Google Git
Sign in
fuchsia / third_party / vulkan-cts / 34639fb3f8b467b261624a1b2863b5e808bb7aef / . / external / vulkancts / modules / vulkan / ray_tracing / vktRayTracingDataSpillTests.cpp
blob: 07617dd4445fd457e64330c69683035ba4a4f28d [file] [log] [blame]
/*-------------------------------------------------------------------------
* Vulkan Conformance Tests
* ------------------------
*
* Copyright (c) 2020 The Khronos Group Inc.
* Copyright (c) 2020 Valve 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
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*//*!
* \file
* \brief Ray Tracing Data Spill tests
*//*--------------------------------------------------------------------*/
#include "vktRayTracingDataSpillTests.hpp"
#include "vktTestCase.hpp"
#include "vkRayTracingUtil.hpp"
#include "vkObjUtil.hpp"
#include "vkBufferWithMemory.hpp"
#include "vkImageWithMemory.hpp"
#include "vkBuilderUtil.hpp"
#include "vkCmdUtil.hpp"
#include "vkTypeUtil.hpp"
#include "vkBarrierUtil.hpp"
#include "tcuStringTemplate.hpp"
#include "tcuFloat.hpp"
#include "deUniquePtr.hpp"
#include "deSTLUtil.hpp"
#include <sstream>
#include <string>
#include <map>
#include <vector>
#include <array>
#include <utility>
using namespace vk;
namespace vkt
{
namespace RayTracing
{
namespace
{
// The type of shader call that will be used.
enum class CallType
{
TRACE_RAY = 0,
EXECUTE_CALLABLE,
REPORT_INTERSECTION,
};
// The type of data that will be checked.
enum class DataType
{
// These can be made an array or vector.
INT32 = 0,
UINT32,
INT64,
UINT64,
INT16,
UINT16,
INT8,
UINT8,
FLOAT32,
FLOAT64,
FLOAT16,
// These are standalone, so the vector type should be scalar.
STRUCT,
IMAGE,
SAMPLER,
SAMPLED_IMAGE,
PTR_IMAGE,
PTR_SAMPLER,
PTR_SAMPLED_IMAGE,
PTR_TEXEL,
OP_NULL,
OP_UNDEF,
};
// The type of vector in use.
enum class VectorType
{
SCALAR = 1,
V2 = 2,
V3 = 3,
V4 = 4,
A5 = 5,
};
struct InputStruct
{
deUint32 uintPart;
float floatPart;
};
constexpr auto kImageFormat = VK_FORMAT_R32_UINT;
const auto kImageExtent = makeExtent3D(1u, 1u, 1u);
// For samplers.
const VkImageUsageFlags kSampledImageUsage = (VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_SAMPLED_BIT);
constexpr size_t kNumImages = 4u;
constexpr size_t kNumSamplers = 4u;
constexpr size_t kNumCombined = 2u;
constexpr size_t kNumAloneImages = kNumImages - kNumCombined;
constexpr size_t kNumAloneSamplers = kNumSamplers - kNumCombined;
// For storage images.
const VkImageUsageFlags kStorageImageUsage = (VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_STORAGE_BIT);
// For the pipeline interface tests.
constexpr size_t kNumStorageValues = 6u;
constexpr deUint32 kShaderRecordSize = sizeof(tcu::UVec4);
// Get the effective vector length in memory.
size_t getEffectiveVectorLength (VectorType vectorType)
{
return ((vectorType == VectorType::V3) ? static_cast<size_t>(4) : static_cast<size_t>(vectorType));
}
// Get the corresponding element size.
VkDeviceSize getElementSize(DataType dataType, VectorType vectorType)
{
const size_t length = getEffectiveVectorLength(vectorType);
size_t dataSize = 0u;
switch (dataType)
{
case DataType::INT32: dataSize = sizeof(deInt32); break;
case DataType::UINT32: dataSize = sizeof(deUint32); break;
case DataType::INT64: dataSize = sizeof(deInt64); break;
case DataType::UINT64: dataSize = sizeof(deUint64); break;
case DataType::INT16: dataSize = sizeof(deInt16); break;
case DataType::UINT16: dataSize = sizeof(deUint16); break;
case DataType::INT8: dataSize = sizeof(deInt8); break;
case DataType::UINT8: dataSize = sizeof(deUint8); break;
case DataType::FLOAT32: dataSize = sizeof(tcu::Float32); break;
case DataType::FLOAT64: dataSize = sizeof(tcu::Float64); break;
case DataType::FLOAT16: dataSize = sizeof(tcu::Float16); break;
case DataType::STRUCT: dataSize = sizeof(InputStruct); break;
case DataType::IMAGE: // fallthrough.
case DataType::SAMPLER: // fallthrough.
case DataType::SAMPLED_IMAGE: // fallthrough.
case DataType::PTR_IMAGE: // fallthrough.
case DataType::PTR_SAMPLER: // fallthrough.
case DataType::PTR_SAMPLED_IMAGE: // fallthrough.
dataSize = sizeof(tcu::Float32); break;
case DataType::PTR_TEXEL: dataSize = sizeof(deInt32); break;
case DataType::OP_NULL: // fallthrough.
case DataType::OP_UNDEF: // fallthrough.
dataSize = sizeof(deUint32); break;
default: DE_ASSERT(false); break;
}
return static_cast<VkDeviceSize>(dataSize * length);
}
// Proper stage for generating default geometry.
VkShaderStageFlagBits getShaderStageForGeometry (CallType type_)
{
VkShaderStageFlagBits bits = VK_SHADER_STAGE_FLAG_BITS_MAX_ENUM;
switch (type_)
{
case CallType::TRACE_RAY: bits = VK_SHADER_STAGE_CLOSEST_HIT_BIT_KHR; break;
case CallType::EXECUTE_CALLABLE: bits = VK_SHADER_STAGE_CALLABLE_BIT_KHR; break;
case CallType::REPORT_INTERSECTION: bits = VK_SHADER_STAGE_INTERSECTION_BIT_KHR; break;
default: DE_ASSERT(false); break;
}
DE_ASSERT(bits != VK_SHADER_STAGE_FLAG_BITS_MAX_ENUM);
return bits;
}
VkShaderStageFlags getShaderStages (CallType type_)
{
VkShaderStageFlags flags = VK_SHADER_STAGE_RAYGEN_BIT_KHR;
switch (type_)
{
case CallType::EXECUTE_CALLABLE:
flags |= VK_SHADER_STAGE_CALLABLE_BIT_KHR;
break;
case CallType::TRACE_RAY:
flags |= VK_SHADER_STAGE_CLOSEST_HIT_BIT_KHR;
break;
case CallType::REPORT_INTERSECTION:
flags |= VK_SHADER_STAGE_INTERSECTION_BIT_KHR;
flags |= VK_SHADER_STAGE_ANY_HIT_BIT_KHR;
break;
default:
DE_ASSERT(false);
break;
}
return flags;
}
// Some test types need additional descriptors with samplers, images and combined image samplers.
bool samplersNeeded (DataType dataType)
{
bool needed = false;
switch (dataType)
{
case DataType::IMAGE:
case DataType::SAMPLER:
case DataType::SAMPLED_IMAGE:
case DataType::PTR_IMAGE:
case DataType::PTR_SAMPLER:
case DataType::PTR_SAMPLED_IMAGE:
needed = true;
break;
default:
break;
}
return needed;
}
// Some test types need an additional descriptor with a storage image.
bool storageImageNeeded (DataType dataType)
{
return (dataType == DataType::PTR_TEXEL);
}
// Returns two strings:
// .first is an optional GLSL additional type declaration (for structs, basically).
// .second is the value declaration inside the input block.
std::pair<std::string, std::string> getGLSLInputValDecl (DataType dataType, VectorType vectorType)
{
using TypePair = std::pair<DataType, VectorType>;
using TypeMap = std::map<TypePair, std::string>;
const std::string varName = "val";
const auto dataTypeIdx = static_cast<int>(dataType);
if (dataTypeIdx >= static_cast<int>(DataType::INT32) && dataTypeIdx <= static_cast<int>(DataType::FLOAT16))
{
// Note: A5 uses the same type as the scalar version. The array suffix will be added below.
const TypeMap map =
{
std::make_pair(std::make_pair(DataType::INT32, VectorType::SCALAR), "int32_t"),
std::make_pair(std::make_pair(DataType::INT32, VectorType::V2), "i32vec2"),
std::make_pair(std::make_pair(DataType::INT32, VectorType::V3), "i32vec3"),
std::make_pair(std::make_pair(DataType::INT32, VectorType::V4), "i32vec4"),
std::make_pair(std::make_pair(DataType::INT32, VectorType::A5), "int32_t"),
std::make_pair(std::make_pair(DataType::UINT32, VectorType::SCALAR), "uint32_t"),
std::make_pair(std::make_pair(DataType::UINT32, VectorType::V2), "u32vec2"),
std::make_pair(std::make_pair(DataType::UINT32, VectorType::V3), "u32vec3"),
std::make_pair(std::make_pair(DataType::UINT32, VectorType::V4), "u32vec4"),
std::make_pair(std::make_pair(DataType::UINT32, VectorType::A5), "uint32_t"),
std::make_pair(std::make_pair(DataType::INT64, VectorType::SCALAR), "int64_t"),
std::make_pair(std::make_pair(DataType::INT64, VectorType::V2), "i64vec2"),
std::make_pair(std::make_pair(DataType::INT64, VectorType::V3), "i64vec3"),
std::make_pair(std::make_pair(DataType::INT64, VectorType::V4), "i64vec4"),
std::make_pair(std::make_pair(DataType::INT64, VectorType::A5), "int64_t"),
std::make_pair(std::make_pair(DataType::UINT64, VectorType::SCALAR), "uint64_t"),
std::make_pair(std::make_pair(DataType::UINT64, VectorType::V2), "u64vec2"),
std::make_pair(std::make_pair(DataType::UINT64, VectorType::V3), "u64vec3"),
std::make_pair(std::make_pair(DataType::UINT64, VectorType::V4), "u64vec4"),
std::make_pair(std::make_pair(DataType::UINT64, VectorType::A5), "uint64_t"),
std::make_pair(std::make_pair(DataType::INT16, VectorType::SCALAR), "int16_t"),
std::make_pair(std::make_pair(DataType::INT16, VectorType::V2), "i16vec2"),
std::make_pair(std::make_pair(DataType::INT16, VectorType::V3), "i16vec3"),
std::make_pair(std::make_pair(DataType::INT16, VectorType::V4), "i16vec4"),
std::make_pair(std::make_pair(DataType::INT16, VectorType::A5), "int16_t"),
std::make_pair(std::make_pair(DataType::UINT16, VectorType::SCALAR), "uint16_t"),
std::make_pair(std::make_pair(DataType::UINT16, VectorType::V2), "u16vec2"),
std::make_pair(std::make_pair(DataType::UINT16, VectorType::V3), "u16vec3"),
std::make_pair(std::make_pair(DataType::UINT16, VectorType::V4), "u16vec4"),
std::make_pair(std::make_pair(DataType::UINT16, VectorType::A5), "uint16_t"),
std::make_pair(std::make_pair(DataType::INT8, VectorType::SCALAR), "int8_t"),
std::make_pair(std::make_pair(DataType::INT8, VectorType::V2), "i8vec2"),
std::make_pair(std::make_pair(DataType::INT8, VectorType::V3), "i8vec3"),
std::make_pair(std::make_pair(DataType::INT8, VectorType::V4), "i8vec4"),
std::make_pair(std::make_pair(DataType::INT8, VectorType::A5), "int8_t"),
std::make_pair(std::make_pair(DataType::UINT8, VectorType::SCALAR), "uint8_t"),
std::make_pair(std::make_pair(DataType::UINT8, VectorType::V2), "u8vec2"),
std::make_pair(std::make_pair(DataType::UINT8, VectorType::V3), "u8vec3"),
std::make_pair(std::make_pair(DataType::UINT8, VectorType::V4), "u8vec4"),
std::make_pair(std::make_pair(DataType::UINT8, VectorType::A5), "uint8_t"),
std::make_pair(std::make_pair(DataType::FLOAT32, VectorType::SCALAR), "float32_t"),
std::make_pair(std::make_pair(DataType::FLOAT32, VectorType::V2), "f32vec2"),
std::make_pair(std::make_pair(DataType::FLOAT32, VectorType::V3), "f32vec3"),
std::make_pair(std::make_pair(DataType::FLOAT32, VectorType::V4), "f32vec4"),
std::make_pair(std::make_pair(DataType::FLOAT32, VectorType::A5), "float32_t"),
std::make_pair(std::make_pair(DataType::FLOAT64, VectorType::SCALAR), "float64_t"),
std::make_pair(std::make_pair(DataType::FLOAT64, VectorType::V2), "f64vec2"),
std::make_pair(std::make_pair(DataType::FLOAT64, VectorType::V3), "f64vec3"),
std::make_pair(std::make_pair(DataType::FLOAT64, VectorType::V4), "f64vec4"),
std::make_pair(std::make_pair(DataType::FLOAT64, VectorType::A5), "float64_t"),
std::make_pair(std::make_pair(DataType::FLOAT16, VectorType::SCALAR), "float16_t"),
std::make_pair(std::make_pair(DataType::FLOAT16, VectorType::V2), "f16vec2"),
std::make_pair(std::make_pair(DataType::FLOAT16, VectorType::V3), "f16vec3"),
std::make_pair(std::make_pair(DataType::FLOAT16, VectorType::V4), "f16vec4"),
std::make_pair(std::make_pair(DataType::FLOAT16, VectorType::A5), "float16_t"),
};
const auto key = std::make_pair(dataType, vectorType);
const auto found = map.find(key);
DE_ASSERT(found != end(map));
const auto baseType = found->second;
const std::string decl = baseType + " " + varName + ((vectorType == VectorType::A5) ? "[5]" : "") + ";";
return std::make_pair(std::string(), decl);
}
else if (dataType == DataType::STRUCT)
{
return std::make_pair(std::string("struct InputStruct { uint val1; float val2; };\n"), std::string("InputStruct val;"));
}
else if (samplersNeeded(dataType))
{
return std::make_pair(std::string(), std::string("float val;"));
}
else if (storageImageNeeded(dataType))
{
return std::make_pair(std::string(), std::string("int val;"));
}
else if (dataType == DataType::OP_NULL || dataType == DataType::OP_UNDEF)
{
return std::make_pair(std::string(), std::string("uint val;"));
}
// Unreachable.
DE_ASSERT(false);
return std::make_pair(std::string(), std::string());
}
class DataSpillTestCase : public vkt::TestCase
{
public:
struct TestParams
{
CallType callType;
DataType dataType;
VectorType vectorType;
};
DataSpillTestCase (tcu::TestContext& testCtx, const std::string& name, const std::string& description, const TestParams& testParams);
virtual ~DataSpillTestCase (void) {}
virtual void initPrograms (vk::SourceCollections& programCollection) const;
virtual TestInstance* createInstance (Context& context) const;
virtual void checkSupport (Context& context) const;
private:
TestParams m_params;
};
class DataSpillTestInstance : public vkt::TestInstance
{
public:
using TestParams = DataSpillTestCase::TestParams;
DataSpillTestInstance (Context& context, const TestParams& testParams);
virtual ~DataSpillTestInstance (void) {}
virtual tcu::TestStatus iterate (void);
private:
TestParams m_params;
};
DataSpillTestCase::DataSpillTestCase (tcu::TestContext& testCtx, const std::string& name, const std::string& description, const TestParams& testParams)
: vkt::TestCase (testCtx, name, description)
, m_params (testParams)
{
switch (m_params.dataType)
{
case DataType::STRUCT:
case DataType::IMAGE:
case DataType::SAMPLER:
case DataType::SAMPLED_IMAGE:
case DataType::PTR_IMAGE:
case DataType::PTR_SAMPLER:
case DataType::PTR_SAMPLED_IMAGE:
case DataType::PTR_TEXEL:
case DataType::OP_NULL:
case DataType::OP_UNDEF:
DE_ASSERT(m_params.vectorType == VectorType::SCALAR);
break;
default:
break;
}
// The code assumes at most one of these is needed.
