[PATCH 16/73] drm/amd/display: cache pwl params and scl_data to avoid extra programming
Harry Wentland
harry.wentland at amd.com
Thu Nov 9 20:05:12 UTC 2017
From: Dmytro Laktyushkin <Dmytro.Laktyushkin at amd.com>
This saves us about 5000 reg writes per full update. This translates to about
40000 writes over the course of single eDP bootup.
Change-Id: If8e0c7bac744d8a3e2c56271c09453152025ccc8
Signed-off-by: Dmytro Laktyushkin <Dmytro.Laktyushkin at amd.com>
Reviewed-by: Tony Cheng <Tony.Cheng at amd.com>
Acked-by: Harry Wentland <harry.wentland at amd.com>
---
drivers/gpu/drm/amd/display/dc/dcn10/dcn10_dpp.c | 41 ++++++++++------------
drivers/gpu/drm/amd/display/dc/dcn10/dcn10_dpp.h | 6 ++--
.../gpu/drm/amd/display/dc/dcn10/dcn10_dpp_dscl.c | 3 ++
.../drm/amd/display/dc/dcn10/dcn10_hw_sequencer.c | 10 +++---
drivers/gpu/drm/amd/display/dc/inc/hw/dpp.h | 8 ++---
5 files changed, 30 insertions(+), 38 deletions(-)
diff --git a/drivers/gpu/drm/amd/display/dc/dcn10/dcn10_dpp.c b/drivers/gpu/drm/amd/display/dc/dcn10/dcn10_dpp.c
index c5f4d5caf976..e9cf9d1514eb 100644
--- a/drivers/gpu/drm/amd/display/dc/dcn10/dcn10_dpp.c
+++ b/drivers/gpu/drm/amd/display/dc/dcn10/dcn10_dpp.c
@@ -178,32 +178,14 @@ void dpp_reset(struct dpp *dpp_base)
dpp->filter_h = NULL;
dpp->filter_v = NULL;
- /* set boundary mode to 0 */
- REG_SET(DSCL_CONTROL, 0, SCL_BOUNDARY_MODE, 0);
+ memset(&dpp->scl_data, 0, sizeof(dpp->scl_data));
+ memset(&dpp->pwl_data, 0, sizeof(dpp->pwl_data));
}
static void dpp1_cm_set_regamma_pwl(
- struct dpp *dpp_base, const struct pwl_params *params)
-{
- struct dcn10_dpp *dpp = TO_DCN10_DPP(dpp_base);
-
- dpp1_cm_power_on_regamma_lut(dpp_base, true);
- dpp1_cm_configure_regamma_lut(dpp_base, dpp->is_write_to_ram_a_safe);
-
- if (dpp->is_write_to_ram_a_safe)
- dpp1_cm_program_regamma_luta_settings(dpp_base, params);
- else
- dpp1_cm_program_regamma_lutb_settings(dpp_base, params);
-
- dpp1_cm_program_regamma_lut(
- dpp_base, params->rgb_resulted, params->hw_points_num);
-}
-
-static void dpp1_cm_set_regamma_mode(
- struct dpp *dpp_base,
- enum opp_regamma mode)
+ struct dpp *dpp_base, const struct pwl_params *params, enum opp_regamma mode)
{
struct dcn10_dpp *dpp = TO_DCN10_DPP(dpp_base);
uint32_t re_mode = 0;
@@ -221,13 +203,27 @@ static void dpp1_cm_set_regamma_mode(
re_mode = 2;
break;
case OPP_REGAMMA_USER:
+ if (memcmp(&dpp->pwl_data, params, sizeof(*params)) == 0)
+ return;
+
+ dpp1_cm_power_on_regamma_lut(dpp_base, true);
+ dpp1_cm_configure_regamma_lut(dpp_base, dpp->is_write_to_ram_a_safe);
+
+ if (dpp->is_write_to_ram_a_safe)
+ dpp1_cm_program_regamma_luta_settings(dpp_base, params);
+ else
+ dpp1_cm_program_regamma_lutb_settings(dpp_base, params);
