| 1 |
/* |
| 2 |
* Copyright © 2023 Imagination Technologies Ltd. |
| 3 |
* |
| 4 |
* Permission is hereby granted, free of charge, to any person obtaining a copy |
| 5 |
* of this software and associated documentation files (the "Software"), to deal |
| 6 |
* in the Software without restriction, including without limitation the rights |
| 7 |
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell |
| 8 |
* copies of the Software, and to permit persons to whom the Software is |
| 9 |
* 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 THE |
| 18 |
* 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 |
|
| 24 |
#include <assert.h> |
| 25 |
#include <stddef.h> |
| 26 |
#include <stdint.h> |
| 27 |
#include <vulkan/vulkan_core.h> |
| 28 |
|
| 29 |
#include "hwdef/rogue_hw_utils.h" |
| 30 |
#include "pvr_bo.h" |
| 31 |
#include "pvr_common.h" |
| 32 |
#include "pvr_device.h" |
| 33 |
#include "pvr_device_info.h" |
| 34 |
#include "pvr_pds.h" |
| 35 |
#include "pvr_physical_device.h" |
| 36 |
#include "pvr_transfer_frag_store.h" |
| 37 |
#include "pvr_types.h" |
| 38 |
#include "pvr_usc.h" |
| 39 |
#include "util/hash_table.h" |
| 40 |
#include "util/macros.h" |
| 41 |
#include "util/ralloc.h" |
| 42 |
#include "util/u_dynarray.h" |
| 43 |
#include "util/u_math.h" |
| 44 |
#include "vk_log.h" |
| 45 |
|
| 46 |
#define PVR_TRANSFER_BYTE_UNWIND_MAX 16U |
| 47 |
|
| 48 |
struct pvr_transfer_frag_store_entry_data { |
| 49 |
pvr_dev_addr_t kick_usc_pds_offset; |
| 50 |
struct pvr_bo *kick_usc_pds_upload; |
| 51 |
|
| 52 |
struct pvr_suballoc_bo *usc_upload; |
| 53 |
struct pvr_tq_frag_sh_reg_layout sh_reg_layout; |
| 54 |
}; |
| 55 |
|
| 56 |
#define to_pvr_entry_data(_entry) \ |
| 57 |
_Generic((_entry), \ |
| 58 |
struct hash_entry *: (struct pvr_transfer_frag_store_entry_data \ |
| 59 |
*)((_entry)->data), \ |
| 60 |
const struct hash_entry *: ( \ |
| 61 |
const struct pvr_transfer_frag_store_entry_data *)((_entry) \ |
| 62 |
->data)) |
| 63 |
|
| 64 |
VkResult pvr_transfer_frag_store_init(struct pvr_device *device, |
| 65 |
struct pvr_transfer_frag_store *store) |
| 66 |
{ |
| 67 |
const struct pvr_device_info *dev_info = &device->pdevice->dev_info; |
| 68 |
|
| 69 |
*store = (struct pvr_transfer_frag_store){ |
| 70 |
.max_multisample = PVR_GET_FEATURE_VALUE(dev_info, max_multisample, 1U), |
| 71 |
.hash_table = _mesa_hash_table_create_u32_keys(NULL), |
| 72 |
}; |
| 73 |
|
| 74 |
if (!store->hash_table) |
| 75 |
return vk_error(device, VK_ERROR_OUT_OF_HOST_MEMORY); |
| 76 |
|
| 77 |
return VK_SUCCESS; |
| 78 |
} |
| 79 |
|
| 80 |
/** |
| 81 |
* \brief Returns a key based on shader properties. |
| 82 |
* |
| 83 |
* Returns a unique key that can be used to uniquely identify a transfer |
| 84 |
* fragment shader based on the provided shader properties. |
| 85 |
* |
| 86 |
* Make sure that the non valid parts of shader_props are memset to 0. Otherwise |
| 87 |
* these bits might appear in the key as uninitialized data and might not |
| 88 |
