Skip to content

release/20.x: [HEXAGON] Fix corner cases for hwloops pass (#135439) #135657

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Apr 25, 2025

Conversation

llvmbot
Copy link
Member

@llvmbot llvmbot commented Apr 14, 2025

Backport da8ce56

Requested by: @androm3da

@llvmbot
Copy link
Member Author

llvmbot commented Apr 14, 2025

@iajbar What do you think about merging this PR to the release branch?

@llvmbot
Copy link
Member Author

llvmbot commented Apr 14, 2025

@llvm/pr-subscribers-backend-hexagon

Author: None (llvmbot)

Changes

Backport da8ce56

Requested by: @androm3da


Full diff: https://github.com/llvm/llvm-project/pull/135657.diff

3 Files Affected:

  • (modified) llvm/lib/Target/Hexagon/HexagonHardwareLoops.cpp (+45-1)
  • (added) llvm/test/CodeGen/Hexagon/hwloop-dist-check.mir (+277)
  • (modified) llvm/test/CodeGen/Hexagon/swp-phi-start.ll (+3-2)
diff --git a/llvm/lib/Target/Hexagon/HexagonHardwareLoops.cpp b/llvm/lib/Target/Hexagon/HexagonHardwareLoops.cpp
index 9334746349240..dd4b240455126 100644
--- a/llvm/lib/Target/Hexagon/HexagonHardwareLoops.cpp
+++ b/llvm/lib/Target/Hexagon/HexagonHardwareLoops.cpp
@@ -731,6 +731,11 @@ CountValue *HexagonHardwareLoops::computeCount(MachineLoop *Loop,
                                                Register IVReg,
                                                int64_t IVBump,
                                                Comparison::Kind Cmp) const {
+  LLVM_DEBUG(llvm::dbgs() << "Loop: " << *Loop << "\n");
+  LLVM_DEBUG(llvm::dbgs() << "Initial Value: " << *Start << "\n");
+  LLVM_DEBUG(llvm::dbgs() << "End Value: " << *End << "\n");
+  LLVM_DEBUG(llvm::dbgs() << "Inc/Dec Value: " << IVBump << "\n");
+  LLVM_DEBUG(llvm::dbgs() << "Comparison: " << Cmp << "\n");
   // Cannot handle comparison EQ, i.e. while (A == B).
   if (Cmp == Comparison::EQ)
     return nullptr;
@@ -846,6 +851,7 @@ CountValue *HexagonHardwareLoops::computeCount(MachineLoop *Loop,
   if (IVBump < 0) {
     std::swap(Start, End);
     IVBump = -IVBump;
+    std::swap(CmpLess, CmpGreater);
   }
   // Cmp may now have a wrong direction, e.g.  LEs may now be GEs.
   // Signedness, and "including equality" are preserved.
@@ -989,7 +995,45 @@ CountValue *HexagonHardwareLoops::computeCount(MachineLoop *Loop,
     CountSR = 0;
   }
 
-  return new CountValue(CountValue::CV_Register, CountR, CountSR);
+  const TargetRegisterClass *PredRC = &Hexagon::PredRegsRegClass;
+  Register MuxR = CountR;
+  unsigned MuxSR = CountSR;
+  // For the loop count to be valid unsigned number, CmpLess should imply
+  // Dist >= 0. Similarly, CmpGreater should imply Dist < 0. We can skip the
+  // check if the initial distance is zero and the comparison is LTu || LTEu.
+  if (!(Start->isImm() && StartV == 0 && Comparison::isUnsigned(Cmp) &&
+        CmpLess) &&
+      (CmpLess || CmpGreater)) {
+    // Generate:
+    //   DistCheck = CMP_GT DistR,  0   --> CmpLess
+    //   DistCheck = CMP_GT DistR, -1   --> CmpGreater
+    Register DistCheckR = MRI->createVirtualRegister(PredRC);
+    const MCInstrDesc &DistCheckD = TII->get(Hexagon::C2_cmpgti);
+    BuildMI(*PH, InsertPos, DL, DistCheckD, DistCheckR)
+        .addReg(DistR, 0, DistSR)
+        .addImm((CmpLess) ? 0 : -1);
+
+    // Generate:
+    //   MUXR = MUX DistCheck, CountR, 1   --> CmpLess
+    //   MUXR = MUX DistCheck, 1, CountR   --> CmpGreater
+    MuxR = MRI->createVirtualRegister(IntRC);
+    if (CmpLess) {
+      const MCInstrDesc &MuxD = TII->get(Hexagon::C2_muxir);
+      BuildMI(*PH, InsertPos, DL, MuxD, MuxR)
+          .addReg(DistCheckR)
+          .addReg(CountR, 0, CountSR)
+          .addImm(1);
+    } else {
+      const MCInstrDesc &MuxD = TII->get(Hexagon::C2_muxri);
+      BuildMI(*PH, InsertPos, DL, MuxD, MuxR)
+          .addReg(DistCheckR)
+          .addImm(1)
+          .addReg(CountR, 0, CountSR);
+    }
+    MuxSR = 0;
+  }
+
+  return new CountValue(CountValue::CV_Register, MuxR, MuxSR);
 }
 