+
+ dpp1_cm_program_regamma_lut(
+ dpp_base, params->rgb_resulted, params->hw_points_num);
+ dpp->pwl_data = *params;
+
re_mode = dpp->is_write_to_ram_a_safe ? 3 : 4;
dpp->is_write_to_ram_a_safe = !dpp->is_write_to_ram_a_safe;
break;
default:
break;
}
-
REG_SET(CM_RGAM_CONTROL, 0, CM_RGAM_LUT_MODE, re_mode);
REG_UPDATE_2(OBUF_CONTROL,
OBUF_BYPASS, obuf_bypass,
@@ -454,7 +450,6 @@ static const struct dpp_funcs dcn10_dpp_funcs = {
.opp_program_regamma_lutb_settings = dpp1_cm_program_regamma_lutb_settings,
.opp_program_regamma_luta_settings = dpp1_cm_program_regamma_luta_settings,
.opp_program_regamma_pwl = dpp1_cm_set_regamma_pwl,
- .opp_set_regamma_mode = dpp1_cm_set_regamma_mode,
.ipp_program_bias_and_scale = dpp1_program_bias_and_scale,
.ipp_set_degamma = dpp1_set_degamma,
.ipp_program_input_lut = dpp1_program_input_lut,
diff --git a/drivers/gpu/drm/amd/display/dc/dcn10/dcn10_dpp.h b/drivers/gpu/drm/amd/display/dc/dcn10/dcn10_dpp.h
index 880e366568a3..8b894ebc4e17 100644
--- a/drivers/gpu/drm/amd/display/dc/dcn10/dcn10_dpp.h
+++ b/drivers/gpu/drm/amd/display/dc/dcn10/dcn10_dpp.h
@@ -54,7 +54,6 @@
SRI(LB_MEMORY_CTRL, DSCL, id), \
SRI(DSCL_AUTOCAL, DSCL, id), \
SRI(SCL_BLACK_OFFSET, DSCL, id), \
- SRI(DSCL_CONTROL, DSCL, id), \
SRI(SCL_TAP_CONTROL, DSCL, id), \
SRI(SCL_COEF_RAM_TAP_SELECT, DSCL, id), \
SRI(SCL_COEF_RAM_TAP_DATA, DSCL, id), \
@@ -194,7 +193,6 @@
TF_SF(DSCL0_DSCL_AUTOCAL, AUTOCAL_PIPE_ID, mask_sh),\
TF_SF(DSCL0_SCL_BLACK_OFFSET, SCL_BLACK_OFFSET_RGB_Y, mask_sh),\
TF_SF(DSCL0_SCL_BLACK_OFFSET, SCL_BLACK_OFFSET_CBCR, mask_sh),\
- TF_SF(DSCL0_DSCL_CONTROL, SCL_BOUNDARY_MODE, mask_sh),\
TF_SF(DSCL0_SCL_TAP_CONTROL, SCL_V_NUM_TAPS, mask_sh),\
TF_SF(DSCL0_SCL_TAP_CONTROL, SCL_H_NUM_TAPS, mask_sh),\
TF_SF(DSCL0_SCL_TAP_CONTROL, SCL_V_NUM_TAPS_C, mask_sh),\
@@ -440,7 +438,6 @@
type AUTOCAL_PIPE_ID; \
type SCL_BLACK_OFFSET_RGB_Y; \
type SCL_BLACK_OFFSET_CBCR; \
- type SCL_BOUNDARY_MODE; \
type SCL_V_NUM_TAPS; \
type SCL_H_NUM_TAPS; \
type SCL_V_NUM_TAPS_C; \
@@ -1038,7 +1035,6 @@ struct dcn_dpp_registers {
uint32_t LB_MEMORY_CTRL;
uint32_t DSCL_AUTOCAL;
uint32_t SCL_BLACK_OFFSET;
- uint32_t DSCL_CONTROL;
uint32_t SCL_TAP_CONTROL;
uint32_t SCL_COEF_RAM_TAP_SELECT;
uint32_t SCL_COEF_RAM_TAP_DATA;
@@ -1284,6 +1280,8 @@ struct dcn10_dpp {
int lb_memory_size;
int lb_bits_per_entry;
bool is_write_to_ram_a_safe;
+ struct scaler_data scl_data;
+ struct pwl_params pwl_data;
};
enum dcn10_input_csc_select {
diff --git a/drivers/gpu/drm/amd/display/dc/dcn10/dcn10_dpp_dscl.c b/drivers/gpu/drm/amd/display/dc/dcn10/dcn10_dpp_dscl.c
index cbad36410b32..242a568294e2 100644
--- a/drivers/gpu/drm/amd/display/dc/dcn10/dcn10_dpp_dscl.c
