1/*
2 * Copyright (C) 2012 Rob Clark <robclark@freedesktop.org>
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21 * SOFTWARE.
22 *
23 * Authors:
24 *    Rob Clark <robclark@freedesktop.org>
25 */
26
27#ifndef FREEDRENO_UTIL_H_
28#define FREEDRENO_UTIL_H_
29
30#include "drm/freedreno_drmif.h"
31#include "drm/freedreno_ringbuffer.h"
32
33#include "pipe/p_format.h"
34#include "pipe/p_state.h"
35#include "util/u_debug.h"
36#include "util/u_math.h"
37#include "util/u_half.h"
38#include "util/u_dynarray.h"
39#include "util/u_pack_color.h"
40
41#include "disasm.h"
42#include "adreno_common.xml.h"
43#include "adreno_pm4.xml.h"
44
45enum adreno_rb_depth_format fd_pipe2depth(enum pipe_format format);
46enum pc_di_index_size fd_pipe2index(enum pipe_format format);
47enum pipe_format fd_gmem_restore_format(enum pipe_format format);
48enum adreno_rb_blend_factor fd_blend_factor(unsigned factor);
49enum adreno_pa_su_sc_draw fd_polygon_mode(unsigned mode);
50enum adreno_stencil_op fd_stencil_op(unsigned op);
51
52#define A3XX_MAX_MIP_LEVELS 14
53/* TBD if it is same on a2xx, but for now: */
54#define MAX_MIP_LEVELS A3XX_MAX_MIP_LEVELS
55
56#define A2XX_MAX_RENDER_TARGETS 1
57#define A3XX_MAX_RENDER_TARGETS 4
58#define A4XX_MAX_RENDER_TARGETS 8
59#define A5XX_MAX_RENDER_TARGETS 8
60#define A6XX_MAX_RENDER_TARGETS 8
61
62#define MAX_RENDER_TARGETS A6XX_MAX_RENDER_TARGETS
63
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/* unused 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_INORDER  0x4000
79#define FD_DBG_BSTAT    0x8000
80#define FD_DBG_NOGROW  0x10000
81#define FD_DBG_LRZ     0x20000
82#define FD_DBG_NOINDR  0x40000
83#define FD_DBG_NOBLIT  0x80000
84#define FD_DBG_HIPRIO 0x100000
85#define FD_DBG_TTILE  0x200000
86#define FD_DBG_PERFC  0x400000
87#define FD_DBG_SOFTPIN 0x800000
88#define FD_DBG_UBWC  0x1000000
89extern int fd_mesa_debug;
90extern bool fd_binning_enabled;
91
92#define DBG(fmt, ...) \
93		do { if (fd_mesa_debug & FD_DBG_MSGS) \
94			debug_printf("%s:%d: "fmt "\n", \
95				__FUNCTION__, __LINE__, ##__VA_ARGS__); } while (0)
96
97/* for conditionally setting boolean flag(s): */
98#define COND(bool, val) ((bool) ? (val) : 0)
99
100#define CP_REG(reg) ((0x4 << 16) | ((unsigned int)((reg) - (0x2000))))
101
102static inline uint32_t DRAW(enum pc_di_primtype prim_type,
103		enum pc_di_src_sel source_select, enum pc_di_index_size index_size,
104		enum pc_di_vis_cull_mode vis_cull_mode,
105		uint8_t instances)
106{
107	return (prim_type         << 0) |
108			(source_select     << 6) |
109			((index_size & 1)  << 11) |
110			((index_size >> 1) << 13) |
111			(vis_cull_mode     << 9) |
112			(1                 << 14) |
113			(instances         << 24);
114}
115
116static inline uint32_t DRAW_A20X(enum pc_di_primtype prim_type,
117		enum pc_di_face_cull_sel faceness_cull_select,
118		enum pc_di_src_sel source_select, enum pc_di_index_size index_size,
119		bool pre_fetch_cull_enable,
120		bool grp_cull_enable,
121		uint16_t count)
122{
123	return (prim_type         << 0) |
124			(source_select     << 6) |
125			(faceness_cull_select << 8) |
126			((index_size & 1)  << 11) |
127			((index_size >> 1) << 13) |
128			(pre_fetch_cull_enable << 14) |
129			(grp_cull_enable << 15) |
130			(count         << 16);
131}
132
133/* for tracking cmdstream positions that need to be patched: */
