/*------------------------------------------------------------------------
 * Vulkan Conformance Tests
 * ------------------------
 *
 * Copyright (c) 2019 The Khronos Group Inc.
 * Copyright (c) 2019 Google Inc.
 * Copyright (c) 2017 Codeplay Software Ltd.
 * Copyright (c) 2018 NVIDIA Corporation
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *      http://www.apache.org/licenses/LICENSE-2.0
 *
 * 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 Subgroups Tests
 */ /*--------------------------------------------------------------------*/

#include "vktSubgroupsPartitionedTests.hpp"
#include "vktSubgroupsScanHelpers.hpp"
#include "vktSubgroupsTestsUtils.hpp"

#include <string>
#include <vector>

using namespace tcu;
using namespace std;
using namespace vk;
using namespace vkt;

namespace
{
enum OpType
{
	OPTYPE_ADD = 0,
	OPTYPE_MUL,
	OPTYPE_MIN,
	OPTYPE_MAX,
	OPTYPE_AND,
	OPTYPE_OR,
	OPTYPE_XOR,
	OPTYPE_INCLUSIVE_ADD,
	OPTYPE_INCLUSIVE_MUL,
	OPTYPE_INCLUSIVE_MIN,
	OPTYPE_INCLUSIVE_MAX,
	OPTYPE_INCLUSIVE_AND,
	OPTYPE_INCLUSIVE_OR,
	OPTYPE_INCLUSIVE_XOR,
	OPTYPE_EXCLUSIVE_ADD,
	OPTYPE_EXCLUSIVE_MUL,
	OPTYPE_EXCLUSIVE_MIN,
	OPTYPE_EXCLUSIVE_MAX,
	OPTYPE_EXCLUSIVE_AND,
	OPTYPE_EXCLUSIVE_OR,
	OPTYPE_EXCLUSIVE_XOR,
	OPTYPE_LAST
};

static Operator getOperator(OpType t)
{
	switch (t)
	{
		case OPTYPE_ADD:
		case OPTYPE_INCLUSIVE_ADD:
		case OPTYPE_EXCLUSIVE_ADD:
			return OPERATOR_ADD;
		case OPTYPE_MUL:
		case OPTYPE_INCLUSIVE_MUL:
		case OPTYPE_EXCLUSIVE_MUL:
			return OPERATOR_MUL;
		case OPTYPE_MIN:
		case OPTYPE_INCLUSIVE_MIN:
		case OPTYPE_EXCLUSIVE_MIN:
			return OPERATOR_MIN;
		case OPTYPE_MAX:
		case OPTYPE_INCLUSIVE_MAX:
		case OPTYPE_EXCLUSIVE_MAX:
			return OPERATOR_MAX;
		case OPTYPE_AND:
		case OPTYPE_INCLUSIVE_AND:
		case OPTYPE_EXCLUSIVE_AND:
			return OPERATOR_AND;
		case OPTYPE_OR:
		case OPTYPE_INCLUSIVE_OR:
		case OPTYPE_EXCLUSIVE_OR:
			return OPERATOR_OR;
		case OPTYPE_XOR:
		case OPTYPE_INCLUSIVE_XOR:
		case OPTYPE_EXCLUSIVE_XOR:
			return OPERATOR_XOR;
		default:
			DE_FATAL("Unsupported op type");
			return OPERATOR_ADD;
	}
}

static ScanType getScanType(OpType t)
{
	switch (t)
	{
		case OPTYPE_ADD:
		case OPTYPE_MUL:
		case OPTYPE_MIN:
		case OPTYPE_MAX:
		case OPTYPE_AND:
		case OPTYPE_OR:
		case OPTYPE_XOR:
			return SCAN_REDUCE;
		case OPTYPE_INCLUSIVE_ADD:
		case OPTYPE_INCLUSIVE_MUL:
		case OPTYPE_INCLUSIVE_MIN:
		case OPTYPE_INCLUSIVE_MAX:
		case OPTYPE_INCLUSIVE_AND:
		case OPTYPE_INCLUSIVE_OR:
		case OPTYPE_INCLUSIVE_XOR:
			return SCAN_INCLUSIVE;
		case OPTYPE_EXCLUSIVE_ADD:
		case OPTYPE_EXCLUSIVE_MUL:
		case OPTYPE_EXCLUSIVE_MIN:
		case OPTYPE_EXCLUSIVE_MAX:
		case OPTYPE_EXCLUSIVE_AND:
		case OPTYPE_EXCLUSIVE_OR:
		case OPTYPE_EXCLUSIVE_XOR:
			return SCAN_EXCLUSIVE;
		default:
			DE_FATAL("Unsupported op type");
			return SCAN_REDUCE;
	}
}

