openwrt/target/linux/bcm27xx/patches-5.15/950-0393-media-rpivid-Remove-the-need-to-have-num_entry_point.patch

From 4b33f988d9da9776bcfe218df4ab9865f4b771a8 Mon Sep 17 00:00:00 2001
From: John Cox <jc@kynesim.co.uk>
Date: Thu, 21 May 2020 11:49:37 +0100
Subject: [PATCH] media: rpivid: Remove the need to have
 num_entry_points set

VAAPI H265 has num entry points but never sets it. Allow a VAAPI
shim to work without requiring rewriting the VAAPI driver.
num_entry_points can be calculated from the slice_segment_addr
of the next slice so delay processing until we have that.

Also includes some minor cosmetics.

Signed-off-by: John Cox <jc@kynesim.co.uk>
---
 drivers/staging/media/rpivid/rpivid_h265.c | 699 +++++++++++----------
 1 file changed, 365 insertions(+), 334 deletions(-)

--- a/drivers/staging/media/rpivid/rpivid_h265.c
+++ b/drivers/staging/media/rpivid/rpivid_h265.c
@@ -202,8 +202,17 @@ struct rpivid_dec_env {
 	unsigned int dpbno_col;
 	u32 reg_slicestart;
 	int collocated_from_l0_flag;
-	unsigned int wpp_entry_x;
-	unsigned int wpp_entry_y;
+	/*
+	 * Last CTB/Tile X,Y processed by (wpp_)entry_point
+	 * Could be in _state as P0 only but needs updating where _state
+	 * is const
+	 */
+	unsigned int entry_ctb_x;
+	unsigned int entry_ctb_y;
+	unsigned int entry_tile_x;
+	unsigned int entry_tile_y;
+	unsigned int entry_qp;
+	u32 entry_slice;
 
 	u32 rpi_config2;
 	u32 rpi_framesize;
@@ -239,22 +248,17 @@ struct rpivid_dec_state {
 	struct v4l2_ctrl_hevc_pps pps;
 
 	// Helper vars & tables derived from sps/pps
-	unsigned int log2_ctb_size; /* log2 width of a CTB */
-	unsigned int ctb_width; /* Width in CTBs */
-	unsigned int ctb_height; /* Height in CTBs */
-	unsigned int ctb_size; /* Pic area in CTBs */
-	unsigned int num_tile_columns;
-	unsigned int num_tile_rows;
-	u8 column_width[member_size(struct v4l2_ctrl_hevc_pps,
-				    column_width_minus1)];
-	u8 row_height[member_size(struct v4l2_ctrl_hevc_pps,
-				  row_height_minus1)];
+	unsigned int log2_ctb_size;     /* log2 width of a CTB */
+	unsigned int ctb_width;         /* Width in CTBs */
+	unsigned int ctb_height;        /* Height in CTBs */
+	unsigned int ctb_size;          /* Pic area in CTBs */
+	unsigned int tile_width;        /* Width in tiles */
+	unsigned int tile_height;       /* Height in tiles */
 
 	int *col_bd;
 	int *row_bd;
 	int *ctb_addr_rs_to_ts;
 	int *ctb_addr_ts_to_rs;
-	int *tile_id;
 
 	// Aux starage for DPB
 	// Hold refs
@@ -274,6 +278,12 @@ struct rpivid_dec_state {
 	unsigned int slice_qp;
 	unsigned int max_num_merge_cand; // 0 if I-slice
 	bool dependent_slice_segment_flag;
+
+	unsigned int start_ts;          /* slice_segment_addr -> ts */
+	unsigned int start_ctb_x;       /* CTB X,Y of start_ts */
+	unsigned int start_ctb_y;
+	unsigned int prev_ctb_x;        /* CTB X,Y of start_ts - 1 */
+	unsigned int prev_ctb_y;
 };
 
 static inline int clip_int(const int x, const int lo, const int hi)
@@ -319,15 +329,16 @@ static int ctb_to_tile(unsigned int ctb,
 	return i - 1;
 }
 
-static int ctb_to_slice_w_h(unsigned int ctb, int ctb_size, int width,
-			    unsigned int *bd, int num)
+static unsigned int ctb_to_tile_x(const struct rpivid_dec_state *const s,
+				  const unsigned int ctb_x)
 {
-	if (ctb < bd[num - 1])
-		return ctb_size;
-	else if (width % ctb_size)
-		return width % ctb_size;
-	else
-		return ctb_size;
+	return ctb_to_tile(ctb_x, s->col_bd, s->tile_width);
+}
+
+static unsigned int ctb_to_tile_y(const struct rpivid_dec_state *const s,
+				  const unsigned int ctb_y)
+{
+	return ctb_to_tile(ctb_y, s->row_bd, s->tile_height);
 }
 
 static void aux_q_free(struct rpivid_ctx *const ctx,
@@ -532,6 +543,15 @@ static void write_prob(struct rpivid_dec
 		p1_apb_write(de, 0x1000 + i,
 			     dst[i] + (dst[i + 1] << 8) + (dst[i + 2] << 16) +
 				     (dst[i + 3] << 24));
+
+	/*
+	 * Having written the prob array back it up
+	 * This is not always needed but is a small overhead that simplifies
+	 * (and speeds up) some multi-tile & WPP scenarios
+	 * There are no scenarios where having written a prob we ever want
+	 * a previous (non-initial) state back
+	 */
+	p1_apb_write(de, RPI_TRANSFER, PROB_BACKUP);
 }
 
 static void write_scaling_factors(struct rpivid_dec_env *const de)
@@ -552,8 +572,8 @@ static inline __u32 dma_to_axi_addr(dma_
 static void write_bitstream(struct rpivid_dec_env *const de,
 			    const struct rpivid_dec_state *const s)
 {
-	// Note that FFmpeg removes emulation prevention bytes, so this is
-	// matched in the configuration here.
+	// Note that FFmpeg V4L2 does not remove emulation prevention bytes,
+	// so this is matched in the configuration here.
 	// Whether that is the correct behaviour or not is not clear in the
 	// spec.
 	const int rpi_use_emu = 1;
@@ -579,78 +599,26 @@ static void write_bitstream(struct rpivi
 
