• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 <stdio.h>
29 
30 #include "pipe/p_state.h"
31 #include "util/u_string.h"
32 #include "util/u_memory.h"
33 #include "util/u_inlines.h"
34 #include "util/format/u_format.h"
35 
36 #include "freedreno_draw.h"
37 #include "freedreno_log.h"
38 #include "freedreno_state.h"
39 #include "freedreno_resource.h"
40 
41 #include "fd6_blitter.h"
42 #include "fd6_gmem.h"
43 #include "fd6_context.h"
44 #include "fd6_draw.h"
45 #include "fd6_emit.h"
46 #include "fd6_program.h"
47 #include "fd6_format.h"
48 #include "fd6_resource.h"
49 #include "fd6_zsa.h"
50 #include "fd6_pack.h"
51 
52 /**
53  * Emits the flags registers, suitable for RB_MRT_FLAG_BUFFER,
54  * RB_DEPTH_FLAG_BUFFER, SP_PS_2D_SRC_FLAGS, and RB_BLIT_FLAG_DST.
55  */
56 void
fd6_emit_flag_reference(struct fd_ringbuffer * ring,struct fd_resource * rsc,int level,int layer)57 fd6_emit_flag_reference(struct fd_ringbuffer *ring, struct fd_resource *rsc,
58 		int level, int layer)
59 {
60 	if (fd_resource_ubwc_enabled(rsc, level)) {
61 		OUT_RELOC(ring, rsc->bo, fd_resource_ubwc_offset(rsc, level, layer), 0, 0);
62 		OUT_RING(ring,
63 				A6XX_RB_MRT_FLAG_BUFFER_PITCH_PITCH(fdl_ubwc_pitch(&rsc->layout, level)) |
64 				A6XX_RB_MRT_FLAG_BUFFER_PITCH_ARRAY_PITCH(rsc->layout.ubwc_layer_size >> 2));
65 	} else {
66 		OUT_RING(ring, 0x00000000);    /* RB_MRT_FLAG_BUFFER[i].ADDR_LO */
67 		OUT_RING(ring, 0x00000000);    /* RB_MRT_FLAG_BUFFER[i].ADDR_HI */
68 		OUT_RING(ring, 0x00000000);
69 	}
70 }
71 
72 static void
emit_mrt(struct fd_ringbuffer * ring,struct pipe_framebuffer_state * pfb,const struct fd_gmem_stateobj * gmem)73 emit_mrt(struct fd_ringbuffer *ring, struct pipe_framebuffer_state *pfb,
74 		const struct fd_gmem_stateobj *gmem)
75 {
76 	unsigned char mrt_comp[A6XX_MAX_RENDER_TARGETS] = {0};
77 	unsigned srgb_cntl = 0;
78 	unsigned i;
79 
80 	unsigned max_layer_index = 0;
81 
82 	for (i = 0; i < pfb->nr_cbufs; i++) {
83 		enum a6xx_format format = 0;
84 		enum a3xx_color_swap swap = WZYX;
85 		bool sint = false, uint = false;
86 		struct fd_resource *rsc = NULL;
87 		struct fdl_slice *slice = NULL;
88 		uint32_t stride = 0;
89 		uint32_t offset;
90 		uint32_t tile_mode;
91 
92 		if (!pfb->cbufs[i])
93 			continue;
94 
95 		mrt_comp[i] = 0xf;
96 
97 		struct pipe_surface *psurf = pfb->cbufs[i];
98 		enum pipe_format pformat = psurf->format;
99 		rsc = fd_resource(psurf->texture);
100 		if (!rsc->bo)
101 			continue;
102 
103 		uint32_t base = gmem ? gmem->cbuf_base[i] : 0;
104 		slice = fd_resource_slice(rsc, psurf->u.tex.level);
105 		format = fd6_pipe2color(pformat);
106 		sint = util_format_is_pure_sint(pformat);
107 		uint = util_format_is_pure_uint(pformat);
108 
109 		if (util_format_is_srgb(pformat))
110 			srgb_cntl |= (1 << i);
111 
112 		offset = fd_resource_offset(rsc, psurf->u.tex.level,
113 				psurf->u.tex.first_layer);
114 
115 		stride = fd_resource_pitch(rsc, psurf->u.tex.level);
116 		swap = fd6_resource_swap(rsc, pformat);
117 
118 		tile_mode = fd_resource_tile_mode(psurf->texture, psurf->u.tex.level);
119 		max_layer_index = psurf->u.tex.last_layer - psurf->u.tex.first_layer;
120 
121 		debug_assert((offset + slice->size0) <= fd_bo_size(rsc->bo));
122 
123 		OUT_REG(ring,
124 			A6XX_RB_MRT_BUF_INFO(i,
125 				.color_format = format,
126 				.color_tile_mode = tile_mode,
127 				.color_swap = swap),
128 			A6XX_RB_MRT_PITCH(i, .a6xx_rb_mrt_pitch = stride),
129 			A6XX_RB_MRT_ARRAY_PITCH(i, .a6xx_rb_mrt_array_pitch = slice->size0),
130 			A6XX_RB_MRT_BASE(i, .bo = rsc->bo, .bo_offset = offset),
131 			A6XX_RB_MRT_BASE_GMEM(i, .unknown = base));
132 
133 		OUT_REG(ring,
134 				A6XX_SP_FS_MRT_REG(i, .color_format = format,
135 						.color_sint = sint, .color_uint = uint));
136 
137 		OUT_PKT4(ring, REG_A6XX_RB_MRT_FLAG_BUFFER(i), 3);
138 		fd6_emit_flag_reference(ring, rsc,
139 				psurf->u.tex.level, psurf->u.tex.first_layer);
140 	}
141 
142 	OUT_REG(ring, A6XX_RB_SRGB_CNTL(.dword = srgb_cntl));
143 	OUT_REG(ring, A6XX_SP_SRGB_CNTL(.dword = srgb_cntl));
144 
145 	OUT_REG(ring, A6XX_RB_RENDER_COMPONENTS(
146 		.rt0 = mrt_comp[0],
147 		.rt1 = mrt_comp[1],
148 		.rt2 = mrt_comp[2],
149 		.rt3 = mrt_comp[3],
150 		.rt4 = mrt_comp[4],
151 		.rt5 = mrt_comp[5],
152 		.rt6 = mrt_comp[6],
153 		.rt7 = mrt_comp[7]));
154 
155 	OUT_REG(ring, A6XX_SP_FS_RENDER_COMPONENTS(
156 		.rt0 = mrt_comp[0],
157 		.rt1 = mrt_comp[1],
158 		.rt2 = mrt_comp[2],
159 		.rt3 = mrt_comp[3],
160 		.rt4 = mrt_comp[4],
161 		.rt5 = mrt_comp[5],
162 		.rt6 = mrt_comp[6],
163 		.rt7 = mrt_comp[7]));
164 
165 	OUT_REG(ring, A6XX_GRAS_MAX_LAYER_INDEX(max_layer_index));
166 }
167 
168 static void
emit_zs(struct fd_ringbuffer * ring,struct pipe_surface * zsbuf,const struct fd_gmem_stateobj * gmem)169 emit_zs(struct fd_ringbuffer *ring, struct pipe_surface *zsbuf,
170 		const struct fd_gmem_stateobj *gmem)
171 {
172 	if (zsbuf) {
173 		struct fd_resource *rsc = fd_resource(zsbuf->texture);
174 		enum a6xx_depth_format fmt = fd6_pipe2depth(zsbuf->format);
175 		uint32_t stride = fd_resource_pitch(rsc, 0);
176 		uint32_t size = fd_resource_slice(rsc, 0)->size0;
177 		uint32_t base = gmem ? gmem->zsbuf_base[0] : 0;
178 		uint32_t offset = fd_resource_offset(rsc, zsbuf->u.tex.level,
179 				zsbuf->u.tex.first_layer);
180 
181 		OUT_REG(ring,
182 			A6XX_RB_DEPTH_BUFFER_INFO(.depth_format = fmt),
183 			A6XX_RB_DEPTH_BUFFER_PITCH(.a6xx_rb_depth_buffer_pitch = stride),
184 			A6XX_RB_DEPTH_BUFFER_ARRAY_PITCH(.a6xx_rb_depth_buffer_array_pitch = size),
185 			A6XX_RB_DEPTH_BUFFER_BASE(.bo = rsc->bo, .bo_offset = offset),
186 			A6XX_RB_DEPTH_BUFFER_BASE_GMEM(.dword = base));
187 
188 		OUT_REG(ring, A6XX_GRAS_SU_DEPTH_BUFFER_INFO(.depth_format = fmt));
189 
190 		OUT_PKT4(ring, REG_A6XX_RB_DEPTH_FLAG_BUFFER_BASE_LO, 3);
191 		fd6_emit_flag_reference(ring, rsc,
192 				zsbuf->u.tex.level, zsbuf->u.tex.first_layer);
193 
194 		if (rsc->lrz) {
195 			OUT_REG(ring,
196 				A6XX_GRAS_LRZ_BUFFER_BASE(.bo = rsc->lrz),
197 				A6XX_GRAS_LRZ_BUFFER_PITCH(.pitch = rsc->lrz_pitch),
198 				// XXX a6xx seems to use a different buffer here.. not sure what for..
199 				A6XX_GRAS_LRZ_FAST_CLEAR_BUFFER_BASE_LO(0),
200 				A6XX_GRAS_LRZ_FAST_CLEAR_BUFFER_BASE_HI(0));
201 		} else {
202 			OUT_PKT4(ring, REG_A6XX_GRAS_LRZ_BUFFER_BASE_LO, 5);
203 			OUT_RING(ring, 0x00000000);
204 			OUT_RING(ring, 0x00000000);
205 			OUT_RING(ring, 0x00000000);     /* GRAS_LRZ_BUFFER_PITCH */
206 			OUT_RING(ring, 0x00000000);     /* GRAS_LRZ_FAST_CLEAR_BUFFER_BASE_LO */
207 			OUT_RING(ring, 0x00000000);
208 		}
209 
210 		/* NOTE: blob emits GRAS_LRZ_CNTL plus GRAZ_LRZ_BUFFER_BASE
211 		 * plus this CP_EVENT_WRITE at the end in it's own IB..