static bool checkVertexPipelineStages(const void* internalData, std::vector<const void*> datas,
									  deUint32 width, deUint32)
{
	DE_UNREF(internalData);
	return vkt::subgroups::check(datas, width, 0xFFFFFF);
}

static bool checkCompute(const void* internalData, std::vector<const void*> datas,
						 const deUint32 numWorkgroups[3], const deUint32 localSize[3],
						 deUint32)
{
	DE_UNREF(internalData);
	return vkt::subgroups::checkCompute(datas, numWorkgroups, localSize, 0xFFFFFF);
}

std::string getOpTypeName(Operator op, ScanType scanType)
{
   return getScanOpName("subgroup", "", op, scanType);
}

std::string getOpTypeNamePartitioned(Operator op, ScanType scanType)
{
   return getScanOpName("subgroupPartitioned", "NV", op, scanType);
}

struct CaseDefinition
{
	Operator			op;
	ScanType			scanType;
	VkShaderStageFlags	shaderStage;
	VkFormat			format;
	de::SharedPtr<bool>	geometryPointSizeSupported;
	deBool				requiredSubgroupSize;
};

std::string getExtHeader(CaseDefinition caseDef)
{
	return	"#extension GL_NV_shader_subgroup_partitioned: enable\n"
			"#extension GL_KHR_shader_subgroup_arithmetic: enable\n"
			"#extension GL_KHR_shader_subgroup_ballot: enable\n" +
			subgroups::getAdditionalExtensionForFormat(caseDef.format);
}

string getTestString(const CaseDefinition &caseDef)
{
	Operator op = caseDef.op;
	ScanType st = caseDef.scanType;

    // NOTE: tempResult can't have anything in bits 31:24 to avoid int->float
    // conversion overflow in framebuffer tests.
    string fmt = subgroups::getFormatNameForGLSL(caseDef.format);
	string bdy =
		"  uint tempResult = 0;\n"
		"  uint id = gl_SubgroupInvocationID;\n";

    // Test the case where the partition has a single subset with all invocations in it.
    // This should generate the same result as the non-partitioned function.
    bdy +=
        "  uvec4 allBallot = mask;\n"
        "  " + fmt + " allResult = " + getOpTypeNamePartitioned(op, st) + "(data[gl_SubgroupInvocationID], allBallot);\n"
        "  " + fmt + " refResult = " + getOpTypeName(op, st) + "(data[gl_SubgroupInvocationID]);\n"
        "  if (" + getCompare(op, caseDef.format, "allResult", "refResult") + ") {\n"
        "      tempResult |= 0x1;\n"
        "  }\n";

	// The definition of a partition doesn't forbid bits corresponding to inactive
	// invocations being in the subset with active invocations. In other words, test that
	// bits corresponding to inactive invocations are ignored.
	bdy +=
		"  if (0 == (gl_SubgroupInvocationID % 2)) {\n"
		"    " + fmt + " allResult = " + getOpTypeNamePartitioned(op, st) + "(data[gl_SubgroupInvocationID], allBallot);\n"
		"    " + fmt + " refResult = " + getOpTypeName(op, st) + "(data[gl_SubgroupInvocationID]);\n"
		"    if (" + getCompare(op, caseDef.format, "allResult", "refResult") + ") {\n"
		"        tempResult |= 0x2;\n"
		"    }\n"
		"  } else {\n"
		"    tempResult |= 0x2;\n"
		"  }\n";

	// Test the case where the partition has each invocation in a unique subset. For
	// exclusive ops, the result is identity. For reduce/inclusive, it's the original value.
	string expectedSelfResult = "data[gl_SubgroupInvocationID]";
	if (st == SCAN_EXCLUSIVE)
		expectedSelfResult = getIdentity(op, caseDef.format);

	bdy +=
		"  uvec4 selfBallot = subgroupPartitionNV(gl_SubgroupInvocationID);\n"
		"  " + fmt + " selfResult = " + getOpTypeNamePartitioned(op, st) + "(data[gl_SubgroupInvocationID], selfBallot);\n"
		"  if (" + getCompare(op, caseDef.format, "selfResult", expectedSelfResult) + ") {\n"
		"      tempResult |= 0x4;\n"
		"  }\n";

