| 1 |
/* |
| 2 |
* Copyright © 2012-2013 Rob Clark <[email protected]> |
| 3 |
* SPDX-License-Identifier: MIT |
| 4 |
* |
| 5 |
* Authors: |
| 6 |
* Rob Clark <[email protected]> |
| 7 |
*/ |
| 8 |
|
| 9 |
#include "pipe/p_state.h" |
| 10 |
#include "util/u_memory.h" |
| 11 |
#include "util/u_prim.h" |
| 12 |
#include "util/u_string.h" |
| 13 |
|
| 14 |
#include "freedreno_resource.h" |
| 15 |
#include "freedreno_state.h" |
| 16 |
|
| 17 |
#include "fd2_context.h" |
| 18 |
#include "fd2_draw.h" |
| 19 |
#include "fd2_emit.h" |
| 20 |
#include "fd2_program.h" |
| 21 |
#include "fd2_util.h" |
| 22 |
#include "fd2_zsa.h" |
| 23 |
|
| 24 |
static inline uint32_t |
| 25 |
pack_rgba(enum pipe_format format, const float *rgba) |
| 26 |
{ |
| 27 |
union util_color uc; |
| 28 |
util_pack_color(rgba, format, &uc); |
| 29 |
return uc.ui[0]; |
| 30 |
} |
| 31 |
|
| 32 |
static void |
| 33 |
emit_cacheflush(struct fd_ringbuffer *ring) |
| 34 |
{ |
| 35 |
unsigned i; |
| 36 |
|
| 37 |
for (i = 0; i < 12; i++) { |
| 38 |
OUT_PKT3(ring, CP_EVENT_WRITE, 1); |
| 39 |
OUT_RING(ring, CACHE_FLUSH); |
| 40 |
} |
| 41 |
} |
| 42 |
|
| 43 |
static void |
| 44 |
emit_vertexbufs(struct fd_context *ctx) assert_dt |
| 45 |
{ |
| 46 |
struct fd_vertex_stateobj *vtx = ctx->vtx.vtx; |
| 47 |
struct fd_vertexbuf_stateobj *vertexbuf = &ctx->vtx.vertexbuf; |
| 48 |
struct fd2_vertex_buf bufs[PIPE_MAX_ATTRIBS]; |
| 49 |
unsigned i; |
| 50 |
|
| 51 |
if (!vtx->num_elements) |
| 52 |
return; |
| 53 |
|
| 54 |
for (i = 0; i < vtx->num_elements; i++) { |
| 55 |
struct pipe_vertex_element *elem = &vtx->pipe[i]; |
| 56 |
struct pipe_vertex_buffer *vb = &vertexbuf->vb[elem->vertex_buffer_index]; |
| 57 |
bufs[i].offset = vb->buffer_offset; |
| 58 |
bufs[i].size = fd_bo_size(fd_resource(vb->buffer.resource)->bo); |
| 59 |
bufs[i].prsc = vb->buffer.resource; |
| 60 |
} |
| 61 |
|
| 62 |
// NOTE I believe the 0x78 (or 0x9c in solid_vp) relates to the |
| 63 |
// CONST(20,0) (or CONST(26,0) in soliv_vp) |
| 64 |
|
| 65 |
fd2_emit_vertex_bufs(ctx->batch->draw, 0x78, bufs, vtx->num_elements); |
| 66 |
fd2_emit_vertex_bufs(ctx->batch->binning, 0x78, bufs, vtx->num_elements); |
| 67 |
} |
| 68 |
|
| 69 |
static void |
| 70 |
draw_impl(struct fd_context *ctx, const struct pipe_draw_info *info, |
| 71 |
const struct pipe_draw_start_count_bias *draw, struct fd_ringbuffer *ring, |
| 72 |
unsigned index_offset, bool binning) assert_dt |
| 73 |
{ |
| 74 |
OUT_PKT3(ring, CP_SET_CONSTANT, 2); |
| 75 |
OUT_RING(ring, CP_REG(REG_A2XX_VGT_INDX_OFFSET)); |
| 76 |
OUT_RING(ring, info->index_size ? 0 : draw->start); |
| 77 |
|
| 78 |
OUT_PKT0(ring, REG_A2XX_TC_CNTL_STATUS, 1); |
| 79 |
OUT_RING(ring, A2XX_TC_CNTL_STATUS_L2_INVALIDATE); |
| 80 |
|
| 81 |
if (is_a20x(ctx->screen)) { |
| 82 |
/* wait for DMA to finish and |
| 83 |
* dummy draw one triangle with indexes 0,0,0. |
| 84 |
* with PRE_FETCH_CULL_ENABLE | GRP_CULL_ENABLE. |
| 85 |
* |
| 86 |
* this workaround is for a HW bug related to DMA alignment: |
| 87 |
* it is necessary for indexed draws and possibly also |
| 88 |
* draws that read binning data |
| 89 |
*/ |
| 90 |
OUT_PKT3(ring, CP_WAIT_REG_EQ, 4); |
| 91 |
OUT_RING(ring, 0x000005d0); /* RBBM_STATUS */ |
| 92 |
