[gcc(refs/vendors/riscv/heads/gcc-13-with-riscv-opts)] RISC-V: Fix bug of VLA SLP auto-vectorization

[Jeff Law <law@gcc.gnu.org>]

https://gcc.gnu.org/g:c57a7b4ef97e78da025c01cc7454628bdeb18d97

commit c57a7b4ef97e78da025c01cc7454628bdeb18d97
Author: Juzhe-Zhong <juzhe.zhong@rivai.ai>
Date:   Tue Jun 13 17:30:55 2023 +0800

    RISC-V: Fix bug of VLA SLP auto-vectorization
    
    Sorry for producing bugs in the previous VLA SLP patch.
    
    Consider this following permutation:
    _85 = VEC_PERM_EXPR <{ 99, 17, ... }, { 11, 80, ... }, { 0, POLY_INT_CST [4, 4], 1, POLY_INT_CST [5, 4], 2, POLY_INT_CST [6, 4], ... }>;
    
    The correct result should be:
    _85 = { 99, 11, 17, 80, ... }
    
    However, I did wrong in the previous patch.
    
    Code sequence before this patch:
    
    set mask = { 0, 1, 0, 1, ... }
    set v0 = { 99, 17, 99, 17, ... }
    set v1 = { 11, 80, 11, 80, ... }
    set index = viota (mask) = { 0, 0, 1, 1, 2, 2, ... }
    set result = vrgather_mu (v0, v1, index, mask) = { 99, 11, 99, 80 }
    The result is incorrect.
    
    After this patch:
    
    set mask = { 0, 1, 0, 1, ... }
    set index = viota (mask) = { 0, 0, 1, 1, 2, 2, ... }
    set v0 = vrgather ({ 99, 17, 99, 17, ... }, index) = { 99, 99, 17, 17, ... }
    set v1 = { 11, 80, 11, 80, ... }
    set result = vrgather_mu (v0, v1, index, mask) = { 99, 11, 17, 80 }
    The result is what we expected.
    
    This issue was discovered in the test I appended in this patch with --param=riscv-autovec-lmul=2.
    
    gcc/ChangeLog:
    
            * config/riscv/riscv-v.cc (emit_vlmax_decompress_insn): Fix bug.
            (shuffle_decompress_patterns): Ditto.
    
    gcc/testsuite/ChangeLog:
    
            * gcc.target/riscv/rvv/autovec/partial/slp-12.c: New test.
            * gcc.target/riscv/rvv/autovec/partial/slp_run-12.c: New test.

Diff:
---
 gcc/config/riscv/riscv-v.cc                        |  8 ++---
 .../gcc.target/riscv/rvv/autovec/partial/slp-12.c  | 35 ++++++++++++++++++++++
 .../riscv/rvv/autovec/partial/slp_run-12.c         | 32 ++++++++++++++++++++
 3 files changed, 71 insertions(+), 4 deletions(-)

diff --git a/gcc/config/riscv/riscv-v.cc b/gcc/config/riscv/riscv-v.cc
index fb970344521..34fdb53c140 100644
--- a/gcc/config/riscv/riscv-v.cc
+++ b/gcc/config/riscv/riscv-v.cc
@@ -863,7 +863,7 @@ emit_vlmax_masked_gather_mu_insn (rtx target, rtx op, rtx sel, rtx mask)
      e q r d c b v a  # v11 destination after vrgather using viota.m under mask
 */
 static void
-emit_vlmax_decompress_insn (rtx target, rtx op, rtx mask)
+emit_vlmax_decompress_insn (rtx target, rtx op0, rtx op1, rtx mask)
 {
   machine_mode data_mode = GET_MODE (target);
   machine_mode sel_mode = related_int_vector_mode (data_mode).require ();
@@ -873,7 +873,8 @@ emit_vlmax_decompress_insn (rtx target, rtx op, rtx mask)
   rtx sel = gen_reg_rtx (sel_mode);
   rtx iota_ops[] = {sel, mask};
   emit_vlmax_insn (code_for_pred_iota (sel_mode), RVV_UNOP, iota_ops);
-  emit_vlmax_masked_gather_mu_insn (target, op, sel, mask);
+  emit_vlmax_gather_insn (target, op0, sel);
+  emit_vlmax_masked_gather_mu_insn (target, op1, sel, mask);
 }
 
