Google Git
Sign in
fuchsia / third_party / vulkan-cts / 8501d2f862d888191fd7a19f188d55d99de8d592 / . / external / vulkancts / modules / vulkan / spirv_assembly / vktSpvAsm16bitStorageTests.cpp
blob: ddf42684de5224515a4e442c21dea98d2c314105 [file] [log] [blame]
/*-------------------------------------------------------------------------
* Vulkan Conformance Tests
* ------------------------
*
* Copyright (c) 2017 Google Inc.
*
* 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 SPIR-V Assembly Tests for the VK_KHR_16bit_storage
*//*--------------------------------------------------------------------*/
// VK_KHR_16bit_storage
//
// \todo [2017-02-08 antiagainst] Additional corner cases to check:
//
// * Test OpAccessChain with subword types
// * For newly enabled types T:
// * For composite types: vector, matrix, structures, array over T:
// 1. Use OpAccessChain to form a pointer to a subword type.
// 2. Load the subword value X16.
// 3. Convert X16 to X32.
// 4. Store X32 to BufferBlock.
// 5. Host inspects X32.
// * Test {StorageInputOutput16} 16-to-16:
// * For newly enabled types T:
// 1. Host creates X16 stream values of type T.
// 2. Shaders have corresponding capability.
// 3. For each viable shader stage:
// 3a. Load X16 Input variable.
// 3b. Store X16 to Output variable.
// 4. Host inspects resulting values.
// * Test {StorageInputOutput16} 16-to-16 one value to two:
// Like the previous test, but write X16 to two different output variables.
// (Checks that the 16-bit intermediate value can be used twice.)
#include "vktSpvAsm16bitStorageTests.hpp"
#include "tcuFloat.hpp"
#include "tcuRGBA.hpp"
#include "tcuStringTemplate.hpp"
#include "tcuTestLog.hpp"
#include "tcuVectorUtil.hpp"
#include "vkDefs.hpp"
#include "vkDeviceUtil.hpp"
#include "vkMemUtil.hpp"
#include "vkPlatform.hpp"
#include "vkPrograms.hpp"
#include "vkQueryUtil.hpp"
#include "vkRef.hpp"
#include "vkRefUtil.hpp"
#include "vkStrUtil.hpp"
#include "vkTypeUtil.hpp"
#include "deRandom.hpp"
#include "deStringUtil.hpp"
#include "deUniquePtr.hpp"
#include "deMath.h"
#include "vktSpvAsmComputeShaderCase.hpp"
#include "vktSpvAsmComputeShaderTestUtil.hpp"
#include "vktSpvAsmGraphicsShaderTestUtil.hpp"
#include "vktTestCaseUtil.hpp"
#include "vktTestGroupUtil.hpp"
#include <limits>
#include <map>
#include <string>
#include <sstream>
#include <utility>
namespace vkt
{
namespace SpirVAssembly
{
using namespace vk;
using std::map;
using std::string;
using std::vector;
using tcu::IVec3;
using tcu::IVec4;
using tcu::RGBA;
using tcu::TestLog;
using tcu::TestStatus;
using tcu::Vec4;
using de::UniquePtr;
using tcu::StringTemplate;
using tcu::Vec4;
namespace
{
struct Capability
{
const char* name;
const char* cap;
const char* decor;
vk::VkDescriptorType dtype;
};
static const Capability CAPABILITIES[] =
{
{"uniform_buffer_block", "StorageUniformBufferBlock16", "BufferBlock", VK_DESCRIPTOR_TYPE_STORAGE_BUFFER},
{"uniform", "StorageUniform16", "Block", VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER},
};
VulkanFeatures get16BitStorageFeatures (const char* cap)
{
VulkanFeatures features;
if (string(cap) == "uniform_buffer_block")
features.ext16BitStorage = EXT16BITSTORAGEFEATURES_UNIFORM_BUFFER_BLOCK;
else if (string(cap) == "uniform")
features.ext16BitStorage = EXT16BITSTORAGEFEATURES_UNIFORM;
else
DE_ASSERT(false && "not supported");
return features;
}
// Batch function to check arrays of 16-bit floats.
//
// For comparing 16-bit floats, we need to consider both RTZ and RTE. So we can only recalculate
// the expected values here instead of get the expected values directly from the test case.
// Thus we need original floats here but not expected outputs.
template<RoundingModeFlags RoundingMode>
bool graphicsCheck16BitFloats (const std::vector<Resource>& originalFloats,
const vector<AllocationSp>& outputAllocs,
const std::vector<Resource>& /* expectedOutputs */,
tcu::TestLog& log)
{
if (outputAllocs.size() != originalFloats.size())
return false;
for (deUint32 outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
{
const deUint16* returned = static_cast<const deUint16*>(outputAllocs[outputNdx]->getHostPtr());
const float* original = static_cast<const float*>(originalFloats[outputNdx].second->data());
const deUint32 count = static_cast<deUint32>(originalFloats[outputNdx].second->getNumBytes() / sizeof(float));
for (deUint32 numNdx = 0; numNdx < count; ++numNdx)
if (!compare16BitFloat(original[numNdx], returned[numNdx], RoundingMode, log))
return false;
}
return true;
}
template<RoundingModeFlags RoundingMode>
bool computeCheck16BitFloats (const std::vector<BufferSp>& originalFloats,
const vector<AllocationSp>& outputAllocs,
const std::vector<BufferSp>& /* expectedOutputs */,
tcu::TestLog& log)
{
if (outputAllocs.size() != originalFloats.size())
return false;
for (deUint32 outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
{
const deUint16* returned = static_cast<const deUint16*>(outputAllocs[outputNdx]->getHostPtr());
const float* original = static_cast<const float*>(originalFloats[outputNdx]->data());
const deUint32 count = static_cast<deUint32>(originalFloats[outputNdx]->getNumBytes() / sizeof(float));
for (deUint32 numNdx = 0; numNdx < count; ++numNdx)
if (!compare16BitFloat(original[numNdx], returned[numNdx], RoundingMode, log))
return false;
}
return true;
}
// Batch function to check arrays of 32-bit floats.
//
// For comparing 32-bit floats, we just need the expected value precomputed in the test case.
// So we need expected outputs here but not original floats.
bool check32BitFloats (const std::vector<Resource>& /* originalFloats */,
const std::vector<AllocationSp>& outputAllocs,
const std::vector<Resource>& expectedOutputs,
tcu::TestLog& log)
{
if (outputAllocs.size() != expectedOutputs.size())
return false;
for (deUint32 outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
{
const float* returnedAsFloat = static_cast<const float*>(outputAllocs[outputNdx]->getHostPtr());
const float* expectedAsFloat = static_cast<const float*>(expectedOutputs[outputNdx].second->data());
const deUint32 count = static_cast<deUint32>(expectedOutputs[outputNdx].second->getNumBytes() / sizeof(float));
for (deUint32 numNdx = 0; numNdx < count; ++numNdx)
if (!compare32BitFloat(expectedAsFloat[numNdx], returnedAsFloat[numNdx], log))
return false;
}
return true;
}
// Overload for compute pipeline
bool check32BitFloats (const std::vector<BufferSp>& /* originalFloats */,
const std::vector<AllocationSp>& outputAllocs,
const std::vector<BufferSp>& expectedOutputs,
tcu::TestLog& log)
{
if (outputAllocs.size() != expectedOutputs.size())
return false;
for (deUint32 outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
{
const float* returnedAsFloat = static_cast<const float*>(outputAllocs[outputNdx]->getHostPtr());
const float* expectedAsFloat = static_cast<const float*>(expectedOutputs[outputNdx]->data());
const deUint32 count = static_cast<deUint32>(expectedOutputs[outputNdx]->getNumBytes() / sizeof(float));
for (deUint32 numNdx = 0; numNdx < count; ++numNdx)
if (!compare32BitFloat(expectedAsFloat[numNdx], returnedAsFloat[numNdx], log))
return false;
}
return true;
}
// Generate and return 32-bit integers.
//
// Expected count to be at least 16.
vector<deInt32> getInt32s (de::Random& rnd, const deUint32 count)
{
vector<deInt32> data;
data.reserve(count);
// Make sure we have boundary numbers.
data.push_back(deInt32(0x00000000)); // 0
data.push_back(deInt32(0x00000001)); // 1
data.push_back(deInt32(0x0000002a)); // 42
data.push_back(deInt32(0x00007fff)); // 32767
data.push_back(deInt32(0x00008000)); // 32768
data.push_back(deInt32(0x0000ffff)); // 65535
data.push_back(deInt32(0x00010000)); // 65536
data.push_back(deInt32(0x7fffffff)); // 2147483647
data.push_back(deInt32(0x80000000)); // -2147483648
data.push_back(deInt32(0x80000001)); // -2147483647
data.push_back(deInt32(0xffff0000)); // -65536
data.push_back(deInt32(0xffff0001)); // -65535
data.push_back(deInt32(0xffff8000)); // -32768
data.push_back(deInt32(0xffff8001)); // -32767
data.push_back(deInt32(0xffffffd6)); // -42
data.push_back(deInt32(0xffffffff)); // -1
DE_ASSERT(count >= data.size());
for (deUint32 numNdx = static_cast<deUint32>(data.size()); numNdx < count; ++numNdx)
data.push_back(static_cast<deInt32>(rnd.getUint32()));
return data;
}
// Generate and return 16-bit integers.
//
// Expected count to be at least 8.
vector<deInt16> getInt16s (de::Random& rnd, const deUint32 count)
{
vector<deInt16> data;
data.reserve(count);
// Make sure we have boundary numbers.
data.push_back(deInt16(0x0000)); // 0
data.push_back(deInt16(0x0001)); // 1
data.push_back(deInt16(0x002a)); // 42
data.push_back(deInt16(0x7fff)); // 32767
data.push_back(deInt16(0x8000)); // -32868
data.push_back(deInt16(0x8001)); // -32767
data.push_back(deInt16(0xffd6)); // -42
data.push_back(deInt16(0xffff)); // -1
DE_ASSERT(count >= data.size());
for (deUint32 numNdx = static_cast<deUint32>(data.size()); numNdx < count; ++numNdx)
data.push_back(static_cast<deInt16>(rnd.getUint16()));
return data;
}
// IEEE-754 floating point numbers:
// +--------+------+----------+-------------+
// | binary | sign | exponent | significand |
// +--------+------+----------+-------------+
// | 16-bit | 1 | 5 | 10 |
// +--------+------+----------+-------------+
// | 32-bit | 1 | 8 | 23 |
// +--------+------+----------+-------------+
//
// 16-bit floats:
//
// 0 000 00 00 0000 0001 (0x0001: 2e-24: minimum positive denormalized)
// 0 000 00 11 1111 1111 (0x03ff: 2e-14 - 2e-24: maximum positive denormalized)
// 0 000 01 00 0000 0000 (0x0400: 2e-14: minimum positive normalized)
//
// 32-bit floats:
//
// 0 011 1110 1 001 0000 0000 0000 0000 0000 (0x3e900000: 0.28125: with exact match in 16-bit normalized)
// 0 011 1000 1 000 0000 0011 0000 0000 0000 (0x38803000: exact half way within two 16-bit normalized; round to zero: 0x0401)
// 1 011 1000 1 000 0000 0011 0000 0000 0000 (0xb8803000: exact half way within two 16-bit normalized; round to zero: 0x8402)
// 0 011 1000 1 000 0000 1111 1111 0000 0000 (0x3880ff00: not exact half way within two 16-bit normalized; round to zero: 0x0403)
// 1 011 1000 1 000 0000 1111 1111 0000 0000 (0xb880ff00: not exact half way within two 16-bit normalized; round to zero: 0x8404)
// Generate and return 32-bit floats
//
// The first 24 number pairs are manually picked, while the rest are randomly generated.
// Expected count to be at least 24 (numPicks).
vector<float> getFloat32s (de::Random& rnd, deUint32 count)
{
vector<float> float32;
float32.reserve(count);
// Zero
float32.push_back(0.f);
float32.push_back(-0.f);
// Infinity
float32.push_back(std::numeric_limits<float>::infinity());
float32.push_back(-std::numeric_limits<float>::infinity());
// SNaN
float32.push_back(std::numeric_limits<float>::signaling_NaN());
float32.push_back(-std::numeric_limits<float>::signaling_NaN());
// QNaN
float32.push_back(std::numeric_limits<float>::quiet_NaN());
float32.push_back(-std::numeric_limits<float>::quiet_NaN());
// Denormalized 32-bit float matching 0 in 16-bit
float32.push_back(deFloatLdExp(1.f, -127));
float32.push_back(-deFloatLdExp(1.f, -127));
// Normalized 32-bit float matching 0 in 16-bit
float32.push_back(deFloatLdExp(1.f, -100));
float32.push_back(-deFloatLdExp(1.f, -100));
// Normalized 32-bit float with exact denormalized match in 16-bit
float32.push_back(deFloatLdExp(1.f, -24)); // 2e-24: minimum 16-bit positive denormalized
float32.push_back(-deFloatLdExp(1.f, -24)); // 2e-24: maximum 16-bit negative denormalized
// Normalized 32-bit float with exact normalized match in 16-bit
float32.push_back(deFloatLdExp(1.f, -14)); // 2e-14: minimum 16-bit positive normalized
float32.push_back(-deFloatLdExp(1.f, -14)); // 2e-14: maximum 16-bit negative normalized
// Normalized 32-bit float falling above half way within two 16-bit normalized
float32.push_back(bitwiseCast<float>(deUint32(0x3880ff00)));
float32.push_back(bitwiseCast<float>(deUint32(0xb880ff00)));
// Normalized 32-bit float falling exact half way within two 16-bit normalized
float32.push_back(bitwiseCast<float>(deUint32(0x38803000)));
float32.push_back(bitwiseCast<float>(deUint32(0xb8803000)));
// Some number
float32.push_back(0.28125f);
float32.push_back(-0.28125f);
// Normalized 32-bit float matching infinity in 16-bit
float32.push_back(deFloatLdExp(1.f, 100));
float32.push_back(-deFloatLdExp(1.f, 100));
const deUint32 numPicks = static_cast<deUint32>(float32.size());
DE_ASSERT(count >= numPicks);
count -= numPicks;
for (deUint32 numNdx = 0; numNdx < count; ++numNdx)
float32.push_back(rnd.getFloat());
return float32;
}
// IEEE-754 floating point numbers:
// +--------+------+----------+-------------+
// | binary | sign | exponent | significand |
// +--------+------+----------+-------------+
// | 16-bit | 1 | 5 | 10 |
// +--------+------+----------+-------------+
// | 32-bit | 1 | 8 | 23 |
// +--------+------+----------+-------------+
//
// 16-bit floats:
//
// 0 000 00 00 0000 0001 (0x0001: 2e-24: minimum positive denormalized)
// 0 000 00 11 1111 1111 (0x03ff: 2e-14 - 2e-24: maximum positive denormalized)
// 0 000 01 00 0000 0000 (0x0400: 2e-14: minimum positive normalized)
//
// 0 000 00 00 0000 0000 (0x0000: +0)
// 0 111 11 00 0000 0000 (0x7c00: +Inf)
// 0 000 00 11 1111 0000 (0x03f0: +Denorm)
// 0 000 01 00 0000 0001 (0x0401: +Norm)
// 0 111 11 00 0000 1111 (0x7c0f: +SNaN)
// 0 111 11 00 1111 0000 (0x7c0f: +QNaN)
// Generate and return 16-bit floats and their corresponding 32-bit values.
//
// The first 14 number pairs are manually picked, while the rest are randomly generated.
