enum adreno_rb_depth_format fd_pipe2depth(enum pipe_format format);
enum pc_di_index_size fd_pipe2index(enum pipe_format format);
+enum pipe_format fd_gmem_restore_format(enum pipe_format format);
enum adreno_rb_blend_factor fd_blend_factor(unsigned factor);
enum adreno_pa_su_sc_draw fd_polygon_mode(unsigned mode);
enum adreno_stencil_op fd_stencil_op(unsigned op);
#define A2XX_MAX_RENDER_TARGETS 1
#define A3XX_MAX_RENDER_TARGETS 4
#define A4XX_MAX_RENDER_TARGETS 8
+#define A5XX_MAX_RENDER_TARGETS 8
+#define A6XX_MAX_RENDER_TARGETS 8
-#define MAX_RENDER_TARGETS A4XX_MAX_RENDER_TARGETS
+#define MAX_RENDER_TARGETS A6XX_MAX_RENDER_TARGETS
#define FD_DBG_MSGS 0x0001
#define FD_DBG_DISASM 0x0002
#define FD_DBG_SHADERDB 0x0800
#define FD_DBG_FLUSH 0x1000
#define FD_DBG_DEQP 0x2000
-#define FD_DBG_NIR 0x4000
+#define FD_DBG_INORDER 0x4000
+#define FD_DBG_BSTAT 0x8000
+#define FD_DBG_NOGROW 0x10000
+#define FD_DBG_LRZ 0x20000
+#define FD_DBG_NOINDR 0x40000
+#define FD_DBG_NOBLIT 0x80000
+#define FD_DBG_HIPRIO 0x100000
+#define FD_DBG_TTILE 0x200000
+#define FD_DBG_PERFC 0x400000
extern int fd_mesa_debug;
extern bool fd_binning_enabled;
(instances << 24);
}
+static inline uint32_t DRAW_A20X(enum pc_di_primtype prim_type,
+ enum pc_di_src_sel source_select, enum pc_di_index_size index_size,
+ enum pc_di_vis_cull_mode vis_cull_mode,
+ uint16_t count)
+{
+ return (prim_type << 0) |
+ (source_select << 6) |
+ ((index_size & 1) << 11) |
+ ((index_size >> 1) << 13) |
+ (vis_cull_mode << 9) |
+ (count << 16);
+}
+
/* for tracking cmdstream positions that need to be patched: */
struct fd_cs_patch {
uint32_t *cs;
}));
}
+/*
+ * NOTE: OUT_RELOC*() is 2 dwords (64b) on a5xx+
+ */
+
static inline void
OUT_RELOC(struct fd_ringbuffer *ring, struct fd_bo *bo,
- uint32_t offset, uint32_t or, int32_t shift)
+ uint32_t offset, uint64_t or, int32_t shift)
{
if (LOG_DWORDS) {
DBG("ring[%p]: OUT_RELOC %04x: %p+%u << %d", ring,
(uint32_t)(ring->cur - ring->last_start), bo, offset, shift);
}
debug_assert(offset < fd_bo_size(bo));
- fd_ringbuffer_reloc(ring, &(struct fd_reloc){
+ fd_ringbuffer_reloc2(ring, &(struct fd_reloc){
.bo = bo,
.flags = FD_RELOC_READ,
.offset = offset,
.or = or,
.shift = shift,
+ .orhi = or >> 32,
});
}
static inline void
OUT_RELOCW(struct fd_ringbuffer *ring, struct fd_bo *bo,
- uint32_t offset, uint32_t or, int32_t shift)
+ uint32_t offset, uint64_t or, int32_t shift)
{
if (LOG_DWORDS) {
DBG("ring[%p]: OUT_RELOCW %04x: %p+%u << %d", ring,
(uint32_t)(ring->cur - ring->last_start), bo, offset, shift);
}
debug_assert(offset < fd_bo_size(bo));
- fd_ringbuffer_reloc(ring, &(struct fd_reloc){
+ fd_ringbuffer_reloc2(ring, &(struct fd_reloc){
.bo = bo,
.flags = FD_RELOC_READ | FD_RELOC_WRITE,
.offset = offset,
.or = or,
.shift = shift,
+ .orhi = or >> 32,
});
}
+static inline void
+OUT_RB(struct fd_ringbuffer *ring, struct fd_ringbuffer *target)
+{
+ fd_ringbuffer_emit_reloc_ring_full(ring, target, 0);
+}
+
static inline void BEGIN_RING(struct fd_ringbuffer *ring, uint32_t ndwords)
{
if (ring->cur + ndwords >= ring->end)
fd_ringbuffer_grow(ring, ndwords);
}
+static inline uint32_t
+__gpu_id(struct fd_ringbuffer *ring)
+{
+ uint64_t val;
+ fd_pipe_get_param(ring->pipe, FD_GPU_ID, &val);
+ return val;
+}
+
static inline void
OUT_PKT0(struct fd_ringbuffer *ring, uint16_t regindx, uint16_t cnt)
{
+ debug_assert(__gpu_id(ring) < 500);
BEGIN_RING(ring, cnt+1);
OUT_RING(ring, CP_TYPE0_PKT | ((cnt-1) << 16) | (regindx & 0x7FFF));
}
static inline void
OUT_PKT2(struct fd_ringbuffer *ring)
{
+ debug_assert(__gpu_id(ring) < 500);
BEGIN_RING(ring, 1);
OUT_RING(ring, CP_TYPE2_PKT);
}
static inline void
OUT_PKT3(struct fd_ringbuffer *ring, uint8_t opcode, uint16_t cnt)
{
+ debug_assert(__gpu_id(ring) < 500);
BEGIN_RING(ring, cnt+1);
OUT_RING(ring, CP_TYPE3_PKT | ((cnt-1) << 16) | ((opcode & 0xFF) << 8));
}
+/*
+ * Starting with a5xx, pkt4/pkt7 are used instead of pkt0/pkt3
+ */
+
+static inline unsigned
+_odd_parity_bit(unsigned val)
+{
+ /* See: http://graphics.stanford.edu/~seander/bithacks.html#ParityParallel
+ * note that we want odd parity so 0x6996 is inverted.
