Re: [Buildroot] [PATCH v2 2/7] package/freescale-imx/imx-dpu-g2d: new package imx-dpu-g2d package is used by i.MX 8 and i.MX 8X series instead of imx-vpu or imx-vpu-hantro.With this commit, imx-dpu-g2d is added and is enabled for i.MX 8X.

[Gary Bisson <gary.bisson@boundarydevices.com>]

Hi Thomas,

On Thu, Aug 05, 2021 at 10:51:02PM +0200, Thomas Petazzoni wrote:
> Hello Gary,
> 
> On Mon, 10 Feb 2020 16:58:01 +0100
> Gary Bisson <bisson.gary@gmail.com> wrote:
> 
> > Sorry but I disagree with the title/log of this commit.
> > 
> > This has nothing to do with vpu, it's all about g2d which is NXP
> > interface for 2D graphics operations (CSC, rotation etc).
> > 
> > On most platforms (like i.MX6 for instance) there is a 3D GPU and a 2D
> > GPU, so for those Vivante package was providing the libg2d libraries
> > that leveraged the 2D GPU.
> > 
> > Now on newer platform such as i.MX8X, there's no 2D GPU but the display
> > IP (here called DPU) can perform the 2D operations so another package
> > was created.
> > 
> > Here is where the selection is made in Yocto:
> > https://source.codeaurora.org/external/imx/meta-fsl-bsp-release/tree/imx/meta-bsp/conf/machine/include/imx-base.inc?h=warrior-4.19.35-1.1.0#n222
> > 
> > So my opinion is that, like Yocto, we should create a virtual g2d
> > package that either be populated by Vivante libs by the platforms that
> > have 2D GPU or by DPU on others.
> 
> Thanks for the explanation, but I do have a number of questions
> following this. Right now the imx-gpu-g2d package is included with:
> 
> if BR2_PACKAGE_FREESCALE_IMX_HAS_VIV_GPU
> source "package/freescale-imx/imx-gpu-g2d/Config.in"
> ...
> endif
> 
> So it means it's visible and accessible to all those platforms:
> 
> config BR2_PACKAGE_FREESCALE_IMX_HAS_VIV_GPU
>         bool
>         default y if BR2_PACKAGE_FREESCALE_IMX_PLATFORM_IMX6Q || \
>                 BR2_PACKAGE_FREESCALE_IMX_PLATFORM_IMX6S || \
>                 BR2_PACKAGE_FREESCALE_IMX_PLATFORM_IMX7 || \
>                 BR2_PACKAGE_FREESCALE_IMX_PLATFORM_IMX8 || \
>                 BR2_PACKAGE_FREESCALE_IMX_PLATFORM_IMX8M || \
>                 BR2_PACKAGE_FREESCALE_IMX_PLATFORM_IMX8MM || \
>                 BR2_PACKAGE_FREESCALE_IMX_PLATFORM_IMX8MN || \
>                 BR2_PACKAGE_FREESCALE_IMX_PLATFORM_IMX8MP || \
>                 BR2_PACKAGE_FREESCALE_IMX_PLATFORM_IMX8X
> 
> But you're saying that imx-gpu-g2d only applies to i.MX6. So it seams
> like we would in fact need a BR2_PACKAGE_FREESCALE_IMX_HAS_VIV_GPU_G2D,
> like:

I was saying it only was supported on i.MX6 series, but since then NXP
added the 2D GPU back on several new SoCs.

> config BR2_PACKAGE_FREESCALE_IMX_HAS_VIV_GPU_G2D
>         bool
>         default y if BR2_PACKAGE_FREESCALE_IMX_PLATFORM_IMX6Q || \
>                 BR2_PACKAGE_FREESCALE_IMX_PLATFORM_IMX6S
> 
> And make imx-gpu-g2d only visible on these ?

I believe I offered something like that in the past but was told to
limit the use of "IMX_HAS_XXX". At the time I believe the reasoning was
that users should know whether they can use g2d or not.

> Could you clarify exactly the list of i.MX6/i.MX7/i.MX8 variants on
> which imx-gpu-g2d makes sense, and on which imx-dpu-g2d makes sense ?
> 
> In Yocto they have this odd:
> 
> MACHINEOVERRIDES_EXTENDER_mx25   = "use-mainline-bsp"
> MACHINEOVERRIDES_EXTENDER_mx6q   = "imxfbdev:imxipu:imxvpu:imxgpu:imxgpu2d:imxgpu3d"
> MACHINEOVERRIDES_EXTENDER_mx6dl  = "imxfbdev:imxpxp:imxipu:imxvpu:imxgpu:imxgpu2d:imxgpu3d:imxepdc"
> MACHINEOVERRIDES_EXTENDER_mx6sx  = "imxfbdev:imxpxp:imxgpu:imxgpu2d:imxgpu3d"
> MACHINEOVERRIDES_EXTENDER_mx6sl  = "imxfbdev:imxpxp:imxgpu:imxgpu2d:imxepdc"
> MACHINEOVERRIDES_EXTENDER_mx6sll = "imxfbdev:imxpxp:imxepdc"
> MACHINEOVERRIDES_EXTENDER_mx6ul  = "imxfbdev:imxpxp"
> MACHINEOVERRIDES_EXTENDER_mx6ull = "imxfbdev:imxpxp:imxepdc"
> MACHINEOVERRIDES_EXTENDER_mx6ulz = "imxfbdev:imxpxp:imxepdc"
> MACHINEOVERRIDES_EXTENDER_mx7d   = "imxfbdev:imxpxp:imxepdc"
> MACHINEOVERRIDES_EXTENDER_mx7ulp = "imxfbdev:imxpxp:imxgpu:imxgpu2d:imxgpu3d"
> MACHINEOVERRIDES_EXTENDER_mx8qm  = "imxdrm:imxdpu:imxgpu:imxgpu2d:imxgpu3d"
> MACHINEOVERRIDES_EXTENDER_mx8mm  = "imxdrm:imxvpu:imxgpu:imxgpu2d:imxgpu3d"
> MACHINEOVERRIDES_EXTENDER_mx8mn  = "imxdrm:imxgpu:imxgpu3d"
> MACHINEOVERRIDES_EXTENDER_mx8mp  = "imxdrm:imxvpu:imxgpu:imxgpu2d:imxgpu3d"
> MACHINEOVERRIDES_EXTENDER_mx8mq  = "imxdrm:imxvpu:imxgpu:imxgpu3d"
> MACHINEOVERRIDES_EXTENDER_mx8qxp = "imxdrm:imxdpu:imxgpu:imxgpu2d:imxgpu3d"
> MACHINEOVERRIDES_EXTENDER_mx8dxl = "imxfbdev"
> 
> But I can't really make much sense out of this.

Well I guess the naming is very confusing, but having imxgpu2d in the
above means that G2D is supported, whether it is with GPU2D or DPU [1].

From that list, if you have both imxgpu2d + imxdpu it means imx-dpu-g2d
is the g2d lib provider[2]. So imx-dpu-g2d is only available for mx8qm
and mx8qxp, all the other platforms with imxgpu2d are using imx-gpu-g2d.

Hope this clarifies things (a bit at least).

Regards,
Gary

[1] https://github.com/Freescale/meta-freescale/blob/hardknott/conf/machine/include/imx-base.inc#L228
[2] https://github.com/Freescale/meta-freescale/blob/hardknott/conf/machine/include/imx-base.inc#L362
_______________________________________________
buildroot mailing list
buildroot@busybox.net
http://lists.busybox.net/mailman/listinfo/buildroot