Test that UBO and SB layout rules are correct and consistent for all backends

Fixed: tint:898
Change-Id: I81633715efd213cf0c566aa44d00bee82aa2c17a
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/54642
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Ben Clayton <bclayton@google.com>
This commit is contained in:
Ben Clayton 2021-06-25 08:51:52 +00:00 committed by Dawn LUCI CQ
parent df4c60546e
commit 17686f78cb
2 changed files with 518 additions and 0 deletions

View File

@ -294,6 +294,7 @@ source_set("dawn_end2end_tests_sources") {
"end2end/CompressedTextureFormatTests.cpp",
"end2end/ComputeCopyStorageBufferTests.cpp",
"end2end/ComputeDispatchTests.cpp",
"end2end/ComputeLayoutMemoryBufferTests.cpp",
"end2end/ComputeSharedMemoryTests.cpp",
"end2end/ComputeStorageBufferBarrierTests.cpp",
"end2end/CopyTests.cpp",

View File

@ -0,0 +1,517 @@
// Copyright 2021 The Dawn Authors
//
// 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.
#include "tests/DawnTest.h"
#include "common/Math.h"
#include "utils/WGPUHelpers.h"
#include <array>
#include <functional>
namespace {
// Helper for replacing all occurrences of substr in str with replacement
std::string ReplaceAll(std::string str,
const std::string& substr,
const std::string& replacement) {
size_t pos = 0;
while ((pos = str.find(substr, pos)) != std::string::npos) {
str.replace(pos, substr.length(), replacement);
pos += replacement.length();
}
return str;
}
// DataMatcherCallback is the callback function by DataMatcher.
// It is called for each contiguous sequence of bytes that should be checked
// for equality.
// offset and size are in units of bytes.
using DataMatcherCallback = std::function<void(uint32_t offset, uint32_t size)>;
// DataMatcher is a function pointer to a data matching function.
// size is the total number of bytes being considered for matching.
// The callback may be called once or multiple times, and may only consider
// part of the interval [0, size)
using DataMatcher = void (*)(uint32_t size, DataMatcherCallback);
// FullDataMatcher is a DataMatcher that calls callback with the interval
// [0, size)
void FullDataMatcher(uint32_t size, DataMatcherCallback callback) {
callback(0, size);
}
// StridedDataMatcher is a DataMatcher that calls callback with the strided
// intervals of length BYTES_TO_MATCH, skipping BYTES_TO_SKIP.
// For example: StridedDataMatcher<2, 4>(18, callback) will call callback
// with the intervals: [0, 2), [6, 8), [12, 14)
template <int BYTES_TO_MATCH, int BYTES_TO_SKIP>
void StridedDataMatcher(uint32_t size, DataMatcherCallback callback) {
uint32_t offset = 0;
while (offset < size) {
callback(offset, BYTES_TO_MATCH);
offset += BYTES_TO_MATCH + BYTES_TO_SKIP;
}
}
// Align returns the WGSL decoration for an explicit structure field alignment
std::string AlignDeco(uint32_t value) {
return "[[align(" + std::to_string(value) + ")]] ";
}
} // namespace
// Field holds test parameters for ComputeLayoutMemoryBufferTests.Fields
struct Field {
const char* type; // Type of the field
uint32_t align; // Alignment of the type in bytes
uint32_t size; // Natural size of the type in bytes
uint32_t padded_size = 0; // Decorated (extended) size of the type in bytes
DataMatcher matcher = &FullDataMatcher; // The matching method
bool storage_buffer_only = false; // This should only be used for storage buffer tests
// Sets the padded_size to value.
// Returns this Field so calls can be chained.
Field& PaddedSize(uint32_t value) {
padded_size = value;
return *this;
}
// Sets the matcher to a StridedDataMatcher<BYTES_TO_MATCH, BYTES_TO_SKIP>.
// Returns this Field so calls can be chained.
template <int BYTES_TO_MATCH, int BYTES_TO_SKIP>
Field& Strided() {
matcher = &StridedDataMatcher<BYTES_TO_MATCH, BYTES_TO_SKIP>;
return *this;
}
// Marks that this should only be used for storage buffer tests.
