppc-asm-full-reg-names \
+; RUN:   -mtriple=powerpc64-unknown-linux-gnu < %s | FileCheck %s
+define dso_local void @calc_buffer() local_unnamed_addr #0 {
+; CHECK-LABEL: calc_buffer:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    ld r3, 0(r3)
+; CHECK-NEXT:    sradi r5, r3, 53
+; CHECK-NEXT:    rldicl r6, r3, 63, 1
+; CHECK-NEXT:    clrldi r7, r3, 63
+; CHECK-NEXT:    clrldi r4, r3, 53
+; CHECK-NEXT:    addi r5, r5, 1
+; CHECK-NEXT:    or r7, r7, r6
+; CHECK-NEXT:    cmpldi r5, 1
+; CHECK-NEXT:    clrldi r5, r7, 53
+; CHECK-NEXT:    addi r4, r4, 2047
+; CHECK-NEXT:    addi r5, r5, 2047
+; CHECK-NEXT:    or r5, r5, r6
+; CHECK-NEXT:    rldicl r6, r3, 10, 54
+; CHECK-NEXT:    or r4, r4, r3
+; CHECK-NEXT:    addi r6, r6, 1
+; CHECK-NEXT:    rldicl r5, r5, 53, 11
+; CHECK-NEXT:    cmpldi cr1, r6, 1
+; CHECK-NEXT:    rldicr r4, r4, 0, 52
+; CHECK-NEXT:    rldicl r5, r5, 11, 1
+; CHECK-NEXT:    bc 12, gt, .LBB0_2
+; CHECK-NEXT:  # %bb.1:
+; CHECK-NEXT:    ori r4, r3, 0
+; CHECK-NEXT:    b .LBB0_2
+; CHECK-NEXT:  .LBB0_2:
+; CHECK-NEXT:    bc 12, 4*cr1+gt, .LBB0_4
+; CHECK-NEXT:  # %bb.3:
+; CHECK-NEXT:    ori r5, r7, 0
+; CHECK-NEXT:    b .LBB0_4
+; CHECK-NEXT:  .LBB0_4:
+; CHECK-NEXT:    cmpdi r3, 0
+; CHECK-NEXT:    std r4, -32(r1)
+; CHECK-NEXT:    std r5, -24(r1)
+; CHECK-NEXT:    bc 12, lt, .LBB0_6
+; CHECK-NEXT:  # %bb.5:
+; CHECK-NEXT:    lfd f0, -32(r1)
+; CHECK-NEXT:    fcfid f0, f0
+; CHECK-NEXT:    frsp f0, f0
+; CHECK-NEXT:    b .LBB0_7
+; CHECK-NEXT:  .LBB0_6:
+; CHECK-NEXT:    lfd f0, -24(r1)
+; CHECK-NEXT:    fcfid f0, f0
+; CHECK-NEXT:    frsp f0, f0
+; CHECK-NEXT:    fadds f0, f0, f0
+; CHECK-NEXT:  .LBB0_7:
+; CHECK-NEXT:    addis r3, r2, .LCPI0_0 at toc@ha
+; CHECK-NEXT:    li r4, 1
+; CHECK-NEXT:    lfs f1, .LCPI0_0 at toc@l(r3)
+; CHECK-NEXT:    rldic r4, r4, 63, 0
+; CHECK-NEXT:    fsubs f2, f0, f1
+; CHECK-NEXT:    fctidz f2, f2
+; CHECK-NEXT:    stfd f2, -8(r1)
+; CHECK-NEXT:    fctidz f2, f0
+; CHECK-NEXT:    stfd f2, -16(r1)
+; CHECK-NEXT:    ld r3, -8(r1)
+; CHECK-NEXT:    ld r5, -16(r1)
+; CHECK-NEXT:    fcmpu cr0, f0, f1
+; CHECK-NEXT:    xor r3, r3, r4
+; CHECK-NEXT:    bc 12, lt, .LBB0_8
+; CHECK-NEXT:    b .LBB0_9
+; CHECK-NEXT:  .LBB0_8:
+; CHECK-NEXT:    addi r3, r5, 0
+; CHECK-NEXT:  .LBB0_9:
+; CHECK-NEXT:    std r3, 0(r3)
+  %load_initial = load i64, ptr poison, align 8
+  %conv39 = uitofp i64 %load_initial to float
+  %add48 = fadd float 0.000000e+00, %conv39
+  %conv49 = fptoui float %add48 to i64
+  store i64 %conv49, ptr poison, align 8
+  unreachable
+}