OUT_RING(ring, 0x00000000); |
| 93 |
OUT_RING(ring, 0x00001000); /* bit: 12: VGT_BUSY_NO_DMA */ |
| 94 |
OUT_RING(ring, 0x00000001); |
| 95 |
|
| 96 |
OUT_PKT3(ring, CP_DRAW_INDX_BIN, 6); |
| 97 |
OUT_RING(ring, 0x00000000); |
| 98 |
OUT_RING(ring, 0x0003c004); |
| 99 |
OUT_RING(ring, 0x00000000); |
| 100 |
OUT_RING(ring, 0x00000003); |
| 101 |
OUT_RELOC(ring, fd_resource(fd2_context(ctx)->solid_vertexbuf)->bo, 64, 0, |
| 102 |
0); |
| 103 |
OUT_RING(ring, 0x00000006); |
| 104 |
} else { |
| 105 |
OUT_WFI(ring); |
| 106 |
|
| 107 |
OUT_PKT3(ring, CP_SET_CONSTANT, 3); |
| 108 |
OUT_RING(ring, CP_REG(REG_A2XX_VGT_MAX_VTX_INDX)); |
| 109 |
OUT_RING(ring, info->index_bounds_valid ? info->max_index |
| 110 |
: ~0); /* VGT_MAX_VTX_INDX */ |
| 111 |
OUT_RING(ring, info->index_bounds_valid ? info->min_index |
| 112 |
: 0); /* VGT_MIN_VTX_INDX */ |
| 113 |
} |
| 114 |
|
| 115 |
/* binning shader will take offset from C64 */ |
| 116 |
if (binning && is_a20x(ctx->screen)) { |
| 117 |
OUT_PKT3(ring, CP_SET_CONSTANT, 5); |
| 118 |
OUT_RING(ring, 0x00000180); |
| 119 |
OUT_RING(ring, fui(ctx->batch->num_vertices)); |
| 120 |
OUT_RING(ring, fui(0.0f)); |
| 121 |
OUT_RING(ring, fui(0.0f)); |
| 122 |
OUT_RING(ring, fui(0.0f)); |
| 123 |
} |
| 124 |
|
| 125 |
enum pc_di_vis_cull_mode vismode = USE_VISIBILITY; |
| 126 |
if (binning || info->mode == MESA_PRIM_POINTS) |
| 127 |
vismode = IGNORE_VISIBILITY; |
| 128 |
|
| 129 |
fd_draw_emit(ctx->batch, ring, ctx->screen->primtypes[info->mode], |
| 130 |
vismode, info, draw, index_offset); |
| 131 |
|
| 132 |
if (is_a20x(ctx->screen)) { |
| 133 |
/* not sure why this is required, but it fixes some hangs */ |
| 134 |
OUT_WFI(ring); |
| 135 |
} else { |
| 136 |
OUT_PKT3(ring, CP_SET_CONSTANT, 2); |
| 137 |
OUT_RING(ring, CP_REG(REG_A2XX_UNKNOWN_2010)); |
| 138 |
OUT_RING(ring, 0x00000000); |
| 139 |
} |
| 140 |
|
| 141 |
emit_cacheflush(ring); |
| 142 |
} |
| 143 |
|
| 144 |
static bool |
| 145 |
fd2_draw_vbo(struct fd_context *ctx, const struct pipe_draw_info *pinfo, |
| 146 |
unsigned drawid_offset, |
| 147 |
const struct pipe_draw_indirect_info *indirect, |
| 148 |
const struct pipe_draw_start_count_bias *pdraw, |
| 149 |
unsigned index_offset) assert_dt |
| 150 |
{ |
| 151 |
if (!ctx->prog.fs || !ctx->prog.vs) |
| 152 |
return false; |
| 153 |
|
| 154 |
if (pinfo->mode != MESA_PRIM_COUNT && !indirect && !pinfo->primitive_restart && |
| 155 |
!u_trim_pipe_prim(pinfo->mode, (unsigned *)&pdraw->count)) |
| 156 |
return false; |
| 157 |
|
| 158 |
if (ctx->dirty & FD_DIRTY_VTXBUF) |
| 159 |
emit_vertexbufs(ctx); |
| 160 |
|
| 161 |
fd_blend_tracking(ctx); |
| 162 |
|
| 163 |
if (fd_binning_enabled) |
| 164 |
fd2_emit_state_binning(ctx, ctx->dirty); |
| 165 |
|
| 166 |
fd2_emit_state(ctx, ctx->dirty); |
| 167 |
|
| 168 |
/* a2xx can draw only 65535 vertices at once |
| 169 |
* on a22x the field in the draw command is 32bits but seems limited too |
| 170 |
* using a limit of 32k because it fixes an unexplained hang |
| 171 |
* 32766 works for all primitives (multiple of 2 and 3) |
| 172 |
*/ |
| 173 |
if (pdraw->count > 32766) { |
| 174 |
/* clang-format off */ |
| 175 |
static const uint16_t step_tbl[MESA_PRIM_COUNT] = { |
| 176 |
[0 ... MESA_PRIM_COUNT - 1] = 32766, |
