Book a Demo!
CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutPoliciesSign UpSign In
PojavLauncherTeam
GitHub Repository: PojavLauncherTeam/mesa
Path: blob/21.2-virgl/src/gallium/drivers/freedreno/a2xx/fd2_gmem.c
4574 views
1
/*
2
* Copyright (C) 2012 Rob Clark <[email protected]>
3
*
4
* Permission is hereby granted, free of charge, to any person obtaining a
5
* copy of this software and associated documentation files (the "Software"),
6
* to deal in the Software without restriction, including without limitation
7
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
8
* and/or sell copies of the Software, and to permit persons to whom the
9
* Software is furnished to do so, subject to the following conditions:
10
*
11
* The above copyright notice and this permission notice (including the next
12
* paragraph) shall be included in all copies or substantial portions of the
13
* Software.
14
*
15
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21
* SOFTWARE.
22
*
23
* Authors:
24
* Rob Clark <[email protected]>
25
*/
26
27
#include "pipe/p_state.h"
28
#include "util/u_inlines.h"
29
#include "util/u_memory.h"
30
#include "util/u_string.h"
31
32
#include "freedreno_draw.h"
33
#include "freedreno_resource.h"
34
#include "freedreno_state.h"
35
36
#include "ir2/instr-a2xx.h"
37
#include "fd2_context.h"
38
#include "fd2_draw.h"
39
#include "fd2_emit.h"
40
#include "fd2_gmem.h"
41
#include "fd2_program.h"
42
#include "fd2_util.h"
43
#include "fd2_zsa.h"
44
45
static uint32_t
46
fmt2swap(enum pipe_format format)
47
{
48
switch (format) {
49
case PIPE_FORMAT_B8G8R8A8_UNORM:
50
case PIPE_FORMAT_B8G8R8X8_UNORM:
51
case PIPE_FORMAT_B5G6R5_UNORM:
52
case PIPE_FORMAT_B5G5R5A1_UNORM:
53
case PIPE_FORMAT_B5G5R5X1_UNORM:
54
case PIPE_FORMAT_B4G4R4A4_UNORM:
55
case PIPE_FORMAT_B4G4R4X4_UNORM:
56
case PIPE_FORMAT_B2G3R3_UNORM:
57
return 1;
58
default:
59
return 0;
60
}
61
}
62
63
static bool
64
use_hw_binning(struct fd_batch *batch)
65
{
66
const struct fd_gmem_stateobj *gmem = batch->gmem_state;
67
68
/* we hardcoded a limit of 8 "pipes", we can increase this limit
69
* at the cost of a slightly larger command stream
70
* however very few cases will need more than 8
71
* gmem->num_vsc_pipes == 0 means empty batch (TODO: does it still happen?)
72
*/
73
if (gmem->num_vsc_pipes > 8 || !gmem->num_vsc_pipes)
74
return false;
75
76
/* only a20x hw binning is implement
77
* a22x is more like a3xx, but perhaps the a20x works? (TODO)
78
*/
79
if (!is_a20x(batch->ctx->screen))
80
return false;
81
82
return fd_binning_enabled && ((gmem->nbins_x * gmem->nbins_y) > 2);
83
}
84
85
/* transfer from gmem to system memory (ie. normal RAM) */
86
87
static void
88
emit_gmem2mem_surf(struct fd_batch *batch, uint32_t base,
89
struct pipe_surface *psurf)
90
{
91
struct fd_ringbuffer *ring = batch->tile_fini;
92
struct fd_resource *rsc = fd_resource(psurf->texture);
93
uint32_t offset =
94
fd_resource_offset(rsc, psurf->u.tex.level, psurf->u.tex.first_layer);
95
enum pipe_format format = fd_gmem_restore_format(psurf->format);
96
uint32_t pitch = fdl2_pitch_pixels(&rsc->layout, psurf->u.tex.level);
97
98
assert((pitch & 31) == 0);
99
assert((offset & 0xfff) == 0);
100
101
if (!rsc->valid)
102
return;
103
104
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
105
OUT_RING(ring, CP_REG(REG_A2XX_RB_COLOR_INFO));
106
OUT_RING(ring, A2XX_RB_COLOR_INFO_BASE(base) |
107
A2XX_RB_COLOR_INFO_FORMAT(fd2_pipe2color(format)));
108
109
OUT_PKT3(ring, CP_SET_CONSTANT, 5);
110