212 		 */
213 		OUT_PKT7(ring, CP_EVENT_WRITE, 1);
214 		OUT_RING(ring, CP_EVENT_WRITE_0_EVENT(UNK_25));
215 
216 		if (rsc->stencil) {
217 			stride = fd_resource_pitch(rsc->stencil, 0);
218 			size = fd_resource_slice(rsc->stencil, 0)->size0;
219 			uint32_t base = gmem ? gmem->zsbuf_base[1] : 0;
220 
221 			OUT_REG(ring,
222 				A6XX_RB_STENCIL_INFO(.separate_stencil = true),
223 				A6XX_RB_STENCIL_BUFFER_PITCH(.a6xx_rb_stencil_buffer_pitch = stride),
224 				A6XX_RB_STENCIL_BUFFER_ARRAY_PITCH(.a6xx_rb_stencil_buffer_array_pitch = size),
225 				A6XX_RB_STENCIL_BUFFER_BASE(.bo = rsc->stencil->bo),
226 				A6XX_RB_STENCIL_BUFFER_BASE_GMEM(.dword = base));
227 		} else {
228 			OUT_REG(ring, A6XX_RB_STENCIL_INFO(0));
229 		}
230 	} else {
231 		OUT_PKT4(ring, REG_A6XX_RB_DEPTH_BUFFER_INFO, 6);
232 		OUT_RING(ring, A6XX_RB_DEPTH_BUFFER_INFO_DEPTH_FORMAT(DEPTH6_NONE));
233 		OUT_RING(ring, 0x00000000);    /* RB_DEPTH_BUFFER_PITCH */
234 		OUT_RING(ring, 0x00000000);    /* RB_DEPTH_BUFFER_ARRAY_PITCH */
235 		OUT_RING(ring, 0x00000000);    /* RB_DEPTH_BUFFER_BASE_LO */
236 		OUT_RING(ring, 0x00000000);    /* RB_DEPTH_BUFFER_BASE_HI */
237 		OUT_RING(ring, 0x00000000);    /* RB_DEPTH_BUFFER_BASE_GMEM */
238 
239 		OUT_REG(ring, A6XX_GRAS_SU_DEPTH_BUFFER_INFO(.depth_format = DEPTH6_NONE));
240 
241 		OUT_PKT4(ring, REG_A6XX_GRAS_LRZ_BUFFER_BASE_LO, 5);
242 		OUT_RING(ring, 0x00000000);    /* RB_DEPTH_FLAG_BUFFER_BASE_LO */
243 		OUT_RING(ring, 0x00000000);    /* RB_DEPTH_FLAG_BUFFER_BASE_HI */
244 		OUT_RING(ring, 0x00000000);    /* GRAS_LRZ_BUFFER_PITCH */
245 		OUT_RING(ring, 0x00000000);    /* GRAS_LRZ_FAST_CLEAR_BUFFER_BASE_LO */
246 		OUT_RING(ring, 0x00000000);    /* GRAS_LRZ_FAST_CLEAR_BUFFER_BASE_HI */
247 
248 		OUT_REG(ring, A6XX_RB_STENCIL_INFO(0));
249 	}
250 }
251 
252 static bool
use_hw_binning(struct fd_batch * batch)253 use_hw_binning(struct fd_batch *batch)
254 {
255 	const struct fd_gmem_stateobj *gmem = batch->gmem_state;
256 
257 	if ((gmem->maxpw * gmem->maxph) > 32)
258 		return false;
259 
260 	return fd_binning_enabled && ((gmem->nbins_x * gmem->nbins_y) >= 2) &&
261 			(batch->num_draws > 0);
262 }
263 
264 static void
patch_fb_read(struct fd_batch * batch)265 patch_fb_read(struct fd_batch *batch)
266 {
267 	const struct fd_gmem_stateobj *gmem = batch->gmem_state;
268 
269 	for (unsigned i = 0; i < fd_patch_num_elements(&batch->fb_read_patches); i++) {
270 		struct fd_cs_patch *patch = fd_patch_element(&batch->fb_read_patches, i);
271 		*patch->cs = patch->val | A6XX_TEX_CONST_2_PITCH(gmem->bin_w * gmem->cbuf_cpp[0]);
272 	}
273 	util_dynarray_clear(&batch->fb_read_patches);
274 }
275 
276 static void
update_render_cntl(struct fd_batch * batch,struct pipe_framebuffer_state * pfb,bool binning)277 update_render_cntl(struct fd_batch *batch, struct pipe_framebuffer_state *pfb, bool binning)
278 {
279 	struct fd_ringbuffer *ring = batch->gmem;
280 	uint32_t cntl = 0;
281 	bool depth_ubwc_enable = false;
282 	uint32_t mrts_ubwc_enable = 0;
283 	int i;
284 
285 	if (pfb->zsbuf) {
286 		struct fd_resource *rsc = fd_resource(pfb->zsbuf->texture);
287 		depth_ubwc_enable = fd_resource_ubwc_enabled(rsc, pfb->zsbuf->u.tex.level);
288 	}
289 
290 	for (i = 0; i < pfb->nr_cbufs; i++) {
291 		if (!pfb->cbufs[i])
292 			continue;
293 
294 		struct pipe_surface *psurf = pfb->cbufs[i];
295 		struct fd_resource *rsc = fd_resource(psurf->texture);
296 		if (!rsc->bo)
297 			continue;
298 
299 		if (fd_resource_ubwc_enabled(rsc, psurf->u.tex.level))
300 			mrts_ubwc_enable |= 1 << i;
301 	}
302 
303 	cntl |= A6XX_RB_RENDER_CNTL_UNK4;
304 	if (binning)
305 		cntl |= A6XX_RB_RENDER_CNTL_BINNING;
306 
307 	OUT_PKT7(ring, CP_REG_WRITE, 3);
308 	OUT_RING(ring, CP_REG_WRITE_0_TRACKER(TRACK_RENDER_CNTL));
309 	OUT_RING(ring, REG_A6XX_RB_RENDER_CNTL);
310 	OUT_RING(ring, cntl |
311 		COND(depth_ubwc_enable, A6XX_RB_RENDER_CNTL_FLAG_DEPTH) |
312 		A6XX_RB_RENDER_CNTL_FLAG_MRTS(mrts_ubwc_enable));
313 }
314 
315 /* extra size to store VSC_DRAW_STRM_SIZE: */
316 #define VSC_DRAW_STRM_SIZE(pitch)  ((pitch) * 32 + 0x100)
317 #define VSC_PRIM_STRM_SIZE(pitch) ((pitch) * 32)
318 
319 static void
update_vsc_pipe(struct fd_batch * batch)320 update_vsc_pipe(struct fd_batch *batch)
321 {
322 	struct fd_context *ctx = batch->ctx;
323 	struct fd6_context *fd6_ctx = fd6_context(ctx);
324 	const struct fd_gmem_stateobj *gmem = batch->gmem_state;
325 	struct fd_ringbuffer *ring = batch->gmem;
326 	int i;
327 
328 	if (batch->draw_strm_bits/8 > fd6_ctx->vsc_draw_strm_pitch) {
329 		if (fd6_ctx->vsc_draw_strm)
330 			fd_bo_del(fd6_ctx->vsc_draw_strm);
331 		fd6_ctx->vsc_draw_strm = NULL;
332 		/* Note: probably only need to align to 0x40, but aligning stronger
333 		 * reduces the odds that we will have to realloc again on the next
334 		 * frame:
335 		 */
336 		fd6_ctx->vsc_draw_strm_pitch = align(batch->draw_strm_bits/8, 0x4000);
337 		debug_printf("pre-resize VSC_DRAW_STRM_PITCH to: 0x%x\n",
338 				fd6_ctx->vsc_draw_strm_pitch);
339 	}
340 
341 	if (batch->prim_strm_bits/8 > fd6_ctx->vsc_prim_strm_pitch) {
342 		if (fd6_ctx->vsc_prim_strm)
343 			fd_bo_del(fd6_ctx->vsc_prim_strm);
344 		fd6_ctx->vsc_prim_strm = NULL;
345 		fd6_ctx->vsc_prim_strm_pitch = align(batch->prim_strm_bits/8, 0x4000);
346 		debug_printf("pre-resize VSC_PRIM_STRM_PITCH to: 0x%x\n",
347 				fd6_ctx->vsc_prim_strm_pitch);
348 	}
349 
350 	if (!fd6_ctx->vsc_draw_strm) {
351 		fd6_ctx->vsc_draw_strm = fd_bo_new(ctx->screen->dev,
352 			VSC_DRAW_STRM_SIZE(fd6_ctx->vsc_draw_strm_pitch),
353 			DRM_FREEDRENO_GEM_TYPE_KMEM, "vsc_draw_strm");
354 	}
355 
356 	if (!fd6_ctx->vsc_prim_strm) {
357 		fd6_ctx->vsc_prim_strm = fd_bo_new(ctx->screen->dev,
358 			VSC_PRIM_STRM_SIZE(fd6_ctx->vsc_prim_strm_pitch),
359 			DRM_FREEDRENO_GEM_TYPE_KMEM, "vsc_prim_strm");
360 	}
361 
362 	OUT_REG(ring,
363 		A6XX_VSC_BIN_SIZE(.width = gmem->bin_w, .height = gmem->bin_h),
364 		A6XX_VSC_DRAW_STRM_SIZE_ADDRESS(
365 			.bo = fd6_ctx->vsc_draw_strm,
366 			.bo_offset = 32 * fd6_ctx->vsc_draw_strm_pitch));
367 
368 	OUT_REG(ring, A6XX_VSC_BIN_COUNT(.nx = gmem->nbins_x,
369 					.ny = gmem->nbins_y));
370 
371 	OUT_PKT4(ring, REG_A6XX_VSC_PIPE_CONFIG_REG(0), 32);
372 	for (i = 0; i < 32; i++) {
373 		const struct fd_vsc_pipe *pipe = &gmem->vsc_pipe[i];
374 		OUT_RING(ring, A6XX_VSC_PIPE_CONFIG_REG_X(pipe->x) |
375 				A6XX_VSC_PIPE_CONFIG_REG_Y(pipe->y) |
376 				A6XX_VSC_PIPE_CONFIG_REG_W(pipe->w) |
377 				A6XX_VSC_PIPE_CONFIG_REG_H(pipe->h));
378 	}
379 
380 	OUT_REG(ring,
381 		A6XX_VSC_PRIM_STRM_ADDRESS(.bo = fd6_ctx->vsc_prim_strm),
382 		A6XX_VSC_PRIM_STRM_PITCH(.dword = fd6_ctx->vsc_prim_strm_pitch),
383 		A6XX_VSC_PRIM_STRM_LIMIT(.dword = fd6_ctx->vsc_prim_strm_pitch - 64));
384 
385 	OUT_REG(ring,
386 		A6XX_VSC_DRAW_STRM_ADDRESS(.bo = fd6_ctx->vsc_draw_strm),
387 		A6XX_VSC_DRAW_STRM_PITCH(.dword = fd6_ctx->vsc_draw_strm_pitch),
388 		A6XX_VSC_DRAW_STRM_LIMIT(.dword = fd6_ctx->vsc_draw_strm_pitch - 64));
389 }
390 
391 /*
392  * If overflow is detected, either 0x1 (VSC_DRAW_STRM overflow) or 0x3
393  * (VSC_PRIM_STRM overflow) plus the size of the overflowed buffer is
394  * written to control->vsc_overflow.  This allows the CPU to
395  * detect which buffer overflowed (and, since the current size is
396  * encoded as well, this protects against already-submitted but
397  * not executed batches from fooling the CPU into increasing the
398  * size again unnecessarily).
