bybrooklyn/linux
1
0
Fork 0
mirror of https://github.com/torvalds/linux.git synced 2026-05-02 21:42:42 -04:00
Code Issues Packages Projects Releases Wiki Activity

drm/amd/display: Optimizations for DML math

Browse Source 
[why]
Conditionals in the DML basic math functions significantly impact mode
enumeration.

[how]
Remove conditionals for floor/ceil operations which are used frequently
in DML and add an assertion for invalid callers using zero granuality.
Fix existing callers that rely on 0 granularity.

Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Reviewed-by: Nevenko Stupar <Nevenko.Stupar@amd.com>
Acked-by: Pavle Kotarac <Pavle.Kotarac@amd.com>
Signed-off-by: Aric Cyr <aric.cyr@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This commit is contained in:
Aric Cyr
2022-08-20 17:20:30 -04:00
committed by Alex Deucher
parent 6da15a236c
commit 17529ea2ac
9 changed files with 53 additions and 61 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
drivers/gpu/drm/amd/display/dc/dml/calcs/dcn_calc_auto.c
View File

@@ -526,10 +526,10 @@ void mode_support_and_system_configuration(struct dcn_bw_internal_vars *v)
}
if (v->max_swath_height_c[k] > 0.0) {
v->swath_width_granularity_c = 256.0 /dcn_bw_ceil2(v->byte_per_pixel_in_detc[k], 2.0) / v->max_swath_height_c[k];
}
v->rounded_up_max_swath_size_bytes_c = (dcn_bw_ceil2(v->swath_width_yper_state[i][j][k] / 2.0 - 1.0, v->swath_width_granularity_c) + v->swath_width_granularity_c) * v->byte_per_pixel_in_detc[k] * v->max_swath_height_c[k];
if (v->source_pixel_format[k] == dcn_bw_yuv420_sub_10) {
v->rounded_up_max_swath_size_bytes_c =dcn_bw_ceil2(v->rounded_up_max_swath_size_bytes_c, 256.0) + 256;
v->rounded_up_max_swath_size_bytes_c = (dcn_bw_ceil2(v->swath_width_yper_state[i][j][k] / 2.0 - 1.0, v->swath_width_granularity_c) + v->swath_width_granularity_c) * v->byte_per_pixel_in_detc[k] * v->max_swath_height_c[k];
if (v->source_pixel_format[k] == dcn_bw_yuv420_sub_10) {
v->rounded_up_max_swath_size_bytes_c = dcn_bw_ceil2(v->rounded_up_max_swath_size_bytes_c, 256.0) + 256;
}
}
if (v->rounded_up_max_swath_size_bytes_y + v->rounded_up_max_swath_size_bytes_c <= v->det_buffer_size_in_kbyte * 1024.0 / 2.0) {
v->swath_height_yper_state[i][j][k] = v->max_swath_height_y[k];
@@ -552,14 +552,14 @@ void mode_support_and_system_configuration(struct dcn_bw_internal_vars *v)
v->lines_in_det_chroma = v->det_buffer_size_in_kbyte * 1024.0 / 3.0 / v->byte_per_pixel_in_dety[k] / (v->swath_width_yper_state[i][j][k] / 2.0);
}
v->effective_lb_latency_hiding_source_lines_luma =dcn_bw_min2(v->max_line_buffer_lines,dcn_bw_floor2(v->line_buffer_size / v->lb_bit_per_pixel[k] / (v->swath_width_yper_state[i][j][k] /dcn_bw_max2(v->h_ratio[k], 1.0)), 1.0)) - (v->vtaps[k] - 1.0);
v->effective_lb_latency_hiding_source_lines_chroma =dcn_bw_min2(v->max_line_buffer_lines,dcn_bw_floor2(v->line_buffer_size / v->lb_bit_per_pixel[k] / (v->swath_width_yper_state[i][j][k] / 2.0 /dcn_bw_max2(v->h_ratio[k] / 2.0, 1.0)), 1.0)) - (v->vta_pschroma[k] - 1.0);
v->effective_detlb_lines_luma =dcn_bw_floor2(v->lines_in_det_luma +dcn_bw_min2(v->lines_in_det_luma * v->required_dispclk[i][j] * v->byte_per_pixel_in_dety[k] * v->pscl_factor[k] / v->return_bw_per_state[i], v->effective_lb_latency_hiding_source_lines_luma), v->swath_height_yper_state[i][j][k]);
v->effective_detlb_lines_chroma =dcn_bw_floor2(v->lines_in_det_chroma +dcn_bw_min2(v->lines_in_det_chroma * v->required_dispclk[i][j] * v->byte_per_pixel_in_detc[k] * v->pscl_factor_chroma[k] / v->return_bw_per_state[i], v->effective_lb_latency_hiding_source_lines_chroma), v->swath_height_cper_state[i][j][k]);
if (v->byte_per_pixel_in_detc[k] == 0.0) {
v->urgent_latency_support_us_per_state[i][j][k] = v->effective_detlb_lines_luma * (v->htotal[k] / v->pixel_clock[k]) / v->v_ratio[k] - v->effective_detlb_lines_luma * v->swath_width_yper_state[i][j][k] *dcn_bw_ceil2(v->byte_per_pixel_in_dety[k], 1.0) / (v->return_bw_per_state[i] / v->no_of_dpp[i][j][k]);
