[igt-dev] [PATCH i-g-t 04/10] lib: Add engine copy support for YUV formats
Kahola, Mika
mika.kahola at intel.com
Mon Dec 30 13:23:41 UTC 2019
On Mon, 2019-12-30 at 05:40 +0200, Imre Deak wrote:
> Add the missing bits to the Vebox copy and AUX pagetable helpers for
> copying YUV FBs with the Vebox engine.
>
> Cc: Mika Kahola <mika.kahola at intel.com>
> Signed-off-by: Imre Deak <imre.deak at intel.com>
Reviewed-by: Mika Kahola <mika.kahola at intel.com>
> ---
> lib/igt_fb.c | 79 ++++++++++++++++++++++++--
> lib/intel_aux_pgtable.c | 121 ++++++++++++++++++++++++++++++++++++
> ----
> lib/intel_batchbuffer.h | 4 ++
> lib/veboxcopy_gen12.c | 58 ++++++++++++++++---
> 4 files changed, 238 insertions(+), 24 deletions(-)
>
> diff --git a/lib/igt_fb.c b/lib/igt_fb.c
> index cc0fb373..e6a3ff07 100644
> --- a/lib/igt_fb.c
> +++ b/lib/igt_fb.c
> @@ -359,6 +359,13 @@ static const struct format_desc_struct
> *lookup_drm_format(uint32_t drm_format)
> return NULL;
> }
>
> +static bool igt_format_is_yuv_semiplanar(uint32_t format)
> +{
> + const struct format_desc_struct *f = lookup_drm_format(format);
> +
> + return igt_format_is_yuv(format) && f->num_planes == 2;
> +}
> +
> /**
> * igt_get_fb_tile_size:
> * @fd: the DRM file descriptor
> @@ -1967,19 +1974,56 @@ static bool use_blitter(const struct igt_fb
> *fb)
> blitter_ok(fb);
> }
>
> +static void init_buf_ccs(struct igt_buf *buf, int ccs_idx,
> + uint32_t offset, uint32_t stride)
> +{
> + buf->ccs[ccs_idx].offset = offset;
> + buf->ccs[ccs_idx].stride = stride;
> +}
> +
> +static void init_buf_surface(struct igt_buf *buf, int surface_idx,
> + uint32_t offset, uint32_t stride, uint32_t
> size)
> +{
> + buf->surface[surface_idx].offset = offset;
> + buf->surface[surface_idx].stride = stride;
> + buf->surface[surface_idx].size = size;
> +}
> +
> +static int yuv_semiplanar_bpp(uint32_t drm_format)
> +{
> + switch (drm_format) {
> + case DRM_FORMAT_NV12:
> + return 8;
> + case DRM_FORMAT_P010:
> + return 10;
> + case DRM_FORMAT_P012:
> + return 12;
> + case DRM_FORMAT_P016:
> + return 16;
> + default:
> + igt_assert_f(0, "Unsupported format: %08x\n",
> drm_format);
> + }
> +}
> +
> static void init_buf(struct fb_blit_upload *blit,
> struct igt_buf *buf,
> const struct igt_fb *fb,
> const char *name)
> {
> + int num_surfaces;
> + int i;
> +
> igt_assert_eq(fb->offsets[0], 0);
>
> buf->bo = gem_handle_to_libdrm_bo(blit->bufmgr, blit->fd,
> name, fb->gem_handle);
> buf->tiling = igt_fb_mod_to_tiling(fb->modifier);
> - buf->surface[0].stride = fb->strides[0];
> buf->bpp = fb->plane_bpp[0];
> - buf->surface[0].size = fb->size;
> + buf->format_is_yuv = igt_format_is_yuv(fb->drm_format);
> + buf->format_is_yuv_semiplanar =
> + igt_format_is_yuv_semiplanar(fb->drm_format);
> + if (buf->format_is_yuv_semiplanar)
> + buf->yuv_semiplanar_bpp = yuv_semiplanar_bpp(fb-
> >drm_format);
>
> if (is_ccs_modifier(fb->modifier)) {
