1/*
2 * Copyright (C) 2016 Rob Clark <robclark@freedesktop.org>
3 * Copyright © 2018 Google, Inc.
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 *
24 * Authors:
25 *    Rob Clark <robclark@freedesktop.org>
26 */
27
28#include "pipe/p_state.h"
29#include "util/u_string.h"
30#include "util/u_memory.h"
31#include "util/u_prim.h"
32
33#include "freedreno_state.h"
34#include "freedreno_resource.h"
35
36#include "fd6_draw.h"
37#include "fd6_context.h"
38#include "fd6_emit.h"
39#include "fd6_program.h"
40#include "fd6_format.h"
41#include "fd6_zsa.h"
42
43/* some bits in common w/ a4xx: */
44#include "a4xx/fd4_draw.h"
45
46static void
47draw_emit_indirect(struct fd_batch *batch, struct fd_ringbuffer *ring,
48				   enum pc_di_primtype primtype,
49				   const struct pipe_draw_info *info,
50				   unsigned index_offset)
51{
52	struct fd_resource *ind = fd_resource(info->indirect->buffer);
53
54	if (info->index_size) {
55		struct pipe_resource *idx = info->index.resource;
56		unsigned max_indicies = idx->width0 / info->index_size;
57
58		OUT_PKT7(ring, CP_DRAW_INDX_INDIRECT, 6);
59		OUT_RINGP(ring, DRAW4(primtype, DI_SRC_SEL_DMA,
60							  fd4_size2indextype(info->index_size), 0),
61				  &batch->draw_patches);
62		OUT_RELOC(ring, fd_resource(idx)->bo,
63				  index_offset, 0, 0);
64		// XXX: Check A5xx vs A6xx
65		OUT_RING(ring, A5XX_CP_DRAW_INDX_INDIRECT_3_MAX_INDICES(max_indicies));
66		OUT_RELOC(ring, ind->bo, info->indirect->offset, 0, 0);
67	} else {
68		OUT_PKT7(ring, CP_DRAW_INDIRECT, 3);
69		OUT_RINGP(ring, DRAW4(primtype, DI_SRC_SEL_AUTO_INDEX, 0, 0),
70				  &batch->draw_patches);
71		OUT_RELOC(ring, ind->bo, info->indirect->offset, 0, 0);
72	}
73}
74
75static void
76draw_emit(struct fd_batch *batch, struct fd_ringbuffer *ring,
77		  enum pc_di_primtype primtype,
78		  const struct pipe_draw_info *info,
79		  unsigned index_offset)
80{
81	if (info->index_size) {
82		assert(!info->has_user_indices);
83
84		struct pipe_resource *idx_buffer = info->index.resource;
85		uint32_t idx_size = info->index_size * info->count;
86		uint32_t idx_offset = index_offset + info->start * info->index_size;
87
88		/* leave vis mode blank for now, it will be patched up when
89		 * we know if we are binning or not
90		 */
91		uint32_t draw = CP_DRAW_INDX_OFFSET_0_PRIM_TYPE(primtype) |
92			CP_DRAW_INDX_OFFSET_0_SOURCE_SELECT(DI_SRC_SEL_DMA) |
93			CP_DRAW_INDX_OFFSET_0_INDEX_SIZE(fd4_size2indextype(info->index_size)) |
94			0x2000;
95
96		OUT_PKT7(ring, CP_DRAW_INDX_OFFSET, 7);
97		OUT_RINGP(ring, draw, &batch->draw_patches);
98		OUT_RING(ring, info->instance_count);    /* NumInstances */
99		OUT_RING(ring, info->count);             /* NumIndices */
100		OUT_RING(ring, 0x0);           /* XXX */
101		OUT_RELOC(ring, fd_resource(idx_buffer)->bo, idx_offset, 0, 0);
102		OUT_RING (ring, idx_size);
103	} else {
104		/* leave vis mode blank for now, it will be patched up when
105		 * we know if we are binning or not
106		 */
107		uint32_t draw = CP_DRAW_INDX_OFFSET_0_PRIM_TYPE(primtype) |
108			CP_DRAW_INDX_OFFSET_0_SOURCE_SELECT(DI_SRC_SEL_AUTO_INDEX) |
109			0x2000;
110
111		OUT_PKT7(ring, CP_DRAW_INDX_OFFSET, 3);
112		OUT_RINGP(ring, draw, &batch->draw_patches);
113		OUT_RING(ring, info->instance_count);    /* NumInstances */
114		OUT_RING(ring, info->count);             /* NumIndices */
115	}
116}
117
118/* fixup dirty shader state in case some "unrelated" (from the state-
119 * tracker's perspective) state change causes us to switch to a
120 * different variant.