OUT_RING(ring, CP_REG(REG_A2XX_RB_COPY_CONTROL));
111
OUT_RING(ring, 0x00000000); /* RB_COPY_CONTROL */
112
OUT_RELOC(ring, rsc->bo, offset, 0, 0); /* RB_COPY_DEST_BASE */
113
OUT_RING(ring, pitch >> 5); /* RB_COPY_DEST_PITCH */
114
OUT_RING(ring, /* RB_COPY_DEST_INFO */
115
A2XX_RB_COPY_DEST_INFO_FORMAT(fd2_pipe2color(format)) |
116
COND(!rsc->layout.tile_mode, A2XX_RB_COPY_DEST_INFO_LINEAR) |
117
A2XX_RB_COPY_DEST_INFO_WRITE_RED |
118
A2XX_RB_COPY_DEST_INFO_WRITE_GREEN |
119
A2XX_RB_COPY_DEST_INFO_WRITE_BLUE |
120
A2XX_RB_COPY_DEST_INFO_WRITE_ALPHA);
121
122
if (!is_a20x(batch->ctx->screen)) {
123
OUT_WFI(ring);
124
125
OUT_PKT3(ring, CP_SET_CONSTANT, 3);
126
OUT_RING(ring, CP_REG(REG_A2XX_VGT_MAX_VTX_INDX));
127
OUT_RING(ring, 3); /* VGT_MAX_VTX_INDX */
128
OUT_RING(ring, 0); /* VGT_MIN_VTX_INDX */
129
}
130
131
fd_draw(batch, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
132
DI_SRC_SEL_AUTO_INDEX, 3, 0, INDEX_SIZE_IGN, 0, 0, NULL);
133
}
134
135
static void
136
prepare_tile_fini_ib(struct fd_batch *batch) assert_dt
137
{
138
struct fd_context *ctx = batch->ctx;
139
struct fd2_context *fd2_ctx = fd2_context(ctx);
140
const struct fd_gmem_stateobj *gmem = batch->gmem_state;
141
struct pipe_framebuffer_state *pfb = &batch->framebuffer;
142
struct fd_ringbuffer *ring;
143
144
batch->tile_fini =
145
fd_submit_new_ringbuffer(batch->submit, 0x1000, FD_RINGBUFFER_STREAMING);
146
ring = batch->tile_fini;
147
148
fd2_emit_vertex_bufs(ring, 0x9c,
149
(struct fd2_vertex_buf[]){
150
{.prsc = fd2_ctx->solid_vertexbuf, .size = 36},
151
},
152
1);
153
154
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
155
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_WINDOW_OFFSET));
156
OUT_RING(ring, 0x00000000); /* PA_SC_WINDOW_OFFSET */
157
158
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
159
OUT_RING(ring, CP_REG(REG_A2XX_VGT_INDX_OFFSET));
160
OUT_RING(ring, 0);
161
162
if (!is_a20x(ctx->screen)) {
163
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
164
OUT_RING(ring, CP_REG(REG_A2XX_VGT_VERTEX_REUSE_BLOCK_CNTL));
165
OUT_RING(ring, 0x0000028f);
166
}
167
168
fd2_program_emit(ctx, ring, &ctx->solid_prog);
169
170
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
171
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_AA_MASK));
172
OUT_RING(ring, 0x0000ffff);
173
174
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
175
OUT_RING(ring, CP_REG(REG_A2XX_RB_DEPTHCONTROL));
176
OUT_RING(ring, A2XX_RB_DEPTHCONTROL_EARLY_Z_ENABLE);
177
178
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
179
OUT_RING(ring, CP_REG(REG_A2XX_PA_SU_SC_MODE_CNTL));
180
OUT_RING(
181
ring,
182
A2XX_PA_SU_SC_MODE_CNTL_PROVOKING_VTX_LAST | /* PA_SU_SC_MODE_CNTL */
183
A2XX_PA_SU_SC_MODE_CNTL_FRONT_PTYPE(PC_DRAW_TRIANGLES) |
184
A2XX_PA_SU_SC_MODE_CNTL_BACK_PTYPE(PC_DRAW_TRIANGLES));
185
186
OUT_PKT3(ring, CP_SET_CONSTANT, 3);
187
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_WINDOW_SCISSOR_TL));
188
OUT_RING(ring, xy2d(0, 0)); /* PA_SC_WINDOW_SCISSOR_TL */
189
OUT_RING(ring, xy2d(pfb->width, pfb->height)); /* PA_SC_WINDOW_SCISSOR_BR */
190
191
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
192
OUT_RING(ring, CP_REG(REG_A2XX_PA_CL_CLIP_CNTL));
193
OUT_RING(ring, 0x00000000);
194
195
OUT_PKT3(ring, CP_SET_CONSTANT, 5);
196
OUT_RING(ring, CP_REG(REG_A2XX_PA_CL_VPORT_XSCALE));
197
OUT_RING(ring, fui((float)gmem->bin_w / 2.0)); /* XSCALE */
198
OUT_RING(ring, fui((float)gmem->bin_w / 2.0)); /* XOFFSET */
199
OUT_RING(ring, fui((float)gmem->bin_h / 2.0)); /* YSCALE */
