From: "Grzegorzek, Dominik" <dominik.grzegorzek@intel.com>
To: "igt-dev@lists.freedesktop.org" <igt-dev@lists.freedesktop.org>,
"Manszewski, Christoph" <christoph.manszewski@intel.com>
Cc: "Wilson, Chris P" <chris.p.wilson@intel.com>
Subject: Re: [igt-dev] [PATCH i-g-t 1/2] lib: Apply x/y offsets to GPGPU fill
Date: Wed, 14 Jun 2023 08:24:10 +0000 [thread overview]
Message-ID: <e9bde4300a66ba82dbc9364a5c42f60f1e86b9ea.camel@intel.com> (raw)
In-Reply-To: <20230609093719.2076046-1-christoph.manszewski@intel.com>
On Fri, 2023-06-09 at 11:37 +0200, Christoph Manszewski wrote:
> Actually honour the caller provided destination region and pass the x/y
> offsets to the thread group.
>
> Signed-off-by: Chris Wilson <chris.p.wilson@intel.com>
> Signed-off-by: Christoph Manszewski <christoph.manszewski@intel.com>
>
Aligns behaviour with interface author inntention. From me it is:
Reviewed-by: Dominik Grzegorzek <dominik.grzegorzek@intel.com>
However, when this is finally working, could we utilize it in gem_gpgpu_fill/xe_gpgpu_fill? :)
> ---
> lib/gpgpu_fill.c | 4 +---
> lib/gpu_cmds.c | 29 +++++++++++++++--------------
> lib/gpu_cmds.h | 1 +
> 3 files changed, 17 insertions(+), 17 deletions(-)
>
> diff --git a/lib/gpgpu_fill.c b/lib/gpgpu_fill.c
> index d854fbf7..eed82187 100644
> --- a/lib/gpgpu_fill.c
> +++ b/lib/gpgpu_fill.c
> @@ -315,8 +315,6 @@ __xehp_gpgpu_fillfunc(int i915,
> {
> struct intel_bb *ibb;
> struct xehp_interface_descriptor_data idd;
> - (void) x;
> - (void) y;
>
> ibb = intel_bb_create(i915, PAGE_SIZE);
> intel_bb_add_intel_buf(ibb, buf, true);
> @@ -335,7 +333,7 @@ __xehp_gpgpu_fillfunc(int i915,
> xehp_emit_state_compute_mode(ibb);
> xehp_emit_state_binding_table_pool_alloc(ibb);
> xehp_emit_cfe_state(ibb, THREADS);
> - xehp_emit_compute_walk(ibb, width, height, &idd, color);
> + xehp_emit_compute_walk(ibb, x, y, width, height, &idd, color);
>
> intel_bb_out(ibb, MI_BATCH_BUFFER_END);
> intel_bb_ptr_align(ibb, 32);
> diff --git a/lib/gpu_cmds.c b/lib/gpu_cmds.c
> index 1f321ae4..aecba928 100644
> --- a/lib/gpu_cmds.c
> +++ b/lib/gpu_cmds.c
> @@ -636,10 +636,10 @@ gen7_emit_gpgpu_walk(struct intel_bb *ibb,
> * Then thread group X = width / 16 (aligned to 16)
> * thread group Y = height;
> */
> - x_dim = (width + 15) / 16;
> - y_dim = height;
> + x_dim = (x + width + 15) / 16;
> + y_dim = y + height;
>
> - tmp = width & 15;
> + tmp = (x + width) & 15;
> if (tmp == 0)
> right_mask = (1 << 16) - 1;
> else
> @@ -657,11 +657,11 @@ gen7_emit_gpgpu_walk(struct intel_bb *ibb,
> 0); /* width:1 */
>
> /* thread group X */
> - intel_bb_out(ibb, 0);
> + intel_bb_out(ibb, x / 16);
> intel_bb_out(ibb, x_dim);
>
> /* thread group Y */
> - intel_bb_out(ibb, 0);
> + intel_bb_out(ibb, y);
> intel_bb_out(ibb, y_dim);
>
> /* thread group Z */
> @@ -693,10 +693,10 @@ gen8_emit_gpgpu_walk(struct intel_bb *ibb,
> * Then thread group X = width / 16 (aligned to 16)
