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
169
170
171
172
173
|
/*
* Copyright (c) 2015-2025 The Khronos Group Inc.
* Copyright (c) 2015-2025 Valve Corporation
* Copyright (c) 2015-2025 LunarG, Inc.
* Copyright (c) 2015-2025 NVIDIA Corporation
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*/
#include "../framework/layer_validation_tests.h"
#include "../framework/pipeline_helper.h"
class PositiveShaderCooperativeVector : public VkLayerTest {};
TEST_F(PositiveShaderCooperativeVector, ConvertCooperativeVectorMatrixNV) {
TEST_DESCRIPTION("Validate using vkConvertCooperativeVectorMatrixNV.");
SetTargetApiVersion(VK_API_VERSION_1_3);
AddRequiredExtensions(VK_NV_COOPERATIVE_VECTOR_EXTENSION_NAME);
AddRequiredFeature(vkt::Feature::cooperativeVector);
RETURN_IF_SKIP(Init());
VkConvertCooperativeVectorMatrixInfoNV info = vku::InitStructHelper();
size_t dstSize = 16 * 32 * 2;
info.srcSize = 16 * 32 * 2;
info.srcData.hostAddress = nullptr;
info.pDstSize = &dstSize;
info.dstData.hostAddress = nullptr;
info.srcComponentType = VK_COMPONENT_TYPE_FLOAT16_KHR;
info.dstComponentType = VK_COMPONENT_TYPE_FLOAT16_KHR;
info.numRows = 16;
info.numColumns = 32;
info.srcLayout = VK_COOPERATIVE_VECTOR_MATRIX_LAYOUT_ROW_MAJOR_NV;
info.srcStride = 64;
info.dstLayout = VK_COOPERATIVE_VECTOR_MATRIX_LAYOUT_ROW_MAJOR_NV;
info.dstStride = 64;
vk::ConvertCooperativeVectorMatrixNV(*m_device, &info);
uint8_t src[16 * 32 * 2], dst[16 * 32 * 2];
info.srcData.hostAddress = src;
info.dstData.hostAddress = dst;
vk::ConvertCooperativeVectorMatrixNV(*m_device, &info);
}
TEST_F(PositiveShaderCooperativeVector, CmdConvertCooperativeVectorMatrixNV) {
TEST_DESCRIPTION("Validate using vkCmdConvertCooperativeVectorMatrixNV.");
SetTargetApiVersion(VK_API_VERSION_1_3);
AddRequiredExtensions(VK_NV_COOPERATIVE_VECTOR_EXTENSION_NAME);
AddRequiredFeature(vkt::Feature::cooperativeVector);
AddRequiredFeature(vkt::Feature::bufferDeviceAddress);
RETURN_IF_SKIP(Init());
VkConvertCooperativeVectorMatrixInfoNV info = vku::InitStructHelper();
size_t dstSize = 16 * 32 * 2;
info.srcSize = 16 * 32 * 2;
info.pDstSize = &dstSize;
info.srcComponentType = VK_COMPONENT_TYPE_FLOAT16_KHR;
info.dstComponentType = VK_COMPONENT_TYPE_FLOAT16_KHR;
info.numRows = 16;
info.numColumns = 32;
info.srcLayout = VK_COOPERATIVE_VECTOR_MATRIX_LAYOUT_ROW_MAJOR_NV;
info.srcStride = 64;
info.dstLayout = VK_COOPERATIVE_VECTOR_MATRIX_LAYOUT_ROW_MAJOR_NV;
info.dstStride = 64;
vkt::Buffer src_buffer(*m_device, 16 * 32 * 2, VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT, vkt::device_address);
vkt::Buffer dst_buffer(*m_device, 16 * 32 * 2, VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT, vkt::device_address);
const VkDeviceAddress src_address = src_buffer.Address();
const VkDeviceAddress dst_address = dst_buffer.Address();
info.srcData.deviceAddress = src_address;
info.dstData.deviceAddress = dst_address;
m_command_buffer.Begin();
vk::CmdConvertCooperativeVectorMatrixNV(m_command_buffer, 1, &info);
m_command_buffer.End();
}
TEST_F(PositiveShaderCooperativeVector, CooperativeVectorSPIRV) {
TEST_DESCRIPTION("Validate Cooperative Vector SPIR-V environment rules.");
SetTargetApiVersion(VK_API_VERSION_1_3);
AddRequiredExtensions(VK_NV_COOPERATIVE_VECTOR_EXTENSION_NAME);
AddRequiredFeature(vkt::Feature::cooperativeVector);
AddRequiredFeature(vkt::Feature::storageBuffer8BitAccess);
AddRequiredFeature(vkt::Feature::shaderFloat16);
AddRequiredFeature(vkt::Feature::shaderInt8);
AddRequiredFeature(vkt::Feature::vulkanMemoryModel);
RETURN_IF_SKIP(Init());