	// Test "random" partitions based on a hash of the invocation id.
	// This "hash" function produces interesting/randomish partitions.
	static const char *idhash = "((id%N)+(id%(N+1))-(id%2)+(id/2))%((N+1)/2)";

	bdy +=
		"  for (uint N = 1; N < 16; ++N) {\n"
		"    " + fmt + " idhashFmt = " + fmt + "(" + idhash + ");\n"
		"    uvec4 partitionBallot = subgroupPartitionNV(idhashFmt) & mask;\n"
		"    " + fmt + " partitionedResult = " + getOpTypeNamePartitioned(op, st) + "(data[gl_SubgroupInvocationID], partitionBallot);\n"
		"      for (uint i = 0; i < N; ++i) {\n"
		"        " + fmt + " iFmt = " + fmt + "(i);\n"
		"        if (" + getCompare(op, caseDef.format, "idhashFmt", "iFmt") + ") {\n"
		"          " + fmt + " subsetResult = " + getOpTypeName(op, st) + "(data[gl_SubgroupInvocationID]);\n"
		"          tempResult |= " + getCompare(op, caseDef.format, "partitionedResult", "subsetResult") + " ? (0x4 << N) : 0;\n"
		"        }\n"
		"      }\n"
		"  }\n"
		// tests in flow control:
		"  if (1 == (gl_SubgroupInvocationID % 2)) {\n"
		"    for (uint N = 1; N < 7; ++N) {\n"
		"      " + fmt + " idhashFmt = " + fmt + "(" + idhash + ");\n"
		"      uvec4 partitionBallot = subgroupPartitionNV(idhashFmt) & mask;\n"
		"      " + fmt + " partitionedResult = " + getOpTypeNamePartitioned(op, st) + "(data[gl_SubgroupInvocationID], partitionBallot);\n"
		"        for (uint i = 0; i < N; ++i) {\n"
		"          " + fmt + " iFmt = " + fmt + "(i);\n"
		"          if (" + getCompare(op, caseDef.format, "idhashFmt", "iFmt") + ") {\n"
		"            " + fmt + " subsetResult = " + getOpTypeName(op, st) + "(data[gl_SubgroupInvocationID]);\n"
		"            tempResult |= " + getCompare(op, caseDef.format, "partitionedResult", "subsetResult") + " ? (0x20000 << N) : 0;\n"
		"          }\n"
		"        }\n"
		"    }\n"
		"  } else {\n"
		"    tempResult |= 0xFC0000;\n"
		"  }\n"
		;

	return bdy;
}

void initFrameBufferPrograms(SourceCollections& programCollection, CaseDefinition caseDef)
{
	const vk::ShaderBuildOptions	buildOptions	(programCollection.usedVulkanVersion, vk::SPIRV_VERSION_1_3, 0u);
	std::ostringstream				bdy;

	std::string extHeader	= getExtHeader(caseDef);

	subgroups::setFragmentShaderFrameBuffer(programCollection);

	if (VK_SHADER_STAGE_VERTEX_BIT != caseDef.shaderStage)
		subgroups::setVertexShaderFrameBuffer(programCollection);

	bdy << getTestString(caseDef);

	if (VK_SHADER_STAGE_VERTEX_BIT == caseDef.shaderStage)
	{
		std::ostringstream vertexSrc;
		vertexSrc << glu::getGLSLVersionDeclaration(glu::GLSL_VERSION_450)<<"\n"
			<< extHeader.c_str()
			<< "layout(location = 0) in highp vec4 in_position;\n"
			<< "layout(location = 0) out float out_color;\n"
			<< "layout(set = 0, binding = 0) uniform Buffer1\n"
			<< "{\n"
			<< "  " << subgroups::getFormatNameForGLSL(caseDef.format) << " data[" << subgroups::maxSupportedSubgroupSize() << "];\n"
			<< "};\n"
			<< "\n"
			<< "void main (void)\n"
			<< "{\n"
			<< "  uvec4 mask = subgroupBallot(true);\n"
			<< bdy.str()
			<< "  out_color = float(tempResult);\n"
			<< "  gl_Position = in_position;\n"
			<< "  gl_PointSize = 1.0f;\n"
			<< "}\n";
		programCollection.glslSources.add("vert")
			<< glu::VertexSource(vertexSrc.str()) << buildOptions;
	}
	else if (VK_SHADER_STAGE_GEOMETRY_BIT == caseDef.shaderStage)
	{
		std::ostringstream geometry;

