shader_validation.h

/* Copyright (c) 2015-2020 The Khronos Group Inc.
 * Copyright (c) 2015-2020 Valve Corporation
 * Copyright (c) 2015-2020 LunarG, Inc.
 * Copyright (C) 2015-2020 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.
 *
 * Author: Chris Forbes <chrisf@ijw.co.nz>
 */
#ifndef VULKAN_SHADER_VALIDATION_H
#define VULKAN_SHADER_VALIDATION_H

#include <cassert>
#include <cstdlib>
#include <cstring>
#include <unordered_map>
#include <unordered_set>
#include <vector>

#include "vulkan/vulkan.h"
#include <SPIRV/spirv.hpp>
#include <generated/spirv_tools_commit_id.h>
#include "spirv-tools/optimizer.hpp"
#include "core_validation_types.h"

// A forward iterator over spirv instructions. Provides easy access to len, opcode, and content words
// without the caller needing to care too much about the physical SPIRV module layout.
struct spirv_inst_iter {
    std::vector<uint32_t>::const_iterator zero;
    std::vector<uint32_t>::const_iterator it;

    uint32_t len() const {
        auto result = *it >> 16;
        assert(result > 0);
        return result;
    }

    uint32_t opcode() { return *it & 0x0ffffu; }

    uint32_t const &word(unsigned n) const {
        assert(n < len());
        return it[n];
    }

    uint32_t offset() { return (uint32_t)(it - zero); }

    spirv_inst_iter() {}

    spirv_inst_iter(std::vector<uint32_t>::const_iterator zero, std::vector<uint32_t>::const_iterator it) : zero(zero), it(it) {}

    bool operator==(spirv_inst_iter const &other) const { return it == other.it; }

    bool operator!=(spirv_inst_iter const &other) const { return it != other.it; }

    spirv_inst_iter operator++(int) {  // x++
        spirv_inst_iter ii = *this;
        it += len();
        return ii;
    }

    spirv_inst_iter operator++() {  // ++x;
        it += len();
        return *this;
    }

    // The iterator and the value are the same thing.
    spirv_inst_iter &operator*() { return *this; }
    spirv_inst_iter const &operator*() const { return *this; }
};

struct shader_stage_attributes {
    char const *const name;
    bool arrayed_input;
    bool arrayed_output;
    VkShaderStageFlags stage;
};

struct decoration_set {
    enum {
        location_bit = 1 << 0,
        patch_bit = 1 << 1,
        relaxed_precision_bit = 1 << 2,
        block_bit = 1 << 3,
        buffer_block_bit = 1 << 4,
        component_bit = 1 << 5,
        input_attachment_index_bit = 1 << 6,
        descriptor_set_bit = 1 << 7,
        binding_bit = 1 << 8,
        nonwritable_bit = 1 << 9,
        builtin_bit = 1 << 10,
    };
    uint32_t flags = 0;
    uint32_t location = static_cast<uint32_t>(-1);
    uint32_t component = 0;
    uint32_t input_attachment_index = 0;
    uint32_t descriptor_set = 0;
    uint32_t binding = 0;
    uint32_t builtin = static_cast<uint32_t>(-1);

    void merge(decoration_set const &other) {
        if (other.flags & location_bit) location = other.location;
        if (other.flags & component_bit) component = other.component;
        if (other.flags & input_attachment_index_bit) input_attachment_index = other.input_attachment_index;
        if (other.flags & descriptor_set_bit) descriptor_set = other.descriptor_set;
        if (other.flags & binding_bit) binding = other.binding;
        if (other.flags & builtin_bit) builtin = other.builtin;
        flags |= other.flags;
    }

    void add(uint32_t decoration, uint32_t value);
};

struct SHADER_MODULE_STATE : public BASE_NODE {
    // The spirv image itself
    std::vector<uint32_t> words;
    // A mapping of <id> to the first word of its def. this is useful because walking type
    // trees, constant expressions, etc requires jumping all over the instruction stream.
    std::unordered_map<unsigned, unsigned> def_index;
    std::unordered_map<unsigned, decoration_set> decorations;
    struct EntryPoint {
        uint32_t offset;
        VkShaderStageFlags stage;
    };
    std::unordered_multimap<std::string, EntryPoint> entry_points;
    bool has_valid_spirv;
    bool has_specialization_constants{false};
    VkShaderModule vk_shader_module;
    uint32_t gpu_validation_shader_id;

    std::vector<uint32_t> PreprocessShaderBinary(uint32_t *src_binary, size_t binary_size, spv_target_env env) {
        std::vector<uint32_t> src(src_binary, src_binary + binary_size / sizeof(uint32_t));