> igt_assert_eq(fb->strides[0] & 127, 0);
> @@ -1994,8 +2038,24 @@ static void init_buf(struct fb_blit_upload
> *blit,
> else
> buf->compression = I915_COMPRESSION_RENDER;
>
> - buf->ccs[0].offset = fb->offsets[1];
> - buf->ccs[0].stride = fb->strides[1];
> + num_surfaces = fb->num_planes / 2;
> + for (i = 0; i < num_surfaces; i++)
> + init_buf_ccs(buf, i,
> + fb->offsets[num_surfaces + i],
> + fb->strides[num_surfaces + i]);
> + } else {
> + num_surfaces = fb->num_planes;
> + }
> +
> + igt_assert(fb->offsets[0] == 0);
> + for (i = 0; i < num_surfaces; i++) {
> + uint32_t end =
> + i == fb->num_planes - 1 ? fb->size : fb-
> >offsets[i + 1];
> +
> + init_buf_surface(buf, i,
> + fb->offsets[i],
> + fb->strides[i],
> + end - fb->offsets[i]);
> }
>
> if (fb->modifier == I915_FORMAT_MOD_Y_TILED_GEN12_RC_CCS_CC)
> @@ -2007,6 +2067,15 @@ static void fini_buf(struct igt_buf *buf)
> drm_intel_bo_unreference(buf->bo);
> }
>
> +static bool use_vebox_copy(const struct igt_fb *src_fb,
> + const struct igt_fb *dst_fb)
> +{
> +
> + return is_gen12_mc_ccs_modifier(dst_fb->modifier) ||
> + igt_format_is_yuv(src_fb->drm_format) ||
> + igt_format_is_yuv(dst_fb->drm_format);
> +}
> +
> /**
> * copy_with_engine:
> * @blit: context for the copy operation
> @@ -2029,7 +2098,7 @@ static void copy_with_engine(struct
> fb_blit_upload *blit,
> igt_render_copyfunc_t render_copy = NULL;
> igt_vebox_copyfunc_t vebox_copy = NULL;
>
> - if (is_gen12_mc_ccs_modifier(dst_fb->modifier))
> + if (use_vebox_copy(src_fb, dst_fb))
> vebox_copy =
> igt_get_vebox_copyfunc(intel_get_drm_devid(blit->fd));
> else
> render_copy =
> igt_get_render_copyfunc(intel_get_drm_devid(blit->fd));
> diff --git a/lib/intel_aux_pgtable.c b/lib/intel_aux_pgtable.c
> index 5addb2e2..fcd24f08 100644
> --- a/lib/intel_aux_pgtable.c
> +++ b/lib/intel_aux_pgtable.c
> @@ -33,6 +33,12 @@
>
> #define max(a, b) ((a) > (b) ? (a) : (b))
>
> +#define AUX_FORMAT_YCRCB 0x03
> +#define AUX_FORMAT_P010 0x07
> +#define AUX_FORMAT_P016 0x08
> +#define AUX_FORMAT_ARGB_8B 0x0A
> +#define AUX_FORMAT_NV12_21 0x0F
> +
> struct pgtable_level_desc {
> int idx_shift;
> int idx_bits;
> @@ -55,6 +61,23 @@ struct pgtable {
> drm_intel_bo *bo;
> };
>
> +static uint64_t last_buf_surface_end(const struct igt_buf *buf)
> +{
> + uint64_t end_offset = 0;
> + int num_surfaces = buf->format_is_yuv_semiplanar ? 2 : 1;
> + int i;
> +
> + for (i = 0; i < num_surfaces; i++) {
> + uint64_t surface_end = buf->surface[i].offset +
> + buf->surface[i].size;
> +
> + if (surface_end > end_offset)
> + end_offset = surface_end;
> + }
> +
> + return end_offset;
> +}
> +
> static int
> pgt_table_count(int address_bits, const struct igt_buf **bufs, int
> buf_count)
> {
> @@ -77,7 +100,7 @@ pgt_table_count(int address_bits, const struct
> igt_buf **bufs, int buf_count)
> /* Avoid double counting for overlapping aligned bufs.