399  */
400 static void
emit_vsc_overflow_test(struct fd_batch * batch)401 emit_vsc_overflow_test(struct fd_batch *batch)
402 {
403 	struct fd_ringbuffer *ring = batch->gmem;
404 	const struct fd_gmem_stateobj *gmem = batch->gmem_state;
405 	struct fd6_context *fd6_ctx = fd6_context(batch->ctx);
406 
407 	debug_assert((fd6_ctx->vsc_draw_strm_pitch & 0x3) == 0);
408 	debug_assert((fd6_ctx->vsc_prim_strm_pitch & 0x3) == 0);
409 
410 	/* Check for overflow, write vsc_scratch if detected: */
411 	for (int i = 0; i < gmem->num_vsc_pipes; i++) {
412 		OUT_PKT7(ring, CP_COND_WRITE5, 8);
413 		OUT_RING(ring, CP_COND_WRITE5_0_FUNCTION(WRITE_GE) |
414 				CP_COND_WRITE5_0_WRITE_MEMORY);
415 		OUT_RING(ring, CP_COND_WRITE5_1_POLL_ADDR_LO(REG_A6XX_VSC_DRAW_STRM_SIZE_REG(i)));
416 		OUT_RING(ring, CP_COND_WRITE5_2_POLL_ADDR_HI(0));
417 		OUT_RING(ring, CP_COND_WRITE5_3_REF(fd6_ctx->vsc_draw_strm_pitch - 64));
418 		OUT_RING(ring, CP_COND_WRITE5_4_MASK(~0));
419 		OUT_RELOC(ring, control_ptr(fd6_ctx, vsc_overflow));  /* WRITE_ADDR_LO/HI */
420 		OUT_RING(ring, CP_COND_WRITE5_7_WRITE_DATA(1 + fd6_ctx->vsc_draw_strm_pitch));
421 
422 		OUT_PKT7(ring, CP_COND_WRITE5, 8);
423 		OUT_RING(ring, CP_COND_WRITE5_0_FUNCTION(WRITE_GE) |
424 				CP_COND_WRITE5_0_WRITE_MEMORY);
425 		OUT_RING(ring, CP_COND_WRITE5_1_POLL_ADDR_LO(REG_A6XX_VSC_PRIM_STRM_SIZE_REG(i)));
426 		OUT_RING(ring, CP_COND_WRITE5_2_POLL_ADDR_HI(0));
427 		OUT_RING(ring, CP_COND_WRITE5_3_REF(fd6_ctx->vsc_prim_strm_pitch - 64));
428 		OUT_RING(ring, CP_COND_WRITE5_4_MASK(~0));
429 		OUT_RELOC(ring, control_ptr(fd6_ctx, vsc_overflow));  /* WRITE_ADDR_LO/HI */
430 		OUT_RING(ring, CP_COND_WRITE5_7_WRITE_DATA(3 + fd6_ctx->vsc_prim_strm_pitch));
431 	}
432 
433 	OUT_PKT7(ring, CP_WAIT_MEM_WRITES, 0);
434 }
435 
436 static void
check_vsc_overflow(struct fd_context * ctx)437 check_vsc_overflow(struct fd_context *ctx)
438 {
439 	struct fd6_context *fd6_ctx = fd6_context(ctx);
440 	struct fd6_control *control = fd_bo_map(fd6_ctx->control_mem);
441 	uint32_t vsc_overflow = control->vsc_overflow;
442 
443 	if (!vsc_overflow)
444 		return;
445 
446 	/* clear overflow flag: */
447 	control->vsc_overflow = 0;
448 
449 	unsigned buffer = vsc_overflow & 0x3;
450 	unsigned size = vsc_overflow & ~0x3;
451 
452 	if (buffer == 0x1) {
453 		/* VSC_DRAW_STRM overflow: */
454 
455 		if (size < fd6_ctx->vsc_draw_strm_pitch) {
456 			/* we've already increased the size, this overflow is
457 			 * from a batch submitted before resize, but executed
458 			 * after
459 			 */
460 			return;
461 		}
462 
463 		fd_bo_del(fd6_ctx->vsc_draw_strm);
464 		fd6_ctx->vsc_draw_strm = NULL;
465 		fd6_ctx->vsc_draw_strm_pitch *= 2;
466 
467 		debug_printf("resized VSC_DRAW_STRM_PITCH to: 0x%x\n",
468 				fd6_ctx->vsc_draw_strm_pitch);
469 
470 	} else if (buffer == 0x3) {
471 		/* VSC_PRIM_STRM overflow: */
472 
473 		if (size < fd6_ctx->vsc_prim_strm_pitch) {
474 			/* we've already increased the size */
475 			return;
476 		}
477 
478 		fd_bo_del(fd6_ctx->vsc_prim_strm);
479 		fd6_ctx->vsc_prim_strm = NULL;
480 		fd6_ctx->vsc_prim_strm_pitch *= 2;
481 
482 		debug_printf("resized VSC_PRIM_STRM_PITCH to: 0x%x\n",
483 				fd6_ctx->vsc_prim_strm_pitch);
484 
485 	} else {
486 		/* NOTE: it's possible, for example, for overflow to corrupt the
487 		 * control page.  I mostly just see this hit if I set initial VSC
488 		 * buffer size extremely small.  Things still seem to recover,
489 		 * but maybe we should pre-emptively realloc vsc_data/vsc_data2
490 		 * and hope for different memory placement?
491 		 */
492 		DBG("invalid vsc_overflow value: 0x%08x", vsc_overflow);
493 	}
494 }
495 
496 /*
497  * Emit conditional CP_INDIRECT_BRANCH based on VSC_STATE[p], ie. the IB
498  * is skipped for tiles that have no visible geometry.
