LCOV - code coverage report
Current view: top level - src - gpu.h (source / functions) Hit Total Coverage
Test: Code coverage Lines: 9 11 81.8 %
Date: 2025-03-29 09:04:10 Functions: 1 1 100.0 %
Legend: Lines: hit not hit | Branches: + taken - not taken # not executed Branches: 8 17 47.1 %

           Branch data     Line data    Source code
       1                 :            : /*
       2                 :            :  * This file is part of libplacebo.
       3                 :            :  *
       4                 :            :  * libplacebo is free software; you can redistribute it and/or
       5                 :            :  * modify it under the terms of the GNU Lesser General Public
       6                 :            :  * License as published by the Free Software Foundation; either
       7                 :            :  * version 2.1 of the License, or (at your option) any later version.
       8                 :            :  *
       9                 :            :  * libplacebo is distributed in the hope that it will be useful,
      10                 :            :  * but WITHOUT ANY WARRANTY; without even the implied warranty of
      11                 :            :  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
      12                 :            :  * GNU Lesser General Public License for more details.
      13                 :            :  *
      14                 :            :  * You should have received a copy of the GNU Lesser General Public
      15                 :            :  * License along with libplacebo.  If not, see <http://www.gnu.org/licenses/>.
      16                 :            :  */
      17                 :            : 
      18                 :            : #pragma once
      19                 :            : 
      20                 :            : #include "common.h"
      21                 :            : #include "log.h"
      22                 :            : 
      23                 :            : #include <libplacebo/gpu.h>
      24                 :            : #include <libplacebo/dispatch.h>
      25                 :            : 
      26                 :            : // To avoid having to include drm_fourcc.h
      27                 :            : #ifndef DRM_FORMAT_MOD_LINEAR
      28                 :            : #define DRM_FORMAT_MOD_LINEAR   UINT64_C(0x0)
      29                 :            : #define DRM_FORMAT_MOD_INVALID  ((UINT64_C(1) << 56) - 1)
      30                 :            : #endif
      31                 :            : 
      32                 :            : // This struct must be the first member of the gpu's priv struct. The `pl_gpu`
      33                 :            : // helpers will cast the priv struct to this struct!
      34                 :            : 
      35                 :            : #define GPU_PFN(name) __typeof__(pl_##name) *name
      36                 :            : struct pl_gpu_fns {
      37                 :            :     // This is a pl_dispatch used (on the pl_gpu itself!) for the purposes of
      38                 :            :     // dispatching compute shaders for performing various emulation tasks (e.g.
      39                 :            :     // partial clears, blits or emulated texture transfers, see below).
      40                 :            :     //
      41                 :            :     // Warning: Care must be taken to avoid recursive calls.
      42                 :            :     pl_dispatch dp;
      43                 :            : 
      44                 :            :     // Internal cache, or NULL. Set by the user (via pl_gpu_set_cache).
      45                 :            :     _Atomic(pl_cache) cache;
      46                 :            : 
      47                 :            :     // Destructors: These also free the corresponding objects, but they
      48                 :            :     // must not be called on NULL. (The NULL checks are done by the pl_*_destroy
      49                 :            :     // wrappers)
      50                 :            :     void (*destroy)(pl_gpu gpu);
      51                 :            :     void (*tex_destroy)(pl_gpu, pl_tex);
      52                 :            :     void (*buf_destroy)(pl_gpu, pl_buf);
      53                 :            :     void (*pass_destroy)(pl_gpu, pl_pass);
      54                 :            :     void (*timer_destroy)(pl_gpu, pl_timer);
      55                 :            : 
      56                 :            :     GPU_PFN(tex_create);
      57                 :            :     GPU_PFN(tex_invalidate); // optional
      58                 :            :     GPU_PFN(tex_clear_ex); // optional if no blittable formats
      59                 :            :     GPU_PFN(tex_blit); // optional if no blittable formats
      60                 :            :     GPU_PFN(tex_upload);
      61                 :            :     GPU_PFN(tex_download);
      62                 :            :     GPU_PFN(tex_poll); // optional: if NULL, textures are always free to use
      63                 :            :     GPU_PFN(buf_create);
      64                 :            :     GPU_PFN(buf_write);
      65                 :            :     GPU_PFN(buf_read);
      66                 :            :     GPU_PFN(buf_copy);
      67                 :            :     GPU_PFN(buf_export); // optional if !gpu->export_caps.buf
      68                 :            :     GPU_PFN(buf_poll); // optional: if NULL, buffers are always free to use
      69                 :            :     GPU_PFN(desc_namespace);
      70                 :            :     GPU_PFN(pass_create);
      71                 :            :     GPU_PFN(pass_run);
      72                 :            :     GPU_PFN(timer_create); // optional
      73                 :            :     GPU_PFN(timer_query); // optional
      74                 :            :     GPU_PFN(gpu_flush); // optional
      75                 :            :     GPU_PFN(gpu_finish);
      76                 :            :     GPU_PFN(gpu_is_failed); // optional
      77                 :            : };
      78                 :            : #undef GPU_PFN
      79                 :            : 
      80                 :            : // All resources such as textures and buffers allocated from the GPU must be
      81                 :            : // destroyed before calling pl_destroy.