200
OUT_RING(ring, fui((float)gmem->bin_h / 2.0)); /* YOFFSET */
201
202
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
203
OUT_RING(ring, CP_REG(REG_A2XX_RB_MODECONTROL));
204
OUT_RING(ring, A2XX_RB_MODECONTROL_EDRAM_MODE(EDRAM_COPY));
205
206
if (batch->resolve & (FD_BUFFER_DEPTH | FD_BUFFER_STENCIL))
207
emit_gmem2mem_surf(batch, gmem->zsbuf_base[0], pfb->zsbuf);
208
209
if (batch->resolve & FD_BUFFER_COLOR)
210
emit_gmem2mem_surf(batch, gmem->cbuf_base[0], pfb->cbufs[0]);
211
212
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
213
OUT_RING(ring, CP_REG(REG_A2XX_RB_MODECONTROL));
214
OUT_RING(ring, A2XX_RB_MODECONTROL_EDRAM_MODE(COLOR_DEPTH));
215
216
if (!is_a20x(ctx->screen)) {
217
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
218
OUT_RING(ring, CP_REG(REG_A2XX_VGT_VERTEX_REUSE_BLOCK_CNTL));
219
OUT_RING(ring, 0x0000003b);
220
}
221
}
222
223
static void
224
fd2_emit_tile_gmem2mem(struct fd_batch *batch, const struct fd_tile *tile)
225
{
226
fd2_emit_ib(batch->gmem, batch->tile_fini);
227
}
228
229
/* transfer from system memory to gmem */
230
231
static void
232
emit_mem2gmem_surf(struct fd_batch *batch, uint32_t base,
233
struct pipe_surface *psurf)
234
{
235
struct fd_ringbuffer *ring = batch->gmem;
236
struct fd_resource *rsc = fd_resource(psurf->texture);
237
uint32_t offset =
238
fd_resource_offset(rsc, psurf->u.tex.level, psurf->u.tex.first_layer);
239
enum pipe_format format = fd_gmem_restore_format(psurf->format);
240
241
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
242
OUT_RING(ring, CP_REG(REG_A2XX_RB_COLOR_INFO));
243
OUT_RING(ring, A2XX_RB_COLOR_INFO_BASE(base) |
244
A2XX_RB_COLOR_INFO_FORMAT(fd2_pipe2color(format)));
245
246
/* emit fb as a texture: */
247
OUT_PKT3(ring, CP_SET_CONSTANT, 7);
248
OUT_RING(ring, 0x00010000);
249
OUT_RING(ring, A2XX_SQ_TEX_0_CLAMP_X(SQ_TEX_WRAP) |
250
A2XX_SQ_TEX_0_CLAMP_Y(SQ_TEX_WRAP) |
251
A2XX_SQ_TEX_0_CLAMP_Z(SQ_TEX_WRAP) |
252
A2XX_SQ_TEX_0_PITCH(
253
fdl2_pitch_pixels(&rsc->layout, psurf->u.tex.level)));
254
OUT_RELOC(ring, rsc->bo, offset,
255
A2XX_SQ_TEX_1_FORMAT(fd2_pipe2surface(format).format) |
256
A2XX_SQ_TEX_1_CLAMP_POLICY(SQ_TEX_CLAMP_POLICY_OGL),
257
0);
258
OUT_RING(ring, A2XX_SQ_TEX_2_WIDTH(psurf->width - 1) |
259
A2XX_SQ_TEX_2_HEIGHT(psurf->height - 1));
260
OUT_RING(ring, A2XX_SQ_TEX_3_MIP_FILTER(SQ_TEX_FILTER_BASEMAP) |
261
A2XX_SQ_TEX_3_SWIZ_X(0) | A2XX_SQ_TEX_3_SWIZ_Y(1) |
262
A2XX_SQ_TEX_3_SWIZ_Z(2) | A2XX_SQ_TEX_3_SWIZ_W(3) |
263
A2XX_SQ_TEX_3_XY_MAG_FILTER(SQ_TEX_FILTER_POINT) |
264
A2XX_SQ_TEX_3_XY_MIN_FILTER(SQ_TEX_FILTER_POINT));
265
OUT_RING(ring, 0x00000000);
266
OUT_RING(ring, A2XX_SQ_TEX_5_DIMENSION(SQ_TEX_DIMENSION_2D));
267
268
if (!is_a20x(batch->ctx->screen)) {
269
OUT_PKT3(ring, CP_SET_CONSTANT, 3);
270
OUT_RING(ring, CP_REG(REG_A2XX_VGT_MAX_VTX_INDX));
271
OUT_RING(ring, 3); /* VGT_MAX_VTX_INDX */
272
OUT_RING(ring, 0); /* VGT_MIN_VTX_INDX */
273
}
274
275
fd_draw(batch, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
276
DI_SRC_SEL_AUTO_INDEX, 3, 0, INDEX_SIZE_IGN, 0, 0, NULL);
277
}
278
279
static void
280
fd2_emit_tile_mem2gmem(struct fd_batch *batch,
281
const struct fd_tile *tile) assert_dt
282
{
283
struct fd_context *ctx = batch->ctx;
284
struct fd2_context *fd2_ctx = fd2_context(ctx);
285
const struct fd_gmem_stateobj *gmem = batch->gmem_state;
286
struct fd_ringbuffer *ring = batch->gmem;
287
struct pipe_framebuffer_state *pfb = &batch->framebuffer;
288
unsigned bin_w = tile->bin_w;
289
unsigned bin_h = tile->bin_h;