+ */
+ val ^= val >> 16;
+ val ^= val >> 8;
+ val ^= val >> 4;
+ val &= 0xf;
+ return (~0x6996 >> val) & 1;
+}
+
+static inline void
+OUT_PKT4(struct fd_ringbuffer *ring, uint16_t regindx, uint16_t cnt)
+{
+ BEGIN_RING(ring, cnt+1);
+ OUT_RING(ring, CP_TYPE4_PKT | cnt |
+ (_odd_parity_bit(cnt) << 7) |
+ ((regindx & 0x3ffff) << 8) |
+ ((_odd_parity_bit(regindx) << 27)));
+}
+
+static inline void
+OUT_PKT7(struct fd_ringbuffer *ring, uint8_t opcode, uint16_t cnt)
+{
+ BEGIN_RING(ring, cnt+1);
+ OUT_RING(ring, CP_TYPE7_PKT | cnt |
+ (_odd_parity_bit(cnt) << 15) |
+ ((opcode & 0x7f) << 16) |
+ ((_odd_parity_bit(opcode) << 23)));
+}
+
static inline void
OUT_WFI(struct fd_ringbuffer *ring)
{
OUT_RING(ring, 0x00000000);
}
+static inline void
+OUT_WFI5(struct fd_ringbuffer *ring)
+{
+ OUT_PKT7(ring, CP_WAIT_FOR_IDLE, 0);
+}
+
static inline void
__OUT_IB(struct fd_ringbuffer *ring, bool prefetch, struct fd_ringbuffer *target)
{
+ if (target->cur == target->start)
+ return;
+
unsigned count = fd_ringbuffer_cmd_count(target);
+ debug_assert(__gpu_id(ring) < 500);
+
/* for debug after a lock up, write a unique counter value
* to scratch6 for each IB, to make it easier to match up
* register dumps to cmdstream. The combination of IB and
emit_marker(ring, 6);
}
+static inline void
+__OUT_IB5(struct fd_ringbuffer *ring, struct fd_ringbuffer *target)
+{
+ if (target->cur == target->start)
+ return;
+
+ unsigned count = fd_ringbuffer_cmd_count(target);
+
+ for (unsigned i = 0; i < count; i++) {
+ uint32_t dwords;
+ OUT_PKT7(ring, CP_INDIRECT_BUFFER, 3);
+ dwords = fd_ringbuffer_emit_reloc_ring_full(ring, target, i) / 4;
+ assert(dwords > 0);
+ OUT_RING(ring, dwords);
+ }
+}
+
/* CP_SCRATCH_REG4 is used to hold base address for query results: */
+// XXX annoyingly scratch regs move on a5xx.. and additionally different
+// packet types.. so freedreno_query_hw is going to need a bit of
+// rework..
#define HW_QUERY_BASE_REG REG_AXXX_CP_SCRATCH_REG4
static inline void
return uc.ui[0];
}
+/*
+ * swap - swap value of @a and @b
+ */
+#define swap(a, b) \
+ do { __typeof(a) __tmp = (a); (a) = (b); (b) = __tmp; } while (0)
+
+#define foreach_bit(b, mask) \
+ for (uint32_t _m = (mask); _m && ({(b) = u_bit_scan(&_m); 1;});)
+
+
+#define BIT(bit) (1u << bit)
+
+/*
+ * a3xx+ helpers:
+ */
+
+static inline enum a3xx_msaa_samples
+fd_msaa_samples(unsigned samples)
+{
+ switch (samples) {
+ default:
+ debug_assert(0);
+ case 1: return MSAA_ONE;
+ case 2: return MSAA_TWO;
+ case 4: return MSAA_FOUR;
+ }
+}
+
+/*
+ * a4xx+ helpers:
+ */
+
+static inline enum a4xx_state_block
+fd4_stage2shadersb(enum shader_t type)
+{
+ switch (type) {
+ case SHADER_VERTEX:
+ return SB4_VS_SHADER;
+ case SHADER_FRAGMENT:
+ return SB4_FS_SHADER;
+ case SHADER_COMPUTE:
+ return SB4_CS_SHADER;
+ default:
+ unreachable("bad shader type");
+ return ~0;
+ }
+}
+
#endif /* FREEDRENO_UTIL_H_ */
projects / mesa.git / blobdiff