Hi,
In the PR we extract non-existent bits/elements from a vector. This is
because we use TYPE_SIZE (vectype) for a boolean vector which returns 8
instead of 4 for RVV's vector (4) <signed-boolean:1>.
The patch uses TYPE_VECTOR_SUBPARTS instead and multiplies its result
with vector_element_bits to get the proper number of elements and size.
Bootstrapped and regtested on x86 and power10. Regtested on riscv64 and
aarch64.
Regards
Robin
PR tree-optimization/123097
gcc/ChangeLog:
* tree-vect-loop.cc (vect_create_epilog_for_reduction):
Calculate vector size by number of elements * bit size per
element.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/rvv/autovec/pr123097-run.c: New test.
* gcc.target/riscv/rvv/autovec/pr123097.c: New test.
---
.../riscv/rvv/autovec/pr123097-run.c | 19 ++++++++++++++++++
.../gcc.target/riscv/rvv/autovec/pr123097.c | 20 +++++++++++++++++++
gcc/tree-vect-loop.cc | 12 +++++------
3 files changed, 45 insertions(+), 6 deletions(-)
create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/autovec/pr123097-run.c
create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/autovec/pr123097.c
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr123097-run.c
b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr123097-run.c
new file mode 100644
index 00000000000..f536c82bcb9
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr123097-run.c
@@ -0,0 +1,19 @@
+/* { dg-do run } */
+/* { dg-require-effective-target riscv_v_ok } */
+/* { dg-options "-O3 -march=rv64gcv -mabi=lp64d" } */
+
+long long a;
+int b[22];
+_Bool c[22];
+
+int main() {
+ for (long f=0; f<10; ++f)
+ b[f] = c[f] = 1;
+ for (int f=0; f<5; f++)
+ for( int g=0; g<5; g++)
+ c[g*2] &= (_Bool)b[2*f];
+ for (long f=0; f<2; ++f)
+ a ^= c[f];
+ if (a != 0)
+ __builtin_abort ();
+}
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr123097.c
b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr123097.c
new file mode 100644
index 00000000000..7f1538940d2
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr123097.c
@@ -0,0 +1,20 @@
+/* { dg-do compile } */
+/* { dg-options "-O3 -march=rv64gcv -mabi=lp64d -fdump-tree-optimized" } */
+
+long long a;
+int b[22];
+_Bool c[22];
+
+int main() {
+ for (long f=0; f<10; ++f)
+ b[f] = c[f] = 1;
+ for (int f=0; f<5; f++)
+ for( int g=0; g<5; g++)
+ c[g*2] &= (_Bool)b[2*f];
+ for (long f=0; f<2; ++f)
+ a ^= c[f];
+ if (a != 0)
+ __builtin_abort ();
+}
+
+/* { dg-final { scan-tree-dump-not "BIT_FIELD_REF <\\{ -1, -1, -1, -1 \\},"
"optimized" } } */
diff --git a/gcc/tree-vect-loop.cc b/gcc/tree-vect-loop.cc
index f9dd88ed824..02ff28d522f 100644
--- a/gcc/tree-vect-loop.cc
+++ b/gcc/tree-vect-loop.cc
@@ -5972,13 +5972,11 @@ vect_create_epilog_for_reduction (loop_vec_info
loop_vinfo,
if (reduce_with_shift && (!slp_reduc || group_size == 1))
{
- tree bitsize = TYPE_SIZE (TREE_TYPE (vectype1));
- int element_bitsize = tree_to_uhwi (bitsize);
+ int element_bitsize = vector_element_bits (vectype1);
/* Enforced by vectorizable_reduction, which disallows SLP reductions
for variable-length vectors and also requires direct target support
for loop reductions. */
- int vec_size_in_bits = tree_to_uhwi (TYPE_SIZE (vectype1));
- int nelements = vec_size_in_bits / element_bitsize;
+ int nelements = TYPE_VECTOR_SUBPARTS (vectype1).to_constant ();
vec_perm_builder sel;
vec_perm_indices indices;
@@ -6019,7 +6017,8 @@ vect_create_epilog_for_reduction (loop_vec_info
loop_vinfo,
"extract scalar result\n");
new_temp = gimple_build (&stmts, BIT_FIELD_REF, TREE_TYPE (vectype1),
- new_temp, bitsize, bitsize_zero_node);
+ new_temp, bitsize_int (element_bitsize),
+ bitsize_zero_node);
new_temp = gimple_convert (&stmts, scalar_type, new_temp);
gsi_insert_seq_before (&exit_gsi, stmts, GSI_SAME_STMT);
scalar_results.safe_push (new_temp);
@@ -6041,8 +6040,9 @@ vect_create_epilog_for_reduction (loop_vec_info
loop_vinfo,
"Reduce using scalar code.\n");
tree compute_type = TREE_TYPE (vectype1);
- unsigned vec_size_in_bits = tree_to_uhwi (TYPE_SIZE (vectype1));
unsigned element_bitsize = vector_element_bits (vectype1);
+ unsigned vec_size_in_bits = element_bitsize
+ * TYPE_VECTOR_SUBPARTS (vectype1).to_constant ();
tree bitsize = bitsize_int (element_bitsize);
gimple_seq stmts = NULL;
FOR_EACH_VEC_ELT (reduc_inputs, i, vec_temp)
--
2.51.1