134struct fd_cs_patch {
135	uint32_t *cs;
136	uint32_t val;
137};
138#define fd_patch_num_elements(buf) ((buf)->size / sizeof(struct fd_cs_patch))
139#define fd_patch_element(buf, i)   util_dynarray_element(buf, struct fd_cs_patch, i)
140
141static inline enum pipe_format
142pipe_surface_format(struct pipe_surface *psurf)
143{
144	if (!psurf)
145		return PIPE_FORMAT_NONE;
146	return psurf->format;
147}
148
149static inline bool
150fd_surface_half_precision(const struct pipe_surface *psurf)
151{
152	enum pipe_format format;
153
154	if (!psurf)
155		return true;
156
157	format = psurf->format;
158
159	/* colors are provided in consts, which go through cov.f32f16, which will
160	 * break these values
161	 */
162	if (util_format_is_pure_integer(format))
163		return false;
164
165	/* avoid losing precision on 32-bit float formats */
166	if (util_format_is_float(format) &&
167		util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 0) == 32)
168		return false;
169
170	return true;
171}
172
173static inline unsigned
174fd_sampler_first_level(const struct pipe_sampler_view *view)
175{
176	if (view->target == PIPE_BUFFER)
177		return 0;
178	return view->u.tex.first_level;
179}
180
181static inline unsigned
182fd_sampler_last_level(const struct pipe_sampler_view *view)
183{
184	if (view->target == PIPE_BUFFER)
185		return 0;
186	return view->u.tex.last_level;
187}
188
189static inline bool
190fd_half_precision(struct pipe_framebuffer_state *pfb)
191{
192	unsigned i;
193
194	for (i = 0; i < pfb->nr_cbufs; i++)
195		if (!fd_surface_half_precision(pfb->cbufs[i]))
196			return false;
197
198	return true;
199}
200
201/* Note sure if this is same on all gens, but seems to be same on the later
202 * gen's
203 */
204static inline unsigned
205fd_calc_guardband(unsigned x)
206{
207	float l = log2(x);
208	if (l <= 8)
209		return 511;
210	return 511 - ((l - 8) * 65);
211}
212
213#define LOG_DWORDS 0
214
215static inline void emit_marker(struct fd_ringbuffer *ring, int scratch_idx);
216
217static inline void
218OUT_RING(struct fd_ringbuffer *ring, uint32_t data)
219{
220	if (LOG_DWORDS) {
221		DBG("ring[%p]: OUT_RING   %04x:  %08x", ring,
222				(uint32_t)(ring->cur - ring->start), data);
223	}
224	fd_ringbuffer_emit(ring, data);
225}
226
227/* like OUT_RING() but appends a cmdstream patch point to 'buf' */
228static inline void
229OUT_RINGP(struct fd_ringbuffer *ring, uint32_t data,
230		struct util_dynarray *buf)
231{
232	if (LOG_DWORDS) {
233		DBG("ring[%p]: OUT_RINGP  %04x:  %08x", ring,
234				(uint32_t)(ring->cur - ring->start), data);
235	}
236	util_dynarray_append(buf, struct fd_cs_patch, ((struct fd_cs_patch){
237		.cs  = ring->cur++,
238		.val = data,
239	}));
240}
241
242/*
243 * NOTE: OUT_RELOC*() is 2 dwords (64b) on a5xx+
244 */
245
246static inline void
247__out_reloc(struct fd_ringbuffer *ring, struct fd_bo *bo,
248		uint32_t offset, uint64_t or, int32_t shift, uint32_t flags)
249{
250	if (LOG_DWORDS) {
251		DBG("ring[%p]: OUT_RELOC   %04x:  %p+%u << %d", ring,
252				(uint32_t)(ring->cur - ring->start), bo, offset, shift);
253	}
254	debug_assert(offset < fd_bo_size(bo));
255	fd_ringbuffer_reloc(ring, &(struct fd_reloc){
256		.bo = bo,
257		.flags = flags,
258		.offset = offset,
259		.or = or,
260		.shift = shift,
261		.orhi = or >> 32,
262	});
