File: bp_cmd_buffer_nv.cpp

package info (click to toggle)
vulkan-validationlayers 1.4.328.1-1
  • links: PTS, VCS
  • area: main
  • in suites: sid
  • size: 49,412 kB
  • sloc: cpp: 615,223; python: 12,115; sh: 24; makefile: 20; xml: 14
file content (225 lines) | stat: -rw-r--r-- 10,554 bytes parent folder | download | duplicates (7)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
/* Copyright (c) 2015-2025 The Khronos Group Inc.
 * Copyright (c) 2015-2025 Valve Corporation
 * Copyright (c) 2015-2025 LunarG, Inc.
 * Modifications Copyright (C) 2020 Advanced Micro Devices, Inc. All rights reserved.
 * Modifications Copyright (C) 2022 RasterGrid Kft.
 *
 * 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 "best_practices/best_practices_validation.h"
#include "best_practices/bp_state.h"
#include <vulkan/utility/vk_format_utils.h>

static std::array<uint32_t, 4> GetRawClearColor(VkFormat format, const VkClearColorValue& clear_value) {
    std::array<uint32_t, 4> raw_color{};
    std::copy_n(clear_value.uint32, raw_color.size(), raw_color.data());

    // Zero out unused components to avoid polluting the cache with garbage
    if (!vkuFormatHasRed(format)) raw_color[0] = 0;
    if (!vkuFormatHasGreen(format)) raw_color[1] = 0;
    if (!vkuFormatHasBlue(format)) raw_color[2] = 0;
    if (!vkuFormatHasAlpha(format)) raw_color[3] = 0;

    return raw_color;
}

static bool IsClearColorZeroOrOne(VkFormat format, const std::array<uint32_t, 4> clear_color) {
    static_assert(sizeof(float) == sizeof(uint32_t), "Mismatching float <-> uint32 sizes");
    const float one = 1.0f;
    const float zero = 0.0f;
    uint32_t raw_one{};
    uint32_t raw_zero{};
    memcpy(&raw_one, &one, sizeof(one));
    memcpy(&raw_zero, &zero, sizeof(zero));

    const bool is_one =
        (!vkuFormatHasRed(format) || (clear_color[0] == raw_one)) && (!vkuFormatHasGreen(format) || (clear_color[1] == raw_one)) &&
        (!vkuFormatHasBlue(format) || (clear_color[2] == raw_one)) && (!vkuFormatHasAlpha(format) || (clear_color[3] == raw_one));
    const bool is_zero = (!vkuFormatHasRed(format) || (clear_color[0] == raw_zero)) &&
                         (!vkuFormatHasGreen(format) || (clear_color[1] == raw_zero)) &&
                         (!vkuFormatHasBlue(format) || (clear_color[2] == raw_zero)) &&
                         (!vkuFormatHasAlpha(format) || (clear_color[3] == raw_zero));
    return is_one || is_zero;
}

bool BestPractices::ValidateZcullScope(const bp_state::CommandBufferSubState& cb_state, const Location& loc) const {
    assert(VendorCheckEnabled(kBPVendorNVIDIA));

    bool skip = false;

    if (cb_state.nv.depth_test_enable) {
        auto& scope = cb_state.nv.zcull_scope;
        skip |= ValidateZcull(cb_state, scope.image, scope.range, loc);
    }

    return skip;
}

bool BestPractices::ValidateZcull(const bp_state::CommandBufferSubState& cb_state, VkImage image,
                                  const VkImageSubresourceRange& subresource_range, const Location& loc) const {
    bool skip = false;

    const char* good_mode = nullptr;
    const char* bad_mode = nullptr;
    bool is_balanced = false;

    const auto image_it = cb_state.nv.zcull_per_image.find(image);
    if (image_it == cb_state.nv.zcull_per_image.end()) {
        return skip;
    }
    const auto& tree = image_it->second;

    auto image_state = Get<vvl::Image>(image);
    ASSERT_AND_RETURN_SKIP(image_state);

    ForEachSubresource(*image_state, subresource_range, [&](uint32_t layer, uint32_t level) {
        if (is_balanced) {
            return;
        }
        const auto& resource = tree.GetState(layer, level);
        const uint64_t num_draws = resource.num_less_draws + resource.num_greater_draws;

        if (num_draws == 0) {
            return;
        }
        const uint64_t less_ratio = (resource.num_less_draws * 100) / num_draws;
        const uint64_t greater_ratio = (resource.num_greater_draws * 100) / num_draws;

        if ((less_ratio > kZcullDirectionBalanceRatioNVIDIA) && (greater_ratio > kZcullDirectionBalanceRatioNVIDIA)) {
            is_balanced = true;

            if (greater_ratio > less_ratio) {
                good_mode = "GREATER";
                bad_mode = "LESS";
            } else {
                good_mode = "LESS";
                bad_mode = "GREATER";
            }
        }
    });

    if (is_balanced) {
        skip |= LogPerformanceWarning(
            "BestPractices-NVIDIA-Zcull-LessGreaterRatio", cb_state.Handle(), loc,
            "%s Depth attachment %s is primarily rendered with depth compare op %s, but some draws use %s. "
            "Z-cull is disabled for the least used direction, which harms depth testing performance. "
            "The Z-cull direction can be reset by clearing the depth attachment, transitioning from VK_IMAGE_LAYOUT_UNDEFINED, "
            "using VK_ATTACHMENT_LOAD_OP_DONT_CARE, or using VK_ATTACHMENT_STORE_OP_DONT_CARE.",
            VendorSpecificTag(kBPVendorNVIDIA), FormatHandle(cb_state.nv.zcull_scope.image).c_str(), good_mode, bad_mode);
    }

