The following also covers TARGET_MEM_REF when decomposing stores from
CTORs to supported elementwise operations. This avoids spilling
and cleans up after vector lowering which doesn't touch loads or
stores. It also mimics what we already do for loads.
Bootstrapped and tested on x86_64-unknown-linux-gnu, pushed.
PR tree-optimization/64731
* tree-ssa-forwprop.cc (pass_forwprop::execute): Also
handle TARGET_MEM_REF destinations of stores from vector
CTORs.
* gcc.target/i386/pr64731.c: New testcase.
---
gcc/testsuite/gcc.target/i386/pr64731.c | 14 +++++++++
gcc/tree-ssa-forwprop.cc | 41 +++++++++++++++----------
2 files changed, 38 insertions(+), 17 deletions(-)
create mode 100644 gcc/testsuite/gcc.target/i386/pr64731.c
diff --git a/gcc/testsuite/gcc.target/i386/pr64731.c
b/gcc/testsuite/gcc.target/i386/pr64731.c
new file mode 100644
index 00000000000..dea5141ad24
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr64731.c
@@ -0,0 +1,14 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -mno-avx" } */
+
+typedef double double4 __attribute__((vector_size(32)));
+
+void fun(double * a, double * b)
+{
+ for (int i = 0; i < 1024; i+=4)
+ *(double4*)&a[i] += *(double4 *)&b[i];
+}
+
+/* We don't want to spill but have both loads and stores lowered
+ to supported SSE operations. */
+/* { dg-final { scan-assembler-not "movap\[sd\].*\[er\]sp" } } */
diff --git a/gcc/tree-ssa-forwprop.cc b/gcc/tree-ssa-forwprop.cc
index 9dc67b5309c..e63d2ab82c9 100644
--- a/gcc/tree-ssa-forwprop.cc
+++ b/gcc/tree-ssa-forwprop.cc
@@ -3236,6 +3236,26 @@ simplify_vector_constructor (gimple_stmt_iterator *gsi)
return true;
}
+/* Prepare a TARGET_MEM_REF ref so that it can be subsetted as
+ lvalue. This splits out an address computation stmt before *GSI
+ and returns a MEM_REF wrapping the address. */
+
+static tree
+prepare_target_mem_ref_lvalue (tree ref, gimple_stmt_iterator *gsi)
+{
+ if (TREE_CODE (TREE_OPERAND (ref, 0)) == ADDR_EXPR)
+ mark_addressable (TREE_OPERAND (TREE_OPERAND (ref, 0), 0));
+ tree ptrtype = build_pointer_type (TREE_TYPE (ref));
+ tree tem = make_ssa_name (ptrtype);
+ gimple *new_stmt
+ = gimple_build_assign (tem, build1 (ADDR_EXPR, TREE_TYPE (tem),
+ unshare_expr (ref)));
+ gsi_insert_before (gsi, new_stmt, GSI_SAME_STMT);
+ ref = build2_loc (EXPR_LOCATION (ref),
+ MEM_REF, TREE_TYPE (ref), tem,
+ build_int_cst (TREE_TYPE (TREE_OPERAND (ref, 1)), 0));
+ return ref;
+}
/* Rewrite the vector load at *GSI to component-wise loads if the load
is only used in BIT_FIELD_REF extractions with eventual intermediate
@@ -3317,20 +3337,7 @@ optimize_vector_load (gimple_stmt_iterator *gsi)
For TARGET_MEM_REFs we have to separate the LEA from the reference. */
tree load_rhs = rhs;
if (TREE_CODE (load_rhs) == TARGET_MEM_REF)
- {
- if (TREE_CODE (TREE_OPERAND (load_rhs, 0)) == ADDR_EXPR)
- mark_addressable (TREE_OPERAND (TREE_OPERAND (load_rhs, 0), 0));
- tree ptrtype = build_pointer_type (TREE_TYPE (load_rhs));
- tree tem = make_ssa_name (ptrtype);
- gimple *new_stmt
- = gimple_build_assign (tem, build1 (ADDR_EXPR, TREE_TYPE (tem),
- unshare_expr (load_rhs)));
- gsi_insert_before (gsi, new_stmt, GSI_SAME_STMT);
- load_rhs = build2_loc (EXPR_LOCATION (load_rhs),
- MEM_REF, TREE_TYPE (load_rhs), tem,
- build_int_cst
- (TREE_TYPE (TREE_OPERAND (load_rhs, 1)), 0));
- }
+ load_rhs = prepare_target_mem_ref_lvalue (load_rhs, gsi);
/* Rewrite the BIT_FIELD_REFs to be actual loads, re-emitting them at
the place of the original load. */
@@ -3823,9 +3830,7 @@ pass_forwprop::execute (function *fun)
&& gimple_store_p (use_stmt)
&& !gimple_has_volatile_ops (use_stmt)
&& !stmt_can_throw_internal (fun, use_stmt)
- && is_gimple_assign (use_stmt)
- && (TREE_CODE (gimple_assign_lhs (use_stmt))
- != TARGET_MEM_REF))
+ && is_gimple_assign (use_stmt))
{
tree elt_t = TREE_TYPE (CONSTRUCTOR_ELT (rhs, 0)->value);
unsigned HOST_WIDE_INT elt_w
@@ -3835,6 +3840,8 @@ pass_forwprop::execute (function *fun)
tree use_lhs = gimple_assign_lhs (use_stmt);
if (auto_var_p (use_lhs))
DECL_NOT_GIMPLE_REG_P (use_lhs) = 1;
+ else if (TREE_CODE (use_lhs) == TARGET_MEM_REF)
+ use_lhs = prepare_target_mem_ref_lvalue (use_lhs, &gsi);
for (unsigned HOST_WIDE_INT bi = 0; bi < n; bi += elt_w)
{
unsigned HOST_WIDE_INT ci = bi / elt_w;
--
2.35.3
