fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-25 13:00:09 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions

intel/dev: Compute pixel pipe information based on geometry topology DRM query.

Browse source 
This changes the intel_device_info calculation to call an additional
DRM query requesting the geometry topology from the kernel, which may
differ from the result of the current topology query on XeHP+
platforms with compute-only and 3D-only DSSes.  This seems more
reliable than the current guesswork done in intel_device_info.c trying
to figure out which DSSes are available for the render CS.

Cc: 22.1 <mesa-stable>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14143>
This commit is contained in:
Francisco Jerez 2021-10-26 17:18:58 -07:00 • committed by Marge Bot
parent bf02bffe15
commit 14cad38b19
1 changed files with 41 additions and 28 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

69
src/intel/dev/intel_device_info.c
View file

@ -1118,7 +1118,7 @@ update_slice_subslice_counts(struct intel_device_info *devinfo)
}
static void
update_pixel_pipes(struct intel_device_info *devinfo)
update_pixel_pipes(struct intel_device_info *devinfo, uint8_t *subslice_masks)
{
if (devinfo->ver < 11)
return;
@ -1147,27 +1147,10 @@ update_pixel_pipes(struct intel_device_info *devinfo)
if (subslice_idx < ARRAY_SIZE(devinfo->subslice_masks))
devinfo->ppipe_subslices[p] =
__builtin_popcount(devinfo->subslice_masks[subslice_idx] & ppipe_mask);
__builtin_popcount(subslice_masks[subslice_idx] & ppipe_mask);
else
devinfo->ppipe_subslices[p] = 0;
}
/* From the "Fusing information" BSpec page regarding DG2 configurations
* where at least a slice has a single pixel pipe fused off:
*
* "Fault disable any 2 DSS in a Gslice and disable that Gslice (incl.
* geom/color/Z)"
*
* XXX - Query geometry topology from hardware once kernel interface is
* available instead of trying to do guesswork here.
*/
if (intel_device_info_is_dg2(devinfo)) {
for (unsigned p = 0; p < INTEL_DEVICE_MAX_PIXEL_PIPES; p++) {
if (devinfo->ppipe_subslices[p] < 2 ||
devinfo->ppipe_subslices[p ^ 1] < 2)
devinfo->ppipe_subslices[p] = 0;
}
}
}
static void
@ -1209,8 +1192,16 @@ update_l3_banks(struct intel_device_info *devinfo)
*/
static void
update_from_single_slice_topology(struct intel_device_info *devinfo,
const struct drm_i915_query_topology_info *topology)
const struct drm_i915_query_topology_info *topology,
const struct drm_i915_query_topology_info *geom_topology)
{
/* An array of bit masks of the subslices available for 3D
* workloads, analogous to intel_device_info::subslice_masks. This
* may differ from the set of enabled subslices on XeHP+ platforms
* with compute-only subslices.
*/
uint8_t geom_subslice_masks[ARRAY_SIZE(devinfo->subslice_masks)] = { 0 };
assert(devinfo->verx10 >= 125);
reset_masks(devinfo);
@ -1229,16 +1220,29 @@ update_from_single_slice_topology(struct intel_device_info *devinfo,
devinfo->eu_subslice_stride = DIV_ROUND_UP(16, 8);
for (uint32_t ss_idx = 0; ss_idx < topology->max_subslices; ss_idx++) {
const uint32_t s = ss_idx / 4;
const uint32_t ss = ss_idx % 4;
/* Determine whether ss_idx is enabled (ss_idx_available) and
* available for 3D workloads (geom_ss_idx_available), which may
* differ on XeHP+ if ss_idx is a compute-only DSS.
*/
const bool ss_idx_available =
(topology->data[topology->subslice_offset + ss_idx / 8] >>
(ss_idx % 8)) & 1;
const bool geom_ss_idx_available =
(geom_topology->data[geom_topology->subslice_offset + ss_idx / 8] >>
(ss_idx % 8)) & 1;
if (geom_ss_idx_available) {
assert(ss_idx_available);
geom_subslice_masks[s * devinfo->subslice_slice_stride +
ss / 8] |= 1u << (ss % 8);
}
if (!ss_idx_available)
continue;
uint32_t s = ss_idx / 4;
uint32_t ss = ss_idx % 4;
devinfo->max_slices = MAX2(devinfo->max_slices, s + 1);
devinfo->slice_masks |= 1u << s;
@ -1261,7 +1265,7 @@ update_from_single_slice_topology(struct intel_device_info *devinfo,
}
update_slice_subslice_counts(devinfo);
update_pixel_pipes(devinfo);
update_pixel_pipes(devinfo, geom_subslice_masks);
update_l3_banks(devinfo);
}
@ -1299,7 +1303,7 @@ update_from_topology(struct intel_device_info *devinfo,
/* Now that all the masks are in place, update the counts. */
update_slice_subslice_counts(devinfo);
update_pixel_pipes(devinfo);
update_pixel_pipes(devinfo, devinfo->subslice_masks);
update_l3_banks(devinfo);
}
@ -1557,10 +1561,19 @@ query_topology(struct intel_device_info *devinfo, int fd)
if (topo_info == NULL)
return false;
if (devinfo->verx10 >= 125)
update_from_single_slice_topology(devinfo, topo_info);
else
if (devinfo->verx10 >= 125) {
struct drm_i915_query_topology_info *geom_topo_info =
intel_i915_query_alloc(fd, DRM_I915_QUERY_GEOMETRY_SUBSLICES, NULL);
if (geom_topo_info == NULL) {
free(topo_info);
return false;
}
update_from_single_slice_topology(devinfo, topo_info, geom_topo_info);
free(geom_topo_info);
} else {
update_from_topology(devinfo, topo_info);
}
free(topo_info);