diff options
Diffstat (limited to 'drivers/gpu/drm/amd/display/dc/dsc/dc_dsc.c')
| -rw-r--r-- | drivers/gpu/drm/amd/display/dc/dsc/dc_dsc.c | 1041 |
1 files changed, 733 insertions, 308 deletions
diff --git a/drivers/gpu/drm/amd/display/dc/dsc/dc_dsc.c b/drivers/gpu/drm/amd/display/dc/dsc/dc_dsc.c index f403d8e84a8c..e4144b244332 100644 --- a/drivers/gpu/drm/amd/display/dc/dsc/dc_dsc.c +++ b/drivers/gpu/drm/amd/display/dc/dsc/dc_dsc.c @@ -22,12 +22,19 @@ * Author: AMD */ -#include <drm/drm_dsc.h> +#include <drm/display/drm_dp_helper.h> +#include <drm/display/drm_dsc_helper.h> #include "dc_hw_types.h" #include "dsc.h" -#include <drm/drm_dp_helper.h> #include "dc.h" #include "rc_calc.h" +#include "fixed31_32.h" + +#include "clk_mgr.h" +#include "resource.h" + +#define DC_LOGGER \ + dsc->ctx->logger /* This module's internal functions */ @@ -39,6 +46,154 @@ static bool dsc_policy_enable_dsc_when_not_needed; static bool dsc_policy_disable_dsc_stream_overhead; +static bool disable_128b_132b_stream_overhead; + +#ifndef MAX +#define MAX(X, Y) ((X) > (Y) ? (X) : (Y)) +#endif +#ifndef MIN +#define MIN(X, Y) ((X) < (Y) ? (X) : (Y)) +#endif + +/* Need to account for padding due to pixel-to-symbol packing + * for uncompressed 128b/132b streams. + */ +static uint32_t apply_128b_132b_stream_overhead( + const struct dc_crtc_timing *timing, const uint32_t kbps) +{ + uint32_t total_kbps = kbps; + + if (disable_128b_132b_stream_overhead) + return kbps; + + if (!timing->flags.DSC) { + struct fixed31_32 bpp; + struct fixed31_32 overhead_factor; + + bpp = dc_fixpt_from_int(kbps); + bpp = dc_fixpt_div_int(bpp, timing->pix_clk_100hz / 10); + + /* Symbols_per_HActive = HActive * bpp / (4 lanes * 32-bit symbol size) + * Overhead_factor = ceil(Symbols_per_HActive) / Symbols_per_HActive + */ + overhead_factor = dc_fixpt_from_int(timing->h_addressable); + overhead_factor = dc_fixpt_mul(overhead_factor, bpp); + overhead_factor = dc_fixpt_div_int(overhead_factor, 128); + overhead_factor = dc_fixpt_div( + dc_fixpt_from_int(dc_fixpt_ceil(overhead_factor)), + overhead_factor); + + total_kbps = dc_fixpt_ceil( + dc_fixpt_mul_int(overhead_factor, total_kbps)); + } + + return total_kbps; +} + +uint32_t dc_bandwidth_in_kbps_from_timing( + const struct dc_crtc_timing *timing, + const enum dc_link_encoding_format link_encoding) +{ + uint32_t bits_per_channel = 0; + uint32_t kbps; + + if (timing->flags.DSC) + return dc_dsc_stream_bandwidth_in_kbps(timing, + timing->dsc_cfg.bits_per_pixel, + timing->dsc_cfg.num_slices_h, + timing->dsc_cfg.is_dp); + + switch (timing->display_color_depth) { + case COLOR_DEPTH_666: + bits_per_channel = 6; + break; + case COLOR_DEPTH_888: + bits_per_channel = 8; + break; + case COLOR_DEPTH_101010: + bits_per_channel = 10; + break; + case COLOR_DEPTH_121212: + bits_per_channel = 12; + break; + case COLOR_DEPTH_141414: + bits_per_channel = 14; + break; + case COLOR_DEPTH_161616: + bits_per_channel = 16; + break; + default: + ASSERT(bits_per_channel != 0); + bits_per_channel = 8; + break; + } + + kbps = timing->pix_clk_100hz / 10; + kbps *= bits_per_channel; + + if (timing->flags.Y_ONLY != 1) { + /*Only YOnly make reduce bandwidth by 1/3 compares to RGB*/ + kbps *= 3; + if (timing->pixel_encoding == PIXEL_ENCODING_YCBCR420) + kbps /= 2; + else if (timing->pixel_encoding == PIXEL_ENCODING_YCBCR422) + kbps = kbps * 2 / 3; + } + + if (link_encoding == DC_LINK_ENCODING_DP_128b_132b) + kbps = apply_128b_132b_stream_overhead(timing, kbps); + + if (link_encoding == DC_LINK_ENCODING_HDMI_FRL && + timing->vic == 0 && timing->hdmi_vic == 0 && + timing->frl_uncompressed_video_bandwidth_in_kbps != 0) + kbps = timing->frl_uncompressed_video_bandwidth_in_kbps; + + return kbps; +} + +/* Forward Declerations */ +static unsigned int get_min_dsc_slice_count_for_odm( + const struct display_stream_compressor *dsc, + const struct dsc_enc_caps *dsc_enc_caps, + const struct dc_crtc_timing *timing); + +static bool decide_dsc_bandwidth_range( + const uint32_t min_bpp_x16, + const uint32_t max_bpp_x16, + const uint32_t num_slices_h, + const struct dsc_enc_caps *dsc_caps, + const struct dc_crtc_timing *timing, + const enum dc_link_encoding_format link_encoding, + struct dc_dsc_bw_range *range); + +static uint32_t compute_bpp_x16_from_target_bandwidth( + const uint32_t bandwidth_in_kbps, + const struct dc_crtc_timing *timing, + const uint32_t num_slices_h, + const uint32_t bpp_increment_div, + const bool is_dp); + +static void get_dsc_enc_caps( + const struct display_stream_compressor *dsc, + struct dsc_enc_caps *dsc_enc_caps, + int pixel_clock_100Hz); + +static bool intersect_dsc_caps( + const struct dsc_dec_dpcd_caps *dsc_sink_caps, + const struct dsc_enc_caps *dsc_enc_caps, + enum dc_pixel_encoding pixel_encoding, + struct dsc_enc_caps *dsc_common_caps); + +static bool setup_dsc_config( + const struct dsc_dec_dpcd_caps *dsc_sink_caps, + const struct dsc_enc_caps *dsc_enc_caps, + int target_bandwidth_kbps, + const struct dc_crtc_timing *timing, + const struct dc_dsc_config_options *options, + const enum dc_link_encoding_format link_encoding, + int min_slice_count, + struct dc_dsc_config *dsc_cfg); + static bool dsc_buff_block_size_from_dpcd(int dpcd_buff_block_size, int *buff_block_size) { @@ -171,30 +326,373 @@ static bool dsc_bpp_increment_div_from_dpcd(uint8_t bpp_increment_dpcd, uint32_t return true; } -static void get_dsc_enc_caps( - const struct display_stream_compressor *dsc, - struct dsc_enc_caps *dsc_enc_caps, - int pixel_clock_100Hz) + + +bool dc_dsc_parse_dsc_dpcd(const struct dc *dc, + const uint8_t *dpcd_dsc_basic_data, + const uint8_t *dpcd_dsc_branch_decoder_caps, + struct dsc_dec_dpcd_caps *dsc_sink_caps) { - // This is a static HW query, so we can use any DSC + if (!dpcd_dsc_basic_data) + return false; + + dsc_sink_caps->is_dsc_supported = + (dpcd_dsc_basic_data[DP_DSC_SUPPORT - DP_DSC_SUPPORT] & DP_DSC_DECOMPRESSION_IS_SUPPORTED) != 0; + if (!dsc_sink_caps->is_dsc_supported) + return false; + + dsc_sink_caps->dsc_version = dpcd_dsc_basic_data[DP_DSC_REV - DP_DSC_SUPPORT]; + + { + int buff_block_size; + int buff_size; + + if (!dsc_buff_block_size_from_dpcd( + dpcd_dsc_basic_data[DP_DSC_RC_BUF_BLK_SIZE - DP_DSC_SUPPORT] & 0x03, + &buff_block_size)) + return false; + + buff_size = dpcd_dsc_basic_data[DP_DSC_RC_BUF_SIZE - DP_DSC_SUPPORT] + 1; + dsc_sink_caps->rc_buffer_size = buff_size * buff_block_size; + } + + dsc_sink_caps->slice_caps1.raw = dpcd_dsc_basic_data[DP_DSC_SLICE_CAP_1 - DP_DSC_SUPPORT]; + if (!dsc_line_buff_depth_from_dpcd(dpcd_dsc_basic_data[DP_DSC_LINE_BUF_BIT_DEPTH - DP_DSC_SUPPORT], + &dsc_sink_caps->lb_bit_depth)) + return false; + + dsc_sink_caps->is_block_pred_supported = + (dpcd_dsc_basic_data[DP_DSC_BLK_PREDICTION_SUPPORT - DP_DSC_SUPPORT] & + DP_DSC_BLK_PREDICTION_IS_SUPPORTED) != 0; + + dsc_sink_caps->edp_max_bits_per_pixel = + dpcd_dsc_basic_data[DP_DSC_MAX_BITS_PER_PIXEL_LOW - DP_DSC_SUPPORT] | + dpcd_dsc_basic_data[DP_DSC_MAX_BITS_PER_PIXEL_HI - DP_DSC_SUPPORT] << 8; + + dsc_sink_caps->color_formats.raw = dpcd_dsc_basic_data[DP_DSC_DEC_COLOR_FORMAT_CAP - DP_DSC_SUPPORT]; + dsc_sink_caps->color_depth.raw = dpcd_dsc_basic_data[DP_DSC_DEC_COLOR_DEPTH_CAP - DP_DSC_SUPPORT]; + + { + int dpcd_throughput = dpcd_dsc_basic_data[DP_DSC_PEAK_THROUGHPUT - DP_DSC_SUPPORT]; + int dsc_throughput_granular_delta; + + dsc_throughput_granular_delta = dpcd_dsc_basic_data[DP_DSC_RC_BUF_BLK_SIZE - DP_DSC_SUPPORT] >> 3; + dsc_throughput_granular_delta *= 2; + + if (!dsc_throughput_from_dpcd(dpcd_throughput & DP_DSC_THROUGHPUT_MODE_0_MASK, + &dsc_sink_caps->throughput_mode_0_mps)) + return false; + dsc_sink_caps->throughput_mode_0_mps += dsc_throughput_granular_delta; + + dpcd_throughput = (dpcd_throughput & DP_DSC_THROUGHPUT_MODE_1_MASK) >> DP_DSC_THROUGHPUT_MODE_1_SHIFT; + if (!dsc_throughput_from_dpcd(dpcd_throughput, &dsc_sink_caps->throughput_mode_1_mps)) + return false; + } + + dsc_sink_caps->max_slice_width = dpcd_dsc_basic_data[DP_DSC_MAX_SLICE_WIDTH - DP_DSC_SUPPORT] * 320; + dsc_sink_caps->slice_caps2.raw = dpcd_dsc_basic_data[DP_DSC_SLICE_CAP_2 - DP_DSC_SUPPORT]; + + if (!dsc_bpp_increment_div_from_dpcd(dpcd_dsc_basic_data[DP_DSC_BITS_PER_PIXEL_INC - DP_DSC_SUPPORT], + &dsc_sink_caps->bpp_increment_div)) + return false; + + if (dc->debug.dsc_bpp_increment_div) { + /* dsc_bpp_increment_div should onl be 1, 2, 4, 8 or 16, but rather than rejecting invalid values, + * we'll accept all and get it into range. This also makes the above check against 0 redundant, + * but that one stresses out the override will be only used if it's not 0. + */ + if (dc->debug.dsc_bpp_increment_div >= 1) + dsc_sink_caps->bpp_increment_div = 1; + if (dc->debug.dsc_bpp_increment_div >= 2) + dsc_sink_caps->bpp_increment_div = 2; + if (dc->debug.dsc_bpp_increment_div >= 4) + dsc_sink_caps->bpp_increment_div = 4; + if (dc->debug.dsc_bpp_increment_div >= 8) + dsc_sink_caps->bpp_increment_div = 8; + if (dc->debug.dsc_bpp_increment_div >= 16) + dsc_sink_caps->bpp_increment_div = 16; + } + + /* Extended caps */ + if (dpcd_dsc_branch_decoder_caps == NULL) { // branch decoder DPCD DSC data can be null for non branch device + dsc_sink_caps->branch_overall_throughput_0_mps = 0; + dsc_sink_caps->branch_overall_throughput_1_mps = 0; + dsc_sink_caps->branch_max_line_width = 0; + return true; + } + + dsc_sink_caps->branch_overall_throughput_0_mps = + dpcd_dsc_branch_decoder_caps[DP_DSC_BRANCH_OVERALL_THROUGHPUT_0 - DP_DSC_BRANCH_OVERALL_THROUGHPUT_0]; + if (dsc_sink_caps->branch_overall_throughput_0_mps == 0) + dsc_sink_caps->branch_overall_throughput_0_mps = 0; + else if (dsc_sink_caps->branch_overall_throughput_0_mps == 1) + dsc_sink_caps->branch_overall_throughput_0_mps = 680; + else { + dsc_sink_caps->branch_overall_throughput_0_mps *= 50; + dsc_sink_caps->branch_overall_throughput_0_mps += 600; + } + + dsc_sink_caps->branch_overall_throughput_1_mps = + dpcd_dsc_branch_decoder_caps[DP_DSC_BRANCH_OVERALL_THROUGHPUT_1 - DP_DSC_BRANCH_OVERALL_THROUGHPUT_0]; + if (dsc_sink_caps->branch_overall_throughput_1_mps == 0) + dsc_sink_caps->branch_overall_throughput_1_mps = 0; + else if (dsc_sink_caps->branch_overall_throughput_1_mps == 1) + dsc_sink_caps->branch_overall_throughput_1_mps = 680; + else { + dsc_sink_caps->branch_overall_throughput_1_mps *= 50; + dsc_sink_caps->branch_overall_throughput_1_mps += 600; + } + + dsc_sink_caps->branch_max_line_width = + dpcd_dsc_branch_decoder_caps[DP_DSC_BRANCH_MAX_LINE_WIDTH - DP_DSC_BRANCH_OVERALL_THROUGHPUT_0] * 320; + ASSERT(dsc_sink_caps->branch_max_line_width == 0 || dsc_sink_caps->branch_max_line_width >= 5120); + + dsc_sink_caps->is_dp = true; + return true; +} + +/* If DSC is possbile, get DSC bandwidth range based on [min_bpp, max_bpp] target bitrate range and + * timing's pixel clock and uncompressed bandwidth. + * If DSC is not possible, leave '*range' untouched. + */ +bool dc_dsc_compute_bandwidth_range( + const struct display_stream_compressor *dsc, + uint32_t dsc_min_slice_height_override, + uint32_t min_bpp_x16, + uint32_t max_bpp_x16, + const struct dsc_dec_dpcd_caps *dsc_sink_caps, + const struct dc_crtc_timing *timing, + const enum dc_link_encoding_format link_encoding, + struct dc_dsc_bw_range *range) +{ + bool is_dsc_possible = false; + unsigned int min_dsc_slice_count; + struct dsc_enc_caps dsc_enc_caps; + struct dsc_enc_caps dsc_common_caps; + struct dc_dsc_config config = {0}; + struct dc_dsc_config_options options = {0}; + + options.dsc_min_slice_height_override = dsc_min_slice_height_override; + options.max_target_bpp_limit_override_x16 = max_bpp_x16; + options.slice_height_granularity = 1; + + get_dsc_enc_caps(dsc, &dsc_enc_caps, timing->pix_clk_100hz); + + min_dsc_slice_count = get_min_dsc_slice_count_for_odm(dsc, &dsc_enc_caps, timing); + + is_dsc_possible = intersect_dsc_caps(dsc_sink_caps, &dsc_enc_caps, + timing->pixel_encoding, &dsc_common_caps); + + if (is_dsc_possible) + is_dsc_possible = setup_dsc_config(dsc_sink_caps, &dsc_enc_caps, 0, timing, + &options, link_encoding, min_dsc_slice_count, &config); + + if (is_dsc_possible) + is_dsc_possible = decide_dsc_bandwidth_range(min_bpp_x16, max_bpp_x16, + config.num_slices_h, &dsc_common_caps, timing, link_encoding, range); + + return is_dsc_possible; +} + +void