#include "pipebuffer/pb_buffer.h"
+#include "amd/common/ac_gpu_info.h"
+#include "amd/common/ac_surface.h"
+
#define RADEON_FLUSH_ASYNC (1 << 0)
#define RADEON_FLUSH_END_OF_FRAME (1 << 1)
RADEON_FLAG_GTT_WC = (1 << 0),
RADEON_FLAG_CPU_ACCESS = (1 << 1),
RADEON_FLAG_NO_CPU_ACCESS = (1 << 2),
+ RADEON_FLAG_HANDLE = (1 << 3), /* the buffer must not be suballocated */
+ RADEON_FLAG_SPARSE = (1 << 4),
};
enum radeon_bo_usage { /* bitfield */
RADEON_USAGE_READ = 2,
RADEON_USAGE_WRITE = 4,
- RADEON_USAGE_READWRITE = RADEON_USAGE_READ | RADEON_USAGE_WRITE
-};
+ RADEON_USAGE_READWRITE = RADEON_USAGE_READ | RADEON_USAGE_WRITE,
-enum radeon_family {
- CHIP_UNKNOWN = 0,
- CHIP_R300, /* R3xx-based cores. */
- CHIP_R350,
- CHIP_RV350,
- CHIP_RV370,
- CHIP_RV380,
- CHIP_RS400,
- CHIP_RC410,
- CHIP_RS480,
- CHIP_R420, /* R4xx-based cores. */
- CHIP_R423,
- CHIP_R430,
- CHIP_R480,
- CHIP_R481,
- CHIP_RV410,
- CHIP_RS600,
- CHIP_RS690,
- CHIP_RS740,
- CHIP_RV515, /* R5xx-based cores. */
- CHIP_R520,
- CHIP_RV530,
- CHIP_R580,
- CHIP_RV560,
- CHIP_RV570,
- CHIP_R600,
- CHIP_RV610,
- CHIP_RV630,
- CHIP_RV670,
- CHIP_RV620,
- CHIP_RV635,
- CHIP_RS780,
- CHIP_RS880,
- CHIP_RV770,
- CHIP_RV730,
- CHIP_RV710,
- CHIP_RV740,
- CHIP_CEDAR,
- CHIP_REDWOOD,
- CHIP_JUNIPER,
- CHIP_CYPRESS,
- CHIP_HEMLOCK,
- CHIP_PALM,
- CHIP_SUMO,
- CHIP_SUMO2,
- CHIP_BARTS,
- CHIP_TURKS,
- CHIP_CAICOS,
- CHIP_CAYMAN,
- CHIP_ARUBA,
- CHIP_TAHITI,
- CHIP_PITCAIRN,
- CHIP_VERDE,
- CHIP_OLAND,
- CHIP_HAINAN,
- CHIP_BONAIRE,
- CHIP_KAVERI,
- CHIP_KABINI,
- CHIP_HAWAII,
- CHIP_MULLINS,
- CHIP_TONGA,
- CHIP_ICELAND,
- CHIP_CARRIZO,
- CHIP_FIJI,
- CHIP_STONEY,
- CHIP_POLARIS10,
- CHIP_POLARIS11,
- CHIP_LAST,
+ /* The winsys ensures that the CS submission will be scheduled after
+ * previously flushed CSs referencing this BO in a conflicting way.
+ */
+ RADEON_USAGE_SYNCHRONIZED = 8
};
-enum chip_class {
- CLASS_UNKNOWN = 0,
- R300,
- R400,
- R500,
- R600,
- R700,
- EVERGREEN,
- CAYMAN,
- SI,
- CIK,
- VI,
-};
+#define RADEON_SPARSE_PAGE_SIZE (64 * 1024)
enum ring_type {
RING_GFX = 0,
enum radeon_value_id {
RADEON_REQUESTED_VRAM_MEMORY,
RADEON_REQUESTED_GTT_MEMORY,
+ RADEON_MAPPED_VRAM,
+ RADEON_MAPPED_GTT,
RADEON_BUFFER_WAIT_TIME_NS,
+ RADEON_NUM_MAPPED_BUFFERS,
RADEON_TIMESTAMP,
- RADEON_NUM_CS_FLUSHES,
+ RADEON_NUM_GFX_IBS,
+ RADEON_NUM_SDMA_IBS,
RADEON_NUM_BYTES_MOVED,
+ RADEON_NUM_EVICTIONS,
+ RADEON_NUM_VRAM_CPU_PAGE_FAULTS,
RADEON_VRAM_USAGE,
+ RADEON_VRAM_VIS_USAGE,
RADEON_GTT_USAGE,
RADEON_GPU_TEMPERATURE, /* DRM 2.42.0 */
RADEON_CURRENT_SCLK,
RADEON_CURRENT_MCLK,
RADEON_GPU_RESET_COUNTER, /* DRM 2.43.0 */
+ RADEON_CS_THREAD_TIME,
};
/* Each group of four has the same priority. */
RADEON_PRIO_DRAW_INDIRECT,
RADEON_PRIO_INDEX_BUFFER,
- RADEON_PRIO_CP_DMA = 8,
-
- RADEON_PRIO_VCE = 12,
+ RADEON_PRIO_VCE = 8,
RADEON_PRIO_UVD,
RADEON_PRIO_SDMA_BUFFER,
RADEON_PRIO_SDMA_TEXTURE,
