pvr: Add support to create subpass load ops.

Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18707>
This commit is contained in:
Rajnesh Kanwal 2022-08-30 15:50:15 +01:00
parent 0756a28f2d
commit f1d61d8161
2 changed files with 195 additions and 61 deletions

View file

@ -205,21 +205,60 @@ VkResult pvr_pds_unitex_state_program_create_and_upload(
}
static VkResult
pvr_load_op_create(struct pvr_device *device,
const VkAllocationCallbacks *allocator,
struct pvr_renderpass_hwsetup_render *hw_render,
struct pvr_load_op **const load_op_out)
pvr_create_subpass_load_op(struct pvr_device *device,
const VkAllocationCallbacks *allocator,
const struct pvr_render_pass *pass,
struct pvr_renderpass_hwsetup_render *hw_render,
uint32_t hw_subpass_idx,
struct pvr_load_op **const load_op_out)
{
const struct pvr_device_info *dev_info = &device->pdevice->dev_info;
const uint32_t cache_line_size = rogue_get_slc_cache_line_size(dev_info);
struct pvr_load_op *load_op;
VkResult result;
const struct pvr_renderpass_hwsetup_subpass *hw_subpass =
&hw_render->subpasses[hw_subpass_idx];
const struct pvr_render_subpass *subpass =
&pass->subpasses[hw_subpass->index];
load_op = vk_zalloc2(&device->vk.alloc,
allocator,
sizeof(*load_op),
8,
VK_SYSTEM_ALLOCATION_SCOPE_DEVICE);
struct pvr_load_op *load_op = vk_zalloc2(&device->vk.alloc,
allocator,
sizeof(*load_op),
8,
VK_SYSTEM_ALLOCATION_SCOPE_DEVICE);
if (!load_op)
return vk_error(device, VK_ERROR_OUT_OF_HOST_MEMORY);
if (hw_subpass->z_replicate != -1 &&
hw_subpass->depth_initop == VK_ATTACHMENT_LOAD_OP_LOAD) {
pvr_finishme("Missing depth 'load' load op");
load_op->load_depth = true;
}
for (uint32_t i = 0; i < subpass->color_count; i++) {
pvr_finishme("Missing color 'clear' and 'load' load ops");
if (hw_subpass->color_initops[i] == VK_ATTACHMENT_LOAD_OP_CLEAR)
load_op->clear_mask |= 1U << i;
else if (hw_subpass->color_initops[i] == VK_ATTACHMENT_LOAD_OP_LOAD)
pvr_finishme("Missing 'load' load op");
}
load_op->is_hw_object = false;
load_op->subpass = subpass;
*load_op_out = load_op;
return VK_SUCCESS;
}
static VkResult
pvr_create_render_load_op(struct pvr_device *device,
const VkAllocationCallbacks *allocator,
struct pvr_renderpass_hwsetup_render *hw_render,
struct pvr_load_op **const load_op_out)
{
struct pvr_load_op *load_op = vk_zalloc2(&device->vk.alloc,
allocator,
sizeof(*load_op),
8,
VK_SYSTEM_ALLOCATION_SCOPE_DEVICE);
if (!load_op)
return vk_error(device, VK_ERROR_OUT_OF_HOST_MEMORY);
@ -232,13 +271,30 @@ pvr_load_op_create(struct pvr_device *device,
pvr_finishme("Missing 'load' load op");
}
result = pvr_gpu_upload_usc(device,
pvr_usc_fragment_shader,
sizeof(pvr_usc_fragment_shader),
cache_line_size,
&load_op->usc_frag_prog_bo);
load_op->is_hw_object = true;
load_op->hw_render = hw_render;
*load_op_out = load_op;
return VK_SUCCESS;
}
static VkResult
pvr_generate_load_op_shader(struct pvr_device *device,
const VkAllocationCallbacks *allocator,
struct pvr_renderpass_hwsetup_render *hw_render,
struct pvr_load_op *load_op)
{
const struct pvr_device_info *dev_info = &device->pdevice->dev_info;
const uint32_t cache_line_size = rogue_get_slc_cache_line_size(dev_info);
VkResult result = pvr_gpu_upload_usc(device,
pvr_usc_fragment_shader,
sizeof(pvr_usc_fragment_shader),
cache_line_size,
&load_op->usc_frag_prog_bo);
if (result != VK_SUCCESS)
goto err_free_load_op;
return result;
result = pvr_pds_fragment_program_create_and_upload(
device,
@ -260,7 +316,6 @@ pvr_load_op_create(struct pvr_device *device,
if (result != VK_SUCCESS)
goto err_free_pds_frag_prog;
load_op->is_hw_object = true;
/* FIXME: These should be based on the USC and PDS programs, but are hard
* coded for now.
*/
@ -269,8 +324,6 @@ pvr_load_op_create(struct pvr_device *device,
load_op->shareds_count = 1;
load_op->temps_count = 1;
*load_op_out = load_op;
return VK_SUCCESS;
err_free_pds_frag_prog:
@ -279,9 +332,6 @@ err_free_pds_frag_prog:
err_free_usc_frag_prog_bo:
pvr_bo_free(device, load_op->usc_frag_prog_bo);
err_free_load_op:
vk_free2(&device->vk.alloc, allocator, load_op);
return result;
}
@ -303,6 +353,35 @@ static void pvr_load_op_destroy(struct pvr_device *device,
__ret; \
})
static bool
pvr_is_load_op_needed(const struct pvr_render_pass *pass,
struct pvr_renderpass_hwsetup_render *hw_render,
const uint32_t subpass_idx)
{
struct pvr_renderpass_hwsetup_subpass *hw_subpass =
&hw_render->subpasses[subpass_idx];
const struct pvr_render_subpass *subpass =
&pass->subpasses[hw_subpass->index];
if (hw_subpass->z_replicate != -1 &&
(hw_subpass->depth_initop == VK_ATTACHMENT_LOAD_OP_LOAD ||
hw_subpass->depth_initop == VK_ATTACHMENT_LOAD_OP_CLEAR)) {
return true;
}
for (uint32_t i = 0; i < subpass->color_count; i++) {
if (subpass->color_attachments[i] == -1)
continue;
if (hw_subpass->color_initops[i] == VK_ATTACHMENT_LOAD_OP_LOAD ||
hw_subpass->color_initops[i] == VK_ATTACHMENT_LOAD_OP_CLEAR) {
return true;
}
}
return false;
}
VkResult pvr_CreateRenderPass2(VkDevice _device,
const VkRenderPassCreateInfo2 *pCreateInfo,
const VkAllocationCallbacks *pAllocator,
@ -513,51 +592,82 @@ VkResult pvr_CreateRenderPass2(VkDevice _device,
if (hw_render->tile_buffers_count)
pvr_finishme("Set up tile buffer table");
if (hw_render->color_init_count == 0U) {
assert(!hw_render->load_op);
continue;
}
assert(!hw_render->load_op);
/* Add a dummy output register use to the HW render setup if it has no
* output registers in use.
*/
if (!pvr_has_output_register_writes(hw_render)) {
const uint32_t last = hw_render->init_setup.num_render_targets;
struct usc_mrt_resource *mrt_resources;
if (hw_render->color_init_count != 0U) {
/* Add a dummy output register use to the HW render setup if it has no
* output registers in use.
*/
if (!pvr_has_output_register_writes(hw_render)) {
const uint32_t last = hw_render->init_setup.num_render_targets;
struct usc_mrt_resource *mrt_resources;
hw_render->init_setup.num_render_targets++;
hw_render->init_setup.num_render_targets++;
mrt_resources = vk_realloc(alloc,
hw_render->init_setup.mrt_resources,
hw_render->init_setup.num_render_targets *
sizeof(*mrt_resources),
8U,
VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
if (!mrt_resources) {
result = vk_error(device, VK_ERROR_OUT_OF_HOST_MEMORY);
mrt_resources =
vk_realloc(alloc,
hw_render->init_setup.mrt_resources,
hw_render->init_setup.num_render_targets *
sizeof(*mrt_resources),
8U,
VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
if (!mrt_resources) {
result = vk_error(device, VK_ERROR_OUT_OF_HOST_MEMORY);
goto err_load_op_destroy;
}
hw_render->init_setup.mrt_resources = mrt_resources;
mrt_resources[last].type = USC_MRT_RESOURCE_TYPE_OUTPUT_REG;
mrt_resources[last].reg.output_reg = 0U;
mrt_resources[last].reg.offset = 0U;
mrt_resources[last].intermediate_size = 4U;