121 */
122static void
123fixup_shader_state(struct fd_context *ctx, struct ir3_shader_key *key)
124{
125	struct fd6_context *fd6_ctx = fd6_context(ctx);
126	struct ir3_shader_key *last_key = &fd6_ctx->last_key;
127
128	if (!ir3_shader_key_equal(last_key, key)) {
129		if (ir3_shader_key_changes_fs(last_key, key)) {
130			ctx->dirty_shader[PIPE_SHADER_FRAGMENT] |= FD_DIRTY_SHADER_PROG;
131			ctx->dirty |= FD_DIRTY_PROG;
132		}
133
134		if (ir3_shader_key_changes_vs(last_key, key)) {
135			ctx->dirty_shader[PIPE_SHADER_VERTEX] |= FD_DIRTY_SHADER_PROG;
136			ctx->dirty |= FD_DIRTY_PROG;
137		}
138
139		fd6_ctx->last_key = *key;
140	}
141}
142
143static bool
144fd6_draw_vbo(struct fd_context *ctx, const struct pipe_draw_info *info,
145             unsigned index_offset)
146{
147	struct fd6_context *fd6_ctx = fd6_context(ctx);
148	struct fd6_emit emit = {
149		.ctx = ctx,
150		.vtx  = &ctx->vtx,
151		.info = info,
152		.key = {
153			.vs = ctx->prog.vp,
154			.fs = ctx->prog.fp,
155			.key = {
156				.color_two_side = ctx->rasterizer->light_twoside,
157				.vclamp_color = ctx->rasterizer->clamp_vertex_color,
158				.fclamp_color = ctx->rasterizer->clamp_fragment_color,
159				.rasterflat = ctx->rasterizer->flatshade,
160				.ucp_enables = ctx->rasterizer->clip_plane_enable,
161				.has_per_samp = (fd6_ctx->fsaturate || fd6_ctx->vsaturate),
162				.vsaturate_s = fd6_ctx->vsaturate_s,
163				.vsaturate_t = fd6_ctx->vsaturate_t,
164				.vsaturate_r = fd6_ctx->vsaturate_r,
165				.fsaturate_s = fd6_ctx->fsaturate_s,
166				.fsaturate_t = fd6_ctx->fsaturate_t,
167				.fsaturate_r = fd6_ctx->fsaturate_r,
168				.vsamples = ctx->tex[PIPE_SHADER_VERTEX].samples,
169				.fsamples = ctx->tex[PIPE_SHADER_FRAGMENT].samples,
170				.sample_shading = (ctx->min_samples > 1),
171				.msaa = (ctx->framebuffer.samples > 1),
172			},
173		},
174		.rasterflat = ctx->rasterizer->flatshade,
175		.sprite_coord_enable = ctx->rasterizer->sprite_coord_enable,
176		.sprite_coord_mode = ctx->rasterizer->sprite_coord_mode,
177	};
178
179	fixup_shader_state(ctx, &emit.key.key);
180
181	if (!(ctx->dirty & FD_DIRTY_PROG)) {
182		emit.prog = fd6_ctx->prog;
183	} else {
184		fd6_ctx->prog = fd6_emit_get_prog(&emit);
185	}
186
187	/* bail if compile failed: */
188	if (!fd6_ctx->prog)
189		return NULL;
190
191	emit.dirty = ctx->dirty;      /* *after* fixup_shader_state() */
192	emit.bs = fd6_emit_get_prog(&emit)->bs;
193	emit.vs = fd6_emit_get_prog(&emit)->vs;
194	emit.fs = fd6_emit_get_prog(&emit)->fs;
195
196	const struct ir3_shader_variant *vp = emit.vs;
197	const struct ir3_shader_variant *fp = emit.fs;
198
199	ctx->stats.vs_regs += ir3_shader_halfregs(vp);
200	ctx->stats.fs_regs += ir3_shader_halfregs(fp);
201
202	/* figure out whether we need to disable LRZ write for binning
203	 * pass using draw pass's fp:
204	 */
205	emit.no_lrz_write = fp->writes_pos || fp->no_earlyz;
206
207	struct fd_ringbuffer *ring = ctx->batch->draw;
208	enum pc_di_primtype primtype = ctx->primtypes[info->mode];
209
210	fd6_emit_state(ring, &emit);
211
212	OUT_PKT4(ring, REG_A6XX_VFD_INDEX_OFFSET, 2);
213	OUT_RING(ring, info->index_size ? info->index_bias : info->start); /* VFD_INDEX_OFFSET */
214	OUT_RING(ring, info->start_instance);   /* VFD_INSTANCE_START_OFFSET */
215
216	OUT_PKT4(ring, REG_A6XX_PC_RESTART_INDEX, 1);
217	OUT_RING(ring, info->primitive_restart ? /* PC_RESTART_INDEX */
218			info->restart_index : 0xffffffff);
219
220	/* for debug after a lock up, write a unique counter value
221	 * to scratch7 for each draw, to make it easier to match up
222	 * register dumps to cmdstream.  The combination of IB
223	 * (scratch6) and DRAW is enough to "triangulate" the
224	 * particular draw that caused lockup.
