@@ -10,6 +10,7 @@ iris-objs += iris_buffer.o \
iris_hfi_gen2_response.o \
iris_hfi_queue.o \
iris_platform_sm8550.o \
+ iris_power.o \
iris_probe.o \
iris_resources.o \
iris_state.o \
@@ -8,6 +8,7 @@
#include "iris_buffer.h"
#include "iris_instance.h"
+#include "iris_power.h"
#include "iris_vpu_buffer.h"
#define PIXELS_4K 4096
@@ -500,6 +501,8 @@ int iris_queue_deferred_buffers(struct iris_inst *inst, enum iris_buffer_type bu
struct iris_buffer *buf;
int ret;
+ iris_scale_power(inst);
+
if (buf_type == BUF_INPUT) {
v4l2_m2m_for_each_src_buf_safe(m2m_ctx, buffer, n) {
buf = to_iris_buffer(&buffer->vb);
@@ -33,6 +33,9 @@
* @state: instance state
* @sub_state: instance sub state
* @once_per_session_set: boolean to set once per session property
+ * @max_input_data_size: max size of input data
+ * @power: structure of power info
+ * @icc_data: structure of interconnect data
* @m2m_dev: a reference to m2m device structure
* @m2m_ctx: a reference to m2m context structure
* @sequence_cap: a sequence counter for capture queue
@@ -60,6 +63,9 @@ struct iris_inst {
enum iris_inst_state state;
enum iris_inst_sub_state sub_state;
bool once_per_session_set;
+ size_t max_input_data_size;
+ struct iris_inst_power power;
+ struct icc_vote_data icc_data;
struct v4l2_m2m_dev *m2m_dev;
struct v4l2_m2m_ctx *m2m_ctx;
u32 sequence_cap;
@@ -20,6 +20,8 @@ struct iris_inst;
#define CODED_FRAMES_PROGRESSIVE 0x0
#define DEFAULT_MAX_HOST_BUF_COUNT 64
#define DEFAULT_MAX_HOST_BURST_BUF_COUNT 256
+#define DEFAULT_FPS 30
+
enum stage_type {
STAGE_1 = 1,
STAGE_2 = 2,
@@ -67,6 +69,10 @@ struct platform_inst_caps {
u32 min_frame_height;
u32 max_frame_height;
u32 max_mbpf;
+ u32 mb_cycles_vsp;
+ u32 mb_cycles_vpp;
+ u32 mb_cycles_fw;
+ u32 mb_cycles_fw_vpp;
u32 num_comv;
};
@@ -106,11 +112,26 @@ struct platform_inst_fw_cap {
enum platform_inst_fw_cap_type cap_id);
};
+struct bw_info {
+ u32 mbs_per_sec;
+ u32 bw_ddr;
+};
+
struct iris_core_power {
u64 clk_freq;
u64 icc_bw;
};
+struct iris_inst_power {
+ u64 min_freq;
+ u32 icc_bw;
+};
+
+struct icc_vote_data {
+ u32 height, width;
+ u32 fps;
+};
+
enum platform_pm_domain_type {
IRIS_CTRL_POWER_DOMAIN,
IRIS_HW_POWER_DOMAIN,
@@ -124,6 +145,8 @@ struct iris_platform_data {
void (*set_preset_registers)(struct iris_core *core);
const struct icc_info *icc_tbl;
unsigned int icc_tbl_size;
+ const struct bw_info *bw_tbl_dec;
+ unsigned int bw_tbl_dec_size;
const char * const *pmdomain_tbl;
unsigned int pmdomain_tbl_size;
const char * const *opp_pd_tbl;
@@ -126,6 +126,9 @@ static struct platform_inst_caps platform_inst_cap_sm8550 = {
.min_frame_height = 96,
.max_frame_height = 8192,
.max_mbpf = (8192 * 4352) / 256,
+ .mb_cycles_vpp = 200,
+ .mb_cycles_fw = 489583,
+ .mb_cycles_fw_vpp = 66234,
.num_comv = 0,
};
@@ -141,6 +144,13 @@ static const struct icc_info sm8550_icc_table[] = {
