1 /* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */
2
3 /*
4 * Copyright (C) 2013 Rob Clark <robclark@freedesktop.org>
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice (including the next
14 * paragraph) shall be included in all copies or substantial portions of the
15 * Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
23 * SOFTWARE.
24 *
25 * Authors:
26 * Rob Clark <robclark@freedesktop.org>
27 */
28
29 #include "pipe/p_state.h"
30 #include "util/u_string.h"
31 #include "util/u_memory.h"
32 #include "util/u_inlines.h"
33 #include "util/u_format.h"
34
35 #include "freedreno_draw.h"
36 #include "freedreno_state.h"
37 #include "freedreno_resource.h"
38
39 #include "fd3_gmem.h"
40 #include "fd3_context.h"
41 #include "fd3_emit.h"
42 #include "fd3_program.h"
43 #include "fd3_format.h"
44 #include "fd3_zsa.h"
45
46 static void
emit_mrt(struct fd_ringbuffer * ring,unsigned nr_bufs,struct pipe_surface ** bufs,uint32_t * bases,uint32_t bin_w,bool decode_srgb)47 emit_mrt(struct fd_ringbuffer *ring, unsigned nr_bufs,
48 struct pipe_surface **bufs, uint32_t *bases, uint32_t bin_w,
49 bool decode_srgb)
50 {
51 enum a3xx_tile_mode tile_mode;
52 unsigned i;
53
54 if (bin_w) {
55 tile_mode = TILE_32X32;
56 } else {
57 tile_mode = LINEAR;
58 }
59
60 for (i = 0; i < A3XX_MAX_RENDER_TARGETS; i++) {
61 enum pipe_format pformat = 0;
62 enum a3xx_color_fmt format = 0;
63 enum a3xx_color_swap swap = WZYX;
64 bool srgb = false;
65 struct fd_resource *rsc = NULL;
66 struct fd_resource_slice *slice = NULL;
67 uint32_t stride = 0;
68 uint32_t base = 0;
69 uint32_t offset = 0;
70
71 if ((i < nr_bufs) && bufs[i]) {
72 struct pipe_surface *psurf = bufs[i];
73
74 rsc = fd_resource(psurf->texture);
75 pformat = psurf->format;
76 /* In case we're drawing to Z32F_S8, the "color" actually goes to
77 * the stencil
78 */
79 if (rsc->stencil) {
80 rsc = rsc->stencil;
81 pformat = rsc->base.format;
82 if (bases)
83 bases++;
84 }
85 slice = fd_resource_slice(rsc, psurf->u.tex.level);
86 format = fd3_pipe2color(pformat);
87 swap = fd3_pipe2swap(pformat);
88 if (decode_srgb)
89 srgb = util_format_is_srgb(pformat);
90 else
91 pformat = util_format_linear(pformat);
92
93 debug_assert(psurf->u.tex.first_layer == psurf->u.tex.last_layer);
94
95 offset = fd_resource_offset(rsc, psurf->u.tex.level,
96 psurf->u.tex.first_layer);
97
98 if (bin_w) {
99 stride = bin_w * rsc->cpp;
100
101 if (bases) {
102 base = bases[i];
103 }
104 } else {
105 stride = slice->pitch * rsc->cpp;
106 }
107 } else if (i < nr_bufs && bases) {
108 base = bases[i];
109 }
110
111 OUT_PKT0(ring, REG_A3XX_RB_MRT_BUF_INFO(i), 2);
112 OUT_RING(ring, A3XX_RB_MRT_BUF_INFO_COLOR_FORMAT(format) |
113 A3XX_RB_MRT_BUF_INFO_COLOR_TILE_MODE(tile_mode) |
114 A3XX_RB_MRT_BUF_INFO_COLOR_BUF_PITCH(stride) |
115 A3XX_RB_MRT_BUF_INFO_COLOR_SWAP(swap) |
116 COND(srgb, A3XX_RB_MRT_BUF_INFO_COLOR_SRGB));
117 if (bin_w || (i >= nr_bufs) || !bufs[i]) {
118 OUT_RING(ring, A3XX_RB_MRT_BUF_BASE_COLOR_BUF_BASE(base));
119 } else {
120 OUT_RELOCW(ring, rsc->bo, offset, 0, -1);
121 }
122
123 OUT_PKT0(ring, REG_A3XX_SP_FS_IMAGE_OUTPUT_REG(i), 1);
124 OUT_RING(ring, COND((i < nr_bufs) && bufs[i],
125 A3XX_SP_FS_IMAGE_OUTPUT_REG_MRTFORMAT(
126 fd3_fs_output_format(pformat))));
127 }
128 }
129
130 static bool
use_hw_binning(struct fd_batch * batch)131 use_hw_binning(struct fd_batch *batch)
132 {
133 struct fd_gmem_stateobj *gmem = &batch->ctx->gmem;
134
135 /* workaround: combining scissor optimization and hw binning
136 * seems problematic. Seems like we end up with a mismatch
137 * between binning pass and rendering pass, wrt. where the hw
138 * thinks the vertices belong. And the blob driver doesn't
139 * seem to implement anything like scissor optimization, so
140 * not entirely sure what I might be missing.
141 *
142 * But scissor optimization is mainly for window managers,
143 * which don't have many vertices (and therefore doesn't
144 * benefit much from binning pass).
145 *
146 * So for now just disable binning if scissor optimization is
147 * used.
148 */
149 if (gmem->minx || gmem->miny)
150 return false;
151
152 if ((gmem->maxpw * gmem->maxph) > 32)
153 return false;
154
155 if ((gmem->maxpw > 15) || (gmem->maxph > 15))
156 return false;
157
158 return fd_binning_enabled && ((gmem->nbins_x * gmem->nbins_y) > 2);
159 }
160
161 /* workaround for (hlsq?) lockup with hw binning on a3xx patchlevel 0 */
162 static void update_vsc_pipe(struct fd_batch *batch);
163 static void
emit_binning_workaround(struct fd_batch * batch)164 emit_binning_workaround(struct fd_batch *batch)
165 {
166 struct fd_context *ctx = batch->ctx;
167 struct fd_gmem_stateobj *gmem = &ctx->gmem;
168 struct fd_ringbuffer *ring = batch->gmem;
169 struct fd3_emit emit = {
170 .debug = &ctx->debug,
171 .vtx = &ctx->solid_vbuf_state,
172 .prog = &ctx->solid_prog,
