[igt-dev] [PATCH i-g-t v2] lib/igt_draw: Change MOCS settings for MTL

Lucas De Marchi lucas.demarchi at intel.com
Tue Dec 20 14:25:19 UTC 2022


On Tue, Dec 20, 2022 at 07:53:18AM +0100, Zbigniew Kempczyński wrote:
>On Mon, Dec 19, 2022 at 06:58:40AM -0800, Lucas De Marchi wrote:
>> On Mon, Dec 19, 2022 at 02:41:54PM +0530, Jeevan B wrote:
>> > On MTL, we want the table entry labelled "UC (GO:Mem)"
>> > which has index 5.  This means that the MOCS value is 10.
>> >
>> > v2: Add define for MOCS settings. (Lucas)
>> >
>> > Signed-off-by: Jeevan B <jeevan.b at intel.com>
>> > ---
>> > lib/igt_draw.c | 12 +++++++++---
>> > 1 file changed, 9 insertions(+), 3 deletions(-)
>> >
>> > diff --git a/lib/igt_draw.c b/lib/igt_draw.c
>> > index 975d65cd..4e19e42a 100644
>> > --- a/lib/igt_draw.c
>> > +++ b/lib/igt_draw.c
>> > @@ -651,13 +651,16 @@ static struct intel_buf *create_buf(int fd, struct buf_ops *bops,
>> > 	return buf;
>> > }
>> >
>> > +#define DG2_MOCS_UC_GO_MEM	2	/* MOCS index 0x1 */
>> > +#define MTL_MOCS_UC_GO_MEM	10	/* MOCS index 0x5 */
>> > +
>> > static void draw_rect_blt(int fd, struct cmd_data *cmd_data,
>> > 			  struct buf_data *buf, struct rect *rect,
>> > 			  uint32_t tiling, uint32_t color)
>> > {
>> > 	struct intel_bb *ibb;
>> > 	struct intel_buf *dst;
>> > -	int blt_cmd_len, blt_cmd_tiling, blt_cmd_depth;
>> > +	int blt_cmd_len, blt_cmd_tiling, blt_cmd_depth, mocs_val;
>> > 	uint32_t devid = intel_get_drm_devid(fd);
>> > 	int gen = intel_gen(devid);
>> > 	int pitch;
>> > @@ -702,8 +705,11 @@ static void draw_rect_blt(int fd, struct cmd_data *cmd_data,
>> > 		pitch = tiling ? buf->stride / 4 : buf->stride;
>> >
>> > 		intel_bb_out(ibb, XY_FAST_COLOR_BLT | blt_cmd_depth);
>> > -		/* DG2 MOCS entry 2 is "UC - Non-Coherent; GO:Memory" */
>> > -		intel_bb_out(ibb, blt_cmd_tiling | 2 << 21 | (pitch-1));
>> > +		if (IS_METEORLAKE(intel_get_drm_devid(fd)))
>> > +			mocs_val = MTL_MOCS_UC_GO_MEM;
>> > +		else
>> > +			mocs_val = DG2_MOCS_UC_GO_MEM;
>>
>> nit: doing this before the calls to intel_bb_out() would be better since
>> it's what is done for all the other fields (e.g. pitch, bld_cmd_depth,
>> etc). With that
>> 	
>> 	Reviewed-by: Lucas De Marchi <lucas.demarchi at intel.com>
>
>I think better would be to extend intel_mocs.c and add appropriate
>function instead of cascaded if's.

indeed. I missed that file. Extending get_mocs_index() and then here
simply callying intel_get_uc_mocs() would be better.

While at it, reordering the if/else ladder in get_mocs_index() to follow
the convention of last platform first.

Lucas De Marchi

>
>--
>Zbigniew
>
>>
>> Lucas De Marchi
>>
>> > +		intel_bb_out(ibb, blt_cmd_tiling | mocs_val << 21 | (pitch-1));
>> > 		intel_bb_out(ibb, (rect->y << 16) | rect->x);
>> > 		intel_bb_out(ibb, ((rect->y + rect->h) << 16) | (rect->x + rect->w));
>> > 		intel_bb_emit_reloc_fenced(ibb, dst->handle, 0,
>> > --
>> > 2.36.0
>> >


More information about the igt-dev mailing list