On Mon, 8 Jun 2026, Monk Chiang wrote:
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/riscv/sifive-7-load-address-bypass.c
> @@ -0,0 +1,23 @@
> +/* Test SiFive-7 load-to-address bypass optimization. */
> +
> +/* { dg-do compile } */
> +/* { dg-options "-mtune=sifive-7-series" } */
> +/* { dg-skip-if "" { *-*-* } { "-O0" "-O1" "-Os" "-Oz" "-Og" "-flto" } } */
> +/* { dg-final { check-function-bodies "**" "" } } */
> +
> +struct node {
> + int value;
> + struct node *next;
> +};
> +
> +/*
> +**load_value:
> +** l[wd] [at][0-9],0\(a0\)
> +** lw a0,0\([at][0-9]\)
> +** ret
> +*/
> +int load_value(struct node **ptr)
> +{
> + struct node *p = *ptr;
> + return p->value;
> +}
Not questioning the pipeline description update itself what is this test
supposed to verify? This code produces the same assembly regardless of
the tuning option used, so under what conditions is it expected to fail?
Maciej