173 .key = {
174 .half_precision = true,
175 },
176 };
177
178 OUT_PKT0(ring, REG_A3XX_RB_MODE_CONTROL, 2);
179 OUT_RING(ring, A3XX_RB_MODE_CONTROL_RENDER_MODE(RB_RESOLVE_PASS) |
180 A3XX_RB_MODE_CONTROL_MARB_CACHE_SPLIT_MODE |
181 A3XX_RB_MODE_CONTROL_MRT(0));
182 OUT_RING(ring, A3XX_RB_RENDER_CONTROL_BIN_WIDTH(32) |
183 A3XX_RB_RENDER_CONTROL_DISABLE_COLOR_PIPE |
184 A3XX_RB_RENDER_CONTROL_ALPHA_TEST_FUNC(FUNC_NEVER));
185
186 OUT_PKT0(ring, REG_A3XX_RB_COPY_CONTROL, 4);
187 OUT_RING(ring, A3XX_RB_COPY_CONTROL_MSAA_RESOLVE(MSAA_ONE) |
188 A3XX_RB_COPY_CONTROL_MODE(0) |
189 A3XX_RB_COPY_CONTROL_GMEM_BASE(0));
190 OUT_RELOCW(ring, fd_resource(ctx->solid_vbuf)->bo, 0x20, 0, -1); /* RB_COPY_DEST_BASE */
191 OUT_RING(ring, A3XX_RB_COPY_DEST_PITCH_PITCH(128));
192 OUT_RING(ring, A3XX_RB_COPY_DEST_INFO_TILE(LINEAR) |
193 A3XX_RB_COPY_DEST_INFO_FORMAT(RB_R8G8B8A8_UNORM) |
194 A3XX_RB_COPY_DEST_INFO_SWAP(WZYX) |
195 A3XX_RB_COPY_DEST_INFO_COMPONENT_ENABLE(0xf) |
196 A3XX_RB_COPY_DEST_INFO_ENDIAN(ENDIAN_NONE));
197
198 OUT_PKT0(ring, REG_A3XX_GRAS_SC_CONTROL, 1);
199 OUT_RING(ring, A3XX_GRAS_SC_CONTROL_RENDER_MODE(RB_RESOLVE_PASS) |
200 A3XX_GRAS_SC_CONTROL_MSAA_SAMPLES(MSAA_ONE) |
201 A3XX_GRAS_SC_CONTROL_RASTER_MODE(1));
202
203 fd3_program_emit(ring, &emit, 0, NULL);
204 fd3_emit_vertex_bufs(ring, &emit);
205
206 OUT_PKT0(ring, REG_A3XX_HLSQ_CONTROL_0_REG, 4);
207 OUT_RING(ring, A3XX_HLSQ_CONTROL_0_REG_FSTHREADSIZE(FOUR_QUADS) |
208 A3XX_HLSQ_CONTROL_0_REG_FSSUPERTHREADENABLE |
209 A3XX_HLSQ_CONTROL_0_REG_RESERVED2 |
210 A3XX_HLSQ_CONTROL_0_REG_SPCONSTFULLUPDATE);
211 OUT_RING(ring, A3XX_HLSQ_CONTROL_1_REG_VSTHREADSIZE(TWO_QUADS) |
212 A3XX_HLSQ_CONTROL_1_REG_VSSUPERTHREADENABLE);
213 OUT_RING(ring, A3XX_HLSQ_CONTROL_2_REG_PRIMALLOCTHRESHOLD(31));
214 OUT_RING(ring, 0); /* HLSQ_CONTROL_3_REG */
215
216 OUT_PKT0(ring, REG_A3XX_HLSQ_CONST_FSPRESV_RANGE_REG, 1);
217 OUT_RING(ring, A3XX_HLSQ_CONST_FSPRESV_RANGE_REG_STARTENTRY(0x20) |
218 A3XX_HLSQ_CONST_FSPRESV_RANGE_REG_ENDENTRY(0x20));
219
220 OUT_PKT0(ring, REG_A3XX_RB_MSAA_CONTROL, 1);
221 OUT_RING(ring, A3XX_RB_MSAA_CONTROL_DISABLE |
222 A3XX_RB_MSAA_CONTROL_SAMPLES(MSAA_ONE) |
223 A3XX_RB_MSAA_CONTROL_SAMPLE_MASK(0xffff));
224
225 OUT_PKT0(ring, REG_A3XX_RB_DEPTH_CONTROL, 1);
226 OUT_RING(ring, A3XX_RB_DEPTH_CONTROL_ZFUNC(FUNC_NEVER));
227
228 OUT_PKT0(ring, REG_A3XX_RB_STENCIL_CONTROL, 1);
229 OUT_RING(ring, A3XX_RB_STENCIL_CONTROL_FUNC(FUNC_NEVER) |
230 A3XX_RB_STENCIL_CONTROL_FAIL(STENCIL_KEEP) |
231 A3XX_RB_STENCIL_CONTROL_ZPASS(STENCIL_KEEP) |
232 A3XX_RB_STENCIL_CONTROL_ZFAIL(STENCIL_KEEP) |
233 A3XX_RB_STENCIL_CONTROL_FUNC_BF(FUNC_NEVER) |
234 A3XX_RB_STENCIL_CONTROL_FAIL_BF(STENCIL_KEEP) |
235 A3XX_RB_STENCIL_CONTROL_ZPASS_BF(STENCIL_KEEP) |
236 A3XX_RB_STENCIL_CONTROL_ZFAIL_BF(STENCIL_KEEP));
237
238 OUT_PKT0(ring, REG_A3XX_GRAS_SU_MODE_CONTROL, 1);
239 OUT_RING(ring, A3XX_GRAS_SU_MODE_CONTROL_LINEHALFWIDTH(0.0));
240
241 OUT_PKT0(ring, REG_A3XX_VFD_INDEX_MIN, 4);
242 OUT_RING(ring, 0); /* VFD_INDEX_MIN */
243 OUT_RING(ring, 2); /* VFD_INDEX_MAX */
244 OUT_RING(ring, 0); /* VFD_INSTANCEID_OFFSET */
245 OUT_RING(ring, 0); /* VFD_INDEX_OFFSET */
246
247 OUT_PKT0(ring, REG_A3XX_PC_PRIM_VTX_CNTL, 1);
248 OUT_RING(ring, A3XX_PC_PRIM_VTX_CNTL_STRIDE_IN_VPC(0) |
249 A3XX_PC_PRIM_VTX_CNTL_POLYMODE_FRONT_PTYPE(PC_DRAW_TRIANGLES) |
250 A3XX_PC_PRIM_VTX_CNTL_POLYMODE_BACK_PTYPE(PC_DRAW_TRIANGLES) |
251 A3XX_PC_PRIM_VTX_CNTL_PROVOKING_VTX_LAST);
252
253 OUT_PKT0(ring, REG_A3XX_GRAS_SC_WINDOW_SCISSOR_TL, 2);
254 OUT_RING(ring, A3XX_GRAS_SC_WINDOW_SCISSOR_TL_X(0) |
255 A3XX_GRAS_SC_WINDOW_SCISSOR_TL_Y(1));
256 OUT_RING(ring, A3XX_GRAS_SC_WINDOW_SCISSOR_BR_X(0) |
257 A3XX_GRAS_SC_WINDOW_SCISSOR_BR_Y(1));
258
259 OUT_PKT0(ring, REG_A3XX_GRAS_SC_SCREEN_SCISSOR_TL, 2);
260 OUT_RING(ring, A3XX_GRAS_SC_SCREEN_SCISSOR_TL_X(0) |
261 A3XX_GRAS_SC_SCREEN_SCISSOR_TL_Y(0));
262 OUT_RING(ring, A3XX_GRAS_SC_SCREEN_SCISSOR_BR_X(31) |
263 A3XX_GRAS_SC_SCREEN_SCISSOR_BR_Y(0));
264
265 fd_wfi(batch, ring);
266 OUT_PKT0(ring, REG_A3XX_GRAS_CL_VPORT_XOFFSET, 6);
267 OUT_RING(ring, A3XX_GRAS_CL_VPORT_XOFFSET(0.0));
268 OUT_RING(ring, A3XX_GRAS_CL_VPORT_XSCALE(1.0));
269 OUT_RING(ring, A3XX_GRAS_CL_VPORT_YOFFSET(0.0));
270 OUT_RING(ring, A3XX_GRAS_CL_VPORT_YSCALE(1.0));
271 OUT_RING(ring, A3XX_GRAS_CL_VPORT_ZOFFSET(0.0));
272 OUT_RING(ring, A3XX_GRAS_CL_VPORT_ZSCALE(1.0));
273
274 OUT_PKT0(ring, REG_A3XX_GRAS_CL_CLIP_CNTL, 1);
275 OUT_RING(ring, A3XX_GRAS_CL_CLIP_CNTL_CLIP_DISABLE |
276 A3XX_GRAS_CL_CLIP_CNTL_ZFAR_CLIP_DISABLE |
277 A3XX_GRAS_CL_CLIP_CNTL_VP_CLIP_CODE_IGNORE |
278 A3XX_GRAS_CL_CLIP_CNTL_VP_XFORM_DISABLE |
279 A3XX_GRAS_CL_CLIP_CNTL_PERSP_DIVISION_DISABLE);
280
281 OUT_PKT0(ring, REG_A3XX_GRAS_CL_GB_CLIP_ADJ, 1);
282 OUT_RING(ring, A3XX_GRAS_CL_GB_CLIP_ADJ_HORZ(0) |
283 A3XX_GRAS_CL_GB_CLIP_ADJ_VERT(0));
