[PATCH 02/27] drm/amd/display: Optimizations for DML math
Lin, Wayne
Wayne.Lin at amd.com
Mon Sep 12 09:01:39 UTC 2022
[AMD Official Use Only - General]
Please remove the file "/drivers/gpu/drm/amd/display/dc/dml/dcn32/display_mode_vba_util_32.c.rej" if it's not necessary.
Thanks,
Wayne
________________________________
From: Kotarac, Pavle <Pavle.Kotarac at amd.com>
Sent: Thursday, September 8, 2022 01:32
To: amd-gfx at lists.freedesktop.org <amd-gfx at lists.freedesktop.org>
Cc: Wentland, Harry <Harry.Wentland at amd.com>; Li, Sun peng (Leo) <Sunpeng.Li at amd.com>; Lakha, Bhawanpreet <Bhawanpreet.Lakha at amd.com>; Siqueira, Rodrigo <Rodrigo.Siqueira at amd.com>; Pillai, Aurabindo <Aurabindo.Pillai at amd.com>; Zhuo, Qingqing (Lillian) <Qingqing.Zhuo at amd.com>; Li, Roman <Roman.Li at amd.com>; Lin, Wayne <Wayne.Lin at amd.com>; Wang, Chao-kai (Stylon) <Stylon.Wang at amd.com>; Chiu, Solomon <Solomon.Chiu at amd.com>; Kotarac, Pavle <Pavle.Kotarac at amd.com>; Gutierrez, Agustin <Agustin.Gutierrez at amd.com>; Cyr, Aric <Aric.Cyr at amd.com>; Stupar, Nevenko <Nevenko.Stupar at amd.com>; Kotarac, Pavle <Pavle.Kotarac at amd.com>
Subject: [PATCH 02/27] drm/amd/display: Optimizations for DML math
From: Aric Cyr <aric.cyr at amd.com>
[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.
Reviewed-by: Nevenko Stupar <Nevenko.Stupar at amd.com>
Acked-by: Pavle Kotarac <Pavle.Kotarac at amd.com>
Signed-off-by: Aric Cyr <aric.cyr at amd.com>
---
.../amd/display/dc/dml/calcs/dcn_calc_auto.c | 22 ++++++-------
.../amd/display/dc/dml/calcs/dcn_calc_math.c | 16 +++++-----
.../dc/dml/dcn20/display_mode_vba_20v2.c | 10 +++---
.../dc/dml/dcn21/display_mode_vba_21.c | 6 ++--
.../dc/dml/dcn30/display_mode_vba_30.c | 8 ++---
.../dc/dml/dcn31/display_mode_vba_31.c | 6 ++--
.../dc/dml/dcn314/display_mode_vba_314.c | 6 ++--
.../dc/dml/dcn32/display_mode_vba_util_32.c | 31 +++++++++----------
.../dml/dcn32/display_mode_vba_util_32.c.rej | 12 +++++++
.../drm/amd/display/dc/dml/dml_inline_defs.h | 9 ++----
10 files changed, 65 insertions(+), 61 deletions(-)
create mode 100644 drivers/gpu/drm/amd/display/dc/dml/dcn32/display_mode_vba_util_32.c.rej
diff --git a/drivers/gpu/drm/amd/display/dc/dml/calcs/dcn_calc_auto.c b/drivers/gpu/drm/amd/display/dc/dml/calcs/dcn_calc_auto.c
index 41284e263325..288d22a16cf2 100644
--- a/drivers/gpu/drm/amd/display/dc/dml/calcs/dcn_calc_auto.c
+++ b/drivers/gpu/drm/amd/display/dc/dml/calcs/dcn_calc_auto.c
@@ -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;
diff --git a/drivers/gpu/drm/amd/display/dc/dml/calcs/dcn_calc_math.c b/drivers/gpu/drm/amd/display/dc/dml/calcs/dcn_calc_math.c
index 07d18e78de49..cac72413a097 100644
--- a/drivers/gpu/drm/amd/display/dc/dml/calcs/dcn_calc_math.c
+++ b/drivers/gpu/drm/amd/display/dc/dml/calcs/dcn_calc_math.c
@@ -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)
diff --git a/drivers/gpu/drm/amd/display/dc/dml/dcn20/display_mode_vba_20v2.c b/drivers/gpu/drm/amd/display/dc/dml/dcn20/display_mode_vba_20v2.c