		geometry << glu::getGLSLVersionDeclaration(glu::GLSL_VERSION_450)<<"\n"
			<< extHeader.c_str()
			<< "layout(points) in;\n"
			<< "layout(points, max_vertices = 1) out;\n"
			<< "layout(location = 0) out float out_color;\n"
			<< "layout(set = 0, binding = 0) uniform Buffer\n"
			<< "{\n"
			<< "  " << subgroups::getFormatNameForGLSL(caseDef.format) << " data[" << subgroups::maxSupportedSubgroupSize() << "];\n"
			<< "};\n"
			<< "\n"
			<< "void main (void)\n"
			<< "{\n"
			<< "  uvec4 mask = subgroupBallot(true);\n"
			<< bdy.str()
			<< "  out_color = float(tempResult);\n"
			<< "  gl_Position = gl_in[0].gl_Position;\n"
			<< (*caseDef.geometryPointSizeSupported ? "  gl_PointSize = gl_in[0].gl_PointSize;\n" : "")
			<< "  EmitVertex();\n"
			<< "  EndPrimitive();\n"
			<< "}\n";

		programCollection.glslSources.add("geometry")
				<< glu::GeometrySource(geometry.str()) << buildOptions;
	}
	else if (VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT == caseDef.shaderStage)
	{
		std::ostringstream controlSource;
		controlSource  << glu::getGLSLVersionDeclaration(glu::GLSL_VERSION_450)<<"\n"
			<< extHeader.c_str()
			<< "layout(vertices = 2) out;\n"
			<< "layout(location = 0) out float out_color[];\n"
			<< "layout(set = 0, binding = 0) uniform Buffer1\n"
			<< "{\n"
			<< "  " << subgroups::getFormatNameForGLSL(caseDef.format) << " data[" << subgroups::maxSupportedSubgroupSize() << "];\n"
			<< "};\n"
			<< "\n"
			<< "void main (void)\n"
			<< "{\n"
			<< "  if (gl_InvocationID == 0)\n"
			<<"  {\n"
			<< "    gl_TessLevelOuter[0] = 1.0f;\n"
			<< "    gl_TessLevelOuter[1] = 1.0f;\n"
			<< "  }\n"
			<< "  uvec4 mask = subgroupBallot(true);\n"
			<< bdy.str()
			<< "  out_color[gl_InvocationID] = float(tempResult);"
			<< "  gl_out[gl_InvocationID].gl_Position = gl_in[gl_InvocationID].gl_Position;\n"
			<< "}\n";


		programCollection.glslSources.add("tesc")
			<< glu::TessellationControlSource(controlSource.str()) << buildOptions;
		subgroups::setTesEvalShaderFrameBuffer(programCollection);
	}
	else if (VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT == caseDef.shaderStage)
	{

		std::ostringstream evaluationSource;
		evaluationSource << glu::getGLSLVersionDeclaration(glu::GLSL_VERSION_450)<<"\n"
			<< extHeader.c_str()
			<< "layout(isolines, equal_spacing, ccw ) in;\n"
			<< "layout(location = 0) out float out_color;\n"
			<< "layout(set = 0, binding = 0) uniform Buffer1\n"
			<< "{\n"
			<< "  " << subgroups::getFormatNameForGLSL(caseDef.format) << " data[" << subgroups::maxSupportedSubgroupSize() << "];\n"
			<< "};\n"
			<< "\n"
			<< "void main (void)\n"
			<< "{\n"
			<< "  uvec4 mask = subgroupBallot(true);\n"
			<< bdy.str()
			<< "  out_color = float(tempResult);\n"
			<< "  gl_Position = mix(gl_in[0].gl_Position, gl_in[1].gl_Position, gl_TessCoord.x);\n"
			<< "}\n";

		subgroups::setTesCtrlShaderFrameBuffer(programCollection);
		programCollection.glslSources.add("tese") << glu::TessellationEvaluationSource(evaluationSource.str()) << buildOptions;
	}
	else
	{
		DE_FATAL("Unsupported shader stage");
	}
}

void initPrograms(SourceCollections& programCollection, CaseDefinition caseDef)
{
	const string extHeader	= getExtHeader(caseDef);
	const string bdy		= getTestString(caseDef);

	if (VK_SHADER_STAGE_COMPUTE_BIT == caseDef.shaderStage)
	{
		std::ostringstream src;

		src << "#version 450\n"
			<< extHeader.c_str()
			<< "layout (local_size_x_id = 0, local_size_y_id = 1, "
			"local_size_z_id = 2) in;\n"
			<< "layout(set = 0, binding = 0, std430) buffer Buffer1\n"
			<< "{\n"
			<< "  uint result[];\n"
			<< "};\n"
			<< "layout(set = 0, binding = 1, std430) buffer Buffer2\n"
			<< "{\n"
			<< "  " << subgroups::getFormatNameForGLSL(caseDef.format) << " data[];\n"
			<< "};\n"
			<< "\n"
			<< "void main (void)\n"
			<< "{\n"
			<< "  uvec3 globalSize = gl_NumWorkGroups * gl_WorkGroupSize;\n"
			<< "  highp uint offset = globalSize.x * ((globalSize.y * "
			"gl_GlobalInvocationID.z) + gl_GlobalInvocationID.y) + "
			"gl_GlobalInvocationID.x;\n"
			<< "  uvec4 mask = subgroupBallot(true);\n"
			<< bdy
			<< "  result[offset] = tempResult;\n"
			<< "}\n";