284
285 OUT_PKT3(ring, CP_DRAW_INDX_2, 5);
286 OUT_RING(ring, 0x00000000); /* viz query info. */
287 OUT_RING(ring, DRAW(DI_PT_RECTLIST, DI_SRC_SEL_IMMEDIATE,
288 INDEX_SIZE_32_BIT, IGNORE_VISIBILITY, 0));
289 OUT_RING(ring, 2); /* NumIndices */
290 OUT_RING(ring, 2);
291 OUT_RING(ring, 1);
292 fd_reset_wfi(batch);
293
294 OUT_PKT0(ring, REG_A3XX_HLSQ_CONTROL_0_REG, 1);
295 OUT_RING(ring, A3XX_HLSQ_CONTROL_0_REG_FSTHREADSIZE(TWO_QUADS));
296
297 OUT_PKT0(ring, REG_A3XX_VFD_PERFCOUNTER0_SELECT, 1);
298 OUT_RING(ring, 0x00000000);
299
300 fd_wfi(batch, ring);
301 OUT_PKT0(ring, REG_A3XX_VSC_BIN_SIZE, 1);
302 OUT_RING(ring, A3XX_VSC_BIN_SIZE_WIDTH(gmem->bin_w) |
303 A3XX_VSC_BIN_SIZE_HEIGHT(gmem->bin_h));
304
305 OUT_PKT0(ring, REG_A3XX_GRAS_SC_CONTROL, 1);
306 OUT_RING(ring, A3XX_GRAS_SC_CONTROL_RENDER_MODE(RB_RENDERING_PASS) |
307 A3XX_GRAS_SC_CONTROL_MSAA_SAMPLES(MSAA_ONE) |
308 A3XX_GRAS_SC_CONTROL_RASTER_MODE(0));
309
310 OUT_PKT0(ring, REG_A3XX_GRAS_CL_CLIP_CNTL, 1);
311 OUT_RING(ring, 0x00000000);
312 }
313
314 /* transfer from gmem to system memory (ie. normal RAM) */
315
316 static void
emit_gmem2mem_surf(struct fd_batch * batch,enum adreno_rb_copy_control_mode mode,bool stencil,uint32_t base,struct pipe_surface * psurf)317 emit_gmem2mem_surf(struct fd_batch *batch,
318 enum adreno_rb_copy_control_mode mode,
319 bool stencil,
320 uint32_t base, struct pipe_surface *psurf)
321 {
322 struct fd_ringbuffer *ring = batch->gmem;
323 struct fd_resource *rsc = fd_resource(psurf->texture);
324 enum pipe_format format = psurf->format;
325 if (stencil) {
326 rsc = rsc->stencil;
327 format = rsc->base.format;
328 }
329 struct fd_resource_slice *slice = fd_resource_slice(rsc, psurf->u.tex.level);
330 uint32_t offset = fd_resource_offset(rsc, psurf->u.tex.level,
331 psurf->u.tex.first_layer);
332
333 debug_assert(psurf->u.tex.first_layer == psurf->u.tex.last_layer);
334
335 OUT_PKT0(ring, REG_A3XX_RB_COPY_CONTROL, 4);
336 OUT_RING(ring, A3XX_RB_COPY_CONTROL_MSAA_RESOLVE(MSAA_ONE) |
337 A3XX_RB_COPY_CONTROL_MODE(mode) |
338 A3XX_RB_COPY_CONTROL_GMEM_BASE(base) |
339 COND(format == PIPE_FORMAT_Z32_FLOAT ||
340 format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT,
341 A3XX_RB_COPY_CONTROL_DEPTH32_RESOLVE));
342
343 OUT_RELOCW(ring, rsc->bo, offset, 0, -1); /* RB_COPY_DEST_BASE */
344 OUT_RING(ring, A3XX_RB_COPY_DEST_PITCH_PITCH(slice->pitch * rsc->cpp));
345 OUT_RING(ring, A3XX_RB_COPY_DEST_INFO_TILE(LINEAR) |
346 A3XX_RB_COPY_DEST_INFO_FORMAT(fd3_pipe2color(format)) |
347 A3XX_RB_COPY_DEST_INFO_COMPONENT_ENABLE(0xf) |
348 A3XX_RB_COPY_DEST_INFO_ENDIAN(ENDIAN_NONE) |
349 A3XX_RB_COPY_DEST_INFO_SWAP(fd3_pipe2swap(format)));
350
351 fd_draw(batch, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
352 DI_SRC_SEL_AUTO_INDEX, 2, 0, INDEX_SIZE_IGN, 0, 0, NULL);
353 }
354
355 static void
fd3_emit_tile_gmem2mem(struct fd_batch * batch,struct fd_tile * tile)356 fd3_emit_tile_gmem2mem(struct fd_batch *batch, struct fd_tile *tile)
357 {
358 struct fd_context *ctx = batch->ctx;
359 struct fd_ringbuffer *ring = batch->gmem;
360 struct pipe_framebuffer_state *pfb = &batch->framebuffer;
361 struct fd3_emit emit = {
362 .debug = &ctx->debug,
363 .vtx = &ctx->solid_vbuf_state,
364 .prog = &ctx->solid_prog,
365 .key = {
366 .half_precision = true,
367 },
368 };
369 int i;
370
371 OUT_PKT0(ring, REG_A3XX_RB_DEPTH_CONTROL, 1);
372 OUT_RING(ring, A3XX_RB_DEPTH_CONTROL_ZFUNC(FUNC_NEVER));
373
374 OUT_PKT0(ring, REG_A3XX_RB_STENCIL_CONTROL, 1);
375 OUT_RING(ring, A3XX_RB_STENCIL_CONTROL_FUNC(FUNC_NEVER) |
376 A3XX_RB_STENCIL_CONTROL_FAIL(STENCIL_KEEP) |
377 A3XX_RB_STENCIL_CONTROL_ZPASS(STENCIL_KEEP) |
378 A3XX_RB_STENCIL_CONTROL_ZFAIL(STENCIL_KEEP) |
379 A3XX_RB_STENCIL_CONTROL_FUNC_BF(FUNC_NEVER) |
380 A3XX_RB_STENCIL_CONTROL_FAIL_BF(STENCIL_KEEP) |
381 A3XX_RB_STENCIL_CONTROL_ZPASS_BF(STENCIL_KEEP) |
382 A3XX_RB_STENCIL_CONTROL_ZFAIL_BF(STENCIL_KEEP));
383
384 OUT_PKT0(ring, REG_A3XX_RB_STENCILREFMASK, 2);
385 OUT_RING(ring, 0xff000000 |
386 A3XX_RB_STENCILREFMASK_STENCILREF(0) |
387 A3XX_RB_STENCILREFMASK_STENCILMASK(0) |
388 A3XX_RB_STENCILREFMASK_STENCILWRITEMASK(0xff));
389 OUT_RING(ring, 0xff000000 |
390 A3XX_RB_STENCILREFMASK_STENCILREF(0) |
391 A3XX_RB_STENCILREFMASK_STENCILMASK(0) |
392 A3XX_RB_STENCILREFMASK_STENCILWRITEMASK(0xff));
393
394 OUT_PKT0(ring, REG_A3XX_GRAS_SU_MODE_CONTROL, 1);
395 OUT_RING(ring, A3XX_GRAS_SU_MODE_CONTROL_LINEHALFWIDTH(0));
396
397 OUT_PKT0(ring, REG_A3XX_GRAS_CL_CLIP_CNTL, 1);
398 OUT_RING(ring, 0x00000000); /* GRAS_CL_CLIP_CNTL */
399
400 fd_wfi(batch, ring);
401 OUT_PKT0(ring, REG_A3XX_GRAS_CL_VPORT_XOFFSET, 6);
402 OUT_RING(ring, A3XX_GRAS_CL_VPORT_XOFFSET((float)pfb->width/2.0 - 0.5));
403 OUT_RING(ring, A3XX_GRAS_CL_VPORT_XSCALE((float)pfb->width/2.0));
404 OUT_RING(ring, A3XX_GRAS_CL_VPORT_YOFFSET((float)pfb->height/2.0 - 0.5));