- RADEON_PRIO_USER_SHADER = 16,
- RADEON_PRIO_INTERNAL_SHADER, /* fetch shader, etc. */
-
- /* gap: 20 */
+ RADEON_PRIO_CP_DMA = 12,
- RADEON_PRIO_CONST_BUFFER = 24,
+ RADEON_PRIO_CONST_BUFFER = 16,
RADEON_PRIO_DESCRIPTORS,
RADEON_PRIO_BORDER_COLORS,
- RADEON_PRIO_SAMPLER_BUFFER = 28,
+ RADEON_PRIO_SAMPLER_BUFFER = 20,
RADEON_PRIO_VERTEX_BUFFER,
- RADEON_PRIO_SHADER_RW_BUFFER = 32,
- RADEON_PRIO_RINGS_STREAMOUT,
- RADEON_PRIO_SCRATCH_BUFFER,
+ RADEON_PRIO_SHADER_RW_BUFFER = 24,
RADEON_PRIO_COMPUTE_GLOBAL,
- RADEON_PRIO_SAMPLER_TEXTURE = 36,
+ RADEON_PRIO_SAMPLER_TEXTURE = 28,
RADEON_PRIO_SHADER_RW_IMAGE,
- RADEON_PRIO_SAMPLER_TEXTURE_MSAA = 40,
+ RADEON_PRIO_SAMPLER_TEXTURE_MSAA = 32,
- RADEON_PRIO_COLOR_BUFFER = 44,
+ RADEON_PRIO_COLOR_BUFFER = 36,
- RADEON_PRIO_DEPTH_BUFFER = 48,
+ RADEON_PRIO_DEPTH_BUFFER = 40,
- RADEON_PRIO_COLOR_BUFFER_MSAA = 52,
+ RADEON_PRIO_COLOR_BUFFER_MSAA = 44,
- RADEON_PRIO_DEPTH_BUFFER_MSAA = 56,
+ RADEON_PRIO_DEPTH_BUFFER_MSAA = 48,
- RADEON_PRIO_CMASK = 60,
+ RADEON_PRIO_CMASK = 52,
RADEON_PRIO_DCC,
RADEON_PRIO_HTILE,
+ RADEON_PRIO_SHADER_BINARY, /* the hw can't hide instruction cache misses */
+
+ RADEON_PRIO_SHADER_RINGS = 56,
+
+ RADEON_PRIO_SCRATCH_BUFFER = 60,
/* 63 is the maximum value */
};
uint64_t used_gart;
};
-struct radeon_info {
- /* PCI info: domain:bus:dev:func */
- uint32_t pci_domain;
- uint32_t pci_bus;
- uint32_t pci_dev;
- uint32_t pci_func;
-
- /* Device info. */
- uint32_t pci_id;
- enum radeon_family family;
- enum chip_class chip_class;
- uint32_t gart_page_size;
- uint64_t gart_size;
- uint64_t vram_size;
- uint64_t max_alloc_size;
- bool has_dedicated_vram;
- bool has_virtual_memory;
- bool gfx_ib_pad_with_type2;
- bool has_sdma;
- bool has_uvd;
- uint32_t uvd_fw_version;
- uint32_t vce_fw_version;
- uint32_t vce_harvest_config;
- uint32_t clock_crystal_freq;
-
- /* Kernel info. */
- uint32_t drm_major; /* version */
- uint32_t drm_minor;
- uint32_t drm_patchlevel;
- bool has_userptr;
-
- /* Shader cores. */
- uint32_t r600_max_quad_pipes; /* wave size / 16 */
- uint32_t max_shader_clock;
- uint32_t num_good_compute_units;
- uint32_t max_se; /* shader engines */
- uint32_t max_sh_per_se; /* shader arrays per shader engine */
-
- /* Render backends (color + depth blocks). */
- uint32_t r300_num_gb_pipes;
- uint32_t r300_num_z_pipes;
- uint32_t r600_gb_backend_map; /* R600 harvest config */
- bool r600_gb_backend_map_valid;
- uint32_t r600_num_banks;
- uint32_t num_render_backends;
- uint32_t num_tile_pipes; /* pipe count from PIPE_CONFIG */
- uint32_t pipe_interleave_bytes;
- uint32_t enabled_rb_mask; /* GCN harvest config */
-
- /* Tile modes. */
- uint32_t si_tile_mode_array[32];
- uint32_t cik_macrotile_mode_array[16];
-};
-
/* Tiling info for display code, DRI sharing, and other data. */
struct radeon_bo_metadata {
/* Tiling flags describing the texture layout for display code
* and DRI sharing.
