device/program.h - GCC Code Coverage Report

device/program.h

Line	Branch	Exec	Source
1			// SPDX-License-Identifier: BSD-3-Clause
2			#pragma once
3
4			#include <chrono>
5			#include <cstdint>
6			#include <map>
7			#include <memory>
8			#include <optional>
9			#include <string>
10			#include <string_view>
11			#include <vector>
12
13			#include "chrono_batch_render/device/adapter.h"
14			#include "chrono_batch_render/device/resource.h"
15			#include "chrono_batch_render/export.h"
16
17			namespace chrono::batch_render {
18			class DiskCache;
19			struct KernelLibrary;
20			struct DeviceGlobals;
21			}
22
23			namespace chrono::batch_render::device {
24
25			struct KernelCode;
26			class Program;
27
28			struct Size3 {
29			uint32_t x = 1;
30			uint32_t y = 1;
31			uint32_t z = 1;
32
33			bool operator==(const Size3&) const = default;
34			};
35
36			inline constexpr uint32_t kMaxGroupsPerAxis = 65535;
37
38		171	constexpr Size3 FoldGroups(uint32_t count) {
39		171	const uint64_t plane = uint64_t(kMaxGroupsPerAxis) * kMaxGroupsPerAxis;
40		171	const uint32_t z = uint32_t((count + plane - 1) / plane);
41		171	const uint64_t perLayer = (uint64_t(count) + z - 1) / z;
42		171	const uint32_t y = uint32_t((perLayer + kMaxGroupsPerAxis - 1) / kMaxGroupsPerAxis);
43		171	const uint32_t x = uint32_t((perLayer + y - 1) / y);
44		171	return {x, y, z};
45			}
46
47		67378	constexpr uint32_t GroupCount(uint64_t threadCount, uint32_t threadsPerGroup) {
48	25/25 ✓ Branch 0 taken 140 times. ✓ Branch 1 taken 36101 times. ✓ Branch 1 taken 15433 times. ✓ Branch 2 taken 947 times. ✓ Branch 3 taken 30 times. ✓ Branch 4 taken 383 times. ✓ Branch 4 taken 40 times. ✓ Branch 4 taken 2994 times. ✓ Branch 5 taken 947 times. ✓ Branch 6 taken 36 times. ✓ Branch 7 taken 1205 times. ✓ Branch 7 taken 43 times. ✓ Branch 7 taken 101 times. ✓ Branch 7 taken 757 times. ✓ Branch 7 taken 1217 times. ✓ Branch 10 taken 7 times. ✓ Branch 10 taken 784 times. ✓ Branch 10 taken 101 times. ✓ Branch 10 taken 811 times. ✓ Branch 13 taken 273 times. ✓ Branch 13 taken 76 times. ✓ Branch 15 taken 8 times. ✓ Branch 16 taken 76 times. ✓ Branch 18 taken 8 times. ✓ Branch 21 taken 8 times.	67378	return uint32_t((threadCount + threadsPerGroup - 1) / threadsPerGroup);
49			}
50
51			CHRONO_BATCH_RENDER_API void RegisterFile(std::string path, std::string source);
52
53			struct ProgramDesc {
54			std::vector<std::string> modules;
55			const KernelLibrary* library = nullptr;
56			std::map<std::string, std::string, std::less<>> sources;
57			std::vector<std::string> entries;
58			bool loadEntries = true;
59			std::string label = "program";
60			Traversal traversal = Traversal::Auto;
61			};
62
63		✗	struct KernelStat {
64			std::string name;
65			std::string value;
66			};
67
68		1117	struct KernelStats {
69			double compileMs = 0;
70			bool cached = false;
71			std::optional<uint32_t> registers;
72			std::optional<uint32_t> spillBytes;
73			std::optional<uint32_t> sharedBytes;
74			std::vector<KernelStat> driver;
75			};
76
77			class CHRONO_BATCH_RENDER_API Kernel {
78			public:
79			virtual ~Kernel();
80			Kernel(const Kernel&) = delete;
81			Kernel& operator=(const Kernel&) = delete;
82
83		3354	const std::string& Name() const {
84	5/16 <unknown function 63>: ✓ Branch 0 taken 1102 times. ✓ Branch 1 taken 15 times. ✗ Branch 1 not taken. ✓ Branch 4 taken 1117 times. ✓ Branch 8 taken 1119 times. <unknown function 83>: ✗ Branch 0 not taken. ✓ Branch 1 taken 1 time. ✗ Branch 1 not taken. ✗ Branch 4 not taken. chrono::batch_render::device::FoldGroups(unsigned int): ✗ Branch 1 not taken. ✗ Branch 4 not taken. ✗ Branch 7 not taken. ✗ Branch 10 not taken. ✗ Branch 13 not taken. ✗ Branch 16 not taken. ✗ Branch 19 not taken.	4471	return m_name;
85			}
86
87		233700	Program& GetProgram() const {
88	1/3 ✗ Branch 0 not taken. ✓ Branch 1 taken 1 time. ✗ Branch 1 not taken.	233700	return *m_program;
89			}
90
91			bool Ready() const;
92
93			void Wait() const;
94
95			Size3 GroupSize() const;
96
97			uint32_t ParamBytes() const;
98
99			bool Foldable() const;
100
101			const KernelStats& Stats() const;
102
103			protected:
104			Kernel();
105
106			virtual KernelStats Load(const KernelCode& code, const DiskCache& cache,
107			const DeviceGlobals& globals) = 0;
108
109			private:
110			friend class Program;
111			struct State;
112
113			std::string m_name;
114			Program* m_program = nullptr;
115			std::unique_ptr<State> m_state;
116			};
117
118			CHRONO_BATCH_RENDER_API void WaitForBuilds();
119
120			class CHRONO_BATCH_RENDER_API Program : public Resource {
121			public:
122			~Program() override;
123
124			Kernel& Entry(std::string_view name);
125
126			private:
127			friend class Device;
128			struct State;
129
130			Program(Device& device, const ProgramDesc& desc,
131			std::shared_ptr<const std::map<std::string, std::string, std::less<>>> sources);
132
133			Kernel& Request(std::string_view name, bool wanted);
134
135			void QueueLoad(Kernel& kernel);
136
137			void Load(Kernel& kernel, const KernelCode& code, const DiskCache& cache,
138			std::chrono::steady_clock::time_point start);
139
140			void Finish(Kernel& kernel, const std::string& label);
141
142			std::unique_ptr<State> m_state;
143			};
144
145			}
146