499  */
500 static void
emit_conditional_ib(struct fd_batch * batch,const struct fd_tile * tile,struct fd_ringbuffer * target)501 emit_conditional_ib(struct fd_batch *batch, const struct fd_tile *tile,
502 		struct fd_ringbuffer *target)
503 {
504 	struct fd_ringbuffer *ring = batch->gmem;
505 
506 	if (target->cur == target->start)
507 		return;
508 
509 	emit_marker6(ring, 6);
510 
511 	unsigned count = fd_ringbuffer_cmd_count(target);
512 
513 	BEGIN_RING(ring, 5 + 4 * count);  /* ensure conditional doesn't get split */
514 
515 	OUT_PKT7(ring, CP_REG_TEST, 1);
516 	OUT_RING(ring, A6XX_CP_REG_TEST_0_REG(REG_A6XX_VSC_STATE_REG(tile->p)) |
517 			A6XX_CP_REG_TEST_0_BIT(tile->n) |
518 			A6XX_CP_REG_TEST_0_WAIT_FOR_ME);
519 
520 	OUT_PKT7(ring, CP_COND_REG_EXEC, 2);
521 	OUT_RING(ring, CP_COND_REG_EXEC_0_MODE(PRED_TEST));
522 	OUT_RING(ring, CP_COND_REG_EXEC_1_DWORDS(4 * count));
523 
524 	for (unsigned i = 0; i < count; i++) {
525 		uint32_t dwords;
526 		OUT_PKT7(ring, CP_INDIRECT_BUFFER, 3);
527 		dwords = fd_ringbuffer_emit_reloc_ring_full(ring, target, i) / 4;
528 		assert(dwords > 0);
529 		OUT_RING(ring, dwords);
530 	}
531 
532 	emit_marker6(ring, 6);
533 }
534 
535 static void
set_scissor(struct fd_ringbuffer * ring,uint32_t x1,uint32_t y1,uint32_t x2,uint32_t y2)536 set_scissor(struct fd_ringbuffer *ring, uint32_t x1, uint32_t y1, uint32_t x2, uint32_t y2)
537 {
538 	OUT_REG(ring,
539 			A6XX_GRAS_SC_WINDOW_SCISSOR_TL(.x = x1, .y = y1),
540 			A6XX_GRAS_SC_WINDOW_SCISSOR_BR(.x = x2, .y = y2));
541 
542 	OUT_REG(ring,
543 			A6XX_GRAS_2D_RESOLVE_CNTL_1(.x = x1, .y = y1),
544 			A6XX_GRAS_2D_RESOLVE_CNTL_2(.x = x2, .y = y2));
545 }
546 
547 static void
set_bin_size(struct fd_ringbuffer * ring,uint32_t w,uint32_t h,uint32_t flag)548 set_bin_size(struct fd_ringbuffer *ring, uint32_t w, uint32_t h, uint32_t flag)
549 {
550 	OUT_REG(ring, A6XX_GRAS_BIN_CONTROL(.binw = w, .binh = h, .dword = flag));
551 	OUT_REG(ring, A6XX_RB_BIN_CONTROL(.binw = w, .binh = h, .dword = flag));
552 	/* no flag for RB_BIN_CONTROL2... */
553 	OUT_REG(ring, A6XX_RB_BIN_CONTROL2(.binw = w, .binh = h));
554 }
555 
556 static void
emit_binning_pass(struct fd_batch * batch)557 emit_binning_pass(struct fd_batch *batch)
558 {
559 	struct fd_ringbuffer *ring = batch->gmem;
560 	const struct fd_gmem_stateobj *gmem = batch->gmem_state;
561 	struct fd_screen *screen = batch->ctx->screen;
562 
563 	debug_assert(!batch->tessellation);
564 
565 	set_scissor(ring, 0, 0, gmem->width - 1, gmem->height - 1);
566 
567 	emit_marker6(ring, 7);
568 	OUT_PKT7(ring, CP_SET_MARKER, 1);
569 	OUT_RING(ring, A6XX_CP_SET_MARKER_0_MODE(RM6_BINNING));
570 	emit_marker6(ring, 7);
571 
572 	OUT_PKT7(ring, CP_SET_VISIBILITY_OVERRIDE, 1);
573 	OUT_RING(ring, 0x1);
574 
575 	OUT_PKT7(ring, CP_SET_MODE, 1);
576 	OUT_RING(ring, 0x1);
577 
578 	OUT_WFI5(ring);
579 
580 	OUT_REG(ring, A6XX_VFD_MODE_CNTL(.binning_pass = true));
581 
582 	update_vsc_pipe(batch);
583 
584 	OUT_PKT4(ring, REG_A6XX_PC_UNKNOWN_9805, 1);
585 	OUT_RING(ring, screen->info.a6xx.magic.PC_UNKNOWN_9805);
586 
587 	OUT_PKT4(ring, REG_A6XX_SP_UNKNOWN_A0F8, 1);
588 	OUT_RING(ring, screen->info.a6xx.magic.SP_UNKNOWN_A0F8);
589 
590 	OUT_PKT7(ring, CP_EVENT_WRITE, 1);
591 	OUT_RING(ring, UNK_2C);
592 
593 	OUT_PKT4(ring, REG_A6XX_RB_WINDOW_OFFSET, 1);
594 	OUT_RING(ring, A6XX_RB_WINDOW_OFFSET_X(0) |
595 			A6XX_RB_WINDOW_OFFSET_Y(0));
596 
597 	OUT_PKT4(ring, REG_A6XX_SP_TP_WINDOW_OFFSET, 1);
598 	OUT_RING(ring, A6XX_SP_TP_WINDOW_OFFSET_X(0) |
599 			A6XX_SP_TP_WINDOW_OFFSET_Y(0));
600 
601 	/* emit IB to binning drawcmds: */
602 	fd_log(batch, "GMEM: START BINNING IB");
603 	fd6_emit_ib(ring, batch->draw);
604 	fd_log(batch, "GMEM: END BINNING IB");
605 
606 	fd_reset_wfi(batch);
607 
608 	OUT_PKT7(ring, CP_SET_DRAW_STATE, 3);
609 	OUT_RING(ring, CP_SET_DRAW_STATE__0_COUNT(0) |
610 			CP_SET_DRAW_STATE__0_DISABLE_ALL_GROUPS |
611 			CP_SET_DRAW_STATE__0_GROUP_ID(0));
612 	OUT_RING(ring, CP_SET_DRAW_STATE__1_ADDR_LO(0));
613 	OUT_RING(ring, CP_SET_DRAW_STATE__2_ADDR_HI(0));
614 
615 	OUT_PKT7(ring, CP_EVENT_WRITE, 1);
616 	OUT_RING(ring, UNK_2D);
617 
618 	fd6_cache_inv(batch, ring);
619 	fd6_cache_flush(batch, ring);
620 	fd_wfi(batch, ring);
621 
622 	OUT_PKT7(ring, CP_WAIT_FOR_ME, 0);
623 
624 	fd_log(batch, "START VSC OVERFLOW TEST");
625 	emit_vsc_overflow_test(batch);
626 	fd_log(batch, "END VSC OVERFLOW TEST");
627 
628 	OUT_PKT7(ring, CP_SET_VISIBILITY_OVERRIDE, 1);
629 	OUT_RING(ring, 0x0);
630 
631 	OUT_PKT7(ring, CP_SET_MODE, 1);
632 	OUT_RING(ring, 0x0);
633 
634 	OUT_WFI5(ring);
635 
636 	OUT_REG(ring,
637 			A6XX_RB_CCU_CNTL(.offset = screen->info.a6xx.ccu_offset_gmem,
638 							 .gmem = true,
639 							 .unk2 = screen->info.a6xx.ccu_cntl_gmem_unk2));
640 }
641 
642 static void
emit_msaa(struct fd_ringbuffer * ring,unsigned nr)643 emit_msaa(struct fd_ringbuffer *ring, unsigned nr)
644 {
645 	enum a3xx_msaa_samples samples = fd_msaa_samples(nr);
646 
647 	OUT_PKT4(ring, REG_A6XX_SP_TP_RAS_MSAA_CNTL, 2);
648 	OUT_RING(ring, A6XX_SP_TP_RAS_MSAA_CNTL_SAMPLES(samples));
649 	OUT_RING(ring, A6XX_SP_TP_DEST_MSAA_CNTL_SAMPLES(samples) |
650 			 COND(samples == MSAA_ONE, A6XX_SP_TP_DEST_MSAA_CNTL_MSAA_DISABLE));
651 
652 	OUT_PKT4(ring, REG_A6XX_GRAS_RAS_MSAA_CNTL, 2);
653 	OUT_RING(ring, A6XX_GRAS_RAS_MSAA_CNTL_SAMPLES(samples));
654 	OUT_RING(ring, A6XX_GRAS_DEST_MSAA_CNTL_SAMPLES(samples) |
655 			 COND(samples == MSAA_ONE, A6XX_GRAS_DEST_MSAA_CNTL_MSAA_DISABLE));
656 
657 	OUT_PKT4(ring, REG_A6XX_RB_RAS_MSAA_CNTL, 2);
658 	OUT_RING(ring, A6XX_RB_RAS_MSAA_CNTL_SAMPLES(samples));
659 	OUT_RING(ring, A6XX_RB_DEST_MSAA_CNTL_SAMPLES(samples) |
660 			 COND(samples == MSAA_ONE, A6XX_RB_DEST_MSAA_CNTL_MSAA_DISABLE));
661 
662 	OUT_PKT4(ring, REG_A6XX_RB_MSAA_CNTL, 1);
663 	OUT_RING(ring, A6XX_RB_MSAA_CNTL_SAMPLES(samples));
664 }
665 
666 static void prepare_tile_setup_ib(struct fd_batch *batch);
667 static void prepare_tile_fini_ib(struct fd_batch *batch);
668 
669 /* before first tile */
670 static void
fd6_emit_tile_init(struct fd_batch * batch)671 fd6_emit_tile_init(struct fd_batch *batch)
672 {
673 	struct fd_ringbuffer *ring = batch->gmem;
674 	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
675 	const struct fd_gmem_stateobj *gmem = batch->gmem_state;
676 	struct fd_screen *screen = batch->ctx->screen;