index 63bbdf8b8678..edd098c7eb92 100644
--- a/drivers/gpu/drm/amd/display/dc/dml/dcn20/display_mode_vba_20v2.c
+++ b/drivers/gpu/drm/amd/display/dc/dml/dcn20/display_mode_vba_20v2.c
@@ -4478,17 +4478,17 @@ void dml20v2_ModeSupportAndSystemConfigurationFull(struct display_mode_lib *mode
locals->EffectiveLBLatencyHidingSourceLinesLuma),
locals->SwathHeightYPerState[i][j][k]);
- locals->EffectiveDETLBLinesChroma = dml_floor(locals->LinesInDETChroma + dml_min(
- locals->LinesInDETChroma * locals->RequiredDISPCLK[i][j] * locals->BytePerPixelInDETC[k] *
- locals->PSCL_FACTOR_CHROMA[k] / locals->ReturnBWPerState[i][0],
- locals->EffectiveLBLatencyHidingSourceLinesChroma),
- locals->SwathHeightCPerState[i][j][k]);
if (locals->BytePerPixelInDETC[k] == 0) {
locals->UrgentLatencySupportUsPerState[i][j][k] = locals->EffectiveDETLBLinesLuma * (locals->HTotal[k] / locals->PixelClock[k])
/ locals->VRatio[k] - locals->EffectiveDETLBLinesLuma * locals->SwathWidthYPerState[i][j][k] *
dml_ceil(locals->BytePerPixelInDETY[k], 1) / (locals->ReturnBWPerState[i][0] / locals->NoOfDPP[i][j][k]);
} else {
+ locals->EffectiveDETLBLinesChroma = dml_floor(locals->LinesInDETChroma + dml_min(
+ locals->LinesInDETChroma * locals->RequiredDISPCLK[i][j] * locals->BytePerPixelInDETC[k] *
+ locals->PSCL_FACTOR_CHROMA[k] / locals->ReturnBWPerState[i][0],
+ locals->EffectiveLBLatencyHidingSourceLinesChroma),
+ locals->SwathHeightCPerState[i][j][k]);
locals->UrgentLatencySupportUsPerState[i][j][k] = dml_min(
locals->EffectiveDETLBLinesLuma * (locals->HTotal[k] / locals->PixelClock[k])
/ locals->VRatio[k] - locals->EffectiveDETLBLinesLuma * locals->SwathWidthYPerState[i][j][k] *
diff --git a/drivers/gpu/drm/amd/display/dc/dml/dcn21/display_mode_vba_21.c b/drivers/gpu/drm/amd/display/dc/dml/dcn21/display_mode_vba_21.c
index 8a7485e21d53..d40d32e380f4 100644
--- a/drivers/gpu/drm/amd/display/dc/dml/dcn21/display_mode_vba_21.c
+++ b/drivers/gpu/drm/amd/display/dc/dml/dcn21/display_mode_vba_21.c
@@ -806,10 +806,12 @@ static bool CalculatePrefetchSchedule(
if (myPipe->SourceScan == dm_horz) {
*swath_width_luma_ub = dml_ceil(SwathWidthY - 1, myPipe->BlockWidth256BytesY) + myPipe->BlockWidth256BytesY;
- *swath_width_chroma_ub = dml_ceil(SwathWidthY / 2 - 1, myPipe->BlockWidth256BytesC) + myPipe->BlockWidth256BytesC;
+ if (myPipe->BlockWidth256BytesC > 0)
+ *swath_width_chroma_ub = dml_ceil(SwathWidthY / 2 - 1, myPipe->BlockWidth256BytesC) + myPipe->BlockWidth256BytesC;
} else {
*swath_width_luma_ub = dml_ceil(SwathWidthY - 1, myPipe->BlockHeight256BytesY) + myPipe->BlockHeight256BytesY;
- *swath_width_chroma_ub = dml_ceil(SwathWidthY / 2 - 1, myPipe->BlockHeight256BytesC) + myPipe->BlockHeight256BytesC;
+ if (myPipe->BlockWidth256BytesC > 0)
+ *swath_width_chroma_ub = dml_ceil(SwathWidthY / 2 - 1, myPipe->BlockHeight256BytesC) + myPipe->BlockHeight256BytesC;
}