// Returns this Field so calls can be chained.
Field& StorageBufferOnly() {
storage_buffer_only = true;
return *this;
}
};
// StorageClass is an enumerator of storage classes used by ComputeLayoutMemoryBufferTests.Fields
enum class StorageClass {
Uniform,
Storage,
};
std::ostream& operator<<(std::ostream& o, StorageClass storageClass) {
switch (storageClass) {
case StorageClass::Uniform:
o << "uniform";
break;
case StorageClass::Storage:
o << "storage";
break;
}
return o;
}
std::ostream& operator<<(std::ostream& o, Field field) {
o << "[[align(" << field.align << "), size("
<< (field.padded_size > 0 ? field.padded_size : field.size) << ")]] " << field.type;
return o;
}
DAWN_TEST_PARAM_STRUCT(ComputeLayoutMemoryBufferTestParams, StorageClass, Field)
class ComputeLayoutMemoryBufferTests
: public DawnTestWithParams<ComputeLayoutMemoryBufferTestParams> {};
TEST_P(ComputeLayoutMemoryBufferTests, Fields) {
// Sentinel value markers codes used to check that the start and end of
// structures are correctly aligned. Each of these codes are distinct and
// are not likely to be confused with data.
constexpr uint32_t kDataHeaderCode = 0xa0b0c0a0u;
constexpr uint32_t kDataFooterCode = 0x40302010u;
constexpr uint32_t kInputHeaderCode = 0x91827364u;
constexpr uint32_t kInputFooterCode = 0x19283764u;
// Byte codes used for field padding. The MSB is set for each of these.
// The field data has the MSB 0.
constexpr uint8_t kDataAlignPaddingCode = 0xfeu;
constexpr uint8_t kFieldAlignPaddingCode = 0xfdu;
constexpr uint8_t kFieldSizePaddingCode = 0xdcu;
constexpr uint8_t kDataSizePaddingCode = 0xdbu;
constexpr uint8_t kInputFooterAlignPaddingCode = 0xdau;
constexpr uint8_t kInputTailPaddingCode = 0xd9u;
// Status codes returned by the shader.
constexpr uint32_t kStatusBadInputHeader = 100u;
constexpr uint32_t kStatusBadInputFooter = 101u;
constexpr uint32_t kStatusBadDataHeader = 102u;
constexpr uint32_t kStatusBadDataFooter = 103u;
constexpr uint32_t kStatusOk = 200u;
const Field& field = GetParam().mField;
const bool isUniform = GetParam().mStorageClass == StorageClass::Uniform;
std::string shader = R"(
struct Data {
header : u32;
[[align({field_align}), size({field_size})]] field : {field_type};
footer : u32;
};
[[block]] struct Input {
header : u32;
{data_align}data : Data;
{footer_align}footer : u32;
};
[[block]] struct Output {
data : {field_type};
};
[[block]] struct Status {
code : u32;
};
[[set(0), binding(0)]] var<{input_qualifiers}> input : Input;
[[set(0), binding(1)]] var<storage, read_write> output : Output;
[[set(0), binding(2)]] var<storage, read_write> status : Status;
[[stage(compute), workgroup_size(1,1,1)]]
fn main() {
if (input.header != {input_header_code}u) {
status.code = {status_bad_input_header}u;
} elseif (input.footer != {input_footer_code}u) {
status.code = {status_bad_input_footer}u;
} elseif (input.data.header != {data_header_code}u) {
status.code = {status_bad_data_header}u;
} elseif (input.data.footer != {data_footer_code}u) {
status.code = {status_bad_data_footer}u;
} else {
status.code = {status_ok}u;
output.data = input.data.field;
}
})";
// https://www.w3.org/TR/WGSL/#alignment-and-size
// Structure size: roundUp(AlignOf(S), OffsetOf(S, L) + SizeOf(S, L))
// https://www.w3.org/TR/WGSL/#storage-class-constraints
// RequiredAlignOf(S, uniform): roundUp(16, max(AlignOf(T0), ..., AlignOf(TN)))
uint32_t dataAlign = isUniform ? std::max(16u, field.align) : field.align;
// https://www.w3.org/TR/WGSL/#structure-layout-rules
// Note: When underlying the target is a Vulkan device, we assume the device does not support
// the scalarBlockLayout feature. Therefore, a data value must not be placed in the padding at
// the end of a structure or matrix, nor in the padding at the last element of an array.