      82                 :            : void pl_gpu_destroy(pl_gpu gpu);
      83                 :            : 
      84                 :            : // Returns true if the device supports interop. This is considered to be
      85                 :            : // the case if at least one of `gpu->export/import_caps` is nonzero.
      86                 :            : static inline bool pl_gpu_supports_interop(pl_gpu gpu)
      87                 :            : {
      88                 :         14 :     return gpu->export_caps.tex ||
      89         [ -  + ]:          2 :            gpu->import_caps.tex ||
      90         [ -  + ]:          2 :            gpu->export_caps.buf ||
      91   [ -  +  #  # ]:          2 :            gpu->import_caps.buf ||
      92   [ +  +  -  + ]:         14 :            gpu->export_caps.sync ||
      93         [ -  + ]:          2 :            gpu->import_caps.sync;
      94                 :            : }
      95                 :            : 
      96                 :            : // Returns the GPU-internal `pl_dispatch` and `pl_cache` objects.
      97                 :            : pl_dispatch pl_gpu_dispatch(pl_gpu gpu);
      98                 :            : pl_cache pl_gpu_cache(pl_gpu gpu);
      99                 :            : 
     100                 :            : // GPU-internal helpers: these should not be used outside of GPU implementations
     101                 :            : 
     102                 :            : // This performs several tasks. It sorts the format list, logs GPU metadata,
     103                 :            : // performs verification and fixes up backwards compatibility fields. This
     104                 :            : // should be returned as the last step when creating a `pl_gpu`.
     105                 :            : pl_gpu pl_gpu_finalize(struct pl_gpu_t *gpu);
     106                 :            : 
     107                 :            : // Look up the right GLSL image format qualifier from a partially filled-in
     108                 :            : // pl_fmt, or NULL if the format does not have a legal matching GLSL name.
     109                 :            : //
     110                 :            : // `components` may differ from fmt->num_components (for emulated formats)
     111                 :            : const char *pl_fmt_glsl_format(pl_fmt fmt, int components);
     112                 :            : 
     113                 :            : // Look up the right fourcc from a partially filled-in pl_fmt, or 0 if the
     114                 :            : // format does not have a legal matching fourcc format.
     115                 :            : uint32_t pl_fmt_fourcc(pl_fmt fmt);
     116                 :            : 
     117                 :            : // Compute the total size (in bytes) of a texture transfer operation
     118                 :            : size_t pl_tex_transfer_size(const struct pl_tex_transfer_params *par);
     119                 :            : 
     120                 :            : // Split a tex transfer into slices. For emulated formats, `texel_fmt` gives
     121                 :            : // the format of the underlying texel buffer.
     122                 :            : //
     123                 :            : // Returns the number of slices, or 0 on error (e.g. no SSBOs available).
     124                 :            : // `out_slices` must be freed by caller (on success).
     125                 :            : int pl_tex_transfer_slices(pl_gpu gpu, pl_fmt texel_fmt,
     126                 :            :                            const struct pl_tex_transfer_params *params,
     127                 :            :                            struct pl_tex_transfer_params **out_slices);
     128                 :            : 
     129                 :            : // Helper that wraps pl_tex_upload/download using texture upload buffers to
     130                 :            : // ensure that params->buf is always set.
     131                 :            : bool pl_tex_upload_pbo(pl_gpu gpu, const struct pl_tex_transfer_params *params);
     132                 :            : bool pl_tex_download_pbo(pl_gpu gpu, const struct pl_tex_transfer_params *params);
     133                 :            : 
     134                 :            : // This requires that params.buf has been set and is of type PL_BUF_TEXEL_*
     135                 :            : bool pl_tex_upload_texel(pl_gpu gpu, const struct pl_tex_transfer_params *params);
     136                 :            : bool pl_tex_download_texel(pl_gpu gpu, const struct pl_tex_transfer_params *params);
     137                 :            : 
     138                 :            : // Both `src` and `dst must be storable. `src` must also be sampleable, if the
     139                 :            : // blit requires linear sampling. Returns false if these conditions are unmet.
