On 5/29/20 10:18 AM, Dikshita Agarwal wrote:
> Add platform specific capabilities and use them
> in place of firmware capabilities.
>
> Signed-off-by: Dikshita Agarwal <dikshita@xxxxxxxxxxxxxx>
> ---
> drivers/media/platform/qcom/venus/core.c | 3 +
> drivers/media/platform/qcom/venus/hfi_helper.h | 3 +-
> drivers/media/platform/qcom/venus/hfi_parser.c | 21 +--
> .../media/platform/qcom/venus/hfi_platform_data.c | 153 ++++++++++++++++++++-
> .../media/platform/qcom/venus/hfi_platform_data.h | 37 ++++-
> 5 files changed, 193 insertions(+), 24 deletions(-)
>
> diff --git a/drivers/media/platform/qcom/venus/core.c b/drivers/media/platform/qcom/venus/core.c
> index 4fde4aa..8221e5c 100644
> --- a/drivers/media/platform/qcom/venus/core.c
> +++ b/drivers/media/platform/qcom/venus/core.c
> @@ -273,6 +273,9 @@ static int venus_probe(struct platform_device *pdev)
> if (ret)
> goto err_venus_shutdown;
>
> + if (core->hfi_data && core->hfi_data->venus_parse_resources)
> + venus_parse_resources(core);
> +
> ret = hfi_core_init(core);
> if (ret)
> goto err_venus_shutdown;
> diff --git a/drivers/media/platform/qcom/venus/hfi_helper.h b/drivers/media/platform/qcom/venus/hfi_helper.h
> index f6613df..1671012 100644
> --- a/drivers/media/platform/qcom/venus/hfi_helper.h
> +++ b/drivers/media/platform/qcom/venus/hfi_helper.h
> @@ -520,7 +520,8 @@
> enum hfi_version {
> HFI_VERSION_1XX,
> HFI_VERSION_3XX,
> - HFI_VERSION_4XX
> + HFI_VERSION_4XX,
> + HFI_VERSION_6XX
> };
>
> struct hfi_buffer_info {
> diff --git a/drivers/media/platform/qcom/venus/hfi_parser.c b/drivers/media/platform/qcom/venus/hfi_parser.c
> index 7f515a4..b4b60e1 100644
> --- a/drivers/media/platform/qcom/venus/hfi_parser.c
> +++ b/drivers/media/platform/qcom/venus/hfi_parser.c
> @@ -10,30 +10,11 @@
> #include "core.h"
> #include "hfi_helper.h"
> #include "hfi_parser.h"
> +#include "hfi_platform_data.h"
>
> typedef void (*func)(struct venus_caps *cap, const void *data,
> unsigned int size);
>
> -static void init_codecs(struct venus_core *core)
> -{
> - struct venus_caps *caps = core->caps, *cap;
> - unsigned long bit;
> -
> - for_each_set_bit(bit, &core->dec_codecs, MAX_CODEC_NUM) {
> - cap = &caps[core->codecs_count++];
> - cap->codec = BIT(bit);
> - cap->domain = VIDC_SESSION_TYPE_DEC;
> - cap->valid = false;
> - }
> -
> - for_each_set_bit(bit, &core->enc_codecs, MAX_CODEC_NUM) {
> - cap = &caps[core->codecs_count++];
> - cap->codec = BIT(bit);
> - cap->domain = VIDC_SESSION_TYPE_ENC;
> - cap->valid = false;
> - }
> -}
> -
Please keep this function in hfi_parse.c and make a copy here. I don't
want to make such cross dependency between hfi_parse and hfi_platform.
Once we are in good shape with hfi_platform we could drop the hfi_parser
completely and don't rely on firmware.
