@@ -70,6 +70,9 @@ struct hantro_irq {
  * @reg_names:			array of register range names
  * @num_regs:			number of register range names in the array
  * @postproc_regs:		&struct hantro_postproc_regs pointer
+ * @scaling:			Set possible scaled output formats.
+ *				Returns zero if OK, a negative value in error cases.
+ *				Optional.
  */
 struct hantro_variant {
 	unsigned int enc_offset;

@@ -91,6 +94,7 @@ struct hantro_variant {
 	const char * const *reg_names;
 	int num_regs;
 	const struct hantro_postproc_regs *postproc_regs;
+	int (*scaling)(struct hantro_ctx *ctx, struct v4l2_frmsizeenum *fsize);
 };
 
 /**

@@ -396,6 +396,17 @@ static void set_ref_pic_list(struct hantro_ctx *ctx)
 	}
 }
 
+static int down_scale_factor(struct hantro_ctx *ctx)
+{
+	const struct hantro_hevc_dec_ctrls *ctrls = &ctx->hevc_dec.ctrls;
+	const struct v4l2_ctrl_hevc_sps *sps = ctrls->sps;
+
+	if (sps->pic_width_in_luma_samples == ctx->dst_fmt.width)
+		return 0;
+
+	return DIV_ROUND_CLOSEST(sps->pic_width_in_luma_samples, ctx->dst_fmt.width);
+}
+
 static int set_ref(struct hantro_ctx *ctx)
 {
 	const struct hantro_hevc_dec_ctrls *ctrls = &ctx->hevc_dec.ctrls;

@@ -409,6 +420,7 @@ static int set_ref(struct hantro_ctx *ctx)
 	size_t mv_offset = hantro_hevc_motion_vectors_offset(sps);
 	size_t compress_luma_offset = hantro_hevc_luma_compress_offset(sps);
 	size_t compress_chroma_offset = hantro_hevc_chroma_compress_offset(sps);
+	int down_scale = down_scale_factor(ctx);
 	u32 max_ref_frames;
 	u16 dpb_longterm_e;
 	static const struct hantro_reg cur_poc[] = {

@@ -511,8 +523,18 @@ static int set_ref(struct hantro_ctx *ctx)
 	hantro_write_addr(vpu, G2_REG_CHR_REF(i), chroma_addr);
 	hantro_write_addr(vpu, G2_REG_DMV_REF(i++), mv_addr);
 
-	hantro_write_addr(vpu, G2_ADDR_DST, luma_addr);
-	hantro_write_addr(vpu, G2_ADDR_DST_CHR, chroma_addr);
+	if (down_scale) {
+		chroma_addr = luma_addr + (cr_offset >> down_scale);
+		hantro_reg_write(vpu, &g2_down_scale_e, 1);
+		hantro_reg_write(vpu, &g2_down_scale_y, down_scale >> 2);
+		hantro_reg_write(vpu, &g2_down_scale_x, down_scale >> 2);
+		hantro_write_addr(vpu, G2_DS_DST, luma_addr);
+		hantro_write_addr(vpu, G2_DS_DST_CHR, chroma_addr);
+	} else {
+		hantro_write_addr(vpu, G2_ADDR_DST, luma_addr);
+		hantro_write_addr(vpu, G2_ADDR_DST_CHR, chroma_addr);
+	}
+
 	hantro_write_addr(vpu, G2_ADDR_DST_MV, mv_addr);
 	hantro_write_addr(vpu, G2_COMP_ADDR_DST, compress_luma_addr);
 	hantro_write_addr(vpu, G2_COMP_CHR, compress_chroma_addr);

@@ -592,6 +614,26 @@ static void hantro_g2_check_idle(struct hantro_dev *vpu)
 	}
 }
 