| 177 |
[MESA_PRIM_LINE_STRIP] = 32765, |
| 178 |
[MESA_PRIM_TRIANGLE_STRIP] = 32764, |
| 179 |
|
| 180 |
/* needs more work */ |
| 181 |
[MESA_PRIM_TRIANGLE_FAN] = 0, |
| 182 |
[MESA_PRIM_LINE_LOOP] = 0, |
| 183 |
}; |
| 184 |
/* clang-format on */ |
| 185 |
|
| 186 |
struct pipe_draw_start_count_bias draw = *pdraw; |
| 187 |
unsigned count = draw.count; |
| 188 |
unsigned step = step_tbl[pinfo->mode]; |
| 189 |
unsigned num_vertices = ctx->batch->num_vertices; |
| 190 |
|
| 191 |
if (!step) |
| 192 |
return false; |
| 193 |
|
| 194 |
for (; count + step > 32766; count -= step) { |
| 195 |
draw.count = MIN2(count, 32766); |
| 196 |
draw_impl(ctx, pinfo, &draw, ctx->batch->draw, index_offset, false); |
| 197 |
draw_impl(ctx, pinfo, &draw, ctx->batch->binning, index_offset, true); |
| 198 |
draw.start += step; |
| 199 |
ctx->batch->num_vertices += step; |
| 200 |
} |
| 201 |
/* changing this value is a hack, restore it */ |
| 202 |
ctx->batch->num_vertices = num_vertices; |
| 203 |
} else { |
| 204 |
draw_impl(ctx, pinfo, pdraw, ctx->batch->draw, index_offset, false); |
| 205 |
draw_impl(ctx, pinfo, pdraw, ctx->batch->binning, index_offset, true); |
| 206 |
} |
| 207 |
|
| 208 |
fd_context_all_clean(ctx); |
| 209 |
|
| 210 |
ctx->batch->num_vertices += pdraw->count * pinfo->instance_count; |
| 211 |
|
| 212 |
return true; |
| 213 |
} |
| 214 |
|
| 215 |
static void |
| 216 |
fd2_draw_vbos(struct fd_context *ctx, const struct pipe_draw_info *info, |
| 217 |
unsigned drawid_offset, |
| 218 |
const struct pipe_draw_indirect_info *indirect, |
| 219 |
const struct pipe_draw_start_count_bias *draws, |
| 220 |
unsigned num_draws, |
| 221 |
unsigned index_offset) |
| 222 |
assert_dt |
| 223 |
{ |
| 224 |
for (unsigned i = 0; i < num_draws; i++) |
| 225 |
fd2_draw_vbo(ctx, info, drawid_offset, indirect, &draws[i], index_offset); |
| 226 |
} |
| 227 |
|
| 228 |
static void |
| 229 |
clear_state(struct fd_batch *batch, struct fd_ringbuffer *ring, |
| 230 |
unsigned buffers, bool fast_clear) assert_dt |
| 231 |
{ |
| 232 |
struct fd_context *ctx = batch->ctx; |
| 233 |
struct fd2_context *fd2_ctx = fd2_context(ctx); |
| 234 |
uint32_t reg; |
| 235 |
|
| 236 |
fd2_emit_vertex_bufs(ring, 0x9c, |
| 237 |
(struct fd2_vertex_buf[]){ |
| 238 |
{.prsc = fd2_ctx->solid_vertexbuf, .size = 36}, |
| 239 |
}, |
| 240 |
1); |
| 241 |
|
| 242 |
OUT_PKT3(ring, CP_SET_CONSTANT, 2); |
| 243 |
OUT_RING(ring, CP_REG(REG_A2XX_VGT_INDX_OFFSET)); |
| 244 |
OUT_RING(ring, 0); |
| 245 |
|
| 246 |
fd2_program_emit(ctx, ring, &ctx->solid_prog); |
| 247 |
|
| 248 |
OUT_PKT0(ring, REG_A2XX_TC_CNTL_STATUS, 1); |
| 249 |
OUT_RING(ring, A2XX_TC_CNTL_STATUS_L2_INVALIDATE); |
| 250 |
|
| 251 |
if (buffers & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL)) { |
| 252 |
OUT_PKT3(ring, CP_SET_CONSTANT, 2); |
| 253 |
OUT_RING(ring, CP_REG(REG_A2XX_RB_DEPTHCONTROL)); |
| 254 |
reg = 0; |
| 255 |
if (buffers & PIPE_CLEAR_DEPTH) { |
| 256 |
reg |= A2XX_RB_DEPTHCONTROL_ZFUNC(FUNC_ALWAYS) | |
| 257 |
A2XX_RB_DEPTHCONTROL_Z_ENABLE | |
| 258 |
A2XX_RB_DEPTHCONTROL_Z_WRITE_ENABLE | |
| 259 |
A2XX_RB_DEPTHCONTROL_EARLY_Z_ENABLE; |
| 260 |
} |
| 261 |
if (buffers & PIPE_CLEAR_STENCIL) { |
| 262 |