DE_ASSERT(!(samplersNeeded(m_params.dataType) && storageImageNeeded(m_params.dataType)));
}
TestInstance* DataSpillTestCase::createInstance (Context& context) const
{
return new DataSpillTestInstance(context, m_params);
}
DataSpillTestInstance::DataSpillTestInstance (Context& context, const TestParams& testParams)
: vkt::TestInstance (context)
, m_params (testParams)
{
}
// General checks for all tests.
void commonCheckSupport (Context& context)
{
context.requireDeviceFunctionality("VK_KHR_acceleration_structure");
context.requireDeviceFunctionality("VK_KHR_ray_tracing_pipeline");
const auto& rtFeatures = context.getRayTracingPipelineFeatures();
if (!rtFeatures.rayTracingPipeline)
TCU_THROW(NotSupportedError, "Ray Tracing pipelines not supported");
const auto& asFeatures = context.getAccelerationStructureFeatures();
if (!asFeatures.accelerationStructure)
TCU_FAIL("VK_KHR_acceleration_structure supported without accelerationStructure support");
}
void DataSpillTestCase::checkSupport (Context& context) const
{
// General checks first.
commonCheckSupport(context);
const auto& features = context.getDeviceFeatures();
const auto& featuresStorage16 = context.get16BitStorageFeatures();
const auto& featuresF16I8 = context.getShaderFloat16Int8Features();
const auto& featuresStorage8 = context.get8BitStorageFeatures();
if (m_params.dataType == DataType::INT64 || m_params.dataType == DataType::UINT64)
{
if (!features.shaderInt64)
TCU_THROW(NotSupportedError, "64-bit integers not supported");
}
else if (m_params.dataType == DataType::INT16 || m_params.dataType == DataType::UINT16)
{
context.requireDeviceFunctionality("VK_KHR_16bit_storage");
if (!features.shaderInt16)
TCU_THROW(NotSupportedError, "16-bit integers not supported");
if (!featuresStorage16.storageBuffer16BitAccess)
TCU_THROW(NotSupportedError, "16-bit storage buffer access not supported");
}
else if (m_params.dataType == DataType::INT8 || m_params.dataType == DataType::UINT8)
{
context.requireDeviceFunctionality("VK_KHR_shader_float16_int8");
context.requireDeviceFunctionality("VK_KHR_8bit_storage");
if (!featuresF16I8.shaderInt8)
TCU_THROW(NotSupportedError, "8-bit integers not supported");
if (!featuresStorage8.storageBuffer8BitAccess)
TCU_THROW(NotSupportedError, "8-bit storage buffer access not supported");
}
else if (m_params.dataType == DataType::FLOAT64)
{
if (!features.shaderFloat64)
TCU_THROW(NotSupportedError, "64-bit floats not supported");
}
else if (m_params.dataType == DataType::FLOAT16)
{
context.requireDeviceFunctionality("VK_KHR_shader_float16_int8");
context.requireDeviceFunctionality("VK_KHR_16bit_storage");
if (!featuresF16I8.shaderFloat16)
TCU_THROW(NotSupportedError, "16-bit floats not supported");
if (!featuresStorage16.storageBuffer16BitAccess)
TCU_THROW(NotSupportedError, "16-bit storage buffer access not supported");
}
else if (samplersNeeded(m_params.dataType))
{
context.requireDeviceFunctionality("VK_EXT_descriptor_indexing");
const auto indexingFeatures = context.getDescriptorIndexingFeatures();
if (!indexingFeatures.shaderSampledImageArrayNonUniformIndexing)
TCU_THROW(NotSupportedError, "No support for non-uniform sampled image arrays");
}
}
void DataSpillTestCase::initPrograms (vk::SourceCollections& programCollection) const
{
const vk::ShaderBuildOptions buildOptions (programCollection.usedVulkanVersion, vk::SPIRV_VERSION_1_4, 0u, true);
const vk::SpirVAsmBuildOptions spvBuildOptions (programCollection.usedVulkanVersion, vk::SPIRV_VERSION_1_4, true);
std::ostringstream spvTemplateStream;
// This SPIR-V template will be used to generate shaders for different
// stages (raygen, callable, etc). The basic mechanism uses 3 SSBOs: one
// used strictly as an input, one to write the check result, and one to
// verify the shader call has taken place. The latter two SSBOs contain just
// a single uint, but the input SSBO typically contains other type of data
// that will be filled from the test instance with predetermined values. The
// shader will expect this data to have specific values that can be combined
// some way to give an expected result (e.g. by adding the 4 components if
// it's a vec4). This result will be used in the shader call to make sure
// input values are read *before* the call. After the shader call has taken
// place, the shader will attempt to read the input buffer again and verify
// the value is still correct and matches the previous one. If the result
// matches, it will write a confirmation value in the check buffer. In the
// mean time, the callee will write a confirmation value in the callee
// buffer to verify the shader call took place.
//
// Some test variants use samplers, images or sampled images. These need
// additional bindings of different types and the interesting value is
// typically placed in the image instead of the input buffer, while the
// input buffer is used for sampling coordinates instead.
//
// Some important SPIR-V template variables:
//
// - INPUT_BUFFER_VALUE_TYPE will contain the type of input buffer data.
// - CALC_ZERO_FOR_CALLABLE is expected to contain instructions that will
// calculate a value of zero to be used in the shader call instruction.
// This value should be derived from the input data.
// - CALL_STATEMENTS will contain the shader call instructions.
// - CALC_EQUAL_STATEMENT is expected to contain instructions that will
// set %equal to true as a %bool if the before- and after- data match.
//
// - %input_val_ptr contains the pointer to the input value.
// - %input_val_before contains the value read before the call.
// - %input_val_after contains the value read after the call.
spvTemplateStream
<< " OpCapability RayTracingKHR\n"
<< "${EXTRA_CAPABILITIES}"
<< " OpExtension \"SPV_KHR_ray_tracing\"\n"
<< "${EXTRA_EXTENSIONS}"
<< " OpMemoryModel Logical GLSL450\n"
<< " OpEntryPoint ${ENTRY_POINT} %main \"main\" %topLevelAS %calleeBuffer %outputBuffer %inputBuffer${MAIN_INTERFACE_EXTRAS}\n"
<< "${INTERFACE_DECORATIONS}"
<< " OpMemberDecorate %InputBlock 0 Offset 0\n"
<< " OpDecorate %InputBlock Block\n"
<< " OpDecorate %inputBuffer DescriptorSet 0\n"
<< " OpDecorate %inputBuffer Binding 3\n"
<< " OpMemberDecorate %OutputBlock 0 Offset 0\n"
<< " OpDecorate %OutputBlock Block\n"
<< " OpDecorate %outputBuffer DescriptorSet 0\n"
<< " OpDecorate %outputBuffer Binding 2\n"
<< " OpMemberDecorate %CalleeBlock 0 Offset 0\n"
<< " OpDecorate %CalleeBlock Block\n"
<< " OpDecorate %calleeBuffer DescriptorSet 0\n"
<< " OpDecorate %calleeBuffer Binding 1\n"
<< " OpDecorate %topLevelAS DescriptorSet 0\n"
<< " OpDecorate %topLevelAS Binding 0\n"
<< "${EXTRA_BINDINGS}"
<< " %void = OpTypeVoid\n"
<< " %void_func = OpTypeFunction %void\n"
<< " %int = OpTypeInt 32 1\n"
<< " %uint = OpTypeInt 32 0\n"
<< " %int_0 = OpConstant %int 0\n"
<< " %uint_0 = OpConstant %uint 0\n"
<< " %uint_1 = OpConstant %uint 1\n"
<< " %uint_2 = OpConstant %uint 2\n"
<< " %uint_3 = OpConstant %uint 3\n"
<< " %uint_4 = OpConstant %uint 4\n"
<< " %uint_5 = OpConstant %uint 5\n"
<< " %uint_255 = OpConstant %uint 255\n"
<< " %bool = OpTypeBool\n"
<< " %float = OpTypeFloat 32\n"
<< " %float_0 = OpConstant %float 0\n"
<< " %float_1 = OpConstant %float 1\n"
<< " %float_9 = OpConstant %float 9\n"
<< " %float_0_5 = OpConstant %float 0.5\n"
<< " %float_n1 = OpConstant %float -1\n"
<< " %v3float = OpTypeVector %float 3\n"
<< " %origin_const = OpConstantComposite %v3float %float_0_5 %float_0_5 %float_0\n"
<< " %direction_const = OpConstantComposite %v3float %float_0 %float_0 %float_n1\n"
<< "${EXTRA_TYPES_AND_CONSTANTS}"
<< " %data_func_ptr = OpTypePointer Function ${INPUT_BUFFER_VALUE_TYPE}\n"
<< "${INTERFACE_TYPES_AND_VARIABLES}"
<< " %InputBlock = OpTypeStruct ${INPUT_BUFFER_VALUE_TYPE}\n"
<< " %_ptr_StorageBuffer_InputBlock = OpTypePointer StorageBuffer %InputBlock\n"
<< " %inputBuffer = OpVariable %_ptr_StorageBuffer_InputBlock StorageBuffer\n"
<< " %data_storagebuffer_ptr = OpTypePointer StorageBuffer ${INPUT_BUFFER_VALUE_TYPE}\n"
<< " %OutputBlock = OpTypeStruct %uint\n"
<< "%_ptr_StorageBuffer_OutputBlock = OpTypePointer StorageBuffer %OutputBlock\n"
<< " %outputBuffer = OpVariable %_ptr_StorageBuffer_OutputBlock StorageBuffer\n"
<< " %_ptr_StorageBuffer_uint = OpTypePointer StorageBuffer %uint\n"
<< " %CalleeBlock = OpTypeStruct %uint\n"
<< "%_ptr_StorageBuffer_CalleeBlock = OpTypePointer StorageBuffer %CalleeBlock\n"
<< " %calleeBuffer = OpVariable %_ptr_StorageBuffer_CalleeBlock StorageBuffer\n"
<< " %as_type = OpTypeAccelerationStructureKHR\n"
<< " %as_uniformconstant_ptr = OpTypePointer UniformConstant %as_type\n"
<< " %topLevelAS = OpVariable %as_uniformconstant_ptr UniformConstant\n"
<< "${EXTRA_BINDING_VARIABLES}"
<< " %main = OpFunction %void None %void_func\n"
<< " %main_label = OpLabel\n"
<< "${EXTRA_FUNCTION_VARIABLES}"
<< " %input_val_ptr = OpAccessChain %data_storagebuffer_ptr %inputBuffer %int_0\n"
<< " %output_val_ptr = OpAccessChain %_ptr_StorageBuffer_uint %outputBuffer %int_0\n"
// Note we use Volatile to load the input buffer value before and after the call statements.
<< " %input_val_before = OpLoad ${INPUT_BUFFER_VALUE_TYPE} %input_val_ptr Volatile\n"
<< "${CALC_ZERO_FOR_CALLABLE}"
<< "${CALL_STATEMENTS}"
<< " %input_val_after = OpLoad ${INPUT_BUFFER_VALUE_TYPE} %input_val_ptr Volatile\n"
<< "${CALC_EQUAL_STATEMENT}"
<< " %output_val = OpSelect %uint %equal %uint_1 %uint_0\n"
<< " OpStore %output_val_ptr %output_val\n"
<< " OpReturn\n"
<< " OpFunctionEnd\n"
;
const tcu::StringTemplate spvTemplate (spvTemplateStream.str());
std::map<std::string, std::string> subs;
std::string componentTypeName;
std::string opEqual;
const int numComponents = static_cast<int>(m_params.vectorType);
const auto isArray = (numComponents > static_cast<int>(VectorType::V4));
const auto numComponentsStr = de::toString(numComponents);
subs["EXTRA_CAPABILITIES"] = "";
subs["EXTRA_EXTENSIONS"] = "";
subs["EXTRA_TYPES_AND_CONSTANTS"] = "";
subs["EXTRA_FUNCTION_VARIABLES"] = "";
subs["EXTRA_BINDINGS"] = "";
subs["EXTRA_BINDING_VARIABLES"] = "";
subs["EXTRA_FUNCTIONS"] = "";
// Take into account some of these substitutions will be updated after the if-block.
if (m_params.dataType == DataType::INT32)
{
componentTypeName = "int";
subs["INPUT_BUFFER_VALUE_TYPE"] = "%int";
subs["EXTRA_TYPES_AND_CONSTANTS"] += " %int_37 = OpConstant %int 37\n";
subs["CALC_ZERO_FOR_CALLABLE"] = " %zero_int = OpISub %int %input_val_before %int_37\n"
" %zero_for_callable = OpBitcast %uint %zero_int\n";
}
else if (m_params.dataType == DataType::UINT32)
{
componentTypeName = "uint";
subs["INPUT_BUFFER_VALUE_TYPE"] = "%uint";
subs["EXTRA_TYPES_AND_CONSTANTS"] += " %uint_37 = OpConstant %uint 37\n";
subs["CALC_ZERO_FOR_CALLABLE"] = " %zero_for_callable = OpISub %uint %input_val_before %uint_37\n";
}
else if (m_params.dataType == DataType::INT64)
{
componentTypeName = "long";
subs["EXTRA_CAPABILITIES"] += " OpCapability Int64\n";
subs["INPUT_BUFFER_VALUE_TYPE"] = "%long";
subs["EXTRA_TYPES_AND_CONSTANTS"] += " %long = OpTypeInt 64 1\n"
" %long_37 = OpConstant %long 37\n";
subs["CALC_ZERO_FOR_CALLABLE"] = " %zero_long = OpISub %long %input_val_before %long_37\n"
" %zero_for_callable = OpSConvert %uint %zero_long\n";
}
else if (m_params.dataType == DataType::UINT64)
{
componentTypeName = "ulong";
subs["EXTRA_CAPABILITIES"] += " OpCapability Int64\n";
subs["INPUT_BUFFER_VALUE_TYPE"] = "%ulong";
subs["EXTRA_TYPES_AND_CONSTANTS"] += " %ulong = OpTypeInt 64 0\n"
" %ulong_37 = OpConstant %ulong 37\n";
subs["CALC_ZERO_FOR_CALLABLE"] = " %zero_ulong = OpISub %ulong %input_val_before %ulong_37\n"
" %zero_for_callable = OpUConvert %uint %zero_ulong\n";
}
else if (m_params.dataType == DataType::INT16)
{
componentTypeName = "short";
subs["EXTRA_CAPABILITIES"] += " OpCapability Int16\n"
" OpCapability StorageBuffer16BitAccess\n";
subs["EXTRA_EXTENSIONS"] += " OpExtension \"SPV_KHR_16bit_storage\"\n";
subs["INPUT_BUFFER_VALUE_TYPE"] = "%short";
subs["EXTRA_TYPES_AND_CONSTANTS"] += " %short = OpTypeInt 16 1\n"
" %short_37 = OpConstant %short 37\n";
subs["CALC_ZERO_FOR_CALLABLE"] = " %zero_short = OpISub %short %input_val_before %short_37\n"
" %zero_for_callable = OpSConvert %uint %zero_short\n";
}
else if (m_params.dataType == DataType::UINT16)
{
componentTypeName = "ushort";
subs["EXTRA_CAPABILITIES"] += " OpCapability Int16\n"
" OpCapability StorageBuffer16BitAccess\n";
subs["EXTRA_EXTENSIONS"] += " OpExtension \"SPV_KHR_16bit_storage\"\n";
subs["INPUT_BUFFER_VALUE_TYPE"] = "%ushort";
subs["EXTRA_TYPES_AND_CONSTANTS"] += " %ushort = OpTypeInt 16 0\n"
" %ushort_37 = OpConstant %ushort 37\n";
subs["CALC_ZERO_FOR_CALLABLE"] = " %zero_ushort = OpISub %ushort %input_val_before %ushort_37\n"
" %zero_for_callable = OpUConvert %uint %zero_ushort\n";
}
else if (m_params.dataType == DataType::INT8)
{
componentTypeName = "char";
subs["EXTRA_CAPABILITIES"] += " OpCapability Int8\n"
" OpCapability StorageBuffer8BitAccess\n";
subs["EXTRA_EXTENSIONS"] += " OpExtension \"SPV_KHR_8bit_storage\"\n";
subs["INPUT_BUFFER_VALUE_TYPE"] = "%char";
subs["EXTRA_TYPES_AND_CONSTANTS"] += " %char = OpTypeInt 8 1\n"
" %char_37 = OpConstant %char 37\n";
subs["CALC_ZERO_FOR_CALLABLE"] = " %zero_char = OpISub %char %input_val_before %char_37\n"
" %zero_for_callable = OpSConvert %uint %zero_char\n";
}
else if (m_params.dataType == DataType::UINT8)
{
componentTypeName = "uchar";
subs["EXTRA_CAPABILITIES"] += " OpCapability Int8\n"
" OpCapability StorageBuffer8BitAccess\n";
subs["EXTRA_EXTENSIONS"] += " OpExtension \"SPV_KHR_8bit_storage\"\n";
subs["INPUT_BUFFER_VALUE_TYPE"] = "%uchar";
subs["EXTRA_TYPES_AND_CONSTANTS"] += " %uchar = OpTypeInt 8 0\n"
" %uchar_37 = OpConstant %uchar 37\n";
subs["CALC_ZERO_FOR_CALLABLE"] = " %zero_uchar = OpISub %uchar %input_val_before %uchar_37\n"
" %zero_for_callable = OpUConvert %uint %zero_uchar\n";
}
else if (m_params.dataType == DataType::FLOAT32)
{
componentTypeName = "float";
subs["INPUT_BUFFER_VALUE_TYPE"] = "%float";
subs["EXTRA_TYPES_AND_CONSTANTS"] += " %float_37 = OpConstant %float 37\n";
subs["CALC_ZERO_FOR_CALLABLE"] = " %zero_float = OpFSub %float %input_val_before %float_37\n"
" %zero_for_callable = OpConvertFToU %uint %zero_float\n";
}
else if (m_params.dataType == DataType::FLOAT64)
{
componentTypeName = "double";
subs["EXTRA_CAPABILITIES"] += " OpCapability Float64\n";
subs["INPUT_BUFFER_VALUE_TYPE"] = "%double";
subs["EXTRA_TYPES_AND_CONSTANTS"] += " %double = OpTypeFloat 64\n"
" %double_37 = OpConstant %double 37\n";
subs["CALC_ZERO_FOR_CALLABLE"] = " %zero_double = OpFSub %double %input_val_before %double_37\n"
" %zero_for_callable = OpConvertFToU %uint %zero_double\n";
}
else if (m_params.dataType == DataType::FLOAT16)
{
componentTypeName = "half";
subs["EXTRA_CAPABILITIES"] += " OpCapability Float16\n"
" OpCapability StorageBuffer16BitAccess\n";
subs["EXTRA_EXTENSIONS"] += " OpExtension \"SPV_KHR_16bit_storage\"\n";
subs["INPUT_BUFFER_VALUE_TYPE"] = "%half";
subs["EXTRA_TYPES_AND_CONSTANTS"] += " %half = OpTypeFloat 16\n"
" %half_37 = OpConstant %half 37\n";
subs["CALC_ZERO_FOR_CALLABLE"] = " %zero_half = OpFSub %half %input_val_before %half_37\n"
" %zero_for_callable = OpConvertFToU %uint %zero_half\n";
}
else if (m_params.dataType == DataType::STRUCT)
{
componentTypeName = "InputStruct";
subs["INPUT_BUFFER_VALUE_TYPE"] = "%InputStruct";
subs["EXTRA_TYPES_AND_CONSTANTS"] += " %InputStruct = OpTypeStruct %uint %float\n"
" %float_37 = OpConstant %float 37\n"
" %uint_part_ptr_type = OpTypePointer StorageBuffer %uint\n"
" %float_part_ptr_type = OpTypePointer StorageBuffer %float\n"
" %uint_part_func_ptr_type = OpTypePointer Function %uint\n"
" %float_part_func_ptr_type = OpTypePointer Function %float\n"
" %input_struct_func_ptr_type = OpTypePointer Function %InputStruct\n"
;
subs["INTERFACE_DECORATIONS"] = " OpMemberDecorate %InputStruct 0 Offset 0\n"
" OpMemberDecorate %InputStruct 1 Offset 4\n";
// Sum struct members, then substract constant and convert to uint.
subs["CALC_ZERO_FOR_CALLABLE"] = " %uint_part_ptr = OpAccessChain %uint_part_ptr_type %input_val_ptr %uint_0\n"
" %float_part_ptr = OpAccessChain %float_part_ptr_type %input_val_ptr %uint_1\n"
" %uint_part = OpLoad %uint %uint_part_ptr\n"
" %float_part = OpLoad %float %float_part_ptr\n"
" %uint_as_float = OpConvertUToF %float %uint_part\n"
" %member_sum = OpFAdd %float %float_part %uint_as_float\n"
" %zero_float = OpFSub %float %member_sum %float_37\n"
" %zero_for_callable = OpConvertFToU %uint %zero_float\n"
;
}
else if (samplersNeeded(m_params.dataType))
{
// These tests will use additional bindings as arrays of 2 elements:
// - 1 array of samplers.