// Expected count to be at least 14 (numPicks).
vector<deFloat16> getFloat16s (de::Random& rnd, deUint32 count)
{
vector<deFloat16> float16;
float16.reserve(count);
// Zero
float16.push_back(deUint16(0x0000));
float16.push_back(deUint16(0x8000));
// Infinity
float16.push_back(deUint16(0x7c00));
float16.push_back(deUint16(0xfc00));
// SNaN
float16.push_back(deUint16(0x7c0f));
float16.push_back(deUint16(0xfc0f));
// QNaN
float16.push_back(deUint16(0x7cf0));
float16.push_back(deUint16(0xfcf0));
// Denormalized
float16.push_back(deUint16(0x03f0));
float16.push_back(deUint16(0x83f0));
// Normalized
float16.push_back(deUint16(0x0401));
float16.push_back(deUint16(0x8401));
// Some normal number
float16.push_back(deUint16(0x14cb));
float16.push_back(deUint16(0x94cb));
const deUint32 numPicks = static_cast<deUint32>(float16.size());
DE_ASSERT(count >= numPicks);
count -= numPicks;
for (deUint32 numIdx = 0; numIdx < count; ++numIdx)
float16.push_back(rnd.getUint16());
return float16;
}
void addCompute16bitStorageUniform16To32Group (tcu::TestCaseGroup* group)
{
tcu::TestContext& testCtx = group->getTestContext();
de::Random rnd (deStringHash(group->getName()));
const int numElements = 128;
const StringTemplate shaderTemplate (
"OpCapability Shader\n"
"OpCapability ${capability}\n"
"OpExtension \"SPV_KHR_16bit_storage\"\n"
"OpMemoryModel Logical GLSL450\n"
"OpEntryPoint GLCompute %main \"main\" %id\n"
"OpExecutionMode %main LocalSize 1 1 1\n"
"OpDecorate %id BuiltIn GlobalInvocationId\n"
"${stride}"
"OpMemberDecorate %SSBO32 0 Offset 0\n"
"OpMemberDecorate %SSBO16 0 Offset 0\n"
"OpDecorate %SSBO32 BufferBlock\n"
"OpDecorate %SSBO16 ${storage}\n"
"OpDecorate %ssbo32 DescriptorSet 0\n"
"OpDecorate %ssbo16 DescriptorSet 0\n"
"OpDecorate %ssbo32 Binding 1\n"
"OpDecorate %ssbo16 Binding 0\n"
"${matrix_decor:opt}\n"
"%bool = OpTypeBool\n"
"%void = OpTypeVoid\n"
"%voidf = OpTypeFunction %void\n"
"%u32 = OpTypeInt 32 0\n"
"%i32 = OpTypeInt 32 1\n"
"%f32 = OpTypeFloat 32\n"
"%uvec3 = OpTypeVector %u32 3\n"
"%fvec3 = OpTypeVector %f32 3\n"
"%uvec3ptr = OpTypePointer Input %uvec3\n"
"%i32ptr = OpTypePointer Uniform %i32\n"
"%f32ptr = OpTypePointer Uniform %f32\n"
"%zero = OpConstant %i32 0\n"
"%c_i32_1 = OpConstant %i32 1\n"
"%c_i32_2 = OpConstant %i32 2\n"
"%c_i32_3 = OpConstant %i32 3\n"
"%c_i32_16 = OpConstant %i32 16\n"
"%c_i32_32 = OpConstant %i32 32\n"
"%c_i32_64 = OpConstant %i32 64\n"
"%c_i32_128 = OpConstant %i32 128\n"
"%i32arr = OpTypeArray %i32 %c_i32_128\n"
"%f32arr = OpTypeArray %f32 %c_i32_128\n"
"${types}\n"
"${matrix_types:opt}\n"
"%SSBO32 = OpTypeStruct %${matrix_prefix:opt}${base32}arr\n"
"%SSBO16 = OpTypeStruct %${matrix_prefix:opt}${base16}arr\n"
"%up_SSBO32 = OpTypePointer Uniform %SSBO32\n"
"%up_SSBO16 = OpTypePointer Uniform %SSBO16\n"
"%ssbo32 = OpVariable %up_SSBO32 Uniform\n"
"%ssbo16 = OpVariable %up_SSBO16 Uniform\n"
"%id = OpVariable %uvec3ptr Input\n"
"%main = OpFunction %void None %voidf\n"
"%label = OpLabel\n"
"%idval = OpLoad %uvec3 %id\n"
"%x = OpCompositeExtract %u32 %idval 0\n"
"%inloc = OpAccessChain %${base16}ptr %ssbo16 %zero %x ${index0:opt}\n"
"%val16 = OpLoad %${base16} %inloc\n"
"%val32 = ${convert} %${base32} %val16\n"
"%outloc = OpAccessChain %${base32}ptr %ssbo32 %zero %x ${index0:opt}\n"
" OpStore %outloc %val32\n"
"${matrix_store:opt}\n"
" OpReturn\n"
" OpFunctionEnd\n");
{ // floats
const char floatTypes[] =
"%f16 = OpTypeFloat 16\n"
"%f16ptr = OpTypePointer Uniform %f16\n"
"%f16arr = OpTypeArray %f16 %c_i32_128\n"
"%v2f16 = OpTypeVector %f16 2\n"
"%v2f32 = OpTypeVector %f32 2\n"
"%v2f16ptr = OpTypePointer Uniform %v2f16\n"
"%v2f32ptr = OpTypePointer Uniform %v2f32\n"
"%v2f16arr = OpTypeArray %v2f16 %c_i32_64\n"
"%v2f32arr = OpTypeArray %v2f32 %c_i32_64\n";
struct CompositeType
{
const char* name;
const char* base32;
const char* base16;
const char* stride;
unsigned count;
};
const CompositeType cTypes[] =
{
{"scalar", "f32", "f16", "OpDecorate %f32arr ArrayStride 4\nOpDecorate %f16arr ArrayStride 2\n", numElements},
{"vector", "v2f32", "v2f16", "OpDecorate %v2f32arr ArrayStride 8\nOpDecorate %v2f16arr ArrayStride 4\n", numElements / 2},
{"matrix", "v2f32", "v2f16", "OpDecorate %m4v2f32arr ArrayStride 32\nOpDecorate %m4v2f16arr ArrayStride 16\n", numElements / 8},
};
vector<deFloat16> float16Data = getFloat16s(rnd, numElements);
vector<float> float32Data;
float32Data.reserve(numElements);
for (deUint32 numIdx = 0; numIdx < numElements; ++numIdx)
float32Data.push_back(deFloat16To32(float16Data[numIdx]));
for (deUint32 capIdx = 0; capIdx < DE_LENGTH_OF_ARRAY(CAPABILITIES); ++capIdx)
for (deUint32 tyIdx = 0; tyIdx < DE_LENGTH_OF_ARRAY(cTypes); ++tyIdx)
{
ComputeShaderSpec spec;
map<string, string> specs;
string testName = string(CAPABILITIES[capIdx].name) + "_" + cTypes[tyIdx].name + "_float";
specs["capability"] = CAPABILITIES[capIdx].cap;
specs["storage"] = CAPABILITIES[capIdx].decor;
specs["stride"] = cTypes[tyIdx].stride;
specs["base32"] = cTypes[tyIdx].base32;
specs["base16"] = cTypes[tyIdx].base16;
specs["types"] = floatTypes;
specs["convert"] = "OpFConvert";
if (strcmp(cTypes[tyIdx].name, "matrix") == 0)
{
specs["index0"] = "%zero";
specs["matrix_prefix"] = "m4";
specs["matrix_types"] =
"%m4v2f16 = OpTypeMatrix %v2f16 4\n"
"%m4v2f32 = OpTypeMatrix %v2f32 4\n"
"%m4v2f16arr = OpTypeArray %m4v2f16 %c_i32_16\n"
"%m4v2f32arr = OpTypeArray %m4v2f32 %c_i32_16\n";
specs["matrix_decor"] =
"OpMemberDecorate %SSBO32 0 ColMajor\n"
"OpMemberDecorate %SSBO32 0 MatrixStride 8\n"
"OpMemberDecorate %SSBO16 0 ColMajor\n"
"OpMemberDecorate %SSBO16 0 MatrixStride 4\n";
specs["matrix_store"] =
"%inloc_1 = OpAccessChain %v2f16ptr %ssbo16 %zero %x %c_i32_1\n"
"%val16_1 = OpLoad %v2f16 %inloc_1\n"
"%val32_1 = OpFConvert %v2f32 %val16_1\n"
"%outloc_1 = OpAccessChain %v2f32ptr %ssbo32 %zero %x %c_i32_1\n"
" OpStore %outloc_1 %val32_1\n"
"%inloc_2 = OpAccessChain %v2f16ptr %ssbo16 %zero %x %c_i32_2\n"
"%val16_2 = OpLoad %v2f16 %inloc_2\n"
"%val32_2 = OpFConvert %v2f32 %val16_2\n"
"%outloc_2 = OpAccessChain %v2f32ptr %ssbo32 %zero %x %c_i32_2\n"
" OpStore %outloc_2 %val32_2\n"
"%inloc_3 = OpAccessChain %v2f16ptr %ssbo16 %zero %x %c_i32_3\n"
"%val16_3 = OpLoad %v2f16 %inloc_3\n"
"%val32_3 = OpFConvert %v2f32 %val16_3\n"
"%outloc_3 = OpAccessChain %v2f32ptr %ssbo32 %zero %x %c_i32_3\n"
" OpStore %outloc_3 %val32_3\n";
}
spec.assembly = shaderTemplate.specialize(specs);
spec.numWorkGroups = IVec3(cTypes[tyIdx].count, 1, 1);
spec.verifyIO = check32BitFloats;
spec.inputTypes[0] = CAPABILITIES[capIdx].dtype;
spec.inputs.push_back(BufferSp(new Float16Buffer(float16Data)));
spec.outputs.push_back(BufferSp(new Float32Buffer(float32Data)));
spec.extensions.push_back("VK_KHR_16bit_storage");
spec.requestedVulkanFeatures = get16BitStorageFeatures(CAPABILITIES[capIdx].name);
group->addChild(new SpvAsmComputeShaderCase(testCtx, testName.c_str(), testName.c_str(), spec));
}
}
{ // Integers
const char sintTypes[] =
"%i16 = OpTypeInt 16 1\n"
"%i16ptr = OpTypePointer Uniform %i16\n"
"%i16arr = OpTypeArray %i16 %c_i32_128\n"
"%v4i16 = OpTypeVector %i16 4\n"
"%v4i32 = OpTypeVector %i32 4\n"
"%v4i16ptr = OpTypePointer Uniform %v4i16\n"
"%v4i32ptr = OpTypePointer Uniform %v4i32\n"
"%v4i16arr = OpTypeArray %v4i16 %c_i32_32\n"
"%v4i32arr = OpTypeArray %v4i32 %c_i32_32\n";
const char uintTypes[] =
"%u16 = OpTypeInt 16 0\n"
"%u16ptr = OpTypePointer Uniform %u16\n"
"%u32ptr = OpTypePointer Uniform %u32\n"
"%u16arr = OpTypeArray %u16 %c_i32_128\n"
"%u32arr = OpTypeArray %u32 %c_i32_128\n"
"%v4u16 = OpTypeVector %u16 4\n"
"%v4u32 = OpTypeVector %u32 4\n"
"%v4u16ptr = OpTypePointer Uniform %v4u16\n"
"%v4u32ptr = OpTypePointer Uniform %v4u32\n"
"%v4u16arr = OpTypeArray %v4u16 %c_i32_32\n"
"%v4u32arr = OpTypeArray %v4u32 %c_i32_32\n";
struct CompositeType
{
const char* name;
bool isSigned;
const char* types;
const char* base32;
const char* base16;
const char* opcode;
const char* stride;
unsigned count;
};
const CompositeType cTypes[] =
{
{"scalar_sint", true, sintTypes, "i32", "i16", "OpSConvert", "OpDecorate %i32arr ArrayStride 4\nOpDecorate %i16arr ArrayStride 2\n", numElements},
{"scalar_uint", false, uintTypes, "u32", "u16", "OpUConvert", "OpDecorate %u32arr ArrayStride 4\nOpDecorate %u16arr ArrayStride 2\n", numElements},
{"vector_sint", true, sintTypes, "v4i32", "v4i16", "OpSConvert", "OpDecorate %v4i32arr ArrayStride 16\nOpDecorate %v4i16arr ArrayStride 8\n", numElements / 4},
{"vector_uint", false, uintTypes, "v4u32", "v4u16", "OpUConvert", "OpDecorate %v4u32arr ArrayStride 16\nOpDecorate %v4u16arr ArrayStride 8\n", numElements / 4},
};
vector<deInt16> inputs = getInt16s(rnd, numElements);
vector<deInt32> sOutputs;
vector<deInt32> uOutputs;
const deUint16 signBitMask = 0x8000;
const deUint32 signExtendMask = 0xffff0000;
sOutputs.reserve(inputs.size());
uOutputs.reserve(inputs.size());
for (deUint32 numNdx = 0; numNdx < inputs.size(); ++numNdx)
{
uOutputs.push_back(static_cast<deUint16>(inputs[numNdx]));
if (inputs[numNdx] & signBitMask)
sOutputs.push_back(static_cast<deInt32>(inputs[numNdx] | signExtendMask));
else
sOutputs.push_back(static_cast<deInt32>(inputs[numNdx]));
}
for (deUint32 capIdx = 0; capIdx < DE_LENGTH_OF_ARRAY(CAPABILITIES); ++capIdx)
for (deUint32 tyIdx = 0; tyIdx < DE_LENGTH_OF_ARRAY(cTypes); ++tyIdx)
{
ComputeShaderSpec spec;
map<string, string> specs;
string testName = string(CAPABILITIES[capIdx].name) + "_" + cTypes[tyIdx].name;
specs["capability"] = CAPABILITIES[capIdx].cap;
specs["storage"] = CAPABILITIES[capIdx].decor;
specs["stride"] = cTypes[tyIdx].stride;
specs["base32"] = cTypes[tyIdx].base32;
specs["base16"] = cTypes[tyIdx].base16;
specs["types"] = cTypes[tyIdx].types;
specs["convert"] = cTypes[tyIdx].opcode;
spec.assembly = shaderTemplate.specialize(specs);
spec.numWorkGroups = IVec3(cTypes[tyIdx].count, 1, 1);
spec.inputTypes[0] = CAPABILITIES[capIdx].dtype;
spec.inputs.push_back(BufferSp(new Int16Buffer(inputs)));
if (cTypes[tyIdx].isSigned)
spec.outputs.push_back(BufferSp(new Int32Buffer(sOutputs)));
else
spec.outputs.push_back(BufferSp(new Int32Buffer(uOutputs)));
spec.extensions.push_back("VK_KHR_16bit_storage");
spec.requestedVulkanFeatures = get16BitStorageFeatures(CAPABILITIES[capIdx].name);
group->addChild(new SpvAsmComputeShaderCase(testCtx, testName.c_str(), testName.c_str(), spec));
}
}
}
void addCompute16bitStoragePushConstant16To32Group (tcu::TestCaseGroup* group)
{
tcu::TestContext& testCtx = group->getTestContext();
de::Random rnd (deStringHash(group->getName()));
const int numElements = 64;
const StringTemplate shaderTemplate (
"OpCapability Shader\n"
"OpCapability StoragePushConstant16\n"
"OpExtension \"SPV_KHR_16bit_storage\"\n"
"OpMemoryModel Logical GLSL450\n"
"OpEntryPoint GLCompute %main \"main\" %id\n"
"OpExecutionMode %main LocalSize 1 1 1\n"
"OpDecorate %id BuiltIn GlobalInvocationId\n"
"${stride}"
"OpDecorate %PC16 Block\n"
"OpMemberDecorate %PC16 0 Offset 0\n"
"OpMemberDecorate %SSBO32 0 Offset 0\n"
"OpDecorate %SSBO32 BufferBlock\n"
"OpDecorate %ssbo32 DescriptorSet 0\n"
"OpDecorate %ssbo32 Binding 0\n"
"${matrix_decor:opt}\n"
"%bool = OpTypeBool\n"
"%void = OpTypeVoid\n"
"%voidf = OpTypeFunction %void\n"
"%u32 = OpTypeInt 32 0\n"
"%i32 = OpTypeInt 32 1\n"
"%f32 = OpTypeFloat 32\n"
"%uvec3 = OpTypeVector %u32 3\n"
"%fvec3 = OpTypeVector %f32 3\n"
"%uvec3ptr = OpTypePointer Input %uvec3\n"
"%i32ptr = OpTypePointer Uniform %i32\n"
"%f32ptr = OpTypePointer Uniform %f32\n"
"%zero = OpConstant %i32 0\n"
"%c_i32_1 = OpConstant %i32 1\n"
"%c_i32_8 = OpConstant %i32 8\n"
"%c_i32_16 = OpConstant %i32 16\n"
"%c_i32_32 = OpConstant %i32 32\n"
"%c_i32_64 = OpConstant %i32 64\n"
"%i32arr = OpTypeArray %i32 %c_i32_64\n"
"%f32arr = OpTypeArray %f32 %c_i32_64\n"