 //////////////////////////////////////////////////////////////////////////////
 
-static void write_slice(struct rpivid_dec_env *const de,
-			const struct rpivid_dec_state *const s,
-			const unsigned int slice_w,
-			const unsigned int slice_h)
-{
-	u32 u32 = (s->sh->slice_type << 12) +
-		  (((s->sh->flags &
-		     V4L2_HEVC_SLICE_PARAMS_FLAG_SLICE_SAO_LUMA) != 0)
-		   << 14) +
-		  (((s->sh->flags &
-		     V4L2_HEVC_SLICE_PARAMS_FLAG_SLICE_SAO_CHROMA) != 0)
-		   << 15) +
-		  (slice_w << 17) + (slice_h << 24);
-
-	u32 |= (s->max_num_merge_cand << 0) + (s->nb_refs[L0] << 4) +
-	       (s->nb_refs[L1] << 8);
-
-	if (s->sh->slice_type == HEVC_SLICE_B)
-		u32 |= ((s->sh->flags &
-			 V4L2_HEVC_SLICE_PARAMS_FLAG_MVD_L1_ZERO) != 0)
-		       << 16;
-	p1_apb_write(de, RPI_SLICE, u32);
-}
-
-//////////////////////////////////////////////////////////////////////////////
-// Tiles mode
-
-static void new_entry_point(struct rpivid_dec_env *const de,
-			    const struct rpivid_dec_state *const s,
-			    const int do_bte,
-			    const int reset_qp_y, const int ctb_addr_ts)
+/*
+ * The slice constant part of the slice register - width and height need to
+ * be ORed in later as they are per-tile / WPP-row
+ */
+static u32 slice_reg_const(const struct rpivid_dec_state *const s)
 {
-	int ctb_col = s->ctb_addr_ts_to_rs[ctb_addr_ts] %
-							de->pic_width_in_ctbs_y;
-	int ctb_row = s->ctb_addr_ts_to_rs[ctb_addr_ts] /
-							de->pic_width_in_ctbs_y;
-
-	int tile_x = ctb_to_tile(ctb_col, s->col_bd, s->num_tile_columns);
-	int tile_y = ctb_to_tile(ctb_row, s->row_bd, s->num_tile_rows);
-
-	int endx = s->col_bd[tile_x + 1] - 1;
-	int endy = s->row_bd[tile_y + 1] - 1;
-
-	u8 slice_w = ctb_to_slice_w_h(ctb_col, 1 << s->log2_ctb_size,
-				      s->sps.pic_width_in_luma_samples,
-				      s->col_bd, s->num_tile_columns);
-	u8 slice_h = ctb_to_slice_w_h(ctb_row, 1 << s->log2_ctb_size,
-				      s->sps.pic_height_in_luma_samples,
-				      s->row_bd, s->num_tile_rows);
-
-	p1_apb_write(de, RPI_TILESTART,
-		     s->col_bd[tile_x] + (s->row_bd[tile_y] << 16));
-	p1_apb_write(de, RPI_TILEEND, endx + (endy << 16));
-
-	if (do_bte)
-		p1_apb_write(de, RPI_BEGINTILEEND, endx + (endy << 16));
+	u32 x = (s->max_num_merge_cand << 0) |
+		(s->nb_refs[L0] << 4) |
+		(s->nb_refs[L1] << 8) |
+		(s->sh->slice_type << 12);
+
+	if (s->sh->flags & V4L2_HEVC_SLICE_PARAMS_FLAG_SLICE_SAO_LUMA)
+		x |= BIT(14);
+	if (s->sh->flags & V4L2_HEVC_SLICE_PARAMS_FLAG_SLICE_SAO_CHROMA)
+		x |= BIT(15);
+	if (s->sh->slice_type == HEVC_SLICE_B &&
+	    (s->sh->flags & V4L2_HEVC_SLICE_PARAMS_FLAG_MVD_L1_ZERO))
+		x |= BIT(16);
 
-	write_slice(de, s, slice_w, slice_h);
-
-	if (reset_qp_y) {
-		unsigned int sps_qp_bd_offset =
-			6 * s->sps.bit_depth_luma_minus8;
-
-		p1_apb_write(de, RPI_QP, sps_qp_bd_offset + s->slice_qp);
-	}
-
-	p1_apb_write(de, RPI_MODE,
-		     (0xFFFF << 0) + (0x0 << 16) +
-			     ((tile_x == s->num_tile_columns - 1) << 17) +
-			     ((tile_y == s->num_tile_rows - 1) << 18));
-
-	p1_apb_write(de, RPI_CONTROL, (ctb_col << 0) + (ctb_row << 16));
+	return x;
 }
 
 //////////////////////////////////////////////////////////////////////////////
@@ -934,197 +902,256 @@ static void pre_slice_decode(struct rpiv
 		       (sh->slice_cb_qp_offset & 31)); // CMD_QPOFF
 }
 