reg |= A2XX_RB_DEPTHCONTROL_STENCILFUNC(FUNC_ALWAYS) | |
| 263 |
A2XX_RB_DEPTHCONTROL_STENCIL_ENABLE | |
| 264 |
A2XX_RB_DEPTHCONTROL_STENCILZPASS(STENCIL_REPLACE); |
| 265 |
} |
| 266 |
OUT_RING(ring, reg); |
| 267 |
} |
| 268 |
|
| 269 |
OUT_PKT3(ring, CP_SET_CONSTANT, 2); |
| 270 |
OUT_RING(ring, CP_REG(REG_A2XX_RB_COLORCONTROL)); |
| 271 |
OUT_RING(ring, A2XX_RB_COLORCONTROL_ALPHA_FUNC(FUNC_ALWAYS) | |
| 272 |
A2XX_RB_COLORCONTROL_BLEND_DISABLE | |
| 273 |
A2XX_RB_COLORCONTROL_ROP_CODE(12) | |
| 274 |
A2XX_RB_COLORCONTROL_DITHER_MODE(DITHER_DISABLE) | |
| 275 |
A2XX_RB_COLORCONTROL_DITHER_TYPE(DITHER_PIXEL)); |
| 276 |
|
| 277 |
OUT_PKT3(ring, CP_SET_CONSTANT, 3); |
| 278 |
OUT_RING(ring, CP_REG(REG_A2XX_PA_CL_CLIP_CNTL)); |
| 279 |
OUT_RING(ring, 0x00000000); /* PA_CL_CLIP_CNTL */ |
| 280 |
OUT_RING( |
| 281 |
ring, |
| 282 |
A2XX_PA_SU_SC_MODE_CNTL_PROVOKING_VTX_LAST | /* PA_SU_SC_MODE_CNTL */ |
| 283 |
A2XX_PA_SU_SC_MODE_CNTL_FRONT_PTYPE(PC_DRAW_TRIANGLES) | |
| 284 |
A2XX_PA_SU_SC_MODE_CNTL_BACK_PTYPE(PC_DRAW_TRIANGLES) | |
| 285 |
(fast_clear ? A2XX_PA_SU_SC_MODE_CNTL_MSAA_ENABLE : 0)); |
| 286 |
|
| 287 |
if (fast_clear) { |
| 288 |
OUT_PKT3(ring, CP_SET_CONSTANT, 2); |
| 289 |
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_AA_CONFIG)); |
| 290 |
OUT_RING(ring, A2XX_PA_SC_AA_CONFIG_MSAA_NUM_SAMPLES(3)); |
| 291 |
} |
| 292 |
|
| 293 |
OUT_PKT3(ring, CP_SET_CONSTANT, 2); |
| 294 |
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_AA_MASK)); |
| 295 |
OUT_RING(ring, 0x0000ffff); |
| 296 |
|
| 297 |
OUT_PKT3(ring, CP_SET_CONSTANT, 2); |
| 298 |
OUT_RING(ring, CP_REG(REG_A2XX_RB_COLOR_MASK)); |
| 299 |
if (buffers & PIPE_CLEAR_COLOR) { |
| 300 |
OUT_RING(ring, A2XX_RB_COLOR_MASK_WRITE_RED | |
| 301 |
A2XX_RB_COLOR_MASK_WRITE_GREEN | |
| 302 |
A2XX_RB_COLOR_MASK_WRITE_BLUE | |
| 303 |
A2XX_RB_COLOR_MASK_WRITE_ALPHA); |
| 304 |
} else { |
| 305 |
OUT_RING(ring, 0x0); |
| 306 |
} |
| 307 |
|
| 308 |
OUT_PKT3(ring, CP_SET_CONSTANT, 2); |
| 309 |
OUT_RING(ring, CP_REG(REG_A2XX_RB_BLEND_CONTROL)); |
| 310 |
OUT_RING(ring, 0); |
| 311 |
|
| 312 |
if (is_a20x(batch->ctx->screen)) |
| 313 |
return; |
| 314 |
|
| 315 |
OUT_PKT3(ring, CP_SET_CONSTANT, 3); |
| 316 |
OUT_RING(ring, CP_REG(REG_A2XX_VGT_MAX_VTX_INDX)); |
| 317 |
OUT_RING(ring, 3); /* VGT_MAX_VTX_INDX */ |
| 318 |
OUT_RING(ring, 0); /* VGT_MIN_VTX_INDX */ |
| 319 |
|
| 320 |
OUT_PKT3(ring, CP_SET_CONSTANT, 3); |
| 321 |
OUT_RING(ring, CP_REG(REG_A2XX_RB_STENCILREFMASK_BF)); |
| 322 |
OUT_RING(ring, |
| 323 |
0xff000000 | A2XX_RB_STENCILREFMASK_BF_STENCILWRITEMASK(0xff)); |
| 324 |
OUT_RING(ring, 0xff000000 | A2XX_RB_STENCILREFMASK_STENCILWRITEMASK(0xff)); |
| 325 |
|
| 326 |
OUT_PKT3(ring, CP_SET_CONSTANT, 2); |
| 327 |
OUT_RING(ring, CP_REG(REG_A2XX_A220_RB_LRZ_VSC_CONTROL)); |
| 328 |
OUT_RING(ring, 0x00000084); |
| 329 |
|
| 330 |
OUT_PKT3(ring, CP_SET_CONSTANT, 2); |
| 331 |
OUT_RING(ring, CP_REG(REG_A2XX_VGT_VERTEX_REUSE_BLOCK_CNTL)); |
| 332 |
OUT_RING(ring, 0x0000028f); |
| 333 |
} |
| 334 |
|
| 335 |
static void |
| 336 |
clear_state_restore(struct fd_context *ctx, struct fd_ringbuffer *ring) |
| 337 |
{ |
| 338 |
if (is_a20x(ctx->screen)) |
| 339 |
return; |
| 340 |
|
| 341 |
OUT_PKT3(ring, CP_SET_CONSTANT, 2); |