"${types}\n"
"${matrix_types:opt}\n"
"%PC16 = OpTypeStruct %${matrix_prefix:opt}${base16}arr\n"
"%pp_PC16 = OpTypePointer PushConstant %PC16\n"
"%pc16 = OpVariable %pp_PC16 PushConstant\n"
"%SSBO32 = OpTypeStruct %${matrix_prefix:opt}${base32}arr\n"
"%up_SSBO32 = OpTypePointer Uniform %SSBO32\n"
"%ssbo32 = OpVariable %up_SSBO32 Uniform\n"
"%id = OpVariable %uvec3ptr Input\n"
"%main = OpFunction %void None %voidf\n"
"%label = OpLabel\n"
"%idval = OpLoad %uvec3 %id\n"
"%x = OpCompositeExtract %u32 %idval 0\n"
"%inloc = OpAccessChain %${base16}ptr %pc16 %zero %x ${index0:opt}\n"
"%val16 = OpLoad %${base16} %inloc\n"
"%val32 = ${convert} %${base32} %val16\n"
"%outloc = OpAccessChain %${base32}ptr %ssbo32 %zero %x ${index0:opt}\n"
" OpStore %outloc %val32\n"
"${matrix_store:opt}\n"
" OpReturn\n"
" OpFunctionEnd\n");
{ // floats
const char floatTypes[] =
"%f16 = OpTypeFloat 16\n"
"%f16ptr = OpTypePointer PushConstant %f16\n"
"%f16arr = OpTypeArray %f16 %c_i32_64\n"
"%v4f16 = OpTypeVector %f16 4\n"
"%v4f32 = OpTypeVector %f32 4\n"
"%v4f16ptr = OpTypePointer PushConstant %v4f16\n"
"%v4f32ptr = OpTypePointer Uniform %v4f32\n"
"%v4f16arr = OpTypeArray %v4f16 %c_i32_16\n"
"%v4f32arr = OpTypeArray %v4f32 %c_i32_16\n";
struct CompositeType
{
const char* name;
const char* base32;
const char* base16;
const char* stride;
unsigned count;
};
const CompositeType cTypes[] =
{
{"scalar", "f32", "f16", "OpDecorate %f32arr ArrayStride 4\nOpDecorate %f16arr ArrayStride 2\n", numElements},
{"vector", "v4f32", "v4f16", "OpDecorate %v4f32arr ArrayStride 16\nOpDecorate %v4f16arr ArrayStride 8\n", numElements / 4},
{"matrix", "v4f32", "v4f16", "OpDecorate %m2v4f32arr ArrayStride 32\nOpDecorate %m2v4f16arr ArrayStride 16\n", numElements / 8},
};
vector<deFloat16> float16Data = getFloat16s(rnd, numElements);
vector<float> float32Data;
float32Data.reserve(numElements);
for (deUint32 numIdx = 0; numIdx < numElements; ++numIdx)
float32Data.push_back(deFloat16To32(float16Data[numIdx]));
for (deUint32 tyIdx = 0; tyIdx < DE_LENGTH_OF_ARRAY(cTypes); ++tyIdx)
{
ComputeShaderSpec spec;
map<string, string> specs;
string testName = string(cTypes[tyIdx].name) + "_float";
specs["stride"] = cTypes[tyIdx].stride;
specs["base32"] = cTypes[tyIdx].base32;
specs["base16"] = cTypes[tyIdx].base16;
specs["types"] = floatTypes;
specs["convert"] = "OpFConvert";
if (strcmp(cTypes[tyIdx].name, "matrix") == 0)
{
specs["index0"] = "%zero";
specs["matrix_prefix"] = "m2";
specs["matrix_types"] =
"%m2v4f16 = OpTypeMatrix %v4f16 2\n"
"%m2v4f32 = OpTypeMatrix %v4f32 2\n"
"%m2v4f16arr = OpTypeArray %m2v4f16 %c_i32_8\n"
"%m2v4f32arr = OpTypeArray %m2v4f32 %c_i32_8\n";
specs["matrix_decor"] =
"OpMemberDecorate %SSBO32 0 ColMajor\n"
"OpMemberDecorate %SSBO32 0 MatrixStride 16\n"
"OpMemberDecorate %PC16 0 ColMajor\n"
"OpMemberDecorate %PC16 0 MatrixStride 8\n";
specs["matrix_store"] =
"%inloc_1 = OpAccessChain %v4f16ptr %pc16 %zero %x %c_i32_1\n"
"%val16_1 = OpLoad %v4f16 %inloc_1\n"
"%val32_1 = OpFConvert %v4f32 %val16_1\n"
"%outloc_1 = OpAccessChain %v4f32ptr %ssbo32 %zero %x %c_i32_1\n"
" OpStore %outloc_1 %val32_1\n";
}
spec.assembly = shaderTemplate.specialize(specs);
spec.numWorkGroups = IVec3(cTypes[tyIdx].count, 1, 1);
spec.verifyIO = check32BitFloats;
spec.pushConstants = BufferSp(new Float16Buffer(float16Data));
spec.outputs.push_back(BufferSp(new Float32Buffer(float32Data)));
spec.extensions.push_back("VK_KHR_16bit_storage");
spec.requestedVulkanFeatures.ext16BitStorage = EXT16BITSTORAGEFEATURES_PUSH_CONSTANT;
group->addChild(new SpvAsmComputeShaderCase(testCtx, testName.c_str(), testName.c_str(), spec));
}
}
{ // integers
const char sintTypes[] =
"%i16 = OpTypeInt 16 1\n"
"%i16ptr = OpTypePointer PushConstant %i16\n"
"%i16arr = OpTypeArray %i16 %c_i32_64\n"
"%v2i16 = OpTypeVector %i16 2\n"
"%v2i32 = OpTypeVector %i32 2\n"
"%v2i16ptr = OpTypePointer PushConstant %v2i16\n"
"%v2i32ptr = OpTypePointer Uniform %v2i32\n"
"%v2i16arr = OpTypeArray %v2i16 %c_i32_32\n"
"%v2i32arr = OpTypeArray %v2i32 %c_i32_32\n";
const char uintTypes[] =
"%u16 = OpTypeInt 16 0\n"
"%u16ptr = OpTypePointer PushConstant %u16\n"
"%u32ptr = OpTypePointer Uniform %u32\n"
"%u16arr = OpTypeArray %u16 %c_i32_64\n"
"%u32arr = OpTypeArray %u32 %c_i32_64\n"
"%v2u16 = OpTypeVector %u16 2\n"
"%v2u32 = OpTypeVector %u32 2\n"
"%v2u16ptr = OpTypePointer PushConstant %v2u16\n"
"%v2u32ptr = OpTypePointer Uniform %v2u32\n"
"%v2u16arr = OpTypeArray %v2u16 %c_i32_32\n"
"%v2u32arr = OpTypeArray %v2u32 %c_i32_32\n";
struct CompositeType
{
const char* name;
bool isSigned;
const char* types;
const char* base32;
const char* base16;
const char* opcode;
const char* stride;
unsigned count;
};
const CompositeType cTypes[] =
{
{"scalar_sint", true, sintTypes, "i32", "i16", "OpSConvert", "OpDecorate %i32arr ArrayStride 4\nOpDecorate %i16arr ArrayStride 2\n", numElements},
{"scalar_uint", false, uintTypes, "u32", "u16", "OpUConvert", "OpDecorate %u32arr ArrayStride 4\nOpDecorate %u16arr ArrayStride 2\n", numElements},
{"vector_sint", true, sintTypes, "v2i32", "v2i16", "OpSConvert", "OpDecorate %v2i32arr ArrayStride 8\nOpDecorate %v2i16arr ArrayStride 4\n", numElements / 2},
{"vector_uint", false, uintTypes, "v2u32", "v2u16", "OpUConvert", "OpDecorate %v2u32arr ArrayStride 8\nOpDecorate %v2u16arr ArrayStride 4\n", numElements / 2},
};
vector<deInt16> inputs = getInt16s(rnd, numElements);
vector<deInt32> sOutputs;
vector<deInt32> uOutputs;
const deUint16 signBitMask = 0x8000;
const deUint32 signExtendMask = 0xffff0000;
sOutputs.reserve(inputs.size());
uOutputs.reserve(inputs.size());
for (deUint32 numNdx = 0; numNdx < inputs.size(); ++numNdx)
{
uOutputs.push_back(static_cast<deUint16>(inputs[numNdx]));
if (inputs[numNdx] & signBitMask)
sOutputs.push_back(static_cast<deInt32>(inputs[numNdx] | signExtendMask));
else
sOutputs.push_back(static_cast<deInt32>(inputs[numNdx]));
}
for (deUint32 tyIdx = 0; tyIdx < DE_LENGTH_OF_ARRAY(cTypes); ++tyIdx)
{
ComputeShaderSpec spec;
map<string, string> specs;
const char* testName = cTypes[tyIdx].name;
specs["stride"] = cTypes[tyIdx].stride;
specs["base32"] = cTypes[tyIdx].base32;
specs["base16"] = cTypes[tyIdx].base16;
specs["types"] = cTypes[tyIdx].types;
specs["convert"] = cTypes[tyIdx].opcode;
spec.assembly = shaderTemplate.specialize(specs);
spec.numWorkGroups = IVec3(cTypes[tyIdx].count, 1, 1);
spec.pushConstants = BufferSp(new Int16Buffer(inputs));
if (cTypes[tyIdx].isSigned)
spec.outputs.push_back(BufferSp(new Int32Buffer(sOutputs)));
else
spec.outputs.push_back(BufferSp(new Int32Buffer(uOutputs)));
spec.extensions.push_back("VK_KHR_16bit_storage");
spec.requestedVulkanFeatures.ext16BitStorage = EXT16BITSTORAGEFEATURES_PUSH_CONSTANT;
group->addChild(new SpvAsmComputeShaderCase(testCtx, testName, testName, spec));
}
}
}
void addGraphics16BitStorageUniformInt32To16Group (tcu::TestCaseGroup* testGroup)
{
de::Random rnd (deStringHash(testGroup->getName()));
map<string, string> fragments;
const deUint32 numDataPoints = 256;
RGBA defaultColors[4];
GraphicsResources resources;
vector<string> extensions;
const StringTemplate capabilities ("OpCapability ${cap}\n");
// inputs and outputs are declared to be vectors of signed integers.
// However, depending on the test, they may be interpreted as unsiged
// integers. That won't be a problem as long as we passed the bits
// in faithfully to the pipeline.
vector<deInt32> inputs = getInt32s(rnd, numDataPoints);
vector<deInt16> outputs;
outputs.reserve(inputs.size());
for (deUint32 numNdx = 0; numNdx < inputs.size(); ++numNdx)
outputs.push_back(static_cast<deInt16>(0xffff & inputs[numNdx]));
resources.inputs.push_back(std::make_pair(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, BufferSp(new Int32Buffer(inputs))));
resources.outputs.push_back(std::make_pair(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, BufferSp(new Int16Buffer(outputs))));
extensions.push_back("VK_KHR_16bit_storage");
fragments["extension"] = "OpExtension \"SPV_KHR_16bit_storage\"";
getDefaultColors(defaultColors);
struct IntegerFacts
{
const char* name;
const char* type32;
const char* type16;
const char* opcode;
const char* isSigned;
};
const IntegerFacts intFacts[] =
{
{"sint", "%i32", "%i16", "OpSConvert", "1"},
{"uint", "%u32", "%u16", "OpUConvert", "0"},
};
const StringTemplate scalarPreMain(
"${itype16} = OpTypeInt 16 ${signed}\n"
"%c_i32_256 = OpConstant %i32 256\n"
" %up_i32 = OpTypePointer Uniform ${itype32}\n"
" %up_i16 = OpTypePointer Uniform ${itype16}\n"
" %ra_i32 = OpTypeArray ${itype32} %c_i32_256\n"
" %ra_i16 = OpTypeArray ${itype16} %c_i32_256\n"
" %SSBO32 = OpTypeStruct %ra_i32\n"
" %SSBO16 = OpTypeStruct %ra_i16\n"
"%up_SSBO32 = OpTypePointer Uniform %SSBO32\n"
"%up_SSBO16 = OpTypePointer Uniform %SSBO16\n"
" %ssbo32 = OpVariable %up_SSBO32 Uniform\n"
" %ssbo16 = OpVariable %up_SSBO16 Uniform\n");
const StringTemplate scalarDecoration(
"OpDecorate %ra_i32 ArrayStride 4\n"
"OpDecorate %ra_i16 ArrayStride 2\n"
"OpMemberDecorate %SSBO32 0 Offset 0\n"
"OpMemberDecorate %SSBO16 0 Offset 0\n"
"OpDecorate %SSBO32 ${indecor}\n"
"OpDecorate %SSBO16 BufferBlock\n"
"OpDecorate %ssbo32 DescriptorSet 0\n"
"OpDecorate %ssbo16 DescriptorSet 0\n"
"OpDecorate %ssbo32 Binding 0\n"
"OpDecorate %ssbo16 Binding 1\n");
const StringTemplate scalarTestFunc(
"%test_code = OpFunction %v4f32 None %v4f32_function\n"
" %param = OpFunctionParameter %v4f32\n"
"%entry = OpLabel\n"
" %i = OpVariable %fp_i32 Function\n"
" OpStore %i %c_i32_0\n"
" OpBranch %loop\n"
" %loop = OpLabel\n"
" %15 = OpLoad %i32 %i\n"
" %lt = OpSLessThan %bool %15 %c_i32_256\n"
" OpLoopMerge %merge %inc None\n"
" OpBranchConditional %lt %write %merge\n"
"%write = OpLabel\n"
" %30 = OpLoad %i32 %i\n"
" %src = OpAccessChain %up_i32 %ssbo32 %c_i32_0 %30\n"
"%val32 = OpLoad ${itype32} %src\n"
"%val16 = ${convert} ${itype16} %val32\n"
" %dst = OpAccessChain %up_i16 %ssbo16 %c_i32_0 %30\n"
" OpStore %dst %val16\n"
" OpBranch %inc\n"
" %inc = OpLabel\n"
" %37 = OpLoad %i32 %i\n"
" %39 = OpIAdd %i32 %37 %c_i32_1\n"
" OpStore %i %39\n"
" OpBranch %loop\n"
"%merge = OpLabel\n"
" OpReturnValue %param\n"
"OpFunctionEnd\n");
const StringTemplate vecPreMain(
"${itype16} = OpTypeInt 16 ${signed}\n"
" %c_i32_64 = OpConstant %i32 64\n"
"%v4itype32 = OpTypeVector ${itype32} 4\n"
"%v4itype16 = OpTypeVector ${itype16} 4\n"
" %up_v4i32 = OpTypePointer Uniform %v4itype32\n"
" %up_v4i16 = OpTypePointer Uniform %v4itype16\n"
" %ra_v4i32 = OpTypeArray %v4itype32 %c_i32_64\n"
" %ra_v4i16 = OpTypeArray %v4itype16 %c_i32_64\n"
" %SSBO32 = OpTypeStruct %ra_v4i32\n"
" %SSBO16 = OpTypeStruct %ra_v4i16\n"
"%up_SSBO32 = OpTypePointer Uniform %SSBO32\n"
"%up_SSBO16 = OpTypePointer Uniform %SSBO16\n"
" %ssbo32 = OpVariable %up_SSBO32 Uniform\n"
" %ssbo16 = OpVariable %up_SSBO16 Uniform\n");
const StringTemplate vecDecoration(
"OpDecorate %ra_v4i32 ArrayStride 16\n"
"OpDecorate %ra_v4i16 ArrayStride 8\n"
"OpMemberDecorate %SSBO32 0 Offset 0\n"
"OpMemberDecorate %SSBO16 0 Offset 0\n"
"OpDecorate %SSBO32 ${indecor}\n"
"OpDecorate %SSBO16 BufferBlock\n"
"OpDecorate %ssbo32 DescriptorSet 0\n"
"OpDecorate %ssbo16 DescriptorSet 0\n"
"OpDecorate %ssbo32 Binding 0\n"
"OpDecorate %ssbo16 Binding 1\n");
const StringTemplate vecTestFunc(
"%test_code = OpFunction %v4f32 None %v4f32_function\n"
" %param = OpFunctionParameter %v4f32\n"
"%entry = OpLabel\n"
" %i = OpVariable %fp_i32 Function\n"
" OpStore %i %c_i32_0\n"
" OpBranch %loop\n"
" %loop = OpLabel\n"
" %15 = OpLoad %i32 %i\n"
" %lt = OpSLessThan %bool %15 %c_i32_64\n"
" OpLoopMerge %merge %inc None\n"
" OpBranchConditional %lt %write %merge\n"
"%write = OpLabel\n"
" %30 = OpLoad %i32 %i\n"
" %src = OpAccessChain %up_v4i32 %ssbo32 %c_i32_0 %30\n"
