Message ID | 1562078787-516-5-git-send-email-amasule@codeaurora.org (mailing list archive) |
---|---|
State | Superseded, archived |
Headers | show |
Series | media: venus: Update clock scaling and core selection | expand |
Hi, On 7/2/19 5:46 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 | 69 +++++++++++++++++++++++--- > drivers/media/platform/qcom/venus/helpers.h | 2 +- > 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 | 2 +- > drivers/media/platform/qcom/venus/venc.c | 2 +- > 6 files changed, 72 insertions(+), 9 deletions(-) > > diff --git a/drivers/media/platform/qcom/venus/helpers.c b/drivers/media/platform/qcom/venus/helpers.c > index 5726d86..321e9f7 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)) if (inst->clk_data.core_id != core_id) I guess will be more readable? > + 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; > +} > + <cut>
On 2019-07-15 21:30, Stanimir Varbanov wrote: > Hi, > > On 7/2/19 5:46 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 | 69 >> +++++++++++++++++++++++--- >> drivers/media/platform/qcom/venus/helpers.h | 2 +- >> 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 | 2 +- >> drivers/media/platform/qcom/venus/venc.c | 2 +- >> 6 files changed, 72 insertions(+), 9 deletions(-) >> >> diff --git a/drivers/media/platform/qcom/venus/helpers.c >> b/drivers/media/platform/qcom/venus/helpers.c >> index 5726d86..321e9f7 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)) > > if (inst->clk_data.core_id != core_id) > > I guess will be more readable? > Yes, I will modify the check. >> + 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; >> +} >> + > > <cut> Regards, Aniket
diff --git a/drivers/media/platform/qcom/venus/helpers.c b/drivers/media/platform/qcom/venus/helpers.c index 5726d86..321e9f7 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; @@ -505,6 +524,16 @@ 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) { @@ -808,19 +837,47 @@ 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, core1_load = 0, core2_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; + core1_load = load_per_core(core, VIDC_CORE_ID_1); + core2_load = load_per_core(core, VIDC_CORE_ID_2); + min_core_id = core1_load < core2_load ? VIDC_CORE_ID_1 : VIDC_CORE_ID_2; + min_load = min(core1_load, core2_load); + cores_max = core_num_max(inst); - return hfi_session_set_property(inst, ptype, &cu); + if (cores_max < VIDC_CORE_ID_2) { + min_core_id = VIDC_CORE_ID_1; + min_load = core1_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/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 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;
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 | 69 +++++++++++++++++++++++--- drivers/media/platform/qcom/venus/helpers.h | 2 +- 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 | 2 +- drivers/media/platform/qcom/venus/venc.c | 2 +- 6 files changed, 72 insertions(+), 9 deletions(-)