| 342 |
OUT_RING(ring, CP_REG(REG_A2XX_RB_COPY_CONTROL)); |
| 343 |
OUT_RING(ring, 0x00000000); |
| 344 |
|
| 345 |
OUT_PKT3(ring, CP_SET_CONSTANT, 2); |
| 346 |
OUT_RING(ring, CP_REG(REG_A2XX_A220_RB_LRZ_VSC_CONTROL)); |
| 347 |
OUT_RING(ring, 0x00000000); |
| 348 |
|
| 349 |
OUT_PKT3(ring, CP_SET_CONSTANT, 2); |
| 350 |
OUT_RING(ring, CP_REG(REG_A2XX_VGT_VERTEX_REUSE_BLOCK_CNTL)); |
| 351 |
OUT_RING(ring, 0x0000003b); |
| 352 |
} |
| 353 |
|
| 354 |
static void |
| 355 |
clear_fast(struct fd_batch *batch, struct fd_ringbuffer *ring, |
| 356 |
uint32_t color_clear, uint32_t depth_clear, unsigned patch_type) |
| 357 |
{ |
| 358 |
BEGIN_RING(ring, 8); /* preallocate next 2 packets (for patching) */ |
| 359 |
|
| 360 |
/* zero values are patched in */ |
| 361 |
OUT_PKT3(ring, CP_SET_CONSTANT, 2); |
| 362 |
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_SCREEN_SCISSOR_BR)); |
| 363 |
OUT_RINGP(ring, patch_type, &batch->gmem_patches); |
| 364 |
|
| 365 |
OUT_PKT3(ring, CP_SET_CONSTANT, 4); |
| 366 |
OUT_RING(ring, CP_REG(REG_A2XX_RB_SURFACE_INFO)); |
| 367 |
OUT_RING(ring, 0x8000 | 32); |
| 368 |
OUT_RING(ring, 0); |
| 369 |
OUT_RING(ring, 0); |
| 370 |
|
| 371 |
/* set fill values */ |
| 372 |
if (!is_a20x(batch->ctx->screen)) { |
| 373 |
OUT_PKT3(ring, CP_SET_CONSTANT, 2); |
| 374 |
OUT_RING(ring, CP_REG(REG_A2XX_CLEAR_COLOR)); |
| 375 |
OUT_RING(ring, color_clear); |
| 376 |
|
| 377 |
OUT_PKT3(ring, CP_SET_CONSTANT, 2); |
| 378 |
OUT_RING(ring, CP_REG(REG_A2XX_RB_COPY_CONTROL)); |
| 379 |
OUT_RING(ring, A2XX_RB_COPY_CONTROL_DEPTH_CLEAR_ENABLE | |
| 380 |
A2XX_RB_COPY_CONTROL_CLEAR_MASK(0xf)); |
| 381 |
|
| 382 |
OUT_PKT3(ring, CP_SET_CONSTANT, 2); |
| 383 |
OUT_RING(ring, CP_REG(REG_A2XX_RB_DEPTH_CLEAR)); |
| 384 |
OUT_RING(ring, depth_clear); |
| 385 |
} else { |
| 386 |
const float sc = 1.0f / 255.0f; |
| 387 |
|
| 388 |
OUT_PKT3(ring, CP_SET_CONSTANT, 5); |
| 389 |
OUT_RING(ring, 0x00000480); |
| 390 |
OUT_RING(ring, fui((float)(color_clear >> 0 & 0xff) * sc)); |
| 391 |
OUT_RING(ring, fui((float)(color_clear >> 8 & 0xff) * sc)); |
| 392 |
OUT_RING(ring, fui((float)(color_clear >> 16 & 0xff) * sc)); |
| 393 |
OUT_RING(ring, fui((float)(color_clear >> 24 & 0xff) * sc)); |
| 394 |
|
| 395 |
// XXX if using float the rounding error breaks it.. |
| 396 |
float depth = ((double)(depth_clear >> 8)) * (1.0 / (double)0xffffff); |
| 397 |
assert((unsigned)(((double)depth * (double)0xffffff)) == |
| 398 |
(depth_clear >> 8)); |
| 399 |
|
| 400 |
OUT_PKT3(ring, CP_SET_CONSTANT, 3); |
| 401 |
OUT_RING(ring, CP_REG(REG_A2XX_PA_CL_VPORT_ZSCALE)); |
| 402 |
OUT_RING(ring, fui(0.0f)); |
| 403 |
OUT_RING(ring, fui(depth)); |
| 404 |
|
| 405 |
OUT_PKT3(ring, CP_SET_CONSTANT, 3); |
| 406 |
OUT_RING(ring, CP_REG(REG_A2XX_RB_STENCILREFMASK_BF)); |
| 407 |
OUT_RING(ring, |
| 408 |
0xff000000 | |
| 409 |
A2XX_RB_STENCILREFMASK_BF_STENCILREF(depth_clear & 0xff) | |
| 410 |
A2XX_RB_STENCILREFMASK_BF_STENCILWRITEMASK(0xff)); |
| 411 |
OUT_RING(ring, 0xff000000 | |
| 412 |
A2XX_RB_STENCILREFMASK_STENCILREF(depth_clear & 0xff) | |
| 413 |
A2XX_RB_STENCILREFMASK_STENCILWRITEMASK(0xff)); |
| 414 |
} |
| 415 |
|
| 416 |
fd_draw(batch, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY, |
| 417 |