prefetch_bw_oto = (PrefetchSourceLinesY * *swath_width_luma_ub * dml_ceil(BytePerPixelDETY, 1) + PrefetchSourceLinesC * *swath_width_chroma_ub * dml_ceil(BytePerPixelDETC, 2)) / Tsw_oto;
diff --git a/drivers/gpu/drm/amd/display/dc/dml/dcn30/display_mode_vba_30.c b/drivers/gpu/drm/amd/display/dc/dml/dcn30/display_mode_vba_30.c
index b7fa003ffe06..c117a9724ae1 100644
--- a/drivers/gpu/drm/amd/display/dc/dml/dcn30/display_mode_vba_30.c
+++ b/drivers/gpu/drm/amd/display/dc/dml/dcn30/display_mode_vba_30.c
@@ -6322,10 +6322,6 @@ static void CalculateSwathWidth(
for (k = 0; k < NumberOfActivePlanes; ++k) {
enum odm_combine_mode MainPlaneODMCombine = 0;
- surface_width_ub_l = dml_ceil(SurfaceWidthY[k], Read256BytesBlockWidthY[k]);
- surface_height_ub_l = dml_ceil(SurfaceHeightY[k], Read256BytesBlockHeightY[k]);
- surface_width_ub_c = dml_ceil(SurfaceWidthC[k], Read256BytesBlockWidthC[k]);
- surface_height_ub_c = dml_ceil(SurfaceHeightC[k], Read256BytesBlockHeightC[k]);
if (SourceScan[k] != dm_vert) {
SwathWidthSingleDPPY[k] = ViewportWidth[k];
@@ -6365,8 +6361,6 @@ static void CalculateSwathWidth(
surface_width_ub_l = dml_ceil(SurfaceWidthY[k], Read256BytesBlockWidthY[k]);
surface_height_ub_l = dml_ceil(SurfaceHeightY[k], Read256BytesBlockHeightY[k]);
- surface_width_ub_c = dml_ceil(SurfaceWidthC[k], Read256BytesBlockWidthC[k]);
- surface_height_ub_c = dml_ceil(SurfaceHeightC[k], Read256BytesBlockHeightC[k]);
if (SourceScan[k] != dm_vert) {
MaximumSwathHeightY[k] = Read256BytesBlockHeightY[k];
@@ -6374,6 +6368,7 @@ static void CalculateSwathWidth(
swath_width_luma_ub[k] = dml_min(surface_width_ub_l, (long) dml_ceil(SwathWidthY[k] - 1,
Read256BytesBlockWidthY[k]) + Read256BytesBlockWidthY[k]);
if (BytePerPixC[k] > 0) {
+ surface_width_ub_c = dml_ceil(SurfaceWidthC[k], Read256BytesBlockWidthC[k]);
swath_width_chroma_ub[k] = dml_min(surface_width_ub_c, (long) dml_ceil(SwathWidthC[k] - 1,
Read256BytesBlockWidthC[k]) + Read256BytesBlockWidthC[k]);
} else {
@@ -6385,6 +6380,7 @@ static void CalculateSwathWidth(
swath_width_luma_ub[k] = dml_min(surface_height_ub_l, (long) dml_ceil(SwathWidthY[k] - 1,
Read256BytesBlockHeightY[k]) + Read256BytesBlockHeightY[k]);
if (BytePerPixC[k] > 0) {
+ surface_height_ub_c = dml_ceil(SurfaceHeightC[k], Read256BytesBlockHeightC[k]);
swath_width_chroma_ub[k] = dml_min(surface_height_ub_c, (long) dml_ceil(SwathWidthC[k] - 1,
Read256BytesBlockHeightC[k]) + Read256BytesBlockHeightC[k]);
} else {
diff --git a/drivers/gpu/drm/amd/display/dc/dml/dcn31/display_mode_vba_31.c b/drivers/gpu/drm/amd/display/dc/dml/dcn31/display_mode_vba_31.c
index d63b4209b14c..8753f94bdd79 100644
--- a/drivers/gpu/drm/amd/display/dc/dml/dcn31/display_mode_vba_31.c
+++ b/drivers/gpu/drm/amd/display/dc/dml/dcn31/display_mode_vba_31.c
@@ -6933,8 +6933,6 @@ static void CalculateSwathWidth(
{
int surface_width_ub_l = dml_ceil(SurfaceWidthY[k], Read256BytesBlockWidthY[k]);
int surface_height_ub_l = dml_ceil(SurfaceHeightY[k], Read256BytesBlockHeightY[k]);
- int surface_width_ub_c = dml_ceil(SurfaceWidthC[k], Read256BytesBlockWidthC[k]);
