mirrors
/
mesa
mirror of https://gitlab.freedesktop.org/mesa/mesa
1
0
Fork 0
Code Issues 9 Packages Projects Releases Wiki Activity
mesa/src/nouveau/vulkan/nvk_cmd_copy.c

824 lines
31 KiB
C
Raw Permalink Blame History

/*
* Copyright © 2022 Collabora Ltd. and Red Hat Inc.
* SPDX-License-Identifier: MIT
*/
#include "nvk_cmd_buffer.h"
#include "nvk_buffer.h"
#include "nvk_device.h"
#include "nvk_device_memory.h"
#include "nvk_entrypoints.h"
#include "nvk_format.h"
#include "nvk_image.h"
#include "nvk_image_view.h"
#include "nvk_physical_device.h"
#include "vk_format.h"
#include "nouveau_bo.h"
#include "nouveau_context.h"
#include "nvtypes.h"
#include "nv_push_cl902d.h"
#include "nv_push_cl90b5.h"
#include "nv_push_clc1b5.h"
static inline uint16_t
nvk_cmd_buffer_copy_cls(struct nvk_cmd_buffer *cmd)
{
struct nvk_device *dev = nvk_cmd_buffer_device(cmd);
struct nvk_physical_device *pdev = nvk_device_physical(dev);
return pdev->info.cls_copy;
}
struct nouveau_copy_buffer {
uint64_t base_addr;
VkImageType image_type;
struct nil_Offset4D_Elements offset_el;
struct nil_Extent4D_Elements extent_el;
uint32_t bpp;
uint32_t row_stride;
uint32_t array_stride;
struct nil_tiling tiling;
};
struct nouveau_copy {
struct nouveau_copy_buffer src;
struct nouveau_copy_buffer dst;
struct nouveau_copy_remap {
uint8_t comp_size;
uint8_t dst[4];
} remap;
struct nil_Extent4D_Elements extent_el;
};
static struct nouveau_copy_buffer
nouveau_copy_rect_buffer(struct nvk_buffer *buf,
VkDeviceSize offset,
struct vk_image_buffer_layout buffer_layout)
{
return (struct nouveau_copy_buffer) {
.base_addr = nvk_buffer_address(buf, offset),
.image_type = VK_IMAGE_TYPE_2D,
.bpp = buffer_layout.element_size_B,
.row_stride = buffer_layout.row_stride_B,
.array_stride = buffer_layout.image_stride_B,
};
}
static struct nil_Offset4D_Pixels
vk_to_nil_offset(VkOffset3D offset, uint32_t base_array_layer)
{
return (struct nil_Offset4D_Pixels) {
.x = offset.x,
.y = offset.y,
.z = offset.z,
.a = base_array_layer
};
}
static struct nil_Extent4D_Pixels
vk_to_nil_extent(VkExtent3D extent, uint32_t array_layers)
{
return (struct nil_Extent4D_Pixels) {
.width = extent.width,
.height = extent.height,
.depth = extent.depth,
.array_len = array_layers,
};
}
static struct nouveau_copy_buffer
nouveau_copy_rect_image(const struct nvk_image *img,
const struct nvk_image_plane *plane,
VkOffset3D offset_px,
const VkImageSubresourceLayers *sub_res)
{
const struct nil_Extent4D_Pixels lvl_extent4d_px =
nil_image_level_extent_px(&plane->nil, sub_res->mipLevel);
offset_px = vk_image_sanitize_offset(&img->vk, offset_px);
const struct nil_Offset4D_Pixels offset4d_px =
vk_to_nil_offset(offset_px, sub_res->baseArrayLayer);
struct nouveau_copy_buffer buf = {
.base_addr = nvk_image_plane_base_address(plane) +
plane->nil.levels[sub_res->mipLevel].offset_B,
.image_type = img->vk.image_type,
.offset_el = nil_offset4d_px_to_el(offset4d_px, plane->nil.format,
plane->nil.sample_layout),
.extent_el = nil_extent4d_px_to_el(lvl_extent4d_px, plane->nil.format,
plane->nil.sample_layout),
.bpp = util_format_get_blocksize(plane->nil.format.p_format),
.row_stride = plane->nil.levels[sub_res->mipLevel].row_stride_B,
.array_stride = plane->nil.array_stride_B,
.tiling = plane->nil.levels[sub_res->mipLevel].tiling,
};
return buf;
}
static struct nouveau_copy_remap
nouveau_copy_remap_format(VkFormat format)
{
/* Pick an arbitrary component size. It doesn't matter what size we
* pick since we're just doing a copy, as long as it's no more than 4B
* and divides the format size.
