video: driver: modify max mapped output count

adjust MAX_MAP_OUTPUT_COUNT based on resolution
to enhance performance and reduce memory pressure.
- For 8K session: count = 20
- For 4K session: count = 32
- For 1080p session: count = 48
- For all remaining sessions: count = 64

reduce MAX_DPB_COUNT to 32.
reduce DEFAULT_MAX_HOST_BUF_COUNT to 64.

Change-Id: I0e6d25121947524b843e9cce96b75871aba174af
Signed-off-by: Darshana Patil <[email protected]>

driver/variant/iris2/src/msm_vidc_iris2.c

@@ -533,34 +533,6 @@ static int __boot_firmware_iris2(struct msm_vidc_core *vidc_core)
 	return rc;
 }
 
-bool res_is_greater_than(u32 width, u32 height,
-	u32 ref_width, u32 ref_height)
-{
-	u32 num_mbs = NUM_MBS_PER_FRAME(height, width);
-	u32 max_side = max(ref_width, ref_height);
-
-	if (num_mbs > NUM_MBS_PER_FRAME(ref_height, ref_width) ||
-		width > max_side ||
-		height > max_side)
-		return true;
-	else
-		return false;
-}
-
-bool res_is_less_than_or_equal_to(u32 width, u32 height,
-	u32 ref_width, u32 ref_height)
-{
-	u32 num_mbs = NUM_MBS_PER_FRAME(height, width);
-	u32 max_side = max(ref_width, ref_height);
-
-	if (num_mbs <= NUM_MBS_PER_FRAME(ref_height, ref_width) &&
-		width <= max_side &&
-		height <= max_side)
-		return true;
-	else
-		return false;
-}
-
 int msm_vidc_decide_work_mode_iris2(struct msm_vidc_inst* inst)
 {
 	u32 work_mode;

driver/vidc/inc/msm_vidc_driver.h

@@ -405,5 +405,9 @@ int msm_vidc_flush_ts(struct msm_vidc_inst *inst);
 const char *buf_name(enum msm_vidc_buffer_type type);
 void msm_vidc_free_capabililty_list(struct msm_vidc_inst *inst,
 	enum msm_vidc_ctrl_list_type list_type);
+bool res_is_greater_than(u32 width, u32 height,
+	u32 ref_width, u32 ref_height);
+bool res_is_less_than_or_equal_to(u32 width, u32 height,
+	u32 ref_width, u32 ref_height);
 #endif // _MSM_VIDC_DRIVER_H_
 

driver/vidc/inc/msm_vidc_inst.h

@@ -153,5 +153,6 @@ struct msm_vidc_inst {
 	bool                               vb2q_init;
 	u32                                max_input_data_size;
 	u32                                dpb_list_payload[MAX_DPB_LIST_ARRAY_SIZE];
+	u32                                max_map_output_count;
 };
 #endif // _MSM_VIDC_INST_H_

driver/vidc/inc/msm_vidc_internal.h

@@ -43,7 +43,7 @@
 #define MAX_CAP_CHILDREN         16
 #define DEFAULT_BITSTREM_ALIGNMENT  16
 #define H265_BITSTREM_ALIGNMENT     32
-#define DEFAULT_MAX_HOST_BUF_COUNT  128
+#define DEFAULT_MAX_HOST_BUF_COUNT  64
 #define DEFAULT_MAX_HOST_BURST_BUF_COUNT 256
 #define BIT_DEPTH_8 (8 << 16 | 8)
 #define BIT_DEPTH_10 (10 << 16 | 10)
@@ -119,14 +119,8 @@
 #define HW_RESPONSE_TIMEOUT_VALUE     (1000)
 #define SW_PC_DELAY_VALUE             (HW_RESPONSE_TIMEOUT_VALUE + 500)
 #define FW_UNLOAD_DELAY_VALUE         (SW_PC_DELAY_VALUE + 1500)
-/*
- * MAX_MAPPED_OUTPUT_COUNT: maximum mappings which can
- * be present in output map list with refcount 1. These
- * mappings exist due to delayed unmap feature. Current
- * threshold is kept as 50 to handle vpp usecases
- * which might have many output buffers.
- */
-#define MAX_MAPPED_OUTPUT_COUNT 64
+
+#define MAX_MAP_OUTPUT_COUNT 64
 #define MAX_DPB_COUNT 32
  /*
   * max dpb count in firmware = 16

driver/vidc/src/msm_vdec.c

@@ -1625,6 +1625,44 @@ static int msm_vdec_subscribe_output_port_settings_change(struct msm_vidc_inst *
 	return rc;
 }
 
