From: Nirmoy Das <nirmoy.das@intel.com>
To: <intel-xe@lists.freedesktop.org>
Cc: "Himal Prasad Ghimiray" <himal.prasad.ghimiray@intel.com>,
"Matthew Auld" <matthew.auld@intel.com>,
"Matthew Brost" <matthew.brost@intel.com>,
"Thomas Hellström" <thomas.hellstrom@linux.intel.com>,
"Akshata Jahagirdar" <akshata.jahagirdar@intel.com>
Subject: Re: [PATCH] drm/xe/migrate: Parameterize ccs and bo data clear in xe_migrate_clear()
Date: Sat, 10 Aug 2024 00:23:48 +0200 [thread overview]
Message-ID: <ef360a74-72c3-42a9-9553-76f2d165613e@intel.com> (raw)
In-Reply-To: <20240809220347.25330-1-nirmoy.das@intel.com>
Extracting this from https://patchwork.freedesktop.org/series/136277/ as
there is a regression in performance number on drm-tip.
I think it will take some time to find out what is going on. This patch
can be applied independently sending it separately.
Regards,
Nirmoy
On 8/10/2024 12:03 AM, Nirmoy Das wrote:
> Parameterize clearing ccs and bo data in xe_migrate_clear() which higher
> layers can utilize. This patch will be used later on when doing bo data
> clear for igfx as well.
>
> v2: Replace multiple params with flags in xe_migrate_clear (Matt B)
> v3: s/CLEAR_BO_DATA_FLAG_*/XE_MIGRATE_CLEAR_FLAG_* and move to
> xe_migrate.h. other nits(Matt B)
>
> Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
> Cc: Matthew Auld <matthew.auld@intel.com>
> Cc: Matthew Brost <matthew.brost@intel.com>
> Cc: "Thomas Hellström" <thomas.hellstrom@linux.intel.com>
> Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
> Signed-off-by: Akshata Jahagirdar <akshata.jahagirdar@intel.com>
> Reviewed-by: Matthew Auld <matthew.auld@intel.com>
> ---
> drivers/gpu/drm/xe/tests/xe_bo.c | 3 ++-
> drivers/gpu/drm/xe/tests/xe_migrate.c | 12 ++++++++----
> drivers/gpu/drm/xe/xe_bo.c | 12 ++++++++++--
> drivers/gpu/drm/xe/xe_migrate.c | 27 +++++++++++++++++++--------
> drivers/gpu/drm/xe/xe_migrate.h | 7 ++++++-
> 5 files changed, 45 insertions(+), 16 deletions(-)
>
> diff --git a/drivers/gpu/drm/xe/tests/xe_bo.c b/drivers/gpu/drm/xe/tests/xe_bo.c
> index 1768483da1b7..df9fd907edd4 100644
> --- a/drivers/gpu/drm/xe/tests/xe_bo.c
> +++ b/drivers/gpu/drm/xe/tests/xe_bo.c
> @@ -36,7 +36,8 @@ static int ccs_test_migrate(struct xe_tile *tile, struct xe_bo *bo,
>
> /* Optionally clear bo *and* CCS data in VRAM. */
> if (clear) {
> - fence = xe_migrate_clear(tile->migrate, bo, bo->ttm.resource);
> + fence = xe_migrate_clear(tile->migrate, bo, bo->ttm.resource,
> + XE_MIGRATE_CLEAR_FLAG_FULL);
> if (IS_ERR(fence)) {
> KUNIT_FAIL(test, "Failed to submit bo clear.\n");
> return PTR_ERR(fence);
> diff --git a/drivers/gpu/drm/xe/tests/xe_migrate.c b/drivers/gpu/drm/xe/tests/xe_migrate.c
> index 4344a1724029..47ae9d0b8864 100644
> --- a/drivers/gpu/drm/xe/tests/xe_migrate.c
> +++ b/drivers/gpu/drm/xe/tests/xe_migrate.c
> @@ -105,7 +105,8 @@ static void test_copy(struct xe_migrate *m, struct xe_bo *bo,
> }
>
> xe_map_memset(xe, &remote->vmap, 0, 0xd0, remote->size);
> - fence = xe_migrate_clear(m, remote, remote->ttm.resource);