 /// Return true if the operation is invalid within hardware loop.
diff --git a/llvm/test/CodeGen/Hexagon/hwloop-dist-check.mir b/llvm/test/CodeGen/Hexagon/hwloop-dist-check.mir
new file mode 100644
index 0000000000000..9f8c14a314309
--- /dev/null
+++ b/llvm/test/CodeGen/Hexagon/hwloop-dist-check.mir
@@ -0,0 +1,277 @@
+# RUN: llc --mtriple=hexagon -run-pass=hwloops %s -o - | FileCheck %s
+
+# CHECK-LABEL: name: f
+# CHECK: [[R1:%[0-9]+]]:predregs = C2_cmpgti [[R0:%[0-9]+]], 0
+# CHECK: [[R3:%[0-9]+]]:intregs = C2_muxir [[R1:%[0-9]+]], [[R2:%[0-9]+]], 1
+# CHECK-LABEL: name: g
+# CHECK: [[R1:%[0-9]+]]:predregs = C2_cmpgti [[R0:%[0-9]+]], 0
+# CHECK: [[R3:%[0-9]+]]:intregs = C2_muxir [[R1:%[0-9]+]], [[R2:%[0-9]+]], 1
+--- |
+  @a = dso_local global [255 x ptr] zeroinitializer, align 8
+
+  ; Function Attrs: minsize nofree norecurse nosync nounwind optsize memory(write, argmem: none, inaccessiblemem: none)
+  define dso_local void @f(i32 noundef %m) local_unnamed_addr #0 {
+  entry:
+    %cond = tail call i32 @llvm.smax.i32(i32 %m, i32 2)
+    %0 = add nsw i32 %cond, -4
+    %1 = shl i32 %cond, 3
+    %cgep = getelementptr i8, ptr @a, i32 %1
+    %cgep36 = bitcast ptr @a to ptr
+    br label %do.body
+
+  do.body:                                          ; preds = %do.body, %entry
+    %lsr.iv1 = phi ptr [ %cgep4, %do.body ], [ %cgep, %entry ]
+    %lsr.iv = phi i32 [ %lsr.iv.next, %do.body ], [ %0, %entry ]
+    %sh.0 = phi i32 [ 256, %entry ], [ %shr, %do.body ]
+    %shr = lshr i32 %sh.0, 1
+    %cgep5 = getelementptr inbounds [255 x ptr], ptr %cgep36, i32 0, i32 %shr
+    store ptr %lsr.iv1, ptr %cgep5, align 4, !tbaa !5
+    %lsr.iv.next = add nsw i32 %lsr.iv, 4
+    %cmp1 = icmp samesign ult i32 %lsr.iv.next, 1073741836
+    %cgep4 = getelementptr i8, ptr %lsr.iv1, i32 32
+    br i1 %cmp1, label %do.body, label %do.end, !llvm.loop !9
+
+  do.end:                                           ; preds = %do.body
+    ret void
+  }
+
+  ; Function Attrs: minsize nofree norecurse nosync nounwind optsize memory(write, argmem: none, inaccessiblemem: none)
+  define dso_local void @g(i32 noundef %m) local_unnamed_addr #0 {
+  entry:
+    %0 = add i32 %m, -4
+    %1 = shl i32 %m, 3
+    %cgep = getelementptr i8, ptr @a, i32 %1
+    %cgep36 = bitcast ptr @a to ptr
+    br label %do.body
+
+  do.body:                                          ; preds = %do.body, %entry
+    %lsr.iv1 = phi ptr [ %cgep4, %do.body ], [ %cgep, %entry ]
+    %lsr.iv = phi i32 [ %lsr.iv.next, %do.body ], [ %0, %entry ]
+    %sh.0 = phi i32 [ 256, %entry ], [ %shr, %do.body ]
+    %shr = lshr i32 %sh.0, 1
+    %cgep5 = getelementptr inbounds [255 x ptr], ptr %cgep36, i32 0, i32 %shr
+    store ptr %lsr.iv1, ptr %cgep5, align 4, !tbaa !5
+    %lsr.iv.next = add i32 %lsr.iv, 4
+    %cmp = icmp slt i32 %lsr.iv.next, 1073741836
+    %cgep4 = getelementptr i8, ptr %lsr.iv1, i32 32
+    br i1 %cmp, label %do.body, label %do.end, !llvm.loop !11
+
+  do.end:                                           ; preds = %do.body
+    ret void
+  }
+
+  ; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+  declare i32 @llvm.smax.i32(i32, i32) #1
+
+  !llvm.module.flags = !{!0, !1, !2, !3}
+  !0 = !{i32 1, !"wchar_size", i32 4}
+  !1 = !{i32 8, !"PIC Level", i32 2}
