[RISCV] Add a test showing that scalable offsets are not handled.

[mgudim]

Currently we encode scalable offsets in CFIs using cfi_escape. But CFIInstrInserter can't handle cfi_escape.

[llvmbot]

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

Author: Mikhail Gudim (mgudim)

Changes

Currently we encode scalable offsets in CFIs using cfi_escape. But CFIInstrInserter can't handle cfi_escape.

---

Full diff: https://github.com/llvm/llvm-project/pull/164480.diff

1 Files Affected:

• (added) llvm/test/CodeGen/RISCV/cfi-inserter-scalable-offset.mir (+94)

diff --git a/llvm/test/CodeGen/RISCV/cfi-inserter-scalable-offset.mir b/llvm/test/CodeGen/RISCV/cfi-inserter-scalable-offset.mir
new file mode 100644
index 0000000000000..17fde2dcc86dc
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/cfi-inserter-scalable-offset.mir
@@ -0,0 +1,94 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
+# RUN: llc %s -mtriple=riscv64 -mattr=+v \
+# RUN: -run-pass=prologepilog,cfi-instr-inserter  \
+# RUN: -riscv-enable-cfi-instr-inserter=true \
+# RUN: -o - | FileCheck %s
+# XFAIL: *
+#
+# In this test prolog will be inserted in bb.3. We need to save the scalable vector register v1.
+# In bb.3 the new rule for computing CFA will be (sp + 16 + 1 * vlenb) which we encode using cfi_escape. Also, $v1 will be saved at (cfa - 1 * vlenb), which we also encode using cfi_escape.
+# Since the only way to get to bb.2 is from bb.3, the same cfi's should be emitted at beginning of bb.2
+#
+# Currently CFIInstrInserter can't handle escape, so we end up with wrong CFI.
+# NOTE: if llc was compiled with NDEBUG, this will just crash. Otherwise, the output will be as in CHECK lines.
+
+--- |
+
+  define riscv_vector_cc void @test0(ptr %p0, ptr %p1) #0 {
+    entry:
+    %v = load <4 x i32>, ptr %p0, align 16
+    store <4 x i32> %v, ptr %p1, align 16
+    ret void
+  }
+
+  attributes #0 = { "target-features"="+v" }
+
+...
+---
+name:            test0
+tracksRegLiveness: true
+frameInfo:
+  savePoint:
+    - point: '%bb.3'
+  restorePoint:
+    - point: '%bb.2'
+body:             |
+  ; CHECK-LABEL: name: test0
+  ; CHECK: bb.0.entry:
+  ; CHECK-NEXT:   successors: %bb.3(0x40000000), %bb.1(0x40000000)
+  ; CHECK-NEXT:   liveins: $x10, $x11, $v1
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   BEQ $x10, $x0, %bb.3
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1:
+  ; CHECK-NEXT:   liveins: $v1
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   PseudoRET
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2:
+  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   CFI_INSTRUCTION def_cfa_offset 16
+  ; CHECK-NEXT:   $x10 = ADDI $x2, 16
+  ; CHECK-NEXT:   $v1 = frame-destroy VL1RE8_V killed $x10 :: (load (<vscale x 1 x s64>) from %stack.0)
+  ; CHECK-NEXT:   $x10 = frame-destroy PseudoReadVLENB
+  ; CHECK-NEXT:   $x2 = frame-destroy ADD $x2, killed $x10
+  ; CHECK-NEXT:   frame-destroy CFI_INSTRUCTION def_cfa $x2, 16
+  ; CHECK-NEXT:   frame-destroy CFI_INSTRUCTION restore $v1
+  ; CHECK-NEXT:   $x2 = frame-destroy ADDI $x2, 16
+  ; CHECK-NEXT:   frame-destroy CFI_INSTRUCTION def_cfa_offset 0
+  ; CHECK-NEXT:   PseudoBR %bb.1
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.3:
+  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
+  ; CHECK-NEXT:   liveins: $x10, $x11, $v1
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   $x2 = frame-setup ADDI $x2, -16
+  ; CHECK-NEXT:   frame-setup CFI_INSTRUCTION def_cfa_offset 16
+  ; CHECK-NEXT:   $x12 = frame-setup PseudoReadVLENB
+  ; CHECK-NEXT:   $x2 = frame-setup SUB $x2, killed $x12
+  ; CHECK-NEXT:   frame-setup CFI_INSTRUCTION escape 0x0f, 0x0d, 0x72, 0x00, 0x11, 0x10, 0x22, 0x11, 0x01, 0x92, 0xa2, 0x38, 0x00, 0x1e, 0x22
+  ; CHECK-NEXT:   $x12 = ADDI $x2, 16
+  ; CHECK-NEXT:   frame-setup VS1R_V killed $v1, killed $x12 :: (store (<vscale x 1 x s64>) into %stack.0)
+  ; CHECK-NEXT:   frame-setup CFI_INSTRUCTION escape 0x10, 0x61, 0x08, 0x11, 0x7f, 0x92, 0xa2, 0x38, 0x00, 0x1e, 0x22
+  ; CHECK-NEXT:   dead $x0 = PseudoVSETIVLI 4, 208 /* e32, m1, ta, ma */, implicit-def $vl, implicit-def $vtype
+  ; CHECK-NEXT:   renamable $v1 = PseudoVLE32_V_M1 undef renamable $v1, killed renamable $x10, 4, 5 /* e32 */, 2 /* tu, ma */, implicit $vl, implicit $vtype
+  ; CHECK-NEXT:   PseudoVSE32_V_M1 killed renamable $v1, killed renamable $x11, 4, 5 /* e32 */, implicit $vl, implicit $vtype
+  ; CHECK-NEXT:   PseudoBR %bb.2
+  bb.0.entry:
+    liveins: $x10, $x11
+    BEQ $x10, $x0, %bb.3
+
+  bb.1:
+    PseudoRET
+
+  bb.2:
+    PseudoBR %bb.1
+
+  bb.3:
+    liveins: $x10, $x11
+    dead $x0 = PseudoVSETIVLI 4, 208, implicit-def $vl, implicit-def $vtype
+    renamable $v1 = PseudoVLE32_V_M1 undef renamable $v1, killed renamable $x10, 4, 5, 2, implicit $vl, implicit $vtype
+    PseudoVSE32_V_M1 killed renamable $v1, killed renamable $x11, 4, 5, implicit $vl, implicit $vtype
+    PseudoBR %bb.2
+...

[ppenzin]

Would that be a problem? Do people usually run tests in debug mode?