Add vp9 single core decode driver to support mt8186. Signed-off-by: Mingjia Zhang <mingjia.zhang@xxxxxxxxxxxx> Signed-off-by: Xiaoyong Lu <xiaoyong.lu@xxxxxxxxxxxx> Signed-off-by: Yunfei Dong <yunfei.dong@xxxxxxxxxxxx> --- This patch depends on "support mt8195 decoder"[1] [1] https://patchwork.kernel.org/project/linux-mediatek/cover/20220507014618.29412-1-yunfei.dong@xxxxxxxxxxxx/ --- changed with v1: - fix sparse and smatch check fail. --- .../vcodec/vdec/vdec_vp9_req_lat_if.c | 197 +++++++++++++++++- .../platform/mediatek/vcodec/vdec_drv_if.c | 2 +- 2 files changed, 195 insertions(+), 4 deletions(-) diff --git a/drivers/media/platform/mediatek/vcodec/vdec/vdec_vp9_req_lat_if.c b/drivers/media/platform/mediatek/vcodec/vdec/vdec_vp9_req_lat_if.c index 3ebdb3c29e0e..f4de6951269f 100644 --- a/drivers/media/platform/mediatek/vcodec/vdec/vdec_vp9_req_lat_if.c +++ b/drivers/media/platform/mediatek/vcodec/vdec/vdec_vp9_req_lat_if.c @@ -439,6 +439,8 @@ struct vdec_vp9_slice_ref { * @init_vsi: vsi used for initialized VP9 instance * @vsi: vsi used for decoding/flush ... * @core_vsi: vsi used for Core stage + * + * @sc_pfc: per frame context single core * @counts_map: used map to counts_helper * @counts_helper: counts table according to newest kernel spec */ @@ -487,6 +489,7 @@ struct vdec_vp9_slice_instance { }; struct vdec_vp9_slice_vsi *core_vsi; + struct vdec_vp9_slice_pfc sc_pfc; struct vdec_vp9_slice_counts_map counts_map; struct v4l2_vp9_frame_symbol_counts counts_helper; }; @@ -693,6 +696,25 @@ static int vdec_vp9_slice_tile_offset(int idx, int mi_num, int tile_log2) return offset < mi_num ? offset : mi_num; } +static +int vdec_vp9_slice_setup_single_from_src_to_dst(struct vdec_vp9_slice_instance *instance) +{ + struct vb2_v4l2_buffer *src; + struct vb2_v4l2_buffer *dst; + + src = v4l2_m2m_next_src_buf(instance->ctx->m2m_ctx); + if (!src) + return -EINVAL; + + dst = v4l2_m2m_next_dst_buf(instance->ctx->m2m_ctx); + if (!dst) + return -EINVAL; + + v4l2_m2m_buf_copy_metadata(src, dst, true); + + return 0; +} + static int vdec_vp9_slice_setup_lat_from_src_buf(struct vdec_vp9_slice_instance *instance, struct vdec_lat_buf *lat_buf) { @@ -1568,6 +1590,33 @@ static int vdec_vp9_slice_update_prob(struct vdec_vp9_slice_instance *instance, return 0; } +static int vdec_vp9_slice_update_single(struct vdec_vp9_slice_instance *instance, + struct vdec_vp9_slice_pfc *pfc) +{ + struct vdec_vp9_slice_vsi *vsi; + + vsi = &pfc->vsi; + memcpy(&pfc->state[0], &vsi->state, sizeof(vsi->state)); + + mtk_vcodec_debug(instance, "Frame %u Y_CRC %08x %08x %08x %08x\n", + pfc->seq, + vsi->state.crc[0], vsi->state.crc[1], + vsi->state.crc[2], vsi->state.crc[3]); + mtk_vcodec_debug(instance, "Frame %u C_CRC %08x %08x %08x %08x\n", + pfc->seq, + vsi->state.crc[4], vsi->state.crc[5], + vsi->state.crc[6], vsi->state.crc[7]); + + vdec_vp9_slice_update_prob(instance, vsi); + + instance->width = vsi->frame.uh.frame_width; + instance->height = vsi->frame.uh.frame_height; + instance->frame_type = vsi->frame.uh.frame_type; + instance->show_frame = vsi->frame.uh.show_frame; + + return 0; +} + static int