*/
- enum radeon_bo_layout microtile;
- enum radeon_bo_layout macrotile;
- unsigned pipe_config;
- unsigned bankw;
- unsigned bankh;
- unsigned tile_split;
- unsigned mtilea;
- unsigned num_banks;
- unsigned stride;
- bool scanout;
+ union {
+ struct {
+ enum radeon_bo_layout microtile;
+ enum radeon_bo_layout macrotile;
+ unsigned pipe_config;
+ unsigned bankw;
+ unsigned bankh;
+ unsigned tile_split;
+ unsigned mtilea;
+ unsigned num_banks;
+ unsigned stride;
+ bool scanout;
+ } legacy;
+
+ struct {
+ /* surface flags */
+ unsigned swizzle_mode:5;
+ } gfx9;
+ } u;
/* Additional metadata associated with the buffer, in bytes.
* The maximum size is 64 * 4. This is opaque for the winsys & kernel.
RADEON_FID_R300_CMASK_ACCESS,
};
-#define RADEON_SURF_MAX_LEVEL 32
-
-#define RADEON_SURF_TYPE_MASK 0xFF
-#define RADEON_SURF_TYPE_SHIFT 0
-#define RADEON_SURF_TYPE_1D 0
-#define RADEON_SURF_TYPE_2D 1
-#define RADEON_SURF_TYPE_3D 2
-#define RADEON_SURF_TYPE_CUBEMAP 3
-#define RADEON_SURF_TYPE_1D_ARRAY 4
-#define RADEON_SURF_TYPE_2D_ARRAY 5
-#define RADEON_SURF_MODE_MASK 0xFF
-#define RADEON_SURF_MODE_SHIFT 8
-#define RADEON_SURF_MODE_LINEAR_ALIGNED 1
-#define RADEON_SURF_MODE_1D 2
-#define RADEON_SURF_MODE_2D 3
-#define RADEON_SURF_SCANOUT (1 << 16)
-#define RADEON_SURF_ZBUFFER (1 << 17)
-#define RADEON_SURF_SBUFFER (1 << 18)
-#define RADEON_SURF_Z_OR_SBUFFER (RADEON_SURF_ZBUFFER | RADEON_SURF_SBUFFER)
-#define RADEON_SURF_HAS_SBUFFER_MIPTREE (1 << 19)
-#define RADEON_SURF_HAS_TILE_MODE_INDEX (1 << 20)
-#define RADEON_SURF_FMASK (1 << 21)
-#define RADEON_SURF_DISABLE_DCC (1 << 22)
-
-#define RADEON_SURF_GET(v, field) (((v) >> RADEON_SURF_ ## field ## _SHIFT) & RADEON_SURF_ ## field ## _MASK)
-#define RADEON_SURF_SET(v, field) (((v) & RADEON_SURF_ ## field ## _MASK) << RADEON_SURF_ ## field ## _SHIFT)
-#define RADEON_SURF_CLR(v, field) ((v) & ~(RADEON_SURF_ ## field ## _MASK << RADEON_SURF_ ## field ## _SHIFT))
-
-struct radeon_surf_level {
- uint64_t offset;
- uint64_t slice_size;
- uint32_t npix_x;
- uint32_t npix_y;
- uint32_t npix_z;
- uint32_t nblk_x;
- uint32_t nblk_y;
- uint32_t nblk_z;
- uint32_t pitch_bytes;
- uint32_t mode;
- uint64_t dcc_offset;
- uint64_t dcc_fast_clear_size;
- bool dcc_enabled;
-};
-
-struct radeon_surf {
- /* These are inputs to the calculator. */
- uint32_t npix_x;
- uint32_t npix_y;
- uint32_t npix_z;
- uint32_t blk_w;
- uint32_t blk_h;
- uint32_t blk_d;
- uint32_t array_size;
- uint32_t last_level;
- uint32_t bpe;
- uint32_t nsamples;
- uint32_t flags;
-
- /* These are return values. Some of them can be set by the caller, but
- * they will be treated as hints (e.g. bankw, bankh) and might be
- * changed by the calculator.
