Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit 249679d8 authored by Praneeth Paladugu's avatar Praneeth Paladugu
Browse files

msm: vidc: Fine tune PMS with on-target profiling 



In Performance Monitoring System, before making clock decision
always follow the below order.

    - If client is starving for buffers, Increase clock.
    - If HW pending count is less, Decrease clock.
    - If none of above, vote for required clock.

Along with this, sanitize CF values from HW for better PMS algo.

CRs-Fixed: 2012520
Change-Id: I31822918013aebe60d7fdac9f91d4cca7bc92345
Signed-off-by: default avatarPraneeth Paladugu <ppaladug@codeaurora.org>
parent e3095a7e

drivers/media/platform/msm/vidc/msm_vidc_clocks.c

+40 −12
Original line number Diff line number Diff line
@@ -16,8 +16,8 @@ 
#include "msm_vidc_debug.h"

#include "msm_vidc_clocks.h"



#define MSM_VIDC_MIN_UBWC_COMPLEXITY_FACTOR 1

#define MSM_VIDC_MAX_UBWC_COMPLEXITY_FACTOR 4

#define MSM_VIDC_MIN_UBWC_COMPLEXITY_FACTOR (1 << 16)

#define MSM_VIDC_MAX_UBWC_COMPLEXITY_FACTOR (4 << 16)



static inline unsigned long int get_ubwc_compression_ratio(

	struct ubwc_cr_stats_info_type ubwc_stats_info)
@@ -104,8 +104,12 @@ static int fill_recon_stats(struct msm_vidc_inst *inst, 
		max_cf = max(max_cf, binfo->CF);

	}

	mutex_unlock(&inst->reconbufs.lock);

	vote_data->compression_ratio = CR;



	/* Sanitize CF values from HW . */

	max_cf = min_t(u32, max_cf, MSM_VIDC_MAX_UBWC_COMPLEXITY_FACTOR);

	min_cf = max_t(u32, min_cf, MSM_VIDC_MIN_UBWC_COMPLEXITY_FACTOR);



	vote_data->compression_ratio = CR;

	vote_data->complexity_factor = max_cf;

	vote_data->use_dpb_read = false;

	if (inst->clk_data.load <= inst->clk_data.load_norm) {
@@ -114,7 +118,7 @@ static int fill_recon_stats(struct msm_vidc_inst *inst, 
	}



	dprintk(VIDC_DBG,

		"Complression Ratio = %d Complexity Factor = %d\n",

		"Compression Ratio = %d Complexity Factor = %d\n",

			vote_data->compression_ratio,

			vote_data->complexity_factor);
@@ -160,7 +164,8 @@ int msm_comm_vote_bus(struct msm_vidc_core *core) 
		list_for_each_entry_safe(temp, next,

				&inst->registeredbufs.list, list) {

			if (temp->vvb.vb2_buf.type ==

				V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE) {

				V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE &&

					temp->deferred) {

				filled_len = max(filled_len,

					temp->vvb.vb2_buf.planes[0].bytesused);

				device_addr = temp->smem[0].device_addr;
@@ -260,6 +265,7 @@ static int msm_dcvs_scale_clocks(struct msm_vidc_inst *inst) 
	int rc = 0;

	int fw_pending_bufs = 0;

	int total_output_buf = 0;

	int min_output_buf = 0;

	int buffers_outside_fw = 0;

	struct msm_vidc_core *core;

	struct hal_buffer_requirements *output_buf_req;
@@ -294,16 +300,37 @@ static int msm_dcvs_scale_clocks(struct msm_vidc_inst *inst) 
	/* Total number of output buffers */

	total_output_buf = output_buf_req->buffer_count_actual;



	min_output_buf = output_buf_req->buffer_count_min;



	/* Buffers outside FW are with display */

	buffers_outside_fw = total_output_buf - fw_pending_bufs;

	dprintk(VIDC_PROF,

		"Counts : total_output_buf = %d fw_pending_bufs = %d buffers_outside_fw = %d\n",

		total_output_buf, fw_pending_bufs, buffers_outside_fw);

		"Counts : total_output_buf = %d Min buffers = %d fw_pending_bufs = %d buffers_outside_fw = %d\n",

		total_output_buf, min_output_buf, fw_pending_bufs,

			buffers_outside_fw);



	if (buffers_outside_fw >=  dcvs->min_threshold)

		dcvs->load = dcvs->load_low;

	else if (buffers_outside_fw <= dcvs->max_threshold)

	/*

	 * PMS decides clock level based on below algo



	 * Limits :

	 * max_threshold : Client extra allocated buffers. Client

	 * reserves these buffers for it's smooth flow.

	 * min_output_buf : HW requested buffers for it's smooth

	 * flow of buffers.

	 * min_threshold : Driver requested extra buffers for PMS.



	 * 1) When buffers outside FW are reaching client's extra buffers,

	 *    FW is slow and will impact pipeline, Increase clock.

	 * 2) When pending buffers with FW are same as FW requested,

	 *    pipeline has cushion to absorb FW slowness, Decrease clocks.

	 * 3) When none of 1) or 2) FW is just fast enough to maintain

	 *    pipeline, request Right Clocks.

	 */



	if (buffers_outside_fw <= dcvs->max_threshold)

		dcvs->load = dcvs->load_high;

	else if (fw_pending_bufs <= min_output_buf)

		dcvs->load = dcvs->load_low;

	else

		dcvs->load = dcvs->load_norm;
@@ -393,7 +420,7 @@ static unsigned long msm_vidc_max_freq(struct msm_vidc_core *core) 


	allowed_clks_tbl = core->resources.allowed_clks_tbl;

	freq = allowed_clks_tbl[0].clock_rate;

	dprintk(VIDC_PROF, "Max rate = %lu", freq);

	dprintk(VIDC_PROF, "Max rate = %lu\n", freq);



	return freq;

}
@@ -571,7 +598,8 @@ int msm_comm_scale_clocks(struct msm_vidc_inst *inst) 
	mutex_lock(&inst->registeredbufs.lock);

	list_for_each_entry_safe(temp, next, &inst->registeredbufs.list, list) {

		if (temp->vvb.vb2_buf.type ==

				V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE) {

				V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE &&

					temp->deferred) {

			filled_len = max(filled_len,

				temp->vvb.vb2_buf.planes[0].bytesused);

			device_addr = temp->smem[0].device_addr;