// - 1 array of images.
// - 1 array of combined image samplers.
// Input values are typically used as texture coordinates (normally zeros)
// Pixels will contain the expected values instead of them being in the input buffer.
subs["INPUT_BUFFER_VALUE_TYPE"] = "%float";
subs["EXTRA_CAPABILITIES"] += " OpCapability SampledImageArrayNonUniformIndexing\n";
subs["EXTRA_EXTENSIONS"] += " OpExtension \"SPV_EXT_descriptor_indexing\"\n";
subs["MAIN_INTERFACE_EXTRAS"] += " %sampledTexture %textureSampler %combinedImageSampler";
subs["EXTRA_BINDINGS"] += " OpDecorate %sampledTexture DescriptorSet 0\n"
" OpDecorate %sampledTexture Binding 4\n"
" OpDecorate %textureSampler DescriptorSet 0\n"
" OpDecorate %textureSampler Binding 5\n"
" OpDecorate %combinedImageSampler DescriptorSet 0\n"
" OpDecorate %combinedImageSampler Binding 6\n";
subs["EXTRA_TYPES_AND_CONSTANTS"] += " %uint_37 = OpConstant %uint 37\n"
" %v4uint = OpTypeVector %uint 4\n"
" %v2float = OpTypeVector %float 2\n"
" %image_type = OpTypeImage %uint 2D 0 0 0 1 Unknown\n"
" %image_array_type = OpTypeArray %image_type %uint_2\n"
" %image_array_type_uniform_ptr = OpTypePointer UniformConstant %image_array_type\n"
" %image_type_uniform_ptr = OpTypePointer UniformConstant %image_type\n"
" %sampler_type = OpTypeSampler\n"
" %sampler_array_type = OpTypeArray %sampler_type %uint_2\n"
"%sampler_array_type_uniform_ptr = OpTypePointer UniformConstant %sampler_array_type\n"
" %sampler_type_uniform_ptr = OpTypePointer UniformConstant %sampler_type\n"
" %sampled_image_type = OpTypeSampledImage %image_type\n"
" %sampled_image_array_type = OpTypeArray %sampled_image_type %uint_2\n"
"%sampled_image_array_type_uniform_ptr = OpTypePointer UniformConstant %sampled_image_array_type\n"
"%sampled_image_type_uniform_ptr = OpTypePointer UniformConstant %sampled_image_type\n"
;
subs["EXTRA_BINDING_VARIABLES"] += " %sampledTexture = OpVariable %image_array_type_uniform_ptr UniformConstant\n"
" %textureSampler = OpVariable %sampler_array_type_uniform_ptr UniformConstant\n"
" %combinedImageSampler = OpVariable %sampled_image_array_type_uniform_ptr UniformConstant\n"
;
if (m_params.dataType == DataType::IMAGE || m_params.dataType == DataType::SAMPLER)
{
// Use the first sampler and sample from the first image.
subs["CALC_ZERO_FOR_CALLABLE"] += "%image_0_ptr = OpAccessChain %image_type_uniform_ptr %sampledTexture %uint_0\n"
"%sampler_0_ptr = OpAccessChain %sampler_type_uniform_ptr %textureSampler %uint_0\n"
"%sampler_0 = OpLoad %sampler_type %sampler_0_ptr\n"
"%image_0 = OpLoad %image_type %image_0_ptr\n"
"%sampled_image_0 = OpSampledImage %sampled_image_type %image_0 %sampler_0\n"
"%texture_coords_0 = OpCompositeConstruct %v2float %input_val_before %input_val_before\n"
"%pixel_vec_0 = OpImageSampleExplicitLod %v4uint %sampled_image_0 %texture_coords_0 Lod|ZeroExtend %float_0\n"
"%pixel_0 = OpCompositeExtract %uint %pixel_vec_0 0\n"
"%zero_for_callable = OpISub %uint %pixel_0 %uint_37\n"
;
}
else if (m_params.dataType == DataType::SAMPLED_IMAGE)
{
// Use the first combined image sampler.
subs["CALC_ZERO_FOR_CALLABLE"] += "%sampled_image_0_ptr = OpAccessChain %sampled_image_type_uniform_ptr %combinedImageSampler %uint_0\n"
"%sampled_image_0 = OpLoad %sampled_image_type %sampled_image_0_ptr\n"
"%texture_coords_0 = OpCompositeConstruct %v2float %input_val_before %input_val_before\n"
"%pixel_vec_0 = OpImageSampleExplicitLod %v4uint %sampled_image_0 %texture_coords_0 Lod|ZeroExtend %float_0\n"
"%pixel_0 = OpCompositeExtract %uint %pixel_vec_0 0\n"
"%zero_for_callable = OpISub %uint %pixel_0 %uint_37\n"
;
}
else if (m_params.dataType == DataType::PTR_IMAGE)
{
// We attempt to create the second pointer before the call.
subs["CALC_ZERO_FOR_CALLABLE"] += "%image_0_ptr = OpAccessChain %image_type_uniform_ptr %sampledTexture %uint_0\n"
"%image_1_ptr = OpAccessChain %image_type_uniform_ptr %sampledTexture %uint_1\n"
"%image_0 = OpLoad %image_type %image_0_ptr\n"
"%sampler_0_ptr = OpAccessChain %sampler_type_uniform_ptr %textureSampler %uint_0\n"
"%sampler_0 = OpLoad %sampler_type %sampler_0_ptr\n"
"%sampled_image_0 = OpSampledImage %sampled_image_type %image_0 %sampler_0\n"
"%texture_coords_0 = OpCompositeConstruct %v2float %input_val_before %input_val_before\n"
"%pixel_vec_0 = OpImageSampleExplicitLod %v4uint %sampled_image_0 %texture_coords_0 Lod|ZeroExtend %float_0\n"
"%pixel_0 = OpCompositeExtract %uint %pixel_vec_0 0\n"
"%zero_for_callable = OpISub %uint %pixel_0 %uint_37\n"
;
}
else if (m_params.dataType == DataType::PTR_SAMPLER)
{
// We attempt to create the second pointer before the call.
subs["CALC_ZERO_FOR_CALLABLE"] += "%sampler_0_ptr = OpAccessChain %sampler_type_uniform_ptr %textureSampler %uint_0\n"
"%sampler_1_ptr = OpAccessChain %sampler_type_uniform_ptr %textureSampler %uint_1\n"
"%sampler_0 = OpLoad %sampler_type %sampler_0_ptr\n"
"%image_0_ptr = OpAccessChain %image_type_uniform_ptr %sampledTexture %uint_0\n"
"%image_0 = OpLoad %image_type %image_0_ptr\n"
"%sampled_image_0 = OpSampledImage %sampled_image_type %image_0 %sampler_0\n"
"%texture_coords_0 = OpCompositeConstruct %v2float %input_val_before %input_val_before\n"
"%pixel_vec_0 = OpImageSampleExplicitLod %v4uint %sampled_image_0 %texture_coords_0 Lod|ZeroExtend %float_0\n"
"%pixel_0 = OpCompositeExtract %uint %pixel_vec_0 0\n"
"%zero_for_callable = OpISub %uint %pixel_0 %uint_37\n"
;
}
else if (m_params.dataType == DataType::PTR_SAMPLED_IMAGE)
{
// We attempt to create the second pointer before the call.
subs["CALC_ZERO_FOR_CALLABLE"] += "%sampled_image_0_ptr = OpAccessChain %sampled_image_type_uniform_ptr %combinedImageSampler %uint_0\n"
"%sampled_image_1_ptr = OpAccessChain %sampled_image_type_uniform_ptr %combinedImageSampler %uint_1\n"
"%sampled_image_0 = OpLoad %sampled_image_type %sampled_image_0_ptr\n"
"%texture_coords_0 = OpCompositeConstruct %v2float %input_val_before %input_val_before\n"
"%pixel_vec_0 = OpImageSampleExplicitLod %v4uint %sampled_image_0 %texture_coords_0 Lod|ZeroExtend %float_0\n"
"%pixel_0 = OpCompositeExtract %uint %pixel_vec_0 0\n"
"%zero_for_callable = OpISub %uint %pixel_0 %uint_37\n"
;
}
else
{
DE_ASSERT(false);
}
}
else if (storageImageNeeded(m_params.dataType))
{
subs["INPUT_BUFFER_VALUE_TYPE"] = "%int";
subs["MAIN_INTERFACE_EXTRAS"] += " %storageImage";
subs["EXTRA_BINDINGS"] += " OpDecorate %storageImage DescriptorSet 0\n"
" OpDecorate %storageImage Binding 4\n"
;
subs["EXTRA_TYPES_AND_CONSTANTS"] += " %uint_37 = OpConstant %uint 37\n"
" %v2int = OpTypeVector %int 2\n"
" %image_type = OpTypeImage %uint 2D 0 0 0 2 R32ui\n"
" %image_type_uniform_ptr = OpTypePointer UniformConstant %image_type\n"
" %uint_img_ptr = OpTypePointer Image %uint\n"
;
subs["EXTRA_BINDING_VARIABLES"] += " %storageImage = OpVariable %image_type_uniform_ptr UniformConstant\n"
;
// Load value from the image, expecting it to be 37 and swapping it with 5.
subs["CALC_ZERO_FOR_CALLABLE"] += "%coords = OpCompositeConstruct %v2int %input_val_before %input_val_before\n"
"%texel_ptr = OpImageTexelPointer %uint_img_ptr %storageImage %coords %uint_0\n"
"%texel_value = OpAtomicCompareExchange %uint %texel_ptr %uint_1 %uint_0 %uint_0 %uint_5 %uint_37\n"
"%zero_for_callable = OpISub %uint %texel_value %uint_37\n"
;
}
else if (m_params.dataType == DataType::OP_NULL)
{
subs["INPUT_BUFFER_VALUE_TYPE"] = "%uint";
subs["EXTRA_TYPES_AND_CONSTANTS"] += " %uint_37 = OpConstant %uint 37\n"
" %constant_null = OpConstantNull %uint\n"
;
// Create a local copy of the null constant global object to work with it.
subs["CALC_ZERO_FOR_CALLABLE"] += "%constant_null_copy = OpCopyObject %uint %constant_null\n"
"%is_37_before = OpIEqual %bool %input_val_before %uint_37\n"
"%zero_for_callable = OpSelect %uint %is_37_before %constant_null_copy %uint_5\n"
;
}
else if (m_params.dataType == DataType::OP_UNDEF)
{
subs["INPUT_BUFFER_VALUE_TYPE"] = "%uint";
subs["EXTRA_TYPES_AND_CONSTANTS"] += " %uint_37 = OpConstant %uint 37\n"
;
// Extract an undef value and write it to the output buffer to make sure it's used before the call. The value will be overwritten later.
subs["CALC_ZERO_FOR_CALLABLE"] += "%undef_var = OpUndef %uint\n"
"%undef_val_before = OpCopyObject %uint %undef_var\n"
"OpStore %output_val_ptr %undef_val_before Volatile\n"
"%zero_for_callable = OpISub %uint %uint_37 %input_val_before\n"
;
}
else
{
DE_ASSERT(false);
}
// Comparison statement for data before and after the call.
switch (m_params.dataType)
{
case DataType::INT32:
case DataType::UINT32:
case DataType::INT64:
case DataType::UINT64:
case DataType::INT16:
case DataType::UINT16:
case DataType::INT8:
case DataType::UINT8:
opEqual = "OpIEqual";
break;
case DataType::FLOAT32:
case DataType::FLOAT64:
case DataType::FLOAT16:
opEqual = "OpFOrdEqual";
break;
case DataType::STRUCT:
case DataType::IMAGE:
case DataType::SAMPLER:
case DataType::SAMPLED_IMAGE:
case DataType::PTR_IMAGE:
case DataType::PTR_SAMPLER:
case DataType::PTR_SAMPLED_IMAGE:
case DataType::PTR_TEXEL:
case DataType::OP_NULL:
case DataType::OP_UNDEF:
// These needs special code for the comparison.
opEqual = "INVALID";
break;
default:
DE_ASSERT(false);
break;
}
if (m_params.dataType == DataType::STRUCT)
{
// We need to store the before and after values in a variable in order to be able to access each member individually without accessing the StorageBuffer again.
subs["EXTRA_FUNCTION_VARIABLES"] = " %input_val_func_before = OpVariable %input_struct_func_ptr_type Function\n"
" %input_val_func_after = OpVariable %input_struct_func_ptr_type Function\n"
;
subs["CALC_EQUAL_STATEMENT"] = " OpStore %input_val_func_before %input_val_before\n"
" OpStore %input_val_func_after %input_val_after\n"
" %uint_part_func_before_ptr = OpAccessChain %uint_part_func_ptr_type %input_val_func_before %uint_0\n"
" %float_part_func_before_ptr = OpAccessChain %float_part_func_ptr_type %input_val_func_before %uint_1\n"
" %uint_part_func_after_ptr = OpAccessChain %uint_part_func_ptr_type %input_val_func_after %uint_0\n"
" %float_part_func_after_ptr = OpAccessChain %float_part_func_ptr_type %input_val_func_after %uint_1\n"
" %uint_part_before = OpLoad %uint %uint_part_func_before_ptr\n"
" %float_part_before = OpLoad %float %float_part_func_before_ptr\n"
" %uint_part_after = OpLoad %uint %uint_part_func_after_ptr\n"
" %float_part_after = OpLoad %float %float_part_func_after_ptr\n"
" %uint_equal = OpIEqual %bool %uint_part_before %uint_part_after\n"
" %float_equal = OpFOrdEqual %bool %float_part_before %float_part_after\n"
" %equal = OpLogicalAnd %bool %uint_equal %float_equal\n"
;
}
else if (m_params.dataType == DataType::IMAGE)
{
// Use the same image and the second sampler with different coordinates (actually the same).
subs["CALC_EQUAL_STATEMENT"] += "%sampler_1_ptr = OpAccessChain %sampler_type_uniform_ptr %textureSampler %uint_1\n"
"%sampler_1 = OpLoad %sampler_type %sampler_1_ptr\n"
"%sampled_image_1 = OpSampledImage %sampled_image_type %image_0 %sampler_1\n"
"%texture_coords_1 = OpCompositeConstruct %v2float %input_val_after %input_val_after\n"
"%pixel_vec_1 = OpImageSampleExplicitLod %v4uint %sampled_image_1 %texture_coords_1 Lod|ZeroExtend %float_0\n"
"%pixel_1 = OpCompositeExtract %uint %pixel_vec_1 0\n"
"%equal = OpIEqual %bool %pixel_0 %pixel_1\n"
;
}
else if (m_params.dataType == DataType::SAMPLER)
{
// Use the same sampler and sample from the second image with different coordinates (but actually the same).
subs["CALC_EQUAL_STATEMENT"] += "%image_1_ptr = OpAccessChain %image_type_uniform_ptr %sampledTexture %uint_1\n"
"%image_1 = OpLoad %image_type %image_1_ptr\n"
"%sampled_image_1 = OpSampledImage %sampled_image_type %image_1 %sampler_0\n"
"%texture_coords_1 = OpCompositeConstruct %v2float %input_val_after %input_val_after\n"
"%pixel_vec_1 = OpImageSampleExplicitLod %v4uint %sampled_image_1 %texture_coords_1 Lod|ZeroExtend %float_0\n"
"%pixel_1 = OpCompositeExtract %uint %pixel_vec_1 0\n"
"%equal = OpIEqual %bool %pixel_0 %pixel_1\n"
;
}
else if (m_params.dataType == DataType::SAMPLED_IMAGE)
{
// Reuse the same combined image sampler with different coordinates (actually the same).
subs["CALC_EQUAL_STATEMENT"] += "%texture_coords_1 = OpCompositeConstruct %v2float %input_val_after %input_val_after\n"
"%pixel_vec_1 = OpImageSampleExplicitLod %v4uint %sampled_image_0 %texture_coords_1 Lod|ZeroExtend %float_0\n"
"%pixel_1 = OpCompositeExtract %uint %pixel_vec_1 0\n"
"%equal = OpIEqual %bool %pixel_0 %pixel_1\n"
;
}
else if (m_params.dataType == DataType::PTR_IMAGE)
{
// We attempt to use the second pointer only after the call.
subs["CALC_EQUAL_STATEMENT"] += "%image_1 = OpLoad %image_type %image_1_ptr\n"
"%sampled_image_1 = OpSampledImage %sampled_image_type %image_1 %sampler_0\n"
"%texture_coords_1 = OpCompositeConstruct %v2float %input_val_after %input_val_after\n"
"%pixel_vec_1 = OpImageSampleExplicitLod %v4uint %sampled_image_1 %texture_coords_1 Lod|ZeroExtend %float_0\n"
"%pixel_1 = OpCompositeExtract %uint %pixel_vec_1 0\n"
"%equal = OpIEqual %bool %pixel_0 %pixel_1\n"
;
}
else if (m_params.dataType == DataType::PTR_SAMPLER)
{
// We attempt to use the second pointer only after the call.