-//////////////////////////////////////////////////////////////////////////////
-// Write STATUS register with expected end CTU address of previous slice
-
-static void end_previous_slice(struct rpivid_dec_env *const de,
-			       const struct rpivid_dec_state *const s,
-			       const int ctb_addr_ts)
-{
-	int last_x =
-		s->ctb_addr_ts_to_rs[ctb_addr_ts - 1] % de->pic_width_in_ctbs_y;
-	int last_y =
-		s->ctb_addr_ts_to_rs[ctb_addr_ts - 1] / de->pic_width_in_ctbs_y;
-
-	p1_apb_write(de, RPI_STATUS, 1 + (last_x << 5) + (last_y << 18));
-}
-
-static void wpp_pause(struct rpivid_dec_env *const de, int ctb_row)
-{
-	p1_apb_write(de, RPI_STATUS, (ctb_row << 18) + 0x25);
-	p1_apb_write(de, RPI_TRANSFER, PROB_BACKUP);
-	p1_apb_write(de, RPI_MODE,
-		     ctb_row == de->pic_height_in_ctbs_y - 1 ?
-							0x70000 : 0x30000);
-	p1_apb_write(de, RPI_CONTROL, (ctb_row << 16) + 2);
-}
-
-static void wpp_end_previous_slice(struct rpivid_dec_env *const de,
-				   const struct rpivid_dec_state *const s,
-				   int ctb_addr_ts)
-{
-	int new_x = s->sh->slice_segment_addr % de->pic_width_in_ctbs_y;
-	int new_y = s->sh->slice_segment_addr / de->pic_width_in_ctbs_y;
-	int last_x =
-		s->ctb_addr_ts_to_rs[ctb_addr_ts - 1] % de->pic_width_in_ctbs_y;
-	int last_y =
-		s->ctb_addr_ts_to_rs[ctb_addr_ts - 1] / de->pic_width_in_ctbs_y;
-
-	if (de->wpp_entry_x < 2 && (de->wpp_entry_y < new_y || new_x > 2) &&
-	    de->pic_width_in_ctbs_y > 2)
-		wpp_pause(de, last_y);
-	p1_apb_write(de, RPI_STATUS, 1 + (last_x << 5) + (last_y << 18));
-	if (new_x == 2 || (de->pic_width_in_ctbs_y == 2 &&
-			   de->wpp_entry_y < new_y))
-		p1_apb_write(de, RPI_TRANSFER, PROB_BACKUP);
+static void write_slice(struct rpivid_dec_env *const de,
+			const struct rpivid_dec_state *const s,
+			const u32 slice_const,
+			const unsigned int ctb_col,
+			const unsigned int ctb_row)
+{
+	const unsigned int cs = (1 << s->log2_ctb_size);
+	const unsigned int w_last = s->sps.pic_width_in_luma_samples & (cs - 1);
+	const unsigned int h_last = s->sps.pic_height_in_luma_samples & (cs - 1);
+
+	p1_apb_write(de, RPI_SLICE,
+		     slice_const |
+		     ((ctb_col + 1 < s->ctb_width || !w_last ?
+				cs : w_last) << 17) |
+		     ((ctb_row + 1 < s->ctb_height || !h_last ?
+				cs : h_last) << 24));
 }
 
-//////////////////////////////////////////////////////////////////////////////
-// Wavefront mode
+#define PAUSE_MODE_WPP  1
+#define PAUSE_MODE_TILE 0xffff
 
-static void wpp_entry_point(struct rpivid_dec_env *const de,
+/*
+ * N.B. This can be called to fill in data from the previous slice so must not
+ * use any state data that may change from slice to slice (e.g. qp)
+ */
+static void new_entry_point(struct rpivid_dec_env *const de,
 			    const struct rpivid_dec_state *const s,
-			    const int do_bte,
-			    const int reset_qp_y, const int ctb_addr_ts)
-{
-	int ctb_size = 1 << s->log2_ctb_size;
-	int ctb_addr_rs = s->ctb_addr_ts_to_rs[ctb_addr_ts];
-
-	int ctb_col = de->wpp_entry_x = ctb_addr_rs % de->pic_width_in_ctbs_y;
-	int ctb_row = de->wpp_entry_y = ctb_addr_rs / de->pic_width_in_ctbs_y;
+			    const bool do_bte,
+			    const bool reset_qp_y,
+			    const u32 pause_mode,
+			    const unsigned int tile_x,
+			    const unsigned int tile_y,
+			    const unsigned int ctb_col,
+			    const unsigned int ctb_row,
+			    const unsigned int slice_qp,
+			    const u32 slice_const)
+{
+	const unsigned int endx = s->col_bd[tile_x + 1] - 1;
+	const unsigned int endy = (pause_mode == PAUSE_MODE_WPP) ?
+		ctb_row : s->row_bd[tile_y + 1] - 1;
 
-	int endx = de->pic_width_in_ctbs_y - 1;
-	int endy = ctb_row;
-
-	u8 slice_w = ctb_to_slice_w_h(ctb_col, ctb_size,
-				      s->sps.pic_width_in_luma_samples,
-				      s->col_bd, s->num_tile_columns);
-	u8 slice_h = ctb_to_slice_w_h(ctb_row, ctb_size,
-				      s->sps.pic_height_in_luma_samples,
-				      s->row_bd, s->num_tile_rows);
-
-	p1_apb_write(de, RPI_TILESTART, 0);
-	p1_apb_write(de, RPI_TILEEND, endx + (endy << 16));
+	p1_apb_write(de, RPI_TILESTART,
+		     s->col_bd[tile_x] | (s->row_bd[tile_y] << 16));
+	p1_apb_write(de, RPI_TILEEND, endx | (endy << 16));
 
 	if (do_bte)
-		p1_apb_write(de, RPI_BEGINTILEEND, endx + (endy << 16));
+		p1_apb_write(de, RPI_BEGINTILEEND, endx | (endy << 16));
 
-	write_slice(de, s, slice_w,
-		    ctb_row == de->pic_height_in_ctbs_y - 1 ?
-							slice_h : ctb_size);
+	write_slice(de, s, slice_const, endx, endy);
 
 	if (reset_qp_y) {
 		unsigned int sps_qp_bd_offset =
 			6 * s->sps.bit_depth_luma_minus8;
 