> */
> start = max(start, end);
>
> - end = ALIGN(buf->bo->offset64 + buf->surface[0].size,
> + end = ALIGN(buf->bo->offset64 +
> last_buf_surface_end(buf),
> 1UL << address_bits);
> igt_assert(end >= start);
>
> @@ -189,7 +212,29 @@ pgt_set_l1_entry(struct pgtable *pgt, uint64_t
> l1_table,
> *l1_entry_ptr = ptr | flags;
> }
>
> -static uint64_t pgt_get_l1_flags(const struct igt_buf *buf)
> +#define DEPTH_VAL_RESERVED 3
> +
> +static int bpp_to_depth_val(int bpp)
> +{
> + switch (bpp) {
> + case 8:
> + return 4;
> + case 10:
> + return 1;
> + case 12:
> + return 2;
> + case 16:
> + return 0;
> + case 32:
> + return 5;
> + case 64:
> + return 6;
> + default:
> + igt_assert_f(0, "invalid bpp %d\n", bpp);
> + }
> +}
> +
> +static uint64_t pgt_get_l1_flags(const struct igt_buf *buf, int
> surface_idx)
> {
> /*
> * The offset of .tile_mode isn't specifed by bspec, it's what
> Mesa
> @@ -213,8 +258,6 @@ static uint64_t pgt_get_l1_flags(const struct
> igt_buf *buf)
> .e = {
> .valid = 1,
> .tile_mode = buf->tiling == I915_TILING_Y ? 1 :
> 0,
> - .depth = 5, /* 32bpp */
> - .format = 0xA, /* B8G8R8A8_UNORM */
> }
> };
>
> @@ -227,7 +270,49 @@ static uint64_t pgt_get_l1_flags(const struct
> igt_buf *buf)
> buf->tiling == I915_TILING_Yf ||
> buf->tiling == I915_TILING_Ys);
>
> - igt_assert(buf->bpp == 32);
> + entry.e.ycr = surface_idx > 0;
> +
> + if (buf->format_is_yuv_semiplanar) {
> + entry.e.depth = bpp_to_depth_val(buf->bpp);
> + switch (buf->yuv_semiplanar_bpp) {
> + case 8:
> + entry.e.format = AUX_FORMAT_NV12_21;
> + entry.e.depth = DEPTH_VAL_RESERVED;
> + break;
> + case 10:
> + entry.e.format = AUX_FORMAT_P010;
> + entry.e.depth = bpp_to_depth_val(10);
> + break;
> + case 12:
> + entry.e.format = AUX_FORMAT_P016;
> + entry.e.depth = bpp_to_depth_val(12);
> + break;
> + case 16:
> + entry.e.format = AUX_FORMAT_P016;
> + entry.e.depth = bpp_to_depth_val(16);
> + break;
> + default:
> + igt_assert(0);
> + }
> + } else if (buf->format_is_yuv) {
> + switch (buf->bpp) {
> + case 16:
> + entry.e.format = AUX_FORMAT_YCRCB;
> + entry.e.depth = DEPTH_VAL_RESERVED;
> + break;
> + default:
> + igt_assert(0);
> + }
> + } else {
> + switch (buf->bpp) {
> + case 32:
> + entry.e.format = AUX_FORMAT_ARGB_8B;
> + entry.e.depth = bpp_to_depth_val(32);
> + break;
> + default:
> + igt_assert(0);
> + }
> + }
>
> return entry.l;
> }
> @@ -253,14 +338,21 @@ static uint64_t pgt_get_lx_flags(void)
> static void
> pgt_populate_entries_for_buf(struct pgtable *pgt,
> const struct igt_buf *buf,
> - uint64_t top_table)
> + uint64_t top_table,
> + int surface_idx)
> {
> - uint64_t surface_addr = buf->bo->offset64;
> - uint64_t surface_end = surface_addr + buf->surface[0].size;
> - uint64_t aux_addr = buf->bo->offset64 + buf->ccs[0].offset;
> - uint64_t l1_flags = pgt_get_l1_flags(buf);
> + uint64_t surface_addr = buf->bo->offset64 +
> + buf->surface[surface_idx].offset;
> + uint64_t surface_end = surface_addr +
> + buf->surface[surface_idx].size;
> + uint64_t aux_addr = buf->bo->offset64 + buf-
> >ccs[surface_idx].offset;
> + uint64_t l1_flags = pgt_get_l1_flags(buf, surface_idx);
> uint64_t lx_flags = pgt_get_lx_flags();
>
> + igt_assert(!(buf->surface[surface_idx].stride % 512));
> + igt_assert_eq(buf->ccs[surface_idx].stride,
> + buf->surface[surface_idx].stride / 512 * 64);
> +
> for (; surface_addr < surface_end;
> surface_addr += MAIN_SURFACE_BLOCK_SIZE,
> aux_addr += AUX_CCS_BLOCK_SIZE) {
> @@ -292,8 +384,13 @@ static void pgt_populate_entries(struct pgtable
> *pgt,
> /* Top level table must be at offset 0. */
> igt_assert(top_table == 0);
>
> - for (i = 0; i < buf_count; i++)
> - pgt_populate_entries_for_buf(pgt, bufs[i], top_table);
> + for (i = 0; i < buf_count; i++) {
> + igt_assert_eq(bufs[i]->surface[0].offset, 0);
> +
> + pgt_populate_entries_for_buf(pgt, bufs[i], top_table,
> 0);
> + if (bufs[i]->format_is_yuv_semiplanar)
> + pgt_populate_entries_for_buf(pgt, bufs[i],