mrt_resources[last].mrt_desc.intermediate_size = 4U;
mrt_resources[last].mrt_desc.component_alignment = 4U;
mrt_resources[last].mrt_desc.priority = 0U;
mrt_resources[last].mrt_desc.valid_mask[0U] = ~0;
mrt_resources[last].mrt_desc.valid_mask[1U] = ~0;
mrt_resources[last].mrt_desc.valid_mask[2U] = ~0;
mrt_resources[last].mrt_desc.valid_mask[3U] = ~0;
}
result =
pvr_create_render_load_op(device, pAllocator, hw_render, &load_op);
if (result != VK_SUCCESS) {
vk_free2(&device->vk.alloc, pAllocator, load_op);
goto err_load_op_destroy;
}
hw_render->init_setup.mrt_resources = mrt_resources;
result =
pvr_generate_load_op_shader(device, pAllocator, hw_render, load_op);
if (result != VK_SUCCESS)
goto err_load_op_destroy;
mrt_resources[last].type = USC_MRT_RESOURCE_TYPE_OUTPUT_REG;
mrt_resources[last].reg.output_reg = 0U;
mrt_resources[last].reg.offset = 0U;
mrt_resources[last].intermediate_size = 4U;
mrt_resources[last].mrt_desc.intermediate_size = 4U;
mrt_resources[last].mrt_desc.component_alignment = 4U;
mrt_resources[last].mrt_desc.priority = 0U;
mrt_resources[last].mrt_desc.valid_mask[0U] = ~0;
mrt_resources[last].mrt_desc.valid_mask[1U] = ~0;
mrt_resources[last].mrt_desc.valid_mask[2U] = ~0;
mrt_resources[last].mrt_desc.valid_mask[3U] = ~0;
hw_render->load_op = load_op;
}
result = pvr_load_op_create(device, pAllocator, hw_render, &load_op);
if (result != VK_SUCCESS)
goto err_load_op_destroy;
for (uint32_t j = 0; j < hw_render->subpass_count; j++) {
if (!pvr_is_load_op_needed(pass, hw_render, j))
continue;
hw_render->load_op = load_op;
result = pvr_create_subpass_load_op(device,
pAllocator,
pass,
hw_render,
j,
&load_op);
if (result != VK_SUCCESS) {
vk_free2(&device->vk.alloc, pAllocator, load_op);
goto err_load_op_destroy;
}
result =
pvr_generate_load_op_shader(device, pAllocator, hw_render, load_op);
if (result != VK_SUCCESS)
goto err_load_op_destroy;
hw_render->subpasses[j].load_op = load_op;
}
}
*pRenderPass = pvr_render_pass_to_handle(pass);
@ -569,6 +679,14 @@ err_load_op_destroy:
struct pvr_renderpass_hwsetup_render *hw_render =
&pass->hw_setup->renders[i];
for (uint32_t j = 0; j < hw_render->subpass_count; j++) {
if (hw_render->subpasses[j].load_op) {
pvr_load_op_destroy(device,
pAllocator,
hw_render->subpasses[j].load_op);
}
}
if (hw_render->load_op)
pvr_load_op_destroy(device, pAllocator, hw_render->load_op);
}
@ -596,7 +714,16 @@ void pvr_DestroyRenderPass(VkDevice _device,
struct pvr_renderpass_hwsetup_render *hw_render =
&pass->hw_setup->renders[i];
pvr_load_op_destroy(device, pAllocator, hw_render->load_op);
for (uint32_t j = 0; j < hw_render->subpass_count; j++) {
if (hw_render->subpasses[j].load_op) {
pvr_load_op_destroy(device,
pAllocator,
hw_render->subpasses[j].load_op);
}
}
if (hw_render->load_op)
pvr_load_op_destroy(device, pAllocator, hw_render->load_op);
}
pvr_destroy_renderpass_hwsetup(pAllocator ? pAllocator : &device->vk.alloc,

View file

@ -1421,6 +1421,8 @@ struct pvr_load_op {
uint32_t clear_mask;
bool load_depth;
struct pvr_bo *usc_frag_prog_bo;
uint32_t const_shareds_count;
uint32_t shareds_dest_offset;
@ -1430,6 +1432,11 @@ struct pvr_load_op {
struct pvr_pds_upload pds_tex_state_prog;
uint32_t temps_count;
union {
const struct pvr_renderpass_hwsetup_render *hw_render;
const struct pvr_render_subpass *subpass;
};
};
uint32_t pvr_calc_fscommon_size_and_tiles_in_flight(