-		p1_apb_write(de, RPI_QP, sps_qp_bd_offset + s->slice_qp);
+		p1_apb_write(de, RPI_QP, sps_qp_bd_offset + slice_qp);
 	}
 
 	p1_apb_write(de, RPI_MODE,
-		     ctb_row == de->pic_height_in_ctbs_y - 1 ?
-							0x60001 : 0x20001);
-	p1_apb_write(de, RPI_CONTROL, (ctb_col << 0) + (ctb_row << 16));
+		     pause_mode |
+			((endx == s->ctb_width - 1) << 17) |
+			((endy == s->ctb_height - 1) << 18));
+
+	p1_apb_write(de, RPI_CONTROL, (ctb_col << 0) | (ctb_row << 16));
+
+	de->entry_tile_x = tile_x;
+	de->entry_tile_y = tile_y;
+	de->entry_ctb_x = ctb_col;
+	de->entry_ctb_y = ctb_row;
+	de->entry_qp = slice_qp;
+	de->entry_slice = slice_const;
 }
 
 //////////////////////////////////////////////////////////////////////////////
 // Wavefront mode
 
+static void wpp_pause(struct rpivid_dec_env *const de, int ctb_row)
+{
+	p1_apb_write(de, RPI_STATUS, (ctb_row << 18) | 0x25);
+	p1_apb_write(de, RPI_TRANSFER, PROB_BACKUP);
+	p1_apb_write(de, RPI_MODE,
+		     ctb_row == de->pic_height_in_ctbs_y - 1 ?
+							0x70000 : 0x30000);
+	p1_apb_write(de, RPI_CONTROL, (ctb_row << 16) + 2);
+}
+
+static void wpp_entry_fill(struct rpivid_dec_env *const de,
+			   const struct rpivid_dec_state *const s,
+			   const unsigned int last_y)
+{
+	const unsigned int last_x = s->ctb_width - 1;
+
+	while (de->entry_ctb_y < last_y) {
+		/* wpp_entry_x/y set by wpp_entry_point */
+		if (s->ctb_width > 2)
+			wpp_pause(de, de->entry_ctb_y);
+		p1_apb_write(de, RPI_STATUS,
+			     (de->entry_ctb_y << 18) | (last_x << 5) | 2);
+
+		/* if width == 1 then the saved state is the init one */
+		if (s->ctb_width == 2)
+			p1_apb_write(de, RPI_TRANSFER, PROB_BACKUP);
+		else
+			p1_apb_write(de, RPI_TRANSFER, PROB_RELOAD);
+
+		new_entry_point(de, s, false, true, PAUSE_MODE_WPP,
+				0, 0, 0, de->entry_ctb_y + 1,
+				de->entry_qp, de->entry_slice);
+	}
+}
+
+static void wpp_end_previous_slice(struct rpivid_dec_env *const de,
+				   const struct rpivid_dec_state *const s)
+{
+	wpp_entry_fill(de, s, s->prev_ctb_y);
+
+	if (de->entry_ctb_x < 2 &&
+	    (de->entry_ctb_y < s->start_ctb_y || s->start_ctb_x > 2) &&
+	    s->ctb_width > 2)
+		wpp_pause(de, s->prev_ctb_y);
+	p1_apb_write(de, RPI_STATUS,
+		     1 | (s->prev_ctb_x << 5) | (s->prev_ctb_y << 18));
+	if (s->start_ctb_x == 2 ||
+	    (s->ctb_width == 2 && de->entry_ctb_y < s->start_ctb_y))
+		p1_apb_write(de, RPI_TRANSFER, PROB_BACKUP);
+}
+
+/* Only main profile supported so WPP => !Tiles which makes some of the
+ * next chunk code simpler
+ */
 static void wpp_decode_slice(struct rpivid_dec_env *const de,
-			     const struct rpivid_dec_state *const s,
-			     const struct v4l2_ctrl_hevc_slice_params *sh,
-			     int ctb_addr_ts)
-{
-	int i, reset_qp_y = 1;
-	int indep = !s->dependent_slice_segment_flag;
-	int ctb_col = s->sh->slice_segment_addr % de->pic_width_in_ctbs_y;
+			     const struct rpivid_dec_state *const s)
+{
+	bool reset_qp_y = true;
+	const bool indep = !s->dependent_slice_segment_flag;
 
-	if (ctb_addr_ts)
-		wpp_end_previous_slice(de, s, ctb_addr_ts);
+	if (s->start_ts)
+		wpp_end_previous_slice(de, s);
 	pre_slice_decode(de, s);
 	write_bitstream(de, s);
-	if (ctb_addr_ts == 0 || indep || de->pic_width_in_ctbs_y == 1)
+
+	if (!s->start_ts || indep || s->ctb_width == 1)
 		write_prob(de, s);
-	else if (ctb_col == 0)
+	else if (!s->start_ctb_x)
 		p1_apb_write(de, RPI_TRANSFER, PROB_RELOAD);
 	else
-		reset_qp_y = 0;
+		reset_qp_y = false;
+
 	program_slicecmds(de, s->slice_idx);
 	new_slice_segment(de, s);
-	wpp_entry_point(de, s, indep, reset_qp_y, ctb_addr_ts);
+	new_entry_point(de, s, indep, reset_qp_y, PAUSE_MODE_WPP,
+			0, 0, s->start_ctb_x, s->start_ctb_y,
+			s->slice_qp, slice_reg_const(s));
 
