search

rockchip-linux / mpp

Media Process Platform (MPP) module
590 stars 171 forks source link

rk3568上mpp解码后由drm显示的问题 #630

Closed zeonman closed 3 months ago

zeonman commented 4 months ago

通过MPP_DEC_SET_OUTPUT_FORMAT设置为MPP_FMT_YUV420SP | MPP_FRAME_FBC_AFBC_V2后, modifier = DRM_FORMAT_MOD_ARM_AFBC(AFBC_FORMAT_MOD_BLOCK_SIZE_16x16 | AFBC_FORMAT_MOD_SPLIT | AFBC_FORMAT_MOD_SPARSE); alg_width = Align(width, 4); alg_height = Align(width, 4); uint32_t strides[4] = { alg_width, 0, 0, 0 }; uint32_t offsets[4] = { 0 }; uint64_t modifiers[4] {modifier, 0, 0, 0 }; drmModeAddFB2WithModifiers(fd, alg_width, alg_height, DRM_FORMAT_NV12,handles, strides, offsets, modifiers,&fb_id, DRM_MODE_FB_MODIFIERS); drmModeAddFB2WithModifiers返回-22,请问是什么原因呢

HermanChen commented 4 months ago

这个是 drm 相关的问题，需要看对应内核的 drm 驱动了，不属于 mpp 内部的问题，不好分析……

HermanChen commented 4 months ago 
/* HACK: Fake hstride for Rockchip DRM driver
 *
 * When importing AFBC dma-bufs, the Rockchip DRM driver would calculate
 * the pixel stride from pitch. But the pitch aligning algorithms are
 * different between MPP and the driver:
 *
 * MPP uses round_up_64(round_up_64(width) * bpp / 8)
 * Rockchip DRM driver expects (pixel_stride * bpp / 8)
 *
 * We need the driver to use the same pixel stride as MPP, so we
 * re-calculate a fake hstride from the pixel stride (i.e. the 64-aligned
 * fake width) here.
 *
 * NOTE: The hstride is not used by others for now.
 */

需要有个 trick 让 drm 能被底层正常计算到