subs["CALC_EQUAL_STATEMENT"] += "%sampler_1 = OpLoad %sampler_type %sampler_1_ptr\n"
"%sampled_image_1 = OpSampledImage %sampled_image_type %image_0 %sampler_1\n"
"%texture_coords_1 = OpCompositeConstruct %v2float %input_val_after %input_val_after\n"
"%pixel_vec_1 = OpImageSampleExplicitLod %v4uint %sampled_image_1 %texture_coords_1 Lod|ZeroExtend %float_0\n"
"%pixel_1 = OpCompositeExtract %uint %pixel_vec_1 0\n"
"%equal = OpIEqual %bool %pixel_0 %pixel_1\n"
;
}
else if (m_params.dataType == DataType::PTR_SAMPLED_IMAGE)
{
// We attempt to use the second pointer only after the call.
subs["CALC_EQUAL_STATEMENT"] += "%sampled_image_1 = OpLoad %sampled_image_type %sampled_image_1_ptr\n"
"%texture_coords_1 = OpCompositeConstruct %v2float %input_val_after %input_val_after\n"
"%pixel_vec_1 = OpImageSampleExplicitLod %v4uint %sampled_image_1 %texture_coords_1 Lod|ZeroExtend %float_0\n"
"%pixel_1 = OpCompositeExtract %uint %pixel_vec_1 0\n"
"%equal = OpIEqual %bool %pixel_0 %pixel_1\n"
;
}
else if (m_params.dataType == DataType::PTR_TEXEL)
{
// Check value 5 was stored properly.
subs["CALC_EQUAL_STATEMENT"] += "%stored_val = OpAtomicLoad %uint %texel_ptr %uint_1 %uint_0\n"
"%equal = OpIEqual %bool %stored_val %uint_5\n"
;
}
else if (m_params.dataType == DataType::OP_NULL)
{
// Reuse the null constant after the call.
subs["CALC_EQUAL_STATEMENT"] += "%is_37_after = OpIEqual %bool %input_val_after %uint_37\n"
"%writeback_val = OpSelect %uint %is_37_after %constant_null_copy %uint_5\n"
"OpStore %input_val_ptr %writeback_val Volatile\n"
"%readback_val = OpLoad %uint %input_val_ptr Volatile\n"
"%equal = OpIEqual %bool %readback_val %uint_0\n"
;
}
else if (m_params.dataType == DataType::OP_UNDEF)
{
// Extract another undef value and write it to the input buffer. It will not be checked later.
subs["CALC_EQUAL_STATEMENT"] += "%undef_val_after = OpCopyObject %uint %undef_var\n"
"OpStore %input_val_ptr %undef_val_after Volatile\n"
"%equal = OpIEqual %bool %input_val_after %input_val_before\n"
;
}
else
{
subs["CALC_EQUAL_STATEMENT"] += " %equal = " + opEqual + " %bool %input_val_before %input_val_after\n";
}
// Modifications for vectors and arrays.
if (numComponents > 1)
{
const std::string vectorTypeName = "v" + numComponentsStr + componentTypeName;
const std::string opType = (isArray ? "OpTypeArray" : "OpTypeVector");
const std::string componentCountStr = (isArray ? ("%uint_" + numComponentsStr) : numComponentsStr);
// Some extra types are needed.
if (!(m_params.dataType == DataType::FLOAT32 && m_params.vectorType == VectorType::V3))
{
// Note: v3float is already defined in the shader by default.
subs["EXTRA_TYPES_AND_CONSTANTS"] += "%" + vectorTypeName + " = " + opType + " %" + componentTypeName + " " + componentCountStr + "\n";
}
subs["EXTRA_TYPES_AND_CONSTANTS"] += "%v" + numComponentsStr + "bool = " + opType + " %bool " + componentCountStr + "\n";
subs["EXTRA_TYPES_AND_CONSTANTS"] += "%comp_ptr = OpTypePointer StorageBuffer %" + componentTypeName + "\n";
// The input value in the buffer has a different type.
subs["INPUT_BUFFER_VALUE_TYPE"] = "%" + vectorTypeName;
// Overwrite the way we calculate the zero used in the call.
// Proper operations for adding, substracting and converting components.
std::string opAdd;
std::string opSub;
std::string opConvert;
switch (m_params.dataType)
{
case DataType::INT32:
case DataType::UINT32:
case DataType::INT64:
case DataType::UINT64:
case DataType::INT16:
case DataType::UINT16:
case DataType::INT8:
case DataType::UINT8:
opAdd = "OpIAdd";
opSub = "OpISub";
break;
case DataType::FLOAT32:
case DataType::FLOAT64:
case DataType::FLOAT16:
opAdd = "OpFAdd";
opSub = "OpFSub";
break;
default:
DE_ASSERT(false);
break;
}
switch (m_params.dataType)
{
case DataType::UINT32:
opConvert = "OpCopyObject";
break;
case DataType::INT32:
opConvert = "OpBitcast";
break;
case DataType::INT64:
case DataType::INT16:
case DataType::INT8:
opConvert = "OpSConvert";
break;
case DataType::UINT64:
case DataType::UINT16:
case DataType::UINT8:
opConvert = "OpUConvert";
break;
case DataType::FLOAT32:
case DataType::FLOAT64:
case DataType::FLOAT16:
opConvert = "OpConvertFToU";
break;
default:
DE_ASSERT(false);
break;
}
std::ostringstream zeroForCallable;
// Create pointers to components and load components.
for (int i = 0; i < numComponents; ++i)
{
zeroForCallable
<< "%component_ptr_" << i << " = OpAccessChain %comp_ptr %input_val_ptr %uint_" << i << "\n"
<< "%component_" << i << " = OpLoad %" << componentTypeName << " %component_ptr_" << i << "\n"
;
}
// Sum components together in %total_sum.
for (int i = 1; i < numComponents; ++i)
{
const std::string previous = ((i == 1) ? "%component_0" : ("%partial_" + de::toString(i-1)));
const std::string resultName = ((i == (numComponents - 1)) ? "%total_sum" : ("%partial_" + de::toString(i)));
zeroForCallable << resultName << " = " << opAdd << " %" << componentTypeName << " %component_" << i << " " << previous << "\n";
}
// Recalculate the zero.
zeroForCallable
<< "%zero_" << componentTypeName << " = " << opSub << " %" << componentTypeName << " %total_sum %" << componentTypeName << "_37\n"
<< "%zero_for_callable = " << opConvert << " %uint %zero_" << componentTypeName << "\n"
;
// Finally replace the zero_for_callable statements with the special version for vectors.
subs["CALC_ZERO_FOR_CALLABLE"] = zeroForCallable.str();
// Rework comparison statements.
if (isArray)
{
// Arrays need to be compared per-component.
std::ostringstream calcEqual;
for (int i = 0; i < numComponents; ++i)
{
calcEqual
<< "%component_after_" << i << " = OpLoad %" << componentTypeName << " %component_ptr_" << i << "\n"
<< "%equal_" << i << " = " << opEqual << " %bool %component_" << i << " %component_after_" << i << "\n";
;
if (i > 0)
calcEqual << "%and_" << i << " = OpLogicalAnd %bool %equal_" << (i - 1) << " %equal_" << i << "\n";
if (i == numComponents - 1)
calcEqual << "%equal = OpCopyObject %bool %and_" << i << "\n";
}
subs["CALC_EQUAL_STATEMENT"] = calcEqual.str();
}
else
{
// Vectors can be compared using a bool vector and OpAll.
subs["CALC_EQUAL_STATEMENT"] = " %equal_vector = " + opEqual + " %v" + numComponentsStr + "bool %input_val_before %input_val_after\n";
subs["CALC_EQUAL_STATEMENT"] += " %equal = OpAll %bool %equal_vector\n";
}
}
if (isArray)
{
// Arrays need an ArrayStride decoration.
std::ostringstream interfaceDecorations;
interfaceDecorations << "OpDecorate %v" << numComponentsStr << componentTypeName << " ArrayStride " << getElementSize(m_params.dataType, VectorType::SCALAR) << "\n";
subs["INTERFACE_DECORATIONS"] = interfaceDecorations.str();
}
const auto inputBlockDecls = getGLSLInputValDecl(m_params.dataType, m_params.vectorType);
std::ostringstream glslBindings;
glslBindings
<< inputBlockDecls.first // Additional data types needed.
<< "layout(set = 0, binding = 0) uniform accelerationStructureEXT topLevelAS;\n"
<< "layout(set = 0, binding = 1) buffer CalleeBlock { uint val; } calleeBuffer;\n"
<< "layout(set = 0, binding = 2) buffer OutputBlock { uint val; } outputBuffer;\n"
<< "layout(set = 0, binding = 3) buffer InputBlock { " << inputBlockDecls.second << " } inputBuffer;\n"
;
if (samplersNeeded(m_params.dataType))
{
glslBindings
<< "layout(set = 0, binding = 4) uniform utexture2D sampledTexture[2];\n"
<< "layout(set = 0, binding = 5) uniform sampler textureSampler[2];\n"
<< "layout(set = 0, binding = 6) uniform usampler2D combinedImageSampler[2];\n"
;
}
else if (storageImageNeeded(m_params.dataType))
{
glslBindings
<< "layout(set = 0, binding = 4, r32ui) uniform uimage2D storageImage;\n"
;
}
const auto glslBindingsStr = glslBindings.str();
const auto glslHeaderStr = "#version 460 core\n"
"#extension GL_EXT_ray_tracing : require\n"
"#extension GL_EXT_shader_explicit_arithmetic_types : require\n";
if (m_params.callType == CallType::TRACE_RAY)
{
subs["ENTRY_POINT"] = "RayGenerationKHR";
subs["MAIN_INTERFACE_EXTRAS"] += " %hitValue";
subs["INTERFACE_DECORATIONS"] += " OpDecorate %hitValue Location 0\n";
subs["INTERFACE_TYPES_AND_VARIABLES"] = " %payload_ptr = OpTypePointer RayPayloadKHR %v3float\n"
" %hitValue = OpVariable %payload_ptr RayPayloadKHR\n";
subs["CALL_STATEMENTS"] = " %as_value = OpLoad %as_type %topLevelAS\n"
" OpTraceRayKHR %as_value %uint_0 %uint_255 %zero_for_callable %zero_for_callable %zero_for_callable %origin_const %float_0 %direction_const %float_9 %hitValue\n";
const auto rgen = spvTemplate.specialize(subs);
programCollection.spirvAsmSources.add("rgen") << rgen << spvBuildOptions;
std::stringstream chit;
chit
<< glslHeaderStr
<< "layout(location = 0) rayPayloadInEXT vec3 hitValue;\n"
<< "hitAttributeEXT vec3 attribs;\n"
<< glslBindingsStr
<< "void main()\n"
<< "{\n"
<< " calleeBuffer.val = 1u;\n"
<< "}\n";
;
programCollection.glslSources.add("chit") << glu::ClosestHitSource(updateRayTracingGLSL(chit.str())) << buildOptions;
}
else if (m_params.callType == CallType::EXECUTE_CALLABLE)
{
subs["ENTRY_POINT"] = "RayGenerationKHR";
subs["MAIN_INTERFACE_EXTRAS"] += " %callableData";
subs["INTERFACE_DECORATIONS"] += " OpDecorate %callableData Location 0\n";
subs["INTERFACE_TYPES_AND_VARIABLES"] = " %callable_data_ptr = OpTypePointer CallableDataKHR %float\n"
" %callableData = OpVariable %callable_data_ptr CallableDataKHR\n";
subs["CALL_STATEMENTS"] = " OpExecuteCallableKHR %zero_for_callable %callableData\n";
const auto rgen = spvTemplate.specialize(subs);
programCollection.spirvAsmSources.add("rgen") << rgen << spvBuildOptions;
std::ostringstream call;
call
<< glslHeaderStr
<< "layout(location = 0) callableDataInEXT float callableData;\n"
<< glslBindingsStr
<< "void main()\n"
<< "{\n"
<< " calleeBuffer.val = 1u;\n"
<< "}\n"
;
programCollection.glslSources.add("call") << glu::CallableSource(updateRayTracingGLSL(call.str())) << buildOptions;
}
else if (m_params.callType == CallType::REPORT_INTERSECTION)
{
subs["ENTRY_POINT"] = "IntersectionKHR";
subs["MAIN_INTERFACE_EXTRAS"] += " %attribs";
subs["INTERFACE_DECORATIONS"] += "";
subs["INTERFACE_TYPES_AND_VARIABLES"] = " %hit_attribute_ptr = OpTypePointer HitAttributeKHR %v3float\n"
" %attribs = OpVariable %hit_attribute_ptr HitAttributeKHR\n";
subs["CALL_STATEMENTS"] = " %intersection_ret = OpReportIntersectionKHR %bool %float_1 %zero_for_callable\n";
const auto rint = spvTemplate.specialize(subs);
programCollection.spirvAsmSources.add("rint") << rint << spvBuildOptions;
std::ostringstream rgen;
rgen
<< glslHeaderStr
<< "layout(location = 0) rayPayloadEXT vec3 hitValue;\n"
<< glslBindingsStr
<< "void main()\n"
<< "{\n"
<< " traceRayEXT(topLevelAS, 0u, 0xFFu, 0, 0, 0, vec3(0.5, 0.5, 0.0), 0.0, vec3(0.0, 0.0, -1.0), 9.0, 0);\n"
<< "}\n"
;
programCollection.glslSources.add("rgen") << glu::RaygenSource(updateRayTracingGLSL(rgen.str())) << buildOptions;
std::stringstream ahit;
ahit
<< glslHeaderStr
<< "layout(location = 0) rayPayloadInEXT vec3 hitValue;\n"
<< "hitAttributeEXT vec3 attribs;\n"
<< glslBindingsStr
<< "void main()\n"
<< "{\n"
<< " calleeBuffer.val = 1u;\n"
<< "}\n";
;
programCollection.glslSources.add("ahit") << glu::AnyHitSource(updateRayTracingGLSL(ahit.str())) << buildOptions;
}
else
{
DE_ASSERT(false);
}
}
using v2i32 = tcu::Vector<deInt32, 2>;
using v3i32 = tcu::Vector<deInt32, 3>;
using v4i32 = tcu::Vector<deInt32, 4>;
using a5i32 = std::array<deInt32, 5>;
using v2u32 = tcu::Vector<deUint32, 2>;
using v3u32 = tcu::Vector<deUint32, 3>;
using v4u32 = tcu::Vector<deUint32, 4>;
using a5u32 = std::array<deUint32, 5>;
using v2i64 = tcu::Vector<deInt64, 2>;
using v3i64 = tcu::Vector<deInt64, 3>;
using v4i64 = tcu::Vector<deInt64, 4>;
using a5i64 = std::array<deInt64, 5>;
using v2u64 = tcu::Vector<deUint64, 2>;
using v3u64 = tcu::Vector<deUint64, 3>;
using v4u64 = tcu::Vector<deUint64, 4>;
using a5u64 = std::array<deUint64, 5>;
using v2i16 = tcu::Vector<deInt16, 2>;
using v3i16 = tcu::Vector<deInt16, 3>;
using v4i16 = tcu::Vector<deInt16, 4>;
using a5i16 = std::array<deInt16, 5>;
using v2u16 = tcu::Vector<deUint16, 2>;
using v3u16 = tcu::Vector<deUint16, 3>;
using v4u16 = tcu::Vector<deUint16, 4>;
using a5u16 = std::array<deUint16, 5>;
using v2i8 = tcu::Vector<deInt8, 2>;
using v3i8 = tcu::Vector<deInt8, 3>;
using v4i8 = tcu::Vector<deInt8, 4>;
using a5i8 = std::array<deInt8, 5>;
using v2u8 = tcu::Vector<deUint8, 2>;
using v3u8 = tcu::Vector<deUint8, 3>;
using v4u8 = tcu::Vector<deUint8, 4>;
using a5u8 = std::array<deUint8, 5>;
using v2f32 = tcu::Vector<tcu::Float32, 2>;
using v3f32 = tcu::Vector<tcu::Float32, 3>;
using v4f32 = tcu::Vector<tcu::Float32, 4>;
using a5f32 = std::array<tcu::Float32, 5>;
using v2f64 = tcu::Vector<tcu::Float64, 2>;
using v3f64 = tcu::Vector<tcu::Float64, 3>;
using v4f64 = tcu::Vector<tcu::Float64, 4>;
using a5f64 = std::array<tcu::Float64, 5>;
using v2f16 = tcu::Vector<tcu::Float16, 2>;
using v3f16 = tcu::Vector<tcu::Float16, 3>;
using v4f16 = tcu::Vector<tcu::Float16, 4>;
using a5f16 = std::array<tcu::Float16, 5>;
// Scalar types get filled with value 37, matching the value that will be substracted in the shader.
#define GEN_SCALAR_FILL(DATA_TYPE) \
do { \
const auto inputBufferValue = static_cast<DATA_TYPE>(37.0); \
deMemcpy(bufferPtr, &inputBufferValue, sizeof(inputBufferValue)); \
} while (0)
// Vector types get filled with values that add up to 37, matching the value that will be substracted in the shader.