-	for (i = 0; i < s->sh->num_entry_point_offsets; i++) {
-		int ctb_addr_rs = s->ctb_addr_ts_to_rs[ctb_addr_ts];
-		int ctb_row = ctb_addr_rs / de->pic_width_in_ctbs_y;
-		int last_x = de->pic_width_in_ctbs_y - 1;
+	if (s->frame_end) {
+		wpp_entry_fill(de, s, s->ctb_height - 1);
+
+		if (de->entry_ctb_x < 2 && s->ctb_width > 2)
+			wpp_pause(de, s->ctb_height - 1);
 
-		if (de->pic_width_in_ctbs_y > 2)
-			wpp_pause(de, ctb_row);
 		p1_apb_write(de, RPI_STATUS,
-			     (ctb_row << 18) + (last_x << 5) + 2);
-		if (de->pic_width_in_ctbs_y == 2)
-			p1_apb_write(de, RPI_TRANSFER, PROB_BACKUP);
-		if (de->pic_width_in_ctbs_y == 1)
-			write_prob(de, s);
-		else
-			p1_apb_write(de, RPI_TRANSFER, PROB_RELOAD);
-		ctb_addr_ts += s->column_width[0];
-		wpp_entry_point(de, s, 0, 1, ctb_addr_ts);
+			     1 | ((s->ctb_width - 1) << 5) |
+				((s->ctb_height - 1) << 18));
 	}
+
 }
 
 //////////////////////////////////////////////////////////////////////////////
 // Tiles mode
 
+static void tile_entry_fill(struct rpivid_dec_env *const de,
+			    const struct rpivid_dec_state *const s,
+			    const unsigned int last_tile_x,
+			    const unsigned int last_tile_y)
+{
+	while (de->entry_tile_y < last_tile_y ||
+	       (de->entry_tile_y == last_tile_y &&
+		de->entry_tile_x < last_tile_x)) {
+		unsigned int t_x = de->entry_tile_x;
+		unsigned int t_y = de->entry_tile_y;
+		const unsigned int last_x = s->col_bd[t_x + 1] - 1;
+		const unsigned int last_y = s->row_bd[t_y + 1] - 1;
+
+		p1_apb_write(de, RPI_STATUS,
+			     2 | (last_x << 5) | (last_y << 18));
+		p1_apb_write(de, RPI_TRANSFER, PROB_RELOAD);
+
+		// Inc tile
+		if (++t_x >= s->tile_width) {
+			t_x = 0;
+			++t_y;
+		}
+
+		new_entry_point(de, s, false, true, PAUSE_MODE_TILE,
+				t_x, t_y, s->col_bd[t_x], s->row_bd[t_y],
+				de->entry_qp, de->entry_slice);
+	}
+}
+
+/*
+ * Write STATUS register with expected end CTU address of previous slice
+ */
+static void end_previous_slice(struct rpivid_dec_env *const de,
+			       const struct rpivid_dec_state *const s)
+{
+	tile_entry_fill(de, s,
+			ctb_to_tile_x(s, s->prev_ctb_x),
+			ctb_to_tile_y(s, s->prev_ctb_y));
+	p1_apb_write(de, RPI_STATUS,
+		     1 | (s->prev_ctb_x << 5) | (s->prev_ctb_y << 18));
+}
+
 static void decode_slice(struct rpivid_dec_env *const de,
-			 const struct rpivid_dec_state *const s,
-			 const struct v4l2_ctrl_hevc_slice_params *const sh,
-			 int ctb_addr_ts)
+			 const struct rpivid_dec_state *const s)
 {
-	int i, reset_qp_y;
+	bool reset_qp_y;
+	unsigned int tile_x = ctb_to_tile_x(s, s->start_ctb_x);
+	unsigned int tile_y = ctb_to_tile_y(s, s->start_ctb_y);
 
-	if (ctb_addr_ts)
-		end_previous_slice(de, s, ctb_addr_ts);
+	if (s->start_ts)
+		end_previous_slice(de, s);
 
 	pre_slice_decode(de, s);
 	write_bitstream(de, s);
 
-#if DEBUG_TRACE_P1_CMD
-	if (p1_z < 256) {
-		v4l2_info(&de->ctx->dev->v4l2_dev,
-			  "TS=%d, tile=%d/%d, dss=%d, flags=%#llx\n",
-			  ctb_addr_ts, s->tile_id[ctb_addr_ts],
-			  s->tile_id[ctb_addr_ts - 1],
-			  s->dependent_slice_segment_flag, sh->flags);
-	}
-#endif
-
-	reset_qp_y = ctb_addr_ts == 0 ||
-		   s->tile_id[ctb_addr_ts] != s->tile_id[ctb_addr_ts - 1] ||
-		   !s->dependent_slice_segment_flag;
+	reset_qp_y = !s->start_ts ||
+		!s->dependent_slice_segment_flag ||
+		tile_x != ctb_to_tile_x(s, s->prev_ctb_x) ||
+		tile_y != ctb_to_tile_y(s, s->prev_ctb_y);
 	if (reset_qp_y)
 		write_prob(de, s);
 
 	program_slicecmds(de, s->slice_idx);
 	new_slice_segment(de, s);
 	new_entry_point(de, s, !s->dependent_slice_segment_flag, reset_qp_y,
-			ctb_addr_ts);
-
-	for (i = 0; i < s->sh->num_entry_point_offsets; i++) {
-		int ctb_addr_rs = s->ctb_addr_ts_to_rs[ctb_addr_ts];
-		int ctb_col = ctb_addr_rs % de->pic_width_in_ctbs_y;
-		int ctb_row = ctb_addr_rs / de->pic_width_in_ctbs_y;
-		int tile_x = ctb_to_tile(ctb_col, s->col_bd,
-					 s->num_tile_columns - 1);
-		int tile_y =
-			ctb_to_tile(ctb_row, s->row_bd, s->num_tile_rows - 1);
-		int last_x = s->col_bd[tile_x + 1] - 1;
-		int last_y = s->row_bd[tile_y + 1] - 1;
+			PAUSE_MODE_TILE,
+			tile_x, tile_y, s->start_ctb_x, s->start_ctb_y,
+			s->slice_qp, slice_reg_const(s));
 