*/
unsigned comp_size = vk_format_get_blocksize(format);
if (comp_size % 3 == 0) {
comp_size /= 3;
assert(util_is_power_of_two_nonzero(comp_size) && comp_size <= 4);
} else {
assert(util_is_power_of_two_nonzero(comp_size) && comp_size <= 16);
comp_size = MIN2(comp_size, 4);
}
return (struct nouveau_copy_remap) {
.comp_size = comp_size,
.dst = { 0, 1, 2, 3 },
};
}
static uint32_t
to_90b5_remap_comp_size(uint8_t comp_size)
{
static const uint8_t to_90b5[] = {
[1] = NV90B5_SET_REMAP_COMPONENTS_COMPONENT_SIZE_ONE,
[2] = NV90B5_SET_REMAP_COMPONENTS_COMPONENT_SIZE_TWO,
[3] = NV90B5_SET_REMAP_COMPONENTS_COMPONENT_SIZE_THREE,
[4] = NV90B5_SET_REMAP_COMPONENTS_COMPONENT_SIZE_FOUR,
};
assert(comp_size > 0 && comp_size < ARRAY_SIZE(to_90b5));
uint32_t size_90b5 = comp_size - 1;
assert(size_90b5 == to_90b5[comp_size]);
return size_90b5;
}
static uint32_t
to_90b5_remap_num_comps(uint8_t num_comps)
{
static const uint8_t to_90b5[] = {
[1] = NV90B5_SET_REMAP_COMPONENTS_NUM_SRC_COMPONENTS_ONE,
[2] = NV90B5_SET_REMAP_COMPONENTS_NUM_SRC_COMPONENTS_TWO,
[3] = NV90B5_SET_REMAP_COMPONENTS_NUM_SRC_COMPONENTS_THREE,
[4] = NV90B5_SET_REMAP_COMPONENTS_NUM_SRC_COMPONENTS_FOUR,
};
assert(num_comps > 0 && num_comps < ARRAY_SIZE(to_90b5));
uint32_t num_comps_90b5 = num_comps - 1;
assert(num_comps_90b5 == to_90b5[num_comps]);
return num_comps_90b5;
}
static void
nouveau_copy_rect(struct nvk_cmd_buffer *cmd, struct nouveau_copy *copy)
{
uint32_t src_bw, dst_bw;
if (copy->remap.comp_size > 0) {
struct nv_push *p = nvk_cmd_buffer_push(cmd, 2);
assert(copy->src.bpp % copy->remap.comp_size == 0);
assert(copy->dst.bpp % copy->remap.comp_size == 0);
uint32_t num_src_comps = copy->src.bpp / copy->remap.comp_size;
uint32_t num_dst_comps = copy->dst.bpp / copy->remap.comp_size;
/* When running with component remapping enabled, most X/Y dimensions
* are in units of blocks.
*/
src_bw = dst_bw = 1;
P_IMMD(p, NV90B5, SET_REMAP_COMPONENTS, {
.dst_x = copy->remap.dst[0],
.dst_y = copy->remap.dst[1],
.dst_z = copy->remap.dst[2],
.dst_w = copy->remap.dst[3],
.component_size = to_90b5_remap_comp_size(copy->remap.comp_size),
.num_src_components = to_90b5_remap_comp_size(num_src_comps),
.num_dst_components = to_90b5_remap_comp_size(num_dst_comps),
});
} else {
/* When component remapping is disabled, dimensions are in units of
* bytes (an implicit block widht of 1B).
*/
assert(copy->src.bpp == copy->dst.bpp);
src_bw = copy->src.bpp;
dst_bw = copy->dst.bpp;
}
assert(copy->extent_el.depth == 1 || copy->extent_el.array_len == 1);
uint32_t layers = MAX2(copy->extent_el.depth, copy->extent_el.array_len);
for (unsigned z = 0; z < layers; z++) {
VkDeviceSize src_addr = copy->src.base_addr;
VkDeviceSize dst_addr = copy->dst.base_addr;
if (copy->src.image_type != VK_IMAGE_TYPE_3D)
src_addr += (z + copy->src.offset_el.a) * copy->src.array_stride;
if (copy->dst.image_type != VK_IMAGE_TYPE_3D)
dst_addr += (z + copy->dst.offset_el.a) * copy->dst.array_stride;
if (!copy->src.tiling.is_tiled) {
src_addr += copy->src.offset_el.x * copy->src.bpp +
copy->src.offset_el.y * copy->src.row_stride;
}
if (!copy->dst.tiling.is_tiled) {
dst_addr += copy->dst.offset_el.x * copy->dst.bpp +
copy->dst.offset_el.y * copy->dst.row_stride;
}
struct nv_push *p = nvk_cmd_buffer_push(cmd, 31);
P_MTHD(p, NV90B5, OFFSET_IN_UPPER);
P_NV90B5_OFFSET_IN_UPPER(p, src_addr >> 32);
P_NV90B5_OFFSET_IN_LOWER(p, src_addr & 0xffffffff);
P_NV90B5_OFFSET_OUT_UPPER(p, dst_addr >> 32);
P_NV90B5_OFFSET_OUT_LOWER(p, dst_addr & 0xffffffff);
P_NV90B5_PITCH_IN(p, copy->src.row_stride);
P_NV90B5_PITCH_OUT(p, copy->dst.row_stride);
P_NV90B5_LINE_LENGTH_IN(p, copy->extent_el.width * src_bw);
P_NV90B5_LINE_COUNT(p, copy->extent_el.height);
uint32_t src_layout = 0, dst_layout = 0;
if (copy->src.tiling.is_tiled) {
P_MTHD(p, NV90B5, SET_SRC_BLOCK_SIZE);
P_NV90B5_SET_SRC_BLOCK_SIZE(p, {
.width = 0, /* Tiles are always 1 GOB wide */
.height = copy->src.tiling.y_log2,
.depth = copy->src.tiling.z_log2,
.gob_height = copy->src.tiling.gob_height_is_8 ?