dc_dsc_dump_encoder_caps(const struct display_stream_compressor *dsc, + const struct dc_crtc_timing *timing) +{ + struct dsc_enc_caps dsc_enc_caps; + + get_dsc_enc_caps(dsc, &dsc_enc_caps, timing->pix_clk_100hz); + + DC_LOG_DSC("dsc encoder caps:"); + DC_LOG_DSC("\tdsc_version 0x%x", dsc_enc_caps.dsc_version); + DC_LOG_DSC("\tslice_caps 0x%x", dsc_enc_caps.slice_caps.raw); + DC_LOG_DSC("\tlb_bit_depth %d", dsc_enc_caps.lb_bit_depth); + DC_LOG_DSC("\tis_block_pred_supported %d", dsc_enc_caps.is_block_pred_supported); + DC_LOG_DSC("\tcolor_formats 0x%x", dsc_enc_caps.color_formats.raw); + DC_LOG_DSC("\tcolor_depth 0x%x", dsc_enc_caps.color_depth.raw); + DC_LOG_DSC("\tmax_total_throughput_mps %d", dsc_enc_caps.max_total_throughput_mps); + DC_LOG_DSC("\tmax_slice_width %d", dsc_enc_caps.max_slice_width); + DC_LOG_DSC("\tbpp_increment_div %d", dsc_enc_caps.bpp_increment_div); +} + +void dc_dsc_dump_decoder_caps(const struct display_stream_compressor *dsc, + const struct dsc_dec_dpcd_caps *dsc_sink_caps) +{ + DC_LOG_DSC("dsc decoder caps:"); + DC_LOG_DSC("\tis_dsc_supported %d", dsc_sink_caps->is_dsc_supported); + DC_LOG_DSC("\tdsc_version 0x%x", dsc_sink_caps->dsc_version); + DC_LOG_DSC("\trc_buffer_size %d", dsc_sink_caps->rc_buffer_size); + DC_LOG_DSC("\tslice_caps1 0x%x", dsc_sink_caps->slice_caps1.raw); + DC_LOG_DSC("\tslice_caps2 0x%x", dsc_sink_caps->slice_caps2.raw); + DC_LOG_DSC("\tlb_bit_depth %d", dsc_sink_caps->lb_bit_depth); + DC_LOG_DSC("\tis_block_pred_supported %d", dsc_sink_caps->is_block_pred_supported); + DC_LOG_DSC("\tedp_max_bits_per_pixel %d", dsc_sink_caps->edp_max_bits_per_pixel); + DC_LOG_DSC("\tcolor_formats 0x%x", dsc_sink_caps->color_formats.raw); + DC_LOG_DSC("\tthroughput_mode_0_mps %d", dsc_sink_caps->throughput_mode_0_mps); + DC_LOG_DSC("\tthroughput_mode_1_mps %d", dsc_sink_caps->throughput_mode_1_mps); + DC_LOG_DSC("\tmax_slice_width %d", dsc_sink_caps->max_slice_width); + DC_LOG_DSC("\tbpp_increment_div %d", dsc_sink_caps->bpp_increment_div); + DC_LOG_DSC("\tbranch_overall_throughput_0_mps %d", dsc_sink_caps->branch_overall_throughput_0_mps); + DC_LOG_DSC("\tbranch_overall_throughput_1_mps %d", dsc_sink_caps->branch_overall_throughput_1_mps); + DC_LOG_DSC("\tbranch_max_line_width %d", dsc_sink_caps->branch_max_line_width); + DC_LOG_DSC("\tis_dp %d", dsc_sink_caps->is_dp); +} + + +static void build_dsc_enc_combined_slice_caps( + const struct dsc_enc_caps *single_dsc_enc_caps, + struct dsc_enc_caps *dsc_enc_caps, + unsigned int max_odm_combine_factor) +{ + /* 1-16 slice configurations, single DSC */ + dsc_enc_caps->slice_caps.raw |= single_dsc_enc_caps->slice_caps.raw; + + /* 2x DSC's */ + if (max_odm_combine_factor >= 2) { + /* 1 + 1 */ + dsc_enc_caps->slice_caps.bits.NUM_SLICES_2 |= single_dsc_enc_caps->slice_caps.bits.NUM_SLICES_1; + + /* 2 + 2 */ + dsc_enc_caps->slice_caps.bits.NUM_SLICES_4 |= single_dsc_enc_caps->slice_caps.bits.NUM_SLICES_2; + + /* 4 + 4 */ + dsc_enc_caps->slice_caps.bits.NUM_SLICES_8 |= single_dsc_enc_caps->slice_caps.bits.NUM_SLICES_4; + + /* 8 + 8 */ + dsc_enc_caps->slice_caps.bits.NUM_SLICES_16 |= single_dsc_enc_caps->slice_caps.bits.NUM_SLICES_8; + } + + /* 3x DSC's */ + if (max_odm_combine_factor >= 3) { + /* 4 + 4 + 4 */ + dsc_enc_caps->slice_caps.bits.NUM_SLICES_12 |= single_dsc_enc_caps->slice_caps.bits.NUM_SLICES_4; + } + + /* 4x DSC's */ + if (max_odm_combine_factor >= 4) { + /* 1 + 1 + 1 + 1 */ + dsc_enc_caps->slice_caps.bits.NUM_SLICES_4 |= single_dsc_enc_caps->slice_caps.bits.NUM_SLICES_1; + + /* 2 + 2 + 2 + 2 */ + dsc_enc_caps->slice_caps.bits.NUM_SLICES_8 |= single_dsc_enc_caps->slice_caps.bits.NUM_SLICES_2; + + /* 3 + 3 + 3 + 3 */ + dsc_enc_caps->slice_caps.bits.NUM_SLICES_12 |= single_dsc_enc_caps->slice_caps.bits.NUM_SLICES_3; + + /* 4 + 4 + 4 + 4 */ + dsc_enc_caps->slice_caps.bits.NUM_SLICES_16 |= single_dsc_enc_caps->slice_caps.bits.NUM_SLICES_4; + } +} + +static void build_dsc_enc_caps( + const struct display_stream_compressor *dsc, + struct dsc_enc_caps *dsc_enc_caps) +{ + unsigned int max_dscclk_khz; + unsigned int num_dsc; + unsigned int max_odm_combine_factor; + struct dsc_enc_caps single_dsc_enc_caps; + + struct dc *dc; + + if (!dsc || !dsc->ctx || !dsc->ctx->dc || !dsc->funcs->dsc_get_single_enc_caps) + return; + + dc = dsc->ctx->dc; + + if (!dc->clk_mgr || !dc->clk_mgr->funcs->get_max_clock_khz || !dc->res_pool || dc->debug.disable_dsc) + return; + + /* get max DSCCLK from clk_mgr */ + max_dscclk_khz = dc->clk_mgr->funcs->get_max_clock_khz(dc->clk_mgr, CLK_TYPE_DSCCLK); + + dsc->funcs->dsc_get_single_enc_caps(&single_dsc_enc_caps, max_dscclk_khz); + + /* global capabilities */ + dsc_enc_caps->dsc_version = single_dsc_enc_caps.dsc_version; + dsc_enc_caps->lb_bit_depth = single_dsc_enc_caps.lb_bit_depth; + dsc_enc_caps->is_block_pred_supported = single_dsc_enc_caps.is_block_pred_supported; + dsc_enc_caps->max_slice_width = single_dsc_enc_caps.max_slice_width; + dsc_enc_caps->bpp_increment_div = single_dsc_enc_caps.bpp_increment_div; + dsc_enc_caps->color_formats.raw = single_dsc_enc_caps.color_formats.raw; + dsc_enc_caps->color_depth.raw = single_dsc_enc_caps.color_depth.raw; + + /* expand per DSC capabilities to global */ + max_odm_combine_factor = dc->caps.max_odm_combine_factor; + num_dsc = dc->res_pool->res_cap->num_dsc; + max_odm_combine_factor = min(max_odm_combine_factor, num_dsc); + dsc_enc_caps->max_total_throughput_mps = + single_dsc_enc_caps.max_total_throughput_mps * + max_odm_combine_factor; + + /* check slice counts possible for with ODM combine */ + build_dsc_enc_combined_slice_caps(&single_dsc_enc_caps, dsc_enc_caps, max_odm_combine_factor); +} + +static inline uint32_t dsc_div_by_10_round_up(uint32_t value) +{ + return (value + 9) / 10; +} + +static unsigned int get_min_dsc_slice_count_for_odm( + const struct display_stream_compressor *dsc, + const struct dsc_enc_caps *dsc_enc_caps, + const struct dc_crtc_timing *timing) +{ + unsigned int max_dispclk_khz; + + /* get max pixel rate and combine caps */ + max_dispclk_khz = dsc_enc_caps->max_total_throughput_mps * 1000; + if (dsc && dsc->ctx->dc) { + if (dsc->ctx->dc->clk_mgr && + dsc->ctx->dc->clk_mgr->funcs->get_max_clock_khz) { + /* dispclk is available */ + max_dispclk_khz = dsc->ctx->dc->clk_mgr->funcs->get_max_clock_khz(dsc->ctx->dc->clk_mgr, CLK_TYPE_DISPCLK); + } + } + + /* validate parameters */ + if (max_dispclk_khz == 0 || dsc_enc_caps->max_slice_width == 0) + return 1; + /* consider minimum odm slices required due to + * 1) display pipe throughput (dispclk) + * 2) max image width per slice + */ + return dc_fixpt_ceil(dc_fixpt_max( + dc_fixpt_div_int(dc_fixpt_from_int(dsc_div_by_10_round_up(timing->pix_clk_100hz)), + max_dispclk_khz), // throughput + dc_fixpt_div_int(dc_fixpt_from_int(timing->h_addressable + timing->h_border_left + timing->h_border_right), + dsc_enc_caps->max_slice_width))); // slice width +} + +static void get_dsc_enc_caps( + const struct display_stream_compressor *dsc, + struct dsc_enc_caps *dsc_enc_caps, + int pixel_clock_100Hz) +{ memset(dsc_enc_caps, 0, sizeof(struct dsc_enc_caps)); - if (dsc) { - if (!dsc->ctx->dc->debug.disable_dsc) - dsc->funcs->dsc_get_enc_caps(dsc_enc_caps, pixel_clock_100Hz); - if (dsc->ctx->dc->debug.native422_support) - dsc_enc_caps->color_formats.bits.YCBCR_NATIVE_422 = 1; + + if (!dsc || !dsc->ctx || !dsc->ctx->dc || dsc->ctx->dc->debug.disable_dsc) + return; + + /* check if reported cap global or only for a single DCN DSC enc */ + if (dsc->funcs->dsc_get_enc_caps) { + dsc->funcs->dsc_get_enc_caps(dsc_enc_caps, pixel_clock_100Hz); + } else { + build_dsc_enc_caps(dsc, dsc_enc_caps); } + + if (dsc->ctx->dc->debug.native422_support) + dsc_enc_caps->color_formats.bits.YCBCR_NATIVE_422 = 1; } -/* Returns 'false' if no intersection was found for at least one capablity. +/* Returns 'false' if no intersection was found for at least one capability. * It also implicitly validates some sink caps against invalid value of zero. */ static bool intersect_dsc_caps( - const struct dsc_dec_dpcd_caps *dsc_sink_caps, - const struct dsc_enc_caps *dsc_enc_caps, - enum dc_pixel_encoding pixel_encoding, - struct dsc_enc_caps *dsc_common_caps) + const struct dsc_dec_dpcd_caps *dsc_sink_caps, + const struct dsc_enc_caps *dsc_enc_caps, + enum dc_pixel_encoding pixel_encoding, + struct dsc_enc_caps *dsc_common_caps) { int32_t max_slices; int32_t total_sink_throughput; @@ -205,10 +703,19 @@ static bool intersect_dsc_caps( if (!dsc_common_caps->dsc_version) return false; - dsc_common_caps->slice_caps.bits.NUM_SLICES_1 = dsc_sink_caps->slice_caps1.bits.NUM_SLICES_1 && dsc_enc_caps->slice_caps.bits.NUM_SLICES_1; - dsc_common_caps->slice_caps.bits.NUM_SLICES_2 = dsc_sink_caps->slice_caps1.bits.NUM_SLICES_2 && dsc_enc_caps->slice_caps.bits.NUM_SLICES_2; - dsc_common_caps->slice_caps.bits.NUM_SLICES_4 = dsc_sink_caps->slice_caps1.bits.NUM_SLICES_4 && dsc_enc_caps->slice_caps.bits.NUM_SLICES_4; - dsc_common_caps->slice_caps.bits.NUM_SLICES_8 = dsc_sink_caps->slice_caps1.bits.NUM_SLICES_8 && dsc_enc_caps->slice_caps.bits.NUM_SLICES_8; + dsc_common_caps->slice_caps.bits.NUM_SLICES_1 = + dsc_sink_caps->slice_caps1.bits.NUM_SLICES_1 && dsc_enc_caps->slice_caps.bits.NUM_SLICES_1; + dsc_common_caps->slice_caps.bits.NUM_SLICES_2 = + dsc_sink_caps->slice_caps1.bits.NUM_SLICES_2 && dsc_enc_caps->slice_caps.bits.NUM_SLICES_2; + dsc_common_caps->slice_caps.bits.NUM_SLICES_4 = + dsc_sink_caps->slice_caps1.bits.NUM_SLICES_4 && dsc_enc_caps->slice_caps.bits.NUM_SLICES_4; + dsc_common_caps->slice_caps.bits.NUM_SLICES_8 = + dsc_sink_caps->slice_caps1.bits.NUM_SLICES_8 && dsc_enc_caps->slice_caps.bits.NUM_SLICES_8; + dsc_common_caps->slice_caps.bits.NUM_SLICES_12 = + dsc_sink_caps->slice_caps1.bits.NUM_SLICES_12 && dsc_enc_caps->slice_caps.bits.NUM_SLICES_12; + dsc_common_caps->slice_caps.bits.NUM_SLICES_16 = + dsc_sink_caps->slice_caps2.bits.NUM_SLICES_16 && dsc_enc_caps->slice_caps.bits.NUM_SLICES_16; + if (!dsc_common_caps->slice_caps.raw) return false; @@ -216,7 +723,8 @@ static bool intersect_dsc_caps( if (!dsc_common_caps->lb_bit_depth) return false; - dsc_common_caps->is_block_pred_supported = dsc_sink_caps->is_block_pred_supported && dsc_enc_caps->is_block_pred_supported; + dsc_common_caps->is_block_pred_supported = + dsc_sink_caps->is_block_pred_supported && dsc_enc_caps->is_block_pred_supported; dsc_common_caps->color_formats.raw = dsc_sink_caps->color_formats.raw & dsc_enc_caps->color_formats.raw; if (!dsc_common_caps->color_formats.raw) @@ -252,56 +760,26 @@ static bool intersect_dsc_caps( if (pixel_encoding == PIXEL_ENCODING_YCBCR422 || pixel_encoding == PIXEL_ENCODING_YCBCR420) dsc_common_caps->bpp_increment_div = min(dsc_common_caps->bpp_increment_div, (uint32_t)8); + dsc_common_caps->edp_sink_max_bits_per_pixel = dsc_sink_caps->edp_max_bits_per_pixel; dsc_common_caps->is_dp = dsc_sink_caps->is_dp; return true; } -static inline uint32_t dsc_div_by_10_round_up(uint32_t value) -{ - return (value + 9) / 10; -} - -static struct fixed31_32 compute_dsc_max_bandwidth_overhead( - const struct dc_crtc_timing *timing, - const int num_slices_h, - const bool is_dp) -{ - struct fixed31_32 max_dsc_overhead; - struct fixed31_32 refresh_rate; - - if (dsc_policy_disable_dsc_stream_overhead || !is_dp) - return dc_fixpt_from_int(0); - - /* use target bpp that can take entire target bandwidth */ - refresh_rate = dc_fixpt_from_int(timing->pix_clk_100hz); - refresh_rate = dc_fixpt_div_int(refresh_rate, timing->h_total); - refresh_rate = dc_fixpt_div_int(refresh_rate, timing->v_total); - refresh_rate = dc_fixpt_mul_int(refresh_rate, 100); - - max_dsc_overhead = dc_fixpt_from_int(num_slices_h); - max_dsc_overhead = dc_fixpt_mul_int(max_dsc_overhead, timing->v_total); - max_dsc_overhead = dc_fixpt_mul_int(max_dsc_overhead, 256); - max_dsc_overhead = dc_fixpt_div_int(max_dsc_overhead, 1000); - max_dsc_overhead = dc_fixpt_mul(max_dsc_overhead, refresh_rate); - - return max_dsc_overhead; -} - static uint32_t compute_bpp_x16_from_target_bandwidth( - const uint32_t bandwidth_in_kbps, - const struct dc_crtc_timing *timing, - const uint32_t num_slices_h, - const uint32_t bpp_increment_div, - const bool is_dp) + const uint32_t bandwidth_in_kbps, + const struct dc_crtc_timing *timing, + const uint32_t