> + fence = xe_migrate_clear(m, remote, remote->ttm.resource,
> + XE_MIGRATE_CLEAR_FLAG_FULL);
> if (!sanity_fence_failed(xe, fence, big ? "Clearing remote big bo" :
> "Clearing remote small bo", test)) {
> retval = xe_map_rd(xe, &remote->vmap, 0, u64);
> @@ -279,7 +280,8 @@ static void xe_migrate_sanity_test(struct xe_migrate *m, struct kunit *test)
> kunit_info(test, "Clearing small buffer object\n");
> xe_map_memset(xe, &tiny->vmap, 0, 0x22, tiny->size);
> expected = 0;
> - fence = xe_migrate_clear(m, tiny, tiny->ttm.resource);
> + fence = xe_migrate_clear(m, tiny, tiny->ttm.resource,
> + XE_MIGRATE_CLEAR_FLAG_FULL);
> if (sanity_fence_failed(xe, fence, "Clearing small bo", test))
> goto out;
>
> @@ -300,7 +302,8 @@ static void xe_migrate_sanity_test(struct xe_migrate *m, struct kunit *test)
> kunit_info(test, "Clearing big buffer object\n");
> xe_map_memset(xe, &big->vmap, 0, 0x11, big->size);
> expected = 0;
> - fence = xe_migrate_clear(m, big, big->ttm.resource);
> + fence = xe_migrate_clear(m, big, big->ttm.resource,
> + XE_MIGRATE_CLEAR_FLAG_FULL);
> if (sanity_fence_failed(xe, fence, "Clearing big bo", test))
> goto out;
>
> @@ -603,7 +606,8 @@ static void test_clear(struct xe_device *xe, struct xe_tile *tile,
>
> kunit_info(test, "Clear vram buffer object\n");
> expected = 0x0000000000000000;
> - fence = xe_migrate_clear(tile->migrate, vram_bo, vram_bo->ttm.resource);
> + fence = xe_migrate_clear(tile->migrate, vram_bo, vram_bo->ttm.resource,
> + XE_MIGRATE_CLEAR_FLAG_FULL);
> if (sanity_fence_failed(xe, fence, "Clear vram_bo", test))
> return;
> dma_fence_put(fence);
> diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
> index 3295bc92d7aa..56a089aa3916 100644
> --- a/drivers/gpu/drm/xe/xe_bo.c
> +++ b/drivers/gpu/drm/xe/xe_bo.c
> @@ -793,8 +793,16 @@ static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict,
> }
> }
> } else {
> - if (move_lacks_source)
> - fence = xe_migrate_clear(migrate, bo, new_mem);
> + if (move_lacks_source) {
> + u32 flags = 0;
> +
> + if (mem_type_is_vram(new_mem->mem_type))
> + flags |= XE_MIGRATE_CLEAR_FLAG_FULL;
> + else if (handle_system_ccs)
> + flags |= XE_MIGRATE_CLEAR_FLAG_CCS_DATA;
> +
> + fence = xe_migrate_clear(migrate, bo, new_mem, flags);
> + }
> else
> fence = xe_migrate_copy(migrate, bo, bo, old_mem,
> new_mem, handle_system_ccs);
> diff --git a/drivers/gpu/drm/xe/xe_migrate.c b/drivers/gpu/drm/xe/xe_migrate.c
> index 6f24aaf58252..a2d0ce3c59bf 100644
> --- a/drivers/gpu/drm/xe/xe_migrate.c
> +++ b/drivers/gpu/drm/xe/xe_migrate.c
> @@ -1037,9 +1037,11 @@ static void emit_clear(struct xe_gt *gt, struct xe_bb *bb, u64 src_ofs,
> * @m: The migration context.
> * @bo: The buffer object @dst is currently bound to.
> * @dst: The dst TTM resource to be cleared.
> + * @clear_flags: flags to specify which data to clear: CCS, BO, or both.
> *
> - * Clear the contents of @dst to zero. On flat CCS devices,
> - * the CCS metadata is cleared to zero as well on VRAM destinations.