405 OUT_RING(ring, A3XX_GRAS_CL_VPORT_YSCALE(-(float)pfb->height/2.0));
406 OUT_RING(ring, A3XX_GRAS_CL_VPORT_ZOFFSET(0.0));
407 OUT_RING(ring, A3XX_GRAS_CL_VPORT_ZSCALE(1.0));
408
409 OUT_PKT0(ring, REG_A3XX_RB_MODE_CONTROL, 1);
410 OUT_RING(ring, A3XX_RB_MODE_CONTROL_RENDER_MODE(RB_RESOLVE_PASS) |
411 A3XX_RB_MODE_CONTROL_MARB_CACHE_SPLIT_MODE |
412 A3XX_RB_MODE_CONTROL_MRT(0));
413
414 OUT_PKT0(ring, REG_A3XX_RB_RENDER_CONTROL, 1);
415 OUT_RING(ring, A3XX_RB_RENDER_CONTROL_DISABLE_COLOR_PIPE |
416 A3XX_RB_RENDER_CONTROL_ENABLE_GMEM |
417 A3XX_RB_RENDER_CONTROL_ALPHA_TEST_FUNC(FUNC_NEVER) |
418 A3XX_RB_RENDER_CONTROL_BIN_WIDTH(ctx->gmem.bin_w));
419
420 OUT_PKT0(ring, REG_A3XX_GRAS_SC_CONTROL, 1);
421 OUT_RING(ring, A3XX_GRAS_SC_CONTROL_RENDER_MODE(RB_RESOLVE_PASS) |
422 A3XX_GRAS_SC_CONTROL_MSAA_SAMPLES(MSAA_ONE) |
423 A3XX_GRAS_SC_CONTROL_RASTER_MODE(1));
424
425 OUT_PKT0(ring, REG_A3XX_PC_PRIM_VTX_CNTL, 1);
426 OUT_RING(ring, A3XX_PC_PRIM_VTX_CNTL_STRIDE_IN_VPC(0) |
427 A3XX_PC_PRIM_VTX_CNTL_POLYMODE_FRONT_PTYPE(PC_DRAW_TRIANGLES) |
428 A3XX_PC_PRIM_VTX_CNTL_POLYMODE_BACK_PTYPE(PC_DRAW_TRIANGLES) |
429 A3XX_PC_PRIM_VTX_CNTL_PROVOKING_VTX_LAST);
430
431 OUT_PKT0(ring, REG_A3XX_GRAS_SC_WINDOW_SCISSOR_TL, 2);
432 OUT_RING(ring, A3XX_GRAS_SC_WINDOW_SCISSOR_TL_X(0) |
433 A3XX_GRAS_SC_WINDOW_SCISSOR_TL_Y(0));
434 OUT_RING(ring, A3XX_GRAS_SC_WINDOW_SCISSOR_BR_X(pfb->width - 1) |
435 A3XX_GRAS_SC_WINDOW_SCISSOR_BR_Y(pfb->height - 1));
436
437 OUT_PKT0(ring, REG_A3XX_VFD_INDEX_MIN, 4);
438 OUT_RING(ring, 0); /* VFD_INDEX_MIN */
439 OUT_RING(ring, 2); /* VFD_INDEX_MAX */
440 OUT_RING(ring, 0); /* VFD_INSTANCEID_OFFSET */
441 OUT_RING(ring, 0); /* VFD_INDEX_OFFSET */
442
443 fd3_program_emit(ring, &emit, 0, NULL);
444 fd3_emit_vertex_bufs(ring, &emit);
445
446 if (batch->resolve & (FD_BUFFER_DEPTH | FD_BUFFER_STENCIL)) {
447 struct fd_resource *rsc = fd_resource(pfb->zsbuf->texture);
448 if (!rsc->stencil || batch->resolve & FD_BUFFER_DEPTH)
449 emit_gmem2mem_surf(batch, RB_COPY_DEPTH_STENCIL, false,
450 ctx->gmem.zsbuf_base[0], pfb->zsbuf);
451 if (rsc->stencil && batch->resolve & FD_BUFFER_STENCIL)
452 emit_gmem2mem_surf(batch, RB_COPY_DEPTH_STENCIL, true,
453 ctx->gmem.zsbuf_base[1], pfb->zsbuf);
454 }
455
456 if (batch->resolve & FD_BUFFER_COLOR) {
457 for (i = 0; i < pfb->nr_cbufs; i++) {
458 if (!pfb->cbufs[i])
459 continue;
460 if (!(batch->resolve & (PIPE_CLEAR_COLOR0 << i)))
461 continue;
462 emit_gmem2mem_surf(batch, RB_COPY_RESOLVE, false,
463 ctx->gmem.cbuf_base[i], pfb->cbufs[i]);
464 }
465 }
466
467 OUT_PKT0(ring, REG_A3XX_RB_MODE_CONTROL, 1);
468 OUT_RING(ring, A3XX_RB_MODE_CONTROL_RENDER_MODE(RB_RENDERING_PASS) |
469 A3XX_RB_MODE_CONTROL_MARB_CACHE_SPLIT_MODE |
470 A3XX_RB_MODE_CONTROL_MRT(MAX2(1, pfb->nr_cbufs) - 1));
471
472 OUT_PKT0(ring, REG_A3XX_GRAS_SC_CONTROL, 1);
473 OUT_RING(ring, A3XX_GRAS_SC_CONTROL_RENDER_MODE(RB_RENDERING_PASS) |
474 A3XX_GRAS_SC_CONTROL_MSAA_SAMPLES(MSAA_ONE) |
475 A3XX_GRAS_SC_CONTROL_RASTER_MODE(0));
476 }
477
478 /* transfer from system memory to gmem */
479
480 static void
emit_mem2gmem_surf(struct fd_batch * batch,uint32_t bases[],struct pipe_surface ** psurf,uint32_t bufs,uint32_t bin_w)481 emit_mem2gmem_surf(struct fd_batch *batch, uint32_t bases[],
482 struct pipe_surface **psurf, uint32_t bufs, uint32_t bin_w)
483 {
484 struct fd_ringbuffer *ring = batch->gmem;
485 struct pipe_surface *zsbufs[2];
486
487 assert(bufs > 0);
488
489 OUT_PKT0(ring, REG_A3XX_RB_MODE_CONTROL, 1);
490 OUT_RING(ring, A3XX_RB_MODE_CONTROL_RENDER_MODE(RB_RENDERING_PASS) |
491 A3XX_RB_MODE_CONTROL_MARB_CACHE_SPLIT_MODE |
492 A3XX_RB_MODE_CONTROL_MRT(bufs - 1));
493
494 emit_mrt(ring, bufs, psurf, bases, bin_w, false);
495
496 if (psurf[0] && (psurf[0]->format == PIPE_FORMAT_Z32_FLOAT ||
497 psurf[0]->format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT)) {
498 /* Depth is stored as unorm in gmem, so we have to write it in using a
499 * special blit shader which writes depth.
500 */
501 OUT_PKT0(ring, REG_A3XX_RB_DEPTH_CONTROL, 1);
502 OUT_RING(ring, (A3XX_RB_DEPTH_CONTROL_FRAG_WRITES_Z |
503 A3XX_RB_DEPTH_CONTROL_Z_WRITE_ENABLE |
504 A3XX_RB_DEPTH_CONTROL_Z_ENABLE |
505 A3XX_RB_DEPTH_CONTROL_EARLY_Z_DISABLE |
506 A3XX_RB_DEPTH_CONTROL_ZFUNC(FUNC_ALWAYS)));
507
508 OUT_PKT0(ring, REG_A3XX_RB_DEPTH_INFO, 2);
509 OUT_RING(ring, A3XX_RB_DEPTH_INFO_DEPTH_BASE(bases[0]) |
510 A3XX_RB_DEPTH_INFO_DEPTH_FORMAT(DEPTHX_32));
511 OUT_RING(ring, A3XX_RB_DEPTH_PITCH(4 * batch->ctx->gmem.bin_w));
512
513 if (psurf[0]->format == PIPE_FORMAT_Z32_FLOAT) {
514 OUT_PKT0(ring, REG_A3XX_RB_MRT_CONTROL(0), 1);
515 OUT_RING(ring, 0);
516 } else {
517 /* The gmem_restore_tex logic will put the first buffer's stencil
518 * as color. Supply it with the proper information to make that
519 * happen.
