Optimize looping pixels in compute_crc() and blend(). Calculate
src_offset in start of looping horizontally and increase it.
It's better than calculating in every pixels.
Cc: Rodrigo Siqueira <[email protected]>
Cc: David Airlie <[email protected]>
Cc: [email protected]
Cc: [email protected]
Signed-off-by: Sidong Yang <[email protected]>
---
drivers/gpu/drm/vkms/vkms_composer.c | 32 +++++++++++++++-------------
1 file changed, 17 insertions(+), 15 deletions(-)
diff --git a/drivers/gpu/drm/vkms/vkms_composer.c b/drivers/gpu/drm/vkms/vkms_composer.c
index 4af2f19480f4..9d2a765ca1fb 100644
--- a/drivers/gpu/drm/vkms/vkms_composer.c
+++ b/drivers/gpu/drm/vkms/vkms_composer.c
@@ -28,14 +28,14 @@ static uint32_t compute_crc(void *vaddr_out, struct vkms_composer *composer)
u32 crc = 0;
for (i = y_src; i < y_src + h_src; ++i) {
- for (j = x_src; j < x_src + w_src; ++j) {
- src_offset = composer->offset
- + (i * composer->pitch)
- + (j * composer->cpp);
+ src_offset = composer->offset + (i * composer->pitch)
+ + (x_src * composer->cpp);
+ for (j = 0 ; j < w_src; ++j) {
/* XRGB format ignores Alpha channel */
memset(vaddr_out + src_offset + 24, 0, 8);
crc = crc32_le(crc, vaddr_out + src_offset,
sizeof(u32));
+ src_offset += composer->cpp;
}
}
@@ -61,7 +61,7 @@ static void blend(void *vaddr_dst, void *vaddr_src,
struct vkms_composer *dest_composer,
struct vkms_composer *src_composer)
{
- int i, j, j_dst, i_dst;
+ int i, j, i_dst;
int offset_src, offset_dst;
int x_src = src_composer->src.x1 >> 16;
@@ -73,21 +73,23 @@ static void blend(void *vaddr_dst, void *vaddr_src,
int w_dst = drm_rect_width(&src_composer->dst);
int y_limit = y_src + h_dst;
- int x_limit = x_src + w_dst;
- for (i = y_src, i_dst = y_dst; i < y_limit; ++i) {
- for (j = x_src, j_dst = x_dst; j < x_limit; ++j) {
- offset_dst = dest_composer->offset
- + (i_dst * dest_composer->pitch)
- + (j_dst++ * dest_composer->cpp);
- offset_src = src_composer->offset
- + (i * src_composer->pitch)
- + (j * src_composer->cpp);
+ for (i = y_src, i_dst = y_dst; i < y_limit; ++i, ++i_dst) {
+ offset_dst = dest_composer->offset
+ + (i_dst * dest_composer->pitch)
+ + (x_dst * dest_composer->cpp);
+ offset_src = src_composer->offset
+ + (i * src_composer->pitch)
+ + (x_src * src_composer->cpp);
+
+ for (j = 0; j < w_dst; ++j) {
memcpy(vaddr_dst + offset_dst,
vaddr_src + offset_src, sizeof(u32));
+
+ offset_dst += dest_composer->cpp;
+ offset_src += src_composer->cpp;
}
- i_dst++;
}
}
--
2.17.1
On Sun, 31 May 2020 at 14:12, Sidong Yang <[email protected]> wrote:
>
> Optimize looping pixels in compute_crc() and blend(). Calculate
> src_offset in start of looping horizontally and increase it.
> It's better than calculating in every pixels.
>
When you say "optimize" have you observed any actual benefits of the
patch - be that smaller binary, faster execution time, etc?
If there are - mentioned them in the commit message. Otherwise, it
doesn't optimise anything.
A while back, I've suggested something similar [1] mostly for cosmetic
purposes - doubt there's much benefits beyond that.
HTH
-Emil
[1] https://patchwork.freedesktop.org/patch/365177/#comment_674314
Hi,
First of all, thanks a lot for all your patch. And thanks Emil for your
feedback.
I have a suggestion here:
Emil:
Could you give me your Acked-by or maybe Reviewed-by for the writeback
series? With that, I can finally apply the series.
Sidong:
Secondly, after applying the writeback series, I would suggest you to
understand Emil's comments (he already provides the link) and prepare a
new patch based on that. Before you submit your patch, I recommend you
to test it with https://patchwork.freedesktop.org/series/68352/ and
kms_flip.
How about that?
Best Regards
On 05/31, Emil Velikov wrote:
> On Sun, 31 May 2020 at 14:12, Sidong Yang <[email protected]> wrote:
> >
> > Optimize looping pixels in compute_crc() and blend(). Calculate
> > src_offset in start of looping horizontally and increase it.
> > It's better than calculating in every pixels.
> >
> When you say "optimize" have you observed any actual benefits of the
> patch - be that smaller binary, faster execution time, etc?
> If there are - mentioned them in the commit message. Otherwise, it
> doesn't optimise anything.
>
> A while back, I've suggested something similar [1] mostly for cosmetic
> purposes - doubt there's much benefits beyond that.
>
> HTH
> -Emil
> [1] https://patchwork.freedesktop.org/patch/365177/#comment_674314
--
Rodrigo Siqueira
https://siqueira.tech
On Mon, 1 Jun 2020 at 01:25, Rodrigo Siqueira
<[email protected]> wrote:
>
> Hi,
>
> First of all, thanks a lot for all your patch. And thanks Emil for your
> feedback.
>
> I have a suggestion here:
>
> Emil:
> Could you give me your Acked-by or maybe Reviewed-by for the writeback
> series? With that, I can finally apply the series.
>
Sure, once the issues highlighted are resolved. Just left you some
more comprehensive feedback.
-Emil
P.S. Something something top posting sucks :-P