[PATCH v3 0/4] media: venus: Update clock scaling and core selection

[PATCH v3 0/4] media: venus: Update clock scaling and core selection

[Aniket Masule]

In this patch series, clock scaling and core selection methods are
updated. Current clock scaling and core selection methods are same
for vpu4 and previous versions. Introducing load calculations using
vpp cycles, which indicates the cycles required by video hardware to
process each macroblock. Clock scaling is now done more precisely using
vpp cycles. Instance is assigned to core with minimum load, instead of
of static assignment.

Changes since v2 addressing the comments:
 - Codec_data and vpp_cycles have renamed to codec_freq_data and
   vpp_freq respectively.
 - Conditional check in patch 4 for checking max cores available
   is updated to VIDC_CORE_ID_2.
 - venus_helper_decide_core renamed to venus_helper_set_core.
 - scale_clocks_vpu4 renamed scale_clocks_v4.

For comment on freq fied in clk_data structure, it is getting initialized
and used in scale_clocks_v4 function. 


Aniket Masule (4):
  media: venus: Add codec data table
  media: venus: Update clock scaling
  media: venus: Add interface for load per core
  media: venus: Update core selection

 drivers/media/platform/qcom/venus/core.c       |  13 ++
 drivers/media/platform/qcom/venus/core.h       |  15 ++
 drivers/media/platform/qcom/venus/helpers.c    | 190 +++++++++++++++++++++++--
 drivers/media/platform/qcom/venus/helpers.h    |   3 +-
 drivers/media/platform/qcom/venus/hfi_helper.h |   1 +
 drivers/media/platform/qcom/venus/hfi_parser.h |   5 +
 drivers/media/platform/qcom/venus/vdec.c       |   6 +-
 drivers/media/platform/qcom/venus/venc.c       |   6 +-
 8 files changed, 226 insertions(+), 13 deletions(-)

-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project

[PATCH v3 1/4] media: venus: Add codec data table

[Aniket Masule]

Add vpp cycles for for different types of codec
It indicates the cycles required by video hardware
to process each macroblock. Initialize the codec
data with core resources.

Signed-off-by: Aniket Masule <amasule@codeaurora.org>
---
 drivers/media/platform/qcom/venus/core.c    | 13 +++++++++++++
 drivers/media/platform/qcom/venus/core.h    | 15 +++++++++++++++
 drivers/media/platform/qcom/venus/helpers.c | 30 +++++++++++++++++++++++++++++
 drivers/media/platform/qcom/venus/helpers.h |  1 +
 drivers/media/platform/qcom/venus/vdec.c    |  4 ++++
 drivers/media/platform/qcom/venus/venc.c    |  4 ++++
 6 files changed, 67 insertions(+)

diff --git a/drivers/media/platform/qcom/venus/core.c b/drivers/media/platform/qcom/venus/core.c
index 7393667..f1597d6 100644
--- a/drivers/media/platform/qcom/venus/core.c
+++ b/drivers/media/platform/qcom/venus/core.c
@@ -473,9 +473,22 @@ static __maybe_unused int venus_runtime_resume(struct device *dev)
 	{  244800, 100000000 },	/* 1920x1080@30 */
 };
 
+static struct codec_freq_data sdm845_codec_freq_data[] =  {
+	{ V4L2_PIX_FMT_H264, VIDC_SESSION_TYPE_ENC, 675 },
+	{ V4L2_PIX_FMT_HEVC, VIDC_SESSION_TYPE_ENC, 675 },
+	{ V4L2_PIX_FMT_VP8, VIDC_SESSION_TYPE_ENC, 675 },
+	{ V4L2_PIX_FMT_MPEG2, VIDC_SESSION_TYPE_DEC, 200 },
+	{ V4L2_PIX_FMT_H264, VIDC_SESSION_TYPE_DEC, 200 },
+	{ V4L2_PIX_FMT_HEVC, VIDC_SESSION_TYPE_DEC, 200 },
+	{ V4L2_PIX_FMT_VP8, VIDC_SESSION_TYPE_DEC, 200 },
+	{ V4L2_PIX_FMT_VP9, VIDC_SESSION_TYPE_DEC, 200 },
+};
+
 static const struct venus_resources sdm845_res = {
 	.freq_tbl = sdm845_freq_table,
 	.freq_tbl_size = ARRAY_SIZE(sdm845_freq_table),
+	.codec_freq_data = sdm845_codec_freq_data,
+	.codec_freq_data_size = ARRAY_SIZE(sdm845_codec_freq_data),
 	.clks = {"core", "iface", "bus" },
 	.clks_num = 3,
 	.max_load = 2563200,
diff --git a/drivers/media/platform/qcom/venus/core.h b/drivers/media/platform/qcom/venus/core.h
index 7a3feb5..2ed6496 100644
--- a/drivers/media/platform/qcom/venus/core.h
+++ b/drivers/media/platform/qcom/venus/core.h
@@ -35,12 +35,20 @@ struct reg_val {
 	u32 value;
 };
 
+struct codec_freq_data {
+	u32 pixfmt;
+	u32 session_type;
+	unsigned int vpp_freq;
+};
+
 struct venus_resources {
 	u64 dma_mask;
 	const struct freq_tbl *freq_tbl;
 	unsigned int freq_tbl_size;
 	const struct reg_val *reg_tbl;
 	unsigned int reg_tbl_size;
+	const struct codec_freq_data *codec_freq_data;
+	unsigned int codec_freq_data_size;
 	const char * const clks[VIDC_CLKS_NUM_MAX];
 	unsigned int clks_num;
 	enum hfi_version hfi_version;
@@ -216,6 +224,12 @@ struct venus_buffer {
 	struct list_head ref_list;
 };
 
+struct clock_data {
+	u32 core_id;
+	unsigned long freq;
+	struct codec_freq_data *codec_freq_data;
+};
+
 #define to_venus_buffer(ptr)	container_of(ptr, struct venus_buffer, vb)
 
 /**
@@ -275,6 +289,7 @@ struct venus_inst {
 	struct list_head list;
 	struct mutex lock;
 	struct venus_core *core;
+	struct clock_data clk_data;
 	struct list_head dpbbufs;
 	struct list_head internalbufs;
 	struct list_head registeredbufs;
diff --git a/drivers/media/platform/qcom/venus/helpers.c b/drivers/media/platform/qcom/venus/helpers.c
index 5cad601..f7f724b 100644
--- a/drivers/media/platform/qcom/venus/helpers.c
+++ b/drivers/media/platform/qcom/venus/helpers.c
@@ -715,6 +715,36 @@ int venus_helper_set_core_usage(struct venus_inst *inst, u32 usage)
 }
 EXPORT_SYMBOL_GPL(venus_helper_set_core_usage);
 
+int venus_helper_init_codec_data(struct venus_inst *inst)
+{
+	const struct codec_data *codec_data;
+	unsigned int i, codec_data_size;
+	u32 pixfmt;
+	int ret = 0;
+
+	if (!IS_V4(inst->core))
+		return 0;
+
+	codec_data = inst->core->res->codec_data;
+	codec_data_size = inst->core->res->codec_data_size;
+	pixfmt = inst->session_type == VIDC_SESSION_TYPE_DEC ?
+			inst->fmt_out->pixfmt : inst->fmt_cap->pixfmt;
+
+	for (i = 0; i < codec_data_size; i++) {
+		if (codec_data[i].pixfmt == pixfmt &&
+		    codec_data[i].session_type == inst->session_type) {
+			inst->clk_data.codec_data = &codec_data[i];
+			break;
+		}
+	}
+
+	if (!inst->clk_data.codec_data)
+		ret = -EINVAL;
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(venus_helper_init_codec_data);
+
 int venus_helper_set_num_bufs(struct venus_inst *inst, unsigned int input_bufs,
 			      unsigned int output_bufs,
 			      unsigned int output2_bufs)
diff --git a/drivers/media/platform/qcom/venus/helpers.h b/drivers/media/platform/qcom/venus/helpers.h
index 2475f284..f9360a8 100644
--- a/drivers/media/platform/qcom/venus/helpers.h
+++ b/drivers/media/platform/qcom/venus/helpers.h
@@ -41,6 +41,7 @@ int venus_helper_set_output_resolution(struct venus_inst *inst,
 				       unsigned int width, unsigned int height,
 				       u32 buftype);
 int venus_helper_set_work_mode(struct venus_inst *inst, u32 mode);
+int venus_helper_init_codec_data(struct venus_inst *inst);
 int venus_helper_set_core_usage(struct venus_inst *inst, u32 usage);
 int venus_helper_set_num_bufs(struct venus_inst *inst, unsigned int input_bufs,
 			      unsigned int output_bufs,
diff --git a/drivers/media/platform/qcom/venus/vdec.c b/drivers/media/platform/qcom/venus/vdec.c
index 282de21..51795fd 100644
--- a/drivers/media/platform/qcom/venus/vdec.c
+++ b/drivers/media/platform/qcom/venus/vdec.c
@@ -660,6 +660,10 @@ static int vdec_init_session(struct venus_inst *inst)
 	if (ret)
 		goto deinit;
 