> static void for_each_codec(struct venus_caps *caps, unsigned int caps_num,
> u32 codecs, u32 domain, func cb, void *data,
> unsigned int size)
> diff --git a/drivers/media/platform/qcom/venus/hfi_platform_data.c b/drivers/media/platform/qcom/venus/hfi_platform_data.c
> index 9d9035f..fc838f5 100644
> --- a/drivers/media/platform/qcom/venus/hfi_platform_data.c
> +++ b/drivers/media/platform/qcom/venus/hfi_platform_data.c
> @@ -5,8 +5,132 @@
> #include "hfi_platform_data.h"
> #include "core.h"
>
> +void init_codecs(struct venus_core *core)
> +{
> + struct venus_caps *caps = core->caps, *cap;
> + unsigned long bit;
> +
> + for_each_set_bit(bit, &core->dec_codecs, MAX_CODEC_NUM) {
> + cap = &caps[core->codecs_count++];
> + cap->codec = BIT(bit);
> + cap->domain = VIDC_SESSION_TYPE_DEC;
> + cap->valid = false;
> + }
> +
> + for_each_set_bit(bit, &core->enc_codecs, MAX_CODEC_NUM) {
> + cap = &caps[core->codecs_count++];
> + cap->codec = BIT(bit);
> + cap->domain = VIDC_SESSION_TYPE_ENC;
> + cap->valid = false;
> + }
> +}
> +
> +static void parse_codecs(struct venus_core *core)
s/parse_codecs/fill_codecs/
> +{
> + const struct venus_hfi_platform_data *hfi_data = core->hfi_data;
> + const struct venus_codecs *codecs = hfi_data->venus_codecs;
> + unsigned int i, codecs_size;
> +
> + core->enc_codecs = 0;
> + core->dec_codecs = 0;
> + codecs_size = hfi_data->venus_codecs_size;
> +
> + for (i = 0; i < codecs_size; i++) {
> + if (codecs[i].domain == VIDC_SESSION_TYPE_ENC)
> + core->enc_codecs |= codecs[i].codecs;
> + else
> + core->dec_codecs |= codecs[i].codecs;
> + }
> + init_codecs(core);
> +}
> +
> +static int fill_raw_fmt(struct venus_capability *hfi_data_caps,
> + struct venus_caps *core_caps)
> +{
> + if ((core_caps->num_pl + 1) == HFI_MAX_PROFILE_COUNT)
num_pl should be num_fmts, and MAX_FMT_ENTRIES
> + return -EINVAL;
> +
> + core_caps->fmts[core_caps->num_fmts].buftype =
> + hfi_data_caps->capability_type;
> + core_caps->fmts[core_caps->num_fmts].fmt =
> + hfi_data_caps->defaul_value;
> + core_caps->num_fmts += 1;
> +
> + return 0;
> +}
> +
> +static int fill_profile(struct venus_capability *hfi_data_caps,
> + struct venus_caps *core_caps)
> +{
> + if ((core_caps->num_pl + 1) == HFI_MAX_PROFILE_COUNT)
> + return -EINVAL;
> +
> + core_caps->pl[core_caps->num_pl].profile =
> + hfi_data_caps->capability_type;
> + core_caps->pl[core_caps->num_pl].level = hfi_data_caps->defaul_value;
> + core_caps->num_pl += 1;
> +
> + return 0;
> +}
> +
> +static int fill_capabilities(struct venus_capability *hfi_data_caps,
> + struct venus_caps *core_caps)
> +{
> + if ((core_caps->num_caps + 1) == MAX_CAP_ENTRIES)
> + return -EINVAL;
> +
> + core_caps->caps[core_caps->num_caps].capability_type =
> + hfi_data_caps->capability_type;
> + core_caps->caps[core_caps->num_caps].min = hfi_data_caps->min;
> + core_caps->caps[core_caps->num_caps].max = hfi_data_caps->max;
> + core_caps->caps[core_caps->num_caps].step_size =
> + hfi_data_caps->step_size;
