/* The top 3 bits specify how the bits of each component are interpreted. */
+/* e.g. ETC2_RGB8 */
+#define MALI_FORMAT_COMPRESSED (0 << 5)
+
/* e.g. R11F_G11F_B10F */
#define MALI_FORMAT_SPECIAL (2 << 5)
#define MALI_CHANNEL_FLOAT 7
enum mali_format {
+ MALI_ETC2_RGB8 = MALI_FORMAT_COMPRESSED | 0x1,
+ MALI_ETC2_R11_UNORM = MALI_FORMAT_COMPRESSED | 0x2,
+ MALI_ETC2_RGBA8 = MALI_FORMAT_COMPRESSED | 0x3,
+ MALI_ETC2_RG11_UNORM = MALI_FORMAT_COMPRESSED | 0x4,
+ MALI_ETC2_R11_SNORM = MALI_FORMAT_COMPRESSED | 0x11,
+ MALI_ETC2_RG11_SNORM = MALI_FORMAT_COMPRESSED | 0x12,
+ MALI_ETC2_RGB8A1 = MALI_FORMAT_COMPRESSED | 0x13,
+ MALI_ASTC_SRGB_SUPP = MALI_FORMAT_COMPRESSED | 0x16,
+ MALI_ASTC_HDR_SUPP = MALI_FORMAT_COMPRESSED | 0x17,
+
MALI_RGB565 = MALI_FORMAT_SPECIAL | 0x0,
MALI_RGB5_A1_UNORM = MALI_FORMAT_SPECIAL | 0x2,
MALI_RGB10_A2_UNORM = MALI_FORMAT_SPECIAL | 0x3,
#define MALI_ALPHA_COVERAGE(clampf) ((uint16_t) (int) (clampf * 15.0f))
#define MALI_GET_ALPHA_COVERAGE(nibble) ((float) nibble / 15.0f)
-/* Applies to midgard1.flags */
+/* Applies to midgard1.flags_lo */
+
+/* Should be set when the fragment shader updates the depth value. */
+#define MALI_WRITES_Z (1 << 4)
/* Should the hardware perform early-Z testing? Normally should be set
* for performance reasons. Clear if you use: discard,
#define MALI_READS_ZS (1 << 8)
#define MALI_READS_TILEBUFFER (1 << 12)
+/* Applies to midgard1.flags_hi */
+
+/* Should be set when the fragment shader updates the stencil value. */
+#define MALI_WRITES_S (1 << 2)
+
/* The raw Midgard blend payload can either be an equation or a shader
* address, depending on the context */
} bifrost1;
struct {
unsigned uniform_buffer_count : 4;
- unsigned flags : 12;
+ unsigned flags_lo : 12;
/* vec4 units */
unsigned work_count : 5;
unsigned uniform_count : 5;
- unsigned unknown2 : 6;
+ unsigned flags_hi : 6;
} midgard1;
};
u64 zero8;
} __attribute__((packed));
-struct bifrost_scratchpad {
- u32 zero;
- u32 flags; // = 0x1f
- /* This is a pointer to a CPU-inaccessible buffer, 16 pages, allocated
- * during startup. It seems to serve the same purpose as the
- * gpu_scratchpad in the SFBD for Midgard, although it's slightly
- * larger.
- */
- mali_ptr gpu_scratchpad;
-} __attribute__((packed));
-
struct mali_vertex_tiler_postfix {
/* Zero for vertex jobs. Pointer to the position (gl_Position) varying
* output from the vertex shader for tiler jobs.
u64 viewport;
u64 occlusion_counter; /* A single bit as far as I can tell */
- /* Note: on Bifrost, this isn't actually the FBD. It points to
- * bifrost_scratchpad instead. However, it does point to the same thing
- * in vertex and tiler jobs.
- */
- mali_ptr framebuffer;
+ /* On Bifrost, this points directly to a mali_shared_memory structure.
+ * On Midgard, this points to a framebuffer (either SFBD or MFBD as
+ * tagged), which embeds a mali_shared_memory structure */
+ mali_ptr shared_memory;
} __attribute__((packed));
struct midgard_payload_vertex_tiler {
unsigned unk3 : 4;
};
-struct mali_single_framebuffer {
- u32 unknown1;
- u32 unknown2;
+/* Shared structure at the start of framebuffer descriptors, or used bare for
+ * compute jobs, configuring stack and shared memory */
+
+struct mali_shared_memory {
+ u32 stack_shift : 4;
+ u32 unk0 : 28;
+
+ /* Configuration for shared memory for compute shaders.
+ * shared_workgroup_count is logarithmic and may be computed for a
+ * compute shader using shared memory as:
+ *
+ * shared_workgroup_count = MAX2(ceil(log2(count_x)) + ... + ceil(log2(count_z), 10)
+ *
+ * For compute shaders that don't use shared memory, or non-compute
+ * shaders, this is set to ~0
+ */
+
+ u32 shared_workgroup_count : 5;
+ u32 shared_unk1 : 3;
+ u32 shared_shift : 4;
+ u32 shared_zero : 20;
+
mali_ptr scratchpad;
- u64 zero1;
- u64 zero0;
+ /* For compute shaders, the RAM backing of workgroup-shared memory. For
+ * fragment shaders on Bifrost, apparently multisampling locations */
+
+ mali_ptr shared_memory;
+ mali_ptr unknown1;
+} __attribute__((packed));
+
+
+struct mali_single_framebuffer {
+ struct mali_shared_memory shared_memory;
struct mali_sfbd_format format;
u32 clear_flags;
/* More below this, maybe */
} __attribute__((packed));
-/* On Midgard, this "framebuffer descriptor" is used for the framebuffer field
- * of compute jobs. Superficially resembles a single framebuffer descriptor */
-
-struct mali_compute_fbd {
- u32 unknown1[8];
-} __attribute__((packed));
-
/* Format bits for the render target flags */
#define MALI_MFBD_FORMAT_MSAA (1 << 1)
#define MALI_MFBD_EXTRA (1 << 13)
struct bifrost_framebuffer {
- u32 stack_shift : 4;
- u32 unk0 : 28;
-
- u32 unknown2; // = 0x1f, same as SFBD
- mali_ptr scratchpad;
+ struct mali_shared_memory shared_memory;
- /* 0x10 */
- mali_ptr sample_locations;
- mali_ptr unknown1;
/* 0x20 */
u16 width1, height1;
u32 zero3;