        // Check if there are any group decoration instructions, and flatten them if found.
        bool has_group_decoration = false;
        bool done = false;

        // Walk through the first part of the SPIR-V module, looking for group decoration and specialization constant instructions.
        // Skip the header (5 words).
        auto itr = spirv_inst_iter(src.begin(), src.begin() + 5);
        auto itrend = spirv_inst_iter(src.begin(), src.end());
        while (itr != itrend && !done) {
            spv::Op opcode = (spv::Op)itr.opcode();
            switch (opcode) {
                case spv::OpDecorationGroup:
                case spv::OpGroupDecorate:
                case spv::OpGroupMemberDecorate:
                    has_group_decoration = true;
                    break;
                case spv::OpSpecConstantTrue:
                case spv::OpSpecConstantFalse:
                case spv::OpSpecConstant:
                case spv::OpSpecConstantComposite:
                case spv::OpSpecConstantOp:
                    has_specialization_constants = true;
                    break;
                case spv::OpFunction:
                    // An OpFunction indicates there are no more decorations
                    done = true;
                    break;
                default:
                    break;
            }
            itr++;
        }

        if (has_group_decoration) {
            spvtools::Optimizer optimizer(env);
            optimizer.RegisterPass(spvtools::CreateFlattenDecorationPass());
            std::vector<uint32_t> optimized_binary;
            // Run optimizer to flatten decorations only, set skip_validation so as to not re-run validator
            auto result =
                optimizer.Run(src_binary, binary_size / sizeof(uint32_t), &optimized_binary, spvtools::ValidatorOptions(), true);
            if (result) {
                return optimized_binary;
            }
        }
        // Return the original module.
        return src;
    }

    SHADER_MODULE_STATE(VkShaderModuleCreateInfo const *pCreateInfo, VkShaderModule shaderModule, spv_target_env env,
                        uint32_t unique_shader_id)
        : words(), def_index(), has_valid_spirv(true), vk_shader_module(shaderModule), gpu_validation_shader_id(unique_shader_id) {
        words = PreprocessShaderBinary((uint32_t *)pCreateInfo->pCode, pCreateInfo->codeSize, env);
        BuildDefIndex();
    }

    SHADER_MODULE_STATE() : has_valid_spirv(false), vk_shader_module(VK_NULL_HANDLE), gpu_validation_shader_id(UINT32_MAX) {}

    decoration_set get_decorations(unsigned id) const {
        // return the actual decorations for this id, or a default set.
        auto it = decorations.find(id);
        if (it != decorations.end()) return it->second;
        return decoration_set();
    }

    // Expose begin() / end() to enable range-based for
    spirv_inst_iter begin() const { return spirv_inst_iter(words.begin(), words.begin() + 5); }  // First insn
    spirv_inst_iter end() const { return spirv_inst_iter(words.begin(), words.end()); }          // Just past last insn
    // Given an offset into the module, produce an iterator there.
    spirv_inst_iter at(unsigned offset) const { return spirv_inst_iter(words.begin(), words.begin() + offset); }

    // Gets an iterator to the definition of an id
    spirv_inst_iter get_def(unsigned id) const {
        auto it = def_index.find(id);
        if (it == def_index.end()) {
            return end();
        }
        return at(it->second);
    }

    void BuildDefIndex();
};

class ValidationCache {
    // hashes of shaders that have passed validation before, and can be skipped.
    // we don't store negative results, as we would have to also store what was
    // wrong with them; also, we expect they will get fixed, so we're less
    // likely to see them again.
    std::unordered_set<uint32_t> good_shader_hashes;
    ValidationCache() {}

  public:
    static VkValidationCacheEXT Create(VkValidationCacheCreateInfoEXT const *pCreateInfo) {
        auto cache = new ValidationCache();
        cache->Load(pCreateInfo);
        return VkValidationCacheEXT(cache);
    }

    void Load(VkValidationCacheCreateInfoEXT const *pCreateInfo) {
        const auto headerSize = 2 * sizeof(uint32_t) + VK_UUID_SIZE;
        auto size = headerSize;
        if (!pCreateInfo->pInitialData || pCreateInfo->initialDataSize < size) return;

