[gcc r15-3791] tree-optimization/116791 - Elementwise SLP vectorization

[Richard Biener via Gcc-cvs]

https://gcc.gnu.org/g:723f7b6db841c1a101a2f5b3b6273d8449dae39e

commit r15-3791-g723f7b6db841c1a101a2f5b3b6273d8449dae39e
Author: Richard Biener <rguent...@suse.de>
Date:   Mon Sep 23 10:13:17 2024 +0200

    tree-optimization/116791 - Elementwise SLP vectorization
    
    The following restricts the elementwise SLP vectorization to the
    single-lane case which is the reason I enabled it to avoid regressions
    with non-SLP.  The PR shows that multi-line SLP loads with elementwise
    accesses require work, I'll open a new bug to track this for the
    future.
    
            PR tree-optimization/116791
            * tree-vect-stmts.cc (get_group_load_store_type): Only
            fall back to elementwise access for single-lane SLP, restore
            hard failure mode for other cases.
    
            * gcc.dg/vect/pr116791.c: New testcase.

Diff:
---
 gcc/testsuite/gcc.dg/vect/pr116791.c | 20 ++++++++++++++++++++
 gcc/tree-vect-stmts.cc               | 23 +++++++++++++++++------
 2 files changed, 37 insertions(+), 6 deletions(-)

diff --git a/gcc/testsuite/gcc.dg/vect/pr116791.c 
b/gcc/testsuite/gcc.dg/vect/pr116791.c
new file mode 100644
index 000000000000..d9700a88fccb
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/vect/pr116791.c
@@ -0,0 +1,20 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-mavx2" { target avx2 } } */
+
+struct nine_context {
+  unsigned tex_stage[8][33];
+};
+struct fvec4 {
+  float x[2];
+};
+void f(struct fvec4 *dst, struct nine_context *context)
+{
+  unsigned s;
+  for (s = 0; s < 8; ++s)
+    {
+      float *rgba = &dst[s].x[0];
+      unsigned color = context->tex_stage[s][0];
+      rgba[0] = (float)((color >> 16) & 0xFF) / 0xFF;
+      rgba[1] = (float)((color >> 8) & 0xFF) / 0xFF;
+    }
+}
diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
index b72b54d66687..ad08fbe55110 100644
--- a/gcc/tree-vect-stmts.cc
+++ b/gcc/tree-vect-stmts.cc
@@ -2190,12 +2190,23 @@ get_group_load_store_type (vec_info *vinfo, 
stmt_vec_info stmt_info,
              && single_element_p
              && maybe_gt (group_size, TYPE_VECTOR_SUBPARTS (vectype)))
            {
-             *memory_access_type = VMAT_ELEMENTWISE;
-             if (dump_enabled_p ())
-               dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
-                                "single-element interleaving not supported "
-                                "for not adjacent vector loads, using "
-                                "elementwise access\n");
+             if (SLP_TREE_LANES (slp_node) == 1)
+               {
+                 *memory_access_type = VMAT_ELEMENTWISE;
+                 if (dump_enabled_p ())
+                   dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
+                                    "single-element interleaving not supported 
"
+                                    "for not adjacent vector loads, using "
+                                    "elementwise access\n");
+               }
+             else
+               {
+                 if (dump_enabled_p ())
+                   dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
+                                    "single-element interleaving not supported 
"
+                                    "for not adjacent vector loads\n");
+                 return false;
+               }
            }
        }
     }