#define GEN_V2_FILL(DATA_TYPE) \
do { \
DATA_TYPE inputBufferValue; \
inputBufferValue.x() = static_cast<DATA_TYPE::Element>(21.0); \
inputBufferValue.y() = static_cast<DATA_TYPE::Element>(16.0); \
deMemcpy(bufferPtr, &inputBufferValue, sizeof(inputBufferValue)); \
} while (0)
#define GEN_V3_FILL(DATA_TYPE) \
do { \
DATA_TYPE inputBufferValue; \
inputBufferValue.x() = static_cast<DATA_TYPE::Element>(11.0); \
inputBufferValue.y() = static_cast<DATA_TYPE::Element>(19.0); \
inputBufferValue.z() = static_cast<DATA_TYPE::Element>(7.0); \
deMemcpy(bufferPtr, &inputBufferValue, sizeof(inputBufferValue)); \
} while (0)
#define GEN_V4_FILL(DATA_TYPE) \
do { \
DATA_TYPE inputBufferValue; \
inputBufferValue.x() = static_cast<DATA_TYPE::Element>(9.0); \
inputBufferValue.y() = static_cast<DATA_TYPE::Element>(11.0); \
inputBufferValue.z() = static_cast<DATA_TYPE::Element>(3.0); \
inputBufferValue.w() = static_cast<DATA_TYPE::Element>(14.0); \
deMemcpy(bufferPtr, &inputBufferValue, sizeof(inputBufferValue)); \
} while (0)
#define GEN_A5_FILL(DATA_TYPE) \
do { \
DATA_TYPE inputBufferValue; \
inputBufferValue[0] = static_cast<DATA_TYPE::value_type>(13.0); \
inputBufferValue[1] = static_cast<DATA_TYPE::value_type>(6.0); \
inputBufferValue[2] = static_cast<DATA_TYPE::value_type>(2.0); \
inputBufferValue[3] = static_cast<DATA_TYPE::value_type>(5.0); \
inputBufferValue[4] = static_cast<DATA_TYPE::value_type>(11.0); \
deMemcpy(bufferPtr, inputBufferValue.data(), de::dataSize(inputBufferValue)); \
} while (0)
void fillInputBuffer (DataType dataType, VectorType vectorType, void* bufferPtr)
{
if (vectorType == VectorType::SCALAR)
{
if (dataType == DataType::INT32) GEN_SCALAR_FILL(deInt32);
else if (dataType == DataType::UINT32) GEN_SCALAR_FILL(deUint32);
else if (dataType == DataType::INT64) GEN_SCALAR_FILL(deInt64);
else if (dataType == DataType::UINT64) GEN_SCALAR_FILL(deUint64);
else if (dataType == DataType::INT16) GEN_SCALAR_FILL(deInt16);
else if (dataType == DataType::UINT16) GEN_SCALAR_FILL(deUint16);
else if (dataType == DataType::INT8) GEN_SCALAR_FILL(deInt8);
else if (dataType == DataType::UINT8) GEN_SCALAR_FILL(deUint8);
else if (dataType == DataType::FLOAT32) GEN_SCALAR_FILL(tcu::Float32);
else if (dataType == DataType::FLOAT64) GEN_SCALAR_FILL(tcu::Float64);
else if (dataType == DataType::FLOAT16) GEN_SCALAR_FILL(tcu::Float16);
else if (dataType == DataType::STRUCT)
{
InputStruct data = { 12u, 25.0f };
deMemcpy(bufferPtr, &data, sizeof(data));
}
else if (dataType == DataType::OP_NULL) GEN_SCALAR_FILL(deUint32);
else if (dataType == DataType::OP_UNDEF) GEN_SCALAR_FILL(deUint32);
else
{
DE_ASSERT(false);
}
}
else if (vectorType == VectorType::V2)
{
if (dataType == DataType::INT32) GEN_V2_FILL(v2i32);
else if (dataType == DataType::UINT32) GEN_V2_FILL(v2u32);
else if (dataType == DataType::INT64) GEN_V2_FILL(v2i64);
else if (dataType == DataType::UINT64) GEN_V2_FILL(v2u64);
else if (dataType == DataType::INT16) GEN_V2_FILL(v2i16);
else if (dataType == DataType::UINT16) GEN_V2_FILL(v2u16);
else if (dataType == DataType::INT8) GEN_V2_FILL(v2i8);
else if (dataType == DataType::UINT8) GEN_V2_FILL(v2u8);
else if (dataType == DataType::FLOAT32) GEN_V2_FILL(v2f32);
else if (dataType == DataType::FLOAT64) GEN_V2_FILL(v2f64);
else if (dataType == DataType::FLOAT16) GEN_V2_FILL(v2f16);
else
{
DE_ASSERT(false);
}
}
else if (vectorType == VectorType::V3)
{
if (dataType == DataType::INT32) GEN_V3_FILL(v3i32);
else if (dataType == DataType::UINT32) GEN_V3_FILL(v3u32);
else if (dataType == DataType::INT64) GEN_V3_FILL(v3i64);
else if (dataType == DataType::UINT64) GEN_V3_FILL(v3u64);
else if (dataType == DataType::INT16) GEN_V3_FILL(v3i16);
else if (dataType == DataType::UINT16) GEN_V3_FILL(v3u16);
else if (dataType == DataType::INT8) GEN_V3_FILL(v3i8);
else if (dataType == DataType::UINT8) GEN_V3_FILL(v3u8);
else if (dataType == DataType::FLOAT32) GEN_V3_FILL(v3f32);
else if (dataType == DataType::FLOAT64) GEN_V3_FILL(v3f64);
else if (dataType == DataType::FLOAT16) GEN_V3_FILL(v3f16);
else
{
DE_ASSERT(false);
}
}
else if (vectorType == VectorType::V4)
{
if (dataType == DataType::INT32) GEN_V4_FILL(v4i32);
else if (dataType == DataType::UINT32) GEN_V4_FILL(v4u32);
else if (dataType == DataType::INT64) GEN_V4_FILL(v4i64);
else if (dataType == DataType::UINT64) GEN_V4_FILL(v4u64);
else if (dataType == DataType::INT16) GEN_V4_FILL(v4i16);
else if (dataType == DataType::UINT16) GEN_V4_FILL(v4u16);
else if (dataType == DataType::INT8) GEN_V4_FILL(v4i8);
else if (dataType == DataType::UINT8) GEN_V4_FILL(v4u8);
else if (dataType == DataType::FLOAT32) GEN_V4_FILL(v4f32);
else if (dataType == DataType::FLOAT64) GEN_V4_FILL(v4f64);
else if (dataType == DataType::FLOAT16) GEN_V4_FILL(v4f16);
else
{
DE_ASSERT(false);
}
}
else if (vectorType == VectorType::A5)
{
if (dataType == DataType::INT32) GEN_A5_FILL(a5i32);
else if (dataType == DataType::UINT32) GEN_A5_FILL(a5u32);
else if (dataType == DataType::INT64) GEN_A5_FILL(a5i64);
else if (dataType == DataType::UINT64) GEN_A5_FILL(a5u64);
else if (dataType == DataType::INT16) GEN_A5_FILL(a5i16);
else if (dataType == DataType::UINT16) GEN_A5_FILL(a5u16);
else if (dataType == DataType::INT8) GEN_A5_FILL(a5i8);
else if (dataType == DataType::UINT8) GEN_A5_FILL(a5u8);
else if (dataType == DataType::FLOAT32) GEN_A5_FILL(a5f32);
else if (dataType == DataType::FLOAT64) GEN_A5_FILL(a5f64);
else if (dataType == DataType::FLOAT16) GEN_A5_FILL(a5f16);
else
{
DE_ASSERT(false);
}
}
else
{
DE_ASSERT(false);
}
}
tcu::TestStatus DataSpillTestInstance::iterate (void)
{
const auto& vki = m_context.getInstanceInterface();
const auto physicalDevice = m_context.getPhysicalDevice();
const auto& vkd = m_context.getDeviceInterface();
const auto device = m_context.getDevice();
const auto queue = m_context.getUniversalQueue();
const auto familyIndex = m_context.getUniversalQueueFamilyIndex();
auto& alloc = m_context.getDefaultAllocator();
const auto shaderStages = getShaderStages(m_params.callType);
// Command buffer.
const auto cmdPool = makeCommandPool(vkd, device, familyIndex);
const auto cmdBufferPtr = allocateCommandBuffer(vkd, device, cmdPool.get(), VK_COMMAND_BUFFER_LEVEL_PRIMARY);
const auto cmdBuffer = cmdBufferPtr.get();
beginCommandBuffer(vkd, cmdBuffer);
// Callee, input and output buffers.
const auto calleeBufferSize = getElementSize(DataType::UINT32, VectorType::SCALAR);
const auto outputBufferSize = getElementSize(DataType::UINT32, VectorType::SCALAR);
const auto inputBufferSize = getElementSize(m_params.dataType, m_params.vectorType);
const auto calleeBufferInfo = makeBufferCreateInfo(calleeBufferSize, VK_BUFFER_USAGE_STORAGE_BUFFER_BIT);
const auto outputBufferInfo = makeBufferCreateInfo(outputBufferSize, VK_BUFFER_USAGE_STORAGE_BUFFER_BIT);
const auto inputBufferInfo = makeBufferCreateInfo(inputBufferSize, VK_BUFFER_USAGE_STORAGE_BUFFER_BIT);
BufferWithMemory calleeBuffer (vkd, device, alloc, calleeBufferInfo, MemoryRequirement::HostVisible);
BufferWithMemory outputBuffer (vkd, device, alloc, outputBufferInfo, MemoryRequirement::HostVisible);
BufferWithMemory inputBuffer (vkd, device, alloc, inputBufferInfo, MemoryRequirement::HostVisible);
// Fill buffers with values.
auto& calleeBufferAlloc = calleeBuffer.getAllocation();
auto* calleeBufferPtr = calleeBufferAlloc.getHostPtr();
auto& outputBufferAlloc = outputBuffer.getAllocation();
auto* outputBufferPtr = outputBufferAlloc.getHostPtr();
auto& inputBufferAlloc = inputBuffer.getAllocation();
auto* inputBufferPtr = inputBufferAlloc.getHostPtr();
deMemset(calleeBufferPtr, 0, static_cast<size_t>(calleeBufferSize));
deMemset(outputBufferPtr, 0, static_cast<size_t>(outputBufferSize));
if (samplersNeeded(m_params.dataType) || storageImageNeeded(m_params.dataType))
{
// The input buffer for these cases will be filled with zeros (sampling coordinates), and the input textures will contain the interesting input value.
deMemset(inputBufferPtr, 0, static_cast<size_t>(inputBufferSize));
}
else
{
// We want to fill the input buffer with values that will be consistently used in the shader to obtain a result of zero.
fillInputBuffer(m_params.dataType, m_params.vectorType, inputBufferPtr);
}
flushAlloc(vkd, device, calleeBufferAlloc);
flushAlloc(vkd, device, outputBufferAlloc);
flushAlloc(vkd, device, inputBufferAlloc);
// Acceleration structures.
de::MovePtr<BottomLevelAccelerationStructure> bottomLevelAccelerationStructure;
de::MovePtr<TopLevelAccelerationStructure> topLevelAccelerationStructure;
bottomLevelAccelerationStructure = makeBottomLevelAccelerationStructure();
bottomLevelAccelerationStructure->setDefaultGeometryData(getShaderStageForGeometry(m_params.callType), VK_GEOMETRY_NO_DUPLICATE_ANY_HIT_INVOCATION_BIT_KHR);
bottomLevelAccelerationStructure->createAndBuild(vkd, device, cmdBuffer, alloc);
topLevelAccelerationStructure = makeTopLevelAccelerationStructure();
topLevelAccelerationStructure->setInstanceCount(1);
topLevelAccelerationStructure->addInstance(de::SharedPtr<BottomLevelAccelerationStructure>(bottomLevelAccelerationStructure.release()));
topLevelAccelerationStructure->createAndBuild(vkd, device, cmdBuffer, alloc);
// Get some ray tracing properties.
deUint32 shaderGroupHandleSize = 0u;
deUint32 shaderGroupBaseAlignment = 1u;
{
const auto rayTracingPropertiesKHR = makeRayTracingProperties(vki, physicalDevice);
shaderGroupHandleSize = rayTracingPropertiesKHR->getShaderGroupHandleSize();
shaderGroupBaseAlignment = rayTracingPropertiesKHR->getShaderGroupBaseAlignment();
}
// Textures and samplers if needed.
de::MovePtr<BufferWithMemory> textureData;
std::vector<de::MovePtr<ImageWithMemory>> textures;
std::vector<Move<VkImageView>> textureViews;
std::vector<Move<VkSampler>> samplers;
if (samplersNeeded(m_params.dataType) || storageImageNeeded(m_params.dataType))
{
// Create texture data with the expected contents.
{
const auto textureDataSize = static_cast<VkDeviceSize>(sizeof(deUint32));
const auto textureDataCreateInfo = makeBufferCreateInfo(textureDataSize, VK_BUFFER_USAGE_TRANSFER_SRC_BIT);
textureData = de::MovePtr<BufferWithMemory>(new BufferWithMemory(vkd, device, alloc, textureDataCreateInfo, MemoryRequirement::HostVisible));
auto& textureDataAlloc = textureData->getAllocation();
auto* textureDataPtr = textureDataAlloc.getHostPtr();
fillInputBuffer(DataType::UINT32, VectorType::SCALAR, textureDataPtr);
flushAlloc(vkd, device, textureDataAlloc);
}
// Images will be created like this with different usages.
VkImageCreateInfo imageCreateInfo =
{
VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO, // VkStructureType sType;
nullptr, // const void* pNext;
0u, // VkImageCreateFlags flags;
VK_IMAGE_TYPE_2D, // VkImageType imageType;
kImageFormat, // VkFormat format;
kImageExtent, // VkExtent3D extent;
1u, // deUint32 mipLevels;
1u, // deUint32 arrayLayers;
VK_SAMPLE_COUNT_1_BIT, // VkSampleCountFlagBits samples;
VK_IMAGE_TILING_OPTIMAL, // VkImageTiling tiling;
kSampledImageUsage, // VkImageUsageFlags usage;
VK_SHARING_MODE_EXCLUSIVE, // VkSharingMode sharingMode;
0u, // deUint32 queueFamilyIndexCount;
nullptr, // const deUint32* pQueueFamilyIndices;
VK_IMAGE_LAYOUT_UNDEFINED, // VkImageLayout initialLayout;
};
const auto imageSubresourceRange = makeImageSubresourceRange(VK_IMAGE_ASPECT_COLOR_BIT, 0u, 1u, 0u, 1u);
const auto imageSubresourceLayers = makeImageSubresourceLayers(VK_IMAGE_ASPECT_COLOR_BIT, 0u, 0u, 1u);
if (samplersNeeded(m_params.dataType))
{
// All samplers will be created like this.
const VkSamplerCreateInfo samplerCreateInfo =
{
VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO, // VkStructureType sType;
nullptr, // const void* pNext;
0u, // VkSamplerCreateFlags flags;
VK_FILTER_NEAREST, // VkFilter magFilter;
VK_FILTER_NEAREST, // VkFilter minFilter;
VK_SAMPLER_MIPMAP_MODE_NEAREST, // VkSamplerMipmapMode mipmapMode;
VK_SAMPLER_ADDRESS_MODE_REPEAT, // VkSamplerAddressMode addressModeU;
VK_SAMPLER_ADDRESS_MODE_REPEAT, // VkSamplerAddressMode addressModeV;
VK_SAMPLER_ADDRESS_MODE_REPEAT, // VkSamplerAddressMode addressModeW;
0.0, // float mipLodBias;
VK_FALSE, // VkBool32 anisotropyEnable;
1.0f, // float maxAnisotropy;
VK_FALSE, // VkBool32 compareEnable;
VK_COMPARE_OP_ALWAYS, // VkCompareOp compareOp;
0.0f, // float minLod;
1.0f, // float maxLod;
VK_BORDER_COLOR_INT_OPAQUE_BLACK, // VkBorderColor borderColor;
VK_FALSE, // VkBool32 unnormalizedCoordinates;
};
// Create textures and samplers.
for (size_t i = 0; i < kNumImages; ++i)
{
textures.emplace_back(new ImageWithMemory(vkd, device, alloc, imageCreateInfo, MemoryRequirement::Any));
textureViews.emplace_back(makeImageView(vkd, device, textures.back()->get(), VK_IMAGE_VIEW_TYPE_2D, kImageFormat, imageSubresourceRange));
}
for (size_t i = 0; i < kNumSamplers; ++i)
samplers.emplace_back(createSampler(vkd, device, &samplerCreateInfo));
// Make sure texture data is available in the transfer stage.
const auto textureDataBarrier = makeMemoryBarrier(VK_ACCESS_HOST_WRITE_BIT, VK_ACCESS_TRANSFER_READ_BIT);
vkd.cmdPipelineBarrier(cmdBuffer, VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0u, 1u, &textureDataBarrier, 0u, nullptr, 0u, nullptr);
const auto bufferImageCopy = makeBufferImageCopy(kImageExtent, imageSubresourceLayers);
// Fill textures with data and prepare them for the ray tracing pipeline stages.
for (size_t i = 0; i < kNumImages; ++i)
{
const auto texturePreCopyBarrier = makeImageMemoryBarrier(0u, VK_ACCESS_TRANSFER_WRITE_BIT, VK_IMAGE_LAYOUT_UNDEFINED, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, textures[i]->get(), imageSubresourceRange);
const auto texturePostCopyBarrier = makeImageMemoryBarrier(VK_ACCESS_TRANSFER_WRITE_BIT, VK_ACCESS_SHADER_READ_BIT, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL, textures[i]->get(), imageSubresourceRange);
vkd.cmdPipelineBarrier(cmdBuffer, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0u, 0u, nullptr, 0u, nullptr, 1u, &texturePreCopyBarrier);
vkd.cmdCopyBufferToImage(cmdBuffer, textureData->get(), textures[i]->get(), VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1u, &bufferImageCopy);
vkd.cmdPipelineBarrier(cmdBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_RAY_TRACING_SHADER_BIT_KHR, 0u, 0u, nullptr, 0u, nullptr, 1u, &texturePostCopyBarrier);
}
}
else if (storageImageNeeded(m_params.dataType))
{
// Image will be used for storage.
imageCreateInfo.usage = kStorageImageUsage;
textures.emplace_back(new ImageWithMemory(vkd, device, alloc, imageCreateInfo, MemoryRequirement::Any));
textureViews.emplace_back(makeImageView(vkd, device, textures.back()->get(), VK_IMAGE_VIEW_TYPE_2D, kImageFormat, imageSubresourceRange));
// Make sure texture data is available in the transfer stage.
const auto textureDataBarrier = makeMemoryBarrier(VK_ACCESS_HOST_WRITE_BIT, VK_ACCESS_TRANSFER_READ_BIT);
vkd.cmdPipelineBarrier(cmdBuffer, VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0u, 1u, &textureDataBarrier, 0u, nullptr, 0u, nullptr);
const auto bufferImageCopy = makeBufferImageCopy(kImageExtent, imageSubresourceLayers);
const auto texturePreCopyBarrier = makeImageMemoryBarrier(0u, VK_ACCESS_TRANSFER_WRITE_BIT, VK_IMAGE_LAYOUT_UNDEFINED, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, textures.back()->get(), imageSubresourceRange);
const auto texturePostCopyBarrier = makeImageMemoryBarrier(VK_ACCESS_TRANSFER_WRITE_BIT, (VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT), VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, VK_IMAGE_LAYOUT_GENERAL, textures.back()->get(), imageSubresourceRange);
// Fill texture with data and prepare them for the ray tracing pipeline stages.
vkd.cmdPipelineBarrier(cmdBuffer, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0u, 0u, nullptr, 0u, nullptr, 1u, &texturePreCopyBarrier);
vkd.cmdCopyBufferToImage(cmdBuffer, textureData->get(), textures.back()->get(), VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1u, &bufferImageCopy);
vkd.cmdPipelineBarrier(cmdBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_RAY_TRACING_SHADER_BIT_KHR, 0u, 0u, nullptr, 0u, nullptr, 1u, &texturePostCopyBarrier);
}
else
{
DE_ASSERT(false);
}
}
// Descriptor set layout.
