From: Francois Dugast <francois.dugast@intel.com>
To: "Zbigniew Kempczyński" <zbigniew.kempczynski@intel.com>
Cc: <igt-dev@lists.freedesktop.org>
Subject: Re: [PATCH i-g-t v3 09/11] lib/intel_blt: add support for matrix mem-copy
Date: Wed, 28 May 2025 10:23:41 +0200 [thread overview]
Message-ID: <aDbIDROnBnbIbMOM@fdugast-desk> (raw)
In-Reply-To: <20250523080126.75295-10-zbigniew.kempczynski@intel.com>
On Fri, May 23, 2025 at 10:01:24AM +0200, Zbigniew Kempczyński wrote:
> Linear copy in intel_blt supports passing large buffers (which
> requires to be spread over couple mem-copies). For matrix this is
> a little bit more complicated so I left simple case in which
Patch LGTM but a nit: impersonal style is preferred over "I did X".
With that:
Reviewed-by: Francois Dugast <francois.dugast@intel.com>
> pitch/width/height must be within mem-copy command limits -
> 18-bit width * 18-bit height gives 64GiB object so testing
> copying bigger buffer would be an overkill.
>
> Cc: Francois Dugast <francois.dugast@intel.com>
> Signed-off-by: Zbigniew Kempczyński <zbigniew.kempczynski@intel.com>
> ---
> lib/intel_blt.c | 69 +++++++++++++++++++++++++++++++++----------------
> 1 file changed, 47 insertions(+), 22 deletions(-)
>
> diff --git a/lib/intel_blt.c b/lib/intel_blt.c
> index 265f5ed50f..77a03aff4e 100644
> --- a/lib/intel_blt.c
> +++ b/lib/intel_blt.c
> @@ -1893,17 +1893,18 @@ static uint64_t emit_blt_mem_copy(int fd, uint64_t ahnd,
> {
> struct xe_mem_copy_data data = {};
> uint64_t dst_offset, src_offset, shift;
> - uint32_t height, width_max, remain;
> + uint32_t width, height, width_max, height_max, remain;
> uint32_t bbe = MI_BATCH_BUFFER_END;
> uint32_t *bb;
>
> if (mem->mode == MODE_BYTE) {
> data.dw01.byte_copy.width = -1;
> - width_max = data.dw01.byte_copy.width + 1;
> + height_max = width_max = data.dw01.byte_copy.width + 1;
> shift = width_max;
> } else {
> data.dw01.page_copy.width = -1;
> width_max = data.dw01.page_copy.width + 1;
> + height_max = 1;
> shift = width_max << 8;
> }
>
> @@ -1914,6 +1915,7 @@ static uint64_t emit_blt_mem_copy(int fd, uint64_t ahnd,
>
> bb = bo_map(fd, mem->bb.handle, mem->bb.size, mem->driver);
>
> + width = mem->src.width;
> height = mem->dst.height;
>
> data.dw00.client = 0x2;
> @@ -1930,34 +1932,57 @@ static uint64_t emit_blt_mem_copy(int fd, uint64_t ahnd,
> data.dw09.src_mocs = mem->src.mocs_index;
> data.dw09.dst_mocs = mem->dst.mocs_index;
>
> - remain = mem->src.width;
> + /* For matrix we don't iterate */
> + if (mem->copy_type == TYPE_MATRIX) {
> + if (width > width_max) {
> + width = width_max;
> + igt_warn("src width is bigger than max width [%u > %u => %u], truncating it\n",
> + mem->src.width, width_max, width);
> + }
>
> - /* Truncate pitches to match operation bits */
> - if (mem->src.pitch > width_max)
> - data.dw03.src_pitch = width_max - 1;
> - else
> - data.dw03.src_pitch = mem->src.pitch;
> + if (height > height_max) {
> + height = height_max;
> + igt_warn("src height is bigger than max height [%u > %u => %u], truncating it\n",
> + mem->src.height, height_max, height);
> + }
>
> - if (mem->dst.pitch > width_max)
> - data.dw04.dst_pitch = width_max - 1;
> - else
> - data.dw04.dst_pitch = mem->dst.pitch;
> -
> - while (remain) {
> - data.dw01.val = min_t(uint32_t, width_max, remain) - 1;
> + data.dw01.byte_copy.width = width - 1;
> + data.dw03.src_pitch = mem->src.pitch - 1;
> + data.dw04.dst_pitch = mem->dst.pitch - 1;
>
> igt_assert(bb_pos + sizeof(data) < mem->bb.size);
> memcpy(bb + bb_pos, &data, sizeof(data));