225	 */
226	emit_marker6(ring, 7);
227
228	if (info->indirect) {
229		draw_emit_indirect(ctx->batch, ring, primtype,
230						   info, index_offset);
231	} else {
232		draw_emit(ctx->batch, ring, primtype,
233				  info, index_offset);
234	}
235
236	emit_marker6(ring, 7);
237	fd_reset_wfi(ctx->batch);
238
239	if (emit.streamout_mask) {
240		struct fd_ringbuffer *ring = ctx->batch->draw;
241
242		for (unsigned i = 0; i < PIPE_MAX_SO_BUFFERS; i++) {
243			if (emit.streamout_mask & (1 << i)) {
244				fd6_event_write(ctx->batch, ring, FLUSH_SO_0 + i, false);
245			}
246		}
247	}
248
249	fd_context_all_clean(ctx);
250
251	return true;
252}
253
254static void
255fd6_clear_lrz(struct fd_batch *batch, struct fd_resource *zsbuf, double depth)
256{
257	struct fd_ringbuffer *ring;
258
259	// TODO mid-frame clears (ie. app doing crazy stuff)??  Maybe worth
260	// splitting both clear and lrz clear out into their own rb's.  And
261	// just throw away any draws prior to clear.  (Anything not fullscreen
262	// clear, just fallback to generic path that treats it as a normal
263	// draw
264
265	if (!batch->lrz_clear) {
266		batch->lrz_clear = fd_submit_new_ringbuffer(batch->submit, 0x1000, 0);
267	}
268
269	ring = batch->lrz_clear;
270
271	emit_marker6(ring, 7);
272	OUT_PKT7(ring, CP_SET_MARKER, 1);
273	OUT_RING(ring, A2XX_CP_SET_MARKER_0_MODE(RM6_BYPASS));
274	emit_marker6(ring, 7);
275
276	OUT_PKT4(ring, REG_A6XX_RB_CCU_CNTL, 1);
277	OUT_RING(ring, 0x10000000);
278
279	OUT_PKT4(ring, REG_A6XX_HLSQ_UPDATE_CNTL, 1);
280	OUT_RING(ring, 0x7ffff);
281
282	emit_marker6(ring, 7);
283	OUT_PKT7(ring, CP_SET_MARKER, 1);
284	OUT_RING(ring, A2XX_CP_SET_MARKER_0_MODE(0xc));
285	emit_marker6(ring, 7);
286
287	OUT_PKT4(ring, REG_A6XX_RB_UNKNOWN_8C01, 1);
288	OUT_RING(ring, 0x0);
289
290	OUT_PKT4(ring, REG_A6XX_SP_PS_2D_SRC_INFO, 13);
291	OUT_RING(ring, 0x00000000);
292	OUT_RING(ring, 0x00000000);
293	OUT_RING(ring, 0x00000000);
294	OUT_RING(ring, 0x00000000);
295	OUT_RING(ring, 0x00000000);
296	OUT_RING(ring, 0x00000000);
297	OUT_RING(ring, 0x00000000);
298	OUT_RING(ring, 0x00000000);
299	OUT_RING(ring, 0x00000000);
300	OUT_RING(ring, 0x00000000);
301	OUT_RING(ring, 0x00000000);
302	OUT_RING(ring, 0x00000000);
303	OUT_RING(ring, 0x00000000);
304
305	OUT_PKT4(ring, REG_A6XX_SP_2D_SRC_FORMAT, 1);
306	OUT_RING(ring, 0x0000f410);
307
308	OUT_PKT4(ring, REG_A6XX_GRAS_2D_BLIT_CNTL, 1);
309	OUT_RING(ring, A6XX_GRAS_2D_BLIT_CNTL_COLOR_FORMAT(RB6_R16_UNORM) |
310			0x4f00080);
311
312	OUT_PKT4(ring, REG_A6XX_RB_2D_BLIT_CNTL, 1);
313	OUT_RING(ring, A6XX_RB_2D_BLIT_CNTL_COLOR_FORMAT(RB6_R16_UNORM) |
314			0x4f00080);
315
316	fd6_event_write(batch, ring, UNK_1D, true);
317	fd6_event_write(batch, ring, PC_CCU_INVALIDATE_COLOR, false);
318
319	OUT_PKT4(ring, REG_A6XX_RB_2D_SRC_SOLID_C0, 4);
320	OUT_RING(ring, fui(depth));
321	OUT_RING(ring, 0x00000000);
322	OUT_RING(ring, 0x00000000);
323	OUT_RING(ring, 0x00000000);
324
325	OUT_PKT4(ring, REG_A6XX_RB_2D_DST_INFO, 9);
326	OUT_RING(ring, A6XX_RB_2D_DST_INFO_COLOR_FORMAT(RB6_R16_UNORM) |