    return skip;
}

static constexpr std::array<VkFormat, 12> kCustomClearColorCompressedFormatsNVIDIA = {
    VK_FORMAT_R8G8B8A8_UNORM,           VK_FORMAT_B8G8R8A8_UNORM,           VK_FORMAT_A8B8G8R8_UNORM_PACK32,
    VK_FORMAT_A2R10G10B10_UNORM_PACK32, VK_FORMAT_A2B10G10R10_UNORM_PACK32, VK_FORMAT_R16G16B16A16_UNORM,
    VK_FORMAT_R16G16B16A16_SNORM,       VK_FORMAT_R16G16B16A16_UINT,        VK_FORMAT_R16G16B16A16_SINT,
    VK_FORMAT_R16G16B16A16_SFLOAT,      VK_FORMAT_R32G32B32A32_SFLOAT,      VK_FORMAT_B10G11R11_UFLOAT_PACK32,
};

void BestPractices::RecordClearColor(VkFormat format, const VkClearColorValue& clear_value) {
    assert(VendorCheckEnabled(kBPVendorNVIDIA));

    const std::array<uint32_t, 4> raw_color = GetRawClearColor(format, clear_value);
    if (IsClearColorZeroOrOne(format, raw_color)) {
        // These colors are always compressed
        return;
    }

    const auto it =
        std::find(kCustomClearColorCompressedFormatsNVIDIA.begin(), kCustomClearColorCompressedFormatsNVIDIA.end(), format);
    if (it == kCustomClearColorCompressedFormatsNVIDIA.end()) {
        // The format cannot be compressed with a custom color
        return;
    }

    // Record custom clear color
    WriteLockGuard guard{clear_colors_lock_};
    if (clear_colors_.size() < kMaxRecommendedNumberOfClearColorsNVIDIA) {
        clear_colors_.insert(raw_color);
    }
}

bool BestPractices::ValidateClearColor(VkCommandBuffer commandBuffer, VkFormat format, const VkClearColorValue& clear_value,
                                       const Location& loc) const {
    assert(VendorCheckEnabled(kBPVendorNVIDIA));

    bool skip = false;

    const std::array<uint32_t, 4> raw_color = GetRawClearColor(format, clear_value);
    if (IsClearColorZeroOrOne(format, raw_color)) {
        return skip;
    }

    const auto it =
        std::find(kCustomClearColorCompressedFormatsNVIDIA.begin(), kCustomClearColorCompressedFormatsNVIDIA.end(), format);
    if (it == kCustomClearColorCompressedFormatsNVIDIA.end()) {
        // The format is not compressible
        std::string format_list;
        for (VkFormat compressed_format : kCustomClearColorCompressedFormatsNVIDIA) {
            if (compressed_format == kCustomClearColorCompressedFormatsNVIDIA.back()) {
                format_list += "or ";
            }
            format_list += string_VkFormat(compressed_format);
            if (compressed_format != kCustomClearColorCompressedFormatsNVIDIA.back()) {
                format_list += ", ";
            }
        }

        skip |= LogPerformanceWarning("BestPractices-NVIDIA-ClearColor-NotCompressed", commandBuffer, loc,
                                      "%s Clearing image with format %s without a 1.0f or 0.0f clear color. "
                                      "The clear will not get compressed in the GPU, harming performance. "
                                      "This can be fixed using a clear color of VkClearColorValue{0.0f, 0.0f, 0.0f, 0.0f}, or "
                                      "VkClearColorValue{1.0f, 1.0f, 1.0f, 1.0f}. Alternatively, use %s.",
                                      VendorSpecificTag(kBPVendorNVIDIA), string_VkFormat(format), format_list.c_str());
    } else {
        // The format is compressible
        bool registered = false;
        {
            ReadLockGuard guard{clear_colors_lock_};
            registered = clear_colors_.find(raw_color) != clear_colors_.end();

            if (!registered) {
                // If it's not in the list, it might be new. Check if there's still space for new entries.
                registered = clear_colors_.size() < kMaxRecommendedNumberOfClearColorsNVIDIA;
            }
        }
        if (!registered) {
            std::string clear_color_str;

            if (vkuFormatIsUINT(format)) {
                clear_color_str = std::to_string(clear_value.uint32[0]) + ", " + std::to_string(clear_value.uint32[1]) + ", " +
                                  std::to_string(clear_value.uint32[2]) + ", " + std::to_string(clear_value.uint32[3]);
            } else if (vkuFormatIsSINT(format)) {
                clear_color_str = std::to_string(clear_value.int32[0]) + ", " + std::to_string(clear_value.int32[1]) + ", " +
                                  std::to_string(clear_value.int32[2]) + ", " + std::to_string(clear_value.int32[3]);
            } else {
                clear_color_str = std::to_string(clear_value.float32[0]) + ", " + std::to_string(clear_value.float32[1]) + ", " +
                                  std::to_string(clear_value.float32[2]) + ", " + std::to_string(clear_value.float32[3]);
            }

            skip |= LogPerformanceWarning(
                "BestPractices-NVIDIA-ClearColor-NotCompressed", commandBuffer, loc,
                "%s Clearing image with unregistered VkClearColorValue{%s}. "
                "This clear will not get compressed in the GPU, harming performance. "
                "The clear color is not registered because too many unique colors have been used. "
                "Select a discrete set of clear colors and stick to those. "
                "VkClearColorValue{0, 0, 0, 0} and VkClearColorValue{1.0f, 1.0f, 1.0f, 1.0f} are always registered.",
                VendorSpecificTag(kBPVendorNVIDIA), clear_color_str.c_str());
        }
    }

    return skip;
}