* match a key for the same shader. |
| 89 |
*/ |
| 90 |
static uint32_t pvr_transfer_frag_shader_key( |
| 91 |
uint32_t max_multisample, |
| 92 |
const struct pvr_tq_shader_properties *shader_props) |
| 93 |
{ |
| 94 |
const struct pvr_tq_layer_properties *layer = &shader_props->layer_props; |
| 95 |
uint32_t resolve_op_num = max_multisample + PVR_RESOLVE_SAMPLE0; |
| 96 |
|
| 97 |
uint32_t num_layers_bits = util_logbase2_ceil(PVR_TRANSFER_MAX_LAYERS + 1U); |
| 98 |
uint32_t layer_float_bits = util_logbase2_ceil(PVR_INT_COORD_SET_FLOATS_NUM); |
| 99 |
uint32_t pixel_src_bits = util_logbase2_ceil(PVR_TRANSFER_PBE_PIXEL_SRC_NUM); |
| 100 |
uint32_t byte_unwind_bits = util_logbase2_ceil(PVR_TRANSFER_BYTE_UNWIND_MAX); |
| 101 |
uint32_t resolve_op_bits = util_logbase2_ceil(resolve_op_num); |
| 102 |
uint32_t sample_cnt_bits = util_last_bit(util_logbase2(max_multisample)); |
| 103 |
uint32_t hash = 0U; |
| 104 |
|
| 105 |
#if MESA_DEBUG |
| 106 |
uint32_t max_shift = 0U; |
| 107 |
# define shift_hash(hash, num) \ |
| 108 |
do { \ |
| 109 |
max_shift += (num); \ |
| 110 |
assert(max_shift <= 32U); \ |
| 111 |
\ |
| 112 |
(hash) <<= (num); \ |
| 113 |
} while (0U) |
| 114 |
#else |
| 115 |
# define shift_hash(hash, num) hash <<= (num) |
| 116 |
#endif |
| 117 |
|
| 118 |
/* Hash layer info. */ |
| 119 |
|
| 120 |
shift_hash(hash, layer_float_bits); |
| 121 |
hash |= (uint32_t)shader_props->layer_props.layer_floats; |
| 122 |
|
| 123 |
shift_hash(hash, 1U); |
| 124 |
hash |= layer->sample; |
| 125 |
|
| 126 |
shift_hash(hash, 1U); |
| 127 |
hash |= (uint32_t)false; |
| 128 |
|
| 129 |
shift_hash(hash, 1U); |
| 130 |
hash |= (uint32_t)false; |
| 131 |
|
| 132 |
shift_hash(hash, pixel_src_bits); |
| 133 |
hash |= (uint32_t)layer->pbe_format; |
| 134 |
|
| 135 |
shift_hash(hash, resolve_op_bits); |
| 136 |
hash |= (uint32_t)layer->resolve_op; |
| 137 |
|
| 138 |
assert(util_is_power_of_two_nonzero(layer->sample_count)); |
| 139 |
shift_hash(hash, sample_cnt_bits); |
| 140 |
hash |= (uint32_t)util_logbase2(layer->sample_count); |
| 141 |
|
| 142 |
shift_hash(hash, 1U); |
| 143 |
hash |= (uint32_t)layer->msaa; |
| 144 |
|
| 145 |
shift_hash(hash, byte_unwind_bits); |
| 146 |
hash |= layer->byte_unwind; |
| 147 |
|
| 148 |
shift_hash(hash, 1U); |
| 149 |
hash |= (uint32_t)layer->linear; |
| 150 |
|
| 151 |
/* End layer info. */ |
| 152 |
|
| 153 |
shift_hash(hash, 1U); |
| 154 |
hash |= (uint32_t)shader_props->full_rate; |
| 155 |
|
| 156 |
shift_hash(hash, 1U); |
| 157 |
hash |= (uint32_t)shader_props->iterated; |
| 158 |
|
| 159 |
shift_hash(hash, 1U); |
| 160 |
hash |= (uint32_t)shader_props->pick_component; |
| 161 |
|
| 162 |
shift_hash(hash, num_layers_bits); |
| 163 |
/* Just 1 layer. */ |
| 164 |
hash |= 1; |
| 165 |
|
| 166 |
shift_hash(hash, 3U); |
| 167 |
/* alpha type none */ |
| 168 |
hash |= 0; |
| 169 |
|
| 170 |
#undef shift_hash |
| 171 |
|
| 172 |
return hash; |
| 173 |
} |
| 174 |
|
| 175 |
#define to_hash_table_key(_key) ((void *)(uintptr_t)(_key)) |