        uint32_t const *data = (uint32_t const *)pCreateInfo->pInitialData;
        if (data[0] != size) return;
        if (data[1] != VK_VALIDATION_CACHE_HEADER_VERSION_ONE_EXT) return;
        uint8_t expected_uuid[VK_UUID_SIZE];
        Sha1ToVkUuid(SPIRV_TOOLS_COMMIT_ID, expected_uuid);
        if (memcmp(&data[2], expected_uuid, VK_UUID_SIZE) != 0) return;  // different version

        data = (uint32_t const *)(reinterpret_cast<uint8_t const *>(data) + headerSize);

        for (; size < pCreateInfo->initialDataSize; data++, size += sizeof(uint32_t)) {
            good_shader_hashes.insert(*data);
        }
    }

    void Write(size_t *pDataSize, void *pData) {
        const auto headerSize = 2 * sizeof(uint32_t) + VK_UUID_SIZE;  // 4 bytes for header size + 4 bytes for version number + UUID
        if (!pData) {
            *pDataSize = headerSize + good_shader_hashes.size() * sizeof(uint32_t);
            return;
        }

        if (*pDataSize < headerSize) {
            *pDataSize = 0;
            return;  // Too small for even the header!
        }

        uint32_t *out = (uint32_t *)pData;
        size_t actualSize = headerSize;

        // Write the header
        *out++ = headerSize;
        *out++ = VK_VALIDATION_CACHE_HEADER_VERSION_ONE_EXT;
        Sha1ToVkUuid(SPIRV_TOOLS_COMMIT_ID, reinterpret_cast<uint8_t *>(out));
        out = (uint32_t *)(reinterpret_cast<uint8_t *>(out) + VK_UUID_SIZE);

        for (auto it = good_shader_hashes.begin(); it != good_shader_hashes.end() && actualSize < *pDataSize;
             it++, out++, actualSize += sizeof(uint32_t)) {
            *out = *it;
        }

        *pDataSize = actualSize;
    }

    void Merge(ValidationCache const *other) {
        good_shader_hashes.reserve(good_shader_hashes.size() + other->good_shader_hashes.size());
        for (auto h : other->good_shader_hashes) good_shader_hashes.insert(h);
    }

    static uint32_t MakeShaderHash(VkShaderModuleCreateInfo const *smci);

    bool Contains(uint32_t hash) { return good_shader_hashes.count(hash) != 0; }

    void Insert(uint32_t hash) { good_shader_hashes.insert(hash); }

  private:
    void Sha1ToVkUuid(const char *sha1_str, uint8_t *uuid) {
        // Convert sha1_str from a hex string to binary. We only need VK_UUID_SIZE bytes of
        // output, so pad with zeroes if the input string is shorter than that, and truncate
        // if it's longer.
        char padded_sha1_str[2 * VK_UUID_SIZE + 1] = {};  // 2 hex digits == 1 byte
        std::strncpy(padded_sha1_str, sha1_str, 2 * VK_UUID_SIZE);

        for (uint32_t i = 0; i < VK_UUID_SIZE; ++i) {
            const char byte_str[] = {padded_sha1_str[2 * i + 0], padded_sha1_str[2 * i + 1], '\0'};
            uuid[i] = static_cast<uint8_t>(std::strtoul(byte_str, nullptr, 16));
        }
    }
};

spirv_inst_iter FindEntrypoint(SHADER_MODULE_STATE const *src, char const *name, VkShaderStageFlagBits stageBits);

// For some analyses, we need to know about all ids referenced by the static call tree of a particular entrypoint. This is
// important for identifying the set of shader resources actually used by an entrypoint, for example.
// Note: we only explore parts of the image which might actually contain ids we care about for the above analyses.
//  - NOT the shader input/output interfaces.
//
// TODO: The set of interesting opcodes here was determined by eyeballing the SPIRV spec. It might be worth
// converting parts of this to be generated from the machine-readable spec instead.
std::unordered_set<uint32_t> MarkAccessibleIds(SHADER_MODULE_STATE const *src, spirv_inst_iter entrypoint);

void ProcessExecutionModes(SHADER_MODULE_STATE const *src, const spirv_inst_iter &entrypoint, PIPELINE_STATE *pipeline);

std::vector<std::pair<descriptor_slot_t, interface_var>> CollectInterfaceByDescriptorSlot(
    SHADER_MODULE_STATE const *src, std::unordered_set<uint32_t> const &accessible_ids, bool *has_writable_descriptor);

uint32_t DescriptorTypeToReqs(SHADER_MODULE_STATE const *module, uint32_t type_id);

#endif  // VULKAN_SHADER_VALIDATION_H