677 
678 	fd6_emit_restore(batch, ring);
679 
680 	fd6_emit_lrz_flush(ring);
681 
682 	if (batch->prologue) {
683 		fd_log(batch, "START PROLOGUE");
684 		fd6_emit_ib(ring, batch->prologue);
685 		fd_log(batch, "END PROLOGUE");
686 	}
687 
688 	fd6_cache_inv(batch, ring);
689 
690 	prepare_tile_setup_ib(batch);
691 	prepare_tile_fini_ib(batch);
692 
693 	OUT_PKT7(ring, CP_SKIP_IB2_ENABLE_GLOBAL, 1);
694 	OUT_RING(ring, 0x0);
695 
696 	/* blob controls "local" in IB2, but I think that is not required */
697 	OUT_PKT7(ring, CP_SKIP_IB2_ENABLE_LOCAL, 1);
698 	OUT_RING(ring, 0x1);
699 
700 	fd_wfi(batch, ring);
701 	OUT_REG(ring,
702 			A6XX_RB_CCU_CNTL(.offset = screen->info.a6xx.ccu_offset_gmem,
703 							 .gmem = true,
704 							 .unk2 = screen->info.a6xx.ccu_cntl_gmem_unk2));
705 
706 	emit_zs(ring, pfb->zsbuf, batch->gmem_state);
707 	emit_mrt(ring, pfb, batch->gmem_state);
708 	emit_msaa(ring, pfb->samples);
709 	patch_fb_read(batch);
710 
711 	if (use_hw_binning(batch)) {
712 		/* enable stream-out during binning pass: */
713 		OUT_REG(ring, A6XX_VPC_SO_DISABLE(false));
714 
715 		set_bin_size(ring, gmem->bin_w, gmem->bin_h,
716 				A6XX_RB_BIN_CONTROL_BINNING_PASS | 0x6000000);
717 		update_render_cntl(batch, pfb, true);
718 		emit_binning_pass(batch);
719 
720 		/* and disable stream-out for draw pass: */
721 		OUT_REG(ring, A6XX_VPC_SO_DISABLE(true));
722 
723 		/*
724 		 * NOTE: even if we detect VSC overflow and disable use of
725 		 * visibility stream in draw pass, it is still safe to execute
726 		 * the reset of these cmds:
727 		 */
728 
729 // NOTE a618 not setting .USE_VIZ .. from a quick check on a630, it
730 // does not appear that this bit changes much (ie. it isn't actually
731 // .USE_VIZ like previous gens)
732 		set_bin_size(ring, gmem->bin_w, gmem->bin_h,
733 				A6XX_RB_BIN_CONTROL_USE_VIZ | 0x6000000);
734 
735 		OUT_PKT4(ring, REG_A6XX_VFD_MODE_CNTL, 1);
736 		OUT_RING(ring, 0x0);
737 
738 		OUT_PKT4(ring, REG_A6XX_PC_UNKNOWN_9805, 1);
739 		OUT_RING(ring, screen->info.a6xx.magic.PC_UNKNOWN_9805);
740 
741 		OUT_PKT4(ring, REG_A6XX_SP_UNKNOWN_A0F8, 1);
742 		OUT_RING(ring, screen->info.a6xx.magic.SP_UNKNOWN_A0F8);
743 
744 		OUT_PKT7(ring, CP_SKIP_IB2_ENABLE_GLOBAL, 1);
745 		OUT_RING(ring, 0x1);
746 	} else {
747 		/* no binning pass, so enable stream-out for draw pass:: */
748 		OUT_REG(ring, A6XX_VPC_SO_DISABLE(false));
749 
750 		set_bin_size(ring, gmem->bin_w, gmem->bin_h, 0x6000000);
751 	}
752 
753 	update_render_cntl(batch, pfb, false);
754 }
755 
756 static void
set_window_offset(struct fd_ringbuffer * ring,uint32_t x1,uint32_t y1)757 set_window_offset(struct fd_ringbuffer *ring, uint32_t x1, uint32_t y1)
758 {
759 	OUT_PKT4(ring, REG_A6XX_RB_WINDOW_OFFSET, 1);
760 	OUT_RING(ring, A6XX_RB_WINDOW_OFFSET_X(x1) |
761 			A6XX_RB_WINDOW_OFFSET_Y(y1));
762 
763 	OUT_PKT4(ring, REG_A6XX_RB_WINDOW_OFFSET2, 1);
764 	OUT_RING(ring, A6XX_RB_WINDOW_OFFSET2_X(x1) |
765 			A6XX_RB_WINDOW_OFFSET2_Y(y1));
766 
767 	OUT_PKT4(ring, REG_A6XX_SP_WINDOW_OFFSET, 1);
768 	OUT_RING(ring, A6XX_SP_WINDOW_OFFSET_X(x1) |
769 			A6XX_SP_WINDOW_OFFSET_Y(y1));
770 
771 	OUT_PKT4(ring, REG_A6XX_SP_TP_WINDOW_OFFSET, 1);
772 	OUT_RING(ring, A6XX_SP_TP_WINDOW_OFFSET_X(x1) |
773 			A6XX_SP_TP_WINDOW_OFFSET_Y(y1));
774 }
775 
776 /* before mem2gmem */
777 static void
fd6_emit_tile_prep(struct fd_batch * batch,const struct fd_tile * tile)778 fd6_emit_tile_prep(struct fd_batch *batch, const struct fd_tile *tile)
779 {
780 	struct fd_context *ctx = batch->ctx;
781 	const struct fd_gmem_stateobj *gmem = batch->gmem_state;
782 	struct fd6_context *fd6_ctx = fd6_context(ctx);
783 	struct fd_ringbuffer *ring = batch->gmem;
784 
785 	emit_marker6(ring, 7);
786 	OUT_PKT7(ring, CP_SET_MARKER, 1);
787 	OUT_RING(ring, A6XX_CP_SET_MARKER_0_MODE(RM6_GMEM));
788 	emit_marker6(ring, 7);
789 
790 	uint32_t x1 = tile->xoff;
791 	uint32_t y1 = tile->yoff;
792 	uint32_t x2 = tile->xoff + tile->bin_w - 1;
793 	uint32_t y2 = tile->yoff + tile->bin_h - 1;
794 
795 	set_scissor(ring, x1, y1, x2, y2);
796 
797 	if (use_hw_binning(batch)) {
798 		const struct fd_vsc_pipe *pipe = &gmem->vsc_pipe[tile->p];
799 
800 		OUT_PKT7(ring, CP_WAIT_FOR_ME, 0);
801 
802 		OUT_PKT7(ring, CP_SET_MODE, 1);
803 		OUT_RING(ring, 0x0);
804 
805 		OUT_PKT7(ring, CP_SET_BIN_DATA5, 7);
806 		OUT_RING(ring, CP_SET_BIN_DATA5_0_VSC_SIZE(pipe->w * pipe->h) |
807 				CP_SET_BIN_DATA5_0_VSC_N(tile->n));
808 		OUT_RELOC(ring, fd6_ctx->vsc_draw_strm,       /* per-pipe draw-stream address */
809 				(tile->p * fd6_ctx->vsc_draw_strm_pitch), 0, 0);
810 		OUT_RELOC(ring, fd6_ctx->vsc_draw_strm,       /* VSC_DRAW_STRM_ADDRESS + (p * 4) */
811 				(tile->p * 4) + (32 * fd6_ctx->vsc_draw_strm_pitch), 0, 0);
812 		OUT_RELOC(ring, fd6_ctx->vsc_prim_strm,
813 				(tile->p * fd6_ctx->vsc_prim_strm_pitch), 0, 0);
814 
815 		OUT_PKT7(ring, CP_SET_VISIBILITY_OVERRIDE, 1);
816 		OUT_RING(ring, 0x0);
817 
818 		set_window_offset(ring, x1, y1);
819 
820 		const struct fd_gmem_stateobj *gmem = batch->gmem_state;
821 		set_bin_size(ring, gmem->bin_w, gmem->bin_h, 0x6000000);
822 
823 		OUT_PKT7(ring, CP_SET_MODE, 1);
824 		OUT_RING(ring, 0x0);
825 	} else {
826 		set_window_offset(ring, x1, y1);
827 
828 		OUT_PKT7(ring, CP_SET_VISIBILITY_OVERRIDE, 1);
829 		OUT_RING(ring, 0x1);
830 
831 		OUT_PKT7(ring, CP_SET_MODE, 1);
832 		OUT_RING(ring, 0x0);
833 	}
834 }
835 
836 static void
set_blit_scissor(struct fd_batch * batch,struct fd_ringbuffer * ring)837 set_blit_scissor(struct fd_batch *batch, struct fd_ringbuffer *ring)
838 {
839 	struct pipe_scissor_state blit_scissor = batch->max_scissor;
840 
841 	blit_scissor.minx = ROUND_DOWN_TO(blit_scissor.minx, 16);
842 	blit_scissor.miny = ROUND_DOWN_TO(blit_scissor.miny, 4);
843 	blit_scissor.maxx = ALIGN(blit_scissor.maxx, 16);
844 	blit_scissor.maxy = ALIGN(blit_scissor.maxy, 4);
845 
846 	OUT_PKT4(ring, REG_A6XX_RB_BLIT_SCISSOR_TL, 2);
847 	OUT_RING(ring,
848 			 A6XX_RB_BLIT_SCISSOR_TL_X(blit_scissor.minx) |
849 			 A6XX_RB_BLIT_SCISSOR_TL_Y(blit_scissor.miny));
850 	OUT_RING(ring,
851 			 A6XX_RB_BLIT_SCISSOR_BR_X(blit_scissor.maxx - 1) |
852 			 A6XX_RB_BLIT_SCISSOR_BR_Y(blit_scissor.maxy - 1));
853 }
854 
855 static void
emit_blit(struct fd_batch * batch,struct fd_ringbuffer * ring,uint32_t base,struct pipe_surface * psurf,bool stencil)856 emit_blit(struct fd_batch *batch,