+int hantro_g2_hevc_dec_scaling(struct hantro_ctx *ctx,
+			       struct v4l2_frmsizeenum *fsize)
+{
+	/**
+	 * G2 scaler can scale down by 0, 2, 4 or 8
+	 * use fsize->index has power of 2 diviser
+	 **/
+	if (fsize->index > 3)
+		return -EINVAL;
+
+	if (!ctx->src_fmt.width || !ctx->src_fmt.height)
+		return -EINVAL;
+
+	fsize->type = V4L2_FRMSIZE_TYPE_DISCRETE;
+	fsize->discrete.width = ctx->src_fmt.width >> fsize->index;
+	fsize->discrete.height = ctx->src_fmt.height >> fsize->index;
+
+	return 0;
+}
+
 int hantro_g2_hevc_dec_run(struct hantro_ctx *ctx)
 {
 	struct hantro_dev *vpu = ctx->dev;

@@ -174,6 +174,10 @@
 #define g2_buswidth		G2_DEC_REG(58, 8,  0x7)
 #define g2_max_burst		G2_DEC_REG(58, 0,  0xff)
 
+#define g2_down_scale_e	G2_DEC_REG(184, 7, 0x1)
+#define g2_down_scale_y	G2_DEC_REG(184, 2, 0x3)
+#define g2_down_scale_x	G2_DEC_REG(184, 0, 0x3)
+
 #define G2_REG_CONFIG				G2_SWREG(58)
 #define G2_REG_CONFIG_DEC_CLK_GATE_E		BIT(16)
 #define G2_REG_CONFIG_DEC_CLK_GATE_IDLE_E	BIT(17)

@@ -193,6 +197,8 @@
 #define G2_TILE_FILTER		(G2_SWREG(179))
 #define G2_TILE_SAO		(G2_SWREG(181))
 #define G2_TILE_BSD		(G2_SWREG(183))
+#define G2_DS_DST		(G2_SWREG(186))
+#define G2_DS_DST_CHR		(G2_SWREG(188))
 #define G2_COMP_ADDR_DST	(G2_SWREG(190))
 #define G2_COMP_ADDR_REF(i)	(G2_SWREG(192) + ((i) * 0x8))
 #define G2_COMP_CHR		(G2_SWREG(224))

@@ -241,6 +241,7 @@ int hantro_hevc_dec_init(struct hantro_ctx *ctx);
 void hantro_hevc_dec_exit(struct hantro_ctx *ctx);
 int hantro_g2_hevc_dec_run(struct hantro_ctx *ctx);
 void hantro_g2_hevc_dec_done(struct hantro_ctx *ctx);
+int hantro_g2_hevc_dec_scaling(struct hantro_ctx *ctx, struct v4l2_frmsizeenum *fsize);
 int hantro_hevc_dec_prepare_run(struct hantro_ctx *ctx);
 dma_addr_t hantro_hevc_get_ref_buf(struct hantro_ctx *ctx, int poc);
 void hantro_hevc_ref_remove_unused(struct hantro_ctx *ctx);

@@ -118,7 +118,7 @@ static int vidioc_enum_framesizes(struct file *file, void *priv,
 	struct hantro_ctx *ctx = fh_to_ctx(priv);
 	const struct hantro_fmt *fmt;
 
-	if (fsize->index != 0) {
+	if (fsize->index != 0 && !ctx->dev->variant->scaling) {
 		vpu_debug(0, "invalid frame size index (expected 0, got %d)\n",
 			  fsize->index);
 		return -EINVAL;

@@ -131,9 +131,13 @@ static int vidioc_enum_framesizes(struct file *file, void *priv,
 		return -EINVAL;
 	}
 
-	/* This only makes sense for coded formats */
-	if (fmt->codec_mode == HANTRO_MODE_NONE)
+	/* For non-coded formats check if scaling is possible */
+	if (fmt->codec_mode == HANTRO_MODE_NONE) {
+		if (ctx->dev->variant->scaling)
+			return ctx->dev->variant->scaling(ctx, fsize);
+
 		return -EINVAL;
+	}
 
 	fsize->type = V4L2_FRMSIZE_TYPE_STEPWISE;
 	fsize->stepwise = fmt->frmsize;

@@ -293,4 +293,5 @@ const struct hantro_variant imx8mq_vpu_g2_variant = {
 	.num_irqs = ARRAY_SIZE(imx8mq_g2_irqs),
 	.clk_names = imx8mq_clk_names,
 	.num_clocks = ARRAY_SIZE(imx8mq_clk_names),
+	.scaling = hantro_g2_hevc_dec_scaling,
 };