"%val32 = OpLoad %v4itype32 %src\n"
"%val16 = ${convert} %v4itype16 %val32\n"
" %dst = OpAccessChain %up_v4i16 %ssbo16 %c_i32_0 %30\n"
" OpStore %dst %val16\n"
" OpBranch %inc\n"
" %inc = OpLabel\n"
" %37 = OpLoad %i32 %i\n"
" %39 = OpIAdd %i32 %37 %c_i32_1\n"
" OpStore %i %39\n"
" OpBranch %loop\n"
"%merge = OpLabel\n"
" OpReturnValue %param\n"
"OpFunctionEnd\n");
struct Category
{
const char* name;
const StringTemplate& preMain;
const StringTemplate& decoration;
const StringTemplate& testFunction;
};
const Category categories[] =
{
{"scalar", scalarPreMain, scalarDecoration, scalarTestFunc},
{"vector", vecPreMain, vecDecoration, vecTestFunc},
};
for (deUint32 catIdx = 0; catIdx < DE_LENGTH_OF_ARRAY(categories); ++catIdx)
for (deUint32 capIdx = 0; capIdx < DE_LENGTH_OF_ARRAY(CAPABILITIES); ++capIdx)
for (deUint32 factIdx = 0; factIdx < DE_LENGTH_OF_ARRAY(intFacts); ++factIdx)
{
map<string, string> specs;
string name = string(CAPABILITIES[capIdx].name) + "_" + categories[catIdx].name + "_" + intFacts[factIdx].name;
specs["cap"] = CAPABILITIES[capIdx].cap;
specs["indecor"] = CAPABILITIES[capIdx].decor;
specs["itype32"] = intFacts[factIdx].type32;
specs["itype16"] = intFacts[factIdx].type16;
specs["signed"] = intFacts[factIdx].isSigned;
specs["convert"] = intFacts[factIdx].opcode;
fragments["pre_main"] = categories[catIdx].preMain.specialize(specs);
fragments["testfun"] = categories[catIdx].testFunction.specialize(specs);
fragments["capability"] = capabilities.specialize(specs);
fragments["decoration"] = categories[catIdx].decoration.specialize(specs);
resources.inputs.back().first = CAPABILITIES[capIdx].dtype;
createTestsForAllStages(name, defaultColors, defaultColors, fragments, resources, extensions, testGroup, get16BitStorageFeatures(CAPABILITIES[capIdx].name));
}
}
void addCompute16bitStorageUniform32To16Group (tcu::TestCaseGroup* group)
{
tcu::TestContext& testCtx = group->getTestContext();
de::Random rnd (deStringHash(group->getName()));
const int numElements = 128;
const StringTemplate shaderTemplate (
"OpCapability Shader\n"
"OpCapability ${capability}\n"
"OpExtension \"SPV_KHR_16bit_storage\"\n"
"OpMemoryModel Logical GLSL450\n"
"OpEntryPoint GLCompute %main \"main\" %id\n"
"OpExecutionMode %main LocalSize 1 1 1\n"
"OpDecorate %id BuiltIn GlobalInvocationId\n"
"${stride}"
"OpMemberDecorate %SSBO32 0 Offset 0\n"
"OpMemberDecorate %SSBO16 0 Offset 0\n"
"OpDecorate %SSBO32 ${storage}\n"
"OpDecorate %SSBO16 BufferBlock\n"
"OpDecorate %ssbo32 DescriptorSet 0\n"
"OpDecorate %ssbo16 DescriptorSet 0\n"
"OpDecorate %ssbo32 Binding 0\n"
"OpDecorate %ssbo16 Binding 1\n"
"${matrix_decor:opt}\n"
"${rounding:opt}\n"
"%bool = OpTypeBool\n"
"%void = OpTypeVoid\n"
"%voidf = OpTypeFunction %void\n"
"%u32 = OpTypeInt 32 0\n"
"%i32 = OpTypeInt 32 1\n"
"%f32 = OpTypeFloat 32\n"
"%uvec3 = OpTypeVector %u32 3\n"
"%fvec3 = OpTypeVector %f32 3\n"
"%uvec3ptr = OpTypePointer Input %uvec3\n"
"%i32ptr = OpTypePointer Uniform %i32\n"
"%f32ptr = OpTypePointer Uniform %f32\n"
"%zero = OpConstant %i32 0\n"
"%c_i32_1 = OpConstant %i32 1\n"
"%c_i32_16 = OpConstant %i32 16\n"
"%c_i32_32 = OpConstant %i32 32\n"
"%c_i32_64 = OpConstant %i32 64\n"
"%c_i32_128 = OpConstant %i32 128\n"
"%i32arr = OpTypeArray %i32 %c_i32_128\n"
"%f32arr = OpTypeArray %f32 %c_i32_128\n"
"${types}\n"
"${matrix_types:opt}\n"
"%SSBO32 = OpTypeStruct %${matrix_prefix:opt}${base32}arr\n"
"%SSBO16 = OpTypeStruct %${matrix_prefix:opt}${base16}arr\n"
"%up_SSBO32 = OpTypePointer Uniform %SSBO32\n"
"%up_SSBO16 = OpTypePointer Uniform %SSBO16\n"
"%ssbo32 = OpVariable %up_SSBO32 Uniform\n"
"%ssbo16 = OpVariable %up_SSBO16 Uniform\n"
"%id = OpVariable %uvec3ptr Input\n"
"%main = OpFunction %void None %voidf\n"
"%label = OpLabel\n"
"%idval = OpLoad %uvec3 %id\n"
"%x = OpCompositeExtract %u32 %idval 0\n"
"%inloc = OpAccessChain %${base32}ptr %ssbo32 %zero %x ${index0:opt}\n"
"%val32 = OpLoad %${base32} %inloc\n"
"%val16 = ${convert} %${base16} %val32\n"
"%outloc = OpAccessChain %${base16}ptr %ssbo16 %zero %x ${index0:opt}\n"
" OpStore %outloc %val16\n"
"${matrix_store:opt}\n"
" OpReturn\n"
" OpFunctionEnd\n");
{ // Floats
const char floatTypes[] =
"%f16 = OpTypeFloat 16\n"
"%f16ptr = OpTypePointer Uniform %f16\n"
"%f16arr = OpTypeArray %f16 %c_i32_128\n"
"%v4f16 = OpTypeVector %f16 4\n"
"%v4f32 = OpTypeVector %f32 4\n"
"%v4f16ptr = OpTypePointer Uniform %v4f16\n"
"%v4f32ptr = OpTypePointer Uniform %v4f32\n"
"%v4f16arr = OpTypeArray %v4f16 %c_i32_32\n"
"%v4f32arr = OpTypeArray %v4f32 %c_i32_32\n";
struct RndMode
{
const char* name;
const char* decor;
ComputeVerifyIOFunc func;
};
const RndMode rndModes[] =
{
{"rtz", "OpDecorate %val16 FPRoundingMode RTZ", computeCheck16BitFloats<ROUNDINGMODE_RTZ>},
{"rte", "OpDecorate %val16 FPRoundingMode RTE", computeCheck16BitFloats<ROUNDINGMODE_RTE>},
{"unspecified_rnd_mode", "", computeCheck16BitFloats<RoundingModeFlags(ROUNDINGMODE_RTE | ROUNDINGMODE_RTZ)>},
};
struct CompositeType
{
const char* name;
const char* base32;
const char* base16;
const char* stride;
unsigned count;
};
const CompositeType cTypes[] =
{
{"scalar", "f32", "f16", "OpDecorate %f32arr ArrayStride 4\nOpDecorate %f16arr ArrayStride 2\n", numElements},
{"vector", "v4f32", "v4f16", "OpDecorate %v4f32arr ArrayStride 16\nOpDecorate %v4f16arr ArrayStride 8\n", numElements / 4},
{"matrix", "v4f32", "v4f16", "OpDecorate %m2v4f32arr ArrayStride 32\nOpDecorate %m2v4f16arr ArrayStride 16\n", numElements / 8},
};
vector<float> float32Data = getFloat32s(rnd, numElements);
vector<deFloat16> float16DummyData (numElements, 0);
for (deUint32 capIdx = 0; capIdx < DE_LENGTH_OF_ARRAY(CAPABILITIES); ++capIdx)
for (deUint32 tyIdx = 0; tyIdx < DE_LENGTH_OF_ARRAY(cTypes); ++tyIdx)
for (deUint32 rndModeIdx = 0; rndModeIdx < DE_LENGTH_OF_ARRAY(rndModes); ++rndModeIdx)
{
ComputeShaderSpec spec;
map<string, string> specs;
string testName = string(CAPABILITIES[capIdx].name) + "_" + cTypes[tyIdx].name + "_float_" + rndModes[rndModeIdx].name;
specs["capability"] = CAPABILITIES[capIdx].cap;
specs["storage"] = CAPABILITIES[capIdx].decor;
specs["stride"] = cTypes[tyIdx].stride;
specs["base32"] = cTypes[tyIdx].base32;
specs["base16"] = cTypes[tyIdx].base16;
specs["rounding"] = rndModes[rndModeIdx].decor;
specs["types"] = floatTypes;
specs["convert"] = "OpFConvert";
if (strcmp(cTypes[tyIdx].name, "matrix") == 0)
{
if (strcmp(rndModes[rndModeIdx].name, "rtz") == 0)
specs["rounding"] += "\nOpDecorate %val16_1 FPRoundingMode RTZ\n";
else if (strcmp(rndModes[rndModeIdx].name, "rte") == 0)
specs["rounding"] += "\nOpDecorate %val16_1 FPRoundingMode RTE\n";
specs["index0"] = "%zero";
specs["matrix_prefix"] = "m2";
specs["matrix_types"] =
"%m2v4f16 = OpTypeMatrix %v4f16 2\n"
"%m2v4f32 = OpTypeMatrix %v4f32 2\n"
"%m2v4f16arr = OpTypeArray %m2v4f16 %c_i32_16\n"
"%m2v4f32arr = OpTypeArray %m2v4f32 %c_i32_16\n";
specs["matrix_decor"] =
"OpMemberDecorate %SSBO32 0 ColMajor\n"
"OpMemberDecorate %SSBO32 0 MatrixStride 16\n"
"OpMemberDecorate %SSBO16 0 ColMajor\n"
"OpMemberDecorate %SSBO16 0 MatrixStride 8\n";
specs["matrix_store"] =
"%inloc_1 = OpAccessChain %v4f32ptr %ssbo32 %zero %x %c_i32_1\n"
"%val32_1 = OpLoad %v4f32 %inloc_1\n"
"%val16_1 = OpFConvert %v4f16 %val32_1\n"
"%outloc_1 = OpAccessChain %v4f16ptr %ssbo16 %zero %x %c_i32_1\n"
" OpStore %outloc_1 %val16_1\n";
}
spec.assembly = shaderTemplate.specialize(specs);
spec.numWorkGroups = IVec3(cTypes[tyIdx].count, 1, 1);
spec.verifyIO = rndModes[rndModeIdx].func;
spec.inputTypes[0] = CAPABILITIES[capIdx].dtype;
spec.inputs.push_back(BufferSp(new Float32Buffer(float32Data)));
// We provided a custom verifyIO in the above in which inputs will be used for checking.
// So put dummy data in the expected values.
spec.outputs.push_back(BufferSp(new Float16Buffer(float16DummyData)));
spec.extensions.push_back("VK_KHR_16bit_storage");
spec.requestedVulkanFeatures = get16BitStorageFeatures(CAPABILITIES[capIdx].name);
group->addChild(new SpvAsmComputeShaderCase(testCtx, testName.c_str(), testName.c_str(), spec));
}
}
{ // Integers
const char sintTypes[] =
"%i16 = OpTypeInt 16 1\n"
"%i16ptr = OpTypePointer Uniform %i16\n"
"%i16arr = OpTypeArray %i16 %c_i32_128\n"
"%v2i16 = OpTypeVector %i16 2\n"
"%v2i32 = OpTypeVector %i32 2\n"
"%v2i16ptr = OpTypePointer Uniform %v2i16\n"
"%v2i32ptr = OpTypePointer Uniform %v2i32\n"
"%v2i16arr = OpTypeArray %v2i16 %c_i32_64\n"
"%v2i32arr = OpTypeArray %v2i32 %c_i32_64\n";
const char uintTypes[] =
"%u16 = OpTypeInt 16 0\n"
"%u16ptr = OpTypePointer Uniform %u16\n"
"%u32ptr = OpTypePointer Uniform %u32\n"
"%u16arr = OpTypeArray %u16 %c_i32_128\n"
"%u32arr = OpTypeArray %u32 %c_i32_128\n"
"%v2u16 = OpTypeVector %u16 2\n"
"%v2u32 = OpTypeVector %u32 2\n"
"%v2u16ptr = OpTypePointer Uniform %v2u16\n"
"%v2u32ptr = OpTypePointer Uniform %v2u32\n"
"%v2u16arr = OpTypeArray %v2u16 %c_i32_64\n"
"%v2u32arr = OpTypeArray %v2u32 %c_i32_64\n";
struct CompositeType
{
const char* name;
const char* types;
const char* base32;
const char* base16;
const char* opcode;
const char* stride;
unsigned count;
};
const CompositeType cTypes[] =
{
{"scalar_sint", sintTypes, "i32", "i16", "OpSConvert", "OpDecorate %i32arr ArrayStride 4\nOpDecorate %i16arr ArrayStride 2\n", numElements},
{"scalar_uint", uintTypes, "u32", "u16", "OpUConvert", "OpDecorate %u32arr ArrayStride 4\nOpDecorate %u16arr ArrayStride 2\n", numElements},
{"vector_sint", sintTypes, "v2i32", "v2i16", "OpSConvert", "OpDecorate %v2i32arr ArrayStride 8\nOpDecorate %v2i16arr ArrayStride 4\n", numElements / 2},
{"vector_uint", uintTypes, "v2u32", "v2u16", "OpUConvert", "OpDecorate %v2u32arr ArrayStride 8\nOpDecorate %v2u16arr ArrayStride 4\n", numElements / 2},
};
vector<deInt32> inputs = getInt32s(rnd, numElements);
vector<deInt16> outputs;
outputs.reserve(inputs.size());
for (deUint32 numNdx = 0; numNdx < inputs.size(); ++numNdx)
outputs.push_back(static_cast<deInt16>(0xffff & inputs[numNdx]));
for (deUint32 capIdx = 0; capIdx < DE_LENGTH_OF_ARRAY(CAPABILITIES); ++capIdx)
for (deUint32 tyIdx = 0; tyIdx < DE_LENGTH_OF_ARRAY(cTypes); ++tyIdx)
{
ComputeShaderSpec spec;
map<string, string> specs;
string testName = string(CAPABILITIES[capIdx].name) + "_" + cTypes[tyIdx].name;
specs["capability"] = CAPABILITIES[capIdx].cap;
specs["storage"] = CAPABILITIES[capIdx].decor;
specs["stride"] = cTypes[tyIdx].stride;
specs["base32"] = cTypes[tyIdx].base32;
specs["base16"] = cTypes[tyIdx].base16;
specs["types"] = cTypes[tyIdx].types;
specs["convert"] = cTypes[tyIdx].opcode;
spec.assembly = shaderTemplate.specialize(specs);
spec.numWorkGroups = IVec3(cTypes[tyIdx].count, 1, 1);
spec.inputTypes[0] = CAPABILITIES[capIdx].dtype;
spec.inputs.push_back(BufferSp(new Int32Buffer(inputs)));
spec.outputs.push_back(BufferSp(new Int16Buffer(outputs)));
spec.extensions.push_back("VK_KHR_16bit_storage");
spec.requestedVulkanFeatures = get16BitStorageFeatures(CAPABILITIES[capIdx].name);
group->addChild(new SpvAsmComputeShaderCase(testCtx, testName.c_str(), testName.c_str(), spec));
}
}
}
void addGraphics16BitStorageUniformFloat32To16Group (tcu::TestCaseGroup* testGroup)
{
de::Random rnd (deStringHash(testGroup->getName()));
map<string, string> fragments;
GraphicsResources resources;
vector<string> extensions;
const deUint32 numDataPoints = 256;
RGBA defaultColors[4];
vector<float> float32Data = getFloat32s(rnd, numDataPoints);
vector<deFloat16> float16DummyData (numDataPoints, 0);
const StringTemplate capabilities ("OpCapability ${cap}\n");
resources.inputs.push_back(std::make_pair(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, BufferSp(new Float32Buffer(float32Data))));
// We use a custom verifyIO to check the result via computing directly from inputs; the contents in outputs do not matter.