857 		  struct fd_ringbuffer *ring,
858 		  uint32_t base,
859 		  struct pipe_surface *psurf,
860 		  bool stencil)
861 {
862 	struct fd_resource *rsc = fd_resource(psurf->texture);
863 	enum pipe_format pfmt = psurf->format;
864 	uint32_t offset;
865 	bool ubwc_enabled;
866 
867 	debug_assert(psurf->u.tex.first_layer == psurf->u.tex.last_layer);
868 
869 	/* separate stencil case: */
870 	if (stencil) {
871 		rsc = rsc->stencil;
872 		pfmt = rsc->base.format;
873 	}
874 
875 	offset = fd_resource_offset(rsc, psurf->u.tex.level,
876 			psurf->u.tex.first_layer);
877 	ubwc_enabled = fd_resource_ubwc_enabled(rsc, psurf->u.tex.level);
878 
879 	debug_assert(psurf->u.tex.first_layer == psurf->u.tex.last_layer);
880 
881 	enum a6xx_format format = fd6_pipe2color(pfmt);
882 	uint32_t stride = fd_resource_pitch(rsc, psurf->u.tex.level);
883 	uint32_t size = fd_resource_slice(rsc, psurf->u.tex.level)->size0;
884 	enum a3xx_color_swap swap = fd6_resource_swap(rsc, pfmt);
885 	enum a3xx_msaa_samples samples =
886 			fd_msaa_samples(rsc->base.nr_samples);
887 	uint32_t tile_mode = fd_resource_tile_mode(&rsc->base, psurf->u.tex.level);
888 
889 	OUT_REG(ring,
890 		A6XX_RB_BLIT_DST_INFO(.tile_mode = tile_mode, .samples = samples,
891 			.color_format = format, .color_swap = swap, .flags = ubwc_enabled),
892 		A6XX_RB_BLIT_DST(.bo = rsc->bo, .bo_offset = offset),
893 		A6XX_RB_BLIT_DST_PITCH(.a6xx_rb_blit_dst_pitch = stride),
894 		A6XX_RB_BLIT_DST_ARRAY_PITCH(.a6xx_rb_blit_dst_array_pitch = size));
895 
896 	OUT_REG(ring, A6XX_RB_BLIT_BASE_GMEM(.dword = base));
897 
898 	if (ubwc_enabled) {
899 		OUT_PKT4(ring, REG_A6XX_RB_BLIT_FLAG_DST_LO, 3);
900 		fd6_emit_flag_reference(ring, rsc,
901 				psurf->u.tex.level, psurf->u.tex.first_layer);
902 	}
903 
904 	fd6_emit_blit(batch, ring);
905 }
906 
907 static void
emit_restore_blit(struct fd_batch * batch,struct fd_ringbuffer * ring,uint32_t base,struct pipe_surface * psurf,unsigned buffer)908 emit_restore_blit(struct fd_batch *batch,
909 				  struct fd_ringbuffer *ring,
910 				  uint32_t base,
911 				  struct pipe_surface *psurf,
912 				  unsigned buffer)
913 {
914 	bool stencil = (buffer == FD_BUFFER_STENCIL);
915 
916 	OUT_REG(ring, A6XX_RB_BLIT_INFO(
917 		.gmem = true, .unk0 = true,
918 		.depth = (buffer == FD_BUFFER_DEPTH),
919 		.integer = util_format_is_pure_integer(psurf->format)));
920 
921 	emit_blit(batch, ring, base, psurf, stencil);
922 }
923 
924 static void
emit_clears(struct fd_batch * batch,struct fd_ringbuffer * ring)925 emit_clears(struct fd_batch *batch, struct fd_ringbuffer *ring)
926 {
927 	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
928 	const struct fd_gmem_stateobj *gmem = batch->gmem_state;
929 	enum a3xx_msaa_samples samples = fd_msaa_samples(pfb->samples);
930 
931 	uint32_t buffers = batch->fast_cleared;
932 
933 	if (buffers & PIPE_CLEAR_COLOR) {
934 
935 		for (int i = 0; i < pfb->nr_cbufs; i++) {
936 			union pipe_color_union *color = &batch->clear_color[i];
937 			union util_color uc = {0};
938 
939 			if (!pfb->cbufs[i])
940 				continue;
941 
942 			if (!(buffers & (PIPE_CLEAR_COLOR0 << i)))
943 				continue;
944 
945 			enum pipe_format pfmt = pfb->cbufs[i]->format;
946 
947 			// XXX I think RB_CLEAR_COLOR_DWn wants to take into account SWAP??
948 			union pipe_color_union swapped;
949 			switch (fd6_pipe2swap(pfmt)) {
950 			case WZYX:
951 				swapped.ui[0] = color->ui[0];
952 				swapped.ui[1] = color->ui[1];
953 				swapped.ui[2] = color->ui[2];
954 				swapped.ui[3] = color->ui[3];
955 				break;
956 			case WXYZ:
957 				swapped.ui[2] = color->ui[0];
958 				swapped.ui[1] = color->ui[1];
959 				swapped.ui[0] = color->ui[2];
960 				swapped.ui[3] = color->ui[3];
961 				break;
962 			case ZYXW:
963 				swapped.ui[3] = color->ui[0];
964 				swapped.ui[0] = color->ui[1];
965 				swapped.ui[1] = color->ui[2];
966 				swapped.ui[2] = color->ui[3];
967 				break;
968 			case XYZW:
969 				swapped.ui[3] = color->ui[0];
970 				swapped.ui[2] = color->ui[1];
971 				swapped.ui[1] = color->ui[2];
972 				swapped.ui[0] = color->ui[3];
973 				break;
974 			}
975 
976 			util_pack_color_union(pfmt, &uc, &swapped);
977 
978 			OUT_PKT4(ring, REG_A6XX_RB_BLIT_DST_INFO, 1);
979 			OUT_RING(ring, A6XX_RB_BLIT_DST_INFO_TILE_MODE(TILE6_LINEAR) |
980 				A6XX_RB_BLIT_DST_INFO_SAMPLES(samples) |
981 				A6XX_RB_BLIT_DST_INFO_COLOR_FORMAT(fd6_pipe2color(pfmt)));
982 
983 			OUT_PKT4(ring, REG_A6XX_RB_BLIT_INFO, 1);
984 			OUT_RING(ring, A6XX_RB_BLIT_INFO_GMEM |
985 				A6XX_RB_BLIT_INFO_CLEAR_MASK(0xf));
986 
987 			OUT_PKT4(ring, REG_A6XX_RB_BLIT_BASE_GMEM, 1);
988 			OUT_RING(ring, gmem->cbuf_base[i]);
989 
990 			OUT_PKT4(ring, REG_A6XX_RB_UNKNOWN_88D0, 1);
991 			OUT_RING(ring, 0);
992 
993 			OUT_PKT4(ring, REG_A6XX_RB_BLIT_CLEAR_COLOR_DW0, 4);
994 			OUT_RING(ring, uc.ui[0]);
995 			OUT_RING(ring, uc.ui[1]);
996 			OUT_RING(ring, uc.ui[2]);
997 			OUT_RING(ring, uc.ui[3]);
998 
999 			fd6_emit_blit(batch, ring);
1000 		}
1001 	}
1002 
1003 	const bool has_depth = pfb->zsbuf;
1004 	const bool has_separate_stencil =
1005 		has_depth && fd_resource(pfb->zsbuf->texture)->stencil;
1006 
1007 	/* First clear depth or combined depth/stencil. */
1008 	if ((has_depth && (buffers & PIPE_CLEAR_DEPTH)) ||
1009 		(!has_separate_stencil && (buffers & PIPE_CLEAR_STENCIL))) {
1010 		enum pipe_format pfmt = pfb->zsbuf->format;
1011 		uint32_t clear_value;
1012 		uint32_t mask = 0;
1013 
1014 		if (has_separate_stencil) {
1015 			pfmt = util_format_get_depth_only(pfb->zsbuf->format);
1016 			clear_value = util_pack_z(pfmt, batch->clear_depth);
1017 		} else {
1018 			pfmt = pfb->zsbuf->format;
1019 			clear_value = util_pack_z_stencil(pfmt, batch->clear_depth,
1020 											  batch->clear_stencil);
1021 		}
1022 
1023 		if (buffers & PIPE_CLEAR_DEPTH)
1024 			mask |= 0x1;
1025 
1026 		if (!has_separate_stencil && (buffers & PIPE_CLEAR_STENCIL))
1027 			mask |= 0x2;
1028 
1029 		OUT_PKT4(ring, REG_A6XX_RB_BLIT_DST_INFO, 1);
1030 		OUT_RING(ring, A6XX_RB_BLIT_DST_INFO_TILE_MODE(TILE6_LINEAR) |
1031 			A6XX_RB_BLIT_DST_INFO_SAMPLES(samples) |
1032 			A6XX_RB_BLIT_DST_INFO_COLOR_FORMAT(fd6_pipe2color(pfmt)));
1033 
1034 		OUT_PKT4(ring, REG_A6XX_RB_BLIT_INFO, 1);
1035 		OUT_RING(ring, A6XX_RB_BLIT_INFO_GMEM |
1036 			// XXX UNK0 for separate stencil ??
