1 /* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */
4 * Copyright (C) 2012 Rob Clark <robclark@freedesktop.org>
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice (including the next
14 * paragraph) shall be included in all copies or substantial portions of the
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
26 * Rob Clark <robclark@freedesktop.org>
29 #ifndef FREEDRENO_UTIL_H_
30 #define FREEDRENO_UTIL_H_
32 #include <freedreno_drmif.h>
33 #include <freedreno_ringbuffer.h>
35 #include "pipe/p_format.h"
36 #include "pipe/p_state.h"
37 #include "util/u_debug.h"
38 #include "util/u_math.h"
39 #include "util/u_half.h"
40 #include "util/u_dynarray.h"
41 #include "util/u_pack_color.h"
44 #include "adreno_common.xml.h"
45 #include "adreno_pm4.xml.h"
47 enum adreno_rb_depth_format
fd_pipe2depth(enum pipe_format format
);
48 enum pc_di_index_size
fd_pipe2index(enum pipe_format format
);
49 enum adreno_rb_blend_factor
fd_blend_factor(unsigned factor
);
50 enum adreno_pa_su_sc_draw
fd_polygon_mode(unsigned mode
);
51 enum adreno_stencil_op
fd_stencil_op(unsigned op
);
53 #define A3XX_MAX_MIP_LEVELS 14
54 /* TBD if it is same on a2xx, but for now: */
55 #define MAX_MIP_LEVELS A3XX_MAX_MIP_LEVELS
57 #define A2XX_MAX_RENDER_TARGETS 1
58 #define A3XX_MAX_RENDER_TARGETS 4
59 #define A4XX_MAX_RENDER_TARGETS 8
60 #define A5XX_MAX_RENDER_TARGETS 8
62 #define MAX_RENDER_TARGETS A5XX_MAX_RENDER_TARGETS
64 #define FD_DBG_MSGS 0x0001
65 #define FD_DBG_DISASM 0x0002
66 #define FD_DBG_DCLEAR 0x0004
67 #define FD_DBG_DDRAW 0x0008
68 #define FD_DBG_NOSCIS 0x0010
69 #define FD_DBG_DIRECT 0x0020
70 #define FD_DBG_NOBYPASS 0x0040
71 #define FD_DBG_FRAGHALF 0x0080
72 #define FD_DBG_NOBIN 0x0100
73 #define FD_DBG_OPTMSGS 0x0200
74 #define FD_DBG_GLSL120 0x0400
75 #define FD_DBG_SHADERDB 0x0800
76 #define FD_DBG_FLUSH 0x1000
77 #define FD_DBG_DEQP 0x2000
78 #define FD_DBG_NIR 0x4000
79 #define FD_DBG_REORDER 0x8000
80 #define FD_DBG_BSTAT 0x10000
82 extern int fd_mesa_debug
;
83 extern bool fd_binning_enabled
;
85 #define DBG(fmt, ...) \
86 do { if (fd_mesa_debug & FD_DBG_MSGS) \
87 debug_printf("%s:%d: "fmt "\n", \
88 __FUNCTION__, __LINE__, ##__VA_ARGS__); } while (0)
90 /* for conditionally setting boolean flag(s): */
91 #define COND(bool, val) ((bool) ? (val) : 0)
93 #define CP_REG(reg) ((0x4 << 16) | ((unsigned int)((reg) - (0x2000))))
95 static inline uint32_t DRAW(enum pc_di_primtype prim_type
,
96 enum pc_di_src_sel source_select
, enum pc_di_index_size index_size
,
97 enum pc_di_vis_cull_mode vis_cull_mode
,
100 return (prim_type
<< 0) |
101 (source_select
<< 6) |
102 ((index_size
& 1) << 11) |
103 ((index_size
>> 1) << 13) |
104 (vis_cull_mode
<< 9) |
109 /* for tracking cmdstream positions that need to be patched: */
114 #define fd_patch_num_elements(buf) ((buf)->size / sizeof(struct fd_cs_patch))
115 #define fd_patch_element(buf, i) util_dynarray_element(buf, struct fd_cs_patch, i)
117 static inline enum pipe_format
118 pipe_surface_format(struct pipe_surface
*psurf
)
121 return PIPE_FORMAT_NONE
;
122 return psurf
->format
;
126 fd_surface_half_precision(const struct pipe_surface
*psurf
)
128 enum pipe_format format
;
133 format
= psurf
->format
;
135 /* colors are provided in consts, which go through cov.f32f16, which will
138 if (util_format_is_pure_integer(format