290
float x0, y0, x1, y1;
291
292
fd2_emit_vertex_bufs(
293
ring, 0x9c,
294
(struct fd2_vertex_buf[]){
295
{.prsc = fd2_ctx->solid_vertexbuf, .size = 36},
296
{.prsc = fd2_ctx->solid_vertexbuf, .size = 24, .offset = 36},
297
},
298
2);
299
300
/* write texture coordinates to vertexbuf: */
301
x0 = ((float)tile->xoff) / ((float)pfb->width);
302
x1 = ((float)tile->xoff + bin_w) / ((float)pfb->width);
303
y0 = ((float)tile->yoff) / ((float)pfb->height);
304
y1 = ((float)tile->yoff + bin_h) / ((float)pfb->height);
305
OUT_PKT3(ring, CP_MEM_WRITE, 7);
306
OUT_RELOC(ring, fd_resource(fd2_ctx->solid_vertexbuf)->bo, 36, 0, 0);
307
OUT_RING(ring, fui(x0));
308
OUT_RING(ring, fui(y0));
309
OUT_RING(ring, fui(x1));
310
OUT_RING(ring, fui(y0));
311
OUT_RING(ring, fui(x0));
312
OUT_RING(ring, fui(y1));
313
314
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
315
OUT_RING(ring, CP_REG(REG_A2XX_VGT_INDX_OFFSET));
316
OUT_RING(ring, 0);
317
318
fd2_program_emit(ctx, ring, &ctx->blit_prog[0]);
319
320
OUT_PKT0(ring, REG_A2XX_TC_CNTL_STATUS, 1);
321
OUT_RING(ring, A2XX_TC_CNTL_STATUS_L2_INVALIDATE);
322
323
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
324
OUT_RING(ring, CP_REG(REG_A2XX_RB_DEPTHCONTROL));
325
OUT_RING(ring, A2XX_RB_DEPTHCONTROL_EARLY_Z_ENABLE);
326
327
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
328
OUT_RING(ring, CP_REG(REG_A2XX_PA_SU_SC_MODE_CNTL));
329
OUT_RING(ring, A2XX_PA_SU_SC_MODE_CNTL_PROVOKING_VTX_LAST |
330
A2XX_PA_SU_SC_MODE_CNTL_FRONT_PTYPE(PC_DRAW_TRIANGLES) |
331
A2XX_PA_SU_SC_MODE_CNTL_BACK_PTYPE(PC_DRAW_TRIANGLES));
332
333
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
334
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_AA_MASK));
335
OUT_RING(ring, 0x0000ffff);
336
337
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
338
OUT_RING(ring, CP_REG(REG_A2XX_RB_COLORCONTROL));
339
OUT_RING(ring, A2XX_RB_COLORCONTROL_ALPHA_FUNC(FUNC_ALWAYS) |
340
A2XX_RB_COLORCONTROL_BLEND_DISABLE |
341
A2XX_RB_COLORCONTROL_ROP_CODE(12) |
342
A2XX_RB_COLORCONTROL_DITHER_MODE(DITHER_DISABLE) |
343
A2XX_RB_COLORCONTROL_DITHER_TYPE(DITHER_PIXEL));
344
345
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
346
OUT_RING(ring, CP_REG(REG_A2XX_RB_BLEND_CONTROL));
347
OUT_RING(ring, A2XX_RB_BLEND_CONTROL_COLOR_SRCBLEND(FACTOR_ONE) |
348
A2XX_RB_BLEND_CONTROL_COLOR_COMB_FCN(BLEND2_DST_PLUS_SRC) |
349
A2XX_RB_BLEND_CONTROL_COLOR_DESTBLEND(FACTOR_ZERO) |
350
A2XX_RB_BLEND_CONTROL_ALPHA_SRCBLEND(FACTOR_ONE) |
351
A2XX_RB_BLEND_CONTROL_ALPHA_COMB_FCN(BLEND2_DST_PLUS_SRC) |
352
A2XX_RB_BLEND_CONTROL_ALPHA_DESTBLEND(FACTOR_ZERO));
353
354
OUT_PKT3(ring, CP_SET_CONSTANT, 3);
355
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_WINDOW_SCISSOR_TL));
356
OUT_RING(ring, A2XX_PA_SC_WINDOW_OFFSET_DISABLE |
357
xy2d(0, 0)); /* PA_SC_WINDOW_SCISSOR_TL */
358
OUT_RING(ring, xy2d(bin_w, bin_h)); /* PA_SC_WINDOW_SCISSOR_BR */
359
360
OUT_PKT3(ring, CP_SET_CONSTANT, 5);
361
OUT_RING(ring, CP_REG(REG_A2XX_PA_CL_VPORT_XSCALE));
362
OUT_RING(ring, fui((float)bin_w / 2.0)); /* PA_CL_VPORT_XSCALE */
363
OUT_RING(ring, fui((float)bin_w / 2.0)); /* PA_CL_VPORT_XOFFSET */
364
OUT_RING(ring, fui(-(float)bin_h / 2.0)); /* PA_CL_VPORT_YSCALE */
365
OUT_RING(ring, fui((float)bin_h / 2.0)); /* PA_CL_VPORT_YOFFSET */
366
367
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
368
OUT_RING(ring, CP_REG(REG_A2XX_PA_CL_VTE_CNTL));
369
OUT_RING(ring, A2XX_PA_CL_VTE_CNTL_VTX_XY_FMT |
370
A2XX_PA_CL_VTE_CNTL_VTX_Z_FMT | // XXX check this???
