Hi Juzhe,
I wasn't yet able to check this locally so just some minor comment nits:
> +/* Return the vectorization machine mode for RVV according to LMUL. */
> +machine_mode
> +preferred_simd_mode (scalar_mode mode)
> +{
> + /* We only enable auto-vectorization when TARGET_MIN_VLEN < 128 &&
> + riscv_autovec_lmul < RVV_M2. Since GCC loop vectorizer report ICE when
> we
> + enable -march=rv64gc_zve32* and -march=rv32gc_zve64*. in the
I believe Kito mentioned this in the last iteration but the comment
here doesn't match the code below. You want >= 128 instead of < 128.
> + /* TODO: Currently, it will produce ICE for --param
> + riscv-autovec-preference=fixed-vlmax. So, we just return NULL_RTX here
> + let GCC genearte loads/stores. Ideally, GCC should either report
> + Warning message to tell user do not use RVV vector type in function
> + arg, or GCC just support function arg calling convention for RVV
> + directly. */
> + if (riscv_v_ext_mode_p (mode))
> + return NULL_RTX;
will produce -> will cause an ICE
genearte -> generate
GCC should either... -> we should either warn the user not to use
an RVV vector type as function argument ... or support the calling
convention....
> +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/fixed-vlmax-1.c
> @@ -0,0 +1,22 @@
> +/* { dg-do compile } */
> +/* { dg-options "-march=rv32gcv -mabi=ilp32 -mpreferred-stack-boundary=3
> -fno-schedule-insns -fno-schedule-insns2 -O3 --param
> riscv-autovec-preference=fixed-vlmax" } */
> +
> +#include "riscv_vector.h"
> +
> +void f (char*);
> +
> +void stach_check_alloca_1 (vuint8m1_t data, uint8_t *base, int y, ...)
> +{
Shouldn't that be stack rather than stach?
> +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/partial/single_rgroup-1.h
> @@ -0,0 +1,106 @@
> +#include <stddef.h>
> +#include <stdint.h>
> +
> +#define N 777
> +
> +#define test_1(TYPE)
> \
> + TYPE a_##TYPE[N];
> \
> + TYPE b_##TYPE[N];
> \
> + void __attribute__ ((noinline, noclone)) test_1_##TYPE (unsigned int n)
> \
Just FYI, you can use ((noipa)) to cover all cases of unwanted "inlining". Not
needed here though.
> diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/v-1.c
> b/gcc/testsuite/gcc.target/riscv/rvv/autovec/v-1.c
> new file mode 100644
> index 00000000000..7ff84f60749
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/v-1.c
> @@ -0,0 +1,4 @@
> +/* { dg-do compile } */
> +/* { dg-options "-march=rv32gcv -mabi=ilp32d --param
> riscv-autovec-preference=scalable -O3 -ftree-vectorize
> -fdump-tree-vect-details -save-temps" } */
> +
> +#include "template-1.h"
I'm a bit wary of these tests not checking anything. Of course we will see
if we ICE or not but that I would expect from a "new feature". Couldn't we
check
something else at least that gives a clue as to what is supposed to happen?
Last time I tried some of those locally, we would not vectorize. In case that's
intended we could check for e.g. "vectorized 0 loops in function". If not, a
comment would still help.
Do we actually need -ftree-vectorize at -O3? In general I would prefer to
split off
common options and set them in rvv.exp already, only giving
dg-additional-options for
each test. Here we don't share too many apart from -O3 -ftree-vectorize so not
yet
necessary.
Regards
Robin
