Hi,
Gathers and scatters get costed as scalar loads/stores. If we don't
include them in the LMUL ratio calculation we're missing the load and
can get ratio mismatches in different loops.
Regtested on rv64gcv_zvl512b. Pushing as obvious.
Regards
Robin
This patch just adds scalar_load and scalar_store.
PR target/127662
gcc/ChangeLog:
* config/riscv/riscv-vector-costs.cc (costs::record_type_sizes):
Add scalar_load and scalar_store.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/rvv/autovec/pr127662.c: New test.
---
gcc/config/riscv/riscv-vector-costs.cc | 7 ++++---
.../gcc.target/riscv/rvv/autovec/pr127662.c | 17 +++++++++++++++++
2 files changed, 21 insertions(+), 3 deletions(-)
create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/autovec/pr127662.c
diff --git a/gcc/config/riscv/riscv-vector-costs.cc
b/gcc/config/riscv/riscv-vector-costs.cc
index b7d22cb505b..924a0330103 100644
--- a/gcc/config/riscv/riscv-vector-costs.cc
+++ b/gcc/config/riscv/riscv-vector-costs.cc
@@ -1559,11 +1559,12 @@ costs::record_type_sizes (vect_cost_for_stmt kind,
stmt_vec_info stmt_info,
if (!m_largest_type || size > tree_to_uhwi (TYPE_SIZE (m_largest_type)))
m_largest_type = type;
- /* ??? What about scalar loads and a vec_construct or similar? */
+ /* We include scalar load and store because strided and gather accesses are
+ costed as scalar loads. */
bool load_p = (kind == vector_load || kind == unaligned_load
- || kind == vector_gather_load);
+ || kind == vector_gather_load || kind == scalar_load);
bool store_p = (kind == vector_store || kind == unaligned_store
- || kind == vector_scatter_store);
+ || kind == vector_scatter_store || kind == scalar_store);
if ((load_p || store_p)
&& (!m_smallest_type
|| size < tree_to_uhwi (TYPE_SIZE (m_smallest_type))))
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr127662.c
b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr127662.c
new file mode 100644
index 00000000000..eac617b360c
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr127662.c
@@ -0,0 +1,17 @@
+/* { dg-do compile } */
+/* { dg-options "-O3 -march=rv64gcv -mabi=lp64d -mrvv-max-lmul=conv-dynamic" }
*/
+
+typedef short a;
+typedef struct { a b, c, d; } e;
+e *f;
+a g;
+long h, i;
+void j (void)
+{
+ for (; i; i++)
+ {
+ h = f[i].c / 0.5;
+ g = 5 * h;
+ f[i].c = g;
+ }
+}
--
2.55.0