+	ret = venus_helper_init_codec_data(inst);
+	if (ret)
+		goto deinit;
+
 	return 0;
 deinit:
 	hfi_session_deinit(inst);
diff --git a/drivers/media/platform/qcom/venus/venc.c b/drivers/media/platform/qcom/venus/venc.c
index 32cff29..792cdce 100644
--- a/drivers/media/platform/qcom/venus/venc.c
+++ b/drivers/media/platform/qcom/venus/venc.c
@@ -847,6 +847,10 @@ static int venc_init_session(struct venus_inst *inst)
 	if (ret)
 		goto deinit;
 
+	ret = venus_helper_init_codec_data(inst);
+	if (ret)
+		goto deinit;
+
 	ret = venc_set_properties(inst);
 	if (ret)
 		goto deinit;
-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project

Re: [PATCH v3 1/4] media: venus: Add codec data table

[Stanimir Varbanov]

On 6/25/19 7:27 PM, Aniket Masule wrote:
> Add vpp cycles for for different types of codec
> It indicates the cycles required by video hardware
> to process each macroblock. Initialize the codec
> data with core resources.
> 
> Signed-off-by: Aniket Masule <amasule@codeaurora.org>
> ---
>  drivers/media/platform/qcom/venus/core.c    | 13 +++++++++++++
>  drivers/media/platform/qcom/venus/core.h    | 15 +++++++++++++++
>  drivers/media/platform/qcom/venus/helpers.c | 30 +++++++++++++++++++++++++++++
>  drivers/media/platform/qcom/venus/helpers.h |  1 +
>  drivers/media/platform/qcom/venus/vdec.c    |  4 ++++
>  drivers/media/platform/qcom/venus/venc.c    |  4 ++++
>  6 files changed, 67 insertions(+)
> 
> diff --git a/drivers/media/platform/qcom/venus/core.c b/drivers/media/platform/qcom/venus/core.c
> index 7393667..f1597d6 100644
> --- a/drivers/media/platform/qcom/venus/core.c
> +++ b/drivers/media/platform/qcom/venus/core.c
> @@ -473,9 +473,22 @@ static __maybe_unused int venus_runtime_resume(struct device *dev)
>  	{  244800, 100000000 },	/* 1920x1080@30 */
>  };
>  
> +static struct codec_freq_data sdm845_codec_freq_data[] =  {
> +	{ V4L2_PIX_FMT_H264, VIDC_SESSION_TYPE_ENC, 675 },
> +	{ V4L2_PIX_FMT_HEVC, VIDC_SESSION_TYPE_ENC, 675 },
> +	{ V4L2_PIX_FMT_VP8, VIDC_SESSION_TYPE_ENC, 675 },
> +	{ V4L2_PIX_FMT_MPEG2, VIDC_SESSION_TYPE_DEC, 200 },
> +	{ V4L2_PIX_FMT_H264, VIDC_SESSION_TYPE_DEC, 200 },
> +	{ V4L2_PIX_FMT_HEVC, VIDC_SESSION_TYPE_DEC, 200 },
> +	{ V4L2_PIX_FMT_VP8, VIDC_SESSION_TYPE_DEC, 200 },
> +	{ V4L2_PIX_FMT_VP9, VIDC_SESSION_TYPE_DEC, 200 },
> +};
> +
>  static const struct venus_resources sdm845_res = {
>  	.freq_tbl = sdm845_freq_table,
>  	.freq_tbl_size = ARRAY_SIZE(sdm845_freq_table),
> +	.codec_freq_data = sdm845_codec_freq_data,
> +	.codec_freq_data_size = ARRAY_SIZE(sdm845_codec_freq_data),
>  	.clks = {"core", "iface", "bus" },
>  	.clks_num = 3,
>  	.max_load = 2563200,
> diff --git a/drivers/media/platform/qcom/venus/core.h b/drivers/media/platform/qcom/venus/core.h
> index 7a3feb5..2ed6496 100644
> --- a/drivers/media/platform/qcom/venus/core.h
> +++ b/drivers/media/platform/qcom/venus/core.h
> @@ -35,12 +35,20 @@ struct reg_val {
>  	u32 value;
>  };
>  
> +struct codec_freq_data {
> +	u32 pixfmt;
> +	u32 session_type;
> +	unsigned int vpp_freq;

isn't unsigned long more suitable?

> +};
> +
>  struct venus_resources {
>  	u64 dma_mask;
>  	const struct freq_tbl *freq_tbl;
>  	unsigned int freq_tbl_size;
>  	const struct reg_val *reg_tbl;
>  	unsigned int reg_tbl_size;
> +	const struct codec_freq_data *codec_freq_data;
> +	unsigned int codec_freq_data_size;
>  	const char * const clks[VIDC_CLKS_NUM_MAX];
>  	unsigned int clks_num;
>  	enum hfi_version hfi_version;
> @@ -216,6 +224,12 @@ struct venus_buffer {
>  	struct list_head ref_list;
>  };
>  
> +struct clock_data {
> +	u32 core_id;
> +	unsigned long freq;
> +	struct codec_freq_data *codec_freq_data;
> +};
> +
>  #define to_venus_buffer(ptr)	container_of(ptr, struct venus_buffer, vb)
>  
>  /**
> @@ -275,6 +289,7 @@ struct venus_inst {
>  	struct list_head list;
>  	struct mutex lock;
>  	struct venus_core *core;
> +	struct clock_data clk_data;
>  	struct list_head dpbbufs;
>  	struct list_head internalbufs;
>  	struct list_head registeredbufs;
> diff --git a/drivers/media/platform/qcom/venus/helpers.c b/drivers/media/platform/qcom/venus/helpers.c
> index 5cad601..f7f724b 100644
> --- a/drivers/media/platform/qcom/venus/helpers.c
> +++ b/drivers/media/platform/qcom/venus/helpers.c
> @@ -715,6 +715,36 @@ int venus_helper_set_core_usage(struct venus_inst *inst, u32 usage)
>  }
>  EXPORT_SYMBOL_GPL(venus_helper_set_core_usage);
>  
> +int venus_helper_init_codec_data(struct venus_inst *inst)
> +{
> +	const struct codec_data *codec_data;

Something is wrong here, there is no prototype of struct codec_data.

-- 
regards,
Stan

Re: [PATCH v3 1/4] media: venus: Add codec data table

[amasule]