GOB_HEIGHT_GOB_HEIGHT_FERMI_8 :
GOB_HEIGHT_GOB_HEIGHT_TESLA_4,
});
/* We use the stride for copies because the copy hardware has no
* concept of a tile width. Instead, we just set the width to the
* stride divided by bpp.
*/
uint32_t src_stride_el = copy->src.row_stride / copy->src.bpp;
P_NV90B5_SET_SRC_WIDTH(p, src_stride_el * src_bw);
P_NV90B5_SET_SRC_HEIGHT(p, copy->src.extent_el.height);
P_NV90B5_SET_SRC_DEPTH(p, copy->src.extent_el.depth);
if (copy->src.image_type == VK_IMAGE_TYPE_3D)
P_NV90B5_SET_SRC_LAYER(p, z + copy->src.offset_el.z);
else
P_NV90B5_SET_SRC_LAYER(p, 0);
if (nvk_cmd_buffer_copy_cls(cmd) >= PASCAL_DMA_COPY_B) {
P_MTHD(p, NVC1B5, SRC_ORIGIN_X);
P_NVC1B5_SRC_ORIGIN_X(p, copy->src.offset_el.x * src_bw);
P_NVC1B5_SRC_ORIGIN_Y(p, copy->src.offset_el.y);
} else {
P_MTHD(p, NV90B5, SET_SRC_ORIGIN);
P_NV90B5_SET_SRC_ORIGIN(p, {
.x = copy->src.offset_el.x * src_bw,
.y = copy->src.offset_el.y
});
}
src_layout = NV90B5_LAUNCH_DMA_SRC_MEMORY_LAYOUT_BLOCKLINEAR;
} else {
src_addr += copy->src.array_stride;
src_layout = NV90B5_LAUNCH_DMA_SRC_MEMORY_LAYOUT_PITCH;
}
if (copy->dst.tiling.is_tiled) {
P_MTHD(p, NV90B5, SET_DST_BLOCK_SIZE);
P_NV90B5_SET_DST_BLOCK_SIZE(p, {
.width = 0, /* Tiles are always 1 GOB wide */
.height = copy->dst.tiling.y_log2,
.depth = copy->dst.tiling.z_log2,
.gob_height = copy->dst.tiling.gob_height_is_8 ?
GOB_HEIGHT_GOB_HEIGHT_FERMI_8 :
GOB_HEIGHT_GOB_HEIGHT_TESLA_4,
});
/* We use the stride for copies because the copy hardware has no
* concept of a tile width. Instead, we just set the width to the
* stride divided by bpp.
*/
uint32_t dst_stride_el = copy->dst.row_stride / copy->dst.bpp;
P_NV90B5_SET_DST_WIDTH(p, dst_stride_el * dst_bw);
P_NV90B5_SET_DST_HEIGHT(p, copy->dst.extent_el.height);
P_NV90B5_SET_DST_DEPTH(p, copy->dst.extent_el.depth);
if (copy->dst.image_type == VK_IMAGE_TYPE_3D)
P_NV90B5_SET_DST_LAYER(p, z + copy->dst.offset_el.z);
else
P_NV90B5_SET_DST_LAYER(p, 0);
if (nvk_cmd_buffer_copy_cls(cmd) >= PASCAL_DMA_COPY_B) {
P_MTHD(p, NVC1B5, DST_ORIGIN_X);
P_NVC1B5_DST_ORIGIN_X(p, copy->dst.offset_el.x * dst_bw);
P_NVC1B5_DST_ORIGIN_Y(p, copy->dst.offset_el.y);
} else {
P_MTHD(p, NV90B5, SET_DST_ORIGIN);
P_NV90B5_SET_DST_ORIGIN(p, {
.x = copy->dst.offset_el.x * dst_bw,
.y = copy->dst.offset_el.y
});
}
dst_layout = NV90B5_LAUNCH_DMA_DST_MEMORY_LAYOUT_BLOCKLINEAR;
} else {
dst_addr += copy->dst.array_stride;
dst_layout = NV90B5_LAUNCH_DMA_DST_MEMORY_LAYOUT_PITCH;
}
P_IMMD(p, NV90B5, LAUNCH_DMA, {
.data_transfer_type = DATA_TRANSFER_TYPE_NON_PIPELINED,
.multi_line_enable = MULTI_LINE_ENABLE_TRUE,
.flush_enable = FLUSH_ENABLE_TRUE,
.src_memory_layout = src_layout,
.dst_memory_layout = dst_layout,
.remap_enable = copy->remap.comp_size > 0,
});
}
}
VKAPI_ATTR void VKAPI_CALL
nvk_CmdCopyBuffer2(VkCommandBuffer commandBuffer,
const VkCopyBufferInfo2 *pCopyBufferInfo)
{
VK_FROM_HANDLE(nvk_cmd_buffer, cmd, commandBuffer);
VK_FROM_HANDLE(nvk_buffer, src, pCopyBufferInfo->srcBuffer);
VK_FROM_HANDLE(nvk_buffer, dst, pCopyBufferInfo->dstBuffer);
for (unsigned r = 0; r < pCopyBufferInfo->regionCount; r++) {
const VkBufferCopy2 *region = &pCopyBufferInfo->pRegions[r];
uint64_t src_addr = nvk_buffer_address(src, region->srcOffset);
uint64_t dst_addr = nvk_buffer_address(dst, region->dstOffset);
uint64_t size = region->size;
while (size) {
struct nv_push *p = nvk_cmd_buffer_push(cmd, 10);
P_MTHD(p, NV90B5, OFFSET_IN_UPPER);
P_NV90B5_OFFSET_IN_UPPER(p, src_addr >> 32);
P_NV90B5_OFFSET_IN_LOWER(p, src_addr & 0xffffffff);
P_NV90B5_OFFSET_OUT_UPPER(p, dst_addr >> 32);
P_NV90B5_OFFSET_OUT_LOWER(p, dst_addr & 0xffffffff);
unsigned bytes = MIN2(size, 1 << 17);
P_MTHD(p, NV90B5, LINE_LENGTH_IN);
P_NV90B5_LINE_LENGTH_IN(p, bytes);
P_NV90B5_LINE_COUNT(p, 1);
P_IMMD(p, NV90B5, LAUNCH_DMA, {
.data_transfer_type = DATA_TRANSFER_TYPE_NON_PIPELINED,
.multi_line_enable = MULTI_LINE_ENABLE_TRUE,
.flush_enable = FLUSH_ENABLE_TRUE,
.src_memory_layout = SRC_MEMORY_LAYOUT_PITCH,
.dst_memory_layout = DST_MEMORY_LAYOUT_PITCH,
});
src_addr += bytes;
dst_addr += bytes;
size -= bytes;
}
}
}
VKAPI_ATTR void VKAPI_CALL
nvk_CmdCopyBufferToImage2(VkCommandBuffer commandBuffer,
const VkCopyBufferToImageInfo2 *pCopyBufferToImageInfo)
{
VK_FROM_HANDLE(nvk_cmd_buffer, cmd, commandBuffer);
VK_FROM_HANDLE(nvk_buffer, src, pCopyBufferToImageInfo->srcBuffer);
VK_FROM_HANDLE(nvk_image, dst, pCopyBufferToImageInfo->dstImage);
for (unsigned r = 0; r < pCopyBufferToImageInfo->regionCount; r++) {
const VkBufferImageCopy2 *region = &pCopyBufferToImageInfo->pRegions[r];
struct vk_image_buffer_layout buffer_layout =
vk_image_buffer_copy_layout(&dst->vk, region);
const VkExtent3D extent_px =
vk_image_sanitize_extent(&dst->vk, region->imageExtent);
const uint32_t layer_count =
vk_image_subresource_layer_count(&dst->vk, &region->imageSubresource);
const struct nil_Extent4D_Pixels extent4d_px =
vk_to_nil_extent(extent_px, layer_count);
const VkImageAspectFlagBits aspects = region->imageSubresource.aspectMask;
uint8_t plane = nvk_image_aspects_to_plane(dst, aspects);
struct nouveau_copy copy = {
.src = nouveau_copy_rect_buffer(src, region->bufferOffset,
buffer_layout),
.dst = nouveau_copy_rect_image(dst, &dst->planes[plane],
region->imageOffset,
&region->imageSubresource),
.extent_el = nil_extent4d_px_to_el(extent4d_px, dst->planes[plane].nil.format,
dst->planes[plane].nil.sample_layout),
};
struct nouveau_copy copy2 = { 0 };
switch (dst->vk.format) {
case VK_FORMAT_D32_SFLOAT_S8_UINT:
if (aspects == VK_IMAGE_ASPECT_DEPTH_BIT) {
copy.remap.comp_size = 4;
copy.remap.dst[0] = NV90B5_SET_REMAP_COMPONENTS_DST_X_SRC_X;
copy.remap.dst[1] = NV90B5_SET_REMAP_COMPONENTS_DST_Y_NO_WRITE;
copy.remap.dst[2] = NV90B5_SET_REMAP_COMPONENTS_DST_Z_NO_WRITE;
copy.remap.dst[3] = NV90B5_SET_REMAP_COMPONENTS_DST_W_NO_WRITE;
} else {
assert(aspects == VK_IMAGE_ASPECT_STENCIL_BIT);
copy2.dst = copy.dst;
copy2.extent_el = copy.extent_el;
copy.dst = copy2.src =
nouveau_copy_rect_image(dst, &dst->stencil_copy_temp,
region->imageOffset,
&region->imageSubresource);
copy.remap.comp_size = 1;
copy.remap.dst[0] = NV90B5_SET_REMAP_COMPONENTS_DST_X_SRC_X;
copy.remap.dst[1] = NV90B5_SET_REMAP_COMPONENTS_DST_Y_NO_WRITE;
copy.remap.dst[2] = NV90B5_SET_REMAP_COMPONENTS_DST_Z_NO_WRITE;
copy.remap.dst[3] = NV90B5_SET_REMAP_COMPONENTS_DST_W_NO_WRITE;
copy2.remap.comp_size = 2;
copy2.remap.dst[0] = NV90B5_SET_REMAP_COMPONENTS_DST_X_NO_WRITE;
copy2.remap.dst[1] = NV90B5_SET_REMAP_COMPONENTS_DST_Y_NO_WRITE;
copy2.remap.dst[2] = NV90B5_SET_REMAP_COMPONENTS_DST_Z_SRC_X;
copy2.remap.dst[3] = NV90B5_SET_REMAP_COMPONENTS_DST_W_NO_WRITE;
}
break;
case VK_FORMAT_D24_UNORM_S8_UINT:
if (aspects == VK_IMAGE_ASPECT_DEPTH_BIT) {
copy.remap.comp_size = 1;
copy.remap.dst[0] = NV90B5_SET_REMAP_COMPONENTS_DST_X_SRC_X;
copy.remap.dst[1] = NV90B5_SET_REMAP_COMPONENTS_DST_Y_SRC_Y;
copy.remap.dst[2] = NV90B5_SET_REMAP_COMPONENTS_DST_Z_SRC_Z;
copy.remap.dst[3] = NV90B5_SET_REMAP_COMPONENTS_DST_W_NO_WRITE;
} else {
assert(aspects == VK_IMAGE_ASPECT_STENCIL_BIT);
copy.remap.comp_size = 1;
copy.remap.dst[0] = NV90B5_SET_REMAP_COMPONENTS_DST_X_NO_WRITE;
copy.remap.dst[1] = NV90B5_SET_REMAP_COMPONENTS_DST_Y_NO_WRITE;
copy.remap.dst[2] = NV90B5_SET_REMAP_COMPONENTS_DST_Z_NO_WRITE;
copy.remap.dst[3] = NV90B5_SET_REMAP_COMPONENTS_DST_W_SRC_X;
}
break;
default:
copy.remap = nouveau_copy_remap_format(dst->vk.format);
break;
}
nouveau_copy_rect(cmd, &copy);
if (copy2.extent_el.width > 0)
nouveau_copy_rect(cmd, &copy2);
vk_foreach_struct_const(ext, region->pNext) {
switch (ext->sType) {
default:
vk_debug_ignored_stype(ext->sType);
break;
}
}
}
vk_foreach_struct_const(ext, pCopyBufferToImageInfo->pNext) {
switch (ext->sType) {
default:
vk_debug_ignored_stype(ext->sType);
break;
}
}
}
VKAPI_ATTR void VKAPI_CALL
nvk_CmdCopyImageToBuffer2(VkCommandBuffer commandBuffer,
const VkCopyImageToBufferInfo2 *pCopyImageToBufferInfo)
{
VK_FROM_HANDLE(nvk_cmd_buffer, cmd, commandBuffer);
VK_FROM_HANDLE(nvk_image, src, pCopyImageToBufferInfo->srcImage);
VK_FROM_HANDLE(nvk_buffer, dst, pCopyImageToBufferInfo->dstBuffer);
for (unsigned r = 0; r < pCopyImageToBufferInfo->regionCount; r++) {
const VkBufferImageCopy2 *region = &pCopyImageToBufferInfo->pRegions[r];
struct vk_image_buffer_layout buffer_layout =
vk_image_buffer_copy_layout(&src->vk, region);
const VkExtent3D extent_px =
vk_image_sanitize_extent(&src->vk, region->imageExtent);
const uint32_t layer_count =
vk_image_subresource_layer_count(&src->vk, &region->imageSubresource);
const struct nil_Extent4D_Pixels extent4d_px =
vk_to_nil_extent(extent_px, layer_count);
const VkImageAspectFlagBits aspects = region->imageSubresource.aspectMask;
uint8_t plane = nvk_image_aspects_to_plane(src, aspects);
struct nouveau_copy copy = {
.src = nouveau_copy_rect_image(src, &src->planes[plane],
region->imageOffset,
&region->imageSubresource),
.dst = nouveau_copy_rect_buffer(dst, region->bufferOffset,
buffer_layout),
.extent_el = nil_extent4d_px_to_el(extent4d_px, src->planes[plane].nil.format,
src->planes[plane].nil.sample_layout),
};
struct nouveau_copy copy2 = { 0 };
switch (src->vk.format) {
case VK_FORMAT_D32_SFLOAT_S8_UINT:
if (aspects == VK_IMAGE_ASPECT_DEPTH_BIT) {
copy.remap.comp_size = 4;
copy.remap.dst[0] = NV90B5_SET_REMAP_COMPONENTS_DST_X_SRC_X;
copy.remap.dst[1] = NV90B5_SET_REMAP_COMPONENTS_DST_Y_NO_WRITE;
copy.remap.dst[2] = NV90B5_SET_REMAP_COMPONENTS_DST_Z_NO_WRITE;
copy.remap.dst[3] = NV90B5_SET_REMAP_COMPONENTS_DST_W_NO_WRITE;
} else {
assert(aspects == VK_IMAGE_ASPECT_STENCIL_BIT);
copy2.dst = copy.dst;
copy2.extent_el = copy.extent_el;
copy.dst = copy2.src =
nouveau_copy_rect_image(src, &src->stencil_copy_temp,
region->imageOffset,
&region->imageSubresource);
copy.remap.comp_size = 2;
copy.remap.dst[0] = NV90B5_SET_REMAP_COMPONENTS_DST_X_SRC_Z;
copy.remap.dst[1] = NV90B5_SET_REMAP_COMPONENTS_DST_Y_NO_WRITE;
copy.remap.dst[2] = NV90B5_SET_REMAP_COMPONENTS_DST_Z_NO_WRITE;
copy.remap.dst[3] = NV90B5_SET_REMAP_COMPONENTS_DST_W_NO_WRITE;
copy2.remap.comp_size = 1;
copy2.remap.dst[0] = NV90B5_SET_REMAP_COMPONENTS_DST_X_SRC_X;
copy2.remap.dst[1] = NV90B5_SET_REMAP_COMPONENTS_DST_Y_NO_WRITE;
copy2.remap.dst[2] = NV90B5_SET_REMAP_COMPONENTS_DST_Z_NO_WRITE;
copy2.remap.dst[3] = NV90B5_SET_REMAP_COMPONENTS_DST_W_NO_WRITE;
}
break;
case VK_FORMAT_D24_UNORM_S8_UINT:
if (aspects == VK_IMAGE_ASPECT_DEPTH_BIT) {
copy.remap.comp_size = 1;
copy.remap.dst[0] = NV90B5_SET_REMAP_COMPONENTS_DST_X_SRC_X;
copy.remap.dst[1] = NV90B5_SET_REMAP_COMPONENTS_DST_Y_SRC_Y;
copy.remap.dst[2] = NV90B5_SET_REMAP_COMPONENTS_DST_Z_SRC_Z;
copy.remap.dst[3] = NV90B5_SET_REMAP_COMPONENTS_DST_W_NO_WRITE;
} else {
assert(aspects == VK_IMAGE_ASPECT_STENCIL_BIT);
copy.remap.comp_size = 1;
copy.remap.dst[0] = NV90B5_SET_REMAP_COMPONENTS_DST_X_SRC_W;
copy.remap.dst[1] = NV90B5_SET_REMAP_COMPONENTS_DST_Y_NO_WRITE;
copy.remap.dst[2] = NV90B5_SET_REMAP_COMPONENTS_DST_Z_NO_WRITE;
copy.remap.dst[3] = NV90B5_SET_REMAP_COMPONENTS_DST_W_NO_WRITE;
}
break;
default:
copy.remap = nouveau_copy_remap_format(src->vk.format);
break;
}
nouveau_copy_rect(cmd, &copy);
if (copy2.extent_el.width > 0)
nouveau_copy_rect(cmd, &copy2);
vk_foreach_struct_const(ext, region->pNext) {
switch (ext->sType) {
default:
vk_debug_ignored_stype(ext->sType);
break;
}
}
}
vk_foreach_struct_const(ext, pCopyImageToBufferInfo->pNext) {
switch (ext->sType) {
default:
vk_debug_ignored_stype(ext->sType);
break;
}
}
}
void
nvk_linear_render_copy(struct nvk_cmd_buffer *cmd,
const struct nvk_image_view *iview,
VkRect2D copy_rect,
bool copy_to_tiled_shadow)
{
const struct nvk_image *image = (struct nvk_image *)iview->vk.image;
const uint8_t ip = iview->planes[0].image_plane;
const struct nvk_image_plane *src_plane = NULL, *dst_plane = NULL;
if (copy_to_tiled_shadow) {
src_plane = &image->planes[ip];
dst_plane = &image->linear_tiled_shadow;
} else {
src_plane = &image->linear_tiled_shadow;
dst_plane = &image->planes[ip];
}
const struct VkImageSubresourceLayers subres = {
.aspectMask = iview->vk.aspects,
.baseArrayLayer = iview->vk.base_array_layer,
.layerCount = iview->vk.layer_count,
.mipLevel = iview->vk.base_mip_level,
};
const VkOffset3D offset_px = {
.x = copy_rect.offset.x,
.y = copy_rect.offset.y,
.z = 0,
};
const struct nil_Extent4D_Pixels extent4d_px = {
.width = copy_rect.extent.width,
.height = copy_rect.extent.height,
.depth = 1,
.array_len = 1,
};
struct nouveau_copy copy = {
.src = nouveau_copy_rect_image(image, src_plane, offset_px, &subres),
.dst = nouveau_copy_rect_image(image, dst_plane, offset_px, &subres),
.extent_el = nil_extent4d_px_to_el(extent4d_px, src_plane->nil.format,
src_plane->nil.sample_layout),
};
copy.remap = nouveau_copy_remap_format(image->vk.format);
nouveau_copy_rect(cmd, &copy);
}
VKAPI_ATTR void VKAPI_CALL
nvk_CmdCopyImage2(VkCommandBuffer commandBuffer,
const VkCopyImageInfo2 *pCopyImageInfo)
{
VK_FROM_HANDLE(nvk_cmd_buffer, cmd, commandBuffer);
VK_FROM_HANDLE(nvk_image, src, pCopyImageInfo->srcImage);
VK_FROM_HANDLE(nvk_image, dst, pCopyImageInfo->dstImage);
for (unsigned r = 0; r < pCopyImageInfo->regionCount; r++) {
const VkImageCopy2 *region = &pCopyImageInfo->pRegions[r];
/* From the Vulkan 1.3.217 spec:
*
* "When copying between compressed and uncompressed formats the
* extent members represent the texel dimensions of the source image
* and not the destination."
*/
const VkExtent3D extent_px =
vk_image_sanitize_extent(&src->vk, region->extent);
const uint32_t layer_count =
vk_image_subresource_layer_count(&src->vk, &region->srcSubresource);
const struct nil_Extent4D_Pixels extent4d_px =
vk_to_nil_extent(extent_px, layer_count);
const VkImageAspectFlagBits src_aspects =
region->srcSubresource.aspectMask;
uint8_t src_plane = nvk_image_aspects_to_plane(src, src_aspects);
const VkImageAspectFlagBits dst_aspects =
region->dstSubresource.aspectMask;
uint8_t dst_plane = nvk_image_aspects_to_plane(dst, dst_aspects);
struct nouveau_copy copy = {
.src = nouveau_copy_rect_image(src, &src->planes[src_plane],
region->srcOffset,
&region->srcSubresource),
.dst = nouveau_copy_rect_image(dst, &dst->planes[dst_plane],
region->dstOffset,
&region->dstSubresource),
.extent_el = nil_extent4d_px_to_el(extent4d_px, src->planes[src_plane].nil.format,
src->planes[src_plane].nil.sample_layout),
};
assert(src_aspects == region->srcSubresource.aspectMask);
switch (src->vk.format) {
case VK_FORMAT_D24_UNORM_S8_UINT:
if (src_aspects == VK_IMAGE_ASPECT_DEPTH_BIT) {
copy.remap.comp_size = 1;
copy.remap.dst[0] = NV90B5_SET_REMAP_COMPONENTS_DST_W_SRC_X;
copy.remap.dst[1] = NV90B5_SET_REMAP_COMPONENTS_DST_Y_SRC_Y;
copy.remap.dst[2] = NV90B5_SET_REMAP_COMPONENTS_DST_Z_SRC_Z;
copy.remap.dst[3] = NV90B5_SET_REMAP_COMPONENTS_DST_W_NO_WRITE;
} else if (src_aspects == VK_IMAGE_ASPECT_STENCIL_BIT) {
copy.remap.comp_size = 1;
copy.remap.dst[0] = NV90B5_SET_REMAP_COMPONENTS_DST_X_NO_WRITE;
copy.remap.dst[1] = NV90B5_SET_REMAP_COMPONENTS_DST_Y_NO_WRITE;
copy.remap.dst[2] = NV90B5_SET_REMAP_COMPONENTS_DST_Z_NO_WRITE;
copy.remap.dst[3] = NV90B5_SET_REMAP_COMPONENTS_DST_W_SRC_W;
} else {
/* If we're copying both, there's nothing special to do */
assert(src_aspects == (VK_IMAGE_ASPECT_DEPTH_BIT |
VK_IMAGE_ASPECT_STENCIL_BIT));
}
break;
default:
copy.remap = nouveau_copy_remap_format(src->vk.format);
break;
}
nouveau_copy_rect(cmd, &copy);
}
}
VKAPI_ATTR void VKAPI_CALL
nvk_CmdFillBuffer(VkCommandBuffer commandBuffer,
VkBuffer dstBuffer,
VkDeviceSize dstOffset,
VkDeviceSize size,
uint32_t data)
{
VK_FROM_HANDLE(nvk_cmd_buffer, cmd, commandBuffer);
VK_FROM_HANDLE(nvk_buffer, dst_buffer, dstBuffer);
uint64_t dst_addr = nvk_buffer_address(dst_buffer, dstOffset);
size = vk_buffer_range(&dst_buffer->vk, dstOffset, size);
uint32_t max_dim = 1 << 15;
struct nv_push *p = nvk_cmd_buffer_push(cmd, 7);
P_IMMD(p, NV90B5, SET_REMAP_CONST_A, data);
P_IMMD(p, NV90B5, SET_REMAP_COMPONENTS, {
.dst_x = DST_X_CONST_A,
.dst_y = DST_Y_CONST_A,
.dst_z = DST_Z_CONST_A,
.dst_w = DST_W_CONST_A,
.component_size = COMPONENT_SIZE_FOUR,
.num_src_components = NUM_SRC_COMPONENTS_ONE,
.num_dst_components = NUM_DST_COMPONENTS_ONE,
});
P_MTHD(p, NV90B5, PITCH_IN);
P_NV90B5_PITCH_IN(p, max_dim * 4);
P_NV90B5_PITCH_OUT(p, max_dim * 4);
while (size >= 4) {
struct nv_push *p = nvk_cmd_buffer_push(cmd, 8);
P_MTHD(p, NV90B5, OFFSET_OUT_UPPER);
P_NV90B5_OFFSET_OUT_UPPER(p, dst_addr >> 32);
P_NV90B5_OFFSET_OUT_LOWER(p, dst_addr & 0xffffffff);
uint64_t width, height;
if (size >= (uint64_t)max_dim * (uint64_t)max_dim * 4) {
width = height = max_dim;
} else if (size >= max_dim * 4) {
width = max_dim;
height = size / (max_dim * 4);
} else {
width = size / 4;
height = 1;
}
uint64_t dma_size = (uint64_t)width * (uint64_t)height * 4;
assert(dma_size <= size);
P_MTHD(p, NV90B5, LINE_LENGTH_IN);
P_NV90B5_LINE_LENGTH_IN(p, width);
P_NV90B5_LINE_COUNT(p, height);
P_IMMD(p, NV90B5, LAUNCH_DMA, {
.data_transfer_type = DATA_TRANSFER_TYPE_NON_PIPELINED,
.multi_line_enable = height > 1,
.flush_enable = FLUSH_ENABLE_TRUE,
.src_memory_layout = SRC_MEMORY_LAYOUT_PITCH,
.dst_memory_layout = DST_MEMORY_LAYOUT_PITCH,
.remap_enable = REMAP_ENABLE_TRUE,
});
dst_addr += dma_size;
size -= dma_size;
}
}
VKAPI_ATTR void VKAPI_CALL
nvk_CmdUpdateBuffer(VkCommandBuffer commandBuffer,
VkBuffer dstBuffer,
VkDeviceSize dstOffset,
VkDeviceSize dataSize,
const void *pData)
{
VK_FROM_HANDLE(nvk_cmd_buffer, cmd, commandBuffer);
VK_FROM_HANDLE(nvk_buffer, dst, dstBuffer);
uint64_t dst_addr = nvk_buffer_address(dst, dstOffset);
uint64_t data_addr;
nvk_cmd_buffer_upload_data(cmd, pData, dataSize, 64, &data_addr);
struct nv_push *p = nvk_cmd_buffer_push(cmd, 10);
P_MTHD(p, NV90B5, OFFSET_IN_UPPER);
P_NV90B5_OFFSET_IN_UPPER(p, data_addr >> 32);
P_NV90B5_OFFSET_IN_LOWER(p, data_addr & 0xffffffff);
P_NV90B5_OFFSET_OUT_UPPER(p, dst_addr >> 32);
P_NV90B5_OFFSET_OUT_LOWER(p, dst_addr & 0xffffffff);
P_MTHD(p, NV90B5, LINE_LENGTH_IN);
P_NV90B5_LINE_LENGTH_IN(p, dataSize);
P_NV90B5_LINE_COUNT(p, 1);
P_IMMD(p, NV90B5, LAUNCH_DMA, {
.data_transfer_type = DATA_TRANSFER_TYPE_NON_PIPELINED,
.multi_line_enable = MULTI_LINE_ENABLE_TRUE,
.flush_enable = FLUSH_ENABLE_TRUE,
.src_memory_layout = SRC_MEMORY_LAYOUT_PITCH,
.dst_memory_layout = DST_MEMORY_LAYOUT_PITCH,
});
}
Reference in New Issue View Git Blame Copy Permalink