- int surface_height_ub_c = dml_ceil(SurfaceHeightC[k], Read256BytesBlockHeightC[k]);
#ifdef __DML_VBA_DEBUG__
dml_print("DML::%s: k=%d surface_width_ub_l=%0d\n", __func__, k, surface_width_ub_l);
@@ -6945,6 +6943,8 @@ static void CalculateSwathWidth(
MaximumSwathHeightC[k] = Read256BytesBlockHeightC[k];
swath_width_luma_ub[k] = dml_min(surface_width_ub_l, (int) dml_ceil(SwathWidthY[k] - 1, Read256BytesBlockWidthY[k]) + Read256BytesBlockWidthY[k]);
if (BytePerPixC[k] > 0) {
+ int surface_width_ub_c = dml_ceil(SurfaceWidthC[k], Read256BytesBlockWidthC[k]);
+
swath_width_chroma_ub[k] = dml_min(
surface_width_ub_c,
(int) dml_ceil(SwathWidthC[k] - 1, Read256BytesBlockWidthC[k]) + Read256BytesBlockWidthC[k]);
@@ -6956,6 +6956,8 @@ static void CalculateSwathWidth(
MaximumSwathHeightC[k] = Read256BytesBlockWidthC[k];
swath_width_luma_ub[k] = dml_min(surface_height_ub_l, (int) dml_ceil(SwathWidthY[k] - 1, Read256BytesBlockHeightY[k]) + Read256BytesBlockHeightY[k]);
if (BytePerPixC[k] > 0) {
+ int surface_height_ub_c = dml_ceil(SurfaceHeightC[k], Read256BytesBlockHeightC[k]);
+
swath_width_chroma_ub[k] = dml_min(
surface_height_ub_c,
(int) dml_ceil(SwathWidthC[k] - 1, Read256BytesBlockHeightC[k]) + Read256BytesBlockHeightC[k]);
diff --git a/drivers/gpu/drm/amd/display/dc/dml/dcn314/display_mode_vba_314.c b/drivers/gpu/drm/amd/display/dc/dml/dcn314/display_mode_vba_314.c
index fc4d7474c111..503d9ede0ac1 100644
--- a/drivers/gpu/drm/amd/display/dc/dml/dcn314/display_mode_vba_314.c
+++ b/drivers/gpu/drm/amd/display/dc/dml/dcn314/display_mode_vba_314.c
@@ -7049,8 +7049,6 @@ static void CalculateSwathWidth(
{
int surface_width_ub_l = dml_ceil(SurfaceWidthY[k], Read256BytesBlockWidthY[k]);
int surface_height_ub_l = dml_ceil(SurfaceHeightY[k], Read256BytesBlockHeightY[k]);
- int surface_width_ub_c = dml_ceil(SurfaceWidthC[k], Read256BytesBlockWidthC[k]);
- int surface_height_ub_c = dml_ceil(SurfaceHeightC[k], Read256BytesBlockHeightC[k]);
#ifdef __DML_VBA_DEBUG__
dml_print("DML::%s: k=%d surface_width_ub_l=%0d\n", __func__, k, surface_width_ub_l);
@@ -7061,6 +7059,8 @@ static void CalculateSwathWidth(
MaximumSwathHeightC[k] = Read256BytesBlockHeightC[k];
swath_width_luma_ub[k] = dml_min(surface_width_ub_l, (int) dml_ceil(SwathWidthY[k] - 1, Read256BytesBlockWidthY[k]) + Read256BytesBlockWidthY[k]);
if (BytePerPixC[k] > 0) {
+ int surface_width_ub_c = dml_ceil(SurfaceWidthC[k], Read256BytesBlockWidthC[k]);
+
swath_width_chroma_ub[k] = dml_min(
surface_width_ub_c,
(int) dml_ceil(SwathWidthC[k] - 1, Read256BytesBlockWidthC[k]) + Read256BytesBlockWidthC[k]);
@@ -7072,6 +7072,8 @@ static void CalculateSwathWidth(
MaximumSwathHeightC[k] = Read256BytesBlockWidthC[k];
swath_width_luma_ub[k] = dml_min(surface_height_ub_l, (int) dml_ceil(SwathWidthY[k] - 1, Read256BytesBlockHeightY[k]) + Read256BytesBlockHeightY[k]);
if (BytePerPixC[k] > 0) {
+ int surface_height_ub_c = dml_ceil(SurfaceHeightC[k], Read256BytesBlockHeightC[k]);
+
