[PATCH] RISC-V: Fix incorrect use of vcompress in permutation auto-vectorization

[Juzhe-Zhong]

This patch fixes following FAILs on zvl512b of RV32 system:

FAIL: gcc.target/riscv/rvv/autovec/struct/struct_vect_run-12.c execution test
FAIL: gcc.target/riscv/rvv/autovec/struct/struct_vect_run-9.c execution test

The root cause is that for permutation indice = {0,3,7,0} use vcompress 
optimization
which is incorrect. Fix vcompress optimization bug.

        PR target/112598

gcc/ChangeLog:

        * config/riscv/riscv-v.cc (shuffle_compress_patterns): Fix vcompress 
bug.

gcc/testsuite/ChangeLog:

        * gcc.target/riscv/rvv/autovec/pr112598-3.c: New test.

---
 gcc/config/riscv/riscv-v.cc                   | 15 ++++++-------
 .../gcc.target/riscv/rvv/autovec/pr112598-3.c | 21 +++++++++++++++++++
 2 files changed, 29 insertions(+), 7 deletions(-)
 create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/autovec/pr112598-3.c

diff --git a/gcc/config/riscv/riscv-v.cc b/gcc/config/riscv/riscv-v.cc
index 7d6d0821d87..7d3e8038dab 100644
--- a/gcc/config/riscv/riscv-v.cc
+++ b/gcc/config/riscv/riscv-v.cc
@@ -3005,14 +3005,15 @@ shuffle_compress_patterns (struct expand_vec_perm_d *d)
   if (compress_point < 0)
     return false;
 
-  /* It must be series increasing from compress point.  */
-  if (!d->perm.series_p (compress_point, 1, d->perm[compress_point], 1))
-    return false;
-
   /* We can only apply compress approach when all index values from 0 to
      compress point are increasing.  */
   for (int i = 1; i < compress_point; i++)
-    if (known_le (d->perm[i], d->perm[i - 1]))
+    if (maybe_le (d->perm[i], d->perm[i - 1]))
+      return false;
+
+  /* It must be series increasing from compress point.  */
+  for (int i = 1 + compress_point; i < vlen; i++)
+    if (maybe_ne (d->perm[i], d->perm[i - 1] + 1))
       return false;
 
   /* Success!  */
@@ -3080,10 +3081,10 @@ shuffle_compress_patterns (struct expand_vec_perm_d *d)
   if (need_slideup_p)
     {
       int slideup_cnt = vlen - (d->perm[vlen - 1].to_constant () % vlen) - 1;
-      rtx ops[] = {d->target, d->op1, gen_int_mode (slideup_cnt, Pmode)};
+      merge = gen_reg_rtx (vmode);
+      rtx ops[] = {merge, d->op1, gen_int_mode (slideup_cnt, Pmode)};
       insn_code icode = code_for_pred_slide (UNSPEC_VSLIDEUP, vmode);
       emit_vlmax_insn (icode, BINARY_OP, ops);
-      merge = d->target;
     }
 
   insn_code icode = code_for_pred_compress (vmode);
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr112598-3.c 
b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr112598-3.c
new file mode 100644
index 00000000000..231a068c680
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr112598-3.c
@@ -0,0 +1,21 @@
+/* { dg-do compile } */
+/* { dg-options "-march=rv32gcv_zvfh_zfh_zvl512b -mabi=ilp32d -O3 
-ftree-vectorize -std=c99 -fno-vect-cost-model" } */
+
+#include <stdint-gcc.h>
+#define TYPE uint64_t
+#define ITYPE int64_t
+
+void __attribute__ ((noinline, noclone))
+foo (TYPE *__restrict a, TYPE *__restrict b, TYPE *__restrict c,
+    TYPE *__restrict d, ITYPE n)
+{
+  for (ITYPE i = 0; i < n; ++i)
+    {
+      d[i * 3] = a[i];
+      d[i * 3 + 1] = b[i];
+      d[i * 3 + 2] = c[i];
+    }
+}
+
+/* We don't want vcompress.vv.  */
+/* { dg-final { scan-assembler-not {vcompress\.vv} } } */
-- 
2.36.3