		programCollection.glslSources.add("comp")
				<< glu::ComputeSource(src.str()) << vk::ShaderBuildOptions(programCollection.usedVulkanVersion, vk::SPIRV_VERSION_1_3, 0u);
	}
	else
	{
		{
			const std::string vertex =
				"#version 450\n"
				+ extHeader +
				"layout(set = 0, binding = 0, std430) buffer Buffer1\n"
				"{\n"
				"  uint result[];\n"
				"};\n"
				"layout(set = 0, binding = 4, std430) readonly buffer Buffer2\n"
				"{\n"
				"  " + subgroups::getFormatNameForGLSL(caseDef.format) + " data[];\n"
				"};\n"
				"\n"
				"void main (void)\n"
				"{\n"
				"  uvec4 mask = subgroupBallot(true);\n"
				+ bdy+
				"  result[gl_VertexIndex] = tempResult;\n"
				"  float pixelSize = 2.0f/1024.0f;\n"
				"  float pixelPosition = pixelSize/2.0f - 1.0f;\n"
				"  gl_Position = vec4(float(gl_VertexIndex) * pixelSize + pixelPosition, 0.0f, 0.0f, 1.0f);\n"
				"  gl_PointSize = 1.0f;\n"
				"}\n";
			programCollection.glslSources.add("vert")
					<< glu::VertexSource(vertex) << vk::ShaderBuildOptions(programCollection.usedVulkanVersion, vk::SPIRV_VERSION_1_3, 0u);
		}

		{
			const std::string tesc =
				"#version 450\n"
				+ extHeader +
				"layout(vertices=1) out;\n"
				"layout(set = 0, binding = 1, std430) buffer Buffer1\n"
				"{\n"
				"  uint result[];\n"
				"};\n"
				"layout(set = 0, binding = 4, std430) readonly buffer Buffer2\n"
				"{\n"
				"  " + subgroups::getFormatNameForGLSL(caseDef.format) + " data[];\n"
				"};\n"
				"\n"
				"void main (void)\n"
				"{\n"
				"  uvec4 mask = subgroupBallot(true);\n"
				+ bdy +
				"  result[gl_PrimitiveID] = tempResult;\n"
				"  if (gl_InvocationID == 0)\n"
				"  {\n"
				"    gl_TessLevelOuter[0] = 1.0f;\n"
				"    gl_TessLevelOuter[1] = 1.0f;\n"
				"  }\n"
				"  gl_out[gl_InvocationID].gl_Position = gl_in[gl_InvocationID].gl_Position;\n"
				"}\n";
			programCollection.glslSources.add("tesc")
				<< glu::TessellationControlSource(tesc) << vk::ShaderBuildOptions(programCollection.usedVulkanVersion, vk::SPIRV_VERSION_1_3, 0u);
		}

		{
			const std::string tese =
				"#version 450\n"
				+ extHeader +
				"layout(isolines) in;\n"
				"layout(set = 0, binding = 2, std430) buffer Buffer1\n"
				"{\n"
				"  uint result[];\n"
				"};\n"
				"layout(set = 0, binding = 4, std430) readonly buffer Buffer2\n"
				"{\n"
				"  " + subgroups::getFormatNameForGLSL(caseDef.format) + " data[];\n"
				"};\n"
				"\n"
				"void main (void)\n"
				"{\n"
				"  uvec4 mask = subgroupBallot(true);\n"
				+ bdy +
				"  result[gl_PrimitiveID * 2 + uint(gl_TessCoord.x + 0.5)] = tempResult;\n"
				"  float pixelSize = 2.0f/1024.0f;\n"
				"  gl_Position = gl_in[0].gl_Position + gl_TessCoord.x * pixelSize / 2.0f;\n"
				"}\n";
			programCollection.glslSources.add("tese")
				<< glu::TessellationEvaluationSource(tese) << vk::ShaderBuildOptions(programCollection.usedVulkanVersion, vk::SPIRV_VERSION_1_3, 0u);
		}

