On Thu, Nov 14, 2019 at 07:16:28PM -0500, Michael Meissner wrote:
> This test tests whether traditional DS and DQ instructions that require the
> bottom 2 bits of the offset to be zero (DS) or the bottom 4 of the offset to
> be
> zero (DQ) properly generate the prefixed form of the instruction instead of
> loading the offset into a GPR and doing an indexed memory operation.
> * gcc.target/powerpc/prefix-odd-memory.c: New test to make sure
> prefixed instructions are generated if an offset would not be
> legal for the non-prefixed DS/DQ instructions.
Same comments as before.
> +/* { dg-require-effective-target powerpc_prefixed_addr_ok } */
What are the plans for getting rid of this test? Do we need it at all
even now, in the tests that use -mcpu=future anyway?
> +/* Tests whether we can generate a prefixed load/store operation for
> addresses
> + that don't meet DS/DQ alignment constraints. */
Test that we *do*, even; and also, test that we *don't* when it isn't
necessary.
And it is about the low bits of the offset on these insns, not about the
alignment.
> + *(float *)(p + 1) = f; /* should generate STF. */
stfs.
> + *(double *)(p + 1) = d; /* should generate STD. */
stfd, instead.
Segher