DescriptorSetLayoutBuilder dslBuilder;
dslBuilder.addBinding(VK_DESCRIPTOR_TYPE_ACCELERATION_STRUCTURE_KHR, 1u, shaderStages, nullptr);
dslBuilder.addBinding(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 1u, shaderStages, nullptr); // Callee buffer.
dslBuilder.addBinding(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 1u, shaderStages, nullptr); // Output buffer.
dslBuilder.addBinding(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 1u, shaderStages, nullptr); // Input buffer.
if (samplersNeeded(m_params.dataType))
{
dslBuilder.addBinding(VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE, 2u, shaderStages, nullptr);
dslBuilder.addBinding(VK_DESCRIPTOR_TYPE_SAMPLER, 2u, shaderStages, nullptr);
dslBuilder.addBinding(VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, 2u, shaderStages, nullptr);
}
else if (storageImageNeeded(m_params.dataType))
{
dslBuilder.addBinding(VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, 1u, shaderStages, nullptr);
}
const auto descriptorSetLayout = dslBuilder.build(vkd, device);
// Pipeline layout.
const auto pipelineLayout = makePipelineLayout(vkd, device, descriptorSetLayout.get());
// Descriptor pool and set.
DescriptorPoolBuilder poolBuilder;
poolBuilder.addType(VK_DESCRIPTOR_TYPE_ACCELERATION_STRUCTURE_KHR);
poolBuilder.addType(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 3u);
if (samplersNeeded(m_params.dataType))
{
poolBuilder.addType(VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE, 2u);
poolBuilder.addType(VK_DESCRIPTOR_TYPE_SAMPLER, 2u);
poolBuilder.addType(VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, 2u);
}
else if (storageImageNeeded(m_params.dataType))
{
poolBuilder.addType(VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, 1u);
}
const auto descriptorPool = poolBuilder.build(vkd, device, VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT, 1u);
const auto descriptorSet = makeDescriptorSet(vkd, device, descriptorPool.get(), descriptorSetLayout.get());
// Update descriptor set.
{
const VkWriteDescriptorSetAccelerationStructureKHR writeASInfo =
{
VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET_ACCELERATION_STRUCTURE_KHR,
nullptr,
1u,
topLevelAccelerationStructure.get()->getPtr(),
};
DescriptorSetUpdateBuilder updateBuilder;
const auto ds = descriptorSet.get();
const auto calleeBufferDescriptorInfo = makeDescriptorBufferInfo(calleeBuffer.get(), 0ull, VK_WHOLE_SIZE);
const auto outputBufferDescriptorInfo = makeDescriptorBufferInfo(outputBuffer.get(), 0ull, VK_WHOLE_SIZE);
const auto inputBufferDescriptorInfo = makeDescriptorBufferInfo(inputBuffer.get(), 0ull, VK_WHOLE_SIZE);
updateBuilder.writeSingle(ds, DescriptorSetUpdateBuilder::Location::binding(0u), VK_DESCRIPTOR_TYPE_ACCELERATION_STRUCTURE_KHR, &writeASInfo);
updateBuilder.writeSingle(ds, DescriptorSetUpdateBuilder::Location::binding(1u), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, &calleeBufferDescriptorInfo);
updateBuilder.writeSingle(ds, DescriptorSetUpdateBuilder::Location::binding(2u), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, &outputBufferDescriptorInfo);
updateBuilder.writeSingle(ds, DescriptorSetUpdateBuilder::Location::binding(3u), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, &inputBufferDescriptorInfo);
if (samplersNeeded(m_params.dataType))
{
// Update textures, samplers and combined image samplers.
std::vector<VkDescriptorImageInfo> textureDescInfos;
std::vector<VkDescriptorImageInfo> textureSamplerInfos;
std::vector<VkDescriptorImageInfo> combinedSamplerInfos;
for (size_t i = 0; i < kNumAloneImages; ++i)
textureDescInfos.push_back(makeDescriptorImageInfo(DE_NULL, textureViews[i].get(), VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL));
for (size_t i = 0; i < kNumAloneSamplers; ++i)
textureSamplerInfos.push_back(makeDescriptorImageInfo(samplers[i].get(), DE_NULL, VK_IMAGE_LAYOUT_UNDEFINED));
for (size_t i = 0; i < kNumCombined; ++i)
combinedSamplerInfos.push_back(makeDescriptorImageInfo(samplers[i + kNumAloneSamplers].get(), textureViews[i + kNumAloneImages].get(), VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL));
updateBuilder.writeArray(ds, DescriptorSetUpdateBuilder::Location::binding(4u), VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE, kNumAloneImages, textureDescInfos.data());
updateBuilder.writeArray(ds, DescriptorSetUpdateBuilder::Location::binding(5u), VK_DESCRIPTOR_TYPE_SAMPLER, kNumAloneSamplers, textureSamplerInfos.data());
updateBuilder.writeArray(ds, DescriptorSetUpdateBuilder::Location::binding(6u), VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, kNumCombined, combinedSamplerInfos.data());
}
else if (storageImageNeeded(m_params.dataType))
{
const auto storageImageDescriptorInfo = makeDescriptorImageInfo(DE_NULL, textureViews.back().get(), VK_IMAGE_LAYOUT_GENERAL);
updateBuilder.writeSingle(ds, DescriptorSetUpdateBuilder::Location::binding(4u), VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, &storageImageDescriptorInfo);
}
updateBuilder.update(vkd, device);
}
// Create raytracing pipeline and shader binding tables.
Move<VkPipeline> pipeline;
de::MovePtr<BufferWithMemory> raygenShaderBindingTable;
de::MovePtr<BufferWithMemory> missShaderBindingTable;
de::MovePtr<BufferWithMemory> hitShaderBindingTable;
de::MovePtr<BufferWithMemory> callableShaderBindingTable;
VkStridedDeviceAddressRegionKHR raygenShaderBindingTableRegion = makeStridedDeviceAddressRegionKHR(DE_NULL, 0, 0);
VkStridedDeviceAddressRegionKHR missShaderBindingTableRegion = makeStridedDeviceAddressRegionKHR(DE_NULL, 0, 0);
VkStridedDeviceAddressRegionKHR hitShaderBindingTableRegion = makeStridedDeviceAddressRegionKHR(DE_NULL, 0, 0);
VkStridedDeviceAddressRegionKHR callableShaderBindingTableRegion = makeStridedDeviceAddressRegionKHR(DE_NULL, 0, 0);
{
const auto rayTracingPipeline = de::newMovePtr<RayTracingPipeline>();
const auto callType = m_params.callType;
// Every case uses a ray generation shader.
rayTracingPipeline->addShader(VK_SHADER_STAGE_RAYGEN_BIT_KHR, createShaderModule(vkd, device, m_context.getBinaryCollection().get("rgen"), 0), 0);
if (callType == CallType::TRACE_RAY)
{
rayTracingPipeline->addShader(VK_SHADER_STAGE_CLOSEST_HIT_BIT_KHR, createShaderModule(vkd, device, m_context.getBinaryCollection().get("chit"), 0), 1);
}
else if (callType == CallType::EXECUTE_CALLABLE)
{
rayTracingPipeline->addShader(VK_SHADER_STAGE_CALLABLE_BIT_KHR, createShaderModule(vkd, device, m_context.getBinaryCollection().get("call"), 0), 1);
}
else if (callType == CallType::REPORT_INTERSECTION)
{
rayTracingPipeline->addShader(VK_SHADER_STAGE_INTERSECTION_BIT_KHR, createShaderModule(vkd, device, m_context.getBinaryCollection().get("rint"), 0), 1);
rayTracingPipeline->addShader(VK_SHADER_STAGE_ANY_HIT_BIT_KHR, createShaderModule(vkd, device, m_context.getBinaryCollection().get("ahit"), 0), 1);
}
else
{
DE_ASSERT(false);
}
pipeline = rayTracingPipeline->createPipeline(vkd, device, pipelineLayout.get());
raygenShaderBindingTable = rayTracingPipeline->createShaderBindingTable(vkd, device, pipeline.get(), alloc, shaderGroupHandleSize, shaderGroupBaseAlignment, 0, 1);
raygenShaderBindingTableRegion = makeStridedDeviceAddressRegionKHR(getBufferDeviceAddress(vkd, device, raygenShaderBindingTable->get(), 0), shaderGroupHandleSize, shaderGroupHandleSize);
if (callType == CallType::EXECUTE_CALLABLE)
{
callableShaderBindingTable = rayTracingPipeline->createShaderBindingTable(vkd, device, pipeline.get(), alloc, shaderGroupHandleSize, shaderGroupBaseAlignment, 1, 1);
callableShaderBindingTableRegion = makeStridedDeviceAddressRegionKHR(getBufferDeviceAddress(vkd, device, callableShaderBindingTable->get(), 0), shaderGroupHandleSize, shaderGroupHandleSize);
}
else if (callType == CallType::TRACE_RAY || callType == CallType::REPORT_INTERSECTION)
{
hitShaderBindingTable = rayTracingPipeline->createShaderBindingTable(vkd, device, pipeline.get(), alloc, shaderGroupHandleSize, shaderGroupBaseAlignment, 1, 1);
hitShaderBindingTableRegion = makeStridedDeviceAddressRegionKHR(getBufferDeviceAddress(vkd, device, hitShaderBindingTable->get(), 0), shaderGroupHandleSize, shaderGroupHandleSize);
}
else
{
DE_ASSERT(false);
}
}
// Use ray tracing pipeline.
vkd.cmdBindPipeline(cmdBuffer, VK_PIPELINE_BIND_POINT_RAY_TRACING_KHR, pipeline.get());
vkd.cmdBindDescriptorSets(cmdBuffer, VK_PIPELINE_BIND_POINT_RAY_TRACING_KHR, pipelineLayout.get(), 0u, 1u, &descriptorSet.get(), 0u, nullptr);
vkd.cmdTraceRaysKHR(cmdBuffer, &raygenShaderBindingTableRegion, &missShaderBindingTableRegion, &hitShaderBindingTableRegion, &callableShaderBindingTableRegion, 1u, 1u, 1u);
// Synchronize output and callee buffers.
const auto memBarrier = makeMemoryBarrier(VK_ACCESS_SHADER_WRITE_BIT, VK_ACCESS_HOST_READ_BIT);
vkd.cmdPipelineBarrier(cmdBuffer, VK_PIPELINE_STAGE_RAY_TRACING_SHADER_BIT_KHR, VK_PIPELINE_STAGE_HOST_BIT, 0u, 1u, &memBarrier, 0u, nullptr, 0u, nullptr);
endCommandBuffer(vkd, cmdBuffer);
submitCommandsAndWait(vkd, device, queue, cmdBuffer);
// Verify output and callee buffers.
invalidateAlloc(vkd, device, outputBufferAlloc);
invalidateAlloc(vkd, device, calleeBufferAlloc);
std::map<std::string, void*> bufferPtrs;
bufferPtrs["output"] = outputBufferPtr;
bufferPtrs["callee"] = calleeBufferPtr;
for (const auto& ptr : bufferPtrs)
{
const auto& bufferName = ptr.first;
const auto& bufferPtr = ptr.second;
deUint32 outputVal;
deMemcpy(&outputVal, bufferPtr, sizeof(outputVal));
if (outputVal != 1u)
return tcu::TestStatus::fail("Unexpected value found in " + bufferName + " buffer: " + de::toString(outputVal));
}
return tcu::TestStatus::pass("Pass");
}
enum class InterfaceType
{
RAY_PAYLOAD = 0,
CALLABLE_DATA,
HIT_ATTRIBUTES,
SHADER_RECORD_BUFFER_RGEN,
SHADER_RECORD_BUFFER_CALL,
SHADER_RECORD_BUFFER_MISS,
SHADER_RECORD_BUFFER_HIT,
};
// Separate class to ease testing pipeline interface variables.
class DataSpillPipelineInterfaceTestCase : public vkt::TestCase
{
public:
struct TestParams
{
InterfaceType interfaceType;
};
DataSpillPipelineInterfaceTestCase (tcu::TestContext& testCtx, const std::string& name, const std::string& description, const TestParams& testParams);
virtual ~DataSpillPipelineInterfaceTestCase (void) {}
virtual void initPrograms (vk::SourceCollections& programCollection) const;
virtual TestInstance* createInstance (Context& context) const;
virtual void checkSupport (Context& context) const;
private:
TestParams m_params;
};
class DataSpillPipelineInterfaceTestInstance : public vkt::TestInstance
{
public:
using TestParams = DataSpillPipelineInterfaceTestCase::TestParams;
DataSpillPipelineInterfaceTestInstance (Context& context, const TestParams& testParams);
~DataSpillPipelineInterfaceTestInstance (void) {}
tcu::TestStatus iterate (void);
private:
TestParams m_params;
};
DataSpillPipelineInterfaceTestCase::DataSpillPipelineInterfaceTestCase (tcu::TestContext& testCtx, const std::string& name, const std::string& description, const TestParams& testParams)
: vkt::TestCase (testCtx, name, description)
, m_params (testParams)
{
}
TestInstance* DataSpillPipelineInterfaceTestCase::createInstance (Context& context) const
{
return new DataSpillPipelineInterfaceTestInstance (context, m_params);
}
DataSpillPipelineInterfaceTestInstance::DataSpillPipelineInterfaceTestInstance (Context& context, const TestParams& testParams)
: vkt::TestInstance (context)
, m_params (testParams)
{
}
void DataSpillPipelineInterfaceTestCase::checkSupport (Context& context) const
{
commonCheckSupport(context);
}
void DataSpillPipelineInterfaceTestCase::initPrograms (vk::SourceCollections& programCollection) const
{
const vk::ShaderBuildOptions buildOptions (programCollection.usedVulkanVersion, vk::SPIRV_VERSION_1_4, 0u, true);
const std::string glslHeader =
"#version 460 core\n"
"#extension GL_EXT_ray_tracing : require\n"
;
const std::string glslBindings =
"layout(set = 0, binding = 0) uniform accelerationStructureEXT topLevelAS;\n"
"layout(set = 0, binding = 1) buffer StorageBlock { uint val[" + std::to_string(kNumStorageValues) + "]; } storageBuffer;\n"
;
if (m_params.interfaceType == InterfaceType::RAY_PAYLOAD)
{
// The closest hit shader will store 100 in the second array position.
// The ray gen shader will store 103 in the first array position using the hitValue after the traceRayExt() call.
std::ostringstream rgen;
rgen
<< glslHeader
<< "layout(location = 0) rayPayloadEXT vec3 hitValue;\n"
<< glslBindings
<< "void main()\n"
<< "{\n"
<< " hitValue = vec3(10.0, 30.0, 60.0);\n"
<< " traceRayEXT(topLevelAS, 0u, 0xFFu, 0, 0, 0, vec3(0.5, 0.5, 0.0), 0.0, vec3(0.0, 0.0, -1.0), 9.0, 0);\n"
<< " storageBuffer.val[0] = uint(hitValue.x + hitValue.y + hitValue.z);\n"
<< "}\n"
;
programCollection.glslSources.add("rgen") << glu::RaygenSource(updateRayTracingGLSL(rgen.str())) << buildOptions;
std::stringstream chit;
chit
<< glslHeader
<< "layout(location = 0) rayPayloadInEXT vec3 hitValue;\n"
<< "hitAttributeEXT vec3 attribs;\n"
<< glslBindings
<< "void main()\n"
<< "{\n"
<< " storageBuffer.val[1] = uint(hitValue.x + hitValue.y + hitValue.z);\n"
<< " hitValue = vec3(hitValue.x + 1.0, hitValue.y + 1.0, hitValue.z + 1.0);\n"
<< "}\n";
;
programCollection.glslSources.add("chit") << glu::ClosestHitSource(updateRayTracingGLSL(chit.str())) << buildOptions;
}
else if (m_params.interfaceType == InterfaceType::CALLABLE_DATA)
{
// The callable shader shader will store 100 in the second array position.
// The ray gen shader will store 200 in the first array position using the callable data after the executeCallableEXT() call.
std::ostringstream rgen;
rgen
<< glslHeader
<< "layout(location = 0) callableDataEXT float callableData;\n"
<< glslBindings
<< "void main()\n"
<< "{\n"
<< " callableData = 100.0;\n"
<< " executeCallableEXT(0, 0);\n"
<< " storageBuffer.val[0] = uint(callableData);\n"
<< "}\n"
;
programCollection.glslSources.add("rgen") << glu::RaygenSource(updateRayTracingGLSL(rgen.str())) << buildOptions;
std::ostringstream call;
call
<< glslHeader
<< "layout(location = 0) callableDataInEXT float callableData;\n"
<< glslBindings
<< "void main()\n"
<< "{\n"
<< " storageBuffer.val[1] = uint(callableData);\n"
<< " callableData = callableData * 2.0;\n"
<< "}\n"
;
programCollection.glslSources.add("call") << glu::CallableSource(updateRayTracingGLSL(call.str())) << buildOptions;
}
else if (m_params.interfaceType == InterfaceType::HIT_ATTRIBUTES)
{
// The ray gen shader will store value 300 in the first storage buffer position.