- */
- uint64_t bo_size;
- uint64_t bo_alignment;
- /* This applies to EG and later. */
- uint32_t bankw;
- uint32_t bankh;
- uint32_t mtilea;
- uint32_t tile_split;
- uint32_t stencil_tile_split;
- struct radeon_surf_level level[RADEON_SURF_MAX_LEVEL];
- struct radeon_surf_level stencil_level[RADEON_SURF_MAX_LEVEL];
- uint32_t tiling_index[RADEON_SURF_MAX_LEVEL];
- uint32_t stencil_tiling_index[RADEON_SURF_MAX_LEVEL];
- uint32_t pipe_config;
- uint32_t num_banks;
- uint32_t macro_tile_index;
- uint32_t micro_tile_mode; /* displayable, thin, depth, rotated */
-
- /* Whether the depth miptree or stencil miptree as used by the DB are
- * adjusted from their TC compatible form to ensure depth/stencil
- * compatibility. If either is true, the corresponding plane cannot be
- * sampled from.
- */
- bool depth_adjusted;
- bool stencil_adjusted;
-
- uint64_t dcc_size;
- uint64_t dcc_alignment;
-};
-
struct radeon_bo_list_item {
uint64_t bo_size;
uint64_t vm_address;
unsigned slice_size,
struct winsys_handle *whandle);
+ /**
+ * Change the commitment of a (64KB-page aligned) region of the given
+ * sparse buffer.
+ *
+ * \warning There is no automatic synchronization with command submission.
+ *
+ * \note Only implemented by the amdgpu winsys.
+ *
+ * \return false on out of memory or other failure, true on success.
+ */
+ bool (*buffer_commit)(struct pb_buffer *buf,
+ uint64_t offset, uint64_t size,
+ bool commit);
+
/**
* Return the virtual address of a buffer.
*
+ * When virtual memory is not in use, this is the offset relative to the
+ * relocation base (non-zero for sub-allocated buffers).
+ *
* \param buf A winsys buffer object
* \return virtual address
*/
uint64_t (*buffer_get_virtual_address)(struct pb_buffer *buf);
+ /**
+ * Return the offset of this buffer relative to the relocation base.
+ * This is only non-zero for sub-allocated buffers.
+ *
+ * This is only supported in the radeon winsys, since amdgpu uses virtual
+ * addresses in submissions even for the video engines.
+ *
+ * \param buf A winsys buffer object
+ * \return the offset for relocations
+ */
+ unsigned (*buffer_get_reloc_offset)(struct pb_buffer *buf);
+
/**
* Query the initial placement of the buffer from the kernel driver.
*/
/**
* Return the index of an already-added buffer.
*
+ * Not supported on amdgpu. Drivers with GPUVM should not care about
+ * buffer indices.
+ *
* \param cs Command stream
* \param buf Buffer
* \return The buffer index, or -1 if the buffer has not been added.
*/
bool (*cs_check_space)(struct radeon_winsys_cs *cs, unsigned dw);
- /**
- * Return true if there is enough memory in VRAM and GTT for the buffers
- * added so far.
- *
- * \param cs A command stream to validate.
- * \param vram VRAM memory size pending to be use
- * \param gtt GTT memory size pending to be use
- */
- bool (*cs_memory_below_limit)(struct radeon_winsys_cs *cs,
- uint64_t vram, uint64_t gtt);
-
/**
* Return the buffer list.
*
+ * This is the buffer list as passed to the kernel, i.e. it only contains
+ * the parent buffers of sub-allocated buffers.
+ *
* \param cs Command stream
* \param list Returned buffer list. Set to NULL to query the count only.
* \return The buffer count.
unsigned flags,
struct pipe_fence_handle **fence);
+ /**
+ * Create a fence before the CS is flushed.
+ * The user must flush manually to complete the initializaton of the fence.
+ * The fence must not be used before the flush.
+ */
+ struct pipe_fence_handle *(*cs_get_next_fence)(struct radeon_winsys_cs *cs);
+
/**
* Return true if a buffer is referenced by a command stream.
*
* Initialize surface
*
* \param ws The winsys this function is called from.
- * \param surf Surface structure ptr
+ * \param tex Input texture description
+ * \param flags Bitmask of RADEON_SURF_* flags
+ * \param bpe Bytes per pixel, it can be different for Z buffers.
+ * \param mode Preferred tile mode. (linear, 1D, or 2D)
+ * \param surf Output structure
*/
int (*surface_init)(struct radeon_winsys *ws,
- struct radeon_surf *surf);
-
- /**
- * Find best values for a surface
- *
- * \param ws The winsys this function is called from.
- * \param surf Surface structure ptr
- */
- int (*surface_best)(struct radeon_winsys *ws,
+ const struct pipe_resource *tex,
+ unsigned flags, unsigned bpe,
+ enum radeon_surf_mode mode,
struct radeon_surf *surf);
uint64_t (*query_value)(struct radeon_winsys *ws,
projects / mesa.git / blobdiff