371
A2XX_PA_CL_VTE_CNTL_VPORT_X_SCALE_ENA |
372
A2XX_PA_CL_VTE_CNTL_VPORT_X_OFFSET_ENA |
373
A2XX_PA_CL_VTE_CNTL_VPORT_Y_SCALE_ENA |
374
A2XX_PA_CL_VTE_CNTL_VPORT_Y_OFFSET_ENA);
375
376
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
377
OUT_RING(ring, CP_REG(REG_A2XX_PA_CL_CLIP_CNTL));
378
OUT_RING(ring, 0x00000000);
379
380
if (fd_gmem_needs_restore(batch, tile, FD_BUFFER_DEPTH | FD_BUFFER_STENCIL))
381
emit_mem2gmem_surf(batch, gmem->zsbuf_base[0], pfb->zsbuf);
382
383
if (fd_gmem_needs_restore(batch, tile, FD_BUFFER_COLOR))
384
emit_mem2gmem_surf(batch, gmem->cbuf_base[0], pfb->cbufs[0]);
385
386
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
387
OUT_RING(ring, CP_REG(REG_A2XX_PA_CL_VTE_CNTL));
388
OUT_RING(ring, A2XX_PA_CL_VTE_CNTL_VTX_W0_FMT |
389
A2XX_PA_CL_VTE_CNTL_VPORT_X_SCALE_ENA |
390
A2XX_PA_CL_VTE_CNTL_VPORT_X_OFFSET_ENA |
391
A2XX_PA_CL_VTE_CNTL_VPORT_Y_SCALE_ENA |
392
A2XX_PA_CL_VTE_CNTL_VPORT_Y_OFFSET_ENA |
393
A2XX_PA_CL_VTE_CNTL_VPORT_Z_SCALE_ENA |
394
A2XX_PA_CL_VTE_CNTL_VPORT_Z_OFFSET_ENA);
395
396
/* TODO blob driver seems to toss in a CACHE_FLUSH after each DRAW_INDX.. */
397
}
398
399
static void
400
patch_draws(struct fd_batch *batch, enum pc_di_vis_cull_mode vismode)
401
{
402
unsigned i;
403
404
if (!is_a20x(batch->ctx->screen)) {
405
/* identical to a3xx */
406
for (i = 0; i < fd_patch_num_elements(&batch->draw_patches); i++) {
407
struct fd_cs_patch *patch = fd_patch_element(&batch->draw_patches, i);
408
*patch->cs = patch->val | DRAW(0, 0, 0, vismode, 0);
409
}
410
util_dynarray_clear(&batch->draw_patches);
411
return;
412
}
413
414
if (vismode == USE_VISIBILITY)
415
return;
416
417
for (i = 0; i < batch->draw_patches.size / sizeof(uint32_t *); i++) {
418
uint32_t *ptr =
419
*util_dynarray_element(&batch->draw_patches, uint32_t *, i);
420
unsigned cnt = ptr[0] >> 16 & 0xfff; /* 5 with idx buffer, 3 without */
421
422
/* convert CP_DRAW_INDX_BIN to a CP_DRAW_INDX
423
* replace first two DWORDS with NOP and move the rest down
424
* (we don't want to have to move the idx buffer reloc)
425
*/
426
ptr[0] = CP_TYPE3_PKT | (CP_NOP << 8);
427
ptr[1] = 0x00000000;
428
429
ptr[4] = ptr[2] & ~(1 << 14 | 1 << 15); /* remove cull_enable bits */
430
ptr[2] = CP_TYPE3_PKT | ((cnt - 2) << 16) | (CP_DRAW_INDX << 8);
431
ptr[3] = 0x00000000;
432
}
433
}
434
435
static void
436
fd2_emit_sysmem_prep(struct fd_batch *batch)
437
{
438
struct fd_context *ctx = batch->ctx;
439
struct fd_ringbuffer *ring = batch->gmem;
440
struct pipe_framebuffer_state *pfb = &batch->framebuffer;
441
struct pipe_surface *psurf = pfb->cbufs[0];
442
443
if (!psurf)
444
return;
445
446
struct fd_resource *rsc = fd_resource(psurf->texture);
447
uint32_t offset =
448
fd_resource_offset(rsc, psurf->u.tex.level, psurf->u.tex.first_layer);
449
uint32_t pitch = fdl2_pitch_pixels(&rsc->layout, psurf->u.tex.level);
450
451
assert((pitch & 31) == 0);
452
assert((offset & 0xfff) == 0);
453
454
fd2_emit_restore(ctx, ring);
455
456
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
457
OUT_RING(ring, CP_REG(REG_A2XX_RB_SURFACE_INFO));
458
OUT_RING(ring, A2XX_RB_SURFACE_INFO_SURFACE_PITCH(pitch));
459
460
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
461
OUT_RING(ring, CP_REG(REG_A2XX_RB_COLOR_INFO));
462
OUT_RELOC(ring, rsc->bo, offset,
463
COND(!rsc->layout.tile_mode, A2XX_RB_COLOR_INFO_LINEAR) |
464
A2XX_RB_COLOR_INFO_SWAP(fmt2swap(psurf->format)) |
465
A2XX_RB_COLOR_INFO_FORMAT(fd2_pipe2color(psurf->format)),
466
0);
467
468
OUT_PKT3(ring, CP_SET_CONSTANT, 3);
469
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_SCREEN_SCISSOR_TL));
470
OUT_RING(ring, A2XX_PA_SC_SCREEN_SCISSOR_TL_WINDOW_OFFSET_DISABLE);
471