> + * Clear the contents of @dst to zero when XE_MIGRATE_CLEAR_FLAG_BO_DATA is set.
> + * On flat CCS devices, the CCS metadata is cleared to zero with XE_MIGRATE_CLEAR_FLAG_CCS_DATA.
> + * Set XE_MIGRATE_CLEAR_FLAG_FULL to clear bo as well as CCS metadata.
> * TODO: Eliminate the @bo argument.
> *
> * Return: Pointer to a dma_fence representing the last clear batch, or
> @@ -1048,18 +1050,27 @@ static void emit_clear(struct xe_gt *gt, struct xe_bb *bb, u64 src_ofs,
> */
> struct dma_fence *xe_migrate_clear(struct xe_migrate *m,
> struct xe_bo *bo,
> - struct ttm_resource *dst)
> + struct ttm_resource *dst,
> + u32 clear_flags)
> {
> bool clear_vram = mem_type_is_vram(dst->mem_type);
> + bool clear_bo_data = XE_MIGRATE_CLEAR_FLAG_BO_DATA & clear_flags;
> + bool clear_ccs = XE_MIGRATE_CLEAR_FLAG_CCS_DATA & clear_flags;
> struct xe_gt *gt = m->tile->primary_gt;
> struct xe_device *xe = gt_to_xe(gt);
> - bool clear_system_ccs = (xe_bo_needs_ccs_pages(bo) && !IS_DGFX(xe)) ? true : false;
> + bool clear_only_system_ccs = false;
> struct dma_fence *fence = NULL;
> u64 size = bo->size;
> struct xe_res_cursor src_it;
> struct ttm_resource *src = dst;
> int err;
>
> + if (WARN_ON(!clear_bo_data && !clear_ccs))
> + return NULL;
> +
> + if (!clear_bo_data && clear_ccs && !IS_DGFX(xe))
> + clear_only_system_ccs = true;
> +
> if (!clear_vram)
> xe_res_first_sg(xe_bo_sg(bo), 0, bo->size, &src_it);
> else
> @@ -1085,7 +1096,7 @@ struct dma_fence *xe_migrate_clear(struct xe_migrate *m,
> batch_size = 2 +
> pte_update_size(m, pte_flags, src, &src_it,
> &clear_L0, &clear_L0_ofs, &clear_L0_pt,
> - clear_system_ccs ? 0 : emit_clear_cmd_len(gt), 0,
> + clear_bo_data ? emit_clear_cmd_len(gt) : 0, 0,
> avail_pts);
>
> if (xe_migrate_needs_ccs_emit(xe))
> @@ -1107,13 +1118,13 @@ struct dma_fence *xe_migrate_clear(struct xe_migrate *m,
> if (clear_vram && xe_migrate_allow_identity(clear_L0, &src_it))
> xe_res_next(&src_it, clear_L0);
> else
> - emit_pte(m, bb, clear_L0_pt, clear_vram, clear_system_ccs,
> + emit_pte(m, bb, clear_L0_pt, clear_vram, clear_only_system_ccs,
> &src_it, clear_L0, dst);
>
> bb->cs[bb->len++] = MI_BATCH_BUFFER_END;
> update_idx = bb->len;
>
> - if (!clear_system_ccs)
> + if (clear_bo_data)
> emit_clear(gt, bb, clear_L0_ofs, clear_L0, XE_PAGE_SIZE, clear_vram);
>
> if (xe_migrate_needs_ccs_emit(xe)) {
> @@ -1172,7 +1183,7 @@ struct dma_fence *xe_migrate_clear(struct xe_migrate *m,
> return ERR_PTR(err);
> }
>
> - if (clear_system_ccs)
> + if (clear_ccs)
> bo->ccs_cleared = true;
>
> return fence;
> diff --git a/drivers/gpu/drm/xe/xe_migrate.h b/drivers/gpu/drm/xe/xe_migrate.h
> index 453e0ecf5034..7929cc2425e8 100644
> --- a/drivers/gpu/drm/xe/xe_migrate.h
