| 1 |
/* |
| 2 |
* Copyright © 2022 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 <stdbool.h> |
| 26 |
#include <stdint.h> |
| 27 |
#include <vulkan/vulkan.h> |
| 28 |
|
| 29 |
#include "pvr_border.h" |
| 30 |
#include "pvr_cmd_buffer.h" |
| 31 |
#include "pvr_csb.h" |
| 32 |
#include "pvr_debug.h" |
| 33 |
#include "pvr_device.h" |
| 34 |
#include "pvr_job_common.h" |
| 35 |
#include "pvr_job_context.h" |
| 36 |
#include "pvr_job_compute.h" |
| 37 |
#include "pvr_macros.h" |
| 38 |
#include "pvr_physical_device.h" |
| 39 |
#include "pvr_types.h" |
| 40 |
#include "pvr_winsys.h" |
| 41 |
#include "util/macros.h" |
| 42 |
|
| 43 |
static void |
| 44 |
pvr_submit_info_stream_init(struct pvr_compute_ctx *ctx, |
| 45 |
struct pvr_sub_cmd_compute *sub_cmd, |
| 46 |
struct pvr_winsys_compute_submit_info *submit_info) |
| 47 |
{ |
| 48 |
const struct pvr_device *const device = ctx->device; |
| 49 |
const struct pvr_physical_device *const pdevice = device->pdevice; |
| 50 |
const struct pvr_device_runtime_info *const dev_runtime_info = |
| 51 |
&pdevice->dev_runtime_info; |
| 52 |
const struct pvr_device_info *const dev_info = &pdevice->dev_info; |
| 53 |
const struct pvr_compute_ctx_switch *const ctx_switch = &ctx->ctx_switch; |
| 54 |
|
| 55 |
uint32_t *stream_ptr = (uint32_t *)submit_info->fw_stream; |
| 56 |
uint32_t *stream_len_ptr = stream_ptr; |
| 57 |
|
| 58 |
/* Leave space for stream header. */ |
| 59 |
stream_ptr += pvr_cmd_length(KMD_STREAM_HDR); |
| 60 |
|
| 61 |
pvr_csb_pack ((uint64_t *)stream_ptr, |
| 62 |
CR_TPU_BORDER_COLOUR_TABLE_CDM, |
| 63 |
value) { |
| 64 |
value.border_colour_table_address = |
| 65 |
device->border_color_table->table->vma->dev_addr; |
| 66 |
} |
| 67 |
stream_ptr += pvr_cmd_length(CR_TPU_BORDER_COLOUR_TABLE_CDM); |
| 68 |
|
| 69 |
pvr_csb_pack ((uint64_t *)stream_ptr, CR_CDM_CTRL_STREAM_BASE, value) { |
| 70 |
value.addr = pvr_csb_get_start_address(&sub_cmd->control_stream); |
| 71 |
} |
| 72 |
stream_ptr += pvr_cmd_length(CR_CDM_CTRL_STREAM_BASE); |
| 73 |
|
| 74 |
pvr_csb_pack ((uint64_t *)stream_ptr, CR_CDM_CONTEXT_STATE_BASE, state) { |
| 75 |
state.addr = ctx_switch->compute_state_bo->vma->dev_addr; |
| 76 |
} |
| 77 |
stream_ptr += pvr_cmd_length(CR_CDM_CONTEXT_STATE_BASE); |
| 78 |
|
| 79 |
pvr_csb_pack (stream_ptr, CR_CDM_CONTEXT_PDS1, state) { |
| 80 |
const uint32_t load_program_data_size = |
| 81 |
PVR_DW_TO_BYTES(ctx_switch->sr[0].pds.load_program.data_size); |
| 82 |
|
| 83 |
state.pds_seq_dep = false; |
| 84 |
state.usc_seq_dep = false; |
| 85 |
state.target = false; |
| 86 |
state.unified_size = ctx_switch->sr[0].usc.unified_size; |
| 87 |
state.common_shared = true; |
| 88 |
state.common_size = |
| 89 |
DIV_ROUND_UP(sub_cmd->num_shared_regs << 2, |
| 90 |
ROGUE_CR_CDM_CONTEXT_PDS1_COMMON_SIZE_UNIT_SIZE); |
| 91 |
state.temp_size = 0; |
| 92 |
|
| 93 |
assert(load_program_data_size % |