		{
			const std::string geometry =
				"#version 450\n"
				+ extHeader +
				"layout(${TOPOLOGY}) in;\n"
				"layout(points, max_vertices = 1) out;\n"
				"layout(set = 0, binding = 3, std430) buffer Buffer1\n"
				"{\n"
				"  uint result[];\n"
				"};\n"
				"layout(set = 0, binding = 4, std430) readonly buffer Buffer2\n"
				"{\n"
				"  " + subgroups::getFormatNameForGLSL(caseDef.format) + " data[];\n"
				"};\n"
				"\n"
				"void main (void)\n"
				"{\n"
				"  uvec4 mask = subgroupBallot(true);\n"
				 + bdy +
				"  result[gl_PrimitiveIDIn] = tempResult;\n"
				"  gl_Position = gl_in[0].gl_Position;\n"
				"  EmitVertex();\n"
				"  EndPrimitive();\n"
				"}\n";
			subgroups::addGeometryShadersFromTemplate(geometry, vk::ShaderBuildOptions(programCollection.usedVulkanVersion, vk::SPIRV_VERSION_1_3, 0u),
													  programCollection.glslSources);
		}

		{
			const std::string fragment =
				"#version 450\n"
				+ extHeader +
				"layout(location = 0) out uint result;\n"
				"layout(set = 0, binding = 4, std430) readonly buffer Buffer2\n"
				"{\n"
				"  " + subgroups::getFormatNameForGLSL(caseDef.format) + " data[];\n"
				"};\n"
				"void main (void)\n"
				"{\n"
				"  uvec4 mask = subgroupBallot(true);\n"
				+ bdy +
				"  result = tempResult;\n"
				"}\n";
			programCollection.glslSources.add("fragment")
				<< glu::FragmentSource(fragment)<< vk::ShaderBuildOptions(programCollection.usedVulkanVersion, vk::SPIRV_VERSION_1_3, 0u);
		}
		subgroups::addNoSubgroupShader(programCollection);
	}
}

void supportedCheck (Context& context, CaseDefinition caseDef)
{
	if (!subgroups::isSubgroupSupported(context))
		TCU_THROW(NotSupportedError, "Subgroup operations are not supported");

	if (!subgroups::isSubgroupFeatureSupportedForDevice(context, VK_SUBGROUP_FEATURE_PARTITIONED_BIT_NV))
		TCU_THROW(NotSupportedError, "Device does not support subgroup partitioned operations");

	if (!subgroups::isFormatSupportedForDevice(context, caseDef.format))
		TCU_THROW(NotSupportedError, "Device does not support the specified format in subgroup operations");

	if (caseDef.requiredSubgroupSize)
	{
		if (!context.requireDeviceFunctionality("VK_EXT_subgroup_size_control"))
			TCU_THROW(NotSupportedError, "Device does not support VK_EXT_subgroup_size_control extension");
		VkPhysicalDeviceSubgroupSizeControlFeaturesEXT subgroupSizeControlFeatures;
		subgroupSizeControlFeatures.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_SIZE_CONTROL_FEATURES_EXT;
		subgroupSizeControlFeatures.pNext = DE_NULL;

		VkPhysicalDeviceFeatures2 features;
		features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2;
		features.pNext = &subgroupSizeControlFeatures;

		context.getInstanceInterface().getPhysicalDeviceFeatures2(context.getPhysicalDevice(), &features);

		if (subgroupSizeControlFeatures.subgroupSizeControl == DE_FALSE)
			TCU_THROW(NotSupportedError, "Device does not support varying subgroup sizes nor required subgroup size");

		if (subgroupSizeControlFeatures.computeFullSubgroups == DE_FALSE)
			TCU_THROW(NotSupportedError, "Device does not support full subgroups in compute shaders");
	}

	*caseDef.geometryPointSizeSupported = subgroups::isTessellationAndGeometryPointSizeSupported(context);
}

tcu::TestStatus noSSBOtest (Context& context, const CaseDefinition caseDef)
{
	if (!subgroups::areSubgroupOperationsSupportedForStage(context, caseDef.shaderStage))
	{
		if (subgroups::areSubgroupOperationsRequiredForStage(caseDef.shaderStage))
		{
			return tcu::TestStatus::fail(
					   "Shader stage " +
					   subgroups::getShaderStageName(caseDef.shaderStage) +
					   " is required to support subgroup operations!");
		}
		else
		{
			TCU_THROW(NotSupportedError, "Device does not support subgroup operations for this stage");
		}
	}

	subgroups::SSBOData inputData;
	inputData.format = caseDef.format;
	inputData.layout = subgroups::SSBOData::LayoutStd140;
	inputData.numElements = subgroups::maxSupportedSubgroupSize();
	inputData.initializeType = subgroups::SSBOData::InitializeNonZero;