520 */
521 zsbufs[0] = zsbufs[1] = psurf[0];
522 psurf = zsbufs;
523 bufs = 2;
524 }
525 } else {
526 OUT_PKT0(ring, REG_A3XX_SP_FS_OUTPUT_REG, 1);
527 OUT_RING(ring, A3XX_SP_FS_OUTPUT_REG_MRT(bufs - 1));
528 }
529
530 fd3_emit_gmem_restore_tex(ring, psurf, bufs);
531
532 fd_draw(batch, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
533 DI_SRC_SEL_AUTO_INDEX, 2, 0, INDEX_SIZE_IGN, 0, 0, NULL);
534 }
535
536 static void
fd3_emit_tile_mem2gmem(struct fd_batch * batch,struct fd_tile * tile)537 fd3_emit_tile_mem2gmem(struct fd_batch *batch, struct fd_tile *tile)
538 {
539 struct fd_context *ctx = batch->ctx;
540 struct fd_gmem_stateobj *gmem = &ctx->gmem;
541 struct fd_ringbuffer *ring = batch->gmem;
542 struct pipe_framebuffer_state *pfb = &batch->framebuffer;
543 struct fd3_emit emit = {
544 .debug = &ctx->debug,
545 .vtx = &ctx->blit_vbuf_state,
546 .sprite_coord_enable = 1,
547 /* NOTE: They all use the same VP, this is for vtx bufs. */
548 .prog = &ctx->blit_prog[0],
549 .key = {
550 .half_precision = fd_half_precision(pfb),
551 },
552 };
553 float x0, y0, x1, y1;
554 unsigned bin_w = tile->bin_w;
555 unsigned bin_h = tile->bin_h;
556 unsigned i;
557
558 /* write texture coordinates to vertexbuf: */
559 x0 = ((float)tile->xoff) / ((float)pfb->width);
560 x1 = ((float)tile->xoff + bin_w) / ((float)pfb->width);
561 y0 = ((float)tile->yoff) / ((float)pfb->height);
562 y1 = ((float)tile->yoff + bin_h) / ((float)pfb->height);
563
564 OUT_PKT3(ring, CP_MEM_WRITE, 5);
565 OUT_RELOCW(ring, fd_resource(ctx->blit_texcoord_vbuf)->bo, 0, 0, 0);
566 OUT_RING(ring, fui(x0));
567 OUT_RING(ring, fui(y0));
568 OUT_RING(ring, fui(x1));
569 OUT_RING(ring, fui(y1));
570
571 fd3_emit_cache_flush(batch, ring);
572
573 for (i = 0; i < 4; i++) {
574 OUT_PKT0(ring, REG_A3XX_RB_MRT_CONTROL(i), 1);
575 OUT_RING(ring, A3XX_RB_MRT_CONTROL_ROP_CODE(ROP_COPY) |
576 A3XX_RB_MRT_CONTROL_DITHER_MODE(DITHER_DISABLE) |
577 A3XX_RB_MRT_CONTROL_COMPONENT_ENABLE(0xf));
578
579 OUT_PKT0(ring, REG_A3XX_RB_MRT_BLEND_CONTROL(i), 1);
580 OUT_RING(ring, A3XX_RB_MRT_BLEND_CONTROL_RGB_SRC_FACTOR(FACTOR_ONE) |
581 A3XX_RB_MRT_BLEND_CONTROL_RGB_BLEND_OPCODE(BLEND_DST_PLUS_SRC) |
582 A3XX_RB_MRT_BLEND_CONTROL_RGB_DEST_FACTOR(FACTOR_ZERO) |
583 A3XX_RB_MRT_BLEND_CONTROL_ALPHA_SRC_FACTOR(FACTOR_ONE) |
584 A3XX_RB_MRT_BLEND_CONTROL_ALPHA_BLEND_OPCODE(BLEND_DST_PLUS_SRC) |
585 A3XX_RB_MRT_BLEND_CONTROL_ALPHA_DEST_FACTOR(FACTOR_ZERO));
586 }
587
588 OUT_PKT0(ring, REG_A3XX_RB_RENDER_CONTROL, 1);
589 OUT_RING(ring, A3XX_RB_RENDER_CONTROL_ALPHA_TEST_FUNC(FUNC_ALWAYS) |
590 A3XX_RB_RENDER_CONTROL_BIN_WIDTH(gmem->bin_w));
591
592 fd_wfi(batch, ring);
593 OUT_PKT0(ring, REG_A3XX_RB_DEPTH_CONTROL, 1);
594 OUT_RING(ring, A3XX_RB_DEPTH_CONTROL_ZFUNC(FUNC_LESS));
595
596 OUT_PKT0(ring, REG_A3XX_RB_DEPTH_INFO, 2);
597 OUT_RING(ring, 0);
598 OUT_RING(ring, 0);
599
600 OUT_PKT0(ring, REG_A3XX_GRAS_CL_CLIP_CNTL, 1);
601 OUT_RING(ring, A3XX_GRAS_CL_CLIP_CNTL_IJ_PERSP_CENTER); /* GRAS_CL_CLIP_CNTL */
602
603 fd_wfi(batch, ring);
604 OUT_PKT0(ring, REG_A3XX_GRAS_CL_VPORT_XOFFSET, 6);
605 OUT_RING(ring, A3XX_GRAS_CL_VPORT_XOFFSET((float)bin_w/2.0 - 0.5));
606 OUT_RING(ring, A3XX_GRAS_CL_VPORT_XSCALE((float)bin_w/2.0));
607 OUT_RING(ring, A3XX_GRAS_CL_VPORT_YOFFSET((float)bin_h/2.0 - 0.5));
608 OUT_RING(ring, A3XX_GRAS_CL_VPORT_YSCALE(-(float)bin_h/2.0));
609 OUT_RING(ring, A3XX_GRAS_CL_VPORT_ZOFFSET(0.0));
610 OUT_RING(ring, A3XX_GRAS_CL_VPORT_ZSCALE(1.0));
611
612 OUT_PKT0(ring, REG_A3XX_GRAS_SC_WINDOW_SCISSOR_TL, 2);
613 OUT_RING(ring, A3XX_GRAS_SC_WINDOW_SCISSOR_TL_X(0) |
614 A3XX_GRAS_SC_WINDOW_SCISSOR_TL_Y(0));
615 OUT_RING(ring, A3XX_GRAS_SC_WINDOW_SCISSOR_BR_X(bin_w - 1) |
616 A3XX_GRAS_SC_WINDOW_SCISSOR_BR_Y(bin_h - 1));
617
618 OUT_PKT0(ring, REG_A3XX_GRAS_SC_SCREEN_SCISSOR_TL, 2);
619 OUT_RING(ring, A3XX_GRAS_SC_SCREEN_SCISSOR_TL_X(0) |
620 A3XX_GRAS_SC_SCREEN_SCISSOR_TL_Y(0));
621 OUT_RING(ring, A3XX_GRAS_SC_SCREEN_SCISSOR_BR_X(bin_w - 1) |
622 A3XX_GRAS_SC_SCREEN_SCISSOR_BR_Y(bin_h - 1));
623
624 OUT_PKT0(ring, REG_A3XX_RB_STENCIL_CONTROL, 1);
625 OUT_RING(ring, 0x2 |
626 A3XX_RB_STENCIL_CONTROL_FUNC(FUNC_ALWAYS) |
627 A3XX_RB_STENCIL_CONTROL_FAIL(STENCIL_KEEP) |
628 A3XX_RB_STENCIL_CONTROL_ZPASS(STENCIL_KEEP) |
629 A3XX_RB_STENCIL_CONTROL_ZFAIL(STENCIL_KEEP) |
630 A3XX_RB_STENCIL_CONTROL_FUNC_BF(FUNC_ALWAYS) |
631 A3XX_RB_STENCIL_CONTROL_FAIL_BF(STENCIL_KEEP) |
632 A3XX_RB_STENCIL_CONTROL_ZPASS_BF(STENCIL_KEEP) |
633 A3XX_RB_STENCIL_CONTROL_ZFAIL_BF(STENCIL_KEEP));
634
635 OUT_PKT0(ring, REG_A3XX_RB_STENCIL_INFO, 2);
636 OUT_RING(ring, 0); /* RB_STENCIL_INFO */
637 OUT_RING(ring, 0); /* RB_STENCIL_PITCH */
638
639 OUT_PKT0(ring, REG_A3XX_GRAS_SC_CONTROL, 1);
640 OUT_RING(ring, A3XX_GRAS_SC_CONTROL_RENDER_MODE(RB_RENDERING_PASS) |
641 A3XX_GRAS_SC_CONTROL_MSAA_SAMPLES(MSAA_ONE) |
642 A3XX_GRAS_SC_CONTROL_RASTER_MODE(1));
643
644 OUT_PKT0(ring, REG_A3XX_PC_PRIM_VTX_CNTL, 1);
645 OUT_RING(ring, A3XX_PC_PRIM_VTX_CNTL_STRIDE_IN_VPC(2) |
646 A3XX_PC_PRIM_VTX_CNTL_POLYMODE_FRONT_PTYPE(PC_DRAW_TRIANGLES) |
647 A3XX_PC_PRIM_VTX_CNTL_POLYMODE_BACK_PTYPE(PC_DRAW_TRIANGLES) |
648 A3XX_PC_PRIM_VTX_CNTL_PROVOKING_VTX_LAST);
649
650 OUT_PKT0(ring, REG_A3XX_VFD_INDEX_MIN, 4);
651 OUT_RING(ring, 0); /* VFD_INDEX_MIN */
652 OUT_RING(ring, 2); /* VFD_INDEX_MAX */
653 OUT_RING(ring, 0); /* VFD_INSTANCEID_OFFSET */
654 OUT_RING(ring, 0); /* VFD_INDEX_OFFSET */
655
656 fd3_emit_vertex_bufs(ring, &emit);
657
658 /* for gmem pitch/base calculations, we need to use the non-
659 * truncated tile sizes:
660 */
661 bin_w = gmem->bin_w;
662 bin_h = gmem->bin_h;
663
664 if (fd_gmem_needs_restore(batch, tile, FD_BUFFER_COLOR)) {
665 emit.prog = &ctx->blit_prog[pfb->nr_cbufs - 1];
666 emit.fp = NULL; /* frag shader changed so clear cache */
667 fd3_program_emit(ring, &emit, pfb->nr_cbufs, pfb->cbufs);
668 emit_mem2gmem_surf(batch, gmem->cbuf_base, pfb->cbufs, pfb->nr_cbufs, bin_w);
669 }
670
671 if (fd_gmem_needs_restore(batch, tile, FD_BUFFER_DEPTH | FD_BUFFER_STENCIL)) {
672 if (pfb->zsbuf->format != PIPE_FORMAT_Z32_FLOAT_S8X24_UINT &&
673 pfb->zsbuf->format != PIPE_FORMAT_Z32_FLOAT) {
674 /* Non-float can use a regular color write. It's split over 8-bit
675 * components, so half precision is always sufficient.
