When we got the simplification of bit-field-ref to view-convert
we lost the ability to detect FMAs since we cannot look through
_1 = {_10};
_11 = VIEW_CONVERT_EXPR<float>(_1);
the following amends the (view_convert CONSTRUCTOR) pattern
to handle this case.
Bootstrapped and tested on x86_64-unknown-linux-gnu, pushed.
2022-06-14 Richard Biener <rguent...@suse.de>
PR middle-end/105965
* match.pd (view_convert CONSTRUCTOR): Handle single-element
CTOR case.
* gcc.target/i386/pr105965.c: New testcase.
---
gcc/match.pd | 17 +++++++++++++----
gcc/testsuite/gcc.target/i386/pr105965.c | 12 ++++++++++++
2 files changed, 25 insertions(+), 4 deletions(-)
create mode 100644 gcc/testsuite/gcc.target/i386/pr105965.c
diff --git a/gcc/match.pd b/gcc/match.pd
index 44a385b912d..776c9c6489a 100644
--- a/gcc/match.pd
+++ b/gcc/match.pd
@@ -3669,12 +3669,21 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
&& TYPE_UNSIGNED (TREE_TYPE (@1)))))
(view_convert @1)))
-/* Simplify a view-converted empty constructor. */
+/* Simplify a view-converted empty or single-element constructor. */
(simplify
(view_convert CONSTRUCTOR@0)
- (if (TREE_CODE (@0) != SSA_NAME
- && CONSTRUCTOR_NELTS (@0) == 0)
- { build_zero_cst (type); }))
+ (with
+ { tree ctor = (TREE_CODE (@0) == SSA_NAME
+ ? gimple_assign_rhs1 (SSA_NAME_DEF_STMT (@0)) : @0); }
+ (switch
+ (if (CONSTRUCTOR_NELTS (ctor) == 0)
+ { build_zero_cst (type); })
+ (if (CONSTRUCTOR_NELTS (ctor) == 1
+ && VECTOR_TYPE_P (TREE_TYPE (ctor))
+ && operand_equal_p (TYPE_SIZE (type),
+ TYPE_SIZE (TREE_TYPE
+ (CONSTRUCTOR_ELT (ctor, 0)->value))))
+ (view_convert { CONSTRUCTOR_ELT (ctor, 0)->value; })))))
/* Re-association barriers around constants and other re-association
barriers can be removed. */
diff --git a/gcc/testsuite/gcc.target/i386/pr105965.c
b/gcc/testsuite/gcc.target/i386/pr105965.c
new file mode 100644
index 00000000000..5bb53790de8
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr105965.c
@@ -0,0 +1,12 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -mfma -mfpmath=sse" } */
+
+typedef float v1sf __attribute__((vector_size(4)));
+
+v1sf
+foo43 (v1sf a, v1sf b, v1sf c)
+{
+ return a * b + c;
+}
+
+/* { dg-final { scan-assembler "fmadd" } } */
--
2.35.3