	if (VK_SHADER_STAGE_VERTEX_BIT == caseDef.shaderStage)
		return subgroups::makeVertexFrameBufferTest(context, VK_FORMAT_R32_UINT, &inputData, 1, DE_NULL, checkVertexPipelineStages);
	else if (VK_SHADER_STAGE_GEOMETRY_BIT == caseDef.shaderStage)
		return subgroups::makeGeometryFrameBufferTest(context, VK_FORMAT_R32_UINT, &inputData, 1, DE_NULL, checkVertexPipelineStages);
	else if (VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT == caseDef.shaderStage)
		return subgroups::makeTessellationEvaluationFrameBufferTest(context, VK_FORMAT_R32_UINT, &inputData, 1, DE_NULL, checkVertexPipelineStages, VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT);
	else if (VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT == caseDef.shaderStage)
		return subgroups::makeTessellationEvaluationFrameBufferTest(context, VK_FORMAT_R32_UINT, &inputData, 1, DE_NULL, checkVertexPipelineStages, VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT);
	else
		TCU_THROW(InternalError, "Unhandled shader stage");
}


tcu::TestStatus test(Context& context, const CaseDefinition caseDef)
{
	if (VK_SHADER_STAGE_COMPUTE_BIT == caseDef.shaderStage)
	{
		if (!subgroups::areSubgroupOperationsSupportedForStage(context, caseDef.shaderStage))
		{
			return tcu::TestStatus::fail(
					"Shader stage " +
					subgroups::getShaderStageName(caseDef.shaderStage) +
					" is required to support subgroup operations!");
		}
		subgroups::SSBOData inputData;
		inputData.format = caseDef.format;
		inputData.layout = subgroups::SSBOData::LayoutStd430;
		inputData.numElements = subgroups::maxSupportedSubgroupSize();
		inputData.initializeType = subgroups::SSBOData::InitializeNonZero;

		if (caseDef.requiredSubgroupSize == DE_FALSE)
			return subgroups::makeComputeTest(context, VK_FORMAT_R32_UINT, &inputData, 1, DE_NULL, checkCompute);

		tcu::TestLog& log = context.getTestContext().getLog();
		VkPhysicalDeviceSubgroupSizeControlPropertiesEXT subgroupSizeControlProperties;
		subgroupSizeControlProperties.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_SIZE_CONTROL_PROPERTIES_EXT;
		subgroupSizeControlProperties.pNext = DE_NULL;
		VkPhysicalDeviceProperties2 properties;
		properties.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2;
		properties.pNext = &subgroupSizeControlProperties;

		context.getInstanceInterface().getPhysicalDeviceProperties2(context.getPhysicalDevice(), &properties);

		log << tcu::TestLog::Message << "Testing required subgroup size range [" <<  subgroupSizeControlProperties.minSubgroupSize << ", "
			<< subgroupSizeControlProperties.maxSubgroupSize << "]" << tcu::TestLog::EndMessage;

		// According to the spec, requiredSubgroupSize must be a power-of-two integer.
		for (deUint32 size = subgroupSizeControlProperties.minSubgroupSize; size <= subgroupSizeControlProperties.maxSubgroupSize; size *= 2)
		{
			tcu::TestStatus result = subgroups::makeComputeTest(context, VK_FORMAT_R32_UINT, &inputData, 1, DE_NULL, checkCompute,
																size, VK_PIPELINE_SHADER_STAGE_CREATE_REQUIRE_FULL_SUBGROUPS_BIT_EXT);
			if (result.getCode() != QP_TEST_RESULT_PASS)
			{
				log << tcu::TestLog::Message << "subgroupSize " << size << " failed" << tcu::TestLog::EndMessage;
				return result;
			}
		}

		return tcu::TestStatus::pass("OK");
	}
	else
	{
		VkPhysicalDeviceSubgroupProperties subgroupProperties;
		subgroupProperties.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_PROPERTIES;
		subgroupProperties.pNext = DE_NULL;

		VkPhysicalDeviceProperties2 properties;
		properties.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2;
		properties.pNext = &subgroupProperties;

		context.getInstanceInterface().getPhysicalDeviceProperties2(context.getPhysicalDevice(), &properties);