> bb_pos += sizeof(data);
> + } else {
> + remain = mem->src.width;
>
> - remain -= remain > width_max ? width_max : remain;
> - src_offset += shift;
> - dst_offset += shift;
> + /* Truncate pitches to match operation bits */
> + if (mem->src.pitch > width_max)
> + data.dw03.src_pitch = width_max - 1;
> + else
> + data.dw03.src_pitch = mem->src.pitch;
>
> - data.dw05.src_address_lo = src_offset;
> - data.dw06.src_address_hi = src_offset >> 32;
> - data.dw07.dst_address_lo = dst_offset;
> - data.dw08.dst_address_hi = dst_offset >> 32;
> + if (mem->dst.pitch > width_max)
> + data.dw04.dst_pitch = width_max - 1;
> + else
> + data.dw04.dst_pitch = mem->dst.pitch;
> +
> + while (remain) {
> + data.dw01.val = min_t(uint32_t, width_max, remain) - 1;
> +
> + igt_assert(bb_pos + sizeof(data) < mem->bb.size);
> + memcpy(bb + bb_pos, &data, sizeof(data));
> + bb_pos += sizeof(data);
> +
> + remain -= remain > width_max ? width_max : remain;
> + src_offset += shift;
> + dst_offset += shift;
> +
> + data.dw05.src_address_lo = src_offset;
> + data.dw06.src_address_hi = src_offset >> 32;
> + data.dw07.dst_address_lo = dst_offset;
> + data.dw08.dst_address_hi = dst_offset >> 32;
> + }
> }
>
> if (emit_bbe) {
> --
> 2.43.0
>
next prev parent reply other threads:[~2025-05-28 8:24 UTC|newest]
Thread overview: 25+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-05-23 8:01 [PATCH i-g-t v3 00/11] Improve mem-copy/mem-set lib and tests Zbigniew Kempczyński
2025-05-23 8:01 ` [PATCH i-g-t v3 01/11] lib/intel_cmds_info: rename M to TYPE in blt_memop_type Zbigniew Kempczyński
2025-05-23 8:01 ` [PATCH i-g-t v3 02/11] lib/intel_blt: separate mem-copy and mem-set Zbigniew Kempczyński
2025-05-23 8:01 ` [PATCH i-g-t v3 03/11] lib/intel_cmds_info: add blt_memop_mode (byte/page) Zbigniew Kempczyński
2025-05-27 12:19 ` Francois Dugast
2025-05-23 8:01 ` [PATCH i-g-t v3 04/11] lib/intel_blt: add emit batchbuffer end Zbigniew Kempczyński
2025-05-23 8:01 ` [PATCH i-g-t v3 05/11] lib/intel_blt: use struct instead of inline coding Zbigniew Kempczyński
2025-05-23 8:01 ` [PATCH i-g-t v3 06/11] tests/xe_copy_basic: replace size to rect which keeps objects geometry Zbigniew Kempczyński
2025-05-27 19:15 ` Francois Dugast
2025-05-28 19:28 ` Zbigniew Kempczyński
2025-05-23 8:01 ` [PATCH i-g-t v3 07/11] tests/xe_copy_basic: add testcase with large buffer size Zbigniew Kempczyński
2025-05-27 19:16 ` Francois Dugast
2025-05-23 8:01 ` [PATCH i-g-t v3 08/11] tests/xe_copy_basic: add subtest to verify mem-copy in pages Zbigniew Kempczyński
2025-05-27 19:23 ` Francois Dugast
2025-05-30 6:10 ` Zbigniew Kempczyński
2025-05-23 8:01 ` [PATCH i-g-t v3 09/11] lib/intel_blt: add support for matrix mem-copy Zbigniew Kempczyński
2025-05-28 8:23 ` Francois Dugast [this message]
2025-05-30 6:17 ` Zbigniew Kempczyński
2025-05-23 8:01 ` [PATCH i-g-t v3 10/11] tests/xe_copy_basic: add mem-copy matrix subtests Zbigniew Kempczyński
2025-05-28 8:26 ` Francois Dugast
2025-05-23 8:01 ` [PATCH i-g-t v3 11/11] lib/intel_blt: add mem-copy debug facility Zbigniew Kempczyński
2025-05-28 8:29 ` Francois Dugast
2025-05-23 10:45 ` ✗ i915.CI.BAT: failure for Improve mem-copy/mem-set lib and tests (rev2) Patchwork
2025-05-23 11:36 ` ✓ Xe.CI.BAT: success " Patchwork
2025-05-23 19:54 ` ✓ Xe.CI.Full: " Patchwork
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=aDbIDROnBnbIbMOM@fdugast-desk \
--to=francois.dugast@intel.com \
--cc=igt-dev@lists.freedesktop.org \
--cc=zbigniew.kempczynski@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