From 17529ea2acfa3e2118f5a9ee911e0daf2d88c13f Mon Sep 17 00:00:00 2001 From: Aric Cyr Date: Sat, 20 Aug 2022 17:20:30 -0400 Subject: 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 Reviewed-by: Nevenko Stupar Acked-by: Pavle Kotarac Signed-off-by: Aric Cyr Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/display/dc/dml/dcn21/display_mode_vba_21.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) (limited to 'drivers/gpu/drm/amd/display/dc/dml/dcn21') 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; -- cgit v1.2.3