OUT_RING(ring, A2XX_PA_SC_SCREEN_SCISSOR_BR_X(pfb->width) |
472
A2XX_PA_SC_SCREEN_SCISSOR_BR_Y(pfb->height));
473
474
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
475
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_WINDOW_OFFSET));
476
OUT_RING(ring,
477
A2XX_PA_SC_WINDOW_OFFSET_X(0) | A2XX_PA_SC_WINDOW_OFFSET_Y(0));
478
479
patch_draws(batch, IGNORE_VISIBILITY);
480
util_dynarray_clear(&batch->draw_patches);
481
util_dynarray_clear(&batch->shader_patches);
482
}
483
484
/* before first tile */
485
static void
486
fd2_emit_tile_init(struct fd_batch *batch) assert_dt
487
{
488
struct fd_context *ctx = batch->ctx;
489
struct fd_ringbuffer *ring = batch->gmem;
490
struct pipe_framebuffer_state *pfb = &batch->framebuffer;
491
const struct fd_gmem_stateobj *gmem = batch->gmem_state;
492
enum pipe_format format = pipe_surface_format(pfb->cbufs[0]);
493
uint32_t reg;
494
495
fd2_emit_restore(ctx, ring);
496
497
prepare_tile_fini_ib(batch);
498
499
OUT_PKT3(ring, CP_SET_CONSTANT, 4);
500
OUT_RING(ring, CP_REG(REG_A2XX_RB_SURFACE_INFO));
501
OUT_RING(ring, gmem->bin_w); /* RB_SURFACE_INFO */
502
OUT_RING(ring, A2XX_RB_COLOR_INFO_SWAP(fmt2swap(format)) |
503
A2XX_RB_COLOR_INFO_FORMAT(fd2_pipe2color(format)));
504
reg = A2XX_RB_DEPTH_INFO_DEPTH_BASE(gmem->zsbuf_base[0]);
505
if (pfb->zsbuf)
506
reg |= A2XX_RB_DEPTH_INFO_DEPTH_FORMAT(fd_pipe2depth(pfb->zsbuf->format));
507
OUT_RING(ring, reg); /* RB_DEPTH_INFO */
508
509
/* fast clear patches */
510
int depth_size = -1;
511
int color_size = -1;
512
513
if (pfb->cbufs[0])
514
color_size = util_format_get_blocksizebits(format) == 32 ? 4 : 2;
515
516
if (pfb->zsbuf)
517
depth_size = fd_pipe2depth(pfb->zsbuf->format) == 1 ? 4 : 2;
518
519
for (int i = 0; i < fd_patch_num_elements(&batch->gmem_patches); i++) {
520
struct fd_cs_patch *patch = fd_patch_element(&batch->gmem_patches, i);
521
uint32_t color_base = 0, depth_base = gmem->zsbuf_base[0];
522
uint32_t size, lines;
523
524
/* note: 1 "line" is 512 bytes in both color/depth areas (1K total) */
525
switch (patch->val) {
526
case GMEM_PATCH_FASTCLEAR_COLOR:
527
size = align(gmem->bin_w * gmem->bin_h * color_size, 0x8000);
528
lines = size / 1024;
529
depth_base = size / 2;
530
break;
531
case GMEM_PATCH_FASTCLEAR_DEPTH:
532
size = align(gmem->bin_w * gmem->bin_h * depth_size, 0x8000);
533
lines = size / 1024;
534
color_base = depth_base;
535
depth_base = depth_base + size / 2;
536
break;
537
case GMEM_PATCH_FASTCLEAR_COLOR_DEPTH:
538
lines =
539
align(gmem->bin_w * gmem->bin_h * color_size * 2, 0x8000) / 1024;
540
break;
541
case GMEM_PATCH_RESTORE_INFO:
542
patch->cs[0] = gmem->bin_w;
543
patch->cs[1] = A2XX_RB_COLOR_INFO_SWAP(fmt2swap(format)) |
544
A2XX_RB_COLOR_INFO_FORMAT(fd2_pipe2color(format));
545
patch->cs[2] = A2XX_RB_DEPTH_INFO_DEPTH_BASE(gmem->zsbuf_base[0]);
546
if (pfb->zsbuf)
547
patch->cs[2] |= A2XX_RB_DEPTH_INFO_DEPTH_FORMAT(
548
fd_pipe2depth(pfb->zsbuf->format));
549
continue;
550
default:
551
continue;
552
}
553
554
patch->cs[0] = A2XX_PA_SC_SCREEN_SCISSOR_BR_X(32) |
555
A2XX_PA_SC_SCREEN_SCISSOR_BR_Y(lines);
556
patch->cs[4] = A2XX_RB_COLOR_INFO_BASE(color_base) |
557
A2XX_RB_COLOR_INFO_FORMAT(COLORX_8_8_8_8);
558
patch->cs[5] = A2XX_RB_DEPTH_INFO_DEPTH_BASE(depth_base) |
559
A2XX_RB_DEPTH_INFO_DEPTH_FORMAT(1);
560
}
561
util_dynarray_clear(&batch->gmem_patches);
562
563
/* set to zero, for some reason hardware doesn't like certain values */
564
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
565
OUT_RING(ring, CP_REG(REG_A2XX_VGT_CURRENT_BIN_ID_MIN));
566
OUT_RING(ring, 0);
567
568
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
569
OUT_RING(ring, CP_REG(REG_A2XX_VGT_CURRENT_BIN_ID_MAX));