vdec_vp9_slice_update_lat(struct vdec_vp9_slice_instance *instance, struct vdec_lat_buf *lat_buf, struct vdec_vp9_slice_pfc *pfc) @@ -1691,6 +1740,40 @@ static int vdec_vp9_slice_setup_core_buffer(struct vdec_vp9_slice_instance *inst return 0; } +static void vdec_vp9_slice_setup_single_buffer(struct vdec_vp9_slice_instance *instance, + struct vdec_vp9_slice_pfc *pfc, + struct vdec_vp9_slice_vsi *vsi, + struct mtk_vcodec_mem *bs, + struct vdec_fb *fb) +{ + int i; + + vsi->bs.buf.dma_addr = bs->dma_addr; + vsi->bs.buf.size = bs->size; + vsi->bs.frame.dma_addr = bs->dma_addr; + vsi->bs.frame.size = bs->size; + + for (i = 0; i < 2; i++) { + vsi->mv[i].dma_addr = instance->mv[i].dma_addr; + vsi->mv[i].size = instance->mv[i].size; + } + for (i = 0; i < 2; i++) { + vsi->seg[i].dma_addr = instance->seg[i].dma_addr; + vsi->seg[i].size = instance->seg[i].size; + } + vsi->tile.dma_addr = instance->tile.dma_addr; + vsi->tile.size = instance->tile.size; + vsi->prob.dma_addr = instance->prob.dma_addr; + vsi->prob.size = instance->prob.size; + vsi->counts.dma_addr = instance->counts.dma_addr; + vsi->counts.size = instance->counts.size; + + vsi->row_info.buf = 0; + vsi->row_info.size = 0; + + vdec_vp9_slice_setup_core_buffer(instance, pfc, vsi, fb, NULL); +} + static int vdec_vp9_slice_setup_core(struct vdec_vp9_slice_instance *instance, struct vdec_fb *fb, struct vdec_lat_buf *lat_buf, @@ -1717,6 +1800,43 @@ static int vdec_vp9_slice_setup_core(struct vdec_vp9_slice_instance *instance, return ret; } +static int vdec_vp9_slice_setup_single(struct vdec_vp9_slice_instance *instance, + struct mtk_vcodec_mem *bs, + struct vdec_fb *fb, + struct vdec_vp9_slice_pfc *pfc) +{ + struct vdec_vp9_slice_vsi *vsi = &pfc->vsi; + int ret; + + ret = vdec_vp9_slice_setup_single_from_src_to_dst(instance); + if (ret) + goto err; + + ret = vdec_vp9_slice_setup_pfc(instance, pfc); + if (ret) + goto err; + + ret = vdec_vp9_slice_alloc_working_buffer(instance, vsi); + if (ret) + goto err; + + vdec_vp9_slice_setup_single_buffer(instance, pfc, vsi, bs, fb); + vdec_vp9_slice_setup_seg_buffer(instance, vsi, &instance->seg[0]); + + ret = vdec_vp9_slice_setup_prob_buffer(instance, vsi); + if (ret) + goto err; + + ret = vdec_vp9_slice_setup_tile_buffer(instance, vsi, bs); + if (ret) + goto err; + + return 0; + +err: + return ret; +} + static int vdec_vp9_slice_update_core(struct vdec_vp9_slice_instance *instance, struct vdec_lat_buf *lat_buf, struct vdec_vp9_slice_pfc *pfc) @@ -1814,8 +1934,8 @@ static int vdec_vp9_slice_flush(void *h_vdec, struct mtk_vcodec_mem *bs, struct vdec_vp9_slice_instance *instance = h_vdec; mtk_vcodec_debug(instance, "flush ...