> top_table, 1);
> + }
> }
>
> static struct pgtable *
> diff --git a/lib/intel_batchbuffer.h b/lib/intel_batchbuffer.h
> index 69580839..fd7ef03f 100644
> --- a/lib/intel_batchbuffer.h
> +++ b/lib/intel_batchbuffer.h
> @@ -235,8 +235,12 @@ struct igt_buf {
> uint32_t tiling;
> enum i915_compression compression;
> uint32_t bpp;
> + uint32_t yuv_semiplanar_bpp;
> uint32_t *data;
> + bool format_is_yuv:1;
> + bool format_is_yuv_semiplanar:1;
> struct {
> + uint32_t offset;
> uint32_t stride;
> uint32_t size;
> } surface[2];
> diff --git a/lib/veboxcopy_gen12.c b/lib/veboxcopy_gen12.c
> index 2f017514..237c43f2 100644
> --- a/lib/veboxcopy_gen12.c
> +++ b/lib/veboxcopy_gen12.c
> @@ -26,7 +26,10 @@
> #include "intel_aux_pgtable.h"
> #include "veboxcopy.h"
>
> +#define YCRCB_NORMAL 0
> +#define PLANAR_420_8 4
> #define R8G8B8A8_UNORM 8
> +#define PLANAR_420_16 12
>
> struct vebox_surface_state {
> struct {
> @@ -129,10 +132,23 @@ struct vebox_tiling_convert {
> };
> } __attribute__((packed));
>
> +static bool format_is_interleaved_yuv(int format)
> +{
> + switch (format) {
> + case YCRCB_NORMAL:
> + case PLANAR_420_8:
> + case PLANAR_420_16:
> + return true;
> + }
> +
> + return false;
> +}
> +
> static void emit_surface_state_cmd(struct intel_batchbuffer *batch,
> int surface_id,
> int width, int height, int bpp,
> - int pitch, uint32_t tiling, int
> format)
> + int pitch, uint32_t tiling, int
> format,
> + uint32_t uv_offset)
> {
> struct vebox_surface_state *ss;
>
> @@ -149,11 +165,15 @@ static void emit_surface_state_cmd(struct
> intel_batchbuffer *batch,
> ss->ss2.width = width - 1;
>
> ss->ss3.surface_format = format;
> + if (format_is_interleaved_yuv(format))
> + ss->ss3.chroma_interleave = 1;
> ss->ss3.surface_pitch = pitch - 1;
> ss->ss3.tile_walk = (tiling == I915_TILING_Y) ||
> (tiling == I915_TILING_Yf);
> ss->ss3.tiled_surface = tiling != I915_TILING_NONE;
>
> + ss->ss4.u_y_offset = uv_offset / pitch;
> +
> ss->ss7.derived_surface_pitch = pitch - 1;
> }
>
> @@ -226,8 +246,7 @@ void gen12_vebox_copyfunc(struct
> intel_batchbuffer *batch,
> {
> struct aux_pgtable_info aux_pgtable_info = { };
> uint32_t aux_pgtable_state;
> -
> - igt_assert(src->bpp == dst->bpp);
> + int format;
>
> intel_batchbuffer_flush_on_ring(batch, I915_EXEC_VEBOX);
>
> @@ -245,18 +264,43 @@ void gen12_vebox_copyfunc(struct
> intel_batchbuffer *batch,
>
> gen12_emit_aux_pgtable_state(batch, aux_pgtable_state, false);
>
> + /* The tiling convert command can't convert formats. */
> + igt_assert_eq(src->format_is_yuv, dst->format_is_yuv);
> + igt_assert_eq(src->format_is_yuv_semiplanar,
> + dst->format_is_yuv_semiplanar);
> + igt_assert_eq(src->bpp, dst->bpp);
> +
> /* TODO: add support for more formats */
> - igt_assert(src->bpp == 32);
> + switch (src->bpp) {
> + case 8:
> + igt_assert(src->format_is_yuv_semiplanar);
> + format = PLANAR_420_8;
> + break;
> + case 16:
> + igt_assert(src->format_is_yuv);
> + format = src->format_is_yuv_semiplanar ? PLANAR_420_16
> :
> + YCRCB_NORMAL;
> + break;
> + case 32:
> + igt_assert(!src->format_is_yuv &&
> + !src->format_is_yuv_semiplanar);
> + format = R8G8B8A8_UNORM;
> + break;
> + default:
> + igt_assert_f(0, "Unsupported bpp: %u\n", src->bpp);
> + }
> +
> + igt_assert(!src->format_is_yuv_semiplanar ||
> + (src->surface[1].offset && dst->surface[1].offset));
> emit_surface_state_cmd(batch, VEBOX_SURFACE_INPUT,
> width, height, src->bpp,
> src->surface[0].stride,
> - src->tiling, R8G8B8A8_UNORM);
> + src->tiling, format, src-
> >surface[1].offset);
>
> - igt_assert(dst->bpp == 32);
> emit_surface_state_cmd(batch, VEBOX_SURFACE_OUTPUT,
> width, height, dst->bpp,
> dst->surface[0].stride,
> - dst->tiling, R8G8B8A8_UNORM);
> + dst->tiling, format, dst-
> >surface[1].offset);
>
> emit_tiling_convert_cmd(batch,
> src->bo, src->tiling, src->compression,
More information about the igt-dev
mailing list