resources.outputs.push_back(std::make_pair(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, BufferSp(new Float16Buffer(float16DummyData))));
extensions.push_back("VK_KHR_16bit_storage");
fragments["extension"] = "OpExtension \"SPV_KHR_16bit_storage\"";
struct RndMode
{
const char* name;
const char* decor;
GraphicsVerifyIOFunc f;
};
getDefaultColors(defaultColors);
{ // scalar cases
fragments["pre_main"] =
" %f16 = OpTypeFloat 16\n"
"%c_i32_256 = OpConstant %i32 256\n"
" %up_f32 = OpTypePointer Uniform %f32\n"
" %up_f16 = OpTypePointer Uniform %f16\n"
" %ra_f32 = OpTypeArray %f32 %c_i32_256\n"
" %ra_f16 = OpTypeArray %f16 %c_i32_256\n"
" %SSBO32 = OpTypeStruct %ra_f32\n"
" %SSBO16 = OpTypeStruct %ra_f16\n"
"%up_SSBO32 = OpTypePointer Uniform %SSBO32\n"
"%up_SSBO16 = OpTypePointer Uniform %SSBO16\n"
" %ssbo32 = OpVariable %up_SSBO32 Uniform\n"
" %ssbo16 = OpVariable %up_SSBO16 Uniform\n";
const StringTemplate decoration (
"OpDecorate %ra_f32 ArrayStride 4\n"
"OpDecorate %ra_f16 ArrayStride 2\n"
"OpMemberDecorate %SSBO32 0 Offset 0\n"
"OpMemberDecorate %SSBO16 0 Offset 0\n"
"OpDecorate %SSBO32 ${indecor}\n"
"OpDecorate %SSBO16 BufferBlock\n"
"OpDecorate %ssbo32 DescriptorSet 0\n"
"OpDecorate %ssbo16 DescriptorSet 0\n"
"OpDecorate %ssbo32 Binding 0\n"
"OpDecorate %ssbo16 Binding 1\n"
"${rounddecor}\n");
fragments["testfun"] =
"%test_code = OpFunction %v4f32 None %v4f32_function\n"
" %param = OpFunctionParameter %v4f32\n"
"%entry = OpLabel\n"
" %i = OpVariable %fp_i32 Function\n"
" OpStore %i %c_i32_0\n"
" OpBranch %loop\n"
" %loop = OpLabel\n"
" %15 = OpLoad %i32 %i\n"
" %lt = OpSLessThan %bool %15 %c_i32_256\n"
" OpLoopMerge %merge %inc None\n"
" OpBranchConditional %lt %write %merge\n"
"%write = OpLabel\n"
" %30 = OpLoad %i32 %i\n"
" %src = OpAccessChain %up_f32 %ssbo32 %c_i32_0 %30\n"
"%val32 = OpLoad %f32 %src\n"
"%val16 = OpFConvert %f16 %val32\n"
" %dst = OpAccessChain %up_f16 %ssbo16 %c_i32_0 %30\n"
" OpStore %dst %val16\n"
" OpBranch %inc\n"
" %inc = OpLabel\n"
" %37 = OpLoad %i32 %i\n"
" %39 = OpIAdd %i32 %37 %c_i32_1\n"
" OpStore %i %39\n"
" OpBranch %loop\n"
"%merge = OpLabel\n"
" OpReturnValue %param\n"
"OpFunctionEnd\n";
const RndMode rndModes[] =
{
{"rtz", "OpDecorate %val16 FPRoundingMode RTZ", graphicsCheck16BitFloats<ROUNDINGMODE_RTZ>},
{"rte", "OpDecorate %val16 FPRoundingMode RTE", graphicsCheck16BitFloats<ROUNDINGMODE_RTE>},
{"unspecified_rnd_mode", "", graphicsCheck16BitFloats<RoundingModeFlags(ROUNDINGMODE_RTE | ROUNDINGMODE_RTZ)>},
};
for (deUint32 capIdx = 0; capIdx < DE_LENGTH_OF_ARRAY(CAPABILITIES); ++capIdx)
for (deUint32 rndModeIdx = 0; rndModeIdx < DE_LENGTH_OF_ARRAY(rndModes); ++rndModeIdx)
{
map<string, string> specs;
string testName = string(CAPABILITIES[capIdx].name) + "_scalar_float_" + rndModes[rndModeIdx].name;
specs["cap"] = CAPABILITIES[capIdx].cap;
specs["indecor"] = CAPABILITIES[capIdx].decor;
specs["rounddecor"] = rndModes[rndModeIdx].decor;
fragments["capability"] = capabilities.specialize(specs);
fragments["decoration"] = decoration.specialize(specs);
resources.inputs.back().first = CAPABILITIES[capIdx].dtype;
resources.verifyIO = rndModes[rndModeIdx].f;
createTestsForAllStages(testName, defaultColors, defaultColors, fragments, resources, extensions, testGroup, get16BitStorageFeatures(CAPABILITIES[capIdx].name));
}
}
{ // vector cases
fragments["pre_main"] =
" %f16 = OpTypeFloat 16\n"
" %c_i32_64 = OpConstant %i32 64\n"
" %v4f16 = OpTypeVector %f16 4\n"
" %up_v4f32 = OpTypePointer Uniform %v4f32\n"
" %up_v4f16 = OpTypePointer Uniform %v4f16\n"
" %ra_v4f32 = OpTypeArray %v4f32 %c_i32_64\n"
" %ra_v4f16 = OpTypeArray %v4f16 %c_i32_64\n"
" %SSBO32 = OpTypeStruct %ra_v4f32\n"
" %SSBO16 = OpTypeStruct %ra_v4f16\n"
"%up_SSBO32 = OpTypePointer Uniform %SSBO32\n"
"%up_SSBO16 = OpTypePointer Uniform %SSBO16\n"
" %ssbo32 = OpVariable %up_SSBO32 Uniform\n"
" %ssbo16 = OpVariable %up_SSBO16 Uniform\n";
const StringTemplate decoration (
"OpDecorate %ra_v4f32 ArrayStride 16\n"
"OpDecorate %ra_v4f16 ArrayStride 8\n"
"OpMemberDecorate %SSBO32 0 Offset 0\n"
"OpMemberDecorate %SSBO16 0 Offset 0\n"
"OpDecorate %SSBO32 ${indecor}\n"
"OpDecorate %SSBO16 BufferBlock\n"
"OpDecorate %ssbo32 DescriptorSet 0\n"
"OpDecorate %ssbo16 DescriptorSet 0\n"
"OpDecorate %ssbo32 Binding 0\n"
"OpDecorate %ssbo16 Binding 1\n"
"${rounddecor}\n");
// ssbo16[] <- convert ssbo32[] to 16bit float
fragments["testfun"] =
"%test_code = OpFunction %v4f32 None %v4f32_function\n"
" %param = OpFunctionParameter %v4f32\n"
"%entry = OpLabel\n"
" %i = OpVariable %fp_i32 Function\n"
" OpStore %i %c_i32_0\n"
" OpBranch %loop\n"
" %loop = OpLabel\n"
" %15 = OpLoad %i32 %i\n"
" %lt = OpSLessThan %bool %15 %c_i32_64\n"
" OpLoopMerge %merge %inc None\n"
" OpBranchConditional %lt %write %merge\n"
"%write = OpLabel\n"
" %30 = OpLoad %i32 %i\n"
" %src = OpAccessChain %up_v4f32 %ssbo32 %c_i32_0 %30\n"
"%val32 = OpLoad %v4f32 %src\n"
"%val16 = OpFConvert %v4f16 %val32\n"
" %dst = OpAccessChain %up_v4f16 %ssbo16 %c_i32_0 %30\n"
" OpStore %dst %val16\n"
" OpBranch %inc\n"
" %inc = OpLabel\n"
" %37 = OpLoad %i32 %i\n"
" %39 = OpIAdd %i32 %37 %c_i32_1\n"
" OpStore %i %39\n"
" OpBranch %loop\n"
"%merge = OpLabel\n"
" OpReturnValue %param\n"
"OpFunctionEnd\n";
const RndMode rndModes[] =
{
{"rtz", "OpDecorate %val16 FPRoundingMode RTZ", graphicsCheck16BitFloats<ROUNDINGMODE_RTZ>},
{"rte", "OpDecorate %val16 FPRoundingMode RTE", graphicsCheck16BitFloats<ROUNDINGMODE_RTE>},
{"unspecified_rnd_mode", "", graphicsCheck16BitFloats<RoundingModeFlags(ROUNDINGMODE_RTE | ROUNDINGMODE_RTZ)>},
};
for (deUint32 capIdx = 0; capIdx < DE_LENGTH_OF_ARRAY(CAPABILITIES); ++capIdx)
for (deUint32 rndModeIdx = 0; rndModeIdx < DE_LENGTH_OF_ARRAY(rndModes); ++rndModeIdx)
{
map<string, string> specs;
string testName = string(CAPABILITIES[capIdx].name) + "_vector_float_" + rndModes[rndModeIdx].name;
specs["cap"] = CAPABILITIES[capIdx].cap;
specs["indecor"] = CAPABILITIES[capIdx].decor;
specs["rounddecor"] = rndModes[rndModeIdx].decor;
fragments["capability"] = capabilities.specialize(specs);
fragments["decoration"] = decoration.specialize(specs);
resources.inputs.back().first = CAPABILITIES[capIdx].dtype;
resources.verifyIO = rndModes[rndModeIdx].f;
createTestsForAllStages(testName, defaultColors, defaultColors, fragments, resources, extensions, testGroup, get16BitStorageFeatures(CAPABILITIES[capIdx].name));
}
}
{ // matrix cases
fragments["pre_main"] =
" %f16 = OpTypeFloat 16\n"
" %c_i32_16 = OpConstant %i32 16\n"
" %v4f16 = OpTypeVector %f16 4\n"
" %m4x4f32 = OpTypeMatrix %v4f32 4\n"
" %m4x4f16 = OpTypeMatrix %v4f16 4\n"
" %up_v4f32 = OpTypePointer Uniform %v4f32\n"
" %up_v4f16 = OpTypePointer Uniform %v4f16\n"
"%a16m4x4f32 = OpTypeArray %m4x4f32 %c_i32_16\n"
"%a16m4x4f16 = OpTypeArray %m4x4f16 %c_i32_16\n"
" %SSBO32 = OpTypeStruct %a16m4x4f32\n"
" %SSBO16 = OpTypeStruct %a16m4x4f16\n"
" %up_SSBO32 = OpTypePointer Uniform %SSBO32\n"
" %up_SSBO16 = OpTypePointer Uniform %SSBO16\n"
" %ssbo32 = OpVariable %up_SSBO32 Uniform\n"
" %ssbo16 = OpVariable %up_SSBO16 Uniform\n";
const StringTemplate decoration (
"OpDecorate %a16m4x4f32 ArrayStride 64\n"
"OpDecorate %a16m4x4f16 ArrayStride 32\n"
"OpMemberDecorate %SSBO32 0 Offset 0\n"
"OpMemberDecorate %SSBO32 0 ColMajor\n"
"OpMemberDecorate %SSBO32 0 MatrixStride 16\n"
"OpMemberDecorate %SSBO16 0 Offset 0\n"
"OpMemberDecorate %SSBO16 0 ColMajor\n"
"OpMemberDecorate %SSBO16 0 MatrixStride 8\n"
"OpDecorate %SSBO32 ${indecor}\n"
"OpDecorate %SSBO16 BufferBlock\n"
"OpDecorate %ssbo32 DescriptorSet 0\n"
"OpDecorate %ssbo16 DescriptorSet 0\n"
"OpDecorate %ssbo32 Binding 0\n"
"OpDecorate %ssbo16 Binding 1\n"
"${rounddecor}\n");
fragments["testfun"] =
"%test_code = OpFunction %v4f32 None %v4f32_function\n"
" %param = OpFunctionParameter %v4f32\n"
"%entry = OpLabel\n"
" %i = OpVariable %fp_i32 Function\n"
" OpStore %i %c_i32_0\n"
" OpBranch %loop\n"
" %loop = OpLabel\n"
" %15 = OpLoad %i32 %i\n"
" %lt = OpSLessThan %bool %15 %c_i32_16\n"
" OpLoopMerge %merge %inc None\n"
" OpBranchConditional %lt %write %merge\n"
" %write = OpLabel\n"
" %30 = OpLoad %i32 %i\n"
" %src_0 = OpAccessChain %up_v4f32 %ssbo32 %c_i32_0 %30 %c_i32_0\n"
" %src_1 = OpAccessChain %up_v4f32 %ssbo32 %c_i32_0 %30 %c_i32_1\n"
" %src_2 = OpAccessChain %up_v4f32 %ssbo32 %c_i32_0 %30 %c_i32_2\n"
" %src_3 = OpAccessChain %up_v4f32 %ssbo32 %c_i32_0 %30 %c_i32_3\n"
"%val32_0 = OpLoad %v4f32 %src_0\n"
"%val32_1 = OpLoad %v4f32 %src_1\n"
"%val32_2 = OpLoad %v4f32 %src_2\n"
"%val32_3 = OpLoad %v4f32 %src_3\n"
"%val16_0 = OpFConvert %v4f16 %val32_0\n"
"%val16_1 = OpFConvert %v4f16 %val32_1\n"
"%val16_2 = OpFConvert %v4f16 %val32_2\n"
"%val16_3 = OpFConvert %v4f16 %val32_3\n"
" %dst_0 = OpAccessChain %up_v4f16 %ssbo16 %c_i32_0 %30 %c_i32_0\n"
" %dst_1 = OpAccessChain %up_v4f16 %ssbo16 %c_i32_0 %30 %c_i32_1\n"
" %dst_2 = OpAccessChain %up_v4f16 %ssbo16 %c_i32_0 %30 %c_i32_2\n"
" %dst_3 = OpAccessChain %up_v4f16 %ssbo16 %c_i32_0 %30 %c_i32_3\n"
" OpStore %dst_0 %val16_0\n"
" OpStore %dst_1 %val16_1\n"
" OpStore %dst_2 %val16_2\n"
" OpStore %dst_3 %val16_3\n"
" OpBranch %inc\n"
" %inc = OpLabel\n"
" %37 = OpLoad %i32 %i\n"
" %39 = OpIAdd %i32 %37 %c_i32_1\n"
" OpStore %i %39\n"
" OpBranch %loop\n"
"%merge = OpLabel\n"
" OpReturnValue %param\n"
"OpFunctionEnd\n";
const RndMode rndModes[] =
{
{"rte", "OpDecorate %val16_0 FPRoundingMode RTE\nOpDecorate %val16_1 FPRoundingMode RTE\nOpDecorate %val16_2 FPRoundingMode RTE\nOpDecorate %val16_3 FPRoundingMode RTE", graphicsCheck16BitFloats<ROUNDINGMODE_RTE>},
{"rtz", "OpDecorate %val16_0 FPRoundingMode RTZ\nOpDecorate %val16_1 FPRoundingMode RTZ\nOpDecorate %val16_2 FPRoundingMode RTZ\nOpDecorate %val16_3 FPRoundingMode RTZ", graphicsCheck16BitFloats<ROUNDINGMODE_RTZ>},
{"unspecified_rnd_mode", "", graphicsCheck16BitFloats<RoundingModeFlags(ROUNDINGMODE_RTE | ROUNDINGMODE_RTZ)>},
};
for (deUint32 capIdx = 0; capIdx < DE_LENGTH_OF_ARRAY(CAPABILITIES); ++capIdx)
for (deUint32 rndModeIdx = 0; rndModeIdx < DE_LENGTH_OF_ARRAY(rndModes); ++rndModeIdx)
{
map<string, string> specs;
string testName = string(CAPABILITIES[capIdx].name) + "_matrix_float_" + rndModes[rndModeIdx].name;
specs["cap"] = CAPABILITIES[capIdx].cap;
specs["indecor"] = CAPABILITIES[capIdx].decor;
specs["rounddecor"] = rndModes[rndModeIdx].decor;
fragments["capability"] = capabilities.specialize(specs);
fragments["decoration"] = decoration.specialize(specs);
resources.inputs.back().first = CAPABILITIES[capIdx].dtype;
resources.verifyIO = rndModes[rndModeIdx].f;
createTestsForAllStages(testName, defaultColors, defaultColors, fragments, resources, extensions, testGroup, get16BitStorageFeatures(CAPABILITIES[capIdx].name));
}
}
}
void addGraphics16BitStorageInputOutputFloat32To16Group (tcu::TestCaseGroup* testGroup)
{
de::Random rnd (deStringHash(testGroup->getName()));
RGBA defaultColors[4];
vector<string> extensions;
map<string, string> fragments = passthruFragments();
const deUint32 numDataPoints = 64;
vector<float> float32Data = getFloat32s(rnd, numDataPoints);
extensions.push_back("VK_KHR_16bit_storage");
fragments["capability"] = "OpCapability StorageInputOutput16\n";
fragments["extension"] = "OpExtension \"SPV_KHR_16bit_storage\"\n";
getDefaultColors(defaultColors);
struct RndMode
{
const char* name;
const char* decor;
RoundingModeFlags flags;
};
const RndMode rndModes[] =
{
{"rtz", "OpDecorate %ret FPRoundingMode RTZ", ROUNDINGMODE_RTZ},
{"rte", "OpDecorate %ret FPRoundingMode RTE", ROUNDINGMODE_RTE},
{"unspecified_rnd_mode", "", RoundingModeFlags(ROUNDINGMODE_RTE | ROUNDINGMODE_RTZ)},
};
struct Case
{
const char* name;
const char* interfaceOpFunc;
const char* preMain;
const char* inputType;
const char* outputType;
deUint32 numPerCase;
deUint32 numElements;
};
const Case cases[] =
{
{ // Scalar cases
"scalar",
"%interface_op_func = OpFunction %f16 None %f16_f32_function\n"
" %io_param1 = OpFunctionParameter %f32\n"
" %entry = OpLabel\n"
" %ret = OpFConvert %f16 %io_param1\n"
" OpReturnValue %ret\n"
" OpFunctionEnd\n",
" %f16 = OpTypeFloat 16\n"
" %op_f16 = OpTypePointer Output %f16\n"
" %a3f16 = OpTypeArray %f16 %c_i32_3\n"
" %op_a3f16 = OpTypePointer Output %a3f16\n"
"%f16_f32_function = OpTypeFunction %f16 %f32\n"
" %a3f32 = OpTypeArray %f32 %c_i32_3\n"
" %ip_a3f32 = OpTypePointer Input %a3f32\n",
"f32",
"f16",
4,
1,
},
{ // Vector cases
"vector",
"%interface_op_func = OpFunction %v2f16 None %v2f16_v2f32_function\n"
" %io_param1 = OpFunctionParameter %v2f32\n"
" %entry = OpLabel\n"
" %ret = OpFConvert %v2f16 %io_param1\n"
" OpReturnValue %ret\n"
" OpFunctionEnd\n",
" %f16 = OpTypeFloat 16\n"
" %v2f16 = OpTypeVector %f16 2\n"
" %op_v2f16 = OpTypePointer Output %v2f16\n"
" %a3v2f16 = OpTypeArray %v2f16 %c_i32_3\n"
" %op_a3v2f16 = OpTypePointer Output %a3v2f16\n"
"%v2f16_v2f32_function = OpTypeFunction %v2f16 %v2f32\n"
" %a3v2f32 = OpTypeArray %v2f32 %c_i32_3\n"
" %ip_a3v2f32 = OpTypePointer Input %a3v2f32\n",
"v2f32",
"v2f16",
2 * 4,
2,
}
};
VulkanFeatures requiredFeatures;
requiredFeatures.ext16BitStorage = EXT16BITSTORAGEFEATURES_INPUT_OUTPUT;
for (deUint32 caseIdx = 0; caseIdx < DE_LENGTH_OF_ARRAY(cases); ++caseIdx)
for (deUint32 rndModeIdx = 0; rndModeIdx < DE_LENGTH_OF_ARRAY(rndModes); ++rndModeIdx)
{
fragments["interface_op_func"] = cases[caseIdx].interfaceOpFunc;
fragments["pre_main"] = cases[caseIdx].preMain;
fragments["decoration"] = rndModes[rndModeIdx].decor;
fragments["input_type"] = cases[caseIdx].inputType;
fragments["output_type"] = cases[caseIdx].outputType;
GraphicsInterfaces interfaces;
const deUint32 numPerCase = cases[caseIdx].numPerCase;
vector<float> subInputs (numPerCase);
vector<deFloat16> subOutputs (numPerCase);
// The pipeline need this to call compare16BitFloat() when checking the result.