| 176 |
|
| 177 |
static VkResult pvr_transfer_frag_store_entry_data_compile( |
| 178 |
struct pvr_device *device, |
| 179 |
struct pvr_transfer_frag_store_entry_data *const entry_data, |
| 180 |
const struct pvr_tq_shader_properties *shader_props, |
| 181 |
uint32_t *const num_usc_temps_out) |
| 182 |
{ |
| 183 |
const uint32_t image_desc_offset = |
| 184 |
offsetof(struct pvr_combined_image_sampler_descriptor, image) / 4; |
| 185 |
const uint32_t sampler_desc_offset = |
| 186 |
offsetof(struct pvr_combined_image_sampler_descriptor, sampler) / 4; |
| 187 |
|
| 188 |
const uint32_t cache_line_size = |
| 189 |
pvr_get_slc_cache_line_size(&device->pdevice->dev_info); |
| 190 |
|
| 191 |
struct pvr_tq_frag_sh_reg_layout *sh_reg_layout = &entry_data->sh_reg_layout; |
| 192 |
uint32_t next_free_sh_reg = 0; |
| 193 |
VkResult result; |
| 194 |
|
| 195 |
/* TODO: Allocate all combined image samplers if needed? Otherwise change the |
| 196 |
* array to a single descriptor. |
| 197 |
*/ |
| 198 |
sh_reg_layout->combined_image_samplers.offsets[0].image = |
| 199 |
next_free_sh_reg + image_desc_offset; |
| 200 |
sh_reg_layout->combined_image_samplers.offsets[0].sampler = |
| 201 |
next_free_sh_reg + sampler_desc_offset; |
| 202 |
sh_reg_layout->combined_image_samplers.count = 1; |
| 203 |
next_free_sh_reg += sizeof(struct pvr_combined_image_sampler_descriptor) / 4; |
| 204 |
|
| 205 |
/* TODO: Handle dynamic_const_regs used for PVR_INT_COORD_SET_FLOATS_{4,6}, Z |
| 206 |
* position, texel unwind, etc. when compiler adds support for them. |
| 207 |
*/ |
| 208 |
sh_reg_layout->dynamic_consts.offset = next_free_sh_reg; |
| 209 |
sh_reg_layout->dynamic_consts.count = 0; |
| 210 |
|
| 211 |
sh_reg_layout->driver_total = next_free_sh_reg; |
| 212 |
|
| 213 |
pco_shader *tq = |
| 214 |
pvr_uscgen_tq(device->pdevice->pco_ctx, shader_props, sh_reg_layout); |
| 215 |
|
| 216 |
*num_usc_temps_out = pco_shader_data(tq)->common.temps; |
| 217 |
|
| 218 |
result = pvr_gpu_upload_usc(device, |
| 219 |
pco_shader_binary_data(tq), |
| 220 |
pco_shader_binary_size(tq), |
| 221 |
cache_line_size, |
| 222 |
&entry_data->usc_upload); |
| 223 |
|
| 224 |
ralloc_free(tq); |
| 225 |
|
| 226 |
if (result != VK_SUCCESS) |
| 227 |
return result; |
| 228 |
|
| 229 |
return VK_SUCCESS; |
| 230 |
} |
| 231 |
|
| 232 |
static uint32_t get_doutu_sample_rate(const struct pvr_device_info *dev_info, |
| 233 |
bool full_rate); |
| 234 |
|
| 235 |
static VkResult pvr_transfer_frag_store_entry_data_create( |
| 236 |
struct pvr_device *device, |
| 237 |
struct pvr_transfer_frag_store *store, |
| 238 |
const struct pvr_tq_shader_properties *shader_props, |
| 239 |
const struct pvr_transfer_frag_store_entry_data **const entry_data_out) |
| 240 |
{ |
| 241 |
struct pvr_pds_kickusc_program kick_usc_pds_prog = { 0 }; |
| 242 |
struct pvr_transfer_frag_store_entry_data *entry_data; |
| 243 |
pvr_dev_addr_t dev_addr; |
| 244 |
uint32_t num_usc_temps; |
| 245 |
VkResult result; |
| 246 |
|
| 247 |