swath_width_chroma_ub[k] = dml_min(
surface_height_ub_c,
(int) dml_ceil(SwathWidthC[k] - 1, Read256BytesBlockHeightC[k]) + Read256BytesBlockHeightC[k]);
diff --git a/drivers/gpu/drm/amd/display/dc/dml/dcn32/display_mode_vba_util_32.c b/drivers/gpu/drm/amd/display/dc/dml/dcn32/display_mode_vba_util_32.c
index dc501ee7d01a..c385c54832cb 100644
--- a/drivers/gpu/drm/amd/display/dc/dml/dcn32/display_mode_vba_util_32.c
+++ b/drivers/gpu/drm/amd/display/dc/dml/dcn32/display_mode_vba_util_32.c
@@ -719,8 +719,8 @@ void dml32_CalculateSwathWidth(
unsigned int surface_width_ub_l;
unsigned int surface_height_ub_l;
- unsigned int surface_width_ub_c;
- unsigned int surface_height_ub_c;
+ unsigned int surface_width_ub_c = 0;
+ unsigned int surface_height_ub_c = 0;
#ifdef __DML_VBA_DEBUG__
dml_print("DML::%s: ForceSingleDPP = %d\n", __func__, ForceSingleDPP);
@@ -784,21 +784,6 @@ void dml32_CalculateSwathWidth(
surface_width_ub_l = dml_ceil(SurfaceWidthY[k], Read256BytesBlockWidthY[k]);
surface_height_ub_l = dml_ceil(SurfaceHeightY[k], Read256BytesBlockHeightY[k]);
- surface_width_ub_c = dml_ceil(SurfaceWidthC[k], Read256BytesBlockWidthC[k]);
- surface_height_ub_c = dml_ceil(SurfaceHeightC[k], Read256BytesBlockHeightC[k]);
-
-#ifdef __DML_VBA_DEBUG__
- dml_print("DML::%s: k=%d surface_width_ub_l=%0d\n", __func__, k, surface_width_ub_l);
- dml_print("DML::%s: k=%d surface_height_ub_l=%0d\n", __func__, k, surface_height_ub_l);
- dml_print("DML::%s: k=%d surface_width_ub_c=%0d\n", __func__, k, surface_width_ub_c);
- dml_print("DML::%s: k=%d surface_height_ub_c=%0d\n", __func__, k, surface_height_ub_c);
- dml_print("DML::%s: k=%d Read256BytesBlockWidthY=%0d\n", __func__, k, Read256BytesBlockWidthY[k]);
- dml_print("DML::%s: k=%d Read256BytesBlockHeightY=%0d\n", __func__, k, Read256BytesBlockHeightY[k]);
- dml_print("DML::%s: k=%d Read256BytesBlockWidthC=%0d\n", __func__, k, Read256BytesBlockWidthC[k]);
- dml_print("DML::%s: k=%d Read256BytesBlockHeightC=%0d\n", __func__, k, Read256BytesBlockHeightC[k]);
- dml_print("DML::%s: k=%d ViewportStationary=%0d\n", __func__, k, ViewportStationary[k]);
- dml_print("DML::%s: k=%d DPPPerSurface=%0d\n", __func__, k, DPPPerSurface[k]);
-#endif
if (!IsVertical(SourceRotation[k])) {
MaximumSwathHeightY[k] = Read256BytesBlockHeightY[k];
@@ -818,6 +803,7 @@ void dml32_CalculateSwathWidth(
Read256BytesBlockWidthY[k]);
}
if (BytePerPixC[k] > 0) {
+ surface_width_ub_c = dml_ceil(SurfaceWidthC[k], Read256BytesBlockWidthC[k]);
if (ViewportStationary[k] && DPPPerSurface[k] == 1) {
swath_width_chroma_ub[k] = dml_min(surface_width_ub_c,
dml_floor(ViewportXStartC[k] + SwathWidthC[k] +
@@ -848,6 +834,7 @@ void dml32_CalculateSwathWidth(
Read256BytesBlockHeightY[k]) + Read256BytesBlockHeightY[k]);
}
if (BytePerPixC[k] > 0) {
+ surface_height_ub_c = dml_ceil(SurfaceHeightC[k], Read256BytesBlockHeightC[k]);
if (ViewportStationary[k] && DPPPerSurface[k] == 1) {
swath_width_chroma_ub[k] = dml_min(surface_height_ub_c,
dml_floor(ViewportYStartC[k] + SwathWidthC[k] +