| 94 |
ROGUE_CR_CDM_CONTEXT_PDS1_DATA_SIZE_UNIT_SIZE == |
| 95 |
0); |
| 96 |
state.data_size = |
| 97 |
load_program_data_size / ROGUE_CR_CDM_CONTEXT_PDS1_DATA_SIZE_UNIT_SIZE; |
| 98 |
state.fence = false; |
| 99 |
} |
| 100 |
stream_ptr += pvr_cmd_length(CR_CDM_CONTEXT_PDS1); |
| 101 |
|
| 102 |
if (PVR_HAS_FEATURE(dev_info, compute_morton_capable)) { |
| 103 |
pvr_csb_pack (stream_ptr, CR_CDM_ITEM, value) { |
| 104 |
value.mode = 0; |
| 105 |
} |
| 106 |
stream_ptr += pvr_cmd_length(CR_CDM_ITEM); |
| 107 |
} |
| 108 |
|
| 109 |
if (PVR_HAS_FEATURE(dev_info, cluster_grouping)) { |
| 110 |
pvr_csb_pack (stream_ptr, CR_COMPUTE_CLUSTER, value) { |
| 111 |
if (PVR_HAS_FEATURE(dev_info, slc_mcu_cache_controls) && |
| 112 |
dev_runtime_info->num_phantoms > 1 && sub_cmd->uses_atomic_ops) { |
| 113 |
/* Each phantom has its own MCU, so atomicity can only be |
| 114 |
* guaranteed when all work items are processed on the same |
| 115 |
* phantom. This means we need to disable all USCs other than |
| 116 |
* those of the first phantom, which has 4 clusters. |
| 117 |
*/ |
| 118 |
value.mask = 0xFU; |
| 119 |
} else { |
| 120 |
value.mask = 0U; |
| 121 |
} |
| 122 |
} |
| 123 |
stream_ptr += pvr_cmd_length(CR_COMPUTE_CLUSTER); |
| 124 |
} |
| 125 |
|
| 126 |
if (PVR_HAS_FEATURE(dev_info, tpu_dm_global_registers)) { |
| 127 |
pvr_csb_pack (stream_ptr, CR_TPU_TAG_CDM_CTRL, value) {} |
| 128 |
stream_ptr += pvr_cmd_length(CR_TPU_TAG_CDM_CTRL); |
| 129 |
} |
| 130 |
|
| 131 |
if (PVR_HAS_FEATURE(dev_info, gpu_multicore_support)) { |
| 132 |
if (device->pdevice->dev_runtime_info.core_count > 1) |
| 133 |
pvr_finishme("Emit execute_count, core_count is greater than one"); |
| 134 |
*stream_ptr = 0; |
| 135 |
stream_ptr++; |
| 136 |
} |
| 137 |
|
| 138 |
submit_info->fw_stream_len = |
| 139 |
(uint8_t *)stream_ptr - (uint8_t *)submit_info->fw_stream; |
| 140 |
assert(submit_info->fw_stream_len <= ARRAY_SIZE(submit_info->fw_stream)); |
| 141 |
|
| 142 |
pvr_csb_pack ((uint64_t *)stream_len_ptr, KMD_STREAM_HDR, value) { |
| 143 |
value.length = submit_info->fw_stream_len; |
| 144 |
} |
| 145 |
} |
| 146 |
|
| 147 |
static void pvr_submit_info_ext_stream_init( |
| 148 |
struct pvr_compute_ctx *ctx, |
| 149 |
struct pvr_winsys_compute_submit_info *submit_info) |
| 150 |
{ |
| 151 |
const struct pvr_device_info *const dev_info = |
| 152 |
&ctx->device->pdevice->dev_info; |
| 153 |
|
| 154 |
uint32_t *stream_ptr = (uint32_t *)submit_info->fw_stream; |
| 155 |
uint32_t main_stream_len = |
| 156 |
pvr_csb_unpack((uint64_t *)stream_ptr, KMD_STREAM_HDR).length; |
| 157 |
uint32_t *ext_stream_ptr = |
| 158 |
(uint32_t *)((uint8_t *)stream_ptr + main_stream_len); |
| 159 |
uint32_t *header0_ptr; |
| 160 |
|
| 161 |
header0_ptr = ext_stream_ptr; |
| 162 |
ext_stream_ptr += pvr_cmd_length(KMD_STREAM_EXTHDR_COMPUTE0); |
| 163 |
|
| 164 |