263}
264
265static inline void
266OUT_RELOC(struct fd_ringbuffer *ring, struct fd_bo *bo,
267		uint32_t offset, uint64_t or, int32_t shift)
268{
269	__out_reloc(ring, bo, offset, or, shift, FD_RELOC_READ);
270}
271
272static inline void
273OUT_RELOCW(struct fd_ringbuffer *ring, struct fd_bo *bo,
274		uint32_t offset, uint64_t or, int32_t shift)
275{
276	__out_reloc(ring, bo, offset, or, shift, FD_RELOC_READ | FD_RELOC_WRITE);
277}
278
279static inline void
280OUT_RELOCD(struct fd_ringbuffer *ring, struct fd_bo *bo,
281		uint32_t offset, uint64_t or, int32_t shift)
282{
283	__out_reloc(ring, bo, offset, or, shift, FD_RELOC_READ | FD_RELOC_DUMP);
284}
285
286static inline void
287OUT_RB(struct fd_ringbuffer *ring, struct fd_ringbuffer *target)
288{
289	fd_ringbuffer_emit_reloc_ring_full(ring, target, 0);
290}
291
292static inline void BEGIN_RING(struct fd_ringbuffer *ring, uint32_t ndwords)
293{
294	if (ring->cur + ndwords > ring->end)
295		fd_ringbuffer_grow(ring, ndwords);
296}
297
298static inline void
299OUT_PKT0(struct fd_ringbuffer *ring, uint16_t regindx, uint16_t cnt)
300{
301	BEGIN_RING(ring, cnt+1);
302	OUT_RING(ring, CP_TYPE0_PKT | ((cnt-1) << 16) | (regindx & 0x7FFF));
303}
304
305static inline void
306OUT_PKT2(struct fd_ringbuffer *ring)
307{
308	BEGIN_RING(ring, 1);
309	OUT_RING(ring, CP_TYPE2_PKT);
310}
311
312static inline void
313OUT_PKT3(struct fd_ringbuffer *ring, uint8_t opcode, uint16_t cnt)
314{
315	BEGIN_RING(ring, cnt+1);
316	OUT_RING(ring, CP_TYPE3_PKT | ((cnt-1) << 16) | ((opcode & 0xFF) << 8));
317}
318
319/*
320 * Starting with a5xx, pkt4/pkt7 are used instead of pkt0/pkt3
321 */
322
323static inline unsigned
324_odd_parity_bit(unsigned val)
325{
326	/* See: http://graphics.stanford.edu/~seander/bithacks.html#ParityParallel
327	 * note that we want odd parity so 0x6996 is inverted.
328	 */
329	val ^= val >> 16;
330	val ^= val >> 8;
331	val ^= val >> 4;
332	val &= 0xf;
333	return (~0x6996 >> val) & 1;
334}
335
336static inline void
337OUT_PKT4(struct fd_ringbuffer *ring, uint16_t regindx, uint16_t cnt)
338{
339	BEGIN_RING(ring, cnt+1);
340	OUT_RING(ring, CP_TYPE4_PKT | cnt |
341			(_odd_parity_bit(cnt) << 7) |
342			((regindx & 0x3ffff) << 8) |
343			((_odd_parity_bit(regindx) << 27)));
344}
345
346static inline void
347OUT_PKT7(struct fd_ringbuffer *ring, uint8_t opcode, uint16_t cnt)
348{
349	BEGIN_RING(ring, cnt+1);
350	OUT_RING(ring, CP_TYPE7_PKT | cnt |
351			(_odd_parity_bit(cnt) << 15) |
352			((opcode & 0x7f) << 16) |
353			((_odd_parity_bit(opcode) << 23)));
354}
355
356static inline void
357OUT_WFI(struct fd_ringbuffer *ring)
358{
359	OUT_PKT3(ring, CP_WAIT_FOR_IDLE, 1);
360	OUT_RING(ring, 0x00000000);
361}
362
363static inline void
364OUT_WFI5(struct fd_ringbuffer *ring)
365{
366	OUT_PKT7(ring, CP_WAIT_FOR_IDLE, 0);
367}
368
369static inline void
370__OUT_IB(struct fd_ringbuffer *ring, bool prefetch, struct fd_ringbuffer *target)
371{
372	if (target->cur == target->start)
373		return;
374
375	unsigned count = fd_ringbuffer_cmd_count(target);
376
377	/* for debug after a lock up, write a unique counter value
378	 * to scratch6 for each IB, to make it easier to match up
379	 * register dumps to cmdstream.  The combination of IB and
380	 * DRAW (scratch7) is enough to "triangulate" the particular
381	 * draw that caused lockup.