static const char * const sm8550_clk_reset_table[] = { "bus" };
+static const struct bw_info sm8550_bw_table_dec[] = {
+ { ((4096 * 2160) / 256) * 60, 1608000 },
+ { ((4096 * 2160) / 256) * 30, 826000 },
+ { ((1920 * 1080) / 256) * 60, 567000 },
+ { ((1920 * 1080) / 256) * 30, 294000 },
+};
+
static const char * const sm8550_pmdomain_table[] = { "venus", "vcodec0" };
static const char * const sm8550_opp_pd_table[] = { "mxc", "mmcx" };
@@ -214,6 +224,8 @@ struct iris_platform_data sm8550_data = {
.icc_tbl_size = ARRAY_SIZE(sm8550_icc_table),
.clk_rst_tbl = sm8550_clk_reset_table,
.clk_rst_tbl_size = ARRAY_SIZE(sm8550_clk_reset_table),
+ .bw_tbl_dec = sm8550_bw_table_dec,
+ .bw_tbl_dec_size = ARRAY_SIZE(sm8550_bw_table_dec),
.pmdomain_tbl = sm8550_pmdomain_table,
.pmdomain_tbl_size = ARRAY_SIZE(sm8550_pmdomain_table),
.opp_pd_tbl = sm8550_opp_pd_table,
new file mode 100644
@@ -0,0 +1,140 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (c) 2022-2024 Qualcomm Innovation Center, Inc. All rights reserved.
+ */
+
+#include <linux/pm_opp.h>
+#include <linux/pm_runtime.h>
+#include <media/v4l2-mem2mem.h>
+
+#include "iris_buffer.h"
+#include "iris_instance.h"
+#include "iris_power.h"
+#include "iris_resources.h"
+#include "iris_vpu_common.h"
+
+static u32 iris_calc_bw(struct iris_inst *inst, struct icc_vote_data *data)
+{
+ const struct bw_info *bw_tbl = NULL;
+ struct iris_core *core = inst->core;
+ u32 num_rows, i, mbs, mbps;
+ u32 icc_bw = 0;
+
+ mbs = DIV_ROUND_UP(data->height, 16) * DIV_ROUND_UP(data->width, 16);
+ mbps = mbs * data->fps;
+ if (mbps == 0)
+ goto exit;
+
+ bw_tbl = core->iris_platform_data->bw_tbl_dec;
+ num_rows = core->iris_platform_data->bw_tbl_dec_size;
+
+ for (i = 0; i < num_rows; i++) {
+ if (i != 0 && mbps > bw_tbl[i].mbs_per_sec)
+ break;
+
+ icc_bw = bw_tbl[i].bw_ddr;
+ }
+
+exit:
+ return icc_bw;
+}
+
+static int iris_set_interconnects(struct iris_inst *inst)
+{
+ struct iris_core *core = inst->core;
+ struct iris_inst *instance;
+ u64 total_bw_ddr = 0;
+ int ret;
+
+ mutex_lock(&core->lock);
+ list_for_each_entry(instance, &core->instances, list) {
+ if (!instance->max_input_data_size)
+ continue;
+
+ total_bw_ddr += instance->power.icc_bw;
+ }
+
+ ret = iris_set_icc_bw(core, total_bw_ddr);
+
+ mutex_unlock(&core->lock);
+
+ return ret;
+}
+
+static int iris_vote_interconnects(struct iris_inst *inst)
+{
+ struct icc_vote_data *vote_data = &inst->icc_data;
+ struct v4l2_format *inp_f = inst->fmt_src;
+
+ vote_data->width = inp_f->fmt.pix_mp.width;
+ vote_data->height = inp_f->fmt.pix_mp.height;
+ vote_data->fps = DEFAULT_FPS;
+
+ inst->power.icc_bw = iris_calc_bw(inst, vote_data);
+
+ return iris_set_interconnects(inst);
+}
+
+static int iris_set_clocks(struct iris_inst *inst)
+{
+ struct iris_core *core = inst->core;