327			A6XX_RB_2D_DST_INFO_TILE_MODE(TILE6_LINEAR) |
328			A6XX_RB_2D_DST_INFO_COLOR_SWAP(WZYX));
329	OUT_RELOCW(ring, zsbuf->lrz, 0, 0, 0);
330	OUT_RING(ring, A6XX_RB_2D_DST_SIZE_PITCH(zsbuf->lrz_pitch * 2));
331	OUT_RING(ring, 0x00000000);
332	OUT_RING(ring, 0x00000000);
333	OUT_RING(ring, 0x00000000);
334	OUT_RING(ring, 0x00000000);
335	OUT_RING(ring, 0x00000000);
336
337	OUT_PKT4(ring, REG_A6XX_GRAS_2D_SRC_TL_X, 4);
338	OUT_RING(ring, A6XX_GRAS_2D_SRC_TL_X_X(0));
339	OUT_RING(ring, A6XX_GRAS_2D_SRC_BR_X_X(0));
340	OUT_RING(ring, A6XX_GRAS_2D_SRC_TL_Y_Y(0));
341	OUT_RING(ring, A6XX_GRAS_2D_SRC_BR_Y_Y(0));
342
343	OUT_PKT4(ring, REG_A6XX_GRAS_2D_DST_TL, 2);
344	OUT_RING(ring, A6XX_GRAS_2D_DST_TL_X(0) |
345			A6XX_GRAS_2D_DST_TL_Y(0));
346	OUT_RING(ring, A6XX_GRAS_2D_DST_BR_X(zsbuf->lrz_width - 1) |
347			A6XX_GRAS_2D_DST_BR_Y(zsbuf->lrz_height - 1));
348
349	fd6_event_write(batch, ring, 0x3f, false);
350
351	OUT_WFI5(ring);
352
353	OUT_PKT4(ring, REG_A6XX_RB_UNKNOWN_8E04, 1);
354	OUT_RING(ring, 0x1000000);
355
356	OUT_PKT7(ring, CP_BLIT, 1);
357	OUT_RING(ring, CP_BLIT_0_OP(BLIT_OP_SCALE));
358
359	OUT_WFI5(ring);
360
361	OUT_PKT4(ring, REG_A6XX_RB_UNKNOWN_8E04, 1);
362	OUT_RING(ring, 0x0);
363
364	fd6_event_write(batch, ring, UNK_1D, true);
365	fd6_event_write(batch, ring, FACENESS_FLUSH, true);
366	fd6_event_write(batch, ring, CACHE_FLUSH_TS, true);
367
368	fd6_cache_inv(batch, ring);
369}
370
371static bool is_z32(enum pipe_format format)
372{
373	switch (format) {
374	case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
375	case PIPE_FORMAT_Z32_UNORM:
376	case PIPE_FORMAT_Z32_FLOAT:
377		return true;
378	default:
379		return false;
380	}
381}
382
383static bool
384fd6_clear(struct fd_context *ctx, unsigned buffers,
385		const union pipe_color_union *color, double depth, unsigned stencil)
386{
387	struct pipe_framebuffer_state *pfb = &ctx->batch->framebuffer;
388	const bool has_depth = pfb->zsbuf;
389	unsigned color_buffers = buffers >> 2;
390	unsigned i;
391
392	/* If we're clearing after draws, fallback to 3D pipe clears.  We could
393	 * use blitter clears in the draw batch but then we'd have to patch up the
394	 * gmem offsets. This doesn't seem like a useful thing to optimize for
395	 * however.*/
396	if (ctx->batch->num_draws > 0)
397		return false;
398
399	foreach_bit(i, color_buffers)
400		ctx->batch->clear_color[i] = *color;
401	if (buffers & PIPE_CLEAR_DEPTH)
402		ctx->batch->clear_depth = depth;
403	if (buffers & PIPE_CLEAR_STENCIL)
404		ctx->batch->clear_stencil = stencil;
405
406	ctx->batch->fast_cleared |= buffers;
407
408	if (has_depth && (buffers & PIPE_CLEAR_DEPTH)) {
409		struct fd_resource *zsbuf = fd_resource(pfb->zsbuf->texture);
410		if (zsbuf->lrz && !is_z32(pfb->zsbuf->format)) {
411			zsbuf->lrz_valid = true;
412			fd6_clear_lrz(ctx->batch, zsbuf, depth);
413		}
414	}
415
416	return true;
417}
418
419void
420fd6_draw_init(struct pipe_context *pctx)
421{
422	struct fd_context *ctx = fd_context(pctx);
423	ctx->draw_vbo = fd6_draw_vbo;
424	ctx->clear = fd6_clear;
425}
426