• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (C) 2016 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 #include "pipe/p_state.h"
28 #include "util/u_string.h"
29 #include "util/u_memory.h"
30 #include "util/u_prim.h"
31 
32 #include "freedreno_state.h"
33 #include "freedreno_resource.h"
34 
35 #include "fd5_draw.h"
36 #include "fd5_context.h"
37 #include "fd5_emit.h"
38 #include "fd5_program.h"
39 #include "fd5_format.h"
40 #include "fd5_zsa.h"
41 
42 
43 static void
draw_impl(struct fd_context * ctx,struct fd_ringbuffer * ring,struct fd5_emit * emit,unsigned index_offset)44 draw_impl(struct fd_context *ctx, struct fd_ringbuffer *ring,
45 		struct fd5_emit *emit, unsigned index_offset)
46 {
47 	const struct pipe_draw_info *info = emit->info;
48 	enum pc_di_primtype primtype = ctx->primtypes[info->mode];
49 
50 	fd5_emit_state(ctx, ring, emit);
51 
52 	if (emit->dirty & (FD_DIRTY_VTXBUF | FD_DIRTY_VTXSTATE))
53 		fd5_emit_vertex_bufs(ring, emit);
54 
55 	OUT_PKT4(ring, REG_A5XX_VFD_INDEX_OFFSET, 2);
56 	OUT_RING(ring, info->index_size ? info->index_bias : info->start); /* VFD_INDEX_OFFSET */
57 	OUT_RING(ring, info->start_instance);   /* VFD_INSTANCE_START_OFFSET */
58 
59 	OUT_PKT4(ring, REG_A5XX_PC_RESTART_INDEX, 1);
60 	OUT_RING(ring, info->primitive_restart ? /* PC_RESTART_INDEX */
61 			info->restart_index : 0xffffffff);
62 
63 	fd5_emit_render_cntl(ctx, false, emit->key.binning_pass);
64 	fd5_draw_emit(ctx->batch, ring, primtype,
65 			emit->key.binning_pass ? IGNORE_VISIBILITY : USE_VISIBILITY,
66 			info, index_offset);
67 }
68 
69 /* fixup dirty shader state in case some "unrelated" (from the state-
70  * tracker's perspective) state change causes us to switch to a
71  * different variant.
72  */
73 static void
fixup_shader_state(struct fd_context * ctx,struct ir3_shader_key * key)74 fixup_shader_state(struct fd_context *ctx, struct ir3_shader_key *key)
75 {
76 	struct fd5_context *fd5_ctx = fd5_context(ctx);
77 	struct ir3_shader_key *last_key = &fd5_ctx->last_key;
78 
79 	if (!ir3_shader_key_equal(last_key, key)) {
80 		if (ir3_shader_key_changes_fs(last_key, key)) {
81 			ctx->dirty_shader[PIPE_SHADER_FRAGMENT] |= FD_DIRTY_SHADER_PROG;
82 			ctx->dirty |= FD_DIRTY_PROG;
83 		}
84 
85 		if (ir3_shader_key_changes_vs(last_key, key)) {
86 			ctx->dirty_shader[PIPE_SHADER_VERTEX] |= FD_DIRTY_SHADER_PROG;
87 			ctx->dirty |= FD_DIRTY_PROG;
88 		}
89 
90 		fd5_ctx->last_key = *key;
91 	}
92 }
93 
94 static bool