 /* Emit merge instruction.  */
@@ -2441,8 +2442,7 @@ shuffle_decompress_patterns (struct expand_vec_perm_d *d)
   rtx const_vec = gen_const_vector_dup (sel_mode, 1);
   rtx mask = gen_reg_rtx (mask_mode);
   expand_vec_cmp (mask, EQ, vid_repeat, const_vec);
-  emit_move_insn (d->target, op0);
-  emit_vlmax_decompress_insn (d->target, op1, mask);
+  emit_vlmax_decompress_insn (d->target, op0, op1, mask);
   return true;
 }
 
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/partial/slp-12.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/partial/slp-12.c
new file mode 100644
index 00000000000..03529f4643a
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/partial/slp-12.c
@@ -0,0 +1,35 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-march=rv32gcv -mabi=ilp32d --param riscv-autovec-preference=scalable -fno-vect-cost-model" } */
+
+#include <stdint-gcc.h>
+
+#define VEC_PERM(TYPE)                                                         \
+  TYPE __attribute__ ((noinline, noclone))                                     \
+  vec_slp_##TYPE (TYPE *restrict a, int n)                                     \
+  {                                                                            \
+    for (int i = 0; i < n; ++i)                                                \
+      {                                                                        \
+	a[i * 8] += 99;                                                        \
+	a[i * 8 + 1] += 11;                                                    \
+	a[i * 8 + 2] += 17;                                                    \
+	a[i * 8 + 3] += 80;                                                    \
+	a[i * 8 + 4] += 63;                                                    \
+	a[i * 8 + 5] += 37;                                                    \
+	a[i * 8 + 6] += 24;                                                    \
+	a[i * 8 + 7] += 81;                                                    \
+      }                                                                        \
+  }
+
+#define TEST_ALL(T)                                                            \
+  T (int8_t)                                                                   \
+  T (uint8_t)                                                                  \
+  T (int16_t)                                                                  \
+  T (uint16_t)                                                                 \
+  T (int32_t)                                                                  \
+  T (uint32_t)                                                                 \
+  T (int64_t)                                                                  \
+  T (uint64_t)
+
+TEST_ALL (VEC_PERM)
+/* This testcase is from aarch64 and floating-point operations are removed.
+   TODO: We will add floating-point operations back and make them as common test in the future. */
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/partial/slp_run-12.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/partial/slp_run-12.c
new file mode 100644
index 00000000000..af892ad7ac4
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/partial/slp_run-12.c
@@ -0,0 +1,32 @@
+/* { dg-do run { target { riscv_vector } } } */
+/* { dg-additional-options "--param riscv-autovec-preference=scalable -fno-vect-cost-model" } */
+
+#include "slp-12.c"
+
+#define N (59 * 8)
+
+#define HARNESS(TYPE)						\
+  {								\
+    TYPE a[N], b[8] = { 99, 11, 17, 80, 63, 37, 24, 81 };	\
+    for (unsigned int i = 0; i < N; ++i)			\
+      {								\
+	a[i] = i * 2 + i % 5;					\
+	asm volatile ("" ::: "memory");				\
+      }								\
+    vec_slp_##TYPE (a, N / 8);					\
+    for (unsigned int i = 0; i < N; ++i)			\
+      {								\
+	TYPE orig = i * 2 + i % 5;				\
+	TYPE expected = orig + b[i % 8];			\
+	if (a[i] != expected)					\
+	  __builtin_abort ();					\
+      }								\
+  }
+
+int __attribute__ ((optimize (1)))
+main (void)
+{
+  TEST_ALL (HARNESS)
+}
+/* This testcase is from aarch64 and floating-point operations are removed.
+   TODO: We will add floating-point operations back and make them as common test in the future. */