VkPhysicalDeviceCooperativeVectorPropertiesNV props = vku::InitStructHelper();
GetPhysicalDeviceProperties2(props);
if (!props.cooperativeVectorTrainingFloat16Accumulation) {
GTEST_SKIP() << "cooperativeVectorTrainingFloat16Accumulation not supported";
}
char const *vt_source = R"glsl(
#version 450
#extension GL_NV_cooperative_vector : enable
#extension GL_KHR_shader_subgroup_basic : enable
#extension GL_KHR_memory_scope_semantics : enable
#extension GL_EXT_shader_explicit_arithmetic_types : enable
layout(set=0, binding=0) buffer B { uint8_t x[]; } b;
void main() {
coopvecNV<float16_t, 16> A;
coopvecNV<float16_t, 32> R;
coopVecMatMulNV(R, A, gl_ComponentTypeFloat16NV, b.x, 0, gl_ComponentTypeFloat16NV, 32, 16, gl_CooperativeVectorMatrixLayoutInferencingOptimalNV, false, 0);
coopVecMatMulAddNV(R, A, gl_ComponentTypeFloat16NV, b.x, 0, gl_ComponentTypeFloat16NV, b.x, 0, gl_ComponentTypeFloat16NV, 32, 16, gl_CooperativeVectorMatrixLayoutInferencingOptimalNV, false, 0);
coopvecNV<int8_t, 16> A2;
coopvecNV<int32_t, 32> R2;
coopVecMatMulNV(R2, A2, gl_ComponentTypeSignedInt8NV, b.x, 0, gl_ComponentTypeSignedInt8NV, 32, 16, gl_CooperativeVectorMatrixLayoutInferencingOptimalNV, false, 0);
}
)glsl";
const std::vector<VkDescriptorSetLayoutBinding> bindings = {
{0, VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC, 1, VK_SHADER_STAGE_COMPUTE_BIT, nullptr},
};
const vkt::DescriptorSetLayout dsl(*m_device, bindings);
const vkt::PipelineLayout pl(*m_device, {&dsl});
CreateComputePipelineHelper pipe(*this);
pipe.cs_ = VkShaderObj(this, vt_source, VK_SHADER_STAGE_COMPUTE_BIT, SPV_ENV_VULKAN_1_3);
pipe.pipeline_layout_ = vkt::PipelineLayout(*m_device, {&dsl});
pipe.CreateComputePipeline();
}
TEST_F(PositiveShaderCooperativeVector, CooperativeVectorTraingingSPIRV) {
TEST_DESCRIPTION("Validate Cooperative Vector SPIR-V environment rules.");
SetTargetApiVersion(VK_API_VERSION_1_3);
AddRequiredExtensions(VK_NV_COOPERATIVE_VECTOR_EXTENSION_NAME);
AddRequiredFeature(vkt::Feature::cooperativeVector);
AddRequiredFeature(vkt::Feature::cooperativeVectorTraining);
AddRequiredFeature(vkt::Feature::storageBuffer8BitAccess);
AddRequiredFeature(vkt::Feature::shaderFloat16);
AddRequiredFeature(vkt::Feature::shaderInt8);
AddRequiredFeature(vkt::Feature::vulkanMemoryModel);
RETURN_IF_SKIP(Init());
VkPhysicalDeviceCooperativeVectorPropertiesNV props = vku::InitStructHelper();
GetPhysicalDeviceProperties2(props);
if (!props.cooperativeVectorTrainingFloat16Accumulation) {
GTEST_SKIP() << "cooperativeVectorTrainingFloat16Accumulation not supported";
}
const std::vector<VkDescriptorSetLayoutBinding> bindings = {
{0, VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC, 1, VK_SHADER_STAGE_COMPUTE_BIT, nullptr},
};
const vkt::DescriptorSetLayout dsl(*m_device, bindings);
const vkt::PipelineLayout pl(*m_device, {&dsl});
char const *vt_source = R"glsl(
#version 450
#extension GL_NV_cooperative_vector : enable
#extension GL_KHR_shader_subgroup_basic : enable
#extension GL_KHR_memory_scope_semantics : enable
#extension GL_EXT_shader_explicit_arithmetic_types : enable
layout(set=0, binding=0) buffer B { uint8_t x[]; } b;
void main() {
coopvecNV<float16_t, 16> A;
coopVecReduceSumAccumulateNV(A, b.x, 0);
coopVecOuterProductAccumulateNV(A, A, b.x, 0, 0, gl_CooperativeVectorMatrixLayoutTrainingOptimalNV, gl_ComponentTypeFloat16NV);
}
)glsl";
CreateComputePipelineHelper pipe(*this);
pipe.cs_ = VkShaderObj(this, vt_source, VK_SHADER_STAGE_COMPUTE_BIT, SPV_ENV_VULKAN_1_3);
pipe.pipeline_layout_ = vkt::PipelineLayout(*m_device, {&dsl});
pipe.CreateComputePipeline();
}
|