// The intersection shader will store value 315 in the second storage buffer position.
// The closes hit shader will store value 330 in the third storage buffer position using the hit attributes.
std::ostringstream rgen;
rgen
<< glslHeader
<< "layout(location = 0) rayPayloadEXT vec3 hitValue;\n"
<< glslBindings
<< "void main()\n"
<< "{\n"
<< " traceRayEXT(topLevelAS, 0u, 0xFFu, 0, 0, 0, vec3(0.5, 0.5, 0.0), 0.0, vec3(0.0, 0.0, -1.0), 9.0, 0);\n"
<< " storageBuffer.val[0] = 300u;\n"
<< "}\n"
;
programCollection.glslSources.add("rgen") << glu::RaygenSource(updateRayTracingGLSL(rgen.str())) << buildOptions;
std::stringstream rint;
rint
<< glslHeader
<< "hitAttributeEXT vec3 attribs;\n"
<< glslBindings
<< "void main()\n"
<< "{\n"
<< " attribs = vec3(140.0, 160.0, 30.0);\n"
<< " storageBuffer.val[1] = 315u;\n"
<< " reportIntersectionEXT(1.0f, 0);\n"
<< "}\n"
;
programCollection.glslSources.add("rint") << glu::IntersectionSource(updateRayTracingGLSL(rint.str())) << buildOptions;
std::stringstream chit;
chit
<< glslHeader
<< "layout(location = 0) rayPayloadInEXT vec3 hitValue;\n"
<< "hitAttributeEXT vec3 attribs;\n"
<< glslBindings
<< "void main()\n"
<< "{\n"
<< " storageBuffer.val[2] = uint(attribs.x + attribs.y + attribs.z);\n"
<< "}\n";
;
programCollection.glslSources.add("chit") << glu::ClosestHitSource(updateRayTracingGLSL(chit.str())) << buildOptions;
}
else if (m_params.interfaceType == InterfaceType::SHADER_RECORD_BUFFER_RGEN)
{
// The ray gen shader will have a uvec4 in the shader record buffer with contents 400, 401, 402, 403.
// The shader will call a callable shader indicating a position in that vec4 (0, 1, 2, 3). For example, let's use position 1.
// The callable shader will return the indicated position+1 modulo 4, so it will return 2 in our case.
// *After* returning from the callable shader, the raygen shader will use that reply to access position 2 and write a 402 in the first output buffer position.
// The callable shader will store 450 in the second output buffer position.
std::ostringstream rgen;
rgen
<< glslHeader
<< "layout(shaderRecordEXT) buffer ShaderRecordStruct {\n"
<< " uvec4 info;\n"
<< "};\n"
<< "layout(location = 0) callableDataEXT uint callableData;\n"
<< glslBindings
<< "void main()\n"
<< "{\n"
<< " callableData = 1u;"
<< " executeCallableEXT(0, 0);\n"
<< " if (callableData == 0u) storageBuffer.val[0] = info.x;\n"
<< " else if (callableData == 1u) storageBuffer.val[0] = info.y;\n"
<< " else if (callableData == 2u) storageBuffer.val[0] = info.z;\n"
<< " else if (callableData == 3u) storageBuffer.val[0] = info.w;\n"
<< "}\n"
;
programCollection.glslSources.add("rgen") << glu::RaygenSource(updateRayTracingGLSL(rgen.str())) << buildOptions;
std::ostringstream call;
call
<< glslHeader
<< "layout(location = 0) callableDataInEXT uint callableData;\n"
<< glslBindings
<< "void main()\n"
<< "{\n"
<< " storageBuffer.val[1] = 450u;\n"
<< " callableData = (callableData + 1u) % 4u;\n"
<< "}\n"
;
programCollection.glslSources.add("call") << glu::CallableSource(updateRayTracingGLSL(call.str())) << buildOptions;
}
else if (m_params.interfaceType == InterfaceType::SHADER_RECORD_BUFFER_CALL)
{
// Similar to the previous case, with a twist:
// * rgen passes the vector position.
// * call increases that by one.
// * subcall increases again and does the modulo operation, also writing 450 in the third output buffer value.
// * call is the one accessing the vector at the returned position, writing 403 in this case to the second output buffer value.
// * call passes this value back doubled to rgen, which writes it to the first output buffer value (806).
std::ostringstream rgen;
rgen
<< glslHeader
<< "layout(location = 0) callableDataEXT uint callableData;\n"
<< glslBindings
<< "void main()\n"
<< "{\n"
<< " callableData = 1u;\n"
<< " executeCallableEXT(0, 0);\n"
<< " storageBuffer.val[0] = callableData;\n"
<< "}\n"
;
programCollection.glslSources.add("rgen") << glu::RaygenSource(updateRayTracingGLSL(rgen.str())) << buildOptions;
std::ostringstream call;
call
<< glslHeader
<< "layout(shaderRecordEXT) buffer ShaderRecordStruct {\n"
<< " uvec4 info;\n"
<< "};\n"
<< "layout(location = 0) callableDataInEXT uint callableDataIn;\n"
<< "layout(location = 1) callableDataEXT uint callableDataOut;\n"
<< glslBindings
<< "void main()\n"
<< "{\n"
<< " callableDataOut = callableDataIn + 1u;\n"
<< " executeCallableEXT(1, 1);\n"
<< " uint outputBufferValue = 777u;\n"
<< " if (callableDataOut == 0u) outputBufferValue = info.x;\n"
<< " else if (callableDataOut == 1u) outputBufferValue = info.y;\n"
<< " else if (callableDataOut == 2u) outputBufferValue = info.z;\n"
<< " else if (callableDataOut == 3u) outputBufferValue = info.w;\n"
<< " storageBuffer.val[1] = outputBufferValue;\n"
<< " callableDataIn = outputBufferValue * 2u;\n"
<< "}\n"
;
programCollection.glslSources.add("call") << glu::CallableSource(updateRayTracingGLSL(call.str())) << buildOptions;
std::ostringstream subcall;
subcall
<< glslHeader
<< "layout(location = 1) callableDataInEXT uint callableData;\n"
<< glslBindings
<< "void main()\n"
<< "{\n"
<< " callableData = (callableData + 1u) % 4u;\n"
<< " storageBuffer.val[2] = 450u;\n"
<< "}\n"
;
programCollection.glslSources.add("subcall") << glu::CallableSource(updateRayTracingGLSL(subcall.str())) << buildOptions;
}
else if (m_params.interfaceType == InterfaceType::SHADER_RECORD_BUFFER_MISS || m_params.interfaceType == InterfaceType::SHADER_RECORD_BUFFER_HIT)
{
// Similar to the previous one, but the intermediate call shader has been replaced with a miss or closest hit shader.
// The rgen shader will communicate with the miss/chit shader using the ray payload instead of the callable data.
// Also, the initial position will be 2, so it will wrap around in this case. The numbers will also change.
std::ostringstream rgen;
rgen
<< glslHeader
<< "layout(location = 0) rayPayloadEXT uint rayPayload;\n"
<< glslBindings
<< "void main()\n"
<< "{\n"
<< " rayPayload = 2u;\n"
<< " traceRayEXT(topLevelAS, 0u, 0xFFu, 0, 0, 0, vec3(0.5, 0.5, 0.0), 0.0, vec3(0.0, 0.0, -1.0), 9.0, 0);\n"
<< " storageBuffer.val[0] = rayPayload;\n"
<< "}\n"
;
programCollection.glslSources.add("rgen") << glu::RaygenSource(updateRayTracingGLSL(rgen.str())) << buildOptions;
std::ostringstream chitOrMiss;
chitOrMiss
<< glslHeader
<< "layout(shaderRecordEXT) buffer ShaderRecordStruct {\n"
<< " uvec4 info;\n"
<< "};\n"
<< "layout(location = 0) rayPayloadInEXT uint rayPayload;\n"
<< "layout(location = 0) callableDataEXT uint callableData;\n"
<< glslBindings
<< "void main()\n"
<< "{\n"
<< " callableData = rayPayload + 1u;\n"
<< " executeCallableEXT(0, 0);\n"
<< " uint outputBufferValue = 777u;\n"
<< " if (callableData == 0u) outputBufferValue = info.x;\n"
<< " else if (callableData == 1u) outputBufferValue = info.y;\n"
<< " else if (callableData == 2u) outputBufferValue = info.z;\n"
<< " else if (callableData == 3u) outputBufferValue = info.w;\n"
<< " storageBuffer.val[1] = outputBufferValue;\n"
<< " rayPayload = outputBufferValue * 3u;\n"
<< "}\n"
;
if (m_params.interfaceType == InterfaceType::SHADER_RECORD_BUFFER_MISS)
programCollection.glslSources.add("miss") << glu::MissSource(updateRayTracingGLSL(chitOrMiss.str())) << buildOptions;
else if (m_params.interfaceType == InterfaceType::SHADER_RECORD_BUFFER_HIT)
programCollection.glslSources.add("chit") << glu::ClosestHitSource(updateRayTracingGLSL(chitOrMiss.str())) << buildOptions;
else
DE_ASSERT(false);
std::ostringstream call;
call
<< glslHeader
<< "layout(location = 0) callableDataInEXT uint callableData;\n"
<< glslBindings
<< "void main()\n"
<< "{\n"
<< " storageBuffer.val[2] = 490u;\n"
<< " callableData = (callableData + 1u) % 4u;\n"
<< "}\n"
;
programCollection.glslSources.add("call") << glu::CallableSource(updateRayTracingGLSL(call.str())) << buildOptions;
}
else
{
DE_ASSERT(false);
}
}
VkShaderStageFlags getShaderStages (InterfaceType type_)
{
VkShaderStageFlags flags = VK_SHADER_STAGE_RAYGEN_BIT_KHR;
switch (type_)
{
case InterfaceType::HIT_ATTRIBUTES:
flags |= VK_SHADER_STAGE_INTERSECTION_BIT_KHR;
// fallthrough.
case InterfaceType::RAY_PAYLOAD:
flags |= VK_SHADER_STAGE_CLOSEST_HIT_BIT_KHR;
break;
case InterfaceType::CALLABLE_DATA:
case InterfaceType::SHADER_RECORD_BUFFER_RGEN:
case InterfaceType::SHADER_RECORD_BUFFER_CALL:
flags |= VK_SHADER_STAGE_CALLABLE_BIT_KHR;
break;
case InterfaceType::SHADER_RECORD_BUFFER_MISS:
flags |= VK_SHADER_STAGE_CALLABLE_BIT_KHR;
flags |= VK_SHADER_STAGE_MISS_BIT_KHR;
break;
case InterfaceType::SHADER_RECORD_BUFFER_HIT:
flags |= VK_SHADER_STAGE_CALLABLE_BIT_KHR;
flags |= VK_SHADER_STAGE_CLOSEST_HIT_BIT_KHR;
break;
default:
DE_ASSERT(false);
break;
}
return flags;
}
// Proper stage for generating default geometry.
VkShaderStageFlagBits getShaderStageForGeometry (InterfaceType type_)
{
VkShaderStageFlagBits bits = VK_SHADER_STAGE_FLAG_BITS_MAX_ENUM;
switch (type_)
{
case InterfaceType::HIT_ATTRIBUTES: bits = VK_SHADER_STAGE_INTERSECTION_BIT_KHR; break;
case InterfaceType::RAY_PAYLOAD: bits = VK_SHADER_STAGE_CLOSEST_HIT_BIT_KHR; break;
case InterfaceType::CALLABLE_DATA: bits = VK_SHADER_STAGE_CALLABLE_BIT_KHR; break;
case InterfaceType::SHADER_RECORD_BUFFER_RGEN: bits = VK_SHADER_STAGE_CALLABLE_BIT_KHR; break;
case InterfaceType::SHADER_RECORD_BUFFER_CALL: bits = VK_SHADER_STAGE_CALLABLE_BIT_KHR; break;
case InterfaceType::SHADER_RECORD_BUFFER_MISS: bits = VK_SHADER_STAGE_MISS_BIT_KHR; break;
case InterfaceType::SHADER_RECORD_BUFFER_HIT: bits = VK_SHADER_STAGE_CLOSEST_HIT_BIT_KHR; break;
default: DE_ASSERT(false); break;
}
DE_ASSERT(bits != VK_SHADER_STAGE_FLAG_BITS_MAX_ENUM);
return bits;
}
void createSBTWithShaderRecord (const DeviceInterface& vkd, VkDevice device, vk::Allocator &alloc,
VkPipeline pipeline, RayTracingPipeline* rayTracingPipeline,
deUint32 shaderGroupHandleSize, deUint32 shaderGroupBaseAlignment,
deUint32 firstGroup, deUint32 groupCount,
de::MovePtr<BufferWithMemory>& shaderBindingTable,
VkStridedDeviceAddressRegionKHR& shaderBindingTableRegion)
{
const auto alignedSize = de::roundUp(shaderGroupHandleSize + kShaderRecordSize, shaderGroupHandleSize);
shaderBindingTable = rayTracingPipeline->createShaderBindingTable(vkd, device, pipeline, alloc, shaderGroupHandleSize, shaderGroupBaseAlignment, firstGroup, groupCount, 0u, 0u, MemoryRequirement::Any, 0u, 0u, kShaderRecordSize);
shaderBindingTableRegion = makeStridedDeviceAddressRegionKHR(getBufferDeviceAddress(vkd, device, shaderBindingTable->get(), 0), alignedSize, groupCount * alignedSize);
// Fill shader record buffer data.
// Note we will only fill the first shader record after the handle.
const tcu::UVec4 shaderRecordData (400u, 401u, 402u, 403u);
auto& sbtAlloc = shaderBindingTable->getAllocation();
auto* dataPtr = reinterpret_cast<deUint8*>(sbtAlloc.getHostPtr()) + shaderGroupHandleSize;
DE_STATIC_ASSERT(sizeof(shaderRecordData) == static_cast<size_t>(kShaderRecordSize));
deMemcpy(dataPtr, &shaderRecordData, sizeof(shaderRecordData));
}
tcu::TestStatus DataSpillPipelineInterfaceTestInstance::iterate (void)
{
const auto& vki = m_context.getInstanceInterface();
const auto physicalDevice = m_context.getPhysicalDevice();
const auto& vkd = m_context.getDeviceInterface();
const auto device = m_context.getDevice();
const auto queue = m_context.getUniversalQueue();
const auto familyIndex = m_context.getUniversalQueueFamilyIndex();
auto& alloc = m_context.getDefaultAllocator();
const auto shaderStages = getShaderStages(m_params.interfaceType);
// Command buffer.
const auto cmdPool = makeCommandPool(vkd, device, familyIndex);
const auto cmdBufferPtr = allocateCommandBuffer(vkd, device, cmdPool.get(), VK_COMMAND_BUFFER_LEVEL_PRIMARY);
const auto cmdBuffer = cmdBufferPtr.get();
beginCommandBuffer(vkd, cmdBuffer);
// Storage buffer.
std::array<deUint32, kNumStorageValues> storageBufferData;
const auto storageBufferSize = de::dataSize(storageBufferData);
const auto storagebufferInfo = makeBufferCreateInfo(storageBufferSize, VK_BUFFER_USAGE_STORAGE_BUFFER_BIT);
BufferWithMemory storageBuffer (vkd, device, alloc, storagebufferInfo, MemoryRequirement::HostVisible);
// Zero-out buffer.
auto& storageBufferAlloc = storageBuffer.getAllocation();
auto* storageBufferPtr = storageBufferAlloc.getHostPtr();
deMemset(storageBufferPtr, 0, storageBufferSize);
flushAlloc(vkd, device, storageBufferAlloc);
// Acceleration structures.
de::MovePtr<BottomLevelAccelerationStructure> bottomLevelAccelerationStructure;
de::MovePtr<TopLevelAccelerationStructure> topLevelAccelerationStructure;
bottomLevelAccelerationStructure = makeBottomLevelAccelerationStructure();
bottomLevelAccelerationStructure->setDefaultGeometryData(getShaderStageForGeometry(m_params.interfaceType), VK_GEOMETRY_NO_DUPLICATE_ANY_HIT_INVOCATION_BIT_KHR);
bottomLevelAccelerationStructure->createAndBuild(vkd, device, cmdBuffer, alloc);
topLevelAccelerationStructure = makeTopLevelAccelerationStructure();
topLevelAccelerationStructure->setInstanceCount(1);
topLevelAccelerationStructure->addInstance(de::SharedPtr<BottomLevelAccelerationStructure>(bottomLevelAccelerationStructure.release()));
topLevelAccelerationStructure->createAndBuild(vkd, device, cmdBuffer, alloc);
// Get some ray tracing properties.
deUint32 shaderGroupHandleSize = 0u;
deUint32 shaderGroupBaseAlignment = 1u;
{
const auto rayTracingPropertiesKHR = makeRayTracingProperties(vki, physicalDevice);
shaderGroupHandleSize = rayTracingPropertiesKHR->getShaderGroupHandleSize();
shaderGroupBaseAlignment = rayTracingPropertiesKHR->getShaderGroupBaseAlignment();
}
// Descriptor set layout.
DescriptorSetLayoutBuilder dslBuilder;
dslBuilder.addBinding(VK_DESCRIPTOR_TYPE_ACCELERATION_STRUCTURE_KHR, 1u, shaderStages, nullptr);
dslBuilder.addBinding(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 1u, shaderStages, nullptr); // Callee buffer.
const auto descriptorSetLayout = dslBuilder.build(vkd, device);
// Pipeline layout.
const auto pipelineLayout = makePipelineLayout(vkd, device, descriptorSetLayout.get());
// Descriptor pool and set.