+  !2 = !{i32 7, !"PIE Level", i32 2}
+  !3 = !{i32 7, !"frame-pointer", i32 2}
+  !5 = !{!6, !6, i64 0}
+  !6 = !{!"any pointer", !7, i64 0}
+  !7 = !{!"omnipotent char", !8, i64 0}
+  !8 = !{!"Simple C/C++ TBAA"}
+  !9 = distinct !{!9, !10}
+  !10 = !{!"llvm.loop.mustprogress"}
+  !11 = distinct !{!11, !10}
+
+...
+---
+name:            f
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+failedISel:      false
+tracksRegLiveness: true
+hasWinCFI:       false
+noPhis:          false
+isSSA:           true
+noVRegs:         false
+hasFakeUses:     false
+callsEHReturn:   false
+callsUnwindInit: false
+hasEHScopes:     false
+hasEHFunclets:   false
+isOutlined:      false
+debugInstrRef:   false
+failsVerification: false
+tracksDebugUserValues: false
+registers:
+  - { id: 0, class: intregs, preferred-register: '', flags: [  ] }
+  - { id: 1, class: intregs, preferred-register: '', flags: [  ] }
+  - { id: 2, class: intregs, preferred-register: '', flags: [  ] }
+  - { id: 3, class: intregs, preferred-register: '', flags: [  ] }
+  - { id: 4, class: intregs, preferred-register: '', flags: [  ] }
+  - { id: 5, class: intregs, preferred-register: '', flags: [  ] }
+  - { id: 6, class: intregs, preferred-register: '', flags: [  ] }
+  - { id: 7, class: intregs, preferred-register: '', flags: [  ] }
+  - { id: 8, class: intregs, preferred-register: '', flags: [  ] }
+  - { id: 9, class: intregs, preferred-register: '', flags: [  ] }
+  - { id: 10, class: intregs, preferred-register: '', flags: [  ] }
+  - { id: 11, class: intregs, preferred-register: '', flags: [  ] }
+  - { id: 12, class: intregs, preferred-register: '', flags: [  ] }
+  - { id: 13, class: predregs, preferred-register: '', flags: [  ] }
+  - { id: 14, class: predregs, preferred-register: '', flags: [  ] }
+  - { id: 15, class: intregs, preferred-register: '', flags: [  ] }
+liveins:
+  - { reg: '$r0', virtual-reg: '%9' }
+frameInfo:
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    1
+  adjustsStack:    false
+  hasCalls:        false
+  stackProtector:  ''
+  functionContext: ''
+  maxCallFrameSize: 4294967295
+  cvBytesOfCalleeSavedRegisters: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+  hasTailCall:     false
+  isCalleeSavedInfoValid: false
+  localFrameSize:  0
+  savePoint:       ''
+  restorePoint:    ''
+fixedStack:      []
+stack:           []
+entry_values:    []
+callSites:       []
+debugValueSubstitutions: []
+constants:       []
+machineFunctionInfo: {}
+body:             |
+  bb.0.entry:
+    successors: %bb.1(0x80000000)
+    liveins: $r0
+
+    %9:intregs = COPY $r0
+    %11:intregs = A2_tfrsi 2
+    %12:intregs = A2_max %9, %11
+    %0:intregs = nsw A2_addi %12, -4
+    %1:intregs = S4_addi_asl_ri @a, %12, 3
+    %2:intregs = A2_tfrsi @a
+    %10:intregs = A2_tfrsi 256
+
+  bb.1.do.body:
+    successors: %bb.1(0x7c000000), %bb.2(0x04000000)
+
+    %3:intregs = PHI %1, %bb.0, %8, %bb.1
+    %4:intregs = PHI %0, %bb.0, %7, %bb.1
+    %5:intregs = PHI %10, %bb.0, %15, %bb.1
+    %15:intregs = S2_extractu %5, 8, 1
+    S4_storeri_rr %2, %15, 2, %3 :: (store (s32) into %ir.cgep5, !tbaa !5)
+    %7:intregs = nsw A2_addi %4, 4
+    %13:predregs = C2_cmpgtui %7, 1073741835
+    %8:intregs = A2_addi %3, 32
+    J2_jumpf %13, %bb.1, implicit-def dead $pc
