[RISCV] Copy call frame size when splitting basic block in emitSelectPseudo.

[topperc]

Fixes #97304.

[llvmbot]

@llvm/pr-subscribers-backend-risc-v

Author: Craig Topper (topperc)

Changes

Fixes #97304.

---

Full diff: https://github.com/llvm/llvm-project/pull/99823.diff

2 Files Affected:

• (modified) llvm/lib/Target/RISCV/RISCVISelLowering.cpp (+5)
• (added) llvm/test/CodeGen/RISCV/pr97304.ll (+51)

diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index e938454b8e642..585d8a060fe94 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -18478,6 +18478,11 @@ static MachineBasicBlock *emitSelectPseudo(MachineInstr &MI,
   F->insert(I, IfFalseMBB);
   F->insert(I, TailMBB);
 
+  // Set the call frame size on entry to the new basic blocks.
+  unsigned CallFrameSize = TII.getCallFrameSizeAt(MI);
+  IfFalseMBB->setCallFrameSize(CallFrameSize);
+  TailMBB->setCallFrameSize(CallFrameSize);
+
   // Transfer debug instructions associated with the selects to TailMBB.
   for (MachineInstr *DebugInstr : SelectDebugValues) {
     TailMBB->push_back(DebugInstr->removeFromParent());
diff --git a/llvm/test/CodeGen/RISCV/pr97304.ll b/llvm/test/CodeGen/RISCV/pr97304.ll
new file mode 100644
index 0000000000000..120a0e787384d
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/pr97304.ll
@@ -0,0 +1,51 @@
+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
+; RUN: llc < %s -mtriple=riscv64 -verify-machineinstrs -stop-after=finalize-isel | FileCheck %s
+
+define i32 @_ZNK2cv12LMSolverImpl3runERKNS_17_InputOutputArrayE(i1 %cmp436) {
+  ; CHECK-LABEL: name: _ZNK2cv12LMSolverImpl3runERKNS_17_InputOutputArrayE
+  ; CHECK: bb.0.entry:
+  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
+  ; CHECK-NEXT:   liveins: $x10
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:gpr = COPY $x10
+  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:gpr = COPY [[COPY]]
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1.for.cond:
+  ; CHECK-NEXT:   successors: %bb.2(0x40000000), %bb.3(0x40000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[ANDI:%[0-9]+]]:gpr = ANDI [[COPY1]], 1
+  ; CHECK-NEXT:   ADJCALLSTACKDOWN 8, 0, implicit-def dead $x2, implicit $x2
+  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:gpr = COPY $x2
+  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:gprjalr = COPY $x0
+  ; CHECK-NEXT:   SD [[COPY3]], [[COPY2]], 0 :: (store (s64))
+  ; CHECK-NEXT:   [[ADDI:%[0-9]+]]:gpr = ADDI $x0, 1
+  ; CHECK-NEXT:   [[ADDI1:%[0-9]+]]:gpr = ADDI $x0, 32
+  ; CHECK-NEXT:   BNE [[ANDI]], $x0, %bb.3
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2.for.cond (call-frame-size 8):
+  ; CHECK-NEXT:   successors: %bb.3(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.3.for.cond (call-frame-size 8):
+  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[PHI:%[0-9]+]]:gpr = PHI [[ADDI1]], %bb.1, [[ADDI]], %bb.2
+  ; CHECK-NEXT:   $x10 = COPY [[COPY3]]
+  ; CHECK-NEXT:   $x11 = COPY [[PHI]]
+  ; CHECK-NEXT:   $x12 = COPY [[COPY3]]
+  ; CHECK-NEXT:   $x13 = COPY [[COPY3]]
+  ; CHECK-NEXT:   $x14 = COPY [[COPY3]]
+  ; CHECK-NEXT:   $x15 = COPY [[COPY3]]
+  ; CHECK-NEXT:   $x16 = COPY [[COPY3]]
+  ; CHECK-NEXT:   $x17 = COPY [[COPY3]]
+  ; CHECK-NEXT:   PseudoCALLIndirect [[COPY3]], csr_ilp32_lp64, implicit-def dead $x1, implicit $x10, implicit $x11, implicit $x12, implicit $x13, implicit $x14, implicit $x15, implicit $x16, implicit $x17, implicit-def $x2, implicit-def $x10
+  ; CHECK-NEXT:   ADJCALLSTACKUP 8, 0, implicit-def dead $x2, implicit $x2
+  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:gpr = COPY $x10
+  ; CHECK-NEXT:   PseudoBR %bb.1
+entry:
+  br label %for.cond
+
+for.cond:                                         ; preds = %for.cond, %entry
+  %conv = select i1 %cmp436, i32 32, i32 1
+  %call479 = call i32 (ptr, ...) null(ptr null, i32 %conv, i32 0, i32 0, double 0.000000e+00, double 0.000000e+00, double 0.000000e+00, double 0.000000e+00, double 0.000000e+00)
+  br label %for.cond
+}

[preames]

LGTM

I'll note the underlying scheme seems very fragile. This may not be the only case like this we need to identify.

[preames]

I'll note the underlying scheme seems very fragile. This may not be the only case like this we need to identify.

We have several other places in the same file where we manually create MI control flow. Don't we need to update every one of those?

[asb]

I'll note the underlying scheme seems very fragile. This may not be the only case like this we need to identify.

We have several other places in the same file where we manually create MI control flow. Don't we need to update every one of those?

I'm also wondering this - is there any reference on when this is necessary?

Broader questions aside, change LGTM.

[topperc]

I'll note the underlying scheme seems very fragile. This may not be the only case like this we need to identify.

We have several other places in the same file where we manually create MI control flow. Don't we need to update every one of those?

Yes we do. I'll try to work on tests for the other places.