676 */
677 emit.prog = &ctx->blit_prog[0];
678 emit.key.half_precision = true;
679 } else {
680 /* Float depth needs special blit shader that writes depth */
681 if (pfb->zsbuf->format == PIPE_FORMAT_Z32_FLOAT)
682 emit.prog = &ctx->blit_z;
683 else
684 emit.prog = &ctx->blit_zs;
685 emit.key.half_precision = false;
686 }
687 emit.fp = NULL; /* frag shader changed so clear cache */
688 fd3_program_emit(ring, &emit, 1, &pfb->zsbuf);
689 emit_mem2gmem_surf(batch, gmem->zsbuf_base, &pfb->zsbuf, 1, bin_w);
690 }
691
692 OUT_PKT0(ring, REG_A3XX_GRAS_SC_CONTROL, 1);
693 OUT_RING(ring, A3XX_GRAS_SC_CONTROL_RENDER_MODE(RB_RENDERING_PASS) |
694 A3XX_GRAS_SC_CONTROL_MSAA_SAMPLES(MSAA_ONE) |
695 A3XX_GRAS_SC_CONTROL_RASTER_MODE(0));
696
697 OUT_PKT0(ring, REG_A3XX_RB_MODE_CONTROL, 1);
698 OUT_RING(ring, A3XX_RB_MODE_CONTROL_RENDER_MODE(RB_RENDERING_PASS) |
699 A3XX_RB_MODE_CONTROL_MARB_CACHE_SPLIT_MODE |
700 A3XX_RB_MODE_CONTROL_MRT(MAX2(1, pfb->nr_cbufs) - 1));
701 }
702
703 static void
patch_draws(struct fd_batch * batch,enum pc_di_vis_cull_mode vismode)704 patch_draws(struct fd_batch *batch, enum pc_di_vis_cull_mode vismode)
705 {
706 unsigned i;
707 for (i = 0; i < fd_patch_num_elements(&batch->draw_patches); i++) {
708 struct fd_cs_patch *patch = fd_patch_element(&batch->draw_patches, i);
709 *patch->cs = patch->val | DRAW(0, 0, 0, vismode, 0);
710 }
711 util_dynarray_resize(&batch->draw_patches, 0);
712 }
713
714 static void
patch_rbrc(struct fd_batch * batch,uint32_t val)715 patch_rbrc(struct fd_batch *batch, uint32_t val)
716 {
717 unsigned i;
718 for (i = 0; i < fd_patch_num_elements(&batch->rbrc_patches); i++) {
719 struct fd_cs_patch *patch = fd_patch_element(&batch->rbrc_patches, i);
720 *patch->cs = patch->val | val;
721 }
722 util_dynarray_resize(&batch->rbrc_patches, 0);
723 }
724
725 /* for rendering directly to system memory: */
726 static void
fd3_emit_sysmem_prep(struct fd_batch * batch)727 fd3_emit_sysmem_prep(struct fd_batch *batch)
728 {
729 struct pipe_framebuffer_state *pfb = &batch->framebuffer;
730 struct fd_ringbuffer *ring = batch->gmem;
731 uint32_t i, pitch = 0;
732
733 for (i = 0; i < pfb->nr_cbufs; i++) {
734 struct pipe_surface *psurf = pfb->cbufs[i];
735 if (!psurf)
736 continue;
737 pitch = fd_resource(psurf->texture)->slices[psurf->u.tex.level].pitch;
738 }
739
740 fd3_emit_restore(batch, ring);
741
742 OUT_PKT0(ring, REG_A3XX_RB_FRAME_BUFFER_DIMENSION, 1);
743 OUT_RING(ring, A3XX_RB_FRAME_BUFFER_DIMENSION_WIDTH(pfb->width) |
744 A3XX_RB_FRAME_BUFFER_DIMENSION_HEIGHT(pfb->height));
745
746 emit_mrt(ring, pfb->nr_cbufs, pfb->cbufs, NULL, 0, true);
747
748 /* setup scissor/offset for current tile: */
749 OUT_PKT0(ring, REG_A3XX_RB_WINDOW_OFFSET, 1);
750 OUT_RING(ring, A3XX_RB_WINDOW_OFFSET_X(0) |
751 A3XX_RB_WINDOW_OFFSET_Y(0));
752
753 OUT_PKT0(ring, REG_A3XX_GRAS_SC_SCREEN_SCISSOR_TL, 2);
754 OUT_RING(ring, A3XX_GRAS_SC_SCREEN_SCISSOR_TL_X(0) |
755 A3XX_GRAS_SC_SCREEN_SCISSOR_TL_Y(0));
756 OUT_RING(ring, A3XX_GRAS_SC_SCREEN_SCISSOR_BR_X(pfb->width - 1) |
757 A3XX_GRAS_SC_SCREEN_SCISSOR_BR_Y(pfb->height - 1));
758
759 OUT_PKT0(ring, REG_A3XX_RB_MODE_CONTROL, 1);
760 OUT_RING(ring, A3XX_RB_MODE_CONTROL_RENDER_MODE(RB_RENDERING_PASS) |
761 A3XX_RB_MODE_CONTROL_GMEM_BYPASS |
762 A3XX_RB_MODE_CONTROL_MARB_CACHE_SPLIT_MODE |
763 A3XX_RB_MODE_CONTROL_MRT(MAX2(1, pfb->nr_cbufs) - 1));
764
765 patch_draws(batch, IGNORE_VISIBILITY);
766 patch_rbrc(batch, A3XX_RB_RENDER_CONTROL_BIN_WIDTH(pitch));
767 }
768
769 static void
update_vsc_pipe(struct fd_batch * batch)770 update_vsc_pipe(struct fd_batch *batch)
771 {
772 struct fd_context *ctx = batch->ctx;
773 struct fd3_context *fd3_ctx = fd3_context(ctx);
774 struct fd_ringbuffer *ring = batch->gmem;
775 int i;
776
777 OUT_PKT0(ring, REG_A3XX_VSC_SIZE_ADDRESS, 1);
778 OUT_RELOCW(ring, fd3_ctx->vsc_size_mem, 0, 0, 0); /* VSC_SIZE_ADDRESS */
779
780 for (i = 0; i < 8; i++) {
781 struct fd_vsc_pipe *pipe = &ctx->vsc_pipe[i];
782
783 if (!pipe->bo) {
784 pipe->bo = fd_bo_new(ctx->dev, 0x40000,
785 DRM_FREEDRENO_GEM_TYPE_KMEM);
786 }
787
788 OUT_PKT0(ring, REG_A3XX_VSC_PIPE(i), 3);
789 OUT_RING(ring, A3XX_VSC_PIPE_CONFIG_X(pipe->x) |
790 A3XX_VSC_PIPE_CONFIG_Y(pipe->y) |
791 A3XX_VSC_PIPE_CONFIG_W(pipe->w) |
792 A3XX_VSC_PIPE_CONFIG_H(pipe->h));
793 OUT_RELOCW(ring, pipe->bo, 0, 0, 0); /* VSC_PIPE[i].DATA_ADDRESS */
794 OUT_RING(ring, fd_bo_size(pipe->bo) - 32); /* VSC_PIPE[i].DATA_LENGTH */
795 }
796 }
797
798 static void
emit_binning_pass(struct fd_batch * batch)799 emit_binning_pass(struct fd_batch *batch)
800 {
801 struct fd_context *ctx = batch->ctx;
802 struct fd_gmem_stateobj *gmem = &ctx->gmem;