uint32_t footerAlign = isUniform ? 16 : 4;
shader = ReplaceAll(shader, "{data_align}", isUniform ? AlignDeco(dataAlign) : "");
shader = ReplaceAll(shader, "{field_align}", std::to_string(field.align));
shader = ReplaceAll(shader, "{footer_align}", isUniform ? AlignDeco(footerAlign) : "");
shader = ReplaceAll(shader, "{field_size}",
std::to_string(field.padded_size > 0 ? field.padded_size : field.size));
shader = ReplaceAll(shader, "{field_type}", field.type);
shader = ReplaceAll(shader, "{input_header_code}", std::to_string(kInputHeaderCode));
shader = ReplaceAll(shader, "{input_footer_code}", std::to_string(kInputFooterCode));
shader = ReplaceAll(shader, "{data_header_code}", std::to_string(kDataHeaderCode));
shader = ReplaceAll(shader, "{data_footer_code}", std::to_string(kDataFooterCode));
shader = ReplaceAll(shader, "{status_bad_input_header}", std::to_string(kStatusBadInputHeader));
shader = ReplaceAll(shader, "{status_bad_input_footer}", std::to_string(kStatusBadInputFooter));
shader = ReplaceAll(shader, "{status_bad_data_header}", std::to_string(kStatusBadDataHeader));
shader = ReplaceAll(shader, "{status_bad_data_footer}", std::to_string(kStatusBadDataFooter));
shader = ReplaceAll(shader, "{status_ok}", std::to_string(kStatusOk));
shader = ReplaceAll(shader, "{input_qualifiers}",
isUniform ? "uniform" //
: "storage, read_write");
// Set up shader and pipeline
auto module = utils::CreateShaderModule(device, shader.c_str());
wgpu::ComputePipelineDescriptor csDesc;
csDesc.compute.module = module;
csDesc.compute.entryPoint = "main";
wgpu::ComputePipeline pipeline = device.CreateComputePipeline(&csDesc);
// Build the input and expected data.
std::vector<uint8_t> inputData; // The whole SSBO data
std::vector<uint8_t> expectedData; // The expected data to be copied by the shader
{
auto PushU32 = [&inputData](uint32_t u32) {
inputData.emplace_back((u32 >> 0) & 0xff);
inputData.emplace_back((u32 >> 8) & 0xff);
inputData.emplace_back((u32 >> 16) & 0xff);
inputData.emplace_back((u32 >> 24) & 0xff);
};
auto AlignTo = [&inputData](uint32_t alignment, uint8_t code) {
uint32_t target = Align(inputData.size(), alignment);
uint32_t bytes = target - inputData.size();
for (uint32_t i = 0; i < bytes; i++) {
inputData.emplace_back(code);
}
};
PushU32(kInputHeaderCode); // Input.header
AlignTo(dataAlign, kDataAlignPaddingCode); // Input.data
{
PushU32(kDataHeaderCode); // Input.data.header
AlignTo(field.align, kFieldAlignPaddingCode); // Input.data.field
for (uint32_t i = 0; i < field.size; i++) {
// The data has the MSB cleared to distinguish it from the
// padding codes.