570
OUT_RING(ring, 0);
571
572
if (use_hw_binning(batch)) {
573
/* patch out unneeded memory exports by changing EXEC CF to EXEC_END
574
*
575
* in the shader compiler, we guarantee that the shader ends with
576
* a specific pattern of ALLOC/EXEC CF pairs for the hw binning exports
577
*
578
* the since patches point only to dwords and CFs are 1.5 dwords
579
* the patch is aligned and might point to a ALLOC CF
580
*/
581
for (int i = 0; i < batch->shader_patches.size / sizeof(void *); i++) {
582
instr_cf_t *cf =
583
*util_dynarray_element(&batch->shader_patches, instr_cf_t *, i);
584
if (cf->opc == ALLOC)
585
cf++;
586
assert(cf->opc == EXEC);
587
assert(cf[ctx->screen->info->num_vsc_pipes * 2 - 2].opc == EXEC_END);
588
cf[2 * (gmem->num_vsc_pipes - 1)].opc = EXEC_END;
589
}
590
591
patch_draws(batch, USE_VISIBILITY);
592
593
/* initialize shader constants for the binning memexport */
594
OUT_PKT3(ring, CP_SET_CONSTANT, 1 + gmem->num_vsc_pipes * 4);
595
OUT_RING(ring, 0x0000000C);
596
597
for (int i = 0; i < gmem->num_vsc_pipes; i++) {
598
/* allocate in 64k increments to avoid reallocs */
599
uint32_t bo_size = align(batch->num_vertices, 0x10000);
600
if (!ctx->vsc_pipe_bo[i] ||
601
fd_bo_size(ctx->vsc_pipe_bo[i]) < bo_size) {
602
if (ctx->vsc_pipe_bo[i])
603
fd_bo_del(ctx->vsc_pipe_bo[i]);
604
ctx->vsc_pipe_bo[i] =
605
fd_bo_new(ctx->dev, bo_size, 0, "vsc_pipe[%u]", i);
606
assert(ctx->vsc_pipe_bo[i]);
607
}
608
609
/* memory export address (export32):
610
* .x: (base_address >> 2) | 0x40000000 (?)
611
* .y: index (float) - set by shader
612
* .z: 0x4B00D000 (?)
613
* .w: 0x4B000000 (?) | max_index (?)
614
*/
615
OUT_RELOC(ring, ctx->vsc_pipe_bo[i], 0, 0x40000000, -2);
616
OUT_RING(ring, 0x00000000);
617
OUT_RING(ring, 0x4B00D000);
618
OUT_RING(ring, 0x4B000000 | bo_size);
619
}
620
621
OUT_PKT3(ring, CP_SET_CONSTANT, 1 + gmem->num_vsc_pipes * 8);
622
OUT_RING(ring, 0x0000018C);
623
624
for (int i = 0; i < gmem->num_vsc_pipes; i++) {
625
const struct fd_vsc_pipe *pipe = &gmem->vsc_pipe[i];
626
float off_x, off_y, mul_x, mul_y;
627
628
/* const to tranform from [-1,1] to bin coordinates for this pipe
629
* for x/y, [0,256/2040] = 0, [256/2040,512/2040] = 1, etc
630
* 8 possible values on x/y axis,
631
* to clip at binning stage: only use center 6x6
632
* TODO: set the z parameters too so that hw binning
633
* can clip primitives in Z too
634
*/
635
636
mul_x = 1.0f / (float)(gmem->bin_w * 8);
637
mul_y = 1.0f / (float)(gmem->bin_h * 8);
638
off_x = -pipe->x * (1.0 / 8.0f) + 0.125f - mul_x * gmem->minx;
639
off_y = -pipe->y * (1.0 / 8.0f) + 0.125f - mul_y * gmem->miny;
640
641
OUT_RING(ring, fui(off_x * (256.0f / 255.0f)));
642
OUT_RING(ring, fui(off_y * (256.0f / 255.0f)));
643
OUT_RING(ring, 0x3f000000);
644
OUT_RING(ring, fui(0.0f));
645
646
OUT_RING(ring, fui(mul_x * (256.0f / 255.0f)));
647
OUT_RING(ring, fui(mul_y * (256.0f / 255.0f)));
648
OUT_RING(ring, fui(0.0f));
649
OUT_RING(ring, fui(0.0f));
650
}
651
652
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
653
OUT_RING(ring, CP_REG(REG_A2XX_VGT_VERTEX_REUSE_BLOCK_CNTL));
654
OUT_RING(ring, 0);
655
656
fd2_emit_ib(ring, batch->binning);
657
658
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
659
OUT_RING(ring, CP_REG(REG_A2XX_VGT_VERTEX_REUSE_BLOCK_CNTL));
660
OUT_RING(ring, 0x00000002);
661
} else {
662
patch_draws(batch, IGNORE_VISIBILITY);
663
}
664
665
util_dynarray_clear(&batch->draw_patches);
666
util_dynarray_clear(&batch->shader_patches);
667
}
668
669
/* before mem2gmem */
670
static void
671
fd2_emit_tile_prep(struct fd_batch *batch, const struct fd_tile *tile)
672
{
673
struct fd_ringbuffer *ring = batch->gmem;
674