Hi Stan,

On 2019-07-01 17:13, Stanimir Varbanov wrote:
> On 6/25/19 7:27 PM, Aniket Masule wrote:
>> Add vpp cycles for for different types of codec
>> It indicates the cycles required by video hardware
>> to process each macroblock. Initialize the codec
>> data with core resources.
>> 
>> Signed-off-by: Aniket Masule <amasule@codeaurora.org>
>> ---
>>  drivers/media/platform/qcom/venus/core.c    | 13 +++++++++++++
>>  drivers/media/platform/qcom/venus/core.h    | 15 +++++++++++++++
>>  drivers/media/platform/qcom/venus/helpers.c | 30 
>> +++++++++++++++++++++++++++++
>>  drivers/media/platform/qcom/venus/helpers.h |  1 +
>>  drivers/media/platform/qcom/venus/vdec.c    |  4 ++++
>>  drivers/media/platform/qcom/venus/venc.c    |  4 ++++
>>  6 files changed, 67 insertions(+)
>> 
>> diff --git a/drivers/media/platform/qcom/venus/core.c 
>> b/drivers/media/platform/qcom/venus/core.c
>> index 7393667..f1597d6 100644
>> --- a/drivers/media/platform/qcom/venus/core.c
>> +++ b/drivers/media/platform/qcom/venus/core.c
>> @@ -473,9 +473,22 @@ static __maybe_unused int 
>> venus_runtime_resume(struct device *dev)
>>  	{  244800, 100000000 },	/* 1920x1080@30 */
>>  };
>> 
>> +static struct codec_freq_data sdm845_codec_freq_data[] =  {
>> +	{ V4L2_PIX_FMT_H264, VIDC_SESSION_TYPE_ENC, 675 },
>> +	{ V4L2_PIX_FMT_HEVC, VIDC_SESSION_TYPE_ENC, 675 },
>> +	{ V4L2_PIX_FMT_VP8, VIDC_SESSION_TYPE_ENC, 675 },
>> +	{ V4L2_PIX_FMT_MPEG2, VIDC_SESSION_TYPE_DEC, 200 },
>> +	{ V4L2_PIX_FMT_H264, VIDC_SESSION_TYPE_DEC, 200 },
>> +	{ V4L2_PIX_FMT_HEVC, VIDC_SESSION_TYPE_DEC, 200 },
>> +	{ V4L2_PIX_FMT_VP8, VIDC_SESSION_TYPE_DEC, 200 },
>> +	{ V4L2_PIX_FMT_VP9, VIDC_SESSION_TYPE_DEC, 200 },
>> +};
>> +
>>  static const struct venus_resources sdm845_res = {
>>  	.freq_tbl = sdm845_freq_table,
>>  	.freq_tbl_size = ARRAY_SIZE(sdm845_freq_table),
>> +	.codec_freq_data = sdm845_codec_freq_data,
>> +	.codec_freq_data_size = ARRAY_SIZE(sdm845_codec_freq_data),
>>  	.clks = {"core", "iface", "bus" },
>>  	.clks_num = 3,
>>  	.max_load = 2563200,
>> diff --git a/drivers/media/platform/qcom/venus/core.h 
>> b/drivers/media/platform/qcom/venus/core.h
>> index 7a3feb5..2ed6496 100644
>> --- a/drivers/media/platform/qcom/venus/core.h
>> +++ b/drivers/media/platform/qcom/venus/core.h
>> @@ -35,12 +35,20 @@ struct reg_val {
>>  	u32 value;
>>  };
>> 
>> +struct codec_freq_data {
>> +	u32 pixfmt;
>> +	u32 session_type;
>> +	unsigned int vpp_freq;
> 
> isn't unsigned long more suitable?
> 
The hard-coded values for this vpp will be in few hundreds.
So, unsigned int would be fine.
>> +};
>> +
>>  struct venus_resources {
>>  	u64 dma_mask;
>>  	const struct freq_tbl *freq_tbl;
>>  	unsigned int freq_tbl_size;
>>  	const struct reg_val *reg_tbl;
>>  	unsigned int reg_tbl_size;
>> +	const struct codec_freq_data *codec_freq_data;
>> +	unsigned int codec_freq_data_size;
>>  	const char * const clks[VIDC_CLKS_NUM_MAX];
>>  	unsigned int clks_num;
>>  	enum hfi_version hfi_version;
>> @@ -216,6 +224,12 @@ struct venus_buffer {
>>  	struct list_head ref_list;
>>  };
>> 
>> +struct clock_data {
>> +	u32 core_id;
>> +	unsigned long freq;
>> +	struct codec_freq_data *codec_freq_data;
>> +};
>> +
>>  #define to_venus_buffer(ptr)	container_of(ptr, struct venus_buffer, 
>> vb)
>> 
>>  /**
>> @@ -275,6 +289,7 @@ struct venus_inst {
>>  	struct list_head list;
>>  	struct mutex lock;
>>  	struct venus_core *core;
>> +	struct clock_data clk_data;
>>  	struct list_head dpbbufs;
>>  	struct list_head internalbufs;
>>  	struct list_head registeredbufs;
>> diff --git a/drivers/media/platform/qcom/venus/helpers.c 
>> b/drivers/media/platform/qcom/venus/helpers.c
>> index 5cad601..f7f724b 100644
>> --- a/drivers/media/platform/qcom/venus/helpers.c
>> +++ b/drivers/media/platform/qcom/venus/helpers.c
>> @@ -715,6 +715,36 @@ int venus_helper_set_core_usage(struct venus_inst 
>> *inst, u32 usage)
>>  }
>>  EXPORT_SYMBOL_GPL(venus_helper_set_core_usage);
>> 
>> +int venus_helper_init_codec_data(struct venus_inst *inst)
>> +{
>> +	const struct codec_data *codec_data;
> 
> Something is wrong here, there is no prototype of struct codec_data.
Something went wrong during git rebase, will correct this.


Regards,
Aniket

[PATCH v3 2/4] media: venus: Update clock scaling

[Aniket Masule]

Current clock scaling calculations are same for vpu4 and
previous versions. For vpu4, Clock scaling calculations
are updated with cycles/mb. This helps in getting precise
clock required.

Signed-off-by: Aniket Masule <amasule@codeaurora.org>
---
 drivers/media/platform/qcom/venus/helpers.c | 111 ++++++++++++++++++++++++----
 drivers/media/platform/qcom/venus/helpers.h |   2 +-
 drivers/media/platform/qcom/venus/vdec.c    |   2 +-
 drivers/media/platform/qcom/venus/venc.c    |   2 +-
 4 files changed, 99 insertions(+), 18 deletions(-)

diff --git a/drivers/media/platform/qcom/venus/helpers.c b/drivers/media/platform/qcom/venus/helpers.c
index f7f724b..e1a0247 100644
--- a/drivers/media/platform/qcom/venus/helpers.c
+++ b/drivers/media/platform/qcom/venus/helpers.c
@@ -348,8 +348,9 @@ static u32 load_per_type(struct venus_core *core, u32 session_type)
 	return mbs_per_sec;
 }
 
-static int load_scale_clocks(struct venus_core *core)
+static int scale_clocks(struct venus_inst *inst)
 {
+	struct venus_core *core = inst->core;
 	const struct freq_tbl *table = core->res->freq_tbl;
 	unsigned int num_rows = core->res->freq_tbl_size;
 	unsigned long freq = table[0].freq;
@@ -398,6 +399,86 @@ static int load_scale_clocks(struct venus_core *core)
 	return ret;
 }
 
+static unsigned long calculate_vpp_freq(struct venus_inst *inst)
+{
+	unsigned long vpp_freq = 0;
+	u32 mbs_per_sec;
+
+	mbs_per_sec = load_per_instance(inst);
+	vpp_freq = mbs_per_sec * inst->clk_data.codec_freq_data->vpp_freq;
+	/* 21 / 20 is overhead factor */
+	vpp_freq += vpp_freq / 20;
+
+	return vpp_freq;
+}
+
+static int scale_clocks_v4(struct venus_inst *inst)
+{
+	struct venus_core *core = inst->core;
+	const struct freq_tbl *table = core->res->freq_tbl;
+	unsigned int num_rows = core->res->freq_tbl_size;
+
+	struct clk *clk = core->clks[0];
+	struct device *dev = core->dev;
+	unsigned int i;
+	unsigned long freq = 0, freq_core0 = 0, freq_core1 = 0;
+	int ret;
+
+	freq = calculate_vpp_freq(inst);
+
+	if (freq > table[0].freq)
+		goto err;
+
+	for (i = 0; i < num_rows; i++) {
+		if (freq > table[i].freq)
+			break;
+		freq = table[i].freq;
+	}
+
+	inst->clk_data.freq = freq;
+
+	mutex_lock(&core->lock);
+	list_for_each_entry(inst, &core->instances, list) {
+		if (inst->clk_data.core_id == VIDC_CORE_ID_1) {
+			freq_core0 += inst->clk_data.freq;
+		} else if (inst->clk_data.core_id == VIDC_CORE_ID_2) {
+			freq_core1 += inst->clk_data.freq;
+		} else if (inst->clk_data.core_id == VIDC_CORE_ID_3) {
+			freq_core0 += inst->clk_data.freq;
+			freq_core1 += inst->clk_data.freq;
+		}
+	}
+	mutex_unlock(&core->lock);
+
+	freq = max(freq_core0, freq_core1);
+
+	ret = clk_set_rate(clk, freq);
+	if (ret)
+		goto err;
+
+	ret = clk_set_rate(core->core0_clk, freq);
+	if (ret)
+		goto err;
+
+	ret = clk_set_rate(core->core1_clk, freq);
+	if (ret)
+		goto err;
+
+	return 0;
+
+err:
+	dev_err(dev, "failed to set clock rate %lu (%d)\n", freq, ret);
+	return ret;
+}
+
+static int load_scale_clocks(struct venus_inst *inst)
+{
+	if (IS_V4(inst->core))
+		return scale_clocks_v4(inst);
+
+	return scale_clocks(inst);
+}
+
 static void fill_buffer_desc(const struct venus_buffer *buf,
 			     struct hfi_buffer_desc *bd, bool response)
 {
@@ -715,35 +796,36 @@ int venus_helper_set_core_usage(struct venus_inst *inst, u32 usage)
 }
 EXPORT_SYMBOL_GPL(venus_helper_set_core_usage);
 
-int venus_helper_init_codec_data(struct venus_inst *inst)
+int venus_helper_init_codec_freq_data(struct venus_inst *inst)
 {
-	const struct codec_data *codec_data;
-	unsigned int i, codec_data_size;
+	const struct codec_freq_data *codec_freq_data;
+	unsigned int i, codec_freq_data_size;
 	u32 pixfmt;
 	int ret = 0;
 
 	if (!IS_V4(inst->core))
 		return 0;
 
-	codec_data = inst->core->res->codec_data;
-	codec_data_size = inst->core->res->codec_data_size;
+	codec_freq_data = inst->core->res->codec_freq_data;
+	codec_freq_data_size = inst->core->res->codec_freq_data_size;
 	pixfmt = inst->session_type == VIDC_SESSION_TYPE_DEC ?
 			inst->fmt_out->pixfmt : inst->fmt_cap->pixfmt;
 