num_slices_h, + const uint32_t bpp_increment_div, + const bool is_dp) { - struct fixed31_32 overhead_in_kbps; + uint32_t overhead_in_kbps; struct fixed31_32 effective_bandwidth_in_kbps; struct fixed31_32 bpp_x16; - overhead_in_kbps = compute_dsc_max_bandwidth_overhead( + overhead_in_kbps = dc_dsc_stream_bandwidth_overhead_in_kbps( timing, num_slices_h, is_dp); effective_bandwidth_in_kbps = dc_fixpt_from_int(bandwidth_in_kbps); - effective_bandwidth_in_kbps = dc_fixpt_sub(effective_bandwidth_in_kbps, + effective_bandwidth_in_kbps = dc_fixpt_sub_int(effective_bandwidth_in_kbps, overhead_in_kbps); bpp_x16 = dc_fixpt_mul_int(effective_bandwidth_in_kbps, 10); bpp_x16 = dc_fixpt_div_int(bpp_x16, timing->pix_clk_100hz); @@ -311,103 +789,108 @@ static uint32_t compute_bpp_x16_from_target_bandwidth( return dc_fixpt_floor(bpp_x16); } -/* Get DSC bandwidth range based on [min_bpp, max_bpp] target bitrate range, and timing's pixel clock - * and uncompressed bandwidth. +/* Decide DSC bandwidth range based on signal, timing, specs specific and input min and max + * requirements. + * The range output includes decided min/max target bpp, the respective bandwidth requirements + * and native timing bandwidth requirement when DSC is not used. */ -static void get_dsc_bandwidth_range( +static bool decide_dsc_bandwidth_range( const uint32_t min_bpp_x16, const uint32_t max_bpp_x16, const uint32_t num_slices_h, const struct dsc_enc_caps *dsc_caps, const struct dc_crtc_timing *timing, + const enum dc_link_encoding_format link_encoding, struct dc_dsc_bw_range *range) { - /* native stream bandwidth */ - range->stream_kbps = dc_bandwidth_in_kbps_from_timing(timing); - - /* max dsc target bpp */ - range->max_kbps = dc_dsc_stream_bandwidth_in_kbps(timing, - max_bpp_x16, num_slices_h, dsc_caps->is_dp); - range->max_target_bpp_x16 = max_bpp_x16; - if (range->max_kbps > range->stream_kbps) { - /* max dsc target bpp is capped to native bandwidth */ - range->max_kbps = range->stream_kbps; - range->max_target_bpp_x16 = compute_bpp_x16_from_target_bandwidth( - range->max_kbps, timing, num_slices_h, - dsc_caps->bpp_increment_div, - dsc_caps->is_dp); + uint32_t preferred_bpp_x16 = timing->dsc_fixed_bits_per_pixel_x16; + + memset(range, 0, sizeof(*range)); + + /* apply signal, timing, specs and explicitly specified DSC range requirements */ + if (preferred_bpp_x16) { + if (preferred_bpp_x16 <= max_bpp_x16 && + preferred_bpp_x16 >= min_bpp_x16) { + range->max_target_bpp_x16 = preferred_bpp_x16; + range->min_target_bpp_x16 = preferred_bpp_x16; + } } + /* TODO - make this value generic to all signal types */ + else if (dsc_caps->edp_sink_max_bits_per_pixel) { + /* apply max bpp limitation from edp sink */ + range->max_target_bpp_x16 = MIN(dsc_caps->edp_sink_max_bits_per_pixel, + max_bpp_x16); + range->min_target_bpp_x16 = min_bpp_x16; + } + else { + range->max_target_bpp_x16 = max_bpp_x16; + range->min_target_bpp_x16 = min_bpp_x16; + } + + /* populate output structure */ + if (range->max_target_bpp_x16 >= range->min_target_bpp_x16 && range->min_target_bpp_x16 > 0) { + /* native stream bandwidth */ + range->stream_kbps = dc_bandwidth_in_kbps_from_timing(timing, link_encoding); - /* min dsc target bpp */ - range->min_kbps = dc_dsc_stream_bandwidth_in_kbps(timing, - min_bpp_x16, num_slices_h, dsc_caps->is_dp); - range->min_target_bpp_x16 = min_bpp_x16; - if (range->min_kbps > range->max_kbps) { - /* min dsc target bpp is capped to max dsc bandwidth*/ - range->min_kbps = range->max_kbps; - range->min_target_bpp_x16 = range->max_target_bpp_x16; + /* max dsc target bpp */ + range->max_kbps = dc_dsc_stream_bandwidth_in_kbps(timing, + range->max_target_bpp_x16, num_slices_h, dsc_caps->is_dp); + + /* min dsc target bpp */ + range->min_kbps = dc_dsc_stream_bandwidth_in_kbps(timing, + range->min_target_bpp_x16, num_slices_h, dsc_caps->is_dp); } + + return range->max_kbps >= range->min_kbps && range->min_kbps > 0; } /* Decides if DSC should be used and calculates target bpp if it should, applying DSC policy. * * Returns: - * - 'true' if DSC was required by policy and was successfully applied - * - 'false' if DSC was not necessary (e.g. if uncompressed stream fits 'target_bandwidth_kbps'), - * or if it couldn't be applied based on DSC policy. + * - 'true' if target bpp is decided + * - 'false' if target bpp cannot be decided (e.g. cannot fit even with min DSC bpp), */ static bool decide_dsc_target_bpp_x16( const struct dc_dsc_policy *policy, + const struct dc_dsc_config_options *options, const struct dsc_enc_caps *dsc_common_caps, const int target_bandwidth_kbps, const struct dc_crtc_timing *timing, const int num_slices_h, + const enum dc_link_encoding_format link_encoding, int *target_bpp_x16) { - bool should_use_dsc = false; struct dc_dsc_bw_range range; - memset(&range, 0, sizeof(range)); - - get_dsc_bandwidth_range(policy->min_target_bpp * 16, policy->max_target_bpp * 16, - num_slices_h, dsc_common_caps, timing, &range); - if (!policy->enable_dsc_when_not_needed && target_bandwidth_kbps >= range.stream_kbps) { - /* enough bandwidth without dsc */ - *target_bpp_x16 = 0; - should_use_dsc = false; - } else if (policy->preferred_bpp_x16 > 0 && - policy->preferred_bpp_x16 <= range.max_target_bpp_x16 && - policy->preferred_bpp_x16 >= range.min_target_bpp_x16) { - *target_bpp_x16 = policy->preferred_bpp_x16; - should_use_dsc = true; - } else if (target_bandwidth_kbps >= range.max_kbps) { - /* use max target bpp allowed */ - *target_bpp_x16 = range.max_target_bpp_x16; - should_use_dsc = true; - } else if (target_bandwidth_kbps >= range.min_kbps) { - /* use target bpp that can take entire target bandwidth */ - *target_bpp_x16 = compute_bpp_x16_from_target_bandwidth( - target_bandwidth_kbps, timing, num_slices_h, - dsc_common_caps->bpp_increment_div, - dsc_common_caps->is_dp); - should_use_dsc = true; - } else { - /* not enough bandwidth to fulfill minimum requirement */ - *target_bpp_x16 = 0; - should_use_dsc = false; + *target_bpp_x16 = 0; + + if (decide_dsc_bandwidth_range(policy->min_target_bpp * 16, policy->max_target_bpp * 16, + num_slices_h, dsc_common_caps, timing, link_encoding, &range)) { + if (target_bandwidth_kbps >= range.stream_kbps) { + if (policy->enable_dsc_when_not_needed || options->force_dsc_when_not_needed) + /* enable max bpp even dsc is not needed */ + *target_bpp_x16 = range.max_target_bpp_x16; + } else if (target_bandwidth_kbps >= range.max_kbps) { + /* use max target bpp allowed */ + *target_bpp_x16 = range.max_target_bpp_x16; + } else if (target_bandwidth_kbps >= range.min_kbps) { + /* use target bpp that can take entire target bandwidth */ + *target_bpp_x16 = compute_bpp_x16_from_target_bandwidth( + target_bandwidth_kbps, timing, num_slices_h, + dsc_common_caps->bpp_increment_div, + dsc_common_caps->is_dp); + } } - return should_use_dsc; + return *target_bpp_x16 != 0; } -#define MIN_AVAILABLE_SLICES_SIZE 4 +#define MIN_AVAILABLE_SLICES_SIZE 6 static int get_available_dsc_slices(union dsc_enc_slice_caps slice_caps, int *available_slices) { int idx = 0; - memset(available_slices, -1, MIN_AVAILABLE_SLICES_SIZE); - if (slice_caps.bits.NUM_SLICES_1) available_slices[idx++] = 1; @@ -420,6 +903,12 @@ static int get_available_dsc_slices(union dsc_enc_slice_caps slice_caps, int *av if (slice_caps.bits.NUM_SLICES_8) available_slices[idx++] = 8; + if (slice_caps.bits.NUM_SLICES_12) + available_slices[idx++] = 12; + + if (slice_caps.bits.NUM_SLICES_16) + available_slices[idx++] = 16; + return idx; } @@ -437,7 +926,7 @@ static int get_max_dsc_slices(union dsc_enc_slice_caps slice_caps) } -// Increment sice number in available sice numbers stops if possible, or just increment if not +// Increment slice number in available slice numbers stops if possible, or just increment if not static int inc_num_slices(union dsc_enc_slice_caps slice_caps, int num_slices) { // Get next bigger num slices available in common caps @@ -461,14 +950,14 @@ static int inc_num_slices(union dsc_enc_slice_caps slice_caps, int num_slices) } } - if (new_num_slices == num_slices) // No biger number of slices found + if (new_num_slices == num_slices) // No bigger number of slices found new_num_slices++; return new_num_slices; } -// Decrement sice number in available sice numbers stops if possible, or just decrement if not. Stop at zero. +// Decrement slice number in available slice numbers stops if possible, or just decrement if not. Stop at zero. static int dec_num_slices(union dsc_enc_slice_caps slice_caps, int num_slices) { // Get next bigger num slices available in common caps @@ -558,14 +1047,14 @@ static bool setup_dsc_config( const struct dsc_enc_caps *dsc_enc_caps, int target_bandwidth_kbps, const struct dc_crtc_timing *timing, - int min_slice_height_override, - int max_dsc_target_bpp_limit_override_x16, + const struct dc_dsc_config_options *options, + const enum dc_link_encoding_format link_encoding, + int min_slices_h, struct dc_dsc_config *dsc_cfg) { struct dsc_enc_caps dsc_common_caps; - int max_slices_h; - int min_slices_h; - int num_slices_h; + int max_slices_h = 0; + int num_slices_h = 0; int pic_width; int slice_width; int target_bpp; @@ -578,7 +1067,7 @@ static bool setup_dsc_config( memset(dsc_cfg, 0, sizeof(struct dc_dsc_config)); - dc_dsc_get_policy_for_timing(timing, max_dsc_target_bpp_limit_override_x16, &policy); + dc_dsc_get_policy_for_timing(timing, options->max_target_bpp_limit_override_x16, &policy, link_encoding); pic_width = timing->h_addressable + timing->h_border_left + timing->h_border_right; pic_height = timing->v_addressable + timing->v_border_top + timing->v_border_bottom; @@ -668,12 +1157,14 @@ static bool setup_dsc_config( if (!is_dsc_possible) goto done; - min_slices_h = pic_width / dsc_common_caps.max_slice_width; - if (pic_width % dsc_common_caps.max_slice_width) - min_slices_h++; + /* increase miniumum slice count to meet sink slice width limitations */ + min_slices_h = dc_fixpt_ceil(dc_fixpt_max( + dc_fixpt_div_int(dc_fixpt_from_int(pic_width), dsc_common_caps.max_slice_width), // sink min + dc_fixpt_from_int(min_slices_h))); // source min min_slices_h = fit_num_slices_up(dsc_common_caps.slice_caps, min_slices_h); + /* increase minimum slice count to meet sink throughput limitations */ while (min_slices_h <= max_slices_h) { int pix_clk_per_slice_khz = dsc_div_by_10_round_up(timing->pix_clk_100hz) / min_slices_h; if (pix_clk_per_slice_khz <= sink_per_slice_throughput_mps * 1000) @@ -682,10 +1173,12 @@ static bool setup_dsc_config( min_slices_h = inc_num_slices(dsc_common_caps.slice_caps, min_slices_h); } - if (pic_width % min_slices_h != 0) - min_slices_h = 0; // DSC TODO: Maybe try increasing the number of slices first? + /* increase minimum slice count to meet divisibility requirements */ + while (pic_width % min_slices_h != 0 && min_slices_h <= max_slices_h) { + min_slices_h = inc_num_slices(dsc_common_caps.slice_caps, min_slices_h); + } - is_dsc_possible = (min_slices_h <= max_slices_h); + is_dsc_possible = (min_slices_h <= max_slices_h) && max_slices_h != 0; if (!is_dsc_possible) goto done; @@ -710,7 +1203,30 @@ static bool setup_dsc_config( else is_dsc_possible = false; } - + // When we force ODM, num dsc h slices must be divisible by num odm h slices + switch (options->dsc_force_odm_hslice_override) { + case 0: + case 1: + break; + case 2: + if (num_slices_h < 2) + num_slices_h = fit_num_slices_up(dsc_common_caps.slice_caps, 2); + break; + case 3: + if (dsc_common_caps.slice_caps.bits.NUM_SLICES_12) + num_slices_h = 12; + else + num_slices_h = 0; + break; + case 4: + if (num_slices_h < 4) + num_slices_h = fit_num_slices_up(dsc_common_caps.slice_caps, 4); + break; + default: + break; + } + if (num_slices_h == 0) + is_dsc_possible = false; if (!is_dsc_possible) goto done; @@ -723,12 +1239,13 @@ static bool setup_dsc_config( // Slice height (i.e. number of slices per column): start with policy and pick the first one that height is divisible by. // For 4:2:0 make sure the slice height is divisible by 2 as well. - if (min_slice_height_override == 0) + if (options->dsc_min_slice_height_override == 0) slice_height = min(policy.min_slice_height, pic_height); else - slice_height = min(min_slice_height_override, pic_height); + slice_height = min((int)(options->dsc_min_slice_height_override), pic_height); while (slice_height < pic_height && (pic_height % slice_height != 0 || + slice_height % options->slice_height_granularity != 0 || (timing->pixel_encoding == PIXEL_ENCODING_YCBCR420 && slice_height % 2 != 0))) slice_height++; @@ -738,29 +1255,33 @@ static bool setup_dsc_config( if (!is_dsc_possible) goto done; - dsc_cfg->num_slices_v = pic_height/slice_height; + if (slice_height > 0) { + dsc_cfg->num_slices_v = pic_height / slice_height; + } else { + is_dsc_possible = false; + goto done; + } if (target_bandwidth_kbps > 0) { is_dsc_possible = decide_dsc_target_bpp_x16( &policy, + options, &dsc_common_caps, target_bandwidth_kbps, timing, num_slices_h, + link_encoding, &target_bpp); dsc_cfg->bits_per_pixel = target_bpp; } if (!is_dsc_possible) goto done; - // Final decission: can we do DSC or not? - if (is_dsc_possible) { - // Fill out the rest of DSC settings - dsc_cfg->block_pred_enable = dsc_common_caps.is_block_pred_supported; - dsc_cfg->linebuf_depth = dsc_common_caps.lb_bit_depth; - dsc_cfg->version_minor = (dsc_common_caps.dsc_version & 0xf0) >> 4; - dsc_cfg->is_dp = dsc_sink_caps->is_dp; - } + /* Fill out the rest of DSC settings */ + dsc_cfg->block_pred_enable = dsc_common_caps.is_block_pred_supported; + dsc_cfg->linebuf_depth = dsc_common_caps.lb_bit_depth; + dsc_cfg->version_minor = (dsc_common_caps.dsc_version & 0xf0) >> 4; + dsc_cfg->is_dp = dsc_sink_caps->is_dp; done: if (!is_dsc_possible) @@ -769,184 +1290,79 @@ done: return is_dsc_possible; } -bool dc_dsc_parse_dsc_dpcd(const struct dc *dc, const uint8_t *dpcd_dsc_basic_data, const uint8_t *dpcd_dsc_branch_decoder_caps, struct dsc_dec_dpcd_caps *dsc_sink_caps) -{ - if (!dpcd_dsc_basic_data) - return false; - - dsc_sink_caps->is_dsc_supported = (dpcd_dsc_basic_data[DP_DSC_SUPPORT - DP_DSC_SUPPORT] & DP_DSC_DECOMPRESSION_IS_SUPPORTED) != 0; - if (!dsc_sink_caps->is_dsc_supported) - return false; - - dsc_sink_caps->dsc_version = dpcd_dsc_basic_data[DP_DSC_REV - DP_DSC_SUPPORT]; - - { - int buff_block_size; - int buff_size; - - if (!dsc_buff_block_size_from_dpcd(dpcd_dsc_basic_data[DP_DSC_RC_BUF_BLK_SIZE - DP_DSC_SUPPORT], &buff_block_size)) - return false; - - buff_size = dpcd_dsc_basic_data[DP_DSC_RC_BUF_SIZE - DP_DSC_SUPPORT] + 1; - dsc_sink_caps->rc_buffer_size = buff_size * buff_block_size; - } - - dsc_sink_caps->slice_caps1.raw = dpcd_dsc_basic_data[DP_DSC_SLICE_CAP_1 - DP_DSC_SUPPORT]; - if (!dsc_line_buff_depth_from_dpcd(dpcd_dsc_basic_data[DP_DSC_LINE_BUF_BIT_DEPTH - DP_DSC_SUPPORT], &dsc_sink_caps->lb_bit_depth)) - return false; - - dsc_sink_caps->is_block_pred_supported = - (dpcd_dsc_basic_data[DP_DSC_BLK_PREDICTION_SUPPORT - DP_DSC_SUPPORT] & DP_DSC_BLK_PREDICTION_IS_SUPPORTED) != 0; - - dsc_sink_caps->edp_max_bits_per_pixel = - dpcd_dsc_basic_data[DP_DSC_MAX_BITS_PER_PIXEL_LOW - DP_DSC_SUPPORT] | - dpcd_dsc_basic_data[DP_DSC_MAX_BITS_PER_PIXEL_HI - DP_DSC_SUPPORT] << 8; - - dsc_sink_caps->color_formats.raw = dpcd_dsc_basic_data[DP_DSC_DEC_COLOR_FORMAT_CAP - DP_DSC_SUPPORT]; - dsc_sink_caps->color_depth.raw = dpcd_dsc_basic_data[DP_DSC_DEC_COLOR_DEPTH_CAP - DP_DSC_SUPPORT]; - - { - int dpcd_throughput = dpcd_dsc_basic_data[DP_DSC_PEAK_THROUGHPUT - DP_DSC_SUPPORT]; - - if (!dsc_throughput_from_dpcd(dpcd_throughput & DP_DSC_THROUGHPUT_MODE_0_MASK, &dsc_sink_caps->throughput_mode_0_mps)) - return false; - - dpcd_throughput = (dpcd_throughput & DP_DSC_THROUGHPUT_MODE_1_MASK) >> DP_DSC_THROUGHPUT_MODE_1_SHIFT; - if (!dsc_throughput_from_dpcd(dpcd_throughput, &dsc_sink_caps->throughput_mode_1_mps)) - return false; - } - - dsc_sink_caps->max_slice_width = dpcd_dsc_basic_data[DP_DSC_MAX_SLICE_WIDTH - DP_DSC_SUPPORT] * 320; - dsc_sink_caps->slice_caps2.raw = dpcd_dsc_basic_data[DP_DSC_SLICE_CAP_2 - DP_DSC_SUPPORT]; - - if (!dsc_bpp_increment_div_from_dpcd(dpcd_dsc_basic_data[DP_DSC_BITS_PER_PIXEL_INC - DP_DSC_SUPPORT], &dsc_sink_caps->bpp_increment_div)) - return false; - - if (dc->debug.dsc_bpp_increment_div) { - /* dsc_bpp_increment_div should onl be 1, 2, 4, 8 or 16, but rather than rejecting invalid values, - * we'll accept all and get it into range. This also makes the above check against 0 redundant, - * but that one stresses out the override will be only used if it's not 0. - */ - if (dc->debug.dsc_bpp_increment_div >= 1) - dsc_sink_caps->bpp_increment_div = 1; - if (dc->debug.dsc_bpp_increment_div >= 2) - dsc_sink_caps->bpp_increment_div = 2; - if (dc->debug.dsc_bpp_increment_div >= 4) - dsc_sink_caps->bpp_increment_div = 4; - if (dc->debug.dsc_bpp_increment_div >= 8) - dsc_sink_caps->bpp_increment_div = 8; - if (dc->debug.dsc_bpp_increment_div >= 16) - dsc_sink_caps->bpp_increment_div = 16; - } - - /* Extended caps */ - if (dpcd_dsc_branch_decoder_caps == NULL) { // branch decoder DPCD DSC data can be null for non branch device - dsc_sink_caps->branch_overall_throughput_0_mps = 0; - dsc_sink_caps->branch_overall_throughput_1_mps = 0; - dsc_sink_caps->branch_max_line_width = 0; - return true; - } - - dsc_sink_caps->branch_overall_throughput_0_mps = dpcd_dsc_branch_decoder_caps[DP_DSC_BRANCH_OVERALL_THROUGHPUT_0 - DP_DSC_BRANCH_OVERALL_THROUGHPUT_0]; - if (dsc_sink_caps->branch_overall_throughput_0_mps == 0) - dsc_sink_caps->branch_overall_throughput_0_mps = 0; - else if (dsc_sink_caps->branch_overall_throughput_0_mps == 1) - dsc_sink_caps->branch_overall_throughput_0_mps = 680; - else { - dsc_sink_caps->branch_overall_throughput_0_mps *= 50; - dsc_sink_caps->branch_overall_throughput_0_mps += 600; - } - - dsc_sink_caps->branch_overall_throughput_1_mps = dpcd_dsc_branch_decoder_caps[DP_DSC_BRANCH_OVERALL_THROUGHPUT_1 - DP_DSC_BRANCH_OVERALL_THROUGHPUT_0]; - if (dsc_sink_caps->branch_overall_throughput_1_mps == 0) - dsc_sink_caps->branch_overall_throughput_1_mps = 0; - else if (dsc_sink_caps->branch_overall_throughput_1_mps == 1) - dsc_sink_caps->branch_overall_throughput_1_mps = 680; - else { - dsc_sink_caps->branch_overall_throughput_1_mps *= 50; - dsc_sink_caps->branch_overall_throughput_1_mps += 600; - } - - dsc_sink_caps->branch_max_line_width = dpcd_dsc_branch_decoder_caps[DP_DSC_BRANCH_MAX_LINE_WIDTH - DP_DSC_BRANCH_OVERALL_THROUGHPUT_0] * 320; - ASSERT(dsc_sink_caps->branch_max_line_width == 0 || dsc_sink_caps->branch_max_line_width >= 5120); - - dsc_sink_caps->is_dp = true; - return true; -} - - -/* If DSC is possbile, get DSC bandwidth range based on [min_bpp, max_bpp] target bitrate range and - * timing's pixel clock and uncompressed bandwidth. - * If DSC is not possible, leave '*range' untouched. - */ -bool dc_dsc_compute_bandwidth_range( - const struct display_stream_compressor *dsc, - uint32_t dsc_min_slice_height_override, - uint32_t min_bpp_x16, - uint32_t max_bpp_x16, - const struct dsc_dec_dpcd_caps *dsc_sink_caps, - const struct dc_crtc_timing *timing, - struct dc_dsc_bw_range *range) -{ - bool is_dsc_possible = false; - struct dsc_enc_caps dsc_enc_caps; - struct dsc_enc_caps dsc_common_caps; - struct dc_dsc_config config; - - get_dsc_enc_caps(dsc, &dsc_enc_caps, timing->pix_clk_100hz); - - is_dsc_possible = intersect_dsc_caps(dsc_sink_caps, &dsc_enc_caps, - timing->pixel_encoding, &dsc_common_caps); - - if (is_dsc_possible) - is_dsc_possible = setup_dsc_config(dsc_sink_caps, &dsc_enc_caps, 0, timing, - dsc_min_slice_height_override, max_bpp_x16, &config); - - if (is_dsc_possible) - get_dsc_bandwidth_range(min_bpp_x16, max_bpp_x16, - config.num_slices_h, &dsc_common_caps, timing, range); - - return is_dsc_possible; -} - bool dc_dsc_compute_config( const struct display_stream_compressor *dsc, const struct dsc_dec_dpcd_caps *dsc_sink_caps, - uint32_t dsc_min_slice_height_override, - uint32_t max_target_bpp_limit_override, + const struct dc_dsc_config_options *options, uint32_t target_bandwidth_kbps, const struct dc_crtc_timing *timing, + const enum dc_link_encoding_format link_encoding, struct dc_dsc_config *dsc_cfg) { bool is_dsc_possible = false; struct dsc_enc_caps dsc_enc_caps; - + unsigned int min_dsc_slice_count; get_dsc_enc_caps(dsc, &dsc_enc_caps, timing->pix_clk_100hz); + + min_dsc_slice_count = get_min_dsc_slice_count_for_odm(dsc, &dsc_enc_caps, timing); + is_dsc_possible = setup_dsc_config(dsc_sink_caps, - &dsc_enc_caps, - target_bandwidth_kbps, - timing, dsc_min_slice_height_override, - max_target_bpp_limit_override * 16, dsc_cfg); + &dsc_enc_caps, + target_bandwidth_kbps, + timing, + options, + link_encoding, + min_dsc_slice_count, + dsc_cfg); return is_dsc_possible; } uint32_t dc_dsc_stream_bandwidth_in_kbps(const struct dc_crtc_timing *timing, - uint32_t bpp_x16, uint32_t num_slices_h, bool is_dp) + uint32_t bpp_x16, uint32_t num_slices_h, bool is_dp) { - struct fixed31_32 overhead_in_kbps; + uint32_t overhead_in_kbps; struct fixed31_32 bpp; struct fixed31_32 actual_bandwidth_in_kbps; - overhead_in_kbps = compute_dsc_max_bandwidth_overhead( - timing, num_slices_h, is_dp); + overhead_in_kbps = dc_dsc_stream_bandwidth_overhead_in_kbps( + timing, num_slices_h, is_dp); bpp = dc_fixpt_from_fraction(bpp_x16, 16); actual_bandwidth_in_kbps = dc_fixpt_from_fraction(timing->pix_clk_100hz, 10); actual_bandwidth_in_kbps = dc_fixpt_mul(actual_bandwidth_in_kbps, bpp); - actual_bandwidth_in_kbps = dc_fixpt_add(actual_bandwidth_in_kbps, overhead_in_kbps); + actual_bandwidth_in_kbps = dc_fixpt_add_int(actual_bandwidth_in_kbps, overhead_in_kbps); return dc_fixpt_ceil(actual_bandwidth_in_kbps); } -void dc_dsc_get_policy_for_timing(const struct dc_crtc_timing *timing, uint32_t max_target_bpp_limit_override_x16, struct dc_dsc_policy *policy) +uint32_t dc_dsc_stream_bandwidth_overhead_in_kbps( + const struct dc_crtc_timing *timing, + const int num_slices_h, + const bool is_dp) +{ + struct fixed31_32 max_dsc_overhead; + struct fixed31_32 refresh_rate; + + if (dsc_policy_disable_dsc_stream_overhead || !is_dp) + return 0; + + /* use target bpp that can take entire target bandwidth */ + refresh_rate = dc_fixpt_from_int(timing->pix_clk_100hz); + refresh_rate = dc_fixpt_div_int(refresh_rate, timing->h_total); + refresh_rate = dc_fixpt_div_int(refresh_rate, timing->v_total); + refresh_rate = dc_fixpt_mul_int(refresh_rate, 100); + + max_dsc_overhead = dc_fixpt_from_int(num_slices_h); + max_dsc_overhead = dc_fixpt_mul_int(max_dsc_overhead, timing->v_total); + max_dsc_overhead = dc_fixpt_mul_int(max_dsc_overhead, 256); + max_dsc_overhead = dc_fixpt_div_int(max_dsc_overhead, 1000); + max_dsc_overhead = dc_fixpt_mul(max_dsc_overhead, refresh_rate); + + return dc_fixpt_ceil(max_dsc_overhead); +} + +void dc_dsc_get_policy_for_timing(const struct dc_crtc_timing *timing, + uint32_t max_target_bpp_limit_override_x16, + struct dc_dsc_policy *policy, + const enum dc_link_encoding_format link_encoding) { uint32_t bpc = 0; @@ -1001,8 +1417,6 @@ void dc_dsc_get_policy_for_timing(const struct dc_crtc_timing *timing, uint32_t return; } - policy->preferred_bpp_x16 = timing->dsc_fixed_bits_per_pixel_x16; - /* internal upper limit, default 16 bpp */ if (policy->max_target_bpp > dsc_policy_max_target_bpp_limit) policy->max_target_bpp = dsc_policy_max_target_bpp_limit; @@ -1012,10 +1426,7 @@ void dc_dsc_get_policy_for_timing(const struct dc_crtc_timing *timing, uint32_t policy->max_target_bpp = max_target_bpp_limit_override_x16 / 16; /* enable DSC when not needed, default false */ - if (dsc_policy_enable_dsc_when_not_needed) - policy->enable_dsc_when_not_needed = dsc_policy_enable_dsc_when_not_needed; - else - policy->enable_dsc_when_not_needed = false; + policy->enable_dsc_when_not_needed = dsc_policy_enable_dsc_when_not_needed; } void dc_dsc_policy_set_max_target_bpp_limit(uint32_t limit) @@ -1032,3 +1443,17 @@ void dc_dsc_policy_set_disable_dsc_stream_overhead(bool disable) { dsc_policy_disable_dsc_stream_overhead = disable; } + +void dc_set_disable_128b_132b_stream_overhead(bool disable) +{ + disable_128b_132b_stream_overhead = disable; +} + +void dc_dsc_get_default_config_option(const struct dc *dc, struct dc_dsc_config_options *options) +{ + options->dsc_min_slice_height_override = dc->debug.dsc_min_slice_height_override; + options->dsc_force_odm_hslice_override = dc->debug.force_odm_combine; + options->max_target_bpp_limit_override_x16 = 0; + options->slice_height_granularity = 1; + options->force_dsc_when_not_needed = false; +} |