Turn vect_create_addr_base_for_vector_ref offset into a byte offset

This changes the offset in elements for vect_create_addr_base_for_vector_ref
and vect_create_data_ref_ptr to an offset in bytes, easing a following
refactoring.

2021-10-26  Richard Biener  <rguenther@suse.de>

	* tree-vect-data-refs.c (vect_create_addr_base_for_vector_ref):
	Take offset in bytes.
	(vect_create_data_ref_ptr): Likewise.
	* tree-vect-loop-manip.c (get_misalign_in_elems): Multiply
	offset by element size.
	(vect_create_cond_for_align_checks): Likewise.
	* tree-vect-stmts.c (get_negative_load_store_type): Likewise.
	(vectorizable_load): Remove duplicate leftover from merge
	conflict.

3 files changed, 19 insertions, 15 deletions

diff --git a/gcc/tree-vect-data-refs.c b/gcc/tree-vect-data-refs.c
index cbcd4b8..46360c5 100644
--- a/gcc/tree-vect-data-refs.c
+++ b/gcc/tree-vect-data-refs.c
@@ -4753,8 +4753,7 @@ vect_duplicate_ssa_name_ptr_info (tree name, dr_vec_info *dr_info)
 	    if LOOP=i_loop:	&in		(relative to i_loop)
 	    if LOOP=j_loop: 	&in+i*2B	(relative to j_loop)
    BYTE_OFFSET: Optional, defaulted to NULL.  If supplied, it is added to the
-	    initial address.  Unlike OFFSET, which is number of elements to
-	    be added, BYTE_OFFSET is measured in bytes.
+	    initial address.  Both OFFSET and BYTE_OFFSET are measured in bytes.
 
    Output:
    1. Return an SSA_NAME whose value is the address of the memory location of
@@ -4777,7 +4776,6 @@ vect_create_addr_base_for_vector_ref (vec_info *vinfo, stmt_vec_info stmt_info,
   tree dest;
   gimple_seq seq = NULL;
   tree vect_ptr_type;
-  tree step = TYPE_SIZE_UNIT (TREE_TYPE (DR_REF (dr)));
   loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (vinfo);
   innermost_loop_behavior *drb = vect_dr_behavior (vinfo, dr_info);
 
@@ -4801,8 +4799,7 @@ vect_create_addr_base_for_vector_ref (vec_info *vinfo, stmt_vec_info stmt_info,
 
   if (offset)
     {
-      offset = fold_build2 (MULT_EXPR, sizetype,
-			    fold_convert (sizetype, offset), step);
+      offset = fold_convert (sizetype, offset);
       base_offset = fold_build2 (PLUS_EXPR, sizetype,
 				 base_offset, offset);
     }
@@ -4860,8 +4857,8 @@ vect_create_addr_base_for_vector_ref (vec_info *vinfo, stmt_vec_info stmt_info,
    2. AGGR_TYPE: the type of the reference, which should be either a vector
         or an array.
    3. AT_LOOP: the loop where the vector memref is to be created.
-   4. OFFSET (optional): an offset to be added to the initial address accessed
-	by the data-ref in STMT_INFO.
+   4. OFFSET (optional): a byte offset to be added to the initial address
+	accessed by the data-ref in STMT_INFO.
    5. BSI: location where the new stmts are to be placed if there is no loop
    6. ONLY_INIT: indicate if ap is to be updated in the loop, or remain
         pointing to the initial address.
@@ -4885,7 +4882,7 @@ vect_create_addr_base_for_vector_ref (vec_info *vinfo, stmt_vec_info stmt_info,
       if OFFSET is not supplied:
          initial_address = &a[init];
       if OFFSET is supplied:
-         initial_address = &a[init + OFFSET];
+	 initial_address = &a[init] + OFFSET;
       if BYTE_OFFSET is supplied:
 	 initial_address = &a[init] + BYTE_OFFSET;
 
@@ -5031,7 +5028,7 @@ vect_create_data_ref_ptr (vec_info *vinfo, stmt_vec_info stmt_info,
   /* (2) Calculate the initial address of the aggregate-pointer, and set
      the aggregate-pointer to point to it before the loop.  */
 