-	for (i = 0; i < codec_data_size; i++) {
-		if (codec_data[i].pixfmt == pixfmt &&
-		    codec_data[i].session_type == inst->session_type) {
-			inst->clk_data.codec_data = &codec_data[i];
+	for (i = 0; i < codec_freq_data_size; i++) {
+		if (codec_freq_data[i].pixfmt == pixfmt &&
+		    codec_freq_data[i].session_type == inst->session_type) {
+			inst->clk_data.codec_freq_data =
+				&codec_freq_data[i];
 			break;
 		}
 	}
 
-	if (!inst->clk_data.codec_data)
+	if (!inst->clk_data.codec_freq_data)
 		ret = -EINVAL;
 
 	return ret;
 }
-EXPORT_SYMBOL_GPL(venus_helper_init_codec_data);
+EXPORT_SYMBOL_GPL(venus_helper_init_codec_freq_data);
 
 int venus_helper_set_num_bufs(struct venus_inst *inst, unsigned int input_bufs,
 			      unsigned int output_bufs,
@@ -1053,7 +1135,7 @@ void venus_helper_vb2_stop_streaming(struct vb2_queue *q)
 
 		venus_helper_free_dpb_bufs(inst);
 
-		load_scale_clocks(core);
+		load_scale_clocks(inst);
 		INIT_LIST_HEAD(&inst->registeredbufs);
 	}
 
@@ -1070,7 +1152,6 @@ void venus_helper_vb2_stop_streaming(struct vb2_queue *q)
 
 int venus_helper_vb2_start_streaming(struct venus_inst *inst)
 {
-	struct venus_core *core = inst->core;
 	int ret;
 
 	ret = intbufs_alloc(inst);
@@ -1081,7 +1162,7 @@ int venus_helper_vb2_start_streaming(struct venus_inst *inst)
 	if (ret)
 		goto err_bufs_free;
 
-	load_scale_clocks(core);
+	load_scale_clocks(inst);
 
 	ret = hfi_session_load_res(inst);
 	if (ret)
diff --git a/drivers/media/platform/qcom/venus/helpers.h b/drivers/media/platform/qcom/venus/helpers.h
index f9360a8..2c13245 100644
--- a/drivers/media/platform/qcom/venus/helpers.h
+++ b/drivers/media/platform/qcom/venus/helpers.h
@@ -41,7 +41,7 @@ int venus_helper_set_output_resolution(struct venus_inst *inst,
 				       unsigned int width, unsigned int height,
 				       u32 buftype);
 int venus_helper_set_work_mode(struct venus_inst *inst, u32 mode);
-int venus_helper_init_codec_data(struct venus_inst *inst);
+int venus_helper_init_codec_freq_data(struct venus_inst *inst);
 int venus_helper_set_core_usage(struct venus_inst *inst, u32 usage);
 int venus_helper_set_num_bufs(struct venus_inst *inst, unsigned int input_bufs,
 			      unsigned int output_bufs,
diff --git a/drivers/media/platform/qcom/venus/vdec.c b/drivers/media/platform/qcom/venus/vdec.c
index 51795fd..d037f80 100644
--- a/drivers/media/platform/qcom/venus/vdec.c
+++ b/drivers/media/platform/qcom/venus/vdec.c
@@ -660,7 +660,7 @@ static int vdec_init_session(struct venus_inst *inst)
 	if (ret)
 		goto deinit;
 
-	ret = venus_helper_init_codec_data(inst);
+	ret = venus_helper_init_codec_freq_data(inst);
 	if (ret)
 		goto deinit;
 
diff --git a/drivers/media/platform/qcom/venus/venc.c b/drivers/media/platform/qcom/venus/venc.c
index 792cdce..cdddc82 100644
--- a/drivers/media/platform/qcom/venus/venc.c
+++ b/drivers/media/platform/qcom/venus/venc.c
@@ -847,7 +847,7 @@ static int venc_init_session(struct venus_inst *inst)
 	if (ret)
 		goto deinit;
 
-	ret = venus_helper_init_codec_data(inst);
+	ret = venus_helper_init_codec_freq_data(inst);
 	if (ret)
 		goto deinit;
 
-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project

Re: [PATCH v3 2/4] media: venus: Update clock scaling

[Stanimir Varbanov]

On 6/25/19 7:27 PM, Aniket Masule wrote:
> Current clock scaling calculations are same for vpu4 and
> previous versions. For vpu4, Clock scaling calculations
> are updated with cycles/mb. This helps in getting precise
> clock required.
> 
> Signed-off-by: Aniket Masule <amasule@codeaurora.org>
> ---
>  drivers/media/platform/qcom/venus/helpers.c | 111 ++++++++++++++++++++++++----
>  drivers/media/platform/qcom/venus/helpers.h |   2 +-
>  drivers/media/platform/qcom/venus/vdec.c    |   2 +-
>  drivers/media/platform/qcom/venus/venc.c    |   2 +-
>  4 files changed, 99 insertions(+), 18 deletions(-)
> 
> diff --git a/drivers/media/platform/qcom/venus/helpers.c b/drivers/media/platform/qcom/venus/helpers.c
> index f7f724b..e1a0247 100644
> --- a/drivers/media/platform/qcom/venus/helpers.c
> +++ b/drivers/media/platform/qcom/venus/helpers.c
> @@ -348,8 +348,9 @@ static u32 load_per_type(struct venus_core *core, u32 session_type)
>  	return mbs_per_sec;
>  }
>  
> -static int load_scale_clocks(struct venus_core *core)
> +static int scale_clocks(struct venus_inst *inst)
>  {
> +	struct venus_core *core = inst->core;
>  	const struct freq_tbl *table = core->res->freq_tbl;
>  	unsigned int num_rows = core->res->freq_tbl_size;
>  	unsigned long freq = table[0].freq;
> @@ -398,6 +399,86 @@ static int load_scale_clocks(struct venus_core *core)
>  	return ret;
>  }
>  
> +static unsigned long calculate_vpp_freq(struct venus_inst *inst)
> +{
> +	unsigned long vpp_freq = 0;
> +	u32 mbs_per_sec;
> +
> +	mbs_per_sec = load_per_instance(inst);
> +	vpp_freq = mbs_per_sec * inst->clk_data.codec_freq_data->vpp_freq;
> +	/* 21 / 20 is overhead factor */
> +	vpp_freq += vpp_freq / 20;
> +
> +	return vpp_freq;
> +}
> +
> +static int scale_clocks_v4(struct venus_inst *inst)
> +{
> +	struct venus_core *core = inst->core;
> +	const struct freq_tbl *table = core->res->freq_tbl;
> +	unsigned int num_rows = core->res->freq_tbl_size;
> +

please remove this blank line.

> +	struct clk *clk = core->clks[0];
> +	struct device *dev = core->dev;
> +	unsigned int i;
> +	unsigned long freq = 0, freq_core0 = 0, freq_core1 = 0;

could you count the cores as it is done for VIDC_CORE_ID_ ?
i.e. start counting from one.

> +	int ret;
> +
> +	freq = calculate_vpp_freq(inst);
> +
> +	if (freq > table[0].freq)
> +		goto err;

if the goto is triggered the error message will be wrong. Infact the
dev_err message is targeted for clk_set_rate failure.

> +
> +	for (i = 0; i < num_rows; i++) {
> +		if (freq > table[i].freq)
> +			break;
> +		freq = table[i].freq;
> +	}
> +
> +	inst->clk_data.freq = freq;
> +
> +	mutex_lock(&core->lock);
> +	list_for_each_entry(inst, &core->instances, list) {
> +		if (inst->clk_data.core_id == VIDC_CORE_ID_1) {
> +			freq_core0 += inst->clk_data.freq;
> +		} else if (inst->clk_data.core_id == VIDC_CORE_ID_2) {
> +			freq_core1 += inst->clk_data.freq;
> +		} else if (inst->clk_data.core_id == VIDC_CORE_ID_3) {
> +			freq_core0 += inst->clk_data.freq;
> +			freq_core1 += inst->clk_data.freq;
> +		}
> +	}
> +	mutex_unlock(&core->lock);
> +
> +	freq = max(freq_core0, freq_core1);
> +
> +	ret = clk_set_rate(clk, freq);
> +	if (ret)
> +		goto err;
> +
> +	ret = clk_set_rate(core->core0_clk, freq);
> +	if (ret)
> +		goto err;
> +
> +	ret = clk_set_rate(core->core1_clk, freq);
> +	if (ret)
> +		goto err;
> +
> +	return 0;
> +
> +err:
> +	dev_err(dev, "failed to set clock rate %lu (%d)\n", freq, ret);
> +	return ret;
> +}
> +
> +static int load_scale_clocks(struct venus_inst *inst)
> +{
> +	if (IS_V4(inst->core))
> +		return scale_clocks_v4(inst);
> +
> +	return scale_clocks(inst);
> +}
> +
>  static void fill_buffer_desc(const struct venus_buffer *buf,
>  			     struct hfi_buffer_desc *bd, bool response)
>  {
> @@ -715,35 +796,36 @@ int venus_helper_set_core_usage(struct venus_inst *inst, u32 usage)
>  }
>  EXPORT_SYMBOL_GPL(venus_helper_set_core_usage);
>  
> -int venus_helper_init_codec_data(struct venus_inst *inst)
> +int venus_helper_init_codec_freq_data(struct venus_inst *inst)
>  {
> -	const struct codec_data *codec_data;
> -	unsigned int i, codec_data_size;

those deletions shouldn't exist once you fix the git rebase issue.