+	/*
+	 * If this is the last slice then fill in the other tile entries
+	 * now, otherwise this will be done at the start of the next slice
+	 * when it will be known where this slice finishes
+	 */
+	if (s->frame_end) {
+		tile_entry_fill(de, s,
+				s->tile_width - 1,
+				s->tile_height - 1);
 		p1_apb_write(de, RPI_STATUS,
-			     2 + (last_x << 5) + (last_y << 18));
-		write_prob(de, s);
-		ctb_addr_ts += s->column_width[tile_x] * s->row_height[tile_y];
-		new_entry_point(de, s, 0, 1, ctb_addr_ts);
+			     1 | ((s->ctb_width - 1) << 5) |
+				((s->ctb_height - 1) << 18));
 	}
 }
 
@@ -1132,13 +1159,12 @@ static void decode_slice(struct rpivid_d
 // Scaling factors
 
 static void expand_scaling_list(const unsigned int size_id,
-				const unsigned int matrix_id, u8 *const dst0,
+				u8 *const dst0,
 				const u8 *const src0, uint8_t dc)
 {
 	u8 *d;
 	unsigned int x, y;
 
-	// FIXME: matrix_id is unused ?
 	switch (size_id) {
 	case 0:
 		memcpy(dst0, src0, 16);
@@ -1199,24 +1225,20 @@ static void populate_scaling_factors(con
 	unsigned int mid;
 
 	for (mid = 0; mid < 6; mid++)
-		expand_scaling_list(0, mid,
-				    de->scaling_factors +
+		expand_scaling_list(0, de->scaling_factors +
 					    scaling_factor_offsets[0][mid],
 				    sl->scaling_list_4x4[mid], 0);
 	for (mid = 0; mid < 6; mid++)
-		expand_scaling_list(1, mid,
-				    de->scaling_factors +
+		expand_scaling_list(1, de->scaling_factors +
 					    scaling_factor_offsets[1][mid],
 				    sl->scaling_list_8x8[mid], 0);
 	for (mid = 0; mid < 6; mid++)
-		expand_scaling_list(2, mid,
-				    de->scaling_factors +
+		expand_scaling_list(2, de->scaling_factors +
 					    scaling_factor_offsets[2][mid],
 				    sl->scaling_list_16x16[mid],
 				    sl->scaling_list_dc_coef_16x16[mid]);
-	for (mid = 0; mid < 2; mid += 1)
-		expand_scaling_list(3, mid,
-				    de->scaling_factors +
+	for (mid = 0; mid < 2; mid++)
+		expand_scaling_list(3, de->scaling_factors +
 					    scaling_factor_offsets[3][mid],
 				    sl->scaling_list_32x32[mid],
 				    sl->scaling_list_dc_coef_32x32[mid]);
@@ -1228,8 +1250,6 @@ static void free_ps_info(struct rpivid_d
 	s->ctb_addr_rs_to_ts = NULL;
 	kfree(s->ctb_addr_ts_to_rs);
 	s->ctb_addr_ts_to_rs = NULL;
-	kfree(s->tile_id);
-	s->tile_id = NULL;
 
 	kfree(s->col_bd);
 	s->col_bd = NULL;
@@ -1237,10 +1257,52 @@ static void free_ps_info(struct rpivid_d
 	s->row_bd = NULL;
 }
 
+static unsigned int tile_width(const struct rpivid_dec_state *const s,
+			       const unsigned int t_x)
+{
+	return s->col_bd[t_x + 1] - s->col_bd[t_x];
+}
+
+static unsigned int tile_height(const struct rpivid_dec_state *const s,
+				const unsigned int t_y)
+{
+	return s->row_bd[t_y + 1] - s->row_bd[t_y];
+}
+
+static void fill_rs_to_ts(struct rpivid_dec_state *const s)
+{
+	unsigned int ts = 0;
+	unsigned int t_y;
+	unsigned int tr_rs = 0;
+
+	for (t_y = 0; t_y != s->tile_height; ++t_y) {
+		const unsigned int t_h = tile_height(s, t_y);
+		unsigned int t_x;
+		unsigned int tc_rs = tr_rs;
+
+		for (t_x = 0; t_x != s->tile_width; ++t_x) {
+			const unsigned int t_w = tile_width(s, t_x);
+			unsigned int y;
+			unsigned int rs = tc_rs;
+
+			for (y = 0; y != t_h; ++y) {
+				unsigned int x;
+
+				for (x = 0; x != t_w; ++x) {
+					s->ctb_addr_rs_to_ts[rs + x] = ts;
+					s->ctb_addr_ts_to_rs[ts] = rs + x;
+					++ts;
+				}
+				rs += s->ctb_width;
+			}
+			tc_rs += t_w;
+		}
+		tr_rs += t_h * s->ctb_width;
+	}
+}
+
 static int updated_ps(struct rpivid_dec_state *const s)
 {
-	unsigned int ctb_addr_rs;
-	int j, x, y, tile_id;
 	unsigned int i;
 
 	free_ps_info(s);
@@ -1259,104 +1321,49 @@ static int updated_ps(struct rpivid_dec_
 