803 struct pipe_framebuffer_state *pfb = &batch->framebuffer;
804 struct fd_ringbuffer *ring = batch->gmem;
805 int i;
806
807 uint32_t x1 = gmem->minx;
808 uint32_t y1 = gmem->miny;
809 uint32_t x2 = gmem->minx + gmem->width - 1;
810 uint32_t y2 = gmem->miny + gmem->height - 1;
811
812 if (ctx->screen->gpu_id == 320) {
813 emit_binning_workaround(batch);
814 fd_wfi(batch, ring);
815 OUT_PKT3(ring, CP_INVALIDATE_STATE, 1);
816 OUT_RING(ring, 0x00007fff);
817 }
818
819 OUT_PKT0(ring, REG_A3XX_VSC_BIN_CONTROL, 1);
820 OUT_RING(ring, A3XX_VSC_BIN_CONTROL_BINNING_ENABLE);
821
822 OUT_PKT0(ring, REG_A3XX_GRAS_SC_CONTROL, 1);
823 OUT_RING(ring, A3XX_GRAS_SC_CONTROL_RENDER_MODE(RB_TILING_PASS) |
824 A3XX_GRAS_SC_CONTROL_MSAA_SAMPLES(MSAA_ONE) |
825 A3XX_GRAS_SC_CONTROL_RASTER_MODE(0));
826
827 OUT_PKT0(ring, REG_A3XX_RB_FRAME_BUFFER_DIMENSION, 1);
828 OUT_RING(ring, A3XX_RB_FRAME_BUFFER_DIMENSION_WIDTH(pfb->width) |
829 A3XX_RB_FRAME_BUFFER_DIMENSION_HEIGHT(pfb->height));
830
831 OUT_PKT0(ring, REG_A3XX_RB_RENDER_CONTROL, 1);
832 OUT_RING(ring, A3XX_RB_RENDER_CONTROL_ALPHA_TEST_FUNC(FUNC_NEVER) |
833 A3XX_RB_RENDER_CONTROL_DISABLE_COLOR_PIPE |
834 A3XX_RB_RENDER_CONTROL_BIN_WIDTH(gmem->bin_w));
835
836 /* setup scissor/offset for whole screen: */
837 OUT_PKT0(ring, REG_A3XX_RB_WINDOW_OFFSET, 1);
838 OUT_RING(ring, A3XX_RB_WINDOW_OFFSET_X(x1) |
839 A3XX_RB_WINDOW_OFFSET_Y(y1));
840
841 OUT_PKT0(ring, REG_A3XX_RB_LRZ_VSC_CONTROL, 1);
842 OUT_RING(ring, A3XX_RB_LRZ_VSC_CONTROL_BINNING_ENABLE);
843
844 OUT_PKT0(ring, REG_A3XX_GRAS_SC_SCREEN_SCISSOR_TL, 2);
845 OUT_RING(ring, A3XX_GRAS_SC_SCREEN_SCISSOR_TL_X(x1) |
846 A3XX_GRAS_SC_SCREEN_SCISSOR_TL_Y(y1));
847 OUT_RING(ring, A3XX_GRAS_SC_SCREEN_SCISSOR_BR_X(x2) |
848 A3XX_GRAS_SC_SCREEN_SCISSOR_BR_Y(y2));
849
850 OUT_PKT0(ring, REG_A3XX_RB_MODE_CONTROL, 1);
851 OUT_RING(ring, A3XX_RB_MODE_CONTROL_RENDER_MODE(RB_TILING_PASS) |
852 A3XX_RB_MODE_CONTROL_MARB_CACHE_SPLIT_MODE |
853 A3XX_RB_MODE_CONTROL_MRT(0));
854
855 for (i = 0; i < 4; i++) {
856 OUT_PKT0(ring, REG_A3XX_RB_MRT_CONTROL(i), 1);
857 OUT_RING(ring, A3XX_RB_MRT_CONTROL_ROP_CODE(ROP_CLEAR) |
858 A3XX_RB_MRT_CONTROL_DITHER_MODE(DITHER_DISABLE) |
859 A3XX_RB_MRT_CONTROL_COMPONENT_ENABLE(0));
860 }
861
862 OUT_PKT0(ring, REG_A3XX_PC_VSTREAM_CONTROL, 1);
863 OUT_RING(ring, A3XX_PC_VSTREAM_CONTROL_SIZE(1) |
864 A3XX_PC_VSTREAM_CONTROL_N(0));
865
866 /* emit IB to binning drawcmds: */
867 ctx->emit_ib(ring, batch->binning);
868 fd_reset_wfi(batch);
869
870 fd_wfi(batch, ring);
871
872 /* and then put stuff back the way it was: */
873
874 OUT_PKT0(ring, REG_A3XX_VSC_BIN_CONTROL, 1);
875 OUT_RING(ring, 0x00000000);
876
877 OUT_PKT0(ring, REG_A3XX_SP_SP_CTRL_REG, 1);
878 OUT_RING(ring, A3XX_SP_SP_CTRL_REG_RESOLVE |
879 A3XX_SP_SP_CTRL_REG_CONSTMODE(1) |
880 A3XX_SP_SP_CTRL_REG_SLEEPMODE(1) |
881 A3XX_SP_SP_CTRL_REG_L0MODE(0));
882
883 OUT_PKT0(ring, REG_A3XX_RB_LRZ_VSC_CONTROL, 1);
884 OUT_RING(ring, 0x00000000);
885
886 OUT_PKT0(ring, REG_A3XX_GRAS_SC_CONTROL, 1);
887 OUT_RING(ring, A3XX_GRAS_SC_CONTROL_RENDER_MODE(RB_RENDERING_PASS) |
888 A3XX_GRAS_SC_CONTROL_MSAA_SAMPLES(MSAA_ONE) |
889 A3XX_GRAS_SC_CONTROL_RASTER_MODE(0));
890
891 OUT_PKT0(ring, REG_A3XX_RB_MODE_CONTROL, 2);
892 OUT_RING(ring, A3XX_RB_MODE_CONTROL_RENDER_MODE(RB_RENDERING_PASS) |
893 A3XX_RB_MODE_CONTROL_MARB_CACHE_SPLIT_MODE |
894 A3XX_RB_MODE_CONTROL_MRT(pfb->nr_cbufs - 1));
895 OUT_RING(ring, A3XX_RB_RENDER_CONTROL_ENABLE_GMEM |
896 A3XX_RB_RENDER_CONTROL_ALPHA_TEST_FUNC(FUNC_NEVER) |
897 A3XX_RB_RENDER_CONTROL_BIN_WIDTH(gmem->bin_w));
898
899 fd_event_write(batch, ring, CACHE_FLUSH);
900 fd_wfi(batch, ring);
901
902 if (ctx->screen->gpu_id == 320) {
903 /* dummy-draw workaround: */
904 OUT_PKT3(ring, CP_DRAW_INDX, 3);
905 OUT_RING(ring, 0x00000000);
906 OUT_RING(ring, DRAW(1, DI_SRC_SEL_AUTO_INDEX,
907 INDEX_SIZE_IGN, IGNORE_VISIBILITY, 0));
908 OUT_RING(ring, 0); /* NumIndices */
909 fd_reset_wfi(batch);
910 }
911
912 OUT_PKT3(ring, CP_NOP, 4);
913 OUT_RING(ring, 0x00000000);
914 OUT_RING(ring, 0x00000000);
915 OUT_RING(ring, 0x00000000);
916 OUT_RING(ring, 0x00000000);
917
918 fd_wfi(batch, ring);
919
920 if (ctx->screen->gpu_id == 320) {
921 emit_binning_workaround(batch);
922 }
923 }
924
925 /* before first tile */
926 static void
fd3_emit_tile_init(struct fd_batch * batch)927 fd3_emit_tile_init(struct fd_batch *batch)
928 {
929 struct fd_ringbuffer *ring = batch->gmem;
930 struct pipe_framebuffer_state *pfb = &batch->framebuffer;
931 struct fd_gmem_stateobj *gmem = &batch->ctx->gmem;
932 uint32_t rb_render_control;
933
934 fd3_emit_restore(batch, ring);
935
936 /* note: use gmem->bin_w/h, the bin_w/h parameters may be truncated
937 * at the right and bottom edge tiles