fd5_draw_vbo(struct fd_context * ctx,const struct pipe_draw_info * info,unsigned index_offset)95 fd5_draw_vbo(struct fd_context *ctx, const struct pipe_draw_info *info,
96              unsigned index_offset)
97 {
98 	struct fd5_context *fd5_ctx = fd5_context(ctx);
99 	struct fd5_emit emit = {
100 		.debug = &ctx->debug,
101 		.vtx  = &ctx->vtx,
102 		.prog = &ctx->prog,
103 		.info = info,
104 		.key = {
105 			.color_two_side = ctx->rasterizer->light_twoside,
106 			.vclamp_color = ctx->rasterizer->clamp_vertex_color,
107 			.fclamp_color = ctx->rasterizer->clamp_fragment_color,
108 			.rasterflat = ctx->rasterizer->flatshade,
109 			.half_precision = ctx->in_blit &&
110 					fd_half_precision(&ctx->batch->framebuffer),
111 			.ucp_enables = ctx->rasterizer->clip_plane_enable,
112 			.has_per_samp = (fd5_ctx->fsaturate || fd5_ctx->vsaturate ||
113 					fd5_ctx->fastc_srgb || fd5_ctx->vastc_srgb),
114 			.vsaturate_s = fd5_ctx->vsaturate_s,
115 			.vsaturate_t = fd5_ctx->vsaturate_t,
116 			.vsaturate_r = fd5_ctx->vsaturate_r,
117 			.fsaturate_s = fd5_ctx->fsaturate_s,
118 			.fsaturate_t = fd5_ctx->fsaturate_t,
119 			.fsaturate_r = fd5_ctx->fsaturate_r,
120 			.vastc_srgb = fd5_ctx->vastc_srgb,
121 			.fastc_srgb = fd5_ctx->fastc_srgb,
122 		},
123 		.rasterflat = ctx->rasterizer->flatshade,
124 		.sprite_coord_enable = ctx->rasterizer->sprite_coord_enable,
125 		.sprite_coord_mode = ctx->rasterizer->sprite_coord_mode,
126 	};
127 
128 	fixup_shader_state(ctx, &emit.key);
129 
130 	unsigned dirty = ctx->dirty;
131 	const struct ir3_shader_variant *vp = fd5_emit_get_vp(&emit);
132 	const struct ir3_shader_variant *fp = fd5_emit_get_fp(&emit);
133 
134 	/* do regular pass first, since that is more likely to fail compiling: */
135 
136 	if (!vp || !fp)
137 		return false;
138 
139 	/* figure out whether we need to disable LRZ write for binning
140 	 * pass using draw pass's fp:
141 	 */
142 	emit.no_lrz_write = fp->writes_pos || fp->has_kill;
143 
144 	emit.key.binning_pass = false;
145 	emit.dirty = dirty;
146 
147 	draw_impl(ctx, ctx->batch->draw, &emit, index_offset);
148 
149 	/* and now binning pass: */
150 	emit.key.binning_pass = true;
151 	emit.dirty = dirty & ~(FD_DIRTY_BLEND);
152 	emit.vp = NULL;   /* we changed key so need to refetch vp */
153 	emit.fp = NULL;
154 	draw_impl(ctx, ctx->batch->binning, &emit, index_offset);
155 
156 	if (emit.streamout_mask) {
157 		struct fd_ringbuffer *ring = ctx->batch->draw;
158 