 	// Inferred parameters
 
+	s->ctb_addr_rs_to_ts = kmalloc_array(s->ctb_size,
+					     sizeof(*s->ctb_addr_rs_to_ts),
+					     GFP_KERNEL);
+	s->ctb_addr_ts_to_rs = kmalloc_array(s->ctb_size,
+					     sizeof(*s->ctb_addr_ts_to_rs),
+					     GFP_KERNEL);
+
 	if (!(s->pps.flags & V4L2_HEVC_PPS_FLAG_TILES_ENABLED)) {
-		s->num_tile_columns = 1;
-		s->num_tile_rows = 1;
-		s->column_width[0] = s->ctb_width;
-		s->row_height[0] = s->ctb_height;
+		s->tile_width = 1;
+		s->tile_height = 1;
 	} else {
-		s->num_tile_columns = s->pps.num_tile_columns_minus1 + 1;
-		s->num_tile_rows = s->pps.num_tile_rows_minus1 + 1;
-		for (i = 0; i < s->num_tile_columns; ++i)
-			s->column_width[i] = s->pps.column_width_minus1[i] + 1;
-		for (i = 0; i < s->num_tile_rows; ++i)
-			s->row_height[i] = s->pps.row_height_minus1[i] + 1;
+		s->tile_width = s->pps.num_tile_columns_minus1 + 1;
+		s->tile_height = s->pps.num_tile_rows_minus1 + 1;
 	}
 
-	s->col_bd = kmalloc((s->num_tile_columns + 1) * sizeof(*s->col_bd),
+	s->col_bd = kmalloc((s->tile_width + 1) * sizeof(*s->col_bd),
 			    GFP_KERNEL);
-	s->row_bd = kmalloc((s->num_tile_rows + 1) * sizeof(*s->row_bd),
+	s->row_bd = kmalloc((s->tile_height + 1) * sizeof(*s->row_bd),
 			    GFP_KERNEL);
 
 	s->col_bd[0] = 0;
-	for (i = 0; i < s->num_tile_columns; i++)
-		s->col_bd[i + 1] = s->col_bd[i] + s->column_width[i];
+	for (i = 1; i < s->tile_width; i++)
+		s->col_bd[i] = s->col_bd[i - 1] +
+			s->pps.column_width_minus1[i - 1] + 1;
+	s->col_bd[s->tile_width] = s->ctb_width;
 
 	s->row_bd[0] = 0;
-	for (i = 0; i < s->num_tile_rows; i++)
-		s->row_bd[i + 1] = s->row_bd[i] + s->row_height[i];
+	for (i = 1; i < s->tile_height; i++)
+		s->row_bd[i] = s->row_bd[i - 1] +
+			s->pps.row_height_minus1[i - 1] + 1;
+	s->row_bd[s->tile_height] = s->ctb_height;
 
-	s->ctb_addr_rs_to_ts = kmalloc_array(s->ctb_size,
-					     sizeof(*s->ctb_addr_rs_to_ts),
-					     GFP_KERNEL);
-	s->ctb_addr_ts_to_rs = kmalloc_array(s->ctb_size,
-					     sizeof(*s->ctb_addr_ts_to_rs),
-					     GFP_KERNEL);
-	s->tile_id = kmalloc_array(s->ctb_size, sizeof(*s->tile_id),
-				   GFP_KERNEL);
-
-	for (ctb_addr_rs = 0; ctb_addr_rs < s->ctb_size; ctb_addr_rs++) {
-		int tb_x = ctb_addr_rs % s->ctb_width;
-		int tb_y = ctb_addr_rs / s->ctb_width;
-		int tile_x = 0;
-		int tile_y = 0;
-		int val = 0;
-
-		for (i = 0; i < s->num_tile_columns; i++) {
-			if (tb_x < s->col_bd[i + 1]) {
-				tile_x = i;
-				break;
-			}
-		}
-
-		for (i = 0; i < s->num_tile_rows; i++) {
-			if (tb_y < s->row_bd[i + 1]) {
-				tile_y = i;
-				break;
-			}
-		}
-
-		for (i = 0; i < tile_x; i++)
-			val += s->row_height[tile_y] * s->column_width[i];
-		for (i = 0; i < tile_y; i++)
-			val += s->ctb_width * s->row_height[i];
-
-		val += (tb_y - s->row_bd[tile_y]) * s->column_width[tile_x] +
-		       tb_x - s->col_bd[tile_x];
-
-		s->ctb_addr_rs_to_ts[ctb_addr_rs] = val;
-		s->ctb_addr_ts_to_rs[val] = ctb_addr_rs;
-	}
-
-	for (j = 0, tile_id = 0; j < s->num_tile_rows; j++)
-		for (i = 0; i < s->num_tile_columns; i++, tile_id++)
-			for (y = s->row_bd[j]; y < s->row_bd[j + 1]; y++)
-				for (x = s->col_bd[i];
-				     x < s->col_bd[i + 1];
-				     x++)
-					s->tile_id[s->ctb_addr_rs_to_ts
-							   [y * s->ctb_width +
-							    x]] = tile_id;
+	fill_rs_to_ts(s);
 
 	return 0;
 }
 
-static int frame_end(struct rpivid_dev *const dev,
-		     struct rpivid_dec_env *const de,
-		     const struct rpivid_dec_state *const s)
-{
-	const unsigned int last_x = s->col_bd[s->num_tile_columns] - 1;
-	const unsigned int last_y = s->row_bd[s->num_tile_rows] - 1;
-	size_t cmd_size;
-
-	if (s->pps.flags & V4L2_HEVC_PPS_FLAG_ENTROPY_CODING_SYNC_ENABLED) {
-		if (de->wpp_entry_x < 2 && de->pic_width_in_ctbs_y > 2)
-			wpp_pause(de, last_y);
-	}
-	p1_apb_write(de, RPI_STATUS, 1 + (last_x << 5) + (last_y << 18));
-
+static int write_cmd_buffer(struct rpivid_dev *const dev,
+			    struct rpivid_dec_env *const de,
+			    const struct rpivid_dec_state *const s)
+{
 	// Copy commands out to dma buf
-	cmd_size = de->cmd_len * sizeof(de->cmd_fifo[0]);
+	const size_t cmd_size = de->cmd_len * sizeof(de->cmd_fifo[0]);
 
