KonstantinSeurer
/
mesa
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

pvr: Move transfer logic to pvr_job_transfer.[ch] 

Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16897>
This commit is contained in:
Rajnesh Kanwal 2022-04-19 17:21:16 +01:00
parent 16d4ca6e14
commit 9acb8ba3db
4 changed files with 184 additions and 64 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
src/imagination/vulkan/meson.build
View File

@ -49,6 +49,7 @@ pvr_files = files(
'pvr_job_compute.c',
'pvr_job_context.c',
'pvr_job_render.c',
'pvr_job_transfer.c',
'pvr_pass.c',
'pvr_pipeline.c',
'pvr_pipeline_cache.c',

111
src/imagination/vulkan/pvr_job_transfer.c Normal file
View File

@ -0,0 +1,111 @@
/*
* Copyright © 2022 Imagination Technologies Ltd.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
#include <stdbool.h>
#include <stdint.h>
#include <stddef.h>
#include <vulkan/vulkan.h>
#include "pvr_job_common.h"
#include "pvr_job_context.h"
#include "pvr_job_transfer.h"
#include "pvr_private.h"
#include "pvr_winsys.h"
#include "util/list.h"
#include "util/macros.h"
#include "vk_sync.h"
/* FIXME: Implement gpu based transfer support. */
VkResult pvr_transfer_job_submit(struct pvr_device *device,
struct pvr_transfer_ctx *ctx,
struct pvr_sub_cmd *sub_cmd,
struct vk_sync **waits,
uint32_t wait_count,
uint32_t *stage_flags,
struct vk_sync *signal_sync)
{
/* Wait for transfer semaphores here before doing any transfers. */
for (uint32_t i = 0U; i < wait_count; i++) {
if (stage_flags[i] & PVR_PIPELINE_STAGE_TRANSFER_BIT) {
VkResult result = vk_sync_wait(&device->vk,
waits[i],
0U,
VK_SYNC_WAIT_COMPLETE,
UINT64_MAX);
if (result != VK_SUCCESS)
return result;
stage_flags[i] &= ~PVR_PIPELINE_STAGE_TRANSFER_BIT;
}
}
list_for_each_entry_safe (struct pvr_transfer_cmd,
transfer_cmd,
&sub_cmd->transfer.transfer_cmds,
link) {
bool src_mapped = false;
bool dst_mapped = false;
void *src_addr;
void *dst_addr;
void *ret_ptr;
/* Map if bo is not mapped. */
if (!transfer_cmd->src->vma->bo->map) {
src_mapped = true;
ret_ptr = device->ws->ops->buffer_map(transfer_cmd->src->vma->bo);
if (!ret_ptr)
return vk_error(device, VK_ERROR_MEMORY_MAP_FAILED);
}
if (!transfer_cmd->dst->vma->bo->map) {
dst_mapped = true;
ret_ptr = device->ws->ops->buffer_map(transfer_cmd->dst->vma->bo);
if (!ret_ptr)
return vk_error(device, VK_ERROR_MEMORY_MAP_FAILED);
}
src_addr =
transfer_cmd->src->vma->bo->map + transfer_cmd->src->vma->bo_offset;
dst_addr =
transfer_cmd->dst->vma->bo->map + transfer_cmd->dst->vma->bo_offset;
for (uint32_t i = 0; i < transfer_cmd->region_count; i++) {
VkBufferCopy2 *region = &transfer_cmd->regions[i];
memcpy(dst_addr + region->dstOffset,
src_addr + region->srcOffset,
region->size);
}
if (src_mapped)
device->ws->ops->buffer_unmap(transfer_cmd->src->vma->bo);
if (dst_mapped)
device->ws->ops->buffer_unmap(transfer_cmd->dst->vma->bo);
}
/* Given we are doing CPU based copy, completion fence should always be
* signaled. This should be fixed when GPU based copy is implemented.
*/
return vk_sync_signal(&device->vk, signal_sync, 0);
}

43
src/imagination/vulkan/pvr_job_transfer.h Normal file
View File

@ -0,0 +1,43 @@
/*
* Copyright © 2022 Imagination Technologies Ltd.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
#ifndef PVR_JOB_TRANSFER_H
#define PVR_JOB_TRANSFER_H
#include <stdint.h>
#include <vulkan/vulkan.h>
struct pvr_device;
struct pvr_sub_cmd;
struct pvr_transfer_ctx;
struct vk_sync;
VkResult pvr_transfer_job_submit(struct pvr_device *device,
struct pvr_transfer_ctx *ctx,
struct pvr_sub_cmd *sub_cmd,
struct vk_sync **waits,
uint32_t wait_count,
uint32_t *stage_flags,
struct vk_sync *signal_sync);
#endif /* PVR_JOB_TRANSFER_H */