> + core_caps->num_caps += 1;
> + return 0;
> +}
> +
> +static int fill_caps(struct venus_capability *hfi_data_caps,
> + struct venus_caps *core_caps)
> +{
> + if (hfi_data_caps->entry_type == CAPABILITY_ENTRY)
> + return fill_capabilities(hfi_data_caps, core_caps);
> + else if (hfi_data_caps->entry_type == PROFILE_ENTRY)
> + return fill_profile(hfi_data_caps, core_caps);
> + else
> + return fill_raw_fmt(hfi_data_caps, core_caps);
> +}
> +
> +static void parse_capabilities(struct venus_core *core)
s/parse_capabilities/fill_capabilities/
> +{
> + unsigned int i, j, ret;
ret should be 'int'
> + const struct venus_hfi_platform_data *hfi_data = core->hfi_data;
> + struct venus_capability *hfi_data_caps = hfi_data->capabilities;
> + unsigned int hfi_data_caps_size = hfi_data->capabilities_size;
> + const unsigned int codecs_size = hfi_data->venus_codecs_size;
> +
> + for (i = 0; i < hfi_data_caps_size; i++) {
> + for (j = 0; j < codecs_size; j++) {
could you loo[ over codecs first and then over hfi_data_caps
> + if ((hfi_data_caps[i].domain &
> + core->caps[j].domain) &&
> + (hfi_data_caps[i].codecs &
> + core->caps[j].codec)) {
could you invert the logic here:
if (!(hfi_data_caps[i].domain & core->caps[j].domain))
continue;
if (!(hfi_data_caps[i].codecs & core->caps[j].codec))
continue;
> + ret = fill_caps(&hfi_data_caps[i],
> + core->caps);
fill_caps could return error but we ignore it here, do we really want
fill_caps to return error?
> + if (!ret)
> + break;
> + }
> + }
> + }
> +}
> +
> +void venus_parse_resources(struct venus_core *core)
Move the content of this function in venus_get_hfi_platform().
> +{
> + parse_codecs(core);
> + parse_capabilities(core);
> +}
> +
> static struct codec_freq_data hfi4_codec_freq_data[] = {
> -{ V4L2_PIX_FMT_H264, VIDC_SESSION_TYPE_ENC, 675, 10 },
> + { V4L2_PIX_FMT_H264, VIDC_SESSION_TYPE_ENC, 675, 10 },
This shouldn't be part of this patch.
> { V4L2_PIX_FMT_HEVC, VIDC_SESSION_TYPE_ENC, 675, 10 },
> { V4L2_PIX_FMT_VP8, VIDC_SESSION_TYPE_ENC, 675, 10 },
> { V4L2_PIX_FMT_MPEG2, VIDC_SESSION_TYPE_DEC, 200, 10 },
> @@ -16,20 +140,45 @@
> { V4L2_PIX_FMT_VP9, VIDC_SESSION_TYPE_DEC, 200, 10 },
> };
>
> +static struct venus_codecs default_codecs[] = { {DEC, H264}, {DEC, HEVC},
> + {DEC, VP8}, {DEC, MPEG2}, {ENC, H264}, {ENC, HEVC}, {ENC, VP8},
> +};
> +
> +static struct venus_capability hfi6_capabilities[] = {
s/hfi6_capabilities/hfi_caps_v6/
> + /* {cap_type, codecs, domains, min, max, step_size, default} */
> + { HFI_CAPABILITY_FRAME_WIDTH, CODECS_ALL, DOMAINS_ALL, 96, 5760, 1,
> + 96, CAPABILITY_ENTRY },
> + { HFI_CAPABILITY_FRAME_HEIGHT, CODECS_ALL, DOMAINS_ALL, 96, 5760, 1,
> + 96, CAPABILITY_ENTRY },
> + { HFI_CAPABILITY_FRAMERATE, CODECS_ALL, DOMAINS_ALL, 1, 480, 1,
> + 1, CAPABILITY_ENTRY },
> + { HFI_CAPABILITY_MAX_VIDEOCORES, CODECS_ALL, DOMAINS_ALL, 1, 2, 1,
> + 1, CAPABILITY_ENTRY },
> +};
> +
> static const struct venus_hfi_platform_data hfi4_data = {