uint8_t code = i & 0x7f;
inputData.emplace_back(code); // Input.data.field
expectedData.emplace_back(code);
}
for (uint32_t i = field.size; i < field.padded_size; i++) {
inputData.emplace_back(kFieldSizePaddingCode); // Input.data.field padding
}
PushU32(kDataFooterCode); // Input.data.footer
AlignTo(field.align, kDataSizePaddingCode); // Input.data padding
}
AlignTo(footerAlign, kInputFooterAlignPaddingCode); // Input.footer [[align]]
PushU32(kInputFooterCode); // Input.footer
AlignTo(256, kInputTailPaddingCode); // Input padding
}
// Set up input storage buffer
wgpu::Buffer inputBuf = utils::CreateBufferFromData(
device, inputData.data(), inputData.size(),
wgpu::BufferUsage::CopySrc | wgpu::BufferUsage::CopyDst |
(isUniform ? wgpu::BufferUsage::Uniform : wgpu::BufferUsage::Storage));
// Set up output storage buffer
wgpu::BufferDescriptor outputDesc;
outputDesc.size = field.size;
outputDesc.usage =
wgpu::BufferUsage::Storage | wgpu::BufferUsage::CopySrc | wgpu::BufferUsage::CopyDst;
wgpu::Buffer outputBuf = device.CreateBuffer(&outputDesc);
// Set up status storage buffer
wgpu::BufferDescriptor statusDesc;
statusDesc.size = 4u;
statusDesc.usage =
wgpu::BufferUsage::Storage | wgpu::BufferUsage::CopySrc | wgpu::BufferUsage::CopyDst;
wgpu::Buffer statusBuf = device.CreateBuffer(&statusDesc);
// Set up bind group and issue dispatch
wgpu::BindGroup bindGroup = utils::MakeBindGroup(device, pipeline.GetBindGroupLayout(0),
{
{0, inputBuf},
{1, outputBuf},
{2, statusBuf},
});
wgpu::CommandBuffer commands;
{
wgpu::CommandEncoder encoder = device.CreateCommandEncoder();
wgpu::ComputePassEncoder pass = encoder.BeginComputePass();
pass.SetPipeline(pipeline);
pass.SetBindGroup(0, bindGroup);
pass.Dispatch(1);
pass.EndPass();
commands = encoder.Finish();
}
queue.Submit(1, &commands);
// Check the status
EXPECT_BUFFER_U32_EQ(kStatusOk, statusBuf, 0) << "status code error" << std::endl
<< "Shader: " << shader;
// Check the data
field.matcher(field.size, [&](uint32_t offset, uint32_t size) {
EXPECT_BUFFER_U8_RANGE_EQ(expectedData.data() + offset, outputBuf, offset, size)
<< "offset: " << offset;
});
}
namespace {
auto GenerateParams() {
auto params = MakeParamGenerator<ComputeLayoutMemoryBufferTestParams>(
{
D3D12Backend({"use_tint_generator"}), MetalBackend({"use_tint_generator"}),
VulkanBackend(),
// TODO(crbug.com/dawn/942)
// There was a compiler error: Buffer block cannot be expressed as any of std430,
// std140, scalar, even with enhanced layouts. You can try flattening this block to
// support a more flexible layout.
// OpenGLBackend(),
// OpenGLESBackend(),
},
{StorageClass::Storage, StorageClass::Uniform},
{
// See https://www.w3.org/TR/WGSL/#alignment-and-size
// Scalar types with no custom alignment or size
Field{"i32", /* align */ 4, /* size */ 4},
Field{"u32", /* align */ 4, /* size */ 4},
Field{"f32", /* align */ 4, /* size */ 4},
// Scalar types with custom alignment
Field{"i32", /* align */ 16, /* size */ 4},
Field{"u32", /* align */ 16, /* size */ 4},
Field{"f32", /* align */ 16, /* size */ 4},
// Scalar types with custom size
Field{"i32", /* align */ 4, /* size */ 4}.PaddedSize(24),
Field{"u32", /* align */ 4, /* size */ 4}.PaddedSize(24),