     140                 :            : bool pl_tex_blit_compute(pl_gpu gpu, const struct pl_tex_blit_params *params);
     141                 :            : 
     142                 :            : // Helper to do a 2D blit with stretch and scale using a raster pass
     143                 :            : void pl_tex_blit_raster(pl_gpu gpu, const struct pl_tex_blit_params *params);
     144                 :            : 
     145                 :            : // Helper for GPU-accelerated endian swapping
     146                 :            : //
     147                 :            : // Note: `src` and `dst` can be the same buffer, for an in-place operation. In
     148                 :            : // this case, `src_offset` and `dst_offset` must be the same.
     149                 :            : struct pl_buf_copy_swap_params {
     150                 :            :     // Source of the copy operation. Must be `storable`.
     151                 :            :     pl_buf src;
     152                 :            :     size_t src_offset;
     153                 :            : 
     154                 :            :     // Destination of the copy operation. Must be `storable`.
     155                 :            :     pl_buf dst;
     156                 :            :     size_t dst_offset;
     157                 :            : 
     158                 :            :     // Number of bytes to copy. Must be a multiple of 4.
     159                 :            :     size_t size;
     160                 :            : 
     161                 :            :     // Underlying word size. Must be 2 (for 16-bit swap) or 4 (for 32-bit swap)
     162                 :            :     int wordsize;
     163                 :            : };
     164                 :            : 
     165                 :            : bool pl_buf_copy_swap(pl_gpu gpu, const struct pl_buf_copy_swap_params *params);
     166                 :            : 
     167                 :            : void pl_pass_run_vbo(pl_gpu gpu, const struct pl_pass_run_params *params);
     168                 :            : 
     169                 :            : // Make a deep-copy of the pass params. Note: cached_program etc. are not
     170                 :            : // copied, but cleared explicitly.
     171                 :            : struct pl_pass_params pl_pass_params_copy(void *alloc, const struct pl_pass_params *params);
     172                 :            : 
     173                 :            : // Helper to compute the size of an index buffer
     174                 :         23 : static inline size_t pl_index_buf_size(const struct pl_pass_run_params *params)
     175                 :            : {
     176      [ +  -  - ]:         23 :     switch (params->index_fmt) {
     177                 :         23 :     case PL_INDEX_UINT16: return params->vertex_count * sizeof(uint16_t);
     178                 :          0 :     case PL_INDEX_UINT32: return params->vertex_count * sizeof(uint32_t);
     179                 :            :     case PL_INDEX_FORMAT_COUNT: break;
     180                 :            :     }
     181                 :            : 
     182                 :          0 :     pl_unreachable();
     183                 :            : }
     184                 :            : 
     185                 :            : // Helper to compute the size of a vertex buffer required to fit all indices
     186                 :            : size_t pl_vertex_buf_size(const struct pl_pass_run_params *params);
     187                 :            : 
     188                 :            : // Utility function for pretty-printing UUIDs
     189                 :            : #define UUID_SIZE 16
     190                 :            : #define PRINT_UUID(uuid) (print_uuid((char[3 * UUID_SIZE]){0}, (uuid)))
     191                 :            : const char *print_uuid(char buf[3 * UUID_SIZE], const uint8_t uuid[UUID_SIZE]);
     192                 :            : 
     193                 :            : // Helper to pretty-print fourcc codes
     194                 :            : #define PRINT_FOURCC(fcc)       \
     195                 :            :     (!(fcc) ? "" : (char[5]) {  \
     196                 :            :         (fcc) & 0xFF,           \
     197                 :            :         ((fcc) >> 8) & 0xFF,    \
     198                 :            :         ((fcc) >> 16) & 0xFF,   \
     199                 :            :         ((fcc) >> 24) & 0xFF    \
     200                 :            :     })
     201                 :            : 
     202                 :            : #define DRM_MOD_SIZE 26
     203                 :            : #define PRINT_DRM_MOD(mod) (print_drm_mod((char[DRM_MOD_SIZE]){0}, (mod)))
     204                 :            : const char *print_drm_mod(char buf[DRM_MOD_SIZE], uint64_t mod);

Generated by: LCOV version 1.16