1037 			A6XX_RB_BLIT_INFO_DEPTH |
1038 			A6XX_RB_BLIT_INFO_CLEAR_MASK(mask));
1039 
1040 		OUT_PKT4(ring, REG_A6XX_RB_BLIT_BASE_GMEM, 1);
1041 		OUT_RING(ring, gmem->zsbuf_base[0]);
1042 
1043 		OUT_PKT4(ring, REG_A6XX_RB_UNKNOWN_88D0, 1);
1044 		OUT_RING(ring, 0);
1045 
1046 		OUT_PKT4(ring, REG_A6XX_RB_BLIT_CLEAR_COLOR_DW0, 1);
1047 		OUT_RING(ring, clear_value);
1048 
1049 		fd6_emit_blit(batch, ring);
1050 	}
1051 
1052 	/* Then clear the separate stencil buffer in case of 32 bit depth
1053 	 * formats with separate stencil. */
1054 	if (has_separate_stencil && (buffers & PIPE_CLEAR_STENCIL)) {
1055 		OUT_PKT4(ring, REG_A6XX_RB_BLIT_DST_INFO, 1);
1056 		OUT_RING(ring, A6XX_RB_BLIT_DST_INFO_TILE_MODE(TILE6_LINEAR) |
1057 				 A6XX_RB_BLIT_DST_INFO_SAMPLES(samples) |
1058 				 A6XX_RB_BLIT_DST_INFO_COLOR_FORMAT(FMT6_8_UINT));
1059 
1060 		OUT_PKT4(ring, REG_A6XX_RB_BLIT_INFO, 1);
1061 		OUT_RING(ring, A6XX_RB_BLIT_INFO_GMEM |
1062 				 //A6XX_RB_BLIT_INFO_UNK0 |
1063 				 A6XX_RB_BLIT_INFO_DEPTH |
1064 				 A6XX_RB_BLIT_INFO_CLEAR_MASK(0x1));
1065 
1066 		OUT_PKT4(ring, REG_A6XX_RB_BLIT_BASE_GMEM, 1);
1067 		OUT_RING(ring, gmem->zsbuf_base[1]);
1068 
1069 		OUT_PKT4(ring, REG_A6XX_RB_UNKNOWN_88D0, 1);
1070 		OUT_RING(ring, 0);
1071 
1072 		OUT_PKT4(ring, REG_A6XX_RB_BLIT_CLEAR_COLOR_DW0, 1);
1073 		OUT_RING(ring, batch->clear_stencil & 0xff);
1074 
1075 		fd6_emit_blit(batch, ring);
1076 	}
1077 }
1078 
1079 /*
1080  * transfer from system memory to gmem
1081  */
1082 static void
emit_restore_blits(struct fd_batch * batch,struct fd_ringbuffer * ring)1083 emit_restore_blits(struct fd_batch *batch, struct fd_ringbuffer *ring)
1084 {
1085 	const struct fd_gmem_stateobj *gmem = batch->gmem_state;
1086 	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
1087 
1088 	if (batch->restore & FD_BUFFER_COLOR) {
1089 		unsigned i;
1090 		for (i = 0; i < pfb->nr_cbufs; i++) {
1091 			if (!pfb->cbufs[i])
1092 				continue;
1093 			if (!(batch->restore & (PIPE_CLEAR_COLOR0 << i)))
1094 				continue;
1095 			emit_restore_blit(batch, ring, gmem->cbuf_base[i], pfb->cbufs[i],
1096 							  FD_BUFFER_COLOR);
1097 		}
1098 	}
1099 
1100 	if (batch->restore & (FD_BUFFER_DEPTH | FD_BUFFER_STENCIL)) {
1101 		struct fd_resource *rsc = fd_resource(pfb->zsbuf->texture);
1102 
1103 		if (!rsc->stencil || (batch->restore & FD_BUFFER_DEPTH)) {
1104 			emit_restore_blit(batch, ring, gmem->zsbuf_base[0], pfb->zsbuf,
1105 							  FD_BUFFER_DEPTH);
1106 		}
1107 		if (rsc->stencil && (batch->restore & FD_BUFFER_STENCIL)) {
1108 			emit_restore_blit(batch, ring, gmem->zsbuf_base[1], pfb->zsbuf,
1109 							  FD_BUFFER_STENCIL);
1110 		}
1111 	}
1112 }
1113 
1114 static void
prepare_tile_setup_ib(struct fd_batch * batch)1115 prepare_tile_setup_ib(struct fd_batch *batch)
1116 {
1117 	if (!(batch->restore || batch->fast_cleared))
1118 		return;
1119 
1120 	batch->tile_setup = fd_submit_new_ringbuffer(batch->submit, 0x1000,
1121 			FD_RINGBUFFER_STREAMING);
1122 
1123 	set_blit_scissor(batch, batch->tile_setup);
1124 
1125 	emit_restore_blits(batch, batch->tile_setup);
1126 	emit_clears(batch, batch->tile_setup);
1127 }
1128 
1129 /*
1130  * transfer from system memory to gmem
1131  */
1132 static void
fd6_emit_tile_mem2gmem(struct fd_batch * batch,const struct fd_tile * tile)1133 fd6_emit_tile_mem2gmem(struct fd_batch *batch, const struct fd_tile *tile)
1134 {
1135 }
1136 
1137 /* before IB to rendering cmds: */
1138 static void
fd6_emit_tile_renderprep(struct fd_batch * batch,const struct fd_tile * tile)1139 fd6_emit_tile_renderprep(struct fd_batch *batch, const struct fd_tile *tile)
1140 {
1141 	if (!batch->tile_setup)
1142 		return;
1143 
1144 	fd_log(batch, "TILE: START CLEAR/RESTORE");
1145 	if (batch->fast_cleared || !use_hw_binning(batch)) {
1146 		fd6_emit_ib(batch->gmem, batch->tile_setup);
1147 	} else {
1148 		emit_conditional_ib(batch, tile, batch->tile_setup);
1149 	}
1150 	fd_log(batch, "TILE: END CLEAR/RESTORE");
1151 }
1152 
1153 static void
emit_resolve_blit(struct fd_batch * batch,struct fd_ringbuffer * ring,uint32_t base,struct pipe_surface * psurf,unsigned buffer)1154 emit_resolve_blit(struct fd_batch *batch,
1155 				  struct fd_ringbuffer *ring,
1156 				  uint32_t base,
1157 				  struct pipe_surface *psurf,
1158 				  unsigned buffer)
1159 {
1160 	uint32_t info = 0;
1161 	bool stencil = false;
1162 
1163 	if (!fd_resource(psurf->texture)->valid)
1164 		return;
1165 
1166 	switch (buffer) {
1167 	case FD_BUFFER_COLOR:
1168 		break;
1169 	case FD_BUFFER_STENCIL:
1170 		info |= A6XX_RB_BLIT_INFO_UNK0;
1171 		stencil = true;
1172 		break;
1173 	case FD_BUFFER_DEPTH:
1174 		info |= A6XX_RB_BLIT_INFO_DEPTH;
1175 		break;
1176 	}
1177 
1178 	if (util_format_is_pure_integer(psurf->format))
1179 		info |= A6XX_RB_BLIT_INFO_INTEGER;
1180 
1181 	OUT_PKT4(ring, REG_A6XX_RB_BLIT_INFO, 1);
1182 	OUT_RING(ring, info);
1183 
1184 	emit_blit(batch, ring, base, psurf, stencil);
1185 }
1186 
1187 /*
1188  * transfer from gmem to system memory (ie. normal RAM)
1189  */
1190 
1191 static void
prepare_tile_fini_ib(struct fd_batch * batch)1192 prepare_tile_fini_ib(struct fd_batch *batch)
1193 {
1194 	const struct fd_gmem_stateobj *gmem = batch->gmem_state;
1195 	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
1196 	struct fd_ringbuffer *ring;
1197 
1198 	batch->tile_fini = fd_submit_new_ringbuffer(batch->submit, 0x1000,
1199 			FD_RINGBUFFER_STREAMING);
1200 	ring = batch->tile_fini;
1201 
1202 	set_blit_scissor(batch, ring);
1203 
1204 	if (batch->resolve & (FD_BUFFER_DEPTH | FD_BUFFER_STENCIL)) {
1205 		struct fd_resource *rsc = fd_resource(pfb->zsbuf->texture);
1206 
1207 		if (!rsc->stencil || (batch->resolve & FD_BUFFER_DEPTH)) {
1208 			emit_resolve_blit(batch, ring,
1209 							  gmem->zsbuf_base[0], pfb->zsbuf,
1210 							  FD_BUFFER_DEPTH);
1211 		}
1212 		if (rsc->stencil && (batch->resolve & FD_BUFFER_STENCIL)) {
1213 			emit_resolve_blit(batch, ring,
1214 							  gmem->zsbuf_base[1], pfb->zsbuf,
1215 							  FD_BUFFER_STENCIL);
1216 		}
1217 	}
1218 
1219 	if (batch->resolve & FD_BUFFER_COLOR) {
1220 		unsigned i;
1221 		for (i = 0; i < pfb->nr_cbufs; i++) {
1222 			if (!pfb->cbufs[i])
1223 				continue;
1224 			if (!(batch->resolve & (PIPE_CLEAR_COLOR0 << i)))
1225 				continue;
1226 			emit_resolve_blit(batch, ring, gmem->cbuf_base[i], pfb->cbufs[i],
1227 							  FD_BUFFER_COLOR);
1228 		}
1229 	}
1230 }
1231 
1232 static void
fd6_emit_tile(struct fd_batch * batch,const struct fd_tile * tile)1233 fd6_emit_tile(struct fd_batch *batch, const struct fd_tile *tile)
1234 {
1235 	if (!use_hw_binning(batch)) {
1236 		fd6_emit_ib(batch->gmem, batch->draw);
1237 	} else {
1238 		emit_conditional_ib(batch, tile, batch->draw);
1239 	}
1240 
1241 	if (batch->epilogue)
1242 		fd6_emit_ib(batch->gmem, batch->epilogue);
1243 }
1244 
1245 static void
fd6_emit_tile_gmem2mem(struct fd_batch * batch,const struct fd_tile * tile)1246 fd6_emit_tile_gmem2mem(struct fd_batch *batch, const struct fd_tile *tile)
1247 {
1248 	struct fd_ringbuffer *ring = batch->gmem;
1249 
1250 	if (use_hw_binning(batch)) {
1251 		OUT_PKT7(ring, CP_SET_MARKER, 1);
1252 		OUT_RING(ring, A6XX_CP_SET_MARKER_0_MODE(RM6_ENDVIS));
1253 	}
1254 
1255 	OUT_PKT7(ring, CP_SET_DRAW_STATE, 3);
1256 	OUT_RING(ring, CP_SET_DRAW_STATE__0_COUNT(0) |
1257 			CP_SET_DRAW_STATE__0_DISABLE_ALL_GROUPS |
1258 			CP_SET_DRAW_STATE__0_GROUP_ID(0));
1259 	OUT_RING(ring, CP_SET_DRAW_STATE__1_ADDR_LO(0));
1260 	OUT_RING(ring, CP_SET_DRAW_STATE__2_ADDR_HI(0));
1261 
1262 	OUT_PKT7(ring, CP_SKIP_IB2_ENABLE_LOCAL, 1);
1263 	OUT_RING(ring, 0x0);
1264 
1265 	emit_marker6(ring, 7);
1266 	OUT_PKT7(ring, CP_SET_MARKER, 1);
1267 	OUT_RING(ring, A6XX_CP_SET_MARKER_0_MODE(RM6_RESOLVE));
1268 	emit_marker6(ring, 7);
1269 
1270 	fd_log(batch, "TILE: START RESOLVE");
1271 	if (batch->fast_cleared || !use_hw_binning(batch)) {
1272 		fd6_emit_ib(batch->gmem, batch->tile_fini);
1273 	} else {
1274 		emit_conditional_ib(batch, tile, batch->tile_fini);
1275 	}
1276 	fd_log(batch, "TILE: END RESOLVE");
1277 }
1278 
1279 static void
fd6_emit_tile_fini(struct fd_batch * batch)1280 fd6_emit_tile_fini(struct fd_batch *batch)
1281 {
1282 	struct fd_ringbuffer *ring = batch->gmem;
1283 
1284 	OUT_PKT4(ring, REG_A6XX_GRAS_LRZ_CNTL, 1);
1285 	OUT_RING(ring, A6XX_GRAS_LRZ_CNTL_ENABLE);
1286 
1287 	fd6_emit_lrz_flush(ring);
1288 
1289 	fd6_event_write(batch, ring, PC_CCU_RESOLVE_TS, true);
1290 
1291 	if (use_hw_binning(batch)) {
1292 		check_vsc_overflow(batch->ctx);
1293 	}
1294 }
1295 
1296 static void
emit_sysmem_clears(struct fd_batch * batch,struct fd_ringbuffer * ring)1297 emit_sysmem_clears(struct fd_batch *batch, struct fd_ringbuffer *ring)
1298 {
1299 	struct fd_context *ctx = batch->ctx;
1300 	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
1301 
1302 	uint32_t buffers = batch->fast_cleared;
1303 
1304 	if (buffers & PIPE_CLEAR_COLOR) {
1305 		for (int i = 0; i < pfb->nr_cbufs; i++) {
1306 			union pipe_color_union *color = &batch->clear_color[i];
1307 
1308 			if (!pfb->cbufs[i])
1309 				continue;
1310 
1311 			if (!(buffers & (PIPE_CLEAR_COLOR0 << i)))
1312 				continue;
1313 
1314 			fd6_clear_surface(ctx, ring,
1315 					pfb->cbufs[i], pfb->width, pfb->height, color);
1316 		}
1317 	}
1318 	if (buffers & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL)) {
1319 		union pipe_color_union value = {};
1320 
1321 		const bool has_depth = pfb->zsbuf;
1322 		struct pipe_resource *separate_stencil =
1323 			has_depth && fd_resource(pfb->zsbuf->texture)->stencil ?