 	if (!de->cmd_copy_gptr->ptr || cmd_size > de->cmd_copy_gptr->size) {
 		size_t cmd_alloc = round_up_size(cmd_size);
@@ -1521,18 +1528,19 @@ static void rpivid_h265_setup(struct rpi
 	struct rpivid_q_aux *dpb_q_aux[V4L2_HEVC_DPB_ENTRIES_NUM_MAX];
 	struct rpivid_dec_state *const s = ctx->state;
 	struct vb2_queue *vq;
-	struct rpivid_dec_env *de;
-	int ctb_addr_ts;
+	struct rpivid_dec_env *de = ctx->dec0;
+	unsigned int prev_rs;
 	unsigned int i;
 	int use_aux;
 	bool slice_temporal_mvp;
 
+	xtrace_in(dev, de);
+
 	pred_weight_table = &sh->pred_weight_table;
 
 	s->frame_end =
 		((run->src->flags & V4L2_BUF_FLAG_M2M_HOLD_CAPTURE_BUF) == 0);
 
-	de = ctx->dec0;
 	slice_temporal_mvp = (sh->flags &
 		   V4L2_HEVC_SLICE_PARAMS_FLAG_SLICE_TEMPORAL_MVP_ENABLED);
 
@@ -1662,6 +1670,13 @@ static void rpivid_h265_setup(struct rpi
 				  s->sps.pic_height_in_luma_samples);
 			goto fail;
 		}
+		if ((s->tile_width != 1 || s->tile_height != 1) &&
+		    (s->pps.flags &
+		     V4L2_HEVC_PPS_FLAG_ENTROPY_CODING_SYNC_ENABLED)) {
+			v4l2_warn(&dev->v4l2_dev,
+				  "Tiles + WPP not supported\n");
+			goto fail;
+		}
 
 		// Fill in ref planes with our address s.t. if we mess
 		// up refs somehow then we still have a valid address
@@ -1760,15 +1775,24 @@ static void rpivid_h265_setup(struct rpi
 	if (s->sps.flags & V4L2_HEVC_SPS_FLAG_SCALING_LIST_ENABLED)
 		populate_scaling_factors(run, de, s);
 
-	ctb_addr_ts = s->ctb_addr_rs_to_ts[sh->slice_segment_addr];
+	// Calc all the random coord info to avoid repeated conversion in/out
+	s->start_ts = s->ctb_addr_rs_to_ts[sh->slice_segment_addr];
+	s->start_ctb_x = sh->slice_segment_addr % de->pic_width_in_ctbs_y;
+	s->start_ctb_y = sh->slice_segment_addr / de->pic_width_in_ctbs_y;
+	// Last CTB of previous slice
+	prev_rs = !s->start_ts ? 0 : s->ctb_addr_ts_to_rs[s->start_ts - 1];
+	s->prev_ctb_x = prev_rs % de->pic_width_in_ctbs_y;
+	s->prev_ctb_y = prev_rs / de->pic_width_in_ctbs_y;
 
 	if ((s->pps.flags & V4L2_HEVC_PPS_FLAG_ENTROPY_CODING_SYNC_ENABLED))
-		wpp_decode_slice(de, s, sh, ctb_addr_ts);
+		wpp_decode_slice(de, s);
 	else
-		decode_slice(de, s, sh, ctb_addr_ts);
+		decode_slice(de, s);
 
-	if (!s->frame_end)
+	if (!s->frame_end) {
+		xtrace_ok(dev, de);
 		return;
+	}
 
 	// Frame end
 	memset(dpb_q_aux, 0,
@@ -1776,8 +1800,9 @@ static void rpivid_h265_setup(struct rpi
 	/*
 	 * Need Aux ents for all (ref) DPB ents if temporal MV could
 	 * be enabled for any pic
-	 * ** At the moment we have aux ents for all pics whether or not
-	 *    they are ref
+	 * ** At the moment we create aux ents for all pics whether or not
+	 *    they are ref - they should then be discarded by the DPB-aux
+	 *    garbage collection code
 	 */
 	use_aux = ((s->sps.flags &
 		  V4L2_HEVC_SPS_FLAG_SPS_TEMPORAL_MVP_ENABLED) != 0);
@@ -1795,7 +1820,7 @@ static void rpivid_h265_setup(struct rpi
 	}
 
 	//        v4l2_info(&dev->v4l2_dev, "rpivid_h265_end of frame\n");
-	if (frame_end(dev, de, s))
+	if (write_cmd_buffer(dev, de, s))
 		goto fail;
 
 	for (i = 0; i < sh->num_active_dpb_entries; ++i) {
@@ -1876,6 +1901,7 @@ static void rpivid_h265_setup(struct rpi
 	}
 
 	de->state = RPIVID_DECODE_PHASE1;
+	xtrace_ok(dev, de);
 	return;
 
 fail:
@@ -1883,6 +1909,7 @@ fail:
 		// Actual error reporting happens in Trigger
 		de->state = s->frame_end ? RPIVID_DECODE_ERROR_DONE :
 					   RPIVID_DECODE_ERROR_CONTINUE;
+	xtrace_fail(dev, de);
 }
 
 //////////////////////////////////////////////////////////////////////////////
@@ -2210,6 +2237,10 @@ static int rpivid_h265_start(struct rpiv
 	size_t pu_alloc;
 	size_t coeff_alloc;
 
+#if DEBUG_TRACE_P1_CMD
+	p1_z = 0;
+#endif
+
 	// Generate a sanitised WxH for memory alloc
 	// Assume HD if unset
 	if (w == 0)