382	 */
383	emit_marker(ring, 6);
384
385	for (unsigned i = 0; i < count; i++) {
386		uint32_t dwords;
387		OUT_PKT3(ring, prefetch ? CP_INDIRECT_BUFFER_PFE : CP_INDIRECT_BUFFER_PFD, 2);
388		dwords = fd_ringbuffer_emit_reloc_ring_full(ring, target, i) / 4;
389		assert(dwords > 0);
390		OUT_RING(ring, dwords);
391		OUT_PKT2(ring);
392	}
393
394	emit_marker(ring, 6);
395}
396
397static inline void
398__OUT_IB5(struct fd_ringbuffer *ring, struct fd_ringbuffer *target)
399{
400	if (target->cur == target->start)
401		return;
402
403	unsigned count = fd_ringbuffer_cmd_count(target);
404
405	for (unsigned i = 0; i < count; i++) {
406		uint32_t dwords;
407		OUT_PKT7(ring, CP_INDIRECT_BUFFER, 3);
408		dwords = fd_ringbuffer_emit_reloc_ring_full(ring, target, i) / 4;
409		assert(dwords > 0);
410		OUT_RING(ring, dwords);
411	}
412}
413
414/* CP_SCRATCH_REG4 is used to hold base address for query results: */
415// XXX annoyingly scratch regs move on a5xx.. and additionally different
416// packet types.. so freedreno_query_hw is going to need a bit of
417// rework..
418#define HW_QUERY_BASE_REG REG_AXXX_CP_SCRATCH_REG4
419
420static inline void
421emit_marker(struct fd_ringbuffer *ring, int scratch_idx)
422{
423	extern unsigned marker_cnt;
424	unsigned reg = REG_AXXX_CP_SCRATCH_REG0 + scratch_idx;
425	assert(reg != HW_QUERY_BASE_REG);
426	if (reg == HW_QUERY_BASE_REG)
427		return;
428	OUT_PKT0(ring, reg, 1);
429	OUT_RING(ring, ++marker_cnt);
430}
431
432static inline uint32_t
433pack_rgba(enum pipe_format format, const float *rgba)
434{
435	union util_color uc;
436	util_pack_color(rgba, format, &uc);
437	return uc.ui[0];
438}
439
440/*
441 * swap - swap value of @a and @b
442 */
443#define swap(a, b) \
444	do { __typeof(a) __tmp = (a); (a) = (b); (b) = __tmp; } while (0)
445
446#define foreach_bit(b, mask) \
447	for (uint32_t _m = (mask); _m && ({(b) = u_bit_scan(&_m); 1;});)
448
449
450#define BIT(bit) (1u << bit)
451
452/*
453 * a3xx+ helpers:
454 */
455
456static inline enum a3xx_msaa_samples
457fd_msaa_samples(unsigned samples)
458{
459	switch (samples) {
460	default:
461		debug_assert(0);
462	case 0:
463	case 1: return MSAA_ONE;
464	case 2: return MSAA_TWO;
465	case 4: return MSAA_FOUR;
466	case 8: return MSAA_EIGHT;
467	}
468}
469
470/*
471 * a4xx+ helpers:
472 */
473
474static inline enum a4xx_state_block
475fd4_stage2shadersb(gl_shader_stage type)
476{
477	switch (type) {
478	case MESA_SHADER_VERTEX:
479		return SB4_VS_SHADER;
480	case MESA_SHADER_FRAGMENT:
481		return SB4_FS_SHADER;
482	case MESA_SHADER_COMPUTE:
483	case MESA_SHADER_KERNEL:
484		return SB4_CS_SHADER;
485	default:
486		unreachable("bad shader type");
487		return ~0;
488	}
489}
490
491#endif /* FREEDRENO_UTIL_H_ */
492