Field{"f32", /* align */ 4, /* size */ 4}.PaddedSize(24),
// Vector types with no custom alignment or size
Field{"vec2<i32>", /* align */ 8, /* size */ 8},
Field{"vec2<u32>", /* align */ 8, /* size */ 8},
Field{"vec2<f32>", /* align */ 8, /* size */ 8},
Field{"vec3<i32>", /* align */ 16, /* size */ 12},
Field{"vec3<u32>", /* align */ 16, /* size */ 12},
Field{"vec3<f32>", /* align */ 16, /* size */ 12},
Field{"vec4<i32>", /* align */ 16, /* size */ 16},
Field{"vec4<u32>", /* align */ 16, /* size */ 16},
Field{"vec4<f32>", /* align */ 16, /* size */ 16},
// Vector types with custom alignment
Field{"vec2<i32>", /* align */ 32, /* size */ 8},
Field{"vec2<u32>", /* align */ 32, /* size */ 8},
Field{"vec2<f32>", /* align */ 32, /* size */ 8},
Field{"vec3<i32>", /* align */ 32, /* size */ 12},
Field{"vec3<u32>", /* align */ 32, /* size */ 12},
Field{"vec3<f32>", /* align */ 32, /* size */ 12},
Field{"vec4<i32>", /* align */ 32, /* size */ 16},
Field{"vec4<u32>", /* align */ 32, /* size */ 16},
Field{"vec4<f32>", /* align */ 32, /* size */ 16},
// Vector types with custom size
Field{"vec2<i32>", /* align */ 8, /* size */ 8}.PaddedSize(24),
Field{"vec2<u32>", /* align */ 8, /* size */ 8}.PaddedSize(24),
Field{"vec2<f32>", /* align */ 8, /* size */ 8}.PaddedSize(24),
Field{"vec3<i32>", /* align */ 16, /* size */ 12}.PaddedSize(24),
Field{"vec3<u32>", /* align */ 16, /* size */ 12}.PaddedSize(24),
Field{"vec3<f32>", /* align */ 16, /* size */ 12}.PaddedSize(24),
Field{"vec4<i32>", /* align */ 16, /* size */ 16}.PaddedSize(24),
Field{"vec4<u32>", /* align */ 16, /* size */ 16}.PaddedSize(24),
Field{"vec4<f32>", /* align */ 16, /* size */ 16}.PaddedSize(24),
// Matrix types with no custom alignment or size
Field{"mat2x2<f32>", /* align */ 8, /* size */ 16},
Field{"mat3x2<f32>", /* align */ 8, /* size */ 24},
Field{"mat4x2<f32>", /* align */ 8, /* size */ 32},
Field{"mat2x3<f32>", /* align */ 16, /* size */ 32}.Strided<12, 4>(),
Field{"mat3x3<f32>", /* align */ 16, /* size */ 48}.Strided<12, 4>(),
Field{"mat4x3<f32>", /* align */ 16, /* size */ 64}.Strided<12, 4>(),
Field{"mat2x4<f32>", /* align */ 16, /* size */ 32},
Field{"mat3x4<f32>", /* align */ 16, /* size */ 48},
Field{"mat4x4<f32>", /* align */ 16, /* size */ 64},
// Matrix types with custom alignment
Field{"mat2x2<f32>", /* align */ 32, /* size */ 16},
Field{"mat3x2<f32>", /* align */ 32, /* size */ 24},
Field{"mat4x2<f32>", /* align */ 32, /* size */ 32},
Field{"mat2x3<f32>", /* align */ 32, /* size */ 32}.Strided<12, 4>(),
Field{"mat3x3<f32>", /* align */ 32, /* size */ 48}.Strided<12, 4>(),
Field{"mat4x3<f32>", /* align */ 32, /* size */ 64}.Strided<12, 4>(),
Field{"mat2x4<f32>", /* align */ 32, /* size */ 32},
Field{"mat3x4<f32>", /* align */ 32, /* size */ 48},
Field{"mat4x4<f32>", /* align */ 32, /* size */ 64},
// Matrix types with custom size
Field{"mat2x2<f32>", /* align */ 8, /* size */ 16}.PaddedSize(128),
Field{"mat3x2<f32>", /* align */ 8, /* size */ 24}.PaddedSize(128),
Field{"mat4x2<f32>", /* align */ 8, /* size */ 32}.PaddedSize(128),
Field{"mat2x3<f32>", /* align */ 16, /* size */ 32}
.PaddedSize(128)
.Strided<12, 4>(),
Field{"mat3x3<f32>", /* align */ 16, /* size */ 48}
.PaddedSize(128)