\n"); - - vdec_msg_queue_wait_lat_buf_full(&instance->ctx->msg_queue); + if (instance->ctx->dev->vdec_pdata->hw_arch != MTK_VDEC_PURE_SINGLE_CORE) + vdec_msg_queue_wait_lat_buf_full(&instance->ctx->msg_queue); return vpu_dec_reset(&instance->vpu); } @@ -1868,6 +1988,63 @@ static int vdec_vp9_slice_get_param(void *h_vdec, enum vdec_get_param_type type, return 0; } +static int vdec_vp9_slice_single_decode(void *h_vdec, struct mtk_vcodec_mem *bs, + struct vdec_fb *fb, bool *res_chg) +{ + struct vdec_vp9_slice_instance *instance = h_vdec; + struct vdec_vp9_slice_pfc *pfc = &(instance->sc_pfc); + struct vdec_vp9_slice_vsi *vsi; + struct mtk_vcodec_ctx *ctx; + int ret; + + if (!instance || !instance->ctx) + return -EINVAL; + ctx = instance->ctx; + + /* bs NULL means flush decoder */ + if (!bs) + return vdec_vp9_slice_flush(h_vdec, bs, fb, res_chg); + + fb = ctx->dev->vdec_pdata->get_cap_buffer(ctx); + if (!fb) + return -EBUSY; + + vsi = &pfc->vsi; + + ret = vdec_vp9_slice_setup_single(instance, bs, fb, pfc); + if (ret) { + mtk_vcodec_err(instance, "Failed to setup VP9 single ret %d\n", ret); + return ret; + } + vdec_vp9_slice_vsi_to_remote(vsi, (void __iomem *)instance->vsi); + + ret = vpu_dec_start(&instance->vpu, NULL, 0); + if (ret) { + mtk_vcodec_err(instance, "Failed to dec VP9 ret %d\n", ret); + return ret; + } + + ret = mtk_vcodec_wait_for_done_ctx(ctx, MTK_INST_IRQ_RECEIVED, + WAIT_INTR_TIMEOUT_MS, MTK_VDEC_CORE); + /* update remote vsi if decode timeout */ + if (ret) { + mtk_vcodec_err(instance, "VP9 decode timeout %d\n", ret); + writel(1, (void __iomem *)&instance->vsi->state.timeout); + } + + vpu_dec_end(&instance->vpu); + + vdec_vp9_slice_vsi_from_remote(vsi, (void __iomem *)instance->vsi, 0); + ret = vdec_vp9_slice_update_single(instance, pfc); + if (ret) { + mtk_vcodec_err(instance, "VP9 decode error: %d\n", ret); + return ret; + } + + instance->ctx->decoded_frame_cnt++; + return 0; +} + static int vdec_vp9_slice_lat_decode(void *h_vdec, struct mtk_vcodec_mem *bs, struct vdec_fb *fb, bool *res_chg) { @@ -1947,6 +2124,20 @@ static int vdec_vp9_slice_lat_decode(void *h_vdec, struct mtk_vcodec_mem *bs, return 0; } +static int vdec_vp9_slice_decode(void *h_vdec, struct mtk_vcodec_mem *bs, + struct vdec_fb *fb, bool *res_chg) +{ + struct vdec_vp9_slice_instance *instance = h_vdec; + int ret; + + if (instance->ctx->dev->vdec_pdata->hw_arch == MTK_VDEC_PURE_SINGLE_CORE) + ret = vdec_vp9_slice_single_decode(h_vdec, bs, fb, res_chg); + else + ret = vdec_vp9_slice_lat_decode(h_vdec, bs, fb, res_chg); + + return ret; +} + static int vdec_vp9_slice_core_decode(struct vdec_lat_buf *lat_buf) { struct vdec_vp9_slice_instance *instance; @@ -2025,7 +2216,7 @@ static int vdec_vp9_slice_core_decode(struct vdec_lat_buf *lat_buf) const struct vdec_common_if vdec_vp9_slice_lat_if = { .init = vdec_vp9_slice_init, - .decode = vdec_vp9_slice_lat_decode, + .decode = vdec_vp9_slice_decode, .get_param = vdec_vp9_slice_get_param, .deinit = vdec_vp9_slice_deinit, }; diff --git a/drivers/media/platform/mediatek/vcodec/vdec_drv_if.c b/drivers/media/platform/mediatek/vcodec/vdec_drv_if.c index 27b4b35039cf..f3807f03d880 100644 --- a/drivers/media/platform/mediatek/vcodec/vdec_drv_if.c +++ b/drivers/media/platform/mediatek/vcodec/vdec_drv_if.c @@ -47,7 +47,7 @@ int vdec_if_init(struct mtk_vcodec_ctx *ctx, unsigned int fourcc) break; case V4L2_PIX_FMT_VP9_FRAME: ctx->dec_if = &vdec_vp9_slice_lat_if; - ctx->hw_id = MTK_VDEC_LAT0; + ctx->hw_id = IS_VDEC_LAT_ARCH(hw_arch) ? MTK_VDEC_LAT0 : MTK_VDEC_CORE; break; default: return -EINVAL; -- 2.18.0