1324 			&fd_resource(pfb->zsbuf->texture)->stencil->base : NULL;
1325 
1326 		if ((has_depth && (buffers & PIPE_CLEAR_DEPTH)) ||
1327 				(!separate_stencil && (buffers & PIPE_CLEAR_STENCIL))) {
1328 			value.f[0] = batch->clear_depth;
1329 			value.ui[1] = batch->clear_stencil;
1330 			fd6_clear_surface(ctx, ring,
1331 					pfb->zsbuf, pfb->width, pfb->height, &value);
1332 		}
1333 
1334 		if (separate_stencil && (buffers & PIPE_CLEAR_STENCIL)) {
1335 			value.ui[0] = batch->clear_stencil;
1336 
1337 			struct pipe_surface stencil_surf = *pfb->zsbuf;
1338 			stencil_surf.format = PIPE_FORMAT_S8_UINT;
1339 			stencil_surf.texture = separate_stencil;
1340 
1341 			fd6_clear_surface(ctx, ring,
1342 					&stencil_surf, pfb->width, pfb->height, &value);
1343 		}
1344 	}
1345 
1346 	fd6_event_write(batch, ring, PC_CCU_FLUSH_COLOR_TS, true);
1347 }
1348 
1349 static void
setup_tess_buffers(struct fd_batch * batch,struct fd_ringbuffer * ring)1350 setup_tess_buffers(struct fd_batch *batch, struct fd_ringbuffer *ring)
1351 {
1352 	struct fd_context *ctx = batch->ctx;
1353 
1354 	batch->tessfactor_bo = fd_bo_new(ctx->screen->dev,
1355 			batch->tessfactor_size,
1356 			DRM_FREEDRENO_GEM_TYPE_KMEM, "tessfactor");
1357 
1358 	batch->tessparam_bo = fd_bo_new(ctx->screen->dev,
1359 			batch->tessparam_size,
1360 			DRM_FREEDRENO_GEM_TYPE_KMEM, "tessparam");
1361 
1362 	OUT_PKT4(ring, REG_A6XX_PC_TESSFACTOR_ADDR_LO, 2);
1363 	OUT_RELOC(ring, batch->tessfactor_bo, 0, 0, 0);
1364 
1365 	batch->tess_addrs_constobj->cur = batch->tess_addrs_constobj->start;
1366 	OUT_RELOC(batch->tess_addrs_constobj, batch->tessparam_bo, 0, 0, 0);
1367 	OUT_RELOC(batch->tess_addrs_constobj, batch->tessfactor_bo, 0, 0, 0);
1368 }
1369 
1370 static void
fd6_emit_sysmem_prep(struct fd_batch * batch)1371 fd6_emit_sysmem_prep(struct fd_batch *batch)
1372 {
1373 	struct fd_ringbuffer *ring = batch->gmem;
1374 	struct fd_screen *screen = batch->ctx->screen;
1375 
1376 	fd6_emit_restore(batch, ring);
1377 	fd6_emit_lrz_flush(ring);
1378 
1379 	if (batch->prologue) {
1380 		fd_log(batch, "START PROLOGUE");
1381 		fd6_emit_ib(ring, batch->prologue);
1382 		fd_log(batch, "END PROLOGUE");
1383 	}
1384 
1385 	/* remaining setup below here does not apply to blit/compute: */
1386 	if (batch->nondraw)
1387 		return;
1388 
1389 	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
1390 
1391 	if (pfb->width > 0 && pfb->height > 0)
1392 		set_scissor(ring, 0, 0, pfb->width - 1, pfb->height - 1);
1393 	else
1394 		set_scissor(ring, 0, 0, 0, 0);
1395 
1396 	set_window_offset(ring, 0, 0);
1397 
1398 	set_bin_size(ring, 0, 0, 0xc00000); /* 0xc00000 = BYPASS? */
1399 
1400 	emit_sysmem_clears(batch, ring);
1401 
1402 	emit_marker6(ring, 7);
1403 	OUT_PKT7(ring, CP_SET_MARKER, 1);
1404 	OUT_RING(ring, A6XX_CP_SET_MARKER_0_MODE(RM6_BYPASS));
1405 	emit_marker6(ring, 7);
1406 
1407 	if (batch->tessellation)
1408 		setup_tess_buffers(batch, ring);
1409 
1410 	OUT_PKT7(ring, CP_SKIP_IB2_ENABLE_GLOBAL, 1);
1411 	OUT_RING(ring, 0x0);
1412 
1413 	/* blob controls "local" in IB2, but I think that is not required */
1414 	OUT_PKT7(ring, CP_SKIP_IB2_ENABLE_LOCAL, 1);
1415 	OUT_RING(ring, 0x1);
1416 
1417 	fd6_event_write(batch, ring, PC_CCU_INVALIDATE_COLOR, false);
1418 	fd6_cache_inv(batch, ring);
1419 
1420 	fd_wfi(batch, ring);
1421 	OUT_REG(ring, A6XX_RB_CCU_CNTL(.offset = screen->info.a6xx.ccu_offset_bypass));
1422 
1423 	/* enable stream-out, with sysmem there is only one pass: */
1424 	OUT_REG(ring, A6XX_VPC_SO_DISABLE(false));
1425 
1426 	OUT_PKT7(ring, CP_SET_VISIBILITY_OVERRIDE, 1);
1427 	OUT_RING(ring, 0x1);
1428 
1429 	emit_zs(ring, pfb->zsbuf, NULL);
1430 	emit_mrt(ring, pfb, NULL);
1431 	emit_msaa(ring, pfb->samples);
1432 
1433 	update_render_cntl(batch, pfb, false);
1434 }
1435 
1436 static void
fd6_emit_sysmem_fini(struct fd_batch * batch)1437 fd6_emit_sysmem_fini(struct fd_batch *batch)
1438 {
1439 	struct fd_ringbuffer *ring = batch->gmem;
1440 
1441 	if (batch->epilogue)
1442 		fd6_emit_ib(batch->gmem, batch->epilogue);
1443 
1444 	OUT_PKT7(ring, CP_SKIP_IB2_ENABLE_GLOBAL, 1);
1445 	OUT_RING(ring, 0x0);
1446 
1447 	fd6_emit_lrz_flush(ring);
1448 
1449 	fd6_event_write(batch, ring, PC_CCU_FLUSH_COLOR_TS, true);
1450 	fd6_event_write(batch, ring, PC_CCU_FLUSH_DEPTH_TS, true);
1451 }
1452 
1453 void
fd6_gmem_init(struct pipe_context * pctx)1454 fd6_gmem_init(struct pipe_context *pctx)
1455 {
1456 	struct fd_context *ctx = fd_context(pctx);
1457 
1458 	ctx->emit_tile_init = fd6_emit_tile_init;
1459 	ctx->emit_tile_prep = fd6_emit_tile_prep;
1460 	ctx->emit_tile_mem2gmem = fd6_emit_tile_mem2gmem;
1461 	ctx->emit_tile_renderprep = fd6_emit_tile_renderprep;
1462 	ctx->emit_tile = fd6_emit_tile;
1463 	ctx->emit_tile_gmem2mem = fd6_emit_tile_gmem2mem;
1464 	ctx->emit_tile_fini = fd6_emit_tile_fini;
1465 	ctx->emit_sysmem_prep = fd6_emit_sysmem_prep;
1466 	ctx->emit_sysmem_fini = fd6_emit_sysmem_fini;
1467 }
1468