+    J2_jump %bb.2, implicit-def dead $pc
+
+  bb.2.do.end:
+    PS_jmpret $r31, implicit-def dead $pc
+
+...
+---
+name:            g
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+failedISel:      false
+tracksRegLiveness: true
+hasWinCFI:       false
+noPhis:          false
+isSSA:           true
+noVRegs:         false
+hasFakeUses:     false
+callsEHReturn:   false
+callsUnwindInit: false
+hasEHScopes:     false
+hasEHFunclets:   false
+isOutlined:      false
+debugInstrRef:   false
+failsVerification: false
+tracksDebugUserValues: false
+registers:
+  - { id: 0, class: intregs, preferred-register: '', flags: [  ] }
+  - { id: 1, class: intregs, preferred-register: '', flags: [  ] }
+  - { id: 2, class: intregs, preferred-register: '', flags: [  ] }
+  - { id: 3, class: intregs, preferred-register: '', flags: [  ] }
+  - { id: 4, class: intregs, preferred-register: '', flags: [  ] }
+  - { id: 5, class: intregs, preferred-register: '', flags: [  ] }
+  - { id: 6, class: intregs, preferred-register: '', flags: [  ] }
+  - { id: 7, class: intregs, preferred-register: '', flags: [  ] }
+  - { id: 8, class: intregs, preferred-register: '', flags: [  ] }
+  - { id: 9, class: intregs, preferred-register: '', flags: [  ] }
+  - { id: 10, class: intregs, preferred-register: '', flags: [  ] }
+  - { id: 11, class: predregs, preferred-register: '', flags: [  ] }
+  - { id: 12, class: predregs, preferred-register: '', flags: [  ] }
+  - { id: 13, class: intregs, preferred-register: '', flags: [  ] }
+liveins:
+  - { reg: '$r0', virtual-reg: '%9' }
+frameInfo:
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    1
+  adjustsStack:    false
+  hasCalls:        false
+  stackProtector:  ''
+  functionContext: ''
+  maxCallFrameSize: 4294967295
+  cvBytesOfCalleeSavedRegisters: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+  hasTailCall:     false
+  isCalleeSavedInfoValid: false
+  localFrameSize:  0
+  savePoint:       ''
+  restorePoint:    ''
+fixedStack:      []
+stack:           []
+entry_values:    []
+callSites:       []
+debugValueSubstitutions: []
+constants:       []
+machineFunctionInfo: {}
+body:             |
+  bb.0.entry:
+    successors: %bb.1(0x80000000)
+    liveins: $r0
+
+    %9:intregs = COPY $r0
+    %0:intregs = A2_addi %9, -4
+    %1:intregs = S4_addi_asl_ri @a, %9, 3
+    %2:intregs = A2_tfrsi @a
+    %10:intregs = A2_tfrsi 256
+
+  bb.1.do.body:
+    successors: %bb.1(0x7c000000), %bb.2(0x04000000)
+
+    %3:intregs = PHI %1, %bb.0, %8, %bb.1
+    %4:intregs = PHI %0, %bb.0, %7, %bb.1
+    %5:intregs = PHI %10, %bb.0, %13, %bb.1
+    %13:intregs = S2_extractu %5, 8, 1
+    S4_storeri_rr %2, %13, 2, %3 :: (store (s32) into %ir.cgep5, !tbaa !5)
+    %7:intregs = A2_addi %4, 4
+    %11:predregs = C2_cmpgti %7, 1073741835
+    %8:intregs = A2_addi %3, 32
+    J2_jumpf %11, %bb.1, implicit-def dead $pc
+    J2_jump %bb.2, implicit-def dead $pc
+
+  bb.2.do.end:
+    PS_jmpret $r31, implicit-def dead $pc
+
+...
diff --git a/llvm/test/CodeGen/Hexagon/swp-phi-start.ll b/llvm/test/CodeGen/Hexagon/swp-phi-start.ll
index 52c258656ec22..6c2b08d83b1c7 100644
--- a/llvm/test/CodeGen/Hexagon/swp-phi-start.ll
+++ b/llvm/test/CodeGen/Hexagon/swp-phi-start.ll
@@ -5,8 +5,9 @@
 ; the same stage.
 