+++ b/drivers/gpu/drm/amd/display/dc/dcn10/dcn10_dpp_dscl.c
@@ -648,6 +648,8 @@ void dpp1_dscl_set_scaler_manual_scale(
bool ycbcr = scl_data->format >= PIXEL_FORMAT_VIDEO_BEGIN
&& scl_data->format <= PIXEL_FORMAT_VIDEO_END;
+ if (memcmp(&dpp->scl_data, scl_data, sizeof(*scl_data)) == 0)
+ return;
/* Recout */
dpp1_dscl_set_recout(dpp, &scl_data->recout);
@@ -699,4 +701,5 @@ void dpp1_dscl_set_scaler_manual_scale(
SCL_H_NUM_TAPS_C, scl_data->taps.h_taps_c - 1);
dpp1_dscl_set_scl_filter(dpp, scl_data, ycbcr);
+ dpp->scl_data = *scl_data;
}
diff --git a/drivers/gpu/drm/amd/display/dc/dcn10/dcn10_hw_sequencer.c b/drivers/gpu/drm/amd/display/dc/dcn10/dcn10_hw_sequencer.c
index 51b7cfe9581f..7579e51761cb 100644
--- a/drivers/gpu/drm/amd/display/dc/dcn10/dcn10_hw_sequencer.c
+++ b/drivers/gpu/drm/amd/display/dc/dcn10/dcn10_hw_sequencer.c
@@ -1235,13 +1235,12 @@ static bool dcn10_set_output_transfer_func(
TF_TYPE_PREDEFINED &&
stream->out_transfer_func->tf ==
TRANSFER_FUNCTION_SRGB) {
- dpp->funcs->opp_set_regamma_mode(dpp, OPP_REGAMMA_SRGB);
+ dpp->funcs->opp_program_regamma_pwl(dpp, NULL, OPP_REGAMMA_SRGB);
} else if (dcn10_translate_regamma_to_hw_format(
stream->out_transfer_func, &dpp->regamma_params)) {
- dpp->funcs->opp_program_regamma_pwl(dpp, &dpp->regamma_params);
- dpp->funcs->opp_set_regamma_mode(dpp, OPP_REGAMMA_USER);
+ dpp->funcs->opp_program_regamma_pwl(dpp, &dpp->regamma_params, OPP_REGAMMA_USER);
} else {
- dpp->funcs->opp_set_regamma_mode(dpp, OPP_REGAMMA_BYPASS);
+ dpp->funcs->opp_program_regamma_pwl(dpp, NULL, OPP_REGAMMA_BYPASS);
}
return true;
@@ -2118,8 +2117,7 @@ static void dcn10_apply_ctx_for_surface(
if (num_planes == 0) {
for (i = dc->res_pool->pipe_count - 1; i >= 0 ; i--) {
- struct pipe_ctx *old_pipe_ctx =
- &dc->current_state->res_ctx.pipe_ctx[i];
+ struct pipe_ctx *old_pipe_ctx = &dc->current_state->res_ctx.pipe_ctx[i];
if (old_pipe_ctx->stream_res.tg && old_pipe_ctx->stream_res.tg->inst == be_idx) {
old_pipe_ctx->stream_res.tg->funcs->set_blank(old_pipe_ctx->stream_res.tg, true);
diff --git a/drivers/gpu/drm/amd/display/dc/inc/hw/dpp.h b/drivers/gpu/drm/amd/display/dc/inc/hw/dpp.h
index 6eca95931ee1..71078d184289 100644
--- a/drivers/gpu/drm/amd/display/dc/inc/hw/dpp.h
+++ b/drivers/gpu/drm/amd/display/dc/inc/hw/dpp.h
@@ -92,11 +92,9 @@ struct dpp_funcs {
const struct pwl_params *params);
void (*opp_program_regamma_pwl)(
- struct dpp *dpp, const struct pwl_params *params);
-
- void (*opp_set_regamma_mode)(
- struct dpp *dpp_base,
- enum opp_regamma mode);
+ struct dpp *dpp,
+ const struct pwl_params *params,
+ enum opp_regamma mode);
void (*ipp_program_bias_and_scale)(
struct dpp *dpp,
--
2.14.1
More information about the amd-gfx
mailing list