93
src/imagination/vulkan/pvr_queue.c
View File

@ -39,6 +39,7 @@
#include "pvr_job_compute.h"
#include "pvr_job_context.h"
#include "pvr_job_render.h"
#include "pvr_job_transfer.h"
#include "pvr_limits.h"
#include "pvr_private.h"
#include "util/macros.h"
@ -358,83 +359,46 @@ pvr_process_compute_cmd(struct pvr_device *device,
return result;
}
/* FIXME: Implement gpu based transfer support. */
static VkResult
pvr_process_transfer_cmds(struct pvr_device *device,
struct pvr_queue *queue,
struct pvr_sub_cmd *sub_cmd,
struct vk_sync **waits,
uint32_t wait_count,
uint32_t *stage_flags,
struct vk_sync *completions[static PVR_JOB_TYPE_MAX])
{
/* Wait for transfer semaphores here before doing any transfers. */
for (uint32_t i = 0U; i < wait_count; i++) {
if (stage_flags[i] & PVR_PIPELINE_STAGE_TRANSFER_BIT) {
VkResult result = vk_sync_wait(&device->vk,
waits[i],
0U,
VK_SYNC_WAIT_COMPLETE,
UINT64_MAX);
if (result != VK_SUCCESS)
return result;
struct vk_sync *sync;
VkResult result;
stage_flags[i] &= ~PVR_PIPELINE_STAGE_TRANSFER_BIT;
}
result = vk_sync_create(&device->vk,
&device->pdevice->ws->syncobj_type,
0U,
0UL,
&sync);
if (result != VK_SUCCESS)
return result;
/* This passes ownership of the wait fences to pvr_transfer_job_submit(). */
result = pvr_transfer_job_submit(device,
queue->transfer_ctx,
sub_cmd,
waits,
wait_count,
stage_flags,
sync);
if (result != VK_SUCCESS) {
vk_sync_destroy(&device->vk, sync);
return result;
}
list_for_each_entry_safe (struct pvr_transfer_cmd,
transfer_cmd,
&sub_cmd->transfer.transfer_cmds,
link) {
bool src_mapped = false;
bool dst_mapped = false;
void *src_addr;
void *dst_addr;
void *ret_ptr;
/* Replace the completion fences. */
if (completions[PVR_JOB_TYPE_TRANSFER])
vk_sync_destroy(&device->vk, completions[PVR_JOB_TYPE_TRANSFER]);
/* Map if bo is not mapped. */
if (!transfer_cmd->src->vma->bo->map) {
src_mapped = true;
ret_ptr = device->ws->ops->buffer_map(transfer_cmd->src->vma->bo);
if (!ret_ptr)
return vk_error(device, VK_ERROR_MEMORY_MAP_FAILED);
}
completions[PVR_JOB_TYPE_TRANSFER] = sync;
if (!transfer_cmd->dst->vma->bo->map) {
dst_mapped = true;
ret_ptr = device->ws->ops->buffer_map(transfer_cmd->dst->vma->bo);
if (!ret_ptr)
return vk_error(device, VK_ERROR_MEMORY_MAP_FAILED);
}
src_addr =
transfer_cmd->src->vma->bo->map + transfer_cmd->src->vma->bo_offset;
dst_addr =
transfer_cmd->dst->vma->bo->map + transfer_cmd->dst->vma->bo_offset;
for (uint32_t i = 0; i < transfer_cmd->region_count; i++) {
VkBufferCopy2 *region = &transfer_cmd->regions[i];
memcpy(dst_addr + region->dstOffset,
src_addr + region->srcOffset,
region->size);
}
if (src_mapped)
device->ws->ops->buffer_unmap(transfer_cmd->src->vma->bo);
if (dst_mapped)
device->ws->ops->buffer_unmap(transfer_cmd->dst->vma->bo);
}
/* Given we are doing CPU based copy, completion fence should always be
* signaled. This should be fixed when GPU based copy is implemented.
*/
return vk_sync_create(&device->vk,
&device->pdevice->ws->syncobj_type,
0U,
1UL,
&completions[PVR_JOB_TYPE_TRANSFER]);
return result;
}
static VkResult
@ -571,6 +535,7 @@ pvr_process_cmd_buffer(struct pvr_device *device,
case PVR_SUB_CMD_TYPE_TRANSFER:
result = pvr_process_transfer_cmds(device,
queue,
sub_cmd,
waits,
wait_count,