+static int msm_vdec_update_max_map_output_count(struct msm_vidc_inst *inst)
+{
+	int rc = 0;
+	struct v4l2_format *f;
+	u32 width, height, count;
+
+	if (!inst) {
+		d_vpr_e("%s: invalid params\n", __func__);
+		return -EINVAL;
+	}
+
+	f = &inst->fmts[OUTPUT_PORT];
+	width = f->fmt.pix_mp.width;
+	height = f->fmt.pix_mp.height;
+
+	/*
+	 * adjust max map output count based on resolution
+	 * to enhance performance.
+	 * For 8K session: count = 20
+	 * For 4K session: count = 32
+	 * For 1080p session: count = 48
+	 * For all remaining sessions: count = 64
+	 */
+	if (res_is_greater_than(width, height, 4096, 2160))
+		count = 20;
+	else if (res_is_greater_than(width, height, 1920, 1080))
+		count = 32;
+	else if (res_is_greater_than(width, height, 1280, 720))
+		count = 48;
+	else
+		count = 64;
+
+	inst->max_map_output_count = count;
+	i_vpr_h(inst, "%s: count: %d\n", __func__, inst->max_map_output_count);
+
+	return rc;
+}
+
 int msm_vdec_streamon_output(struct msm_vidc_inst *inst)
 {
 	int rc = 0;
@@ -1642,6 +1680,10 @@ int msm_vdec_streamon_output(struct msm_vidc_inst *inst)
 		return -EINVAL;
 	}
 
+	rc = msm_vdec_update_max_map_output_count(inst);
+	if (rc)
+		return rc;
+
 	rc = msm_vdec_set_output_properties(inst);
 	if (rc)
 		goto error;
@@ -1884,7 +1926,7 @@ static int msm_vidc_unmap_excessive_mappings(struct msm_vidc_inst *inst)
 			refcount_one_bufs_count++;
 	}
 
-	if (refcount_one_bufs_count <= MAX_MAPPED_OUTPUT_COUNT)
+	if (refcount_one_bufs_count <= inst->max_map_output_count)
 		return 0;
 
 	/* unmap these buffers as they are stale entries */
@@ -2528,6 +2570,7 @@ int msm_vdec_inst_init(struct msm_vidc_inst *inst)
 			inst->buffers.output.min_count +
 			inst->buffers.output.extra_count;
 	inst->buffers.output.size = f->fmt.pix_mp.plane_fmt[0].sizeimage;
+	inst->max_map_output_count = MAX_MAP_OUTPUT_COUNT;
 
 	f = &inst->fmts[OUTPUT_META_PORT];
 	f->type = OUTPUT_META_PLANE;

driver/vidc/src/msm_vidc_driver.c

@@ -1027,6 +1027,34 @@ struct msm_vidc_allocations *msm_vidc_get_allocations(
 	}
 }
 
+bool res_is_greater_than(u32 width, u32 height,
+	u32 ref_width, u32 ref_height)
+{
+	u32 num_mbs = NUM_MBS_PER_FRAME(height, width);
+	u32 max_side = max(ref_width, ref_height);
+
+	if (num_mbs > NUM_MBS_PER_FRAME(ref_height, ref_width) ||
+		width > max_side ||
+		height > max_side)
+		return true;
+	else
+		return false;
+}
+
+bool res_is_less_than_or_equal_to(u32 width, u32 height,
+	u32 ref_width, u32 ref_height)
+{
+	u32 num_mbs = NUM_MBS_PER_FRAME(height, width);
+	u32 max_side = max(ref_width, ref_height);
+
+	if (num_mbs <= NUM_MBS_PER_FRAME(ref_height, ref_width) &&
+		width <= max_side &&
+		height <= max_side)
+		return true;
+	else
+		return false;
+}
+
 int msm_vidc_change_core_state(struct msm_vidc_core *core,
 	enum msm_vidc_core_state request_state, const char *func)
 {