 ; CHECK-DAG: [[REG3:(r[0-9]+)]] = add([[REG1:(r[0-9]+)]],#-1)
-; CHECK-DAG: [[REG2:(r[0-9]+)]] = add([[REG1]],#-1)
-; CHECK-DAG: loop0(.LBB0_[[LOOP:.]],[[REG3]])
+; CHECK-DAG: [[REG2:(r[0-9]+)]] = add([[REG4:(r[0-9]+)]],#-1)
+; CHECK-DAG: loop0(.LBB0_[[LOOP:.]],[[REG2]])
+; CHECK-NOT: = [[REG3]]
 ; CHECK-NOT: = [[REG2]]
 ; CHECK: .LBB0_[[LOOP]]:
 ; CHECK: }{{[ \t]*}}:endloop

@tstellar tstellar moved this from Needs Triage to Needs Review in LLVM Release Status Apr 14, 2025
@androm3da
Copy link
Member

@iajbar, please review.

@github-project-automation github-project-automation bot moved this from Needs Review to Needs Merge in LLVM Release Status Apr 22, 2025
@iajbar
Copy link
Contributor

iajbar commented Apr 22, 2025

LGTM.

Add check to make sure Dist > 0 or Dist < 0 for appropriate cmp cases to
hexagon hardware loops pass. The change modifies the
HexagonHardwareLoops pass to add runtime checks to make sure that
end_value > initial_value for less than comparisons and end_value <
initial_value for greater than comparisons.

Fix for llvm#133241

@androm3da @iajbar PTAL

---------

Co-authored-by: aankit-quic <[email protected]>
(cherry picked from commit da8ce56)
@tstellar tstellar merged commit 89adc2d into llvm:release/20.x Apr 25, 2025
7 of 10 checks passed
@github-project-automation github-project-automation bot moved this from Needs Merge to Done in LLVM Release Status Apr 25, 2025
Copy link

@androm3da (or anyone else). If you would like to add a note about this fix in the release notes (completely optional). Please reply to this comment with a one or two sentence description of the fix. When you are done, please add the release:note label to this PR.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
Development

Successfully merging this pull request may close these issues.

5 participants