))
141 /* avoid losing precision on 32-bit float formats */
142 if (util_format_is_float(format
) &&
143 util_format_get_component_bits(format
, UTIL_FORMAT_COLORSPACE_RGB
, 0) == 32)
149 static inline unsigned
150 fd_sampler_first_level(const struct pipe_sampler_view
*view
)
152 if (view
->target
== PIPE_BUFFER
)
154 return view
->u
.tex
.first_level
;
157 static inline unsigned
158 fd_sampler_last_level(const struct pipe_sampler_view
*view
)
160 if (view
->target
== PIPE_BUFFER
)
162 return view
->u
.tex
.last_level
;
166 fd_half_precision(struct pipe_framebuffer_state
*pfb
)
170 for (i
= 0; i
< pfb
->nr_cbufs
; i
++)
171 if (!fd_surface_half_precision(pfb
->cbufs
[i
]))
179 static inline void emit_marker(struct fd_ringbuffer
*ring
, int scratch_idx
);
180 static inline void emit_marker5(struct fd_ringbuffer
*ring
, int scratch_idx
);
183 OUT_RING(struct fd_ringbuffer
*ring
, uint32_t data
)
186 DBG("ring[%p]: OUT_RING %04x: %08x", ring
,
187 (uint32_t)(ring
->cur
- ring
->last_start
), data
);
189 fd_ringbuffer_emit(ring
, data
);
192 /* like OUT_RING() but appends a cmdstream patch point to 'buf' */
194 OUT_RINGP(struct fd_ringbuffer
*ring
, uint32_t data
,
195 struct util_dynarray
*buf
)
198 DBG("ring[%p]: OUT_RINGP %04x: %08x", ring
,
199 (uint32_t)(ring
->cur
- ring
->last_start
), data
);
201 util_dynarray_append(buf
, struct fd_cs_patch
, ((struct fd_cs_patch
){
208 * NOTE: OUT_RELOC*() is 2 dwords (64b) on a5xx+
212 OUT_RELOC(struct fd_ringbuffer
*ring
, struct fd_bo
*bo
,
213 uint32_t offset
, uint64_t or, int32_t shift
)
216 DBG("ring[%p]: OUT_RELOC %04x: %p+%u << %d", ring
,
217 (uint32_t)(ring
->cur
- ring
->last_start
), bo
, offset
, shift
);
219 debug_assert(offset
< fd_bo_size(bo
));
220 fd_ringbuffer_reloc2(ring
, &(struct fd_reloc
){
222 .flags
= FD_RELOC_READ
,
231 OUT_RELOCW(struct fd_ringbuffer
*ring
, struct fd_bo
*bo
,
232 uint32_t offset
, uint64_t or, int32_t shift
)
235 DBG("ring[%p]: OUT_RELOCW %04x: %p+%u << %d", ring
,
236 (uint32_t)(ring
->cur
- ring
->last_start
), bo
, offset
, shift
);
238 debug_assert(offset
< fd_bo_size(bo
));
239 fd_ringbuffer_reloc2(ring
, &(struct fd_reloc
){
241 .flags
= FD_RELOC_READ
| FD_RELOC_WRITE
,
249 static inline void BEGIN_RING(struct fd_ringbuffer
*ring
, uint32_t ndwords
)
251 if (ring
->cur
+ ndwords
>= ring
->end
)
252 fd_ringbuffer_grow(ring
, ndwords
);
255 static inline uint32_t
256 __gpu_id(struct fd_ringbuffer
*ring
)
259 fd_pipe_get_param(ring
->pipe
, FD_GPU_ID
, &val
);
264 OUT_PKT0(struct fd_ringbuffer
*ring
, uint16_t regindx
, uint16_t cnt
)
266 debug_assert(__gpu_id(ring
) < 500);
267 BEGIN_RING(ring
, cnt
+1);
268 OUT_RING(ring
, CP_TYPE0_PKT
| ((cnt
-1) << 16) | (regindx
& 0x7FFF));
272 OUT_PKT2(struct fd_ringbuffer
*ring
)
274 debug_assert(__gpu_id(ring
) < 500);
276 OUT_RING(ring
, CP_TYPE2_PKT
);
280 OUT_PKT3(struct fd_ringbuffer
*ring
, uint8_t opcode
, uint16_t cnt
)
282 debug_assert(__gpu_id(ring
) < 500);
283 BEGIN_RING(ring
, cnt
+1);
284 OUT_RING(ring
, CP_TYPE3_PKT
| ((cnt
-1) << 16) | ((opcode
& 0xFF) << 8));
288 * Starting with a5xx, pkt4/pkt7 are used instead of pkt0/pkt3
291 static inline unsigned
292 _odd_parity_bit(unsigned val
)
294 /* See: http://graphics.stanford.edu/~seander/bithacks.html#ParityParallel
295 * note that we want odd parity so 0x6996 is inverted.