entry_data = ralloc(store->hash_table, __typeof__(*entry_data)); |
| 248 |
if (!entry_data) |
| 249 |
return vk_error(device, VK_ERROR_OUT_OF_HOST_MEMORY); |
| 250 |
|
| 251 |
result = pvr_transfer_frag_store_entry_data_compile(device, |
| 252 |
entry_data, |
| 253 |
shader_props, |
| 254 |
&num_usc_temps); |
| 255 |
if (result != VK_SUCCESS) |
| 256 |
goto err_free_entry; |
| 257 |
|
| 258 |
dev_addr = entry_data->usc_upload->dev_addr; |
| 259 |
dev_addr.addr -= device->heaps.usc_heap->base_addr.addr; |
| 260 |
|
| 261 |
pvr_pds_setup_doutu(&kick_usc_pds_prog.usc_task_control, |
| 262 |
dev_addr.addr, |
| 263 |
num_usc_temps, |
| 264 |
get_doutu_sample_rate(&device->pdevice->dev_info, |
| 265 |
shader_props->full_rate), |
| 266 |
false); |
| 267 |
|
| 268 |
pvr_pds_kick_usc(&kick_usc_pds_prog, NULL, 0U, false, PDS_GENERATE_SIZES); |
| 269 |
|
| 270 |
result = pvr_bo_alloc(device, |
| 271 |
device->heaps.pds_heap, |
| 272 |
PVR_DW_TO_BYTES(kick_usc_pds_prog.data_size + |
| 273 |
kick_usc_pds_prog.code_size), |
| 274 |
16, |
| 275 |
PVR_BO_ALLOC_FLAG_CPU_MAPPED, |
| 276 |
&entry_data->kick_usc_pds_upload); |
| 277 |
if (result != VK_SUCCESS) |
| 278 |
goto err_free_usc_upload; |
| 279 |
|
| 280 |
pvr_pds_kick_usc(&kick_usc_pds_prog, |
| 281 |
entry_data->kick_usc_pds_upload->bo->map, |
| 282 |
0U, |
| 283 |
false, |
| 284 |
PDS_GENERATE_CODEDATA_SEGMENTS); |
| 285 |
|
| 286 |
dev_addr = entry_data->kick_usc_pds_upload->vma->dev_addr; |
| 287 |
dev_addr.addr -= device->heaps.pds_heap->base_addr.addr; |
| 288 |
entry_data->kick_usc_pds_offset = dev_addr; |
| 289 |
|
| 290 |
*entry_data_out = entry_data; |
| 291 |
|
| 292 |
return VK_SUCCESS; |
| 293 |
|
| 294 |
err_free_usc_upload: |
| 295 |
pvr_bo_suballoc_free(entry_data->usc_upload); |
| 296 |
|
| 297 |
err_free_entry: |
| 298 |
ralloc_free(entry_data); |
| 299 |
|
| 300 |
return result; |
| 301 |
} |
| 302 |
|
| 303 |
static void inline pvr_transfer_frag_store_entry_data_destroy_no_ralloc_free( |
| 304 |
struct pvr_device *device, |
| 305 |
const struct pvr_transfer_frag_store_entry_data *entry_data) |
| 306 |
{ |
| 307 |
pvr_bo_free(device, entry_data->kick_usc_pds_upload); |
| 308 |
pvr_bo_suballoc_free(entry_data->usc_upload); |
| 309 |
} |
| 310 |
|
| 311 |
static void inline pvr_transfer_frag_store_entry_data_destroy( |
| 312 |
struct pvr_device *device, |
| 313 |
const struct pvr_transfer_frag_store_entry_data *entry_data) |
| 314 |
{ |
| 315 |
pvr_transfer_frag_store_entry_data_destroy_no_ralloc_free(device, |
| 316 |
entry_data); |
| 317 |
/* Casting away the const :( */ |
| 318 |
ralloc_free((void *)entry_data); |
| 319 |
} |
| 320 |
|
| 321 |
static VkResult pvr_transfer_frag_store_get_entry( |
| 322 |
struct pvr_device *device, |
| 323 |
struct pvr_transfer_frag_store *store, |
| 324 |
const struct pvr_tq_shader_properties *shader_props, |
| 325 |
const struct pvr_transfer_frag_store_entry_data **const entry_data_out) |
| 326 |
{ |
| 327 |
const uint32_t key = |
| 328 |