DI_SRC_SEL_AUTO_INDEX, 3, 0, INDEX_SIZE_IGN, 0, 0, NULL); |
| 418 |
} |
| 419 |
|
| 420 |
static bool |
| 421 |
fd2_clear_fast(struct fd_context *ctx, unsigned buffers, |
| 422 |
const union pipe_color_union *color, double depth, |
| 423 |
unsigned stencil) assert_dt |
| 424 |
{ |
| 425 |
/* using 4x MSAA allows clearing ~2x faster |
| 426 |
* then we can use higher bpp clearing to clear lower bpp |
| 427 |
* 1 "pixel" can clear 64 bits (rgba8+depth24+stencil8) |
| 428 |
* note: its possible to clear with 32_32_32_32 format but its not faster |
| 429 |
* note: fast clear doesn't work with sysmem rendering |
| 430 |
* (sysmem rendering is disabled when clear is used) |
| 431 |
* |
| 432 |
* we only have 16-bit / 32-bit color formats |
| 433 |
* and 16-bit / 32-bit depth formats |
| 434 |
* so there are only a few possible combinations |
| 435 |
* |
| 436 |
* if the bpp of the color/depth doesn't match |
| 437 |
* we clear with depth/color individually |
| 438 |
*/ |
| 439 |
struct fd2_context *fd2_ctx = fd2_context(ctx); |
| 440 |
struct fd_batch *batch = ctx->batch; |
| 441 |
struct fd_ringbuffer *ring = batch->draw; |
| 442 |
struct pipe_framebuffer_state *pfb = &batch->framebuffer; |
| 443 |
uint32_t color_clear = 0, depth_clear = 0; |
| 444 |
enum pipe_format format = pipe_surface_format(&pfb->cbufs[0]); |
| 445 |
int depth_size = -1; /* -1: no clear, 0: clear 16-bit, 1: clear 32-bit */ |
| 446 |
int color_size = -1; |
| 447 |
|
| 448 |
/* TODO: need to test performance on a22x */ |
| 449 |
if (!is_a20x(ctx->screen)) |
| 450 |
return false; |
| 451 |
|
| 452 |
if (buffers & PIPE_CLEAR_COLOR) |
| 453 |
color_size = util_format_get_blocksizebits(format) == 32; |
| 454 |
|
| 455 |
if (buffers & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL)) { |
| 456 |
/* no fast clear when clearing only one component of depth+stencil buffer */ |
| 457 |
if (!(buffers & PIPE_CLEAR_DEPTH)) |
| 458 |
return false; |
| 459 |
|
| 460 |
if ((pfb->zsbuf.format == PIPE_FORMAT_Z24_UNORM_S8_UINT || |
| 461 |
pfb->zsbuf.format == PIPE_FORMAT_S8_UINT_Z24_UNORM) && |
| 462 |
!(buffers & PIPE_CLEAR_STENCIL)) |
| 463 |
return false; |
| 464 |
|
| 465 |
depth_size = fd_pipe2depth(pfb->zsbuf.format) == DEPTHX_24_8; |
| 466 |
} |
| 467 |
|
| 468 |
assert(color_size >= 0 || depth_size >= 0); |
| 469 |
|
| 470 |
if (color_size == 0) { |
| 471 |
color_clear = pack_rgba(format, color->f); |
| 472 |
color_clear = (color_clear << 16) | (color_clear & 0xffff); |
| 473 |
} else if (color_size == 1) { |
| 474 |
color_clear = pack_rgba(format, color->f); |
| 475 |
} |
| 476 |
|
| 477 |
if (depth_size == 0) { |
| 478 |
depth_clear = (uint32_t)(0xffff * depth); |
| 479 |
depth_clear |= depth_clear << 16; |
| 480 |
} else if (depth_size == 1) { |
| 481 |
depth_clear = (((uint32_t)(0xffffff * depth)) << 8); |
| 482 |
depth_clear |= (stencil & 0xff); |
| 483 |
} |
| 484 |
|
| 485 |
/* disable "window" scissor.. */ |
| 486 |
OUT_PKT3(ring, CP_SET_CONSTANT, 3); |
| 487 |
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_WINDOW_SCISSOR_TL)); |
| 488 |
OUT_RING(ring, xy2d(0, 0)); |
| 489 |
OUT_RING(ring, xy2d(0x7fff, 0x7fff)); |
| 490 |
|
| 491 |
/* make sure we fill all "pixels" (in SCREEN_SCISSOR) */ |
| 492 |