301 return (~0x6996 >> val
) & 1;
305 OUT_PKT4(struct fd_ringbuffer
*ring
, uint16_t regindx
, uint16_t cnt
)
307 BEGIN_RING(ring
, cnt
+1);
308 OUT_RING(ring
, CP_TYPE4_PKT
| cnt
|
309 (_odd_parity_bit(cnt
) << 7) |
310 ((regindx
& 0x3ffff) << 8) |
311 ((_odd_parity_bit(regindx
) << 27)));
315 OUT_PKT7(struct fd_ringbuffer
*ring
, uint8_t opcode
, uint16_t cnt
)
317 BEGIN_RING(ring
, cnt
+1);
318 OUT_RING(ring
, CP_TYPE7_PKT
| cnt
|
319 (_odd_parity_bit(cnt
) << 15) |
320 ((opcode
& 0x7f) << 16) |
321 ((_odd_parity_bit(opcode
) << 23)));
325 OUT_WFI(struct fd_ringbuffer
*ring
)
327 OUT_PKT3(ring
, CP_WAIT_FOR_IDLE
, 1);
328 OUT_RING(ring
, 0x00000000);
332 OUT_WFI5(struct fd_ringbuffer
*ring
)
334 OUT_PKT7(ring
, CP_WAIT_FOR_IDLE
, 0);
338 __OUT_IB(struct fd_ringbuffer
*ring
, bool prefetch
, struct fd_ringbuffer
*target
)
340 unsigned count
= fd_ringbuffer_cmd_count(target
);
342 debug_assert(__gpu_id(ring
) < 500);
344 /* for debug after a lock up, write a unique counter value
345 * to scratch6 for each IB, to make it easier to match up
346 * register dumps to cmdstream. The combination of IB and
347 * DRAW (scratch7) is enough to "triangulate" the particular
348 * draw that caused lockup.
350 emit_marker(ring
, 6);
352 for (unsigned i
= 0; i
< count
; i
++) {
354 OUT_PKT3(ring
, prefetch
? CP_INDIRECT_BUFFER_PFE
: CP_INDIRECT_BUFFER_PFD
, 2);
355 dwords
= fd_ringbuffer_emit_reloc_ring_full(ring
, target
, i
) / 4;
357 OUT_RING(ring
, dwords
);
361 emit_marker(ring
, 6);
365 __OUT_IB5(struct fd_ringbuffer
*ring
, struct fd_ringbuffer
*target
)
367 unsigned count
= fd_ringbuffer_cmd_count(target
);
369 /* for debug after a lock up, write a unique counter value
370 * to scratch6 for each IB, to make it easier to match up
371 * register dumps to cmdstream. The combination of IB and
372 * DRAW (scratch7) is enough to "triangulate" the particular
373 * draw that caused lockup.
375 emit_marker5(ring
, 6);
377 for (unsigned i
= 0; i
< count
; i
++) {
379 OUT_PKT7(ring
, CP_INDIRECT_BUFFER
, 3);
380 dwords
= fd_ringbuffer_emit_reloc_ring_full(ring
, target
, i
) / 4;
382 OUT_RING(ring
, dwords
);
385 emit_marker5(ring
, 6);
388 /* CP_SCRATCH_REG4 is used to hold base address for query results: */
389 // XXX annoyingly scratch regs move on a5xx.. and additionally different
390 // packet types.. so freedreno_query_hw is going to need a bit of
392 #define HW_QUERY_BASE_REG REG_AXXX_CP_SCRATCH_REG4
395 emit_marker(struct fd_ringbuffer
*ring
, int scratch_idx
)
397 extern unsigned marker_cnt
;
398 unsigned reg
= REG_AXXX_CP_SCRATCH_REG0
+ scratch_idx
;
399 assert(reg
!= HW_QUERY_BASE_REG
);
400 if (reg
== HW_QUERY_BASE_REG
)
402 OUT_PKT0(ring
, reg
, 1);
403 OUT_RING(ring
, ++marker_cnt
);
407 emit_marker5(struct fd_ringbuffer
*ring
, int scratch_idx
)
409 extern unsigned marker_cnt
;
410 //XXX unsigned reg = REG_A5XX_CP_SCRATCH_REG(scratch_idx);
411 unsigned reg
= 0x00000b78 + scratch_idx
;
412 assert(reg
!= HW_QUERY_BASE_REG
);
413 if (reg
== HW_QUERY_BASE_REG
)
416 OUT_PKT4(ring
, reg
, 1);
417 OUT_RING(ring
, ++marker_cnt
);
421 /* helper to get numeric value from environment variable.. mostly
422 * just leaving this here because it is helpful to brute-force figure
423 * out unknown formats, etc, which blob driver does not support:
425 static inline uint32_t env2u(const char *envvar
)
427 char *str
= getenv(envvar
);
429 return strtoul(str
, NULL
, 0);
433 static inline uint32_t
434 pack_rgba(enum pipe_format format
, const float *rgba
)
437 util_pack_color(rgba
, format
, &uc
);
442 * swap - swap value of @a and @b
445 do { __typeof(a) __tmp = (a); (a) = (b); (b) = __tmp; } while (0)
447 #define foreach_bit(b, mask) \
448 for (uint32_t _m = (mask); _m && ({(b) = u_bit_scan(&_m); 1;});)
450 #endif /* FREEDRENO_UTIL_H_ */