interfaces.setRoundingMode(rndModes[rndModeIdx].flags);
for (deUint32 caseNdx = 0; caseNdx < numDataPoints / numPerCase; ++caseNdx)
{
string testName = string(cases[caseIdx].name) + numberToString(caseNdx) + "_" + rndModes[rndModeIdx].name;
for (deUint32 numNdx = 0; numNdx < numPerCase; ++numNdx)
{
subInputs[numNdx] = float32Data[caseNdx * numPerCase + numNdx];
// We derive the expected result from inputs directly in the graphics pipeline.
subOutputs[numNdx] = 0;
}
interfaces.setInputOutput(std::make_pair(IFDataType(cases[caseIdx].numElements, NUMBERTYPE_FLOAT32), BufferSp(new Float32Buffer(subInputs))),
std::make_pair(IFDataType(cases[caseIdx].numElements, NUMBERTYPE_FLOAT16), BufferSp(new Float16Buffer(subOutputs))));
createTestsForAllStages(testName, defaultColors, defaultColors, fragments, interfaces, extensions, testGroup, requiredFeatures);
}
}
}
void addGraphics16BitStorageInputOutputFloat16To32Group (tcu::TestCaseGroup* testGroup)
{
de::Random rnd (deStringHash(testGroup->getName()));
RGBA defaultColors[4];
vector<string> extensions;
map<string, string> fragments = passthruFragments();
const deUint32 numDataPoints = 64;
vector<deFloat16> float16Data (getFloat16s(rnd, numDataPoints));
vector<float> float32Data;
float32Data.reserve(numDataPoints);
for (deUint32 numIdx = 0; numIdx < numDataPoints; ++numIdx)
float32Data.push_back(deFloat16To32(float16Data[numIdx]));
extensions.push_back("VK_KHR_16bit_storage");
fragments["capability"] = "OpCapability StorageInputOutput16\n";
fragments["extension"] = "OpExtension \"SPV_KHR_16bit_storage\"\n";
getDefaultColors(defaultColors);
struct Case
{
const char* name;
const char* interfaceOpFunc;
const char* preMain;
const char* inputType;
const char* outputType;
deUint32 numPerCase;
deUint32 numElements;
};
Case cases[] =
{
{ // Scalar cases
"scalar",
"%interface_op_func = OpFunction %f32 None %f32_f16_function\n"
" %io_param1 = OpFunctionParameter %f16\n"
" %entry = OpLabel\n"
" %ret = OpFConvert %f32 %io_param1\n"
" OpReturnValue %ret\n"
" OpFunctionEnd\n",
" %f16 = OpTypeFloat 16\n"
" %ip_f16 = OpTypePointer Input %f16\n"
" %a3f16 = OpTypeArray %f16 %c_i32_3\n"
" %ip_a3f16 = OpTypePointer Input %a3f16\n"
"%f32_f16_function = OpTypeFunction %f32 %f16\n"
" %a3f32 = OpTypeArray %f32 %c_i32_3\n"
" %op_a3f32 = OpTypePointer Output %a3f32\n",
"f16",
"f32",
4,
1,
},
{ // Vector cases
"vector",
"%interface_op_func = OpFunction %v2f32 None %v2f32_v2f16_function\n"
" %io_param1 = OpFunctionParameter %v2f16\n"
" %entry = OpLabel\n"
" %ret = OpFConvert %v2f32 %io_param1\n"
" OpReturnValue %ret\n"
" OpFunctionEnd\n",
" %f16 = OpTypeFloat 16\n"
" %v2f16 = OpTypeVector %f16 2\n"
" %ip_v2f16 = OpTypePointer Input %v2f16\n"
" %a3v2f16 = OpTypeArray %v2f16 %c_i32_3\n"
" %ip_a3v2f16 = OpTypePointer Input %a3v2f16\n"
"%v2f32_v2f16_function = OpTypeFunction %v2f32 %v2f16\n"
" %a3v2f32 = OpTypeArray %v2f32 %c_i32_3\n"
" %op_a3v2f32 = OpTypePointer Output %a3v2f32\n",
"v2f16",
"v2f32",
2 * 4,
2,
}
};
VulkanFeatures requiredFeatures;
requiredFeatures.ext16BitStorage = EXT16BITSTORAGEFEATURES_INPUT_OUTPUT;
for (deUint32 caseIdx = 0; caseIdx < DE_LENGTH_OF_ARRAY(cases); ++caseIdx)
{
fragments["interface_op_func"] = cases[caseIdx].interfaceOpFunc;
fragments["pre_main"] = cases[caseIdx].preMain;
fragments["input_type"] = cases[caseIdx].inputType;
fragments["output_type"] = cases[caseIdx].outputType;
GraphicsInterfaces interfaces;
const deUint32 numPerCase = cases[caseIdx].numPerCase;
vector<deFloat16> subInputs (numPerCase);
vector<float> subOutputs (numPerCase);
for (deUint32 caseNdx = 0; caseNdx < numDataPoints / numPerCase; ++caseNdx)
{
string testName = string(cases[caseIdx].name) + numberToString(caseNdx);
for (deUint32 numNdx = 0; numNdx < numPerCase; ++numNdx)
{
subInputs[numNdx] = float16Data[caseNdx * numPerCase + numNdx];
subOutputs[numNdx] = float32Data[caseNdx * numPerCase + numNdx];
}
interfaces.setInputOutput(std::make_pair(IFDataType(cases[caseIdx].numElements, NUMBERTYPE_FLOAT16), BufferSp(new Float16Buffer(subInputs))),
std::make_pair(IFDataType(cases[caseIdx].numElements, NUMBERTYPE_FLOAT32), BufferSp(new Float32Buffer(subOutputs))));
createTestsForAllStages(testName, defaultColors, defaultColors, fragments, interfaces, extensions, testGroup, requiredFeatures);
}
}
}
void addGraphics16BitStorageInputOutputInt32To16Group (tcu::TestCaseGroup* testGroup)
{
de::Random rnd (deStringHash(testGroup->getName()));
RGBA defaultColors[4];
vector<string> extensions;
map<string, string> fragments = passthruFragments();
const deUint32 numDataPoints = 64;
// inputs and outputs are declared to be vectors of signed integers.
// However, depending on the test, they may be interpreted as unsiged
// integers. That won't be a problem as long as we passed the bits
// in faithfully to the pipeline.
vector<deInt32> inputs = getInt32s(rnd, numDataPoints);
vector<deInt16> outputs;
outputs.reserve(inputs.size());
for (deUint32 numNdx = 0; numNdx < inputs.size(); ++numNdx)
outputs.push_back(static_cast<deInt16>(0xffff & inputs[numNdx]));
extensions.push_back("VK_KHR_16bit_storage");
fragments["capability"] = "OpCapability StorageInputOutput16\n";
fragments["extension"] = "OpExtension \"SPV_KHR_16bit_storage\"\n";
getDefaultColors(defaultColors);
const StringTemplate scalarInterfaceOpFunc(
"%interface_op_func = OpFunction %${type16} None %${type16}_${type32}_function\n"
" %io_param1 = OpFunctionParameter %${type32}\n"
" %entry = OpLabel\n"
" %ret = ${convert} %${type16} %io_param1\n"
" OpReturnValue %ret\n"
" OpFunctionEnd\n");
const StringTemplate scalarPreMain(
" %${type16} = OpTypeInt 16 ${signed}\n"
" %op_${type16} = OpTypePointer Output %${type16}\n"
" %a3${type16} = OpTypeArray %${type16} %c_i32_3\n"
" %op_a3${type16} = OpTypePointer Output %a3${type16}\n"
"%${type16}_${type32}_function = OpTypeFunction %${type16} %${type32}\n"
" %a3${type32} = OpTypeArray %${type32} %c_i32_3\n"
" %ip_a3${type32} = OpTypePointer Input %a3${type32}\n");
const StringTemplate vecInterfaceOpFunc(
"%interface_op_func = OpFunction %${type16} None %${type16}_${type32}_function\n"
" %io_param1 = OpFunctionParameter %${type32}\n"
" %entry = OpLabel\n"
" %ret = ${convert} %${type16} %io_param1\n"
" OpReturnValue %ret\n"
" OpFunctionEnd\n");
const StringTemplate vecPreMain(
" %i16 = OpTypeInt 16 1\n"
" %u16 = OpTypeInt 16 0\n"
" %v4i16 = OpTypeVector %i16 4\n"
" %v4u16 = OpTypeVector %u16 4\n"
" %op_${type16} = OpTypePointer Output %${type16}\n"
" %a3${type16} = OpTypeArray %${type16} %c_i32_3\n"
" %op_a3${type16} = OpTypePointer Output %a3${type16}\n"
"%${type16}_${type32}_function = OpTypeFunction %${type16} %${type32}\n"
" %a3${type32} = OpTypeArray %${type32} %c_i32_3\n"
" %ip_a3${type32} = OpTypePointer Input %a3${type32}\n");
struct Case
{
const char* name;
const StringTemplate& interfaceOpFunc;
const StringTemplate& preMain;
const char* type32;
const char* type16;
const char* sign;
const char* opcode;
deUint32 numPerCase;
deUint32 numElements;
};
Case cases[] =
{
{"scalar_sint", scalarInterfaceOpFunc, scalarPreMain, "i32", "i16", "1", "OpSConvert", 4, 1},
{"scalar_uint", scalarInterfaceOpFunc, scalarPreMain, "u32", "u16", "0", "OpUConvert", 4, 1},
{"vector_sint", vecInterfaceOpFunc, vecPreMain, "v4i32", "v4i16", "1", "OpSConvert", 4 * 4, 4},
{"vector_uint", vecInterfaceOpFunc, vecPreMain, "v4u32", "v4u16", "0", "OpUConvert", 4 * 4, 4},
};
VulkanFeatures requiredFeatures;
requiredFeatures.ext16BitStorage = EXT16BITSTORAGEFEATURES_INPUT_OUTPUT;
for (deUint32 caseIdx = 0; caseIdx < DE_LENGTH_OF_ARRAY(cases); ++caseIdx)
{
map<string, string> specs;
specs["type32"] = cases[caseIdx].type32;
specs["type16"] = cases[caseIdx].type16;
specs["signed"] = cases[caseIdx].sign;
specs["convert"] = cases[caseIdx].opcode;
fragments["pre_main"] = cases[caseIdx].preMain.specialize(specs);
fragments["interface_op_func"] = cases[caseIdx].interfaceOpFunc.specialize(specs);
fragments["input_type"] = cases[caseIdx].type32;
fragments["output_type"] = cases[caseIdx].type16;
GraphicsInterfaces interfaces;
const deUint32 numPerCase = cases[caseIdx].numPerCase;
vector<deInt32> subInputs (numPerCase);
vector<deInt16> subOutputs (numPerCase);
for (deUint32 caseNdx = 0; caseNdx < numDataPoints / numPerCase; ++caseNdx)
{
string testName = string(cases[caseIdx].name) + numberToString(caseNdx);
for (deUint32 numNdx = 0; numNdx < numPerCase; ++numNdx)
{
subInputs[numNdx] = inputs[caseNdx * numPerCase + numNdx];
subOutputs[numNdx] = outputs[caseNdx * numPerCase + numNdx];
}
if (strcmp(cases[caseIdx].sign, "1") == 0)
{
interfaces.setInputOutput(std::make_pair(IFDataType(cases[caseIdx].numElements, NUMBERTYPE_INT32), BufferSp(new Int32Buffer(subInputs))),
std::make_pair(IFDataType(cases[caseIdx].numElements, NUMBERTYPE_INT16), BufferSp(new Int16Buffer(subOutputs))));
}
else
{
interfaces.setInputOutput(std::make_pair(IFDataType(cases[caseIdx].numElements, NUMBERTYPE_UINT32), BufferSp(new Int32Buffer(subInputs))),
std::make_pair(IFDataType(cases[caseIdx].numElements, NUMBERTYPE_UINT16), BufferSp(new Int16Buffer(subOutputs))));
}
createTestsForAllStages(testName, defaultColors, defaultColors, fragments, interfaces, extensions, testGroup, requiredFeatures);
}
}
}
void addGraphics16BitStorageInputOutputInt16To32Group (tcu::TestCaseGroup* testGroup)
{
de::Random rnd (deStringHash(testGroup->getName()));
RGBA defaultColors[4];
vector<string> extensions;
map<string, string> fragments = passthruFragments();
const deUint32 numDataPoints = 64;
// inputs and outputs are declared to be vectors of signed integers.
// However, depending on the test, they may be interpreted as unsiged
// integers. That won't be a problem as long as we passed the bits
// in faithfully to the pipeline.
