drm/amd/display: Optimizations for DML math

[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>

1 files changed, 14 insertions, 17 deletions

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 8f1c0e12dd86..36e4a4a9296b 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]);