> * thread group Y = height;
> */
> - x_dim = (width + 15) / 16;
> - y_dim = height;
> + x_dim = (x + width + 15) / 16;
> + y_dim = y + height;
>
> - tmp = width & 15;
> + tmp = (x + width) & 15;
> if (tmp == 0)
> right_mask = (1 << 16) - 1;
> else
> @@ -715,12 +715,12 @@ gen8_emit_gpgpu_walk(struct intel_bb *ibb,
> 0); /* width:1 */
>
> /* thread group X */
> - intel_bb_out(ibb, 0);
> + intel_bb_out(ibb, x / 16);
> intel_bb_out(ibb, 0);
> intel_bb_out(ibb, x_dim);
>
> /* thread group Y */
> - intel_bb_out(ibb, 0);
> + intel_bb_out(ibb, y);
> intel_bb_out(ibb, 0);
> intel_bb_out(ibb, y_dim);
>
> @@ -948,6 +948,7 @@ xehp_emit_state_base_address(struct intel_bb *ibb)
>
> void
> xehp_emit_compute_walk(struct intel_bb *ibb,
> + unsigned int x, unsigned int y,
> unsigned int width, unsigned int height,
> struct xehp_interface_descriptor_data *pidd,
> uint8_t color)
> @@ -965,8 +966,8 @@ xehp_emit_compute_walk(struct intel_bb *ibb,
> * Then thread group X = width / 16 (aligned to 16)
> * thread group Y = height;
> */
> - x_dim = (width + 15) / 16;
> - y_dim = height;
> + x_dim = (x + width + 15) / 16;
> + y_dim = y + height;
>
> intel_bb_out(ibb, XEHP_COMPUTE_WALKER | 0x25);
>
> @@ -994,8 +995,8 @@ xehp_emit_compute_walk(struct intel_bb *ibb,
> intel_bb_out(ibb, 1); //dw9
>
> /* group id x/y/z */
> - intel_bb_out(ibb, 0); //dw10
> - intel_bb_out(ibb, 0); //dw11
> + intel_bb_out(ibb, x / 16); //dw10
> + intel_bb_out(ibb, y); //dw11
> intel_bb_out(ibb, 0); //dw12
>
> /* partition id / partition size */
> diff --git a/lib/gpu_cmds.h b/lib/gpu_cmds.h
> index 92cbbde9..b7ed64f8 100644
> --- a/lib/gpu_cmds.h
> +++ b/lib/gpu_cmds.h
> @@ -138,6 +138,7 @@ xehp_emit_state_base_address(struct intel_bb *ibb);
>
> void
> xehp_emit_compute_walk(struct intel_bb *ibb,
> + unsigned int x, unsigned int y,
> unsigned int width, unsigned int height,
> struct xehp_interface_descriptor_data *pidd,
> uint8_t color);
prev parent reply other threads:[~2023-06-14 8:24 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-06-09 9:37 [igt-dev] [PATCH i-g-t 1/2] lib: Apply x/y offsets to GPGPU fill Christoph Manszewski
2023-06-09 9:37 ` [igt-dev] [PATCH i-g-t 2/2] lib: Apply rightmost execution mask to xehp gpu walker Christoph Manszewski
2023-06-14 8:25 ` Grzegorzek, Dominik
2023-06-09 10:25 ` [igt-dev] ✗ GitLab.Pipeline: warning for series starting with [i-g-t,1/2] lib: Apply x/y offsets to GPGPU fill Patchwork
2023-06-09 17:31 ` [igt-dev] ✓ Fi.CI.BAT: success " Patchwork
2023-06-10 22:27 ` [igt-dev] ✓ Fi.CI.IGT: " Patchwork
2023-06-14 8:24 ` Grzegorzek, Dominik [this message]
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=e9bde4300a66ba82dbc9364a5c42f60f1e86b9ea.camel@intel.com \
--to=dominik.grzegorzek@intel.com \
--cc=chris.p.wilson@intel.com \
--cc=christoph.manszewski@intel.com \
--cc=igt-dev@lists.freedesktop.org \
/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