vector<deInt16> inputs = getInt16s(rnd, numDataPoints);
vector<deInt32> sOutputs;
vector<deInt32> uOutputs;
const deUint16 signBitMask = 0x8000;
const deUint32 signExtendMask = 0xffff0000;
sOutputs.reserve(inputs.size());
uOutputs.reserve(inputs.size());
for (deUint32 numNdx = 0; numNdx < inputs.size(); ++numNdx)
{
uOutputs.push_back(static_cast<deUint16>(inputs[numNdx]));
if (inputs[numNdx] & signBitMask)
sOutputs.push_back(static_cast<deInt32>(inputs[numNdx] | signExtendMask));
else
sOutputs.push_back(static_cast<deInt32>(inputs[numNdx]));
}
extensions.push_back("VK_KHR_16bit_storage");
fragments["capability"] = "OpCapability StorageInputOutput16\n";
fragments["extension"] = "OpExtension \"SPV_KHR_16bit_storage\"\n";
getDefaultColors(defaultColors);
const StringTemplate scalarIfOpFunc (
"%interface_op_func = OpFunction %${type32} None %${type32}_${type16}_function\n"
" %io_param1 = OpFunctionParameter %${type16}\n"
" %entry = OpLabel\n"
" %ret = ${convert} %${type32} %io_param1\n"
" OpReturnValue %ret\n"
" OpFunctionEnd\n");
const StringTemplate scalarPreMain (
" %${type16} = OpTypeInt 16 ${signed}\n"
" %ip_${type16} = OpTypePointer Input %${type16}\n"
" %a3${type16} = OpTypeArray %${type16} %c_i32_3\n"
" %ip_a3${type16} = OpTypePointer Input %a3${type16}\n"
"%${type32}_${type16}_function = OpTypeFunction %${type32} %${type16}\n"
" %a3${type32} = OpTypeArray %${type32} %c_i32_3\n"
" %op_a3${type32} = OpTypePointer Output %a3${type32}\n");
const StringTemplate vecIfOpFunc (
"%interface_op_func = OpFunction %${type32} None %${type32}_${type16}_function\n"
" %io_param1 = OpFunctionParameter %${type16}\n"
" %entry = OpLabel\n"
" %ret = ${convert} %${type32} %io_param1\n"
" OpReturnValue %ret\n"
" OpFunctionEnd\n");
const StringTemplate vecPreMain (
" %i16 = OpTypeInt 16 1\n"
" %u16 = OpTypeInt 16 0\n"
" %v4i16 = OpTypeVector %i16 4\n"
" %v4u16 = OpTypeVector %u16 4\n"
" %ip_${type16} = OpTypePointer Input %${type16}\n"
" %a3${type16} = OpTypeArray %${type16} %c_i32_3\n"
" %ip_a3${type16} = OpTypePointer Input %a3${type16}\n"
"%${type32}_${type16}_function = OpTypeFunction %${type32} %${type16}\n"
" %a3${type32} = OpTypeArray %${type32} %c_i32_3\n"
" %op_a3${type32} = OpTypePointer Output %a3${type32}\n");
struct Case
{
const char* name;
const StringTemplate& interfaceOpFunc;
const StringTemplate& preMain;
const char* type32;
const char* type16;
const char* sign;
const char* opcode;
deUint32 numPerCase;
deUint32 numElements;
};
Case cases[] =
{
{"scalar_sint", scalarIfOpFunc, scalarPreMain, "i32", "i16", "1", "OpSConvert", 4, 1},
{"scalar_uint", scalarIfOpFunc, scalarPreMain, "u32", "u16", "0", "OpUConvert", 4, 1},
{"vector_sint", vecIfOpFunc, vecPreMain, "v4i32", "v4i16", "1", "OpSConvert", 4 * 4, 4},
{"vector_uint", vecIfOpFunc, vecPreMain, "v4u32", "v4u16", "0", "OpUConvert", 4 * 4, 4},
};
VulkanFeatures requiredFeatures;
requiredFeatures.ext16BitStorage = EXT16BITSTORAGEFEATURES_INPUT_OUTPUT;
for (deUint32 caseIdx = 0; caseIdx < DE_LENGTH_OF_ARRAY(cases); ++caseIdx)
{
map<string, string> specs;
specs["type32"] = cases[caseIdx].type32;
specs["type16"] = cases[caseIdx].type16;
specs["signed"] = cases[caseIdx].sign;
specs["convert"] = cases[caseIdx].opcode;
fragments["pre_main"] = cases[caseIdx].preMain.specialize(specs);
fragments["interface_op_func"] = cases[caseIdx].interfaceOpFunc.specialize(specs);
fragments["input_type"] = cases[caseIdx].type16;
fragments["output_type"] = cases[caseIdx].type32;
GraphicsInterfaces interfaces;
const deUint32 numPerCase = cases[caseIdx].numPerCase;
vector<deInt16> subInputs (numPerCase);
vector<deInt32> subOutputs (numPerCase);
for (deUint32 caseNdx = 0; caseNdx < numDataPoints / numPerCase; ++caseNdx)
{
string testName = string(cases[caseIdx].name) + numberToString(caseNdx);
for (deUint32 numNdx = 0; numNdx < numPerCase; ++numNdx)
{
subInputs[numNdx] = inputs[caseNdx * numPerCase + numNdx];
if (cases[caseIdx].sign[0] == '1')
subOutputs[numNdx] = sOutputs[caseNdx * numPerCase + numNdx];
else
subOutputs[numNdx] = uOutputs[caseNdx * numPerCase + numNdx];
}
if (strcmp(cases[caseIdx].sign, "1") == 0)
{
interfaces.setInputOutput(std::make_pair(IFDataType(cases[caseIdx].numElements, NUMBERTYPE_INT16), BufferSp(new Int16Buffer(subInputs))),
std::make_pair(IFDataType(cases[caseIdx].numElements, NUMBERTYPE_INT32), BufferSp(new Int32Buffer(subOutputs))));
}
else
{
interfaces.setInputOutput(std::make_pair(IFDataType(cases[caseIdx].numElements, NUMBERTYPE_UINT16), BufferSp(new Int16Buffer(subInputs))),
std::make_pair(IFDataType(cases[caseIdx].numElements, NUMBERTYPE_UINT32), BufferSp(new Int32Buffer(subOutputs))));
}
createTestsForAllStages(testName, defaultColors, defaultColors, fragments, interfaces, extensions, testGroup, requiredFeatures);
}
}
}
void addGraphics16BitStoragePushConstantFloat16To32Group (tcu::TestCaseGroup* testGroup)
{
de::Random rnd (deStringHash(testGroup->getName()));
map<string, string> fragments;
RGBA defaultColors[4];
vector<string> extensions;
GraphicsResources resources;
PushConstants pcs;
const deUint32 numDataPoints = 64;
vector<deFloat16> float16Data (getFloat16s(rnd, numDataPoints));
vector<float> float32Data;
VulkanFeatures requiredFeatures;
float32Data.reserve(numDataPoints);
for (deUint32 numIdx = 0; numIdx < numDataPoints; ++numIdx)
float32Data.push_back(deFloat16To32(float16Data[numIdx]));
extensions.push_back("VK_KHR_16bit_storage");
requiredFeatures.ext16BitStorage = EXT16BITSTORAGEFEATURES_PUSH_CONSTANT;
fragments["capability"] = "OpCapability StoragePushConstant16\n";
fragments["extension"] = "OpExtension \"SPV_KHR_16bit_storage\"";
pcs.setPushConstant(BufferSp(new Float16Buffer(float16Data)));
resources.outputs.push_back(std::make_pair(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, BufferSp(new Float32Buffer(float32Data))));
resources.verifyIO = check32BitFloats;
getDefaultColors(defaultColors);
const StringTemplate testFun (
"%test_code = OpFunction %v4f32 None %v4f32_function\n"
" %param = OpFunctionParameter %v4f32\n"
"%entry = OpLabel\n"
" %i = OpVariable %fp_i32 Function\n"
" OpStore %i %c_i32_0\n"
" OpBranch %loop\n"
" %loop = OpLabel\n"
" %15 = OpLoad %i32 %i\n"
" %lt = OpSLessThan %bool %15 ${count}\n"
" OpLoopMerge %merge %inc None\n"
" OpBranchConditional %lt %write %merge\n"
"%write = OpLabel\n"
" %30 = OpLoad %i32 %i\n"
" %src = OpAccessChain ${pp_type16} %pc16 %c_i32_0 %30 ${index0:opt}\n"
"%val16 = OpLoad ${f_type16} %src\n"
"%val32 = OpFConvert ${f_type32} %val16\n"
" %dst = OpAccessChain ${up_type32} %ssbo32 %c_i32_0 %30 ${index0:opt}\n"
" OpStore %dst %val32\n"
"${store:opt}\n"
" OpBranch %inc\n"
" %inc = OpLabel\n"
" %37 = OpLoad %i32 %i\n"
" %39 = OpIAdd %i32 %37 %c_i32_1\n"
" OpStore %i %39\n"
" OpBranch %loop\n"
"%merge = OpLabel\n"
" OpReturnValue %param\n"
"OpFunctionEnd\n");
{ // Scalar cases
fragments["pre_main"] =
" %f16 = OpTypeFloat 16\n"
" %c_i32_64 = OpConstant %i32 64\n" // Should be the same as numDataPoints
" %a64f16 = OpTypeArray %f16 %c_i32_64\n"
" %a64f32 = OpTypeArray %f32 %c_i32_64\n"
" %pp_f16 = OpTypePointer PushConstant %f16\n"
" %up_f32 = OpTypePointer Uniform %f32\n"
" %SSBO32 = OpTypeStruct %a64f32\n"
"%up_SSBO32 = OpTypePointer Uniform %SSBO32\n"
" %ssbo32 = OpVariable %up_SSBO32 Uniform\n"
" %PC16 = OpTypeStruct %a64f16\n"
" %pp_PC16 = OpTypePointer PushConstant %PC16\n"
" %pc16 = OpVariable %pp_PC16 PushConstant\n";
fragments["decoration"] =
"OpDecorate %a64f16 ArrayStride 2\n"
"OpDecorate %a64f32 ArrayStride 4\n"
"OpDecorate %SSBO32 BufferBlock\n"
"OpMemberDecorate %SSBO32 0 Offset 0\n"
"OpDecorate %PC16 Block\n"
"OpMemberDecorate %PC16 0 Offset 0\n"
"OpDecorate %ssbo32 DescriptorSet 0\n"
"OpDecorate %ssbo32 Binding 0\n";
map<string, string> specs;
specs["count"] = "%c_i32_64";
specs["pp_type16"] = "%pp_f16";
specs["f_type16"] = "%f16";
specs["f_type32"] = "%f32";
specs["up_type32"] = "%up_f32";
fragments["testfun"] = testFun.specialize(specs);
createTestsForAllStages("scalar", defaultColors, defaultColors, fragments, pcs, resources, extensions, testGroup, requiredFeatures);
}
{ // Vector cases
fragments["pre_main"] =
" %f16 = OpTypeFloat 16\n"
" %v4f16 = OpTypeVector %f16 4\n"
" %c_i32_16 = OpConstant %i32 16\n"
" %a16v4f16 = OpTypeArray %v4f16 %c_i32_16\n"
" %a16v4f32 = OpTypeArray %v4f32 %c_i32_16\n"
" %pp_v4f16 = OpTypePointer PushConstant %v4f16\n"
" %up_v4f32 = OpTypePointer Uniform %v4f32\n"
" %SSBO32 = OpTypeStruct %a16v4f32\n"
"%up_SSBO32 = OpTypePointer Uniform %SSBO32\n"
" %ssbo32 = OpVariable %up_SSBO32 Uniform\n"
" %PC16 = OpTypeStruct %a16v4f16\n"
" %pp_PC16 = OpTypePointer PushConstant %PC16\n"
" %pc16 = OpVariable %pp_PC16 PushConstant\n";
fragments["decoration"] =
"OpDecorate %a16v4f16 ArrayStride 8\n"
"OpDecorate %a16v4f32 ArrayStride 16\n"
"OpDecorate %SSBO32 BufferBlock\n"
"OpMemberDecorate %SSBO32 0 Offset 0\n"
"OpDecorate %PC16 Block\n"
"OpMemberDecorate %PC16 0 Offset 0\n"
"OpDecorate %ssbo32 DescriptorSet 0\n"
"OpDecorate %ssbo32 Binding 0\n";
map<string, string> specs;
specs["count"] = "%c_i32_16";
specs["pp_type16"] = "%pp_v4f16";
specs["f_type16"] = "%v4f16";
specs["f_type32"] = "%v4f32";
specs["up_type32"] = "%up_v4f32";
fragments["testfun"] = testFun.specialize(specs);
createTestsForAllStages("vector", defaultColors, defaultColors, fragments, pcs, resources, extensions, testGroup, requiredFeatures);
}
{ // Matrix cases
fragments["pre_main"] =
" %c_i32_8 = OpConstant %i32 8\n"
" %f16 = OpTypeFloat 16\n"
" %v4f16 = OpTypeVector %f16 4\n"
" %m2v4f16 = OpTypeMatrix %v4f16 2\n"
" %m2v4f32 = OpTypeMatrix %v4f32 2\n"
"%a8m2v4f16 = OpTypeArray %m2v4f16 %c_i32_8\n"
"%a8m2v4f32 = OpTypeArray %m2v4f32 %c_i32_8\n"
" %pp_v4f16 = OpTypePointer PushConstant %v4f16\n"
" %up_v4f32 = OpTypePointer Uniform %v4f32\n"
" %SSBO32 = OpTypeStruct %a8m2v4f32\n"
"%up_SSBO32 = OpTypePointer Uniform %SSBO32\n"
" %ssbo32 = OpVariable %up_SSBO32 Uniform\n"
" %PC16 = OpTypeStruct %a8m2v4f16\n"
" %pp_PC16 = OpTypePointer PushConstant %PC16\n"
" %pc16 = OpVariable %pp_PC16 PushConstant\n";
fragments["decoration"] =
"OpDecorate %a8m2v4f16 ArrayStride 16\n"
"OpDecorate %a8m2v4f32 ArrayStride 32\n"
"OpDecorate %SSBO32 BufferBlock\n"
"OpMemberDecorate %SSBO32 0 Offset 0\n"
"OpMemberDecorate %SSBO32 0 ColMajor\n"
"OpMemberDecorate %SSBO32 0 MatrixStride 16\n"
"OpDecorate %PC16 Block\n"
"OpMemberDecorate %PC16 0 Offset 0\n"
"OpMemberDecorate %PC16 0 ColMajor\n"
"OpMemberDecorate %PC16 0 MatrixStride 8\n"
"OpDecorate %ssbo32 DescriptorSet 0\n"
"OpDecorate %ssbo32 Binding 0\n";
map<string, string> specs;
specs["count"] = "%c_i32_8";
specs["pp_type16"] = "%pp_v4f16";
specs["up_type32"] = "%up_v4f32";
specs["f_type16"] = "%v4f16";
specs["f_type32"] = "%v4f32";
specs["index0"] = "%c_i32_0";
specs["store"] =
" %src_1 = OpAccessChain %pp_v4f16 %pc16 %c_i32_0 %30 %c_i32_1\n"
"%val16_1 = OpLoad %v4f16 %src_1\n"
"%val32_1 = OpFConvert %v4f32 %val16_1\n"
" %dst_1 = OpAccessChain %up_v4f32 %ssbo32 %c_i32_0 %30 %c_i32_1\n"
" OpStore %dst_1 %val32_1\n";
fragments["testfun"] = testFun.specialize(specs);
createTestsForAllStages("matrix", defaultColors, defaultColors, fragments, pcs, resources, extensions, testGroup, requiredFeatures);
}
}
void addGraphics16BitStoragePushConstantInt16To32Group (tcu::TestCaseGroup* testGroup)
{
de::Random rnd (deStringHash(testGroup->getName()));
map<string, string> fragments;
RGBA defaultColors[4];
const deUint32 numDataPoints = 64;
vector<deInt16> inputs = getInt16s(rnd, numDataPoints);
vector<deInt32> sOutputs;
vector<deInt32> uOutputs;
PushConstants pcs;
GraphicsResources resources;
vector<string> extensions;
const deUint16 signBitMask = 0x8000;
const deUint32 signExtendMask = 0xffff0000;
VulkanFeatures requiredFeatures;
sOutputs.reserve(inputs.size());
uOutputs.reserve(inputs.size());
for (deUint32 numNdx = 0; numNdx < inputs.size(); ++numNdx)
{
uOutputs.push_back(static_cast<deUint16>(inputs[numNdx]));
if (inputs[numNdx] & signBitMask)
sOutputs.push_back(static_cast<deInt32>(inputs[numNdx] | signExtendMask));
else
sOutputs.push_back(static_cast<deInt32>(inputs[numNdx]));
}
extensions.push_back("VK_KHR_16bit_storage");
requiredFeatures.ext16BitStorage = EXT16BITSTORAGEFEATURES_PUSH_CONSTANT;
fragments["capability"] = "OpCapability StoragePushConstant16\n";
fragments["extension"] = "OpExtension \"SPV_KHR_16bit_storage\"";
pcs.setPushConstant(BufferSp(new Int16Buffer(inputs)));
getDefaultColors(defaultColors);
const StringTemplate testFun (
"%test_code = OpFunction %v4f32 None %v4f32_function\n"
" %param = OpFunctionParameter %v4f32\n"
"%entry = OpLabel\n"
" %i = OpVariable %fp_i32 Function\n"
" OpStore %i %c_i32_0\n"
" OpBranch %loop\n"
" %loop = OpLabel\n"
" %15 = OpLoad %i32 %i\n"
" %lt = OpSLessThan %bool %15 %c_i32_${count}\n"
" OpLoopMerge %merge %inc None\n"
" OpBranchConditional %lt %write %merge\n"
"%write = OpLabel\n"
" %30 = OpLoad %i32 %i\n"
" %src = OpAccessChain %pp_${type16} %pc16 %c_i32_0 %30\n"
"%val16 = OpLoad %${type16} %src\n"
"%val32 = ${convert} %${type32} %val16\n"
" %dst = OpAccessChain %up_${type32} %ssbo32 %c_i32_0 %30\n"
" OpStore %dst %val32\n"
" OpBranch %inc\n"
" %inc = OpLabel\n"
" %37 = OpLoad %i32 %i\n"
" %39 = OpIAdd %i32 %37 %c_i32_1\n"
" OpStore %i %39\n"
" OpBranch %loop\n"
"%merge = OpLabel\n"
" OpReturnValue %param\n"
"OpFunctionEnd\n");
{ // Scalar cases
const StringTemplate preMain (
" %${type16} = OpTypeInt 16 ${signed}\n"
" %c_i32_${count} = OpConstant %i32 ${count}\n" // Should be the same as numDataPoints
"%a${count}${type16} = OpTypeArray %${type16} %c_i32_${count}\n"
"%a${count}${type32} = OpTypeArray %${type32} %c_i32_${count}\n"
" %pp_${type16} = OpTypePointer PushConstant %${type16}\n"
" %up_${type32} = OpTypePointer Uniform %${type32}\n"
" %SSBO32 = OpTypeStruct %a${count}${type32}\n"
" %up_SSBO32 = OpTypePointer Uniform %SSBO32\n"
" %ssbo32 = OpVariable %up_SSBO32 Uniform\n"
" %PC16 = OpTypeStruct %a${count}${type16}\n"
" %pp_PC16 = OpTypePointer PushConstant %PC16\n"
" %pc16 = OpVariable %pp_PC16 PushConstant\n");
const StringTemplate decoration (
"OpDecorate %a${count}${type16} ArrayStride 2\n"