> +++ b/drivers/gpu/drm/xe/xe_migrate.h
> @@ -102,9 +102,14 @@ struct dma_fence *xe_migrate_copy(struct xe_migrate *m,
> struct ttm_resource *dst,
> bool copy_only_ccs);
>
> +#define XE_MIGRATE_CLEAR_FLAG_BO_DATA BIT(0)
> +#define XE_MIGRATE_CLEAR_FLAG_CCS_DATA BIT(1)
> +#define XE_MIGRATE_CLEAR_FLAG_FULL (XE_MIGRATE_CLEAR_FLAG_BO_DATA | \
> + XE_MIGRATE_CLEAR_FLAG_CCS_DATA)
> struct dma_fence *xe_migrate_clear(struct xe_migrate *m,
> struct xe_bo *bo,
> - struct ttm_resource *dst);
> + struct ttm_resource *dst,
> + u32 clear_flags);
>
> struct xe_vm *xe_migrate_get_vm(struct xe_migrate *m);
>
next prev parent reply other threads:[~2024-08-09 22:24 UTC|newest]
Thread overview: 30+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-08-09 22:03 [PATCH] drm/xe/migrate: Parameterize ccs and bo data clear in xe_migrate_clear() Nirmoy Das
2024-08-09 22:23 ` Nirmoy Das [this message]
2024-08-12 8:55 ` Thomas Hellström
2024-08-12 9:04 ` Nirmoy Das
2024-08-16 14:25 ` Nirmoy Das
2024-08-09 22:25 ` ✓ CI.Patch_applied: success for " Patchwork
2024-08-09 22:25 ` ✓ CI.checkpatch: " Patchwork
2024-08-09 22:26 ` ✓ CI.KUnit: " Patchwork
2024-08-09 22:38 ` ✓ CI.Build: " Patchwork
2024-08-09 22:40 ` ✓ CI.Hooks: " Patchwork
2024-08-09 22:41 ` ✓ CI.checksparse: " Patchwork
2024-08-09 23:02 ` ✗ CI.BAT: failure " Patchwork
2024-08-10 2:05 ` ✗ CI.FULL: " Patchwork
2024-08-10 19:08 ` ✓ CI.Patch_applied: success for drm/xe/migrate: Parameterize ccs and bo data clear in xe_migrate_clear() (rev2) Patchwork
2024-08-10 19:08 ` ✓ CI.checkpatch: " Patchwork
2024-08-10 19:09 ` ✓ CI.KUnit: " Patchwork
2024-08-10 19:21 ` ✓ CI.Build: " Patchwork
2024-08-10 19:23 ` ✓ CI.Hooks: " Patchwork
2024-08-10 19:25 ` ✓ CI.checksparse: " Patchwork
2024-08-10 19:45 ` ✓ CI.BAT: " Patchwork
2024-08-10 20:41 ` ✗ CI.FULL: failure " Patchwork
2024-08-12 17:32 ` ✓ CI.Patch_applied: success for drm/xe/migrate: Parameterize ccs and bo data clear in xe_migrate_clear() (rev3) Patchwork
2024-08-12 17:32 ` ✓ CI.checkpatch: " Patchwork
2024-08-12 17:33 ` ✓ CI.KUnit: " Patchwork
2024-08-12 17:45 ` ✓ CI.Build: " Patchwork
2024-08-12 17:47 ` ✓ CI.Hooks: " Patchwork
2024-08-12 17:49 ` ✓ CI.checksparse: " Patchwork
2024-08-12 18:09 ` ✓ CI.BAT: " Patchwork
2024-08-12 22:46 ` ✗ CI.FULL: failure " Patchwork
2024-08-13 8:22 ` Nirmoy Das
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=ef360a74-72c3-42a9-9553-76f2d165613e@intel.com \
--to=nirmoy.das@intel.com \
--cc=akshata.jahagirdar@intel.com \
--cc=himal.prasad.ghimiray@intel.com \
--cc=intel-xe@lists.freedesktop.org \
--cc=matthew.auld@intel.com \
--cc=matthew.brost@intel.com \
--cc=thomas.hellstrom@linux.intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox