1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
|
#pragma once
#include "bionic_fg/config.hpp"
#include "vulkan/vk_types.hpp"
#include "shaders_registry.hpp"
#include <vulkan/vulkan.h>
#ifdef __ANDROID__
#include <android/hardware_buffer.h>
#endif
#include <cstdint>
#include <memory>
#include <vector>
namespace bionic_fg {
// ─── UBO layouts (matching embedded shader expectations) ─────────────────
struct SynthUBO { float flowScale; float alpha; float epsilon; };
struct FlowUBO { float flowScale; float pad0; float pad1; float pad2; };
struct PyramidUBO{ uint32_t scale; uint32_t aspect; uint32_t pad0; uint32_t pad1; };
// ─── Pass ────────────────────────────────────────────────────────────────────
class Pass {
public:
Pass() = default;
Pass(const vk::Device& dev,
VkDescriptorPool descPool,
int embeddedShaderIdx,
const std::vector<vk::DescriptorBinding>& bindings);
void destroy(const vk::Device& dev);
// Static bindings write BOTH parity sets; the per-frame input bindings use
// the parity variants so the set an in-flight graph is reading is never
// touched (the async run pipeline keeps up to two frames on the GPU).
void bindUBO (const vk::Device& dev, uint32_t binding, const vk::Buffer& buf);
void bindSampled(const vk::Device& dev, uint32_t binding, const vk::Image& img,
const vk::Sampler& sampler);
void bindSampledGeneral(const vk::Device& dev, uint32_t binding, const vk::Image& img,
const vk::Sampler& sampler);
void bindStorage(const vk::Device& dev, uint32_t binding, const vk::Image& img);
void bindSampledAt(const vk::Device& dev, uint32_t parity, uint32_t binding,
const vk::Image& img, const vk::Sampler& sampler);
void bindSampledGeneralAt(const vk::Device& dev, uint32_t parity, uint32_t binding,
const vk::Image& img, const vk::Sampler& sampler);
void dispatch(VkCommandBuffer cmd, uint32_t gx, uint32_t gy, uint32_t parity = 0) const;
bool valid() const { return pipeline_.valid(); }
private:
vk::DescriptorSetLayout layout_;
vk::DescriptorSet descSets_[2];
vk::ComputePipeline pipeline_;
};
// ─── FramegenContext ─────────────────────────────────────────────────────────
class FramegenContext {
public:
FramegenContext() = default;
#ifdef __ANDROID__
// `device` is the application's own VkDevice (wrapped, not owned). Running
// the interpolation on the SAME device the swapchain/AHBs live on avoids the
// cross-device fence deadlock that hangs a standalone second device on
// wrapper-ICD stacks (Winlator/Turnip). See BIONIC_FG_INTEGRATION_REPORT.md.
// Null entries in prevAhb/currAhb/outputAhbs allocate plain device-local
// images instead of importing AHBs.
static std::unique_ptr<FramegenContext> create(
const vk::Device& device,
AHardwareBuffer* prevAhb,
AHardwareBuffer* currAhb,
const std::vector<AHardwareBuffer*>& outputAhbs,
VkExtent2D extent,
VkFormat format,
const Config& cfg);
// Single-device layer mode: the context owns device-local input/output
// images (no AHB round-trip, no external queue-family transfers); the
// layer blits the swapchain into currInput() and out of outputAt(k).
static std::unique_ptr<FramegenContext> create(
const vk::Device& device,
uint32_t provisionedOutputs,
VkExtent2D extent,
VkFormat format,
const Config& cfg);
void present(AHardwareBuffer* newPrevAhb, AHardwareBuffer* newCurrAhb);
void run() { present(nullptr, nullptr); }
// Rotate frame inputs after a presented frame: the previous "current"
// becomes "previous" and its image is reused as the next blit target.
void swapFrameInputs();
vk::Image& prevInput() { return prevFrame_; }
vk::Image& currInput() { return currFrame_; }
vk::Image& outputAt(size_t k) { return outputImages_[k]; }
size_t outputCount() const { return outputImages_.size(); }
#endif
void updateConfig(const Config& cfg);
void waitIdle();
void destroy();
bool valid() const { return device_.valid(); }
std::string describe() const;
private:
// ── Vulkan core ──────────────────────────────────────────────────────────
Config cfg_;
VkExtent2D extent_ = {};
VkFormat format_ = VK_FORMAT_R8G8B8A8_UNORM;
vk::Device device_;
vk::CommandPool cmdPool_;
vk::Sampler linearSampler_;
vk::Sampler nearestSampler_;
VkDescriptorPool descPool_ = VK_NULL_HANDLE;
// ── AHB-backed images (external, not owned) ───────────────────────────
vk::Image prevFrame_;
vk::Image currFrame_;
std::vector<vk::Image> outputImages_;
// Shared-graph resources (descriptor labels dNNN.bXX from the traced
// model-1 edge map; both models bind into the same graph).
std::vector<vk::Image> model1Resources_;
std::vector<Pass> model1GraphPasses_; // runtime-confirmed dispatch graph
std::vector<VkExtent2D> model1GraphDispatch_; // per-pass group counts
// Per-pass image read/write sets, used to emit barriers only at true
// dependencies instead of after every dispatch (~100 pipeline drains).
std::vector<std::vector<VkImage>> model1Reads_;
std::vector<std::vector<VkImage>> model1Writes_;
size_t model1FinalPassStart_ = 0;
// ── UBO buffers ──────────────────────────────────────────────────────────
vk::Buffer uboPyramid_;
vk::Buffer uboFlow_;
std::vector<vk::Buffer> uboSynth_;
// Flow-magnitude probe (model 0): spread samples of the three flow fields
// shader_04 consumes, copied to a host-visible buffer each run and logged
// every 120 frames so "is the flow alive" is observable instead of
// inferred from artifacts.
vk::Buffer dbgFlowBuf_;
uint32_t dbgLogCounter_ = 0;
size_t dbgFlowIdx_[3] = {SIZE_MAX, SIZE_MAX, SIZE_MAX};
// ── Per-frame command buffer + fence ─────────────────────────────────────
struct Frame {
vk::CommandBuffer cmd;
vk::Fence fence;
};
Frame frames_[2];
uint32_t frameIdx_ = 0;
#ifdef __ANDROID__
AHardwareBuffer* prevAhbPtr_ = nullptr;
AHardwareBuffer* currAhbPtr_ = nullptr;
void rebindFrameInputs();
#endif
};
} // namespace bionic_fg
|