		VkShaderStageFlagBits stages = (VkShaderStageFlagBits)(caseDef.shaderStage  & subgroupProperties.supportedStages);

		if (VK_SHADER_STAGE_FRAGMENT_BIT != stages && !subgroups::isVertexSSBOSupportedForDevice(context))
		{
			if ( (stages & VK_SHADER_STAGE_FRAGMENT_BIT) == 0)
				TCU_THROW(NotSupportedError, "Device does not support vertex stage SSBO writes");
			else
				stages = VK_SHADER_STAGE_FRAGMENT_BIT;
		}

		if ((VkShaderStageFlagBits)0u == stages)
			TCU_THROW(NotSupportedError, "Subgroup operations are not supported for any graphic shader");

		subgroups::SSBOData inputData;
		inputData.format			= caseDef.format;
		inputData.layout			= subgroups::SSBOData::LayoutStd430;
		inputData.numElements		= subgroups::maxSupportedSubgroupSize();
		inputData.initializeType	= subgroups::SSBOData::InitializeNonZero;
		inputData.binding			= 4u;
		inputData.stages			= stages;

		return subgroups::allStages(context, VK_FORMAT_R32_UINT, &inputData, 1, DE_NULL, checkVertexPipelineStages, stages);
	}
}
}

namespace vkt
{
namespace subgroups
{
tcu::TestCaseGroup* createSubgroupsPartitionedTests(tcu::TestContext& testCtx)
{
	de::MovePtr<tcu::TestCaseGroup> graphicGroup(new tcu::TestCaseGroup(
		testCtx, "graphics", "Subgroup partitioned category tests: graphics"));
	de::MovePtr<tcu::TestCaseGroup> computeGroup(new tcu::TestCaseGroup(
		testCtx, "compute", "Subgroup partitioned category tests: compute"));
	de::MovePtr<tcu::TestCaseGroup> framebufferGroup(new tcu::TestCaseGroup(
		testCtx, "framebuffer", "Subgroup partitioned category tests: framebuffer"));

	const VkShaderStageFlags stages[] =
	{
		VK_SHADER_STAGE_VERTEX_BIT,
		VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT,
		VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT,
		VK_SHADER_STAGE_GEOMETRY_BIT,
	};

	const std::vector<VkFormat> formats = subgroups::getAllFormats();

	for (size_t formatIndex = 0; formatIndex < formats.size(); ++formatIndex)
	{
		const VkFormat format = formats[formatIndex];

		for (int opTypeIndex = 0; opTypeIndex < OPTYPE_LAST; ++opTypeIndex)
		{
			bool isBool = subgroups::isFormatBool(format);
			bool isFloat = subgroups::isFormatFloat(format);

			OpType opType = static_cast<OpType>(opTypeIndex);
			Operator op = getOperator(opType);
			ScanType st = getScanType(opType);

			bool isBitwiseOp = (op == OPERATOR_AND || op == OPERATOR_OR || op == OPERATOR_XOR);

			// Skip float with bitwise category.
			if (isFloat && isBitwiseOp)
				continue;

			// Skip bool when its not the bitwise category.
			if (isBool && !isBitwiseOp)
				continue;

			const std::string name = de::toLower(getOpTypeName(op, st)) + "_" + subgroups::getFormatNameForGLSL(format);

			{
				CaseDefinition caseDef = {op, st, VK_SHADER_STAGE_COMPUTE_BIT, format, de::SharedPtr<bool>(new bool), DE_FALSE};
				addFunctionCaseWithPrograms(computeGroup.get(), name,
											"", supportedCheck, initPrograms, test, caseDef);
				caseDef.requiredSubgroupSize = DE_TRUE;
				addFunctionCaseWithPrograms(computeGroup.get(), name + "_requiredsubgroupsize",
											"", supportedCheck, initPrograms, test, caseDef);
			}

			{
				const CaseDefinition caseDef = {op, st, VK_SHADER_STAGE_ALL_GRAPHICS, format, de::SharedPtr<bool>(new bool), DE_FALSE};
				addFunctionCaseWithPrograms(graphicGroup.get(), name,
											"", supportedCheck, initPrograms, test, caseDef);
			}

			for (int stageIndex = 0; stageIndex < DE_LENGTH_OF_ARRAY(stages); ++stageIndex)
			{
				const CaseDefinition caseDef = {op, st, stages[stageIndex], format, de::SharedPtr<bool>(new bool), DE_FALSE};
				addFunctionCaseWithPrograms(framebufferGroup.get(), name +
											"_" + getShaderStageName(caseDef.shaderStage), "",
											supportedCheck, initFrameBufferPrograms, noSSBOtest, caseDef);
			}
		}
	}

	de::MovePtr<tcu::TestCaseGroup> group(new tcu::TestCaseGroup(
		testCtx, "partitioned", "Subgroup partitioned category tests"));

	group->addChild(graphicGroup.release());
	group->addChild(computeGroup.release());
	group->addChild(framebufferGroup.release());

	return group.release();
}
} // subgroups
} // vkt