159 		for (unsigned i = 0; i < PIPE_MAX_SO_BUFFERS; i++) {
160 			if (emit.streamout_mask & (1 << i)) {
161 				OUT_PKT7(ring, CP_EVENT_WRITE, 1);
162 				OUT_RING(ring, FLUSH_SO_0 + i);
163 			}
164 		}
165 	}
166 
167 	fd_context_all_clean(ctx);
168 
169 	return true;
170 }
171 
is_z32(enum pipe_format format)172 static bool is_z32(enum pipe_format format)
173 {
174 	switch (format) {
175 	case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
176 	case PIPE_FORMAT_Z32_UNORM:
177 	case PIPE_FORMAT_Z32_FLOAT:
178 		return true;
179 	default:
180 		return false;
181 	}
182 }
183 
184 static void
fd5_clear_lrz(struct fd_batch * batch,struct fd_resource * zsbuf,double depth)185 fd5_clear_lrz(struct fd_batch *batch, struct fd_resource *zsbuf, double depth)
186 {
187 	struct fd_ringbuffer *ring;
188 	uint32_t clear = util_pack_z(PIPE_FORMAT_Z16_UNORM, depth);
189 
190 	// TODO mid-frame clears (ie. app doing crazy stuff)??  Maybe worth
191 	// splitting both clear and lrz clear out into their own rb's.  And
192 	// just throw away any draws prior to clear.  (Anything not fullscreen
193 	// clear, just fallback to generic path that treats it as a normal
194 	// draw
195 
196 	if (!batch->lrz_clear) {
197 		batch->lrz_clear = fd_ringbuffer_new(batch->ctx->pipe, 0x1000);
198 		fd_ringbuffer_set_parent(batch->lrz_clear, batch->gmem);
199 	}
200 
201 	ring = batch->lrz_clear;
202 
203 	OUT_WFI5(ring);
204 
205 	OUT_PKT4(ring, REG_A5XX_RB_CCU_CNTL, 1);
206 	OUT_RING(ring, 0x10000000);
207 
208 	OUT_PKT4(ring, REG_A5XX_HLSQ_UPDATE_CNTL, 1);
209 	OUT_RING(ring, 0x20fffff);
210 
211 	OUT_PKT4(ring, REG_A5XX_GRAS_SU_CNTL, 1);
212 	OUT_RING(ring, A5XX_GRAS_SU_CNTL_LINEHALFWIDTH(0.0));
213 
214 	OUT_PKT4(ring, REG_A5XX_GRAS_CNTL, 1);
215 	OUT_RING(ring, 0x00000000);
216 
217 	OUT_PKT4(ring, REG_A5XX_GRAS_CL_CNTL, 1);
218 	OUT_RING(ring, 0x00000181);
219 
220 	OUT_PKT4(ring, REG_A5XX_GRAS_LRZ_CNTL, 1);
221 	OUT_RING(ring, 0x00000000);
222 
223 	OUT_PKT4(ring, REG_A5XX_RB_MRT_BUF_INFO(0), 5);
224 	OUT_RING(ring, A5XX_RB_MRT_BUF_INFO_COLOR_FORMAT(RB5_R16_UNORM) |
225 			A5XX_RB_MRT_BUF_INFO_COLOR_TILE_MODE(TILE5_LINEAR) |
226 			A5XX_RB_MRT_BUF_INFO_COLOR_SWAP(WZYX));
227 	OUT_RING(ring, A5XX_RB_MRT_PITCH(zsbuf->lrz_pitch * 2));
228 	OUT_RING(ring, A5XX_RB_MRT_ARRAY_PITCH(fd_bo_size(zsbuf->lrz)));
229 	OUT_RELOCW(ring, zsbuf->lrz, 0x1000, 0, 0);
230 
231 	OUT_PKT4(ring, REG_A5XX_RB_RENDER_CNTL, 1);
232 	OUT_RING(ring, 0x00000000);
233 
234 	OUT_PKT4(ring, REG_A5XX_RB_DEST_MSAA_CNTL, 1);