+ struct iris_inst *instance;
+ u64 freq = 0;
+ int ret;
+
+ mutex_lock(&core->lock);
+ list_for_each_entry(instance, &core->instances, list) {
+ if (!instance->max_input_data_size)
+ continue;
+
+ freq += instance->power.min_freq;
+ }
+
+ core->power.clk_freq = freq;
+ ret = dev_pm_opp_set_rate(core->dev, freq);
+ mutex_unlock(&core->lock);
+
+ return ret;
+}
+
+static int iris_scale_clocks(struct iris_inst *inst)
+{
+ const struct vpu_ops *vpu_ops = inst->core->iris_platform_data->vpu_ops;
+ struct v4l2_m2m_ctx *m2m_ctx = inst->m2m_ctx;
+ struct v4l2_m2m_buffer *buffer, *n;
+ struct iris_buffer *buf;
+ size_t data_size = 0;
+
+ v4l2_m2m_for_each_src_buf_safe(m2m_ctx, buffer, n) {
+ buf = to_iris_buffer(&buffer->vb);
+ data_size = max(data_size, buf->data_size);
+ }
+
+ inst->max_input_data_size = data_size;
+ if (!inst->max_input_data_size)
+ return 0;
+
+ inst->power.min_freq = vpu_ops->calc_freq(inst, inst->max_input_data_size);
+
+ return iris_set_clocks(inst);
+}
+
+int iris_scale_power(struct iris_inst *inst)
+{
+ struct iris_core *core = inst->core;
+ int ret;
+
+ if (pm_runtime_suspended(core->dev)) {
+ ret = pm_runtime_resume_and_get(core->dev);
+ if (ret < 0)
+ return ret;
+
+ pm_runtime_put_autosuspend(core->dev);
+ }
+
+ ret = iris_scale_clocks(inst);
+ if (ret)
+ return ret;
+
+ return iris_vote_interconnects(inst);
+}
new file mode 100644
@@ -0,0 +1,13 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (c) 2022-2024 Qualcomm Innovation Center, Inc. All rights reserved.
+ */
+
+#ifndef __IRIS_POWER_H__
+#define __IRIS_POWER_H__
+
+struct iris_inst;
+
+int iris_scale_power(struct iris_inst *inst);
+
+#endif
@@ -10,6 +10,7 @@
#include "iris_instance.h"
#include "iris_vb2.h"
#include "iris_vdec.h"
+#include "iris_power.h"
static int iris_check_core_mbpf(struct iris_inst *inst)
{
@@ -187,6 +188,8 @@ int iris_vb2_start_streaming(struct vb2_queue *q, unsigned int count)
goto error;
}
+ iris_scale_power(inst);
+
ret = iris_check_session_supported(inst);
if (ret)
goto error;
@@ -9,6 +9,7 @@
#include "iris_buffer.h"
#include "iris_ctrls.h"
#include "iris_instance.h"
+#include "iris_power.h"
#include "iris_vdec.h"
#include "iris_vpu_buffer.h"
@@ -360,6 +361,8 @@ static int iris_vdec_process_streamon_input(struct iris_inst *inst)
enum iris_inst_sub_state set_sub_state = 0;
int ret;
+ iris_scale_power(inst);
+
ret = hfi_ops->session_start(inst, V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE);
if (ret)
return ret;
@@ -427,6 +430,8 @@ static int iris_vdec_process_streamon_output(struct iris_inst *inst)
enum iris_inst_sub_state clear_sub_state = 0;
int ret = 0;
+ iris_scale_power(inst);
+
drain_active = inst->sub_state & IRIS_INST_SUB_DRAIN &&
inst->sub_state & IRIS_INST_SUB_DRAIN_LAST;
@@ -573,6 +578,8 @@ int iris_vdec_qbuf(struct iris_inst *inst, struct vb2_v4l2_buffer *vbuf)
