alex/vello
1
0
Fork 0
mirror of https://github.com/italicsjenga/vello.git synced 2024-10-17 23:11:30 +11:00
Code Issues Packages Projects Releases Wiki Activity
vello/piet-gpu/shader/gen/draw_leaf.hlsl

239 lines
8.1 KiB
HLSL
Raw Normal View History

Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
struct DrawMonoid
{
uint path_ix;
uint clip_ix;
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
uint scene_offset;
uint info_offset;
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
};
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
struct Alloc
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
{
uint offset;
};
struct Config
{
uint n_elements;
uint n_pathseg;
uint width_in_tiles;
uint height_in_tiles;
Alloc tile_alloc;
Alloc bin_alloc;
Alloc ptcl_alloc;
Alloc pathseg_alloc;
Alloc anno_alloc;
Alloc trans_alloc;
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
Alloc path_bbox_alloc;
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
Alloc drawmonoid_alloc;
New clip implementation This PR reworks the clip implementation. The highlight is that clip bounding box accounting is now done on GPU rather than CPU. The clip mask is also rasterized on EndClip rather than BeginClip, which decreases memory traffic needed for the clip stack. This is a pretty good working state, but not all cleanup has been applied. An important next step is to remove the CPU clip accounting (it is computed and encoded, but that result is not used). Another step is to remove the Annotated structure entirely. Fixes #88. Also relevant to #119
2022-02-18 11:25:41 +11:00
Alloc clip_alloc;
Alloc clip_bic_alloc;
Alloc clip_stack_alloc;
Alloc clip_bbox_alloc;
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
Alloc draw_bbox_alloc;
Alloc drawinfo_alloc;
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
uint n_trans;
Beginnings of new element pipeline This successfully renders the tiger; fills and strokes are supported. Other parts of the imaging model, not yet. Progress toward #119
2021-12-03 10:07:33 +11:00
uint n_path;
New clip implementation This PR reworks the clip implementation. The highlight is that clip bounding box accounting is now done on GPU rather than CPU. The clip mask is also rasterized on EndClip rather than BeginClip, which decreases memory traffic needed for the clip stack. This is a pretty good working state, but not all cleanup has been applied. An important next step is to remove the CPU clip accounting (it is computed and encoded, but that result is not used). Another step is to remove the Annotated structure entirely. Fixes #88. Also relevant to #119
2022-02-18 11:25:41 +11:00
uint n_clip;
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
uint trans_offset;
uint linewidth_offset;
Beginnings of new element pipeline This successfully renders the tiger; fills and strokes are supported. Other parts of the imaging model, not yet. Progress toward #119
2021-12-03 10:07:33 +11:00
uint pathtag_offset;
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
uint pathseg_offset;
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
uint drawtag_offset;
uint drawdata_offset;
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
};
Adjust workgroup sizes Make max workgroup size 256 and respect LG_WG_FACTOR. Because the monoid scans only support a height of 2, this will reduce the maximum scene complexity we can render. But it also increases compatibility. Supporting larger scans is a TODO.
2021-12-09 05:42:35 +11:00
static const uint3 gl_WorkGroupSize = uint3(256u, 1u, 1u);
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
static const DrawMonoid _23 = { 0u, 0u, 0u, 0u };
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
ByteAddressBuffer _92 : register(t1, space0);
ByteAddressBuffer _102 : register(t2, space0);
ByteAddressBuffer _202 : register(t3, space0);
RWByteAddressBuffer _284 : register(u0, space0);
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
static uint3 gl_WorkGroupID;
static uint3 gl_LocalInvocationID;
static uint3 gl_GlobalInvocationID;
struct SPIRV_Cross_Input
{
uint3 gl_WorkGroupID : SV_GroupID;
uint3 gl_LocalInvocationID : SV_GroupThreadID;
uint3 gl_GlobalInvocationID : SV_DispatchThreadID;
};
Adjust workgroup sizes Make max workgroup size 256 and respect LG_WG_FACTOR. Because the monoid scans only support a height of 2, this will reduce the maximum scene complexity we can render. But it also increases compatibility. Supporting larger scans is a TODO.
2021-12-09 05:42:35 +11:00
groupshared DrawMonoid sh_scratch[256];
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
DrawMonoid map_tag(uint tag_word)
{
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
uint has_path = uint(tag_word != 0u);
DrawMonoid _75 = { has_path, tag_word & 1u, tag_word & 28u, (tag_word >> uint(4)) & 28u };
return _75;
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
}
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
DrawMonoid combine_draw_monoid(DrawMonoid a, DrawMonoid b)
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
{
DrawMonoid c;
c.path_ix = a.path_ix + b.path_ix;
c.clip_ix = a.clip_ix + b.clip_ix;
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
c.scene_offset = a.scene_offset + b.scene_offset;
c.info_offset = a.info_offset + b.info_offset;
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
return c;
}
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
DrawMonoid draw_monoid_identity()
Beginnings of new element pipeline This successfully renders the tiger; fills and strokes are supported. Other parts of the imaging model, not yet. Progress toward #119
2021-12-03 10:07:33 +11:00
{
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
return _23;
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
}
void comp_main()
{
uint ix = gl_GlobalInvocationID.x * 8u;
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
uint drawtag_base = _92.Load(100) >> uint(2);
uint tag_word = _102.Load((drawtag_base + ix) * 4 + 0);
uint param = tag_word;
DrawMonoid agg = map_tag(param);
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
DrawMonoid local[8];
local[0] = agg;
for (uint i = 1u; i < 8u; i++)
{
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
tag_word = _102.Load(((drawtag_base + ix) + i) * 4 + 0);
uint param_1 = tag_word;
DrawMonoid param_2 = agg;
DrawMonoid param_3 = map_tag(param_1);
agg = combine_draw_monoid(param_2, param_3);
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
local[i] = agg;
}
sh_scratch[gl_LocalInvocationID.x] = agg;
Adjust workgroup sizes Make max workgroup size 256 and respect LG_WG_FACTOR. Because the monoid scans only support a height of 2, this will reduce the maximum scene complexity we can render. But it also increases compatibility. Supporting larger scans is a TODO.
2021-12-09 05:42:35 +11:00
for (uint i_1 = 0u; i_1 < 8u; i_1++)
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
{
GroupMemoryBarrierWithGroupSync();
if (gl_LocalInvocationID.x >= (1u << i_1))
{
DrawMonoid other = sh_scratch[gl_LocalInvocationID.x - (1u << i_1)];
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
DrawMonoid param_4 = other;
DrawMonoid param_5 = agg;
agg = combine_draw_monoid(param_4, param_5);
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
}
GroupMemoryBarrierWithGroupSync();
sh_scratch[gl_LocalInvocationID.x] = agg;
}
GroupMemoryBarrierWithGroupSync();
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
DrawMonoid row = draw_monoid_identity();
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
if (gl_WorkGroupID.x > 0u)
{
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
DrawMonoid _208;
_208.path_ix = _202.Load((gl_WorkGroupID.x - 1u) * 16 + 0);
_208.clip_ix = _202.Load((gl_WorkGroupID.x - 1u) * 16 + 4);
_208.scene_offset = _202.Load((gl_WorkGroupID.x - 1u) * 16 + 8);
_208.info_offset = _202.Load((gl_WorkGroupID.x - 1u) * 16 + 12);
row.path_ix = _208.path_ix;
row.clip_ix = _208.clip_ix;
row.scene_offset = _208.scene_offset;
row.info_offset = _208.info_offset;
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
}
if (gl_LocalInvocationID.x > 0u)
{
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
DrawMonoid param_6 = row;
DrawMonoid param_7 = sh_scratch[gl_LocalInvocationID.x - 1u];
row = combine_draw_monoid(param_6, param_7);
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
}
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
uint drawdata_base = _92.Load(104) >> uint(2);
uint drawinfo_base = _92.Load(68) >> uint(2);
Beginnings of new element pipeline This successfully renders the tiger; fills and strokes are supported. Other parts of the imaging model, not yet. Progress toward #119
2021-12-03 10:07:33 +11:00
uint out_ix = gl_GlobalInvocationID.x * 8u;
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
uint out_base = (_92.Load(44) >> uint(2)) + (out_ix * 4u);
uint clip_out_base = _92.Load(48) >> uint(2);
Beginnings of new element pipeline This successfully renders the tiger; fills and strokes are supported. Other parts of the imaging model, not yet. Progress toward #119
2021-12-03 10:07:33 +11:00
float4 mat;
float2 translate;
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
for (uint i_2 = 0u; i_2 < 8u; i_2++)
{
New clip implementation This PR reworks the clip implementation. The highlight is that clip bounding box accounting is now done on GPU rather than CPU. The clip mask is also rasterized on EndClip rather than BeginClip, which decreases memory traffic needed for the clip stack. This is a pretty good working state, but not all cleanup has been applied. An important next step is to remove the CPU clip accounting (it is computed and encoded, but that result is not used). Another step is to remove the Annotated structure entirely. Fixes #88. Also relevant to #119
2022-02-18 11:25:41 +11:00
DrawMonoid m = row;
if (i_2 > 0u)
{
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
DrawMonoid param_8 = m;
DrawMonoid param_9 = local[i_2 - 1u];
m = combine_draw_monoid(param_8, param_9);
New clip implementation This PR reworks the clip implementation. The highlight is that clip bounding box accounting is now done on GPU rather than CPU. The clip mask is also rasterized on EndClip rather than BeginClip, which decreases memory traffic needed for the clip stack. This is a pretty good working state, but not all cleanup has been applied. An important next step is to remove the CPU clip accounting (it is computed and encoded, but that result is not used). Another step is to remove the Annotated structure entirely. Fixes #88. Also relevant to #119
2022-02-18 11:25:41 +11:00
}
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
_284.Store((out_base + (i_2 * 4u)) * 4 + 8, m.path_ix);
_284.Store(((out_base + (i_2 * 4u)) + 1u) * 4 + 8, m.clip_ix);
_284.Store(((out_base + (i_2 * 4u)) + 2u) * 4 + 8, m.scene_offset);
_284.Store(((out_base + (i_2 * 4u)) + 3u) * 4 + 8, m.info_offset);
uint dd = drawdata_base + (m.scene_offset >> uint(2));
uint di = drawinfo_base + (m.info_offset >> uint(2));
tag_word = _102.Load(((drawtag_base + ix) + i_2) * 4 + 0);
if ((((tag_word == 68u) || (tag_word == 276u)) || (tag_word == 72u)) || (tag_word == 5u))
Beginnings of new element pipeline This successfully renders the tiger; fills and strokes are supported. Other parts of the imaging model, not yet. Progress toward #119
2021-12-03 10:07:33 +11:00
{
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
uint bbox_offset = (_92.Load(40) >> uint(2)) + (6u * m.path_ix);
float bbox_l = float(_284.Load(bbox_offset * 4 + 8)) - 32768.0f;
float bbox_t = float(_284.Load((bbox_offset + 1u) * 4 + 8)) - 32768.0f;
float bbox_r = float(_284.Load((bbox_offset + 2u) * 4 + 8)) - 32768.0f;
float bbox_b = float(_284.Load((bbox_offset + 3u) * 4 + 8)) - 32768.0f;
Beginnings of new element pipeline This successfully renders the tiger; fills and strokes are supported. Other parts of the imaging model, not yet. Progress toward #119
2021-12-03 10:07:33 +11:00
float4 bbox = float4(bbox_l, bbox_t, bbox_r, bbox_b);
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
float linewidth = asfloat(_284.Load((bbox_offset + 4u) * 4 + 8));
Beginnings of new element pipeline This successfully renders the tiger; fills and strokes are supported. Other parts of the imaging model, not yet. Progress toward #119
2021-12-03 10:07:33 +11:00
uint fill_mode = uint(linewidth >= 0.0f);
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
if ((linewidth >= 0.0f) || (tag_word == 276u))
Beginnings of new element pipeline This successfully renders the tiger; fills and strokes are supported. Other parts of the imaging model, not yet. Progress toward #119
2021-12-03 10:07:33 +11:00
{
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
uint trans_ix = _284.Load((bbox_offset + 5u) * 4 + 8);
uint t = (_92.Load(36) >> uint(2)) + (6u * trans_ix);
mat = asfloat(uint4(_284.Load(t * 4 + 8), _284.Load((t + 1u) * 4 + 8), _284.Load((t + 2u) * 4 + 8), _284.Load((t + 3u) * 4 + 8)));
if (tag_word == 276u)
Beginnings of new element pipeline This successfully renders the tiger; fills and strokes are supported. Other parts of the imaging model, not yet. Progress toward #119
2021-12-03 10:07:33 +11:00
{
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
translate = asfloat(uint2(_284.Load((t + 4u) * 4 + 8), _284.Load((t + 5u) * 4 + 8)));
Beginnings of new element pipeline This successfully renders the tiger; fills and strokes are supported. Other parts of the imaging model, not yet. Progress toward #119
2021-12-03 10:07:33 +11:00
}
}
if (linewidth >= 0.0f)
{
linewidth *= sqrt(abs((mat.x * mat.w) - (mat.y * mat.z)));
}
switch (tag_word)
{
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
case 68u:
case 72u:
Beginnings of new element pipeline This successfully renders the tiger; fills and strokes are supported. Other parts of the imaging model, not yet. Progress toward #119
2021-12-03 10:07:33 +11:00
{
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
_284.Store(di * 4 + 8, asuint(linewidth));
Beginnings of new element pipeline This successfully renders the tiger; fills and strokes are supported. Other parts of the imaging model, not yet. Progress toward #119
2021-12-03 10:07:33 +11:00
break;
}
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
case 276u:
Beginnings of new element pipeline This successfully renders the tiger; fills and strokes are supported. Other parts of the imaging model, not yet. Progress toward #119
2021-12-03 10:07:33 +11:00
{
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
_284.Store(di * 4 + 8, asuint(linewidth));
uint index = _102.Load(dd * 4 + 0);
float2 p0 = asfloat(uint2(_102.Load((dd + 1u) * 4 + 0), _102.Load((dd + 2u) * 4 + 0)));
float2 p1 = asfloat(uint2(_102.Load((dd + 3u) * 4 + 0), _102.Load((dd + 4u) * 4 + 0)));
p0 = ((mat.xy * p0.x) + (mat.zw * p0.y)) + translate;
p1 = ((mat.xy * p1.x) + (mat.zw * p1.y)) + translate;
Beginnings of new element pipeline This successfully renders the tiger; fills and strokes are supported. Other parts of the imaging model, not yet. Progress toward #119
2021-12-03 10:07:33 +11:00
float2 dxy = p1 - p0;
float scale = 1.0f / ((dxy.x * dxy.x) + (dxy.y * dxy.y));
float line_x = dxy.x * scale;
float line_y = dxy.y * scale;
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
float line_c = -((p0.x * line_x) + (p0.y * line_y));
_284.Store((di + 1u) * 4 + 8, asuint(line_x));
_284.Store((di + 2u) * 4 + 8, asuint(line_y));
_284.Store((di + 3u) * 4 + 8, asuint(line_c));
Beginnings of new element pipeline This successfully renders the tiger; fills and strokes are supported. Other parts of the imaging model, not yet. Progress toward #119
2021-12-03 10:07:33 +11:00
break;
}
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
case 5u:
New clip implementation This PR reworks the clip implementation. The highlight is that clip bounding box accounting is now done on GPU rather than CPU. The clip mask is also rasterized on EndClip rather than BeginClip, which decreases memory traffic needed for the clip stack. This is a pretty good working state, but not all cleanup has been applied. An important next step is to remove the CPU clip accounting (it is computed and encoded, but that result is not used). Another step is to remove the Annotated structure entirely. Fixes #88. Also relevant to #119
2022-02-18 11:25:41 +11:00
{
break;
}
Beginnings of new element pipeline This successfully renders the tiger; fills and strokes are supported. Other parts of the imaging model, not yet. Progress toward #119
2021-12-03 10:07:33 +11:00
}
}
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
if ((tag_word == 5u) || (tag_word == 37u))
New clip implementation This PR reworks the clip implementation. The highlight is that clip bounding box accounting is now done on GPU rather than CPU. The clip mask is also rasterized on EndClip rather than BeginClip, which decreases memory traffic needed for the clip stack. This is a pretty good working state, but not all cleanup has been applied. An important next step is to remove the CPU clip accounting (it is computed and encoded, but that result is not used). Another step is to remove the Annotated structure entirely. Fixes #88. Also relevant to #119
2022-02-18 11:25:41 +11:00
{
uint path_ix = ~(out_ix + i_2);
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
if (tag_word == 5u)
Beginnings of new element pipeline This successfully renders the tiger; fills and strokes are supported. Other parts of the imaging model, not yet. Progress toward #119
2021-12-03 10:07:33 +11:00
{
New clip implementation This PR reworks the clip implementation. The highlight is that clip bounding box accounting is now done on GPU rather than CPU. The clip mask is also rasterized on EndClip rather than BeginClip, which decreases memory traffic needed for the clip stack. This is a pretty good working state, but not all cleanup has been applied. An important next step is to remove the CPU clip accounting (it is computed and encoded, but that result is not used). Another step is to remove the Annotated structure entirely. Fixes #88. Also relevant to #119
2022-02-18 11:25:41 +11:00
path_ix = m.path_ix;
Beginnings of new element pipeline This successfully renders the tiger; fills and strokes are supported. Other parts of the imaging model, not yet. Progress toward #119
2021-12-03 10:07:33 +11:00
}
Variable size encoding of draw objects This patch switches to a variable size encoding of draw objects. In addition to the CPU-side scene encoding, it changes the representation of intermediate per draw object state from the `Annotated` struct to a variable "info" encoding. In addition, the bounding boxes are moved to a separate array (for a more "structure of "arrays" approach). Data that's unchanged from the scene encoding is not copied. Rather, downstream stages can access the data from the scene buffer (reducing allocation and copying). Prefix sums, computed in `DrawMonoid` track the offset of both scene and intermediate data. The tags for the CPU-side encoding have been split into their own stream (again a change from AoS to SoA style). This is not necessarily the final form. There's some stuff (including at least one piet-gpu-derive type) that can be deleted. In addition, the linewidth field should probably move from the info to path-specific. Also, the 1:1 correspondence between draw object and path has not yet been broken. Closes #152
2022-03-03 09:44:03 +11:00
_284.Store((clip_out_base + m.clip_ix) * 4 + 8, path_ix);
Beginnings of new element pipeline This successfully renders the tiger; fills and strokes are supported. Other parts of the imaging model, not yet. Progress toward #119
2021-12-03 10:07:33 +11:00
}
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
}
}
Adjust workgroup sizes Make max workgroup size 256 and respect LG_WG_FACTOR. Because the monoid scans only support a height of 2, this will reduce the maximum scene complexity we can render. But it also increases compatibility. Supporting larger scans is a TODO.
2021-12-09 05:42:35 +11:00
[numthreads(256, 1, 1)]
Add draw object stage This is one of the stages in the new element pipeline. It's a simple one, just a prefix sum of a couple counts, and some of it will probably get merged with a downstream stage, but we'll do it separately for now for convenience. This patch also contains an update to Vulkan tools 1.2.198, which accounts for the large diff of translated shaders.
2021-12-03 03:41:41 +11:00
void main(SPIRV_Cross_Input stage_input)
{
gl_WorkGroupID = stage_input.gl_WorkGroupID;
gl_LocalInvocationID = stage_input.gl_LocalInvocationID;
gl_GlobalInvocationID = stage_input.gl_GlobalInvocationID;
comp_main();
}
Reference in a new issue Copy permalink