DescriptorPoolBuilder poolBuilder;
poolBuilder.addType(VK_DESCRIPTOR_TYPE_ACCELERATION_STRUCTURE_KHR);
poolBuilder.addType(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER);
const auto descriptorPool = poolBuilder.build(vkd, device, VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT, 1u);
const auto descriptorSet = makeDescriptorSet(vkd, device, descriptorPool.get(), descriptorSetLayout.get());
// Update descriptor set.
{
const VkWriteDescriptorSetAccelerationStructureKHR writeASInfo =
{
VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET_ACCELERATION_STRUCTURE_KHR,
nullptr,
1u,
topLevelAccelerationStructure.get()->getPtr(),
};
const auto ds = descriptorSet.get();
const auto storageBufferDescriptorInfo = makeDescriptorBufferInfo(storageBuffer.get(), 0ull, VK_WHOLE_SIZE);
DescriptorSetUpdateBuilder updateBuilder;
updateBuilder.writeSingle(ds, DescriptorSetUpdateBuilder::Location::binding(0u), VK_DESCRIPTOR_TYPE_ACCELERATION_STRUCTURE_KHR, &writeASInfo);
updateBuilder.writeSingle(ds, DescriptorSetUpdateBuilder::Location::binding(1u), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, &storageBufferDescriptorInfo);
updateBuilder.update(vkd, device);
}
// Create raytracing pipeline and shader binding tables.
const auto interfaceType = m_params.interfaceType;
Move<VkPipeline> pipeline;
de::MovePtr<BufferWithMemory> raygenShaderBindingTable;
de::MovePtr<BufferWithMemory> missShaderBindingTable;
de::MovePtr<BufferWithMemory> hitShaderBindingTable;
de::MovePtr<BufferWithMemory> callableShaderBindingTable;
VkStridedDeviceAddressRegionKHR raygenShaderBindingTableRegion = makeStridedDeviceAddressRegionKHR(DE_NULL, 0, 0);
VkStridedDeviceAddressRegionKHR missShaderBindingTableRegion = makeStridedDeviceAddressRegionKHR(DE_NULL, 0, 0);
VkStridedDeviceAddressRegionKHR hitShaderBindingTableRegion = makeStridedDeviceAddressRegionKHR(DE_NULL, 0, 0);
VkStridedDeviceAddressRegionKHR callableShaderBindingTableRegion = makeStridedDeviceAddressRegionKHR(DE_NULL, 0, 0);
{
const auto rayTracingPipeline = de::newMovePtr<RayTracingPipeline>();
// Every case uses a ray generation shader.
rayTracingPipeline->addShader(VK_SHADER_STAGE_RAYGEN_BIT_KHR, createShaderModule(vkd, device, m_context.getBinaryCollection().get("rgen"), 0), 0);
if (interfaceType == InterfaceType::RAY_PAYLOAD)
{
rayTracingPipeline->addShader(VK_SHADER_STAGE_CLOSEST_HIT_BIT_KHR, createShaderModule(vkd, device, m_context.getBinaryCollection().get("chit"), 0), 1);
}
else if (interfaceType == InterfaceType::CALLABLE_DATA || interfaceType == InterfaceType::SHADER_RECORD_BUFFER_RGEN)
{
rayTracingPipeline->addShader(VK_SHADER_STAGE_CALLABLE_BIT_KHR, createShaderModule(vkd, device, m_context.getBinaryCollection().get("call"), 0), 1);
}
else if (interfaceType == InterfaceType::HIT_ATTRIBUTES)
{
rayTracingPipeline->addShader(VK_SHADER_STAGE_INTERSECTION_BIT_KHR, createShaderModule(vkd, device, m_context.getBinaryCollection().get("rint"), 0), 1);
rayTracingPipeline->addShader(VK_SHADER_STAGE_CLOSEST_HIT_BIT_KHR, createShaderModule(vkd, device, m_context.getBinaryCollection().get("chit"), 0), 1);
}
else if (interfaceType == InterfaceType::SHADER_RECORD_BUFFER_CALL)
{
rayTracingPipeline->addShader(VK_SHADER_STAGE_CALLABLE_BIT_KHR, createShaderModule(vkd, device, m_context.getBinaryCollection().get("call"), 0), 1);
rayTracingPipeline->addShader(VK_SHADER_STAGE_CALLABLE_BIT_KHR, createShaderModule(vkd, device, m_context.getBinaryCollection().get("subcall"), 0), 2);
}
else if (interfaceType == InterfaceType::SHADER_RECORD_BUFFER_MISS)
{
rayTracingPipeline->addShader(VK_SHADER_STAGE_MISS_BIT_KHR, createShaderModule(vkd, device, m_context.getBinaryCollection().get("miss"), 0), 1);
rayTracingPipeline->addShader(VK_SHADER_STAGE_CALLABLE_BIT_KHR, createShaderModule(vkd, device, m_context.getBinaryCollection().get("call"), 0), 2);
}
else if (interfaceType == InterfaceType::SHADER_RECORD_BUFFER_HIT)
{
rayTracingPipeline->addShader(VK_SHADER_STAGE_CLOSEST_HIT_BIT_KHR, createShaderModule(vkd, device, m_context.getBinaryCollection().get("chit"), 0), 1);
rayTracingPipeline->addShader(VK_SHADER_STAGE_CALLABLE_BIT_KHR, createShaderModule(vkd, device, m_context.getBinaryCollection().get("call"), 0), 2);
}
else
{
DE_ASSERT(false);
}
pipeline = rayTracingPipeline->createPipeline(vkd, device, pipelineLayout.get());
if (interfaceType == InterfaceType::SHADER_RECORD_BUFFER_RGEN)
{
createSBTWithShaderRecord (vkd, device, alloc, pipeline.get(), rayTracingPipeline.get(), shaderGroupHandleSize, shaderGroupBaseAlignment,
0u, 1u, raygenShaderBindingTable, raygenShaderBindingTableRegion);
}
else
{
raygenShaderBindingTable = rayTracingPipeline->createShaderBindingTable(vkd, device, pipeline.get(), alloc, shaderGroupHandleSize, shaderGroupBaseAlignment, 0, 1);
raygenShaderBindingTableRegion = makeStridedDeviceAddressRegionKHR(getBufferDeviceAddress(vkd, device, raygenShaderBindingTable->get(), 0), shaderGroupHandleSize, shaderGroupHandleSize);
}
if (interfaceType == InterfaceType::CALLABLE_DATA || interfaceType == InterfaceType::SHADER_RECORD_BUFFER_RGEN)
{
callableShaderBindingTable = rayTracingPipeline->createShaderBindingTable(vkd, device, pipeline.get(), alloc, shaderGroupHandleSize, shaderGroupBaseAlignment, 1, 1);
callableShaderBindingTableRegion = makeStridedDeviceAddressRegionKHR(getBufferDeviceAddress(vkd, device, callableShaderBindingTable->get(), 0), shaderGroupHandleSize, shaderGroupHandleSize);
}
else if (interfaceType == InterfaceType::RAY_PAYLOAD || interfaceType == InterfaceType::HIT_ATTRIBUTES)
{
hitShaderBindingTable = rayTracingPipeline->createShaderBindingTable(vkd, device, pipeline.get(), alloc, shaderGroupHandleSize, shaderGroupBaseAlignment, 1, 1);
hitShaderBindingTableRegion = makeStridedDeviceAddressRegionKHR(getBufferDeviceAddress(vkd, device, hitShaderBindingTable->get(), 0), shaderGroupHandleSize, shaderGroupHandleSize);
}
else if (interfaceType == InterfaceType::SHADER_RECORD_BUFFER_CALL)
{
createSBTWithShaderRecord (vkd, device, alloc, pipeline.get(), rayTracingPipeline.get(), shaderGroupHandleSize, shaderGroupBaseAlignment,
1u, 2u, callableShaderBindingTable, callableShaderBindingTableRegion);
}
else if (interfaceType == InterfaceType::SHADER_RECORD_BUFFER_MISS)
{
createSBTWithShaderRecord (vkd, device, alloc, pipeline.get(), rayTracingPipeline.get(), shaderGroupHandleSize, shaderGroupBaseAlignment,
1u, 1u, missShaderBindingTable, missShaderBindingTableRegion);
// Callable shader table.
callableShaderBindingTable = rayTracingPipeline->createShaderBindingTable(vkd, device, pipeline.get(), alloc, shaderGroupHandleSize, shaderGroupBaseAlignment, 2, 1);
callableShaderBindingTableRegion = makeStridedDeviceAddressRegionKHR(getBufferDeviceAddress(vkd, device, callableShaderBindingTable->get(), 0), shaderGroupHandleSize, shaderGroupHandleSize);
}
else if (interfaceType == InterfaceType::SHADER_RECORD_BUFFER_HIT)
{
createSBTWithShaderRecord (vkd, device, alloc, pipeline.get(), rayTracingPipeline.get(), shaderGroupHandleSize, shaderGroupBaseAlignment,
1u, 1u, hitShaderBindingTable, hitShaderBindingTableRegion);
// Callable shader table.
callableShaderBindingTable = rayTracingPipeline->createShaderBindingTable(vkd, device, pipeline.get(), alloc, shaderGroupHandleSize, shaderGroupBaseAlignment, 2, 1);
callableShaderBindingTableRegion = makeStridedDeviceAddressRegionKHR(getBufferDeviceAddress(vkd, device, callableShaderBindingTable->get(), 0), shaderGroupHandleSize, shaderGroupHandleSize);
}
else
{
DE_ASSERT(false);
}
}
// Use ray tracing pipeline.
vkd.cmdBindPipeline(cmdBuffer, VK_PIPELINE_BIND_POINT_RAY_TRACING_KHR, pipeline.get());
vkd.cmdBindDescriptorSets(cmdBuffer, VK_PIPELINE_BIND_POINT_RAY_TRACING_KHR, pipelineLayout.get(), 0u, 1u, &descriptorSet.get(), 0u, nullptr);
vkd.cmdTraceRaysKHR(cmdBuffer, &raygenShaderBindingTableRegion, &missShaderBindingTableRegion, &hitShaderBindingTableRegion, &callableShaderBindingTableRegion, 1u, 1u, 1u);
// Synchronize output and callee buffers.
const auto memBarrier = makeMemoryBarrier(VK_ACCESS_SHADER_WRITE_BIT, VK_ACCESS_HOST_READ_BIT);
vkd.cmdPipelineBarrier(cmdBuffer, VK_PIPELINE_STAGE_RAY_TRACING_SHADER_BIT_KHR, VK_PIPELINE_STAGE_HOST_BIT, 0u, 1u, &memBarrier, 0u, nullptr, 0u, nullptr);
endCommandBuffer(vkd, cmdBuffer);
submitCommandsAndWait(vkd, device, queue, cmdBuffer);
// Verify storage buffer.
invalidateAlloc(vkd, device, storageBufferAlloc);
deMemcpy(storageBufferData.data(), storageBufferPtr, storageBufferSize);
// These values must match what the shaders store.
std::vector<deUint32> expectedData;
if (interfaceType == InterfaceType::RAY_PAYLOAD)
{
expectedData.push_back(103u);
expectedData.push_back(100u);
}
else if (interfaceType == InterfaceType::CALLABLE_DATA)
{
expectedData.push_back(200u);
expectedData.push_back(100u);
}
else if (interfaceType == InterfaceType::HIT_ATTRIBUTES)
{
expectedData.push_back(300u);
expectedData.push_back(315u);
expectedData.push_back(330u);
}
else if (interfaceType == InterfaceType::SHADER_RECORD_BUFFER_RGEN)
{
expectedData.push_back(402u);
expectedData.push_back(450u);
}
else if (interfaceType == InterfaceType::SHADER_RECORD_BUFFER_CALL)
{
expectedData.push_back(806u);
expectedData.push_back(403u);
expectedData.push_back(450u);
}
else if (interfaceType == InterfaceType::SHADER_RECORD_BUFFER_MISS || interfaceType == InterfaceType::SHADER_RECORD_BUFFER_HIT)
{
expectedData.push_back(1200u);
expectedData.push_back( 400u);
expectedData.push_back( 490u);
}
else
{
DE_ASSERT(false);
}
size_t pos;
for (pos = 0u; pos < expectedData.size(); ++pos)
{
const auto& stored = storageBufferData.at(pos);
const auto& expected = expectedData.at(pos);
if (stored != expected)
{
std::ostringstream msg;
msg << "Unexpected output value found at position " << pos << " (expected " << expected << " but got " << stored << ")";
return tcu::TestStatus::fail(msg.str());
}
}
// Expect zeros in unused positions, as filled on the host.
for (; pos < storageBufferData.size(); ++pos)
{
const auto& stored = storageBufferData.at(pos);
if (stored != 0u)
{
std::ostringstream msg;
msg << "Unexpected output value found at position " << pos << " (expected 0 but got " << stored << ")";
return tcu::TestStatus::fail(msg.str());
}
}
return tcu::TestStatus::pass("Pass");
}
} // anonymous namespace
tcu::TestCaseGroup* createDataSpillTests(tcu::TestContext& testCtx)
{
de::MovePtr<tcu::TestCaseGroup> group(new tcu::TestCaseGroup(testCtx, "data_spill", "Ray tracing tests for data spilling and unspilling around shader calls"));
struct
{
CallType callType;
const char* name;
} callTypes[] =
{
{ CallType::EXECUTE_CALLABLE, "execute_callable" },
{ CallType::TRACE_RAY, "trace_ray" },
{ CallType::REPORT_INTERSECTION, "report_intersection" },
};
struct
{
DataType dataType;
const char* name;
} dataTypes[] =
{
{ DataType::INT32, "int32" },
{ DataType::UINT32, "uint32" },
{ DataType::INT64, "int64" },
{ DataType::UINT64, "uint64" },
{ DataType::INT16, "int16" },
{ DataType::UINT16, "uint16" },
{ DataType::INT8, "int8" },
{ DataType::UINT8, "uint8" },
{ DataType::FLOAT32, "float32" },
{ DataType::FLOAT64, "float64" },
{ DataType::FLOAT16, "float16" },
{ DataType::STRUCT, "struct" },
{ DataType::SAMPLER, "sampler" },
{ DataType::IMAGE, "image" },
{ DataType::SAMPLED_IMAGE, "combined" },
{ DataType::PTR_IMAGE, "ptr_image" },
{ DataType::PTR_SAMPLER, "ptr_sampler" },
{ DataType::PTR_SAMPLED_IMAGE, "ptr_combined" },
{ DataType::PTR_TEXEL, "ptr_texel" },
{ DataType::OP_NULL, "op_null" },
{ DataType::OP_UNDEF, "op_undef" },
};
struct
{
VectorType vectorType;
const char* prefix;
} vectorTypes[] =
{
{ VectorType::SCALAR, "" },
{ VectorType::V2, "v2" },
{ VectorType::V3, "v3" },
{ VectorType::V4, "v4" },
{ VectorType::A5, "a5" },
};
for (int callTypeIdx = 0; callTypeIdx < DE_LENGTH_OF_ARRAY(callTypes); ++callTypeIdx)
{
const auto& entryCallTypes = callTypes[callTypeIdx];
de::MovePtr<tcu::TestCaseGroup> callTypeGroup(new tcu::TestCaseGroup(testCtx, entryCallTypes.name, ""));
for (int dataTypeIdx = 0; dataTypeIdx < DE_LENGTH_OF_ARRAY(dataTypes); ++dataTypeIdx)
{
const auto& entryDataTypes = dataTypes[dataTypeIdx];
for (int vectorTypeIdx = 0; vectorTypeIdx < DE_LENGTH_OF_ARRAY(vectorTypes); ++vectorTypeIdx)
{
const auto& entryVectorTypes = vectorTypes[vectorTypeIdx];
if ((samplersNeeded(entryDataTypes.dataType)
|| storageImageNeeded(entryDataTypes.dataType)
|| entryDataTypes.dataType == DataType::STRUCT
|| entryDataTypes.dataType == DataType::OP_NULL
|| entryDataTypes.dataType == DataType::OP_UNDEF)
&& entryVectorTypes.vectorType != VectorType::SCALAR)
{
continue;
}
DataSpillTestCase::TestParams params;
params.callType = entryCallTypes.callType;
params.dataType = entryDataTypes.dataType;
params.vectorType = entryVectorTypes.vectorType;
const auto testName = std::string(entryVectorTypes.prefix) + entryDataTypes.name;
callTypeGroup->addChild(new DataSpillTestCase(testCtx, testName, "", params));
}
}
group->addChild(callTypeGroup.release());
}
// Pipeline interface tests.
de::MovePtr<tcu::TestCaseGroup> pipelineInterfaceGroup(new tcu::TestCaseGroup(testCtx, "pipeline_interface", "Test data spilling and unspilling of pipeline interface variables"));
struct
{
InterfaceType interfaceType;
const char* name;
} interfaceTypes[] =
{
{ InterfaceType::RAY_PAYLOAD, "ray_payload" },
{ InterfaceType::CALLABLE_DATA, "callable_data" },
{ InterfaceType::HIT_ATTRIBUTES, "hit_attributes" },
{ InterfaceType::SHADER_RECORD_BUFFER_RGEN, "shader_record_buffer_rgen" },
{ InterfaceType::SHADER_RECORD_BUFFER_CALL, "shader_record_buffer_call" },
{ InterfaceType::SHADER_RECORD_BUFFER_MISS, "shader_record_buffer_miss" },
{ InterfaceType::SHADER_RECORD_BUFFER_HIT, "shader_record_buffer_hit" },
};
for (int idx = 0; idx < DE_LENGTH_OF_ARRAY(interfaceTypes); ++idx)
{
const auto& entry = interfaceTypes[idx];
DataSpillPipelineInterfaceTestCase::TestParams params;
params.interfaceType = entry.interfaceType;
pipelineInterfaceGroup->addChild(new DataSpillPipelineInterfaceTestCase(testCtx, entry.name, "", params));
}
group->addChild(pipelineInterfaceGroup.release());
return group.release();
}
} // RayTracing
} // vkt