> +	const struct codec_freq_data *codec_freq_data;
> +	unsigned int i, codec_freq_data_size;

could you rename the variables to shorter?

>  	u32 pixfmt;
>  	int ret = 0;
>  
>  	if (!IS_V4(inst->core))
>  		return 0;
>  
> -	codec_data = inst->core->res->codec_data;
> -	codec_data_size = inst->core->res->codec_data_size;
> +	codec_freq_data = inst->core->res->codec_freq_data;
> +	codec_freq_data_size = inst->core->res->codec_freq_data_size;
>  	pixfmt = inst->session_type == VIDC_SESSION_TYPE_DEC ?
>  			inst->fmt_out->pixfmt : inst->fmt_cap->pixfmt;
>  
> -	for (i = 0; i < codec_data_size; i++) {
> -		if (codec_data[i].pixfmt == pixfmt &&
> -		    codec_data[i].session_type == inst->session_type) {
> -			inst->clk_data.codec_data = &codec_data[i];
> +	for (i = 0; i < codec_freq_data_size; i++) {
> +		if (codec_freq_data[i].pixfmt == pixfmt &&
> +		    codec_freq_data[i].session_type == inst->session_type) {
> +			inst->clk_data.codec_freq_data =
> +				&codec_freq_data[i];
>  			break;
>  		}
>  	}
>  
> -	if (!inst->clk_data.codec_data)
> +	if (!inst->clk_data.codec_freq_data)
>  		ret = -EINVAL;
>  
>  	return ret;
>  }
> -EXPORT_SYMBOL_GPL(venus_helper_init_codec_data);
> +EXPORT_SYMBOL_GPL(venus_helper_init_codec_freq_data);
>  

-- 
regards,
Stan

Re: [PATCH v3 2/4] media: venus: Update clock scaling

[amasule]

Hi Stan,

On 2019-07-01 18:41, Stanimir Varbanov wrote:
> On 6/25/19 7:27 PM, Aniket Masule wrote:
>> Current clock scaling calculations are same for vpu4 and
>> previous versions. For vpu4, Clock scaling calculations
>> are updated with cycles/mb. This helps in getting precise
>> clock required.
>> 
>> Signed-off-by: Aniket Masule <amasule@codeaurora.org>
>> ---
>>  drivers/media/platform/qcom/venus/helpers.c | 111 
>> ++++++++++++++++++++++++----
>>  drivers/media/platform/qcom/venus/helpers.h |   2 +-
>>  drivers/media/platform/qcom/venus/vdec.c    |   2 +-
>>  drivers/media/platform/qcom/venus/venc.c    |   2 +-
>>  4 files changed, 99 insertions(+), 18 deletions(-)
>> 
>> diff --git a/drivers/media/platform/qcom/venus/helpers.c 
>> b/drivers/media/platform/qcom/venus/helpers.c
>> index f7f724b..e1a0247 100644
>> --- a/drivers/media/platform/qcom/venus/helpers.c
>> +++ b/drivers/media/platform/qcom/venus/helpers.c
>> @@ -348,8 +348,9 @@ static u32 load_per_type(struct venus_core *core, 
>> u32 session_type)
>>  	return mbs_per_sec;
>>  }
>> 
>> -static int load_scale_clocks(struct venus_core *core)
>> +static int scale_clocks(struct venus_inst *inst)
>>  {
>> +	struct venus_core *core = inst->core;
>>  	const struct freq_tbl *table = core->res->freq_tbl;
>>  	unsigned int num_rows = core->res->freq_tbl_size;
>>  	unsigned long freq = table[0].freq;
>> @@ -398,6 +399,86 @@ static int load_scale_clocks(struct venus_core 
>> *core)
>>  	return ret;
>>  }
>> 
>> +static unsigned long calculate_vpp_freq(struct venus_inst *inst)
>> +{
>> +	unsigned long vpp_freq = 0;
>> +	u32 mbs_per_sec;
>> +
>> +	mbs_per_sec = load_per_instance(inst);
>> +	vpp_freq = mbs_per_sec * inst->clk_data.codec_freq_data->vpp_freq;
>> +	/* 21 / 20 is overhead factor */
>> +	vpp_freq += vpp_freq / 20;
>> +
>> +	return vpp_freq;
>> +}
>> +
>> +static int scale_clocks_v4(struct venus_inst *inst)
>> +{
>> +	struct venus_core *core = inst->core;
>> +	const struct freq_tbl *table = core->res->freq_tbl;
>> +	unsigned int num_rows = core->res->freq_tbl_size;
>> +
> 
> please remove this blank line.
> 
>> +	struct clk *clk = core->clks[0];
>> +	struct device *dev = core->dev;
>> +	unsigned int i;
>> +	unsigned long freq = 0, freq_core0 = 0, freq_core1 = 0;
> 
> could you count the cores as it is done for VIDC_CORE_ID_ ?
> i.e. start counting from one.
> 
Sure, I was aligning it with clock handle name in core, but
aligning it with VIDC_CORE_ID_ would give good readability.
>> +	int ret;
>> +
>> +	freq = calculate_vpp_freq(inst);
>> +
>> +	if (freq > table[0].freq)
>> +		goto err;
> 
> if the goto is triggered the error message will be wrong. Infact the
> dev_err message is targeted for clk_set_rate failure.
> 
I will add separate exit for this.
>> +
>> +	for (i = 0; i < num_rows; i++) {
>> +		if (freq > table[i].freq)
>> +			break;
>> +		freq = table[i].freq;
>> +	}
>> +
>> +	inst->clk_data.freq = freq;
>> +
>> +	mutex_lock(&core->lock);
>> +	list_for_each_entry(inst, &core->instances, list) {
>> +		if (inst->clk_data.core_id == VIDC_CORE_ID_1) {
>> +			freq_core0 += inst->clk_data.freq;
>> +		} else if (inst->clk_data.core_id == VIDC_CORE_ID_2) {
>> +			freq_core1 += inst->clk_data.freq;
>> +		} else if (inst->clk_data.core_id == VIDC_CORE_ID_3) {
>> +			freq_core0 += inst->clk_data.freq;
>> +			freq_core1 += inst->clk_data.freq;
>> +		}
>> +	}
>> +	mutex_unlock(&core->lock);
>> +
>> +	freq = max(freq_core0, freq_core1);
>> +
>> +	ret = clk_set_rate(clk, freq);
>> +	if (ret)
>> +		goto err;
>> +
>> +	ret = clk_set_rate(core->core0_clk, freq);
>> +	if (ret)
>> +		goto err;
>> +
>> +	ret = clk_set_rate(core->core1_clk, freq);
>> +	if (ret)
>> +		goto err;
>> +
>> +	return 0;
>> +
>> +err:
>> +	dev_err(dev, "failed to set clock rate %lu (%d)\n", freq, ret);
>> +	return ret;
>> +}
>> +
>> +static int load_scale_clocks(struct venus_inst *inst)
>> +{
>> +	if (IS_V4(inst->core))
>> +		return scale_clocks_v4(inst);
>> +
>> +	return scale_clocks(inst);
>> +}
>> +
>>  static void fill_buffer_desc(const struct venus_buffer *buf,
>>  			     struct hfi_buffer_desc *bd, bool response)
>>  {
>> @@ -715,35 +796,36 @@ int venus_helper_set_core_usage(struct 
>> venus_inst *inst, u32 usage)
>>  }
>>  EXPORT_SYMBOL_GPL(venus_helper_set_core_usage);
>> 
>> -int venus_helper_init_codec_data(struct venus_inst *inst)
>> +int venus_helper_init_codec_freq_data(struct venus_inst *inst)
>>  {
>> -	const struct codec_data *codec_data;
>> -	unsigned int i, codec_data_size;
> 
> those deletions shouldn't exist once you fix the git rebase issue.
> 
>> +	const struct codec_freq_data *codec_freq_data;
>> +	unsigned int i, codec_freq_data_size;
> 
> could you rename the variables to shorter?
> 
Yes
>>  	u32 pixfmt;
>>  	int ret = 0;
>> 
>>  	if (!IS_V4(inst->core))
>>  		return 0;
>> 
>> -	codec_data = inst->core->res->codec_data;
>> -	codec_data_size = inst->core->res->codec_data_size;
>> +	codec_freq_data = inst->core->res->codec_freq_data;
>> +	codec_freq_data_size = inst->core->res->codec_freq_data_size;
>>  	pixfmt = inst->session_type == VIDC_SESSION_TYPE_DEC ?
>>  			inst->fmt_out->pixfmt : inst->fmt_cap->pixfmt;
>> 
>> -	for (i = 0; i < codec_data_size; i++) {
>> -		if (codec_data[i].pixfmt == pixfmt &&
>> -		    codec_data[i].session_type == inst->session_type) {
>> -			inst->clk_data.codec_data = &codec_data[i];
>> +	for (i = 0; i < codec_freq_data_size; i++) {
>> +		if (codec_freq_data[i].pixfmt == pixfmt &&
>> +		    codec_freq_data[i].session_type == inst->session_type) {
>> +			inst->clk_data.codec_freq_data =
>> +				&codec_freq_data[i];
>>  			break;
>>  		}
>>  	}
>> 
>> -	if (!inst->clk_data.codec_data)
>> +	if (!inst->clk_data.codec_freq_data)
>>  		ret = -EINVAL;
>> 
>>  	return ret;
>>  }
>> -EXPORT_SYMBOL_GPL(venus_helper_init_codec_data);
>> +EXPORT_SYMBOL_GPL(venus_helper_init_codec_freq_data);
>> 