235 	OUT_RING(ring, A5XX_RB_DEST_MSAA_CNTL_SAMPLES(MSAA_ONE));
236 
237 	OUT_PKT4(ring, REG_A5XX_RB_BLIT_CNTL, 1);
238 	OUT_RING(ring, A5XX_RB_BLIT_CNTL_BUF(BLIT_MRT0));
239 
240 	OUT_PKT4(ring, REG_A5XX_RB_CLEAR_CNTL, 1);
241 	OUT_RING(ring, A5XX_RB_CLEAR_CNTL_FAST_CLEAR |
242 			A5XX_RB_CLEAR_CNTL_MASK(0xf));
243 
244 	OUT_PKT4(ring, REG_A5XX_RB_CLEAR_COLOR_DW0, 1);
245 	OUT_RING(ring, clear);  /* RB_CLEAR_COLOR_DW0 */
246 
247 	OUT_PKT4(ring, REG_A5XX_VSC_RESOLVE_CNTL, 2);
248 	OUT_RING(ring, A5XX_VSC_RESOLVE_CNTL_X(zsbuf->lrz_width) |
249 			 A5XX_VSC_RESOLVE_CNTL_Y(zsbuf->lrz_height));
250 	OUT_RING(ring, 0x00000000);   // XXX UNKNOWN_0CDE
251 
252 	OUT_PKT4(ring, REG_A5XX_RB_CNTL, 1);
253 	OUT_RING(ring, A5XX_RB_CNTL_BYPASS);
254 
255 	OUT_PKT4(ring, REG_A5XX_RB_RESOLVE_CNTL_1, 2);
256 	OUT_RING(ring, A5XX_RB_RESOLVE_CNTL_1_X(0) |
257 			A5XX_RB_RESOLVE_CNTL_1_Y(0));
258 	OUT_RING(ring, A5XX_RB_RESOLVE_CNTL_2_X(zsbuf->lrz_width - 1) |
259 			A5XX_RB_RESOLVE_CNTL_2_Y(zsbuf->lrz_height - 1));
260 
261 	fd5_emit_blit(batch->ctx, ring);
262 }
263 
264 static bool
fd5_clear(struct fd_context * ctx,unsigned buffers,const union pipe_color_union * color,double depth,unsigned stencil)265 fd5_clear(struct fd_context *ctx, unsigned buffers,
266 		const union pipe_color_union *color, double depth, unsigned stencil)
267 {
268 	struct fd_ringbuffer *ring = ctx->batch->draw;
269 	struct pipe_framebuffer_state *pfb = &ctx->batch->framebuffer;
270 	struct pipe_scissor_state *scissor = fd_context_get_scissor(ctx);
271 
272 	if ((buffers & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL)) &&
273 			is_z32(pfb->zsbuf->format))
274 		return false;
275 
276 	ctx->batch->max_scissor.minx = MIN2(ctx->batch->max_scissor.minx, scissor->minx);
277 	ctx->batch->max_scissor.miny = MIN2(ctx->batch->max_scissor.miny, scissor->miny);
278 	ctx->batch->max_scissor.maxx = MAX2(ctx->batch->max_scissor.maxx, scissor->maxx);
279 	ctx->batch->max_scissor.maxy = MAX2(ctx->batch->max_scissor.maxy, scissor->maxy);
280 
281 	fd5_emit_render_cntl(ctx, true, false);
282 
283 	if (buffers & PIPE_CLEAR_COLOR) {
284 		for (int i = 0; i < pfb->nr_cbufs; i++) {
285 			union util_color uc = {0};
286 
287 			if (!pfb->cbufs[i])
288 				continue;
289 
290 			if (!(buffers & (PIPE_CLEAR_COLOR0 << i)))
291 				continue;
292 
293 			enum pipe_format pfmt = pfb->cbufs[i]->format;
294 
295 			// XXX I think RB_CLEAR_COLOR_DWn wants to take into account SWAP??