pvr_transfer_frag_shader_key(store->max_multisample, shader_props); |
| 329 |
const struct hash_entry *entry; |
| 330 |
VkResult result; |
| 331 |
|
| 332 |
entry = _mesa_hash_table_search(store->hash_table, to_hash_table_key(key)); |
| 333 |
if (!entry) { |
| 334 |
/* Init so that gcc stops complaining. */ |
| 335 |
const struct pvr_transfer_frag_store_entry_data *entry_data = NULL; |
| 336 |
|
| 337 |
result = pvr_transfer_frag_store_entry_data_create(device, |
| 338 |
store, |
| 339 |
shader_props, |
| 340 |
&entry_data); |
| 341 |
if (result != VK_SUCCESS) |
| 342 |
return result; |
| 343 |
|
| 344 |
assert(entry_data); |
| 345 |
|
| 346 |
entry = _mesa_hash_table_insert(store->hash_table, |
| 347 |
to_hash_table_key(key), |
| 348 |
(void *)entry_data); |
| 349 |
if (!entry) { |
| 350 |
pvr_transfer_frag_store_entry_data_destroy(device, entry_data); |
| 351 |
return vk_error(device, VK_ERROR_OUT_OF_HOST_MEMORY); |
| 352 |
} |
| 353 |
} |
| 354 |
|
| 355 |
*entry_data_out = to_pvr_entry_data(entry); |
| 356 |
|
| 357 |
return VK_SUCCESS; |
| 358 |
} |
| 359 |
|
| 360 |
VkResult pvr_transfer_frag_store_get_shader_info( |
| 361 |
struct pvr_device *device, |
| 362 |
struct pvr_transfer_frag_store *store, |
| 363 |
const struct pvr_tq_shader_properties *shader_props, |
| 364 |
pvr_dev_addr_t *const pds_dev_addr_out, |
| 365 |
const struct pvr_tq_frag_sh_reg_layout **const reg_layout_out) |
| 366 |
{ |
| 367 |
/* Init so that gcc stops complaining. */ |
| 368 |
const struct pvr_transfer_frag_store_entry_data *entry_data = NULL; |
| 369 |
VkResult result; |
| 370 |
|
| 371 |
result = pvr_transfer_frag_store_get_entry(device, |
| 372 |
store, |
| 373 |
shader_props, |
| 374 |
&entry_data); |
| 375 |
if (result != VK_SUCCESS) |
| 376 |
return result; |
| 377 |
|
| 378 |
*pds_dev_addr_out = entry_data->kick_usc_pds_offset; |
| 379 |
*reg_layout_out = &entry_data->sh_reg_layout; |
| 380 |
|
| 381 |
return VK_SUCCESS; |
| 382 |
} |
| 383 |
|
| 384 |
void pvr_transfer_frag_store_fini(struct pvr_device *device, |
| 385 |
struct pvr_transfer_frag_store *store) |
| 386 |
{ |
| 387 |
hash_table_foreach_remove(store->hash_table, entry) |
| 388 |
{ |
| 389 |
/* ralloc_free() in _mesa_hash_table_destroy() will free each entry's |
| 390 |
* memory so let's not waste extra time freeing them one by one and |
| 391 |
* unliking. |
| 392 |
*/ |
| 393 |
pvr_transfer_frag_store_entry_data_destroy_no_ralloc_free( |
| 394 |
device, |
| 395 |
to_pvr_entry_data(entry)); |
| 396 |
} |
| 397 |
|
| 398 |
_mesa_hash_table_destroy(store->hash_table, NULL); |
| 399 |
} |
| 400 |
|
| 401 |
/* Leave this at the very end, to avoid leakage of HW-defs here */ |
| 402 |
#define PVR_BUILD_ARCH_ROGUE |
| 403 |
#include "pvr_csb.h" |
| 404 |
|
| 405 |
static uint32_t get_doutu_sample_rate(const struct pvr_device_info *dev_info, |
| 406 |
bool full_rate) |
| 407 |
{ |
| 408 |
return full_rate ? ROGUE_PDSINST_DOUTU_SAMPLE_RATE_FULL |
| 409 |
: ROGUE_PDSINST_DOUTU_SAMPLE_RATE_INSTANCE; |
| 410 |
} |