return 0;
}
+ iris_scale_power(inst);
+
return iris_queue_buffer(inst, buf);
}
@@ -6,6 +6,33 @@
#include "iris_instance.h"
#include "iris_vpu_common.h"
+static u64 iris_vpu2_calc_freq(struct iris_inst *inst, size_t data_size)
+{
+ struct platform_inst_caps *caps = inst->core->iris_platform_data->inst_caps;
+ struct v4l2_format *inp_f = inst->fmt_src;
+ u32 mbs_per_second, mbpf, height, width;
+ unsigned long vpp_freq, vsp_freq;
+ u32 fps = DEFAULT_FPS;
+
+ width = max(inp_f->fmt.pix_mp.width, inst->crop.width);
+ height = max(inp_f->fmt.pix_mp.height, inst->crop.height);
+
+ mbpf = NUM_MBS_PER_FRAME(height, width);
+ mbs_per_second = mbpf * fps;
+
+ vpp_freq = mbs_per_second * caps->mb_cycles_vpp;
+
+ /* 21 / 20 is overhead factor */
+ vpp_freq += vpp_freq / 20;
+ vsp_freq = mbs_per_second * caps->mb_cycles_vsp;
+
+ /* 10 / 7 is overhead factor */
+ vsp_freq += ((fps * data_size * 8) * 10) / 7;
+
+ return max(vpp_freq, vsp_freq);
+}
+
const struct vpu_ops iris_vpu2_ops = {
.power_off_hw = iris_vpu_power_off_hw,
+ .calc_freq = iris_vpu2_calc_freq,
};
@@ -79,6 +79,44 @@ static void iris_vpu3_power_off_hardware(struct iris_core *core)
iris_vpu_power_off_hw(core);
}
+static u64 iris_vpu3_calculate_frequency(struct iris_inst *inst, size_t data_size)
+{
+ struct platform_inst_caps *caps = inst->core->iris_platform_data->inst_caps;
+ struct v4l2_format *inp_f = inst->fmt_src;
+ u32 height, width, mbs_per_second, mbpf;
+ u64 fw_cycles, fw_vpp_cycles;
+ u64 vsp_cycles, vpp_cycles;
+ u32 fps = DEFAULT_FPS;
+
+ width = max(inp_f->fmt.pix_mp.width, inst->crop.width);
+ height = max(inp_f->fmt.pix_mp.height, inst->crop.height);
+
+ mbpf = NUM_MBS_PER_FRAME(height, width);
+ mbs_per_second = mbpf * fps;
+
+ fw_cycles = fps * caps->mb_cycles_fw;
+ fw_vpp_cycles = fps * caps->mb_cycles_fw_vpp;
+
+ vpp_cycles = mult_frac(mbs_per_second, caps->mb_cycles_vpp, (u32)inst->fw_caps[PIPE].value);
+ /* 21 / 20 is minimum overhead factor */
+ vpp_cycles += max(div_u64(vpp_cycles, 20), fw_vpp_cycles);
+
+ /* 1.059 is multi-pipe overhead */
+ if (inst->fw_caps[PIPE].value > 1)
+ vpp_cycles += div_u64(vpp_cycles * 59, 1000);
+
+ vsp_cycles = fps * data_size * 8;
+ vsp_cycles = div_u64(vsp_cycles, 2);
+ /* VSP FW overhead 1.05 */
+ vsp_cycles = div_u64(vsp_cycles * 21, 20);
+
+ if (inst->fw_caps[STAGE].value == STAGE_1)
+ vsp_cycles = vsp_cycles * 3;
+
+ return max3(vpp_cycles, vsp_cycles, fw_cycles);
+}
+
const struct vpu_ops iris_vpu3_ops = {
.power_off_hw = iris_vpu3_power_off_hardware,
+ .calc_freq = iris_vpu3_calculate_frequency,
};
@@ -13,6 +13,7 @@ extern const struct vpu_ops iris_vpu3_ops;
struct vpu_ops {
void (*power_off_hw)(struct iris_core *core);
+ u64 (*calc_freq)(struct iris_inst *inst, size_t data_size);
};
int iris_vpu_boot_firmware(struct iris_core *core);