296 			union pipe_color_union swapped;
297 			switch (fd5_pipe2swap(pfmt)) {
298 			case WZYX:
299 				swapped.ui[0] = color->ui[0];
300 				swapped.ui[1] = color->ui[1];
301 				swapped.ui[2] = color->ui[2];
302 				swapped.ui[3] = color->ui[3];
303 				break;
304 			case WXYZ:
305 				swapped.ui[2] = color->ui[0];
306 				swapped.ui[1] = color->ui[1];
307 				swapped.ui[0] = color->ui[2];
308 				swapped.ui[3] = color->ui[3];
309 				break;
310 			case ZYXW:
311 				swapped.ui[3] = color->ui[0];
312 				swapped.ui[0] = color->ui[1];
313 				swapped.ui[1] = color->ui[2];
314 				swapped.ui[2] = color->ui[3];
315 				break;
316 			case XYZW:
317 				swapped.ui[3] = color->ui[0];
318 				swapped.ui[2] = color->ui[1];
319 				swapped.ui[1] = color->ui[2];
320 				swapped.ui[0] = color->ui[3];
321 				break;
322 			}
323 
324 			if (util_format_is_pure_uint(pfmt)) {
325 				util_format_write_4ui(pfmt, swapped.ui, 0, &uc, 0, 0, 0, 1, 1);
326 			} else if (util_format_is_pure_sint(pfmt)) {
327 				util_format_write_4i(pfmt, swapped.i, 0, &uc, 0, 0, 0, 1, 1);
328 			} else {
329 				util_pack_color(swapped.f, pfmt, &uc);
330 			}
331 
332 			OUT_PKT4(ring, REG_A5XX_RB_BLIT_CNTL, 1);
333 			OUT_RING(ring, A5XX_RB_BLIT_CNTL_BUF(BLIT_MRT0 + i));
334 
335 			OUT_PKT4(ring, REG_A5XX_RB_CLEAR_CNTL, 1);
336 			OUT_RING(ring, A5XX_RB_CLEAR_CNTL_FAST_CLEAR |
337 					A5XX_RB_CLEAR_CNTL_MASK(0xf));
338 
339 			OUT_PKT4(ring, REG_A5XX_RB_CLEAR_COLOR_DW0, 4);
340 			OUT_RING(ring, uc.ui[0]);  /* RB_CLEAR_COLOR_DW0 */
341 			OUT_RING(ring, uc.ui[1]);  /* RB_CLEAR_COLOR_DW1 */
342 			OUT_RING(ring, uc.ui[2]);  /* RB_CLEAR_COLOR_DW2 */
343 			OUT_RING(ring, uc.ui[3]);  /* RB_CLEAR_COLOR_DW3 */
344 
345 			fd5_emit_blit(ctx, ring);
346 		}
347 	}
348 
349 	if (pfb->zsbuf && (buffers & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL))) {
350 		uint32_t clear =
351 			util_pack_z_stencil(pfb->zsbuf->format, depth, stencil);
352 		uint32_t mask = 0;
353 
354 		if (buffers & PIPE_CLEAR_DEPTH)
355 			mask |= 0x1;
356 
357 		if (buffers & PIPE_CLEAR_STENCIL)
358 			mask |= 0x2;
359 
360 		OUT_PKT4(ring, REG_A5XX_RB_BLIT_CNTL, 1);
361 		OUT_RING(ring, A5XX_RB_BLIT_CNTL_BUF(BLIT_ZS));
362 
363 		OUT_PKT4(ring, REG_A5XX_RB_CLEAR_CNTL, 1);
364 		OUT_RING(ring, A5XX_RB_CLEAR_CNTL_FAST_CLEAR |
365 				A5XX_RB_CLEAR_CNTL_MASK(mask));
366 
367 		OUT_PKT4(ring, REG_A5XX_RB_CLEAR_COLOR_DW0, 1);
368 		OUT_RING(ring, clear);    /* RB_CLEAR_COLOR_DW0 */
369 
370 		fd5_emit_blit(ctx, ring);
371 
372 		if (pfb->zsbuf && (buffers & PIPE_CLEAR_DEPTH)) {
373 			struct fd_resource *zsbuf = fd_resource(pfb->zsbuf->texture);
374 			if (zsbuf->lrz) {
375 				zsbuf->lrz_valid = true;
376 				fd5_clear_lrz(ctx->batch, zsbuf, depth);
377 			}
378 		}
379 	}
380 
381 	/* disable fast clear to not interfere w/ gmem->mem, etc.. */
382 	OUT_PKT4(ring, REG_A5XX_RB_CLEAR_CNTL, 1);
383 	OUT_RING(ring, 0x00000000);   /* RB_CLEAR_CNTL */
384 
385 	return true;
386 }
387 
388 void
fd5_draw_init(struct pipe_context * pctx)389 fd5_draw_init(struct pipe_context *pctx)
390 {
391 	struct fd_context *ctx = fd_context(pctx);
392 	ctx->draw_vbo = fd5_draw_vbo;
393 	ctx->clear = fd5_clear;
394 }
395