OUT_PKT3(ring, CP_SET_CONSTANT, 5); |
| 493 |
OUT_RING(ring, CP_REG(REG_A2XX_PA_CL_VPORT_XSCALE)); |
| 494 |
OUT_RING(ring, fui(4096.0f)); |
| 495 |
OUT_RING(ring, fui(4096.0f)); |
| 496 |
OUT_RING(ring, fui(4096.0f)); |
| 497 |
OUT_RING(ring, fui(4096.0f)); |
| 498 |
|
| 499 |
clear_state(batch, ring, ~0u, true); |
| 500 |
|
| 501 |
if (color_size >= 0 && depth_size != color_size) |
| 502 |
clear_fast(batch, ring, color_clear, color_clear, |
| 503 |
GMEM_PATCH_FASTCLEAR_COLOR); |
| 504 |
|
| 505 |
if (depth_size >= 0 && depth_size != color_size) |
| 506 |
clear_fast(batch, ring, depth_clear, depth_clear, |
| 507 |
GMEM_PATCH_FASTCLEAR_DEPTH); |
| 508 |
|
| 509 |
if (depth_size == color_size) |
| 510 |
clear_fast(batch, ring, color_clear, depth_clear, |
| 511 |
GMEM_PATCH_FASTCLEAR_COLOR_DEPTH); |
| 512 |
|
| 513 |
clear_state_restore(ctx, ring); |
| 514 |
|
| 515 |
OUT_PKT3(ring, CP_SET_CONSTANT, 2); |
| 516 |
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_AA_CONFIG)); |
| 517 |
OUT_RING(ring, 0); |
| 518 |
|
| 519 |
/* can't patch in SCREEN_SCISSOR_BR as it can be different for each tile. |
| 520 |
* MEM_WRITE the value in tile_renderprep, and use CP_LOAD_CONSTANT_CONTEXT |
| 521 |
* the value is read from byte offset 60 in the given bo |
| 522 |
*/ |
| 523 |
OUT_PKT3(ring, CP_LOAD_CONSTANT_CONTEXT, 3); |
| 524 |
OUT_RELOC(ring, fd_resource(fd2_ctx->solid_vertexbuf)->bo, 0, 0, 0); |
| 525 |
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_SCREEN_SCISSOR_BR)); |
| 526 |
OUT_RING(ring, 1); |
| 527 |
|
| 528 |
OUT_PKT3(ring, CP_SET_CONSTANT, 4); |
| 529 |
OUT_RING(ring, CP_REG(REG_A2XX_RB_SURFACE_INFO)); |
| 530 |
OUT_RINGP(ring, GMEM_PATCH_RESTORE_INFO, &batch->gmem_patches); |
| 531 |
OUT_RING(ring, 0); |
| 532 |
OUT_RING(ring, 0); |
| 533 |
return true; |
| 534 |
} |
| 535 |
|
| 536 |
static bool |
| 537 |
fd2_clear(struct fd_context *ctx, enum fd_buffer_mask buffers, |
| 538 |
const union pipe_color_union *color, double depth, |
| 539 |
unsigned stencil) assert_dt |
| 540 |
{ |
| 541 |
struct fd_ringbuffer *ring = ctx->batch->draw; |
| 542 |
struct pipe_framebuffer_state *fb = &ctx->batch->framebuffer; |
| 543 |
|
| 544 |
if (fd2_clear_fast(ctx, buffers, color, depth, stencil)) |
| 545 |
goto dirty; |
| 546 |
|
| 547 |
/* set clear value */ |
| 548 |
if (is_a20x(ctx->screen)) { |
| 549 |
if (buffers & FD_BUFFER_COLOR) { |
| 550 |
/* C0 used by fragment shader */ |
| 551 |
OUT_PKT3(ring, CP_SET_CONSTANT, 5); |
| 552 |
OUT_RING(ring, 0x00000480); |
| 553 |
OUT_RING(ring, color->ui[0]); |
| 554 |
OUT_RING(ring, color->ui[1]); |
| 555 |
OUT_RING(ring, color->ui[2]); |
| 556 |
OUT_RING(ring, color->ui[3]); |
| 557 |
} |
| 558 |
|
| 559 |
if (buffers & FD_BUFFER_DEPTH) { |
| 560 |
/* use viewport to set depth value */ |
| 561 |
OUT_PKT3(ring, CP_SET_CONSTANT, 3); |
| 562 |
OUT_RING(ring, CP_REG(REG_A2XX_PA_CL_VPORT_ZSCALE)); |
| 563 |
OUT_RING(ring, fui(0.0f)); |
| 564 |
OUT_RING(ring, fui(depth)); |
| 565 |
} |
| 566 |
|
| 567 |
if (buffers & FD_BUFFER_STENCIL) { |
| 568 |
OUT_PKT3(ring, CP_SET_CONSTANT, 3); |
| 569 |
OUT_RING(ring, CP_REG(REG_A2XX_RB_STENCILREFMASK_BF)); |
| 570 |
OUT_RING(ring, 0xff000000 | |
| 571 |
A2XX_RB_STENCILREFMASK_BF_STENCILREF(stencil) | |