@@ -866,6 +853,16 @@ void dml32_CalculateSwathWidth(
}
#ifdef __DML_VBA_DEBUG__
+ dml_print("DML::%s: k=%d surface_width_ub_l=%0d\n", __func__, k, surface_width_ub_l);
+ dml_print("DML::%s: k=%d surface_height_ub_l=%0d\n", __func__, k, surface_height_ub_l);
+ dml_print("DML::%s: k=%d surface_width_ub_c=%0d\n", __func__, k, surface_width_ub_c);
+ dml_print("DML::%s: k=%d surface_height_ub_c=%0d\n", __func__, k, surface_height_ub_c);
+ dml_print("DML::%s: k=%d Read256BytesBlockWidthY=%0d\n", __func__, k, Read256BytesBlockWidthY[k]);
+ dml_print("DML::%s: k=%d Read256BytesBlockHeightY=%0d\n", __func__, k, Read256BytesBlockHeightY[k]);
+ dml_print("DML::%s: k=%d Read256BytesBlockWidthC=%0d\n", __func__, k, Read256BytesBlockWidthC[k]);
+ dml_print("DML::%s: k=%d Read256BytesBlockHeightC=%0d\n", __func__, k, Read256BytesBlockHeightC[k]);
+ dml_print("DML::%s: k=%d ViewportStationary=%0d\n", __func__, k, ViewportStationary[k]);
+ dml_print("DML::%s: k=%d DPPPerSurface=%0d\n", __func__, k, DPPPerSurface[k]);
dml_print("DML::%s: k=%d swath_width_luma_ub=%0d\n", __func__, k, swath_width_luma_ub[k]);
dml_print("DML::%s: k=%d swath_width_chroma_ub=%0d\n", __func__, k, swath_width_chroma_ub[k]);
dml_print("DML::%s: k=%d MaximumSwathHeightY=%0d\n", __func__, k, MaximumSwathHeightY[k]);
diff --git a/drivers/gpu/drm/amd/display/dc/dml/dcn32/display_mode_vba_util_32.c.rej b/drivers/gpu/drm/amd/display/dc/dml/dcn32/display_mode_vba_util_32.c.rej
new file mode 100644
index 000000000000..dff2badbf820
--- /dev/null
+++ b/drivers/gpu/drm/amd/display/dc/dml/dcn32/display_mode_vba_util_32.c.rej
@@ -0,0 +1,12 @@
+diff a/drivers/gpu/drm/amd/display/dc/dml/dcn32/display_mode_vba_util_32.c b/drivers/gpu/drm/amd/display/dc/dml/dcn32/display_mode_vba_util_32.c (rejected hunks)
+@@ -727,8 +727,8 @@ void dml32_CalculateSwathWidth(
+ enum odm_combine_mode MainSurfaceODMMode;
+ unsigned int surface_width_ub_l;
+ unsigned int surface_height_ub_l;
+- unsigned int surface_width_ub_c;
+- unsigned int surface_height_ub_c;
++ unsigned int surface_width_ub_c = 0;
++ unsigned int surface_height_ub_c = 0;
+ unsigned int k, j;
+
+ #ifdef __DML_VBA_DEBUG__
diff --git a/drivers/gpu/drm/amd/display/dc/dml/dml_inline_defs.h b/drivers/gpu/drm/amd/display/dc/dml/dml_inline_defs.h
index 479d7d83220c..072bd0539605 100644
--- a/drivers/gpu/drm/amd/display/dc/dml/dml_inline_defs.h
+++ b/drivers/gpu/drm/amd/display/dc/dml/dml_inline_defs.h
@@ -76,14 +76,9 @@ static inline double dml_floor(double a, double granularity)
static inline double dml_round(double a)
{
- double round_pt = 0.5;
- double ceil = dml_ceil(a, 1);
- double floor = dml_floor(a, 1);
+ const double round_pt = 0.5;
- if (a - floor >= round_pt)
- return ceil;
- else
- return floor;
+ return dml_floor(a + round_pt, 1);
}
/* float
--
2.34.1
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/amd-gfx/attachments/20220912/be08d8de/attachment-0001.htm>
More information about the amd-gfx
mailing list