"OpDecorate %a${count}${type32} ArrayStride 4\n"
"OpDecorate %SSBO32 BufferBlock\n"
"OpMemberDecorate %SSBO32 0 Offset 0\n"
"OpDecorate %PC16 Block\n"
"OpMemberDecorate %PC16 0 Offset 0\n"
"OpDecorate %ssbo32 DescriptorSet 0\n"
"OpDecorate %ssbo32 Binding 0\n");
{ // signed int
map<string, string> specs;
specs["type16"] = "i16";
specs["type32"] = "i32";
specs["signed"] = "1";
specs["count"] = "64";
specs["convert"] = "OpSConvert";
fragments["testfun"] = testFun.specialize(specs);
fragments["pre_main"] = preMain.specialize(specs);
fragments["decoration"] = decoration.specialize(specs);
resources.outputs.clear();
resources.outputs.push_back(std::make_pair(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, BufferSp(new Int32Buffer(sOutputs))));
createTestsForAllStages("sint_scalar", defaultColors, defaultColors, fragments, pcs, resources, extensions, testGroup, requiredFeatures);
}
{ // signed int
map<string, string> specs;
specs["type16"] = "u16";
specs["type32"] = "u32";
specs["signed"] = "0";
specs["count"] = "64";
specs["convert"] = "OpUConvert";
fragments["testfun"] = testFun.specialize(specs);
fragments["pre_main"] = preMain.specialize(specs);
fragments["decoration"] = decoration.specialize(specs);
resources.outputs.clear();
resources.outputs.push_back(std::make_pair(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, BufferSp(new Int32Buffer(uOutputs))));
createTestsForAllStages("uint_scalar", defaultColors, defaultColors, fragments, pcs, resources, extensions, testGroup, requiredFeatures);
}
}
{ // Vector cases
const StringTemplate preMain (
" %${base_type16} = OpTypeInt 16 ${signed}\n"
" %${type16} = OpTypeVector %${base_type16} 2\n"
" %c_i32_${count} = OpConstant %i32 ${count}\n"
"%a${count}${type16} = OpTypeArray %${type16} %c_i32_${count}\n"
"%a${count}${type32} = OpTypeArray %${type32} %c_i32_${count}\n"
" %pp_${type16} = OpTypePointer PushConstant %${type16}\n"
" %up_${type32} = OpTypePointer Uniform %${type32}\n"
" %SSBO32 = OpTypeStruct %a${count}${type32}\n"
" %up_SSBO32 = OpTypePointer Uniform %SSBO32\n"
" %ssbo32 = OpVariable %up_SSBO32 Uniform\n"
" %PC16 = OpTypeStruct %a${count}${type16}\n"
" %pp_PC16 = OpTypePointer PushConstant %PC16\n"
" %pc16 = OpVariable %pp_PC16 PushConstant\n");
const StringTemplate decoration (
"OpDecorate %a${count}${type16} ArrayStride 4\n"
"OpDecorate %a${count}${type32} ArrayStride 8\n"
"OpDecorate %SSBO32 BufferBlock\n"
"OpMemberDecorate %SSBO32 0 Offset 0\n"
"OpDecorate %PC16 Block\n"
"OpMemberDecorate %PC16 0 Offset 0\n"
"OpDecorate %ssbo32 DescriptorSet 0\n"
"OpDecorate %ssbo32 Binding 0\n");
{ // signed int
map<string, string> specs;
specs["base_type16"] = "i16";
specs["type16"] = "v2i16";
specs["type32"] = "v2i32";
specs["signed"] = "1";
specs["count"] = "32"; // 64 / 2
specs["convert"] = "OpSConvert";
fragments["testfun"] = testFun.specialize(specs);
fragments["pre_main"] = preMain.specialize(specs);
fragments["decoration"] = decoration.specialize(specs);
resources.outputs.clear();
resources.outputs.push_back(std::make_pair(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, BufferSp(new Int32Buffer(sOutputs))));
createTestsForAllStages("sint_vector", defaultColors, defaultColors, fragments, pcs, resources, extensions, testGroup, requiredFeatures);
}
{ // signed int
map<string, string> specs;
specs["base_type16"] = "u16";
specs["type16"] = "v2u16";
specs["type32"] = "v2u32";
specs["signed"] = "0";
specs["count"] = "32";
specs["convert"] = "OpUConvert";
fragments["testfun"] = testFun.specialize(specs);
fragments["pre_main"] = preMain.specialize(specs);
fragments["decoration"] = decoration.specialize(specs);
resources.outputs.clear();
resources.outputs.push_back(std::make_pair(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, BufferSp(new Int32Buffer(uOutputs))));
createTestsForAllStages("uint_vector", defaultColors, defaultColors, fragments, pcs, resources, extensions, testGroup, requiredFeatures);
}
}
}
void addGraphics16BitStorageUniformInt16To32Group (tcu::TestCaseGroup* testGroup)
{
de::Random rnd (deStringHash(testGroup->getName()));
map<string, string> fragments;
const deUint32 numDataPoints = 256;
RGBA defaultColors[4];
vector<deInt16> inputs = getInt16s(rnd, numDataPoints);
vector<deInt32> sOutputs;
vector<deInt32> uOutputs;
GraphicsResources resources;
vector<string> extensions;
const deUint16 signBitMask = 0x8000;
const deUint32 signExtendMask = 0xffff0000;
const StringTemplate capabilities ("OpCapability ${cap}\n");
sOutputs.reserve(inputs.size());
uOutputs.reserve(inputs.size());
for (deUint32 numNdx = 0; numNdx < inputs.size(); ++numNdx)
{
uOutputs.push_back(static_cast<deUint16>(inputs[numNdx]));
if (inputs[numNdx] & signBitMask)
sOutputs.push_back(static_cast<deInt32>(inputs[numNdx] | signExtendMask));
else
sOutputs.push_back(static_cast<deInt32>(inputs[numNdx]));
}
resources.inputs.push_back(std::make_pair(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, BufferSp(new Int16Buffer(inputs))));
extensions.push_back("VK_KHR_16bit_storage");
fragments["extension"] = "OpExtension \"SPV_KHR_16bit_storage\"";
getDefaultColors(defaultColors);
struct IntegerFacts
{
const char* name;
const char* type32;
const char* type16;
const char* opcode;
bool isSigned;
};
const IntegerFacts intFacts[] =
{
{"sint", "%i32", "%i16", "OpSConvert", true},
{"uint", "%u32", "%u16", "OpUConvert", false},
};
const StringTemplate scalarPreMain (
"${itype16} = OpTypeInt 16 ${signed}\n"
" %c_i32_256 = OpConstant %i32 256\n"
" %up_i32 = OpTypePointer Uniform ${itype32}\n"
" %up_i16 = OpTypePointer Uniform ${itype16}\n"
" %ra_i32 = OpTypeArray ${itype32} %c_i32_256\n"
" %ra_i16 = OpTypeArray ${itype16} %c_i32_256\n"
" %SSBO32 = OpTypeStruct %ra_i32\n"
" %SSBO16 = OpTypeStruct %ra_i16\n"
"%up_SSBO32 = OpTypePointer Uniform %SSBO32\n"
"%up_SSBO16 = OpTypePointer Uniform %SSBO16\n"
" %ssbo32 = OpVariable %up_SSBO32 Uniform\n"
" %ssbo16 = OpVariable %up_SSBO16 Uniform\n");
const StringTemplate scalarDecoration (
"OpDecorate %ra_i32 ArrayStride 4\n"
"OpDecorate %ra_i16 ArrayStride 2\n"
"OpMemberDecorate %SSBO32 0 Offset 0\n"
"OpMemberDecorate %SSBO16 0 Offset 0\n"
"OpDecorate %SSBO32 BufferBlock\n"
"OpDecorate %SSBO16 ${indecor}\n"
"OpDecorate %ssbo32 DescriptorSet 0\n"
"OpDecorate %ssbo16 DescriptorSet 0\n"
"OpDecorate %ssbo32 Binding 1\n"
"OpDecorate %ssbo16 Binding 0\n");
const StringTemplate scalarTestFunc (
"%test_code = OpFunction %v4f32 None %v4f32_function\n"
" %param = OpFunctionParameter %v4f32\n"
"%entry = OpLabel\n"
" %i = OpVariable %fp_i32 Function\n"
" OpStore %i %c_i32_0\n"
" OpBranch %loop\n"
" %loop = OpLabel\n"
" %15 = OpLoad %i32 %i\n"
" %lt = OpSLessThan %bool %15 %c_i32_256\n"
" OpLoopMerge %merge %inc None\n"
" OpBranchConditional %lt %write %merge\n"
"%write = OpLabel\n"
" %30 = OpLoad %i32 %i\n"
" %src = OpAccessChain %up_i16 %ssbo16 %c_i32_0 %30\n"
"%val16 = OpLoad ${itype16} %src\n"
"%val32 = ${convert} ${itype32} %val16\n"
" %dst = OpAccessChain %up_i32 %ssbo32 %c_i32_0 %30\n"
" OpStore %dst %val32\n"
" OpBranch %inc\n"
" %inc = OpLabel\n"
" %37 = OpLoad %i32 %i\n"
" %39 = OpIAdd %i32 %37 %c_i32_1\n"
" OpStore %i %39\n"
" OpBranch %loop\n"
"%merge = OpLabel\n"
" OpReturnValue %param\n"
"OpFunctionEnd\n");
const StringTemplate vecPreMain (
"${itype16} = OpTypeInt 16 ${signed}\n"
"%c_i32_128 = OpConstant %i32 128\n"
"%v2itype16 = OpTypeVector ${itype16} 2\n"
"%v2itype32 = OpTypeVector ${itype32} 2\n"
" %up_v2i32 = OpTypePointer Uniform %v2itype32\n"
" %up_v2i16 = OpTypePointer Uniform %v2itype16\n"
" %ra_v2i32 = OpTypeArray %v2itype32 %c_i32_128\n"
" %ra_v2i16 = OpTypeArray %v2itype16 %c_i32_128\n"
" %SSBO32 = OpTypeStruct %ra_v2i32\n"
" %SSBO16 = OpTypeStruct %ra_v2i16\n"
"%up_SSBO32 = OpTypePointer Uniform %SSBO32\n"
"%up_SSBO16 = OpTypePointer Uniform %SSBO16\n"
" %ssbo32 = OpVariable %up_SSBO32 Uniform\n"
" %ssbo16 = OpVariable %up_SSBO16 Uniform\n");
const StringTemplate vecDecoration (
"OpDecorate %ra_v2i32 ArrayStride 8\n"
"OpDecorate %ra_v2i16 ArrayStride 4\n"
"OpMemberDecorate %SSBO32 0 Offset 0\n"
"OpMemberDecorate %SSBO16 0 Offset 0\n"
"OpDecorate %SSBO32 BufferBlock\n"
"OpDecorate %SSBO16 ${indecor}\n"
"OpDecorate %ssbo32 DescriptorSet 0\n"
"OpDecorate %ssbo16 DescriptorSet 0\n"
"OpDecorate %ssbo32 Binding 1\n"
"OpDecorate %ssbo16 Binding 0\n");
const StringTemplate vecTestFunc (
"%test_code = OpFunction %v4f32 None %v4f32_function\n"
" %param = OpFunctionParameter %v4f32\n"
"%entry = OpLabel\n"
" %i = OpVariable %fp_i32 Function\n"
" OpStore %i %c_i32_0\n"
" OpBranch %loop\n"
" %loop = OpLabel\n"
" %15 = OpLoad %i32 %i\n"
" %lt = OpSLessThan %bool %15 %c_i32_128\n"
" OpLoopMerge %merge %inc None\n"
" OpBranchConditional %lt %write %merge\n"
"%write = OpLabel\n"
" %30 = OpLoad %i32 %i\n"
" %src = OpAccessChain %up_v2i16 %ssbo16 %c_i32_0 %30\n"
"%val16 = OpLoad %v2itype16 %src\n"
"%val32 = ${convert} %v2itype32 %val16\n"
" %dst = OpAccessChain %up_v2i32 %ssbo32 %c_i32_0 %30\n"
" OpStore %dst %val32\n"
" OpBranch %inc\n"
" %inc = OpLabel\n"
" %37 = OpLoad %i32 %i\n"
" %39 = OpIAdd %i32 %37 %c_i32_1\n"
" OpStore %i %39\n"
" OpBranch %loop\n"
"%merge = OpLabel\n"
" OpReturnValue %param\n"
"OpFunctionEnd\n");
struct Category
{
const char* name;
const StringTemplate& preMain;
const StringTemplate& decoration;
const StringTemplate& testFunction;
};
const Category categories[] =
{
{"scalar", scalarPreMain, scalarDecoration, scalarTestFunc},
{"vector", vecPreMain, vecDecoration, vecTestFunc},
};
for (deUint32 catIdx = 0; catIdx < DE_LENGTH_OF_ARRAY(categories); ++catIdx)
for (deUint32 capIdx = 0; capIdx < DE_LENGTH_OF_ARRAY(CAPABILITIES); ++capIdx)
for (deUint32 factIdx = 0; factIdx < DE_LENGTH_OF_ARRAY(intFacts); ++factIdx)
{
map<string, string> specs;
string name = string(CAPABILITIES[capIdx].name) + "_" + categories[catIdx].name + "_" + intFacts[factIdx].name;
specs["cap"] = CAPABILITIES[capIdx].cap;
specs["indecor"] = CAPABILITIES[capIdx].decor;
specs["itype32"] = intFacts[factIdx].type32;
specs["itype16"] = intFacts[factIdx].type16;
if (intFacts[factIdx].isSigned)
specs["signed"] = "1";
else
specs["signed"] = "0";
specs["convert"] = intFacts[factIdx].opcode;
fragments["pre_main"] = categories[catIdx].preMain.specialize(specs);
fragments["testfun"] = categories[catIdx].testFunction.specialize(specs);
fragments["capability"] = capabilities.specialize(specs);
fragments["decoration"] = categories[catIdx].decoration.specialize(specs);
resources.inputs.back().first = CAPABILITIES[capIdx].dtype;
resources.outputs.clear();
if (intFacts[factIdx].isSigned)
resources.outputs.push_back(std::make_pair(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, BufferSp(new Int32Buffer(sOutputs))));
else
resources.outputs.push_back(std::make_pair(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, BufferSp(new Int32Buffer(uOutputs))));
createTestsForAllStages(name, defaultColors, defaultColors, fragments, resources, extensions, testGroup, get16BitStorageFeatures(CAPABILITIES[capIdx].name));
}
}
void addGraphics16BitStorageUniformFloat16To32Group (tcu::TestCaseGroup* testGroup)
{
de::Random rnd (deStringHash(testGroup->getName()));
map<string, string> fragments;
GraphicsResources resources;
vector<string> extensions;
const deUint32 numDataPoints = 256;
RGBA defaultColors[4];
const StringTemplate capabilities ("OpCapability ${cap}\n");
vector<deFloat16> float16Data = getFloat16s(rnd, numDataPoints);
vector<float> float32Data;
float32Data.reserve(numDataPoints);
for (deUint32 numIdx = 0; numIdx < numDataPoints; ++numIdx)
float32Data.push_back(deFloat16To32(float16Data[numIdx]));
resources.inputs.push_back(std::make_pair(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, BufferSp(new Float16Buffer(float16Data))));
resources.outputs.push_back(std::make_pair(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, BufferSp(new Float32Buffer(float32Data))));
resources.verifyIO = check32BitFloats;
extensions.push_back("VK_KHR_16bit_storage");
fragments["extension"] = "OpExtension \"SPV_KHR_16bit_storage\"";
getDefaultColors(defaultColors);
{ // scalar cases
fragments["pre_main"] =
" %f16 = OpTypeFloat 16\n"
"%c_i32_256 = OpConstant %i32 256\n"
" %up_f32 = OpTypePointer Uniform %f32\n"
" %up_f16 = OpTypePointer Uniform %f16\n"
" %ra_f32 = OpTypeArray %f32 %c_i32_256\n"
" %ra_f16 = OpTypeArray %f16 %c_i32_256\n"
" %SSBO32 = OpTypeStruct %ra_f32\n"
" %SSBO16 = OpTypeStruct %ra_f16\n"
"%up_SSBO32 = OpTypePointer Uniform %SSBO32\n"
"%up_SSBO16 = OpTypePointer Uniform %SSBO16\n"
" %ssbo32 = OpVariable %up_SSBO32 Uniform\n"
" %ssbo16 = OpVariable %up_SSBO16 Uniform\n";
const StringTemplate decoration (
"OpDecorate %ra_f32 ArrayStride 4\n"
"OpDecorate %ra_f16 ArrayStride 2\n"
"OpMemberDecorate %SSBO32 0 Offset 0\n"
"OpMemberDecorate %SSBO16 0 Offset 0\n"
"OpDecorate %SSBO32 BufferBlock\n"
"OpDecorate %SSBO16 ${indecor}\n"
"OpDecorate %ssbo32 DescriptorSet 0\n"
"OpDecorate %ssbo16 DescriptorSet 0\n"
"OpDecorate %ssbo32 Binding 1\n"
"OpDecorate %ssbo16 Binding 0\n");
// ssbo32[] <- convert ssbo16[] to 32bit float
fragments["testfun"] =
"%test_code = OpFunction %v4f32 None %v4f32_function\n"
" %param = OpFunctionParameter %v4f32\n"
"%entry = OpLabel\n"
" %i = OpVariable %fp_i32 Function\n"
" OpStore %i %c_i32_0\n"
" OpBranch %loop\n"
" %loop = OpLabel\n"
" %15 = OpLoad %i32 %i\n"
" %lt = OpSLessThan %bool %15 %c_i32_256\n"
" OpLoopMerge %merge %inc None\n"
" OpBranchConditional %lt %write %merge\n"
"%write = OpLabel\n"
" %30 = OpLoad %i32 %i\n"
" %src = OpAccessChain %up_f16 %ssbo16 %c_i32_0 %30\n"
"%val16 = OpLoad %f16 %src\n"
"%val32 = OpFConvert %f32 %val16\n"
" %dst = OpAccessChain %up_f32 %ssbo32 %c_i32_0 %30\n"
" OpStore %dst %val32\n"
" OpBranch %inc\n"
" %inc = OpLabel\n"
" %37 = OpLoad %i32 %i\n"
" %39 = OpIAdd %i32 %37 %c_i32_1\n"
" OpStore %i %39\n"
" OpBranch %loop\n"
"%merge = OpLabel\n"
" OpReturnValue %param\n"
"OpFunctionEnd\n";
for (deUint32 capIdx = 0; capIdx < DE_LENGTH_OF_ARRAY(CAPABILITIES); ++capIdx)
{
map<string, string> specs;
string testName = string(CAPABILITIES[capIdx].name) + "_scalar_float";
specs["cap"] = CAPABILITIES[capIdx].cap;
specs