Thanks,
Aniket

[PATCH v3 3/4] media: venus: Add interface for load per core

[Aniket Masule]

Add and interface to calculate load per core. Also,
add an interface to get maximum cores available with
video. This interface is preparation for updating core
selection.

Signed-off-by: Aniket Masule <amasule@codeaurora.org>
---
 drivers/media/platform/qcom/venus/helpers.c    | 19 +++++++++++++++++++
 drivers/media/platform/qcom/venus/hfi_helper.h |  1 +
 drivers/media/platform/qcom/venus/hfi_parser.h |  5 +++++
 3 files changed, 25 insertions(+)

diff --git a/drivers/media/platform/qcom/venus/helpers.c b/drivers/media/platform/qcom/venus/helpers.c
index e1a0247..b79e83a 100644
--- a/drivers/media/platform/qcom/venus/helpers.c
+++ b/drivers/media/platform/qcom/venus/helpers.c
@@ -26,6 +26,7 @@
 #include "helpers.h"
 #include "hfi_helper.h"
 #include "hfi_venus_io.h"
+#include "hfi_parser.h"
 
 struct intbuf {
 	struct list_head list;
@@ -331,6 +332,24 @@ static u32 load_per_instance(struct venus_inst *inst)
 	return mbs * inst->fps;
 }
 
+static u32 load_per_core(struct venus_core *core, u32 core_id)
+{
+	struct venus_inst *inst = NULL;
+	u32 mbs_per_sec = 0, load = 0;
+
+	mutex_lock(&core->lock);
+	list_for_each_entry(inst, &core->instances, list) {
+		if (!(inst->clk_data.core_id == core_id))
+			continue;
+
+		mbs_per_sec = load_per_instance(inst);
+		load = mbs_per_sec * inst->clk_data.codec_freq_data->vpp_freq;
+	}
+	mutex_unlock(&core->lock);
+
+	return load;
+}
+
 static u32 load_per_type(struct venus_core *core, u32 session_type)
 {
 	struct venus_inst *inst = NULL;
diff --git a/drivers/media/platform/qcom/venus/hfi_helper.h b/drivers/media/platform/qcom/venus/hfi_helper.h
index 34ea503..f3d1018 100644
--- a/drivers/media/platform/qcom/venus/hfi_helper.h
+++ b/drivers/media/platform/qcom/venus/hfi_helper.h
@@ -559,6 +559,7 @@ struct hfi_bitrate {
 #define HFI_CAPABILITY_LCU_SIZE				0x14
 #define HFI_CAPABILITY_HIER_P_HYBRID_NUM_ENH_LAYERS	0x15
 #define HFI_CAPABILITY_MBS_PER_SECOND_POWERSAVE		0x16
+#define HFI_CAPABILITY_MAX_VIDEOCORES		0x2B
 
 struct hfi_capability {
 	u32 capability_type;
diff --git a/drivers/media/platform/qcom/venus/hfi_parser.h b/drivers/media/platform/qcom/venus/hfi_parser.h
index 3e931c7..264e6dd 100644
--- a/drivers/media/platform/qcom/venus/hfi_parser.h
+++ b/drivers/media/platform/qcom/venus/hfi_parser.h
@@ -107,4 +107,9 @@ static inline u32 frate_step(struct venus_inst *inst)
 	return cap_step(inst, HFI_CAPABILITY_FRAMERATE);
 }
 
+static inline u32 core_num_max(struct venus_inst *inst)
+{
+	return cap_max(inst, HFI_CAPABILITY_MAX_VIDEOCORES);
+}
+
 #endif
-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project

Re: [PATCH v3 3/4] media: venus: Add interface for load per core

[Stanimir Varbanov]

Hi Aniket,

On 6/25/19 7:27 PM, Aniket Masule wrote:
> Add and interface to calculate load per core. Also,
> add an interface to get maximum cores available with
> video. This interface is preparation for updating core
> selection.
> 
> Signed-off-by: Aniket Masule <amasule@codeaurora.org>
> ---
>  drivers/media/platform/qcom/venus/helpers.c    | 19 +++++++++++++++++++
>  drivers/media/platform/qcom/venus/hfi_helper.h |  1 +
>  drivers/media/platform/qcom/venus/hfi_parser.h |  5 +++++
>  3 files changed, 25 insertions(+)
> 
> diff --git a/drivers/media/platform/qcom/venus/helpers.c b/drivers/media/platform/qcom/venus/helpers.c
> index e1a0247..b79e83a 100644
> --- a/drivers/media/platform/qcom/venus/helpers.c
> +++ b/drivers/media/platform/qcom/venus/helpers.c
> @@ -26,6 +26,7 @@
>  #include "helpers.h"
>  #include "hfi_helper.h"
>  #include "hfi_venus_io.h"
> +#include "hfi_parser.h"
>  
>  struct intbuf {
>  	struct list_head list;
> @@ -331,6 +332,24 @@ static u32 load_per_instance(struct venus_inst *inst)
>  	return mbs * inst->fps;
>  }
>  
> +static u32 load_per_core(struct venus_core *core, u32 core_id)
> +{
> +	struct venus_inst *inst = NULL;
> +	u32 mbs_per_sec = 0, load = 0;
> +
> +	mutex_lock(&core->lock);
> +	list_for_each_entry(inst, &core->instances, list) {
> +		if (!(inst->clk_data.core_id == core_id))
> +			continue;
> +
> +		mbs_per_sec = load_per_instance(inst);
> +		load = mbs_per_sec * inst->clk_data.codec_freq_data->vpp_freq;
> +	}
> +	mutex_unlock(&core->lock);
> +
> +	return load;
> +}
> +

Please squash this patch with the next one where load_per_core() is used
to avoid compiler warnings.

-- 
regards,
Stan

[PATCH v3 4/4] media: venus: Update core selection

[Aniket Masule]

Present core assignment is static. Introduced load balancing
across the cores. Load on earch core is calculated and core
with minimum load is assigned to given instance.

Signed-off-by: Aniket Masule <amasule@codeaurora.org>
---
 drivers/media/platform/qcom/venus/helpers.c | 52 +++++++++++++++++++++++++----
 drivers/media/platform/qcom/venus/helpers.h |  2 +-
 drivers/media/platform/qcom/venus/vdec.c    |  2 +-
 drivers/media/platform/qcom/venus/venc.c    |  2 +-
 4 files changed, 49 insertions(+), 9 deletions(-)

diff --git a/drivers/media/platform/qcom/venus/helpers.c b/drivers/media/platform/qcom/venus/helpers.c
index b79e83a..ef35fd8 100644
--- a/drivers/media/platform/qcom/venus/helpers.c
+++ b/drivers/media/platform/qcom/venus/helpers.c
@@ -498,6 +498,16 @@ static int load_scale_clocks(struct venus_inst *inst)
 	return scale_clocks(inst);
 }
 
+int set_core_usage(struct venus_inst *inst, u32 usage)
+{
+	const u32 ptype = HFI_PROPERTY_CONFIG_VIDEOCORES_USAGE;
+	struct hfi_videocores_usage_type cu;
+
+	cu.video_core_enable_mask = usage;
+
+	return hfi_session_set_property(inst, ptype, &cu);
+}
+
 static void fill_buffer_desc(const struct venus_buffer *buf,
 			     struct hfi_buffer_desc *bd, bool response)
 {
@@ -801,19 +811,49 @@ int venus_helper_set_work_mode(struct venus_inst *inst, u32 mode)
 }
 EXPORT_SYMBOL_GPL(venus_helper_set_work_mode);
 
