[RISCV] Precommit test to show wrong way to pass scalable FP vector on stack

[kito-cheng]

Summary

Test case to demo scaleable vector on stack will cause stack corruption.

Detail explan what happened:

tail call void @foo(i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0,
                    <vscale x 16 x float> zeroinitializer,
                    <vscale x 16 x float> zeroinitializer,
                    <vscale x 16 x float> zeroinitializer)

foo take 13 arguments, first 8 argument pass in GPR, and next 2 LMUL 8 vector
arguments passed in v8-v23, and now we run out of argument register for GPR and
vector register, so we must pass last LMUL 8 vector argument by stack.

However LLVM only reserve 8 byte on stack for the LMUL 8 vector
argument, it will cause stack corruption when we try to store that into
stack.

[kito-cheng]

Add comments for easier understand what's wrong for the current code gen.

llvm/test/CodeGen/RISCV/calling-conv-vector-on-stack.ll

3

-treat-scalable-fixed-error-as-warning to prevent llvm crash

5

This should not happened, will turn into CHECK-NOT in follow up patch.

10

Stack only reserve 16 byte, 8 byte for backup sp...which mean only left 8 byte for a LMUL 8 value.

12

Set vl to vlmax here

14

And then store an LMUL 8 value to stack, stack corruption here.

[craig.topper]

llvm/test/CodeGen/RISCV/calling-conv-vector-on-stack.ll

2

This wasn't fully generated by the script. You have a custom line for the warning. I think the script would delete that? I'd remove this comment.

[kito-cheng]

Chagnes:

• Drop UTC_ARGS: --version 2

[kito-cheng]

llvm/test/CodeGen/RISCV/calling-conv-vector-on-stack.ll

2

You mean the ; CHECK: warning: Invalid size request on a scalable vector? the file other than the check is generated by update_llc_test_checks.py, and it will preserved :)

[craig.topper]

llvm/test/CodeGen/RISCV/calling-conv-vector-on-stack.ll

2

I meant the comment about being generated by the script so that people won't naively run the script. But if the script doesn't delete the warning check then its fine. I just the script was pretty aggressive about deleting things it doesn't know about.

[craig.topper]

LGTM

llvm/test/CodeGen/RISCV/calling-conv-vector-on-stack.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv64 -mattr=+v -verify-machineinstrs -treat-scalable-fixed-error-as-warning < %s 2>&1 | FileCheck %s
+
+; CHECK: warning: Invalid size request on a scalable vector
+
+define void @bar() nounwind {
+; CHECK-LABEL: bar:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    addi sp, sp, -16
+; CHECK-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    vsetvli a0, zero, e32, m8, ta, ma
+; CHECK-NEXT:    vmv.v.i v8, 0
+; CHECK-NEXT:    vs8r.v v8, (sp)
+; CHECK-NEXT:    li a0, 0
+; CHECK-NEXT:    li a1, 0
+; CHECK-NEXT:    li a2, 0
+; CHECK-NEXT:    li a3, 0
+; CHECK-NEXT:    li a4, 0
+; CHECK-NEXT:    li a5, 0
+; CHECK-NEXT:    li a6, 0
+; CHECK-NEXT:    li a7, 0
+; CHECK-NEXT:    vmv.v.i v16, 0
+; CHECK-NEXT:    call foo@plt
+; CHECK-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    addi sp, sp, 16
+; CHECK-NEXT:    ret
+entry:
+  tail call void @foo(i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, <vscale x 16 x float> zeroinitializer, <vscale x 16 x float> zeroinitializer, <vscale x 16 x float> zeroinitializer)
+  ret void
+}
+
+declare void @foo(i32, i32, i32, i32, i32, i32, i32, i32, <vscale x 16 x float>, <vscale x 16 x float>, <vscale x 16 x float>) nounwind