pvr_csb_pack (header0_ptr, KMD_STREAM_EXTHDR_COMPUTE0, header0) { |
| 165 |
if (PVR_HAS_QUIRK(dev_info, 49927)) { |
| 166 |
header0.has_brn49927 = true; |
| 167 |
|
| 168 |
pvr_csb_pack (ext_stream_ptr, CR_TPU, value) { |
| 169 |
value.tag_cem_4k_face_packing = true; |
| 170 |
} |
| 171 |
ext_stream_ptr += pvr_cmd_length(CR_TPU); |
| 172 |
} |
| 173 |
} |
| 174 |
|
| 175 |
if ((*header0_ptr & ROGUE_KMD_STREAM_EXTHDR_DATA_MASK) != 0) { |
| 176 |
submit_info->fw_stream_len = |
| 177 |
(uint8_t *)ext_stream_ptr - (uint8_t *)submit_info->fw_stream; |
| 178 |
assert(submit_info->fw_stream_len <= ARRAY_SIZE(submit_info->fw_stream)); |
| 179 |
} |
| 180 |
} |
| 181 |
|
| 182 |
static void |
| 183 |
pvr_submit_info_flags_init(const struct pvr_device_info *const dev_info, |
| 184 |
const struct pvr_sub_cmd_compute *const sub_cmd, |
| 185 |
struct pvr_winsys_compute_submit_flags *flags) |
| 186 |
{ |
| 187 |
*flags = (struct pvr_winsys_compute_submit_flags){ |
| 188 |
.prevent_all_overlap = sub_cmd->uses_barrier, |
| 189 |
.use_single_core = PVR_HAS_FEATURE(dev_info, gpu_multicore_support) && |
| 190 |
sub_cmd->uses_atomic_ops, |
| 191 |
}; |
| 192 |
} |
| 193 |
|
| 194 |
static void pvr_compute_job_ws_submit_info_init( |
| 195 |
struct pvr_compute_ctx *ctx, |
| 196 |
struct pvr_sub_cmd_compute *sub_cmd, |
| 197 |
struct vk_sync *wait, |
| 198 |
struct pvr_winsys_compute_submit_info *submit_info) |
| 199 |
{ |
| 200 |
const struct pvr_device *const device = ctx->device; |
| 201 |
const struct pvr_device_info *const dev_info = &device->pdevice->dev_info; |
| 202 |
|
| 203 |
memset(submit_info, 0, sizeof(*submit_info)); |
| 204 |
|
| 205 |
submit_info->frame_num = device->global_queue_present_count; |
| 206 |
submit_info->job_num = device->global_cmd_buffer_submit_count; |
| 207 |
|
| 208 |
submit_info->wait = wait; |
| 209 |
|
| 210 |
pvr_submit_info_stream_init(ctx, sub_cmd, submit_info); |
| 211 |
pvr_submit_info_ext_stream_init(ctx, submit_info); |
| 212 |
pvr_submit_info_flags_init(dev_info, sub_cmd, &submit_info->flags); |
| 213 |
} |
| 214 |
|
| 215 |
VkResult pvr_arch_compute_job_submit(struct pvr_compute_ctx *ctx, |
| 216 |
struct pvr_sub_cmd_compute *sub_cmd, |
| 217 |
struct vk_sync *wait, |
| 218 |
struct vk_sync *signal_sync) |
| 219 |
{ |
| 220 |
struct pvr_winsys_compute_submit_info submit_info; |
| 221 |
struct pvr_device *device = ctx->device; |
| 222 |
|
| 223 |
pvr_compute_job_ws_submit_info_init(ctx, sub_cmd, wait, &submit_info); |
| 224 |
|
| 225 |
#ifdef PVR_BUILD_ARCH_ROGUE |
| 226 |
if (PVR_IS_DEBUG_SET(DUMP_CONTROL_STREAM)) { |
| 227 |
pvr_csb_dump(&sub_cmd->control_stream, |
| 228 |
submit_info.frame_num, |
| 229 |
submit_info.job_num); |
| 230 |
} |
| 231 |
#endif |
| 232 |
|
| 233 |
return device->ws->ops->compute_submit(ctx->ws_ctx, |
| 234 |
&submit_info, |
| 235 |
&device->pdevice->dev_info, |
| 236 |
signal_sync); |
| 237 |
} |