-int venus_helper_set_core_usage(struct venus_inst *inst, u32 usage)
+int venus_helper_set_core(struct venus_inst *inst)
 {
-	const u32 ptype = HFI_PROPERTY_CONFIG_VIDEOCORES_USAGE;
-	struct hfi_videocores_usage_type cu;
+	struct venus_core *core = inst->core;
+	u32 min_core_id = 0, core0_load = 0, core1_load = 0;
+	unsigned long min_load, max_freq, cur_inst_load;
+	u32 cores_max;
+	int ret;
 
 	if (!IS_V4(inst->core))
 		return 0;
 
-	cu.video_core_enable_mask = usage;
+	core0_load = load_per_core(core, VIDC_CORE_ID_1);
+	core1_load = load_per_core(core, VIDC_CORE_ID_2);
 
-	return hfi_session_set_property(inst, ptype, &cu);
+	min_core_id = core0_load < core1_load ? VIDC_CORE_ID_1 : VIDC_CORE_ID_2;
+	min_load = min(core0_load, core1_load);
+	cores_max = core_num_max(inst);
+
+	if (cores_max < VIDC_CORE_ID_2) {
+		min_core_id = VIDC_CORE_ID_1;
+		min_load = core0_load;
+	}
+
+	cur_inst_load = load_per_instance(inst) *
+		inst->clk_data.codec_freq_data->vpp_freq;
+	max_freq = core->res->freq_tbl[0].freq;
+
+	if ((cur_inst_load + min_load)	> max_freq) {
+		dev_warn(core->dev, "HW is overloaded, needed: %lu max: %lu\n",
+			 cur_inst_load, max_freq);
+		return -EINVAL;
+	}
+
+	ret = set_core_usage(inst, min_core_id);
+
+	if (ret)
+		return ret;
+
+	inst->clk_data.core_id = min_core_id;
+
+	return 0;
 }
-EXPORT_SYMBOL_GPL(venus_helper_set_core_usage);
+EXPORT_SYMBOL_GPL(venus_helper_set_core);
 
 int venus_helper_init_codec_freq_data(struct venus_inst *inst)
 {
diff --git a/drivers/media/platform/qcom/venus/helpers.h b/drivers/media/platform/qcom/venus/helpers.h
index 2c13245..1034111 100644
--- a/drivers/media/platform/qcom/venus/helpers.h
+++ b/drivers/media/platform/qcom/venus/helpers.h
@@ -42,7 +42,7 @@ int venus_helper_set_output_resolution(struct venus_inst *inst,
 				       u32 buftype);
 int venus_helper_set_work_mode(struct venus_inst *inst, u32 mode);
 int venus_helper_init_codec_freq_data(struct venus_inst *inst);
-int venus_helper_set_core_usage(struct venus_inst *inst, u32 usage);
+int venus_helper_set_core(struct venus_inst *inst);
 int venus_helper_set_num_bufs(struct venus_inst *inst, unsigned int input_bufs,
 			      unsigned int output_bufs,
 			      unsigned int output2_bufs);
diff --git a/drivers/media/platform/qcom/venus/vdec.c b/drivers/media/platform/qcom/venus/vdec.c
index d037f80..620e060 100644
--- a/drivers/media/platform/qcom/venus/vdec.c
+++ b/drivers/media/platform/qcom/venus/vdec.c
@@ -551,7 +551,7 @@ static int vdec_output_conf(struct venus_inst *inst)
 	if (ret)
 		return ret;
 
-	ret = venus_helper_set_core_usage(inst, VIDC_CORE_ID_1);
+	ret = venus_helper_set_core(inst);
 	if (ret)
 		return ret;
 
diff --git a/drivers/media/platform/qcom/venus/venc.c b/drivers/media/platform/qcom/venus/venc.c
index cdddc82..28e76cc 100644
--- a/drivers/media/platform/qcom/venus/venc.c
+++ b/drivers/media/platform/qcom/venus/venc.c
@@ -660,7 +660,7 @@ static int venc_set_properties(struct venus_inst *inst)
 	if (ret)
 		return ret;
 
-	ret = venus_helper_set_core_usage(inst, VIDC_CORE_ID_2);
+	ret = venus_helper_set_core(inst);
 	if (ret)
 		return ret;
 
-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project

Re: [PATCH v3 4/4] media: venus: Update core selection

[Stanimir Varbanov]

Hi,

On 6/25/19 7:27 PM, Aniket Masule wrote:
> Present core assignment is static. Introduced load balancing
> across the cores. Load on earch core is calculated and core
> with minimum load is assigned to given instance.
> 
> Signed-off-by: Aniket Masule <amasule@codeaurora.org>
> ---
>  drivers/media/platform/qcom/venus/helpers.c | 52 +++++++++++++++++++++++++----
>  drivers/media/platform/qcom/venus/helpers.h |  2 +-
>  drivers/media/platform/qcom/venus/vdec.c    |  2 +-
>  drivers/media/platform/qcom/venus/venc.c    |  2 +-
>  4 files changed, 49 insertions(+), 9 deletions(-)
> 
> diff --git a/drivers/media/platform/qcom/venus/helpers.c b/drivers/media/platform/qcom/venus/helpers.c
> index b79e83a..ef35fd8 100644
> --- a/drivers/media/platform/qcom/venus/helpers.c
> +++ b/drivers/media/platform/qcom/venus/helpers.c
> @@ -498,6 +498,16 @@ static int load_scale_clocks(struct venus_inst *inst)
>  	return scale_clocks(inst);
>  }
>  
> +int set_core_usage(struct venus_inst *inst, u32 usage)
> +{
> +	const u32 ptype = HFI_PROPERTY_CONFIG_VIDEOCORES_USAGE;
> +	struct hfi_videocores_usage_type cu;
> +
> +	cu.video_core_enable_mask = usage;
> +
> +	return hfi_session_set_property(inst, ptype, &cu);
> +}
> +
>  static void fill_buffer_desc(const struct venus_buffer *buf,
>  			     struct hfi_buffer_desc *bd, bool response)
>  {
> @@ -801,19 +811,49 @@ int venus_helper_set_work_mode(struct venus_inst *inst, u32 mode)
>  }
>  EXPORT_SYMBOL_GPL(venus_helper_set_work_mode);
>  
> -int venus_helper_set_core_usage(struct venus_inst *inst, u32 usage)
> +int venus_helper_set_core(struct venus_inst *inst)
>  {
> -	const u32 ptype = HFI_PROPERTY_CONFIG_VIDEOCORES_USAGE;
> -	struct hfi_videocores_usage_type cu;
> +	struct venus_core *core = inst->core;
> +	u32 min_core_id = 0, core0_load = 0, core1_load = 0;

the same comment, please use the same counting scheme as for VIDC_CORE_ID_

> +	unsigned long min_load, max_freq, cur_inst_load;
> +	u32 cores_max;
> +	int ret;
>  
>  	if (!IS_V4(inst->core))
>  		return 0;
>  
> -	cu.video_core_enable_mask = usage;
> +	core0_load = load_per_core(core, VIDC_CORE_ID_1);
> +	core1_load = load_per_core(core, VIDC_CORE_ID_2);
>  
> -	return hfi_session_set_property(inst, ptype, &cu);
> +	min_core_id = core0_load < core1_load ? VIDC_CORE_ID_1 : VIDC_CORE_ID_2;
> +	min_load = min(core0_load, core1_load);
> +	cores_max = core_num_max(inst);
> +
> +	if (cores_max < VIDC_CORE_ID_2) {
> +		min_core_id = VIDC_CORE_ID_1;
> +		min_load = core0_load;
> +	}
> +
> +	cur_inst_load = load_per_instance(inst) *
> +		inst->clk_data.codec_freq_data->vpp_freq;
> +	max_freq = core->res->freq_tbl[0].freq;
> +
> +	if ((cur_inst_load + min_load)	> max_freq) {
> +		dev_warn(core->dev, "HW is overloaded, needed: %lu max: %lu\n",
> +			 cur_inst_load, max_freq);
> +		return -EINVAL;
> +	}
> +
> +	ret = set_core_usage(inst, min_core_id);
> +