struct pipe_framebuffer_state *pfb = &batch->framebuffer;
675
enum pipe_format format = pipe_surface_format(pfb->cbufs[0]);
676
677
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
678
OUT_RING(ring, CP_REG(REG_A2XX_RB_COLOR_INFO));
679
OUT_RING(ring, A2XX_RB_COLOR_INFO_SWAP(1) | /* RB_COLOR_INFO */
680
A2XX_RB_COLOR_INFO_FORMAT(fd2_pipe2color(format)));
681
682
/* setup screen scissor for current tile (same for mem2gmem): */
683
OUT_PKT3(ring, CP_SET_CONSTANT, 3);
684
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_SCREEN_SCISSOR_TL));
685
OUT_RING(ring, A2XX_PA_SC_SCREEN_SCISSOR_TL_X(0) |
686
A2XX_PA_SC_SCREEN_SCISSOR_TL_Y(0));
687
OUT_RING(ring, A2XX_PA_SC_SCREEN_SCISSOR_BR_X(tile->bin_w) |
688
A2XX_PA_SC_SCREEN_SCISSOR_BR_Y(tile->bin_h));
689
}
690
691
/* before IB to rendering cmds: */
692
static void
693
fd2_emit_tile_renderprep(struct fd_batch *batch,
694
const struct fd_tile *tile) assert_dt
695
{
696
struct fd_context *ctx = batch->ctx;
697
struct fd2_context *fd2_ctx = fd2_context(ctx);
698
struct fd_ringbuffer *ring = batch->gmem;
699
struct pipe_framebuffer_state *pfb = &batch->framebuffer;
700
enum pipe_format format = pipe_surface_format(pfb->cbufs[0]);
701
702
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
703
OUT_RING(ring, CP_REG(REG_A2XX_RB_COLOR_INFO));
704
OUT_RING(ring, A2XX_RB_COLOR_INFO_SWAP(fmt2swap(format)) |
705
A2XX_RB_COLOR_INFO_FORMAT(fd2_pipe2color(format)));
706
707
/* setup window scissor and offset for current tile (different
708
* from mem2gmem):
709
*/
710
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
711
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_WINDOW_OFFSET));
712
OUT_RING(ring, A2XX_PA_SC_WINDOW_OFFSET_X(-tile->xoff) |
713
A2XX_PA_SC_WINDOW_OFFSET_Y(-tile->yoff));
714
715
/* write SCISSOR_BR to memory so fast clear path can restore from it */
716
OUT_PKT3(ring, CP_MEM_WRITE, 2);
717
OUT_RELOC(ring, fd_resource(fd2_ctx->solid_vertexbuf)->bo, 60, 0, 0);
718
OUT_RING(ring, A2XX_PA_SC_SCREEN_SCISSOR_BR_X(tile->bin_w) |
719
A2XX_PA_SC_SCREEN_SCISSOR_BR_Y(tile->bin_h));
720
721
/* set the copy offset for gmem2mem */
722
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
723
OUT_RING(ring, CP_REG(REG_A2XX_RB_COPY_DEST_OFFSET));
724
OUT_RING(ring, A2XX_RB_COPY_DEST_OFFSET_X(tile->xoff) |
725
A2XX_RB_COPY_DEST_OFFSET_Y(tile->yoff));
726
727
/* tile offset for gl_FragCoord on a20x (C64 in fragment shader) */
728
if (is_a20x(ctx->screen)) {
729
OUT_PKT3(ring, CP_SET_CONSTANT, 5);
730
OUT_RING(ring, 0x00000580);
731
OUT_RING(ring, fui(tile->xoff));
732
OUT_RING(ring, fui(tile->yoff));
733
OUT_RING(ring, fui(0.0f));
734
OUT_RING(ring, fui(0.0f));
735
}
736
737
if (use_hw_binning(batch)) {
738
struct fd_bo *pipe_bo = ctx->vsc_pipe_bo[tile->p];
739
740
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
741
OUT_RING(ring, CP_REG(REG_A2XX_VGT_CURRENT_BIN_ID_MIN));
742
OUT_RING(ring, tile->n);
743
744
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
745
OUT_RING(ring, CP_REG(REG_A2XX_VGT_CURRENT_BIN_ID_MAX));
746
OUT_RING(ring, tile->n);
747
748
/* TODO only emit this when tile->p changes */
749
OUT_PKT3(ring, CP_SET_DRAW_INIT_FLAGS, 1);
750
OUT_RELOC(ring, pipe_bo, 0, 0, 0);
751
}
752
}
753
754
void
755
fd2_gmem_init(struct pipe_context *pctx) disable_thread_safety_analysis
756
{
757
struct fd_context *ctx = fd_context(pctx);
758
759
ctx->emit_sysmem_prep = fd2_emit_sysmem_prep;
760
ctx->emit_tile_init = fd2_emit_tile_init;
761
ctx->emit_tile_prep = fd2_emit_tile_prep;
762
ctx->emit_tile_mem2gmem = fd2_emit_tile_mem2gmem;
763
ctx->emit_tile_renderprep = fd2_emit_tile_renderprep;
764
ctx->emit_tile_gmem2mem = fd2_emit_tile_gmem2mem;
765
}
766
767