This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: preliminary patch for data prefetch support
- To: Janis Johnson <janis187 at us dot ibm dot com>
- Subject: Re: preliminary patch for data prefetch support
- From: Richard Henderson <rth at redhat dot com>
- Date: Wed, 7 Nov 2001 00:12:42 -0800
- Cc: gcc-patches at gcc dot gnu dot org
- References: <20011106153540.B7156@us.ibm.com>
On Tue, Nov 06, 2001 at 03:35:40PM -0800, Janis Johnson wrote:
> +/* Memory prefetch, with attributes supported on some machines.
> + Operand 1 is the address of the memory to fetch.
> + Operand 2 is 1 for a write access, 0 otherwise.
> + Operand 3 is the level of temporal locality; 0 means there is no
> + temporal locality and 1, 2, and 3 are for increasing levels of temporal
> + locality.
> +
> + The attributes specified by operands 2 and 3 are ignored for targets
> + whose prefetch instructions do not support them. */
> +
> +DEF_RTL_EXPR(PREFETCH, "prefetch", "eee", 'x')
If op2 and op3 must be constants, and not particularly manipulatable
by normal rtl jiggery pokery, why not make the format "eii"?
Are you considering the semi-related operation of allocating a dirty
cache line? I.e. clobber the entire cache line, avoiding the read
fetch?
r~