please, delete this blank line

> +	if (ret)
> +		return ret;
> +
> +	inst->clk_data.core_id = min_core_id;
> +
> +	return 0;
>  }
> -EXPORT_SYMBOL_GPL(venus_helper_set_core_usage);
> +EXPORT_SYMBOL_GPL(venus_helper_set_core);
>  
>  int venus_helper_init_codec_freq_data(struct venus_inst *inst)
>  {
> diff --git a/drivers/media/platform/qcom/venus/helpers.h b/drivers/media/platform/qcom/venus/helpers.h
> index 2c13245..1034111 100644
> --- a/drivers/media/platform/qcom/venus/helpers.h
> +++ b/drivers/media/platform/qcom/venus/helpers.h
> @@ -42,7 +42,7 @@ int venus_helper_set_output_resolution(struct venus_inst *inst,
>  				       u32 buftype);
>  int venus_helper_set_work_mode(struct venus_inst *inst, u32 mode);
>  int venus_helper_init_codec_freq_data(struct venus_inst *inst);
> -int venus_helper_set_core_usage(struct venus_inst *inst, u32 usage);
> +int venus_helper_set_core(struct venus_inst *inst);
>  int venus_helper_set_num_bufs(struct venus_inst *inst, unsigned int input_bufs,
>  			      unsigned int output_bufs,
>  			      unsigned int output2_bufs);
> diff --git a/drivers/media/platform/qcom/venus/vdec.c b/drivers/media/platform/qcom/venus/vdec.c
> index d037f80..620e060 100644
> --- a/drivers/media/platform/qcom/venus/vdec.c
> +++ b/drivers/media/platform/qcom/venus/vdec.c
> @@ -551,7 +551,7 @@ static int vdec_output_conf(struct venus_inst *inst)
>  	if (ret)
>  		return ret;
>  
> -	ret = venus_helper_set_core_usage(inst, VIDC_CORE_ID_1);
> +	ret = venus_helper_set_core(inst);
>  	if (ret)
>  		return ret;
>  
> diff --git a/drivers/media/platform/qcom/venus/venc.c b/drivers/media/platform/qcom/venus/venc.c
> index cdddc82..28e76cc 100644
> --- a/drivers/media/platform/qcom/venus/venc.c
> +++ b/drivers/media/platform/qcom/venus/venc.c
> @@ -660,7 +660,7 @@ static int venc_set_properties(struct venus_inst *inst)
>  	if (ret)
>  		return ret;
>  
> -	ret = venus_helper_set_core_usage(inst, VIDC_CORE_ID_2);
> +	ret = venus_helper_set_core(inst);
>  	if (ret)
>  		return ret;
>  
> 

-- 
regards,
Stan

Re: [PATCH v3 4/4] media: venus: Update core selection

[amasule]

Hi Stan,

On 2019-07-01 19:28, Stanimir Varbanov wrote:
> Hi,
> 
> On 6/25/19 7:27 PM, Aniket Masule wrote:
>> Present core assignment is static. Introduced load balancing
>> across the cores. Load on earch core is calculated and core
>> with minimum load is assigned to given instance.
>> 
>> Signed-off-by: Aniket Masule <amasule@codeaurora.org>
>> ---
>>  drivers/media/platform/qcom/venus/helpers.c | 52 
>> +++++++++++++++++++++++++----
>>  drivers/media/platform/qcom/venus/helpers.h |  2 +-
>>  drivers/media/platform/qcom/venus/vdec.c    |  2 +-
>>  drivers/media/platform/qcom/venus/venc.c    |  2 +-
>>  4 files changed, 49 insertions(+), 9 deletions(-)
>> 
>> diff --git a/drivers/media/platform/qcom/venus/helpers.c 
>> b/drivers/media/platform/qcom/venus/helpers.c
>> index b79e83a..ef35fd8 100644
>> --- a/drivers/media/platform/qcom/venus/helpers.c
>> +++ b/drivers/media/platform/qcom/venus/helpers.c
>> @@ -498,6 +498,16 @@ static int load_scale_clocks(struct venus_inst 
>> *inst)
>>  	return scale_clocks(inst);
>>  }
>> 
>> +int set_core_usage(struct venus_inst *inst, u32 usage)
>> +{
>> +	const u32 ptype = HFI_PROPERTY_CONFIG_VIDEOCORES_USAGE;
>> +	struct hfi_videocores_usage_type cu;
>> +
>> +	cu.video_core_enable_mask = usage;
>> +
>> +	return hfi_session_set_property(inst, ptype, &cu);
>> +}
>> +
>>  static void fill_buffer_desc(const struct venus_buffer *buf,
>>  			     struct hfi_buffer_desc *bd, bool response)
>>  {
>> @@ -801,19 +811,49 @@ int venus_helper_set_work_mode(struct venus_inst 
>> *inst, u32 mode)
>>  }
>>  EXPORT_SYMBOL_GPL(venus_helper_set_work_mode);
>> 
>> -int venus_helper_set_core_usage(struct venus_inst *inst, u32 usage)
>> +int venus_helper_set_core(struct venus_inst *inst)
>>  {
>> -	const u32 ptype = HFI_PROPERTY_CONFIG_VIDEOCORES_USAGE;
>> -	struct hfi_videocores_usage_type cu;
>> +	struct venus_core *core = inst->core;
>> +	u32 min_core_id = 0, core0_load = 0, core1_load = 0;
> 
> the same comment, please use the same counting scheme as for 
> VIDC_CORE_ID_
> 
Yes, I will align this with VIDC_CORE_ID_.
>> +	unsigned long min_load, max_freq, cur_inst_load;
>> +	u32 cores_max;
>> +	int ret;
>> 
>>  	if (!IS_V4(inst->core))
>>  		return 0;
>> 
>> -	cu.video_core_enable_mask = usage;
>> +	core0_load = load_per_core(core, VIDC_CORE_ID_1);
>> +	core1_load = load_per_core(core, VIDC_CORE_ID_2);
>> 
>> -	return hfi_session_set_property(inst, ptype, &cu);
>> +	min_core_id = core0_load < core1_load ? VIDC_CORE_ID_1 : 
>> VIDC_CORE_ID_2;
>> +	min_load = min(core0_load, core1_load);
>> +	cores_max = core_num_max(inst);
>> +
>> +	if (cores_max < VIDC_CORE_ID_2) {
>> +		min_core_id = VIDC_CORE_ID_1;
>> +		min_load = core0_load;
>> +	}
>> +
>> +	cur_inst_load = load_per_instance(inst) *
>> +		inst->clk_data.codec_freq_data->vpp_freq;
>> +	max_freq = core->res->freq_tbl[0].freq;
>> +
>> +	if ((cur_inst_load + min_load)	> max_freq) {
>> +		dev_warn(core->dev, "HW is overloaded, needed: %lu max: %lu\n",
>> +			 cur_inst_load, max_freq);
>> +		return -EINVAL;
>> +	}
>> +
>> +	ret = set_core_usage(inst, min_core_id);
>> +
> 
> please, delete this blank line
> 
>> +	if (ret)
>> +		return ret;
>> +
>> +	inst->clk_data.core_id = min_core_id;
>> +
>> +	return 0;
>>  }
>> -EXPORT_SYMBOL_GPL(venus_helper_set_core_usage);
>> +EXPORT_SYMBOL_GPL(venus_helper_set_core);
>> 
>>  int venus_helper_init_codec_freq_data(struct venus_inst *inst)
>>  {
>> diff --git a/drivers/media/platform/qcom/venus/helpers.h 
>> b/drivers/media/platform/qcom/venus/helpers.h
>> index 2c13245..1034111 100644
>> --- a/drivers/media/platform/qcom/venus/helpers.h
>> +++ b/drivers/media/platform/qcom/venus/helpers.h
>> @@ -42,7 +42,7 @@ int venus_helper_set_output_resolution(struct 
>> venus_inst *inst,
>>  				       u32 buftype);
>>  int venus_helper_set_work_mode(struct venus_inst *inst, u32 mode);
>>  int venus_helper_init_codec_freq_data(struct venus_inst *inst);
>> -int venus_helper_set_core_usage(struct venus_inst *inst, u32 usage);
>> +int venus_helper_set_core(struct venus_inst *inst);
>>  int venus_helper_set_num_bufs(struct venus_inst *inst, unsigned int 
>> input_bufs,
>>  			      unsigned int output_bufs,
>>  			      unsigned int output2_bufs);
>> diff --git a/drivers/media/platform/qcom/venus/vdec.c 
>> b/drivers/media/platform/qcom/venus/vdec.c
>> index d037f80..620e060 100644
>> --- a/drivers/media/platform/qcom/venus/vdec.c
>> +++ b/drivers/media/platform/qcom/venus/vdec.c
>> @@ -551,7 +551,7 @@ static int vdec_output_conf(struct venus_inst 
>> *inst)
>>  	if (ret)
>>  		return ret;
>> 
>> -	ret = venus_helper_set_core_usage(inst, VIDC_CORE_ID_1);
>> +	ret = venus_helper_set_core(inst);
>>  	if (ret)
>>  		return ret;
>> 
>> diff --git a/drivers/media/platform/qcom/venus/venc.c 
>> b/drivers/media/platform/qcom/venus/venc.c
>> index cdddc82..28e76cc 100644
>> --- a/drivers/media/platform/qcom/venus/venc.c
>> +++ b/drivers/media/platform/qcom/venus/venc.c
>> @@ -660,7 +660,7 @@ static int venc_set_properties(struct venus_inst 
>> *inst)
>>  	if (ret)
>>  		return ret;
>> 
>> -	ret = venus_helper_set_core_usage(inst, VIDC_CORE_ID_2);
>> +	ret = venus_helper_set_core(inst);
>>  	if (ret)
>>  		return ret;
>> 
>> 

Regards,
Aniket