.Strided<12, 4>(),
Field{"mat4x3<f32>", /* align */ 16, /* size */ 64}
.PaddedSize(128)
.Strided<12, 4>(),
Field{"mat2x4<f32>", /* align */ 16, /* size */ 32}.PaddedSize(128),
Field{"mat3x4<f32>", /* align */ 16, /* size */ 48}.PaddedSize(128),
Field{"mat4x4<f32>", /* align */ 16, /* size */ 64}.PaddedSize(128),
// Array types with no custom alignment, size or stride
// Note: The use of StorageBufferOnly() is due to UBOs requiring 16 byte alignment
// of array elements. See https://www.w3.org/TR/WGSL/#storage-class-constraints
Field{"array<u32, 1>", /* align */ 4, /* size */ 4}.StorageBufferOnly(),
Field{"array<u32, 2>", /* align */ 4, /* size */ 8}.StorageBufferOnly(),
Field{"array<u32, 3>", /* align */ 4, /* size */ 12}.StorageBufferOnly(),
Field{"array<u32, 4>", /* align */ 4, /* size */ 16}.StorageBufferOnly(),
Field{"[[stride(16)]] array<u32, 1>", /* align */ 4, /* size */ 16}
.StorageBufferOnly()
.Strided<4, 12>(),
Field{"[[stride(16)]] array<u32, 2>", /* align */ 4, /* size */ 32}
.StorageBufferOnly()
.Strided<4, 12>(),
Field{"[[stride(16)]] array<u32, 3>", /* align */ 4, /* size */ 48}
.StorageBufferOnly()
.Strided<4, 12>(),
Field{"[[stride(16)]] array<u32, 4>", /* align */ 4, /* size */ 64}
.StorageBufferOnly()
.Strided<4, 12>(),
Field{"array<vec3<u32>, 4>", /* align */ 16, /* size */ 64}.Strided<12, 4>(),
Field{"[[stride(32)]] array<vec3<u32>, 4>", /* align */ 16, /* size */ 128}
.Strided<12, 20>(),
// Array types with custom alignment
Field{"array<u32, 1>", /* align */ 32, /* size */ 4}.StorageBufferOnly(),
Field{"array<u32, 2>", /* align */ 32, /* size */ 8}.StorageBufferOnly(),
Field{"array<u32, 3>", /* align */ 32, /* size */ 12}.StorageBufferOnly(),
Field{"array<u32, 4>", /* align */ 32, /* size */ 16}.StorageBufferOnly(),
Field{"[[stride(16)]] array<u32, 1>", /* align */ 32, /* size */ 16}
.Strided<4, 12>(),
Field{"[[stride(16)]] array<u32, 2>", /* align */ 32, /* size */ 32}
.Strided<4, 12>(),
Field{"[[stride(16)]] array<u32, 3>", /* align */ 32, /* size */ 48}
.Strided<4, 12>(),
Field{"[[stride(16)]] array<u32, 4>", /* align */ 32, /* size */ 64}
.Strided<4, 12>(),
Field{"array<vec3<u32>, 4>", /* align */ 32, /* size */ 64}.Strided<12, 4>(),
// Array types with custom size
Field{"array<u32, 1>", /* align */ 4, /* size */ 4}
.PaddedSize(128)
.StorageBufferOnly(),
Field{"array<u32, 2>", /* align */ 4, /* size */ 8}
.PaddedSize(128)
.StorageBufferOnly(),
Field{"array<u32, 3>", /* align */ 4, /* size */ 12}
.PaddedSize(128)
.StorageBufferOnly(),
Field{"array<u32, 4>", /* align */ 4, /* size */ 16}
.PaddedSize(128)
.StorageBufferOnly(),
Field{"array<vec3<u32>, 4>", /* align */ 16, /* size */ 64}
.PaddedSize(128)
.Strided<12, 4>(),
});
std::vector<ComputeLayoutMemoryBufferTestParams> filtered;
for (auto param : params) {
if (param.mStorageClass != StorageClass::Storage && param.mField.storage_buffer_only) {
continue;
}
filtered.emplace_back(param);
}
return filtered;
}
INSTANTIATE_TEST_SUITE_P(
,
ComputeLayoutMemoryBufferTests,
::testing::ValuesIn(GenerateParams()),
DawnTestBase::PrintToStringParamName("ComputeLayoutMemoryBufferTests"));
GTEST_ALLOW_UNINSTANTIATED_PARAMETERIZED_TEST(ComputeLayoutMemoryBufferTests);
} // namespace