[llvm] 1ffc369 - [RISCV] Add a test showing an incorrect vsetvli insertion

[Craig Topper via llvm-commits]

Author: Fraser Cormack
Date: 2021-07-23T09:27:06-07:00
New Revision: 1ffc3693949ce4ea6e2159ecb7d0d7386258e21f

URL: https://github.com/llvm/llvm-project/commit/1ffc3693949ce4ea6e2159ecb7d0d7386258e21f
DIFF: https://github.com/llvm/llvm-project/commit/1ffc3693949ce4ea6e2159ecb7d0d7386258e21f.diff

LOG: [RISCV] Add a test showing an incorrect vsetvli insertion

This patch adds a reduced test case which identifies an illegal vsetvli
inserted by the compiler. The compiler emits a vsetvli which is intended
to preserve VL with the SEW/LMUL ratio e32/m1 when in fact the VL could
have been set by e64/m2 in a predecessor block.

Differential Revision: https://reviews.llvm.org/D106286

Added: 
    llvm/test/CodeGen/RISCV/rvv/vsetvli-regression.ll

Modified: 
    

Removed: 
    


################################################################################
diff  --git a/llvm/test/CodeGen/RISCV/rvv/vsetvli-regression.ll b/llvm/test/CodeGen/RISCV/rvv/vsetvli-regression.ll
new file mode 100644
index 000000000000..dd760f3e6aba
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/rvv/vsetvli-regression.ll
@@ -0,0 +1,22 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=riscv64 -mattr=+experimental-v | FileCheck %s
+
+; This test checks a regression in the vsetvli insertion pass. We used to
+; prserve the VL on the second vsetvli with ratio e32/m1, when the the last
+; update of VL was the vsetvli with e64/m4. Changing VTYPE here changes VLMAX
+; which may make the original VL invalid. Instead of preserving it we use 0.
+
+define i32 @illegal_preserve_vl(<vscale x 2 x i32> %a, <vscale x 4 x i64> %x, <vscale x 4 x i64>* %y) {
+; CHECK-LABEL: illegal_preserve_vl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli a1, zero, e64, m4, ta, mu
+; CHECK-NEXT:    vadd.vv v28, v12, v12
+; CHECK-NEXT:    vs4r.v v28, (a0)
+; CHECK-NEXT:    vsetivli zero, 0, e32, m1, ta, mu
+; CHECK-NEXT:    vmv.x.s a0, v8
+; CHECK-NEXT:    ret
+  %index = add <vscale x 4 x i64> %x, %x
+  store <vscale x 4 x i64> %index, <vscale x 4 x i64>* %y
+  %elt = extractelement <vscale x 2 x i32> %a, i64 0
+  ret i32 %elt
+}