[Mesa-dev] [PATCH 1/2] i965: Fix broxton 2x6 l3 config

[Francisco Jerez]

Anuj Phogat <anuj.phogat at gmail.com> writes:

> The new table added in this patch matches with the table
> in gfxspecs. We were programming the wrong values earlier.
>
> Signed-off-by: Anuj Phogat <anuj.phogat at gmail.com>
> Cc: Francisco Jerez <currojerez at riseup.net>
> Cc: "17.1" <mesa-stable at lists.freedesktop.org>
> ---
>  src/intel/common/gen_device_info.c |  1 +
>  src/intel/common/gen_device_info.h |  1 +
>  src/intel/common/gen_l3_config.c   | 19 +++++++++++++++++++
>  3 files changed, 21 insertions(+)
>
> diff --git a/src/intel/common/gen_device_info.c b/src/intel/common/gen_device_info.c
> index 75284a6..eccb464 100644
> --- a/src/intel/common/gen_device_info.c
> +++ b/src/intel/common/gen_device_info.c
> @@ -502,6 +502,7 @@ static const struct gen_device_info gen_device_info_bxt = {
>  
>  static const struct gen_device_info gen_device_info_bxt_2x6 = {
>     GEN9_LP_FEATURES_2X6,
> +   .is_broxton_2x6 = 1,
>     .l3_banks = 1,
>  };
>  /*
> diff --git a/src/intel/common/gen_device_info.h b/src/intel/common/gen_device_info.h
> index 6207630..4fe1b21 100644
> --- a/src/intel/common/gen_device_info.h
> +++ b/src/intel/common/gen_device_info.h
> @@ -41,6 +41,7 @@ struct gen_device_info
>     bool is_haswell;
>     bool is_cherryview;
>     bool is_broxton;
> +   bool is_broxton_2x6;

I don't think this boolean flag is useful.  The hardware spec refers to
the validated configurations that apply to BXT 2x6 as relevant to all
products with a single bank, which means you should probably check for 
gen >= 9 and l3_banks == 1 instead.

>     bool is_kabylake;
>  
>     bool has_hiz_and_separate_stencil;
> diff --git a/src/intel/common/gen_l3_config.c b/src/intel/common/gen_l3_config.c
> index ae31d08..e17994b 100644
> --- a/src/intel/common/gen_l3_config.c
> +++ b/src/intel/common/gen_l3_config.c
> @@ -102,6 +102,23 @@ static const struct gen_l3_config chv_l3_configs[] = {
>  };
>  
>  /**
> + * BXT 2x6 validated L3 configurations.  \sa ivb_l3_configs.
> + * Number of ways =
> + *    Allocation in KB for SKU / (Way size per bank * Number of banks).

Is this formula relevant?  The way size per bank is not really
documented in the BSpec so it doesn't really give the reader any
additional useful information.

> + * For BXT 2x6: Banks = 1, Way size per bank = 4.
> + */
> +static const struct gen_l3_config bxt_2x6_l3_configs[] = {
> +   /*SLM URB  All DC  RO  IS   C   T */
> +   {{  0, 32, 48,  0,  0,  0,  0,  0 }},
> +   {{  0, 32,  0,  8, 40,  0,  0,  0 }},
> +   {{  0, 32,  0, 32, 16,  0,  0,  0 }},
> +   {{ 16, 16, 48,  0,  0,  0,  0,  0 }},
> +   {{ 16, 16,  0, 40,  8,  0,  0,  0 }},
> +   {{ 16, 16,  0, 16, 32,  0,  0,  0 }},
> +   {{ 0 }}
> +};
> +
> +/**
>   * Return a zero-terminated array of validated L3 configurations for the
>   * specified device.
>   */
> @@ -117,6 +134,8 @@ get_l3_configs(const struct gen_device_info *devinfo)
>  
>     case 9:
>     case 10:
> +      if (devinfo->is_broxton_2x6)
> +         return bxt_2x6_l3_configs;
>        return chv_l3_configs;
>  
>     default:
> -- 
> 2.9.3
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 212 bytes
Desc: not available
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20170612/eaf5c849/attachment.sig>