> .codec_freq_data = hfi4_codec_freq_data,
> .codec_freq_data_size = ARRAY_SIZE(hfi4_codec_freq_data),
> };
>
> +static const struct venus_hfi_platform_data hfi6_data = {
> + .venus_codecs = default_codecs,
> + .venus_codecs_size = ARRAY_SIZE(default_codecs),
> + .capabilities = hfi6_capabilities,
> + .capabilities_size = ARRAY_SIZE(hfi6_capabilities),
> + .venus_parse_resources = venus_parse_resources,
> +};
> +
> const struct venus_hfi_platform_data *venus_get_hfi_platform
> (enum hfi_version version)
> {
> switch (version) {
> case HFI_VERSION_4XX:
> return &hfi4_data;
> + case HFI_VERSION_6XX:
> + return &hfi6_data;
> default:
> return NULL;
> }
> return NULL;
> }
> -
> diff --git a/drivers/media/platform/qcom/venus/hfi_platform_data.h b/drivers/media/platform/qcom/venus/hfi_platform_data.h
> index 1b4bfb6..55de59b 100644
> --- a/drivers/media/platform/qcom/venus/hfi_platform_data.h
> +++ b/drivers/media/platform/qcom/venus/hfi_platform_data.h
> @@ -8,6 +8,19 @@
>
> #include "core.h"
>
> +#define ENC VIDC_SESSION_TYPE_ENC
> +#define DEC VIDC_SESSION_TYPE_DEC
> +#define H264 HFI_VIDEO_CODEC_H264
> +#define MPEG2 HFI_VIDEO_CODEC_MPEG2
> +#define HEVC HFI_VIDEO_CODEC_HEVC
> +#define VP8 HFI_VIDEO_CODEC_VP8
please don't re-define defines.
> +#define DOMAINS_ALL (HFI_DOMAIN_BASE_VENC | HFI_DOMAIN_BASE_VDEC)
> +#define CODECS_ALL (HFI_VIDEO_CODEC_H264 | HFI_VIDEO_CODEC_H264 \
> + | HFI_VIDEO_CODEC_VP8)
The CODECS_ALL should have the hfi version in the name, because
different versions supports different codecs.
> +#define CAPABILITY_ENTRY 0
> +#define PROFILE_ENTRY 1
> +#define FMT_ENTRY 2
> +
> struct codec_freq_data {
> u32 pixfmt;
> u32 session_type;
> @@ -15,13 +28,35 @@ struct codec_freq_data {
> unsigned long vsp_freq;
> };
>
> +struct venus_codecs {
can we call this hfi_platform_codecs ?
> + u32 codecs;
> + u32 domain;
> +};
> +
> +struct venus_capability {
hfi_platform_caps ?
> + u32 capability_type;
> + u32 codecs;
> + u32 domain;
maybe 'domains' ?
> + u32 min;
> + u32 max;
> + u32 step_size;
> + u32 defaul_value;
> + u32 entry_type;
can we avoid this entry_type by grouping the const arrays for caps,
profile/level and raw formats?
> +};
> +
> struct venus_hfi_platform_data {
> const struct codec_freq_data *codec_freq_data;
> unsigned int codec_freq_data_size;
> + struct venus_codecs *venus_codecs;
> + unsigned int venus_codecs_size;
> + struct venus_capability *capabilities;
> + unsigned int capabilities_size;
> + void (*venus_parse_resources)(struct venus_core *core);
I'd like to see the function pointers (aka ops) in separate structure.
I think struct hfi_platform_ops is a good name.
Also I think it would be better to get the capabilities when
venus_get_hfi_platform (hfi_platform_get) is called.
> };
>
> const struct venus_hfi_platform_data *venus_get_hfi_platform
> (enum hfi_version version);
> +void venus_parse_resources(struct venus_core *core);
> +void init_codecs(struct venus_core *core);
>
> #endif
> -
>
--
regards,
Stan