938 */
939 OUT_PKT0(ring, REG_A3XX_VSC_BIN_SIZE, 1);
940 OUT_RING(ring, A3XX_VSC_BIN_SIZE_WIDTH(gmem->bin_w) |
941 A3XX_VSC_BIN_SIZE_HEIGHT(gmem->bin_h));
942
943 update_vsc_pipe(batch);
944
945 fd_wfi(batch, ring);
946 OUT_PKT0(ring, REG_A3XX_RB_FRAME_BUFFER_DIMENSION, 1);
947 OUT_RING(ring, A3XX_RB_FRAME_BUFFER_DIMENSION_WIDTH(pfb->width) |
948 A3XX_RB_FRAME_BUFFER_DIMENSION_HEIGHT(pfb->height));
949
950 if (use_hw_binning(batch)) {
951 /* emit hw binning pass: */
952 emit_binning_pass(batch);
953
954 patch_draws(batch, USE_VISIBILITY);
955 } else {
956 patch_draws(batch, IGNORE_VISIBILITY);
957 }
958
959 rb_render_control = A3XX_RB_RENDER_CONTROL_ENABLE_GMEM |
960 A3XX_RB_RENDER_CONTROL_BIN_WIDTH(gmem->bin_w);
961
962 patch_rbrc(batch, rb_render_control);
963 }
964
965 /* before mem2gmem */
966 static void
fd3_emit_tile_prep(struct fd_batch * batch,struct fd_tile * tile)967 fd3_emit_tile_prep(struct fd_batch *batch, struct fd_tile *tile)
968 {
969 struct fd_ringbuffer *ring = batch->gmem;
970 struct pipe_framebuffer_state *pfb = &batch->framebuffer;
971
972 OUT_PKT0(ring, REG_A3XX_RB_MODE_CONTROL, 1);
973 OUT_RING(ring, A3XX_RB_MODE_CONTROL_RENDER_MODE(RB_RENDERING_PASS) |
974 A3XX_RB_MODE_CONTROL_MARB_CACHE_SPLIT_MODE |
975 A3XX_RB_MODE_CONTROL_MRT(MAX2(1, pfb->nr_cbufs) - 1));
976 }
977
978 /* before IB to rendering cmds: */
979 static void
fd3_emit_tile_renderprep(struct fd_batch * batch,struct fd_tile * tile)980 fd3_emit_tile_renderprep(struct fd_batch *batch, struct fd_tile *tile)
981 {
982 struct fd_context *ctx = batch->ctx;
983 struct fd3_context *fd3_ctx = fd3_context(ctx);
984 struct fd_ringbuffer *ring = batch->gmem;
985 struct fd_gmem_stateobj *gmem = &ctx->gmem;
986 struct pipe_framebuffer_state *pfb = &batch->framebuffer;
987
988 uint32_t x1 = tile->xoff;
989 uint32_t y1 = tile->yoff;
990 uint32_t x2 = tile->xoff + tile->bin_w - 1;
991 uint32_t y2 = tile->yoff + tile->bin_h - 1;
992
993 uint32_t reg;
994
995 OUT_PKT0(ring, REG_A3XX_RB_DEPTH_INFO, 2);
996 reg = A3XX_RB_DEPTH_INFO_DEPTH_BASE(gmem->zsbuf_base[0]);
997 if (pfb->zsbuf) {
998 reg |= A3XX_RB_DEPTH_INFO_DEPTH_FORMAT(fd_pipe2depth(pfb->zsbuf->format));
999 }
1000 OUT_RING(ring, reg);
1001 if (pfb->zsbuf) {
1002 struct fd_resource *rsc = fd_resource(pfb->zsbuf->texture);
1003 OUT_RING(ring, A3XX_RB_DEPTH_PITCH(rsc->cpp * gmem->bin_w));
1004 if (rsc->stencil) {
1005 OUT_PKT0(ring, REG_A3XX_RB_STENCIL_INFO, 2);
1006 OUT_RING(ring, A3XX_RB_STENCIL_INFO_STENCIL_BASE(gmem->zsbuf_base[1]));
1007 OUT_RING(ring, A3XX_RB_STENCIL_PITCH(rsc->stencil->cpp * gmem->bin_w));
1008 }
1009 } else {
1010 OUT_RING(ring, 0x00000000);
1011 }
1012
1013 if (use_hw_binning(batch)) {
1014 struct fd_vsc_pipe *pipe = &ctx->vsc_pipe[tile->p];
1015
1016 assert(pipe->w * pipe->h);
1017
1018 fd_event_write(batch, ring, HLSQ_FLUSH);
1019 fd_wfi(batch, ring);
1020
1021 OUT_PKT0(ring, REG_A3XX_PC_VSTREAM_CONTROL, 1);
1022 OUT_RING(ring, A3XX_PC_VSTREAM_CONTROL_SIZE(pipe->w * pipe->h) |
1023 A3XX_PC_VSTREAM_CONTROL_N(tile->n));
1024
1025
1026 OUT_PKT3(ring, CP_SET_BIN_DATA, 2);
1027 OUT_RELOCW(ring, pipe->bo, 0, 0, 0); /* BIN_DATA_ADDR <- VSC_PIPE[p].DATA_ADDRESS */
1028 OUT_RELOCW(ring, fd3_ctx->vsc_size_mem, /* BIN_SIZE_ADDR <- VSC_SIZE_ADDRESS + (p * 4) */
1029 (tile->p * 4), 0, 0);
1030 } else {
1031 OUT_PKT0(ring, REG_A3XX_PC_VSTREAM_CONTROL, 1);
1032 OUT_RING(ring, 0x00000000);
1033 }
1034
1035 OUT_PKT3(ring, CP_SET_BIN, 3);
1036 OUT_RING(ring, 0x00000000);
1037 OUT_RING(ring, CP_SET_BIN_1_X1(x1) | CP_SET_BIN_1_Y1(y1));
1038 OUT_RING(ring, CP_SET_BIN_2_X2(x2) | CP_SET_BIN_2_Y2(y2));
1039
1040 emit_mrt(ring, pfb->nr_cbufs, pfb->cbufs, gmem->cbuf_base, gmem->bin_w, true);
1041
1042 /* setup scissor/offset for current tile: */
1043 OUT_PKT0(ring, REG_A3XX_RB_WINDOW_OFFSET, 1);
1044 OUT_RING(ring, A3XX_RB_WINDOW_OFFSET_X(tile->xoff) |
1045 A3XX_RB_WINDOW_OFFSET_Y(tile->yoff));
1046
1047 OUT_PKT0(ring, REG_A3XX_GRAS_SC_SCREEN_SCISSOR_TL, 2);
1048 OUT_RING(ring, A3XX_GRAS_SC_SCREEN_SCISSOR_TL_X(x1) |
1049 A3XX_GRAS_SC_SCREEN_SCISSOR_TL_Y(y1));
1050 OUT_RING(ring, A3XX_GRAS_SC_SCREEN_SCISSOR_BR_X(x2) |
1051 A3XX_GRAS_SC_SCREEN_SCISSOR_BR_Y(y2));
1052 }
1053
1054 void
fd3_gmem_init(struct pipe_context * pctx)1055 fd3_gmem_init(struct pipe_context *pctx)
1056 {
1057 struct fd_context *ctx = fd_context(pctx);
1058
1059 ctx->emit_sysmem_prep = fd3_emit_sysmem_prep;
1060 ctx->emit_tile_init = fd3_emit_tile_init;
1061 ctx->emit_tile_prep = fd3_emit_tile_prep;
1062 ctx->emit_tile_mem2gmem = fd3_emit_tile_mem2gmem;
1063 ctx->emit_tile_renderprep = fd3_emit_tile_renderprep;
1064 ctx->emit_tile_gmem2mem = fd3_emit_tile_gmem2mem;
1065 }
1066