-  /* Create: (&(base[init_val+offset]+byte_offset) in the loop preheader.  */
+  /* Create: (&(base[init_val]+offset+byte_offset) in the loop preheader.  */
 
   new_temp = vect_create_addr_base_for_vector_ref (vinfo,
 						   stmt_info, &new_stmt_list,
diff --git a/gcc/tree-vect-loop-manip.c b/gcc/tree-vect-loop-manip.c
index 378b102..72d5831 100644
--- a/gcc/tree-vect-loop-manip.c
+++ b/gcc/tree-vect-loop-manip.c
@@ -1625,7 +1625,9 @@ get_misalign_in_elems (gimple **seq, loop_vec_info loop_vinfo)
   bool negative = tree_int_cst_compare (DR_STEP (dr_info->dr),
 					size_zero_node) < 0;
   tree offset = (negative
-		 ? size_int (-TYPE_VECTOR_SUBPARTS (vectype) + 1)
+		 ? size_int ((-TYPE_VECTOR_SUBPARTS (vectype) + 1)
+			     * TREE_INT_CST_LOW
+				 (TYPE_SIZE_UNIT (TREE_TYPE (vectype))))
 		 : size_zero_node);
   tree start_addr = vect_create_addr_base_for_vector_ref (loop_vinfo,
 							  stmt_info, seq,
@@ -3227,7 +3229,9 @@ vect_create_cond_for_align_checks (loop_vec_info loop_vinfo,
       bool negative = tree_int_cst_compare
 	(DR_STEP (STMT_VINFO_DATA_REF (stmt_info)), size_zero_node) < 0;
       tree offset = negative
-	? size_int (-TYPE_VECTOR_SUBPARTS (vectype) + 1) : size_zero_node;
+	? size_int ((-TYPE_VECTOR_SUBPARTS (vectype) + 1)
+		    * TREE_INT_CST_LOW (TYPE_SIZE_UNIT (TREE_TYPE (vectype))))
+	: size_zero_node;
 
       /* create: addr_tmp = (int)(address_of_first_vector) */
       addr_base =
diff --git a/gcc/tree-vect-stmts.c b/gcc/tree-vect-stmts.c
index c28c937..1f56e10 100644
--- a/gcc/tree-vect-stmts.c
+++ b/gcc/tree-vect-stmts.c
@@ -1996,6 +1996,11 @@ get_negative_load_store_type (vec_info *vinfo,
       return VMAT_ELEMENTWISE;
     }
 
+  /* For backward running DRs the first access in vectype actually is
+     N-1 elements before the address of the DR.  */
+  *poffset = ((-TYPE_VECTOR_SUBPARTS (vectype) + 1)
+	      * TREE_INT_CST_LOW (TYPE_SIZE_UNIT (TREE_TYPE (vectype))));
+
   int misalignment = dr_misalignment (dr_info, vectype);
   alignment_support_scheme
     = vect_supportable_dr_alignment (vinfo, dr_info, vectype, misalignment);
@@ -2006,6 +2011,7 @@ get_negative_load_store_type (vec_info *vinfo,
 	dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
 			 "negative step but alignment required.\n");
       return VMAT_ELEMENTWISE;
+      *poffset = 0;
     }
 
   if (vls_type == VLS_STORE_INVARIANT)
@@ -2014,7 +2020,6 @@ get_negative_load_store_type (vec_info *vinfo,
 	dump_printf_loc (MSG_NOTE, vect_location,
 			 "negative step with invariant source;"
 			 " no permute needed.\n");
-      *poffset = -TYPE_VECTOR_SUBPARTS (vectype) + 1;
       return VMAT_CONTIGUOUS_DOWN;
     }
 
@@ -2023,10 +2028,10 @@ get_negative_load_store_type (vec_info *vinfo,
       if (dump_enabled_p ())
 	dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
 			 "negative step and reversing not supported.\n");
+      *poffset = 0;
       return VMAT_ELEMENTWISE;
     }
 
-  *poffset = -TYPE_VECTOR_SUBPARTS (vectype) + 1;
   return VMAT_CONTIGUOUS_REVERSE;
 }
 
@@ -9315,8 +9320,6 @@ vectorizable_load (vec_info *vinfo,
   tree offset = NULL_TREE;
   if (!known_eq (poffset, 0))
     offset = size_int (poffset);
-  if (memory_access_type == VMAT_CONTIGUOUS_REVERSE)
-    offset = size_int (-TYPE_VECTOR_SUBPARTS (vectype) + 1);
 
   tree bump;
   tree vec_offset = NULL_TREE;