| 572 |
A2XX_RB_STENCILREFMASK_BF_STENCILWRITEMASK(0xff)); |
| 573 |
OUT_RING(ring, 0xff000000 | |
| 574 |
A2XX_RB_STENCILREFMASK_STENCILREF(stencil) | |
| 575 |
A2XX_RB_STENCILREFMASK_STENCILWRITEMASK(0xff)); |
| 576 |
} |
| 577 |
} else { |
| 578 |
if (buffers & FD_BUFFER_COLOR) { |
| 579 |
OUT_PKT3(ring, CP_SET_CONSTANT, 2); |
| 580 |
OUT_RING(ring, CP_REG(REG_A2XX_CLEAR_COLOR)); |
| 581 |
OUT_RING(ring, pack_rgba(PIPE_FORMAT_R8G8B8A8_UNORM, color->f)); |
| 582 |
} |
| 583 |
|
| 584 |
if (buffers & (FD_BUFFER_DEPTH | FD_BUFFER_STENCIL)) { |
| 585 |
uint32_t clear_mask, depth_clear; |
| 586 |
switch (fd_pipe2depth(fb->zsbuf.format)) { |
| 587 |
case DEPTHX_24_8: |
| 588 |
clear_mask = ((buffers & FD_BUFFER_DEPTH) ? 0xe : 0) | |
| 589 |
((buffers & FD_BUFFER_STENCIL) ? 0x1 : 0); |
| 590 |
depth_clear = |
| 591 |
(((uint32_t)(0xffffff * depth)) << 8) | (stencil & 0xff); |
| 592 |
break; |
| 593 |
case DEPTHX_16: |
| 594 |
clear_mask = 0xf; |
| 595 |
depth_clear = (uint32_t)(0xffffffff * depth); |
| 596 |
break; |
| 597 |
default: |
| 598 |
UNREACHABLE("invalid depth"); |
| 599 |
break; |
| 600 |
} |
| 601 |
|
| 602 |
OUT_PKT3(ring, CP_SET_CONSTANT, 2); |
| 603 |
OUT_RING(ring, CP_REG(REG_A2XX_RB_COPY_CONTROL)); |
| 604 |
OUT_RING(ring, A2XX_RB_COPY_CONTROL_DEPTH_CLEAR_ENABLE | |
| 605 |
A2XX_RB_COPY_CONTROL_CLEAR_MASK(clear_mask)); |
| 606 |
|
| 607 |
OUT_PKT3(ring, CP_SET_CONSTANT, 2); |
| 608 |
OUT_RING(ring, CP_REG(REG_A2XX_RB_DEPTH_CLEAR)); |
| 609 |
OUT_RING(ring, depth_clear); |
| 610 |
} |
| 611 |
} |
| 612 |
|
| 613 |
/* scissor state */ |
| 614 |
OUT_PKT3(ring, CP_SET_CONSTANT, 3); |
| 615 |
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_WINDOW_SCISSOR_TL)); |
| 616 |
OUT_RING(ring, xy2d(0, 0)); |
| 617 |
OUT_RING(ring, xy2d(fb->width, fb->height)); |
| 618 |
|
| 619 |
/* viewport state */ |
| 620 |
OUT_PKT3(ring, CP_SET_CONSTANT, 5); |
| 621 |
OUT_RING(ring, CP_REG(REG_A2XX_PA_CL_VPORT_XSCALE)); |
| 622 |
OUT_RING(ring, fui((float)fb->width / 2.0f)); |
| 623 |
OUT_RING(ring, fui((float)fb->width / 2.0f)); |
| 624 |
OUT_RING(ring, fui((float)fb->height / 2.0f)); |
| 625 |
OUT_RING(ring, fui((float)fb->height / 2.0f)); |
| 626 |
|
| 627 |
/* common state */ |
| 628 |
clear_state(ctx->batch, ring, buffers, false); |
| 629 |
|
| 630 |
fd_draw(ctx->batch, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY, |
| 631 |
DI_SRC_SEL_AUTO_INDEX, 3, 0, INDEX_SIZE_IGN, 0, 0, NULL); |
| 632 |
|
| 633 |
clear_state_restore(ctx, ring); |
| 634 |
|
| 635 |
dirty: |
| 636 |
ctx->dirty |= FD_DIRTY_ZSA | FD_DIRTY_VIEWPORT | FD_DIRTY_RASTERIZER | |
| 637 |
FD_DIRTY_SAMPLE_MASK | FD_DIRTY_PROG | FD_DIRTY_CONST | |
| 638 |
FD_DIRTY_BLEND | FD_DIRTY_FRAMEBUFFER | FD_DIRTY_SCISSOR; |
| 639 |
|
| 640 |
ctx->dirty_shader[MESA_SHADER_VERTEX] |= FD_DIRTY_SHADER_PROG; |
| 641 |
ctx->dirty_shader[MESA_SHADER_FRAGMENT] |= |
| 642 |
FD_DIRTY_SHADER_PROG | FD_DIRTY_SHADER_CONST; |
| 643 |
|
| 644 |
return true; |
| 645 |
} |
| 646 |
|
| 647 |
void |
| 648 |
fd2_draw_init(struct pipe_context *pctx) disable_thread_safety_analysis |
| 649 |
{ |
| 650 |
struct fd_context *ctx = fd_context(pctx); |
| 651 |
ctx->draw_vbos = fd2_draw_vbos; |
| 652 |
ctx->clear = fd2_clear; |
| 653 |
} |