}
else {
v->urgent_latency_support_us_per_state[i][j][k] =dcn_bw_min2(v->effective_detlb_lines_luma * (v->htotal[k] / v->pixel_clock[k]) / v->v_ratio[k] - v->effective_detlb_lines_luma * v->swath_width_yper_state[i][j][k] *dcn_bw_ceil2(v->byte_per_pixel_in_dety[k], 1.0) / (v->return_bw_per_state[i] / v->no_of_dpp[i][j][k]), v->effective_detlb_lines_chroma * (v->htotal[k] / v->pixel_clock[k]) / (v->v_ratio[k] / 2.0) - v->effective_detlb_lines_chroma * v->swath_width_yper_state[i][j][k] / 2.0 *dcn_bw_ceil2(v->byte_per_pixel_in_detc[k], 2.0) / (v->return_bw_per_state[i] / v->no_of_dpp[i][j][k]));
v->effective_lb_latency_hiding_source_lines_chroma = dcn_bw_min2(v->max_line_buffer_lines, dcn_bw_floor2(v->line_buffer_size / v->lb_bit_per_pixel[k] / (v->swath_width_yper_state[i][j][k] / 2.0 / dcn_bw_max2(v->h_ratio[k] / 2.0, 1.0)), 1.0)) - (v->vta_pschroma[k] - 1.0);
v->effective_detlb_lines_chroma = dcn_bw_floor2(v->lines_in_det_chroma + dcn_bw_min2(v->lines_in_det_chroma * v->required_dispclk[i][j] * v->byte_per_pixel_in_detc[k] * v->pscl_factor_chroma[k] / v->return_bw_per_state[i], v->effective_lb_latency_hiding_source_lines_chroma), v->swath_height_cper_state[i][j][k]);
v->urgent_latency_support_us_per_state[i][j][k] = dcn_bw_min2(v->effective_detlb_lines_luma * (v->htotal[k] / v->pixel_clock[k]) / v->v_ratio[k] - v->effective_detlb_lines_luma * v->swath_width_yper_state[i][j][k] * dcn_bw_ceil2(v->byte_per_pixel_in_dety[k], 1.0) / (v->return_bw_per_state[i] / v->no_of_dpp[i][j][k]), v->effective_detlb_lines_chroma * (v->htotal[k] / v->pixel_clock[k]) / (v->v_ratio[k] / 2.0) - v->effective_detlb_lines_chroma * v->swath_width_yper_state[i][j][k] / 2.0 * dcn_bw_ceil2(v->byte_per_pixel_in_detc[k], 2.0) / (v->return_bw_per_state[i] / v->no_of_dpp[i][j][k]));
}
}
}
@@ -1146,10 +1146,10 @@ void display_pipe_configuration(struct dcn_bw_internal_vars *v)
}
if (v->maximum_swath_height_c > 0.0) {
v->swath_width_granularity_c = 256.0 /dcn_bw_ceil2(v->byte_per_pix_detc, 2.0) / v->maximum_swath_height_c;
}
v->rounded_up_max_swath_size_bytes_c = (dcn_bw_ceil2(v->swath_width / 2.0 - 1.0, v->swath_width_granularity_c) + v->swath_width_granularity_c) * v->byte_per_pix_detc * v->maximum_swath_height_c;
if (v->source_pixel_format[k] == dcn_bw_yuv420_sub_10) {
v->rounded_up_max_swath_size_bytes_c =dcn_bw_ceil2(v->rounded_up_max_swath_size_bytes_c, 256.0) + 256;
v->rounded_up_max_swath_size_bytes_c = (dcn_bw_ceil2(v->swath_width / 2.0 - 1.0, v->swath_width_granularity_c) + v->swath_width_granularity_c) * v->byte_per_pix_detc * v->maximum_swath_height_c;
if (v->source_pixel_format[k] == dcn_bw_yuv420_sub_10) {
v->rounded_up_max_swath_size_bytes_c = dcn_bw_ceil2(v->rounded_up_max_swath_size_bytes_c, 256.0) + 256;
}
}
if (v->rounded_up_max_swath_size_bytes_y + v->rounded_up_max_swath_size_bytes_c <= v->det_buffer_size_in_kbyte * 1024.0 / 2.0) {
v->swath_height_y[k] = v->maximum_swath_height_y;

16
drivers/gpu/drm/amd/display/dc/dml/calcs/dcn_calc_math.c
View File

@@ -23,6 +23,7 @@
*
*/
#include "os_types.h"
#include "dcn_calc_math.h"
#define isNaN(number) ((number) != (number))
@@ -69,8 +70,8 @@ float dcn_bw_max2(const float arg1, const float arg2)
float dcn_bw_floor2(const float arg, const float significance)
{
if (significance == 0)
return 0;
ASSERT(significance != 0);
return ((int) (arg / significance)) * significance;
}
float dcn_bw_floor(const float arg)
@@ -80,17 +81,14 @@ float dcn_bw_floor(const float arg)
float dcn_bw_ceil(const float arg)
{
float flr = dcn_bw_floor2(arg, 1);
return flr + 0.00001 >= arg ? arg : flr + 1;
return (int) (arg + 0.99999);
}
float dcn_bw_ceil2(const float arg, const float significance)
{
float flr = dcn_bw_floor2(arg, significance);
if (significance == 0)
return 0;
return flr + 0.00001 >= arg ? arg : flr + significance;
ASSERT(significance != 0);
return ((int) (arg / significance + 0.99999)) * significance;
}
float dcn_bw_max3(float v1, float v2, float v3)